2.

0
28.0
Laboratorio Bases Moleculares de la Proliferacin Celular
Departamento de Biologa Molecular y Bioqumica
Facultad de Ciencias
Universidad de Mlaga
Tesis Doctoral
Aproximaciones computacionales, dinmicas y
topolgicas al estudio del metabolismo
Ral Montaez Martnez
Mlaga, 9 de marzo de 2010
2.0
28.0
Aproximaciones computacionales, dinmicas y
topolgicas al estudio del metabolismo
Memoria presentada para optar al grado de Doctor por la
Universidad de Mlaga
Fdo.: Ral Montaez Martnez
Directores

Fdo.: Franscisca M Snchez Jimnez
Fdo.: Miguel ngel Medina Torres
Fdo.: Carlos F. Rodrguez Caso

2.0
28.0
Laboratorio Bases Moleculares de la Proliferacin Celular
Departamento de Biologa Molecular y Bioqumica
Facultad de Ciencias
Universidad de Mlaga
Francisca M Snchez Jimnez, Catedrtica de Bioqumica y Biologa Molecular de la
Universidad de Mlaga, Miguel ngel Medina Torres Catedrtico de Bioqumica y
Biologa Molecular de la Universidad de Mlaga y Carlos F. Rodrguez Caso,
Investigador Post-doctoral de la Universitat Pompeu Fabra,
CERTIFICAN
Que Ral Montaez Martnez, Licenciado en Biologa por la Universidad de Mlaga,
ha realizado bajo nuestra direccin el trabajo de investigacin correspondiente a su
Tesis Doctoral que lleva por ttulo Aproximaciones computacionales, dinmicas y
topolgicas al estudio del metabolismo.
Una vez revisado el trabajo estimamos que puede ser presentado ante el tribunal
encargado de juzgarlo.
Para que conste a efecto de lo establecido en el Artculo 8 del Real Decreto 778/1998,
regulador de los estudios de Tercer Ciclo- Doctorado, AUTORIZAMOS la
presentacin de esta Tesis Doctoral en la Universidad de Mlaga.
Mlaga, enero de 2010
Fdo.: Francisca M Snchez Jimnez Fdo.: Miguel ngel Medina Torres Fdo.: Carlos F. Rodrguez Caso
2.0
28.0
Este trabajo ha sido nanciado por el Programa de Formacin del Profesorado
Universitario (FPU) del actual Ministerio de Educacin, los proyectos P07-
CVI-02999, CVI-657 y el grupo BIO-267 de la Junta de Andaluca, el proyecto SAF
2005-01812 y SAF 2008-02522 del Ministerio de Ciencia e Innovacin, la Fundacin
Ramn Areces, EU 6th framework project ComplexDis (NEST-043241, CRC) y el
CIBER de Enfermedades Raras. CIBERER es una iniciativa del ISCIII (Espaa).
Agradecimientos
La gratitud, virtud o hipocresa?. La gratitud debera ser esa reconfortante
sensacin de saber que hay alguien que te acompaa y apoya, que confa en ti
ciegamente y te cuida. Lamentablemente, la realidad es otra, la gratitud es con
demasiada frecuencia la ofrenda a la falsa cordialidad que nos permite vivir en
sociedad. Confo en ser lo suficientemente hipcrita como para que nadie note
la diferencia.
Quiero agradecer en primer lugar a las personas que me acogieron en sus
grupos durante las dos estancias que he realizado y que me han permitido
sumergirme un poco ms en el ocano del conocimiento. Dr. Ramnek, Thomas,
Sren, Tejal, y dems, gracias por vuestra amabilidad y dedicacin durante los
meses que pas en el Center of Biological Sequences, esos tres meses a
vuestro lado me han enseado que hay otra manera de hacer ciencia. A vosotros,
Ricar, Bernat, Sergi, Andrea, Javier y dems integrantes del Laboratorio de
Sistemas Complejos muchsimas gracias. Cada caf, cada almuerzo, cada cigarro
a vuestro lado han sido fascinantes y divertidos y con ellos habis reavivado una
curiosidad que olvid en algn bolsillo tiempo atrs. He disfrutado mucho a
vuestro lado.
A mis directores, a todos ellos los habra ahogado en un cubo un par de
veces durante estos aos. Por suerte el litio y la lobotoma frontal me han
mantenido estable.
Kikita hija, Gracias a ti he aprendido, tal vez, lo ms desagradable, lo que nadie
quiere mostrar, me has enseado que la ciencia no es una sociedad idlica en la
que todos cooperan en pro del conocimiento, me has enseado que no siempre
es bueno ir de cabeza y a pecho descubierto, que este mundo es sibilino y los
francos son presa fcil. Me has ayudado, guiado y defendiendo cada una de mis
ideas. Gracias por creer en mi.
Miguel ngel, tras tu aparente frialdad e inexpresividad hay una persona buena.
De ti he aprendido la otra cara, la cara hermosa de la ciencia, la ciencia
platnica, la que no es ms que un juego, la de sistemas auto-organizados, la
que no obvia y sabe que cada parte es importante, la que es bella por ser
inalcanzable. Muchas gracias Miguel ngel por acercarme a un mundo tan
maravilloso.
Carlos, me enseaste cuando eras compaero y me rescataste cuando me
encontr perdido. Has sido mentor y eres amigo, me has escuchado y
2.0
28.0
aconsejado, en la ciencia y en la vida, y me has hecho rer en los momentos
difciles. Me has enseado que sin preguntas la ciencia es slo un mero
desarrollo tecnolgico. Muchas gracias Carlos.
Dani, el director de mi tesina y quien me brind el contexto para hacer mi
primera aportacin escrita a la ciencia, ese impulso fue definitivo Dani, muchas
gracias.
Con respecto a los restantes jefes, quiero empezar por JL, he conocido
pocas personas tan nobles y desinteresadas, muchas gracias por estar hay
siempre. Ignacio, capaz de hacerse odiar en un instante y arrancarte una sonrisa
al siguiente, cido y elegante, metdico como un obsesivo compulsivo, gracias
por soportarme. Juan Antonio, tu eres jefe o no?. Bueno, seas lo que seas,
gracias por dedicarme tu escaso tiempo en las mltiples charlas que hemos
tenido.
En cuanto a mis compaeros, Luisio y Javi el rubio, juntos fuimos el
grandioso Imperio Redox capaz colonizar el laboratorio en un instante y de
reducirse a una caja de cartn en el siguiente, nos hemos redo y hemos
sufrido largas horas de cinticas junto al Shimatsu. A vosotros, a mis
primeros compaeros de ciencia y otras muchas cosas, gracias.
Alejandro, la bondad tentada por el lado ms oscuro del comportamiento
humano, sin ti este trabajo que se presenta no habra sido posible. Armando, el
chico del cerebro inquieto, muchas gracias a ambos por acompaarme en el da
y en la noche haciendo del paso del tiempo algo ms divertido.
Hicham, tu generosidad y buen corazn han conseguido que un xenfobo
portero de discoteca ahora lamente alejarse de ti. Muchas gracias por tu
amistad y por cuidar de Catalina.
Aurelio, Almudena Ian, Gianni, Esther, Flor Beatriz, Luis Gustavo, Lucas,
Casimiro, Auxi, Pepe, Ana, Mara, Mara Victoria (una mujer a la que admiro
sinceramente), Mara Jesus, Javi, Melissa, Germn y dems, vosotros conformis
el contexto, ayudis a que el grupo sea lo que es y a que sea agradable
pertenecer a el y acudir cada maana. Me he divertido y aprendido mucho a
vuestro lado. Muchas gracias, gracias por vuestras enseanzas y por llegar a
considerarme casi como un amigo. Muchas gracias a todos.
Pero si realmente hay personas que han hecho posible este trabajo son los
que me han generado, en primer lugar como ente fsico y posteriormente
conformando la persona que soy, esas personas son mi pequea familia.
Gracias madre por ensearme que las caadas reales no son el nico camino,
que aunque el rebao da calor y acompaa, solo pace y camina. Por ensearme a
2.0
28.0
ser critico con la realidad y a no conformarme. Por ensearme que el
conocimiento y las ideas propias son las que te hacen libre.
Gracias padre por ensearme a sentirme orgulloso del trabajo bien hecho.
Gracias Sergio, mi hermano, mi hermano mayor, mi referencia. Tu cuidaste de
mi y me llevaste de la mano por una infancia difcil sin separarte nunca.
Gracias Dama, mi hermana, mi pequea hermana, fortaleza y ternura en dosis
letales. Te quiero pequea.
Gracias Rubn, Adam, Daniela Mara y Mario, en diferentes etapas de mi vida
habis sido y sois aun la infancia renovada y las ganas de jugar.
Gracias a mis tos, Manolo y Fali, me habis cuidado como a un hijo y al
igual que mi madre me habis enseado a ser yo mismo.
Gracias Bo, el espalda plateada, el abuelo, aun recuerdo tu mano en mi
cabeza mientras caminabas despacio, tu entereza y tu ganas de vivir.
Gracias, Karin, Fali, Krishna, y Concha por cuidar de las personas a las que
quiero y hacerlos felices.
Finalmente, agradecer a la persona que me cuida a diario y me ha mostrado,
quizs lo ms importante, cmo se ha de querer a una persona. Gracias
Catalina.
2.0
28.0
2.0
28.0
A mi madre
El da que nos cansemos de ser,
seremos simplemente uno ms

2.0
28.0
Prembulo
2.0
28.0
Puede parecer dramtico pensar que
cada uno de nuestros pasos no es ms que
una gota en el mar, mas esa es la verdad.
Este trabajo no ser ms que una efmera
aportacin al conocimiento. Pero sin gotas
no habra mar.
Siempre me ha maravillado pensar en el
devenir del universo. Cmo las partculas
subatmicas burlaron el estado de mxima
entropa formando tomos. Cmo los to-
mos colapsaron en estrellas y cmo estas
enriquecieron el universo de nuevos to-
mos hasta generar entidades fsicas tan
sorprendentes como los seres vivos. Curiosa
inconsistencia con la inercia entrpica que
determina el sentido del tiempo.
The irreversibility of time is the mecha-
nism that brings order out of chaos.
Ilya Prigogine
En el interior de una clula la materia y la
energa son transformadas y ordenadas
generando vida. La periodicidad, intensi-
dad y composicin de esta materia y
3
2.0
28.0
Prembulo
Cada uno de nuestros pasos es el inicio de un insondable
viaje atravesando el jardn de Borges hacia la singular cer-
teza del equilibrio.
energa es interpretada por la clula reajus-
tando su composicin y variables de esta-
do, apresurndose a cambiar para mante-
nerse constante. Es sorprendente cmo la
vida elude incesantemente el equilibrio,
incrementando la entropa de su entorno
para generar su propio orden gracias a los
miles de procesos fsico-qumicos que se
dan en el interior y la frontera de las clu-
las.
Para favorecer su comprensin, la espesa
trama de procesos interconectados fue
cortada por la navaja de Occam. Se defi-
nieron y se estudiaron as el metabolismo,
los sistemas de biosealizacin, los procesos
de replicacin, reparacin, transcripcin y
traduccin del ADN, etc. Pero, ms all de
nuestra propia necesidad de comprensin,
tal frontera no existe y en ocasiones es dif-
cil discernir un lmite claro entre cada uno
de estos mdulos funcionales.
Lo que pretend, en origen, con el trabajo
que mostrar a continuacin es alcanzar
un mayor nivel de comprensin de la di-
nmica de la vida. Una gran meta, un r-
bol de elucubraciones propias de un nefi-
to en el quehacer cientfico. Inevitable-
mente la realidad ha de acotarse en ma-
yor o menor grado. Necesitamos definir
sistemas reales cuyos sistemas conceptua-
les y resultados derivados puedan abarcar-
se empleando nuestra limitada capacidad
de comprensin.
Estos cuatro aos de trabajo me han
ayudado a comprender que la necesidad
de acotar y definir un sistema no ha de
conducirnos necesariamente al estricto
reduccionismo. Que, con las abstracciones
apropiadas, la madeja de incertidumbre
puede ir desenredndose poco a poco.
El trabajo de mi tesis se inici ya con una
interpretacin holista del quehacer cientfi-
co, aunque en origen supeditado a la tra-
yectoria histrica y el mbito de conoci-
miento del grupo en el que me integraba.
El laboratorio de Proliferacin Celular del
Departamento de Biologa Molecular y
Bioqumica de la Universidad de Mlaga
era por aquel entonces un laboratorio cen-
trado en el estudio del metabolismo y fun-
ciones biolgicas de las aminas bigenas
en clulas animales proliferantes.
El grupo parta de un enfoque puramente
experimental y reduccionista. Enfoque que
permiti la caracterizacin de propiedades
de diversos elementos de estas rutas meta-
blicas: ornitina descarboxilasa (ODC, una
de las enzimas limitantes de la biosntesis de
poliaminas) y su regulacin por aminas,
histidina descarboxilasa (HDC, enzima res-
ponsable de la biosntesis de histamina),
modos preferenciales de unin entre ami-
nas y cidos nucleicos, sistemas de trans-
porte de aminas y respuestas fenomenol-
gicas del metabolismo de aminas en mo-
delos celulares y animales donde estos
compuestos tienen un papel fisiopatolgi-
co relevante [10-12]. Pero en el seno del
grupo se senta ya la necesidad de un en-
foque ms amplio.
Como refleja la cronologa de los artcu-
los que dan soporte a esta tesis, mi trabajo
comenz con la consolidacin de un estu-
dio ya iniciado por uno de mis actuales
directores de Tesis en el que se pretenda
hacer una abstraccin matemtica que
caracterizase la dinmica de un sistema
compuesto por las principales enzimas que
conforman el biciclo de las poliaminas y
caracterizar las enzimas y metabolitos rele-
vantes en el control de su dinmica. En
aquel trabajo se daban la mano antiguas
disciplinas, que hoy parecen olvidadas ba-
jo la sobretecnificacin de la actual biolo-
ga molecular, tales como la enzimologa
[3], la teora general de sistemas [1] o inclu-
so la ciberntica [9], para poner de mani-
fiesto que el metabolismo pueden ser re-
ducido a un sistema de ecuaciones capaz
no slo de reproducir, sino de predecir su
comportamiento. En nuestro caso, repro-
ducamos el comportamiento del metabo-
lismo de poliaminas en ratones transgni-
cos y el efecto de las principales drogas
empleadas en esta ruta y fuimos capaces
de predecir la dependencia en la activi-
Ral Montaez Martnez
4
2.0
28.0
dad del ciclo de las poliaminas de la dis-
ponibilidad de S-adenosilmetionina y de
acetil-CoA [Captulo 3.1], prediccin que
fue corroborada despus [4].
Este modelo fue mi primera aproximacin
a la abstraccin de un sistema biolgico y
su implementacin en un programa infor-
mtico. Adems de la experiencia meto-
dolgica, los mltiples intentos de ajustar
los parmetros de las diferentes variables se
convirtieron en la comprobacin de facto
de que incluso un micro-sistema como era
el modelo de las poliaminas poda mostrar
un elevado nmero de correlaciones de
larga distancia, lo que despert mi curiosi-
dad por los sistemas complejos. De este
modo, el modelo de las poliaminas se mos-
tr no slo como una importante expe-
riencia procedimental sino tambin filosfi-
ca.
Todo esto reforzado por la esperanza en
una emergente metodologa que pareca
consolidarse poco a poco: la Biologa de
Sistemas se convirti en la base de una
apuesta mucho ms amplia en la que se
intent ir incrementando la abstraccin
inicial de modo modular. Si fuera cierta la
propuesta segn la cual el metabolismo es
un sistema modular y cada uno de los m-
dulos una unidad funcional independiente
[15], podra reconstruirse un modelo cinti-
co cada vez mayor mediante la adicin
recurrente de modelos construidos y vali-
dados por separado. Para tal fin, se gene-
r un nuevo modelo para el catabolismo
de la arginina [Captulo 3.2] y se readapta-
ron modelos existentes [14, 17]. Los resulta-
dos de esta integracin se presentan de
modo resumido en la seccin de posibili-
dades futuras del captulo 3.
Se pueden mencionar muchas dificulta-
des en el camino trazado, cuyos intentos
de solucin han constituido buena parte
del entrenamiento y la experiencia adqui-
ridos:

La recopilacin de la informacin ne-

cesaria para comprender el conjunto de
elementos implicados en cada uno de los
subsistemas y las relaciones entre ellos, in-
cluyendo la caracterizacin de las dinmi-
cas de cada una de las enzimas, los modi-
ficadores de su actividad y la determina-
cin de los parmetros cinticos necesarios
para acotar el dominio de esas funciones.

La consolidacin de la necesidad de
aplicar aproximaciones realmente holistas,
que contemplasen en su conjunto la ele-
vada complejidad de los sistemas biolgi-
cos.

La creciente acumulacin en bases de

datos de informacin no estructurada ni
consensuada que entorpece la localiza-
cin y aprovechamiento eficiente de la
informacin.
Como apoyo tecnolgico indispensable
para hacer nuestra tarea ms eficiente, se
requeran nuevas herramientas de recopi-
lacin y anlisis de la informacin [Captulo
1.1]. En los ltimos aos, Internet se ha con-
solidado como el principal repositorio de
informacin biolgica proveniente de las
tcnicas de anlisis de alto rendimiento.
Esto inspir la idea de desarrollar un sistema
autnomo capaz de aplicar un criterio ex-
perto. Para la mayora de los usuarios, esta
informacin es accesible manualmente
mediante entornos de interfaces webs.
Sin embargo, la flexibilidad y accesibilidad
de las consultas manuales puede ser una
traba cuando pretende conectarse a ellas
una aplicacin software automatizada.
En este punto de encuentro era necesa-
rio hallar el soporte tecnolgico ms apro-
piado para hacer viable la idea (que pu-
diera tildarse de ambiciosa) de desarrollar
un sistema autnomo capaz de generar las
consultas apropiadas con las que recupe-
rar la informacin requerida.
Gracias a la paciente colaboracin y
asesoramiento de Jos Aldana Montes,
investigador principal del grupo KHAOS
(adscritos al Departamento de Lenguajes y
Ciencias de la Computacin, de la Univer-
5
Prembulo
2.0
28.0
sidad de Mlaga), se puso en marcha el
proyecto de un mediador semntico me-
diante el cual se facilitara la recuperacin
de informacin biolgica en muy diferentes
reas.
El principal propsito de la mediacin
semntica consiste precisamente en reali-
zar consultas automticas sobre los conte-
nidos almacenados en internet. Para tal fin
muchas de las bases de datos o de cono-
cimiento estructuran su informacin acorde
a un patrn unvoco resultado del consen-
so entre consorcios web, haciendo expl-
cita la semntica del contenido para un
dominio concreto de conocimiento.
En mi caso, particip previamente en una
experiencia piloto en la que se gener una
aplicacin software con la que recuperar
la informacin referente a las interacciones
entre protenas, (PROTOPIA) [16] cuyas ba-
ses procedimentales se plasmaron en el
artculo presentado en [Captulo 1.3]. El
tiempo dedicado a este trabajo nos puso
en antecedentes acerca de la rpida ob-
solescencia de este tipo de herramientas,
as como de las deficiencias de buena par-
te de las bases de datos a nuestro alcance
en aquel momento: un sustancial grado de
incompletitud en la informacin, falta de
consenso en la nomenclatura, referencias
cruzadas incorrectas, redundancia, cadu-
cidad, informacin no validada, etc. Para
solventar este importante escollo, se plan-
te la posibilidad de emplear contextuali-
zaciones del conocimiento preexistentes,
tales como Gene Ontology, Tambis, Cell
Ontology o Proten Ontology. El inconve-
niente que planteaban estos contextos de
conocimiento era su pobreza de relacio-
nes, al consistir en jerarquizaciones por sub-
suncin de conceptos. Era evidente que
deba generar mi propio andamio concep-
tual, una ontologa que definiese, de modo
concreto y explcito, el contexto de cono-
cimiento en el que pretenda trabajar.
Defino as AMMO (The AMine Metabolism
Ontology), una ontologa que representa
los elementos claves del metabolismo, su
regulacin, su dinmica, llegando incluso a
su estructura tridimensional. En esta ontolo-
ga los nodos son los conceptos implicados
en el dominio de conocimiento de inters
para mi trabajo, los arcos no identificados
son las relaciones jerrquicas que estructu-
ran la informacin, siendo los identificados
los que definen el modo en que los con-
ceptos se relacionan entre s.
AMMO era slo el soporte sobre el que
nuestro conocimiento biolgico cobraba
orden y sentido. Como toda ontologa, esta
debe ser fruto del consenso de opiniones y
ser aprobada por una comunidad de ex-
pertos. De este modo, mientras AMMO se
enriqueca con las aportaciones de la co-
munidad cientfica, nosotros debamos
comprobar su validez y viabilidad para
nuestros propsitos.
La ontologa permita que, de modo or-
denado, un sistema de gestin de consul-
tas determinase qu informacin era re-
querida para dar una respuesta lo suficien-
temente completa a una consulta. Ahora
tocaba conectar los diferentes conceptos
y relaciones contenidos en la ontologa al
mundo de conocimiento de las bases de
dat os. Se i ni ci a as el pr oceso de
mapping, operacin que asocia cada
elemento de un conjunto (el dominio) con
uno o ms elementos de un segundo con-
junto (su rango). En nuestro caso, los ele-
mentos y relaciones de la ontologa eran
correlacionados con campos o hiperenla-
ces concretos de las bases de datos, as
como con los conceptos equivalentes en
otras ontologas. De este modo, el conjunto
de nuestra ontologa, junto con el conjunto
de ontologas con conceptos correlacio-
nados con la nuestra, los campos y relacio-
nes de las bases de datos, y dos ontologas
nter-relacionadas que describen las se-
mnticas del conjunto, constituyen el direc-
torio semntico de nuestra aplicacin. Los
metadatos contenidos en el directorio se-
mntico pueden ser gestionados por un
repertorio de herramientas que van desde
simples parsers para archivos en formato
Ral Montaez Martnez
6
2.0
28.0
OWL (Web Ontology Language Reference)
hasta elaborados razonadores ontolgicos.
Basndonos en esta potente herramienta
de gestin de la informacin y el contexto
de conocimiento, planteamos la posibili-
dad de contrastar su funcionalidad desa-
rrollando una aplicacin capaz de aportar
el mejor modelo molecular posible de las
principales protenas implicadas en el me-
tabolismo de poliaminas [Captulo 2.1]. El
xito conseguido con esta pequea prue-
ba controlada permiti ir un paso ms all
y afrontar la tarea original para la que se
haba planteado la aplicacin.
Comienzo as a planificar y dar forma a lo
que ser el System Biology Metabolic Mo-
deling Assistant (SBMM) [Captulo 2.2], una
herramienta capaz de facilitar la genera-
cin de modelos metablicos cinticos
exportables en SBML y compatible, por tan-
to, con distintas aplicaciones matemticas
y de representacin de modelos metabli-
cos de uso estandarizados. Para adaptar
nuestros recursos a la generacin del
SBMM, refin los conceptos implicados en
la ontologa y las relaciones entre estos pa-
ra que se ajustasen mucho mejor a nuestras
necesidades y fuesen coherentemente
gestionadas por el razonador de ontologas
sin posibilidades de ambigedad. Junto a
mi compaero Armando Reyes Palomares,
revis las correlaciones ms correctas con
las bases de datos. Con la ayuda de mi
amigo Wenceslao Vereda, diseamos el
entorno de la aplicacin a la que finalmen-
te dio forma e integr con el mediador mi
genial amigo y compaero Alejandro del-
Real.
Durante esta singladura, continu inda-
gando sobre los sistemas complejos, asis-
tiendo a congresos de Biologa de Sistemas
y comprobando que, como en todo aflo-
ramiento ecolgico, la abundancia de re-
cursos haba propiciado el desplazamiento
de los especialistas por los generalistas. La
Biologa de Sistemas era por muchos en-
tendida como Biologa Molecular a gran
escala o trabajos puramente computacio-
nales o de ingeniera.
Siempre he sido, quizs, demasiado crti-
co y consider que deba de manifestarse
que ese proceder no era el apropiado y
que se requera una visin real de los siste-
mas biolgicos como sistemas integrados.
Slo de este modo se procedera y se
avanzara hacia una autentica Biologa de
Sistemas.
De este modo, aprovechando una invi-
tacin de Biochemical Society Transactions,
convenc a uno de mis directores, la Dra.
Snchez-Jimnez, acerca de la necesidad
de plasmar esa opinin. Surge de este mo-
do la primera de mis aportaciones concep-
tuales al contexto del conocimiento cient-
fico en la que se planteaba una interpre-
tacin de lo que debera ser una autntica
aproximacin sistmica a un problema bio-
lgico [1.1]. La segunda llegara un ao
ms tarde, con la colaboracin de mi pri-
mer mentor oficial, el Dr. Medina. En sta se
haca una vez ms referencia a la necesi-
dad de aplicar la Biologa de Sistemas al
tema de las poliaminas, pero adems se
plasmaba de un modo ms crtico la err-
nea concepcin de la Biologa de Sistemas
y la necesidad de una revisin de la misma
[1.2].
Como es evidente, he intentado que mi
trabajo se desarrollase acorde a mi con-
cepcin de cmo haba de enfrentarse un
problema biolgico. De ese modo, poco a
poco, me fui alejando del metabolismo de
las aminas bigenas para ir teniendo una
visin ms amplia del metabolismo y de su
funcionamiento integrado.
Todo este trabajo me permiti aprender
acerca de la estructura de las bases de
datos, entender sus estructuras relacionales
y familiarizarme con las herramientas con
las que poder gestionar su informacin. De
ese modo, la informacin se desplegaba
ante m como un extenso abanico de po-
sibilidades y preguntas.
7
Prembulo
2.0
28.0
Las diferentes lecturas y el asesoramiento
de mi director ms joven, el Dr. Rodrguez-
Caso, coetneo mo, favorecieron que
tendiese a integrar toda esa informacin
en grafos. Al enfocar mi atencin a la teo-
ra de grafos, descubr nuevas inconsisten-
cias en el modo de proceder de la comu-
nidad cientfica. En primer lugar, los siste-
mas biolgicos eran muchas veces abstra-
dos por ingenieros o fsicos, los cuales por
tradicin dominan el soporte conceptual y
de procedimiento de la metodologa, pero
frecuentemente caen en el error de des-
cuidar o sobresimplificar el problema que
pretenden afrontar. Por otro lado, los bilo-
gos desconocamos el sentido y las impli-
caciones que estos ingenieros y fsicos ex-
traan de sus anlisis. Ambas circunstancias
han favorecido que la comunidad cientfi-
ca adoptase un importante nmero de
generalizaciones, basadas en una topolo-
ga comn, sin tan siquiera plantearse qu
significado tenan. Este problema se ha
planteado por diferentes autores, incluidos
nosotros [Captulo 4.1][2, 5, 6, 8, 13].
Dos ejemplos de generalizaciones son:
1) Las distribuciones de ley de potencia
negativa. Estas distribuciones y su justifica-
cin mediante el modelo de unin prefe-
rencial se han asumido como una ley que
se ha de cumplir y que cuando no se cum-
ple es justificada por la incompletitud de los
datos en lugar de asumir que falla el mode-
lo [6]. Se han consolidado como la cons-
triccin esencial que ha guiado la evolu-
cin y se ha empleado para generar mlti-
ples modelos de evolucin [7].
2) El concepto de robustez o tolerancia a
errores, aplicado a las redes biolgicas sin
la consideracin espacio-temporal propia
de los organismos vivos. La robustez del
proteoma, aplicada a una red general en
la que todas las protenas sintetizables por
un organismo y capaces de interaccionar
son valoradas, ignora la expresin diferen-
cial de genes en las diferentes fases celula-
res y fisiolgicas de los organismos, la com-
partimentacin o el nmero limitado de
dominios de unin de las protenas. En el
caso del metabolismo, adicionalmente, se
ignora que para suprimir un determinado
metabolito, el nmero de reacciones que
se deben silenciar (aun ms en el caso de
metabolitos altamente conectados) puede
alcanzar el centenar y evidentemente el
organismo no ser viable. Si la valoracin
se hace sobre reacciones, protenas enzi-
mticas o genes metablicos, el problema
es similar, ya que los hubs surgen de un
artefacto introducido por un proceso ma-
temtico llamado proyeccin. El proceso
de proyeccin es aquel por el cual una red
naturalmente bipartita es colapsada a una
red con solo uno de los dos tipos de nodos,
conectados stos por los nodos comunes
entre pares.
3) Por otro lado, los parmetros que ca-
racterizan a las redes unipartitas se asumen
como parmetros capaces de caracterizar
cualquier tipo de red. Sin embargo, la in-
certidumbre en las abstracciones introdu-
cidas por las diferencias entre redes basa-
das en reacciones, en enzimas, segn c-
digo EC, en protenas con actividad enzi-
mtica o en genes, condiciona notable-
mente las caractersticas globales de los
grafos generados.
Intent solventar todas estas incertidum-
bres con el empleo de una abstraccin en
la que el metabolismo era modelado em-
pleando ambos tipos de nodos, reacciones
y metabolitos. Aplicando esta aproxima-
cin he podido no slo evidenciar algunos
de los artefactos antes mencionados, sino
que he podido alcanzar el propsito inicial
de comprender un poco mejor cmo fun-
ciona el metabolismo.
Los! periodos de formacin acadmica
son y deben ser finitos, aunque a menu-
do! tan ficticios como las fronteras de co-
nocimiento o las separaciones del metabo-
lismo en vas metablicas y, adems,! suele
ocurrir que en un! tema de investigacin
que permanece vivo, cuanto ms se traba-
ja ms cuestiones suelen abrirse y que en
Ral Montaez Martnez
8
2.0
28.0
definitiva, como deca Mayor-Zaragoza hay que
"emprender
!y reemprender
por las soadas
sendas...
sabiendo!
que siempre
todo
queda inacabado"
("Sentir". F. Mayor-Zaragoza, Terral, Ed. Litoral)
9
Prembulo
2.0
28.0
Bibliografa
1.! Bertalanffy. An Outline of General System Theory. British Journal for the Philosophy of Science.
1950;1: 139-164.
2.! Doyle JC, Alderson DL, Li L, Low S, Roughan M, Shalunov S, Tanaka R, and Willinger W. The
"robust yet fragile" nature of the Internet. Proc Natl Acad Sci U S A. 2005;102(41): 14497-
14502.
3.! Fell DA. Enzymes, metabolites and uxes. J Exp Bot. 2005;56(410): 267-272.
4.! Jell J, Merali S, Hensen ML, Mazurchuk R, Spernyak JA, Diegelman P, Kisiel ND, Barrero C,
Deeb KK, Alhonen L, Patel MS, and Porter CW. Genetically altered expression of spermidine/s-
permine N1-acetyltransferase affects fat metabolism in mice via acetyl-CoA. J Biol Chem.
2007;282(11): 8404-8413.
5.! Keller EF. Revisiting "scale-free" networks. Bioessays. 2005;27(10): 1060-1068.
6.! Khanin R and Wit E. How scale-free are biological networks. J Comput Biol. 2006;13(3): 810-
818.
7.! Light S, Kraulis P, and Elofsson A. Preferential attachment in the evolution of metabolic net-
works. BMC Genomics. 2005;6: 159.
8.! Lima-Mendez G and van Helden J. The powerful law of the power law and other myths in net-
work biology. Mol Biosyst. 2009;5(12): 1482-1493.
9.! Maturana H and Varela F, The Tree of Knowledge 1998, Boston.: Shambhala Press, .
10.!Medina MA, Quesada AR, Nunez de Castro I, and Sanchez-Jimenez F. Histamine, polyamines,
and cancer. Biochem Pharmacol. 1999;57(12): 1341-1344.
11.!Medina MA, Urdiales JL, Rodriguez-Caso C, Ramirez FJ, and Sanchez-Jimenez F. Biogenic
amines and polyamines: similar biochemistry for different physiological missions and biomedical
applications. Crit Rev Biochem Mol Biol. 2003;38(1): 23-59.
12.!Moya-Garcia AA, Medina MA, and Sanchez-Jimenez F. Mammalian histidine decarboxylase:
from structure to function. Bioessays. 2005;27(1): 57-63.
13.!Newman ME and Park J. Why social networks are different from other types of networks. Phys
Rev E Stat Nonlin Soft Matter Phys. 2003;68(3 Pt 2): 036122.
14.!Nijhout HF, Reed MC, Budu P, and Ulrich CM. A mathematical model of the folate cycle: new
insights into folate homeostasis. J Biol Chem. 2004;279(53): 55008-55016.
15.!Ravasz E, Somera AL, Mongru DA, Oltvai ZN, and Barabasi AL. Hierarchical organization of
modularity in metabolic networks. Science. 2002;297(5586): 1551-1555.
16.!Real-Chicharro A, Ruiz-Mostazo I, Navas-Delgado I, Kerzazi A, Chniber O, Sanchez-Jimenez F,
Medina MA, and Aldana-Montes JF. Protopia: a protein-protein interaction tool. BMC Bioinfor-
matics. 2009;10 Suppl 12: S17.
17.!Reed MC, Nijhout HF, Sparks R, and Ulrich CM. A mathematical model of the methionine cycle.
J Theor Biol. 2004;226(1): 33-43.
Ral Montaez Martnez
10
2.0
28.0

2.0
28.0
Sinopsis
2.0
28.0
En esta Tesis Doctoral se presentan 4 ca-
ptulos en los que se plasma parte de mi
trabajo y mi modo de entender el queha-
cer cientfico.
Resumen de los contenidos del captulo 1
El primer captulo, titulado Mi concep-
cin de la ciencia (haciendo referencia al
magnfico libro de Erwin Schrdinger Mi
concepcin del mundo) [12], consta de
tres artculos de opinin: The usefulness of
post-genomics tools for characterization of
the amine cross-talk in mammalian cells
[Captulo 1.1][10], Polyamines: metabolism
to systems biology and beyond[Captulo
1.2][7] e Information integration of pro-
teinprotein interactions as essential tools
for immunomics[Captulo 1.3][6].
En el primero de estos trabajos [Captulo
1.1] se refleja el papel multifuncional de las
aminas bigenas y su relacin con mltiples
procesos fisiopatolgicos. La aminas se
muestran as como un sistema biolgico
complejo de un elevado nmero de inte-
Sinopsis
13
2.0
28.0
Sinopsis
Nuestro conocimiento es necesariamente finito, mientras que nues-
tra ignorancia es necesariamente infinita
Karl Raimund Popper
rrelaciones no lineales, lo que crea la nece-
sidad de abordad su estudio bajo una p-
tica integrada.
Se describe la necesidad de conjugar las
tecnologas in silicio con las tcnologas
micas para poder establecer una di-
nmica cruzada de prediccin y valida-
cin absolutamente necesaria para avan-
zar de forma firme y eficiente en el cono-
cimiento de un sistema biolgico. Se hace
una revisin de aquellos sistemas celulares
y animales donde se han detectado fen-
menos de interrelacin metablica en el
metabolismo de aminas.
En la figura tres de este trabajo se presen-
ta de modo esquemtico una propuesta
de aproximacin sistmica al estudio de un
problema biolgico y se expone la estrate-
gia del grupo para avanzar en una visin
ms holista y contextualizada de ese co-
nocimiento.
La elegancia de la navaja de Occam
con la que se pretenda descomponer to-
do problema hasta las unidades mnimas
de comprensin se convierten aqu en ale-
gora del eterno retorno. El conocimiento
en el que nos basamos surge de la aplica-
cin de la simplificacin y la hiptesis final
habr de ser contrastada y refutada en
sistemas experimentales simplificados en los
que la nica variable libre nos permita
asociar el comportamiento observado en
las variables dependientes al comporta-
miento de la variable libre.
Mi posicin como segundo autor en esta
publicacin, refleja mi participacin direc-
ta en esas reflexiones. La Dra. Snchez, uno
de mis directores, fue la persona invitada a
realizar la revisin, y como representante
espaola en la accin COST que gener la
publicacin deba ir a la cabeza. Mis otros
dos directores tambin constan como au-
tores, lo que prueba que las bases sobre las
que se desarrollara mi proyecto de Tesis
Doctoral, estaban consensuadas y plas-
madas por escrito desde tiempo atrs.
En el segundo trabajo [Captulo 1.2],
manifestamos nuevamente la necesidad
de aplicar aproximaciones sistmicas al
estudio de las poliaminas y el metabolismo.
Pero en este artculo no slo proponemos
un modo de aproximarse a el estudio, sino
que manifestamos nuestra desaprobacin
hacia las desacertadas interpretaciones de
la biologa de sistemas:
In the wikipedia, systems biology is defined as an
academic field that seeks to integrate different levels of
information to understand how biological systems
function. The identification of systems biology as an
academic field (namely, an additional speciality of
biology) is suspicious. In fact, since the objects of
study of biology (cells, organisms, populations, ecosys-
tems) are all of them complex systems, should not the
whole biology be considered as a systems biology?
We are afraid that most of the scientists converted
to systems biology make an instrumental use of it to
manage the huge amount of information provided by
new omics technologies and bioinformatics. Al-
though this instrumental approach is indeed also requi-
red, this cannot surpass the limitations of reductio-
nism. It is our understanding that an authentic holistic
and systemic view of biology should be based on the
analysis of the relationships among elements of a
biological system in a given steady-state, as well as
along the response of the system against any perturba-
tion of its environment, with the final aim to know not
only the system itself but also its dynamics. To achieve
this goal, the approaches provided by the general
theory of systems, the theory of dynamical systems
and, in general, the new sciences of complexity should
be applied. Complexity is a keyword, which was
identified as one of the three main challenges of bio-
logy (the others are consilience and communica-
tion) for the incoming century in an influential edito-
rial published in BioEssays (1999).
El tercer y ltimo trabajo [1.3], surge nue-
vamente como consecuencia de una invi-
tacin. En 2006, nuestro grupo fue invitado
a participar en el Congreso Immunomics
(Budapest), y all presentamos el plantea-
miento inicial y las hiptesis con las que
trabajaramos para la consecucin del
proyecto Amine System Project [1].
Este fue el comienzo del proyecto de
mediacin semntica y la contextualiza-
cin de una idea en una ontologa en la
Ral Montaez Martnez
14
2.0
28.0
que los mltiples fragmentos del reduccio-
nismo se integraban y contextualizaban
dando soporte a una herramienta que
permitiera capturar informacin metabli-
ca y molecular desde distintas bases de
datos internacionales. Algunos de esos re-
sultados se expondrn en el captulo 3.
En ese mismo Congreso, nuestro grupo
ofreci un tutorial del manejo y funciona-
miento de una herramienta con la que po-
der integrar informacin sobre interaccio-
nes protena-protena. El resultado fue el
manuscrito que aqu se presenta, en el que
se reflejan mis ideas personales con rela-
cin a las caracterizaciones protemicas
de los organismos y su importancia en el
metabolismo de cualquier clula viva, En el
trabajo se hace referencia a las inconsis-
tencias en la informacin depositada en
las bases de datos y en qu modo esta
informacin debe ponderarse sobre la ba-
se de un criterio de valoracin. En aquella
poca, consideraba que cualquier esfuer-
zo para poder ayudar en este sentido, si-
quiera sea en la integracin de la informa-
cin que ya exista, podra ser interesante.
Aunque por decisin propia no fui final-
mente autor de la herramienta publicada
por el grupo para la captura de esta infor-
macin [9], la realidad es que junto con mi
director Carlos Rodrguez-Caso, contribui-
mos en gran medida a la concepcin de
la herramienta, la procedencia de la infor-
macin que deba integrarse, su interpre-
tacin, los criterios de refinado de la infor-
macin y a la generacin de la ontologa
que subyace a esta herramienta. Con la
perspectiva que da el tiempo transcurrido
desde la preparacin de esta revisin, hay
que reconocer que los esfuerzos realizados
por los distintos grupos internacionales para
mejorar las herramientas de integracin de
la informacin sobre interacciones protei-
cas les dan la razn a muchas de las afir-
maciones hechas en este artculo, mientras
que otros puntos an permanecen sin re-
solver.
Resumen de los contenidos del captulo 2
El capitulo 2 rene los trabajos de inte-
gracin de informacin automatizada. A
este captulo lo he titulado Buscando la
verosimilitud en un mundo abierto.
Hacer realidad la construccin de un
mediador semntico me oblig a trabajar
codo con codo con un grupo de inform-
ticos. Esto fue una experiencia no siempre
grata pero me permiti familiarizarme con
un nuevo lenguaje y una forma de trabajar
muy diferente a la que habitualmente yo
desarrollaba. De manos del grupo Khaos
aprend a contextualizar mi conocimiento
en un sistema lgico formal y estructurado,
las ontologas. Aprend su programacin
facilitada y su visualizacin empleando el
entorno Protg [4] y el visualizador Jamba-
laya as como, los estndares del Web On-
tology Language Reference (OWL).
Puesto que en el primero de los artculos
que componen esta seccin [Captulo
2.1][8] se explican minuciosamente tanto
los antecedentes como el objetivo general
de construccin de un director semntico,
incluyendo su organizacin y componen-
tes, obviar la repeticin de todos estos
conceptos en esta seccin. Cualquier lec-
tor interesado en estos aspectos puede
seguir ilustrndose a partir de esta referen-
cia, de la pgina web del grupo Khaos, y
de las pginas webs donde las herramien-
tas siguen activas, a las que se puede ac-
ceder tambin a travs de [1].
En esta Memoria se incluyen dos de las
herramientas que se desarrollaron a partir
de la generacin del directorio semntico
inicial. Ambas se han publicado en revistas
de Bioinformtica y de ambas soy primer
coautor. A continuacin expondr breve-
mente el sentido de cada una de ellas, y
mi contribucin especifica a su desarrollo:
El primero de los trabajos de mediacin
semntica [Captulo 2.1] constituy el ex-
perimento piloto de nuestro mediador se-
mntico. AMMO-Prot es un mediador se-
mntico que pretende facilitar la recupe-
racin de estructuras tridimensionales de
15
Sinopsis
2.0
28.0
protenas. A la aplicacin se accede me-
diante una interface de usuario centrada
en las principales protenas del metabolis-
mo de poliaminas e histamina, AMMO-Prot
busca para el organismo definido por el
usuario, las estructuras tridimensionales, de
la protena seleccionada, almacenadas en
el Protein Data Bank (PDB) [2]. En caso de
no encontrarse un registro adecuado mo-
dela, a partir del homlogo ms cercano
encontrado, la estructura tridimensional de
dicha protena empleando la aplicacin
de modelado comparativo o por homolo-
ga Modeller [3]. La herramienta devuelve
finalmente un fichero en formato .pdb
compatible con cualquier visualizador de
molculas. Esta herramienta se present en
el Seventh International Workshop on Net-
work Tools and Applications in Biology (NET-
TAB 2007) Pisa, Italy. 12-15 June 2007 y co-
mo consecuencia de esta comunicacin,
nos invitaron a preparar esta publicacin
para la revista [8].
Como se ha explicado anteriormente,
gener la ontologa y defin las rutinas ne-
cesarias para acceder a los distintos pro-
gramas y pginas de forma automatizada
(Swiss-Prot, Modeler, Protein Data Bank) y
comprob la validez de los resultados ob-
tenidos en colaboracin con mi compae-
ra Almudena Pino ngeles. Pese a la res-
triccin impuesta por la interface, esta
herramienta puede emplearse para cual-
quier protena de igual modo que lo hace
Swiss-Model [5].
En el segundo de los trabajos de media-
cin semntica [Captulo 2.2] se consolida
el propsito inicial de generar una herra-
mienta general (no restringida a va meta-
blica alguna) que facilitase de modo au-
tomatizado tareas esenciales para la gene-
racin de modelos metablicos predictivos
aplicando un protocolo de razonamiento
basado en la estructura lgica definida en
la ontologa. Se trata de un asistente que
captura la informacin cintica, desde las
relaciones entre las enzimas, pasando por
los moduladores o las ecuaciones cinti-
cas, hasta el conjunto de parmetros ne-
cesarios para acotar su dinmica. La infor-
macin es recuperada de las bases de da-
tos KEGG, Brenda y SABIORK, sobre cual-
quier enzima de cualquier organismo, y
otra informacin complementaria captu-
rada de PubChem o Uniprot. El usuario tie-
ne la posibilidad de editar los datos obte-
nidos. Tambin prepara los datos y ecua-
ciones para facilitar la generacin de mo-
delos matemticos cinticos. Este asistente
se ha usado ya con distintos fines; entre
ellos, ha servido para actualizar y enrique-
cer una base de datos sobre el metabolo-
ma peroxisomal, recientemente publicada
[11].
Resumen de los contenidos del captulo 3
Durante las primeras dcadas del siglo
XX, investigadores como Leonor Michaelis,
Maud Menten y John Burdon Sanderson
Haldane, generaban hiptesis y trabajaban
sobre la posibilidad de formalizar en ecua-
ciones matemticas la dinmica de los
procesos vitales. Gracias a ese esfuerzo, la
cintica de las reacciones enzimticas pu-
do expresarse en el lenguaje de las mate-
mticas. La formalizacin de los mecanis-
mos de reaccin como funciones permiti
el avance de una Enzimologa formal que
abra las puertas de teoras generalistas,
como la teora general de sistemas, el an-
lisis e incluso la prediccin del comporta-
miento de mltiples macromolculas biol-
gicas.
Una va metablica no es sino el torrente
termodinmico que concatena las trans-
formaciones moleculares dentro de los se-
res vivos (reacciones metablicas) confor-
mando el conjunto una red de mnimos
energticos similar a los paisajes de plega-
miento.
Cada hebra termodinmica es una
creoda modelable por separado pero im-
posible de escindir de su conjunto, pues
depende de los aportes de este entorno.
Desde esta perspectiva, puede pensarse
en las distintas rutas metablicas como en
las piezas de un lego, cuya estructura apor-
Ral Montaez Martnez
16
2.0
28.0
ta informacin sobre el control general del
sistema. No sera descabellado pensar en
la posibilidad, por tanto, de ir descubriendo
las propiedades emergentes del sistema a
partir de la conexin de distintos mdulos
metablicos siguiendo una aproximacin
bottom-up.
Bajo esta hiptesis se comenz el trabajo
contenido en este captulo que he titulado
Metabolizando nmeros, en el que se
pretenda el modelado de dos mdulos
metablicos interesantes para los proyec-
tos financiados del grupo: catabolismo de
la arginina por clulas endoteliales y meta-
bolismo de poliaminas en clulas de mam-
feros. Este trabajo dio lugar a las dos publi-
caciones que forman parte de este captu-
lo.
El primero de los trabajos de modelado
cintico del metabolismo [Captulo 3.1] se
centr en el metabolismo de poliaminas. El
objetivo de este trabajo era encontrar las
bases del control homeosttico del meta-
bolismo de poliaminas. Pese a los importan-
tes esfuerzos por entender el comporta-
miento de los diferentes elementos impli-
cados en la homeostasis de esta ruta, su
compleja regulacin (transcripcional, tra-
ducional y metablica) as como su estre-
cha relacin con mltiples procesos celula-
res, dificultaron la comprensin de sus me-
canismos de regulacin conjunta. Para sol-
ventar esta limitacin se aplico una apro-
ximacin de tipo Bottom-up en la que la
ruta era modelada matemticamente a
partir de las cinticas individuales de cada
enzima, extradas de la informacin biblio-
grfica. Como resultado, el modelo captu-
raba las tendencias observadas en ratones
transgnicos para las principales enzimas
implicadas as como el efecto de los prin-
cipales inhibidores empleados tradicional-
mente en esta ruta. Adems, el modelo
evidenci la implicacin directa de S-ade-
nosilmetionina y acetil-CoA en la homeos-
tasis de los niveles de poliaminas.
En el segundo de los trabajos de mode-
lado cintico del metabolismo [Captulo
3.2] aplicamos la misma metodologa apli-
cada en el trabajo anterior [Captulo 3.1]
para modelar los principales elementos
implicados en el catabolismo de arginina.
El modelo era consistente con los resultados
experimentales descritos en la bibliografa.
Los anlisis de control de flujo metablico
permitieron predecir los coeficientes de
control de todos y cada uno de los ele-
mentos implicados en esta ruta, cumplien-
do el teorema del sumatorio.
Sin embargo, ya desde el comienzo tena
claro que estos dos trabajos que aqu se
presentan deban ser nicamente el co-
mienzo de un proyecto ms ambicioso que
tratara de contextualizar estas vas meta-
blicas en un entorno de funcionamiento
global. Los resultados obtenidos, especial-
mente los correspondientes al metabolismo
de poliaminas remarcaron esa necesidad.
Nuestras predicciones, validadas por los
resultados experimentales de distintos gru-
pos internacionales (fundamentalmente los
de los laboratorios de los Dres. Anthony E.
Pegg, Juhane Janne y Carl W.. Porter, todos
ellos pioneros en el campo de las poliami-
nas), indicaban que el metabolismo de
poliaminas debe estar ntimamente ligado,
no slo a la capacidad proliferativa celular
a travs de su unin directa a los cidos
nucleicos y la importancia de la metilacin
de la cromatina, sino tambin al metabo-
lismo oxidativo celular.
Aunque no se incluye en esta memoria,
comenz a trabajar en la integracin del
metabolismo de aminocidos catinicos
con el metabolismo de aminocidos azu-
frados, y el ciclo de los metilos. Fue la lle-
gada de mi compaero Armando Reyes
Palomares la que me sirvi de relevo en
esta tarea y me permiti afrontar la ltima
etapa de esta Tesis. En su mano qued el
reto de consolidar esa integracin e inte-
grar las rutas fundamentales del metabo-
lismo oxidativo de clulas de mamferos
(Esquema).
Tambin es preciso especificar que, por
mantener el orden lgico de todo trabajo
17
Sinopsis
2.0
28.0
cientfico, he optado por presentar nuestro
anlisis inicial (antecedentes en forma de
revisiones) en el Captulo 1, y las herramien-
tas desarrolladas (tecnologa de integra-
cin de la informacin) en el captulo 2;
dejando en este tercer captulo los trabajos
de modelado metablico aplicados al me-
tabolismo de aminocidos catinicos. Sin
embargo, como puede deducirse de las
fechas de las publicaciones, concretamen-
te estos dos artculos de modelado comen-
zaron sin la ayuda de las herramientas des-
critas en el captulo 2, que estaban an en
construccin, de manera que la bsqueda
de datos se realiz de forma manual. Esta
experiencia fue la que me hizo ver la nece-
sidad de desarrollar nuevas herramientas
de integracin, manipulacin y anlisis au-
tomtico de la informacin. Puede decirse,
entonces, que el desarrollo de los conteni-
dos de los captulos 2 y 3 coincidieron du-
rante gran parte en el tiempo, y que es en
la actualidad cuando las herramientas pre-
sentadas en esta Memoria, especialmente
el SBBM Assistant, nos estn facilitando la
captura y manipulacin de la informacin
de otros proyectos ms ambiciosos que se
estn desarrollando en la actualidad en
nuestro grupo, y en algunos casos en cola-
boracin con otros laboratorios y redes de
grupos, como por ejemplo el CIBER de En-
fermedades Raras.
Resumen de los contenidos del captulo 4
En el captulo 4 que titulo Simples lneas y
puntos, el foco se abre, el metabolismo es
ahora abstrado como un sistema en su
conjunto. Un nuevo propsito, entender
qu constricciones han condicionado la
estructura del metabolismo, comprender
cmo las relaciones de las diferentes trans-
formaciones metablicas se integran en un
todo funcional, se convierte en el nuevo
objetivo.
ste, pese a ser el enfoque ms grosero
(que aporta un menor grado de detalle),
es uno de los que permite trazar lneas ge-
nerales. El metabolismo, tal como lo cono-
cemos actualmente, es un sistema alta-
mente imbricado. La simplificacin del
entramado metablico a simples puntos
unidos por lneas ha puesto de manifiesto
que las relaciones entre sus elementos no
surgen al azar. Los nodos del metabolismo
muestran una distribucin de conexiones
que se ajusta a una ley de potencia nega-
tiva, es libre de escala, mundo pequeo y
jerrquicamente modular. Actualmente
sabemos que estas generalidades, pese a
encajar con posibles caractersticas del
metabolismo, son en realidad fruto de la
densificacin que introduce el proceso de
proyeccin. Durante la proyeccin, las re-
des se densifica, incrementando el nmero
de nodos. El agrupamiento aparece como
una correlacin de esta densificacin.
Como demuestran lo trabajos del captulo
4, estas anomalas resultantes del proceso
de proyeccin han condicionado una vi-
sin alterada de la realidad metablica.
En el primero de los trabajos de anlisis
topolgico del metabolismo [Captulo 4.1]
se muestran los principales problemas de la
sobre-simplificacin de una red natural-
mente bipartita (como lo es el metabolis-
mo) y se plantea la necesidad de refutar
las conclusiones extradas a partir de las
propiedades topolgicas ms sensibles a
los artefactos derivados de la proyeccin.
En el segundo de los trabajos [Captulo
4.2], se trata de dar respuesta a las pregun-
tas planteadas en el captulo anterior. Se
verifica la presencia de artefactos en la
caracterizacin derivados de la inflacin y
las correlaciones locales que la proyeccin
incorpora a las redes proyectadas. Por otro
lado se revisan las propiedades topolgicas
bajo una ptica puramente biolgica, in-
tentando hacer lo ms comprensible posi-
ble las medidas.
En el tercero de los trabajos [Captulo
4.3], se revisa de modo independiente las
medidas que se emplean para caracterizar
a las redes metablicas como redes de
tipo mundo pequeo (Small-world). El me-
tabolismo ha sido considerado desde el
ao 2000 un ejemplo paradigmtico de
Ral Montaez Martnez
18
2.0
28.0
mundo pequeo. Sus diferentes represen-
taciones unipartitas (metabolitos conecta-
dos a travs de reacciones o reacciones
conectadas por metabolitos en comn)
presentan un elevado grado de conexio-
nes locales en comparacin a sus equiva-
lentes aleatorias. Ahora sabemos que esta
sobre-representacin de conexiones entre
nodos vecinos en las proyecciones del me-
tabolismo vienen de la manos del propio
proceso de proyeccin. Es este trabajo se
evala en que grado el sesgo afecta a la
caracterizacin del metabolismo como
mundo pequeo y se trata de identificar las
causas de este comportamiento.
Bibliografa
1.! ASP. Available from: http://asp.uma.es.
2.! Berman HM, Westbrook J, Feng Z, Gilliland G, Bhat TN, Weissig H, Shindyalov IN, and Bourne
PE. The Protein Data Bank. Nucleic Acids Res. 2000;28(1): 235-242.
3.! Eswar N, Webb B, Marti-Renom MA, Madhusudhan MS, Eramian D, Shen MY, Pieper U, and
Sali A. Comparative protein structure modeling using Modeller. Curr Protoc Bioinformatics.
2006;Chapter 5: Unit 5 6.
4.! H. K, Musen MA, and AlanL. Rector AL. Editing description logic ontologies with the Protg
owl plugin. in Proc. of International Conf. on Description Logics. 2004.
5.! Kiefer F, Arnold K, Kunzli M, Bordoli L, and Schwede T. The SWISS-MODEL Repository and
associated resources. Nucleic Acids Res. 2009;37(Database issue): D387-392.
6.! Montanez R, Navas-Delgado I, Medina MA, Aldana-Montes JF, and Sanchez-Jimenez F. Infor-
mation integration of protein-protein interactions as essential tools for immunomics. Cell Immu-
nol. 2006;244(2): 84-86.
7.! Montanez R, Sanchez-Jimenez F, Aldana-Montes JF, and Medina MA. Polyamines: metabolism
to systems biology and beyond. Amino Acids. 2007;33(2): 283-289.
8.! Navas-Delgado I, Montanez R, Pino-Angeles A, Moya-Garcia AA, Urdiales JL, Sanchez-Jime-
nez F, and Aldana-Montes JF. AMMO-Prot: amine system project 3D-model nder. BMC Bioin-
formatics. 2008;9 Suppl 4: S5.
9.! Real-Chicharro A, Ruiz-Mostazo I, Navas-Delgado I, Kerzazi A, Chniber O, Sanchez-Jimenez F,
Medina MA, and Aldana-Montes JF. Protopia: a protein-protein interaction tool. BMC Bioinfor-
matics. 2009;10 Suppl 12: S17.
10.!Sanchez-Jimenez F, Montanez R, Correa-Fiz F, Chaves P, Rodriguez-Caso C, Urdiales JL, Al-
dana JF, and Medina MA. The usefulness of post-genomics tools for characterization of the
amine cross-talk in mammalian cells. Biochem Soc Trans. 2007;35(Pt 2): 381-385.
11.!Schluter A, Real-Chicharro A, Gabaldon T, Sanchez-Jimenez F, and Pujol A. PeroxisomeDB 2.0:
an integrative view of the global peroxisomal metabolome. Nucleic Acids Res. 38(Database
issue): D800-805.
12.!Schrdinger E, Mi concepcin del mundo. Metatemas, ed. Tusquets. 1988.
19
Sinopsis
2.0
28.0

