Академический Документы
Профессиональный Документы
Культура Документы
BIOCHEMISTRY
Analytical Biochemistry 342 (2005) 300–311
www.elsevier.com/locate/yabio
Abstract
Peptide arrays prepared by the SPOT synthesis technology have emerged as a proteomic tool to study molecular recognition and
identify biologically active peptides. However, it was previously not clear how accurately signal intensities obtained by probing pep-
tide arrays for protein binding really reflect the dissociation constants of the protein–peptide complexes. Using the monoclonal anti-
body CB4-1 as a model system, we systematically compared dissociation constants of antibody–peptide complexes with signal
intensities obtained using the SPOT technology. By analyzing a set of peptides possessing different affinities to the antibody, we
determined the strengths of the SPOT screening method. The accuracy of the measured results was improved by taking regional
trends in the membrane surface into account. A model based on the mass action law compares well with the experimental results.
Interestingly, the applied concentrations of the binding partners do not directly correspond to the effective concentrations in the
assay. We show that the SPOT technology is an accurate method for assigning the spotsÕ measured signal intensities to three different
binding affinity classes. The dissociation constants of the intermediate region were found to be between pKdis = 5 and pKdis = 7.
Altering the experimental parameters causes a directed change of this region.
2005 Elsevier Inc. All rights reserved.
Keywords: SPOT technology; Synthetic peptide arrays; Binding affinity; Dissociation constant; Effective concentration; Mass action law;
Competition
0003-2697/$ - see front matter 2005 Elsevier Inc. All rights reserved.
doi:10.1016/j.ab.2005.04.033
Array-based measurement of peptides / A.A. Weiser et al. / Anal. Biochem. 342 (2005) 300–311 301
and dissociation of all the functional components. ping peptides [10], randomly generated peptide libraries
Therefore, more or less quantitative readout for map- [19], transition pathway libraries that reveal evolution-
ping protein–protein interaction in a high-throughput ary connections between different peptides with similar
manner is of general interest. activities [20], and peptide arrays to elucidate the
Synthetic peptide arrays offer the benefit of studying cross-reactivity of mAb CB4-1 [21].
asymmetric protein–protein interactions: a domain of
partner A acts as a receptor for a linear peptide in part-
ner B. In fact, a fairly large set of cellular protein inter- Materials and methods
actions are mediated by families of relatively small
protein interaction domains (PID),2 such as SH2, SH3, Synthesis of the cellulose-membrane-bound peptide arrays
GYF, WW, or PDZ, which act as receptors accommo-
dating short peptides in their binding pockets. The Cellulose-bound peptide arrays were semiautomati-
SPOT technology is an ideal tool for preparing synthetic cally prepared according to standard SPOT synthesis
peptide arrays, because it is an easy-to-use, robust, and protocols using a SPOT synthesizer (Intavis, Koeln,
inexpensive method. Furthermore, the technique and Germany) as described in detail [22]. The peptides were
many of its applications have been reviewed extensively synthesized on amino-functionalized cellulose mem-
[8–10]. Recently, we successfully applied SPOT technol- branes of the ester type [10] prepared by modifying a cel-
ogy to studying the interaction network of PIDs [11–14]. lulose paper (Whatman 50, Whatman, Maidstone, UK)
Landgraf et al. [12] describes how we were able to dem- with Fmoc-b-alanine as the first anchor residue. Load-
onstrate for the first time that measured spot signal ing of the amino-functionalized cellulose membrane
intensities provide an approximation of the affinities of was determined as described [22]. In the second step of
the several thousand interactions analyzed simulta- coupling the next anchor position, a mixture of different
neously on a peptide array. In principle, signal intensi- ratios of a 0.3 M solution of Fmoc-b-alanine-OPfp and
ties can be used to distinguish between several a 0.3 M solution of N-acetyl-b-alanine-OPfp in dimethyl
interactions, with different affinities, mediated by the sulfoxide was used (100, 50, 25, and 10 Fmoc-b-alanine-
same protein. We first showed this for peptide–antibody OPfp) leading to membranes with a 100, 50, 25, and 10%
