Developments and Applications of Functional Protein Microarrays*

The proteome is the entire set of proteins that can be expressed by a genome. The development of a purified proteome microarray usually requires assembly of a genome-wide collection of open reading frames (ORFs) cloned into an expression vector, expression of the encoded proteins in cells, individual protein purification in a high-throughput fashion, and immobilization of the proteins on a microarray. Advances in purified proteome microarrays for model organisms, such as S. cerevisiae, E. coli, humans, and Arabidopsis thaliana, have propelled functional and biochemical studies of proteins to a proteomic level. The first of its kind is the S. cerevisiae (budding yeast) proteome array, developed by the Snyder group in 2001 and containing 5,800 full length yeast proteins (). Currently, there are many purified proteome microarrays covering a wide variety of model systems, including coronaviruses (), flaviviruses (), human herpesviruses (), M. tuberculosis (), E. coli K12 (), S. cerevisiae (), Arabidopsis thaliana (, ), and humans (, ). Because of the coverage of ORF collections and the efficiency of protein expression/purification, the proteome coverage on such arrays ranges from 56% to 95% (Table I). The choice of protein expression system greatly influences post-translational modifications and can affect the success rate of protein purification. For example, because of a lack of eukaryotic posttranslational modifications and chaperones, proteins encoded by C. elegans were poorly expressed in E. coli, with an expression rate of 48%. Of this 48%, only 15% were soluble (). Therefore, homologous expression systems are generally preferred to obtain the highest protein activity and expression efficiency. The S. cerevisiae, E. coli, and Arabidopsis thaliana proteome arrays are three of the best examples for use as homologues expression systems. In some cases, especially with mammalian cells, it is difficult and expensive to transfect cells, and thus one can use an alternative expression system, such as budding yeast, to accommodate protein production pipelines. Indeed, the human proteome microarray (i.e. HuProt) is one of the best examples to use a heterologous expression system, as it exhibits the most comprehensive human proteome collection purified from yeast (81% proteome coverage). Another commercial human proteome microarray, called ProtoArray, contained >9,000 human proteins purified from insect cells (43% proteome coverage), but was discontinued in 2018.

A protein family microarray is designed to interrogate specialized groups of proteins for their biochemical functions. Today, there are many different protein family microarrays, each used for different purposes. For example, one can utilize a G-protein coupled receptor (GPCR) array for pharmaceutical applications (), a membrane/secreted protein array for profiling autoantibodies (), a hemagglutinin antigen array for investigating influenza vaccines (, ), and a gp120/140 array from HIV for analyzing immune responses (). Because most protein family microarrays have a relatively small number of proteins, the expression system can be tailored for desired qualities and quantities. For example, the GPCR array is developed using Virion Display (VirD) technology () to maintain the seven transmembrane structure and to obtain the best GPCR expression in several mammalian cell lines, including Vero, HEL, HeLa, and 293T cells ().

Alternatively, protein domain microarrays can be designed to analyze certain regions, domains, or epitopes within the proteins. These arrays often involve the careful design of desired gene sequences before entering the protein production pipeline. Protein domain arrays, Protein Epitope Signature Tag (PrEST) arrays, and consensus sequence protein arrays are the three best examples of this sort. The protein domain arrays reported by Jones et al. contain all the human Src homology 2 and phosphotyrosine binding domains to profile the interaction networks for tyrosine phosphorylation on ErbB receptors (). PrEST arrays contain the unique signature in the human proteome developed by the Human Protein Atlas Consortium for identifying multiple sclerosis autoantibodies () or for validating antibody specificity (). In a consensus sequence protein array, Qi et al. summarize 44 consensus serotype sequences out of 3604 different dengue strains and construct a protein array accordingly for dengue serotyping (). Overall, both purified proteome, protein family, and protein domain arrays have a wide variety of applications in basic and translational research, as well as pharmaceutical industry.