2.0
28.0
Summary
2.0
28.0
Thi s work enti tl ed Computati onal ,
Dynamics and Topological Approaches to
Study Metabolism started more than 5
years ago within a biochemical and
molecular biology research group. One of
the major objectives of the group is to
locate elements that can be considered
new and more effi ci ent tar gets for
intervention.
The massive data acquisition in this post-
genome era has given rise to an over-
accumul at i on of huge amount of
information. These data, far from the
understanding of individual researchers,
requi re of the use of computati onal
techniques but also, of new perspectives
a r oadmap - wher e dat a can be
incorporated in an integrative way. Such a
view requires a molecular system to be
treated as a whole and this framework is
provided from a Systemic Biology discipline.
Data integration, mathematical modeling
and experimental testing are the key point
wher e these emer gent di sci pl i ne i s
grounded.
Biogenic amines are involved in several
physi opathol ogi cal probl ems (mai nl y
23
2.0
28.0
Summary
Our knowledge can only be finite, while our ignorance
must necessarily be infinite.
Karl Raimund Popper
cancer and inflammation). The pleiotropic
biological activities of these compounds,
acting at very different levels of the cell
physiology has inspired the turn to a new
more systemic view in our group research.
Here, where reductionistic approaches
were not enough, biogenic amines should
be understood within the framework of a
multi-scale view: from molecular structure to
cell response.
In the light of this view, this Thesis was
conceived to respond to the need of a
integrative view of our study system.
However, although in most cases the
research in this work is oriented to the
understanding of biogenic amines, this
work is not restricted exclusively to this
system and a broader perspective and
more general works can be found in this
document. In this way, for the specialized
reader focused on the study of biogenic
amines, this work is a rigorous exploration
about the contribution in this field using a
systemic biology perspective; on the other
hand the systems biologist will appreciate
this system as a case study and the starting
point for more general considerations.
This work has been possible by the
collaborative interaction with computer
science and theoretical research groups.
The final body of work is divided into 4
sections:
Chapter 1, My conception of science.
This first chapter is devoted to present the
ideas, aims and planning of our group to
advance in a more efficient way in a more
comprehensive view of biogenic amine
metabolism and their physiopathological
roles. Somehow, these reviews reflect the
state of the art in this biological topic, as
well as the hypothesis and aims of the
present project (and those of other group
member projects, too). The weight of my
contri buti on to these refl exi ons and
planning can be deduced from the order
of the authors (first/second authorship).
These references are:
Chapter 1.1 The usefulness of post-
genomics tools for characterization of the
amine cross-talk in mammalian cells [11].
Evi dence i s growi ng i n favour of a
relationship between cancer and chronic
inammation, and particularly of the role of
a polyamine and histamine metabolic
i n t e r p l a y i n v o l v e d i n t h e s e
physiopathological problems, which are
indeed highly complex biological systems.
Decodication of the complex inter- and
intra-cellular signalling mechanisms that
control these effects is not an easy task,
which must be helped by systems biology
technologies, including new tools for
location and integration of database-
s t or ed i nf or mat i on and pr edi ct i ve
mathematical models, as well as functional
genomi cs and ot her exper i ment al
mol ecul ar approaches necessary for
hypothesis validation. We review the state
of the art and present our latest efforts in
t hi s ar ea, f ocus ed on t he ami ne
metabolism eld.
Chapter 1.2 Polyamines: metabolism to
s ys t ems bi ol ogy and beyond [ 4] .
Pol yami nes and the metabol i c and
physiopathological processes in which they
are involved represent an active eld of
research that has been conti nuousl y
growing since the seventies. In the last
years, the trends in the focused areas of
interest within this eld since the 1970s have
been conrmed. The impact of -omics in
polyamine research remains too low in
comparison with its deep impact on other
biological research areas. These high-
throughput approaches, along with systems
biology and, in general, more systemic and
holistic approaches should contribute to a
renewal of this research area in the near
future.
Chapter1.3 Information integration of
protein-protein interactions as essential
tools for immunomics [2]. After a brief
introduction to point out the necessity to
Ral Montaez Martnez
24
2.0
28.0
advance for a global understanding of the
macromolecular interactions occurring
during the immune system development
and responses, Section 2 will be devoted to
analyze the current tools for an automatic
location of information on these protein
protein interactions in the web. In the next
section (Section 3), we will point out
diferent action lines to improve these tools
and, consequentl y, to i ncrease the
efciency to establish (to understand) the
protein network skeleton that controls
our immune responses. Finally, we will briey
present our current strategy and work to
advance towards this goal.
Chapter 2, Looking for verisimilitude in
an open world.
Papers included in this section describe
the need and the characteristics of a
semantic mediator built as the technical
support of a biocomputational platform
able to integrate information from different
international respositories of information
concerning macromolecular structures and
metabolic data. The mediator, shortly
named as AMMO, was initially applied to
develop AMMO-Prot, a tool able to locate
and to represent the 3D structure of proteins
related to amine metabolism of any
species, if they are available in Protein Data
Bank. Alternatively, the system can find a
putative template and build a first 3D
model assisted by Modeller [1]. Then,
AMMO was used to develop the Systems
Biology Metabolic Modeling Assistant, able
to l ocate ki neti c data on enzymes,
metabolites and regulators of any enzyme
of any species, and to process the data to
be understood by other biocomputational
tools of metabolic modeling and dynamic
analysis. The references of these works are:
Chapter 2.1 AMMO-Prot: amine system
project 3D-model finder [8]. Amines are
biogenic amino acid derivatives, which
play pleiotropic and very important yet
complex roles in animal physiology. For
many ot her r el evant bi omol ecul es,
biochemical and molecular data are being
accumulated, which need to be integrated
in order to be effective in the advance of
biological knowledge in the field. For this
purpose, a multidisciplinary group has
started an ontology-based system named
the Amine System Project (ASP) for which
amine-related information is the validation
bench.
In this paper, we describe the Ontology-
Based Mediator developed in the Amine
System Project (http://asp.uma.es) using
the infrastructure of Semantic Directories,
and how this system has been used to solve
a case related to amine metabolism-
related protein structures.
This infrastructure is used to publish and
manage not only ontologies and their
relationships, but also metadata relating to
t he r es our ces commi t t ed wi t h t he
ontol ogi es. The system devel oped i s
av ai l abl e at h t t p: / / as p. u ma. es /
WebMediator.
Chapter 2.2 Systems biology metabolic
modeling assistant: an ontology-based tool
for the integration of metabolic data in
kinetic modeling [9]. We present Systems
Bi ol ogy Metabol i c Model i ng Assi stant
(SBMM Assistant), a tool built using an
ontology-based mediator, and designed to
facilitate metabolic modeling through the
integration of data from repositories that
contain valuable metabolic information.
Thi s sof t war e can be used f or t he
visualization, design and management of
metabolic networks; selection, integration
and storage of metabolic information; and
as an assistant for kinetic modeling.
SBMM Assistant for academic use is freely
available at
http://www.sbmm.uma.es.
Chapter 3 Metabolizing numbers
Thi s secti on i ncl udes the two fi r st
published works of a more ambitious and
25
Summary
2.0
28.0
long-term bottom-up approach (currently
funded in our lab) to the dynamic modeling
of both amino acid metabolism and energy
metabolism with predictive capabilities and
potentially useful in different pathologies
related to proliferating cells and rare
genetic syndromes. Mathematical models
pr edi ct t he behavi or descr i bed by
independent and reputated international
groups. The references are:
Chapter 3.1 Mathematical Modeling of
Polyamine Metabolism in Mammals [10].
Polyamines are considered as essential
compounds in living cells, since they are
involved in cell proliferation, transcription,
and translation processes. Furthermore,
polyamine homeostasis is necessary to cell
survival, and its deregulation is involved in
relevant processes, such as cancer and
neurodegenerative disorders. Great efforts
have been made to elucidate the nature
of polyamine homeostasis, giving rise to
relevant information concerning the be-
havior of the different components of
pol yami ne metabol i sm, and a great
amount of information has been gener-
ated. However, a complex regulation at
transcriptional, translational, and metabolic
levels as well as the strong relationship be-
tween polyamines and essential cell proc-
esses make it difficult to discriminate the
role of polyamine regulation itself from the
whole cell response when an experimental
approach is given in vivo. To overcome this
limitation, a bottom-up approach to model
mathematically metabolic pathways could
allow us to elucidate the systemic behavior
from individual kinetic and molecular prop-
erties. In this paper, we propose a mathe-
matical model of polyamine metabolism
from kinetic constants and both metabolite
and enzyme levels extracted from biblio-
graphic sources. This model captures the
tendencies observed in transgenic mice for
the so-called key enzymes of polyamine
metabolism, ornithine decarboxylase, S-
adenosylmethionine decarboxylase and
spermine spermidine N-acetyl transferase.
Furthermore, the model shows a relevant
role of S-adenosylmethionine and acetyl-
CoA availability in polyamine homeostasis,
which are not usually considered in sys-
temic experimental studies.
Chapter 3.2 In silico analysis of arginine
catabolism as a source of nitric oxide or
polyamines in endothelial cells [3].
We use a modeling and simulation
approach to carry out an in silico analysis of
the metabolic pathways involving arginine
as a precursor of nitric oxide or polyamines
in aorta endothelial cells. Our model
predicts conditions of physiological steady
state, as well as the response of the system
to changes in the control parameter,
external arginine concentration. Metabolic
ux control analysis allowed us to predict
the values of ux control coefcients for all
the transporters and enzymes included in
the model. This analysis fullls the ux control
coefcient summation theorem and shows
that both the low afnity transporter and
arginase share the control of the uxes
through these metabolic pathways.
Chapter 4 Simple dots and lines
Papers in this section provide a view
about the application of graph theory on
the study of the organization of metabolism.
In a initial paper we analyze the impact of
the metabolic network definition in our
topological understanding of metabolism .
Since metabolism consists of reactions and
metabolites, its natural representation
agrees with a bipartite graph. However,
these networks have been commonly sim-
plified in one-mode versions through the
application of a process called projection.
We demonstrate in this work that this treat-
ment introduces a bias in the results. In par-
ticular we highlight the fact that a simple
bipartite random network is just enough! to
exhibit similar topological properties such as
small-world pattern and hierarchical order
by the simple effect of projection. At this
point, a revision of the topological organi-
zation of metabolism is required. According
to this need we reproduce in a compara-
Ral Montaez Martnez
26
2.0
28.0
tive study the most fundamental topologi-
cal statistics using metabolic networks for
different species attending to their bipartite
representation. After that, we explore and
define a concept of small-world for bipar-
tite networks.!
Chapter 4.1 When metabolism meets
topol ogy:reconci l i ng metabol i te and
reaction networks [5]
The search for a systems-level picture of
metabol i sm as a web of mol ecul ar
i nteracti ons provi des a paradi gmati c
example of how the methods used to
characteri ze a system can bi as the
interpretation of this functional meaning.
Metabolic maps have been analyzed using
novel techniques from network theory,
revealing some non-trivial, functionally
relevant properties. These include a small-
world structure and hierarchical modularity.
However, as discussed here, some of these
properties might actually result from an
inappropriate way of defining network
interactions. Starting from the so-called
bipartite organization of metabolism, where
the two meaningful subsets (reactions and
metabolites) are considered, most current
works use only one of the subsets by means
of s o cal l ed gr aph pr oj ect i on s .
Unfortunately, projected graphs often
ignore relevant biological and chemical
constraints, thus leading to statistical
artifacts. Some of these drawbacks and
al ter nati ve approaches need to be
properly addressed.
Chapter 4.2 Analyzing the bipartite
structure of the metabolic network [6].
Metabolism is a natural bipartite network
that has always been treated as an one-
mode net wor k wi t h t he f ocus on
compounds or reactions. Projections of
bipartite networks to obtain one-mode
versions reduce the information contents in
the system and modi f y topol ogi cal
properties, such as density, degree and
clustering coefficient. In consequence,
some topological generalizations assumed
for metabolic networks must be reviewed in
the light of a bipartite graph analysis.
Herein, we analyze E. coli, S. cerevisiae and
H. sapiens bipartite reactions-metabolites
networ ks, and hi ghl i ght di f f er ences
between both graph abstractions and the
advantages of bipartite measures such as
strength or squares cliquishness, on the
bases of a biological interpretation of the
topological parameter.
Chapt er 4. 3 I s metabol i sm smal l -
world? [7]
Universal patterns as Small-World can
emerge as an artifactual consequence of
the inappropriate use of the methodology.
Its noteworthy that the projection of
bipartite networks increases the density of
local connections in the resulting networks.
Consequently, as we show in this work,
even completely random bipartite networks
could seem small-worlds when projected,
this contributing to consider erroneously the
small-world pattern as an universal one. In
this article, we develop an alternative
method, inspired in the fraction of cycles
wi th si ze four, to per for m a correct
evaluation of the small-world characteristics
of bipartite networks, and it is applied to
metabolic networks of 11 species. The
results obtained indicate that, indeed,
metabolism maintains the small world
features even when analyzed as a bipartite
network.
Bibliography
27
Summary
2.0
28.0
1.! Eswar N, et al. Comparative protein structure modeling using Modeller. Curr Protoc Bioinforma-
tics. 2006;Chapter 5: Unit 5 6.
2.! Montanez R, et al. Information integration of protein-protein interactions as essential tools for
immunomics. Cell Immunol. 2006;244(2): 84-86.
3.! Montanez R, et al. In silico analysis of arginine catabolism as a source of nitric oxide or polya-
mines in endothelial cells. Amino Acids. 2008;34(2): 223-229.
4.! Montanez R, et al. Polyamines: metabolism to systems biology and beyond. Amino Acids.
2007;33(2): 283-289.
5.! Montaez R, et al. When metabolism meets topology:reconciling metabolite and reaction net-
works. BioEssays. 2010;In-Press.
6.! Montaez R, et al. Analyzing the bipartite structure of the metabolic network. 2010;Manuscript.
7.! Montaez R, et al. Is small-worldness still despite their abstraction as a bipartite graph?
2010;Manuscript.
8.! Navas-Delgado I, et al. AMMO-Prot: amine system project 3D-model nder. BMC Bioinforma-
tics. 2008;9 Suppl 4: S5.
9.! Reyes-Palomares A, et al. Systems biology metabolic modeling assistant: an ontology-based
tool for the integration of metabolic data in kinetic modeling. Bioinformatics. 2009;25(6): 834-
835.
10.!Rodriguez-Caso C, et al. Mathematical modeling of polyamine metabolism in mammals. J Biol
Chem. 2006;281(31): 21799-21812.
11.!Sanchez-Jimenez F, et al. The usefulness of post-genomics tools for characterization of the
amine cross-talk in mammalian cells. Biochem Soc Trans. 2007;35(Pt 2): 381-385.
Ral Montaez Martnez
28
2.0
28.0

2.0
28.0
Hiptesis y
Objetivos
2.0
28.0
Planteamiento general
La investigacin descrita en el presente documento responde a la necesidad de ex-
plorar aquellos aspectos planteados en el prembulo que, desde un punto de vista for-
mal, constituye la proposicin y contrastacin de las siguientes hiptesis:
Hiptesis 1.- La definicin adecuada de una ontologa orientada a la organizacin
semntica dentro del campo de la biologa molecular permite la integracin de la
informacin biolgica de una forma automatizada, requisito previo para el diseo
de plataformas informticas que gestionen y proporcionen de manera eficiente la
informacin necesaria para el desarrollo de proyectos de Biologa de Sistemas.
Hiptesis 2.- La integracin de modelos matemticos de vas! metablicas basada
en una adicin progresiva de mdulos individuales! constituye una estrategia bo-
ttom-up capaz de revelar nuevas propiedades emergentes del metabolismo. Esta
aproximacin puede aplicarse para obtener una visin integrada del metabolismo
secundario de amincidos (y derivados), generando predicciones cuantitativas
contrastables experimentalmente.
Hiptesis 3.- La teoria de grafos constituye una excelente aproximacin top-down
al estudio del metabolismo. Sin embargo, la forma en que se defina la red metab-
lica es determinante, pues una sobresimplificacin en la misma introduce artefactos
en las medidas topolgicas de la red que pueden distorsionar nuestro conocimien-
to acerca de la organizacin del metabolismo.
!
En base a estas hiptesis, se establecieron los siguientes objetivos:
Objetivo 1.- Desarrollar un mediador semntico capaz de capturar informacin so-
bre propiedades de elementos metablicos, as como validar estas capacidades
del mediador desarrollando herramientas piloto que fuesen de utilidad para los
proyectos del grupo en curso, centrados fundamentalmente en problemas estruc-
turales y funcionales de elementos implicados en el metabolismo de aminocidos y
derivados.
Objetivo 2.- Establecer las bases para la construccin de un modelo metablico
integrado del metabolismo de aminocidos y metabolismo oxidativo con capaci-
dad predictiva mediante la implementacin computacional de los mdulos esen-
ciales de dicho macromodelo. En concreto, crear los mdulos esenciales corres-
pondientes al catabolismo de la arginina y el metabolismo de las poliaminas.
Objetivo 3.- Evaluar el efecto de la proyeccin en redes metablicas en distintas
medidas topolgicas tales como el patrn de mundo pequeo y la libertad de es-
cala, as como reproducir el anlisis topolgico de redes metablicas en su versin
bipartita para distintas especies.
31
2.0
28.0

2.0
28.0
Hypothesis
&
Goals
2.0
28.0
General approach
This work was planned and executed on the bases of the following hypotheses:
Hypothesis 1.-!!! Integration of biological information by semantic web technologies
can be a helpful effort for efficient advance of Systems Biology projects.
Hypothesis 2.-! ! ! Modeling of metabolic systems by integration of mathematical
models of single pathways (metabolic modules) can be considered a valid bottom-
up strategy. This approach applied to secondary metabolism of amino acids and
their derivatives can provide a comprehensive view of the relationships among ele-
ments with potential predictive value, which can be validated and/or refined by
experimentation.
Hypothesis 3.-!!! Graph Theory is an excellent top-down approach to study metabo-
lism. However, a careful definition of metabolic network is required. Any oversimpli-
fication of this definition can introduce artifacts in the topological measurements of
the network that can affect to the quality of our knowledge on metabolic organiza-
tion.
!
According to these hypotheses, the following objectives were established:
!
Goal 1.-! ! ! To develop a semantic mediator able to capture information on the
properties of the metabolic elements. In order to validate the mediator capabilities,
pilot tools should be developed being useful for the current projects of our research
group. Thus, these pilot projects should solve structural and functional questions on
elements related to amino acid (and their derivatives) metabolism and physiopa-
thological conditions.
Goal 2.-! ! ! To start a long-term project on a predictive metabolic macro-model,
containing elements related to both amino acid and oxidative metabolism in
mammalian tissues, with a bottom-up strategy (by sequential addition of new
modules).! Specifically, within the present Project, we wanted to start working on
modeling of arginine and polyamine-related pathways.
Goal 3.-! ! ! To evaluate the effect of the metabolic network projections on different
topological measures, such as the small-world pattern and scale-freeness and to
reproduce the topological analysis for metabolic networks from different species in
their bipartite versions.
35
2.0
28.0

2.0
28.0
Conceptos
2.0
28.0
En este captulo, todos los trabajos, ade-
ms de revisar el papel de poliaminas e
histamina [C1.1] en mltiples procesos fisio-
patolgicos, la escasa repercusin de las
micas en el campo de la poliaminas
[C1.2] o la falta de estandarizacin en las
bases datos de interacciones [C1.3]),
transmiten una idea. Esta idea subyacente
en todos los trabajos propugna una biolo-
ga de sistemas en la que la meta sea en-
contrar las propiedades emergentes de los
sistemas y comprender su origen. Hacer ver
que las micas, por s solas, no son otra
cosa que biologa molecular de alta a gran
escala en la que los sistemas se diseccio-
nan en sus componentes individuales para
ser estudiados.
39
2.0
28.0
Mi concepcin de la ciencia:
El punto de partida de un recorrido bottom-up hacia la Biologa de Sistemas
La supresin real de la metafsica convierte al arte y a la ciencia en
ptreos esqueletos sin almas, incapaces del ms mnimo progresos
Erwin Schrdinger, Mi concepcin del mundo
1
Ral Montaez Martnez
40
2.0
28.0
41
Captulo 1
2.0
28.0
Health Implications of Dietary Amines 381
The usefulness of post-genomics tools
for characterization of the amine cross-talk
in mammalian cells
F. S anchez-Jim enez*
1
, R. Monta nez*, F. Correa-Fiz*, P. Chaves*, C. Rodr guez-Caso*
2
, J.L. Urdiales*, J.F. Aldana
and M.A. Medina*
*Department of Molecular Biology and Biochemistry, Campus de Teatinos, University of M alaga, 29071 M alaga, and CIBER of Rare Diseases, Spain, and
Department of Computer Languages and Computer Sciences, Campus de Teatinos, University of M alaga, 29071 M alaga, Spain
Abstract
Evidence is growing in favour of a relationship between cancer and chronic inammation, and particularly of
the role of a polyamine and histamine metabolic interplay involved in these physiopathological problems,
which are indeed highly complex biological systems. Decodication of the complex inter- and intra-cellular
signalling mechanisms that control these effects is not an easy task, which must be helped by systems
biology technologies, including new tools for location and integration of database-stored information
and predictive mathematical models, as well as functional genomics and other experimental molecular
approaches necessary for hypothesis validation. We review the state of the art and present our latest efforts
in this area, focused on the amine metabolism eld.
A brief overview of amine metabolism in
immune system and cancer cells
Decarboxylation of amino acids produces biogenic amines
that play important biosignalling roles in mammalian cells.
Those derived from arginine/ornithine (polyamines) are
essential for proliferation in every living cell (especially in
cancer and other highly proliferative cells and tissues), as ex-
plained elsewhere in this issue of Biochemical Society Trans-
actions. Histamine (the product of the histidine decarboxyl-
ase reaction) is mainly synthesized in immune cells (mast
cells, basophils, macrophages, platelets, T-cells and dendri-
tic cells), where it is considered as a proinflammatory medi-
ator [1]. Mast cells, the major producer of histamine in the
human body, are bone-marrow-derived cells expressing a
variety of phenotypic features as determined by the local
environment. They are located in connective tissue (skin and
peritoneal cavity) and mucosa, and synthesize inflammatory
mediators releasedtothe environment inresponse todifferent
stimuli [2]. Some of these mast cell mediators are accumulated
into granules and released by exocytosis, as is the case with
specific proteinases (tryptase) and histamine. Histamine is
synthesized by HDC (histidine decarboxylase) [3]. Other
immune cells (for instance, macrophages) produce but do
not store histamine into granules. Of course, these immune
cells also produce polyamines, as essential compounds for
macromolecular synthesis and cell survival. Thus they can be
Key words: cancer, histamine, inammation, metabolic modelling, polyamine, protein
modelling.
Abbreviations used: HDC, histidine decarboxylase; ODC, ornithine decarboxylase; SAM, S-
adenosylmethione; SSAT, spermine spermidine N
1
-acetyltransferase.
1
To whom correspondence should be addressed (email kika@uma.es).
2
Present address: Complex System Lab., University Pompeu-Fabra, Barcelona, Spain.
considered as complex amine-producing and amine-handling
cell models.
Nowadays, clear relationships are being established be-
tween chronic inflammation and several types of tumours [4].
Histamine- and polyamine-producing cells are often close
in vivo (for instance, during carcinoma and/or leukaemia
growth), and malignant forms of histamine-producing cells
have been reported (mastocytosis, basophilic leukaemias and
some types of gastric cancer, among others) [58]. Specific
roles of histamine and polyamines in the cross-talk between
mast cells and tumour cells have been reported [9]: histamine
can promote or inhibit tumour growth depending on the
receptor expressed by the target tumour cell type. In turn,
polyamines and their oxidized products are related to the
mast cell proliferation/death equilibrium and the liberation
dynamics of mast cell granule mediators to the medium[9,10].
Working on murine mast cell models, we have previously
observed antagonistic relationships between histamine and
polyamine metabolisms [11,12]. We also observed that mast
cells are particularly sensitive to the apoptosis inducers
produced from polyamines by serum amine oxidases [10,13].
All of these facts are particularly interesting for gut physio-
logy, since diet can provide compounds (i.e. drugs, biogenic
amines, enterobacterial products, etc.) affecting mast cell
proliferation/death and, consequently, mast cell roles in gut
inflammatory and neoplastic diseases.
From a biochemical point of view, interferences between
histamine and polyamine metabolic pathways are not sur-
prising, since both types of biogenic amines share structural
properties and common metabolic steps, as observed with
purified enzymes or animal cultured cell models (Figure 1)
[14]: (i) histamine interferes with spermine transport systems
and has a co-operative effect with inducers of the key enzyme
C
2007 Biochemical Society
Ral Montaez Martnez
42
2.0
28.0
382 Biochemical Society Transactions (2007) Volume 35, part 2
Figure 1 Interplay between polyamine metabolism and histamine metabolism
Black arrows, reactions; dotted and dashed lines, inductions; dotted lines, inhibitions/repressions. In the amine structures,
1,4-diamine moieties are represented in bold lines. Monomeric structures of the human enzymic monomers are taken
from the ModBase (http://www.modbase.compbio.ucsf.edu/). In boxes are processes related to amine metabolism.
ADH, alcohol dehydrogenase; DAO, diamine oxidase; dcSAM, decarboxylated S-adenosylmethionine; FA, fatty acids; Glc,
glucose; HMT, histidine N-methyltransferase; MAO, monoamine oxidase; N-ac Spd/Spm, N-acetylspermidine/spermine; PAO,
polyamine oxidase; SAMdc, S-adenosylmethionine decarboxylase; SpdS, spermidine synthase; SpmS, spermine synthase;
TGase, transglutaminase.
of polyamine degradation, SSAT (spermine spermidine N
1
-
acetyltransferase), so reducing the intracellular polyamine
content in mouse C57 mast cells [12]; (ii) some diamines
inhibit the uptake of cationic amino acids by the y
+
system, as
observed in Ehrlich ascitic tumour cells [15]; (iii) polyamines
and histamine can be covalently cross-linked to proteins by
the action of transglutaminases [16]; (iv) some amino oxidases
degrade bothpolyamine- andhistamine-producing toxic pro-
ducts (aldehydes and oxygen free radicals) [17]. In addition,
SAM(S-adenosylmethione), the aminopropyl donor for both
spermidine and spermine synthesis, is also required for
methylation of histamine, as a previous step for degrad-
ation in many cell types [1]. Since SAM is also a methyl
donor for DNA methylases, it could constitute a metabolic
node connecting epigenetic response and amine metabolism
(among other pathways). Experimental evidence for a
relationship between DNA methylation and both polyamine
and histamine metabolisms, has also been reported [18,19].
The physiopathological problem: a highly
complex biological system
At the molecular level, the picture of the problem in vivo is
extremely complex [20]. As stated in the first updated report
of COSTAction 922, many proteins (enzymes, receptors and
transporters) are involved in the amine cross-talk between
inflammatory and cancer cells [21]. In addition, many other
elements involved in intercellular communication and trans-
duction signal elements, having a high degree of tissue
specificity, should be characterized and taken into account
C
2007 Biochemical Society
43
Captulo 1
2.0
28.0
Health Implications of Dietary Amines 383
Figure 2 Intercellular communication among mast cells, cancer cells and endothelial cells
On the left, paracrine signal mediators positive for tumour progression; on the right, paracrine signal mediators negative
for tumour progression. The processes affected by the mediators are specied within parentheses. Arrows go from the
mediator-synthesizing cell to the target cell. Continuous lines, effects on tumour cells; dotted and dashed lines, effects on
mast cells; dotted lines, effects on endothelial cells.
to provide a full set of information on the problem. Figure 2
shows a brief scheme of the most relevant effects reported
so far. Histamine, tryptase and proteoglycans have been
described as essential for mast cell differentiation [22,23],
and as both angiogenesis and tumour progression modulators
[24,25]. Mast cell degranulation is followed by secretion of
different interleukins and growth factors that also can affect
cancer cell growth (and polyamine metabolism) even with
opposite roles: mast cells have therefore been described as
Dr Jekyll and Mr Hyde for tumour growth [9]. Of course,
the effects elicited by these factors are dependent on the signal
transduction pathways operative in the target cells. Mast
cells also produce invasiveness factors (for instance, matrix
metalloproteinases and other proteases), since they need these
activities to infiltrate into tissues during the final stages of
their maturation process. Nevertheless, these proteins could
also help in tumour invasion. From these findings, it is
deduced that another cell type must be added to this picture:
the endothelial cell, responsible for new vessel formation
(angiogenesis), an essential process for tumour survival,
invasion and metastasis [26]. From the same proteinogenic
precursor as polyamines, arginine, endothelial cells and some
tumours can produce nitric oxide, described as a regulator
of amine metabolism, inflammation and cell proliferation
[27].
The present degree of knowledge was reached with an
enormous quantity of work necessary for molecular charac-
terization of individual elements and metabolic pathways,
as well as for observations of biological effect. However,
all the useful information that could contribute to complete
the picture has been distributed between different research
areas (for instance, immunology, oncology, cardiovascular
and basic molecular biology and biochemical research),
which makes it more difficult to find its location, and con-
sequently restricts the perspective for formulation of further
experimental hypotheses.
The need for post-genomics tool
development
During the few last years, we have been asking for more holi-
stic approaches to characterize this highly complex biological
system [21,28]. The information generated by Genome Pro-
jects, as well as the results of many molecular and functional
genomics projects, is being accumulated in different public
databanks andbibliographies. It provides animportant source
of knowledge which needs the development of more efficient
tools to be located and analysed properly. It is clear that
biomathematics and bioinformatics (that is, systems biology
technologies) could help to reach this goal. The most recent
C
2007 Biochemical Society
Ral Montaez Martnez
44
2.0
28.0
384 Biochemical Society Transactions (2007) Volume 35, part 2
Figure 3 A proposal for a systemic approach to amine biochemistry, molecular biology and physiopathology
bibliography shares this analysis of the situation and supports
this assessment [20,2932].
Figure 3 shows a scheme of our proposal for combined
biocomputational and experimental approaches in order to
gain efficiency in the advance of a more systemic knowledge
of amine metabolism. More details on this proposal can be
obtained from the website http://www.asp.uma.es. In this
sense, we have worked on different action lines. Metabolic
modelling allows us to predict the limiting step(s) and the
more efficient way(s) for intervention under a given meta-
bolic state. Using this hypothesis, the first mathematical
predictive model for polyamine metabolism in mammals has
been developed and validated [33]. New biomodules, for
instance, arginine [34], histidine and methionine pathways,
as well as oxidative metabolism (as the source of acetyl-CoA)
(Figure 2) could be added to this first one, in order to get
a more integrated metabolic view of the amine metabolism.
This approach would indeed be helped by development of
improved methods for text mining, since most of the kinetic
constants are reported in the literature.
With respect to both intercellular communication and
intracellular signalling mechanisms that control gene ex-
pression, they could be decoded using a combination of
graph-theory-based interactomics studies and functional
genomics approaches. This approach needs to be assisted by
automated tools for integration of the information stored
in data banks (for instance, promoter cis elements, experi-
mentally assessed macromolecular interactions, and ex-
perimental functional genomics results on specific species,
tissues and physiopathological circumstances). At present, we
are also working on different projects on the development
of new biocomputational tools for these purposes [35].
Nevertheless, some specific hypotheses derived from in silico
predictions on amine-related genes could need further ex-
perimental validation. We advancedthe generationof a cDNA
macroarray of human amine-related probes for simultaneous
detection of gene expression alterations caused by a given
stimulus. At present, the array contains cDNA probes for
the most important proteins of the following metabolic path-
ways: polyamines, histamine metabolism (including recep-
tors), arginine (including urea cycle enzymes and inducible
nitric oxide synthase), methionine cycle and other inflam-
mation, cell migration and proliferation-related probes, many
of them mentioned in Figures 1 and 2. This first array is being
validated on human mast cells, a cell type with an important
lack of information concerning its intracellular signalling
network, and for which functional genomics studies have
not been frequent so far [3638]. Nevertheless, the array is
continuously growing and could be applied to many other
human cell types and physiopathological conditions.
In any case, once a specific macromolecular element (pro-
tein or nucleic acid) is located as the best target for inter-
vention in a given pathway, a deeper characterization of
structurefunction relationships of this macromolecule can
also be assisted in silico by macromolecular modelling and
molecular dynamics calculations that could provide valuable
insight for design of new and more specific structural and
functional modulators (Figure 3) [39].
This work was supported by Ram on-Areces Foundation, Grants
SAF2005-01812 (MEC, Spain), PI05-0327, CIBER of Rare Diseases
and the Spanish Network of Mastocytosis (MSC, Spain), as well as
Grants CVI-657 and CVI-267 (Andalusian Resarch Programme, PAI,
C
2007 Biochemical Society
45
Captulo 1
2.0
28.0
Health Implications of Dietary Amines 385
Andalusia, Spain). We thank Dr I. Fajardo for useful comments. F.C.-F,
P.C. and R.M. are predoctoral fellowship recipients from MEC, MSC
and PAI respectively. Thanks are due to COST 922 members for very
useful information provided to our group during the Action. We also
thank M. C anovas for pre-submission language correction.
References
1 Dy, M. and Schneider, E. (2004) Cytokine Growth Factor Rev. 15,
393410
2 Metcalfe, D.D., Baram, D. and Mekori, Y.A. (1997) Physiol. Rev. 77,
10331079
3 Moya-Garc a, A.A., Medina, M.A. and S anchez-Jim enez, F. (2005)
BioEssays 27, 5763
4 Coussens, L.M. and Werb, Z. (2002) Nature 420, 860867
5 Ai, W., Liu, Y., Langlois, M. and Wang, T.C. (2004) J. Biol. Chem. 279,
86848693
6 Akin, C. and Metcalfe, D.D. (2004) Annu. Rev. Med. 55, 419432
7 Garc a-Montero, A.C., Jara-Acevedo, M., Teodosio, C., S anchez, M.L.,
N u nez, R., Prados, A., Aldanondo, I., S anchez, L., Dom nguez, M., Botana,
L.M. et al. (2006) Blood 108, 23662372
8 Krauth, M.T., Agis, H., Aichberger, K.J., Simonitsch-Klupp, I., Mullauer, L.,
Mayerhofer, M., Bohm, A., Horny, H.P. and Valent, P. (2006)
Hum. Pathol. 37, 439447
9 Theoharides, T.C. and Conti, P. (2004) Trends Immunol. 25, 235241
10 Fajardo, I., Urdiales, J.L., Paz, J.C., Medina, M.A. and S anchez-Jim enez, F.
(2001) In COST Action 917 Biogenically Active Amines in Food.
Polyamines and Tumour Growth: Biogenically Active Amines in Food
Processing and Amines Produced by Bacteria (Morgan, D.M.L.,
Milovic, V., Krizek, M. and White, A., eds.), pp. 1013, Ofce for Ofcial
Publications of the European Communities, Luxembourg
11 Fajardo, I., Urdiales, J.L., Medina, M.A. and S anchez-Jim enez, F. (2001)
Biochem. Pharmacol. 61, 11011106
12 Fajardo, I., Urdiales, J.L., Paz, J.C., Chavarr a, T., S anchez-Jim enez, F. and
Medina, M.A. (2001) Eur. J. Biochem. 268, 768773
13 Seiler, N. and Raul, F. (2005) J. Cell. Mol. Med. 9, 623642
14 Medina, M.A., Urdiales, J.L., Rodr guez-Caso, C., Ram rez, F.J. and
S anchez-Jim enez, F. (2003) Crit. Rev. Biochem. Mol. Biol. 38, 2359
15 Medina, M.A., Urdiales, J.L., Mates, J.M., Nunez de Castro, I. and
Sanchez-Jimenez, F. (1991) Biochem. J. 280, 825827
16 Facchiano, F., Facchiano, A. and Facchiano, A.M. (2006) Front. Biosci. 11,
17581773
17 Seiler, N. (2004) Amino Acids 26, 217233
18 Frostesjo, L., Holm, I., Grahn, B., Page, A.W., Bestor, T.H. and Heby, O.
(1997) J. Biol. Chem. 272, 43594366
19 Suzuki-Ishigaki, S., Numayama-Tsuruta, K., Kuramasu, A., Sakurai, E.,
Makabe, Y., Shimura, S., Shirato, K., Igarashi, K., Watanabe, T. and
Ohtsu, H. (2000) Nucleic Acids Res. 28, 26272633
20 Hornberg, J.J., Bruggeman, F.J., Westerhoff, H.V. and Lankelma, J. (2006)
Biosystems 83, 8190
21 S anchez-Jim enez, F., Medina, M.A., Urdiales, J.L., Moya-Garc a, A.A.,
Rodr guez-Caso, C. and Rodr guez-Agudo, D., (2004) In Health
Implications of Dietary Amines: Review of Current Status (Wallace, H.M.
and Hughes, A., eds), pp, 4554, Ofce for Ofcial Publications of the
European Communities, Luxembourg
22 Hallgren, J. and Pejler, G. (2006) FEBS J. 273, 18711895
23 Ohtsu, H., Tanaka, S., Terui, T., Hori, Y., Makabe-Kobayashi, Y., Pejler, G.,
Tchougounova, E., Hellman, L., Gertsenstein, M., Hirasawa, N. et al.
(2001) FEBS Lett. 502, 5356
24 Ghosh, A.K., Hirasawa, N., Ohtsu, H., Watanabe, T. and Ohuchi, K. (2002)
J. Exp. Med. 195, 973982
25 Pos, Z., Safrany, G., Muller, K., Toth, S., Falus, A. and Hegyesi, H. (2005)
Cancer Res. 65, 44584466
26 Quesada, A.R., Mu noz-Ch apuli, R. and Medina, M.A. (2006)
Med. Res. Rev. 26, 483530
27 Satriano, J., Ishizuka, S., Archer, D.C., Blantz, R.C. and Kelly, C.J. (1999)
Am. J. Physiol. 276, C892C899
28 Medina, M.A., Correa-Fiz, F., Rodr guez-Caso, C. and S anchez-Jim enez, F.
(2005) J. Cell Mol. Med. 9, 854864
29 Carlberg, C. and Dunlop, T.W. (2006) Crit. Rev. Eukaryotic
Gene Expression 16, 122
30 Dopazo, J. (2006) Clin. Transl. Oncol. 8, 409415
31 Kolch, W., Calder, M. and Gilbert, D. (2005) FEBS Lett. 579, 18911895
32 Liu, E.T., Kuznetsov, V.A. and Miller, L.D. (2006) Cancer Cells 9,
245247
33 Rodr guez-Caso, C., Monta nez, R., Cascante, M., S anchez-Jim enez, F. and
Medina, M.A. (2006) J. Biol. Chem. 281, 2179921812
34 Monta nez, R., Rodr guez-Caso, C., S anchez-Jim enez, F. and Medina, M.A.
(2007) Amino Acids, in the press
35 Rodr guez-Caso, C., Medina, M.A. and Sol e, R.V. (2005) FEBS J. 272,
64236434
36 Inomata, N., Tomita, H., Ikezawa, Z. and Saito, H. (2005) Immunol. Lett.
98, 265271
37 Kuramasu, A., Kubota, Y., Matsumoto, K., Nakajima, T., Sun, X.M.,
Watanabe, T., Saito, H. and Ohtsu, H. (2001) FEBS Lett. 498, 3741
38 Nakajima, T., Matsumoto, K., Suto, H., Tanaka, K., Ebisawa, M.,
Tomita, H., Yuki, K., Katsunuma, T., Akasawa, A., Hashida, R. et al. (2001)
Blood 98, 11271134
39 Schames, J.R., Henchman, R.H., Siegel, J.S., Sotriffer, C.A., Ni, H. and
McCammon, J.A. (2004) J. Med. Chem. 47, 18791881
Received 25 September 2006
C
2007 Biochemical Society
Ral Montaez Martnez
46
2.0
28.0
47
Captulo 1
2.0
28.0
Amino Acids (2007) 33: 283289
DOI 10.1007/s00726-007-0521-4
Printed in The Netherlands
Polyamines: metabolism to systems biology and beyond
Minireview Article
R. Montan ez
1
, F. Sanchez-Jimenez
1
, J. F. Aldana-Montes
2
, and M. A

. Medina
1
1
Procel Group, Department of Molecular Biology and Biochemistry, University of Malaga, and Ciberer, Malaga, Spain
2
Khaos Group, Department of Computer Languages and Computing Sciences, University of Malaga, Malaga, Spain
Received November 29, 2006
Accepted February 1, 2007
Published online May 21, 2007; # Springer-Verlag 2007
Summary. Polyamines and the metabolic and physiopathological pro-
cesses in which they are involved represent an active eld of research that
has been continuously growing since the seventies. In the last years, the
trends in the focused areas of interest within this eld since the 1970s have
been conrmed. The impact of -omics in polyamine research remains
too low in comparison with its deep impact on other biological research
areas. These high-throughput approaches, along with systems biology and,
in general, more systemic and holistic approaches should contribute to a
renewal of this research area in the near future.
Keywords: Polyamines Systems biology Functional genomics
Proteomics Metabolomics Sematic web
Abbreviations: ADME=Tox, absorption, distribution, metabolism, ex-
creton and toxicity; HTML, hypertext markup language; HTTP, hypertext
transfer protocol; SAM, S-adenosyl methionine; siRNA, small interfer-
ent RNA
Introduction
Polyamines (putrescine, spermidine and spermine) are
aliphatic polycations present in almost all living species,
the exceptions being two orders of Archaea, namely,
Methanobacteriales and Halobacteriales (Hamana and
Matsuzaki, 1992). Their ubiquity and conservation across
evolution point to their importance for cell biology. In
fact, polyamines have pleiotropic effects with relevant
regulatory roles in macromolecular synthesis and cell pro-
liferation rates (Cohen, 1998; Thomas and Thomas, 2001;
Medina et al., 2003).
Although the prehistory of polyamine research can
be drawn back to the famous letter from Antonie van
Leeuwenhoek to the Royal Society of London in 1678,
the history of the scientic study of polyamines does
begin in the last quarter of the 19th century, with the
identication of spermine (1878), cadaverine (1886) and
putrescine (1889). Discovery and synthesis of spermidine
was achieved in 1927 (Cohen, 1998). The evolution of
this research topic along the last 86 years of its history
is depicted in Fig. 1, showing a continuous increase in the
publication rate, with a steady rise in the number of pub-
lications in the last 35 years. Two key cornerstone refer-
ences clearly show the evolution of polyamine research in
this last period of time: the monographic Polyamines
volume of Methods in Enzymology edited by Herbert
and Celia White Tabor in 1983 and the comprehensive
A Guide to Polyamines written in 1997 by Seymour
Cohen and published in 1998 (Tabor and Tabor, 1983;
Cohen, 1998). According to the seach criteria used in
Fig. 1, up to 1970 around 500 references on polyamines
had accumulated, meaning a mean value of 7 new refer-
ences per year. From 1971 to the date of publication of the
volume Polyamines, the publication rate drastically
increased to more than 300 new references per year. From
1984 to the date of publication of A Guide to Polyamines
there was a further increase in the publication rate up to
more than 500 references per year. Since then up to the
moment of writting this review, the rate has suffered a
new increase to more than 600 new references per year.
The growing accumulated number of references contain-
ing data on polyamines is currently around 18,000 ref-
Ral Montaez Martnez
48
2.0
28.0
erences according to the search criteria used in Fig. 1
(increased to 70,000 references according to a general
search using polyamine as a keyword).
The contents of the monographic Polyamines volume
of Methods in Enzymology edited by Herbert and Celia
White Tabor (1983) show that the main topics of poly-
amine research were as follows: a) Analytical and prepara-
tive methods for amines and metabolically related com-
pounds. b) Physiopathological and metabolic studies on
the effects of polyamine treatments or genetic manipula-
tion of their metabolism. c) Isolation and kinetic charac-
terization of the enzymes involved in polyamine metabo-
lism, including the study of inhibitors. d) Study of analogs
and derivatives.
According to the contents of A Guide to Polyamines by
Seymour Cohen (1998), the previously mentioned re-
search topics remained in the mainstream, but it can also
be observed a renewed interest for the study of: 1) poly-
amine metabolism in mammals, other animals, plants,
fungi and bacteria; 2) the role of polyamines in cancer;
and 3) the interactions of polyamines with macrobiomo-
lecules (proteins, RNA and DNA).
The rest of this minireview is devoted to analyse
the trends of polyamine research since the publication
date of A Guide to Polyamines and to describe future
prospects.
What has been published for the last 9 years
concerning polyamines?
A selective search (by adding new restriction criteria)
within the more than 6,000 references found to be
published since the publication date of A Guide to Poly-
amines allows to depict a classication of current re-
search trends (Fig. 2). Some clear trends deserve to be
commented:
1. The trends mentioned for the previous period (1984
97) conrm their prevalence within the preference of
the polyamine researchers.
2. More than 2=3 of the published work is devoted to
some aspects of metabolism.
3. Enzyme studies have still a remarkable yield (30% of
total production). Within this area, more than 40% of
papers are devoted to kinetic aspects. Ornithine decar-
boxylase remains the most popular enzyme. The num-
ber of studies devoted to this enzyme exceeds by
4-fold that of papers concerning spermidine, spermine
acetyltransferase and by 5-fold that of papers devoted
to study antizyme. The relative number of papers on
S-adenosyl methionine decarboxylase and polyamine
oxidases remains low.
4. The transport of polyamines remains a topic full of
uncertainties (Medina et al., 2003) and emerges as a
rapidly growing research subarea (more than 10% of
polyamine research total production within this period
of time).
5. The enzymologic, metabolic, physiological and phar-
macological studies concerning polyamine analogs and
derivatives remain in the mainstream.
6. Genetic manipulation with transgenic mice and mod-
ied cell lines provides new valuable information on
the pathophysiological roles of polyamines and the
enzymes involved in their metabolism. This subarea
is continuosly growing. It is noteworthy that in the last
three years the use of the potent technologies of siRNA
has entered with strength in the polyamine eld, pro-
viding the rst thirty-something papers based in the
use of this technology (see, for instance, Choi et al.,
2005; Loopez-Contreras et al., 2006).
7. Another strong subarea of research is that related to
the study of the interactions of polyamines with nu-
cleic acids or proteins, yielding more than 150 new
papers.
8. Finally, in relation with pathophysiological processes,
the connection of polyamines with cancer remains the
most productive topic, yelding almost the 15% of
newly released papers. Inammation, neurodegenera-
Fig. 1. Evolution of the rate of publication within polyamine research
since 1940. Data are the total number of publications (squares) and the
number of reviews (circles) per year, according to a search within the
PubMed database using the unrestricted query polyamine

