Вы находитесь на странице: 1из 5

Plant, Cell and Environment (2007) 30, 887891

doi: 10.1111/j.1365-3040.2007.01687.x

Methods for improving the visualization and deconvolution of isotopic signals


GUILLAUME TCHERKEZ1,2, JALEH GHASHGHAIE2 & HOWARD GRIFFITHS3
1 Plateforme Mtabolisme-Mtabolome, IFR 87, Bt. 630, 2Laboratoire dcophysiologie vgtale, CNRS UMR 8079, Bt. 362, Universit Paris Sud-XI, 91405 Orsay cedex, France and 3Department of Plant Sciences, University of Cambridge, CB2 3EA, UK

ABSTRACT
Stable isotopes and their associated mechanistic frameworks have provided the means to model biological transformations from inorganic sources to organic sinks, and their impact on long-term terrestrial carbon reservoirs. However, stable isotopes have also the potential to diagnose the mechanisms by which biological systems operate, when using multidimensional analyses of the isotopic outputs. For this purpose, we suggest that isotopic signals may be treated as mathematical vectors to reveal their interrelationships. These visualizations may be achieved, thanks to multidimensional representations (the isotopology method), or by appreciating the colinearity of isotopic vectors to develop a clustering analysis (the isotopomic method) similar to that used in molecular biology. Both methods converge to the same mathematical form, i.e. both lead to a covariation approach. Using simple practical examples, we argue that such procedures allow the deconvolution of plant biological systems by revealing the hierarchy of contributory physiological processes. Key-words: stables isotopes; fractionation; vectors; array; clustering analysis.

NO3-) (Ledgard, Woo & Bergersen 1985; Werner & Schmidt 2002; Tcherkez and Farquhar 2006). There have been a number of recent reviews which have evaluated biological and ecological stable isotope transformations, their impact on carbon reservoirs in the atmosphere, plant and soil, and coupling with the hydrological cycle (Grifths 1998; Dawson et al. 2002; Flanagan, Pataki & Ehleringer 2005; Grifths & Jarvis 2005). Most recently, it has been suggested that transformations and isotopic distributions can be viewed as isoscapes which trace interannual variations or geographical linkages (West et al. 2006). In this paper, we propose that isotope data should be visualized more clearly to reveal the connectivity between isotopic signals, and diagnose underlying mechanisms within plant systems. This may be achieved with isotopologies, which represent multidimensional maps of observed isotopic distributions, or isotopomics, in which systematic isotopic enrichments and depletions (isotopic shifts) within biological systems are correlated as vectors using multiarray, differential displays and cluster analysis.
15

A MULTIDIMENSIONAL ISOTOPOLOGY
The clearest representation of a global isotopic output is the annual dance of photosynthetic carbon uptake and sequestration in summer, followed by respiratory CO2 release in winter, which drives seasonal variations in CO2 concentration. Particularly in the Northern Hemisphere, the heterogeneity and temporary disequilibrium in isotopic mass balance of atmospheric CO2 lead to a striking intraannual variation in 12C/13C isotope composition (denoted as d13C). This interaction is classically represented as a rug or ying carpet diagram (whose data are available at the NOAA CMDL Carbon Cycle Cooperative Global Air Sampling Network website: http://wwwsrv.cmdl.noaa.gov/ccgg/iadv), and encapsulates our vision of an isotopic landscape or an isotopology. We suggest that this kind of approach can be used for other isotopic data, which are normally only shown as one-dimensional isotopic diagrams.

INTRODUCTION
Earth and plant systems biology requires an understanding of both interannual biospheric processes and its contributory molecular mechanisms, and stable isotopes can be used to integrate such geochemical and biological systems (Yakir 2002). Systematic shifts in natural isotopic ratios (such as 13C/12C or 18O/16O) are based on fractionation processes which lead to discriminations between isotopes in biological cornerstone reactions like photosynthesis [in which ribulose-1,5-bisphosphate carboxylase/oxygenase (Rubisco) discriminates against 13CO2] (Farquhar, Ehleringer & Hubick 1989; Tcherkez, Farquhar & Andrews 2006), leaf and canopy transpiration (discrimination against H218O and exchange as either C18O16O or water vapour) (Farquhar, Barbour & Henry 1998; Helliker & Grifths 2007) or plant nitrate assimilation (discrimination against
Correspondence: u-psud.fr G. Tcherkez. E-mail: guillaume.tcherkez@