interactions [15,16]. Here, we focus in more detail on the amino-functionalization quotient (FQ) at the spot posi-
correlation between spot signal intensities measured in tions [16]. The cellulose-bound peptide arrays were
the array format and dissociation constants determined assembled on these membranes by using 0.3 M solutions
for the appropriate interactions. The question we want of Fmoc-amino acid-OPfp in NMP (in case of Ser and
to address is the quantitative reliability of the SPOT Thr the ODNp-esters were used). Side-chain protection
technique. First, we investigated the reproducibility of of the used Fmoc-amino acids was as follows: Glu, Asp
signals measured on cellulose membrane-bound peptide (OtBu); Ser, Thr, Tyr (tBu); His, Lys, Trp (Boc); Asn,
arrays prepared by the SPOT technology. Second, we Gln, Cys (Trt); Arg (Pbf). The sequence files for the ar-
assessed the correlation between the measured signals ray design were generated with the in-house tool Pool
and the known dissociation constants. Third, we investi- 1.0 and translated into chemistry files for the automated
gated the accuracy with which measured signals can be synthesis using the in-house software LISA 1.571. For
used to classify peptides into binding affinity classes. Fi- synthesis quality control of the peptide library, a repre-
nally, we discuss the experimental requirements needed sentative selection of peptides was prepared (spot size
to obtain reliable binding affinity class predictions. To 0.25 cm2) and cleaved by ammonia vapor in the dry
study the quantitative aspects of the SPOT technology state [22,23]. Subsequently, identity was verified by ma-
we chose the interaction of the anti-p24 (HIV-1) human trix-assisted laser desorption ionization mass spectrome-
monoclonal antibody (mAb) CB4-1 [17,18] with 38 well- try (MALDI-MS) (LaserTec BenchTop II, Applied
known peptides as a model system. The recognition pro- Biosystems, Foster City, CA, USA) and peptide quality
file of mAb CB4-1 is a well-established experimental sys- was tested by analytical reversed-phase HPLC (Waters
tem that has already been used to evaluate a variety of 600, Waters, Eschborn, Germany) on a C-18 column
structurally different types of peptide arrays prepared (Vydac, Hesparia, CA, USA). According to the general
by the SPOT technology. This includes scans of overlap- protocol a standard peptide array (dimension
6 cm · 8 cm) was synthesized on cellulose membranes
containing 37 different peptides (Table 1, entries 2–38)
2
Abbreviations used: AuC, area under curve; BG, background signal with an average of 13 repeats of each peptide and 120
intensity; FQ, membrane functionality quotient; Kdis, dissociation times the peptide GATPEDLNQKLAGN (Table 1, en-
constant; pKdis, negative decadic logarithm of Kdis; mAb, monoclonal try 1). The peptides were equally distributed on the ar-
antibody; ROC, receiver operator characteristic; CM, cellulose mem-
brane; SCM, standard cellulose membrane; SI, signal intensity; PID,
ray, resulting in a cellulose membrane that contains
protein interaction domain; MALD, matrix-assisted laser description 607 spots. Such an array was generated on membranes
ionization. with four different FQ values (100, 50, 25, and 10%)
302 Array-based measurement of peptides / A.A. Weiser et al. / Anal. Biochem. 342 (2005) 300–311
Table 1
Dissociation constants obtained for mAb CB4-1/peptide complexes
No. Sequence pKdis No. Sequence pKdis
a a
1 GATPEDLNQKLAGN 9.0 20 RDFDKEWNLIEQNS 5.7
2 GATPQDLkTMLb 9.0 21 LELIQDLNQKLQDGFd 5.7
3 GATPEDLNQKLc 8.2 22 TTMEWFRITDGARIMd 5.7
4 sATPwDLkTsle 7.7 23 sAGPwDLksslb 5.3
5 FATPEDLNQKLc 7.7 24 LEMKQDLNQKLQDGFd 5.3
6 GATPQDLNTMLe 7.3 25 LEMKQDLNQMLQDGFd 5.2
7 GLKEWGGARITc 7.1 26 DYFDLTQDNIITRRLd 5.0
8 DATPEDLNAKLe 7.0 27 LEMKQDLNIKLQDGFd 5.0
9 DATPEDLNARLe 7.0 28 efslkGpllqwrsGa 4.7
10 GATPQDLkTMle 6.9 29 DYFDLTQDNIITRRNd 4.7
11 GATPEDLNAKLe 6.8 30 LEMKQDLNPMLQDGFd 4.7
12 FDKEWGGIRITc 6.8 31 sAGdwwLksslb 4.5
13 DALYEWGGARIc 6.7 32 saGdwwGksslb 4.4
14 GLYEWGGARITNTDa 6.7 33 sAGdwDLksslb 4.3
15 sATPwDLkTMle 6.6 34 saGdwwLksslb 4.2
16 DALPEWGGARIe 6.2 35 ITDGARIMd 4.0
17 GATPwDLkTMle 6.1 36 DYFDLTQDNIIERRNd 4.0
18 sATPwDLksslb 6.0 37 LEMKQDLNIMLQDGFd 4.0
19 EAWVLEGAMILWKTDc 6.0 38 EAWVLRGAMILWKTDd 3.5
a
[21].
b
[19].
c
[20].
d
[15].
e
BIAcore studies.
leading to the standard cellulose membranes SCM-100, centrations in T-TBS blocking buffer for 16 h at 4 C.
