1.    Introduction

Antibody microarrays have an unparalleled potential for the simultaneous monitoring of thousands of proteins found in crude unfractionated bio fluids and lysates from cells and tissues, with sensitivities that surpass that of any of the competing technologies. Detection of target proteins within the 1-50 pg/ml range has been successfully achieved with antibody microarrays [1-4]. Microarrays with as many as 5264individual antibody spots have been successfully produced, as demonstrated with our latest generation Kinex^™ KAM-1300 slides from Kinexus. Not only can antibody microarrays be used to track changes in protein expression, but also alterations in post-translational modifications, protein-protein, and protein-drug interactions.

The earliest descriptions of antibody arrays in the scientific literature were by Tse Wen Chang, when he developed matrices of up to 400 antibody spots printed on glass cover slips to sort and quantify immune cells based on their expression of CD and HLA allotypic antigens [5]. However, it was not until the work of Roger Ekins and his colleagues [6-8] and improvements in microarray printing technologies in the 1990s that general interest in the development and applications of antibody microarrays really began to galvanize in the field. As illustrated in Figure 1, the number of annual publications related to “antibody microarrays” rapidly escalated after 2001, but then plateaued after 2008. During this growth period, several commercial antibody microarrays were launched (Figure 1), and this supported many of the published studies that covered a wide range of applications of antibody microarrays, especially for disease biomarker discovery. However, over the last decade, there has been no further growth in the use of antibody microarrays for biomedical research, and several vendors have since discontinued their sales of high content antibody microarrays (Table 1).

2.    Antibodies Probes

Molecular biology-based methods such as DNA amplification with the polymerase chain reaction (PCR), DNA sequencing, site-directed mutagenesis, gene transfection, and recombinant protein expression have provided profound insights into the inner workings of cells. Around 21,300 human genes that encode proteins have become identified, but little is known about the roles and regulation of more than a third of all of these proteins despite their discovery more than 18 years ago. As the main dynamic components that mediate cell structure and function, proteins are highly finicky macromolecules that exhibit diverse structures, enzymatic and other functional activities, and they are subject to a myriad of confounding regulatory controls. A range of immunological techniques that depend on antibodies have been developed to single out proteins so that they can be individually tracked with respect to their expression levels, post-translational modifications, tissue and subcellular locations, and interactions with other proteins. However, this is a daunting task that has actually progressed relatively slowly despite the euphoric adoption of genomic solutions to proteomics problems. While the cost for sequencing entire genomes of humans and other species has plummeted by a million-fold over the last two decades, the pricing of antibodies has not really changed in the last 30 years.

The development of gene microarray technology in which as many as 20,000 complementary oligonucleotide probes for mRNA sequences can be arrayed as individual spots on a 1-inch x 3-inch glass slide actually paved the way for the creation of high density antibody microarrays. In principle, only 10 µg of a particular antibody may be sufficient to be featured as a spot printed with ~0.1 pg of IgG on over 100,000,000 slides. Measurement of a target protein on an antibody microarray requires about 100-times less antibody than required for a typical Enzyme-Linked Immune Sorbent Assay (ELISA) [9]. Consequently, the price of a particular antibody is really secondary to its specificity and potency, and suitability on a microarray.

Over two million antibodies are commercially available, although it is very common for the same antibody to be repackaged and offered by multiple vendors. At Kinexus, we have tested over 6000 commercial antibodies by Western blotting, and have determined that less than a quarter of these have sufficient selectivity and sensitivity to be useful for Western blotting in most applications. Others have found similar results and commented on this problem[10-12]. Nevertheless, with so many antibodies on the market, and the ease with which new antibodies can be produced, these are likely to remain the best reagents for selective binding and detection of proteins and other analytes in the foreseeable future.

The reproducibility and supply of “individual” antibodies have been significant factors in the drive to produce monoclonal antibodies or recombinant antibodies. For example, human recombinant single-chain (Fv) antibodies selected from a large phage display library were successfully incorporated into microarrays [13,14]. However, the costs of monoclonal antibody production are typically 5-times higher than polyclonal antibody production, and monoclonal antibodies are much less humane to produce. Furthermore, in our experience, most of the monoclonal antibodies that we have rigorously tested by Western blotting are no more potent and specific than many of the polyclonal antibodies that are generated against short peptides and affinity-purified. In theory, since the avidity and stability of antibody-antigen complexes is determined by both the affinity of their binding as well as by the valencies of the binding, polyclonal antibodies with similar specificity to monoclonal antibodies should have a better antigen-capturing capacity than monoclonal antibodies due to the formation of more stable complexes with their targets by binding to multiple epitopes. In fact, at least one independent study reported that polyclonal antibodies displayed superiority over monoclonal antibodies in capturing toxins on microarray slides even when they had similar affinities as determined by ELISA [15].

While recombinant antibodies have the potential to be mutated to further improve specificity and potency, in practice, their high costs render their development suitable for primarily therapeutic purposes. These likely accounts for why so few recombinant antibodies are commercially sold for the research market despite the availability of this technology for over 20 years [16]. Nevertheless, some recombinant antibodies have been successfully deployed in low content custom microarrays [17,18].

3.    Proteome Complexity

Although about 21,300 genes encode proteins in the human genome, the actual number of distinct protein entities or proteoforms in the human proteome may exceed 10 million due to alternative splicing and post-translational modifications. About 95% of the multi-exon genes in the human genome have been estimated to undergo approximately 100,000 intermediate- to alternative splicing events [19,20]. Amongst more than 50 types of post-translational modifications documented in proteins, phosphorylation is the predominant reversible regulatory mechanism in eukaryotic cells. Reputable on-line databases such as PhosphoSite Plus (www.phosphosite.org) and PhosphoNET (www.phosphonet.ca) list over 250,000 non-redundant human phosphosites with approximately 60% on serine residues, 25% on threonine residues and 15.5% on tyrosine residues. A significant portion of the studies that have identified phosphosites by mass spectrometry (MS) used selected enrichment of phosphotyrosine containing peptides with generic anti-phosphotyrosine antibodies. This is because even in Rous sarcoma virus-transformed fibroblasts that overexpress the protein-tyrosine kinase Src in a highly active form, only about 0.3% of the protein phosphorylation by weight is on tyrosine residues (93% on serine and 7% on threonine residues) [21]. Using alternative enrichment strategies such as strong cation exchange chromatography or immobilized metal ion affinity chromatography (IMAC), the ratio of the numbers of phosphoserine: phosphothreonine: phosphotyrosine sites is closer to 19:5:1. Based on predictions algorithms developed at Kinexus [22], we have posted about 960,000 known and predicted human phosphosites on the PhosphoNET website. Since over 44,000 human phosphotyrosine sites have already been confirmed by MS, and there are apparently 24 phosphoserines and phosphothreonine sites for every phosphotyrosine site in the human phosphoproteome, this supports the existence of over a million human phosphosites. However, about three-quarters of the reported phosphosites from MS have only been documented once or twice as evident on PhosphoSite Plus [23]. Only about 1.5% of these phosphosites have been commonly recorded in at least 100 separate MS studies, with about 3% of this particular subset of phosphosites detected in high frequency over a thousand MS experiments. Amongst the most frequently phosphorylated human proteins are: the cytoskeletal proteins actins (ACTA1-Y55, ACTA2-Y55, ACTB-Y53, ACTG1-Y53, ACTG2-Y54),calponins (TAGLN2-Y192, TAGLN3-Y192), paxillin-Y88+Y118, talin 1-Y70, VASP-Y39, vimentin-Y53+Y61, and vinculin-Y822; the scaffolding proteins Crk-Y221,CrkL-Y207, CTNND1-Y228+Y904, DLG3-Y673, Dok1-Y449, Gab1-Y406+Y659, Hrs-Y216, PDLIM5-Y251, N-WASP-Y256, PZR-Y263, Shc1-Y427, and ZO2-Y1118; the metabolic enzymes enolases (ENO1-Y44, ENO2-Y44, ENO3-Y44), LDH-A-Y239, G6PD-Y503, pyruvate dehydrogenases (PDHA1-Y301, PDHA2-Y299), and phosphoglycerate mutases (PGAM1-Y92, PGAM2-Y92, PGAM4-Y92); and the regulatory proteins calmodulin-Y100, eEF1A1-Y29+Y141, eEF1A2-Y29+Y141, phosphoinositide-3-kinase regulatory subunits (PIK3R1-Y467, PIK3R2-Y464), PTPRA-Y798, Rin1-Y36, SHP2-Y62+Y584, and STAT’s (STAT3-Y705, STAT5A-Y694, STAT5B-Y699).As pointed out earlier, since MS studies commonly use anti- phosphotyrosine antibodies to enrich phosphotyrosine-containing peptides, the prevalence of phosphotyrosine in these most frequently phosphorylated proteins may be somewhat biased. However, in view of their ubiquitous presence in cells and tissues in high abundance, they are unlikely to be selective biomarkers for diagnostic purposes.