Multiparametric representation: visualizing the multidependence of a given isotopic signal


To illustrate this point practically at the leaf level, we start from a multidimensional representation of exported
887

2007 The Authors Journal compilation 2007 Blackwell Publishing Ltd

888 G. Tcherkez et al.

15

d C of phloem sucrose ()

16 17 18 19 20 21 22 5 10 15 20

Time (h)

sy Star nt ch he si s

0.09 0.08 0.07 0.06 0.05 0.04 0.03 0.02

Figure 1. Multidimensional isotopic representation showing the


modelled time-course of phloem sucrose in a leaf, using a sinusodal function of 13C-enriched sucrose by night and 13 C-depleted sucrose by day. Note that the exported carbon slips to more 13C-depleted values when starch synthesis (in moles of hexoses per mole of carbon xed) increases. Calculations from Tcherkez et al. (2004).

13

assimilates that are subsequently distributed in several pools and could be used for respiration or growth (Fig. 1). It shows how the time-course of the d13C signal in phloem sucrose depends on both circadian metabolic changes and internal parameters (such as the rate of starch synthesis) that are in turn modulated by environmental conditions. Briey, the fructose-producing aldolase reaction of the chloroplast favours 13C (enriching starch in this isotope). The remaining triose phosphates are used in the cytosol to produce (13C-depleted) sucrose, so that sucrose found in the phloem during the day is 13C-depleted, contrary to sucrose produced during the night, simply because the latter comes from depolymerization of starch. This example typies how the compartmentation of metabolic reactions (chloroplastic starch synthesis versus cytoplasmic sucrose synthesis), when associated with an isotope effect, scales up to the plant level. Multidimensional representations involving isotopes are useful to deconvolute the complexity of physiological processes, and more importantly, challenge the common interpretation of delta values; as can be seen from Fig. 1, a given d13C of sucrose may correspond to several possible drivers (starch synthesis, relative time), as well as environmental effects such as the impact of temperature on metabolic processes. Such diel variations will be, among others, a factor contributing to the cellulose isotope signatures of tree rings, which in turn integrate seasonal climatic conditions (Schleser et al. 1999; MacCarroll & Loader 2004). Thus, as a further multidimensional representation, we provide a simple conceptual framework depicting the physiological complexity of ring deposition by the cambium (Fig. 2). A physical analogy involving transfer functions has also been proposed to improve the analyses of isotopic signals in rings (Schleser et al. 1999). Cellulose deposition is inuenced by the (leaf) carbon source (as determined by cc/ca, the ratio of intracellular/external CO2), and also by the contribution that reserves make as substrate for cellulose deposition (whether sucrose from starch in the dark, or sucrose during the day) and respiratory losses (Fig. 2a).

When modelled as a three-dimensional representation (Fig. 2b), the deviation of the d13C value of ring cellulose from the d13C value of xed CO2 can be seen as a function of the proportion of day sucrose (versus night sucrose) as a carbon source (L) and the relative respiratory rate (r), assuming a xed respiratory discrimination of -6. It is clear that both parameters have an effect on the carbon isotope signature of cellulose, and a quite large deviation is obtained under certain conditions (Fig. 2b). For example, the larger the respiratory losses (13C-enriched), the more 13 C depleted the remaining deposited material (so that d*dC increases). Further, the larger the contribution of (13C-depleted) day sucrose to cellulose deposition, the lower the isotope composition of cellulose (so that d*dC increases as well). This representation integrates the key determinants of internal carbon signature, which may in turn be changed by environmental conditions (e.g. the increase of respiration with temperature, etc.). The multidimensional representations of both Figs 1 and 2 are classical representations, in which the isotopic signal is a function of physiological drivers. Still, they are useful to deconvolute the complexity of physiological processes, and visualize how a given isotopic signal value may correspond to several possible values of the drivers. That said, multidimensional representations plotting several isotopic signals are perhaps more useful for deconvoluting interactions between bio(geo)chemical pathways, as we now show.

