Chemical Tools To Monitor and Manipulate The Adaptive Immune System

Abstract

The ability to monitor and manipulate antigen-specific immune responses would have a major impact on several areas of biology and medicine. In this perspective, I consider pharmacological methods to do this with a focus on the development of abiological “antigen surrogates” capable of binding to the antigen-binding sites of antibodies and B cell receptors with high affinity and selectivity. I will describe application of combinatorial library screening to identify antigen surrogates for monoclonal antibodies of therapeutic interest using chronic lymphocytic leukemia (CLL) as an example. Furthermore, I discuss the use of multiplexed assays for the quantification of antigen surrogate-antibody complexes as diagnostic tools and antigen surrogate discovery via serum screening. Although “antigen surrogates” are a fairly new concept I argue that they will open new avenues for both basic and clinical research and expect major advances over the next few years.

Antigen-Specific Immune Responses In Therapeutics and Diagnostics

The manipulation of antigen-specific immune responses is common in clinical medicine. By far the most important example is vaccination. Most vaccines introduce to the host immune system antigens derived from a pathogen. The resultant proliferation of antibodies and T cells that recognize these antigens affords protection from a subsequent infection by that pathogen. Extension of the vaccine concept to non-infectious diseases, especially cancers, is an active area of research. The idea is to identify tumor-specific antigens and vaccinate people with these to hyper activate cancer-specific immune responses(Palucka and Banchereau, 2014). There has also been exciting recent progress in engineering artificial antigen-specific immune responses by introducing into the patients own T cells engineered chimeric receptors (CARs) that recognize specific cancer antigens and trigger activation of the T cell. The engineered cells are then reintroduced to the patient where they attack the tumor(Barrett et al., 2014). The technologies mentioned above are focused on stimulating an immune response to a particular antigen. The flip side, eliminating or dampening responses to particular antigens through tolerization strategies (Roep et al., 2013), is of interest for the treatment of autoimmune disease.

All of the above technologies utilize biological strategies to manipulate antigen-specific immune responses. A little explored alternative strategy would be to develop drugs that do so. This would require “antigen surrogates”, that is synthetic compounds capable of binding tightly and selectively to the antigen-binding site of an antibody, B cell receptor (BCR) or T cell receptor (TCR) (Fig. 1). A high affinity ligand of this type could potentially block access of the antigen to its cognate receptor. Alternatively, the antigen surrogate could be tethered to some effector molecule, for example a toxin, resulting in a chimeric reagent capable of killing only pathogenic lymphocytes (Fig. 1). This would represent an interesting advance over the current state of the art in pharmacological manipulation of lymphocytes, such as the ability of Rituximab, an anti-CD20 therapeutic monoclonal antibody, to kill all B cells (Edwards et al., 2004) (Fig. 1). Alternatively, it might be possible to vaccinate patients with an antigen surrogate (Caulfield et al., 2010; Knittelfelder et al., 2009). Antibodies that recognize the surrogate might also have significant affinity for the native antigen of interest. This synthetic vaccine strategy would be quite useful in eliciting an immune response against a poorly immunogenic antigen or one that is difficult to prepare in large quantities.

Fig. 1.

A potential therapeutic application of antigen surrogates to monitor or treat chronic lymphocytic leukemia (CLL). A. A single antigen-specific B lymphocyte is amplified relentlessly in CLL. Yet because CLL B cells are deficient in differentiation into plasmablasts, the soluble antibody form of the B cell receptor (BCR) of the pathogenic cell is not present in the circulation (Chiorazzi et al., 2005). B. The state of the art in current pharmacological manipulation of B cells results in killing all CD20+ through the use of Rituximab or similar monoclonal antibodies (red). An antigen surrogate capable coupled to a toxin or a molecule that recruits effector functions (Murelli et al., 2009) could, in theory, eliminate only pathogenic B cells without affecting the healthy function of the humoral immune system.