Over 85% of the 21,300 known human proteins have now been reported to be phosphorylated [23]. Consequently, with potentially a million human phosphosites, many proteins will in fact feature more than 50 phosphosites. Most of these phosphosites tend to be clustered in their distribution in proteins, and are most frequently located right next to each other [22]. This has important ramifications with respect to the utility of phosphosite-specific antibodies, since the presence or absence of a phospho moiety in a flanking phosphosite in a protein could interfere with the recognition of the intended phosphosite by the antibody.

In the selection of suitable phosphorylation sites to target for antibody development, priority has been assigned to phosphosites that have been demonstrated to be functionally important in the literature or are likely to be based on high similarity with phosphosites in related, well characterized proteins. Repeated observation of the phosphorylation of a site in numerous MS studies is also desirable, although there are many instances of phosphosites that are well documented to be regulatory but poorly detected by MS. From a query of the PhosphoSite Plus website with 2742 known phosphosites that have been shown to be regulatory in the literature, we found that 602 (22%) have not been reported by random MS studies, and another 434 (16%) had only been documented in one MS experiment. To identify potentially important phosphosites, we have also investigated the evolutionary conservation of phosphorylation sites in diverse species, since these sites are likely to be retained if they play a regulatory role. The PhosphoNET website provides data on the evolutionary conservation of 960,000 known and predicted human phosphosites [24].

Mass spectrometry remains the method of choice by many in the field for identifying novel phosphorylation sites and quantifying phosphorylation events in exploratory studies despite the high price of equipment and consumable costs associated with this technology. Using stable isotope tag labelling, MS has been very successfully used to monitor fluctuations in the levels of phosphorylated proteins with high specificity. Indeed, the sequencing of the human genome was critical for this, since phosphopeptides can be easily identified from the charge to mass ratio of predicted peptides from protease digested proteins [25]. However, the requirements of sample pre-processing, including protease cleavages and phosphopeptide enrichment, also pose as potential sources of data inaccuracy and bias, as evidenced by the absence of many previously characterized phosphorylation sites in the largest datasets and by the low data reproducibility observed in replicate samples [26,27]. The lack of antibody reagents for following up on promising leads from MS analyses further reduces confidence in these findings without independent validation. By contrast, antibody microarrays not only define potentially important biomarkers, but also the antibody reagents that can serve to further track them by immune blotting, immune precipitation, immunohistochemistry, ELISA and other antibody-based methods.

Mass spectrometry has also been a powerful tool for the discovery of other types of covalent modifications of proteins. While phosphorylation is the most prevalent post-translational modification, at the time of the preparation of this review, the PhosphoSitePlus website had already annotated 77,335 ubiquitination sites, 38,139 acetylation sites, 18,614 methylation sites, 8,171succinylation sites and 10,242 glycosylation sites in proteins found in humans, rats and/or mice by MS [23].

Despite the enthusiasm for application of MS approaches to track changes in the post-translational modifications of proteins, there are serious limitations that compromise the use of this technology for biomarker discovery and tracking cell signalling systems in addition to the high expense in conducting these analyses. While MS is very sensitive, extensive enrichment and fractionation are required to analyze low abundance cell signaling proteins from tissue and cell lysates. The need for substantial amounts of starting sample lysate protein in milligram amounts has particularly curtailed broad adoption of this methodology in phosphoproteomics studies where clinical specimens are to be analyzed. For phosphor protein profiling, ideally 2 mg or more lysate protein are required. Phosphoproteins are tracked by serendipity, since only a small portion of the phosphoproteome can be monitored within an MS experiment, and the more abundantly expressed proteins are more likely to be detected after enrichment for phosphor peptides prior to MS analyses. This is why the majority of the most frequently detected phosphor proteins are cytoskeletal proteins, scaffolding proteins, or metabolic pathway enzymes as exemplified above. However, the biggest concern is that this technique is unable to distinguish whether any change in phosphorylation detected is due to an alteration in the total concentration of a particular protein or a change in the stoichiometry of phosphorylation of the protein or some combination of both. Ultimately, the observed change must be validated by use of pan-specific (detects both phosphorylated and non-phosphorylated target proteins) and phosphosite-specific antibodies, if these are even available for the given target.

The sheer complexity of the human proteome within a single tissue or cell type poses a serious problem for proteomics analysis of specimens obtained from most of the tissues and cells found in the human body. In one recent study associated with the Human Proteome Atlas Project, across 56 diverse human cell lines that were tested for mRNA transcripts for 19,628 genes, about 11,490 genes were typically expressed as proteins in most cells, and about half of these were “housekeeping” proteins that were commonly found in all of the cell lines [28]. In another report from the Human Proteome Atlas Project, immunohistochemistry studies with 44 human tissues and organs with 24,028 antibodies corresponding to 16,975 protein-encoding genes in addition to mRNA transcription analyses for 32 of the tissues, similarly revealed expression of approximately half of the protein-coding genes in each analyzed tissue [29]. About 60% of all of the enzymes involved in metabolism were expressed in all of the tissues. The largest number of protein-coding genes were expressed in the brain, testis and liver [29].

Collectively, these studies demonstrate that with alternative splicing and post-translational modifications, more than 100,000 proteoforms may exist within any particular tissue or cell lysate. Under the most optimal conditions, 2D gel electrophoretic techniques such as two-dimensional difference gel electrophoresis (DIGE) [30] can permit the resolution and visualization of under 8,000 protein spots with the most sensitive fluorescent stains. Consequently, tracking changes in protein expression and phosphorylation by 2D gel electrophoresis in earlier proteomics studies was never really up to the challenge and was restricted to primarily monitoring the most abundant of cellular proteins.              

Yet another major complicating factor is the high degree of protein-protein interactions that exist within cells and their dynamic nature. With most antibody microarrays, the samples of lysate proteins are incubated with the slides in their native states with only the presence of mild detergents to reduce non-specific interactions (Figure 2). The usage of stronger detergents and other agents that denature or precipitate proteins interferes with capture antibody-antigen interactions on the microarray slides. Any proteins that are in homodimeric or higher homo-oligomeric states are less problematic, since the signal from the label (e.g. biotin or dye) of the lysate proteins will ideally reflect only the target protein, unless the capture antibody also cross-reacts with off target proteins as well. However, it is also possible that many antibody epitopes could be masked when these proteins reside in homo-multimeric complexes.        

Of greater concern is the number of proteins that are bound to other types of proteins in heteromeric complexes. Some of the larger protein complexes include the 40S and 60S ribosomes, proteasomes, nucleosomes, DNA polymerases, RNA polymerases, the TFIIH transcription complex, splicosomes, the anaphase promoting complex, and the mitochondrial respiratory supercomplex. Several enzymes, including many protein kinases and protein phosphatases, are associated with regulatory subunits, and all of the receptor-tyrosine kinases and receptor-tyrosine phosphatases appear to form signalling complexes with other proteins. Capture antibodies on microarrays that target one of the components of these complexes may indirectly retain other proteins, and these interactions within complexes may be subject to regulation. Since these associated proteins are also tagged, they will contribute to the total signal that is detected within an antibody spot on a microarray. Consequently, an apparent increase in the level of a target protein may actually reflect the increased binding of one or more of its partners in a complex, and would represent a false positive event (Figure 2).

Protein phosphorylation, especially on tyrosine residues is well recognized to drive complex formation between signaling proteins by virtue of phosphotyrosine recognition binding domains such as Src-homology-2 (SH2) and protein-tyrosine binding (PTB) domains [31]. Apparently 119 SH2 domains are found in 109 different human proteins and 54 PTB domains reside in another 54 human proteins [22]. Proteins also feature domains that recognize phosphorylated serine or threonine residues for binding. In particular, the seven human 14-3-3 proteins have been reported to interact with over 200 different proteins in complexes [32], and the actual number is likely to be far greater. The formation of these phosphorylation-driven complexes is likely to render the interacting phosphorylation sites inaccessible to capture antibodies on microarrays and generate false negatives. 

In addition to problems of protein-protein interactions, the location of proteins embedded in cellular membranes and organelles presents yet another challenge to the detection of proteins of interest with antibodies on microarrays. Approximately 30% of the proteins in the human proteome reside in membranes, and many of these are important therapeutic drug targets [33]. However, in one study, only 1 in 20 commercially available antibodies that were tested were determined to be suitable for targeting detergent-extracted membrane proteins in a microarray [34]. Some proteins, such as protein kinase C isoforms and CTP: Phosphocholinecytidylyl Transferase (PCYT1) can exhibit profound changes in their location between cytosolic and microsomal preparations of lysates depending on how these enzymes are stimulated in cells [35,36]. Likewise, many cytoskeletal proteins may be insoluble in their polymerized forms, but soluble in their monomeric forms. The states of polymerization of many cytoskeletal proteins can change profoundly during the cell cycle in proliferating cells. This is one of the reasons why abundant cytoskeletal proteins such as actins and tubulins are not reliable loading controls for Western blotting studies of lysates prepared for cell signaling research. 