OR putres-
cine OR spermidine OR spermine
284 R. Montanez et al.
49
Captulo 1
2.0
28.0
ive diseases and aging are the following topics and the
study of connections of polyamines with angiogenesis
emerges as a new topic of interest, with 33 articles
published in this period of time.
An especial mention should be given concerning review
papers. Almost a 10% of all the new references containing
data on polyamines are reviews. Among the 132 review
articles containing the words polyamine, putrescine,
spermidine or spermine, the general trends mentioned
above are conrmed. Table 1 contains a selection of repre-
sentative reviews related to the different subareas men-
tioned above.
What about the -omics world?
The launch of genome projects in the last decade of 20th
century marked the beginning of the -omics era within
biological research. Genomics, functional genomics,
transcriptomics, interactomics, metabolomics and other
-omics pervade current biochemical and molecular
biology research. It has been claimed that the advent of
this new -omics technologies will contribute to the
identication of new polyamine-regulated genes (Wallace
et al., 2003) and, in general, it is expected that these ap-
proaches will contribute to a new burst of additional data
concerning polyamines. In high contrast with these ex-
pectations, it is remarkable that up to the moment the
-omics approaches have had only a testimonial pres-
ence in the production within polyamine research area.
Within the more than 6,000 references published since
the publication date of A Guide to Polyamines, only six
papers are rescued when proteomics is added as a key-
Fig. 2. A simple classication of current research trends within the polyamine eld, according to the number of publications devoted since the date of
publication of A Guide to Polyamines to different areas and subareas. The query used in Fig. 1 was restricted by the addition of a keyword (describing
the area or subarea) to the original query by the connecting Boolean AND
Table 1. Representative reviews published in the last years and focused
on some of the different subareas within the polyamine research topic
Metabolism Seiler (2004)
Enzyme studies Binda et al. (2002)
ODC Kubota (1999)
Antizyme Mangold (2005)
SAMDC Pegg et al. (1998)
Transport Reguera et al. (2005)
Pharmacological studies Liang et al. (2006), Bienz et al.
(2005) and Seiler (2005)
Genetic studies Jaanne et al. (2005) and Pegg et al.
(2003)
Macromolecular interactions DAgostino et al. (2006) and
Bachrach (2004)
Pathophysiological processes Mionard et al. (2005)
Cancer Samejima (2006), Gerner et al.
(2004) and Bachrach (2004)
Inammation Satriano (2004)
Immunodegenerative diseases Jeevanandam et al. (2001)
Ageing Wu et al. (2000)
Polyamines: metabolism to systems biology and beyond 285
Ral Montaez Martnez
50
2.0
28.0
word, and only two of them are research papers focused
on polyamines in bacteria and plants (Franceschetti et al.,
2004; Pessione et al., 2005). The situation is even worst
for functional genomics, yielding eight papers and
none of them is a research paper focused on the study
of polyamines. Concerning metabolomics, only two
references are rescued, both of them corresponding to
metabolic proling studies carried out in plants, and only
one of them yielding some relevant results related to poly-
amines (Parr et al., 2005).
This scarcity of research papers devoted to study some
aspects of polyamines with the tools of the different new
-omics clearly points to a delay in the general incor-
poration of these tools by polyamine research groups.
This situation should change deeply in the near future.
Our research group is contributing within this issue with
a rst functional genomics approach to the study of
polyamine=histamine metabolic cross-talk (Chaves et al.,
2007).
Systems biology enters into scene
Much of last century biology was an attempt to reduce
biological phenomena to the behavior of molecules. In
spite of the great success of this approach, most biological
functions arise from interactions among many compo-
nents, yielding nonlinear behavior that has been ne tuned
by natural selection to achieve specic functional proper-
ties (Hartwell et al., 1999; Alberghina et al., 2004). There-
fore, a comprehensive understanding of biological func-
tions requires new systemic approaches, as those provided
by systems biology. In fact, systems biology uses to be
described as the analysis of the relationships among the
elements in a system in response to genetic or environ-
mental perturbations, with the goal of understanding the
system as a whole (Weston and Hood, 2004; Yang et al.,
2005). Systems biology approaches are hypothesis-driven
and involve iterative rounds of model building, predic-
tion, experimentation, model renement and development
(Kitano, 2001, 2002a, b; Weston and Hood, 2004). Within
this framework, computational biology contributes with
former and newly designed tools for both knowledge dis-
covery (or data-mining) and model-and-simulation-based
analysis.
During the last years, we have been claiming for more
holistic approaches to characterize the biological roles of
biogenic amines and in particular- polyamines (Medina
et al., 2003, 2005; Sanchez-Jimenez et al., 2007). Due to
the pleiotropic and important roles of polyamines, their
metabolism has long been the focus of biochemical stud-
ies that have provided extensive and detailed information
concerning each of the enzymes and metabolites of the
pathway (Wallace et al., 2003; Pegg et al., 2003; Pegg,
2006). However, most of this information is very disperse
and not integrated in a comprehensive, systemic frame-
work. On the other hand, many therapeutic strategies
based on the specic inhibition of one of the key enzymes
of polyamine metabolism have failed, mostly due to the
presence of compensating mechanisms in polyamine
metabolism that contribute to the buffering of those ef-
fects elicited when only an enzyme is the target (Medina
et al., 2005). The structure of the reaction diagram of
polyamine metabolism in mammals is relatively complex,
consisting of a bi-cycle having two required entrances
(ornithine and S-adenosylmethionine) and several alterna-
tive outwards. For most of the reactions, both activities
and turnover rates of enzymes depend on polyamine con-
centrations in a non-linear way. Therefore, the behavior of
the full pathway in response to genetic and environmental
perturbations cannot be easily deduced from the reaction
diagram itself. Nevertheless, if the behavior of the ele-
ments of a system is known, they can be assembled in a
model to acquire a global knowledge of the system. This
is known as bottom-up approach. In this case, the global
behavior of polyamine metabolism taken as a biomodule
(Hartwell et al., 1999) can be investigated with a mathe-
matical model, which describes the reactions and interac-
tions among its components. We have recently described
the rst basic mathematical modeling of polyamine me-
tabolism in mammals based on available experimental
metabolite concentrations and kinetic data (Rodr guez-
Caso et al., 2006). In this work, we show that polyamine
homeostasis is not only controlled by the key enzymes
but also acetyl-CoA and S-adenosylmethionine (SAM)
availability, suggesting metabolic connections with other
biological processes, as aerobic glucose (and fatty acid)
consumption and processes involving SAM as a methyl
donor (including gene regulation by DNA=histone methyl-
ation). New metabolic modules will be added to this rst
model, and we are working on this line (Montanez et al.,
2007). This and other systemic approaches to the study
of polyamines will allow to ascertain whether potential
strong modulators of polyamine metabolism are expected
to induce relevant effects administered either alone or in
combination.
Systems biology can also be used for the prediction
or determination of absorption, distribution, metabolism,
excretion and toxicity (known as ADME=Tox in pharma-
cology jargon) properties of compounds before they reach
the clinic (Ekins et al., 2005). This emergent area of re-
286 R. Montanez et al.
51
Captulo 1
2.0
28.0
search could be applied for the study of new modulators
of polyamine metabolism with potential pharmacological
interest. This has been the case of previously described
metabolically programmed polyamine analog antidiar-
rheals (Bergeron et al., 1996).
The identication of potentially relevant regulatory ele-
ments will establish the interconnections among amine-
related proteins and genes, as well as the transcription
factors that control their regulation. This can be assessed
by the use of algorithms to analyze the amine-related
macromolecules in combination with experimental ap-
proaches (e.g. expression microarrays, proteomics, and
other experimental methods to assess proteinprotein inter-
actions and=or coexpressions). The topological approaches
provided by graph theory is a suitable approximation to
extract valuable information from cellular interaction net-
works. The rst efforts have been done on the human
transcription factor network (Rodr guez-Caso et al., 2005).
Similar studies can be carried out for other aspects of
the intracellular and intercellular communication network
involving polyamines. To improve the efciency of this
task, development of new tools for automatic integration
of the information are very helpful and it is one of our
current priorities. From the analysis of the state of the art
on the polyamine eld, we detect the following needs that
could be solved by systems biology technologies:
1. Development and validation of new computational tools
for integration of biochemical and molecular infor-
mation stored in public databases containing properties
of individual elements (metabolites, proteins, genes), as
well as for assessed interactions among them.
2. Development of a new database containing bibliogra-
phic information on biochemistry, molecular biology
and physiopatological roles of biogenic amines and
related compounds by using text mining techniques.
3. New bioinformatic tools for structural and functional
predictions on the molecules and pathways related to
amine metabolism and biological missions.
To contribute to fulll these purposes, we have an-
nounced (Medina et al., 2007) and started the so-called
Amine System Project (www.asp.uma.es). This project
combines our most recent efforts (mentioned above) in
development of predictive models (item 3), with our pres-
ent aims for development of a prototype system for inte-
gration of metabolic and functional data concerning bio-
genic amine-related molecules and pathways (items 1 and
2). Since most of the control mechanisms of the essential
processes for any biological system (maintenance of the
genetic material, gene expression and signalling) are sup-
ported on proteinprotein interactions, the automated
search and integration of proteinprotein interactions con-
cerning the polyamine eld is another goal of our project
in the long-term. To achieve a maximum efciency in this
automated search and integration, an optimum system
should be able to reach a high degree of integration
among any information repository, as well as the maxi-
mum interoperability among the different data analysis
tools. Furthermore, it should avoid redundant information,
discriminate (to detect and to score) a condence degree
for each interaction, and provide the user with a solution
vector compatible with as much independent graphic tools
as possible. Finally, an additional facility could be ob-
tained by integration of a minimum algorithm set to the
system to carry out some graph theory calculations (for
instance, connectivity and clustering). In our opinion, dev-
elopment of ontology-based systems could help to ad-
vance in development of new and better information inte-
gration services. Ontologies provide a formal representa-
tion of the real world, shared by a sufcient amount of
users, by dening concepts and relationships among them.
In order to provide semantics to web resources, elements
(instances) of such concepts and relationships are used to
annotate them. These annotations over the resources are
the foundation of the Semantic Web. Semantic Web tech-
nologies provide a natural and exible solution for com-
bining two levels of abstraction, the data level and the
knowledge level, which are related by means of metadata.
Thus, ontologies provide necessary elements to make
explicit the database semantics, allowing the manage-
ment of great amounts of knowledge distributed among
different databases. According to the World Wide Web
inventor Tim Berners-Lee, the Semantic Web could be
the key to unlocking scientic data thats sequestered by
disparate applications formats and organizational lim-
itations, and could allow scientists to harness computa-
tions full power. Furthermore, Berners-Lee (who also
invented key components of the World Wide Web such
as HTTP Hypertext Transfer Protocol and HTML
Hypertext Markup Language in the late 1980s) thinks
that life scientists in particular could nd the Semantic
Web a useful tool, and in so doing, provide leadership to
lots of other elds in implementing this next-generation
Web technology.
. . . And beyond: the need for more systemic
and holistic approaches to polyamines in biology
Nowadays, in our opinion, systems biology is a contro-
versial forum of open and active discussion, mostly due to
Polyamines: metabolism to systems biology and beyond 287
Ral Montaez Martnez
52
2.0
28.0
the many different interpretations of its aims and tools. In
the wikipedia, systems biology is dened as an aca-
demic eld that seeks to integrate different levels of infor-
mation to understand how biological systems function.
The identication of systems biology as an academic
eld (namely, an additional speciality of biology) is sus-
picious. In fact, since the objects of study of biology
(cells, organisms, populations, ecosystems) are all of them
complex systems, should not the whole biology be con-
sidered as a systems biology? We are afraid that most
of the scientists converted to systems biology make an
instrumental use of it to manage the huge amount of
information provided by new -omics technologies and
bioinformatics. Although this instrumental approach is
indeed also required, this cannot surpass the limitations
of reductionism. It is our understanding that an authentic
holistic and systemic view of biology should be based on
the analysis of the relationships among elements of a
biological system in a given steady-state, as well as along
the response of the system against any perturbation of its
environment, with the nal aim to know not only the
system itself but also its dynamics. To achieve this goal,
the approaches provided by the general theory of systems,
the theory of dynamical systems and, in general, the new
sciences of complexity should be applied. Complexity
is a keyword, which was identied as one of the three
main challenges of biology (the others are consilience
and communication) for the incoming century in an
inuential editorial published in BioEssays (1999).
Acknowledgements
This work was supported by a grant from Fundacioon Ramoon Areces,
Grants SAF2005-01812 and TIN2005-09098-C05-01 (Ministry of Edu-
cation and Sciences, Spain), and CVI-267 and CVI-657 (Andalusian
Resarch Programme, PAI, Andalusia, Spain).
References
Alberghina L, Chiaradonna F, Vanoni M (2004) Systems biology and the
molecular circuits of cancer. Chembiochemistry 5: 13221333
Bachrach U (2004) Polyamines and cancer. Amino Acids 26: 307309
Bergeron RJ, Yao GW, Yao H, Weimar WR, Sninsky CA, Raisler B, Feng
Y, Wu Q, Gao F (1996) Metabolically programmed polyamine analo-
gue antidiarrheals. J Med Chem 39: 24612471
Bienz S, Bisegger P, Guggisberg A, Hesse M (2005) Polyamine alkaloids.
Nat Prod Rep 22: 647658
Binda C, Mattevi A, Edmondson DE (2002) Structure-function relation-
ships in avoenzyme-dependent amine oxidations: a comparison
of polyamine oxidase and monoamine oxidase. J Biol Chem 277:
2397323976
Chaves P, Correa-Fiz AMF, Melgarejo E, Urdiales JL, Medina MA,
Sanchez-Jimenez F (2007) Development of an expression macroarray
for amine metabolism-related genes. Amino Acids (in press)
Choi KS, Suh YH, Kim WH, Lee TH, Jung MH (2005) Stable siRNA-
mediated silencing of antizyme inhibitor: regulation of ornithine
decarboxylase activity. Biochem Biophys Res Commun 328: 206212
Cohen SS (1998) A guide to polyamines. Oxford University Press,
New York
DAgostino L, di Pietro M, di Luccia A (2006) Nuclear aggregates of
polyamines. IUBMB Life 58: 7582
Editorial (1999) Consilience, complexity and communication: three
challanges at the start of the new century. BioEssays 21: 983984
Ekins S, Nikolsky Y, Nikolskaya T (2005) Techniques: Application of
systems biology to absorption, distribution, metabolism, excretion and
toxicity. Trends Pharmacol Sci 26: 202209
Franceschetti M, Perry B, Thompson B, Hanfrey C, Michael AJ (2004)
Expression proteomics identies biochemical adaptations and defense
responses in transgenic plants with perturbed polyamine metabolism.
FEBS Lett 576: 477480
Gerner EW, Meyskens FL Jr (2004) Polyamines and cancer: old mole-
cules, new understanding. Nat Rev Cancer 4: 781792
Hamana K, Matsuzaki S (1992) Polyamines as a chomotaxonomic marker
in bacterial systematics. Crit Rev Microbiol 18: 261283
Hartwell LH, Hopeld JJ, Leibler S, Murray AW (1999) From molecular
to modular cell biology. Nature 402: C47C52
Jaanne J, Alhonen L, Keinanen TA, pietila M, Uimari A, Pirinen E,
Hyvonen MT, Jarvinen A (2005) Animal disease models generated
by genetic engineering of polyamine metabolism. J Cell Mol Med 9:
865882
Jeevanandam M, Petersen SR (2001) Clinical role of polyamine analysis:
problem and promise. Curr Opin Clin Nutr Metab Care 4: 385390
Kitano H (2001) Foundations of systems biology. MIT Press, London
Kitano H (2002a) Computational systems biology. Nature 420: 206210
Kitano H (2002b) Systems biology: a brief overview. Science 295:
16621664
Kubota S (1999) Polyamine and ODC. Nippon Rinsho 57: 570574
Liang F, Wan S, Li Z, Xiong X, Yang L, Zhou X, Wu C (2006)
Medical applications of macrocyclic polyamines. Curr Med Chem
13: 711727
Loopez-Contreras AJ, Loopez-Garc a C, Jimmenez-Cervantes C, Cremades
A, Penael R (2006) Mouse ornithine decarboxylase-like gene encodes
an antizyme inhibitor devoid of ornithine and arginine decarboxylase
activity. J Biol Chem 281: 3089630906
Mangold U (2005) The antizyme family: polyamines and beyond.
IUBMB Life 57: 671676
Medina MA, Aldana JF, Villatoro FR, Claros G, Urdiales JL, Trelles O,
Sanchez-Jimenez F (2007) Amine system project. Inamm Res
(in press)
Medina MA, Correa-Fiz F, Rodr guez-Caso C, Sanchez-Jimenez F (2005)
A comprehensive view of polyamine and histamine metabolism to the
light of new technologies. J Cell Mol Med 9: 854864
Medina MA, Urdiales JL, Rodr guez-Caso C, Ram rez FJ, Sanchez-
Jimenez F (2003) Biogenic amines and polyamines: similar biochem-
istry for different physiological missions and biomedical applications.
Crit Rev Biochem Mol Biol 38: 2359
Moinard C, Cynober L, de Bandt JP (2005) Polyamines: metabolism and
implications in human diseases. Clin Nutr 24: 184197
Montanez R, Rodr guez-Caso C, Sanchez-Jimenez F, Medina MA (2007)
In silico analysis of arginine catabolism as a source of nitric oxide or
polyamines in endothelial cells. Amino Acids (in press)
Parr AJ, Mellon FA, Colquhoun IJ, Davies HV (2005) Dihydrocaffeoyl
polyamines (kukoamine and allies) in potato (Solanum tuberosum)
tubers detected during metabolite proling. J Agric Food Chem 53:
54615466
Pegg AE (2006) Regulation of ornithine decarboxylase. J Biol Chem 281:
1452914532
Pegg AE, Feith DJ, Fong LY, Coleman CS, OBrien TG, Shantz LM
(2003) Transgenic mouse models for studies of the role of polyamines
288 R. Montanez et al.
53
Captulo 1
2.0
28.0
in normal, hypertrophic and neoplastic growth. Biochem Soc Trans 31:
356360
Pegg AE, Xiong H, Feith DJ, Shantz LM (1998) S-adenosymethionine
decarboxilase: structure, function and regulation by polyamines. Bio-
chem Soc Trans 26: 580586
Pessione E, Mazzoli R, Giuffrida MG, Lamberti C, Garc a-Moruno E,
Barello C, Conti A, Giunta C (2005) A proteomic approach to studying
biogenic amine producing lactic acid bacteria. Proteomics 5: 687698
Reguera RM, Tekwani BL, Balana-Fouce R(2005) Polyamine transport in
parasites: a potential target for new antiparasitic drug development.
Comp Biochem Physiol C: Toxicol Pharmacol 140: 151164
Rodr guez-Caso C, Medina MA, Sole RV (2005) Topology, tinkering and
evolution of the human transcription factor network. FEBS J 272:
64236434
Rodr guez-Caso C, Montanez R, Cascante M, Sanchez-Jimenez F,
Medina MA (2006) Mathematical modeling of polyamine metabolism
in mammals. J Biol Chem 281: 2179921812
Samejima K (2006) Regulation of polyamine and cancer. Yakugaku
Zasshi 126: 529542
Sanchez-Jimenez F, Montanez R, Correa-Fiz F, Chaves P, Rodr guez-Caso
R, Urdiales JL, Aldana JF, Medina MA (2007) The usefulness of post-
genomics tools for characterization of the amine crosstalk in mamma-
lian cells. Biochem Soc Trans (in press)
Satriano J (2004) Arginine pathways and the inammatory response:
interregulation of nitric oxide and polyamines. Amino Acids 26: 321329
Seiler N (2004) Catabolism of polyamines. Amino Acids 26: 217233
Seiler N (2005) Pharmacological aspects of cytotoxic polyamine analogs
and derivates for cancer therapy. Pharmacol Ther 107: 99109
Tabor H, Tabor CW (eds) (1983) Polyamines. Methods in Ezymology,
vol 94. Academic Press, New York
Thomas T, Thomas TH (2001) Polyamines in cell growth and cell death:
molecular mechanisms and therapeutic applications. Cell Mol Life Sci
58: 244258
Wallace HM, Fraser AV, Hughes A (2003) A perspective of polyamine
metabolism. Biochem J 376: 114
Weston AD, Hood L (2004) Systems biology, proteomics, and the future
of health care: toward predictive, preventative, and personalized med-
icine. J Proteome Res 3: 179196
Wu G, Meininger CJ (2000) Arginine nutrition and cardiovascular func-
tion. J Nutr 130: 26262629
Yang CR, Shapiro BE, Mjolsness ED, Hateld GW (2005) An enzyme
mechanism language for the mathematical modeling of metabolic
pathways. Bioinformatics 21: 774780
Authors address: Miguel A

ngel Medina, Procel Group, Department of

Molecular Biology and Biochemistry, Faculty of Sciences, University of
Malaga, E-29071 Malaga, Spain,
E-mail: medina@uma.es
Polyamines: metabolism to systems biology and beyond 289
Ral Montaez Martnez
54
2.0
28.0
55
Captulo 1
2.0
28.0
Information integration of proteinprotein interactions as essential
tools for immunomics
Rau l Montan ez
a
, Ismael Navas-Delgado
b
, Miguel A

ngel Medina
a
,
Jose F. Aldana-Montes
b
, Francisca Sanchez-Jimenez
a,
*
a
Department of Molecular Biology and Biochemistry, Centre for Biomedical Research on Rare Diseases, University of Malaga, E-29071 Malaga, Spain
b
Department of Computer Languages and Computing Science, University of Malaga, E-29071 Malaga, Spain
Received 18 November 2006; accepted 11 December 2006
Available online 17 April 2007
Abstract
After a brief introduction to point out the necessity to advance for a global understanding of the macromolecular interactions occur-
ring during the immune system development and responses, Section 2 will be devoted to analyse the current tools for an automatic loca-
tion of information on these proteinprotein interactions in the web. In the next section (Section 3), we will point out dierent action lines
to improve these tools and, consequently, to increase the eciency to establish (to understand) the protein network skeleton that con-
trols our immune responses. Finally, we will briey present our current strategy and work to advance towards this goal.
2007 Elsevier Inc. All rights reserved.
Keywords: Proteinprotein interactions; Data mining; Databanks; Networks; Signaling; Histamine; Inammation
1. Introduction
Most of the control mechanisms of the essential pro-
cesses for any biological system (maintenance of the genetic
material, gene expression and signaling) are supported on
proteinprotein interactions, as the basic skeleton for living
organism self-organization and homeostasis [1]. In this sys-
temic view of life, it is not enough to characterize the bind-
ing of a given protein to its corresponding target, or the
capability of several enzymes to form a signaling complex
or metabolon, but it is also important to know the global
interaction pattern, because self-organization cannot be
deduced from the properties of the single elements.
Consequently, the understanding of the structural data
concerning this global network skeleton, as well as those
of its one-to-one element interactions, is an essential
knowledge (but still a dawning one) for an ecient advance
in Systems Biology.
The immune system is an excellent example of the
concepts exposed above, an extremely rened biological
system that involves intercellular communication net-
works (antigenantibody interactions, cytokine and
growth factor secretions and receptions, etc), as well as
intracellular signaling pathways and other reactions also
based on proteinprotein interactions (protein phosphor-
ylation, degradation, sorting, etc). Important eorts are
being made to collect and to predict inter-peptidic inter-
actions relevant for immunology (see this issue). Net-
works of molecular interactions are widely studied to
reveal the complex roles played by gene products and
cellular environments in biological processes. The com-
plexity of interactions in a given biological system are
being analysed by graph theory [2].
2. What we currently have got for storage and automated
analysis of information on proteinprotein interactions
When a new interactome is built by the current informa-
tion integration tools, nodes represent macromolecules and
connecting segments represent specic interactions. Nodes
0008-8749/$ - see front matter 2007 Elsevier Inc. All rights reserved.
doi:10.1016/j.cellimm.2006.12.008
*
Corresponding author. Fax: +34 952131674.
E-mail address: kika@uma.es (F. Sanchez-Jimenez).
URL: www.asp.uma.es (F. Sanchez-Jimenez).
www.elsevier.com/locate/ycimm
Cellular Immunology 244 (2006) 8486
Ral Montaez Martnez
56
2.0
28.0
are associated with additional information about the genes/
proteins; for instance, chromosome number and gene loca-
tion, intracellular location of the protein, or their Gene
Ontology (GO) classications. All this information is
extensively stored in several data warehouse (Expasy, Gene
Bank, or Gene Ontology, among others) [35]. But the
problem is the edge. In the graph, an edge could have many
dierent meanings: co-expression and/or co-regulation in
dierent cellular processes, the binding of a transcription
factor to a promoter/regulatory region, recorded evidence
of macromolecular interaction or inclusion in multi-protein
complexes, participation in the same signaling or metabolic
pathway, or even molecular co-evolution. It is possible to
dene six interaction subtypes in proteinprotein interac-
tion only [6].
Protein interactions have usually been analysed indi-
vidually by genetic, biochemical and biophysical tech-
niques. However, recently, new methods have been
developed for high-throughput macromolecular interac-
tion analysis, including both experimental and biocompu-
tational approaches [6]. The high speed for the
generation of new data has generated a problem, that
is, standardization of their notations in databases [7].
In the most signicant interaction databases [8], the
information provided by these methods is stored, curated
and commonly linked to node data warehouses, but the
type of interaction is not specied. This lack of precision
in interaction description is a major problem of the
high-throughput data repositories. Further eorts are
therefore essential to improve the quality of the pro-
teinprotein interactions databases and, consequently,
the biological information that can be inferred from
the interactome graphs.
Each data base has its own extraction, curation and
storage protocols, and not all of them explore the same sci-
entic papers. In fact, intersection and overlap among
these databases are small and, therefore, the information
is complementary in many cases; thus it should be unied
to increase and to improve the knowledge about interac-
tion networks. In this way, several databases are working
on trying to integrate all this information on exible plat-
forms: BiologicalNetwork, Agile Protein Interaction Data-
Analyzer (APID), KEEG or Protein Launge [912].
Nevertheless, the present approaches are not enough to
provide systematic information about the interaction
characteristics.
3. What we need for ecient automated analysis of
information on proteinprotein interactions. Some
suggestions to get it
In addition to a more systematic system to indicate the
properties of the notated interactions, we also consider
the following facts to achieve the maximum eciency in
the automated search and integration of proteinprotein
interactions information:
An optimum system should be able to reach a high
degree of integration among any information repository,
as well as the maximum interoperability among the dif-
ferent data analysis tools.
To avoid redundant information. An important source
of redundancy is the diversity in the identier and nota-
tions used for a same protein in dierent repositories. In
this sense, an optimum system should be able to locate
and unify the information on a given molecule stored
under any of its identiers.
To discriminate (to detect and to score) a condence
degree for each interaction. A simple (but slanting)
approach to it could be provided by the quotient
between the number of repositories-containing informa-
tion on the interaction and the total number of queried
databases. Nevertheless, it is clear that other facts
should be taken into account for a more accurate esti-
mation of condence; for instance, the method(s) used
to deduce such an interaction (in silico, in vitro and/or
in vivo). This information is often available in databases,
but the dierent technologies (for instance, bioinfor-
matic predictions, 2-hybrids, pool-down, uorescence
correlation spectroscopy, uorescence resonance energy
transfer) should (need to) be ranked (by consensus
among the scientic community) and their values inte-
grated in a global condence score algorithm for each
given result upon request by the user.
The graphic representation of the results is also an
important issue. Thus, an ideal system should provide
the user with a solution vector compatible with as much
independent graphic tools as possible, for instance,
Cytoscape, Osprey and VisANT [1315]. In agreement
with our initial assessment, the possibility to represent
dierences among interaction types is stressed.
Finally, an additional facility could be obtained by the
integration of a minimum algorithm set to the system
in order to carry out some graph theory calculations
(connectivity, clustering, etc).
In our opinion, the development of ontology-based sys-
tems could help to advance in development of new and bet-
ter information integration services, in general, due to the
following reasons. Ontologies provide a formal representa-
tion of the real world, shared by a sucient amount of
users, by dening concepts and relationships among them.
In order to provide semantics to web resources, elements
(instances) of such concepts and relationships are used to
annotate them. These annotations over the resources are
the foundation of the Semantic Web [16]. Thus, ontology
denition languages such as OWL [17] provide mechanisms
to dene classes (for example, bound polypeptide, cell,
organism...) and properties (for example, experimental
technology, reference. . .). Semantic Web technologies pro-
vide a natural and exible solution for the combination of
two levels of abstraction, the data level and the knowledge
level, which are related by means of metadata. Thus,
ontologies provide necessary elements to make explicit
R. Montanez et al. / Cellular Immunology 244 (2006) 8486 85
57
Captulo 1
2.0
28.0
the database semantics, allowing to manage great amounts
of knowledge distributed among these dierent databases.
Consequently, it goes beyond XML, RDF or RDF-S in
its ability to represent a machine interpretable content on
the web, so making possible to provide the necessary ele-
ments to achieve the level of interoperability required by
the highly dynamic and integrated bioinformatics applica-
tions. The use of ontology-based systems compels to use
standard pre-dened ontologies, as Gene Ontology or Pro-
tein Ontology [5,18] that are enriched with new relation-
ships, so that any of the problems mentioned above
could be surpassed.
We advance that we are currently working on a proto-
type of proteinprotein interaction tool that is able to inte-
grate results from dierent repositories; in addition, the
user is provided with a condence score for the interactions
and a vector compatible with many other independent gra-
phic tools. The tool will be further moved to an ontology-
based system and will be distributed by the web page
www.asp.es. A rst version of it was used to generate the
human transcription factor network [19]. This work
revealed that PPAR-gamma behaves as a connector
between modules of inammation- and cancer-related
genes, both modules being also related to histamine. In
its present stage, the prototype is being used to analyze
intracellular transduction pathways of pro-inammatory
and oxidative stress signals, which are also highly related
to both amine metabolism and immune system physiopath-
ological responses.
Acknowledgments
This work is part of a multidisciplinary excellence Grant
(CVI-657) Amine System Project (ASP) funded by the
Andalusian Government Research Programme, and Pro-
jects SAF 2005-01812 and TIN2005-09098-C05-01 (Span-
ish Ministry of Education and Science). The present
activities are coordinated by the respective seniors of the
dierent Science and Technology elds: JF Aldana-Montes
(www.khaos.es and www.bmbq.uma.es/procel). Thanks
are due to F. Villatoro, C. Rodr guez-Caso, M.M. Rojano,
I. Morillas, R. Fernandez de la Cruz, and M.G. Claros for
their helpful inputs and suggestions during evolution of
these activities.
References
[1] T. Misteli, The concept of self-organization in cellular architecture, J.
Cell Biol. 155 (2) (2001) 181185.
[2] M.E.J. Newman, The structure and function of complex networks,
SIAM Rev. 45 (2003) 167256.
[3] E. Gasteiger, A. Gattiker, C. Hoogland, I. Ivanyi, R.D. Appel, A.
Bairoch, ExPASy: the proteomics server for in-depth protein knowl-
edge and analysis, Nucleic Acids Res. 31 (2003) 37843788.
[4] D.A. Benson, I. Karsch-Mizrachi, D.J. Lipman, J. Ostell, D.L.
Wheeler, GenBank, Nucleic Acids Res. 34 (Database issue) (2006) 16
20.
[5] M. Ashburner, C.A. Ball, J.A. Blake, D. Botstein, H. Butler, J.M.
Cherry, A.P. Davis, K. Dolinski, S.S. Dwight, J.T. Eppig, M.A.
Harris, D.P. Hill, L. Issel-Tarver, A. Kasarskis, S. Lewis, J.C. Matese,
J.E. Richardson, M. Ringwald, G.M. Rubin, G. Sherlock, The Gene
Ontology Consortium. Gene Ontology: tool for the unication of
biology, Nat. Genet. 25 (2000) 2529.
[6] J. de Las Rivas, A. de Luis, Interactome data and databases: dierent
types of protein interaction, Comp. Funct. Genom. 5 (2004) 173178.
[7] C. von Mering, R. Krause, B. Snel, M. Cernell, S.G. Oliver, S. Fields,
P. Bork, Comparative assessment of large-scale data sets of protein
protein interactions, Nature 417 (2002) 399403.
[8] Data Base Issue (January), Nucleic Acid Res. 34 (2006).
[9] M. Baitaluk, M. Sedova, A. Ray, A. Gupta, BiologicalNetworks:
visualization and analysis tool for systems biology, Nucleic Acids
Res. 34 (Web Server issue) (2006) 466471.
[10] C. Prieto, J. De Las Rivas, APID: Agile Protein Interaction
DataAnalyzer, Nucleic Acids Res. 34 (Web Server issue) (2006)
298302.
[11] M. Kanehisa, S. Goto, M. Hattori, K.F. Aoki-Kinoshita, M. Itoh, S.
Kawashima, T. Katayama, M. Araki, M. Hirakawa, From genomics
to chemical genomics: new developments in KEGG, Nucleic Acids
Res. 34 (Database issue) (2006) 354357.
[12] Available from: <http://www.proteinlounge.com/default.asp/>.
[13] P. Shannon, A. Markiel, O. Ozier, N.S. Baliga, J.T. Wang, D.
Ramage, N. Amin, B. Schwikowski, T. Ideker, Cytoscape: a software
environment for integrated models of biomolecular interaction
networks, Genome Res. 13 (11) (2003) 24982504.
[14] P.M. Gordon, C.W. Sensen, Osprey: a comprehensive tool employing
novel methods for the design of oligonucleotides for DNA sequencing
and microarrays, Nucleic Acids Res. 32 (17) (2004) e133.
[15] Z. Hu, J. Mellor, J. Wu, C. DeLisi, VisANT: an online visualization
and analysis tool for biological interaction data, BMC Bioinformatics
5 (17) (2004).
[16] T. Berners-Lee, J. Hendler, O. Lassila, The Semantic Web, in:
Scientic American, 2001.
[17] D.L. McGuinness, F. van Harmelen, Owl web ontology language
overview. w3c recommendation, 2004. [Online]. Available from:
<http://www.w3.org/TR/owl-features/>.
[18] Available from: <http://proteinontology.info/>.
[19] C. Rodr guez-Caso, M.A. Medina, R.V. Sole, Topology, tinkering
and evolution of the human transcription factor network, FEBS J.
272 (24) (2005) 64236434.
86 R. Montanez et al. / Cellular Immunology 244 (2006) 8486

2.0
28.0
Buscando la ve-
rosimilitud en un
mundo abierto
2.0
28.0
La supervivencia en Ciencia o Tecnologa
parece haberse convertido en una compe-
ticin encarnizada. Todos necesitamos
producir conocimiento o servicios originales
lo antes posible para que nuestro impacto
sea notorio. En el caso de las bases de da-
tos, la prisa inicial por dejar plasmadas las
autoras y colonizar nuevos nichos de co-
nocimiento, en mi opinin, impidi consen-
suar qu informacin era importante y c-
mo deba de identificarse y estructurarse.
Por otro lado, propici un cierto grado de
redundancia justificada por pequeos por-
centajes de cobertura, informacin conte-
nida, etc. En la actualidad, esos problemas
estn siendo resueltos por los grandes con-
sorcios de bases de datos. Pese a ello,
cremos que poder disponer de una apli-
cacin automatizada capaz de: i) recupe-
rar la informacin, ii) prever qu informa-
cin puede ser til al usuario, y iii) consen-
suarla bajo un estndar de referencia, sera
de gran utilidad y facilitara la correcta abs-
traccin de modelos biolgicos (no slo a
nosotros, sino a otros bilogos de sistemas).
Sinopsis
61
2.0
28.0
Buscando la verosimilitud en un mundo abierto:
Sistemas de integracin de la informacin basados en tecnologa de web semn-
tica.
Por qu esta magnfica tecnologa cientfica, que ahorra trabajo y
nos hace la vida mas fcil, nos aporta tan poca felicidad? La repues-
ta es esta, simplemente: porque an no hemos aprendido a usarla
con tino
Albert Einstein
2
Ral Montaez Martnez
62
2.0
28.0
63
Captulo 2
2.0
28.0
BioMed Central
Page 1 of 14
(page number not for citation purposes)
BMC Bioinformatics
Open Access
Research
AMMO-Prot: amine system project 3D-model finder
Ismael Navas-Delgado
1
, Ral Montaez
2
, Almudena Pino-ngeles
2
,
Aurelio A Moya-Garca
2
, Jos Luis Urdiales
2
, Francisca Snchez-Jimnez
2