Many investigators also believe that the adaptive immune response is a potential treasure trove of diagnostic biomarkers(Anderson and LaBaer, 2005). The underlying hypothesis is that many disease states are likely to produce molecules that are not present in healthy people, such as unusual post-translationally modified proteins, and that the adaptive immune system will react to these species as foreign antigens. The resultant disease antigen-specific antibodies or cells would thus serve as attractive biomarkers. As will be discussed below powerful genomic and proteomic methods to identify these putative antibody biomarkers are being explored, but these methods do not shed light on the native antigen. Yet to develop a practical and inexpensive clinical test to measure the levels of these antibodies, one requires a “capture agent” that can be immobilized on an ELISA plate or the like to retain the biomarker antibody from the serum. High affinity and selectivity antigen surrogates would be ideal for this application.

This perspective will present progress to date in the discovery and utilization of effective antigen surrogates, as well as discuss likely future directions in this area. I will focus entirely on targeting soluble antibodies and BCRs. TCR targeting, which is also feasible (Gocke et al., 2009) will not be discussed here.

Identification of antigen surrogates for monoclonal antibodies of therapeutic interest through combinatorial library screening

The simplest route to antigen surrogates would be to screen a suitable compound collection against a monoclonal antibody that one would like to target for therapeutic or diagnostic purposes. The identification of an antigen surrogate would be most important when the native antigen is unknown or impractical to employ as a targeting agent or diagnostic tool.

On the therapeutic side, a good example would be chronic lymphocytic leukemia (CLL) (Chiorazzi et al., 2005). In CLL a single antigen-specific B cell clone is amplified relentlessly, crowding out healthy B cells from lymph nodes and other immune centers, eventually forming a tumor. Moreover, CLL B cells are defective for differentiation into antibody-producing plasma cells, so these patients lack high levels of soluble antibodies corresponding to the pathogenic BCRs. This makes the CLL BCR a potentially interesting target for drugs that would deliver a cytotoxic moiety to them, sparing healthy B cells and thus allowing for continued function of normal humoral immune responses during treatment (Fig. 1). DNA sequencing of the heavy chain CDR3 region of thousands of CLL BCRs has revealed that about 35% can be grouped into sequence-related families called stereotypes (Agathangelidis et al., 2012), strongly suggesting that the amplification of these B cells is the result of a response to a limited number of autoantigens. However, with one exception (Chu et al., 2010), the identities of these suspected antigens are unknown. Thus, one would require antigen surrogates to develop the type of targeted toxins mentioned above. One could also imagine roles for compounds that target antigen-specific IgE monoclonal antibodies in blocking allergic responses (Handlogten et al., 2011).

Antigen surrogates that bind monoclonal antibodies will also be interesting reagents for diagnostic purposes. As mentioned above, there is significant interest in mining the adaptive immune response, and especially antibodies, for disease-specific biomarkers. Recently, advances have been made in methods to sort many thousands of individual B cells into small wells allowing native heavy and light pairing to be retained (Laserson et al., 2014; Tan et al., 2014). The CDR sequences of either the heavy or light chain, or both, of each BCR clone can then be determined by sequencing the corresponding cDNA (Georgiou et al., 2014). If this experiment is done with several case and control patients, then antibody consensus sequences restricted to the case population can be discovered and patient-to-patient variability can be assessed. While sequencing BCR-encoding genes could possibly be done as a clinical assay, it would be preferable to develop a simple ELISA protocol. Unfortunately, this genomics-driven discovery approach does not provide any insight into the native antigens recognized by the disease-restricted antibodies. Therefore, a high affinity, selective antigen surrogate would be quite useful in developing a simple ELISA-like diagnostic test in which it is used as the capture agent to retain the antibody of interest from a patient serum sample.

There is no question that the discovery of antigen surrogates is feasible. Indeed there is a considerable literature on peptide “mimitopes” isolated from different types of combinatorial peptide libraries that have a modest affinity for antibodies whose native antigens are not peptides (sugars, etc.) (Knittelfelder et al., 2009). Relevant to the CLL example provided above, a recent paper reported the identification of peptide ligands for the soluble IgG form of a CLL BCR from a phage display library (Seiler et al., 2009). However, the feasibility of identifying non-peptidic, more drug-like compounds with high affinity and selectivity for antigen-binding sites is less clear.