The cell-free protein/peptide microarray is designed to display a short peptide or full-length protein using a cell-free system. Cell-free expression is designed to bypass the expensive and often tedious work of cell-based protein production. To construct protein assays with an in vitro expression, many expression systems, including expression lysate from E. coli, insect cells, wheat germs, and human cells, are commercially available. For instance, the Felgner Lab established various pathogen arrays ranging from viruses to bacteria and yeasts by using an in vitro transcription and translation (IVTT) system adopted from E. coli (Table I and footnote) (, ). On the other hand, the LaBaer group utilized a DNA array, dubbed as the Nucleic Acid Programmable Protein Array (NAPPA), to construct human proteome arrays using in vitro transcription/translation system (, , ). Because cell-free expression lacks regulated protein folding, segregated cellular compartments, and coordinated post-translational modifications (PTMs), the protein functions are not guaranteed (). The IVTT system also suffers from a lower yield of larger proteins (e.g. >50 kDa), potential contamination by other proteins presented in the lysates, and low array density (e.g. ∼2,000 features per array) (). Nevertheless, protein arrays produced by cell-free expression are quite useful to analyze immune responses (, , ).

Functional protein microarrays, especially purified proteome microarrays, are useful for profiling proteome-wide molecular interactions and allow for a comprehensive, unbiased screening. In basic research, researchers have been using functional protein microarrays to study protein-protein interactions, protein-lipid interactions, protein-cell/lysates, protein-DNA interactions, protein-RNA interactions, small molecule binding, and PTMs, such as glycosylation, ubiquitylation, SUMOylation, acetylation, phosphorylation, and methylation (Fig. 1A–1G). In Table II, we summarize representative studies based on the research applications illustrated in Fig. 1. Here, we review research studies based on the proteomes immobilized on microarrays.

Zhu et al. constructed the very first proteome microarray, the yeast proteome microarray, and utilized it to investigate protein-protein interactions and protein-lipid interactions. The array was probed with biotinylated calmodulin and 33 new calmodulin binding proteins with new common motifs were identified (). In the same study, the yeast proteome array was probed with fluorescently labeled liposomes carrying various phosphatidyl-inositides and more than 150 phospholipid binding proteins were identified (). Huang et al. used the same microarray to identify binding proteins for two small molecule inhibitors of rapamycin, SMIR3 and SMIR4, and identified 8 and 30 protein targets, respectively. Most target proteins were involved in PI_3,4P₂ signaling (). Hall et al. used the yeast proteome microarray to profile DNA binding proteins and revealed a mitochondrial enzyme, Arg5,6, can regulate both nuclear and mitochondrial gene expression (). Similarly, Zhu et al. used the yeast proteome microarray to profile RNA hairpin binding proteins and identified two proteins: Pus4 and App1. Their antiviral activity against the spread of brome mosaic virus was demonstrated in tobacco (). The Zhu lab further demonstrated the utility of proteome arrays by performing covalent enzymatic reactions on the arrays. They were the first to establish the protein acetylation reactions using the yeast NuA4 complex, and two parallel signaling pathways in yeast aging were discovered (). It has also been applied to determine the substrates of a HECT domain ubiquitin E3 ligase Rsp5 (). These studies demonstrate the usefulness of the yeast proteome microarray in basic research.

Chen et al. established a purified E. coli proteome microarray in 2008, comprising of 4256 unique proteins and applied it to identify potential new players in the DNA damage response. The E. coli proteome microarray was probed with several short DNA probes containing mismatched base pairs or abasic sites, and two DNA repair proteins were identified: YbaZ and YbcN (). In another study the same array was used to detect DNA binding proteins to the promoter of type 1 fimbriae and identified Spr as a phase switch for type 1 fimbria expression (). Ho et al. probed several antimicrobial peptides using the E. coli proteome array and identified many intracellular targets. Among the four antimicrobial peptides, they identified some shared and unique targets and suggested a synergistic effect on LfcinB and Bac7, as well as LfcinB and PR-39 (). Hsiao et al. probed the E. coli proteome array with four glycosaminoglycans that are common on host cells and identified a hundred protein targets. They further validated ycbS as a bacterial factor for cell entry (). Xu et al. probed the E. coli proteome array with an important bacterial second messenger, cyclic di-GMP, and identified CobB as a strong binder. Because CobB is a deacetylation enzyme, they subsequently found that cyclic di-GMP inhibits the enzymatic activity and forms a novel feedback loop to the cyclic di-GMP production (). Feng et al. used E. coli proteome microarray to investigate protein-cell interactions by probing the human brain microvascular endothelial cells (HBMEC) on the array. They identified 23 target proteins and validated YojI as a protein for E. coli invasion. Moreover, they purified Yojl, probed using HuProt, and further identified interferon-alpha receptor as a host receptor for Yojl (). Besides various binding assays, the E. coli proteome microarray has also been applied to identify substrates, including substrates of glycoproteins (), tyrosine sulfation (), and ClpYQ protease (). As demonstrated by these representative works, the E. coli proteome microarray is widely used to study bacterial physiology as well as host-microbial interactions.