and Jos F Aldana-Montes*
1
Address:
1
Computer Languages and Computing Science Department, University of Mlaga, Mlaga, 29071, Spain and
2
Molecular Biology and
Biochemistry Department. University of Mlaga, and CIBER of Rare Diseases, National Institute of Health, Mlaga, 29071, Spain
Email: Ismael Navas-Delgado - ismael@lcc.uma.es; Ral Montaez - raulemm@gmail.com; Almudena Pino-ngeles - almupino@gmail.com;
Aurelio A Moya-Garca - amoyag@uma.es; Jos Luis Urdiales - jlurdial@uma.es; Francisca Snchez-Jimnez - kika@uma.es; Jos F Aldana-
Montes* - jfam@lcc.uma.es
* Corresponding author Equal contributors
Abstract
Background: Amines are biogenic amino acid derivatives, which play pleiotropic and very
important yet complex roles in animal physiology. For many other relevant biomolecules,
biochemical and molecular data are being accumulated, which need to be integrated in order to be
effective in the advance of biological knowledge in the field. For this purpose, a multidisciplinary
group has started an ontology-based system named the Amine System Project (ASP) for which
amine-related information is the validation bench.
Results: In this paper, we describe the Ontology-Based Mediator developed in the Amine System
Project (http://asp.uma.es) using the infrastructure of Semantic Directories, and how this system
has been used to solve a case related to amine metabolism-related protein structures.
Conclusions: This infrastructure is used to publish and manage not only ontologies and their
relationships, but also metadata relating to the resources committed with the ontologies. The
system developed is available at http://asp.uma.es/WebMediator.
Background
Over the last few years the internet has become a large
information repository that is accessed manually in the
vast majority of cases. However, the flexibility and open-
ness of the internet in computer systems is lacking when
software applications are connected. The main aim of the
Semantic Web is to automatically perform tasks done in
the current Contents Web. This will be done by making
explicit the semantics of the contents, thereby providing
unambiguous knowledge to Web documents and applica-
tions. For any field of knowledge, and particularly in Life
Sciences, research on Semantic Web infrastructures and
applications can be especially helpful to improve effi-
ciency in the finding, collection and organization of data
stored in the growing number of resources which make
their semantics explicit.
from Seventh International Workshop on Network Tools and Applications in Biology (NETTAB 2007)
Pisa, Italy. 12-15 June 2007
Published: 25 April 2008
BMC Bioinformatics 2008, 9(Suppl 4):S5 doi:10.1186/1471-2105-9-S4-S5
<supplement> <title> <p>A Semantic Web for Bioinformatics: Goals, Tools, Systems, Applications</p> </title> <editor>Paolo Romano, Michael Schroeder, Nicola Cannata and Roberto Marangoni</editor> <note>Research</note> </supplement>
This article is available from: http://www.biomedcentral.com/1471-2105/9/S4/S3
2008 Navas-Delgado et al.; licensee BioMed Central Ltd.
This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0),
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Ral Montaez Martnez
64
2.0
28.0
BMC Bioinformatics 2008, !(Suppl 4):S5 http://www.biomedcentral.com/1471-2105/9/S4/S3
Page 2 of 14
(page number not for citation purposes)
In the context of Life Sciences, the frame of Systems Biol-
ogy is being merged [1]. It is supported by all high-
throughput methods which generate large amounts of
data that cannot be covered simply by the human mind.
This field includes a wide variety of concepts and methods
but, in general, it can be considered the analysis of living
systems, through the study of the relationships among the
elements in response to genetic or environmental pertur-
bations, with the ultimate goal of understanding the sys-
tem as a whole. A system can be considered at different
levels, from a metabolic pathway or gene regulatory net-
work to a cell, tissue, organism or ecosystem. The number
of information repositories and services for biological ele-
ments (molecules, cells, etc) is growing exponentially.
Consequently, Systems Biology is the prototype of a
knowledge-intensive application domain for which the
Semantic Web should be particularly interesting.
Initially, our system will comprise the biochemistry,
molecular biology and physiopathology related to
amines. We are using this system to develop, validate and
apply our infrastructure and are focusing at this stage
mainly on amines derived from cationic amino acids.
They are histamine and polyamines, which have been the
main area of research in our laboratory for the last 15
years [2-6]. Their biosynthetic pathways start with the
alfa-decarboxylation of their respective amino acid pre-
cursors by enzymes that cannot be easily purified from
their native sources.
These metabolic pathways also involve different proteins
with transferase, oxidase and dehydrogenase activities
that are not fully-characterized yet [7]. However, these
pathways are considered well-defined modules of second-
ary nitrogen metabolism, with minimum input and out-
put from/to other biochemical modules, where most of
their components have at least been defined. Concerning
their physiological roles, these compounds play pleio-
tropic roles in human and animal physiology, being
involved in many different physiopathological condi-
tions. Histamine has been related to allergies and inflam-
mation, gastric acid secretion, neurotransmission and
tumour progression [8]. Polyamines are essential for cell
growth and their levels are closely linked to the survival of
every living cell. Thus, their metabolism is a promising
target for anti-proliferative strategies of chemoprevention
and the treatment of cancer and parasitic infections. In
addition, they also act as differentiation and neurotrans-
mission regulators [9]. From this, we can deduce that the
choice of these biomodules as pilots for our integration
project has the following advantages:
The physiopathological problems (cancer, allergic and
inflammatory processes, as well as many other emerging
and rare diseases) associated with these compounds and
their metabolic pathways are global and affect most of
humanity at some stage of their lives. These circumstances
explain the growing interest in the information obtained
from this project.
The information on amine-related processes is dispersed
among specific bibliographies of many different scientific
areas: Oncology, Immunology and Haematology, Neuro-
biology, Pharmacology and Basic Biophysics, Biochemical
and Molecular Biology Research, which makes manual
integration of all the valuable data more difficult. Thus,
this project can contribute to the efficiency of the scientific
advances in the field.
Many molecular questions still remain to be solved with
respect to the structure/function relationships of their
components, the extracellular and intracellular communi-
cation pathways involved in regulating these metabolic
pathways and the physiological effects of these amines.
Therefore, the integration of information (as a part of Sys-
tems Biology techniques) can provide the best perspective
for analyzing these complex molecular relationships and
networks established in living systems [1].
Many of the ontologies and tools developed throughout
this pilot project could be easily extended to solve other
biological problems, as deduced from the most recent
bibliography [10-13].
Finally, this interdisciplinary group combines experi-
mental and bioinformatics approaches for studies in this
field. Thus, any result or prediction can be easily checked
and even experimentally validated [2-4].
We propose a generic infrastructure for publishing and
managing knowledge and information on the Semantic
Web. This infrastructure is based on a resource directory,
called Semantic Directory, containing information about
web resource semantics. This paper focuses on the resolu-
tion of data integration problems by concentrating on our
proposal of a generic infrastructure architecture. We have
developed an Ontology-Based Mediator, which has been
applied to solve a data integration problem in this biolog-
ical domain (Amine System Project, http://asp.uma.es).
The main advantage of this mediator is that data sources
and services can be easily plugged into the system (by
describing the semantics with respect to registered ontol-
ogies). Furthermore, the semantic description is a simple
process, thanks to the use of the proposed infrastructure
(see Section Ontology-Based Mediator: An application
example). This mediator is currently available through a
use case published as a Web site [14].
65
Captulo 2
2.0
28.0
BMC Bioinformatics 2008, !(Suppl 4):S5 http://www.biomedcentral.com/1471-2105/9/S4/S3
Page 3 of 14
(page number not for citation purposes)
Related works
This section describes some related works which are
grouped depending on their goals. Thus, we first describe
data integration systems (from traditional to ontology-
based ones, ending with those focused on bioinformatics
resources). The second group is composed of systems that
have been developed to solve biological problems, and
which provide solutions from different points of view:
Web Services, Workflows, and Web Portals.
Our system is a data integration proposal with a test Web
interface, which uses ontologies to describe resource
semantics and these resources are published as Web Serv-
ices. The workflow described in the use case has been
hand coded (but each step of the workflow is a query that
is automatically solved), and in the future, it will be inter-
esting to include workflow management capabilities.
Our main goal is to provide an infrastructure for interop-
erating applications in the Semantic Web. This infrastruc-
ture can be used to build different kinds of applications,
so we have developed as a use case the Ontology-Based
Mediation system, a system for locating Semantic Web
Services [15] and an ontology clustering algorithm [16].
In this paper, we focus on data integration solutions.
Thus, we have proposed a mediation architecture based
on the wrapper-mediator approach, which has some
interesting characteristics:
It follows an ontology-based approach, in which the
AMMO ontology is used as integration schema, but we
have also used this ontology to perform basic reasoning
processes (class-subclass inference).
Wrappers are published as Web Services to enable their
distribution and use in different applications.
The mediator is divided into components to enable its
extension by including new components (such as query
optimization algorithms).
Information about relationships between resources and
the AMMO ontology is distributed using a generic infra-
structure for Semantic Web applications.
These characteristics have been proposed in previous
works, and have been successfully combined to build a
useful application in Systems Biology for solving real sci-
entific problems. The use of this application by ASP mem-
bers has provided them with an easy-to-use way of solving
daily tasks by integrating different data sources. Further-
more, the system provides an intuitive interface, in which
the user can click on enzymes in a specific metabolic path-
way. This approach can be extended to other metabolic
pathways, and can include specialized data sources as
KEGG, Reactome, etc. (in which we are currently work-
ing).
Data integration approaches
Data integration systems are formally defined as a triple
<G,S,M> where G is the global (or mediated) schema, S is
the heterogeneous set of source schemas, and M is the
mapping that maps queries between the source and the
global schemas. Both G and S are expressed in languages
over alphabets comprised of symbols for each of their
respective relationships. The mapping M consists of asser-
tions between queries over G and queries over S. When
users send queries to the data integration system, they
describe those queries over G and the mapping then
asserts connections between the elements in the global
schema and the source schemas.
The most important proposal to solve the data integration
problem is the wrapper/mediator architecture. In this
architecture, a mediator (an intermediate virtual database
with a schema G according to a previous definition of the
data integration system) is established between data
sources (with a set of schemas S) and applications. A
wrapper is a data source interface that translates data into
a common data model used by the mediator. The user
accesses the data sources through one or several mediator
systems which present high-level abstractions (views) of
combinations of source data. The user does not know
where the data comes from but is able to retrieve it using
a common mediator query language.
Mediator-based integration has query translation as its
main task. A mediator in our context is an application that
has to solve queries formulated by the user at runtime in
terms of either a single or an integrated schema. These
queries are re-written in terms of the data source schemas
in order to delegate the query resolution to the data
sources. Thus, expressing the relationships between the
integrated schema and the data source schemas is a crucial
step in the mediation system development. The two main
approaches for determining the relationships (mapping)
between the integration schema (G) and the data source
schemas (S) are: global-as-view (GAV) and local-as-view
(LAV)[17].
In GAV, each element in the integration schema should be
described in terms of a view (a query) over the data
sources. In other words, the mapping makes it explicit
how to retrieve data from several elements in the integra-
tion schema. This approach is effective when the set of
data sources is stable (i.e., it remains relatively
unchanged).
It is noteworthy that elements in the integration schema
are defined in terms of the data sources, so the addition of
Ral Montaez Martnez
66
2.0
28.0
BMC Bioinformatics 2008, !(Suppl 4):S5 http://www.biomedcentral.com/1471-2105/9/S4/S3
Page 4 of 14
(page number not for citation purposes)
a new data source implies the redefinition of some ele-
ments in the integration schema. This approach benefits
from easier rewriting methods.
In LAV, each element in the data source schemas should
be described in terms of the integration schema. This kind
of approach is effective when the integration schema is
stable and well established in the domain/application. In
this case, the extension of the system is easy because it
only implies adding the description of the new data
source in terms of the integration schema. This approach
implies a more difficult query reformulation and evalua-
tion, which contrasts with the benefits of greater scalabil-
ity.
Data integration systems
Data integration systems [18-22] deal with problems that
could be solved with the infrastructure presented in this
paper. Thus, we propose to relate the semantics (domain
ontologies), with the resources' data schemas using map-
pings, as is done in mediation applications. Our proposal
uses this information to solve a wide range of problems,
in which mediation is a sub-range. Consequently, it is fea-
sible to develop new information integration applica-
tions, by adding new components able to solve specific
tasks.
The wrapper-mediator approach provides an interface to a
group of (semi) structured data sources, combining their
local schemas into a global one and integrating the infor-
mation of local sources. Therefore, the views of the data
that mediators offer are coherent. These mediators per-
form semantic reconciliation of the common data model
representations provided by the wrappers. Some good
examples of wrapper-mediator systems are TSIMMIS [23]
and Manifold [24]. Several improvements have been
made on traditional mediators. One of the most impor-
tant is the use of standard representation languages, like
XML. Thus, the MIX [25] (the successor to the TSIMMIS
project) and MOCHA [26] projects are XML-based. How-
ever, these kinds of systems are usually built as monolithic
systems in which reuse is not possible. Besides, the meta-
data used to integrate the data sources is not made
explicit, so the relationships between resources and inte-
gration schemas are not public. This implies that the prov-
enance of the integrated data is unavailable to the users.
Our proposal allows this knowledge to be available. The
publication of the different components as Web Services
therefore makes it possible to reuse them.
The next level of abstraction on Web integration corre-
sponds to ontology-based systems. Their main advantage
over mediators is their capacity to manage schemas that
are unknown a priori. This is achieved by means of a
mechanism that allows contents and query capabilities of
the data source to be described declaratively. OBSERVER
[27] uses different ontologies to represent data source
information. Users explicitly select the ontology to be
used for query evaluation. The existence of mappings
between ontologies allows the user to change the ontol-
ogy initially selected. The main disadvantage is that the
wrappers are developed for a specific mediator, so they
cannot be reused in other mediators. Model-Based Medi-
ation [28] is a paradigm for data integration in which data
sources can be integrated, using auxiliary expert knowl-
edge. This knowledge includes information about the
domain and is the glue that joins data source schemas
together. The expert knowledge is captured in a data struc-
ture called Knowledge Map. In Model-Based Mediation,
the mediation architecture is extended, taking data
sources from the data level without semantics to the con-
ceptual model level. This architecture introduces seman-
tics into data sources and mediators, but it is not
published nor is it accessible to agents or applications.
Mediators are monolithic systems and they are strongly
coupled to wrappers, limiting dynamic integration and
interoperability.
In the specific field of biological data there are the follow-
ing examples: TAMBIS [29], BioDataServer [30], KIND
[31], BioZoom [32], BioKleisli [33], DiscoveryLink [34],
BioBroker [35] and BioMoby [36].
Web Services based systems
BioMoby [36] is a project with the goal of producing an
open-source, simple, extensible platform to enable the
discovery, representation, integration, and retrieval of
biological data from widely disparate data hosts and anal-
ysis services. In this platform, data and data analysis tools
(for analyzing or transforming data) are distributed in
Web Services. Resources are registered in a central server
called MOBY central. BioMOBY objects are lightweight
XML coded data used as query input and output values.
In summary, the primary components of this infrastruc-
ture are MOBY Services (bioinformatics software tools),
MOBY Objects (input and output data for the services)
and MOBY Central (a register of all resources). This sys-
tem also offers Object and Service hierarchies in order to
classify information and services, helping users to under-
stand the meaning of data required by services. This pro-
posal also includes how to easily develop web services in
order to access biological data. Although it introduces the
use of web services, it is not exactly an integration archi-
tecture, because it is not possible to solve problems
directly, but requires access to different databases or serv-
ices. Furthermore, this proposal does not provide a work-
flow definition and execution system, so several proposals
are being developed in order to define and execute work-
67
Captulo 2
2.0
28.0
BMC Bioinformatics 2008, !(Suppl 4):S5 http://www.biomedcentral.com/1471-2105/9/S4/S3
Page 5 of 14
(page number not for citation purposes)
flows based on BioMOBY services (as described in the
next sub-section, Workflow management systems).
Semantic MOBY [37] is an extension created to solve the
problems discovered in MOBY. This proposal defines four
roles for software agents: Service Providers, Ontology Pro-
viders, Discovery Servers and Service Consumers. The
architecture is based on an ontology that describes the
relationships between these elements, and allows users to
annotate services with different ontologies. The main idea
in this proposal is to build RDF graphs with the service
descriptions and to locate services using graph patterns.
Thus, the main functionality it proposes is to match RDF
graphs with user queries represented as graph patterns.
The National Institute for Bioinformatics (INB) in Spain
has addressed the development of a Web client for locat-
ing and executing BioMOBY services [38]. The description
of biological input/output objects is coordinated and
standardized by means of a data type taxonomy in such a
way that services can communicate with each other, wir-
ing natural bioinformatics workflows. Automatic inter-
faces and help system builders have been incorporated
into the architecture to make it more cohesive and to facil-
itate user communication. Beyond traditional bioinfor-
matics platforms, data persistence systems, user
management and scheduling abilities have produced a
new generation of bioinformatics platforms.
Workflow management systems
The Taverna project [39] has developed a tool for the com-
position and execution of bioinformatics workflows. This
tool includes a graphical interface for the creation and exe-
cution of workflows, which are described using a language
called the Simple conceptual unified flow language
(Scufl), where each step within a workflow represents one
atomic task. This platform has been adapted in order to
access BioMOBY services.
Remora [40] was designed to create and launch BioMoby
workflows. The interface was simplified using the stand-
ard scheme of the traffic lights colour code (red, green and
amber). It can only use its own workflow created by its
own web interface. BIOWep [41] allows users to execute
predefined workflows. It supports workflow annotation
by using a simple ontology for bioinformatics processors
(domain, task, i/o, etc.) and implements the search and
selection of workflows on the basis of their annotation. It
also supports retrieval of workflows on the basis of users'
profiling. Biowep provides a semantic workflow reposi-
tory. The user can search this repository using a graphical
interface defining complex queries to find the desired
workflow. Nowadays, this is the only portal/WMS that
uses semantic annotation in the workflow systems.
GPIPE [42] offers a set of syntactic and algebraic operators
which are able to represent analytical workflows in bioin-
formatics. Iteration, recursion, the use of conditional
statements, and management of suspend/resume tasks
have traditionally been implemented on an ad hoc and
hard-coded basis. GPIPE is a prototype graphic pipeline
generator for PISE that allows the definition of a pipeline,
parameterization of its component methods, and storage
of metadata in XML formats. GPIPE has been imple-
mented and tested on the EMBOSS package. In order to
employ other algorithms, it is necessary to describe the
command-line user interface by means of XML. Thus, this
proposal is only applicable to command-line applica-
tions, which should be in the local machine in which the
workflow is going to be built and executed. It is not possi-
ble to use external applications published as web pages or
web services.
BioWBI and WEE [43] have been designed to assist
researchers in defining their data sources, drawing graph-
ically and executing analysis workflows. These tools con-
stitute the basic components of a much more general
bioinformatics e-workplace, available via a web-browser
that provides a collaborative space which is able to sup-
port both their analysis activities concerning the data
management and the design and execution of their analy-
sis processes. This proposal includes an XML description
of algorithms, so their parameters are represented as data
types. Thus, the only knowledge that a user needs to know
in order to connect two applications is the parameter
types, and it is not possible to check the consistency of the
built workflow. This information would not be enough to
build useful workflows, because parameters of the same
type (strings for example) may have different semantics.
Results
Generic infrastructure
This section presents the generic infrastructure used as the
core for the resolution of data integration problems in
Systems Biology. This infrastructure is based on a resource
directory, called Semantic Directory (Figure 1). We define
the Semantic Directory as a server to register semantics
about available web resources, one or more registered
ontologies, mappings between resources and these ontol-
ogies, and it provides services to browse all the registered
semantics.
The internal elements of the Semantic Directory are
described by means of metadata. In order to deal with this
metadata, the Semantic Directory is composed of two
inter-related ontologies (OMV [44] and SDMO), which
describe the internal semantics of the Semantic Directory
(see Figure 1). This metadata can be managed by tools
that can range from a simple OWL parser to a complex
ontology reasoner. Nowadays, most of the ontologies are
Ral Montaez Martnez
68
2.0
28.0
BMC Bioinformatics 2008, !(Suppl 4):S5 http://www.biomedcentral.com/1471-2105/9/S4/S3
Page 6 of 14
(page number not for citation purposes)
published without additional information such as who
the owner is. This problem makes it difficult to identify, or
to use and reuse published ontologies. For this reason we
use OMV to register additional information about ontol-
ogies to help users locate and use them.
SDMO is the ontology in charge of registering informa-
tion about resources and relationships between these
resources and ontologies registered in the Semantic Direc-
tory. SDMO and OMV are related by a class included in
SDMO, which provides a way of relating resources
(SDMO instances) with registered ontologies (OMV
instances). The current version of SDMO is composed of
five classes:
OMV: this class is used to link resources with registered
ontologies (as instances of the OMV ontology).
Resource: this class is used to store information (query
capabilities, schema, query interface, name and URI)
about resources.
Mapping: this class is used to set the relationships
between resources and ontologies. Each mapping is
related with a similarity instance that establishes the sim-
ilarity between ontology concepts and resource elements.
Similarity: the similarity class contains three properties
(concept1, concept2 and similarity Value) to establish the
similarity between an ontology concept and a resource
element. The similarity value is a real value between 0 and
1, indicating the probability of two concepts being the
same. When adding manual mapping, the similarity value
is 1, but if we use an automatic matching tool, this value
may be less than 1. This similarity is used in the mediator
to filter those mappings that are not taken into account in
the query rewriting.
User: this class is added in order to deal with users in the
applications.
The Semantic Directory provides three interfaces repre-
senting the main tasks it can perform: (1) Resource Meta-
data Repository; (2) Ontology Metadata Repository; and
(3) Semantic Register. These components have been
described by means of an API. This API has been repre-
sented as three Java interfaces (see Figure 2). This first
implementation uses Jena to register and search for infor-
mation. Registry methods are available to resource own-
ers, and there are three possibilities:
To make explicit the relationships with one of the ontol-
ogies registered in the Semantic Directory. These map-
pings will contain two parts: an expression in terms of the
resource structure and an expression in terms of the ontol-
ogy. The syntax of the mappings depends on the kind of
resources registered and the applications developed over
the semantic directory. Thus, a different kind of applica-
tion needs a different extension of the semantic directory
with a different mapping syntax. The next section presents
an Ontology-Based Mediator which uses a Semantic
Directory, so we will describe the syntax of the mappings
for this case there.
Semantic Directories API Figure 2
Semantic Directories API. Semantic Register is the inter-
face in charge of providing methods for registering resources
and ontologies in the semantic directory. Resource Metadata
Repository provides methods for obtaining information
about registered resources. Ontology Metadata Repository
has several methods for obtaining information about regis-
tered ontologies.
Semantic Directory internal elements Figure 1
Semantic Directory internal elements. Two metadata
ontologies provide the required elements to keep the meta-
data about registered ontologies (OMV: Ontology Metadata
Vocabulary) and data sources (SDMO: Semantic Directory
Metadata Ontology).
69
Captulo 2
2.0
28.0
BMC Bioinformatics 2008, !(Suppl 4):S5 http://www.biomedcentral.com/1471-2105/9/S4/S3
Page 7 of 14
(page number not for citation purposes)
To take advantage of a tool for finding mappings
between resource structures and one of the ontologies reg-
istered in the Semantic Directory. The first implementa-
tion contains a tool (Matching Tool, MaF) which can find
mappings between XML documents and OWL ontologies,
and between pairs of OWL ontologies. However, the
Semantic Directory interface can be implemented using
different tools.
To take advantage of a tool for finding mappings
between resource structures and all the ontologies regis-
tered in the Semantic Directory. The response time in this
case depends on the number of ontologies registered and
the tool used to compare structures.
Our goal is to provide applications which will make the
semantics of the resources explicit through its commit-
ment with an ontology registered in the Semantic Direc-
tory. The applications that can be developed using the
Semantic Directory components depend on the extension
of the infrastructure using new components (built on top
of the Semantic Directory). Thus, semantic aware applica-
tions use the Semantic Directory to find the semantics of
the resources registered in order to access the relevant
information. These resources have to be registered in the
Semantic Directory, but this will not involve making
changes in them.
Ontology-Based Mediator: an application example
As we propose the use of an ontology which is supposed
to formalize a shared and consensus knowledge, the
ontology used to integrate the data will be stable. For this
reason, we have chosen a GAV approach. In GAV, each
element in the data source schemas should be related with
the terms of the integration schema.
In order to benefit from the semantics, we have decided to
develop an ontology-based mediator, which will take
advantage of the generic infrastructure (described in the
previous section) for dealing with semantics. Thus, we
focus on a Mediation architecture that uses resources reg-
istered in a semantic directory. Registering of resources in
the Semantic Directory is a key step towards the develop-
ment of the integration solution, and this task is helped by
ontologies. The architecture of the proposed Ontology-
Based Mediator (Figure 3) is composed of four main com-
ponents:
Controller: the main task of this component is to interact
with the user interface, providing solutions described in
terms of one of the ontologies registered in the semantic
directory.
Query Planner: the task of this component is to find a
query plan (QP) for the user query. The current planner
has been implemented including the most basic reason-
ing mechanisms to take advantage of described semantics
(subsumption and classification). Thus, if a query
includes a concept this query will be expanded to include
the semantic descendants. The mappings are also impor-
tant in this process and they are used to find if the query
pattern matches one or more patterns in the mappings. A
bucket algorithm has been applied in this component, but
we are working on more complex algorithms that could
benefit more from reasoning mechanisms. Thus, more
complex reasoning mechanisms can be included in this
component, but considerable work still remains to be
done to establish the consequences of its application.
Query Solver: this component analyzes the query plan
(QP), and performs the corresponding call to the data
services involved in the sub-queries (SQ
1
, !,SQ
n
) of the
query plan (R
1
, !,R
n
). This component will obtain a set
of XML documents from different data services.
Integrator: Results from data services (R
1
, !, R
n
) are com-
posed by this component, in this way, obtaining the
results of the user query. The current implementation of
this component uses the mappings to translate the XML
document to ontology instances, and then a conjunctive
query evaluator is applied to the set of instances found.
Future versions can include a reasoner, but this will imply
taking care of the formal consequences of retrieving cer-
tain instances.
Ontology-Base Mediator Architecture Figure 3
Ontology-Base Mediator Architecture. The mediator is
composed of: a Controller (which controls the data flow in
the system; a Query Planner in charge of finding a way of
solving user queries using the Semantic Directory; a Query
Plan Solver that executes the query plan making the corre-
sponding calls to the data services; and an Integrator which
integrates all the data retrieved from the data services to
provide the user with the results of his/her query.
Ral Montaez Martnez
70
2.0
28.0
BMC Bioinformatics 2008, !(Suppl 4):S5 http://www.biomedcentral.com/1471-2105/9/S4/S3
Page 8 of 14
(page number not for citation purposes)
In our proposal (Figure 3), the sources are made available
by publishing them as Web Services (named Data Serv-
ices). Our primary goal here is to integrate databases
accessible via internet pages. In this context, wrappers are
an important part of the internal elements of data services.
A wrapper is an interface to a data source that translates
data into the common data model used by the mediator.
In our case, we have chosen XML as the common data
model. The development of Data Services that require the
development of a wrapper has been studied in previous
work [45]. However, biological data sources are usually
public and downloadable. In these cases we have
designed some patterns to retrieve a data source stored as
a flat file to store it in an XML database. In summary, data
services, independently of the development process, are
distributed software applications that receive queries in
XQuery and return XML documents.
In the context of mediator development, the process of
registering resources in a semantic directory implies find-
ing a set of mappings between one or several ontologies
and the data service schema (usually expressed as an
XMLSchema document). These mappings will be the key
elements to integrate all the data sources, and these map-
ping will be the way in which the resource semantics are
made explicit. The mappings used are defined as a pair
(P,Q). P is a set of path expressions on the resource
schema, and Q a query expression in terms of the ontol-
ogy. In a first approach we have chosen XPath as the lan-
guage to express P, and conjunctive queries to Q. For
example to set the mappings between the Swiss-Prot data
service and the ontology we have established the follow-
ing mappings (registered in the Semantic Directory simply
calling the register Resource method):
! /Result/polypeptides/polypeptides_name
" Polypeptides(P)
! /Result/polypeptides/organism_name
" Organism(O)
! /Result/polypeptides/amino_acid_sequence
" Amino_acids_Sequence(A)
! /Result/polypeptides/polypeptides_name AND /Result/
polypeptides/id
" Polypeptides(P) AND SWISSPROT_id(P,i)
! /Result/polypeptides/polypeptides_name AND /Result/
polypeptides/molecular_weight
" Polypeptides(P) AND
molecular_weight(P,m)
! /Result/polypeptides/polypeptides_name AND /Result/
polypeptides/PDB
" Polypeptides(P) AND PDB_id(P,i)
! /Result/polypeptides/polypeptides_name AND /Result/
polypeptides/synonym_polypeptides_name
" Polypeptides(P) AND
name_of_entities(P,n)
! /Result/polypeptides/organism_name AND /Result/
polypeptides/organism_name
" Organism(O) AND Organism_name(O,n)
! /Result/polypeptides/organism_name AND /Result/
polypeptides/Taxon
" Organism(O) AND Taxon_id(O,i)
! /Result/polypeptides/amino_acid_sequence AND /
Result/polypeptides/ amino_acid_sequence
" Amino_acids_Sequence(A) AND
sequence(A,s)
! /Result/polypeptides/amino_acid_sequence AND /
Result/polypeptides/ length_sequence
" Amino_acids_Sequence(A) AND
length(A,l)
! /Result/polypeptides/polypeptides_name AND /Result/
polypeptides/ organism_name
" belong_to(P,O)
! /Result/polypeptides/polypeptides_name AND /Result/
polypeptides/ amino_acid_sequence
" amino_acids_sequence(P,A)
The Amine System Project: integration of data on
biological amine-related information
In this pilot project (ASP Model Finder[14]) we have devel-
oped and tested the Ontology-Based Mediator described
in the previous section. The long term goal of this project
is to register the most relevant public databases at differ-
ent levels of study: metabolite properties and concentra-
tions, macromolecular structures, assigned functions,
docking among biomolecules and information on bio-
71
Captulo 2
2.0
28.0
BMC Bioinformatics 2008, !(Suppl 4):S5 http://www.biomedcentral.com/1471-2105/9/S4/S3
Page 9 of 14
(page number not for citation purposes)
chemical pathways. This proposal is flexible as has been
shown in the previous section, but in order to validate its
viability for solving specific and real issues, we have tack-
led the resolution of a well known bioinformatics prob-
lem by integrating a limited but increasingly growing
number of databases. The initial problem to be solved is
summarized as following, and the use case developed to
solve it is named AMMO-Prot:
Problem: A common and useful strategy to determine
the 3D structure of a protein, which cannot be
obtained by its crystallization, is to apply comparative
modelling techniques. These techniques start working
with the primary sequence of the target protein to
finally predict its 3D structure by comparing the target
polypeptide to those of solved homologous proteins
[46].
In order to solve this problem, protein structure data and
some tools to compare them are required from databases.
These databases have been used to validate our integra-
tion tool. First of all, we need to define the domain ontol-
ogy in order to relate it to the resource semantics. There
are several ontologies related to this domain, the most
representative being GO (http://www.geneontology.org)
and the molecular biology ontology TAMBIS (http://
www.daml.org/ontologies/99). However, these ontolo-
gies are very large and describe very light weight semantics
(only describing concept hierarchy). Thus, we have devel-
oped our own ontology (The AMine Metabolism Ontol-
ogy, AMMO) which is based on the Gene Ontology
concepts but with enriched relationships which improve
its semantics (see Figure 4). The definition of the relation-
ships between concepts will allow us to retrieve interre-
lated information (for example polypeptides and the
organisms in which they can be found). In addition,
future versions of the mediator will benefit from the
improved semantics to infer new knowledge.
Being based on the GO ontology guarantees interopera-
bility with other applications also based on this ontology.
In Figure 4, we have shown only the relevant concepts for
our domain helping users to understand the domain
semantics. The AMMO is used as the pivot that will inte-
grate the whole domain, which includes concepts/rela-
tionships necessary for the use case described above and
other ongoing projects. This ontology has been described
with OWL.
At present, the AMMO ontology has been used to register
the semantics of different consolidated resources/tools in
bioinformatics: SWISS-Prot [47], PDB [48], Modeller [49]
and JMol (http://www.jmol.org). They allow us to retrieve
information about protein structures which is used at an
initial analysis stage. This information will subsequently
be the basis of future developments to extend the possibil-
ities of our system, and to allow users to retrieve informa-
tion on many other queries on metabolic, signalling and
molecular interactions and relationships i.e., to further
progress in automatic data integration resources on a
given biochemical problem. In order to achieve this goal
new databases (PubChem, Kegg, Brenda and Prosite) are
being wrapped.
In its present stage, the developed Web tool has as its goal
to show the viability of the proposal and its application in
the real case mentioned above (location/prediction of an
amine-related protein structure), so we have divided the
problem into a set of simple steps. Initially, the species
must be selected by filling its (common or scientific)
names in the organism field (Figure 5). On the other
hand, the Web interface shows a pathway that is used as
the entry point (Figure 5) to retrieve structural informa-
tion on the target by clicking on the protein picture. The
process for information searching and integration is illus-
trated by the two examples of human polypeptides
explained below.
AMMO Ontology Figure 4
AMMO Ontology. This ontology represents the key con-
cepts in the use case we present below and other on-going
use cases. Nodes are concepts, unlabeled arrows are is_a
relationships (concept hierarchy), and labelled arrows are the
properties defined to relate concepts. The main concepts in
the example presented in this paper are Organism, Polypep-
tides, Homologue_Sequence, Amino_acids_Sequence and
Molecular_3D_Structure.
Ral Montaez Martnez
72
2.0
28.0
BMC Bioinformatics 2008, !(Suppl 4):S5 http://www.biomedcentral.com/1471-2105/9/S4/S3
Page 10 of 14
(page number not for citation purposes)
In the first example, we click on the enzyme SSAT (Sper-
midine/Spermine N-acetyltransferase)(synonymous of
diamine acetyltransferase). Thus, the first query to be
solved is:
P ! Polypeptide(P) and name(P, Spermidine/Spermine N-
acetyltransferase) and Organism(O) and name(O,
Human) and belongs_to(P,O).
The query built inside the Web application is decomposed
into two sub-queries to be sent to the Swiss-Prot and PDB
databases, which are the databases related to the polypep-
tide concept. First the sub-query will return data as an
instance of the Polypeptide concept. Thus, the query
results comprise information available in the Swiss-Prot
database on two isoforms for the human SSAT protein.
Then, the graphical Web interface will allow the users to
manually select from information (including the primary
sequence) for one or another isoform. In this specific case,
both structures have been previously determined by
experimental methods, so that the application will make
use of the second sub-query to retrieve information from
the PDB database, which could be downloaded or dis-
played using the JMol tool (http://www.jmol.org). When
launching a new query about an unsolved structure (or
one not annotated in PDB), as is the case for TGase (Pro-
tein-glutamine gamma-glutamyltransferase), an initial
search for information about this enzyme would start in
the SwissProt database as in the previous example. Results
will show all the entries related with our query, which cor-
respond with different tissue-specific TGase types. In this
task, we are going to focus on TGase Z, which is widely
expressed in humans. As there is no entry in PDB for this
protein, the next step involves executing the Modeller tool
[49] to predict a 3D structure of the polypeptide. Modeller
is an automated homology modelling tool, which per-
forms the necessary steps to carry out: searching for
homologous proteins, target-template sequence align-
ments, model building on the bases of the template coor-
dinates and basic geometry optimization. In our example
(TGase Z) the user query for this process is:
Q(HS): Polypeptide (P1) and Amino_acid_sequence(AA) and
sequence (AA, MDQVATLRLESV") and
amino_acid_sequence (P1,AA) and Polypeptide (P2) and
Homologue_Sequence (HS) and
homologue_sequence_aas(HS,AA) and
homologue_sequence_polypeptide(HS,P2).
Then, using the primary sequence retrieved from Swiss-
Prot, a search on PDB is performed to obtain a set of
homologous proteins of our target, corresponding to
AMMO-Prot interface Figure 5
AMMO-Prot interface. The pathway is used as the entry point in the developed use case (http://asp.uma.es/WebMediator).
Each protein/enzyme is a link to a workflow that will capture the available structural information on the polypeptide in a given
species.
73
Captulo 2
2.0
28.0
BMC Bioinformatics 2008, !(Suppl 4):S5 http://www.biomedcentral.com/1471-2105/9/S4/S3
Page 11 of 14
(page number not for citation purposes)
structures that have been previously determined experi-
mentally by using mainly X-ray diffraction or NMR tech-
niques. We obtain a set of template candidates that are
filtered by the sequence identity shared with the target
sequence. In a homology modelling process the quality of
the final model is highly dependent on the identity
between target and template sequences, so those tem-
plates sharing an identity below 30% are going to be dis-
carded by the application [46]. In this first version, the
template showing the highest identity with the target pro-
tein will be selected for the next steps. Once the most suit-
able template is chosen, further steps of alignment and
modelling are performed by the automated tool. Tem-
plate and target sequences are aligned before the model-
ling process, where spatial coordinates of the template are
extrapolated to build the target structure.
Finally, the results of this process are five different models
of the same target protein, which can be selected in the
Web interface. Subsequently, the model can be either
downloaded or viewed using JMol, in order to analyze
whether the solution is realistic or not (Figure 6). This vis-
ualization tool is included as part of the application, pro-
viding added-value features.
3D model predicted by AMMO-Prot for human TGase Z Figure 6
3D model predicted by AMMO-Prot for human TGase Z. Figure created with JMol [http://www.jmol.org].
Ral Montaez Martnez
74
2.0
28.0
BMC Bioinformatics 2008, !(Suppl 4):S5 http://www.biomedcentral.com/1471-2105/9/S4/S3
Page 12 of 14
(page number not for citation purposes)
Discussion and conclusions
This paper presents an Ontology-Based Mediator that uses
an infrastructure for building applications in the Semantic
Web. It is being validated in a specific biological context,
that of amine-related biochemistry, and consequently it
has been named the Amine System Project (http://
asp.uma.es). Of course, this Semantic Mediator can be
adapted very easily to obtain structural information for
many other biochemical problems as can be deduced
form the AMMO ontology organization (Figure 4). The
proposed ontology is an extension of the Gene Ontology,
so the developed system can interoperate with other sys-
tems using this ontology (using shared vocabulary). If the
AMMO ontology is modified to add new concepts, exist-
ing mappings will still be valid, ensuring system scalabil-
ity. However, if a concept is changed, related mappings
should be redefined. In addition, the developed proposal
is a generic infrastructure and mediator, so the ontology
used can be changed to be used in other domains (adding
new mappings between resources and registered ontolo-
gies). In the development of the mediator and in previous
works [34][37][45], we have discovered some limitations.
The main one is the maintenance of data services, because
the services developed use public databases that are not
under our control. Thus, the long term success of this pro-
posal and similar ones relies on the collaboration of data
and tool owners. For this reason, the data services inte-
grated into this proposal have been developed from stable
data sources, providing their data as files, data services or
databases (system that are not affected by aspect changes,
as in web interfaces). In addition, the developed data serv-
ices include the use of internal cache systems to prevent
source unavailability.
The proposed solution is based on the registration of the
resources' semantics by relating them with ontologies.
However, the location of relevant ontologies in a specific
domain is an open problem which is being dealt with by
relating ontologies (ontology alignment) and organizing
them. In this way, we are studying how to extend our pro-
posal to include mechanisms for organizing ontologies in
order to facilitate their location.
Protein structures contain fundamental information
regarding their function, location and interactions, which
is most of the information in their biological missions.
Our use case (AMMO-Prot) returns the correct informa-
tion for the protein structures included initially in the
pilot project. The pilot could be easily adapted to any
other metabolic pathway. In addition, queries can be
defined in a user friendly way for biochemists, that is, by
clicking on the protein symbols organized as a metabolic
pathway scheme after definition of the required species.
Combining information integration with prediction tech-
niques results in efficient information retrieval and
expands the spectrum of applicability of structural bioin-
formatics techniques to non-experienced users. Therefore,
the problem presented in this paper is important in this
context, as automatic performance of the process will
reduce the effort required to solve questions on protein
structures. All of these characteristics will expand the spec-
trum of applicability of structural bioinformatics tech-
niques to non-experienced users. Genomic Projects have
exponentially increased the number of known polypep-
tide sequences. Thus, any effort to improve efficiency for
the extraction of structural information at its highest level
should help advance many on-going Systems Biology
projects. Our project fulfils all of these aims and objec-
tives.
List of abbreviations used
ASP: Amine System Project
AMMO: the Amine Metabolism Ontology
KEGG: Kyoto Encyclopedia of Genes and Genomes
LAV: Local As View
GAV: Global As View
TSIMMIS: The StanfordIBM Manager of Multiple Informa-
tion Sources
XML: eXtensible Markup Language
MOCHA: Middleware Based On a Code SHipping Archi-
tecture
TAMBIS: Transparent Access to Multiple Bioinformatics
Information Sources
RDF: Resource Description Framework
INB: National Institute for Bioinformatics
BIOWeP: Workflow Enactment Portal for Bioinformatics
WMS: Workflow Management System
PISE: Pasteur Institute Software Environment
EMBOSS: European Molecular Biology Open Software
Suite
BioWBI: Bioinformatic Workflow Builder Interface
WEE: Workflow Execution Engine
75
Captulo 2
2.0
28.0
BMC Bioinformatics 2008, !(Suppl 4):S5 http://www.biomedcentral.com/1471-2105/9/S4/S3
Page 13 of 14
(page number not for citation purposes)
OMV: Ontology Metadata Vocabulary
SDMO: Semantic Directory Metadata Ontology
OWL: Web Ontology Language
URI: Uniform Resource Identifier
API: Application Programming Interface
MaF: Matching Framework
XQuery: XML Query Language
XPath: XML Path Language
GO: Gene Ontology
PDB: Protein Data Bank
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
IND designed the infrastructure of Semantic Directories,
carried out the implementation of the system and drafted
the manuscript. RM designed the biological problem to be
solved. RM and APA carried out the study of the databases
and analyzed result provided by AMMO-Prot. AMG devel-
oped the study of mechanisms to predict structures. JLU
and FSJ participated in the design of the study, coordi-
nated the test phases and performed analysis of the
results. FSJ and helped to draft the manuscript. JFAM con-
ceived the infrastructure, participated in its design and
coordination and helped to draft the manuscript. All
authors read and approved the final manuscript.
Acknowledgements
Supported by Grants CVI-267 and CVI-657 (Junta de Andaluca), Grants
SAF2005-01812 and TIN2005-09098-C05-01 (Spanish Ministry of Educa-
tion and Science), and Ramn Areces Foundation.
This article has been published as part of BMC Bioinformatics Volume 9 Sup-
plement 4, 2008: A Semantic Web for Bioinformatics: Goals, Tools, Sys-
tems, Applications. The full contents of the supplement are available online
at http://www.biomedcentral.com/1471-2105/9?issue=S4.
References
1. Kitano H: Systems biology: a brief overview. Science 2002,
295(5560):1662-1664.
2. Viguera E, Trelles O, Urdiales JL, Mats JM, Snchez-Jimnez F: Mam-
malian L-amino acid decarboxylases producing 1,4-diamines:
analogies among differences. Trends Biochem. Sci. 1994,
19:318-319.
3. Rodrguez-Caso C, Rodrguez-Agudo D, Moya-Garca A, Fajardo I,
Medina MA, Subramaniam V, Snchez-Jimnez F: Local changes in
the catalytic site of mammalian histidine decarboxylase can
affect its global conformation and stability. Eur. J. Biochem
2003, 270:4376-4387.
4. Rodrguez-Caso C, Montaez R, Cascante M, Snchez-Jimnez F,
Medina MA: Mathematical modeling of polyamine metabo-
lism in mammals. J. Biol. Chem. 2006, 281(31):21799-812.
5. Medina MA, Urdiales JL, Mats JM, Nez de Castro I, Snchez-
Jimnez F: Diamines interfere with the transport of L-orni-
thine in Ehrlich-cell plasma-membrane vesicles. Biochem. J.
1991, 280:825-827.
6. Medina MA, Quesada AR, Nez de Castro I, Snchez-Jimnez F: His-
tamine, polyamines and cancer. Biochem. Pharmacol. 1999,
57:1341-1344.
7. Medina MA, Urdiales JL, Rodrguez-Caso C, Ramrez FJ, Snchez-
Jimnez F: Biogenic amines and polyamines: similar biochem-
istry for different physiological missions and biomedical
applications. Crit. Rev. Biochem. Mol. Biol. 2003, 38:23-59.
8. Moya-Garca AA, Medina MA, Snchez-Jimnez F: Mammalian his-
tidine decarboxylase: from structure to function. BioEssays
2005, 27:57-63.
9. Urdiales JL, Mats JM, Nez de Castro I, Snchez-Jimnez F: Chlo-
rpheniramine inhibits the ornithine decarboxylase induction
of Ehrlich carcinoma growing in vivo. FEBS Lett. 1992,
305:260-264.
10. Dopazo J: Bioinformatics and cancer: an essential alliance. Clin
Transl Oncol. 2006, 8:409-15.
11. Overall CM, Dean RA: Degradomics: systems biology of the
protease web. Pleiotropic roles of MMPs in cancer. Cancer Metastasis
Rev. 2006, 25:69-75.
12. Liu ET, Kuznetsov VA, Miller LD: In the pursuit of complexity:
systems medicine in cancer biology. Cancer Cell. 2006, 9:245-7.
13. Carlberg C, Dunlop TW: An integrated biological approach to
nuclear receptor signaling in physiological control and dis-
ease. Crit Rev Eukaryot Gene Expr. 2006, 16:1-22.
14. AMMO-Prot. . [http://asp.uma.es/WebMediator]
15. Brogi A, Corfini S, Aldana-Montes JF, Navas-Delgado I: Automated
Discovery of Compositions of Services Described with Sepa-
rate Ontologies. In Proceedings of the 4th International Conference on
Service Oriented Computing: 4-7 December 2006; Chicago Edited by:
Georgakopoulos D, Ritter N, Benatallah B, Zirpins C, Feuerlicht G, Schoen-
herr M, Motahari-Nezhad HR. Springer; 2006:509-514.
16. Navas-Delgado I, Sanz I, Aldana-Montes JF, Berlanga Llavori R: Auto-
matic Generation of Semantic Fields for Resource Discovery
in the Semantic Web. In Proceedings of the 16th International Con-
ference on Database and Expert Systems Applications: 22-26 August 2005;
Copenhagen Edited by: Andersen KV, Debenham J, Wagner R. Springer;
2005:706-715.
17. Halevy AY: Answering queries using views: A survey. VLDB Jour-
nal: Very Large Data Bases 2001, 10(4):270-294.
18. Grahne G: Information integration and incomplete informa-
tion. IEEE Data Eng. Bull. 2002, 25(3):46-52.
19. Ullman JD: Information integration using logical views. Theo-
retical Computer Science 2000, 239(2):189-210.
20. Levy AY: Logic-based techniques in data integration. In Pro-
ceedings of the Workshop on Logic-Based Artificial Intelligence: 14-16 June
1999; Washington, DC Edited by: Minker J. Computer Science Depart-
ment, University of Maryland; 1999:575-595.
21. Lenzerini M: Data integration: A theoretical perspective. In
Proceedings of the 21st ACM SIGACT-SIGMOD-SIGART Symposium on
Principles of Database Systems: 3-6 June 2002; Madison Edited by: Popa
L. ACM; 2002:233-246.
22. Bertossi L, Chomicki J: Query answering in inconsistent data-
bases. In Logics for Emerging Applications of Databases Edited by:
Chomicki J, Meyden R, Saake G. Springer-Verlag; 2003:43-84.
23. Chawathe SS, Garcia-Molina H, Widom J: Flexible constraint
management for autonomous distributed databases. Data
Engineering Bulletin 1994, 17(2):23-27.
24. Levy A: The information manifold approach to data integra-
tion. IEEE Intelligent Systems 1998, 13:12-16.
25. Bornhovd C: Mix - a representation model for the integration
of web-based data. In Tech. Rep. Dep. CS, Darmstadt University of
Technology, Germany; 1998.
26. Rodriguez-Martinez M, Roussopoulos N: Mocha: a self-extensible
database middleware system for distributed data sources. In
Proceedings of the ACM SIGMOD International Conference on Manage-
ment of Data: 16-18 May 2000; Dallas Edited by: Chen W, Naughton JF,
Bernstein PA. ACM; 2000:213-224.
27. Mena E, Kashyap V, Illarramendi A, Sheth A: Imprecise answers on
highly open and distributed environments: An approach
Ral Montaez Martnez
76
2.0
28.0
Publish with BioMed Central and every
scientist can read your work free of charge
"BioMed Central will be the most significant development for
disseminating the results of biomedical research in our lifetime."
Sir Paul Nurse, Cancer Research UK
Your research papers will be:
available free of charge to the entire biomedical community
peer reviewed and published immediately upon acceptance
cited in PubMed and archived on PubMed Central
yours you keep the copyright
Submit your manuscript here:
http://www.biomedcentral.com/info/publishing_adv.asp
BioMedcentral
BMC Bioinformatics 2008, !(Suppl 4):S5 http://www.biomedcentral.com/1471-2105/9/S4/S3
Page 14 of 14
(page number not for citation purposes)
based on information loss for multiontology based query
processing. International Journal of Cooperative Information Systems
(IJCIS) 2000:403-425.
28. Ascher L, Gupta B, Martone A: A model-based mediator system
for scientific data management. In Bioinformatics: Managing Scien-
tific Data Edited by: Critchlow T, Lacroix Z. Morgan Kaufmann;
2003:335-369.
29. Stevens R, Baker P, Bechhofer S, Ng G, Jacoby A, Paton NW, Goble
CA, Brass A: TAMBIS: Transparent access to multiple bioin-
formatics information sources. Bioinformatics 2000, 16:184-186.
30. Lange M, Freier A, Scholz U, Stephanik A: A computational sup-
port for access to integrated molecular biology data. In Pro-
ceeding of the German Conference on Bioinformatics: 7-10 October 2001;
Braunschweig Edited by: Wingender E. In Silico Biology; 2001:203-205.
31. Gupta A, Ludscher B, Martone ME: Knowledge-based integra-
tion of neuroscience data sources. In Proceedings of the 12th Inter-
national Conference on Scientific and Statistical Database Management
(SSDBM): 26-28 July 2000; Berlin Edited by: Gnther O, Lenz HJ. IEEE
Computer Society; 2000:39-52.
32. Liu L, Buttler D, Critchlow T, Han W, Paques H, Pu C, Rocco D: Bio-
Zoom: Exploiting source-capability information for inte-
grated access to multiple bioinformatics data sources.
Proceedings of the 3rd IEEE Symposium on BioInformatics and BioEngineer-
ing: 10-12 March 2003; Washington, DC 2003.
33. Davidson S, Overton C, Tannen V, Wong L: BioKleisli: A digital
library for biomedical researchers. International Journal of Digital
Libraries 1997, 1:36-53.
34. IBM Corp. DiscoveryLink . [http://researchweb.watson.ibm.com/
journal/sj/402/haas.html].
35. Aldana JF, Roldn-Castro M, Navas-Delgado I, Roldn-Garca MM,
Hidalgo-Conde M, Trelles O: Bio-Broker: a tool for integration
of biological data sources and data analysis tools. Software:
Practice and Experience 2006, 36(14):1585-1604.
36. Wilkinson MD, Gessler D, Farmer A, Stein L: The Bio-MOBY
Project Explores Open-Source, Simple, Extensible Protocols
for Enabling Biological Database Inter-operability. Proceedings
of the Virtual Conference Genomic and Bioinformatics: 16-19 September
2003 2003:16-26.
37. Schiltz G, Gessler D, Stein L: Semantic MOBY. In Proceedings of the
W3C Workshop on Semantic Web for Life Sciences: 27-28 October 2004
Edited by: W3C. Cambridge; 2004.
38. Navas-Delgado I, Rojano-Muoz MM, Ramrez S, Prez AJ, Andrs
Len E, Aldana-Montes JF, Trelles O: Intelligent client for inte-
grating bioinformatics services. Bioinformatics 2006,
22(1):106-111.
39. Oinn T, Addis M, Ferris J, Marvin D, Senger M, Greenwood M, Carver
T, Glover K, Pocock MR, Wipat A, Li P: Taverna: A tool for the
composition and enactment of bioinformatics workflows.
Bioinformatics 2004, 20(17):3045-3054.
40. Carrere S, Gouzy J: REMORA: a pilot in the ocean of BioMoby
web-services. Bioinformatics 2006, 22(7):900-901.
41. Romano P, Bartocci E, Bertolini G, De Paoli F, Marra D, Mauri G,
Merelli E, Milanesi L: Biowep: a workflow enactment portal for
bioinformatics applications. BMC Bioinformatics 2007, 8(Suppl
1):S19.
42. Garcia Castro A, Thoraval S, Garcia LJ, Ragan MA: Workflows in
bioinformatics: meta-analysis and prototype implementa-
tion of a workflow generator. BMC Bioinformatics 2005, 6:87.
43. Leo P, Marinelli C, Pappad G, Scioscia G, Zanchetta L: BioWBI: an
Integrated Tool for building and executing Bioinformatic
Analysis Workflows. Bioinformatics Italian Society Meeting: 26-27
March 2004; Padova 2004.
44. Hartmann J, Sure Y, Haase P, Palma R, Suarez-Figueroa M: Omv -
ontology metadata vocabulary. In Proceedings of the 4th Interna-
tional Semantic Web Conference: 6-10 November 2005; Galway Edited
by: Gil Y, Motta E, Benjamins VR, Musen MA. Springer; 2005.
45. Navas Delgado I, Roldn-Garca MM, Dianes Mazorra D, Aldana-Mon-
tes JF: Developing Data Services. In Proceegings of the 17th Confer-
ence on Advanced Information Systems Engineering. Data Integration and
the Semantic Web: 13-17 June 2005; Porto Edited by: Castro J, Teniente
E. FEUP Edies; 2005:287-301.
46. Baker D, Sali A: Protein structure prediction and structural
genomics. Science Date 2001, 294(5540):93-6.
47. Boeckmann B, Bairoch A, Apweiler R, Blatter MC, Estreicher A,
Gasteiger E, Martin MJ, Michoud K, O'Donovan C, Phan I, Pilbout S,
Schneider M: The swiss-prot protein knowledgebase and its
supplement trembl in 2003. Nucleic Acids Res 2003,
31(1):365-370.
48. Kellenberger E, Muller P, Schalon C, Bret G, Foata N, Rognan D: sc-
pdb: an annotated database of druggable binding sites from
the protein data bank. J Chem Inf Model 2006, 46(2):717-727.
49. Marti-Renom MA, Stuart A, Fiser A, Snchez R, Melo F, Sali A: Com-
parative protein structure modeling of genes and genomes.
Annu. Rev. Biophys. Biomol. Struct. 2000, 29:291-325.
77
Captulo 2
2.0
28.0
BIOINFORMATICS APPLICATIONS NOTE
Vol. 25 no. 6 2009, pages 834835
doi:10.1093/bioinformatics/btp061
Systems biology
Systems biology metabolic modeling assistant: an
ontology-based tool for the integration of metabolic data in
kinetic modeling
Armando Reyes-Palomares
1,
, Raul Montaez
1,
, Alejando Real-Chicharro
1
,
Othmane Chniber
2
, Amine Kerzazi
2
, Ismael Navas-Delgado
2
, Miguel ngel Medina
1
,
Jos F. Aldana-Montes
2
and Francisca Snchez-Jimnez
1,
1
Department of Molecular Biology and Biochemistry and unit 741 of CIBER de Enfermedades Raras and
2
Department of Computer Languages and Computational Sciences, Campus de Teatinos, University
of Malaga, 29071, Spain
Received and revised on December 27, 2008; accepted on January 24, 2009
Advance Access publication February 2, 2009
Associate Editor: Trey Ideker
ABSTRACT
Summary: We present Systems Biology Metabolic Modeling
Assistant (SBMM Assistant), a tool built using an ontology-based
mediator, and designed to facilitate metabolic modeling through the
integration of data from repositories that contain valuable metabolic
information. This software can be used for the visualization, design
and management of metabolic networks; selection, integration and
storage of metabolic information; and as an assistant for kinetic
modeling.
Availability: SBMM Assistant for academic use is freely available at
http://www.sbmm.uma.es.
Contact: kika@uma.es
1 INTRODUCTION
The ability to perform dynamic analysis of biochemical reactions
networks is essential for understanding the intrinsic complexity
of biological systems. Systems biology platforms now use a
structured standard of biological information (http://www.sbml.org)
and software for the analysis of biological networks, e.g.
Cytoscape (http://www.cytoscape.org), and the dynamic behavior
of biochemical reactions e.g. COPASI (http://www.copasi.org)
and CellDesigner (http://www.celldesigner.org). These advances
correlate with the increased rate of growth in available biological
information and the reorganization of data service architectures.
For metabolic modeling, diverse information is required, from the
characteristics of enzymes, metabolites and modulators to the global
structure and dynamics of networks. One of the acknowledged
barriers to efcient kinetic modeling is the lack of availability
of kinetic parameters. To alleviate this problem, repositories
for kinetic data have been built e.g. BRENDA (http://www.
brenda.uni-koeln.de); and SABIO-RK (http://sabio.villa-bosch.de)

To whom correspondence should be addressed.

The authors wish it to be known that, in their opinion, the rst two authors
should be regarded as joint First Authors.
(Schomburg et al., 2004; Wittig et al., 2006). On the other hand,
semantic web technologies can address the shortcomings in the
workows used for metabolic modeling (Lee et al., 2008; Oinn et
al., 2004).
In the present note, we describe Systems Biology Metabolic
Modeling Assistant (SBMM Assistant), an ontology-based
application designed to facilitate the process of metabolic modeling.
SBMM Assistant is an SBML-compatible (Hucka et al.,2003) and
user-friendly tool that gives both the novice or experienced user
the ability to capture, enrich, generate and visualize biological
networks, to make basic queries about enzymatic kinetics and
regulation, and to annotate this information using MIRIAM (Le
Novre et al. 2005) specications. Furthermore, SBMM Assistant
facilitates friendly cross-talk among different resources and tools.
These features provide SBMM Assistant with capabilities not
present in other, previously reported applications.
2 DESCRIPTION
SBMM Assistant uses an ontology-based mediator developed
to integrate data from KEGG (Kanehisa and Goto 2000),
ChEBI (http://www.ebi.ac.uk/chebi) (Brooksbank et al., 2005),
BRENDA and SABIO-RK. KEGG provides information about
pathways, enzymes, reactions and compounds. SABIO-RK has
data on reactions and enzymes, including kinetic parameters and
equations. BRENDAcontains information about kinetic parameters
and modulators. ChEBI has the general chemical properties of
metabolites and modulators, including molecular composition and
molecular weight. The mediator architecture is composed of three
main components: the Controller (which receives user requests and
coordinates the mediator components), the Query Planner (which
elaborates one or several query plans to retrieve data for the user
from different data sources) and the Evaluator/Integrator (which
analyzes the query plan and performs the necessary calls to the
data services involved in the sub-queries of the query plan). For
further information about the computational work and architecture of
SBMM Assistant, refer to http://sbmm.uma.es/ and previous works
(Navas-Delgado et al., 2008).
834 The Author 2009. Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oxfordjournals.org
Ral Montaez Martnez
78
2.0
28.0
Systems biology metabolic modeling assistant
3 APPLICATIONS
3.1 Metabolic network visualization and management
SBMM Assistant uses an interactive graphic interface to facilitate
queries and capture information. Biochemical reaction networks
can be visualized following the Systems Biology Graphical
Notation (http://www.sbgn.org) with CellDesigner labels. Therefore,
reactions loaded as SBML les, or queried online, can be edited in
a friendly, intuitive and standardized environment.
3.2 Selection, integration and storage of the metabolic
information
The use of a mediator enables the user to only access information
related to the structures and kinetic properties of the reactions being
studied. All of this information is integrated by the mediator and can
be edited by the user, as mentioned in the previous section.
The integrated information is annotated in a SBMLle, according
to MIRIAM (Le Novre et al. 2005). Because SBMMAssistant was
designed to promote cross-talk with other metabolic modeling tools,
every compound is identied by both its KEGG compound number
and its ChEBI identier, and every enzyme/reaction is identied by
its EC code as well as its KEGG reaction number.
3.3 Design of metabolic networks
SBMM Assistant makes it possible to design metabolic networks
using data from a KEGG pathway, a SBML le or a list of EC
numbers. KEGG pathways can be referred to by ID (e.g. hsa00010)
or name (e.g. glycolysis). SBML les can be obtained from
databases such as BIOMODELS (http://www.ebi.ac.uk/biomodels-
main/static-pages.do?page=home) or can be uploaded by the user
(as long as each reaction and compound is identied according
to MIRIAM specications). The application can also generate
metabolic networks by integrating information from a list of EC
numbers. Once the new metabolic structure is built, it can be
manually modied to obtain the nal desired conguration for the
network.
3.4 Assistance with kinetic modeling
After a series of easy steps, a user can construct a formal
implementation of the kinetics of a given reaction using SBMM
Assistant. First, the user must dene the reaction stoichiometry
and the regulatory elements. Next, the kinetic law describing the
reaction can be found by choosing the relevant kinetics equations
from SABIO-RK, or alternatively, dened by user. The following
step allows user to query BRENDAor SABIO-RK to nd the values
of the kinetic parameters of the enzymes involved in the reaction.
It is possible to assign each parameter either as a constant or as a
time-dependent condition, and to describe each parameter as global
or local. The user can also read the abstract of references associated
with each parameter in the repositories to evaluate its accuracy under
the desired simulation conditions.
4 FUTURE AND SCOPE
The aim of SBMM Assistant is to help users overcome the major
problems encountered in metabolic modeling, in a standardized
environment. It is a semantic, web-based tool that will provide
integrated, up-to-date metabolic information that the user can add
to, as well as consult in a friendly and interactive way. Thus,
it is useful for both experienced metabolic modelers and novice
biochemists. This tool could be potentially used to establish the
variation in each kinetic parameter in a reaction, with available
parameter optimization methods. It is fully compatible with the
major systems biology standards, and so it can be integrated into the
current platforms that are used by the preexisting biocomputational
systems biology infrastructure. Finally, it is worth noting that the
present architecture will allow us to easily integrate data from more
repositories, to enrich subsequent versions of the application and that
the use of an ontology-based mediator could allow us to improve
the integration process, as well as to infer new knowledge.
ACKNOWLEDGEMENTS
This work has been carried out by the unit 741 of the CIBER
de Enfermedades Raras (Rare Diseases). The CIBER de
Enfermedades Raras is an initiative of the ISCIII (Spanish Ministry
of Health).
Funding: Plan Andaluz de Investigacin (BIO-267 and Grants
of Excellence 2999); The Ramn Areces Foundation; Ministry
of Sciences and Innovation, Spain (SAF2008-02522); Spanish
Ministry of Education and Science (TIN2005-09098-C05-01); Junta
de Andaluca project P07-TIC-02978; CIBER (Grant 741.1).
Conict of Interest: none declared.
REFERENCES
Brooksbank,C. et al. (2005) The European Bioinformatics Institutes data resources:
towards systems biology. Nucleic Acids Res., 33, D46D53.
Hucka,M.et al. (2003) The systems biology markup language (SBML): a medium for
representation and exchange of biochemical network models. Bioinformatics, 19,
524531.
Kanehisa,M. and Goto,S. (2000) KEGG: kyoto encyclopedia of genes and genomes.
Nucleic Acids Res., 28, 2730.
Lee,D.-Y. et al. (2008) Web-based applications for building, managing and analysing
kinetic models of biological systems. Brief. Bioinform. [Epub ahead of print,
doi:10.1093/bib/bbn039, September 19, 2008].
Le Novre,N. et al. (2005) Minimum information requested in the annotation of
biochemical models (MIRIAM). Nat. Biotechnol., 23, 15091515.
Navas-Delgado,I. et al. (2008) AMMO-Prot: amine system project 3D-model nder.
BMC Bioinformatics, 9(Suppl 4), S5.
Oinn,T. et al. (2004) Taverna: a tool for the composition and enactment of bioinformatics
workows. Bioinformatics, 20, 30453054.
Schomburg,I. et al. (2004) BRENDA: the enzyme database: updates and major new
developments. Nucleic Acids Res., 32, D431D433.
Wittig,U. et al. (2006) SABIO-RK: integration and curation of reaction kinetics data.
Lect. Notes Bioinform., 4075, 94103.
835
79
Captulo 2
2.0
28.0

2.0
28.0
Metabolizando
nmeros
2.0
28.0
El fluir termodinmico que condiciona
qu reacciones sern posibles bajo cada
circunstancia conforma la estructura disi-
pativa que somos los seres vivos. Cada ruta
metablica es modelable por separado
pero imposible de escindir de su conjunto,
pues depende del entorno, as como de su
propia dinmica. Desde esta perspectiva,
puede pensarse en las distintas rutas meta-
blicas como en las piezas de un lego,
cuya estructura aporta informacin sobre
el control general del sistema.
No es entonces un absurdo pensar en la
posibilidad, por tanto, de ir descubriendo
las propiedades emergentes del sistema a
partir de la conexin de distintos mdulos
metablicos siguiendo una aproximacin
bottom-up.
83
2.0
28.0
Metabolizando nmeros:
Modelos matemticos del metabolismo
Es la matemtica invencin o descubrimiento? elaboradas cons-
trucciones sin autenticidad pero cuya elegancia engaa incluso a sus
inventores? o estn descubriendo verdades que ya estaban ah?
Roger Penrose
La nueva mente del emperador
3
Ral Montaez Martnez
84
2.0
28.0
Mathematical Modeling of Polyamine Metabolism in
Mammals
*
S
Received for publication, March 23, 2006, and in revised form, May 17, 2006 Published, JBC Papers in Press, May 18, 2006, DOI 10.1074/jbc.M602756200
Carlos Rodr guez-Caso
1
, Rau l Montan ez

, Marta Cascante

, Francisca Sa nchez-Jime nez

, and Miguel A. Medina

2
Fromthe

Departamento de Biolog a Molecular y Bioqu mica, Facultad de Ciencias, Universidad de Malaga, Malaga E-29071,
Spain and

Departamento de Bioqu mica, Facultad de Qu mica, Universidad de Barcelona, Barcelona E-08028, Spain
Polyamines are considered as essential compounds in living
cells, since they are involved in cell proliferation, transcription,
and translation processes. Furthermore, polyamine homeosta-
sis is necessary tocell survival, andits deregulationis involvedin
relevant processes, such as cancer and neurodegenerative disor-
ders. Great efforts have been made to elucidate the nature of
polyamine homeostasis, giving rise to relevant information con-
cerning the behavior of the different components of polyamine
metabolism, and a great amount of information has been gener-
ated. However, a complex regulation at transcriptional, transla-
tional, and metabolic levels as well as the strong relationship
between polyamines and essential cell processes make it diffi-
cult to discriminate the role of polyamine regulation itself from
the whole cell response when an experimental approach is given
in vivo. To overcome this limitation, a bottom-up approach to
model mathematically metabolic pathways couldallowus to eluci-
date the systemic behavior from individual kinetic and molecular
properties. In this paper, we propose a mathematical model of
polyaminemetabolismfromkineticconstants andbothmetabolite
and enzyme levels extracted from bibliographic sources. This
model captures the tendencies observed in transgenic mice
for the so-called key enzymes of polyamine metabolism, orni-
thine decarboxylase, S-adenosylmethionine decarboxylase
and spermine spermidine N-acetyl transferase. Furthermore,
the model shows a relevant role of S-adenosylmethionine and
acetyl-CoA availability in polyamine homeostasis, which are
not usually considered in systemic experimental studies.
During much of the last century, biology was an attempt to
reduce biological phenomena to the behavior of molecules.
Despite the enormous success of this approach, most biological
functions arise from interactions among many components,
yielding nonlinear behavior that has been fine tuned by natural
selection to achieve specific functional properties (1, 2). There-
fore, a comprehensive understanding of biological functions
requires a new systemic perspective. In the last few years, sys-
tems biology and synthetic biology have emerged to fulfill this
goal (35). Systems biology approaches are hypothesis-driven
and involve iterative rounds of model building, prediction,
experimentation, and model refinement and development (6,
7). Computer science development is allowing faster and faster
numerical simulations of mathematical models. Interesting and
relevant mathematical models of several well known metabolic
pathways have been published very recently (810). Far from
replacing knowledge acquisition from experimental evidence,
mathematical modeling allows to test and define more accurately
hypothesis that have to be experimentally tested later. Further-
more, modelingallows tobuilda comprehensive frameworkbased
on a compilation of the experimental data provided by the scien-
tific community. With this philosophy, we propose and study a
mathematical model of polyamine metabolism. We will see that
polyamine metabolism exhibits some features that, on one hand,
makepossiblethemodelingtaskand, ontheother, makemodeling
an excellent tool to propose novel hypotheses to be tested.
Ornithine-derived polyamines (putrescine, spermidine, and
spermine) are biogenic low molecular weight organic polyca-
tions that are present in all living organisms. They have pleio-
tropic effects, with a recognized major role as metabolic regu-
lators of the rate and fidelity of macromolecular synthesis, cell
proliferation/death balance, and cell differentiation, among
others (1115). Direct interactions between polyamines and
nucleic acids (16, 17), proteins (1821), and biological mem-
branes (22) have been proposed to explain some of these bio-
logical functions. In fact, estimations of noncovalent polyamine
binding to macromolecules suggest that only around 10% of
total polyamines are free in mammalian cells (23, 24). Further-
more, it is noteworthy that polyamines are involved in the
regulation of the expression of certain genes by means of
polyamine response elements in their promoters (25, 26). Intra-
cellular levels of polyamines must be maintained within narrow
limits, since a decrease of polyamine levels interferes with cell
growth, whereas an excess appears to be toxic (2729). In fact,
a deregulation of polyamine metabolism is related to different
pathologies, such as cancer (16), psoriasis (30), and neurode-
generative diseases (19, 20). Polyamine-cancer connections
have been extensively studied for decades, pointing to inhibi-
tion of polyamine synthesis or activation of polyamine catabo-
lism as a potential cancer chemopreventive strategy (3133).
Polyamine metabolismin cancer is a good example of a process
showing a strong interweaving between interacting genes and
metabolites, since specific oncogenes and tumor suppressor are
* This work was supported by Andalusian Government Grants SAF2005-
01812 and funds to group CVI-267 and to the Amine System Project
(CVI-657) (to M. A. M. and F. S.-J.) and Ministerio de Ciencia y Tecnolog a,
Spain, Grant SAF2005-01627 (to M. C.). The costs of publication of this arti-
cle were defrayedinpart by the payment of page charges. This article must
therefore be hereby marked advertisement in accordance with 18 U.S.C.
Section 1734 solely to indicate this fact.
S
The on-line versionof this article (available at http://www.jbc.org) contains
Equations 118 and Tables S1 and S2.
1
Recipient of a fellowship from the Spanish Ministry of Education and
Science. Present address: Complex Systems Lab, Universitat Pompeu
Fabra, Dr. Aiguader 80, Barcelona 08003, Spain.
2
To whom correspondence should be addressed. Tel.: 34-95-2137132; Fax:
34-95-2132000; E-mail: medina@uma.es.
THE JOURNAL OF BIOLOGICAL CHEMISTRY VOL. 281, NO. 31, pp. 2179921812, August 4, 2006
2006 by The American Society for Biochemistry and Molecular Biology, Inc. Printed in the U.S.A.
AUGUST 4, 2006 VOLUME 281 NUMBER 31 JOURNAL OF BIOLOGICAL CHEMISTRY 21799

a
t

U
N
I
V
E
R
S
I
D
A
D

D
E

M

L
A
G
A

-

T
r
i
a
l

S
u
b
s
c
r
i
p
t
i
o
n
,

o
n

J
a
n
u
a
r
y

7
,

2
0
1
0
w
w
w
.
j
b
c
.
o
r
g
D
o
w
n
l
o
a
d
e
d

f
r
o
m

http://www.jbc.org/content/suppl/2006/06/21/M602756200.DC1.html
Supplemental Material can be found at:
85
Captulo 3
2.0
28.0
regulators of polyamine metabo-
lism, and polyamine levels affect the
rate of cell proliferation (3437).
Due to the pleiotropic and impor-
tant roles of polyamines, their
metabolism has long been the focus
of biochemical studies that have
provided extensive and detailed
information concerning each of the
enzymes and metabolites of the
pathway (22, 38, 39). However, most
of this information is very disperse
and not integrated in a comprehen-
sive, systemic framework. On the
other hand, many therapeutic strat-
egies based on the specific inhibi-
tion of one of the key enzymes of polyamine metabolism have
failed, mostly due to the presence of compensating mechanisms
in polyamine metabolism that contribute to the buffering of
those effects elicited when only an enzyme is the target (15, 40).
The structure of the reactiondiagramof polyamine metabolism
in mammals is relatively complex, consisting of a bi-cycle hav-
ing two required entrances (ornithine and S-adenosylmethi-
onine) and several alternative exits (Fig. 1). For most of the
reactions, both activities and turnover rates of enzymes depend
onpolyamine concentrations ina nonlinear way. Therefore, the
behavior of the full pathway in response to genetic and environ-
mental perturbations cannot be easily deduced from the reac-
tiondiagramitself. Nevertheless, if the behavior of the elements
of a system is known, they can be assembled in a model to
acquire a global knowledge of the system. This is known as a
bottom-up approach. In this case, the global behavior of poly-
amine metabolismtaken as a biomodule (1) can be investigated
with a mathematical model, which describes the reactions and
interactions among its components. This will allowus to ascer-
tain whether potential strong modulators of polyamine metab-
olism are expected to induce relevant effects administered
either alone or in combination.
In this paper, the first basic mathematical modeling of poly-
amine metabolisminmammals is described. It is basedonavail-
able experimental metabolite concentrations and kinetic data.
This work is carried out under the philosophy of a modeling-
testing recursive operation, in order to obtain the minimal
model able to capture the majority of tendencies observed in
the literature. By means of this process, starting fromthe model
shown in Fig. 1A (which includes the enzymes and metabolites
usually determined in experimental studies concerning poly-
amine metabolism), we obtained a model that considers the
study system shown in Fig. 1B, which includes S-adenosylmethi-
onine (SAM)
3
as a variable and takes into account acetyl-CoA
availability. Our model does not, at this stage, include details, such
as polyamine and cationic amino acid transport, compartmental-
ization, and detailed gene expression regulation (22, 41). Despite
these simplifications, this basic model reproduces and explains
many experimental findings. Furthermore, this modeling con-
tributes to and aims at understanding the emergent properties
of this nonlinear complex-interconnected biomodule and its
responses to genetic and environmental variations.
MATERIALS ANDMETHODS
Study System Definition and Mathematical Model Descrip-
tionWe built a mathematical model of mammalian poly-
amine metabolism (Fig. 1) from metabolite concentrations and
enzymic constants (Michaelis constants, inhibitory constants,
and activities) derived from relevant published studies carried
out by expert groups actively involved in polyamine research.
The limits of the system under study were defined by an itera-
tive process of model refinement in order to achieve the sim-
plest model able to capture the more relevant features of poly-
amine metabolism. We started with a basic model, where we
only considered the bi-cyclic central core for the interconver-
sion of putrescine, spermidine, and spermine, with ornithine
and SAM as entries of the system (Fig. 1A). The initial versions
of the model consider the enzyme activities and polyamine con-
centrations usually taken into account and determined in
experimental studies. As a result of this iterative modeling
process, we propose in this paper an extended version of this
study system (Fig. 1B), where entries are through ornithine
decarboxylase (ODC) and methionine adenosyltransferase.
Their respective substrates, ornithine and methionine, were
considered parameters inour model. We also took into account
the role of acetyl-CoAas a cosubstrate in polyamine acetylation
reactions.
The study systemproposed in Fig. 1B was modeled by taking
into account some assumptions. The entries of the model (orni-
thine, methionine, and acetyl-CoA) were considered as param-
eters. Putrescine and acetylspermidine were the unique molec-
ular species able to go out of the model. This simplification was
done based on studies with cultured cells showing that, indeed,
putrescine and acetylspermidine are the major forms of poly-
amine release (42). These releases were modeled by simple, lin-
ear equations. Terminal catabolism of putrescine by diamine
3
The abbreviations usedare: SAM, S-adenosylmethionine; SSAT, spermidine/
spermine acetyltransferase; P, putrescine; D, spermidine; S, spermine; A,
decarboxylated S-adenosylmethionine; aD, N-acetylspermidine; aS,
N-acetyl-spermine; Antz, antizyme; MAT, methionine adenosyltransferase;
ODC, ornithine decarboxylase; SAMdc, S-adenosylmethionine decarboxy-
lase; SpdS, spermidine synthase; SpmS, spermine synthase; PAO, poly-
amine oxidase; 5MTA, 5-methylthioadenosine; DFMO, difluoromethyl-
ornithine; MGBG, methylglyoxalbis(guanylhidrazone).
FIGURE 1. Polyamine metabolism. Components considered in the initial (A) and further versions (B) of the
model are shown. The acetyl-CoA/CoA recycling, only considered in the final version of the model, is repre-
sented as a dotted box.
Mammalian Polyamine MetabolismModel
21800 JOURNAL OF BIOLOGICAL CHEMISTRY VOLUME 281 NUMBER 31 AUGUST 4, 2006

a
t

U
N
I
V
E
R
S
I
D
A
D

D
E

M

L
A
G
A

-

T
r
i
a
l

S
u
b
s
c
r
i
p
t
i
o
n
,

o
n

J
a
n
u
a
r
y

7
,

2
0
1
0
w
w
w
.
j
b
c
.
o
r
g
D
o
w
n
l
o
a
d
e
d

f
r
o
m

http://www.jbc.org/content/suppl/2006/06/21/M602756200.DC1.html
Supplemental Material can be found at:
Ral Montaez Martnez
86
2.0
28.0
TABLE 1
Rate equations for enzymes and processes included in the model
Mammalian Polyamine MetabolismModel
AUGUST 4, 2006 VOLUME 281 NUMBER 31 JOURNAL OF BIOLOGICAL CHEMISTRY 21801

a
t

U
N
I
V
E
R
S
I
D
A
D

D
E

M

L
A
G
A

-

T
r
i
a
l

S
u
b
s
c
r
i
p
t
i
o
n
,

o
n

J
a
n
u
a
r
y

7
,

2
0
1
0
w
w
w
.
j
b
c
.
o
r
g
D
o
w
n
l
o
a
d
e
d

f
r
o
m

http://www.jbc.org/content/suppl/2006/06/21/M602756200.DC1.html
Supplemental Material can be found at:
87
Captulo 3
2.0
28.0
oxidase was not included in the model, since this activity is
unevenly distributed in mammals and terminal catabolism
appears to be restricted to selected tissues (43, 44). In any case,
due to the nature of the efflux inour model, putrescine release
can also be considered as a combination of putrescine efflux
plus terminal catabolism for this diamine. Antizyme, a small
regulatory protein regulated by polyamine levels (45) with a key
role in ODC degradation (28, 39, 46), was included in the
model. SAM availability was taken into account by introducing
in the model the rate equation assumed in a methionine cycle
model published recently (8). For acetyl-CoA availability, we
assumed a linear interconversion between acetyl-CoA and
CoA, taking the acetyl-CoA CoA pool and the recycling rate
Ras parameters. This assumptionis based onthe postulate that,
under certain conditions, SSAT inductions can deplete acetyl-
CoA levels. Other aspects, such as polyamine and cationic
amino acid transport, compartmentalization, and detailed gene
expression regulatory features (22, 41), were not included in the
model at the present stage.
Table 1 gives the rate equations for all of the enzymes
included in the model, along with a brief description of their
features. For the bisubstrate enzymes spermidine and spermine
synthases (SpdSandSpmS), theinhibitoryeffect of thesubproduct
5-methylthioadenosine (5MTA) was omitted. We assumed in
the model that 5MTA was virtually zero. This assumption was
supported by the fact that 5MTA is quickly removed by 5MTA
phosphorylase (47, 48). Table 2 gives the differential equations for
the different metabolites and other time-dependent variables.
More details on relevant features of the enzymes included in the
model are given in the supplemental materials.
Model Implementation and Initial ConditionsThe model
was implemented in Perl language (available on the World
TABLE 2
Differential equations for the different metabolites and other time-dependent variables included in the model
Metabolite-dependent velocities are indicated by combining both enzyme and metabolite abbreviations as the subscripted notations. (e.g. V
SSATD
, SSAT velocity for
spermidine; V
PAOaS
, PAO velocity for acetylated spermine, etc.) AcCoA, acetyl-CoA.
Mammalian Polyamine MetabolismModel
21802 JOURNAL OF BIOLOGICAL CHEMISTRY VOLUME 281 NUMBER 31 AUGUST 4, 2006