Finally, in the case of phosphoproteomics studies, the post-homogenisation instability of phosphorylation sites due to endogenous phosphatases in lysates of cells and tissues is a major problem that is poorly appreciated in the field. We recently demonstrated that most of the phosphorylations of the EGF receptor and other downstream signaling proteins were lost when human A431 epidermoid carcinoma cells, which over-expressed this receptor-tyrosine kinase, were homogenised in buffers that contained a wide range of standard protease and phosphatase inhibitors that are routinely used in most cell signaling studies[37]. We found that these phosphorylation events were preserved if the cells were homogenised directly into SDS-PAGE sample buffer, which contains 1% Sodium Dodecylsulphate (SDS). However, such a concentration of SDS almost completely abolishes antibody-antigen interactions on antibody microarrays. Based on our experiments, we believe that most phosphoproteomics studies performed in the pastin other laboratories and ours have substantially under-estimated in vivo protein phosphorylation. One exception appears to be in MS studies where cells and tissues were directly homogenised in strong chaotropic agents such as urea and guanidine hydrochloride, and then subjected to trypsin treatment to generate phosphopeptides that were later enriched, resolved and analyzed by MS. These harsh treatments provided better recovery of phosphopeptides. 

4.         Biomarker Discovery               

One of the most promising applications of antibody microarrays is for the discovery of biomarkers for disease diagnosis. Historically, most of these efforts have been focused on the early detection of cancer, especially in the pancreas, ovaries, and lungs, where prognosis is very poor by the time the cancers are more symptomatic. Table 2 provides a summary of many of the reported studies where antibody microarrays were employed to search for biomarkers of cancer and a few other diseases. Many of these reports utilized bio fluids such as serum, Cerebral Spinal Fluid (CSF) and in one study exhaled breath [54].

Due to its large volume, high protein concentration (typically 60-80 mg/ml) and easy access with relatively little discomfort to patients, blood serum has been the bio fluid of choice for most studies in the search for disease biomarkers. Serum (after removal of fibrin from plasma) can serve as a systemic mirror of more distantly localized tumours that release proteins or protein-containing exosomes [86]. Plasma may contain as many as 40,000 different proteoforms generated from a thousand distinct gene products, but their individual concentrations can range over a trillion-fold.  Only a few proteins actually account for the bulk of the total protein mass in plasma. About 55% of the total protein concentration of human blood plasma comes from albumins, another 38% from globulins such as antibodies, and 7% from fibrinogen. Consequently, these abundant proteins are typically removed to enrich the remaining proteins prior to most proteomics analyses [87]. This can be achieved by using commercial kits [9,72] orcibacron blue chromatography, which has a high affinity for albumin [63]. Significantly, in one antibody microarray study [9] that used the ProteoPrep 20 Plasma Immuno depletion kit from Sigma-Aldrich, which removes 20 of the highly abundant serum proteins with antibodies immobilised on agarose beads, the levels of more than 196 proteins (representing 26% of the detected proteins) were statistically affected, with most of these showing reductions rather than enrichments. While the removal of the most abundant serum proteins may be necessary for MS analyses, unfractionated serum can be directly applied to antibody microarrays due to the insensitivity of most antibodies to these serum proteins. Other human biofluids with proteins such as urine (normally 0-2 mg/ml) [11], CSF (normally 0.2-0.6 mg/ml) [82] and prostatic fluid (~50 mg/ml) [78] have also been successfully interrogated with antibody microarrays. 

One of the most commonly used biomarkers for prostate cancer is the prostate specific antigen (PSA), which is the androgen-related serine protease Kallikrein-related peptidase 3 (KLK3). PSA is mainly expressed in the prostate epithelium and is found in seminal fluid at about 1 mg/ml, but its concentration in serum is normally more than 500,000-times lower at 2ng/ml or less.It is likely that other disease biomarkers also have to be detectable at concentrations in serum that are in the order of 100,000- to a million-fold lower levels than found in the organs from which they originate. Therefore, it may be more instructive to initially identify biomarker proteins in diseased organs and tissues, and then subsequently search for the most promising lead biomarkers in serum specimens. Table 3 identifies several studies with antibody microarrays that utilized tumours, normal tissue,and diverse cancer cell lines to discern potential cancer biomarkers. It is evident from inspection of this list, that a few of the studies were contradictory with respect to some of the observed changes in biomarkers in the same types of cancer (e.g. serum C3, C4, Cathespin D, MCP-1 and VEGF in pancreatic cancer, and serum von Wille brand factor in prostate cancer). However, often the same proteins were similarly affected in diverse cancers. It should be appreciated that no single biomarker, including PSA, is reliable by itself for disease diagnosis. However, monitoring a panel of validated biomarkers can provide diagnosis with high accuracy, which further supports the need for platforms like antibody microarrays that are able to monitor multiple analytes of interest.

Unfortunately, many promising disease biomarkers from proteomics studies ultimately do not pan out in a clinical context, because there are often large differences in individual protein levels even within the healthy, normal population. In one carefully controlled study performed with groups of ten littermate mice of the same gender, we observed fluctuations that ranged in variation by 2-fold in the levels of potential biomarker lung proteins for cigarette smoke particulates exposure amongst the individual animals in the same untreated groups [89]. Pooling lysate specimens from within the same groups and testing the pooled specimens with antibody microarrays is one strategy to find especially robust changes to define biomarker proteins. This can also help to reduce the costs of such analyses. However, for biopsies from lung tumours obtained from the same organ, we have commonly observed profound variations in protein expression and phosphorylation amongst individual patients as determined by both antibody microarrays and by Western blotting. 

Another confounding factor in biomarker discovery is the occurrence of what has been referred to as repeatedly identified differentially expressed proteins (RIDEPs), which were originally flagged from primarily 2D gel and MS studies [90-92]. These RIDEPs in human specimens most frequently included enolase 1 and heat shock protein 27 (Hsp27), whilst members of the keratins, annexins and peroxiredoxins were also commonly altered in expression in different experimental model systems. In antibody microarray studies [93], Hsp27, enolase 1, and zyxin were the most commonly observed RIDEPs, and 14-3-3 proteins, BclxL, BID, MAGI3, MyD88, protein kinase C isoforms, SLIPR, Siah2, and Smad4 were also affected in several diverse studies. Most of these proteins are involved in cellular stress responses, and many not be very distinguishing biomarkers for specific pathologies.        

The search for serum biomarkers, particularly in the context of cancer, has uncovered a wide range of interesting biomarkers as evident in Table 3. However, the identified proteins have overall proved to have relatively little diagnostic value, because most of the elevated proteins in these studies appear to be derived from activated immune cells or are produced by many different cell types. The same proteins may be produced in response to cancer, autoimmune diseases, and infection with viruses and bacteria. To find more discriminating biomarkers, it may be more sensible to identify proteins that are very tissue-specific in their expression, and then assess how their levels are altered under pathological circumstances. Antibody microarrays can be utilised to rapidly screen for antibodies that uncover such tumour- or tissue-specific proteins, which can then be used to evaluate their reliability as biomarkers in larger trials with more patients.        

Most studies with antibody microarrays in the past have focused on tracking changes in protein levels in biofluids, and lysates from tissues and cells. However, the altered phosphorylation status of proteins may provide greater insights as these can serve as indicators of the activity states of these proteins. A few commercial antibody microarrays have been developed to track the phosphorylation status of proteins (Table 2). One of the advantages of commercial antibody microarrays is that with increasing reports of the results of the applications of these microarrays in different model systems, it becomes easier upon comparison of data sets to identify promising biomarkers that are more restricted in the circumstances under which they reveal changes in expression or phosphorylation. Towards this end, Kinexus created the KiNETDataBank (www.kinet-am.ca), which is an open-access website that contains the data from over 2000 different studies performed with Kinex™ KAM antibody microarrays. 

5.    Improving Antibody Microarrays 

For antibody microarrays to become better accepted as reliable tools for biomarker identification and the study of cell signaling systems, there are several strategies that can be adopted. These include improving the capture antibodies and microarray slide substrates, the preparation of lysate samples, the detection of captured proteins on microarrays, and the validation of the antibody leads from microarrays. 

One of the strategies that we have developed to reduce false positive results on antibody microarrays has been to chemically cleave lysate proteins at cysteine residues by successive treatment with Tris(2-Carboxyethyl) Phosphine Hydrochloride (TCEP) and 2-Nitro-5-Thiocyanatobenzoic Acid (NTCB) [37,94] (Figure 3). Many of the epitope sequences used to generate the antibodies used on antibody microarrays avoid the inclusion of cysteine residues (since this is usually used only at the N- or C-termini of peptides for coupling them to keyhole limpet hemocyanin for immunisation and to thio-agarose affinity resins for antibody purification), and so this affords the opportunity to generate peptide fragments that retained the epitopes for these antibodies. While the length of peptides produced from Cysteine Chemical Cleavage (CCC) of the human proteome are predicted to vary markedly, most will feature about 30-40 amino acid residues. These are larger peptides than would be typically generated by treatment with a protease like trypsin, and since many protein phosphorylation sites contain flanking arginine and lysine residues, trypsin digestion would destroy many of the epitopes that phosphosite-specific antibodies will recognize.

