Printing Proteins as Microarrays for High-Throughput Function Determination

See allHide authors and affiliations

Science  08 Sep 2000:
Vol. 289, Issue 5485, pp. 1760-1763
DOI: 10.1126/science.289.5485.1760


Systematic efforts are currently under way to construct defined sets of cloned genes for high-throughput expression and purification of recombinant proteins. To facilitate subsequent studies of protein function, we have developed miniaturized assays that accommodate extremely low sample volumes and enable the rapid, simultaneous processing of thousands of proteins. A high-precision robot designed to manufacture complementary DNA microarrays was used to spot proteins onto chemically derivatized glass slides at extremely high spatial densities. The proteins attached covalently to the slide surface yet retained their ability to interact specifically with other proteins, or with small molecules, in solution. Three applications for protein microarrays were demonstrated: screening for protein-protein interactions, identifying the substrates of protein kinases, and identifying the protein targets of small molecules.

Historically, genome-wide screens for protein function have been carried out with random cDNA libraries. Most frequently, the libraries are prepared in phage vectors and the expressed proteins immobilized on a membrane by a plaque lift procedure. This method has been effective for a variety of applications (1–4), but it has several limitations. Most clones in the library do not encode proteins in the correct reading frame, and most proteins are not full-length. Bacterial expression of eukaryotic genes frequently fails to yield correctly folded proteins, and products derived from abundant transcripts are overrepresented. Moreover, because plaque lifts are not amenable to miniaturization on the micrometer scale, it is hard to imagine screening all the proteins of an organism hundreds or thousands of times by this approach.

With the advent of high-throughput molecular biology, it is now possible to prepare large, normalized collections of cloned genes. UniGene sets in the form of polymerase chain reaction products have been used extensively over the past decade to construct DNA microarrays for the study of transcriptional regulation (5). Recently, spatially segregated clones in expression vectors were used to study protein function in vivo using the yeast two-hybrid system (6) and in vitro using biochemical assays (7). We have built on these efforts by developing microarray-based methods to study protein function.

To accomplish these goals, it is necessary to immobilize proteins on a solid support in a way that preserves their folded conformations. One group has described methods of arraying functionally active proteins, using microfabricated polyacrylamide gel pads to capture their samples and microelectrophoresis to accelerate diffusion (8). In contrast, we have immobilized proteins by covalently attaching them to the smooth, flat surface of glass microscope slides. One of our primary objectives in pursuing this approach was to make the technology easily accessible and compatible with standard instrumentation. We use a variety of chemically derivatized slides that can be printed and imaged by commercially available arrayers and scanners. For most applications, we use slides that have been treated with an aldehyde-containing silane reagent (9). The aldehydes react readily with primary amines on the proteins to form a Schiff's base linkage. Because typical proteins display many lysines on their surfaces as well as the generally more reactive α-amine at their NH2-termini, they can attach to the slide in a variety of orientations, permitting different sides of the protein to interact with other proteins or small molecules in solution.

To fabricate protein microarrays, we use a high-precision contact-printing robot (10) to deliver nanoliter volumes of protein samples to the slides, yielding spots about 150 to 200 μm in diameter (1600 spots per square centimeter). The proteins are printed in phosphate-buffered saline with 40% glycerol included to prevent evaporation of the nanodroplets. It is important that the proteins remain hydrated throughout this and subsequent steps to prevent denaturation. After a 3-hour incubation, the slides are immersed in a buffer containing bovine serum albumin (BSA). This step not only quenches the unreacted aldehydes on the slide, but also forms a molecular layer of BSA that reduces nonspecific binding of other proteins in subsequent steps.

Although appropriate for most applications, aldehyde slides cannot be used when peptides or very small proteins are printed, presumably because the BSA obscures the molecules of interest. For such applications, we use BSA-NHS (BSA-N-hydroxysuccinimide) slides that are fabricated by first attaching a molecular layer of BSA to the surface of glass slides and then activating the BSA withN,N′-disuccinimidyl carbonate (11). The activated lysine, aspartate, and glutamate residues on the BSA react readily with surface amines on the printed proteins to form covalent urea or amide linkages. The slides are then quenched with glycine. In contrast to the aldehyde slides, proteins or peptides printed on BSA-NHS slides are displayed on top of the BSA monolayer, rendering them accessible to macromolecules in solution.

As a first application of protein microarrays, we have looked at protein-protein interactions. Until now, only the yeast two-hybrid system has been used to investigate such interactions systematically on a genome-wide scale (6). This in vivo method, although easy to implement and of great utility, has several limitations. Proteins that function as transcriptional activators yield false positives when expressed as DNA binding domain fusions. False negatives are encountered when proteins are displayed inappropriately or when the DNA binding domain fusions are produced in excess. Proteins that do not fold correctly in yeast are inaccessible, and posttranslational modifications (such as phosphorylation or glycosylation) cannot be controlled. Finally, it is impossible to control the environment (e.g., ion concentration, presence or absence of cofactors, temperature) during the experiment.