a
t

U
N
I
V
E
R
S
I
D
A
D

D
E

M

L
A
G
A

-

T
r
i
a
l

S
u
b
s
c
r
i
p
t
i
o
n
,

o
n

J
a
n
u
a
r
y

7
,

2
0
1
0
w
w
w
.
j
b
c
.
o
r
g
D
o
w
n
l
o
a
d
e
d

f
r
o
m

http://www.jbc.org/content/suppl/2006/06/21/M602756200.DC1.html
Supplemental Material can be found at:
Ral Montaez Martnez
88
2.0
28.0
Wide Web at www.perl.com/). Perl is a multiplatform, free, and
easy-to-learn programming language commonly used by bioin-
formaticians. Furthermore, the use of Perl allows the solutionof
complex differential equation systems with low computational
cost, due to its simplicity and its capability to work without the
requirement of a graphic interface. Nonetheless, other pro-
gramming languages, such as Python, can also be used for
implementation of the model. Whatever the programming lan-
guage used, our program is a script of instructions with the
following basic structure: 1) parameter definition; 2) definition
of initial conditions for all of the variables; 3) beginning of an
iterative loop ((a) calculation of velocities from differential
equations providing the current values of variables; (b) updat-
ing of the variable values by multiplying velocities by step size;
(c) calculationof the current simulationtime by multiplying the
number of iterations by the step size; (d) collection of variables
and time in a file); 4) end of program.
Simulations were carried out by the iterative Euler method,
witha time increment of 0.01 min/iteration. For numerical sim-
ulations of the model, a PC with an AMD-K7 Athlon 1800 XP
microprocessor with Linux as the operating system and an
Apple Macintosh Powerbook G4 with MacOS X were used
interchangeably.
Table S1 shows the ranges of values available in literature for
the parameters considered in our model as well as our initial
choices for the simulations. For those parameters with no avail-
able reference in the literature, the values included in the model
for simulations were chosen by initial tuning, an optimization
process to achieve steady states within the physiological range
of values for the elements included in the system. Physiological
levels of polyamines are found in cultured cells in the submilli-
molar range, and activities are in the range of M min
1
(23,
4954). In order to analyze the behavior of these tuned
parameters in the acquisition of steady states, we carried out
additional simulations, changing their values by a factor of 0.1
or 100.
On the other hand, since polyamines are distributed among
free and noncovalently bound polyamine pools in vivo, we
tuned the model to respond to free polyamines, taking the free
polyamine/total polyamine ratio as a constant. As previously
mentioned, 10% of total polyamines (spermine and spermi-
dine) are estimated to be free in living cells according to the
literature (23, 24). In this work, [S] and [D] values (spermine
and spermidine in mathematical model notation, respectively)
are shown as free polyamine concentrations.
Since activities and polyamine levels available in the litera-
ture are expressedwithrespect to the total amount of proteinor
number of cells, we converted these units to M min
1
(activi-
ties) and M (metabolites). For these conversions, we took 2 pl
as a mean cellular volume, as well as the equivalence of 100 g
of protein contained in 10
6
cells (55).
In all of the simulations, steady state was considered to be
reached when the change of any variable was lower than 10
12
within an interval of 3 h (18,000 iterations).
Sensitivity AnalysisSensitivity analysis determines the
relative importance of the different parameters to induce
TABLE 3
Comparison of the predictive capabilities of the different versions of the model
In all cases, overexpressions were simulated by increasing 100 times the basal level of the respective activity.
Description ODC overexpression SSAT overexpression SAMdc overexpression
Version 1
Components are shown in Fig. 1A. Polyamine homeostasis and high
putrescine increases
a
Polyamine depletion and putrescine
homeostasis
Fast increases to
nonphysiological levels
of polyamines
Activities are considered as parameters.
Version 2
Components are shown in Fig. 1B. Polyamine homeostasis and high
putrescine increases
a
Polyamine depletion and putrescine
homeostasis
Slow responses tending to
increase polyamine
levels
Activities, [acetyl-CoA] and [CoA], are
considered as parameters.
Version 3
Components are shown in Fig. 1A. Polyamine homeostasis and high
putrescine increases
a
Polyamine depletion and increases
of both ODC and putrescine
a
Increases of polyamine
levels
ODC, SAMdc, and SSAT activities are
considered as time-dependent
variables.
Version 4 (final model)
Components are shown in Fig. 1B. Polyamine homeostasis and high
putrescine increases
a
Polyamine depletion or homeostasis
(depending on the acetyl-CoA
availability) and increases of both
ODC and putrescine
a
Polyamine homeostasis
a
ODC, SAMdc, and SSAT activities are
considered as time-dependent
variables.
[Acetyl-CoA]/[CoA] recycling is
considered.
a
Results according to the major tendencies reported in the literature.
TABLE 4
Polyamine concentrations and enzyme activities considered as the basal conditions in the final model compared with those calculated from
data in Ref. 58
Free polyamine concentrations Enzyme activities
[Antz]
Putrescine Spermidine Spermine ODC SSAT SAMdc
M M min
1
M
Data from Ref. 58 131.9 91.4 38.5 9.5 Not given 0.30 Not given
Data from the model 109.4 73.4 53.4 1.6 1.3 0.9 0.6
Mammalian Polyamine MetabolismModel
AUGUST 4, 2006 VOLUME 281 NUMBER 31 JOURNAL OF BIOLOGICAL CHEMISTRY 21803

a
t

U
N
I
V
E
R
S
I
D
A
D

D
E

M

L
A
G
A

-

T
r
i
a
l

S
u
b
s
c
r
i
p
t
i
o
n
,

o
n

J
a
n
u
a
r
y

7
,

2
0
1
0
w
w
w
.
j
b
c
.
o
r
g
D
o
w
n
l
o
a
d
e
d

f
r
o
m

http://www.jbc.org/content/suppl/2006/06/21/M602756200.DC1.html
Supplemental Material can be found at:
89
Captulo 3
2.0
28.0
changes in time-dependent variables. From an initial steady
state, an infinitesimal change in the value of a parameter is
introduced, and the global changes in the system are evalu-
ated after acquiring a new steady state. Detailed information
on how this kind of analysis was carried out can be found in
the supplemental material.
In Silico Simulations of Experiments Representing Gene or
Environmental ChangesWe focused our studies on induc-
tions and inhibitions of the three enzymes described as rate-
limiting of the overall metabolic pathway, namely, ODC,
SAMdc, and SSAT (13). As detailed in the supplemental
material, these three key enzymes and the regulator anti-
zyme have turnover rates that depend on polyamine concen-
trations, a key feature taken into account in our model. To
simulate gene manipulation leading to induction (overex-
pression), two different approaches were used. We tested the
effects of different increases (11000-fold) in the values of
the respective k
s
parameters. In other cases, inductions were
simulated by direct increase of the respective values of activ-
ities in the basal conditions, taking those activities not as
time-dependent variables but, on the contrary, as time-inde-
pendent parameters. Down-regulations of rate-limiting
enzyme activities were simulated by up to 10-fold decreases
of k
s
values. We also simulated pharmacological interven-
tions leading to inhibition of ODC and SAMdc by difluoro-
methylornithine (DFMO) and methylglyoxalbis(guanyl-
hidrazone) (MGBG), respectively (22). For ODC inhibition
with DFMO, its activity was treated as a time-independent
parameter and received a value of zero, since DFMO is
described as a irreversible inhibitor of ODC (56). However,
concerning SAMdc inhibition, the competitive inhibitor
MGBG (57) was considered to produce a potent decrease
FIGURE 2. Effects of changes inODCk
S
(AandB) andAntz k
S
(CandD) ontime-dependent activities (AandC) (ODC(empty circles), Antz (crosses), SSAT
(black diamonds), and SAMdc (empty squares)) and on metabolite levels (B and D) (putrescine (empty circles), spermidine (empty squares), and
spermine (full circles)), relative to the basal condition steady state.
Mammalian Polyamine MetabolismModel
21804 JOURNAL OF BIOLOGICAL CHEMISTRY VOLUME 281 NUMBER 31 AUGUST 4, 2006

a
t

U
N
I
V
E
R
S
I
D
A
D

D
E

M

L
A
G
A

-

T
r
i
a
l

S
u
b
s
c
r
i
p
t
i
o
n
,

o
n

J
a
n
u
a
r
y

7
,

2
0
1
0
w
w
w
.
j
b
c
.
o
r
g
D
o
w
n
l
o
a
d
e
d

f
r
o
m

http://www.jbc.org/content/suppl/2006/06/21/M602756200.DC1.html
Supplemental Material can be found at:
Ral Montaez Martnez
90
2.0
28.0
(90%) of activity. To simulate ODC, SAMdc, and SSAT alter-
ations caused by the polyamine analog DENSPM (N
1
,N
11
-
diethylnorspermidine) at a 1 mM intracellular concentration
(58), we considered it as a structural analog of higher poly-
amines; thus, we replaced ([D] [S]) by ([D] [S] 1000) in
the equations for the time-dependent enzymes described in
Table 2. All of the time response effects were simulated for
96 h (576,000 iterations).
RESULTS ANDDISCUSSION
From the Core System Model to Our Final Model: Iterative
Model RefinementA main goal of modeling is to discern the
minimal relevant processes able to explain the majority of
experimental data concerning the behavior of the modeled sys-
tem. As mentioned above, we started with a core system model
(Fig. 1A) based on those elements of polyamine metabolism
usually taken into account in experimental studies. This core
system was properly modeled defining the involved rate veloc-
ity equations, maintaining fixed all involved enzymic activities;
later, it was extended to include differential equations of time-
dependent variables and initial conditions. The procedure was
similar to that described under Materials and Methods and
shown in Tables 1, 2 and S1) for our final model (Fig. 1B). For
the sake of simplicity, herein we only show details concerning
the final model, but simulations of the core system model
yielded a physiological basal steady state and sensitivity analysis
showing that this model was robust (results not shown). Fur-
thermore, this core systemmodel was good enough to simulate
properly much of the available experimental data concerning
modulation (overexpression or inhibition) of three key
enzymes (see Table 3), with the exception of some features of
SAMdc and SSATmodulation that could not be well simulated
(discussedbelow). However, this is not a limitationof modeling;
on the contrary, this kind of disagreement regarding available
experimental results is a sign of the predictive potential of the
modeling approach, since it points out some essential aspects
not usually taken into account in experimental studies of poly-
amine metabolism modulation. Concretely, simulations of the
core systemmodel describedinFig. 1Asuggestedthat SAMand
acetyl-CoA (two metabolites not included in the core system
model) couldplay a mainrole inthis metabolic pathway. To test
this hypothesis, we carried out an iterative model refinement
leading to the model shown in Fig. 1B. More details are given in
the supplemental material. Table 3 summarizes the compara-
tive predictive capabilities of the models generated and tested
during this iterative refinement process. As stated above, from
now on, we only present details concerning the model repre-
sented in Fig. 1B.
Initial Conditions, Steady State Acquisition, and Sensitivity
Analysis of the ModelAs showninTable S1 of the supplemen-
tal materials and mentioned above, several parameters of the
model were not directly available in the scientific literature.
They include the equilibrium constants for the interaction of
polyamines with the enzymes regulated by their concentrations
(ODC, SAMdc, and SSAT), the velocity constants for synthesis
and degradation in the modified Schimke equations, the inhi-
bition constant for spermidine on spermine synthase activity,
and the velocity constants in the linear equations for effluxes.
For these parameters, the model reached steady state over a
wide range of initial values tested. However, as described under
Materials and Methods, the specific values finally included as
initial conditions (Table S1) were selected by a tuning of the
model in order to obtain physiological levels of polyamines and
enzyme activities. A lack of experimental data frequently
imposes such kinds of adjustments for the proper performance
of mathematical models of metabolic pathways (9). Nonethe-
less, half-life values for the short lived enzymes ODC, SAMdc,
and SSATare available in the literature, and they can be used to
check whether the k
D
values selected by tuning of the model
were out of range or not. From these values, theoretical half-
lives of ODC, SAMdc, and SSAT are calculated to be 14, 9, and
10 min, respectively. These three values agree with those pre-
viously reported in experimental works. In fact, mammalian
ODC is described as having a half-life of minutes, as short as
5 min (59); for SAMdc, half-life determined experimentally
is around 20 min (59); and, finally, SSAT overexpressed in
COS-7 cells has a half-life of 20 min (60).
Simulations of the model described in Fig. 1B and in Tables 1
and 2, starting fromthe initial conditions described in Table S1
(included in the supplemental material) yielded a steady state
that we call a basal condition. Table 4 shows that free polyamine
concentrations and enzyme activities in this basal condition fit
remarkably well to actual values estimated from experimental
available data (58). Furthermore, simulations with different ini-
tial conditions for time-dependent variables and for different
time increments gave the same basal condition steady state,
indicating that our model is not sensitive to initial conditions
and does not exhibit numerical artifacts.
To check the influence of the tuned parameters on the model
behavior, we tested simulations with variations in their values
covering 4 orders of magnitude. For this, in each new simula-
tion, one (and only one) of the tuned parameter values was
FIGURE 3. DFMO effects on polyamine levels in both time-response sim-
ulations anddataextractedfromRef. 58. Insilicoexperiments areshownas
follows: putrescine (continuous line), spermidine (dotted line), and spermine
(segment and dotted line). Experimental data are shownas follows: putrescine
(empty circles), spermidine (empty squares), and spermine (full circles).
Mammalian Polyamine MetabolismModel
AUGUST 4, 2006 VOLUME 281 NUMBER 31 JOURNAL OF BIOLOGICAL CHEMISTRY 21805

a
t

U
N
I
V
E
R
S
I
D
A
D

D
E

M

L
A
G
A

-

T
r
i
a
l

S
u
b
s
c
r
i
p
t
i
o
n
,

o
n

J
a
n
u
a
r
y

7
,

2
0
1
0
w
w
w
.
j
b
c
.
o
r
g
D
o
w
n
l
o
a
d
e
d

f
r
o
m

http://www.jbc.org/content/suppl/2006/06/21/M602756200.DC1.html
Supplemental Material can be found at:
91
Captulo 3
2.0
28.0
multiplied by 0.1 or 100. In all cases, steady state conditions
were reached.
A way to test the capabilities of the model to respond to
changes is to study the influence of the different parameters
on the behavior of time-dependent variables (i.e. to carry out
a sensitivity analysis) (61). Moreover, great changes in sen-
sitivity coefficients can be taken as a probe of model failure,
since a great change in a time-dependent variable caused by
a little change in a parameter is not frequently found in this
kind of mathematical model (6264). The sensitivity analy-
sis for all of the parameters and variables shows the robust-
ness of our model, as detailed in the supplemental materials
(under Results from the Sensitivity Analysis of the Model;
see Table S2).
Model Predicts Compensatory Mechanisms to Keep Poly-
amine Homeostasis in Response to ODC Alterations by either
Genetic or Pharmacological InterventionMuch experimental
data has shown that changes in ODC expression have a prog-
nostic value concerning cell transformation (65, 66). Polyamine
depletion prevents cell proliferation, and ODC has been sug-
gested as a therapeutic target in proliferative disorders, since it
is a main factor responsible for the de novo synthesis of poly-
amines (31, 67, 68). In fact, knock-out ODC mice were not
viable by failure in embryogenesis at the earliest stages (69, 70),
showing that de novo synthesis must be essential in some cellu-
lar and embryonic processes. However, polyamine levels have
proved to be very robust against alterations of ODC contents.
ODCcan be inactivated by its irreversible inhibitor DFMO. Many
FIGURE 4. Effects of changes in SAMdc k
S
(A and B) and k
S
for both SAMdc and ODC at the same time (C and D) on time-dependent activities (A and C)
(ODC(emptycircles), Antz(crosses), SSAT(blackdiamonds), andSAMdc (emptysquares)) andonmetabolitelevels (BandD) (putrescine(emptycircles),
spermidine (empty squares), andspermine (full circles)), relative to the basal condition steady state. B (inset), changes in the concentrations of SAMand
decarboxylated SAM (A).
Mammalian Polyamine MetabolismModel
21806 JOURNAL OF BIOLOGICAL CHEMISTRY VOLUME 281 NUMBER 31 AUGUST 4, 2006

a
t

U
N
I
V
E
R
S
I
D
A
D

D
E

M

L
A
G
A

-

T
r
i
a
l

S
u
b
s
c
r
i
p
t
i
o
n
,

o
n

J
a
n
u
a
r
y

7
,

2
0
1
0
w
w
w
.
j
b
c
.
o
r
g
D
o
w
n
l
o
a
d
e
d

f
r
o
m

http://www.jbc.org/content/suppl/2006/06/21/M602756200.DC1.html
Supplemental Material can be found at:
Ral Montaez Martnez
92
2.0
28.0
studies using DFMOhave shown effective depletion of putrescine
and spermidine, whereas spermine content is often unaffected.
DFMO has been described as a cytostatic rather than cytotoxic
drug in cancer therapy (71). Moreover, its use in clinical chemo-
therapyhas not shownas great results as initiallyexpected(31). On
the other hand, overexpressing ODCtransgenic mice donot show
increased levels of spermidine and spermine (72).
In our model, an increase of putrescine as a consequence of
ODC expression modulation could be reached either by simu-
lating an ODC overexpression or by simulating the effects of
polyamines on antizyme translation (45) by changing their k
S
values. Putrescine could also be increased by affecting enzymes
related to back conversionof polyamines. However, inthis case,
the putrescine increase would be a consequence and not a cause
of polyamine depletion, as we will see later. Fig. 2 (A and B for
ODC and C and D for antizyme) shows the relative changes of
the steady state values as compared with basal condition when
k
S
for each enzyme regulated by the Schimke equation is mod-
ified in a range that covers 5 orders of magnitude (0.11000-
fold increases of its value in basal conditions, as mentioned
under Materials and Methods). Antizyme decreased ODC
levels, affecting neither the other enzymes nor higher poly-
amine (spermidine and spermine) levels. Acetyl-CoAand SAM
levels remained unaffected (results not shown). ODC and anti-
zyme increases only affected putrescine levels, as suggested by
the sensitivity analysis for these enzymes (see Table S2). This is
in agreement with experimental data on overexpressing ODC
transgenic mice, where polyamine levels are unaltered while
putrescine is highly increased (72, 73). In our model, only when
putrescine appeared to be limiting (simulated by ODC down-
regulation), spermine levels were slightly modified(Fig. 2, Band
D), in agreement with DFMO treatment simulation results, as
shown and discussed below.
To check the influence of ODC suppression on the behavior
of the system, we simulated a treatment with DFMO. Fig. 3
shows the time responses to a DFMO treatment obtained with
simulations of our model andcomparedwithexperimental data
(58). Our simulations predict very well the real experimental
responses, with a drastic decrease of putrescine that caused a
slight increase of SAM (results not shown), followed by a
decrease of spermidine and a slight increase of spermine. The
same tendencies, but more attenuated, were also observed for
0.1-fold ODC activity down-regulation (Fig. 2). In fact, such
results were already seen in simulations of the core system
model (Table 3 and results not shown).
Summarizing the results of simulated ODCmodulations, our
model predicts very well the available experimental data, show-
ing that dramatic changes in putrescine levels are not accom-
panied by similar changes in the levels of higher polyamines
(spermidine and spermine), at least in the short term. This
robustness of polyamine metabolism against alterations in the
key enzyme ODC has been suggested to be due to compensat-
ing alterations in the rate of exogenous polyamine uptake, since
antizyme behaves as an inhibitor of the transporter (74). In
addition to this transport-dependent effect, our simulations
suggest that a transport-independent component (the regula-
tory mechanism of the bi-cyclic pathway) also contributes to
this robustness. This fact and the proven importance of ODCin
cell transformation could suggest that putrescine could serve
not only as a requested substrate for spermidine (and spermine)
synthesis but also as a regulatory signal of cell proliferation. It
has been shown that peaks of putrescine occur in different
phases of the cell cycle (29, 75). Furthermore, putrescine stim-
ulates tyrosine phosphorylation by tyrosine kinases and the
expression of the nuclear oncogenes c-fos and c-jun (35).
Model Predicts the Effects of SAMdc Modulation in Accord-
ance with Experimental Data Obtained Using Transgenic Mice
andInhibitorsSynthesis of spermidine andspermine does not
only require putrescine but also an aminopropyl donor, namely
decarboxylated S-adenosylmethionine, the product of SAMdc
activity. Therefore, it could be argued that SAMdc induction
would cause an increase in the levels of higher polyamines.
Under this hypothesis, great efforts have been devoted to
obtaining SAMdc-overexpressing or knocked out transgenic
mice. As it happens with ODC, SAMdc knock-outs are lethal at
the earliest stages of embryonic development (76). SAMdc-
overexpressing mice show modest inductions (5-fold increase)
(49), with only small changes of higher polyamine levels. In a
recently published work, 100-fold inductions have been
obtained with little variation of polyamine content (77). Such
works and others with SAMdc-overexpressing cells (7880)
show that putrescine decreases. However, a cross-breeding of
ODC- and SAMdc-overexpressing transgenic mice shows that
putrescine is not limiting in polyamine synthesis, and those
mice still maintain a polyamine homeostasis. Heljasvaara et al.
(49) suggest that, since polyamine accumulation is toxic, a
homeostasis is not unexpected. Indeed, they proved that this
homeostasis is reached by an increased rate of the bi-cycle, with
a faster polyamine synthesis followed by an increase of acety-
lated polyamine efflux, although they could not find any direct
evidence of increased SSAT activity (49).
Simulations of SAMdc and SAMdc/ODC overexpression
(Fig. 4) predict a very remarkable buffering of the levels of
FIGURE 5. MGBG effects on polyamine levels in both time-response sim-
ulations anddataextractedfromRef. 58. Insilicoexperiments areshownas
follows: putrescine (continuous line), spermidine (dotted line), and spermine
(segment and dotted line). Experimental data are shownas follows: putrescine
(empty circles), spermidine (empty squares), and spermine (full circles).
Mammalian Polyamine MetabolismModel
AUGUST 4, 2006 VOLUME 281 NUMBER 31 JOURNAL OF BIOLOGICAL CHEMISTRY 21807

a
t

U
N
I
V
E
R
S
I
D
A
D

D
E

M

L
A
G
A

-

T
r
i
a
l

S
u
b
s
c
r
i
p
t
i
o
n
,

o
n

J
a
n
u
a
r
y

7
,

2
0
1
0
w
w
w
.
j
b
c
.
o
r
g
D
o
w
n
l
o
a
d
e
d

f
r
o
m

http://www.jbc.org/content/suppl/2006/06/21/M602756200.DC1.html
Supplemental Material can be found at:
93
Captulo 3
2.0
28.0
higher polyamines, in agreement with available experimental
data (49, 77). However, we only observed such a tendency when
we extended the core system model to consider the production
of SAM from methionine. In our model depicted in Fig. 1B,
simulations of SAMdc activity overexpression showed that
SAM levels changed linearly with variations in the relative
levels of SAMdc, maintaining well buffered levels of decar-
boxylated SAM to feed the polyamine metabolism bi-cycle
(see inset in Fig. 4B). Based on these observations, we suggest
that the branch of SAM production in the methyl cycle path-
way could be relevant for polyamine homeostasis. However,
there are no available experimental data to test this hypoth-
esis currently.
The use of the competitive inhibitor MGBG is an alternative
approach to analyze the effects of an inhibition of SAMdc activ-
ity. MGBG increases putrescine levels and diminishes higher
polyamines (58, 71, 81). Actual experimental data on pharma-
cological inhibition with high doses of the reversible inhibitor
MGBG are also remarkably well predicted with simulations of
partial inhibition of SAMdc activity (Fig. 5).
Simulations of Changes of Acetyl-CoA Availability Predict
the Responses to SSAT Induction Observed in Experimental
ModelsSSAT is described as the key enzyme for polyamine
catabolism (43, 82). In SSAT-induced experimental models, a
remarkable accumulation of putrescine and decreases of sper-
midine and spermine levels are observed (8386). However, it
should be stressed that steady state polyamine levels in experi-
mental SSAT-overexpressing models can vary in a wide range,
depending on the actual SSAT induction reached. This is true
even for data published by the same group (51, 87).
In our model (Fig. 1B), acetyl-CoA availability is controlled
by the R parameter (used to estimate the rate of recycling).
FIGURE 6. Effects of changes in SSAT k
S
for two values of the R parameter, that of basal conditions (panels A and B) and that corresponding to a 25%
of the basal conditions value (C and D) on time-dependent activities (A and C) (ODC (empty circles), Antz (cross), SSAT (full triangles), and SAMdc
(empty squares)) andonmetabolite levels (BandD) (putrescine (empty circles), spermidine (empty squares), andspermine (full circles)) relative tothe
basal condition steady state.
Mammalian Polyamine MetabolismModel
21808 JOURNAL OF BIOLOGICAL CHEMISTRY VOLUME 281 NUMBER 31 AUGUST 4, 2006

a
t

U
N
I
V
E
R
S
I
D
A
D

D
E

M

L
A
G
A

-

T
r
i
a
l

S
u
b
s
c
r
i
p
t
i
o
n
,

o
n

J
a
n
u
a
r
y

7
,

2
0
1
0
w
w
w
.
j
b
c
.
o
r
g
D
o
w
n
l
o
a
d
e
d

f
r
o
m

http://www.jbc.org/content/suppl/2006/06/21/M602756200.DC1.html
Supplemental Material can be found at:
Ral Montaez Martnez
94
2.0
28.0
Simulations of SSAT level variations exhibited two different
behaviors depending on the availability of acetyl-CoA for the
degradation pathway. Fig. 6 (A and B) shows that polyamine
depletion is allowed when the requirements of acetyl-CoA are
satisfied. This occurs in our model when Ris high (Fig. 7). How-
ever, when the R value is reduced to 25%of its value under basal
conditions, a new physiological steady state slightly different
from the previous one is achieved (78 M putrescine, 91 M
spermidine, and 76 M spermine). Under these new basal con-
ditions, a SSAT overexpression causes only modest changes in
the levels of polyamines (Fig. 6, Cand D), as seems to be the case
for most of the available experimental data with transgenic
mice (51, 86, 88, 89). Acetyl-CoA levels do not change with
increasing SSATexpression under basal conditions of recycling
(Fig. 6B). In contrast, for R values reduced to 25% of the value
under basal conditions, acetyl-CoA levels drop with increasing
SSAT expression (Fig. 6D). This observation suggests that, in
fact, acetyl-CoA availability determines the actual changes in
polyamine induced by SSAT overexpression.
Model Predicts the Effects of the Polyamine Analog DENSPMon
Polyamine Metabolism Observed with Experimental Data
DENSPM is a polyamine analog described as one of the most
potent inducers of SSAT (90). Furthermore, it also reduces
ODCand SAMdc activities (58, 90). These joined effects lead to
polyamine depletion (87, 91). To simulate the effects of
DENSPM in our model, we assumed that this unmetabolizable
analog causes its effect by a modulation of synthesis and degra-
dation constants for those enzymes ruled by the Schimke equa-
tion. Experimental measurements of intracellular DENSPM
after 24 h of incubation in culture cells were around 1000 M, as
calculated from data described by Korhonen et al. (58). Thus,
we modified the Schimke equation for each enzyme by adding
this concentration value to the polyamine pool (i.e. (S D) was
changed by (S D1000), and simulation was restarted from
basal conditions of the modified model. Fig. 8B shows that the
time response of a DENSPM treatment on polyamine levels
predicted by simulations of our model fits well to experimental
data (58, 90, 92). It should be stressed that higher polyamine
depletion predicted by DENSPM treatment is more drastic
than those caused by a SSAT overexpression (compare Fig. 8B
with Fig. 6) and a SAMdc inhibition (compare Fig. 8B with Fig.
5). Furthermore, in this case of DENSPM treatment, all three
polyamines, including putrescine, were depleted. Fig. 8Ashows
that DENSPM treatment increased SSAT and decreased ODC
and SAMdc activities, in agreement with the tendencies
described by real experimental data (58).
FIGURE 7. Effects of acetyl-CoArecyclingrate inthe free polyamine levels
once steady state is reached. Continuous line, basal conditions; dashed line,
10-fold induced SSAT basal activity conditions. In this case, SSAT activity is
considered a time-independent parameter. Vertical dashed lines define the R
conditions used for the SSAT induction simulations described under Results
and Discussion.
FIGURE 8. DENSPMeffects onshort livedenzymes andpolyamine levels. A, predicted effects on enzymes, ODC(continuous line), Antz (dashed line), SAMdc
(dotted line), and SSAT (segment and dotted line). B, effects on polyamines in both time-response simulations and data extracted from Ref. 58. In silico
experiments are shown as follows: putrescine (continuous line), spermidine (dotted line), and spermine (segment and dotted line). Experimental data are shown
as follows: putrescine (empty circles), spermidine (empty squares), and spermine (full circles).
Mammalian Polyamine MetabolismModel
AUGUST 4, 2006 VOLUME 281 NUMBER 31 JOURNAL OF BIOLOGICAL CHEMISTRY 21809

a
t

U
N
I
V
E
R
S
I
D
A
D

D
E

M

L
A
G
A

-

T
r
i
a
l

S
u
b
s
c
r
i
p
t
i
o
n
,

o
n

J
a
n
u
a
r
y

7
,

2
0
1
0
w
w
w
.
j
b
c
.
o
r
g
D
o
w
n
l
o
a
d
e
d

f
r
o
m

http://www.jbc.org/content/suppl/2006/06/21/M602756200.DC1.html
Supplemental Material can be found at:
95
Captulo 3
2.0
28.0
ConclusionHerein, we describe the first mathematical
model proposed for mammalian polyamine metabolism. Mod-
eling is a process of simplification of real biological phenomena
that imposes severe limitations and restrictions. However, the
essence of modeling is also to obtain the simplest model able to
explain the majority of cases. In this context, a model as com-
plex as reality is useless to acquire a comprehensive under-
standing of experimental evidence. The ultimate goal of this
work is not just to mimic the biochemistry of polyamine metab-
olism but also to define the minimal study system that we
should have to take into account, in both experimental and
theoretical approaches, to understand how this complex path-
way performs its biological function and responds to genetic
and environmental perturbations. Aminimal study systemdef-
inition is possible based on the modular architecture of metab-
olism (93). The topology of the system (shown in Fig. 1) is rel-
atively complex, with a central bi-cycle, two requested
entrances, and two alternative exits. Furthermore, some of the
enzymes involved inthis metabolic pathway exhibit remarkable
regulatory features, which have been taken into account to for-
mulate the basic model. On the other hand, the unhomoge-
neous quality of the sparse and disperse experimental data con-
cerning polyamine metabolism in mammals is another
limitation. Despite all of these constraints, the described math-
ematical model fulfils the initial goals, contributing to increase
our understanding of polyamine metabolismand to predict the
adaptative response of this pathway to genetic and environ-
mental perturbations. In addition, our iterative model refine-
ment process has allowed us to suggest a minimal study system
of polyamine metabolism in which SAM and acetyl-CoA avail-
ability appears as an additional factor to explain controversial
results, such as the diversity of behaviors observed when SSAT
is induced and polyamine homeostasis when SAMdc is
induced. Obviously, this model does not try to mimic the reality
in all of the situations, and the results have to be taken as a
plausible explanation and integration of sparse data compiled
throughalmost 30 years of polyamine research. Incontrast with
top-down metabolic flux control analysis methods (61, 94), our
model represents a bottom-upapproachto anunderstanding of
complex metabolic networks, as is the case for other recently
published models (810).
We have shown that our model of polyamine interconver-
sion reflects some critical features observed in experimental
approaches, such as ODC modulation, polyamine depletion by
SSAT induction, and responses to polyamine analog exposure,
among others. Results from our simulations include predic-
tions on changes in antizyme contents (Figs. 4 and 6, A and C)
not discussed in this work, which could be tested experimen-
tally. The fact that polyamine analogs can produce such potent
depletions of polyamine levels as those described in literature
and predicted by our model can be due to synergic effects on
more thanone enzyme (87). This suggests that the combination
of targets in order to alter polyamine levels could be a useful
strategy to prevent polyamine-related diseases. However, from
a technical point of view, the number of conditions needed to
experimentally test the influence of n drugs is as high as 2
n
.
Furthermore, that huge number is correct considering only one
replica, variable, and concentration to be tested. The use of
mathematical models represents a helpful tool to restrict the
actual number of real experiments to only those producing
interesting results, according to the predictions based onmodel
simulations. Moreover, a deeper analysis concerning parameter
weights under different conditions could help us to identify
conditions and targets to control polyamine levels.
As we would expect from systems and synthetic biology
approaches, the predictions of this model could contribute to
propose new hypotheses to be experimentally tested. In turn,
the results of these experiments could contribute to the model
refinement. In fact, this has already occurred, since some of the
predictions of the core system model suggested that we had to
introduce some changes to improve it, as verified with our pro-
posed model. In this paper, we have proposed that alternative
behaviors of SSAT overexpression can depend on acetyl-
CoA availability, in agreement with experimental evidence
(95). We have also suggested an explanation for the observed
compatibility of ODC and SAMdc induction with polyamine
homeostasis.
Such in silico results open new hypotheses to be tested by
experimental work and can suggest predictions for conditions
where viable transgenic mice have not yet been obtained.
According to this, we suggest that SAM levels and acetyl-CoA
availability should be taken into account in future experimental
work to get a more complete view of this metabolism, as has
already been the case in a study of the effects of activated poly-
amine catabolismin an in vivo model of prostate cancer (95). As
more quantitative information about polyamine metabolism
becomes available, it can be incorporated into the model, and
further hypotheses can be tested. Of course, the model pre-
sented here is incomplete for many reasons, including those
simplifications assumed a priori and mentioned in the descrip-
tion of the model. However, the modularity of complex, hierar-
chical biological systems allows relatively simple extensions of a
metabolic model by the aggregation of additional modules,
such as the one suggested in relation to the role of S-adenosyl-
methionine. Additional features, such as the expected relevant
roles of polyamine uptake and detailed gene expression regula-
tory mechanisms would demand future modifications of the
present model.
AcknowledgmentsWe thank Dr. A. Pegg for useful suggestions dur-
ing manuscript preparation, Dr. C. Cotta for help during model con-
struction, and M. Canovas and V. Boucard-Mathey for language cor-
rections. Valuable information input was also received from other
member groups of European Cooperation in the field of Scientific and
Technical Research Action 922 (supported by the European Science
Foundation).
REFERENCES
1. Hartwell, L. H., Hopfield, J. J., Leibler, S., and Murray, A. W. (1999) Nature
402, C47C52
2. Alberghina, L., Chiaradonna, F., and Vanoni, M. (2004) Chembiochem 5,
13221333
3. Kitano, H. (2001) Foundations of Systems Biology, The MIT press, Lon-
don, UK
4. Kitano, H. (2002) Science 295, 16621664
5. Benner, S. A., and Sismour, A. M. (2005) Nat. Rev. Genet. 6, 533543
6. Kitano, H. (2002) Nature 420, 206210
Mammalian Polyamine MetabolismModel
21810 JOURNAL OF BIOLOGICAL CHEMISTRY VOLUME 281 NUMBER 31 AUGUST 4, 2006

a
t

U
N
I
V
E
R
S
I
D
A
D

D
E

M

L
A
G
A

-

T
r
i
a
l

S
u
b
s
c
r
i
p
t
i
o
n
,

o
n

J
a
n
u
a
r
y

7
,

2
0
1
0
w
w
w
.
j
b
c
.
o
r
g
D
o
w
n
l
o
a
d
e
d

f
r
o
m

http://www.jbc.org/content/suppl/2006/06/21/M602756200.DC1.html
Supplemental Material can be found at:
Ral Montaez Martnez
96
2.0
28.0
7. Weston, A. D., and Hood, L. (2004) J. Proteome Res. 3, 179196
8. Reed, M. C., Nijhout, H. F., Sparks, R., and Ulrich, C. M. (2004) J. Theor.
Biol. 226, 3343
9. Nijhout, H. F., Reed, M. C., Budu, P., and Ulrich, C. M. (2004) J. Biol. Chem.
279, 5500855016
10. Yang, C. R., Shapiro, B. E., Hung, S. P., Mjolsness, E. D., and Hatfield, G. W.
(2005) J. Biol. Chem. 280, 1122411232
11. Tabor, C. W., and Tabor, H. (1984) Annu. Rev. Biochem. 53, 749790
12. Heby, O., and Persson, L. (1990) Trends Biochem. Sci. 15, 153158
13. Urdiales, J. L., Medina, M. A., and Sanchez-Jimenez, F. (2001) Eur. J. Gas-
troenterol. Hepatol. 13, 10151019
14. Medina, M. A., Urdiales, J. L., Rodriguez-Caso, C., Ramirez, F. J., and
Sanchez-Jimenez, F. (2003) Crit. Rev. Biochem. Mol. Biol. 38, 2359
15. Medina, M. A., Correa-Fiz, F., Rodriguez-Caso, C., and Sanchez-Jime-
nez, F. (2005) J. Cell Mol. Med. 9, 854864
16. Cohen, S. S. (1998) A Guide of Polyamines, Oxford University Press, New
York
17. Ruiz-Chica, J., Medina, M. A., Sanchez-Jimenez, F., and Ramirez, F. J.
(2001) Biophys. J. 80, 443454
18. Brandes, L. J., Queen, G. M., and LaBella, F. S. (1998) J. Cell. Biochem. 69,
233243
19. Lesort, M., Tucholski, J., Miller, M. L., and Johnson, G. V. (2000) Prog.
Neurobiol. 61, 439463
20. Esposito, C., and Caputo, I. (2005) FEBS J. 272, 615631
21. Seiler, N. (2005) Pharmacol. Ther. 107, 99119
22. Wallace, H. M., Fraser, A. V., and Hughes, A. (2003) Biochem. J. 376, 114
23. Watanabe, S., Kusama-Eguchi, K., Kobayashi, H., and Igarashi, K. (1991)
J. Biol. Chem. 266, 2080320809
24. Igarashi, K., and Kashiwagi, K. (2000) Biochem. Biophys. Res. Commun.
271, 559564
25. Wang, Y., Xiao, L., Thiagalingam, A., Nelkin, B. D., and Casero, R. A., Jr.
(1998) J. Biol. Chem. 273, 3462334630
26. Wang, Y., Devereux, W., Woster, P. M., Stewart, T. M., Hacker, A., and
Casero, R. A., Jr. (2001) Cancer Res. 61, 53705373
27. Davis, R. H. (1990) J. Cell. Biochem. 44, 199205
28. Coffino, P. (2000) Proc. Natl. Acad. Sci. U. S. A. 97, 44214423
29. Thomas, T., and Thomas, T. J. (2001) Cell Mol. Life Sci. 58, 244258
30. Pietila, M., Pirinen, E., Keskitalo, S., Juutinen, S., Pasonen-Seppanen, S.,
Keinanen, T., Alhonen, L., and Janne, J. (2005) J. Invest. Dermatol. 124,
596601
31. Gerner, E. W., and Meyskens, F. L., Jr. (2004) Nat. Rev. Cancer 4, 781792
32. Coffino, P., and Poznanski, A. (1991) J. Cell. Biochem. 45, 5458
33. Nilsson, J. A., Keller, U. B., Baudino, T. A., Yang, C., Norton, S., Old, J. A.,
Nilsson, L. M., Neale, G., Kramer, D. L., Porter, C. W., and Cleveland, J. L.
(2005) Cancer Cell 7, 433444
34. Kramer, D. L., Vujcic, S., Diegelman, P., Alderfer, J., Miller, J. T., Black,
J. D., Bergeron, R. J., and Porter, C. W. (1999) Cancer Res. 59, 12781286
35. Bachrach, U., Wang, Y. C., and Tabib, A. (2001) News Physiol. Sci. 16,
106109
36. Kutuzov, M. A., Andreeva, A. V., and Voyno-Yasenetskaya, T. A. (2005)
J. Biol. Chem. 280, 2538825395
37. Hu, X., Washington, S., Verderame, M. F., and Manni, A. (2005) Cancer
Res. 65, 1102611033
38. Pegg, A. E., Feith, D. J., Fong, L. Y., Coleman, C. S., OBrien, T. G., and
Shantz, L. M. (2003) Biochem. Soc. Trans. 31, 356360
39. Pegg, A. E. (2006) J. Biol. Chem. 281, 1452914532
40. Huang, Y., Pledgie, A., Casero, R. A., Jr., and Davidson, N. E. (2005) Anti-
cancer Drugs 16, 229241
41. Childs, A. C., Mehta, D. J., and Gerner, E. W. (2003) Cell Mol. Life Sci. 60,
13941406
42. Wallace, H. M., and Mackarel, A. J. (1998) Biochem. Soc. Trans. 26,
571575
43. Seiler, N. (2004) Amino Acids 26, 217233
44. Seiler, N., Delcros, J. G., and Moulinoux, J. P. (1996) Int. J. Biochem. Cell
Biol. 28, 843861
45. Ivanov, I. P., Matsufuji, S., Murakami, Y., Gesteland, R. F., and Atkins, J. F.
(2000) EMBO J. 19, 19071917
46. Murakami, Y., Matsufuji, S., Hayashi, S., Tanahashi, N., and Tanaka, K.
(2000) Biochem. Biophys. Res. Commun. 267, 16
47. Raina, A., Hyvonen, T., Eloranta, T., Voutilainen, M., Samejima, K., and
Yamanoha, B. (1984) Biochem. J. 219, 9911000
48. Pajula, R. L. (1983) Biochem. J. 215, 669676
49. Heljasvaara, R., Veress, I., Halmekyto, M., Alhonen, L., Janne, J., Laajala, P.,
and Pajunen, A. (1997) Biochem. J. 323, 457462
50. Vujcic, S., Liang, P., Diegelman, P., Kramer, D. L., and Porter, C. W. (2003)
Biochem. J. 370, 1928
51. Alhonen, L., Karppinen, A., Uusi-Oukari, M., Vujcic, S., Korhonen, V. P.,
Halmekyto, M., Kramer, D. L., Hines, R., Janne, J., and Porter, C. W. (1998)
J. Biol. Chem. 273, 19641969
52. Shappell, N. W., Fogel-Petrovic, M. F., and Porter, C. W. (1993) FEBS Lett.
321, 179183
53. Fogel-Petrovic, M., Shappell, N. W., Bergeron, R. J., and Porter, C. W.
(1993) J. Biol. Chem. 268, 1911819125
54. Shantz, L. M., Holm, I., Janne, O. A., and Pegg, A. E. (1992) Biochem. J. 288,
511518
55. Bogle, R. G., Mann, G. E., Pearson, J. D., and Morgan, D. M. L. (1994) Am. J.
Physiol. 284, C776C783
56. Pegg, A. E., McGovern, K. A., and Wiest, L. (1987) Biochem. J. 241,
305307
57. Seppanen, P., Fagerstrom, R., Alhonen-Hongisto, L., Elo, H., Lumme, P.,
and Janne, J. (1984) Biochem. J. 221, 483488
58. Korhonen, V. P., Niiranen, K., Halmekyto, M., Pietila, M., Diegelman, P.,
Parkkinen, J. J., Eloranta, T., Porter, C. W., Alhonen, L., and Janne, J. (2001)
Mol. Pharmacol. 59, 231238
59. Berntsson, P. S., Alm, K., and Oredsson, S. M. (1999) Biochem. Biophys.
Res. Commun. 263, 1316
60. Parry, L., Balana Fouce, R., and Pegg, A. E. (1995) Biochem. J. 305, 451458
61. Fell, D. (1997) in Understanding the Control of Metabolism (Snell, K., ed)
Portland Press, London
62. Savageau, M. A. (1971) Nature 229, 542544
63. Savageau, M. A., Kotre, A. M., and Sakamoto, N. (1972) Biochem. Biophys.
Res. Commun. 48, 4147
64. Morohashi, M., Winn, A. E., Borisuk, M. T., Bolouri, H., Doyle, J., and
Kitano, H. (2002) J. Theor. Biol. 216, 1930
65. Auvinen, M., Laine, A., Paasinen-Sohns, A., Kangas, A., Kangas, L.,
Saksela, O., Andersson, L. C., and Holtta, E. (1997) Cancer Res. 57,
30163025
66. Moshier, J. A., Dosescu, J., Skunca, M., and Luk, G. D. (1993) Cancer Res.
53, 26182622
67. Pegg, A. E. (1988) Cancer Res. 48, 759774
68. Thomas, T., and Thomas, T. J. (2003) J. Cell Mol. Med. 7, 113126
69. Pendeville, H., Carpino, N., Marine, J. C., Takahashi, Y., Muller, M., Mar-
tial, J. A., and Cleveland, J. L. (2001) Mol. Cell. Biol. 21, 65496558
70. Janne, J., Alhonen, L., Pietila, M., and Keinanen, T. A. (2004) Eur. J. Bio-
chem. 271, 877894
71. Wallace, H. M., and Fraser, A. V. (2004) Amino Acids 26, 353365
72. Halmekyto, M., Alhonen, L., Alakuijala, L., and Janne, J. (1993) Bio-
chem. J. 291, 505508
73. Halmekyto, M., Alhonen, L., Wahlfors, J., Sinervirta, R., Eloranta, T., and
Janne, J. (1991) Biochem. J. 278, 895898
74. Mitchell, J. L., Simkus, C. L., Thane, T. K., Tokarz, P., Bonar, M. M.,
Frydman, B., Valasinas, A. L., Reddy, V. K., and Marton, L. J. (2004) Bio-
chem. J. 384, 271279
75. Bettuzzi, S., Davalli, P., Astancolle, S., Pinna, C., Roncaglia, R., Boraldi, F.,
Tiozzo, R., Sharrard, M., and Corti, A. (1999) FEBS Lett. 446, 1822
76. Nishimura, K., Nakatsu, F., Kashiwagi, K., Ohno, H., Saito, T., andIgarashi,
K. (2002) Genes Cells 7, 4147
77. Nisenberg, O., Pegg, A. E., Welsh, P. A., Keefer, K., and Shantz, L. M.
(2006) Biochem. J. 393, 295302
78. Kramer, D., Mett, H., Evans, A., Regenass, U., Diegelman, P., and Porter,
C. W. (1995) J. Biol. Chem. 270, 21242132
79. Suzuki, T., Sadakata, Y., Kashiwagi, K., Hoshino, K., Kakinuma, Y., Shira-
hata, A., and Igarashi, K. (1993) Eur. J. Biochem. 215, 247253
80. Manni, A., Badger, B., Grove, R., Kunselman, S., and Demers, L. (1995)
Cancer Lett. 95, 2328
81. Ferioli, M. E., Bottone, M. G., Soldani, C., and Pellicciari, C. (2004) Cell
Mammalian Polyamine MetabolismModel
AUGUST 4, 2006 VOLUME 281 NUMBER 31 JOURNAL OF BIOLOGICAL CHEMISTRY 21811

a
t

U
N
I
V
E
R
S
I
D
A
D

D
E

M

L
A
G
A

-

T
r
i
a
l

S
u
b
s
c
r
i
p
t
i
o
n
,

o
n

J
a
n
u
a
r
y

7
,

2
0
1
0
w
w
w
.
j
b
c
.
o
r
g
D
o
w
n
l
o
a
d
e
d

f
r
o
m

http://www.jbc.org/content/suppl/2006/06/21/M602756200.DC1.html
Supplemental Material can be found at:
97
Captulo 3
2.0
28.0
Mol. Life Sci. 61, 27672773
82. Casero, R. A., Jr., and Pegg, A. E. (1993) FASEB J. 7, 653661
83. Fogel-Petrovic, M., Vujcic, S., Miller, J., and Porter, C. W. (1996) FEBS
Lett. 391, 8994
84. Fogel-Petrovic, M., Vujcic, S., Brown, P. J., Haddox, M. K., and Porter,
C. W. (1996) Biochemistry 35, 1443614444
85. Pietila, M., Alhonen, L., Halmekyto, M., Kanter, P., Janne, J., and Porter,
C. W. (1997) J. Biol. Chem. 272, 1874618751
86. Suppola, S., Heikkinen, S., Parkkinen, J. J., Uusi-Oukari, M., Korhonen,
V. P., Keinanen, T., Alhonen, L., and Janne, J. (2001) Biochem. J. 358,
343348
87. Alhonen, L., Pietila, M., Halmekyto, M., Kramer, D. L., Janne, J., and Por-
ter, C. W. (1999) Mol. Pharmacol. 55, 693698
88. Coleman, C. S., Pegg, A. E., Megosh, L. C., Guo, Y., Sawicki, J. A., and
OBrien, T. G. (2002) Carcinogenesis 23, 359364
89. Tucker, J. M., Murphy, J. T., Kisiel, N., Diegelman, P., Barbour, K. W.,
Davis, C., Medda, M., Alhonen, L., Janne, J., Kramer, D. L., Porter, C. W.,
and Berger, F. G. (2005) Cancer Res. 65, 53905398
90. Pegg, A. E., Wechter, R., Pakala, R., and Bergeron, R. J. (1989) J. Biol. Chem.
264, 1174411749
91. Schipper, R. G., Deli, G., Deloyer, P., Lange, W. P., Schalken, J. A., and
Verhofstad, A. A. (2000) Prostate 44, 313321
92. Hector, S., Porter, C. W., Kramer, D. L., Clark, K., Prey, J., Kisiel, N.,
Diegelman, P., Chen, Y., and Pendyala, L. (2004) Mol. Cancer Ther. 3,
813822
93. Ravasz, E., Somera, A. L., Mongru, D. A., Oltvai, Z. N., and Barabasi, A. L.
(2002) Science 297, 15511555
94. Edwards, J. S., Ibarra, R. U., and Palsson, B. O. (2001) Nat. Biotechnol. 19,
125130
95. Kee, K., Foster, B. A., Merali, S., Kramer, D. L., Hensen, M. L., Diegelman,
P., Kisiel, N., Vujcic, S., Mazurchuk, R. V., and Porter, C. W. (2004) J. Biol.
Chem. 279, 4007640083
96. Pegg, A. E., and Poso, H. (1983) Methods Enzymol. 94, 234239
97. Ragione, F. D., and Pegg, A. E. (1982) Biochemistry 21, 61526158
98. Martinov, M. V., Vitvitsky, V. M., Mosharov, E. V., Banerjee, R., and
Ataullakhanov, F. I. (2000) J. Theor. Biol. 204, 521532
Mammalian Polyamine MetabolismModel
21812 JOURNAL OF BIOLOGICAL CHEMISTRY VOLUME 281 NUMBER 31 AUGUST 4, 2006

a
t

U
N
I
V
E
R
S
I
D
A
D

D
E

M

L
A
G
A

-

T
r
i
a
l

S
u
b
s
c
r
i
p
t
i
o
n
,

o
n

J
a
n
u
a
r
y

7
,

2
0
1
0
w
w
w
.
j
b
c
.
o
r
g
D
o
w
n
l
o
a
d
e
d

f
r
o
m

http://www.jbc.org/content/suppl/2006/06/21/M602756200.DC1.html
Supplemental Material can be found at:
Ral Montaez Martnez
98
2.0
28.0
18
SUPPLEMENTARY MATERIAL
FOR METHODS SECTION
Details on the rate equations for SSAT and PAO.
Polyamine oxidase (PAO) and spermidine/spermine N
1
-acetyltransferase (SSAT) are able to use two
different, alternative substrates (see Table 1for final expressions). These equations are derived from standard
Michaelis equations. PAO can use either acetyl-spermidine (aD) or acetyl-spermine (aS) as a substrate.
Therefore, we can state:
v
PAOaD
=k
cat
aD
PAO aD | | [1]
v
PAOaS
=k
cat
aS
PAO aS | | [2]
PAO | |
T
= PAO | |+ PAO aD | |+ PAO aS | |+ PAO D | |+ PAO S | | [3]
and, from these equations, assuming that each substrate behaves as a competititve inhibitor for the reaction
involving the other substrate, we can obtain
v
PAOaD
=
V
max
PAO
aD | |
K
MaD
PAO
1+
aD | |
K
MaD
PAO
+
aS | |
K
MaS
PAO
+
D | |
K
MD
PAO
+
S | |
K
MS
PAO
|
\

|
.
|
[4]
v
PAOaS
=
V
max
PAO
aS | |
K
MaS
PAO
1+
aD | |
K
MaD
PAO
+
aS | |
K
MaS
PAO
+
D | |
K
MD
PAO
+
S | |
K
MS
PAO
|
\

|
.
|
[5]
The equations for SSAT can be deduced in a similar way. Since SSSAT uses acetyl-CoA as a cosubstrate, in
this case the simplest kinetic model for bisubstrate enzymes was assumed. We also considered the efferent
efficiency shown by SSAT to catalyze an acetyl group transfer from acetyl-CoA to either spermidine or
spermine by introducing a correction factor C (see Table 1), in agreement with experimental observations (97).
Details on the differential equations for time-dependent variables.
Three of the enzymes included in the model (namely, ODC, SAMdc, and SSAT) are among the
enzymes with the shortest half-lives in mammals (13,22). Furthermore, the actual activities of these three
enzymes and the actual contents of the key regulator of ODC degradation (antizyme) are regulated by the
levels of polyamines (52). Therefore, in our model ODC, SAMdc and SSAT activities, and antizyme
concentration are treated as time-dependent variables. Their respective differential equations (Table 2) are
derived by using a modified Schimke equation. Schimke equation [6] describes the time-dependent
concentration of an enzyme as a function of its synthesis and degradation rates (83,99-101). Synthesis is
considered to follow zero order kinetics (k
S
) and degradation is linearly dependent on enzyme concentration
([Ez]), where k
D
represent the kinetic constant for degradation,
d Ez | |
dt
=k
S
k
D
Ez | | [6]
Multipliying both terms of equation [6] by the catalytic constant of the enzyme (k
cat
) we obtain the Schimke
equation as a function of V
max
,
dV
max
dt
=k
S
k
cat
k
D
V
max
[7]
This modification allows us to modify directly the activity during simulation. Equation [7] was used for ODC,
SSAT and SAMdc but not for antizyme. Antizyme cannot be formally considered as an enzyme, at least in our
model, since its ODC degradation kinetics was not defined by a Michaelis-Menten or saturable behavior.
99
Captulo 3
2.0
28.0
19
Therefore, equation [6] was used for the time-dependent antizyme concentration in our model. On the other
hand, since antizyme is a key regulator of ODC degradation, it affects the degradation term of the Schimke
equation for ODC in our model:
dV
max
ODC
dt
=k
S
ODC
k
cat
ODC
k
D
ODC
V
max
k
Antz
Antz > @ [8]
Equations [7] and [8] can be simplified, as follows,
dV
max
dt
=k'
S
k
D
V
max
[9]
dV
max
ODC
dt
=k'
S
ODC
k'
D
ODC
V
max
Antz > @ [10]
To model the polyamine effect on the modulation of the synthesis and degradation rates of ODC,
SAMdc, SSAT and antizyme, we assumed that these effects are due to binding. We also assumed that free
polyamines are in a rapid equilibrium with polyamines non-covalently bound to macromolecules. Therefore,
changes in free polyamine levels would alter the macromolecule-bound pool of polyamines, and this could be
used as a modulator of those k
S
and k
D
according to data available in bibliography (23).
Putrescine is not considered as such a modulator, since its binding to macromolecules is much weaker
than those of spermidine and spermine (24)According to this, an equilibrium
Pol > @+ M > @ Pol M > @ [11]
can be considered, where Pol represents free polyamines (spermine +spermidine), PolM is polyamine non-
covalently bound to macromolecules, and M means free macromolecules.
The equilibrium constant for [11] is defined as
K
eq
=
Pol M > @
Pol > @ M > @
[12]
Taking the total macromolecule pool as a constant, we can write:
M > @
T
= M > @+ Pol M > @ [13]
From [12] and [13], we get equations [14] and [15]:
M > @
M > @
T
=
1
1+K
eq
Pol > @
[14]
Pol M > @
M > @
T
=1
1
1+K
eq
Pol > @
[15]
An increase of polyamines is described to inhibit both ODC and SAMdc synthesis; therefore, in our
model k
S
values for these enzymes were affected by equation [14]. On the other hand, an increase of
polyamines favors processes governed by k
S
and k
D
for SSAT and k
S
for antizyme; thus, in our model, their
values were affected by equation [15].
Sensitivity analysis of the model.
Sensitivity analysis was done according to its definition (61):
S
Par
Var
=
wVar
wPar

