Profiling receptor tyrosine kinase activation by using Ab microarrays

Abstract

Signal transduction in mammalian cells is mediated by complex networks of interacting proteins. Understanding these networks at a circuit level requires devices to measure the amounts and activities of multiple proteins in a rapid and accurate manner. Ab microarrays have previously been applied to the quantification of labeled recombinant proteins and proteins in serum. The development of methods to analyze intracellular signaling molecules on microarrays would make Ab arrays widely useful in systems biology. Here we describe the fabrication of multiplex Ab arrays sensitive to the amounts and modification states of signal transduction proteins in crude cell lysates and the integration of these arrays with 96-well microtiter plate technology to create microarrays in microplates. We apply the Ab arrays to monitoring the activation, uptake, and signaling of ErbB receptor tyrosine kinases in human tumor cell lines. Data obtained from multicolor ratiometric microarrays correlate well with data obtained by using traditional approaches, but the arrays are faster and simpler to use. The integration of microplate and microarray methods for crude cell lysates should make it possible to identify and analyze small molecule inhibitors of signal transduction processes with unprecedented speed and precision. We demonstrate the future potential of this approach by characterizing the action of the epidermal growth factor receptor inhibitor PD153035 on cells by using Ab arrays; direct scale-up to array-based screening in 96- and 384-well plates should allow small molecules to be identified with specific inhibitory profiles against a signaling network.

The signal transduction systems that control cellular physiology are comprised of biochemical networks with shared components, common inputs, and overlapping outputs. Understanding how signals flow through these pathways, how the pathways vary among cell types, and how normal and diseased tissues differ requires information on signaling networks as a whole rather than simply on one or two components. To make network (or systems) biology possible, we need devices that can probe the activities of signaling proteins in a parallel and reliable manner. We envision these as a biological analog of the multiprobe “bed of nails” testers that are a mainstay of the electronics industry. Bed of nails testers can monitor printed circuit boards at enough locations to fully trace and test a circuit.

In this paper we describe the development of an Ab microarray integrated with 96-well microtiter plates that can quantify the amounts and modification states of ErbB receptors in crude cell lysates. Ab microarrays are an extension of DNA microarrays. In both cases, ratiometric comparisons derived from differentially labeled control and experimental samples are an effective way to standardize measurements among and within experiments (1). Ab arrays have the potential to reveal the amounts and modification states of proteins and also, when integrated with fractionation steps, subcellular protein compartmentalization. The use of Ab arrays has previously been described to quantify proteins in serum and to measure the levels of fluorescently labeled recombinant proteins (2–6). It might be assumed that building arrays for cell signaling processes represents a direct extension of this technology. However, we and others (7) have discovered that reducing array-based analysis of signaling proteins to practice has required new fabrication and experimental methods.

To determine the critical steps in fabricating Ab arrays for signal transduction, we have focused on early events in ErbB receptor activation (8). The epidermal growth factor receptor (EGFR or ErbB1) is a prototypical receptor tyrosine kinase whose intracellular domain becomes phosphorylated on a series of tyrosine residues after activation by EGF (9). ErbB2 (also known as HER2) is a structurally related protein that does not appear to bind extracellular ligands but is a potent oncogene (10, 11). ErbB2 is phosphorylated in response to EGFR activation (12), and EGFR and ErbB2 act together to regulate cellular proliferation. Misregulation of EGFR and ErbB2 is implicated in a wide variety of cancers, and a humanized mAb against ErbB2, Herceptin, is effective for the treatment of metastatic breast cancer (13). We show here that Abs specific for EGFR, ErbB2, and their tyrosine-phosphorylated forms can be used to monitor the levels and activities of receptor tyrosine kinases in a multiplexed, ratiometric microarray format. We use Ab microarrays and a panel of tumor cell lines to demonstrate five applications of microarrays to the study of ErbB signaling: (i) profiling protein abundance, (ii) profiling the functional state of a signaling system, (iii) analyzing the kinetics of ligand-activated signaling, (iv) measuring the in vivo inhibitory constant of a small molecule EGFR inhibitor, and (v) array-based profiling in 96-well plates for screening. We find that cell lines differ markedly not only in the levels of ErbB proteins that they express, as expected, but also in their responsiveness to EGF activation. A direct scale-up of the Ab microarrays described here to 10–20 independent elements will permit the systematic analysis of a complete signal transduction system in normal and diseased tissues in a rapid and parallel fashion.