SCM-50, SCM-25, and SCM-10. The membrane The following concentrations of mAb CB4-1 were used:
SCM-50 was prepared three times and each replica 0.1, 1, and 10 lg/ml. The concentration of 1 lg/ml was
was incubated with a different antibody concentration. used with the membranes SCM-10, SCM-25, SCM-50,
A membrane with the same FQ and the same peptides and SCM-100. After washing three times for 10 min
as SCM-50 was prepared but with about the three- to with T-TBS, the second anti-mouse IgG peroxidase-la-
fourfold number of each peptide resulting in the mem- beled antibody (Sigma, Deisenhofen, Germany) was
brane CM2090-50 with 2090 spots. Furthermore, the cel- added at a final concentration of 1 lg/ml in T-TBS
lulose membrane CM6000-50 (FQ = 50%) was prepared blocking buffer for 2 h, followed by washing three times
using six different peptides from Table 1 (entries: 1, 6, with T-TBS. Analysis and quantification of peptide-
14, 20, 26, 36) but with each peptide equally distributed bound mAb CB4-1 was carried out using a chemilumi-
as 1000 reiterations on the array (size: 18 · 24.4 cm). nescence substrate and a Lumi-Imager (Roche Diagnos-
The peptides cover the affinity spectrum from pKdis = 9 tics, Basel, Switzerland). All steps were carried out at
to pKdis = 3.5. room temperature, unless stated otherwise.
All incubation and washing steps were usually carried Analysis and quantification of spot signal intensities
out under gentle shaking. After washing the membrane were executed with the software Genespotter (MicroDis-
with ethanol once (10 min) and three times for 10 min covery GmbH, Berlin, Germany). Genespotter has a
with Tween–Tris-buffered saline (T-TBS, 50 mM fully automatic grid finding routine resulting in repro-
Tris(hydroxymethyl)aminomethane, 137 mM NaCl, ducible signal intensities. The spot signal is calculated
2.7 mM KCl, adjusted to pH 8 with HCl/0.05% Tween from a circular region around the spot center detected
20), the membrane-bound peptide arrays were blocked on the image. The background signal for each spot is
(3 h) with blocking buffer, i.e., blocking reagent (CRB, determined with a safety margin to this circular region.
Northwich, UK) diluted 1:10 in T-TBS containing 5%
(w/v) sucrose, and then washed with T-TBS Standard solid-phase peptide synthesis
(1 · 10 min). Subsequently, the peptide arrays were
incubated with the anti-p24 (HIV-1) monoclonal anti- Soluble peptides were synthesized (50 l mol scale) as
body CB4-1 (mouse IgG2a/k) [17,18] at different con- amides on a multiple synthesizer AMS 422 (Abimed,
Array-based measurement of peptides / A.A. Weiser et al. / Anal. Biochem. 342 (2005) 300–311 303
Fig. 2. (A) Neighborhood of a spot (gray). The size of the neighborhood N indicates that it consists of N · N spots. The black spots represent
reference spots. (B–D) Mapping of the membrane CM6000-50 (18 cm · 24.4 cm) containing 6000 spots (67 rows · 90 columns). Light areas denote
relatively high and dark areas low signal intensity values. (B) Relative signals of the high-affinity reference peptide (about 1000 data points,
pKdis = 9), (C) background (6000 data points), (D) a weak-binding peptide (about 1000 data points, pKdis = 4). Note that maxima and minima of
mapping (B) have different locations compared to (C and D). The scale gives the relation between the local and the global mean signal.
necessary—one peptide with high-affinity and one with were calculated for the CM6000-50 membrane using
low-affinity to the binding partner. The algorithm for background signals as the low-affinity reference and
improving measured SIs involves the following five considering different neighborhood sizes N and different
steps: first, calculating the mean SI values of all high- percentages of high-affinity reference peptide replicas
ðH global Þ and low-affinity references ðL global Þ; second, (Fig. 3B). As expected, increasing the number of high-
defining a neighborhood i for each spot with N spots affinity reference peptides improves the efficiency of
around it (Fig. 2A); third, calculating the mean SI values the method and reduces the optimal neighborhood size
of the high- ðH local;i Þ and low-affinity references ðL local;i Þ N (Fig. 3B). The results suggest that good adjustment
within each neighborhood i; fourth, calculating the vari- for regional trends can be achieved when around 4%
ables a and b with the equations L global ¼ ai L local;i þ bi of the spots on a membrane are high-affinity reference
and H global ¼ ai H local;i þ bi ; fifth, the improved signal peptides and by using a neighborhood with N · N spots,
intensity for each spot in its neighborhood i (SIimproved) when N is between 10 and 20.
can then be calculated from the measured signal intensi- In practice, there are often no replicas available on a
ties (SImeasured). membrane or no high-affinity peptide is known before-
hand. In such cases, SIimproved can be calculated by using
SIimproved ¼ ai SImeasured þ bi . ð1Þ
only the background signal around each spot according
improved
The standard deviation of the SI values for to
replicas of the same peptide will be smaller than that global;i
L
of SImeasured. As an alternative to a low-affinity reference SIimproved ¼ SImeasured . ð2Þ
peptide, the background signals of each spot can also be Llocal;i
used as references. The resulting standard deviations are The improvements achieved by applying Eqs. (1) and
shown for the SCM-50 membrane in Fig. 3A (stars). The (2) are compared in Fig. 3C. Note that the decrease in
efficiency of improvement is influenced by the neighbor- standard deviation is smaller when using Eq. (2), espe-
hood size N and the percentage of reference peptides cially for peptides with high signal intensities.