Par
Var
[16]
where, for steady state conditions, Var is a time-dependent variable and Par is a parameter. Numerically, we
applied finite increments,
Ral Montaez Martnez
100
2.0
28.0
20
S
Par
Var
=
'Var
Var
'Par
Par
=
Var
1
Var
0

Var
0
Par
1
Par
0

Par
0
[17]
Negative coefficients indicate a decrease in the variable in response to an increase in the parameter.
According to this definition, a parameter with high sensitivity coefficients for all the variables would point an
important controlling element for the analyzed steady state.
Simulations progressed until the steady state acquisition for a relative change of each parameter of the
model was equal to 0.001. That is,
Par
1
Par
0

Par
0
0.001 [18]
The initial steady state chosen for this analysis was the state described as basal condition in Results
section.
FOR RESULTS SECTION
Results from the sensitivity analysis of the model.
The sensitivity analysis for all the parameters and variables of our basic model represented a
combination of 46 parameters and 11 time-dependent variables, that is, a total of 506 sensitivity coefficients.
All of them were within the (1.05, 1.06) range, whereas 91% remained within the narrower (-0.5, 0.5) range,
and 61% of the coefficients had values within the (-0.1, 0.1) range. Table S2 summarizes the whole analysis,
where it can be observed that the most influential parameters are those related to the synthesis and degradation
of the enzymes ODC, SAMdc, SSAT and antizyme. This is in complete agreement with their recognized
regulatory role in polyamine metabolism (22,68). Furthermore, from data in Table S2, it is also possible to
extract some noteworthy aspects that could indicate their influence in the control of this metabolism. Antizyme
and ODC exhibited opposite coefficients since antizyme decreased ODC levels with a sensitivity coefficient
close to -1. Furthermore, ODC and antizyme caused remarkable effects in putrescine, while the changes were
much lower for spermine and irrelevant for spermidine. This agrees with the experimental observation that
ODC overexpressing mice exhibited a putrescine overproduction without affecting spermine and spermidine
levels (38).
On the other hand, SAMdc and SSAT synthesis and degradation constants had remarkable effects on
all the time-dependent variables. Therefore, we could say that, at least in the basal condition, SAMdc and
SSAT were the most influential enzymes on the behavior of the whole system. This is noteworthy, since much
literature stresses the relevance of ODC, and not always that of SAMdc and SSAT.
Changes in PAO activity had little effect on the system. Spermidine and spermine synthases had only
a little effect on spermidine, but not on spermine. Efflux parameters had no important effect on the whole
system; in fact, putrescine level concentration was the only time-dependent variable that responded to its own
efflux, as it could be expected and in contrast with the lack of response of acetylated spermidine to changes in
its efflux parameter.
101
Captulo 3
2.0
28.0
REFENCES FOR SUPPLEMENTAL MATERIAL
99. Doherty, F. J., and Mayer, R. J. (1992) Intracellular protein degradation, IRL Press,
New York
100. Ruan, H., Shantz, L. M., Pegg, A. E., and Morris, D. R. (1996) J Biol Chem 271,
29576-29582
101. Howard, M. T., Shirts, B. H., Zhou, J., Carlson, C. L., Matsufuji, S., Gesteland, R. F.,
Weeks, R. S., and Atkins, J. F. (2001) Genes Cells 6, 931-941
102. Marquez, J., Mates, J. M., Quesada, A. R., Medina, M. A., Nunez de Castro, I., and
Sanchez-Jimenez, F. (1989) Life Sci 45, 1877-1884
103. Baydoun, A. R., Emery, P. W., Pearson, J. D., and Mann, G. E. (1990) Biochem
Biophys Res Commun 173, 940-948
104. Cynober, L. A. (2002) Nutrition 18, 761-766
105. Albe, K. R., Butler, M. H., and Wright, B. E. (1990) J Theor Biol 143, 163-195.
106. Kitani, T., and Fujisawa, H. (1984) Biochim Biophys Acta 784, 164-167
107. Flamigni, F., Guarnieri, C., and Caldarera, C. M. (1984) Biochim Biophys Acta 802,
245-252
108. Ikeguchi, Y., Wang, X., McCloskey, D. E., Coleman, C. S., Nelson, P., Hu, G., Shantz,
L. M., and Pegg, A. E. (2004) Biochem J 381, 701-707
109. Seiler, N. (1990) Digestion 46 Suppl 2, 319-330
110. Samejima, K., Raina, A., Yamanoha, B., and Eloranta, T. (1983) Methods Enzymol 94,
270-276
111. Kajander, E. O., Kauppinen, L. I., Pajula, R. L., Karkola, K., and Eloranta, T. O. (1989)
Biochem J 259, 879-886
112. Hibasami, H., Kawase, M., Tsukada, T., Maekawa, S., Sakurai, M., and Nakashima, K.
(1988) FEBS Lett 229, 243-246
113. Samejima, K., and Yamanoha, B. (1982) Arch Biochem Biophys 216, 213-222
114. Pajula, R. L., Raina, A., and Eloranta, T. (1979) Eur J Biochem 101, 619-626.
115. Raina, A., Pajula, R. L., and Eloranta, T. (1983) Methods Enzymol 94, 276-279
116. Hannonen, P., Janne, J., and Raina, A. (1972) Biochim Biophys Acta 289, 225-231
117. Pegg, A. E., Shuttleworth, K., and Hibasami, H. (1981) Biochem J 197, 315-320.
118. Baillon, J. G., Mamont, P. S., Wagner, J., Gerhart, F., and Lux, P. (1988) Eur J
Biochem 176, 237-242
119. Tunici, P., Sessa, A., Rabellotti, E., Grant, G., Bardocz, S., and Perin, A. (1999)
Biochim Biophys Acta 1428, 219-224
120. Holtta, E. (1983) Methods Enzymol 94, 306-311
121. Coleman, C. S., and Pegg, A. E. (2001) Biochem J 358, 137-145
122. Bras, A. P., Janne, J., Porter, C. W., and Sitar, D. S. (2001) Drug Metab Dispos 29,
676-680
123. Sakai, T., Hori, C., Kano, K., and Oka, T. (1979) Biochemistry 18, 5541-5548.
124. Feldman, M. J., Levy, C. C., and Russell, D. H. (1972) Biochemistry 11, 671-677
125. Poso, H., and Pegg, A. E. (1982) Biochemistry 21, 3116-3122
126. Suppola, S., Pietila, M., Parkkinen, J. J., Korhonen, V. P., Alhonen, L., Halmekyto, M.,
Porter, C. W., and Janne, J. (1999) Biochem J 338 ( Pt 2), 311-316
127. Pegg, A. E., and Bennett, R. A. (1983) Methods Enzymol 94, 69-72
Ral Montaez Martnez
102
2.0
28.0
21
Table S1. Values chosen for simulations in silico and their respective experimental data available in literature.
Description Symbol In silico Experimental range References
Orn > @ 300 M 220-410 M (102,103)
Met > @ 50 M 25-200 M (8,102,104)
AcCoA > @ 39.5 M 9.6 - 39 M (105)
Metabolite parameters
CoA > @ 160 M 43 - 195 M (105)
M
ODC
K 60 M 60 M (106) ODC kinetic constants
ip
ODC
K 1300 M 1300 M (107)
d
ODC
k 0.05 M
-1
min
-1
-- Tuned ODC synthesis and
degradation constants
s
ODC
k 5 M
-1
min
-2
-- Tuned
d
Antz
k 0.02 min
-1
-- Tuned Antz synthesis and
degradation constants
s
Antz
k 0.02 M
-1
min
-1
-- Tuned
max
SpdS
V 10.95 M min
-1
0.43-10.95 M min
-1
(58,108,109)
A
SpdS
K 0.3 M 0.3-1100 M (47,110-112)
ia
SpdS
K 0.8 M 0.8 M (47)
P
SpdS
K 40 M 30-100 M (47,110-113)
SpdS kinetic constants
iD
SpdS
K 100 M 100 M (47)
max
SpmS
V 3.23 M min
-1
0-42 M min
-1
(58,108,109)
A
SpmS
K 0.1 M 0.1-5 M (48,111,114-116)
ia
SpmS
K 0.06 M 0.6 M (48)
D
SpmS
K 60 M 6-200 M (48,111,114-118)
SpmS kinetic constants
iS
SpmS
K 25 M 10-50 M (48)
max
PAO
V 10.35 M 0.5-10.35 M min
-1
(109,119)
MaS
PAO
K 0.6 M 0.6 M (120)
MaD
PAO
K 14 M 10-50 M (120)
MS
PAO
K 15 M 10-20 M (26,120)
PAO kinetic constants
MD
PAO
K 50 M 10-50 M (120)
MD
SSAT
K 130 M 50-260 M (60,97,121,122) SSAT kinetic constants
MS
SSAT
K 35 M 20-35 M (60,97)
d
SSAT
k 0.2 min
-1
-- Tuned SSAT synthesis and
degradation constants
s
SSAT
k' 0,001 M
-1
min
-2
-- Tuned
is
SAMdc
K 500 M 500 M (123)
M
SAMdc
K 50 M 20-100 M (96,123-125)
aP
SAMdc
K 0.5 M 0.5 M (123)
SAMdc kinetic constants
iA
SAMdc
K 2.5 M 1-5 M (96)
d
SAMdc
k 0.02 min
-1
-- Tuned SAMdc synthesis and
degradation constants
s
SAMdc
k 1 M
-1
min
-2
-- Tuned
P
efflux
k 0.01 min
-1
-- Tuned Efflux parameters
aD
efflux
k 0.01 min
-1
-- Tuned
Acetyl-CoA/CoA recycling
acCoA k
R min
-1
-- Tuned
CoA k
3R min
-1
-- Tuned
R 0.004 -- Tuned
eq

K
1 M
-1
-- Tuned
Other parameters (*)
C 4.44 -- (97)
P > @ 0.01 M 0-910 M (49,51,52,58,85,86,126)
D > @ 0.01 M 0-1700 M (49,51,52,58,85,86,126)
S > @ 0.01 M 0-1040 M (49,51,52,58,85,86,126)
Polyamine and S-adenosyl-
L-methionine
concentrations (initial
values) SAM > @ 0.01 M 20-100 M (98,127)
Values considered for simulations in silico are taken from bold references.
(*)
eq
means the parameter values for , , and . The correction factor "C" was explained in
Table 1.

K eq
ODC
K eq
SAMdc
K eq
SSAT
K eq
Antz
K
103
Captulo 3
2.0
28.0
22
Table S2. Sensitivity analysis for parameters and time-dependent variables.
Time-dependent variables
Parameters
max
ODC
V
Antz
max
SSAT
V max
SAMdc
V
P > @ D > @ S > @ SAM > @ A > @ aD > @ aS > @
Orn > @ -- -- -- -- 0.19 -- -- -- -- -- --
Met > @ -0.34 -- 0.24 -0.24 -0.34 0.19 0.3 1.11 0.49 0.44 0.57 Metabolite parameters
acCoA CoA > @ -- -- -- -- -- -- -- -- -- -- --
M
ODC
K -- -- -- -- -0.19 -- -- -- -- -- --
ODC kinetic constants
ip
ODC
K -- -- -- -- -- -- -- -- -- -- --
d
ODC
k' -1.05 -- -- -- -1.05 -- 0.21 -- 0.35 -- 0.27 ODC synthesis and
degradation constants
s
ODC
k' 1.06 -- -- -- 1.05 -- -0.21 -- -0.35 -- -0.27
d
Antz
k 1.06 -1.02 -- -- 1.05 -- -0.21 -- -0.35 -- -0.27 Antz synthesis and
degradation constants
s
Antz
k -1.05 1.02 -- -- -1.05 -- 0.21 -- 0.35 -- 0.27
max
SpdS
V 0.11 -- -- -- 0.11 0.18 -0.43 -0.15 -0.7 -- -0.54
A
SpdS
K -- -- -- -- -- -- 0.21 -- 0.34 -- 0.26
ia
SpdS
K -- -- -- -- -- -- 0.21 -- 0.35 -- 0.27
P
SpdS
K -- -- -- -- -- -- 0.22 -- 0.35 -- 0.27
SpdS kinetic constants
iD
SpdS
K -- -- -- -- -- -- -0.18 -- -0.3 -- -0.23
max
SpmS
V -0.12 -- -- -- -0.12 -0.19 0.44 0.16 -0.32 -- 0.55
A
SpmS
K -- -- -- -- -- 0.12 -0.29 -- 0.21 -- -0.36
ia
SpmS
K -- -- -- -- -- -- -0.14 -- -- -- -0.17
D
SpmS
K -- -- -- -- -- -- -0.14 -- -- -- -0.18
SpmS kinetic constants
iS
SpmS
K -- -- -- -- -- -0.13 0.3 0.11 -0.22 -- 0.37
max
PAO
V -- -- -- -- -- -- -- -- -- -0.93 -1.02
MaS
PAO
K -- -- -- -- -- -- -- -- -- -- 1.00
MaD
PAO
K -- -- -- -- -- -- -- -- -- 0.92 --
MS
PAO
K -- -- -- -- -- -- -- -- -- -0.55 -0.6
PAO kinetic constants
MD
PAO
K -- -- -- -- -- -- -- -- -- -0.22 -0.24
MD
SSAT
K -0.25 -- 0.18 -0.18 -0.25 0.40 -0.11 0.2 -- -0.23 0.27
MS
SSAT
K -- -- -- -- -- -0.25 0.34 -- -- 0.28 -0.12
MacCoA
SSAT
K
-0.14 -- -- -- -0.14 -- 0.12 0.13 -- -- --
SSAT kinetic constants
iCoA
SSAT
K
0.11 -- -- -- 0.11 -- -- -0.11 -- -- --
d
SSAT
k -0.85 0.26 -0.40 -0.60 -0.85 0.48 0.76 0.84 0.26 0.15 0.47 SSAT synthesis and
degradation constants
s
SSAT
k' 0.85 -0.26 0.40 0.60 0.85 -0.48 -0.76 -0.84 -0.26 -0.15 -0.47
is
SAMdc
K -- -- -- -- -- -- -- -- -- -- --
M
SAMdc
K 0.17 -- -0.12 0.12 0.17 -- -0.15 0.43 -0.25 -0.23 -0.29
aP
SAMdc
K -- -- -- -- -- -- -- -- -- -- --
SAMdc kinetic constants
iA
SAMdc
K -- -- -- -- -- -- -- -- -- -- --
d
SAMdc
k 0.35 -0.11 -0.25 -0.75 0.36 -0.20 -0.31 0.89 -0.51 -0.46 -0.59 SAMdc synthesis and
degradation constants
s
SAMdc
k' -0.35 0.11 0.25 0.75 -0.36 0.20 0.31 -0.88 0.51 0.46 0.59
P
efflux
k -- -- -- -- -1.04 -- 0.21 -- 0.35 -- 0.27
Efflux parameters
aD
efflux
k -- -- -- -- -- -- -- -- -- -- --
R -- -- -- -- -- -- -- -- -- -- --
acCoA k 0.17 -- -0.12 0.12 0.17 -- 0.15 -0.16 -- -- --
Acetyl-CoA/CoA
recycling
CoA k -0.12 -- -- -- -0.12 -- 0.11 0.12 -- -- --
eq
ODC
K -1.05 -- -- -- -1.04 -- 0.21 -- 0.35 -- 0.27
eq
SAMdc
K 0.35 -0.11 -0.25 -0.75 0.35 -0.2 -0.31 0.88 -0.51 -0.46 -0.59
eq
SSAT
K -- -- -- -- -- -- -- -- -- -- --
eq
Antz
K -0.45 0.43 -- -- 0.45 -- -- -- 0.15 -- 0.11
Other parameters
C -0.45 0.13 0.31 -0.31 -0.45 -- .073 0.48 0.18 0.35 0.12
For abbreviations, see Fig. 1.
Ral Montaez Martnez
104
2.0
28.0
105
Captulo 3
2.0
28.0
Amino Acids (2008) 34: 223229
DOI 10.1007/s00726-007-0502-7
Printed in The Netherlands
In silico analysis of arginine catabolism as a source of nitric oxide
or polyamines in endothelial cells
R. Montanez

, C. Rodr guez-Caso
;
, F. Sanchez-Jimenez, and M.

AA. Medina
Procel Group, Department of Molecular Biology and Biochemistry, University of Malaga, and CIBERER, Malaga, Spain
Received September 30, 2006
Accepted January 31, 2007
Published online May 23, 2007; # Springer-Verlag 2007
Summary. We use a modeling and simulation approach to carry out an
in silico analysis of the metabolic pathways involving arginine as a pre-
cursor of nitric oxide or polyamines in aorta endothelial cells. Our model
predicts conditions of physiological steady state, as well as the response of
the system to changes in the control parameter, external arginine concen-
tration. Metabolic ux control analysis allowed us to predict the values
of ux control coefcients for all the transporters and enzymes included in
the model. This analysis fullls the ux control coefcient summation
theorem and shows that both the low afnity transporter and arginase
share the control of the uxes through these metabolic pathways.
Keywords: Arginine Nitric oxide Ornithine Polyamines Meta-
bolic control analysis
Introduction
L-Arginine is a conditionally essential amino acid for
adult mammals, since it should be supplied by diet under
physiological or pathological conditions in which its re-
quirement exceeds its velocity of production. Its meta-
bolic relevance is stressed by the fact of being a precursor
for a wide variety of biomolecules playing very important
metabolic, regulatory, and physiological roles. Figure 1
summarizes the main metabolic pathways in which argi-
nine is involved. Exogenous arginine is taken up via sev-
eral isoforms of the y

system for cationic amino acids

(Grillo and Colombatto, 2004a). Arginine is the imme-
diate precursor of urea as a substrate of arginase in the
urea cycle. It also participates in the synthesis of creatine,
which in phosphorylated state plays a relevant role in
muscle bioenergetics. Arginase pathway bifurcates at the
level of ornithine, either leading to L-proline biosynthesis
(required for collagen synthesis) or to the polyamine syn-
thetic pathway. Arginine=ornithine-derived polyamines
are ubiquitous aliphatic polycations with pleiotropic
biological effects, including a relevant regulatory role in
macromolecular synthesis and cell proliferation rate
(Medina et al., 2003, 2005). Direct decarboxylation of
arginine could produce agmatine, which can act as a bio-
signaling molecule and as an intermediate in an alternative
pathway of polyamine synthesis (Grillo and Colombatto,
2004b; Morris, 2004). In the last fteen years, arginine
has attracted renewed attention, since it is the immediate
precursor of nitric oxide (NO) through the nitric oxide
synthase reaction. NO is a bio-active gas involved in
many processes, including neurotransmission, immune
response, vasodilation and adhesion of platelets and
leukocytes (Wu and Morris, 1998; Grillo and Colombatto,
2004a; Bronte and Zanovello, 2005). Therefore, arginine
metabolism is particularly relevant in physio-pathological
contexts such as the regulation of immune responses
(Bronte and Zanovello, 2005) and endothelial=vascular
health=disease (Cooke and Tsao, 1997; Gornik and
Creager, 2004).
In spite of its relevance, arginine catabolism is not fully
understood. A comprehensive understanding of arginine
metabolism biological functions should require new ho-
listic approaches, as claimed by the emergent systems bi-
ology (Kitano, 2002; Medina et al., 2005). Metabolic ux
control analysis is frequently used to access global regu-
latory features of whole metabolic pathways (Fell, 1997).
On the other hand, mathematical modeling and simulation
could provide an operational way to explore in silico these

Both authors equally contributed to this work.

Current address: ICREA-Complex Systems Laboratory, Universitat

Pompeu Fabra, Barcelona, Spain.
Ral Montaez Martnez
106
2.0
28.0
global regulatory features, as shown by relevant mathe-
matical models of several well-known metabolic path-
ways published recently (Nijhout et al., 2004; Reed et al.,
2004; Yang et al., 2005). In this work, we describe the
modeling and in silico analysis of the branch of arginine
metabolism leading to either NO or polyamines in aorta
endothelial cells (see Fig. 2), based on known enzymol-
ogy and available, experimentally measured intermediate
metabolite concentrations and kinetic data on the compo-
nents included in the model. The model allows for the use
of metabolic ux control analysis to determine the relative
contribution of the different components to the regulation
of the ux through the system.
Materials and methods
Description of the model
We delimited our system of study as the portion of arginine metabolism
leading from external arginine to either NO or putrescine (the rst argi-
Fig. 1. Overview of arginine catabolism. ADC Arginine decarboxylase;
AL argininesuccinate lyase; AS argininesuccinate synthase; carbamoyl-
P carbamoyl-phosphate; CAT-1 high afnity arginine transporter;
CAT-2a low afnity arginine transporter; CPS-I carbamoyl-phosphate
synthetase I; NO nitric oxide; NOS nitric oxide synthase; OAT or-
nithine aminetransferase; ODC ornithine decarboxylase; P5C pyrro-
line-5-carboxylate
Fig. 2. Simplied model of the catabolic branch of arginine as a pre-
cursor of NO and polyamines, including described inhibitions and re-
lease. See text for details
Table 1. Rate equations for enzymes and transports included in the model
Equations and parameters Description
Arginine transport
tranp
Argex
K
Hat
m
Argex

V
Hat
max
1

Orn
K
Hat
iO

Argin
K
Hat
m

Argex
K
Lat
m
Argex

V
Lat
max
1

Orn
K
Hat
iO

Argin
K
Lat
m
Equation (with parameters
taken from Brenda enzyme
database
a
) that assembles the
activities of the high and low
activity transporters
Arginase
Arg
V
Arg
max
Argin
K
Arg
m

Orn
K
Arg
iO

Argin
Michaelis-Menten equation
for competitive inhibition
by ornithine
ODC
ODC
V
ODC
max
Orn
K
ODC
M
Orn
Michaelis-Menten equation
NOS
NOS1
V
NOS1
max
Argin
K
NOS1
m
Argin
Michaelis-Menten equation
Ornithine efux
effluxO
V
effl:Hat
max
1

Arg
ex

K
Hat
m

Orn
K
Hat
iOrn

Arg
in

K
Hat
m

Orn

V
effl:Lat
max
1

Arg
ex

K
Lat
m

Orn
K
effl:Lat
m

Arg
in

K
Lat
m

Orn
Equation (with parameters
taken from Brenda enzyme
database
a
) that assembles the
activities of the high and low
activity transporters
a
Source: www.brenda.uni-koeln.de
224 R. Montanez et al.
107
Captulo 3
2.0
28.0
nine=ornithine-derived polyamine). Due to the relevance of arginine me-
tabolism to endothelial function and to the availability of a relatively
homogenous set of experimental data obtained using aorta endothelial
cells, we adopted our model to the known enzymology of arginine metab-
olism in endothelial cells. Two transporters of the y

family are con-

sidered to model arginine uptake and ornithine efux: the high afnity
CAT-1 and the low afnity CAT-2a transport systems (Mann et al., 2003;
Grillo and Colombatto, 2004a; Verrey et al., 2004; Rotoli et al., 2005).
Three enzymes are included in the model: arginase, which splits arginine
to urea plus ornithine; ornithine decarboxylase, which decarboxylates
ornithine to putrescine; and nitric oxide synthase, wihich splits arginine
to citrullin plus NO. Although endothelial cells express two arginase
isozymes, mitochondrial arginase II is not included in the model, since
aorta endothelial cells only express it as a response to lipopolysaccharide
activation (Buga et al., 1996). Basic described inhibitions are also taken
into account in the model, including inhibition of both transporters by both
internal arginine and ornithine, and inhibition of arginase by ornithine. For
all the transporters and enzymes included in the model, a very simple
Michaelian kinetic behavior is considered.
Since aorta endothelial cells from different sources are easily cultured
and components of culture media can be controlled, we used external
arginine concentration as the control parameter of the system. Table 1
shows the rate equations for enzymes and transporters included in the
model. Arginine uptake and ornithine efux velocities are considered to be
the addition of velocities through both the high and low afnity transport-
ers. Table 2 shows the differential equations for the two time-dependent
variables (internal arginine and ornithine) considered in the model. Table 3
shows the range of experimentally-determined values described in the
literature for the different kinetic parameters of the model and the values
selected to be included in the model. The values of the parameters describ-
ing ornithine efux are considered to be equal to those describing arginine
uptake. The selected external concentration of arginine is that contained in
DMEM culture medium, frequently used to culture endothelial cells
(Baydoun et al., 1990).
The models were implemented in Perl language (http:==www.perl.
com=). Simulations were carried out by the iterarative Euler method, with
a time increment of 0.01min per iteration. In all the simulations, steady
state was considered to be reached when the change of any variable was
lower than 10
10
within an interval of 3 h (18000 iterations). To simulate
the response of the system to changes in the control parameter, different
simulations were carried out changing the values of external arginine con-
centration within a four order of magnitude range (from 10mM to 10mM),
as experimentally carried out (Arnal et al., 1995).
Consideration of metabolic ux control analysis
coefcients and theorem
Metabolic ux control analysis theory is an extension of classical enzy-
mology to a metabolic context, allowing for a deterministic and quanti-
tative description of the contribution of all metabolites and modulators
(both called as structural elements) and enzymes or functional proteins
(called operator elements) taking part in the control of a metabolic
pathway (Fell, 1997). This theory is built on the ground of a number of
Table 2. Differential equations for time-dependent variables included in
the model
Time dependent variables Differential equation
Argin
dArgin
dt
transp Arg NOS
Inner Arginine
Orn
dOrn
dt
Arg ODC effluxO
Ornithine
Table 3. Model parameter values and initial values
Description Symbol Model Range References
High afnity transport V
Hat
max
160.5 mM min
1
36.8161.4 mM min
1
Kikuta et al. (1998)
K
Hat
m
70 mM 68140 mM Kikuta et al. (1998)
K
Hat
i;Orn
380 mM 360 35 mM Bogle et al. (1996)
Low afnity transport V
Lat
max
420 mM min
1
112435 mM min
1
Kikuta et al. (1998)
K
Lat
m
847 mM 847 mM Kikuta et al. (1998)
Arginase V
Arg
max
110 mM min
1
40116 mM min
1
Buga et al. (1996)
K
Arg
m
1500 mM 10003000 mM Buga et al. (1996)
K
Arg
i;Orn
1000 mM 1000 mM Daghigh et al. (1994),
Buga et al. (1996)
ODC V
ODC
max
0.013 mM 0.12 10
3
0.013 mM min
1
Osterman et al. (1995)
K
ODC
M
60 mM 60 mM Osterman et al. (1995)
Ornithine efux V
effl:Hat
max
160.5 mM min
1

V
effl:Lat
max
420 mM min
1

K
effl:Hat
m
380 mM
K
effl:Lat
m
847 mM
NOS V
NOS
max
1.33 mM min
1
1.216 mM min
1
Hecker et al. (1994),
Gerber et al. (1997)
K
NOS
m
16 mM 3.922 mM Buga et al. (1996), Gerber et al.
(1997), Leber et al. (1999)
Metabolite parameter Argex 330 mM
a
50100 mM Marquez et al. (1989), Darblade
et al. (2001), Cynober (2002)
a
Concentration in DMEM culture medium
In silico study of arginine catabolism in endothelial cells 225
Ral Montaez Martnez
108
2.0
28.0
axioms and considering several coefcients (related to local and systemic
properties), yielding several theorems (deduced and demonstrated on
the basis of the axioms). In this study, we have taken into account two
systemic coefcients and a theorem.
Flux control coefcient is described as the percentual variation of the
ux through a pathway induced by a percentual variation in one of the
operator elements of the system:
C
J
Ei d J=J=dEi=Ei, with J is the ux and Ei is an operator element
of the system. This coefcient gives information on the relative contri-
bution of every enzyme or transporter of a pathway to the control of the
ux through it.
Response coefcient is described as the percentual variation of the ux
through a pathway induced by a percentual variation in one of the struc-
tural elements of the system:
R
J
x d J=J=dEi=Ei, with J is the ux and x is a structural element
of the system. This coeffcient shows the relative importance of any
modulator or metabolite in a pathway.
The summation theorem of ux control coefcients establishes that in
any metabolic pathway under steady state conditions the total sum of the
C
J
Ei
for all the Ei of the system has to be equal to 1.
To determine the values of ux control coefcients, percentual changes
of the activities of each enzyme or transporter were introduced in the model
and simulations yielded new data allowing for the estimation of the cor-
respondent percentual changes in uxes. In the case of response coef-
cients, we only determined the response of the systemto percentual changes
in the value of the control parameter, external arginine concentration.
Results
Simulations of our model from initial conditions yielded
a physiological steady-state condition in less than 1 h,
as shown both in the evolution of inner arginine and
ornithine concentrations (Fig. 3A) and enzyme activities
(Fig. 3B). These values are within the range of experi-
mental values so far reported (Table 4). This steady state
was considered as the basal condition for the application
of ux control analysis theory (see below). Furthermore,
additional simulations changing the value of the control
parameter, external arginine, up to four orders of magni-
tude (from 10 mM to 10 mM) gave also rise to steady state
conditions after no more than 100 min (Fig. 4). At least
within three orders of magnitude, the steady state con-
ditions as determined by inner arginine levels (Table 5)
were within the described range of values in a previous
experimental study (Arnal et al., 1995).
Fig. 3. Simulations of the model yield a physiological steady state. A
Evolution of the concentrations of inner arginine (triangles) and or-
nithine (diamonds). B Evolution of activities: high afnity transporter
(open triangles); low afnity transporter (open squares); arginase
(diamonds); nitric oxide synthase (closed grey squares); ornithine de-
carboxylase (closed triangles); efux (closed black squares)
Table 4. Simulation of the model yields a physiological basal steady-state
Description Symbol Model Range References
Inner arginine concentration Argin 1196 (mM) 100800 in-culture Baydoun et al. (1990), Arnal et al. (1995)
8002000 in-vivo (mM)
Ornithine concentration Orn 314 (mM) 220410 (mM) Marquez et al. (1989), Casey et al. (2000)
Transport activity transp 43 (mM=min) 34.180.7 (mM=min) Kikuta et al. (1998), Casey et al. (2000)
Arginase activity Arg 42 (mM=min) 293 (mM=min) Buga et al. (1996)
ODC activity ODC 0.01 (mM=min) 00.17 (mM=min) Morrison and Seidel (1995)
NOS activity
NOS
1.31 (mM=min) 1.33 (mM=min) Hecker et al. (1994)
Efux activity EffluxO 42 (mM=min)
226 R. Montanez et al.
109
Captulo 3
2.0
28.0
Application of ux control analysis theory to the basal
conditions achieved in simulations of the model allowed
us to determine ux control coefcient for all the operator
elements included in the model (Table 6), namely, the
high afnity transproter, the low afnity transporter, argi-
nase, ornithine decarboxylase, nitric oxide synthase and
the combined action of transporters for ornithine efux.
Furthermore, Table 6 shows that the summation theorem
for ux control coefcients is fullled.
The study of the response of the system to changes in the
control parameter (external arginine) was a second applica-
tion of ux control analysis theory in this work (Fig. 5).
The response curve showed a three-steps pattern, with a
fast decay in the R
J
x
value from 0.001 to 0.25% eternal
arginine (taking 0.33 mM arginine as 100%), a wide range
of external arginine values (from 25 to 250%) with small
changes in the response coefcient value, and a second
region of fast decay in the response coefcient value at
external arginine concentrations higher than 250%.
Discussion
In spite of the paramount biological importance of ar-
ginine=ornithine derived polyamines and nitric oxide (Wu
and Morris, 1998; Medina et al., 2003; Grillo and
Colombatto, 2004a), the relative contribution of their bio-
synthetic enzymes to the regulation of arginine catabolism
pathway is not well described. To get new insights in this
topic by in silico application of ux control analysis was a
nal goal of this work. To achieve this goal, we had previ-
ously to build a model of the pathway and to show that the
model yielded physiological steady states and was robust.
We built an extremely simple model of arginine cata-
bolism, taking only into account the branches leading to
NO and putrescine (Fig. 2). Nonetheless, in spite of its
simplicity, our model managed to simulate properly the
behavior of this pathway in endothelial cells, yielding a
basal steady state with physiological values of metabolite
concentrations and enzyme activities (Fig. 3 and Table 4)
within the range of available experimental data. Further-
more, our model was very robust, as shown by its ability to
reach steady states for a range of external arginine concen-
trations covering four orders of magnitude (Fig. 4). This
in silico experiment simulated the experiments carried
out by Arnal et al. to determine the effect of changes in
extracellular arginine concentrations on intracellular argi-
nine levels (Arnal et al., 1995). The values of inner argi-
nine concentration once the steady states were reached in
our in silico experiment were within the experimental
range of values for extracellular arginine concentration
values of 0.01, 0.1 and 1 mM (Table 5). Only for an
external arginine concentration value as high as 10 mM
(far away from the physiological situation, since plasma
concentrations of arginine are in the 60100 mM range)
(Marquez et al., 1989; Darblade et al., 2001; Cynober,
2002), the steady state value of inner arginine yielded by
Fig. 4. The model is robust. Inner arginine steady state values mono-
tonically increase with increasing external arginine levels (10, 100, 330,
1000 and 10000 mM)
Table 5. Simulations t well with experimental data within 3 orders of
magnitude of external arginine
Outside arginine
(mM)
Inner arginine
(in-vivo) (mM)
a
Inner arginine
(in-silico) (mM)
10 200340 142
100 350610 628
330 8002000 1268
1000 21603440 2147
10000 920011400 3391
a
Data from Arnal et al. (1995)
Table 6. Flux control coefcients for enzymes and transporters included in the model
H.A.T L.A.T Arginase ODC NOS Efux Summation
CE
J
0.0964 0.3604 0.3548 0.0000 0.0173 0.1937 0.9992
In silico study of arginine catabolism in endothelial cells 227
Ral Montaez Martnez
110
2.0
28.0
simulations of our model was out of the range of reported
experimental values (Arnal et al., 1995).
As expected from systems biology approaches, the pre-
dictions of this simple model could contribute to propose
new hypothesis to be experimentally tested. This has been
the case, since our model has allowed us to apply ux
control analysis, obtaining results that could be experi-
mentally tested in the future. The use of mathematical
models represents a helpful tool to minimize the actual
number of real experiments to only those producing inter-
esting results, according to the predictions of the model
simulations. The deduced ux control coefcient values,
obtained for the operator elements included in the model
(Table 6) show us that endothelial cell arginine catabolic
rate is mainly controled by the low afnity arginine trans-
porter and arginase (C
J
Ei
values of 0.360 and 0.355,
respectively), with a remarkable contribution of ornithine
efux (C
J
Ei
0:197). It should be stressed that, although
ornithine decarboxylase and nitric oxide synthase are
usually taken as key enzymes in the biosynthetic path-
ways of polyamines and NO, respectively, their contribu-
tion to the control of the ux through arginine catabolic
pathway is negligible. The fulllment of the summatory
theorem for ux control coefcients can be taken as a
support for the condence on the results obtained in the
simulations of our model.
Finally, the curve describing the changes in the res-
ponse coefcent with changing values of the control param-
eter (external arginine) shows that the system is more
sensible to these changes at low external concentrations
of arginine (<80 mM) and becomes almost insensitive to
changes in external arginine under conditions well above
(80800 mM) the usual physiological range.
Since metabolic networks have been shown to be hier-
archical and modular, relative simple extensions of a
metabolic model are allowed by aggregation of additional
modules. This modular approach should provide in the
future more comprehensive and detailed models of argi-
nine metabolism.
Acknowledgements
This work was supported by a grant from Fundacioon Ramoon Areces,
Grants SAF2005-01812 and TIN2005-09098-C05-01 (Ministry of Educa-
tion and Sciences, Spain), and CVI-267 and CVI-657 (Andalusian
Research Programme, PAI, Andalusia, Spain).
References
Arnal JF, Munzel T, Venema RC, James NL, Bai CL, Mitch WE, Harrison
DG (1995) Interactions between L-arginine and L-glutamine change
endothelial NO production. An effect independent of NO synthase
substrate availability. J Clin Invest 95: 25652572
Baydoun AR, Emery PW, Pearson JD, Mann GE (1990) Substrate-
dependent regulation of intracellular amino acid concentrations in
cultured bovine aortic endothelial cells. Biochem Biophys Res Com-
mun 173: 940948
Bogle RG, Baydoun AR, Pearson JD, Mann GE (1996) Regulation of
L-arginine transport and nitric oxide release in superfused porcine
aortic endothelial cells. J Physiol 490: 229241
Bronte V, Zanovello P (2005) Regulation of immune responses by
L-arginine metabolism. Nat Rev Immunol 5: 641654
Buga GM, Singh R, Pervin S, Rogers NE, Schmitz DA, Jenkinson CP,
Cederbaum SD, Ignarro LJ (1996) Arginase activity in endothelial
cells: inhibition by NG-hydroxy-L-arginine during high-output NO
production. Am J Physiol 271: H1988H1998
Casey TE, Harrison AC, Zimmerman JK, Hilderman RH(2000) Studies of
L-arginine transport in bovine aortic endothelial cells. Arch Biochem
Biophys 379: 283291
Cooke JP, Tsao PS (1997) Arginine: a new therapy for atherosclerosis?
Circulation 95: 311312
Cynober LA (2002) Plasma amino acid levels with a note on membrane
transport: characteristics, regulation, and metabolic signicance. Nutri-
tion 18: 761766
Daghigh F, Fukuto JM, Ash DE (1994) Inhibition of rat liver arginase
by an intermediate in NO biosynthesis, NG-hydroxy-L-arginine: impli-
cations for the regulation of nitric oxide biosynthesis by arginase.
Biochem Biophys Res Commun 202: 174180
Darblade B, Batkai S, Causse E, Gourdy P, Fouque MJ, Rami J, Arnal JF
(2001) Failure of L-nitroarginine to inhibit the activity of aortic
inducible nitric oxide synthase. J Vasc Res 38: 266275
Fell DA (1997) Understanding the control of metabolism. Portland Press,
London
Gerber NC, Nishida CR, Ortiz de Montellano PR (1997) Characterization
of human liver inducible nitric oxide synthase expressed in Escherichia
coli. Arch Biochem Biophys 343: 249253
Gornik HL, Creager MA (2004) Arginine and endothelial and vascular
health. J Nutr 134: 2880S2887S; discussion 2895S
Grillo MA, Colombatto S (2004a) Arginine revisited: minireview article.
Amino Acids 26: 345351
Fig. 5. Curve showing the evolution of the response coefcient with
changes in the levels of external arginine (taking the 330 mM concentra-
tion included in the initial conditions as 100%)
228 R. Montanez et al.
111
Captulo 3
2.0
28.0
Grillo MA, Colombatto S (2004b) Metabolism and function in animal
tissues of agmatine, a biogenic amine formed from arginine. Amino
Acids 26: 38
Hecker M, Mulsch A, Bassenge E, Forstermann U, Busse R (1994) Sub-
cellular localization and characterization of nitric oxide synthase(s) in
endothelial cells: physiological implications. Biochem J 299: 247252
Kikuta K, Sawamura T, Miwa S, Hashimoto N, Masaki T (1998) High-
afnity arginine transport of bovine aortic endothelial cells is impaired
by lysophosphatidylcholine. Circ Res 83: 10881096
Kitano H (2002) Systems biology: a brief overview. Science 295:
16621664
Leber A, Hemmens B, Klosch B, Goessler W, Raber G, Mayer B, Schmidt
K (1999) Characterization of recombinant human endothelial nitric-
oxide synthase puried fromthe yeast Pichia pastoris. J Biol Chem274:
3765837664
Mann GE, Yudilevich DL, Sobrevia L (2003) Regulation of amino acid
and glucose transporters in endothelial and smooth muscle cells.
Physiol Rev 83: 183252
Marquez J, Sanchez-Jimenez F, Medina MA, Quesada AR, Nunez de
Castro I (1989) Nitrogen metabolism in tumor bearing mice. Arch
Biochem Biophys 268: 667675
Medina MA, Urdiales JL, Rodriguez-Caso C, Ramirez FJ, Sanchez-
Jimenez F (2003) Biogenic amines and polyamines: similar biochem-
istry for different physiological missions and biomedical applications.
Crit Rev Biochem Mol Biol 38: 2359
Medina MA, Correa-Fiz F, Rodriguez-Caso C, Sanchez-Jimenez F (2005)
A comprehensive view of polyamine and histamine metabolism to the
light of new technologies. J Cell Mol Med 9: 854864
Morris SM Jr (2004) Recent advances in arginine metabolism. Curr Opin
Clin Nutr Metab Care 7: 4551
Morrison RF, Seidel ER (1995) Vascular endothelial cell proliferation:
regulation of cellular polyamines. Cardiovasc Res 29: 841847
Nijhout HF, Reed MC, Budu P, Ulrich CM (2004) A mathematical model
of the folate cycle: new insights into folate homeostasis. J Biol Chem
279: 5500855016
Osterman AL, Kinch LN, Grishin NV, Phillips MA(1995) Acidic residues
important for substrate binding and cofactor reactivity in eukaryotic
ornithine decarboxylase identied by alanine scanning mutagenesis. J
Biol Chem 270: 1179711802
Reed MC, Nijhout HF, Sparks R, Ulrich CM (2004) A mathematical
model of the methionine cycle. J Theor Biol 226: 3343
Rotoli BM, Bussolati O, Sala R, Gazzola GC, DallAsta V (2005) The
transport of cationic amino acids in human airway cells: expression of
system y L activity and transepithelial delivery of NOS inhibitors.
Faseb J 19: 810812
Verrey F, Closs EI, Wagner CA, Palacin M, Endou H, Kanai Y (2004)
CATs and HATs: the SLC7 family of amino acid transporters. Pugers
Arch 447: 532542
Wu G, Morris SM Jr (1998) Arginine metabolism: nitric oxide and
beyond. Biochem J 336: 117
Yang CR, Shapiro BE, Hung SP, Mjolsness ED, Hateld GW (2005)
A mathematical model for the branched chain amino acid bio-
synthetic pathways of Escherichia coli K12. J Biol Chem 280:
1122411232
Authors address: Dr. Miguel

AAngel Medina, Departamento de Biolog a
Molecular y Bioqu mica, Facultad de Ciencias, Universidad de Malaga,
E-29071 Malaga, Spain,
Fax: 34-952131674, E-mail: medina@uma.es
In silico study of arginine catabolism in endothelial cells 229
Ral Montaez Martnez
112
2.0
28.0
Anexo
Modelado modular del metabolismo.
Con la intencin de refutar la Hiptesis 2
y cumplir con el objetivo 2, se modelaron e
integraron cuatro rutas metablicas.
Cada una de estos modelos ha sido ge-
nerado independientemente y publicados
en diferentes revistas de investigacin [1-
4], proviniendo cada uno de ellos de dife-
rentes grupos de investigacin sin relacin.
Los diferentes modelos han sido repro-
ducidos en una programa que permite la
integracin de las diferentes ecuaciones
diferenciales mediante el mtodo de inte-
gracin de Euler.
Los modelos integrados fueron: el meta-
bolismo de las poliaminas [4], el catabo-
lismo de arginina [1], readaptando las
constantes cinticas de clulas endotelia-
les a clulas hepticas. El ciclo de los meti-
los activados[3] y el ciclo de los folatos [2].
La metodologa que se emple en la
integracin fue la misma que se emple
para la consolidacin de los modelos de
poliaminas y arginina y que se detallan en
la figura 1.
La integracin de los diferentes mdulos
se muestra como un sistema estable, ca-
paz de alcanzar un estado estacionario
con valores fisiolgicos y en un intervalo
de tiempo que se corresponde con los
comportamientos descritos en la bibliogra-
fa. De este modo se demuestra que es
posible ir integrando recursivamente m-
dulos de diferentes rutas metablicas,
como piezas de un LEGO previamente
modeladas (Figura 2 y Tabla 1).
Como es evidente, la integracin ha de
ser coherente, esto significa que no de-
bemos de integrar un modelo de diferen-
tes organismos o diferentes tejidos, siempre
que sea posible readaptar el modelo a las
reacciones especficas del organismo o
113
Captulo 3
2.0
28.0
Figura 1.-Metodologa aplicada para le integracin
y evaluacin de la informacin enzimolgica, rela-
tiva a las diferentes rutas modeladas, y la consoli-
dacin de los mdulos aadidos sequencialmente al
mdulo de las poliaminas
Figura 2.-Concentracin en el tiempo de todos los metabolitos
definidos como variables libres del modelo integrado.
tejido y a las diferentes constantes cinti-
cas descritas para tales.
Los modelos integrados van mostrando,
a medida que se adiciona un nuevo m-
dulo una nueva dependencia. en este
sentido se cumple la hiptesis de que a
medida que se van adicionando mdulos
vamos descubriendo nuevas propiedades
emergentes del sistema.
Los anlisis de sensibilidad a las condi-
ciones iniciales realizados sobre el modelo
integrado corroboran que el modelos es
estable. Como se detalla en la tabla 2, el
sistema no es extremadamente depen-
diente de ninguna variable libre o par-
metro.
Perspectivas futuras
El trabajo de integracin se ha plantea-
do como la posible herramienta con la
Ral Montaez Martnez
114
2.0
28.0
DHF THF CH2THF CH=THF 10fTHF 5mTHF Metionina Homocisteina SAH SAM Put Spdn Spmn
Modelo 0,026 5,62 0,91 1,13 6,25 5,10 50,16 0,54 93,6 61,29 104,93 79,58 57,16
Bibliografa 0,023-0,12 1,8-6,8 1-2,5 2,7-11,2 1-16 4,6-8 45-80 0,1-1 1-30 20-100 131,9 91-4 38,5
Tabla 1.-Concentracin micromolar (!M) de cada uno de los metabolitos implicados en el modelo integrado, mostrndose la
concentracin alcanzada en el modelo y el rango de concentraciones descritos para cada uno de ellos en la bibliografa.
Figura 3.-Representacin esquemtica de los diferentes metabolitos y actividades contenidas en el modelo integrado.
Catabolismo de arginina (gris), Biciclo de las poliaminas (Rojo), ciclo de la metionina (verde) y ciclo de los fola-
tos(azul)
115
Captulo 3
2.0
28.0
Tabla 2.- Anlisis de sensibilidad a condiciones iniciales tal como se desarrollan en los artculos 1 y 2 de este captulo. Las
columnas hacen referencia a las variables libres mientras que las filas referencian los parmetros.
que analizar las relaciones del metabolis-
mo de las poliaminas e histamina con el
metabolismo oxidativo y la proliferacin
celular. En ese sentido, La tarea de mi
compaero Armando Reyes contempla la
adicin de mdulos del ciclo de los cidos
tricarboxlicos, la glucolisis, el metabolismo
de histamina, la actividad de las diferentes
lanzaderas mitocondriales, etc (Figura 3).
Ral Montaez Martnez
116
2.0
28.0
Figura 3.- Representacin de las rutas ha incluir en el modelo integrado.
Bibliografa
1.! Montanez R, Rodriguez-Caso C, Sanchez-Jimenez F, and Medina
MA. In silico analysis of arginine catabolism as a source of nitric
oxide or polyamines in endothelial cells. Amino Acids. 2008;34(2):
223-229.
2.! Nijhout HF, Reed MC, Budu P, and Ulrich CM. A mathematical mo-
del of the folate cycle: new insights into folate homeostasis. J Biol
Chem. 2004;279(53): 55008-55016.
3.! Reed MC, Nijhout HF, Sparks R, and Ulrich CM. A mathematical
model of the methionine cycle. J Theor Biol. 2004;226(1): 33-43.
4.! Rodriguez-Caso C, Montanez R, Cascante M, Sanchez-Jimenez F,
and Medina MA. Mathematical modeling of polyamine metabolism
in mammals. J Biol Chem. 2006;281(31): 21799-21812.
117
Captulo 3
2.0
28.0