Because there is no obvious high-throughput functional assay with which to screen for antibody/BCR ligands, the most straightforward way to approach this problem is to employ a binding assay using small molecules displayed in either a microarray format (MacBeath et al., 1999) or on the surface of hydrophilic beads (Liu et al., 2002). The latter format is especially attractive since large one-bead one-compund (OBOC) libraries of certain types of unnatural oligomeric compounds such as peptoids or β-peptides can be produced readily using solid-phase split and pool synthesis (Figliozzi et al., 1996). For example, my laboratory has elaborated the classical “sub-monomer” synthesis of peptoids (Zuckermann, 1992) (oligomers of N-substituted glycines) to facilitate the synthesis of peptoid-inspired compounds with more chemically complex and conformationally restricted main chain scaffolds (Fig. 2) (Aditya and Kodadek, 2012; Aquino et al., 2011; Gao and Kodadek, 2013; Sarma and Kodadek, 2011; Suwal and Kodadek, 2013). Very recently, we screened one such library (Fig. 2) against several soluble IgG forms of CLL BCRs(Sarkar et al., 2014). This library was largely built using a COPA (chiral oligomers of pentenoic amides) scaffold, developed by Micalizio and co-workers (Aquino et al., 2011). Because of strong allylic 1,3 strain interactions, the COPA units greatly stiffen the main chain of the molecule relative to peptides or peptoids. Thus even small COPA oligomers fold into stable structures determined by the absolute stereochemistry at the chiral center (Aquino et al., 2011). About 1.3 million beads displaying diverse tetramers were screened by first incubating the OBOC library with a mixture of antibodies from patients that did not suffer from CLL. Beads that retained significant amounts of antibody were visualized by the addition of a fluorescently labeled secondary antibody and the fluorescent beads were removed from the population (Fig. 2). The remainder of the library was then incubated with three monoclonal CLL IgGs and, again, the hits were identified using a fluorescent secondary antibody. Antigen surrogates with good affinity and selectivity were identified for two of the three targets. For example, the compound KMS5, shown in Fig. 2, bound to the CLL 169 IgG with a K_D of approximately 90 nM and did not show significant affinity for other IgGs with different antigen-binding sites. Efforts are underway to improve the affinity of the molecule for the BCR. Importantly, when KMS5 was mounted onto a biotinylated dextran polymer, it bound with high affinity and selectivity to patient-derived CLL 169 B cells, but not to B cells displaying other antigen-specific antibodies (Sarkar et al., 2014). To the best of my knowledge, this is the first example of a synthetic, unnatural molecule recognizing an antigen-specific lymphocyte. As such, it constitutes a critical first step in the development of chimeric molecules of the type shown in Fig. 1B. It is interesting to note that the CLL 169 IgG was also employed as the target of a screen using a 12-residue phage-displayed peptide library (Seiler et al., 2009). The best ligands to arise from this screen were of similar affinity to the much smaller, proteolytically stable synthetic oligomer.

Fig. 2.

Peptoid-inspired, conformationally constrained oligomers provide a valuable source of antigen surrogates. A. Structures of acid sub-monomers that have been developed for use in the peptoid submonomer scheme (box). The 2-oxopiperazine and diketopiperazine structures are made via a multi-step sequence on the resin rather than being true sub-monomers (Suwal and Kodadek, 2013). B. A combinatorial library that was employed to screen for antigen surrogates that bind to the soluble IgG form of CLL BCRs. The library is comprised of a peptoid unit followed by three COPA units. The structures of the amine submonomers employed are shown. The library contained approximately 1.3 million different compounds (Sarkar et al., 2014). C. Schematic representation of the screening protocol employed to mine antigen surrogates from a combinatorial library. The OBOC library is first exposed to a large number of human IgG antibodies obtained from healthy volunteers. Beads that retain significant levels of antibody are detected by subsequent incubation with a red quantum dot-labeled secondary antibody. These beads, which display antibody ligands that are not of interest, are removed from the library. The remainder of the beads are then screened against the antibody of interest, in this case the soluble IgG form of a BCR from a CLL patient. The hits are again identified and collected. The compounds are released from the bead via CNBr-mediated cleavage of the methionine residue in the conserved linker and the structure is determined by mass spectrometry (Sarkar et al., 2013). The structure of the highest affinity hit, KMS5, which was obtained in a screen against CLL 169 (a particular patient-derived BCR), is shown. This proved to be a 500 nM ligand for the IgG form of the BCR in solution and 90 nM when immobilized on an ELISA plate, the difference presumably reflecting avidity effects on the plate.