regularly distributed on the membrane. To determine We compared SI measurements made from two iden-
optimal parameters, standard deviations of SIimproved tical membranes prepared with the same peptides, the
Array-based measurement of peptides / A.A. Weiser et al. / Anal. Biochem. 342 (2005) 300–311 305
Fig. 3. Standard deviation of the signal intensity versus size of the neighborhood and the mean signal intensities, respectively, after applying the noise
reduction algorithm. (A) Standard deviations of the signal intensities for each of the 38 different peptides (10–15 spots each) on the SCM-10
membrane (Fig. 1) versus their mean values (circles). Gray diamonds denote the standard deviation after application of the noise reduction algorithm
using only background SI values as a reference. Black stars show the effect of the noise reduction algorithm using background SI values and high-
affinity peptide repeats (14% of total spots) as references. The standard deviation obtained for the corrected signal intensities depends on the size of
the chosen local neighborhood (N) and on the number of reference peptides in it. (B) and (C) show the standard deviation of corrected signal
intensities for two peptides, one with low and one with high binding affinity, using the background and a high-affinity reference peptide (B) or only
the background as reference (C). In (B), four cases where the numbers of high-affinity reference peptides are 14, 7, 4, and 2% of the total number of
spots on the membrane are shown. The signals are derived from membrane CM6000-50.
same number of spots, the same concentration of anti- Correlation between signal intensities and dissociation
body, the same FQ, and the same incubation time. We constants
found signal intensity deviations of around a factor of
10. Nevertheless, the relations of the signals were fairly A representative repertoire of 38 peptides was se-
constant for all peptides, suggesting that, to compare lected from a set of over 100 peptides, all known to
SIs from different membranes, these values should first interact with mAb CB4-1 (Table 1). All chosen peptides
be normalized using are well described in the literature and dissociation con-
stants (pKdis) for the various peptide/antibody com-
SImeasured SImin
SInorm ¼ . ð3Þ plexes span a range of more than five orders of
SImax SImin magnitude (pKdis = 9 to pKdis = 3.5). As an example,
Note that, below, this equation will be used only for bet- mean values of measured SIs obtained from a SCM-10
ter visualization of the intermediate region (Fig. 4), with membrane were used to draw an SI/pKdis scatter plot
SI being the mean signal intensity values for each recur- (Fig. 4A). It demonstrates that the obtained SIs corre-
rent peptide. late with given pKdis. To describe the SI/pKdis depen-
306 Array-based measurement of peptides / A.A. Weiser et al. / Anal. Biochem. 342 (2005) 300–311
Fig. 4. Mean signal intensities versus dissociation constants of all measured peptides. All signal intensities are normalized according to Eq. (3). The
gray region marks all signals that are weaker than the mean background signal. (A) Mean values of the 38 different peptides on the SCM-10 membrane
with an antibody concentration of [AT] nM. The solid curve was derived by fitting Eqs. (4)–(7) to the data. The scale of the SI axis is logarithmic. (B–D)
Fitting of the results from several experimental setups with all parameters held constant except one: (B) two membranes prepared with a different
number of peptide replicas, (C) two membranes loaded with different FQ, (D) two membranes incubated with different antibody concentrations.
dency mathematically and elucidate understand which define a total effective antibody concentration ½Aeff T for
parameters influence the correlation, we defined a mass this space. Furthermore, the surface density of accessi-
action law model. It is based on the assumption that the ble peptides on a spot is not known. It may be influ-
system is in equilibrium and that the mass action law is enced by positive cooperativeness: cluster formation
the underlying effect. Furthermore, we have to take into of several peptide molecules within a spot may decrease
account that competition among the peptides for mAb its concentration [25,26]. We define a total effective
CB4-1 might take place, especially in cases when peptide peptide concentration per spot ½P effT and assume that
concentrations or the number of spots are high com- it is the same for all spots. Unfortunately, it is not pos-
eff
pared to the antibody concentration. sible to determine the parameters ½P eff
T and ½AT exper-
Since cellulose membranes are porous, the concen- imentally. We therefore estimate these effective
tration of antibody within the fibers may be different parameters by the proposed mass action law model.
from the antibody concentration [AT] in the bulk solu- To formulate the model, we first define Eqs. (4)–(6).
eff
tion. The inter-fiber spaces could be on the order of Here, ½P eff
i and [A ] denote the concentrations of un-
magnitude of antibody molecules in size. As described bound peptide within a spot i (i = 1, . . . , n) and un-
by Minton [24], this could lead to confinement effects bound antibody, respectively.