2.0
28.0
Simples lneas y
puntos
2.0
28.0
Hasta qu punto podemos simplificar un
sistema?. Es evidente que en sistemas
complejos, sistemas con un elevado nme-
ro de componentes y relaciones, la simplifi-
cacin es la nica alternativa de compren-
sin. Pero puede darse la circunstancia de
que en el afn por comprender dejemos
atrs la esencia de nuestra curiosidad. El
empleo de la teora de grafos en sistemas
de elevado nmero de elementos interac-
tuantes ha demostrado sobradamente su
vala. La fuerza de esta metodologa radica
en la correcta abstraccin de nuestro sis-
tema de estudio en un grafo. En nuestro
estudio mostramos que a pesar de lo rigu-
rosos que podemos ser en la aplicacin de
la teora esto carece de valor si la red que
representa nuestro sistema no es fiel a su
naturaleza. En el estudio de redes es cada
da ms una realidad la admisin de que
las proyecciones de redes naturalmente
bipartitas eliminan informacin relevante
121
2.0
28.0
Simples lneas y puntos:
estudiando el metabolismo como una red bipartita
Pasarn millones de aos hasta que lleguemos a alguna compren-
sin, y an entonces no ser completo, porque nos enfrentamos al
infinito.
Paul Erds
4
sobre cmo se conectan los elementos de
la red. Nosotros hemos querido incorporar
este conocimiento al estudio de las redes
del metabolismo. Redes naturalmente bi-
partitas en las que prescindir de los sustra-
tos o los metabolitos condiciona nuestra
comprensin de la realidad que estudia-
mos.
122
2.0
28.0
Ral Montaez Martnez
123
Captulo 4
2.0
28.0
When metabolism meets topology:
Reconciling metabolite and reaction networks
Raul Montanez,
1
Miguel Angel Medina,
1
Ricard V. Sole,
2,3
and Carlos Rodrguez-Caso
2
*
1
Departamento de Biolog a Molecular y Bioqu mica, Facultad de Ciencias, Universidad de Ma laga, E-29071 Ma laga, and
CIBER de Enfermedades Raras (CIBERER), Ma laga, Spain
2
ICREA-Complex Systems Lab, Universitat Pompeu Fabra. Parc de Recerca Biome` dica de Barcelona. Dr. Aiguader 88, 08003.
Barcelona, Spain
3
Santa Fe Institute 1399 Hyde Park Road, Santa Fe, NM 87501, USA
The search for a systems-level picture of metabolismas a
web of molecular interactions provides a paradigmatic
example of how the methods used to characterize a
systemcan bias the interpretation of its functional mean-
ing. Metabolic maps have been analyzed using novel
techniques from network theory, revealing some non-
trivial, functionally relevant properties. These include a
small-world structure and hierarchical modularity. How-
ever, as discussed here, some of these properties might
actually result from an inappropriate way of dening
network interactions. Starting from the so-called bipar-
tite organization of metabolism, where the two mean-
ingful subsets (reactions and metabolites) are
considered, most current works use only one of the
subsets by means of so-called graph projections. Unfor-
tunately, projected graphs often ignore relevant biologi-
cal and chemical constraints, thus leading to statistical
artifacts. Some of these drawbacks and alternative
approaches need to be properly addressed.
Keywords: bipartite networks; hierarchical modularity;
metabolic networks; scale-free; small-world
Introduction
More than a thousand chemical reactions occur within our
cells, providing the building blocks and the fuel for life. They
are linked together forming an intricate, complex network in
which metabolites are transformed according to the laws of
chemistry and thermodynamics. Metabolism has been
traditionally organized in terms of pathways that are
interlinked forming a connected roadmap. This roadmap
was elucidated through the joint efforts of generations of
biochemists during the 20th century. The dispersed informa-
tion was initially compiled into the famous Boehringers
metabolic map by Gerhard Michal in 1993.
(1,2)
However, it is
only in the last decade that, thanks to new advances in
computational methods, metabolic pathways have been
conveniently organized and standardized within databases
such as KEGG
(3)
and MetaCyc.
(4)
Nowadays, metabolic data
for a variety of organisms are available.
Metabolism is the best-known cellular network. However,
its topological organization the global pattern of connections
of the graph has not raised the interest of biologists until
recently.
(57)
This shift was motivated by the nding that real
networks display a number of previously unknown traits such
as small-world
(8)
and scale-free organization.
(9,10)
Such a
picture has permeated a large part of the literature in the
eld,
(1113)
even at the level of standard cell biology
textbooks.
(14)
The small-world property tells us, roughly speaking, that
any two nodes in a network are on average separated by just a
few intermediate links, despite the fact that networks are
large and sparse. Such short paths could have an
immediate impact on network functionality, since they
enhance the propagation of changes through the system.
Additionally, small-world graphs have a high cliquishness,
much longer than expected from random (see Box 1 and
Box 2 for a formal denition). Therefore, a small world
combines a far-from-random local association with a short-
path structure.
The second property is related to the high heterogeneity in
the number of links that a given node can display. Specically,
it was observed that the number of nodes having k
connections N(k) is such that most of the elements of the
net have just one or two links, whereas a handful of them (the
hubs) have many connections. This pattern can be described
using a well-dened fat-tailed distribution. The presence of
hubs has several consequences, from favoring the small-
world behavior to inducing internal fragilities associated with
their failure. Moreover, the presence of these patterns can be
measured, modeled, and used to explore the potential paths
followed by these webs through their evolution.
(1519)
Network
thinking has inuenced the study of very diverse systems,
from proteomes to the Internet, revealing that very disparate
systems share common patterns of organization that can be
characterized by their topology.
Problems and paradigms DOI 10.1002/bies.200900145
*Correspondence to: C. Rodr guez-Caso, ICREA-Complex Systems Lab,
Universitat Pompeu Fabra. Parc de Recerca Biome` dica de Barcelona. Dr.
Aiguader 88, 08003. Barcelona, Spain.
E-mail: carlos.rodriguez@upf.edu
246 BioEssays 32:246256, 2010 Wiley Periodicals, Inc.
124
2.0
28.0
Ral Montaez Martnez
In a nutshell, a graph can be described as a mathematical
abstraction of reality in which nodes are individual units
(metabolites of a reaction, species, proteins, actors of a lm,
or websites) that appear linked by an edge if some type of
relation exists among them. In this process, the real system is
represented by a connection pattern, thus ignoring most
Box 1
A graph is the mathematical representation of the relation
between the elements of a system. Elements are nodes or
vertices in the graph. Two nodes are connected in the graph by an
link (or edge) if they establish some type of relationship.
(61)
The
number of nodes (N) and edges (E) dene the size of the graph.
A bipartite graph is a graph consisting of two disjoint sets of
nodes related through a set of edges. In this graph, connections
between nodes of the same set are forbidden. This contrasts to
unipartite graphs, where elements of a set of nodes establish the
connections among them.
The degree (k1) or rst neighborhood of a node is its number of
edges or connections. In the bipartite metabolic network, the degree
(k1 of a metabolite indicates the number of reactions in which it
participates, whereas the degree of a reaction node represents the
number of substrates and products of this reaction. Average degree
hki is calculated as a global estimator of the network from degree
node information. The abundance of the number of nodes of a
certain degree is the degree distribution. Notice that two hki and
two degree distributions can be dened for a bipartite graph. A
suitable generalization of the concept of degree for bipartite graphs
is just the total number of paths connecting some given node with all
nearest-neighboring nodes of the same type, namely strength.
The second neighborhood (k2) is the number of the
neighbors neighbors of a node. In bipartite metabolic networks,
the k2 of a metabolite A indicates the number of metabolites that
can be combined in some reaction with A.
Projection of a bipartite graph is the mathematical operation that
allows constructing a unipartite graph where nodes of only one
type are present. These nodes are connected in the new graph if
they share a common neighbor (of the other type) in the bipartite
graph. According to this denition, two alternative projections can
be done from a bipartite graph (see also Box 2).
A set of nodes forms a clique if all possible connections among
them are present. Projections of bipartite graphs are enriched of
cliques.
Clustering coefcient is a measure of the local association or
cliquishness (fromclique) of a node. The clustering coefcient of
a node is the number of connected neighbors normalized by the
number of all possible combinations. Average clustering
coefcient hCi is zero in bipartite graphs. However, it is very high
in projections. Calculating the average of clustering coefcient for
every k, the clustering coefcient distribution is obtained.
A path is dened as a sequence of connected nodes. Minimal
path average is a global estimator of the network. Average is
calculated from the minimum of every possible pair of nodes. This
measure is closely related to the diameter of a network.
Module is a set of nodes that presents a closer relation, usually
dened by a more dense connection, than with the remaining
network.
Box 2: Graphs and models
Bipartite graph and projections: For a bipartite graph with two
disjoint sets of nodes A and B, the projection of A on B is the
unipartite graph constructed with A nodes. In this projection, two A
nodes establish a link if they are connected to, at least one,
common B node in the bipartite graph (see Fig. 1). Analogously,
the projection of B on A can be constructed. For metabolic
networks, the two alternative projections of the bipartite graph are
known as metabolite and reaction network.
Erdo s-Re nyi (ER) model
(62)
(A) is a random unipartite graph
of N nodes and E edges. The
probability P denes the likelihood that a pair of nodes is
connected in the graph. ER model follows a Poisson degree
distribution where hki is the mean of distribution. Given Nand P, for
large enough ER graphs EP[N(N1)]/2, hkiPN and hCiP.
Prior to Baraba sis work on scale-free networks, it was commonly
accepted that the ER model tted the structure of very large
networks. Analogously, the Erdo s-Re nyi Bipartite model (B) can
be dened introducing one additional constraint by separating
nodes of the network in two groups (labeled in Bas white and black
nodes). In this network a link between two nodes is established
with a probability (P) provided that no pair is established between
nodes the same groups. In this network a Poisson degree
distribution is given for both sets of nodes and hCi 0.
Scale-free graph: This is a network that
follows a power-law degree distribution
P(k)k
-g
where gamma ranges between 2
and 3. A handful of nodes (the hubs) have
many connections, whereas the vast
majority has very few ones. From a
theoretical perspective, a consequence of a power-lawdistribution
is that a characteristic scale cannot be dened.
Small-world model: Watts and
Strogatz showed in 1999
(8)
that
the small-world phenomenon
described by Milgram
(63)
can
be represented by a simple
experiment of disordering a regular lattice at random. Rewiring
only a handful of links is sufcient to produce a path length similar
(L) to that observed in an ER model, but keeping the original local
organization of the real network. This local organization is
estimated through hCi. This model allows the small-world criterion
for any (unipartite) network to be dened. Real hCi and L are
compared with the expected values of an ER model with the same
number of nodes and links. Formally a graph is small-world when
hCi
(real)
>>hCi
(ER)
and L
(real)
L
(ER)
.
Hierarchical modular model: This is a
deterministic model where nodes are
clustered in modules in a hierarchical
organization. This model, proposed by
Ravasz et al.
(31)
presents a C(k)k
-1
. This
property has been usually associated
in real networks as an indicator of a
hierarchical modular organization. Inter-
estingly, this model also exhibits scale-free and small-world
behavior.
R. Montan ez et al. Problems and paradigms
BioEssays 32:246256, 2010 Wiley Periodicals, Inc. 247
125
Captulo 4
2.0
28.0
details in favor of a systemic perspective. Nevertheless, these
simplied pictures of reality have proven helpful in providing
useful insights into ecology,
(20)
genomics,
(11)
neuro-
science,
(21)
or language.
(22)
However, as shown here, the
process of graph construction is an important issue that can
bias the conclusions derived from the topological approach.
In this paper we critically examine this view within the
context of metabolism. Metabolic maps, along with collabora-
tion,
(2326)
mutualistic,
(27,28)
gene-disease,
(29)
or drug-
target
(30)
networks belong to the class of so-called bipartite
graphs (see Fig. 1 and Box 1 for denitions). Metabolism
involves both metabolites and reactions, but metabolic
networks have usually been treated as unipartite graphs
(networks with only one type of element acting as nodes).
(7,3134)
In the seminal works concerning the topological organization
of metabolism, such a limitation was overcome by a graph
transformation called projection of the bipartite network in
its unipartite version.
(7,31)
As summarized in Fig. 1, two
alternative graph projections [Pa(G) and Pb(G)] can be
obtained from a bipartite graph G. We can construct the
so-called metabolite projection in which nodes are metabo-
lites that are linked to each other if they participate in the
same reaction (see Box 2). Alternatively, the reaction
network is constructed considering nodes as reactions: two
reactions are connected if they share a common metabolite.
Further work revealed that network projections introduce a
strong bias in the results of the topological analysis.
(3538)
However, such biases were only addressed within the eld of
social networks. As shown here, a biologically consistent
denition of metabolic network according to its bipartite nature
provides a more accurate biological interpretation of its
topology.
Metabolism as a complex network: What
do we know about metabolic webs?
Metabolism was one of the rst real networks identied as
both scale-free
(6)
and small-world
(7)
with some contro-
versy.
(39)
Metabolic networks are also considered a paradig-
matic example of hierarchical modular organization.
(31)
Where does this scale-free, modular pattern come from?
Models help us to understand reality and the study of
metabolisms organization has not been an exception. Some
simple, toy models of reality have provided useful insights
into their origin and evolution. An illustrative example of this
type of model is the so-called preferential attachment (or
rich-gets-richer) mechanism. It has been shown that a graph
growing under preferential attachment (see Box 2 for model
description) can produce a scale-free network.
(40)
This
simple rule captures the effect of popularity occurring in
some real networks such as the Internet
(9)
or social networks
exhibiting scale-free distributions.
(10)
Another paradigmatic
example is given in the paper of Watts and Strogatz, which
explains the small-word effect using a very simple network
model.
(8)
This example illustrates the success of a network
approximation by giving a very simple explanation about how
this pattern can emerge. Finally, hierarchical modularity
(31)
is
another example of a key property displayed by real
networks. Roughly speaking, it involves the presence of a
nested set of hierarchically assembled modular parts. Such
a pattern is reminiscent of fractal structures in nature, and
some key properties of real webs can actually be recovered
from a fractal-like iterative model (Box 2).
Topological properties are usually evaluated by compar-
ison with null models. The most studied random model is the
so-called Erdo s-Re nyi (ER) graph in which nodes of a set
are simply linked by a single probability (see Box 2). The
outcome of this model drastically differs from most of the real
networks, and it constitutes a reference of what we expect by
chance. This model is used to establish the presence of the
small-world condition. Analogously, a bipartite version of an
ERmodel is constructed by considering two separated sets of
nodes, provided that no connection between nodes of the
same group is present (see Box 2). As no other process but
randomness is implied in its generation, this model constitutes
an excellent null reference for our purposes since it allows us
to illustrate the impact of the graph projection on topological
measures.
Figure 1. Representation of a bipartite graph. Links establish the
connection between the members two disjoint sets: top nodes (A) and
bottom nodes (B). Connections between nodes of the same set are
forbidden. Two alternative projections (see Box 1 for definition) are
possible by considering nodes of one set as connectors of the other
one. Notice that triangle connection, a feature measured by the so-
called clustering coefficient (see Box 1 for definition) is forbidden in
the bipartite graph but not in its projections. Bipartite graph and
projections are defined in Boxes 1 and 2.
Problems and paradigms R. Montan ez et al.
248 BioEssays 32:246256, 2010 Wiley Periodicals, Inc.
126
2.0
28.0
Ral Montaez Martnez
Interestingly, scale-free, small-world, and hierarchical
modularity topological properties match well with the standard
view of metabolism from a biological perspective. ATP and
NADH

metabolites are examples of hubs (the most

connected nodes) in the metabolic network
(6)
and their
topological importance matches their relevant roles as the
coins for energetic and reduction power in metabolism,
respectively. They participate in most metabolic pathways,
whereas many metabolites only take part in single steps of
particular pathways, in agreement with the scale-free
behavior. The organization of metabolismin coupled reactions
and its traditional classication in pathways agrees with the
idea of local association and modularity. In this context, cross-
links between pathways and the existence of ubiquitous
metabolites the hubs would contribute to the small-world
effect and a hierarchical modular organization.
However, as mentioned above, a network is an abstraction
of reality and its construction can determine the conclusions
derived from it. At this point, the choice of either a bipartite or
unipartite (projected) representation of metabolism acquires
special importance. In fact, small world, hierarchy and scale-
free features of metabolic networks have been reported using
some type of projection. Although these properties seem to
be reasonable for metabolism, in light of evidence from
other bipartite network studies, the nal conclusion is
that the topological analysis of metabolic webs should be
revisited.
Bipartite nature of metabolic networks
and the problem of graph denition
Within metabolic maps, each reaction is not just a causal link
between molecules. The ow of matter, kinetic parameters,
and time scales can be properly measured or estimated. The
global picture is thus rather complete and meaningful. As
mentioned above, a graph is an abstraction of reality where
only the pattern of connection matters. Figure 2A depicts the
standard representation of a metabolic pathway found in any
Biochemistry handbook. Although molecular details, enzyme
kinetics and most of the complexity of the real system are
absent, this representation is enough to sketch the wiring of
this particular metabolic process. Thus, a graph based on the
metabolic map information is arguably the best choice when
looking at how metabolism is connected. However, different
graph constructions are possible. In this context, Albert
Einsteins famous quote Make everything as simple as
possible, but not simpler captures the key point of our work.
According to this, a metabolic network should contain the
essential information but not less, regarding topological
questions, since it provides the clues for a suitable biological
interpretation.
Since metabolic networks consist of metabolites that
participate in reactions, a special feature of these networks
is that metabolites do not interact in pairs, as seen for example
with the protein interaction map. In the later, we can use a graph
construction based on nodes of the same type interacting in
pairs. Metabolic networks requirethe denition of both reactions
and metabolites as separated sets of nodes to reach a
meaningful description. In a reaction, a number of substrates
(usually more than one) give a number of products. In this case,
the connection of metabolites by pairs offers a poorer
representation of reality since it cannot properly capture the
metabolite association (conversion) in reactions, as bipartite
graphs do (see Fig. 2). This bipartite view of metabolism has
been commonly addressed in different works.
(6,7,4143)
A number of analyses describing the topological organiza-
tion of metabolism have introduced substantial simplications
Figure 2. Graph representations of three-step metabolic pathway reproducing the gamma-phosphate transference fromATP to UMPand CTP
formation. (A) The standard view of a metabolic pathway in Biochemistrys handbooks. (B) The metabolite-reaction representation in a bipartite
graph.
(6)
(C) Projections of the bipartite graph according to;
(7)
metabolite projection: all metabolites of a reaction are fully connected (white-node
graph in C); and reaction projection: two reactions are connected if they share a common metabolite (black-node graph in C). (D) Alternative
metabolite-projection used in ref.
(31)
Given a reaction, substrates are connected to products, but substrates (and products) are not connected
among them. (E) Representation eliminating the ATP and ADP metabolic network hubs applied in ref.
(7)
and ref.
(31)
with slight modifications.
R. Montan ez et al. Problems and paradigms
BioEssays 32:246256, 2010 Wiley Periodicals, Inc. 249
127
Captulo 4
2.0
28.0
to metabolic network denition. In most of them, such
simplications involve the generation of unipartite networks
with only metabolites as the nodes of the graph. Figure 2
shows some of these unipartite versions. The standard
projection (see Box 1 and Box 2 for denition, and Fig. 2C) is
probably one of the most inuential network simplications
since it was the rst used to dene the small-world character
of metabolism.
(7)
In addition, it has been used to describe the
power-law distribution of metabolite and reaction unipartite
networks.
(32)
However, more rened projections have been proposed. A
widely used alternative is depicted in Fig. 2D. In this projection
the connections occur only between the substrates and
products of a reaction.
(31,33,44,45)
This modication can be
justied as a way to capture molecular transformations.
Nevertheless, the opposite projection criterion, i.e., the
connection between substrates on one side and products
on the other (of a reaction), may be similarly justied by the
fact that molecular collisions actually occurs between
substrates of a reaction. However, as far as we know such
a simplication has never been applied.
Additional simplications in the projection process, such as
the elimination of the most connected metabolites due to its
ubiquitous nature,
(7)
or even the elimination of metabolites
displaying only one connection and the merging of linear
paths in single nodes,
(31)
have been used for different
purposes. In conclusion, a repertoire of unipartite versions of
metabolic networks has been used to highlight one particular
trait of metabolic organization. However, what determines the
correct level of simplication when using projections is an
open question. A problem arises when all the pieces of
information obtained fromdifferent network denitions are put
together, ignoring the possibility that such simplications may
introduce a bias.
However, unipartite versions of metabolism are also
possible without the use of a projection. An alternative
construction of a metabolic network proposed in ref.
(39)
is
probably a fair view of a natural unipartite network. In this
representation metabolites are nodes and two metabolites
are linked if a carbon transfer occurs between them. In this
case, links are not a consequence of a mathematical
transformation and they have a physical meaning, even if
reactions are not explicit in such a representation. The main
limitation of this network is that only a small part of the
whole chemical process is considered since not only carbon
but other atoms, electrons and energy are actually
transferred between molecules. However, extending this
idea to any type of transfer would produce a chemically
realistic unipartite network. As a side effect, we would lose
the information associated with molecular relations due to
reactions. However, the combined information of bipartite
and this unipartite network would provide a very good
insight into the global organization of metabolic topology.
Unfortunately, as far as we know, no such network has yet
been constructed.
In the next section, we showthat simplications of bipartite
metabolic networks produce a loss of information and even
generate contradictory results in the case of small-world
patterns,
(7,39)
introducing a bias in the biological interpretation
of systemic functional traits.
Too much simple metabolic networks?
The problem of graph projection
Wagner and Fell
(7)
suggested that bipartite graphs are much
less intuitive constructs, and less obviously treatable than
their projections. It is worth noting that at that time, the
analysis of bipartite networks was less developed than for
unipartite graphs. Therefore, the topological analysis of the
metabolite projection was considered as a reasonable
starting point for the study of metabolism as a network.
As we mentioned above, a bipartite network can be
transformed in two alternative unipartite versions by means of
projection of a graph (see Fig. 1 for a general denition of a
graph projection). In metabolite networks, a metabolite
projection graph is formed only by metabolites. In this graph,
two metabolites are connected by a link if they participate in
the same reaction. In contrast, a reaction projection involves
reactions as nodes. In this case, two reactions are connected
by a link if they share a common metabolite.
As shown in Fig. 3, the same graph projection can be
associated with very different bipartite networks. This
indicates that only attending to projections, the relation
among metabolites through reactions cannot be recovered,
and therefore part of the information present in the bipartite
graph is lost in the transformation process. Additionally,
projections usually exhibit a markedly higher value of
average degree in relation to the bipartite graph. As a
consequence of projection, average degree has a different
interpretation than the bipartite one. As Box 1 indicates, the
degree in bipartite graphs has a natural meaning, i.e., the
number of partners in a reaction or the number of reactions
in which one metabolite participates. In projections as
depicted in Fig. 3C, the degree of a node (for example the
UTP node) indicates the number of different metabolites that
can be related to it. However, compared with the bipartite
representation, in such projections we know neither the
number of reactions in which UTP is involved nor the number
of times that metabolite partners of UTP are repeated in
such reactions. Interestingly, the degree of a given node in
this (metabolite) projection is recovered by its so-called
second neighborhood (see Box 1 for denition) in the
bipartite network. Accordingly, it should be stressed that
bipartite graphs contain information that cannot be inferred
from projections.
Problems and paradigms R. Montan ez et al.
250 BioEssays 32:246256, 2010 Wiley Periodicals, Inc.
128
2.0
28.0
Ral Montaez Martnez
Another important limitation appears when we look at the
degree distribution. Although the degree distribution (see
Box 1) for metabolites falls off as a power law,
(6)
the fact that
reactions follow instead a binomial distribution has been
largely ignored. To illustrate this, we show in Fig. 4 the degree
of distributions for the E. coli metabolic network used in ref.
(6)
in its bipartite version. Figure 4D shows the reaction degree
distribution, with an average of 3.73 metabolites and a
maximum of 8. This is reasonable since one reaction is a
single event, where the number of participants is restricted to
a few reactants. We can argue that the minimum number of
participants in a reaction is two, corresponding to a tautomeric
conversion. However, as Fig. 4D reveals, only a very small
fraction of reactions with just one metabolite are present in the
network. This is not consistent with the biological interpreta-
tion and it may be due to a problem of data curation or
incompleteness.
In spite of these small discrepancies, the key problem is
that degree distributions have a simple interpretation in the
bipartite graph but not in its projections. Similarly, metabolite
and reaction distributions are strongly canalized by biological
and chemical constraints. In the case of metabolites, the
appearance of a new reaction in evolution can be fairly
justied by the appearance of new enzymatic activities by a
process of gene duplication and diversication. Such a new
reaction should occur using existent metabolites and with
some probability they may form new compounds that would
be incorporated to metabolism. According to this view, a
scale-free metabolite distribution is likely to happen and can
be indirectly deduced from current unipartite network models
of metabolic evolution.
(46)
However, as reactions mean
transformation by molecular collision satisfying both thermo-
dynamic and chemical constraints, the addition or elimination
of one molecule to reactions is unlikely to happen. This
constraint is more compatible with an exponential/normal
distribution than with a scale-free one. This situation has been
reported from the study of association networks, where the
networks exhibit a scale-free distribution for one node type
and normal distributions for the other one. This occurs just
because one type of nodes allows a very large range of
connections compared to the other.
(47,48)
As illustrated in Figs. 1 and 3, a complete picture of the two
types of nodes is only captured using bipartite representation,
whereas none of the graph projections recover it. In this
context, it has been observed that, if one node type follows a
power law in the bipartite representation, both projections
exhibit power-law degree distributions.
(47)
This has caused
wrong interpretations of the results for reactions, as they have
been described to also follow a power-law fashion.
(32,49)
Finally, the clustering coefcient (C) is probably the most
problematic parameter. Clustering is zero in bipartite networks
by denition but not in their projections. An immediate feature
derived from the theory is that projections of bipartite graphs
are enriched of cliques (see Box 1 for denition) and cliques
markedly increase the value of the average clustering (hCi).
For an ER bipartite graph (see Box 2 for description), its
projection presents a high hCi (see Table 1). However, graph
projections do not correspond with a unipartite ER. In fact, hCi
is much greater in the ER bipartite projection than in an ER
unipartite version constructed with the same number of nodes
and links of the graph projection (see Table 1 for numerical
comparison). This indicates that simple aggregation in a
graph induces the clustering after projection. It could be
argued that, indeed, clustering is capturing this aggregation,
Figure 3. More than one bipartite network can produce the same metabolite projection. (A) Metabolite projection of a three-step pathway
representing the gamma-phosphate transference from ATP to CDP. (B) The original bipartite graph with its standard biochemical representation
(below). (C, D) Different bipartite graphs originating fromnon-real reactions that produce metabolic projection like the real one. Notice that a loss
of information occurs in the projection.
R. Montan ez et al. Problems and paradigms
BioEssays 32:246256, 2010 Wiley Periodicals, Inc. 251
129
Captulo 4
2.0
28.0
Figure 4. E. coli metabolic network, degree, and clustering distributions recalculated from
(6)
(AD) and its ER random version (EH). Both
networks present 1509 reactions (labeled in black), 766 metabolites (red), and 5627 links. Node size is proportional to its degree. Power-law
decay of hCi versus k distribution is shown for metabolite projection E. coli metabolic network and for its respective ER model as calculated in
ref.
(31)
Logarithmic binning inset (B and F, respectively), reveals the power-law dependence. As a guide for the eye, the k
1
slope is indicated in
solid line. (C, G) The degree metabolite distribution and respective degree distribution for the metabolite projection inset. (D, H) Degree reaction
distribution and degree distribution of the reaction projection inset. ER distributions show the mean and standard deviation of a set of 100 ER
graphs.
Problems and paradigms R. Montan ez et al.
252 BioEssays 32:246256, 2010 Wiley Periodicals, Inc.
130
2.0
28.0
Ral Montaez Martnez
but actually this gives no additional information to the one
already captured by the bipartite graph. An interesting
question not yet answered is the likelihood that two
metabolites participating in a reaction can also participate
in another one. In this case, it seems obvious that clustering
measures specically thought for bipartite networks could
contribute to answer this question. In this context, a number of
approximations using bipartite graphs has been applied to
other real systems
(5053)
but not yet translated to metabolism.
As previously mentioned, the standard clustering may lead us
to wrong interpretations. This becomes more relevant when
properties based on clustering such as small-world and
hierarchical modularity are evaluated, as seen in the following
sections.
Is the metabolic small-world behavior an
artifact?
Small-world behavior is a common characteristic of real
networks. Its denition requires using both the average of
clustering coefcient (C) and minimal path length (L) (see
Box 1). For simple (non-bipartite) graphs, the small-world
property is often determined by comparing the original graph
with a random ER model having the same number of nodes
and links.
(8,54)
However, a problem arises when we deal with
bipartite graphs. Since bipartite networks do not exhibit
clustering, the presence of a small-world behavior is here
meaningless and only unipartite networks can be properly
evaluated. Therefore, a projection is required. Once the
projected network is obtained, it can be compared with its
associated graph obtained from the ER model. Notice that
such an ERmodel has the same number than nodes and links
of the projected network. But, what happens if we do the same
with the ER bipartite graph?
As mentioned above, clustering appears as a result of the
projection process and this also occurs when ER bipartite
graphs are projected. According to this, to establish a small-
world criterion we must compare this network with its
respective ER model derived from the projected bipartite
ER graph. A remarkable fact is that the projection of an ER
bipartite network, roughly satises a small-world pattern (see
Table 1 for a numerical example). Some questions now arise:
can we ensure that metabolic or even any bipartite graph is
small world when its random bipartite version also satises
this condition? Is the simple aggregation by reactions an
appropriate requisite to conclude that metabolism is small
world, or is it a side effect of the graph construction algorithm?
As Table 1 indicates, the E. coli metabolite projection
reveals a small-world behavior but also its ER bipartite
correlates when they were projected as described in Fig. 2C
(see also ref.
(7)
). According to the arguments provided here,
we cannot guarantee that the small world of metabolic
networks has a biologically meaningful interpretation (such as
an optimization of the metabolic ux) since a null bipartite
model also fullls such a condition. On the other hand, we
cannot reject the idea that metabolisms are organized within
the small-world constraints. Clustering of real networks is
greater than that of their bipartite random null model
counterparts (Table 1), thus providing indirect evidence of
some local association of metabolism beyond a simple
aggregation effect. However, as small world is a qualitative
behavior, such a comparison cannot be properly done.
Alternative projections have been described, such as that
displayed in Fig. 2D, that also share small-world beha-
vior.
(33,44,45)
However, the impact of such a projection on
clustering has not yet been explored in depth. The main
limitation is that it requires additional considerations beyond
the standard ER bipartite model, such the estimation of what
number of substrates and products are expected by chance
Table 1. Small-world criterion is satised in both E. coli metabolic network and in its random null model counterpart. Metabolite and reaction
projections of the E. coli bipartite network (1), and the respective projections of an ER bipartite model with an identical number of metabolites,
reactions, and links of the original network (2), are compared with an ER random graph obtained from the (metabolite or reaction) projection (3).
Data from the E. coli network construction were obtained from ref.
(6)
and the resulting network is depicted in Fig. 3. According to the denition,
small-world criterion (hCi
(1 or 2)
>>hCi
ER
and L
(1 or 2)
$L
ER
) is evaluated by comparison with an ERrandomgraph obtained fromthe (metabolite or
reaction) projection (3). In other words, an ER unipartite graph is constructed preserving the same number of nodes and links of the projected
network. One hundred of random graphs were generated in each case. Mean and SD for each set were calculated for hCi and L values. The
results reveal that a real network (1), but also its random bipartite null model (2), are small-worlds when projected. Remarkably, the bipartite
nature of a graph is in this case just enough to satisfy the small-world criterion.
Real network
projection (1)
ER bipartite graph
projection (2)
ER from projected
graph (3)
L hCi L hCi L hCi
Metabolite projection
(762 nodes, 5627 links)
2.57 0.67 2.380.007 0.190.003 3.170.003 0.010.001
Reaction projection
(1506 nodes, 170973 links)
1.82 0.72 2.620.006 0.360.006 1.840.0 0.150.001
R. Montan ez et al. Problems and paradigms
BioEssays 32:246256, 2010 Wiley Periodicals, Inc. 253
131
Captulo 4
2.0
28.0
for every reaction. Further research in this direction would
contribute to addressing our open questions.
Summarizing, the main problem is that the standard
denition of small world is not suitable for bipartite networks
and, therefore, the search for a more convenient small-world
criterion for these graphs may provide an answer as to
whether metabolic networks are really small worlds. Accord-
ing to this idea, some clustering coefcient denitions have
been proposed for bipartite graphs, but little of its relevance
for the small-world property has been explored.
(52,53)
It is
worth stressing that other arguments, such as modularity, go
in favor of the local association required for small-worldness
but as seen below, the effect of hierarchical modularity also
may be affected by a wrong interpretation of clustering in the
graph projections.
Do metabolic networks exhibit
hierarchical modularity?
It has been suggested (and widely used) that an indicator of a
hierarchical organization is the power-law dependence of hCi
versus k.
(31,37,49,55)
Clustering captures the cliquishness of a
node (measuring the probability that two nodes sharing a
common neighbor are also connected among them). It was
found that degree and clustering are negatively correlated:
nodes with low k are more likely to form clusters than those
with high degree. This idea is captured by a toy model of graph
generation. This is illustrated in Box 2 where a complex,
fractal-like network is generated by iteratively reconnecting an
initial subgraph. As this model is far from realistic, it has been
used as a metaphor for complex networks displaying modular
and hierarchical scale-free organization. The topological
analysis of this web reveals a C(k) decay, namely:
Ck $ k
a
with a1. Such a scaling lawhas been observed in a number
of real systems, although with different scaling exponents,
(56)
and this power-law dependence has been widely adopted as
an indicative for hierarchy. In addition, it has been suggested
that scale-free behavior is necessary, but not sufcient,
condition to produce such scaling behavior.
(55)
Actually,
metabolic networks (or rather metabolite projections) exhibit
such a scale-freeness and a clustering decay close to that
predicted by the model. Although graph construction can
enhance clustering, we show in Fig. 4B that a standard
metabolite projection (as shown in Fig. 2B) is quite close to a
power-law decay with a1.
Considering the effect of graph projection on clustering,
the following question has to be formulated. Does projection
introduce some bias in the C(k) distribution? As with the small-
world pattern, Fig. 4F indicates that projection is just enough
to produce a power-lawdependence on C(k). It is worth noting
that such a decay in the projected ERbipartite model is not so
pronounced as in the real (projected) network (see Figs. 4B
and F). Actually, such a dependence occurs in the ERbipartite
null model even without satisfying the scale-free condition and
being dramatically distant from the prototypical hierarchical
one (see Box 2). Is the hierarchical criterion correct? Beyond
this general question, since we cannot establish the
contribution of projection on clustering decay, we cannot
properly talk about hierarchy in metabolic networks. Conse-
quently, a criterion for hierarchy avoiding the use of projections
is required and further research in this direction would
contribute to clarifying whether metabolic networks are truly
hierarchical.
Similarly, modular measures, particularly those based on
clustering coefcient measure,
(31,57,58)
may be affected by the
projection effect. In this context, modularity detection in
bipartite graphs would contribute to clarifying the modular
character of metabolism. Methods to detect modularity in
bipartite networks have been recently published,
(59,60)
representing a promising alternative to previous approaches.
The bipartite network approach
As discussed above, some well-established measures for
bipartite graphs offer an ideal framework for the study of
metabolism, keeping the relation between metabolites and
reactions.
An alternative to commonly applied measures in unipartite
graphs is the strength
(48)
(see Box 1 for denition). In the
case of metabolism, it is possible to gure out howimportant a
metabolite can be through its degree. However, one can
imagine that, for instance, not only the ATP hub but also the
ATP-ADP pair is biologically relevant, since it represents a
pair transformation shared by large number of reactions. This
information is captured by the strength, and its biological
interpretation is straightforward. As metabolism occurs by
transformations, strength would indicate the importance of a
substrate-product pair due to its participation in a large
number of reactions. However, in spite of the potential of
strength, it has not been applied to metabolism so far.
Strength and the use of a suitable version of bipartite
clustering could provide new insights on metabolic organiza-
tion. In addition, a number of properties such as modularity or
small-world behavior should be revisited according to a
bipartite view, avoiding the undesirable effects associated
with projections.
In addition, the bipartite representation offers a better
picture of metabolism since direction and stoichiometry can
be included to produce a more reliable view of metabolism.
The study of directed bipartite paths, cycle detection, and
degree correlations represent still unexplored ways in the
study of metabolic maps.
Problems and paradigms R. Montan ez et al.
254 BioEssays 32:246256, 2010 Wiley Periodicals, Inc.
132
2.0
28.0
Ral Montaez Martnez
Conclusions and perspectives
Topological studies of metabolism have been developed over
the last few years by using graph representations of
pathways. In this work we have shown that a network
description imposes strong constraints to our abstraction of
reality. In the case of metabolism, the use of unipartite
versions can lead to wrong interpretations of some of the most
relevant graph attributes. Average degree and degree
distributions are paradigmatic examples of this problem.
Instead, the bipartite view offers a cleaner interpretation of
topological features. A note of caution is thus needed in
relation to the effect of projection on clustering and how this
produces some properties for free, such as small-world
structure and hierarchy. Finding universalities in complex
networks is an intriguing and fascinating issue. However, this
is meaningful only when real systems are conveniently
mapped into a network abstraction. In this context, the more
natural representation of metabolism as a bipartite graph and
the development of appropriate measures based on such a
representation are much needed.
Acknowledgments: This work was supported by the EU 6
th
framework project ComplexDis (NEST-043241, CRC), James
S. McDonnell Foundation and Santa Fe Institute (RVS),
grants SAF 2008-02522 (Spanish Ministry of Science and
Innovation), Fundacio n Ramo n Areces, P07-CVI-02999 and
group BIO-267 (Andalusian Government). The CIBER de
Enfermedades Raras is an initiative of the ISCIII (Spain). The
authors would like to thank Bernat Corominas-Murtra,
Francisca Sa nchez-Jime nez, and Alejandro Real-Chicharro
for useful comments.
References
1. Michal G. 1993. Biochemical Pathways (poster). Boehringer Mannheim:
Penzberg.
2. Michal G. 1998. On representation of metabolic pathways. Biosystems 47:
17.
3. Kanehisa M, Araki M, Goto S, et al. 2008. KEGG for linking genomes to
life and the environment. Nucleic Acids Res 36: D48084.
4. Caspi R, Foerster H, Fulcher CA, et al. 2008. The MetaCyc Database of
metabolic pathways and enzymes and the BioCyc collection of Pathway/
Genome Databases. Nucleic Acids Res 36: 62331.
5. Ouzounis CA, Karp PD. 2000. Global properties of the metabolic map of
Escherichia coli. Genome Res 10: 56876.
6. Jeong H, Tombor B, Albert R, et al. 2000. The large-scale organization of
metabolic networks. Nature Oct 5; 407: 65154.
7. Wagner A, Fell DA. 2001. The small world inside large metabolic net-
works. Proc Biol Sci 268: 180310.
8. Watts DJ, Strogatz SH. 1998. Collective dynamics of small-world net-
works. Nature 393: 44042.
9. Albert R, Jeong H, Barabasi AL. 1999. Internet: diameter of the world-
wide web. Nature 401: 13031.
10. Baraba si AL. 2009. Scale-free networks: a decade and beyond. Science
325: 41213.
11. Koonin EV, Wolf YI, Karev GP. 2006. Power Laws, Scale-Free Networks
and Genome Biology (Molecular Biology Intelligence Unit). Berlin:
Springer.
12. Csermely P. 2006. Weak links. Stabilizers of complex systems from
proteins to social networks. Berlin: Springer.
13. Ke pe` s F. 2007. Biological networks, Complex systems and interdisciplin-
ary science Vol. 3. New Jersey: World Scientic.
14. Alberts B, Johnson A, Lewis J, et al. 2002. Molecular Biology of the Cell,
4th edn. New York: Garland Science.
15. Wagner A. 2002. Asymmetric functional divergence of duplicate genes.
Mol Biol Evol 19: 176068.
16. Sole RV, Pastor-Satorras R, Smith ED, et al. 2002. A model of large-
scale proteome evolution. Adv Complex Syst. 5: 4354.
17. Rodr guez-Caso C, Medina MA, Sole RV. 2005. Topology, tinkering and
evolution of the human transcription factor network. FEBS J 272: 642334.
18. Rodr guez-Caso C, Corominas-Murtra B, Sole RV. 2009. On the basic
computational structure of gene regulatory networks. Mol Biosyst 5:
161729.
19. Knight CG, Pinney JW. 2009. Making the right connections: biological
networks in the light of evolution. BioEssays 31: 108090.
20. Montoya JM, Pimm SL, Sole RV. 2006. Ecological networks and their
fragility. Nature 442: 25964.
21. Bullmore E, Sporns O. 2009. Complex brain networks: graph theoretical
analysis of structural and functional systems. Nat Rev Neurosci Mar; 10:
18698.
22. Sole RV, Corminas-Murtra B, Valverde S, et al. Language networks:
their structure, function and evolution. Complexity (in press).
23. Mariolis P. 1975. Interlocking directorates and control of corporations: the
theory of bank control. Soc Sci Q 56: 42539.
24. Newman MEJ. 2001a. Who is the best connected scientist? A study of
scientic coauthorship networks. Phys Rev E 64: 016131.
25. Newman MEJ. 2001b. Scientic collaboration networks. II. Shortest
paths, weighted networks, and centrality. Phys Rev E 64: 016132.
26. Davis GF, Yoo M, Baker WE. 2001. The small world of the corporate elite.
New Orleans: Social Networks Sunbelt XXI Conference.
27. Bascompte J, Jordano P. 2007. Plant-animal mutualistic networks: the
architecture of biodiversity. Annu Rev Ecol Evol Syst 38: 56793.
28. Rezende E, Lavabre J, Guimaraes P, et al. 2007. Non-random coex-
tinctions in phylogenetically structured mutualistic networks. Nature 448:
9258.
29. Goh KI, Cusick ME, Valle D, et al. 2007. The human disease network.
Proc Natl Acad Sci U S A 104: 868590.
30. Yildirim MA, Goh KI, Cusick ME, et al. 2007. Drug-target network. Nat
Biotechnol Oct; 25: 111926.
31. Ravasz E, Somera AL, Mongru DA, et al. 2002. Hierarchical organization
of modularity in metabolic networks. Science 297: 15515.
32. Burgos E, Ceva H, Hernandez L, et al. 2008. Two classes of bipartite
networks: nested biological and social systems. Phys Rev E 78: 046113.
33. Ma HW, Zeng AP. 2003. The connectivity structure, giant strong
component and centrality of metabolic networks. Bioinformatics 19:
142330.
34. Bilke S, Peterson C. 2001. Topological properties of citation and meta-
bolic networks. Phys Rev E Stat Nonlin Soft Matter Phys 64: 036106.
35. Newman ME, Watts DJ, Strogatz SH. 2002. Random graph models of
social networks. Proc Natl Acad Sci U S A 99: 256672.
36. Newman ME, Park J. 2003. Why social networks are different from other
types of networks. Phys Rev E 68: 036122.
37. Va zquez A. 2003. Growing network with local rules: preferential attach-
ment, clustering hierarchy, and degree correlations. Phys Rev E 67:
056104.
38. Ramasco JJ, Dorogovtsev SN, Pastor-Satorras R. 2004. Self-organiza-
tion of collaboration networks. Phys Rev E 70: 036106.
39. Arita M. 2004. The metabolic world of Escherichia coli is not small. Proc
Natl Acad Sci U S A 101: 15437.
40. Baraba si AL, Albert R. 1999. Emergence of scaling in random networks.
Science 286: 50912.
41. van Helden J, Wernisch L, Gilbert D, et al. 2002. Graph-based analysis
of metabolic networks. Ernst Schering Res Found Workshop 24574.
42. Zhou T, Ren J, Medo M, et al. 2007. Bipartite network projection and
personal recommendation. Phys Rev E 76: 046115.
R. Montan ez et al. Problems and paradigms
BioEssays 32:246256, 2010 Wiley Periodicals, Inc. 255
133
Captulo 4
2.0
28.0
43. Lacroix V, Cottret L, The bault P, et al. 2008. An introduction to metabolic
networks and their structural analysis. IEEE/ACM Trans Comput Biol
Bioinform 5: 594617.
44. Zhao J, Yu H, Luo JH, et al. 2006. Hierarchical modularity of nested bow-
ties in metabolic networks. BMC Bioinformatics 7: 386.
45. Szymanski J, Jozefczuk S, Nikoloski Z, et al. 2009. Stability of metabolic
correlations under changing environmental conditions in Escherichia
coli A systems approach. PLoS One 4: e7441.
46. Light S, Kraulis P, Elofsson A. 2005. Preferential attachment in the
evolution of metabolic networks. BMC Genomics Nov 10; 6: 159.
47. Guillaume JL, Latapy M. 2006. Bipartite graphs as models of complex
networks. Physica A 371: 795813.
48. Tanaka R. 2005. Scale-rich metabolic networks. Phys Rev Lett 94:
168101.
49. Albert R. 2005. Scale-free networks in cell biology. J Cell Sci 118: 494757.
50. Borgatti SP, Everett MG. 1997. Network analysis of 2-mode data. Soc
Networks 19: 24369.
51. Robins G, Alexander M. 2004. Small worlds among interlocking direc-
tors: network structure and distance in bipartite graphs. Comput Math
Organization Theory 10: 6994.
52. Lind PG, Gonza lez MC, Herrmann HJ. 2005. Cycles and clustering in
bipartite networks. Phys Rev E 72: 056127.
53. Zhang P, Wang J, Li X, et al. 2008. Clustering coefcient and community
structure of bipartite networks. Physica A 387: 686975.
54. Dorogovtsev S, Mendes J. 2003. Evolution of networks: From biological
nets to the Internet and WWW. Oxford: Oxford University Press.
55. Baraba si AL, Oltvai ZN. 2004. Network biology: understanding the cells
functional organization. Nat Rev Genet Feb; 5: 10113.
56. Ravasz E, Barabasi AL. 2003. Hierarchical organization in complex
networks. Phys Rev E 67: 026112.
57. Clauset A, Newman MEJ, Moore C. 2004. Finding community structure in
very large networks. Phys Rev E 70: 066111.
58. Yip AM, Horvath S. 2007. Gene network interconnectedness and
the generalized topological overlap measure. BMC Bioinformatics 8:
22.
59. Guimera` R, Sales-Pardo M, Amaral LA. 2007. Module identication in
bipartite and directed networks. Phys Rev E Stat Nonlin Soft Matter Phys
76: 036102.
60. Lehmann S, Schwartz M, Hansen LK. 2008. Biclique communities. Phys
Rev E 78: 016108.
61. Bolloba s B. 1979. Graph theory: an introductory course. New York:
Springer Verlag.
62. Erdo s P, Renyi A. 1959. On random graphs. I Publ Math 6: 290.
63. Milgram S. 1967. The small-world problem. Psycol Today 2: 607.
Problems and paradigms R. Montan ez et al.
256 BioEssays 32:246256, 2010 Wiley Periodicals, Inc.
134
2.0
28.0
Ral Montaez Martnez
135
Captulo 4
2.0
28.0
Analyzing the bipartite structure of the metabolic network
Montaez R
1
, Real-Chicharro A
1
, Snchez-Jimnez F
1
, Medina MA
1
. and Rodrguez-Caso C
2
.
1
Departamento de Biologa Molecular y Bioqumica, Facultad de Ciencias, Universidad de Mlaga, E-29071 Mlaga, and CIBER de Enfermedades Raras
(CIBERER), Mlaga, SPAIN.
2
ICREA-Complex Systems Lab, Universitat Pompeu Fabra. Parc de Recerca Biomdica de Barcelona. Dr. Aiguader 88, 08003. Barcelona Spain.
Abstract
Metabolism is a natural bipartite network that has always been treated as a one-mode network with the focus on
compounds or reactions. Projections of bipartite networks to obtain one-mode versions reduce the information
contents in the system and modify topological properties, such as density, degree and clustering coefcient. In
consequence, some topological generalizations assumed for metabolic networks must be reviewed in the light of a
bipartite graph analysis. In this work, we provide a comparative analysis of one-mode and bipartite statistics
performed in E. coli, S. cerevisiae and H. sapiens metabolic networks. We show that some bipartite-oriented
measures such as strength or bipartite clustering recover some relevant information loss in one-mode
representations. Our work highlight that a biologically consistent denition of metabolic network according to its
bipartite nature provides a more accurate biological interpretation of its topology and more tted values of its
topological features.
Introduction
Metabolism is by far the best-known cellular
network. However, its global pattern of connections
has not raised the interest of biologists until recently
[10, 12, 26]. The reason of this shift was the nding
that real networks display a number of previously
unknown universal patterns shared by very
disparate systems that can be characterized by their
graph topology. In this way, patterns such as small-
world [38] and scale-free organization [1, 5] are
measured, modelled and used in order to explore
the potential constrains that have guided these
webs through their evolution [31-33, 36]. Such a
picture has pervaded a large part of the literature in
the eld [7, 14, 15], even at the level of standard cell
biology textbooks [3]. It is noteworthy that
metabolism was one of the rst real networks
identied as both a scale-free [12] and a small-world
system [37], with some controversy [4], and it is also
considered a paradigmatic example of hierarchical
modular organization [28].
In a nutshell, a graph can be described as a
mathematical abstraction of reality where nodes are
individual units (metabolites of a reaction, species,
proteins, actors of a lm or websites) that appear
linked by an edge if some type of relation exists
between them. In this process, the real system is
represented by a connection pattern, thus ignoring
most of the details, such as kinetics parameters or
stoichiometric rates of change, in favour of a
systemic perspective. However, the process of
graph construction is an important issue that can
bias the conclusions derived from the topological
approach.
Interestingly, the topological properties resulting of
these abstractions (scale-free, small-world and
hierarchical modularity) t with the widely accepted
view of metabolism from a biological perspective.
ATP and NADH are examples of hubs (the most
connected nodes) in the metabolic network [12] and
their topological importance matches their relevant
roles in metabolism. These metabolites participate
in most of metabolic pathways, whereas many
metabolites take part in unique steps of particular
pathways, in agreement with the scale-free
behaviour. The organization of metabolism in
coupled reactions and its traditional classication in
pathways agree with the idea of local association
and modularity. In this context, cross-links between
pathways and the exi stence of ubi qui tous
metabolites -the hubs- would contribute to the small-
world effect and a hierarchical modular organization
that could be the justiable argument of optimality in
the evolution process.
However, as mentioned above, the metabolic
network is an abstraction of reality and the way to
build it can determine possible derived conclusions.
Metabolic maps, along with collaboration [8, 19, 21,
22, 25], mutualistic [6, 29], gene-disease [11] or
drug-target [39] networks belong to the class of so-
called bipartite graphs. Metabolites are transformed
by reactions. A special feature of these networks is
that metabolites are not interacting in pairs, as it
happens for example with the protein interaction
map. In this way, metabolic networks can be
abstracted as graphs with two disjoint sets of nodes,
1
136
2.0
28.0
Ral Montaez Martnez
metabolites and reactions, for a more accurate
biological description. In a reaction, a number of
substrates (usually more than one) react to produce
a number of products. According to this, the
connection of metabolites or reactions by pairs
provides a poor representation of reality since this
does not capture properly the metabolite conversion
by reactions, as bipartite graphs do. This bipartite
view of metabolism has been commonly addressed
in different works [12, 16, 34, 37, 41]. In spite of this,
the seminal works concerning the topological
organization of metabolism (small-world, hierarchy
and scale-free features) were performed using
projections from the bipartite network to its one-
mode counterparts [28, 37]. Although these
properties seem to be reasonable for metabolism,
the truth is that network projections introduce a
strong bias in the results of the topological analysis
[23, 24, 27, 35]. However, such biases have been
addressed only within the eld of social networks
[23, 24, 27, 35]. The aim of the present work is to
show that a biologically consistent denition of
metabolic network according to its bipartite nature
provides a more accurate biological interpretation of
its topology and more tted values of its topological
features.
Methodology
Data obtaining
Metabolic networks used here have been retrieved
from the Kyoto Encyclopaedia of Genes and
Genomes (KEGG) [13]. We have included in our
metabolic system only those processes contained in
the metabolic map labelled by 01100 (Metabolic
Pathway) in KEGG database. This label was used to
recover the reaction information for each one of the
three organisms included in this research: E. coli, S.
cerevisiae and H. sapiens. According to this,
reactions related to cell signalling and, in general,
any protein or DNA modication were excluded for
our study. In order to avoid unrealistic processes, all
the reactions included in every network were
identied to occur in the respective organism.
Additionally, we checked for every analyzed
metabolic network that all the reactions are non-
generic reactions and that any reaction includes
generic compounds as alcohol (C00069) or peptide
(C00012).
Metabolic network denition
Acquired metabolic data were organized in
bipartite graphs according to the relation between
metabolites and reactions for every studied
organism
(1)
. The resulting graphs (G) were
undirected networks with a set of reaction nodes R=
{r1,r2,...,r|R|} and a set of metabolite nodes M=
{m1,m2,...,m|M|} interconnected by a set of edges E=
{e1,e2,...,e|E|}. A metabolite node mi is linked to a
reaction node rj if this metabolite is a reactant in this
reaction. Topological analyses were applied to the
largest connected component. As we shall see, this
is especially relevant to provide a meaningful value
of measures such as the average path length.
Elements within every set of nodes, labelled with the
KEGG nomenclature identier (R and ve numbers
for reacti ons and C and ve numbers for
metabolites) are presented as they are commonly
referred in biochemistry's handbooks. It is worth to
stress that metabolic networks used in this work
included the so-called ubiquitous metabolites such
as water ATP or H
+
, among others, not always
included in topological studies of metabolism [12,
37].
Null Model construction
A common way to evaluate the topological
properties of real networks is by comparison with
null models in which a number of topological
invariants are preserved. These null models are
treated and measured in the same way as the real
network, giving rise to a proper comparison of the
topological measures. In this work, real networks
were compared with two different null models. The
rst one represents a raw model in which only the
|M|, |R|, and |E| is conserved. This model is
equivalent to the so-called Erds-Rnyi (ER) [9]
graph, where nodes of a set are simple linked by a
single probability but formalized for bipartite
networks. The second one conserves, besides |M|,
|R|, and |E|, the degree sequence of the graph, i.e.
the exact number of links for every particular node
of the graph, giving an identical degree distribution
of the original network but destroying any local
correlation between nodes connectivity. Hereafter
we refer to these models as ER-rand and k-rand,
respectively.
In order to evaluate the bias introduced by the
projection process, we performed two alternative
methods to use the one-mode networks resulting
from a bipartite projection. In the rst method
(Method I), we rstly randomized the bipartite graph
and latter we projected the resulting graph. In the
second one (Method II), we projected the graph and
then randomized the one-mode projection.
Topological Measures
In this section we present the topological
denitions of the measures used in this work.
Given an undirected bipartite graph G(R,M,E), two
alternative one-mode graphs applying projection
2
(1) We use the standard notation for set theory. For the not-specialized reader we remind that R, M and E are sets
while the notation |...|, such as |R|, |M| and |E|, makes reference to set cardinalities, i.e. the number of elements of such
set).
137
Captulo 4
2.0
28.0
process can be performed. They are reaction and
metabolite graphs, noted respectively by GR(R,ER)
and GM(R,EM). In GR two reactions are connected if
they share a common metabolite; in GM two
metabolites are linked if they take part in the same
reaction.
Given an undirected bipartite graph G(R,M,E), the
degree -also called connectivity- of a node ni (where
n can be either m or r) , noted by k(ni), is dened as
the number of edges connected to this node.
Depending on the node type, two kinds of degree
can be dened. The degree of a metabolite
indicates the number of reactions in which it
participates, whereas the degree of a reaction
informs about the number of reactants. Formally, k
(ri) denotes the number of metabolites connected to
the reaction i whereas k(mi) indicates the number of
reactions connected to the metabolite i. According to
node degree, a highly connected node is a hub.
From individual nodes, the average degree for the
entire network is calculated as
<k>=2|E|/(|M|+|R|).
Alternatively, average degree can be calculated
for every node type as follows:
<kR>=|E|/|R|
and
<kM>=|E|/|M|.
In the same way, it is possible to dene the
probability distribution by dening the likelihood to
nd a node having k links. According to the bipartite
nature of the network, degree distributions for
reactions, p(kR), and metabolites, p(kM), are
dened.
Within the most basic graph statistics, the bipartite
density of a graph is dened as the fraction of
existing links with respect to the possible ones,
namely
!G=|E|/(|M||R|).
It has been recently suggested that the second
neighbourhood, k2, of a node could be an useful
measure in bipartite graphs [17] since it recovers the
degree information obtained from the analysis of
projected graphs. Given ni in a bipartite graph G
(M,R,E), k2(ni) is the number of nodes at distance
two from i (not including i). Analogous to degree
concept, <k2> and distribution (pk2) can be easily
calculated. Depending on which set of nodes we are
interested in, we can calculate these distributions for
reacti ons, pk2(R), or for metabol i tes, pk2(M).
Interesting it can be probed that
pk2(R)=pk(GR)
and
pk2(R)=pk(GM)
where GR and GM correspond with the reaction
graph projection and metabolite graph projection,
respectively.
Is noteworthy that the <k2> of the graph is a
measure that needs to estimate rstly the number of
edges in each node projection before to calculate it,
thus <k2>=2(|ER|+|EM|)/(|R|+|M|), being |ER| the
number of edges in the reaction projection and |EM|
the number of edges in the metabolites projection.
These statistics therefore offer a way to study how
node degrees appear in the projections, and to
distinguish between different behaviours, by plotting
the correlation between k and k2.
In order to measure distances between two nodes,
we dened two types of paths. Firstly, a path
between ni and nj -in principle no mattering the type
of node we are talking about- is a set of edges that
connect ni to nj. Then, the path length LG(r(i),r(j)) is
easi l y esti mated as the number of edges
participating in this path. In bipartite graphs, a
measure of path lengths between nodes of the
same type has a direct correlation in the measure of
path length in projected graph. The relation between
the path length between nR(i) and nR(j) in bipartite
and reaction-projected networks are established as
follows:
LG(r(i),r(j))=2LG(R)(r(i),r(j)),
where LG(R)(r(i),r(j)) denotes the path length in the
reaction network. A similar relation can be obtained
for metabolite projection. For random bipartite
graphs, the average path length follows the
expression
<LG>=log(|R|+|M|)/log(|E|).
The most distant nodes within the network are
measured with the diameter of the graph (DG) as the
minimal path length between the most distant
couple of nodes in the graph. The most distant
nodes in the metabolic graph must be always two
metabolites.
Strength (S) is a suitable generalization of the
degree concept for bipartite graphs since it
measures the total number of paths connecting
some given node with their second nearest
neighbourhood. We can measure strength for every
pair of nodes (Sij) within a given set, or the total
strength of a concrete node (Si) as Sij/k2i, where Sij
is the strength of all the possible pairs of nodes
connected to i in the second nearest neighbourhood
and being k2i the degree at distance two of i.
The clustering coefcient that has been used in
the Watts-Strogatz (WS) model measures the
probability that two neighbours of a selected node
are linked together. In a bipartite network, this
measure is always zero because connections
between nodes of the same set are forbidden. An
alternative possibility to measure the cliquishness in
3
138
2.0
28.0
Ral Montaez Martnez
bipartite graphs of a typical neighbourhood is
assessed by the probability that two neighbours of ni
are connected to another common neighbour.
Notice that this special connection pattern
corresponds wi th a square of four nodes.
Calculating the number of squares in which ni
participates, noted by Q(ni), clustering coefcient
(C4) for ni in bipartite graphs is dened as follows:
C4(ni)=2Q(ni)/(k(ni)(k(ni)-1))*k2(ni))
Where (ka) is the number of neighbours in the
opposite set directly linked to node i, or degree of i,
and (k2a) is the number of nodes in the same set of i
that i can achieve crossing the nearest nodes on the
opposite set, or second neighbourhood of node i.
This C4 clustering coefcient is based on Robins
square coefcient [30] reviewed previously [18, 40].
This measure can be performed over every set of
nodes; in consequence, the average clustering for
every set of nodes (<C4M> or <C4R>) and for the
complete graph <C4> can be determined.
For ER bipartite graphs, C4G follows the
expression C4G(ER)=<kG>/2*(nM+nR). Where <kG> is
the average connectivity of the graph. For a
particular set (for example, nodes of the metabolite
set) of a ER bipartite graph model it is established
that
C4M(ER)=2*<kM>/(<kM>nM + <kR>nR).
In a similar way, it can be calculated the same
measure for the reaction set.
Results and discussion
In this work, we analyzed the metabolic networks
of three organisms (H. sapiens, S. cerevisiae and E.
coli) in their bipartite and one-mode (projected)
representations. Because of the similarity in the
topologies,for illustration, results for Saccharomyces
cerevisiae are mainly presented in the text of this
paper.
The obtained graph of S. cerevisiae is formed by |
R|=697, |M|=641 and |E|=3,021 (corresponding to
697 reaction and 641 metabolite nodes connected
by 3,021 edges). The network shows three
connected components. The largest connected
component of this network embraces 1330 nodes
(695 reactions and 635 metabolites) and 3,015
edges. The remaining connected components were
much smaller: one of them has 5 nodes (R03867,
C00045, C02166, C03363 and C05951) and 4
edges, and the second one has only 3 nodes
(R05976, C00621 and C03021) and 2 edges. Notice
that both of them reproduce single reactions, that
they probably are wrongly annotated for this
organism in KEGG repository, since they appeared
associated to other reactions when they were
manually tested in the KEGG database. We can
assume that the resulting picture of yeast
metabolism (as well as in the other organisms
included in this study) is in fact a single component.
The rst trait that took our attention is the fact that
the number of reactions is similar to or even greater
than the number of metabolites. A common
observation in any handbook of biochemistry is that
most of the reactions detailed there display more
than two metabolites, usually four (this value
approximately corresponds with the average
number of metabolites per reaction, as we shall see
later). According to this perception, one may gure
out that the metabolism has more metabolites than
reactions. As we see, this is not true and the only
one possibility to have this ratio of metabolite/
reaction is by a very high reuse of the same
molecules in different reactions. In particular, those
that have an especial role in energy or electron
transfer such as ATP, ADP, NAD(P) and NAD(P)H
among others. As expected, they are hubs (highly
connected nodes) in our networks.
The topological analysis of the metabolic network
of S. cerevisiae revealed that its largest connected
component presented a very sparse density,
!G=0.0068, and a low average degree, <kR(G)
>=4.33 and <kM(G)>=4.74. One of the problems in
network projections is the increase of edges in the
resulting network. The metabolite projection of S.
cerevisiae gave a slight increase in the number of
edges (|EM|=3,440), producing a minor impact on
the density (!(GM)=0.009) but doubling the average
degree (<k(GM)>=10.83). This bias changes
dramatically the reaction projection giving a very
highly connected network with |R|=697 and |ER|
=45,999. This huge increase in the number of links
also produces a !(GR)=0.1 and <k(GR)>=132.37.
The differential behaviour of these alternative
projections is motivated by the distinct degree
distribution of the two sets of nodes. As it has been
previously reported [12, 20], metabolite degree
distribution exhibits a power-law distribution "=-2.1,
while the reaction distribution can be tted to a quite
narrow binomial function ranging from a minimum of
2 and a maximum of 8 (Figure 1 A and C). It can be
calculated that the contribution in the number of
links of a node mi with k(mi) in the GR goes as k(mi)
(k(mi)-1)/2. Thus, a large degree of removed nodes
introduces a non-linear and non-negligible bias in
the connectivity and degree distribution of projected
nodes in GR. For instance, H2O is a hub in the
bipartite network, participating in 197 reactions;
according to the previous equation, it was
responsible for 19,306 edges in the reaction
projection. This number of edges is not produced,
by far, by any reaction within the metabolite network.
In fact, the number of links that a reaction can
establish is rather negligible compared with
contributions of metabolite hubs in their respective
projection. In particular, a reaction with 8 reactants
(the maximum number observed in the network)
4
139
Captulo 4
2.0
28.0
only contributes with 28 edges in the metabolite
projection.
Another observation derived from the application
of projected networks is that the non-linear increase
of the number of edges of the projected network
pr oduces degr ee di st r i but i ons of di f cul t
interpretation, contrasting with the highly intuitive
picture derived from the bipartite network analysis
(see Figure 1). It is noteworthy that, bipartite degree
distributions give us straightforward information
about the relation between metabolites and
reacti ons. In contrast, proj ecti ons gi ve us
information that is already captured in the degree
distribution obtained for the second neighbourhood,
i.e. p(k2). The topological analysis of the yeast
metabolic network revealed a binomial distribution
with <k>R=4.34. Interestingly, this p(k) for reactions
in the bipartite network is a result of a chemical
r est r i ct i on. Most r eact i ons r epr esent t he
transformation by chemical collisions; these
chemical collisions make probabilities of multi-
reactants reactions very low. By this reason, they
have a limit in the number of reactants involved in
the transformation, from 3 to 8 or 9 (in human).
Besides, the case of unimolecular reactions, as
isomerization, is expected. The analysis of
metabolic networks revealed a maximum of 8
5
Figure 1.- Degree distribution of S. cerevisiae metabolic network A) Degree distribution of metabolite nodes in the bipartite
network, B) Degree distribution of metabolites in their projection counterpart, C) Degree distribution of reaction nodes in the
bipartite network, D) Degree distribution of reactions in their projection counterpart and E) their equivalent distributions in H.
sapiens and E. coli
E
140
2.0
28.0
Ral Montaez Martnez
reactants for S. cerevisiae. This is a reaction
catalysed by the O-acetyl-L-homoserine acetate-
lyase (O-Acetyl-L-homoserine + Thiosulfate +
Thioredoxin + H+ <=> L-Homocysteine + Sulte +
Thioredoxin disulde + Acetate).
When we observe the metabolite distribution
without the restrictions that operate in reactions, we
observe a heterogeneous distribution with a large
range of connections. This distribution follows a
power-law decay (Figure 1A) indicating that a
handful of metabolites have a very large number of
connections while most of the metabolite exhibits
very few connections. Inorganic metabolites as
water, oxygen or CO2, and cofactors as ATP, NAD or
SAM act as hubs, involved in an unbounded number
of specic types of chemical modications, applied
to a large density of substrates (such as hydrolysis,
oxidoreduction, and energy transfer or methyl
carrier).
Fell and Wagner justied power law distribution as
a consequence of a preferential attachment and
proposed this as the model of metabolic evolution: if
early in the evolution of life, metabolic network grew
by adding new metabolites, then the most highly
connected metabol i tes shoul d al so be the
phylogenetically oldest [10]. This scenario seems
reasonable for some of the highly connected
compounds involved in intermediary metabolism
(e.g. oxaloacetate, pyruvate, glutamate), as well as
for some amino acids pointed by the authors.
However, the strict application of this model would
lead to impossibilities, since it would imply that ATP
6
Figure 2.- Strength and k2 related to degree of S. cerevisiae. A) Strength of metabolites in the real network, B) Strength of
randomized network preserving the degree distribution, averaged over 1000 repetitions. C) Strength of a randomized network
just preserving the number of nodes and links of the original bipartite network, averaged over 1000 repetitions. D, E and F are
the equivalent plots for reactions. G) k2-k correlation of metabolites in the real network, H) k2-k correlation in k-rand networks
of metabolites, I) k2-k correlation in ER-rand networks of metabolites, J, K and L are the equivalent correlation for reactions.
141
Captulo 4
2.0
28.0
appeared before adenosi ne, S-adenosyl -L-
homocysteine before cysteine, etc. The preferential
attachment model may thus partly explain some
relationships between central and peripheral
metabolism, but should certainly not be considered
as the reason for the topological properties of the
whole network.
Based on the power-law and scale-free behaviour
of the degree distribution of metabolites, the
robustness and error tolerance of metabolism has
been proposed [2]. However, the number of
connections among metabolites has no relation with
robustness. If all the acetyl-CoA of a cell is removed,
the cell machinery stops. The suppression of a
single hub from the metabolic network requires
deleting or inactivating several hundreds of coding
genes or enzymes that produce this highly
connected metabolite. After a handful of such
mutations, the cell would suffer from the depletion of
its main enzymatic products (among which is the
hub and many other metabolites that are generally
poorly connected compounds) and would die. Thus,
the suppression of metabolic hubs from the network
by natural or even directed mutations is an
unrealistic scenario.
Table 1 presents a ranking of the metabolites with
the highest strengths and the most connected
metabolites as k(ni) and k2(ni). It should be kept in
mind that, as previously commented, k2 degree
behaviour in a bipartite network corresponds to k
behaviour in their projection. As Table 1 illustrates,
both k and k2 rankings of metabolites are very
similar, in spite of some minor variations. However,
in bipartite networks an additional information can
be obtained calculating the strength. The ranking of
strength compared with k and k2 provides the actual
relevance of metabolites such as NAD(P) and NAD
(P)H. We observe that these molecules are on the
top of the ranking although they are not the highest
connected ones. ATP and water are the highest
connected metabolites but they are not on the top of
the strength ranking. This observation stresses the
fact that the high degree is not enough to have a
large value of strength. We found that metabolites
such as allysine or saccharoporine (essentials for
lysine biosynthesis and degradation in yeast) and
7
Metabolite Strength (S) Metabolite Degree (k) Metabolite K2 degree (k2)
H+ 3,18 H2O 197 H2O 320
NAD
+
2,49 H+ 163 H+ 235
NADPH 2,47 ATP 100 ATP 166
NADP
+
2,47 NADPH 86 NADPH 158
NADH 2,47 NADP
+
86 NADP
+
158
Allysine 2,45 ADP 77 H3PO4 139
Dolichyl D-mannosyl phosphate 2,40 H3PO4 68 ADP 135
Saccharopine 2,40 NAD
+
66 NAD
+
122
L-Glutamine 2,32 NADH 64 NADH 118
L-Glutamate 2,31 Pyrophosphate 54 CO2 102
Folate 2,29 CO2 51 Pyrophosphate 99
ATP 2,27 L-Glutamate 48 L-Glutamate 88
H2O 2,27 CoA 46 CoA 87
H2O2 2,13 NH3 39 NH3 80
1-Pyrroline-5-carboxylate 2,10 2-Oxoglutarate 33 2-Oxoglutarate 61
1-Pyrroline-3-hydroxy-5-carboxylate 2,10 O2 26 O2 56
O2 2,09 AMP 26 Acetyl-CoA 55
ADP 2,06 Acetyl-CoA 26 AMP 52
2-Oxoglutarate 2,05 Pyruvate 22 Pyruvate 47
Reaction catalyzed by Strength (S) Reaction Degree (k) Reaction K2 degree (k2)
R01251,EC.1.5.1.2 2,0485 R00257,EC.6.3.5.1 8 R00257,EC.6.3.5.1 390
R03293,EC.1.5.1.2 2,0485 R00573,EC.6.3.4.2 8 R00177,EC.2.5.1.6 358
R04390,EC.1.2.1.31 2,0485 R00575,EC.6.3.5.5 8 R00248,EC.1.4.1.4 354
R02236,EC.1.5.1.3 2,0482 R00578,EC.6.3.5.4 8 R00243,EC.1.4.1.2 353
R00114,EC.1.4.1.13 2,0383 R01078,EC,2.8.1.6 8 R00578,EC.6.3.5.4 352
R01759,EC.1.1.1.21 2,0366 R01231,EC.6.3.5.2 8 R01231,EC.6.3.5.2 349
R05639,EC.1.3.1.70 2,0366 R02026,EC.2.5.1.49 8 R00573,EC.6.3.4.2 348
R07506,EC.1.14.14.- 2,0366 R03905,EC.6.3.5.7 8 R00707,EC.1.5.1.12 345
R08379,EC.1.1.1.- 2,0366 R04463,EC.6.3.5.3 8 R00708,EC.1.5.1.12 345
R01431,EC.1.1.1.21 2,0305 R04859,EC.2.5.1.47,2.5.1.49 8 R02313,EC.1.5.1.9 344
R05641,EC.1.3.1.71 2,0305 R05640,EC.1.14.13.70 8 R02315,EC.1.5.1.10 344
R07483,EC.1.3.1.70 2,0305 R05731,EC.1.14.13.70 8 R04463,EC.6.3.5.3 338
R07495,EC.1.1.1.270 2,0305 R00210,EC.1.2.1.51 7 R00575,EC.6.3.5.5 337
R01775,EC.1.1.1.3 2,0301 R00243,EC.1.4.1.2 7 R03905,EC.6.3.5.7 337
R00746,EC.1.1.1.71,EC.1.1.1.2 2,0298 R00248,EC.1.4.1.4 7 R00715,EC.1.5.1.7 336
R03458,EC.1.1.1.193 2,0244 R00352,EC.2.3.3.8 7 R00716,EC.1.5.1.8 336
R07505,EC.1.3.3.- 2,0242 R00688,EC.1.4.1.20 7 R00199,EC.2.7.9.2 335
R03821,EC.1.8.1.12 2,0241 R00715,EC.1.5.1.7 7 R05640,EC.1.14.13.70 326
R04430,EC.1.3.1.10 2,0241 R00716,EC.1.5.1.8 7 R05731,EC.1.14.13.70 325
R04725,EC.1.3.1.10,EC. 2,0241 R01088,EC.1.4.1.9 7 R00713,EC.1.2.1.16 324
Table 1.-Top 20 ranking of metabolites and reactions, according to their strength (S), metabolite degree (k) and to the second nearest
neighbourhood (k2). We remind that in EC codes the first number corresponds to: 1 oxidoreductase, 2 transferase and 6 ligase activities.
142
2.0
28.0
Ral Montaez Martnez
dolichyl D-mannosyl phosphate (a key element to
the GPI anchor in protein glycosylation) present a
high value of strength. With respect to reactions, it is
not ewor t hy t o hi ghl i ght how r eact i ons of
oxidoreduction, the foundations of anabolic and
catabolic processes, are in the top, attending to the
strength. However, degree gives us an uneven
group of reactions. On the other hand, the bias of
highly connected metabolites over the k2 could be
the basis of the differences between k and k2 top
reactions.
The relevance of a node in a graph is commonly
associated to its degree (either k1 or k2). However,
as we can observe for bipartite networks, strength
provides us complementary information. In Figure 2
we show that the neighbour correlations among
nodes of the same set are directly associated to
connectivity in metabolite projection, because all the
reactions have in average the same degree,
However, in GR the connectivity are related with the
degree of t hei r subst rat es and product s.
Correlations between k2(ni) k(ni), in real and k-rand
metabolic networks, shown as that k2 distribution in
metabolic networks is a simple consequence of the
degree distribution of each set of nodes. However,
correlation between S(ni) and k(ni) gives us
information about the topology of the network that
itself goes far beyond its degree distribution.
Cliquishness is another concept commonly
applied in graph organization. As we mentioned
above, the most common measure capturing this
idea is the clustering coefcient [38]. However, this
measure is meaningless when applied to bipartite
graphs directly, and projection is usually applied to
study cliquishness. Thus, C4 denition (see methods
section) allows us to avoid projections. In this way,
cliquishness can be directly estimated from bipartite
networks. Table 2 illustrates the behaviour of
clustering coefcient (C) and C4 when metabolite
network is compared with the set of different
randomized networks described in methodology
section. In a rst look, we observe that any measure
of cliquishness has a higher value than the
observed in random graph versions. This provides
a strong evidence for a more clustered organization
in metabolic network than expected by chance.
When we compared the real and ER bipartite
networks, the absolute values for C4 are quite
smaller than for C. However, the differences
between C4 values are widely greater than the
observed when clustering coefcients are evaluated
in the projected networks. In this case, C4, as they
are directly calculated from the original bipartite
network produces a more reliable estimation of
cliquishness in both set.
The evaluation of clustering coefcient, together
with minimal path length average (as in Table 2), is
a very important issue since it makes part of the
t opol ogi cal cri t eri on f or smal l -worl d graph
identication (see method section). Metabolic
network has been proposed as a paradigmatic
example of small-world [37]. However, this result
comes from the use of projection where C is an
artifact of graph projection. In fact, we have recently
reported that graph projection process may induce a
wrong percept i on of smal l -worl dness [ 20]
suggesting that the small-world character of
metabolism might be revisited. As we have shown,
C4 is an alternative to overcome the effect of
projection. Our results provide strong evidences that
local organization is higher than the expected by
chance. However, minimal path length must be
veried to be similar than the expected in a ER. At
this point, further research on the ability of C4 to
reproduce the small-world effect, as described in
[37], is required to establish a proper evaluation of
the small-world in bipartite graphs.
Concluding remarks
Problems -raised from the social science- related
to the use of one-mode networks instead of their
natural two-mode counterpart include the loss of
information stored in the abstraction and biased
8
Method II <C|M|> <C|M|-ER> <C|M|-ER>/ <C|M|> <LGM> <LGM-ER> <C|R|> <C|R|-ER> <C|R|-ER>/ <C|R|> <LGR> <LGR-ER>
E. coli 0.69 0.02 0.03
2.52 2.96 0.77
0.24 0.31
1.95 1.81
S. cerevisiae 0.70 0.02 0.03
2.56 2.94 0.78 0.25
0.32
1.97 1.80
H. sapiens 0.71 0.01 0.01
2.59 3.08 0.82 0.27
0.33
2.0 1.80
Method I <C|M|> <C|M|-ER> <C|M|-ER>/ <C|M|> <LGM> <LGR-EM> <C|R|> <C|R|-ER> <C|R|-ER>/ <C|R|> <LGR> <LGR-ER>
E. coli 0.69 0.26 0.36
2.52 2.68 0.77
0.27 0.35
1.95 2.63
S. cerevisiae 0.70 0.27 0.39
2.56 2.64 0.78 0.27
0.35
1.97 2.57
H. sapiens 0.71 0.24 0.34
2.59 2.70 0.82 0.28
0.34
2.0 2.67
Bipartite <C4|M|> <C4|M|-ER> <C4|M|-ER>/ <C4|M|> <L|M|> <L|M|-ER> <C4|R|> <C4|R|-ER> <C4|R|-ER>/ <C4|R|> <L|R|> <L|R|-ER>
E. coli 0.05 1.2E-3 0.02
2.52 2.68 0.05
1.3E-3 0.03
1.95 2.63
S. cerevisiae 0.07 1.4E-3 0.02
2.56 2.64 0,06 1.4E-3
0.03
1.97 2.58
H. sapiens 0.08 1E-3 0.01
2.59 2.70 0.06 1.1E-3
0.02
2.0 2.68
Table 2.- , Difference on cliquishness behaviours and distances among organisms. Every bipartite node set was compared with there equivalent
counterparts obtained from randomization process, Method I (ER-rand of bipartite network and subsequent projection) and Method II (Projection
follow of the ER-rand process). Differences was compared using the rate of change C4|N|-ER/ C4|N|, where C4|N|-ER is the average bipartite clustering
of ER-rand networks, for every set of nodes, and C4|N| is the average bipartite clustering of real network (Real Values are outside of the standard
deviations over 100 repetitions).
143
Captulo 4
2.0
28.0
behaviours, with the subsequent risks of delivering
wrong conclusions. The ination suffered by one-
mode pr oj ect ed net wor ks has i mpor t ant
consequences for density, degree distributions,
clustering coefcients and, of course, the biological
interpretation of their topological characterization.
Results shown in the present work help to
understand metabolism as an integrated network
with dependence of the relation between substrates
and products. For example, the average degree of
reactions is expected to be around four neighbours,
two reactions and two products. Metabolites
involved in general transformations are expected to
have a high number of neighbours, something
already written in general biochemistry texts. To
understand which constraints determine the
relationships between reactions and metabolites in
the global organization of metabolism and how they
can affect the topology of the projected networks
would be the next aim to be fullled.
Acknowledgments
This work was supported by the EU 6th framework
project ComplexDis (NEST-043241, CRC), grants
SAF 2008-02522 (Spanish Ministry of Science and
Innovation), Fundacin Ramn Areces, P07-
CVI-02999 and group BIO-267 (Andalusian
Government). The CIBER de Enfermedades
Raras is an initiative of the ISCIII (Spain). The
authors would like to thank Bernat Corominas-
Murtra for useful comments.
Bibliography
1.! Albert R, Jeong H, and Barabasi AL. Diameter of
the World-Wide Web. Nature. 1999;401: 130-131.
2.! Albert R, Jeong H, and Barabasi AL. Error and
attack tolerance of complex networks. Nature.
2000;406(6794): 378-382.
3.! Alberts B, Jhonson A, Lewis J, Raff M, Roberts K,
and Walter P, Molecular Biology of the Cell. 5 ed.
Vol. 1. 2008: Garland Science.
4.! Arita M. The metabolic world of Escherichia coli is
not small. Proc Natl Acad Sci U S A. 2004;101(6):
1543-1547.
5.! Barabasi AL. Scale-free networks: a decade and
beyond. Science. 2009;325(5939): 412-413.
6.! Bascompte J and Jordano P. Plant-Animal
Mutualistic Networks: The Architecture of
Biodiversity Annual Review of Ecology,
Evolution, and Systematics. 2007;38: 567-593.
7.! Csermely P, Weak Links: Stabilizers of Complex
Systems from Proteins to Social Networks. 1 ed. Vol.
1. 2006: Springer.
8.! Davis GF, Yoo M, and Baker W. The Small World
of the American Corporate Elite, 1981-2001.
Strategic Organization. 2003;1(3): 301-326.
9.! Erds P and Rnyi A. On Random Graphs. I.
Publicationes Mathematicae. 1959;6(290-297).
10.!Fell DA and Wagner A. The small world of
metabolism. Nat Biotechnol. 2000;18(11):
1121-1122.
11.!Goh KI, Cusick ME, Valle D, Childs B, Vidal M,
and Barabasi AL. The human disease network. Proc
Natl Acad Sci U S A. 2007;104(21): 8685-8690.
12.!Jeong H, Tombor B, Albert R, Oltvai ZN, and
Barabasi AL. The large-scale organization of
metabolic networks. Nature. 2000;407(6804):
651-654.
13.!Kanehisa M, Araki M, Goto S, Hattori M,
Hirakawa M, Itoh M, Katayama T, Kawashima S,
Okuda S, Tokimatsu T, and Yamanishi Y. KEGG
for linking genomes to life and the environment.
Nucleic Acids Res. 2008;36(Database issue):
D480-484.
14.!Kps F, Biological Networks (Complex Systems and
Interdisciplinary Science). 1 ed. Vol. 1. 2007, New
Jersey: Worls Scientic publishing Company.
15.!Koonin EV, Wolf YI, and P. KG, Power Laws, Scale-
Free Networks and Genome Biology. 1 ed. Molecular
Biology Intelligence Unit, ed. Springer. 2006.
16.!Lacroix V, Cottret L, Thebault P, and Sagot MF.
An introduction to metabolic networks and their
structural analysis. IEEE/ACM Trans Comput
Biol Bioinform. 2008;5(4): 594-617.
17.!Latapy M, Magnien C, and Del Vecchio N.
BasiNotions for the analysis of large two-mode
networks. Social Networks. 2008;30: 31-48.
18.!Lind PG, Gonzalez MC, and Herrmann HJ.
Cycles and clustering in bipartite networks. Phys
Rev E Stat Nonlin Soft Matter Phys. 2005;72(5 Pt
2): 056127.
19.!Mariolis P. Interlocking directorates and control of
corporations: the theory of bank control. Soc. Sci. Q.
1975;56: 425-439.
20.!Montaez R, Medina MA, Sol R, and
Rodrguez-Caso C. When metabolism meets
topology:reconciling metabolite and reaction
networks. BioEssays. 2010;In-Press.
21.!Newman ME. Scientic collaboration networks. I.
Network construction and fundamental results. Phys
Rev E Stat Nonlin Soft Matter Phys. 2001;64(1 Pt
2): 016131.
22.!Newman ME. Scientic collaboration networks. II.
Shortest paths, weighted networks, and centrality.
Phys Rev E Stat Nonlin Soft Matter Phys. 2001;64
(1 Pt 2): 016132.
23.!Newman ME and Park J. Why social networks are
different from other types of networks. Phys Rev E
Stat Nonlin Soft Matter Phys. 2003;68(3 Pt 2):
036122.
24.!Newman ME, Watts DJ, and Strogatz SH.
Random graph models of social networks. Proc Natl
Acad Sci U S A. 2002;99 Suppl 1: 2566-2572.
9
144
2.0
28.0
Ral Montaez Martnez
25.!Newman MEJ. Who is the best connected scientist?
A study of scientic coauthorship network.
arXiv:cond-mat/0011144v2. 2000.
26.!Ouzounis CA and Karp PD. Global properties of
the metabolic map of Escherichia coli. Genome Res.
2000;10(4): 568-576.
27.!Ramasco JJ, Dorogovtsev SN, and Pastor-
Satorras R. Self-organization of collaboration
networks. Phys Rev E Stat Nonlin Soft Matter
Phys. 2004;70(3 Pt 2): 036106.
28.!Ravasz E, Somera AL, Mongru DA, Oltvai ZN,
and Barabasi AL. Hierarchical organization of
modularity in metabolic networks. Science. 2002;297
(5586): 1551-1555.
29.!Rezende EL, Lavabre JE, Guimaraes PR, Jordano
P, and Bascompte J. Non-random coextinctions in
phylogenetically structured mutualistic networks.
Nature. 2007;448(7156): 925-928.
30.!Robins G and Malcolm A. Small Worlds Among
Interlocking Directors: Network Structure and
Distance in Bipartite Graphs. Comput. Math.
Organ. Theory. 2004;10(1): 69-94.
31.!Rodriguez-Caso C, Corominas-Murtra B, and
Sole RV. On the basic computational structure of
gene regulatory networks. Mol Biosyst. 2009.
32.!Rodriguez-Caso C, Medina MA, and Sole RV.
Topology, tinkering and evolution of the human
transcription factor network. FEBS J. 2005;272(24):
6423-6434.
33.!Sol R, Pastor-Satorras R, Smith ED, and T. K. A
model of large-scale proteome evolution. Advances in
Complex Systems. 2002;5: 43-54.
34.!van Helden J, Wernisch L, Gilbert D, and Wodak
SJ. Graph-based analysis of metabolic networks. Ernst
Schering Res Found Workshop. 2002;(38):
245-274.
35.!Vazquez A. Growing network with local rules:
preferential attachment, clustering hierarchy, and
degree correlations. Phys Rev E Stat Nonlin Soft
Matter Phys. 2003;67(5 Pt 2): 056104.
36.!Wagner A. Asymmetric functional divergence of
duplicate genes in yeast. Mol Biol Evol. 2002;19(10):
1760-1768.
37.!Wagner A and Fell DA. The small world inside
large metabolic networks. Proc Biol Sci. 2001;268
(1478): 1803-1810.
38.!Watts DJ and Strogatz SH. Collective dynamics of
'small-world' networks. Nature. 1998;393(6684):
440-442.
39.!Yildirim MA, Goh KI, Cusick ME, Barabasi AL,
and Vidal M. Drug-target network. Nat Biotechnol.
2007;25(10): 1119-1126.
40.!Zhang P, Wang J, Li X, Di Z, and Fan Y. The
clustering coefcient and community structure of
bipartite networks. arXiv:0710.0117. 2007.
41.!Zhou T, Ren J, Medo M, and Zhang YC. Bipartite
network projection and personal recommendation.
Phys Rev E Stat Nonlin Soft Matter Phys. 2007;76
(4 Pt 2): 046115.
10
145
Captulo 4
2.0
28.0