Materials and Methods

Ab Array Methods. BSA-N-hydroxysuccinimide slides were prepared as described (14). Anti-ErbB2 mAb clone 3B5, anti-pY1248-ErbB2 mAb clone PN2A, anti-EGFR mAb clones 199.12, and anti-EGFR 111.6 were purchased from Lab Vision (Fremont, CA). Anti-transferrin receptor (TfR) mAb clones 7F8 and 11F5 were purchased from Research Diagnostics (Flanders, NJ). Polyclonal rabbit anti-pY1068-EGFR was purchased from BioSource International (Camarillo, CA). Recombinant humanized mAb Herceptin was obtained from the pharmacy. Anti-ErbB2 scFv F5 was produced as described (15) (a kind gift from James D. Marks, University of California, San Francisco). BSA-N-hydroxysuccinimide slides were prepared as described (14) and affixed to bottomless 96-well plates (Greiner, Nurtingen, Germany) by using adherent precut gaskets (Grace BioLabs, Bend, OR). Abs clones 111.6, 11F5, and Herceptin were spotted at 0.5 mg/ml in PBS containing 40% glycerol by using a GMS 417 Arrayer (Affymetrix, Santa Clara, CA), and the resulting Ab microarrays were stored at 4°C. Typically, 12 Ab arrays (or 96 for plates) were spotted per slide and separated with a hydrophobic pen or silicone gasket. Slides were blocked with glycine and BSA immediately before use (14). Detection Abs 3B5, 199.12, 7F8, anti-pY1068, and F5 scFv were labeled with Cy3, Cy5 (Amersham Biosciences), or Alexa 488 (Molecular Probes) as recommended by the manufacturer. Recombinant extracellular domain (ECD) of ErbB2 and EGFR was expressed in Chinese hamster ovary cells [kind gifts of James D. Marks and Greg Adams (Fox Chase Cancer Center, Philadelphia), respectively]. Purified transferrin receptor was obtained from Research Diagnostics. Recombinant human EGF was obtained from PeproTech (Rocky Hill, NJ). The small molecule inhibitor PD153035 (4-[(3′-bromophenyl)amino]-6,7-dimethoxyquinazoline) of EGF receptor kinase was purchased from Calbiochem and dissolved at 2 mM in DMSO before use. All cell lines were obtained from American Type Culture Collection and cultured in the recommended media. Extracts of cells grown in 6- or 12-well tissue culture plates were prepared by passing cells five times through a 27-gauge needle in lysis buffer (20 mM Tris, pH 7.5/150 mM NaCl/1 mM EDTA/1 mM EGTA/1% Triton X-100/0.5% Nonidet P-40/10 mM; alternatively, the lysis buffer contained 0.25% SDS in place of Triton X-100 and Nonidet P-40) containing phosphatase inhibitors (β-glycerolphosphate/10 mM NaF/1 mM Na₃VO₄) to minimize changes in the phosphorylation state after lysis. In addition, protease inhibitors were added (1 mM PMSF/1 μg/ml leupeptin/1 μg/ml pepstatin) to reduce protein degradation, and incubations were carried out on ice. For fluorescent cell surface labeling, cells were washed five times in cold PBS, incubated with 10 mg/ml of fluorescein-polyethylene glycol 2000-N-hydroxysuccinimide (Shearwater Polymers, Huntsville, AL) on ice for 2 h, and then washed five times with cold PBS before blocking with 100 mM glycine in PBS. Before lysis, cells were washed twice in cold PBS. Arrays were incubated with lysates mixed 1:1 with 2% BSA in PBS containing 0.1% Tween for 3 h at 4°C before washing three times in PBS containing 0.1% Tween and three times in PBS. Arrays in 96-well plates were washed with an automated plate washer (BioTek). When the sandwich assay was performed, the arrays were further incubated with fluorescently labeled detection Abs for 1 h at room temperature and washed again. Arrays were quantified by using an Applied Precision (Issaquah, WA) ArrayWoRx scanner or on a Tecan (Durham, NC) LS400 (in the case of 96-well plates). Spots were quantified by using arrayworx software (Applied Precision) and fluorescent signals corrected for local background.