Multiplexed Assays for the quantification of antigen surrogate-antibody complexes as diagnostic tools

As mentioned above, next generation DNA sequencing-based methods to analyze the repertoire of large numbers of individual Ig-producing cells is proceeding at a rapid rate (Georgiou et al., 2014). While few such studies have been done yet, one can imagine that this technique will soon be employed to identify antibodies that distinguish case and control populations for many disease states. These antibodies could then be expressed recombinantly and serve as targets for antigen surrogates screens of the type discussed above for CLL. While these are still early days, the promising results obtained in the CLL study (Sarkar et al., 2014) raises the hope that, when presented with a monoclonal antibody, the identification of antigen surrogates with good affinity and selectivity will soon be fairly routine. In anticipation of this workflow becoming more important, my research group has developed a highly multiplexed “liquid array” platform optimized for measuring interactions between small molecules and serum antibodies (Fig. 3) (Doran and Kodadek, 2013). This system was inspired by the Luminex platform (Vignali, 2000), in which latex microspheres displaying a particular ligand on their surface (usually a nucleic acid or antibody) are encoded by adsorption of specific concentrations and ratios of two different colored dyes in the interior of the bead. This system is useful for multiplexed sandwich assays in which an immobilized antibody binds an analyte, whose level is then measured by the addition of a labeled sandwich antibody. The analysis is done using a specialized flow cytometer-like instrument with three lasers, two of which “read” the bead's encoding dyes with the third irradiating the dye on the sandwich antibody. This system is sub-optimal for the analysis of serum antibodies using immobilized small molecules however (Doran and Kodadek, 2013). First, when one attempts to do organic chemistry on these beads, the dyes leach out rapidly in organic solvents, resulting in loss of the color cod. Second, there is a high degree of non-specific binding of serum antibodies to the beads. We solved both of these problems by using 10 μm TentaGel microspheres as the solid support, which consist of an amine-functionalized polystyrene core with a thick outer coating of amine-terminated polyethylene glycol (PEG) chains. Lam and co-workers developed a protocol by which one can do chemistry selectively on the interior or exterior domains of these beads (Liu et al., 2002). We used this technique to immobilize particular ratios of Pacific Blue and Pacific Orange covalently to the interior of the bead, which is not exposed to proteins, to encode. The small molecule capture agent was then affixed to the hydrophilic outer layer. The thick PEG coating results in a very low degree of non-specific binding. Moreover, the density of the small molecule ligand on the surface of the bead is sufficient that both arms of an antibody engage ligands, providing higher affinity through an avidity effect (Doran and Kodadek, 2013). After incubation of the beads with a serum sample, they are washed, incubated with a labeled secondary antibody labeled with a third color, washed again, then analyzed using a standard flow cytometer. In our published work, we demonstrated the ability to analyze 24 different antibody-small molecule complexes simultaneously(Doran and Kodadek, 2013), but further efforts are under way to extend this number further.

Fig. 3.

Luminex-like liquid array platform for the multiplexed analysis of small molecule-antibody interactions. A. Schematic of the biphasic TentaGel microsphere-based system. The hydrophobic interior domain of the beads is modified covalently with a particular concentration and ratio of two dyes, Pacific Blue and Pacific Orange. The hydrophilic outer layer of the bead, which is exposed to solution, is modified with the antigen surrogate capture agent. When exposed to serum, the amount of antibody captured by the antigen surrogate is quantified by subsequent incubation with a secondary antibody labeled with a third color. B. Illustration of bead sorting and fluorescence “reading” using a common flow cytometer requiring two excitation lasers and three detectors. C) A typical data plot showing the ratiometric emission intensities of the two encoding dyes. Each subpopulation can be separated from a batch of differentially dyed microspheres. d) Upon gating a designated subpopulation, binding can be quantified as the relative intensity of a reporter fluorochrome (dye 3). e) FACS dot plot of Pacific Orange vs. Pacific Blue emission intensities for a batch of 24 subpopulations of encoded microspheres.

F. Comparison of Luminex vs. TentaGel platforms in serological measurements. a) Binding isotherms generated for the detection of anti-ADP3 IgY in chicken serum using ADP3 immobilized onto TentaGel microspheres. Binding was quantified by measuring the mean fluorescence intensity (MFI) of a PE-conjugated anti-IgY antibody. b) The same experiment was performed using ADP3 immobilized onto Luminex microspheres. Reprinted with permission from ref. (Doran and Kodadek, 2013).