About the small-world concept in bipartite
networks.

Montaez R
1
, Real-Chicharro A
1
, Snchez-Jimnez F
1
, Medina MA
1
. and Rodrguez-Caso C
2
.
1
Departamento de Biologa Molecular y Bioqumica, Facultad de Ciencias, Universidad de Mlaga, E-29071 Mlaga, and CIBER de Enfermedades Raras (CIBERER),
Mlaga, SPAIN.
2
ICREA-Complex Systems Lab, Universitat Pompeu Fabra. Parc de Recerca Biomdica de Barcelona. Dr. Aiguader 88, 08003. Barcelona Spain.


Abstract
Universal patterns as Small-World in bipartite networks can emerge as an artifactual consequence of the inappropriate use of
the methodology. It is noteworthy that the projection of bipartite networks has as a consequence the increase of local
connections in the resulting graph. Thus, even completely random bipartite networks could seem small-worlds when projected,
as we show in this work. In this article, we develop an alternative method, inspired in the Watts-Strogatz model to perform a
bipartite evaluation of the small-world characteristics of bipartite networks. This methodology is tested applying it to metabolic
networks of 11 species. The obtained results indicate that, indeed, metabolism maintains the small-world features even when
analyzed as a bipartite network.

I. Introduction
In the last few years, a great effort has been devoted to
understand complex network organization [5, 7, 8, 26,
30, 33] and how natural systems can be described in
terms of elements and their connections [3, 5, 12, 27].
Topological organization -the global pattern of
connections of the graph of these systems- displays a
number of traits, such as small-world [8] and scale-free
organization [12, 13].
The small-world property tells us that any two nodes in
a network are on average separated by just a few
intermediate links, even when the networks are large,
sparse and clustered. Such short paths could have an
immediate impact on network functionality, since they
enhance the propagation of changes through the system.
Additionally, small-world graphs have a high frequency of
triangles indicating that the neighbours of a node tend to
be also connected among them, much more than
expected from randomness. Therefore, the small-world
property combines both local associations and very short
distances among nodes in a same topology. The over-
representation of local associations among the nearest
neighbourhood of a node and itself, is a clue of an order
higher than the expected from random.
An important statistical tool required to estimate this
property order beyond a random pattern is the clustering
coefficient [33]. This statistical test measures local
associations or cliquishness among the neighbourhood
of a node. Defined as the number of connected
neighbours normalized by the number of all possible
combinations, this coefficient has been extensively used
to characterize, for instance, the small-world pattern [25],
or the hierarchical modularity in complex one-mode
networks [27].
Unfortunately, not all the systems can be correctly
abstracted as one-mode networks (also named unipartite
networks). Instead, they should be modelled as bipartite
networks [16], a particular class of graphs whose nodes
are divided into two sets of nodes, top (T) and bottom
nodes (B). In these networks, connections between
nodes of the same type are forbidden (Figure 1A). For
instance, collaboration networks such as actors co-
starring in the same movie [1], scientists connected by
co-authoring a scientific paper [21-23], and opinion
networks, -for example customers and the books that
they buy (e.g. Amazon.com) [6]- are described as
bipartite networks. More recently, the relevance of
considering metabolism as a bipartite network has been
discussed, taking into account that metabolites are
transformed by reactions carried out by protein enzymes
[13, 15, 19].
Bipartite representations have important restrictions,
since connections between nodes of a same set are
forbidden. Triangles (a node linked to two nodes that are
also connected between them) are not possible in
bipartite networks [12]. The triangle is the minimal clique
146
2.0
28.0
Ral Montaez Martnez


in a graph. This motif is key for the quantification of the
cliquishness of a node, using the so-called clustering
coefficient. However, as no triangles can be formed in
bipartite networks, clustering coefficient is zero in these
graphs. To solve this drawback, bipartite networks have
been traditionally simplified by one-mode projections [4,
21-23], as depicted in Figure 1.
It is noteworthy that, although relevant topological
properties can be obtained from these projections, the
conceptual simplification of the process has a high cost.
Firstly, the obvious loss of information [17], evidenced by
the impossibility to reconstruct the bipartite network from
its G
T
and G
B
projections, even if we keep weighted
projections. Secondly, although real networks are
typically sparse, the projection procedure decreases the
sparseness into the resultant projected networks. This is
especially problematic when one node set is significantly
larger than the other, since each node that is going to be
projected induces k(n
i
)(k(n
i
)!1)/2 links in the projection
of the opposite nodes, and vice versa [16]. Furthermore,
it imposes correlations among projected node degrees
[31]. In the previous equation, k(n
i
) is the number of links
that this node contains in the original graph. The non-
linear dependence between k(n
i
) and the projected k
i
,
besides correlations among node degrees, may mislead
to a wrong interpretation of cliquishness. In fact, it has
been shown that an uncorrelated network where no
cliquishness is expected (as it is the case of the so-called
Erds-Renyi bipartite networks) exhibits a high clustering
coefficient when projected [11, 24]. Therefore, high
clustering coefficients in projected graphs should not be
viewed as meaningful properties. It is noteworthy that
these undesirable consequences of projection processes
has been ignored in almost all the published studies
concerning the bipartite nature of some networks, such
as the metabolic ones. For this reason, the aim of this
work is to provide a suitable evaluation of the small-world
condition for bipartite graphs, avoiding the use of
projections.
According to this objective, the remaining of this paper
is organised as follows: Section II presents the
topological measures used in this work to define the
concept of cliquishness for bipartite graphs and to
reproduce the small-world effect in a bipartite regular
lattice in a similar way to that described in [33] for one-
mode graphs. The effect of rewiring of the bipartite lattice
is then compared with the effect of rewiring of a
previously projected bipartite lattice according to the
original article. Finally, section III describes the
application of the resulting small-world condition to a set
of 11 metabolic networks, in order to elucidate whether
the small-world pattern is an intrinsic feature of
metabolism or a consequence of the bias introduced by
the projection.

II. The Model
Basic definitions
The clustering coefficient that has been used in the
WS-model measures the probability that two neighbours
of a selected node are linked together forming a triangle.
An alternative for bipartite graphs (avoiding the process
of projections) to measure the cliquishness of a given
node t
i
is assessed by the probability that two neighbours
(belonging to the B set) of t
i
were linked to a common
node in T. In this case, cliquishness deals with square
motifs of four nodes. Two alternative clustering values
can be calculated for bipartite graphs. Formally, a given
bipartite graph G(T,B,E) consists of two disjoint sets of
nodes T={t
1
,t
2
,...,t
|T|
} and B={b
1
,b
2
,...,b
|T|
} and a set of links,
E={e
1
,e
2
,,e
|E|
} where e
i
={t
j
,b
k
} is an undirected pair.
Clustering of the node t
i
is defined as
where k(t
i
) is the degree of t
i
(i.e, the number of
neighbours in the opposite set directly linked to node t
i
)

and k
2
(t
i
) is the second neighbourhood of node t
i
, (it is the
number of nodes in T that t
i
can achieve crossing through
its k(t
i
) nearest in B). This clustering coefficient is based
on the square coefficient of Robins [28], and reviewed in
[18, 35].
Calculating the corresponding coefficient for every
node of the graph, we can obtain the average clustering
for every set of nodes (<C
4
T
> or <C
4
B
>) and for the
complete graph <C
4
G
> if we do not attend to the type of
the node.
Since the small-world condition is evaluated by
comparison with a completely random graph [33], as
Erds Reny graph [8, 11, 25], we select a bipartite ER
for a suitable comparison.
For Erdos-Renyi (ER) bipartite graphs, C
4
G
follows the
expression
where <k
G
> is the average degree of the complete graph.
For a given set of nodes within random bipartite graphs
where C
4
B

and <k
B
> are the clustering coefficient and the
average degree for the B set, for the calculation of the
average clustering for B set. Notice that the same
equation is applicable to the nodes of the T set changing
the notation.
The average distance between pairs of nodes (a
measure required to estimate the small-world condition)
is quantified using the minimal path length average <L
G
>.
This measure provides an average of the minimal path
length calculated from every pair of nodes in the network.
When node type of the network is not considered, the
path-length is simply the number of nodes that connects
a pair of nodes in a chain. However, if we want to
distinguish between both types of nodes, T and B, we
must define a path length only considering as possible
pairs those with nodes of the same type; for instance,
given a specific node path, L(t
i
,t
j
), we count the minimal
number of nodes belonging to B required to connect t
i
to
t
j
. According to this measure, it is also possible to
calculate the average minimal path-length for every node
type, <L
T
> and <L
B
>. For ER bipartite graphs, the
average path length follows the expression:
L
G
=log(n
T
+n
B
); and for the specific sets of nodes,
L
T
=log(n
T
) and L
B
=log(n
B
).
147
Captulo 4
2.0
28.0


The Bipartite Small-World Model

!

To reproduce the interpolation between regular and
random networks of Watts and Strogatz model (WS-
model) [33], we used the same random rewiring
procedure with the constrain that any connection
between T and B is forbidden. Figure 2 reproduces the
Watts-Strogatz experiment attending to the bipartite
nature of the regular lattice.
Let us consider a regular lattice forming a ring with two
disjoint sets of nodes, T and B, and a set of links, E,
linking them. In order to construct this graph we consider
an initial linear (bipartite) circular chain with |T|=|B| nodes
and |E|=2|B| links. In this graph every node t has two b
neighbours. If we establish a connection with the n
nearest neighbours in the chain we can construct a
circular lattice of arbitrary regular degree. In general we
can say that, given T and B sets, where |T|=|B|, a circular
lattice with n nearest neighbours presents |E|=2n|B| links
and it is satisfied that

<k>=k(t
i
)=k(b
i
)=2n.!
Figure 2 illustrates a particular example of this regular
graph. Once defined this regular lattice, we proceed to
rewire the set of links. For this purpose, we start from a
given node t
k
(note that the process is equivalent for b
k
)
and we chose one of the two edges, e
i
={t
k
,b
j
}, that
connects t
k
to one of the nearest neighbours (n=1). Then,
with a probability p, we evaluate the likelihood of e
i
={t
k
,b
j
}
to change to e
i
={t
k
,b
m
}. Here, b
m
is a randomly chosen
node from B set, provided that this new link configuration
does not already exist in the set E. We apply the same
process to the edge connecting the next nearest nodes.
This process is applied node by node in a clockwise
motion around the ring. Once a round is given, we
evaluate the second and third nearest neighbourhoods,
and so on until the n
th
one.


III. Results
The basic concepts presented in the previous section
were used for the reproduction of the small-world effect in
bipartite networks and the evaluation of this property in
metabolic networks.!
The bipartite Watts-Strogatz model
Figure 3 illustrates the effect of rewiring of a regular
bipartite lattice applying the probability p, as described in
the previous section, showing the means and standard
deviation for one hundred replicas of this process,
ranging p from p=0 to p=1. In this figure, measures of C
4

and L applied directly to the bipartite resulting graphs are
compared with the standard measure of clustering and L
of their projected counterparts. In other words, every
resulted rewired bipartite network was projected and
measure for comparison.
!
148
2.0
28.0
Ral Montaez Martnez

!

Comparison of bipartite and projected networks was
possible by normalizing every value of clustering and
length by the initial values of the respective original
regular lattices.
Figure 4 reveals that C
4
and L values of bipartite
graphs capture the small-world effect, determined by the
dramatical impact of rewiring (p) on L, keeping
unaffected the normalized clustering. This is specially
appreciated for p=0.01 where C
4
(p)~C
4
(0) and L(p)<<L(0).
Values of C and L for projected (unipartite) networks
show the same behaviour. Although the decay of
L(p)/L(0) is comparable in both experiments and it is in
agreement to the observed in the WS original model, a
remarkable difference is observed in clustering measures
at high p: whilst C
4
(p)/C
4
(0) falls to a values close to zero,
C(p)/C(0) in projections shows a lower-boundary close to
0.5 with a little (but significant) increase for high values of
p.
Notice that the p=1 condition represents a completely
bipartite ER network. Differences with its projected
counterparts at this point are maxima, thus revealing that
projection has an impact on clustering that is not
observed in the bipartite network.
According to these results we can establish a
condition of small-world pattern by analogy with the
original experiment [33]. Then, we can say that a bipartite
SW network satisfies
C
4
(real graph)>>C
4
(bipartite ER)
L(real graph)~L(bipartite ER)

The small-world pattern in bipartite
metabolic networks
Once defined the SW conditions for bipartite graphs, in
this section we proceed to evaluate this condition of
metabolism.
Metabolic networks used here have been retrieved
from the Kyoto Encyclopedia of Genes and Genomes
(KEGG) [14], as described in [20]. In the main text we
show the results achieved analyzing 10 organisms:
(Mycoplasma pneumoniae (mpn), Escherichia coli (eco),
Bacillus subtilis (bsu), Saccharomyces cerevisiae (sce),
Schizosaccharomyces pombe (spo), Dictyostelium
discoideum (ddi), Plasmodium falciparum (pfa),
Arabidopsis thaliana (ath), Drosophila melanogaster
(dme), Mus musculus (mmu) and Homo sapiens (has)).
Metabolic information was used to construct the set of
bipartite undirected networks studied in this work. These
networks consist of two sets of nodes: metabolite and
reactions, noted as M and R, respectively. A metabolite
node is linked to a reaction node by an undirected edge if
it participates in such a reaction. The edges set is noted
as E. Notice that, to satisfy the bipartite nature of the
network, edges between elements of M (or between
elements of R) are forbidden.
For each network, we analyzed the biggest connected
component ensuring a corrected measure of the average
path length. In our network construction, we included the
13 ubiquitous metabolites traditionally excluded of
metabolic network reconstructions [32].
The small-world evaluation using the direct analysis of
the bipartite network for the 11 metabolic networks is
summarized in Table 1. These networks are sparse, with
average path lengths similar to the random ones but with
a higher value of C
4
. We observe that in all of them the
small-world bipartite condition is satisfied. However, it is
noteworthy that L is slightly smaller in real networks
(excluding the case of Mycoplasma pneumoniae) than in
ER bipartite counterparts. These evidences are
discussed in the next section.
Discussion
The results presented in this work confirm that
metabolic networks can be considered as small-worlds.
This is in agreement with Fell and Wagners first work of
the small-world in metabolism. [9, 32]. Small-worldness
enables to minimize the transition times between
metabolic states, and therefore it contains clues of the
evolutionary history of metabolism. This may be applied
to our results. However, a more sophisticated network
representation of metabolism, as that proposed by Arita,
suggested a non small-world behaviour for metabolism
[2]. In Aritas representation the compounds themselves
are represented as nodes, and two nodes are linked if a
carbon transfer occurs between them. However, the main
limitation of this work is that not only carbon atoms are
transferred in metabolism and this may be a too strong
simplification.
Our bipartite representation allows to overcome the
problem of projection and to construct a more realistic
network. In this work, we show how, in a natural bipartite
graph of metabolism, the relative number of square
motifs is much higher than the expected in a random
graph. This, along with the similar values of average path
lengths, can be correlated with the small-world
characteristics in a graph. In this way, several
interpretations of the cliquishness can be applied, and all
Table 1.- , Difference of <C4> and <LM> in the selected 11 organisms. Every bipartite node set was compared with their equivalent ER counterparts. Differences in
<C4> were compared using the rate of change <C4M-ER/ C4M>, where <C4M-ER> is the average <C4> over every specific M set of ER-rand networks and <C4M> over
real network. (Real Values are outside of the standard deviations over 1000 repetitions). Distances are compared using <LM> and <LM-ER>. <kM> is the degree of
every set of nodes. In a similar way C4 ans L were calculated for R set.

Bipartite <C4M> <C4M-ER> <C4M-ER>/ <C4M> <LM> <LM-ER> <C4R> <C4R-ER> <C4R-ER>/ <C4R> <LR> <LR-ER> <kM> <kR>
M. pneumoniae 0.06 6.7E
-3
0.11 2.73 2.42 0.07 6E
-3
0.09 2.16 2.22 3.39 4.16
E. coli 0.05 1.2E
-3
0.02 2.52 2.68 0.05 1.3E
-3
0.03 1.95 2.63 4.23 4.49
B. subtilis 0.06 1.5E
-3
0.03 2.56 2.68 0.05 1.5E
-3
0.03 1.95 2.61 4.61 4.26
S. cerevisiae 0.07 1.4E
-3
0.02 2.56 2.64 0,06 1.4E
-3
0.03 1.97 2.58 4.75 4.34
S. pombe 0.07 1.6E
-3
0.02 2.56 2.61 0.06 1.6E
-3
0.03 1.96 2.54 4.74 4.40
D. discoideum 0.07 1.6E
-3
0.02 2.59 2.66 0.06 1.6E
-3
0.03 1.96 2.59 4.59 4.28
P. falciparum 0.06 2.5E
-3
0.04 2.69 2.64 0.06 2.3E
-3
0.04 1.99 2.49 3.92 4.32
A. thaliana 0.07 1.E
-3
0.01 2.58 2.75 0.05 1E
-3
0.02 1.97 2.71 4.77 4.31
D. melanogaster 0.08 1.3E
-3
0.02 2.56 2.68 0.06 1.4E
-3
0.03 1.96 2.63 4.84 4.23
M. musculus 0.08 1.E
-3
0.01 2.59 2.71 0.06 1.1E
-3
0.02 2.0 2.68 5.06 4.26
H. sapiens 0.08 1E
-3
0.01 2.59 2.70 0.06 1.1E
-3
0.02 2.0 2.68 5.08 4.25
149
Captulo 4
2.0
28.0


of them have somehow a functional interpretation. A
square motif could be the transformation of one
metabolite into another by two different reactions, but it
also could be two metabolites that are products of the
first reaction and at the same time substrates of a second
one. Another possibility would be a small cycle of two
forward and reverse reactions that turn a metabolite to
another. From our current analysis, we cannot be sure
that this high density of squares correlates with one of
these interpretations and a deeper exploration in this
direction is required. On the other hand, it is noteworthy
that the analysis of projected graph has induced an
interpretation that fits with our results but, as we show in
this article, such a conclusion is derived from an artefact
of the methodology [20].
An interesting fact derived from our study is related to
minimal path length. We observed that the values for the
measure (excluding one case) are smaller than the
observed in their respective random counterparts. This
could be associated to an evolutionary trait related to
some kind of flux optimization. Interestingly, the only
organism that does not satisfy this observation is the
endocellular parasite Mycoplasma pneumoniae. This
organism preserves a minimal genome, since it has a
strong dependence of the metabolic processes derived
from the cell host. In this case, the rich metabolic
environment where this organism lives may have relaxed
the need of an autonomous control of the metabolic flux
within the parasite, since most of the metabolites are
available from the host.
Bipartite networks are, at the present, the rawest
version of a natural abstraction of metabolism. Building
up from this basis, including directionality [34],
stoichiometry [29] or thermodynamics restrictions [10]
would provide relevant insights about the relation
between topological organization and dynamics.

Acknowledgments
This work was supported by the EU 6th framework
project ComplexDis (NEST-043241, CRC), grants SAF
2008-02522 (Spanish Ministry of Science and
Innovation), Fundacin Ramn Areces, P07-CVI-02999
and group BIO-267 (Andalusian Government). The
CIBER de Enfermedades Raras is an initiative of the
ISCIII (Spain). The authors would like to thank Bernat
Corominas-Murtra for useful comments.

Bibliography

1. Internet movie Database. Available from:
www.imdb.com.
2. Arita M. The metabolic world of Escherichia coli is not
small. Proc Natl Acad Sci U S A. 2004;101(6): 1543-
1547.
3. Barabasi AL and Albert R. Emergence of scaling in
random networks. Science. 1999;286(5439): 509-
512.
4. Barabasi AL, Jeong H, Neda Z, Ravasz E, Schubert
A, and Vicsek T. Evolution of the social network of
scientific collaborations Physica A. 2002;311: 590-
614.
5. Bollobs B and Riordan O, Mathematical Results on
Scale-free Random Graphs, in Handbook of Graphs
and Networks: From the Genome to the Internet, S.
Bornholdt and H.G. Schuster, Editors. 2002, Wiley
Journals.
6. Cattuto C, Loreto V, and Pietronero L. Semiotic
dynamics and collaborative tagging. Proc Natl Acad
Sci U S A. 2007;104(5): 1461-1464.
7. Dorogovtsev SN and Mendes JFF. Evolution of
networks. Adv. Phys. 2002;51: 1079.
8. Erds P and Rnyi A. On Random Graphs. I.
Publicationes Mathematicae. 1959;6(290-297).
9. Fell DA and Wagner A. The small world of
metabolism. Nat Biotechnol. 2000;18(11): 1121-1122.
10. Fierke CA, Kuchta RD, Johnson KA, and Benkovic
SJ. Implications for enzymic catalysis from free-
energy reaction coordinate profiles. Cold Spring Harb
Symp Quant Biol. 1987;52: 631-638.
11. Guillaume JL and Latapy M. Bipartite graphs as
models of complex networks. Information Processing
Letters. 2004;90(5): 215-221.
12. Holme P, Liljeros F, Edling CR, and Kim BJ. Network
bipartivity. Phys Rev E Stat Nonlin Soft Matter Phys.
2003;68(5 Pt 2): 056107.
13. Jeong H, Tombor B, Albert R, Oltvai ZN, and
Barabasi AL. The large-scale organization of
metabolic networks. Nature. 2000;407(6804): 651-
654.
14. Kanehisa M, Araki M, Goto S, Hattori M, Hirakawa M,
Itoh M, Katayama T, Kawashima S, Okuda S,
Tokimatsu T, and Yamanishi Y. KEGG for linking
genomes to life and the environment. Nucleic Acids
Res. 2008;36(Database issue): D480-484.
15. Lacroix V, Cottret L, Thebault P, and Sagot MF. An
introduction to metabolic networks and their structural
analysis. IEEE/ACM Trans Comput Biol Bioinform.
2008;5(4): 594-617.
16. Latapy M, Magnien C, and Del Vecchio N.
BasiNotions for the analysis of large two-mode
networks. Social Networks. 2008;30: 31-48.
17. Lehmann S, Schwartz M, and Hansen LK. Biclique
communities. Phys Rev E Stat Nonlin Soft Matter
Phys. 2008;78(1 Pt 2): 016108.
18. Lind PG, Gonzalez MC, and Herrmann HJ. Cycles
and clustering in bipartite networks. Phys Rev E Stat
Nonlin Soft Matter Phys. 2005;72(5 Pt 2): 056127.
19. Montaez R, Medina MA, Sol R, and Rodrguez-
Caso C. When metabolism meets
topology:reconciling metabolite and reaction
networks. BioEssays. 2010;In-Press.
20. Montaez R, Real-Chicharro A, Sanchez-Jimenez F,
Medina-Tores MA, and Rodriguez-Caso C. Analyzing
the bipartite structure of the metabolic network.
Manuscript. 2010.
21. Newman ME. Scientific collaboration networks. I.
Network construction and fundamental results. Phys
Rev E Stat Nonlin Soft Matter Phys. 2001;64(1 Pt 2):
016131.
22. Newman ME. Scientific collaboration networks. II.
Shortest paths, weighted networks, and centrality.
Phys Rev E Stat Nonlin Soft Matter Phys. 2001;64(1
Pt 2): 016132.
23. Newman ME. The structure of scientific collaboration
networks. Proc Natl Acad Sci U S A. 2001;98(2): 404-
409.
24. Newman ME, Strogatz SH, and Watts DJ. Random
graphs with arbitrary degree distributions and their
applications. Phys Rev E Stat Nonlin Soft Matter
Phys. 2001;64(2 Pt 2): 026118.
150
2.0
28.0
Ral Montaez Martnez


25. Newman ME, Watts DJ, and Strogatz SH. Random
graph models of social networks. Proc Natl Acad Sci
U S A. 2002;99 Suppl 1: 2566-2572.
26. Newman MEJ. The structure and function of complex
networks. SIAM review. 2003;45: 167-256.
27. Ravasz E, Somera AL, Mongru DA, Oltvai ZN, and
Barabasi AL. Hierarchical organization of modularity
in metabolic networks. Science. 2002;297(5586):
1551-1555.
28. Robins G and Malcolm A. Small Worlds Among
Interlocking Directors: Network Structure and
Distance in Bipartite Graphs. Comput. Math. Organ.
Theory. 2004;10(1): 69-94.
29. Stelling J, Klamt S, Bettenbrock K, Schuster S, and
Gilles ED. Metabolic network structure determines
key aspects of functionality and regulation. Nature.
2002;420(6912): 190-193.
30. Strogatz SH. Exploring complex networks. Nature.
2001;410(6825): 268-276.
31. Vazquez A. Growing network with local rules:
preferential attachment, clustering hierarchy, and
degree correlations. Phys Rev E Stat Nonlin Soft
Matter Phys. 2003;67(5 Pt 2): 056104.
32. Wagner A and Fell DA. The small world inside large
metabolic networks. Proc Biol Sci. 2001;268(1478):
1803-1810.
33. Watts DJ and Strogatz SH. Collective dynamics of
'small-world' networks. Nature. 1998;393(6684): 440-
442.
34. Wunderlich Z and Mirny LA. Using the topology of
metabolic networks to predict viability of mutant
strains. Biophys J. 2006;91(6): 2304-2311.
35. Zhang P, Wang J, Li X, Di Z, and Fan Y. The
clustering coefficient and community structure of
bipartite networks. arXiv:0710.0117. 2007.


151
Captulo 4
2.0
28.0
152
2.0
28.0
Ral Montaez Martnez

2.0
28.0
Conclusiones
2.0
28.0
Conclusiones
Conclusin 1.- Se ha demostrado que el mediador semntico AMMO es capaz de
localizar automticamente la informacin dispersa, que pueda estar contenida en
las bases de datos, en relacin a las estructuras de protenas y propiedades cinti-
cas de las protenas demandadas. Ambas herramientas derivadas de AMMO,
AMMO-Prot y SBMM Assistant, cumple con sus objetivos y pueden ser fcilmente
adaptadas para trabajar integrando informacin procedente de otras plataformas
dedicadas a la protemica, la metabolmica y la biologa de sistemas, en general.
Conclusin 2.- Los resultados obtenidos con los modelos matemticos del metabo-
lismo de aminocidos catinicos y sus derivados en clulas de mamferos, constitu-
yen claros ejemplos de como la integracin metablica puede revelar nuevas pro-
piedades reguladoras. Estos resultados nos animan a proseguir con una estrategia
de integracin ascendente que contribuya a aclarar el papel de las aminas bige-
nas en las adaptaciones metablicas globales de clulas proliferantes.
Conclusin 3.- Por ltimo, evaluando la red metablica de un enfoque descenden-
te, dentro del marco conceptual de la teora de grafos, exploramos el hecho de
que una definicin simplista de la red metablica puede provocar una distorsin en
nuestra percepcin de la organizacin del metabolismo. Los anlisis efectuados en
varias especies revelaron que las versiones unipartitas de las redes metablicas,
comnmente aplicadas en los estudios topolgicos del metabolismo, producen
artefactos estadsticos. Por el contrario, la representacin bipartita del metabolismo
proporciona una abstraccin ms natural del metabolismo y una representacin
biolgicamente ms significativa.
155
2.0
28.0

2.0
28.0
Conclusions
2.0
28.0
Conclusions
Conclusion 1.- We demonstrated that our Semantic Mediator, named as AMMO,
can locate automatically relevant information on protein structures and kynetic
properties among disperse repositories. Both AMMO-derived tools, AMMO-Prot and
SBMM Assistant, can be easily adapted to work integrated with other platforms de-
voted to Proteomics, Metabolomics and Systems Biology, in general.
Conclusion 2.- Results obtained from mathematical metabolic models of cationic
amino acids and their derivatives in mammalian models clearly show how the inte-
gration of metabolic modules can reveal new regulatory properties. These results
encourage us to a further research focussed on a bottom-up strategy in order to
integrate new metabolic modules aiming of clarify the role of the biogenic amines
within the global metabolic adaptation occurring in proliferating cells.
Conclusion 3.- Finally, looking at the metabolic network from a top-down approach
within the framework of Graph Theory, we explore the fact that an oversimplified
definition of metabolic network can induce a bias in our perception of the metabo-
lisms organization. Analyses carried out on several species revealed that the one-
mode versions of metabolic networks, commonly applied in topological studies pro-
duce statistic artifacts. In contrast, the bipartite metabolite-reaction representation
provides a more natural representation of metabolism and a more biologically
meaningful representation.
159
2.0
28.0