Immunoblotting and Flow Cytometry. Abs PN2A and anti-pY1068 were used for immunoblotting at 1:1,000 dilution and detected with goat–anti-mouse-horseradish peroxidase or goat–anti-rabbit-horseradish peroxidase, respectively. Ab clones 111.6, 11F5, and Herceptin were used for flow cytometry at 2 μg/ml and detected with goat–anti-mouse FITC or goat–anti-human FITC (Sigma).

Results

Developing an Ab Microarray for Cell Signaling Molecules. To build an Ab array, mAbs specific to EGFR, ErbB2, and TfR were printed onto BSA-coated glass slides at a density of ≈1,600 spots per cm² by using a contact printing robot (14). The activated BSA on the slides serves to passivate the surface and to covalently link the Abs. We first tested the sensitivity and linearity of the Ab arrays by using recombinant proteins labeled directly with dye (Fig. 1A). Extracellular domains of the three receptors were conjugated to fluorophores as follows: ErbB2 to Alexa 488 (blue), EGFR to Cy5 (red), and TfR to Cy3 (green; TfR serves as a control for a receptor that is essentially unaffected by EGF stimulation). Specific binding of labeled antigens to arrays could be detected over a range of 0.1–100 ng/ml (Fig. 1 A and C). Next, we tested an indirect detection method based on sandwich assays. Unlabeled antigens, individually or in a mixture, were bound to “capture” Abs linked to BSA-coated slides. The slides were then washed and the captured antigens detected by the addition of a mixture of labeled secondary Abs (detection Abs; Fig. 1B). One detection Ab is required for each antigen, and it must bind to a different epitope than that recognized by the capture Ab, permitting an Ab sandwich to form. When we used this indirect microsandwich detection, we could quantify purified antigens over a 1,000-fold range and down to ≈1 ng/ml, making indirect detection ≈10-fold less sensitive than direct labeling (Fig. 1C). We conclude from these data that our high-density Abs microarrays, using either direct or indirect detection, are linear over a wide-range of antigen concentrations and at least as sensitive as standard ELISA methods (6).

Fig. 1.

Multiplex detection of protein antigens on Ab microarrays. (A) Detection of directly labeled recombinant antigen. Ab arrays were incubated with 25 ng/ml purified Cy3-labeled TfR ECD (green), Alexa 488-labeled ErbB2 ECD (blue), and Cy5-labeled EGFR ECD (red) individually or with the proteins as a mixture (fourth panel). (B) Detection of recombinant antigen by using microsandwich assays. Arrays were incubated with unlabeled purified antigens at 25 ng/ml, individually or as a mixture, and detected with a mixture of Cy3-labeled anti-TfR Ab (clone 7F8, green), Alexa 488-labeled anti-ErbB2 Ab (clone F5, blue), and Cy5-labeled anti-EGF receptor Ab (clone 199.12, red). (C) Sensitivity of EGFR detection. Slides were incubated with varying amounts of labeled recombinant EGFR ECD (Left), and fluorescence was quantified. Alternatively, varying amounts of unlabeled recombinant EGFR ECD (Right) was bound to arrays and detected by using fluorescently labeled anti-EGFR. All data represent triplicate spots from a single array; error bars represent SDs. (D) Detection of antigens in cell lysates by direct labeling of antigen or by microsandwich assay. A-431 and SK-BR-3 cancer cell lines were surface-labeled with polyethylene glycol-conjugated fluorescent dyes and antigens detected on Ab microarrays. Alternatively, antigens in cell lysates from A431 cells were detected with fluorescently labeled second Abs. (E) Multiplex detection of antigens in lysates of tumor cell lines by using microsandwich assays. Slides were incubated with unlabeled cell lysates from various cell lines and detected with a mixture of Alexa 488-labeled anti-TfR Ab (clone 7F8, blue), Cy5-labeled anti-ErbB2 (clone 3B5, red), and Cy5-labeled anti-EGFR (clone 1991.2, red). For comparison, the relative abundance of each receptor was measured on whole cells by flow cytometry by using the same Abs (Right).