The human proteome microarray is the most widely used array in basic research, translational research, and in the pharmaceutical industry. There are three popular human proteome microarrays: HuProt, ProtoArray, and NAPPA. HuProt contains ∼21,000 individual purified human proteins in full-length, which is by far the most comprehensive human proteome collection. ProtoArray contained ∼9000 human proteins purified from insect cells, but was discontinued commercially in 2018. NAPPA is an in vitro expression system that has been applied to express 10,000 human proteins.

The HuProt array was not made overnight. In its early stages, it contained 4191 unique human proteins, mostly transcription factors and co-factors. Hu et al. performed a large scale DNA-binding assay with 460 DNA motifs on this array and found 17,718 protein-DNA interactions. Not only were numerous known protein-DNA interactions recovered, but they also found many unconventional DNA-binding proteins, including a mitogen-activated protein kinase (MAPK), Erk2. In-depth mutagenesis studies and cell-based assays demonstrated that Erk2 acts as a transcriptional repressor in the regulation of interferon-gamma signaling (). In 2012, the Zhu lab published the construction of HuProt version I, which contained 16,368 individual purified human proteins in full-length and demonstrated that it could serve as a useful tool to identify highly specific monoclonal antibodies (). The work laid the foundation for the NIH-funded Protein Capture Reagents Program (PCRP; https://commonfund.nih.gov/proteincapture).

The birth of HuProt arrays expanded researchers' arsenal for interrogation of a great fraction of the entire human proteome for specific biochemical properties. For example, Liu et al. profiled the binding specificities of 13 long noncoding RNAs (lncRNAs) on HuProt to determine potential players in lncRNA-mediated biological processes. Ultimately, 671 lncRNA-binding proteins were found, 525 of which lacked any known RNA-binding domains. A novel RNA binding protein, IDH1, was further validated in cells and shown to bind thousands of RNA transcripts (). Similarly, Fan et al. probed HuProt with miR-122 and identified 40 target proteins. Because miR-122 is required for hepatitis C virus (HCV) replication, they further validated the target hnRNP K as a repressor for HCV replication (). Therefore, the human proteome microarray is a valuable tool to study the complex regulatory networks of protein-DNA and -RNA interactions (Fig. 1F and 1G).

The human proteome microarray is also useful for the analysis of protein-protein interactions, especially for determining players involved in pathogen-host interactions (Fig. 1A). Park et al. probed the nonstructural 5A protein from HCV on ProtoArray and identified 90 proteins. They further validated one of these proteins, Pim1, as a factor involved in HCV cell entry (). Yoon et al. constructed a Zika virus-host protein-protein interaction network using a similar approach and compared its dengue virus counterparts to determine Zika virus-specific interactions (). Further orthogonal large-scale screenings allowed them to pinpoint drug targets in the host involved in Zika virus replication. Yang et al. investigated the binding events of T. gondii virulence factor ROP18 using HuProt and identified 68 targets. They subsequently validated the crucial role of ROP18 on p53, p38, UBE2N, and SMAD1 through phosphorylation-dependent degradation (). Wu et al. investigated the binding events of PknG, an important kinase in M. tuberculosis (MTB), using HuProt and identified 128 binding proteins. They further validated that one of these binding proteins, CypA, is degraded upon phosphorylation and subsequently inhibits inflammatory responses (). Using human NAPPA, Yu et al. identified 18, 20, and 8 host proteins that interact with L. pneumophila effectors SidM, LidA, and AnkX, respectively (, ).