and modified diffusion lengths. Since peptide–antibody
interactions primarily occur in the inter-fiber spaces, we ½Aeff ½P eff
i ¼ K disi ½AP i ; ð4Þ
Array-based measurement of peptides / A.A. Weiser et al. / Anal. Biochem. 342 (2005) 300–311 307
½P eff eff
T ¼ ½P i þ ½AP i ; ð5Þ of the fitted total effective peptide concentration from
280 to 42 pM. This result is supported by the fact that
X
n
absolute measured SIs decrease with increasing FQ also
½Aeff eff
T ¼ ½A þ ½AP i . ð6Þ
(data not shown). The decrease of effective accessible
i¼1
peptide concentration with increasing FQ can be ex-
Eq. (4) simply states the mass action law for one sin- plained by a clustering effect [25,26], suggesting that
gle spot, Eq. (5) sums the unbound and bound fractions much lower FQs than those used here could also yield
of the peptide population in spot i, and Eq. (6) sums the good results.
concentration of free antibody with the sum of all com- Second, the fitted total effective antibody concentra-
plexes to the given total antibody concentration. This tion ½Aeff
T does not increase with the same order of mag-
describes a nonlinear system of 2n + 1 equations with nitude as the concentration [AT] originally applied in the
eff
the 2n + 1 variables ½P effi , [APi], and [A ] and the bulk solution. While [AT] increases by a factor of 100,
eff eff
parameters ½AT and ½P T . Furthermore, it is reasonable ½Aeff
T increases only by a factor of approximately 4. Most
to assume that the concentration of the antibody–pep- likely, the total effective concentration of the antibody
tide complex [APi] within a spot is proportional to its saturates within the membrane and one consequence is
measured signal intensity and to introduce the back- a nonlinear correlation between the total effective con-
ground signal as an additive factor. Thus, we addition- centration within the spots and the concentration of
ally define Eq. (7) with the proportionality factor a the antibody in solution.
and the background parameter b: Third, several competition effects were observed. Two
SIi ¼ f ðpK disi Þ ¼ a ½AP i þ b. ð7Þ membranes with identical antibody concentration and
FQ and the same peptides, but with a different number
Eqs. (4)–(7) are used to fit the free parameters ½Aeff T ,
of spots, i.e., different reiteration of the peptides (see
½P eff
T , and b to the measured experimental data using Materials and methods), show very different signal
the functions ‘‘nls’’ and ‘‘nls.lm’’ from the R language behavior. This can be understood if one considers that
environment [27]. We exclude parameter a from the fit- many spots with high-affinity bind many antibody mole-
ting procedure to avoid redundancies and due to the fact cules, thus depleting the solution of free antibodies. Few-
that a differs from membrane to membrane. Instead, it er antibody molecules are then available for binding to
max
was estimated by the formula a ¼ SI½P 0 , where ½P 0T de- peptides with lower affinity, so they will have lower SIs
T
notes the sweeping start parameter of the fitting process compared to when there are only a few high-affinity spots
for ½P eff
T .
and therefore less competition for antibody (Fig. 4B).
To evaluate the relevance of the model, we specifi- Furthermore, competition for antibodies may also
cally varied several experimental parameters: the num- possibly explain an effect described by Kramer et al.
ber of spots on a membrane, the FQ, and the antibody [16]: signal intensities either increase or decrease with
concentration. The fitting process was accomplished the FQ of the membrane, depending on the peptide
for each binding experiment. The correlation coefficients investigated. In a simulation of this experiment applying
between the mean values of the measured SI and the fit- Eqs. (4)–(7), we compared two model results where only
ted SI were found to be between 0.83 and 0.91. Table 2 the peptide concentration was varied: ½P eff T ¼ 100 nM
eff
summarizes all fitting results and the experimental con- and ½P eff
T ¼ 400 nM, both with ½A T ¼ 10 lM and seven
ditions used. Figs. 4B–D show the best fits to the ob- peptides (pKdis = 9 up to pKdis = 3) each replicated 15
tained experimental data after normalization with Eq. times. The results shown in Fig. 5 illustrate that peptides
(3). with high binding affinity show decreasing signal inten-
First, we observed that the effective peptide concen- sities for decreasing ½P eff
T (and hence increasing FQ),
tration increases with decreasing FQ. Increasing the while the converse applies to peptides with low binding
membrane FQ from 10 to 100% resulted in a decrease affinity.
Table 2
Several fitted and experimental parameters for different experimental setups
No. Membrane Spots FQ (%) [AT] (nM) a b ½P eff
T ðpMÞ ½Aeff
T ðnMÞ Correlation
1 SCM-10 607 10 6.3 5.2e+13 2100 280 180 0.88
2 SCM-25 607 25 6.3 6.6e+13 3100 140 190 0.86
3 SCM-50 607 50 0.63 8.7e+13 1400 130 70 0.89
4 SCM-50 607 50 6.3 7.4e+13 2800 90 190 0.86
5 SCM-50 607 50 63 6.9e+13 2500 100 260 0.85
6 SCM-100 607 100 6.3 1.0e+14 2400 40 190 0.83
7 CM2090-50 2090 50 6.3 9.8e+12 300 100 100 0.91
308 Array-based measurement of peptides / A.A. Weiser et al. / Anal. Biochem. 342 (2005) 300–311
Table 3
Calculated pKdis and SI borders with their corresponding contingency coefficients (R) and AuC values
No. Membrane FQ (%) [AT] (nM) R SI1 SI2 pK 1dis pK 2dis AuC
1 SCM-10 10 6.3 0.92 4500 8900 6.8 7.2 0.97
2 SCM-25 25 6.3 0.92 3300 7400 5.3 6.8 0.97
3 SCM-50 50 0.63 0.96 3700 9200 6.8 7.2 0.98
4 SCM-50 50 6.3 0.90 3700 6500 5.3 6.8 0.96
5 SCM-50 50 63 0.93 3200 7600 5.3 6.8 0.95
6 SCM-100 100 6.3 0.83 3200 4800 5.3 6.8 0.90
For calculating the AuC value the SI border of the ROC curve was the mean value of the background signals plus three times its standard deviation
(see Fig. 6B).