Next, we tried to measure the levels of EGFR, ErbB2, and TfR in A-431 squamous carcinoma and SK-BR-3 breast cancer cell lysates (Fig. 1D). A-431 cells have high EGFR and low ErbB2 levels, whereas the opposite is true of SK-BR-3 cells. TfR serves as a control protein whose levels should not vary among experiments. We first attempted direct detection because it appeared to be more sensitive and because it represents a direct extension of previous work. After considerable testing and optimization, we determined that our Ab arrays worked best when cells were labeled with fluorescent dyes conjugated to polyethylene glycol and extracts prepared by lysis in dilute SDS buffer. When A-431 and SK-BR-3 cells were labeled with activated fluorescein-polyethylene glycol and then analyzed on Ab arrays, ErbB2 and EGFR could be detected in the expected relative amounts (Fig. 1D). Sample-to-sample variation was high, however, and TfR was undetectable. In contrast, when unlabeled cell extracts were analyzed by using microsandwich assays, all three cell surface proteins could be detected, the coefficient of variation among repeat determinations was typically <10%, and a very good correlation (r > 0.99 for ErbB2 and EGFR) was observed between receptor levels measured by using microarrays and the receptor levels measured by using conventional flow cytometry (Fig. 1 D and E). Thus, although both direct labeling and microsandwich methods can be used with Ab arrays to detect cell surface receptors expressed at endogenous levels, the microsandwich method appears to work better with cells and cell extracts and, given the variability of cell-surface labeling reactions, is better suited to high-throughput biology. Experiments with purified proteins suggest that direct labeling has a higher ultimate sensitivity. We suspect low dye-to-receptor stoichiometry to be the primary problem with direct labeling of cell lysates. It is also possible that coupling receptors to dyes via lysine residues interferes with their binding to Abs (although this did not appear to be a problem in experiments with recombinant proteins).

Ratiometric Profiling ErbB-Signaling Activity. For microarray-based profiling to provide useful data on cell signaling proteins, it is important to monitor their state of activation. We therefore asked whether levels of phosphotyrosine 1068 (pY1068) on EGFR and phosphotyrosine 1248 (pY1248) on ErbB2 could be detected on microarrays by using phospho-specific Abs. The phosphorylations at EGFR-Y1068 and ErbB2-Y1248 are excellent measures of receptor activation because the sites are modified, in trans, by receptor autophosphorylation and are binding sites for the Grb2 adapter protein that initiates downstream signaling via Ras (16, 17).

To monitor activation, receptors in cell lysates were captured on arrays by using pan-specific capture Abs that are insensitive to the state of receptor tyrosine phosphorylation. The arrays were then probed with a mixture of two detection Abs, a Cy3-labeled phospho-specific Ab and a Cy5-labeled pan-specific Ab different from the capture Ab (Fig. 2). No crossreactivity was observed when the detection Abs were used individually (results not shown). The ratio of Cy3 to Cy5 fluorescence at each spot is proportional to the fraction of receptor that is phosphorylated at Y1068 (EGFR) or Y1248 (ErbB2). The amount of TfR (detected by using an Alexa 488 conjugate) controls for variations in extract preparation. In most experiments with a single cell line, TfR signals were within 15% of each other when a constant amount of cell extract was applied to arrays, as judged by total cell protein (data not shown). Thus, both total protein and TfR intensities can be used for normalization, with similar results.

Fig. 2.