Looking into the future then, the rapidly expanding power of immunogenomics is likely to provide increasing numbers of diagnostically useful monoclonal antibodies. I believe it is likely that over the next few years the ability to identify high affinity and selectivity antigen surrogates capable of retaining these antibodies from serum will become routine. Combined with the existing analytical technology mentioned directly above, one can foresee an exciting future in which more and more important diseases are diagnosed through the use of multiplexed immunoassays using synthetic capture agents. Finally, since there is a reasonable expectation that immune responses against disease states are likely to occur early in the pathogenic process, there is the hope that this type of technology would enable the development of a new kind of pre-symptomatic diagnostics in which these kind of highly multiplexed antibody detection assays could be performed each time a patient comes in for an annual physical.

Antigen Surrogates From Serum Screening: The Promise and Challenge of Screening Combinatorial Libraries Against Complex, Polyclonal Antibody Populations

As mentioned directly above, deep sequencing of lymphocyte populations is likely to facilitate the discovery of diagnostically interesting antibodies. But it is not the only path to this goal. Indeed, an issue with this approach, as pointed out by Georgiou and colleagues (Georgiou et al., 2014), is that the readily accessible, circulating peripheral B cells represent only a small fraction of the immune repertoire. What one would really like to do is to screen the circulating antibody population itself for biomarkers. Therefore, there is continued interest in other methods to tackle this problem. Some efforts are already being made to employ deep sequencing data to enable interpretation of mass spectra of antibody-derived tryptic peptides to find those corresponding to the variable regions. It will be interesting to see how applicable this approach is to biomarker discovery (Boutz et al., 2014). The most common approach to antibody profiling currently available is to screen case and control antibody populations against some collection of candidate disease-specific antigens and identify one that binds antibodies rich in the case population but absent in the control population (Fig. 4). In other words, it is a search for the antigen, not an antigen surrogate. This type of experiment has been done with proteome arrays (Lueking et al., 2003; Nagele et al., 2011; Robinson et al., 2003), lipid arrays (Kanter et al., 2006), arrays created from chromatographic fractions of tissue lysates (Qiu et al., 2004), and a variety of other tools. Somewhat similar in approach is the use of peptide libraries for this application (Restrepo et al., 2011; Robinson et al., 2002; Wang et al., 2005). Of course, large random peptide libraries or those made from mRNA via shearing-based methods will be comprised mostly of sequences not found in the proteome (Wang et al., 2005), but nonetheless the hope in most such efforts is to identify peptides that would closely resemble a native linear epitope.

Fig. 4.

Searching for Ig biomarkers via hybridization of serum to arrays of antigens or antigen surrogates. All such experiments involve exposing some collection of molecules arrayed on chemically modified glass slides to case and control serum samples that contain all circulating antibodies. After washing, the degree of antibody binding to each feature of the array is determined by subsequent hybridization of a labeled secondary antibody (a green dye in this schematic) and scanning of the array. The goal is to identify features on the array that capture significantly more antibody from the case sera than the control sera. Three spots are highlighted (indicated by arrows) to illustrate this idea. The hypothetical biomarker antibody is colored pink in this figure.

While these efforts have had some success, a rush of useful antibody biomarkers has not resulted from this work. Why this is so is not clear. One hypothesis is that, outside of infectious diseases and some autoimmune conditions, the most useful antibody biomarkers might result from an adaptive immune response against unusually modified proteins or other antigens, for example oxidized species. If so, these antigens would not be present in collections of “vanilla” proteins, peptides or most other biomolecule collections and the search would largely fail. With this idea in mind, my research group decided to explore a different idea, which was to carry out differential case vs. control screens on libraries of unnatural molecules in hopes that these libraries would contain surrogates (not mimics) of the most interesting disease-specific antigens (Reddy et al., 2011).