The CCC procedure has several advantages over commonly adopted protocols, including providing for dissociation of the antibody target protein from associated proteins and facilitating exposure of previously masked epitopes. Epitope-containing peptides tend to be in the same size ranges, so following CCC, the amount of signal recorded for an antibody spot is more reflective of the molar concentration of the target protein and less influenced by its size. The average number of amino acids in the 21,300 native full-length proteins found in the human proteome is about 700 residues, but the sizes of different proteins can vary by several orders of magnitude. The largest human protein is the kinase titin, which features 34,350 amino acid residues.

        ysate protein samples that are subject to CCC are stable at room temperature for over a week. The CCC protocol leads to the destruction of the activities of kinases, phosphatases, proteases and other enzymes that can lead to changes post-homogenisation such as alterations in protein phosphorylation. If the CCC       procedure is undertaken right at the time of homogenisation of tissues and cells, this best preserves the state of protein phosphorylation. In a recent study in which we used the Kinex^™ KAM-900 antibody microarray to explore changes in phosphorylation in EGF treated human A431 cervical carcinoma cells, we were       able to increase the number of detected lead changes by over 40-fold with the chemical cleavage step performed during lysate preparation as compared to implementing it afterwards [37].

    Western blotting has traditionally been the main validation approach to confirm the lead findings from antibody microarray studies. However, antibody microarrays can be several magnitudes more sensitive for detection of target proteins than immunoblots [95] or reverse lysate microarrays. In one antibody     microarray study, it was found that 7000-times more cell lysate was required to visualise the lead proteins by Western blotting [18]. With reverse microarrays, the lysate proteins from different tissues or cells are printed as individual spots on slides, and each slide is probed with a different detection antibody. Less     than 1 ng of total lysate protein is usually printed as a spot on a reverse microarray, and consequently the sensitivity of antibody detection remains a significant issue. With Western blots, typically 25 µg of lysate protein are deposited in a single lane on an SDS-PAGE gel, but the signal is spread out over the width of     the lane. Despite this, Western blotting has successfully confirmed the leads from many different antibody microarray studies [40,63,67,73,95-99].

    In our experience with using Western blotting to confirm leads from antibody microarray experiments, we have found that about 40% of the time, the signals from immune blots performed with the same antibodies used on the microarrays are just too weak for detection. These same antibodies had been previously     shown to successfully detect their target proteins in lysates from other cells and tissues, but not in the particular experimental model system under investigation with an antibody microarray. To overcome this issue, especially for validating changes in protein expression, we have tried to incorporate into our     microarrays more redundant capture antibodies that target different epitopes in the same protein. When a protein change is observed with multiple antibodies for the same target, this internal cross-validation improves confidence that the change is likely to arise from the intended protein and not an antibody     cross-    reactive off-target protein. This strategy is more difficult to implement for the analysis of specific phosphosites. In such cases, it may be necessary to enrich the target phosphor protein first. This might be achieved by subcellular fractionation, or by immune precipitation with a pan-specific antibody for the     target protein prior to Western blotting with the phosphosite-specific antibody.

    Often, from the Western blotting validation exercise, the change observed with an antibody microarray can arise from a cross-reactive protein. In this event, since it is retained by the microarray and detectable on an immunoblot with a capture antibody, it can probably be immune precipitated. The immune     precipitated protein can be subjected to SDS-polyacrylamide gel electrophoresis, and the excised protein from the gel can subsequently be identified by MS. With the availability of an antibody for the identified protein, it is feasible to rapidly follow up with other experiments with larger numbers of specimens to     establish how reliable and robust the biomarker may be.

    Due to limitations of the amounts of antibodies that are spotted on the antibody microarrays, and the low levels of expression of most cell signaling proteins, the signals generated for many target proteins can be very weak despite the high sensitivity of antibody microarrays. When antibody microarrays were first d    eveloped, it was common to directly tag lysate proteins with fluorescent dyes such as Cy3 and Cy5, often in a competitive two dye format. With this approach, the lysate proteins in a “control” sample might be labelled with Cy3, whereas the lysate proteins in a “treatment” sample would be separately labelled with C    y5. Subsequently, the samples are mixed and incubated on the same field of antibody spots. With equal amounts of a target protein in the control and treatment lysates, effectively half of the antibodies on a spot would bind the Cy3-labelled protein and the other half of the antibodies in the same spot would bind t    he Cy5-labelled protein. The presence of the two populations of labelled target proteins are monitored separately at the two wavelengths that are optimal for Cy3 and Cy5, but the amplitude of the signals will be reduced in half of what would be achieved with a single dye, non-competitive method. This two dyes, c    ompetitive approach reduces the costs of such microarray analyses, but extreme caution needs to be exercised as different proteins do not bind to different dyes uniformly [100,101]. Often, the experiment is repeated by switching the dyes with the two lysate samples undergoing testing, and checking that the same l    eads for altered expression or phosphorylation between the lysate samples are observed. Where this has been done, it is claimed that most of the leads are still reproducible [45,83,102,103].

    To improve the signal to background noise ratio, we prefer using a non-competitive, single dye approach in which a blend of dyes that fluoresce maximally at the same wavelength are used to label all of the different lysate protein samples under investigation. The use of blended dyes increases the opportunity for     more uniform labelling of the lysate proteins in view of the differential binding of dyes by proteins. Each sample is tested with a separate field of spotted antibodies, typically with two separate fields per microarray slide, one field for the “control” sample and the other field for the “treatment” sample. Usually replicate     printing of each antibody spot is performed in each field to check reproducibility of signals, and the replicate spots are not printed side-by-side to reduce the chances of problems with both replicates due to misprinting, spot mixing, dust, and other background issues.

    One of the major issues is the removal of free fluorescent dye remaining even after gel filtration or dialysis of the lysate protein following its labelling step. This free dye can be partly neutralized by the inclusion of ethanolamine or hydroxyurea during the incubation of the dye-labelled lysate samples with the     antibody microarray slide. However, there is often still some direct labelling of capture antibodies with remaining free dye that can increase background signals.

    To improve sensitivity of detection with antibody microarrays and reduce background noise, initial labelling of lysate protein with biotin instead of fluorescent dyes is an alternative strategy that we and others have adopted [62,73,82,104-106] (Figure 3). The biotinylated lysate proteins captured on the antibody     microarray are typically detected with streptavidin that is conjugated to a fluorescent dye. However, avidin exhibits non-specific interactions directly with lysate proteins on the microarray, although NeutrAvidin is a diglycosylated form of avidin that purportedly possesses higher binding specificity and reduced     non-   specific binding [107]. We prefer using dye-conjugated anti-biotin antibody for detection of biotinylated lysate proteins, because this yields much less non-specific binding than dye-labelled avidin, and since IgG is three times the size of avidin, it binds three times more dye to yield stronger signals [37].

    To further improve the detection of captured biotinylated lysate proteins on antibody microarrays, it is possible to use Rolling-Circle Amplification (RCA) for signal generation. In this strategy, the 5’end of an oligonucleotide primer is attached to an anti-biotin antibody. After binding of the oligonucleotide conjugated     anti-biotin antibody to the captured biotinylated lysate proteins on an antibody microarray, the oligonucleotide is enzymatically extended using a circular DNA sequence as template. Complementary, fluorescently labelled short oligonucleotides are then hybridised to the extended DNA to decorate each of the bound     antibodies with thousands of fluorophores. Using RCA assay, it should be feasible to detect captured proteins in the sub-pg/ml range.

6.    Expansion of Antibody Microarray Applications 

While antibody microarrays have long been used to track the relative differential levels of target proteins and their phosphorylation at specific sites, antibody microarrays can also be adapted to monitor changes in different types of post-translational modifications, protein-protein, and drug-protein interactions. It can additionally be used to monitor changes in protein kinase activities. Figure 4 illustrates some of the ways that different post-translational modifications of proteins can be detected.