Ratiometric detection of receptor tyrosine phosphorylation and expression. The abundance and phosphorylation of EGFR and ErbB2 was measured in the cell lines A-431, SK-BR-3, MCF-7, and ErbB2-transfected MCF-7 before and after 5 min of treatment with EGF. (A) Analyses of the response of SK-BR-3 cells to treatment with EGF. Ab arrays were incubated with cell lysates, and the captured antigens were analyzed with a mixture of five fluorescently labeled detection Abs: Alexa 488-labeled anti-TfR Ab (blue), Cy5-labeled pan-specific anti-ErbB2 (clone 3B5) and anti-EGFR (clone 199.12) Abs (red), and Cy3-labeled phospho-specific anti-pY1248 ErbB2 (clone PN2A) and anti-pY1068 EGFR (pAb pY1068) Abs (green). Signals were quantified, and the ratio of Cy3 signal (phospho-specific Ab) to Cy5 signal (pan-specific Ab) was determined. Immunoblotting was performed on the same SK-BR-3 cell lysates by using pan- and phospho-specific Abs. (B and C) Analyses of response of A-431 and MCF-7 cells transfected with ErbB2 to EGF, as described in A.(D) Linearity and sensitivity of the microsandwich assays as determined by dilution experiments with lysates. Lysates of MCF-7 cells transfected with ErbB2 were diluted into MCF-7 lysates to determine the linearity of array-based determinations of ErbB2 protein and ErbB2 phosphorylation levels and the sensitivity of measurements of the stoichiometry of phosphorylation to sample dilution. Error bars represent SDs of three replicate spots.

To activate signaling, cells were treated with EGF for 5 min and extracts analyzed for the amount of total and phosphorylated EGFR and ErbB2. Three tumor lines were compared to uncover cell-type specific differences. In SK-BR-3 cells, tyrosine phosphorylation was stimulated on EGFR but not on ErbB2; in A-431 cells, the levels of ErbB2 and EGFR tyrosine phosphorylation increased 3- to 4-fold on EGF treatment; and in ErbB2-transfected MCF-7 cells, the levels of pY1248-ErbB2 decreased on EGF stimulation (Fig. 2; EGFR levels are undetectable in ErbB2-transfected MCF-7 cells by flow cytometry and on arrays; data not shown). In all cases, microarray data correlated well with data from the same extracts obtained by immunoblotting (Fig. 2 A and data not shown, r > 0.98). It is possible to view the microarray data in several different ways. For example, when normalized to TfR (or total protein), the levels of tyrosine-phosphorylated EGFR and ErbB2 (5 min after EGF stimulation) were observed to vary 8-fold or more among different cell types. The most reproducible measurements of protein phosphorylation from experiment to experiment (with a coefficient of variation typically ≤5%) were obtained by calculating the ratio of signals for phospho-dependent and phospho-independent Abs to generate ratiometric data that correlate with the fraction of receptor activated.

To demonstrate that ratiometric measurements on receptor phosphorylation lie within the linear range of the microsandwich method, lysates from ErbB2-transfected MCF-7 cells were diluted into lysates from nontransfected MCF-7 cells, which have very low endogenous levels of ErbB2 (Fig. 2D). When the diluted lysates were analyzed on arrays, Cy5 and Cy3 fluorescence measures of ErbB2 abundance and Y1248 phosphorylation were observed to vary linearly over at least the 16-fold range that spans our assay conditions. The ratio of Cy3 to Cy5 fluorescence, which is proportional to the fraction of phosphorylated receptor, varied <40% over the same 16-fold range of receptor concentrations (Fig. 2D). Thus, ratiometric measurements from Ab microarrays are relatively insensitive to variables such as sample loading and constitute reliable measures of receptor phosphorylation (with an error of approximately ±25%). In our hands, this level of accuracy is at least as good as that obtained with Western blots.

We conclude from these data that Ab microarrays represent a simple, accurate, and rapid method to assay the protein phosphorylation events associated with cell signaling. The rapidity and accuracy make it possible to detect clear differences in ErbB signaling systems among different cell types. Extending such analysis to human breast cancer samples would make it possible to characterize ErbB signaling systems in patients before treatment with therapeutics directed against the ErbB family of receptors.