Multiple isotopic signals: partitioning biochemical pathways from isotopic compositions


When partitioning occurs during metabolism, graphical representations can be used to visualize the network of correlations. This can also be allied with additional techniques such as correlative clustering analysis, developed as follows. It will become apparent that both analyses graphical multidimensional and clustering are similar in the way that

2007 The Authors Journal compilation 2007 Blackwell Publishing Ltd, Plant, Cell and Environment, 30, 887891

Improving the visualization and deconvolution of isotopic signals 889


(a) (b)

Leaves

ci /ca

Fixed CO2 (d *)
d dC

0 2 4 6 0.40 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00

Tree ring

Starch 1L

Suc L

Other

Ring carbohydrates

e, r
Reserves CO2

8 0.8 0.6 0.4

0.2 0.0

Ring cellulose (d C)
Figure 2. (a) The carbon isotope composition of cellulose in tree rings (dC) does not simply reect that of xed CO2 [given by the internal-to-external CO2 molar fraction ratio ci/ca (or cc/ca) denoted as d*], because of so many intermediary stages. Namely, the proportion of sucrose produced by day (Suc) versus that at night (denoted as L) to feed the metabolism of ring cells, as well as respiration (relative rate r), which liberates CO2 that may be enriched or depleted (the fractionation is denoted as e). These internal factors may in turn be inuenced by external parameters (curved arrows): the respiration rate responds positively to temperature, the allocation of xed carbon to starch or sucrose during the day depends on ci, etc. (b) As an example, the difference between d* and dC is plotted as a double function of L and r, with a xed respiratory fractionation (e) of -6. Numbers were obtained using an extension of the model of Tcherkez et al. (2004).

the nal mathematical formalism converges, but differ in terms of their visual realization. As an example, we reanalyse data generated from the isotope composition of different metabolites extracted from a leaf during a series of labelling experiments (Nogus et al. 2004). A conventional multidimensional representation would show each compound on a separate axis. However, to display the interrelationships between compounds in such a graphical presentation, such data can be interpreted using a Bayesian analysis; for a given isotopic signal in compound number i, one may look at (the) other compound(s) in the pathway which covary in association with that signal (Fig. 3). Thus, in mathematical terms, two isotopic signals i and j are best correlated if they follow each other in the graphical presentation, independently of others (denoted here as n). This means that signals n are in a perpendicular space, with the scalar product of i and j maximal, and the scalar products of i and any n, or j and any n, zero. In its general form, the scalar product would be as follows:

xi , xj = xik xjk
k =1

where isotopic signal xik is that of compound i obtained during the measurement (or experiment) number k. This is very similar to what the similarity-based clustering analysis discussed as follows gives. The associated visualization may be, however, complicated when more than three isotopic signals are considered. While a classical bidimensional graph may be used for all the isotopic signals, a polar graph is, perhaps, a more convenient way to associate isotopic shifts in the different compounds, and determine whether they follow the same pathway. Figure 3 gives such an example using the data of Nogus et al. (2004), collected during labelling experiments allowing Phaseolus leaves to x varying quantities of isotopically dened CO2. We now plot the offset in isotopic signals between respired CO2 (after illumination) and compound-specic signals in sucrose, starch, organic acids,

2007 The Authors Journal compilation 2007 Blackwell Publishing Ltd, Plant, Cell and Environment, 30, 887891

(1)

890 G. Tcherkez et al. clusters show the isotopic proximity of compounds or partners within a network of uxes comprising the particular biological system of interest. This provides an effective mean of seeing the correlative network between compounds in the given biological system. In other words, we use here isotopes not only as classical tracers, but additionally as correlators. From a mathematical point of view, the principle is similar to the genetic clustering analysis, i.e. uses correlations of isotopic shift vectors as calculated, for example, by the normalized scalary product. If we denote x as the vectors of isotopic shifts (array conditions or experiments are numbered from 1 to K), the similarity (or colinearity) index (denoted as S) between two vectors i and j has the following general form:

90

CO2 Suc Starch OA Lip Prot

12 11 10 9 8 7 6 5 4 3 2 1 0

60

30

Experiments

Figure 3. Representation of the isotopic shifts in the


compound-specic d13C values after labelling leaves with industrial CO2 (data from Nogus et al. 2004), plotted as a polar graph in which each increment corresponds to a different experiment (increasing xed C amount) and the angles (denoted as q in the main text) represent the isotopic shifts normalized to 100 angular degrees. Suc, sucrose; OA, organic acids; Lip, lipids; Prot, heat-precipitated proteins.