    With the availability of generic phosphotyrosine-site antibodies, these have been used as affinity reagents to enrich phosphotyrosine containing peptides prior to MS analyses. However, they can also be used as reporter antibodies to detect differential phosphorylation of lysate proteins on antibody microarrays. This     procedure was initially testedfor detection of protein-tyrosine phosphorylation with 35 capture antibodies on a microarray incubated with lysate proteins from Bcr-Abl-expressing RT10+ cells and EGF treated HeLa cells probed with Cy5-labelled custom anti-phosphotyrosine monoclonal antibody [108]. Reductions of     at least 45% in the tyrosine phosphorylation signals for 23 of the antibody targets were recorded when the RT10+ cells were pretreated for 2 hours with the protein-tyrosine kinase inhibitor Gleevec, and increases of 2-fold in tyrosine phosphorylation signals were recorded for 16 antibodies following1 hour of     exposure of HeLa cells to EGF after they had been serum-starved previously for 24 hours [108].We have also detected increased tyrosine phosphorylation of lysate proteins in A431 cells exposed to EGF for 5 minutes using a similar approach with the Kinex^™ KAM-1150 antibody microarray and our generic PYK     polyclonal anti-phosphotyrosine antibody used as a reporter antibody (L. Yue and S. Pelech, manuscript in preparation). Figure 5 shows an image of identical grids of antibodies spots with captured lysate proteins from untreated and EGF-treated A431 cells probed with fluorescent dye-labelled PYK antibody. We were     able to further increase the sensitivity for tyrosine phosphorylation by using biotinylated PYK antibody, and detecting the presence of this reporter antibody with fluorescent dye-conjugated anti-biotin antibody. 

Other types of covalent modification of proteins can also be monitored on antibody microarrays using the appropriate reporter antibodies. For example, a custom antibody microarray for 24 target proteins (primarily growth factors and cytokines) was used to investigate their tyrosine nitrosylation in plasma samples from cigarette smokers, those with chronic obstructive pulmonary disease, and non-smokers [109]. In this study, a generic anti-3-nitrotyrosine antibody was used as a biotin-labelled reporter antibody, and this was subsequently detected with fluorescent-dye tagged streptavidin. We have recently begun to monitor changes in the ubiquitination of proteins using the Kinex™ KAM-1150 antibody microarray using dye-labelled anti-ubiquitin antibody as a reporter antibody (Figure 6).

Many proteins in cells are found in transient complexes with adapter and scaffolding proteins and chaperonins such as heat shock proteins. Reporter antibodies that are specific for these proteins could identify binding partner proteins captured on antibody microarrays and reveal changes in their associations as a consequence of perturbations of experimental model systems (Figure 7A). While this methodology for exploring protein-protein interactions on microarrays has not yet really been exploited, it should prove to be very fruitful.        

Finally, antibody microarrays can be adapted to monitor the activity states of proteins, especially as more biosensor molecules become available that selectively detect the inhibited or activated forms of enzymes. One of the most promising opportunities in this direction is for tracking the activation states of protein kinases. Assaying the enzymatic activity of protein kinases in crude cell and tissue lysates is a formidable task without some sort of affinity purification. At least 536 different human protein kinases are documented on our KinaseNET (www.kinasenet.ca) website, and they have highly overlapping substrate specificities. Approximately 484 typical protein kinases of the human kinome feature a common catalytic domain that includes a highly conserved lysine residue in the kinase subdomain II region in the “AXK” motif [110], which is located in the active site of these kinases and is critical for phosphotransferase activity by transferring the gamma-phosphate of ATP on to the hydroxyl group of phosphorylatable residues on protein substrates. Affinity labelling with the ATP analogue 5’-(p-flurosulphonylbenzoyl) adenosine (FSBA) inhibits protein kinases by covalently modifying the kinase subdomain II lysine residue [111]. The active site of protein kinases need to be open and accessible for FSBA to bind, so it can be a useful probe to determine whether kinases are in a state where they may be catalytically active. We have successfully used a biotinylated version of FSBA to interrogate the activation states of protein kinases as illustrated in Figure 7B.              

By using preparations of cell lysates in which the endogenous protein kinases have been preactivated in vivo, and able to bind FSBA, it is also feasible to identify subsets of protein kinases that may be sensitive to inhibition by test compounds. As documented in our DrugKiNET website (www.drugkinet.ca), most of the known protein kinase inhibitors bind directly to the active sites of protein kinases, and are partly analogues of ATP. Preincubation of captured protein kinases on an antibody microarray with a kinase inhibitor that occupies the active sites of sensitive kinases will pre-empt the binding of the biotinylated FSBA (Figure 7C). The reduction of biotinylated FSBA binding to the array can be monitored with dye-labelled anti-biotin antibody. Such approaches could be very useful for counter screening promising kinase inhibitor leads with hundreds of kinases for off target effects. It is also possible that the biotinylated FSBA probe may be helpful for identifying possible peptide and protein substrates of kinases, since the binding of these substrates may also interfere with the interaction with the biotinylated FSBA.

For all of these types of investigations, full length native lysate proteins would be desirable, so the cysteine chemical cleavage process would not be used. Therefore, caution will be necessary in interpretation of the results since the proteins will likely reside in complexes, and some types of covalent modifications may be prone to rapid removal post-homogenisation.

7.        Conclusion

In this review, we have described many applications of antibody microarray technology in biomedical research while emphasizing its utility and unrealized potential as a quick, sensitive and cost-effective platform that is compatible with various types of biological samples for flexible analysis of different proteomics aspects. Antibody microarrays are burdened by the challenge of availability of high-quality antibodies suitable for the platform, and are also subject to many of the same problems that confound other proteomics assays, such as the nature of the sheer complexity of proteomes, the high degree of protein-protein interactions, and over a million-fold differentially expressed classes of proteins. Nevertheless, antibody microarray technology is superior to mass spectrometry due to the high sensitivity of antibodies as biosensors, affording more quantitative analyses with low sample amount requirements without the need for enrichment of low-abundance proteins. In this review, we have suggested how analyses can be further improved by strategies in sample preparation, detection methods, and also study designs. Ultimately, any findings with the antibody microarrays should be confirmed by alternative methods. However, the high sensitivity, breadth of coverage, economy and speed of antibody microarray analyses will ultimately advance this technology to eventually prevail as the preferred methodology for proteomics investigations into the foreseeable future.

8.       Conflict of Interest Declaration

S.P. and his family are the majority shareholders of Kinexus Bioinformatics Corporation.

Figure 1: Time course of annual rate for antibody microarray publications. The search term was “antibody microarray” and redundant entries from Google Scholar were removed.

Figure 2: Possible issues with antibody microarrays. True positives occur when the target protein that is labelled with a fluorescent dye or another tag specifically binds to the capture antibody in a monomeric form (Panel A), although the target protein can also be bound in a homodimeric or homopolymeric forms. False positives can be generated from signals being generated from associated proteins (Panel B), and cross-reactive off target proteins with the Fab (Panel C) and Fc (Panel D) regions of the capture antibodies. False negative signals can be generated by the binding epitope on the target protein being blocked by interaction with other proteins (Panel E).

Figure 3: Chemical cleavage of proteins and their detection on antibody microarrays. In most published studies, proteins are typically directly dye-labelled in their native form as shown in Panel A. Chemical cleavage at cysteine residues fragments proteins and only the cleaved portion that retains the epitope is bound to the capture antibody on the microarray. The protein fragment can be directly labelled with fluorescent dye (Panel B) or biotinylated (Panel C). Captured biotinylated proteins on the microarray can be detected with fluorescent dye-labelled avidin or anti-biotin antibody (Panel C).  

Figure 4: Monitoring post-translational modifications of target proteins on antibody microarrays. Following capture of proteins by primary antibodies printed onto antibody microarrays, in a sandwich format, fluorescent dye conjugated reported antibodies that target phosphotyrosine (Panel A) and ubiquitin (Panel C) can be used to track changes in the tyrosine phosphorylation and ubiquitination of proteins, respectively. Biotinylation of the reporter antibodies can be used in combination with dye-labelled anti-biotin antibodies to further amplify signal detection (Panels B and D).

Figure 5: Kinex^™ KAM-1150 antibody microarray analyses of the expression and tyrosine phosphorylation of signalling proteins in overnight serum-starved human A431 cervical carcinoma cells treated without (Panels A, C, F, H, K, and M) and with 100 ng/ml epidermal growth factor (EGF) for 5 minutes (Panels B, D, G, I, L and N). Cell lysates were prepared with 1.0% Triton X-100 and buffers containing protein phosphatase inhibitors and protease inhibitors and subjected to antibody microarray analyses as described [37]. Approximately 100 µg of each A431 cell lysate were subjected to cysteine chemical cleavage, dye-labelled then and incubated with one of two fields of antibodies printed on each KAM-1150 chip (Panels A, B, F, G, K and L). Other 100 µg aliquots of the same lysates were not chemically cleaved nor tagged, but following capture on one of the grids of a KAM-1150 chip, were probed with dye-conjugated PYK (Cat. No. PG001, Kinexus) generic anti-phosphotyrosine antibody (Panels C, D, H, I, M and N).  Only 3 representative grids of the 16 distinct grids of antibodies printed in a single field are shown as scanned images in this figure; each field was replicated four times per array. The strongest signals appear red and the weakest signals are blue. As a control, one field was incubated with only the dye-labelled anti-phosphotyrosine polyclonal rabbit antibody in the absence of cell lysates, and the immunoreactive antibodies are circled (Panels E, J and D). There were little if any changes apparent in protein expression in A431 cells with short term exposure to EGF (Panels A, B, F, G, K and L), but the treatment increased the phosphorylation signals for several of the captured proteins (Panels C, D, H, I, M and N).