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 expð2 IðSI; pK dis ÞÞ between low- and high-binding affinity classes. Fur-
RðSI; pK dis Þ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi . thermore, we determine the border for separating
1 expð2 minðH ðSIÞ; H ðpK dis ÞÞÞ
low and high signal intensities (SI1) with the criteria
ð12Þ described above (background signal plus three times
Note that, when the classifications SI and pKdis are inde- its standard deviation), finally resulting in four quad-
pendent, then H(SI) + H(pKdis) = H(SI, pKdis) and rants nij (i,j = 1,2). With these, we calculate true posi-
therefore R(SI, pKdis) = 0. The larger the value of the tive (TP) rates and false positive (FP) rates
contingency coefficient R, the more information ðTP ¼ n11nþn
11
21
; FP ¼ n12nþn
12
22
Þ. Moving the border pK 1dis
the two classifications have in common. We calculate over the range of all possible values will change TP
the R values for a sweep of the parameters pK 1dis , and FP. Plotting (TP, FP) points into a diagram
pK 2dis , SI1, and SI2 and determine the borders with the and connecting those gives a ROC curve (Fig. 6B).
best R value. The area under the ROC curve is a measure of the
This procedure was applied to several cellulose mem- quality of discrimination between the two classes.
branes that we examined and the resulting contingency Random prediction would result in a bisecting ROC
coefficients lay between 0.83 and 0.96 (Table 3) showing line, giving an AuC value of 0.5. The closer the
high correlation for the distinction of the three defined AuC is to 1.0, the better is the classification. The
classes of peptide binding affinities. SIs larger than SI2 SI1 border of background signal plus three times its
may be associated with a dissociation constant smaller standard deviation results in AuC values of up to
than pK 2dis and SIs smaller than SI1 may be associated 0.98 for the differently synthesized membranes (Table
with a dissociation constant larger than pK 1dis . The inter- 3), suggesting that signal intensity is an excellent clas-
mediate class between the two borders pK 1dis and pK 2dis sifier for high- and low-affinity binders. Note that the
isolates a small region of dissociation constants (‘‘dy- best pK 1dis value for the distinction of high and low-af-
namic range’’), so that measured signals inside the corre- finity is defined by the experimental setup and will
sponding SI class may be associated with an usually not be known by the experimenter. The exam-
approximate dissociation constant. Under the often ple ROC curve for membrane SCM-25 shown in Fig.
used condition (FQ = 50% and [AT] = 6.3 nM), the dy- 6B suggests a good pK 1dis value at 6.7: 90% of peptides
namic range of the curve is between pK 1dis ¼ 5.3 and classified as having high affinity in fact belong to the
pK 2dis ¼ 6.8. class of peptides with pK dis < pK 1dis , while only 7%
Applying the transinformation to more than the three of the peptides in this class were wrongly classified
classes described above led to singleton classes (data not as being low-affinity binders ðpK dis > pK 1dis Þ. An even
shown). This suggests that any other attempt to classify higher accuracy of classification can be achieved when
the results of the SPOT technology into more classes mean SI values of replica peptides are used.
would be meaningless. For instance, it seems to be
impossible to assign a dissociation constant to each
measured signal. This is due to the extreme variability Conclusions
of the data in the central class and the saturation taking
place far off the dynamic range. We describe a method to reduce the standard devia-
In the data investigated we found that the lower tion of signals measured with the SPOT technique. It
border for signal intensities (SI1) never exceeds the is able to approximately halve the standard deviation
background signal plus three times its standard devia- (Fig. 3A), resulting in improved signal intensities for fur-
tion. To confirm that this value is an objective border ther analysis. To make the best use of this method, we
for distinguishing between the two classes of high- and suggest that around 4% of the spots on a membrane
low-affinity peptides we applied ROC statistics to the should be homogeneously distributed repeats of a refer-
measured data. Contrary to the above, we need to de- ence peptide possessing high-affinity toward the ligand,
fine only one border pK 1dis as the border distinguishing if available. This leads to (i) optimal results in the algo-
310 Array-based measurement of peptides / A.A. Weiser et al. / Anal. Biochem. 342 (2005) 300–311
rithm for reducing signal noise due to regional trends changing the FQ or using a differently coated mem-
and (ii) optimal conditions for automatic grid position- brane, e.g., CAPE, as described [23]. Furthermore,
ing by the software that measures signal intensities. For the ligand concentration in the bulk solution can be
experiments where no strong binding peptides are varied.
known beforehand, the noise reduction algorithm can
still be applied with only background signals as refer-
ences. In this case noise reduction is suboptimal (Fig. Acknowledgments
3A).