To determine whether microarrays could be used for these types of studies, we selected three pairs of proteins that are known to interact: protein G and immunoglobulin G (IgG) (12); p50 (of the nuclear factor NF-κB complex) and the NF-κB inhibitor IκBα (13); and the FKBP12-rapamycin binding (FRB) domain of FKBP-rapamycin–associated protein (FRAP) and the human immunophilin FKBP12 (12 kD FK506-binding protein) (14). The first two interactions occur without special requirements, whereas the third interaction depends on the presence of the small molecule rapamycin (14). We arrayed the first protein of each pair in quadruplicate on five aldehyde slides and probed each slide with a different fluorescently labeled protein (11).

The slide in Fig. 1A was probed with BODIPY-FL–conjugated IgG, washed, and scanned with an ArrayWoRx fluorescence slide scanner (15). As anticipated, only the spots containing protein G were visible, indicating that the immobilized protein is able to retain its functional properties on the glass surface. Similarly, only the p50-containing spots were visible on the slide probed with Cy3-IκBα (Fig. 1B) (15). For Cy5-FKBP12, binding to FRB was observed only when rapamycin was added (Fig. 1, C and D). Because the three fluorophores used for these studies have nonoverlapping excitation and emission spectra, we were also able to detect these interactions simultaneously (Fig. 1E).

Figure 1

Detecting protein-protein interactions on glass slides. (A) Slide probed with BODIPY-FL-IgG (0.5 μg/ml). (B) Slide probed with Cy3-IκBα (0.1 μg/ml). (C) Slide probed with Cy5-FKBP12 (0.5 μg/ml) and 100 nM rapamycin. (D) Slide probed with Cy5-FKBP12 (0.5 μg/ml) and no rapamycin. (E) Slide probed with BODIPY-FL-IgG (0.5 μg/ml), Cy3-IκBα (0.1 μg/ml), Cy5-FKBP12 (0.5 μg/ml), and 100 nM rapamycin. In all panels, BODIPY-FL, Cy3, and Cy5 fluorescence were false-colored blue, green, and red, respectively.

By varying the concentration of FRB (the protein being immobilized), we found that at concentrations above 1 mg/ml, the fluorescence of the spots began to saturate. Below this, fluorescence scaled linearly with decreasing concentrations of FRB. All proteins immobilized on the slides described here were spotted at 100 μg/ml. Because only a few microliters of each protein are sufficient to fabricate thousands of microarrays, purified proteins may be readily obtained by high-throughput expression and purification, or even by in vitro transcription/translation (16).

Much lower concentrations are needed for the solution-phase protein. In the case of Cy5-FKBP12, fluorescence scaled linearly with protein concentration over four orders of magnitude (11). Specific binding could be detected using Cy5-FKBP12 concentrations as low as 150 pg/ml (∼12.5 pM). Concentrations in this range are accessible not only with purified proteins, but also with fluorescently labeled proteins from cell lysates. Thus, specific interactions, once defined, may potentially be exploited to quantify protein abundance and modification in whole cells or tissues.

At the spot density used for these studies, it was possible to fit more than 10,000 samples in about half the area of a standard (2.5 cm by 7.5 cm) slide. To investigate the feasibility of detecting a single specific interaction in this larger context, we prepared a slide containing 60 rows and 180 columns of spatially separated spots. Protein G was spotted 10,799 times on this slide, with a single spot of FRB in row 27, column 109. The slide was then probed with a mixture of BODIPY-FL-IgG and Cy5-FKBP12, with 100 nM rapamycin included in the buffer. Figure 2 shows the single FRB spot, clearly visible in the sea of protein G spots.

Figure 2

A single slide holding 10,800 spots. Protein G was printed 10,799 times. A single spot of FRB was printed in row 27, column 109. The slide was probed with BODIPY-FL-IgG (0.5 μg/ml), Cy5-FKBP12 (0.5 μg/ml), and 100 nM rapamycin. BODIPY-FL and Cy5 fluorescence were false-colored blue and red, respectively.

Although it is of great value to identify stable protein-protein interactions in a system such as a cell or tissue, it is equally important to define the transient interactions that occur between enzymes and their substrates. Protein microarrays offer an ideal system, for example, for the rapid and parallel identification of the substrates of protein kinases. To investigate this application, we chose three different kinase-substrate pairs: adenosine 3′,5′-monophosphate–dependent protein kinase (PKA) and Kemptide (a peptide substrate for PKA) (17); casein kinase II (CKII) and protein phosphatase inhibitor 2 (I-2) (18); and p42 mitogen-activated protein (MAP) kinase (Erk2) and Elk1 (19). The protein substrates of each pair were spotted in quadruplicate onto three BSA-NHS slides, and each slide was incubated with a different kinase in the presence of [γ-33P]adenosine triphosphate.