Figure 6: Kinex^™ KAM-1150 antibody microarray analyses of the ubiquitination of signaling proteins from human HeLa cervical carcinoma cells treated without (Panels A, Cand E) and with 50µM of theproteasome inhibitor MG-132 for 4 hours (Panels B, Dand F). Cell lysates with 1.0% Triton-X100 were prepared and100 µg were incubated with the KAM-1150 chip, which was subsequently probed with anti-ubiquitin antibody (Cat. No. SC-8017, Santa Cruz Biotechnology) that was dye-conjugated. Inhibition of protein degradation with the MG-132 resulted in increased detection of several proteins on the KAM-1150 antibody microarray as shown in 3 representative grids of distinct 16 grids of antibody spots, which were each replicated in four fields on the array.

Figure 7: Monitoring protein-protein and protein kinase-drug interactions on antibody microarrays. A sandwich antibody microarray format, can be used to detect interactions between two different proteins depicted as A and B (Panel A). The active site of a captured protein kinase on an antibody microarray can become covalently bound to a biotinylated version of the ATP analogue FSBA if the kinase is in an open, active conformation (Panel B). Preincubation of active, captured protein kinases on antibody microarrays with inhibitory drugs can sterically block subsequent access of the FBSA-biotin probe, and through a reduction of signal from dye-conjugated anti-biotin antibody binding observed in the absence of the inhibitor reveal those kinases that are sensitive to the drug (Panel C).

Table 1: Commercial high content antibody microarrays. “Pan-Ab” refers to antibodies that bind to target proteins independently of their phosphorylation states, whereas “Phospho-Ab” refers to antibodies that are phosphorylation site-specific. In most cases, each antibody was printed with two replicate antibody spots per array, except for the Kinex^™ KAM antibody microarrays, which feature four replicates of each antibody.

Table 2: Examples of reported studies in the literature to identify biomarkers of disease using antibody microarrays and human clinical specimens.

Table 3: Common protein biomarkers of human disease identified using antibody microarrays and clinical specimens. “Change” refers to how the level of the biomarker was affected with the disease relative to biofluid specimens from healthy individuals or from patient-matched control tissues.

1.       Nielsen UB, Geierstanger BH (2004) Multiplexed sandwich assays in microarray format. J Immunol Methods 290: 107-120.

2.       Usui-Aoki K, Shimada K, Koga H (2007) A novel antibody microarray format using non-covalent antibody immobilization with chemiluminescent detection. Molecular Biosystems 3: 36-42.

3.       Olle E, Deogracias M, Messamore J, McClintock S, Barron A, et al. (2007) Screening of serum samples from Wegener's granulomatosis patients using antibody microarrays. Proteomics Clin Appl 1: 1212-1220.

4.       Tessler LA, Mitra RD (2011) Sensitive single‐molecule protein quantification and protein complex detection in a microarray format. Proteomics 11: 4731-4735.

5.       Chang TW (1983) Binding of cells to matrixes of distinct antibodies coated on solid surface. J. Immunol. Methods 65: 217-223.

6.       Ekins RP (1989) Multi-analyte immunoassay. J. Pharm. Biomed. Anal 7: 155-168.

7.       Ekins RP, Chu FW (1991) Multianalyte microspot immunoassay-microanalytical "compact disk" of the future. Clin Chem 37: 1955-1967.

8.       Ekins RP (1998) Ligand assays: from electrophoresis to miniaturized microarrays. Clin Chemistry 44: 2015-2030.

9.       Schröder C, Jacob A, Tonack S, Radon TP, Sill M, et al. (2010) Dual-color proteomic Profiling of complex samples with a microarray of 810 cancer-related antibodies.  Mol Cell Proteomics 9: 1271-1280.

10.    Baker M (2015) Reproducibility crisis: Blame it on the antibodies. Nature 521: 274-276.

11.    Weller MG (2016) Quality issues of research antibodies. Anal Chem Insights 11: 21-27.

12.    Brumbaugh K, Johnson W, Liao W-C, Lin M-S, Houchins JP, et al. (2017) Overview of the generation, validation, and application of phosphosite-specific antibodies. In: Kalyuzhny A. (eds) Signal Transduction Immunohistochemistry Pg No: 3-43. Methods in Molecular Biology: 717. Humana Press, New York, NY.

13.    Borrebaeck CAK, Wingren C (2007) High-throughput proteomics using antibody microarrays: an update. Exp Rev Mol Diagn. 7: 673-686.

14.    Borrebaeck CAK, Wingren C (2009) Design of high-density antibody microarrays for disease proteomics: Key technological issues. J Proteomics 72: 928-935.

15.    Lian W, Wu D, Lim DV, Jin S (2010) Sensitive detection of multiplex toxins using antibody microarray. Anal Biochem 401: 271-279.

16.    Winter G, Griffiths AD, Hawkins RE, Hoogenboom HR (1994) Making antibodies by phage display technology. Annu Rev Immunol. 12: 433-455.

17.    Böhm D, Keller K, Boehm N, Lebrecht A, Schmidt M, et al. (2011) Antibody microarray analysis of the serum proteome in primary breast cancer patients. Cancer Biology & Therapy 12: 772-779.

18.    Dexlin-Mellby L, Sandström A, Antberg L, Gunnarsson J, Hansson SR, et al. (2011) Design of recombinant antibody microarrays for membrane protein profiling of cell lysates and tissue extracts. Proteomics 11: 1550-1554.

19.    Pan Q, Shai O, Lee LJ, Frey BJ, Blencowe, BJ (2008) Deep surveying of alternative splicing complexity in the human transcriptome by high-throughput sequencing. Nature Genetics 40: 1413-1415. 

20.    Aebersold R, Agar JN, Amster IJ, Baker MS, Bertozzi CR, et al. (2018) How many human proteoforms are there? Nat. Chem. Biol14: 206-214.

21.    Hunter T, Sefton BM (1980) Transforming gene product of Rous sarcoma virus phosphorylates tyrosine. Proc. Natl. Acad. Sci. USA 77: 1311-1315.

22.    SafaeiMehranpour J (2015) Modelling human cell protein phosphorylation networks. Ph.D. thesis. University of British Columbia.

23.    Hornbeck PV, Zhang B, Murray B, Kornhause JM, Latham V, et al. (2015) PhosphoSitePlus, 2014: mutations, PTMs and recalibrations. Nucleic Acids Res 43: 512-520.

24.    Safaei J, Manuch J, Gupta A, Stacho L, Pelech S (2011) Evolutionary conservation of human phosphorylation sites. Bioinformatics and Biomedicine (BIBM), 2011 IEEE International Conference on Nov 12-15: 222-227.

25.    Bodenmiller B, Aebersold R (2010) Quantitative analysis of protein phosphorylation on a system-wide scale by mass spectrometry-based proteomics. Methods Enzymol 470: 3173-3134.

26.    Oyama M, Kozuka-Hata H, Tasaki S, Semba K, Hattori S, et al. (2009) Temporal perturbation of tyrosine phosphoproteome dynamics reveals the system-wide regulatory networks. Mol Cell Proteomics 8: 226-231.

27.    Olsen JV, Vermeulen M, Santamaria A, Kumar C, Miller ML, et al. (2010) Quantitative phosphoproteomics reveals widespread full phosphorylation site occupancy during mitosis. Sci Signal 3: 3.

28.    Thul PJ, Åkesson L, Wiking M, Mahdessian D, Geladaki A, et al. (2017) A subcellular map of the human proteome. Science 356: 3321.

29.    Uhlén M, Fagerberg L, Hallström BM, Lindskog C, Oksvold P, et al. (2015) Proteomics. Tissue-based map of the human proteome. Science 347: 1260419.

30.    Diez R, Herbstreith M, Osorio C, Alzate O (2010) 2-D fluorescence difference gel electrophoresis (DIGE) in neuroproteomics. In Alzate, O., editor. Source Neuroproteomics. Boca Raton (FL): CRC Press/Taylor & Francis; Chapter 4. Frontiers in Neuroscience.

31.    Wagner MJ, Stacey MM, Liu BA, Pawson T (2013) Molecular mechanisms of SH2- and PTB-domain-containing proteins in receptor tyrosine kinase signaling. Cold Spring HarbPerspect Biol 5: 008987.

32.    Mhawech P (2005) 14-3-3 proteins--an update. Cell Res 15: 228-236.

33.    Tan S, Ta, HT, Chung MC (2008) Membrane proteins and membrane proteomics. Proteomics 8: 3924-3932.

34.    Nielsen UB, Cardone MH, Sinskey AJ, MacBeath G, Sorger PK (2003) Profiling receptor tyrosine kinase activation by using Ab microarrays. Proc Natl Acad Sci USA 100: 9330-9335.

