The serum antibody repertoire is a unique repository of information regarding past immune encounters. Antibodies are produced in response to exposures and bind specifically to their target antigens. Antibodies associated with infection and disease can serve as diagnostic biomarkers upon detection. However, many disease-specific antibodies and their targets remain undiscovered. Therefore, to discover antibody biomarkers, reagents must be developed to specifically bind the disease-specific antibodies. Peptides are suitable reagents as they can often mimic native antibody epitopes with high affinity and specificity. Moreover, the sequences of antibody-binding peptides can be determined and used to identify the original antibody targets, revealing previously unknown antigens that contribute to disease progression and creating new opportunities for therapeutic development. Here, we developed and applied high-throughput methods to discover and characterize peptide epitopes from immune-related diseases.

A large bacterial display peptide library composed of randomized peptides was constructed to screen against human serum specimens and discover peptides that bind antibodies. For each specimen, millions of antibody-binding peptide sequences were determined using next-generation sequencing and analyzed computationally to reveal disease-specific binding motifs. This screening methodology was first employed to discover and characterize epitopes from two highly similar viruses, herpes simplex virus type 1 (HSV-1) and type 2 (HSV-2). Following antibody repertoire analysis, we discovered HSV type-specific motifs that could be used as diagnostic classifiers to achieve 100% diagnostic accuracy when distinguishing HSV-1 from HSV-2 and vice versa. Furthermore, numerous type-specific motifs were mapped to HSV antigens using protein sequence database alignments, including known antigens such as glycoproteins G and D as well as previously unreported antigens. We then applied this screening methodology to age-related macular degeneration (AMD). We identified a large panel of antibody-binding motifs associated with the onset of advanced AMD and developed a classifier with 84% accuracy when distinguishing specimens at high risk of advanced AMD from those at low risk. However, as the risk of developing advanced AMD increased, the classifier performance worsened suggesting a unique antibody signature associated with AMD progression.

Additional methods were developed to improve the discovery and characterization of epitopes from the antibody repertoire. We developed a method to selectively deplete highly abundant antibodies from a specimen to reduce repertoire complexity, improve the limits of detection, and enable the discovery of rare antibodies and their targets. Additionally, we developed a novel computational algorithm to rapidly determine binding motifs for epitopes by utilizing the full extent of next-generation sequencing datasets. Ultimately, these tools for analyzing the antibody repertoire at great depth can be applied to a broad range of infectious, autoimmune, and allergic diseases. The discovery of disease-associated epitopes will create new opportunities for the development of diagnostics, vaccines, and therapeutics.

Discovery of biologic molecules specific to a diseased state, or biomarkers, can lead to diagnostic development, therapeutic target identification, and improved understanding of disease pathogenesis. Antibodies remain an attractive class of biomarkers given their amplification by the immune system, stability, and current clinical use. While the antibody repertoire represents a rich source for biomarker discovery, it has been difficult to impartially identify which molecules from this repertoire are associated with disease. This work demonstrates three molecular discovery processes centered on the utility of bacterial displayed peptide libraries and fluorescence activated cell sorting (FACS) for identifying novel antibody biomarkers and their targets. We applied these methods to discover and characterize disease-associated antibodies for pre-eclampsia (PE), a condition with unknown etiology that affects 5-8% of pregnancies, using an unbiased approach.

Applying three quantitative screening strategies against a set of PE and healthy-outcome pregnancies (HOP) identified unique disease-associated antibody binding peptides from a fully random 15 amino acid peptide library. With a two-color screening method, we used antibody fractions enriched from plasma to isolate significantly PE cross-reactive and specific peptides distinct from a previously identified PE-associated antibody specificity. We used a panel of these antibody-detecting peptides to train and validate an Adaptive Boosting classification algorithm that achieved high specificity (95%) and a validated overall 80% diagnostic accuracy. To more closely replicate the native antibody binding environment, a second screening method used unprocessed, diluted plasma. This approach sequentially enriched peptides binding to PE antibodies and removed HOP antibody binders, resulting in a strong consensus motif that we further expanded through directed evolution. Importantly, we linked this motif to a region of a common viral antigen, Epstein-Barr virus nuclear antigen 1, and a human G protein-coupled receptor, GPR50, presenting a novel case for molecular mimicry. Thus, this method enabled unbiased identification of a disease-associated antibody and characterization of its targets. Finally, we developed and applied a unique methodology that combines bacterial displayed library screening with next-generation sequencing to profile the antibody repertoire of individual PE patients and HOP samples. This analysis re-identified the viral antigen-linked motif among several distinct PE- and HOP-associated antibody specificities, providing broader insights into alterations to the immune repertoire in PE.