Initially, this experiment was done using an animal model system for MS (multiple sclerosis) called EAE (experimental autoimmune encephalomyelitis). This involves immunizing a mouse with a 21-residue peptide derived from the murine nerve sheath protein MOG (myelin oligodendrocyte glycoprotein) along with an appropriate adjuvant to break tolerance and thus drive an autoimmune attack, reproducing some aspects of MS. Serum from EAE mice or control mice were hybridized to planar glass microarrays displaying 8600 octameric peptoids taken from a combinatorial library created by solid-phase split and pool chemistry (Fig. 4). Several peptoids were identified that were subsequently shown to bind to the anti-MOG peptide antibodies, but not to other murine antibodies with different antigen-binding sites (Reddy et al., 2011). Excellent diagnostic sensitivity and specificity could be achieved by monitoring IgG antibody binding to these peptoids. These data demonstrated that the antigen surrogate concept is indeed extendable to serum antibody screening and biomarker discovery, at least in this simple disease model.

Of course, murine EAE is a very simple disease model and it was not difficult to imagine why this approach might fail when applied to human disease. In the EAE model, the antigen is a single, modest-sized peptide. For any human disease, there will surely exist many antigens, and many might be large proteins with several epitopes. If so, one would expect a complex, polyclonal immune response. While I have used the term antigen surrogate in this article, the modest-sized molecules that we are using would most likely serve as a surrogate for a particular epitope of a larger antigen. If the antibody population is highly polyclonal, it might be difficult to see a strong signal in the case vs. control screen since each molecule might bind only a small fraction of this polyclonal spectrum. Also, since these antigen surrogates must interact with a given antigen-binding pocket differently than does the native antigen, it could be that the antigen surrogate relies on contacts with residues that are unimportant for native antigen recognition. If so, then it is possible that an antigen surrogate might have a highly restricted binding profile even to the polyclonal population of antibodies that all bind the same native epitope. This raises the specter that an antigen surrogate which recognizes an antibody in patient A would not cross-react with an antibody in patient B even if patients A and B have antibodies to the same native epitope. Indeed, this complex issue of polyclonality makes antigen surrogate discovery via serum screening a completely different and far more difficult game than library screens against a single protein target. Finally, there are potentially major issues with the clinical samples themselves. Those provided by different centers may have been collected or stored under different conditions, affecting antibody stability. Or the clinical diagnosis could simply be wrong. This is a major concern in many neurological conditions and other areas of medicine where diagnoses are largely made on the basis of symptoms. Even in diseases where a reliable gold standard for diagnosis exists, such as detection of colon cancer by colonoscopy, how many different molecular pathways might exist to produce that phenotype? If it is 10, then presumably the adaptive immune system might react to each by producing different antibodies and a simple case vs. control screen with 10 patients would be unlikely to provide obvious hits.

Even given all of these concerns, we conducted a modest test of the idea in Alzheimer's disease (AD). Again, an array of 8600 peptoids was used to screen IgG antibodies in 6 AD patients, six Parkinson's disease (PD) patients and six non-demented, but age-matched, controls. Three peptoids were found to bind at least 4-fold more antibodies from all of the AD patients than any of the PD or control samples (Reddy et al., 2011). On the microarray surface, these peptoids, especially one called ADP3, provided good diagnostic sensitivity and specificity in a small preliminary, open label study of 50 patients. At about the same time, two other reports appeared that also suggested that there are AD-specific autoantibodies (Nagele et al., 2011; Restrepo et al., 2011), though neither found good single markers, but rather relied on a fingerprint or algorithmic treatment of multiple antibody-peptide or antibody-protein complexes formed on arrays. This generated considerable excitement, although some more recent unpublished result from my lab suggest that the ADP3 peptoid is not a sufficiently good marker to consider for clinical application, largely due to an unacceptable level of false positives. Also, significantly different results were observed using samples collected at different institutions, indicating that more standardization is required. Even more problematic is the fact that even mediocre results are only obtained on the microarray platform, which was custom made and had a thick layer of PEG on the surface to block any non-specific antibody binding. These would be difficult to deploy for large scale testing since their manufacture is a tedious process with significant batch-to-batch variation. When ADP3 or the other peptoids were immobilized on ELISA plates, the results were disappointing, with almost all of the signal being due to non-specific IgG binding to the peptoid-coated plate (unpublished observations). This is very likely due to the low affinity of the peptoid for the ADP3-binding antibodies and/or the low levels of these antibodies in the serum, with the small amount of specific signal being overwhelmed by non-specific binding. These experiments were done prior to the development of the Luminex-inspired assay, and future efforts will employ this platform, though we are currently focused on the identification of much improved antibody ligands. We also screened a similar peptoid library against serum samples from patients with the autoimmune neuroinflammatory disease neuromyelitis optica (NMO) (Raveendra et al., 2013). About 75% of NMO patients have autoantibodies against Aquaporin 4 (AQP4) (Lennon et al., 2005), a water transport protein found on the surface of cells that line the optic nerve and are attacked in this disease. This study used a different screening protocol quite similar to that shown in Fig. 2C in which a library of about 100,000 peptoids synthesized on TentaGel beads were exposed first to serum from normal patients and then, after clearing beads that retained antibodies from the control samples, to serum from NMO patients. This bead-based sequential screening protocol has the advantage of allowing far more compounds to be screened than is possible using the microarray format. One of the hits, called NMOP6 proved to be a ligand for anti-AQP4 autoantibodies and, when displayed on a microarray surface, provided excellent diagnostic sensitivity and specificity for the diagnosis of NMO in a blinded study, comparable to that achieved using the native antigen AQP4 as the probe (Raveendra et al., 2013). NMOP6 was also moderately effective was used in an ELISA format. Again significant non-specific IgG binding was observed, but this constituted only about 50% of the true signal rather than overwhelming it as was the case for Alzheimer's disease (B. Raveendra, W. Hao and T.K., in preparation), despite the fact that the affinity of NMOP6 for a monoclonal antibody against AQP4 is weak (>10 μM). The ability of the specific signal to rise above the noise in the ELISA assay for NMO, but not AD, may reflect the high levels of the NMO autoantibodies in the serum.