Although isotopic labeling of the protein spots is the most direct way to identify phosphorylation, the challenge lies in detecting the radioactive decay. Neither x-ray film nor conventional PhosphorImagers offer sufficient spatial resolution to visualize the spots, which are 150 to 200 μm in diameter. Borrowing from the technique of isotopic in situ hybridization, we dipped the slides in a photographic emulsion and developed them manually; this resulted in the deposition of silver grains directly on the glass surface. The slides were then visualized using an automated light microscope (20) and individual frames were stitched together. As anticipated, only the specific substrates for each enzyme were phosphorylated (Fig. 3).

Figure 3

Detecting the substrates of protein kinases on glass slides. (A) Slide incubated with the catalytic subunit of PKA. (B) Slide incubated with CKII. (C) Slide incubated with p42 MAP kinase (Erk2).

As the third and most demanding application, we sought to use protein microarrays to identify protein–small molecule interactions. With the advent of high-throughput, cell-based screening, more and more compounds are being identified on the basis of their biological activity. Once a “hit” is obtained, the daunting task of target identification remains. Several innovative techniques have been developed to address this bottleneck (4, 21–23), but they all suffer from the common limitations imposed by using random cDNA libraries. As an alternative, we sought to develop microarray-based assays that use purified, full-length, correctly folded proteins.

To test this approach, we chose three unrelated small molecules for which specific protein receptors are available: DIG, a derivative of the steroid digoxigenin that is recognized by a mouse monoclonal antibody (24); biotin, a common vitamin recognized by the bacterial protein streptavidin (25); and AP1497 (Fig. 4), a synthetic pipecolyl α-ketoamide designed to be recognized by FKBP12 (26). The proteins from all three pairs were spotted in quadruplicate on four aldehyde slides, and each slide was probed with a different small molecule. Rather than labeling the compounds directly, each ligand was coupled to BSA that had previously been labeled with a unique fluorophore (Alexa488, Cy3, or Cy5) (15). As anticipated, fluorescence localized to the appropriate spots in all three cases (Fig. 5, A to C). Because the fluorophores used for these studies have nonoverlapping excitation and emission spectra, we were also able to detect all three interactions simultaneously (Fig. 5D).

Figure 4

Synthetic ligands for FKBP12. The compounds were coupled to BSA through their carboxyl groups (via a flexible linker).

Figure 5

Detecting the targets of small molecules on glass slides. (A) Slide probed with Alexa488-BSA-DIG. (B) Slide probed with Cy5-BSA-biotin. (C) Slide probed with Cy3-BSA-AP1497. (D) Slide probed with Alexa488-BSA-DIG, Cy5-BSA-biotin, and Cy3-BSA-AP1497. All conjugates were used at a concentration of 10 μg/ml. In all panels, Alexa488, Cy3, and Cy5 fluorescence were false-colored blue, green, and red, respectively.

To investigate our ability to detect low-affinity interactions, we prepared Cy3-BSA conjugates of compounds AP1497, AP1767, and AP1780 (Fig. 4; dissociation constants for FKBP12 of 8.8 nM, 140 nM, and 2.6 μM, respectively). When three identical slides displaying FKBP12 were probed in parallel, spots with comparable fluorescence intensities were obtained for all three conjugates (11). This means that interactions in the micromolar range can easily be observed. The fact that the intensity of the fluorescence did not vary appreciably as the affinity of the interaction was lowered can be attributed to the multivalency of the BSA conjugates (avidity effects). In the context of small-molecule microarrays (27), we have previously shown that when these three compounds are immobilized on a glass surface and then probed with Cy5-labeled FKBP12 (a monomeric protein), the intensity of the fluorescence correlates very well with the affinity of the interaction. Thus, by controlling the valency of the probe, we can choose whether to observe differences in affinity or to favor the detection of low-affinity interactions. The combination of these two approaches may prove useful in the identification of both primary and secondary drug targets.

Although traditional biochemical methods have yielded invaluable insight into protein function on a case-by-case basis, they cannot realistically be applied to the study of every protein in a cell, tissue, or organism. If we hope to assign function on a broader level, we must turn to miniaturized assays that can be performed in a highly parallel format. It is certainly a daunting task to express and purify thousands of different proteins, and some proteins will inevitably prove refractory to biochemical manipulation. Nonetheless, the effort will be worthwhile if the many proteins that are amenable can be assayed both simultaneously and repeatedly. By fabricating protein microarrays, we can fulfill both these criteria, facilitating the in vitro study of protein function on a genome-wide level.

  • * To whom correspondence should be addressed. E-mail: gavin_macbeath{at}


View Abstract

Navigate This Article