This work demonstrates the utility of screening bacterial displayed peptide libraries to profile the antibody repertoire and identify new markers of disease. These disease-associated antibody-detecting peptide reagents enable development of molecular diagnostics and discovery of antibody binding target(s) to improve understanding of disease etiology and potentially elucidate therapeutic targets.

The adaptive immune system produces antibodies to specifically target antigens. Identifying and characterizing epitopes on target antigens enables numerous medical applications such as vaccine design, diagnostic discovery, and therapeutic antibody development. We have developed computational tools that characterize epitopes using large sets of antibody-binding peptides. To identify epitopes in target proteins, we used an approach termed K-mer Tiling of Protein Epitopes (K-TOPE). In this approach, we divided protein sequences into short overlapping subsequences of length k (k-mers). Then, we defined and scored epitopes using each k-mer’s enrichment in the sets of antibody-binding peptides. Using K-TOPE, we accurately identified epitopes for monoclonal and polyclonal antibodies. Next, using 250 specimens, we identified commonly targeted epitopes in nearly 3,000 viral proteins as well as two bacterial proteomes. Importantly, these epitopes agreed with previously reported results. To map antibody binding in epitopes, we developed Multiplexed Epitope Substitution Analysis (MESA). In this approach, target epitopes were divided into short overlapping k-mers. Then, each k-mer was exhaustively substituted with all amino acids. The effects of these substitutions were used to identify amino acid preferences at important binding positions in the epitopes. By applying this method to monoclonal antibodies and multiple sets of specimens, we identified binding motifs which agreed with an alternative computational approach. Finally, K-TOPE and MESA were used to characterize epitopes and antigens in age-related macular degeneration (AMD), herpes simplex virus (HSV), and Chagas disease. We identified 42 AMD-specific epitopes, 30 HSV2-specific epitopes, and 222 Chagas-specific epitopes. Several epitopes were in validated antigens while many were novel. MESA demonstrated that generally only 4-5 positions in these epitopes were important for binding. Future application of these approaches could enhance our understanding of the role that antibodies play in disease progression.

Proteases regulate many biological processes through their ability to activate or inactivate their target substrates and therefore present unique opportunities for therapeutic application. Because proteases catalytically turnover proteins, they could potentially be used at lower doses in therapy, reducing the cost of treatment. However, many proteases are capable of cleaving multiple physiological substrates. Therefore their activity, expression, and localization are tightly controlled to prevent unwanted proteolysis that could lead to side effects. Currently approved protease therapeutics rely on naturally evolved specificities and are often used for protease replacement therapy in genetically deficient patients. The clinical use of proteases in replacement therapy has been successful due to the narrow substrate specificity of these enzymes, which limits their toxicity. However, the application of proteases in therapy could be extended beyond their native biological functions. The emergence of methods for engineering proteases with new activities and narrow specificities toward substrates relevant in disease could greatly expand their therapeutic potential.

Here we have developed a novel intracellular screen in yeast for redesigning the specificity of human secreted proteases which we have termed protease evolution via cleavage of an intracellular substrate (PrECISE). Using PrECISE, a protease library and a target substrate flanked by fluorescent proteins CyPet and YPet, capable of Fӧrster resonance energy transfer (FRET), are co-expressed in the endoplasmic reticulum (ER) of yeast. The ER provides an oxidizing environment for the formation of disulfide bonds common to human secreted proteases and the co-localization of the protease and substrate promotes cleavage. Protease activity on the selection substrate leads to loss of intracellular FRET and increases cyan fluorescence enabling screening of large protease libraries using fluorescence activated cell sorting (FACS). As a model system, we screened randomly mutated and rationally designed libraries of the secreted protease human kallikrein 7 (hK7) using PrECISE to isolate variants with improved selectivity toward the hydrophobic core of the amyloid beta peptide (Aβ8: KLVF↓F↓AED).

Sequential rounds of low error rate random mutagenesis were found to be most effective in altering protease selectivity by incrementally introducing and screening for beneficial substitutions. Findings from our work emphasize the importance of screening large libraries during protease evolution since multiple substitutions were required to alter hK7 selectivity for Aβ. The substitutions found to improve hK7 selectivity would be impossible to predict since the majority were located far from the hK7 active site. Interestingly, improvements in selectivity were accompanied by a reduction in toxicity of the protease variant toward mammalian cells and improved resistance to wild-type inhibitors. Analysis of the crystal structures of improved variants provided insights to the potential mechanisms that affected hK7 activity and selectivity. The PrECISE method and techniques developed here can be broadly applied to evolve human proteases for specific degradation of toxic proteins involved in disease to enable their greater use in therapy.

Purpose

Experimental design

Results

Conclusions