Another major problem that has limited our progress over the last few years is the high level of false positives that the bead screening protocol provides (Lian et al., 2013). By false positives, I mean compounds that, at the level of the initial bead screen, appear to be excellent ligands for antibodies found only in the case samples, but when re-synthesized and tested on other analytical platforms, including the TentaGel bead flow assay, display poor binding. This results in an enormous amount of wasted time and effort re-synthesizing and characterizing screening hits that ultimately prove useless. Thankfully, we have recently solved this problem as well. It was shown that these false positives are extremely low affinity antibody ligands that happen to be displayed on a TentaGel bead of unusually high density (the loading of individual beads in a batch of TentaGel beads varies by over 20-fold) (Doran et al., 2014). This likely traps bivalent antibodies kinetically in a microenvironment of high ligand density even though the intrinsic affinity is terrible. Fortunately, there is a simple solution, which is to employ redundant OBOC libraries in the screen and only devote resources post-screening to compounds isolated more than once. The idea is that the super-high density beads causing the problem are rare in the population and thus it is highly unlikely that the same ligand found on several different beads is a false positive. Indeed we have found that hits isolated more than once from redundant libraries are almost always of high quality (Doran et al., 2014). This has accelerated our work tremendously.

Concluding remarks

In summary, a great deal of progress has been made solving vexing technical issues in the serum screening process and a new, highly multiplexed analytical platform has been developed that simplifies the analysis of antigen surrogate-antibody interactions. Another important advance has been the development of new classes of oligomers that are far more conformationally constrained than the floppy peptoids (Aditya and Kodadek, 2012; Aquino et al., 2011; Gao and Kodadek, 2013; Sarma and Kodadek, 2011; Suwal and Kodadek, 2013) (Fig. 2A). Several screens against antibodies and other proteins have indicated that these new libraries are a source of much higher affinity ligands than are peptoid libraries (Aquino et al., 2011; Gao and Kodadek, 2013). This will be important to allow the analysis of lower titer antibodies. Other improvements are currently being explored. For example, unrestricted use of all of the building blocks we have developed as diversity elements in library synthesis is constrained by the current need to deduce hit structures by mass spectrometry. Some linkages fragment better than others, so the interpretation of fragmentation patterns of molecules with mixed backbones can be challenging (Sarkar et al., 2013). This is unfortunate, since greater scaffold diversity in these libraries is highly desirable. Thus, we are moving towards the development of DNA-encoded(Brenner and Lerner, 1992; Clark, 2010; Scheuermann and Neri, 2010) bead libraries that will remove this limitation. Thus, with much improved protocols and libraries in hand, the next 1-2 years should provide a fair test of the real utility of this approach for the discovery of antibody biomarkers for human disease.