We report good agreement between model results This work was in part supported by the Deutsche
based on the mass action law (Fig. 4A). The model takes Forschungsgemeinschaft (SFB 618 and SFB 449).
into account competition among peptides for free anti- M.O.G and V.T. thank the Volkswagen Foundation
body molecules. It describes several observed effects, for funding. We thank W. Hoehne (Institut für Bioche-
such as a shift in the ‘‘dynamic’’ range and hence the mie, Berlin, Germany) for providing the CB4-1 anti-
border separating high- and low-affinity classes when body. Excellent technical assistance by I. Kretzschmar
only the number of peptide replicas on a membrane is is gratefully acknowledged. We thank N. Bruni for help-
changed (Fig. 4B) and the apparently contradictory ful criticism and comments.
behavior of different peptidesÕ SIs when experiments per-
formed with different amino-functionalization (FQ) val- References
ues are compared. This observation suggests that
competition effects do indeed play a role in the usual [1] M.B. Eisen, P.O. Brown, DNA arrays for analysis of gene
experimental setup and that, due to this effect, inter- expression, Methods Enzymol. 303 (1999) 179–205.
membrane SI comparisons have to be handled with care [2] E. Fehrenbach, D. Zieker, A.M. Niess, E. Moeller, S. Russwurm,
even after normalization of the values (but note that this H. Northoff, Microarray technology - the future analyses tool in
exercise physiology, Exerc. Immunol. Rev. 9 (2003) 58–69.
effect does not disrupt the ability to classify peptides [3] S.P. Gygi, Y. Rochon, B.R. Franza, R. Aebersold, Correlation
according to their affinity when the SIs used were mea- between protein and mRNA abundance in yeast, Mol. Cell. Biol.
sured on a single membrane). It also suggests that a de- 19 (1999) 1720–1730.
crease in effective peptide concentration, possibly by [4] Y. Ho, A. Gruhler, A. Heilbut, et al., Systematic identification of
dramatically decreasing FQ, could help diminish this protein complexes in Saccharomyces cerevisiae by mass spectrom-
etry, Nature 415 (2002) 180–183.
effect. [5] A.C. Gavin, M. Bosche, R. Krause, et al., Functional organiza-
Fitting of the model to experimental data to estimate tion of the yeast proteome by systematic analysis of protein
experimental parameters suggests that the effective con- complexes, Nature 415 (2002) 141–147.
centration of accessible peptides on the cellulose mem- [6] L. Giot, J.S. Bader, C. Brouwer, et al., A protein interaction
brane decreases for increasing FQ. It is expected that map of Drosophila melanogaster, Science 302 (2003)
1727–1736.
for an FQ much smaller than 10% this behavior can [7] S. Li, C.M. Armstrong, N. Bertin, et al., A map of the
be reversed. Moreover, the fitted parameters suggest interactome network of the metazoan C. elegans, Science 303
that the effective ligand concentration within the mem- (2004) 540–543.
brane fibers increases sublinearly with increased ligand [8] R. Frank, The SPOT-synthesis technique. Synthetic peptide arrays
concentration in the bulk solution. on membrane supports—principles and applications, J. Immunol.
Methods 267 (2002) 13–26.
Classification of peptides into two classes representing [9] U. Reineke, R. Volkmer-Engert, J. Schneider-Mergener, Appli-
high and low binding affinities is possible with high pre- cations of peptide arrays prepared by the SPOT-technology, Curr.
dictive power using the measured signal intensities as cri- Opin. Biotechnol. 12 (2001) 59–64.
teria. The SI threshold separating both classes is best [10] H. Wenschuh, R. Volkmer-Engert, M. Schmidt, M. Schulz, J.
chosen as the mean value of the SIs of the spot back- Schneider-Mergener, U. Reineke, Coherent membrane supports
for parallel microsynthesis and screening of bioactive peptides,
ground signals plus three times its standard deviation. Re- Biopolymers 55 (2000) 188–206.
cently, the method derived here was successfully applied [11] L. Otte, U. Wiedemann, B. Schlegel, J.R. Pires, M. Beyermann,
to characterizing PDZ domain/ligand specificity [13,14]. P. Schmieder, G. Krause, R. Volkmer-Engert, J. Schneider-
The SPOT technique also allows differentiation be- Mergener, H. Oschkinat, WW domain sequence activity relation-
tween three classes of dissociation constants: high, ships identified using ligand recognition propensities of 42 WW
domains, Protein Sci. 12 (2003) 491–500.
low, and intermediate. The most frequently used [12] C. Landgraf, S. Panni, L. Montecchi-Palazzi, L. Castagnoli, J.
experimental condition (FQ = 50% and [AT] = 6.3 nM) Schneider-Mergener, R. Volkmer-Engert, G. Cesareni, Protein
yields an intermediate class whose affinities lie between interaction networks by proteome peptide scanning, PLoS Biol. 2
pK 1dis ¼ 5.3 and pK 2dis ¼ 6.8. This dynamic range, and (2004) 94–103 (Epub 2004 Jan 20).
with that the pK 1dis border of classification in two clas- [13] P. Boisguerin, R. Leben, B. Ay, G. Radziwill, K. Moelling, L.