Kinetics of ErbB Signaling. Cell signaling is a highly dynamic process in which time-dependent changes in protein activities are critical. To explore the utility of Ab arrays in analyzing the kinetics of ErbB activation, we analyzed extracts from SK-BR-3 cells over the course of 2 h after treatment with EGF (Fig. 3). We observed that TfR levels remained steady throughout the experiment and that the coefficient of variation of these measurements was <5% (over three determinations of six time points; Fig. 3B). The total amounts of detectable EGFR and ErbB2 dropped steadily over the course of the EGF treatment (Fig. 3C), reflecting receptor internalization and degradation. The half-life of EGFR as measured on microarrays was ≈1.5 h, identical to the value reported previously, and the half-life of ErbB2 was ≈6–7 h, also consistent with previous data (18). The amount of phospho-Y1248 ErbB2 was highest at t = 0 and then decreased steadily, whereas phospho-Y1068 EGFR rose 5-fold by 1 min and then fell steadily. The fraction (rather than absolute amount) of EGFR that was phosphorylated remained above basal levels throughout the experiment, consistent with the idea that EGF signaling has an early acute phase and then a subsequent sustained phase (19). Overall, these data reveal that Ab microarrays can be used to monitor the earliest steps of ErbB receptor signaling.

Fig. 3.

Microarray analysis of receptor activation. (A) SK-BR-3 cells were treated with EGF for 2 h, and cell lysates were analyzed at different times during this treatment by using Ab microarrays and microsandwich detection, as described in the Fig. 2 legend. (B) Quantification of transferrin receptor levels as a control for repeatability. (C) Quantification of protein abundance, phosphorylation, and the fraction of phosphorylated receptor for EGFR and ErbB2. See the Fig. 2 legend for details.

Microarray Analyses After Perturbation with Small Molecule Inhibitors. The connectivity of signaling networks is often investigated by selectively perturbing the networks by using small molecules. Moreover, these small molecules can be of considerable pharmaceutical interest. To investigate whether Ab microarrays might be used in combination with small molecule inhibitors for this purpose, we treated A-431 cells with varying amounts of the EGFR tyrosine kinase inhibitor PD153035 (20) (Fig. 4A). As expected, the phosphorylation of EGFR on Y1068 decreased with increasing inhibitor concentration, as did ErbB2 PY1248, consistent with the idea that ErbB2 is activated by EGFR and that both are inhibited in vivo by PD153035 (12, 21). The apparent in vivo K_i for PD153035 as estimated by the array method was ≈20 nM, well above the reported in vitro K_i for EGFR of 5 pM but consistent with the IC₅₀ of 14 nM (20). It is thought that PD153035 is much less potent on cells than on purified receptors as a consequence of competition between the drug and endogenous ATP.

Fig. 4.

Microarray analysis of a tyrosine kinase inhibitor and integration of microarrays and microplates. (A) Inhibition of EGFR and ErbB2 phosphorylation in cells treated with the small molecule PD 153035. Error bars represent SDs of three replicate spots. (B) MIMs. Ab arrays were printed on glass slides and then attached to bottomless 96-well plates. A common pool of lysate from A431 cells was then added to each well, the plates were washed by using a BIOTEK plate washer, and bound antigen was detected by using a mixture of fluorescently labeled Abs as described in Fig. 1. Each well contained 18 spots, and portions of four wells are shown magnified. The coefficient of variation among similar spots in different wells was ≈20%. To the right, the IC₅₀ for PD 153035 on EGFR-PY1068 was determined from cells grown in 24-well multi-plates, lysed in situ, and analyzed by using MIMs. Error bars represent the mean and range of triplicate spots.

Integrating Microarrays in Microtiter Plates (MIMs). High-throughput biological analysis requires efficient handling of multiple samples. This is accomplished most effectively by using 96- and 384-well microtiter plates. To determine whether our Ab microarrays can produce reliable data from samples prepared in parallel in microtiter plates, we printed 18 spots at the bottom of each well in a 96-well plate, added a common stock of unfractionated cell lysate by using a liquid handling robot, and then washed arrays by using automated plate washers. Plates were then imaged in reader capable of scanning 96-well plates (Fig. 4B). When ratiometric measurements were performed on receptor abundance and phosphorylation the accuracy of the measurements was high (the well-to-well coefficient of variation was ≈20%). Similarly high accuracy was observed when cells were grown in 24-well plates, treated with various cytokines and small molecules, lysed in situ, and then transferred to arrays. By way of illustration, we have determined the IC₅₀ of PD153035 by the MIM method to be 18 nM. These and similar data demonstrate that MIMs will be accurate enough for cell profiling and screening of small molecules. The importance of ratiometric measurements for MIMs emphasizes the value of multispectral fluorescence detection as opposed to chemiluminescence or single-color fluorescence methods (5, 22).