35.    Pelech SL, Cook HW, Paddon HB, Vance DE (1984) Membrane-bound CTP: phosphocholinecytidylyl transferase regulates the rate of phosphatidyl choline synthesis in HeLa cells treated with unsaturated fatty acids.  Biochim Biophys Acta 795: 433-440.

36.    Hasham MI, Pelech SL, Koide H, Krieger C (1997) Activation of protein kinase C and calcium/calmodulin-dependent protein kinase II by intracellular free calcium in the motoneuron cell line NSC-19. Biochim Biophys Acta 1360: 177-191.

37.    Yue L, Sam C, Arora N, Winkler FHD, Pelech S (2017) Antibody microarray and immunoblotting analyses of the EGF signaling phosphorylation network in human A431 epidermoid carcinoma cells Clin. Prot Bioinformatics 2: 1-10.

38.    Sanchez-Carbayo M, Socci ND, Lozano JJ, Haab BB, Cordon-Cardo C (2006) Profiling bladder cancer using targeted antibody arrays. Am J Pathol 168: 93-103.

39.    Srinivasan H, Allory Y, Sill M, Vordos D, Alhamdani MSS, et al. (2014) Prediction of recurrence of non-muscle‐invasive bladder cancer by means of a protein signature identified by antibody microarray analyses. Proteomics 14: 1333-1342.

40.    Lee M, Kim J, Lee J, Yun S, Kim W, et al. (2015) Prognostic significance of CREB-binding protein and CD81 expression in primary high grade non-muscle invasive bladder cancer: Identification of novel biomarkers for bladder cancer using antibody microarray. Plos One 10: 0125405.

41.    Carlsson A, Wingren C, Ingvarsson J, Ellmark P, Baldertorp B, et al. (2008) Serum proteome profiling of metastatic breast cancer using recombinant antibody microarrays. Eur J Cancer 44: 472-480.

42.    Li CI, Miru JE, Zhang Y, Ramirez AB, Ladd JJ, et al. (2012) Discovery and preliminary confirmation of novel early detection biomarkers for triple-negative breast cancer using preclinical plasma samples from the Women’s Health Initiative observational study. Breast Cancer Res Treat 135: 611-618.

43.    Song XC, Fu G, Yang X, Jiang Z, Wang Y, et al. (2008) Protein expression profiling of breast cancer cells by dissociable antibody microarray (DAMA) staining. Mol Cell Prot 7: 163-169.

44.    Fares A, Miranda M, Wei S, Peter T, Simpson PT, et al. (2014) Kinome profiling reveals breast cancer heterogeneity and identifies targeted therapeutic opportunities for triple negative breast cancer. Oncotarget 5: 3145-3158.

45.    Kit OI, Vodolazhsky DI, Kutilin DS, Enin YS, Gevorkyan YA, et al. (2017) A proteomics analysis reveals 9 up-regulated proteins associated with altered cell signaling in colon cancer patients. The Protein Journal 36: 513-522.

46.    Sreekumar A, Nyati MK, Varambally S, Barrette TR, Ghosh D, et al. (2001) Profiling of cancer cells using protein microarrays: discovery of novel radiation-regulated proteins. Cancer Res 61: 7585-7593.

47.    Hao Y, Yu Y, Wang L, Yan M, Ji J, et al. (2008) IPO-38 is identified as a novel serum biomarker of gastric cancer based on clinical proteomics technology. J Proteome Res 7: 3668-3677.

48.    Cui J. Chen Y, Chou WC, Sun L, Chen L, et al. (2011) An integrated transcriptomic and computational analysis for biomarker ide-tification in gastric cancer. Nucleic Acids Res 39: 1197-1207.

49.    Lukesova S, Kopecky O, Vroblova V, Hlavkova D, Andrys C, et al. (2008) Determination of angiogenic factors in serum by protein array in patients with renal cell carcinoma. Folia Biol (Praha) 54: 134-140.

50.    Reiman A, Lu X, Seabra L, Boora U, Nahorski MS, et al. (2012) Anticancer Res 32: 4663-460.

51.    Nonomura T, Masaki T, Morishita A, Jian G, Uchida N, et al. (2007) Identification of c-Yes expression in the nuclei of hepatocellular carcinoma cells: involvement in the early stages of hepatocarcinogenesis. Int J Oncol 30: 105-111.

52.    Di Masi A, Viganotti M, Antoccia A, Magrelli A, Salvatore M, et al. (2010) Characterization of HUH6, HEP3B, HEPG2 and HLE liver cancer cell lines by WNT/B-catenin pathway, microRNA expression and protein expression profile. Cell Mol Bio l56: 1299-1317.

53.    Gao WM, Kuick R, Orchekowski RP, Misek DE, Qiu J, et al. (2005) Distinctive serum protein profiles involving abundant proteins in lung cancer patients based upon antibody microarray analysis. BMC Cancer 5: 110.

54.    Kullmann T, Barta I, Csiszer E, Antus B, Horvath I (2008) Differential cytokine pattern in the exhaled breath of patients with lung cancer. Pathol Oncol Res 14: 481-483.

55.    Bartling B, Hofmann HS, Boettger T, Hansen G, Burdach S, et al. (2005) Comparative application of antibody and gene array for expression profiling in human squamous cell lung carcinoma. Lung Cancer 49: 145-154.

56.    Schröder C, Srinivasan H, Sill M, Linseisen J, Fellenberg K, et al. (2013) Plasma protein analysis of patients with different B-cell lymphomas using high-content antibody microarrays. Prot. Clin. Applications 7: 802-812.

57.    Ghobrial IM, McCormick DJ, Kaufmann SH, Leontovich AA, Loegering DA, et al. (2005) Proteomic analysis of mantle-cell lymphoma by protein microarray. Blood 105:  3722-3730.

58.    Díez P, Lorenzo S, Dégano RM, Ibarrola N, González‐González M, et al. (2016) Multipronged functional proteomics approaches for global identification of altered cell signaling pathways in B‐cell chronic lymphocytic leukaemia. Proteomics 16: 1193-1203.

59.    Kaufman KL, Belov L, Huang P, Mactier S, Scolyer R.A, et al. (2010) An extended antibody microarray for surface profiling metastatic melanoma. Journal of Immunological Methods 358: 23-34.

60.    Moschos SJ, Smith AP, Mandic M, Athanassiou C, Watson-Hurst K, et al. (2007) SAGE and antibody array analysis of melanoma-infiltrated lymph nodes: identification of Ubc9 as an important molecule in advanced-stage melanomas. Oncogene 26: 4216-4225.

61.    Ibrahim SO, Miron T, Krohn M, Amaratunga AN, Warnakulasauriya S, et al. (2005) Tumour-associated proteins in oral squamous cell carcinomas by proteomics. Cancer Genomics and Proteomics 2: 353-364.

62.    Zhu Z, Tang J, Gang D, Wang M, Wang J, et al. (2015) Antibody microarray profiling of osteosarcoma cell serum for identifying potential biomarkers. Mol Medicine Reports 12: 1157-1162.

63.    Loch CM, Ramirez AB, Liu Y, Sather CL, Delrow JJ, et al. (2007) Use of high density antibody arrays to validate and discover cancer serum biomarkers. Molecular Oncology 1: 313-320.

64.    Ramirez AB, Loch CM, Zhang Y, Liu Y, Wang X, et al. (2010) Use of a single-chain antibody library for ovarian cancer biomarker discovery. Mol Cell Proteomics 9: 1449-1460.

65.    Ramirez AB and Lampe PD (2011) Discovery and validation of ovarian cancer biomarkers utlilizing high density antibody microarrays. Cancer Biomarkers 8: 293-307.

66.    Hamelinck D, Zhou H, Li L, Verweij C, Dillon D, et al, (2005) Optimized normalization for antibody microarrays and application to serum-protein profiling. Mol. Cell. Proteomics 4: 773-784.

67.    Orchekowski R, Hamelinck D, Li L, Gliwa E, VanBrocklin M, et al. (2005) Antibody microarray profiling reveals individual and combined serum proteins associated with pancreatic cancer. Cancer Res 65: 11193-11202.

68.    Chen S, LaRoche T, Hamelinck D, Bergsma D, Brenner D, et al. (2007) Multiplexed analysis of glycan variation on native proteins captured by antibody microarrays. Nat Methods 4: 437-444.

69.    Ingvarsson J, Wingren C, Carlsson A, Ellmark P, Wahren B, et al. (2008) Detection of pancreatic cancer using antibody microarray-based serum protein profiling. Proteomics 8: 2211-2219.

70.    Li C, Simeone DM, Brenner DE, Anderson MA., Shedden KA, et al. (2009) Pancreatic cancer serum detection using a lectin/glyco-antibody array method. J Proteome Res 8: 483-492.

71.    Gerdtsson AS, Malats N, Säll A, Real FX, Porta M, et al. (2015) A multicenter trial defining a serum protein signature associated with panceatic ductal adenocarcinoma. Int J Proteomics 1-10.