Sij =

x
k =1

ik

xjk

xik 2
k =1

xjk 2
k =1

(2)

proteins and lipids. The isotopic shifts are normalized to 100 (angular degrees) and plotted as angles (denoted as q just below), and each labelling condition (experiment) corresponds to a radial increment. This allows one to see whether the isotopic pathways have the same shape, or rather, diverge. From a quantitative point of view, we note that two pathways A and B are exactly similar if the angles (qA and qB) are similar for each radial increment, i.e. if the corresponding cosine [cos(qA - qB)] value is 1, then the scalar product of the associated vectors is maximal. It can be seen in Fig. 3 that the pathway followed by CO2 is closer to that of carbohydrates and correlates less well with other compounds. In other words, this indicates that the origin of CO2 respired by leaves after illumination is carbohydrates. Less apparent is, however, the molecule from which CO2 is derived (starch or sucrose). Another representation, coupled to a more quantitative correlative technique, is thus described as follows.

The xik values may be relative isotopic shifts in delta values or ratios or % in 13C (in case of strong labelling), etc. All the Sij values generate a matrix which may be in turn treated by the clustering programme to generate clusters. As a simple example, such an approach can be used at the leaf scale with compound-specic isotope composition following isotopic labelling (Fig. 4). Again, data from Nogus et al. (2004) (in which leaves were labelled with 13Cdepleted CO2) are used to generate a matrix of relative isotopic shifts from the initial value. Array conditions (columns) correspond to different labelling level (denoted as Lab_i where i is the amount of carbon xed during the labelling period in Fig. 4). The colour brightness indicates the amplitude of the relative isotopic shift. The cluster analysis can be used to generate a phylogenetic-style tree indicated on the left-hand side of Fig. 4. In this particular example, metabolic associations can be easily concluded

ISOTOPOMICS
The isotopomics concept is illustrated by Fig. 4. Here, we show how statistical methods derived from genomics can be used to deconvolute isotope signals. Traditional cluster analysis of gene transcription using microarrays (Eisen et al. 1998) indicates the abundance of many transcripts by colours. The cluster analysis of genes, where correlation coefcients or related mathematical parameters use similarities in transcription patterns to detect the coregulation of gene expression, can also be used to infer metabolic interactions and associations (Gachon et al. 2005). Similarly, arrays can be used to show isotopic modications for different compounds of a biological system (in rows), induced by different environmental conditions (in columns); the

Figure 4. Example of isotopomics at the leaf level. The array


has been obtained with isotopic shifts in the compound-specic d13C values after labelling leaves with industrial CO2 (data from Nogus et al. 2004). The array has been used to derive a cluster analysis (left) indicating the correlations between isotopic shifts. Each Lab_i indicates a different labelling experiment, in which i is the amount of C net xed (in mmol C m-2 s-1). The array and the cluster have been generated with the Cluster and Treeview softwares (M. Eisen, Standford University). Suc, sucrose; Prot, heat-precipitated proteins; OA, organic acids; Lip, lipids.