Human proteome microarrays have also been widely used to study PTMs (Fig. 1E). Song et al. established methods to detect global tyrosine phosphorylation, lysine acetylation, ubiquitylation, and SUMOylation on HuProt. The HuProt arrays were incubated with cell lysates diluted in different PTM reaction buffers to perform covalent protein modifications, and the modified proteins on the array were visualized using the corresponding PTM antibodies. Among the complex regulation of PTMs in cancers, they validated the hyperactivities of PTK2 and PTK2B kinases in ovarian cancer (). Xu et al. surveyed the substrate for ppGalNAc-Ts using HuProt and identified 128 common substrates for O-GalNAc glycosylation (). Yu et al. used human NAPPA to identify the 20 and 21 AMPylation substrates for VopS and IbpAFic2, respectively (). Overall, the human proteome microarray serves as an unbiased platform for studying many kinds of binding events and enzyme-substrate relationships.

The two major pharmaceutical applications of the human proteome microarray are drug target identification (Fig. 1D) and specificity tests for monoclonal antibodies (mAbs) (Fig. 1H). HuProt was used to identify the targets of arsenic, a cancer drug, and 360 potential binders were identified. Hexokinase was validated to bind arsenic, and this binding event was further shown to result in the inhibition of glycolysis (). With a similar strategy, Cheng et al. screened the targets of 6-O-angeloylplenolin, a drug that induces cell cycle arrest, and identified 99 proteins. The proteins Skp1 and STAT3 were further validated to show involvement in cell cycle arrest (). Because the mAb-based biologicals are one of the fastest growing therapeutic modalities, quality control is extremely important. Many commercial mAbs, however, exhibit poor quality and have wasted $350 million annually in the United States alone (). HuProt arrays are an ideal platform to screen mAbs for mono-specificity (). As such, Venkataraman et al. established a production pipeline for the mAbs against transcription factors and adapted HuProt as a primary validation tool for specificity tests (). Of the 5882 mAbs tested on HuProt arrays, 2000 passed the specificity tests, 1462 of which eventually passed the secondary cell-based validation for their ability to perform Western blot analysis and/or immunoprecipitation.

Proteome microarrays other than those already mentioned can also be quite useful in basic research, such as the Arabidopsis proteome array () and the MTB proteome array () to name a few. Popescu et al. established the Arabidopsis proteome microarray and profiled the binding of calmodulin and calmodulin-like proteins (). Deng et al. developed the MTB proteome microarray and used it to identify the binding partners of PknG and the protein interactions of second messenger cyclic di-GMP (). Wu et al. probed 11 serine/threonine protein kinases on the MTB proteome array and identified 492 binding proteins with 1027 network interactions ().

Serological biomarkers are valuable tools for diagnosis, prognosis and companion diagnosis in various autoimmune diseases, cancers, and infectious diseases (, ). One of the early applications of functional protein microarrays was to discover new serological biomarkers for autoimmune diseases because they can serve as antigen surveying platforms to detect subtle changes in antibody composition. In a dysregulated immune system, the antibodies that are generated by humoral immunity and react with self-antigens are referred to as autoantibodies (AAbs). When a functional protein array covers most of the human proteome (e.g. HuProt), a specific AAb signature can be readily detected by probing the array with a diluted patient serum/plasma sample. When this approach is used to profile AAb signatures for a large cohort, subsequent statistical analysis can reveal potential biomarkers associated with a disease of interest (Table III). This approach has three major advantages. First, patient samples are easy to obtain and store because they are mostly in the forms of serum, plasma or body fluid. Second, detection of AAbs on a human proteome array is very sensitive and quantitative, only requiring several microliters of samples. Finally, the presence of AAbs is detectable before symptoms can be identified, making early diagnosis possible.