Discussion

In this paper we demonstrate the fabrication and use of Ab microarrays to study ErbB signal transduction in human cells. The application of Ab microarrays to the detection of recombinant proteins and proteins in serum has been described (2, 5, 22), and it might be assumed that fabricating arrays that can monitor intracellular signal transduction is a simple extension of these methods. However, we are not aware of other reports using Ab microarrays to monitor signaling in cells. Assaying endogenous proteins in crude lysates by using microarrays requires effective surface passivation, ratiometric measurements (necessitating multiwavelength fluorescence, as opposed to chemiluminescence) (5), and Abs with high selectivity and affinity. Here we show that suitably fabricated Ab microarrays can be used to determine the abundance of receptors in cells, profile signaling in different cells types, study small molecule inhibitors, and monitor the kinetics of signal transduction. When compared to DNA microarrays, Ab microarrays provide much more data per element. They also appear to be about as accurate. The overall coefficient of variation for repeat measurements of protein abundance and phosphorylation levels for the data in this paper was ≈15%.

Both direct fluorescent labeling and indirect sandwich detection can be used for quantifying proteins bound to microarrays. Each method has strengths and weaknesses. The most obvious advantage of direct labeling is that it requires only one Ab per antigen. However, our data suggest that it may be difficult to incorporate sufficient label into low abundance proteins to make direct labeling routinely useful. Presumably, this problem will be overcome in the future through the use of instruments designed for label-free detection (23). In our hands, indirect microsandwich assays are clearly superior to direct labeling in sensitivity and accuracy (as determined by the variation between repeated experiments). The use of detection Abs specific for different states of an antigen also makes it possible to determine the abundance and modification states of a protein simultaneously. It might be argued that Ab arrays based on fluorescence microsandwich assays will be useful in the short term but will eventually be replaced by direct label-free detection. However, the use of two Abs to generate a signal has important advantages in selectivity that cannot be matched by direct detection by using even the most exotic instrumentation. Very few Abs, including those used in this paper, are truly monospecific when used to probe cell extracts; the vast majority also bind to at least one other cellular antigen. The microsandwich assay achieves exquisite selectivity without the size fractionation afforded by immunoblotting because the specificities of two different Abs are exploited. Very rarely will the capture and detection Abs bind to the same extraneous protein. Unless affinity reagents more effective than Abs can be developed, sandwich methods will enjoy fundamental advantages for the analysis of very complex protein mixtures.

Although polyclonal, monoclonal, and recombinant Abs can be used for detection in microsandwich assays, we are disappointed to be describing arrays with so few elements. In very recent work, we have managed to add additional elements for several other kinases and signaling proteins. However, fewer than one in 20 of the commercial Abs we have tested are suitable for microarray-based analysis of cell lysates, and it appears that the requirements for the capture Ab are the most demanding (the detection Ab appears to be less critical). Thus, we are currently exploring conventional and recombinant methods to isolated new capture Abs by using MIMs for automated screening (Fig. 4B). If these efforts succeed, then MIMs capable of a few dozen measurements will become as useful for intracellular signaling as cytokine arrays have become from serum profiling. We do not believe that the reliance of microsandwich methods on solution-phase detection will prove a significant hurdle. When MIMs are used, microsandwich-based Ab microarrays can be split among several wells, each of which contains perhaps 20 immobilized capture Abs in triplicate and 20–40 detection Abs (22), thereby avoiding background problems introduced by Ab mixtures.

In conclusion, we have successfully used a prototype Ab microarray to analyze signal transduction in mammalian cells. Arrays containing a modest increase in the number of independent elements are easily within reach. Future Ab microarrays with several dozen elements should be able to monitor information flow within a signal transduction system with unprecedented precision. Such a network view of cell signaling will be invaluable for cell biologists, for drug discovery, and for investigating the differences between normal and diseased tissues.