2007 The Authors Journal compilation 2007 Blackwell Publishing Ltd, Plant, Cell and Environment, 30, 887891

Improving the visualization and deconvolution of isotopic signals 891 with this technique: CO2 evolved in darkness correlates more closely with starch and sucrose; and lipids, on the other hand, show little correlation with other compounds because of their slow turnover. Thus, starch appears to be the most likely substrate feeding leaf respiration following illumination.
Proceedings of the National Academy of Sciences of the United States of America 95, 1486314868. Farquhar G.D., Ehleringer J.R. & Hubick K.T. (1989) Carbon isotope discrimination and photosynthesis. Annual Review of Plant Physiology and Plant Molecular Biology 40, 503537. Farquhar G.D., Barbour M.M., Henry B.K. (1998) Interpretation of oxygen isotope composition of leaf material. In Stable Isotopes, Integration of Biological, Ecological and Geochemical Processes (ed. H. Grifths), pp. 2762. Bios Scientic Publishers, Oxford, UK. Flanagan L.B., Pataki D.E. & Ehleringer J.R. (2005) Stable Isotopes and Biosphere Atmosphere Interactions: Processes and Ecological Controls. Elsevier, San Diego, CA, USA. Gachon C.M.M., Langlois-Meurinne M., Henry Y. & Saindrenan P. (2005) Transcriptional co-regulation of secondary metabolism enzymes in Arabidopsis: functional and evolutionary implications. Plant Molecular Biology 58, 229245. Grifths H. (1998) Stable Isotopes and the Integration of Biological, Ecological and Geochemical Processes. Bios Scientic Publishers, Oxford, UK. Grifths H. & Jarvis P.G. (2005) The Carbon Balance of Forest Biomes. Taylor and Francis, Oxford, UK. Helliker B. & Grifths H. (2007) Towards a plant-based proxy for the 18O content of atmospheric water vapour. Global Change Biology, in press. Knohl A., Werner R.A., Brand W.A. & Buchmann N. (2005) Short term variations in d13C of ecosystem respiration reveals links between assimilation and respiration in a deciduous forest. Oecologia 142, 7082. Ledgard S.F., Woo K.C. & Bergersen F.J. (1985) Isotopic fractionation during reduction of nitrate and nitrite by extracts of spinach leaves. Australian Journal of Plant Physiology 12, 631 640. MacCarroll D. & Loader N.J. (2004) Stable isotopes in tree rings. Quaternary Science Reviews 23, 771801. Nogus S., Tcherkez G., Cornic G. & Ghashghaie J. (2004) Respiratory carbon metabolism following illumination in intact French bean leaves using 12C/13C labelling. Plant Physiology 136, 32453254. Schleser G.H., Helle G., Lucke A. & Vos H. (1999) Isotope signals as climate proxies: the role of transfer functions in the study of terrestrial archives. Quaternary Science Reviews 18, 927943. Tcherkez G. & Farquhar G.D. (2006) Isotopic fractionation by plant nitrate reductase, twenty years later. Functional Plant Biology 33, 531537. Tcherkez G., Farquhar G.D., Badeck F. & Ghashghaie J. (2004) Theoretical considerations about carbon isotope distribution in glucose of C3 plants. Functional Plant Biology 31, 857877. Tcherkez G., Farquhar G.D. & Andrews T.J. (2006) Despite slow catalysis and confused substrate specicity, all ribulose bisphosphate carboxylases may be nearly perfectly optimized. Proceedings of the National Academy of Sciences of the United States of America 103, 72467251. Werner R.A. & Schmidt H.L. (2002) The in vivo nitrogen isotope discrimination among organic compounds. Phytochemistry 61, 465484. West J.B., Bowen G.J., Cerling T.E. & Ehleringer J.R. (2006) Stable isotopes as one of natures ecological recorders. Trends in Ecology & Evolution 21, 408414. Yakir D. (2002) Sphere of inuence. Nature 416, 795. Received 19 April 2007; received in revised form 27 April 2007; accepted for publication 29 April 2007

PERSPECTIVES
At the outset, we emphasized the current use of isotopes to understand the architecture of plant metabolism and communities in ecosystems. Isotopes indeed provide a key to these approaches, linking global carbon mass balance to basic enzymatic discrimination (e.g. Rubisco and respiratory processes) via elementary physiological events, which ultimately dene the CO2 signature in the atmosphere. However, more correlative approaches using isotopes, which have been undertaken in genomic and metabolomic approaches, have not been realistically conducted to date. If integrative methods such as isotopomics and isotopologies were to be adopted, this would help to understanding connectivity between metabolite pools or drivers of ecosytem dynamics.This would help to elucidate interactions between biogeochemical factors controlling carbon sequestration and respiratory release. In the present paper, we have used leaf-scale data as an example to explain the principle of the approaches. In a similar manner, we propose that isotopomic approaches could be extended to larger-scale studies, to investigate the architecture of the carbon trafcking networks in soil or ecosystem scales, showing, for example, transfer of carbon from leaves to soil-respired CO2 (Bowling et al. 2002; Knohl et al. 2005). We nevertheless recognize that this would require a large body of isotopic data associated with several components of ecosystems. However, there is no doubt that this may be overcome in the very near future, thanks to the use of automated continuous-ow spectrometric measurements or tunable diode laser techniques (as CO2 isotope composition is concerned) (Bowling et al. 2002). Ultimately, we suggest that by mapping the distribution of isotopic signals in biological systems, these approaches will improve our understanding of ecological processes by weighting the contributory physiological processes, and provide a more effective means to visualize their interrelationships.

REFERENCES
Bowling D.R., McDowel N.G., Bond B.J., Law B.E. & Ehleringer J.R. (2002) 13C content of ecosystem respiration is linked to precipitation and vapor pressure decit. Oecologia 131, 113 124. Dawson T.E., Mambelli S., Plamboeck A.H., Templer P.H. & Tu K.P. (2002) Stable isotopes in plant ecology. Annual Review of Ecology and Systematics 33, 507559. Eisen M.B., Spellman P.T., Brown P.O. & Botstein D. (1998) Cluster analysis and display of genome-wide expression patterns.

2007 The Authors Journal compilation 2007 Blackwell Publishing Ltd, Plant, Cell and Environment, 30, 887891