Dong, R. Volkmer-Engert, An improved method for the synthesis
ses, can be shifted by altering the effective concentra- of cellulose membrane-bound peptides with free C termini is
tions of the binding partners. There are two useful for PDZ domain binding studies, Chem. Biol. 11 (2004)
possibilities for varying the peptide concentration: 449–459.
Array-based measurement of peptides / A.A. Weiser et al. / Anal. Biochem. 342 (2005) 300–311 311
[14] U. Wiedemann, P. Boisguerin, R. Leben, D. Leitner, G. Krause, [20] U. Hoffmüller, T. Knaute, M. Hahn, W. Höhne, J. Schneider-
K. Moelling, R. Volkmer-Engert, H. Oschkinat, Quantification Mergener, A. Kramer, Evolutionary transition pathways for
of PDZ domain specificity, prediction of ligand affinity and changing peptide ligand specificity and structure, EMBO J. 19
rational design of super-binding peptides, J. Mol. Biol. 343 (2004) (2000) 4866–4874.
703–718. [21] A. Kramer, T. Keitel, K. Winkler, W. Stöcklein, W. Höhne, J.
[15] R. Volkmer-Engert, B. Ehrhard, J. Hellwig, A. Kramer, W. Schneider-Mergener, Molecular basis for the binding promiscuity
Höhne, J. Schneider-Mergener, Preparation, analysis and anti- of an anti-p24 (HIV-1) monoclonal antibody, Cell 91 (1997) 799–
body binding studies of simultaneously synthesized soluble and 809.
cellulose-bound HIV-1 p24 peptide epitope libraries, LIPS 1 [22] H. Wenschuh, H. Gausepohl, L. Germeroth, M. Ulbricht, H.
(1994) 243–253. Matuschewski, A. Kramer, R. Volkmer-Engert, N. Heine, T.
[16] A. Kramer, U. Reineke, L. Dong, B. Hoffmann, U. Hoffmüller, Ast, D. Scharn, J. Schneider-Mergener, Positionally addressable
D. Winkler, R. Volkmer-Engert, J. Schneider-Mergener, Spot parallel synthesis on continuous membranes, in: H. Fenniri (Ed.),
synthesis: observation and optimizations, J. Peptide Res. 54 (1999) Combinatorial Chemistry: A Practical Approach, Oxford Univer-
319–327. sity Press, Oxford, UK, 2000, pp. 95–116.
[17] R. Grunow, R. Giese, T. Porstmann, H. Döpel, K. Händel, R. [23] R. Volkmer-Engert, B. Hoffmann, J. Schneider-Mergener, Stable
Von Baehr, Developement and biological testing of human and attachment of the HMB-linker to continuous cellulose membranes
murine monoclonal antibodies against HIV antigens, Z. Klin. for parallel solid phase spot synthesis, Tetrahedron Lett. 38 (1997)
Med. 45 (1990) 367–369. 1029–1032.
[18] W.E. Höhne, G. Kuttner, S. Kiessig, G. Hausdorf, R. Grunow, [24] A.P. Minton, The influence of macromolecular crowding and
K. Winkler, H. Wessner, E. Giessmann, R. Stigler, J. Schneider- macromolecular confinement on biochemical reactions in physi-
Mergener, R. von Baehr, D. Schomburg, Structural base of the ological media, J. Biol. Chem. 276 (2001) 10577–10580.
interaction of a monoclonal antibody against p24 of HIV-1 with [25] A.P. Minton, Effects of excluded surface area and adsorbate
its peptide epitope, Mol. Immunol. 30 (1993) 1213–1221. clustering on surface adsorption of proteins I. Equilibrium
[19] U. Reineke, C. Ivascu, M. Schlief, C. Landgraf, S. Gerike, G. models, Biophys. Chem. 86 (2000) 239–247.
Zahn, H. Herzel, R. Volkmer-Engert, J. Schneider-Mergener, [26] P. Minton, Effects of excluded surface area and adsorbate
Identification of antibody epitopes and mimotopes applying a clustering on surface adsorption of proteins. II. Kinetic models,
small single peptide array of 5520 randomly generated sequences, Biophys. J. 80 (2001) 1641–1648.
J. Immunol. Methods 267 (2002) 37–51. [27] http://www.r-project.org.