72.    Mirus J, Zhang Y, Li C, Lokshin A, Prentice R, et al. (2015) Cross-species antibody microarray interrogation identifies a 3-protein panel of plasma biomarker for early diagnosis of pancreas cancer. Clin Cancer Res 21: 1764-1771.

73.    Shi W, Meng Z, Chen Z, Luo J.  Liu L (2011) Proteome analysis of human pancreatic cancer cell lines with highly liver metastatic potential by antibody microarray. Mol Cell Biochem 347: 117-125.

74.    Alhamdani M, Youns M, Buchholz M, Gress T, Beckers M, et al. (2012) Immunoassay-based proteome profiling of 24 pancreatic cancer cell lines. J Proteomics 75: 3747-3759.

75.    Miller JC, Zhou H, Kwekel J, Cavallo R, Burke J, et al. (2003) Antibody microrray profiling of human prostate cancer sera: antibody screening and identification of potential biomarkers. Proteomics 3: 56-63.

76.    Shafer MW, Mangold L, Partin AW, Haab BB (2007) Antibody array profiling reveals serum TSP-1 as a marker to distinguish benign from malignant prostatic disease. Prostate 67: 255-267.

77.    Wan F, Qin X, Zhang G, Lu X, ZhuY, et al (2015) Oxidized low-density lipoprotein is associated with advanced-stage prostate cancer. Tumor Biol 36:3573.

78.    Fujita K, Ewing CM, Sokoll LJ, Elliott DJ, Cunningham M, et al. (2008) Cytokine profiling of prostatic fluid from cancerous prostate glands identifies cytokines associated with extent of tumor and inflammation. Prostate 68: 872-882.

79.    El-Haibi CP, Singh R, Gupta P, Sharma PK, Greenleaf KN, et al. (2012) Antibody microarray analysis of signaling networks regulated by Cxcl13 and Cxcr5 in prostate cancer. J Prot Bioinform 5: 177-184.

80.    Yoshikawa R, Yanagi H, Shen C-S, Fujiwara Y, Noda M, et al. (2006) ECA39 is a novel distant metastasis-related biomarker in colorectal cancer. World J Gastroenterol 12: 5884-5889.

81.    Madoz-Gurpide J, Canamero M, Sanchez L, Solano J, Alfonso P, et al. (2007) A proteomics analysis of cell signaling alterations in colorectal cancer. Mol Cell Proteomics 6: 2150-2164.

82.    Olah Z, Kalman J, Toth M, Zvara A, Santha M, et al. (2015) Proteomic analysis of cerebrospinal fluid in alzheimer's disease: Wanted dead or alive. J Alzheimer’s Dis 44: 1303-1312.

83.    Arnold S, Louneva N, Cao K, Wang L, Han L, et al. (2013) Cellular, synaptic, and biochemical features of resilient cognition in Alzheimer's disease. Neurobiology of Aging 34: 157-168.

84.    Srivastava M, Eidelman O, Jozwik C, Paweletz C, Huang W, et al. (2006) Serum proteomic signature for cystic fibrosis using an antibody microarray platform. Mol. Genet. Metabolism 87: 303-310.

85.    Sacca SC, Centofanti M, Izzotti A (2012) New proteins as vascular biomarkers in primary open angle glaucomatous aqueous humor. Investigative Ophthalmology & Visual Science 53: 4242-4253.

86.    Guo W, Gao Y, Li N, Shao F, Wang C, et al. (2017) Exosomes: New players in cancer. Oncol Rep 38: 665-675.

87.    Whiteaker JR, Zhang H, Eng JK, Fang R, Piening BD, et al. (2007) Head-to-head comparison of serum fractionation techniques. J. Proteome Res 6: 828-836.

88.    Mahlknecht P, Stemberger S, Sprenger F, Rainer J, Hametner E, et al. (2012) An antibody microarray analysis of serum cytokines in neurodegenerative parkinsonian syndromes. Proteome Science 10: 71-71.

89.    Carter CA, Misra M, Pelech S (2011) Proteomic analyses of lung lysates from short-term exposure of Fischer 344 rats to cigarette smoke. J. Proteome Res 10 : 3720-3731.

90.    Rifai N, Gillette MA, Carr SA (2006) Protein biomarker discovery and validation: the long and uncertain path to clinical utility. Nat Biotechnol 24: 971-983.

91.    Paulovich AG, Whiteaker JR, Hoofnagle AN, Wang P (2008) The interface between biomarker discovery and clinical validation: the tar pit of the protein biomarker pipeline. Proteomics Clin Appl 2: 1386-1402.

92.    Petrak J, Ivanek R, Toman O, Cmejla R, Cmejlova J, et al. (2008) Déjà vu in proteomics. A hit parade of repeatedly identified differentially expressed proteins. Proteomics 8: 1744-1749.

93.    Hodgkinson VC, ELFadl D, Drew PJ, Lind MJ, Cawkwell L (2011) Repeatedly identified differentially expressed proteins (RIDEPs) from antibody microarray proteomic analysis. J Proteomics 74: 698-703.

94.    Zhang H, Shi X, Pelech S (2016) Monitoring protein kinase expression and phosphorylation in cell lysates with antibody microarrays. In: Zegzouti H, Goueli S (eds) Protein kinase screening and Profiling. Methods in Molecular Biology 1360: 107-122.  Humana Press, New York, NY.

95.    Olle EW, Sreekumar A, Warner RL, McClintock SD, Chinnaiyan AM, et al. (2005) Development of an internally controlled antibody microarray. Mol Cell Proteomics 4: 1664-1672.

96.    Dewing P, Christensen A, Bondar G, Micevych P (2008) Protein kinase C signaling in the hypothalamic arcuate nucleus regulates sexual receptivity in female rats. Endocrinology149: 5934-5942.

97.    Repka V, Fiala R, Čiamporová M, Martinka M, Pavlovkin J (2015) Antibody microarray expression profiling of maize roots treated with cadmium and nickel. Agriculture (Polnohospodárstvo) 61: 41-49.

98.    Yang T, Yao H, He G, Song L, Liu N, et al. (2016) Effects of lovastatin on MDA-MB-231 breast cancer cells: An antibody microarray analysis. J Cancer 7: 192-199.

99.    Yim JH, Yun JM, Kim JY, Nam SY, Kim CS (2017) Estimation of low-dose radiation-responsive proteins in the absence of genomic instability in normal human fibroblast cells. Int J Radiation Biology 93: 1197-1206.

100. Wingren C, Ingvarsson J, Dexlin L, Szul D, Borrebaeck CAK (2007) Design of recombinant antibody microarrays for complex proteome analysis: choice of sample labeling-tag and solid support. Proteomics 7: 3055-3065.

101. Wingren C, Borrebaeck C (2008) Antibody microarray analysis of directly labelled complex proteomes. Current Opinion in Biotechnology 19: 55-61.

102. Hudelist G, Pacher-zavisin M, Singer C, Holper T, Kubista E, et al. (2004) Use of high-throughput protein array for profiling of differentially expressed proteins in normal and malignant breast tissue. Breast Cancer Research and Treatment 86: 283-293.

103. Izzotti A, Bagnasco M, Cartiglia C, Longobardi M, Balansky RM, et al. (2005) Chemoprevention of genome, transcriptome, and proteome alterations induced by cigarette smoke in rat lung. Eur J Cancer 41: 1864-1874.

104. Lv LL, Liu BC, Zhang CX, Tang ZM, Zhang L, et al. (2007) Construction of an antibody microarray based on agarose-coated slides. Electrophoresis 28: 406-413.

105. Zhong D, Liu X, Khuri F, Sun S, Vertino P, et al. (2008) LKB1 is necessary for Akt-mediated phosphorylation of proapoptotic proteins. Cancer Res 68: 7270-7277.

106. Delfani P, Mellby LD, Nordström M, Holmér A, Ohlsson M, et al. (2016) Technical advances of the recombinant antibody microarray technology platform for clinical immunoproteomics. PLoS One 11: e0159138.

107. Gehring AG, Brewster JD, He Y, Irwin PL, Paoli GC, et al. (2015) Antibody microarray for E. coli O157:H7 and shiga toxin in microtiter plates. Sensors 15: 30429-30442.

108. Gembitsky DS, Lawlor K, Jacovina A, Yaneva M, Tempst P (2004) A prototype antibody microarray platform to monitor changes in protein tyrosine phosphorylation. Mol Cell Proteomics 3: 1102-1118.

109. Jin H, Webb-Robertson B, Peterson ES, Tan R, Bigelow DJ, et al. (2011) Smoking, COPD, and 3-nitrotyrosine levels of plasma proteins. Environmental Health Perspectives 119: 1314-1320.

110. Manning G, Whyte DB, Martinez R, Hunter T, Sudarsanam S (2002) The protein kinase complement of the human genome. Science 298: 1912-1934.

111. Kamps MP, Taylor SS, Sefton B (1984) Direct evidence that oncogenic tyrosine kinases and cyclic AMP-dependent protein kinase have homologous ATP-binding sites. Nature 310: 589-592.