Phage display of environmental protein toxins and virulence factors reveals the prevalence, persistence, and genetics of antibody responses

Abstract

Microbial exposures are crucial environmental factors that impact healthspan by sculpting the immune system and microbiota. Antibody profiling via Phage ImmunoPrecipitation Sequencing (PhIP-Seq) provides a high-throughput, cost-effective approach for detecting exposure and response to microbial protein products. We designed and constructed a library of 95,601 56-amino acid peptide tiles spanning 14,430 proteins with "toxin" or "virulence factor" keyword annotations. We used PhIP-Seq to profile the antibodies of ∼1,000 individuals against this "ToxScan" library. In addition to enumerating immunodominant antibody epitopes, we studied the age-dependent stability of the ToxScan profile and used a genome-wide association study to find that the MHC-II locus modulates bacterial epitope selection. We detected previously described anti-flagellin antibody responses in a Crohn's disease cohort and identified an association between anti-flagellin antibodies and juvenile dermatomyositis. PhIP-Seq with the ToxScan library is thus an effective tool for studying the environmental determinants of health and disease at cohort scale.

Keywords: Crohn's; ToxScan; antibody epitoe profiling; disease; humoral immunity; immunogenetics; juvenile dermatomyositis; microbiome.

Figure 1.. Serum antibody profiling via T7 phage display of 14,430 toxins and virulence factors detects hundreds reactivities per individual.

A. (i) Protein sequences with “toxin” (collapsed to 90% sequence identity) or “virulence factor” (unique sequences) were downloaded from UniProt. (ii) The proteins were represented as 95,601 56 aa peptide tiles with 28 aa overlaps and encoded as 200-mer oligonucleotides. The library was cloned into the T7 phage display system and mixed with serum or plasma antibodies. (iii) Antibody and bound phage are immunocaptured using magnetic beads. (iv) The library inserts are amplified by PCR and sequenced. B. ToxScan library QC: 99.3% of all library members were detected; 92.6% of the library was within one log (plus or minus) of the mean (indicated by vertical dashed lines). C. Number of ToxScan protein reactivities detected per individual in a healthy cohort (n = 598, median = 144, sd = 66). D. Number of VRC samples (n = 598) reactive to at least 3 virulence factor or toxin peptides produced by an organism; the number of peptides from the organism in the ToxScan library, and the number of peptides enriched at a threshold of 5% prevalence (reactive in 30 or more individuals). See also Table S1.

Figure 2.. PhIP-Seq with the ToxScan library reveals hundreds of immunoprevalent reactivities in a healthy human cohort (n = 598).

A. Heatmap showing the 848 immunoprevalent ToxScan peptide reactivities (at a threshold of 5% seroprevalence). Each row is a peptide and each column is a sample. Column annotations indicate the age, race, and sex of each individual. The color intensity of each cell indicates the fold change of each peptide versus mock immunoprecipitations (IPs). “Example cluster 1” is composed of Staphylococcus aureus and S. epidermidis serine-aspartate repeat-containing proteins E, F, and G (SdrE, SdrF, and SdrG), and clumping factors A and B (ClfA and ClfB). “Example cluster 2” is composed of Shigella flexneri peptides, including the invasion plasmid antigens A, B, and C (IpaA, IpaB, and IpaC). B. The 848 immunoprevalent peptides were aligned using blastp and a shared epitope network graph was constructed, with yellow dots corresponding to immunoprevalent peptides, red dots corresponding to “Example cluster 1” peptides and blue dots corresponding to “Example cluster 2” peptides. See also Table S1.

Figure 3.. Prevalent antibodies target proteins from three commonly encountered bacteria.

Data in this figure are bacterial species-specific subsets of the VRC reactivities targeting the 848 immunoprevalent peptides. Column annotations and colors are as in Fig. 2. Column order is determined by hierarchical clustering. Row order is determined by hierarchical clustering, then subgrouped by protein family. A. Reactivities to S. pneumoniae: immunoglobulin A1 protease (IgA1P), pneumococcal histidine triads D and E (PhtD, PhtE), surface protein PspA, and zinc metalloproteases B and C (ZmpB, ZmpC). B. Shared epitope network graph depicting S. pneumoniae immunoprevalent ToxScan peptides and linear IEDB antibody epitopes. C. Reactivities to S. aureus: bifunctional autolysin (Atl), clumping factor (Clf), fibronectin binding protein A (FnBPA), immunoglobulin-binding protein Sbi, iron-regulated surface determinant B (IsdB) protein, N-acetylmuramoyl-L-alanine amidase (Sle1), and Staphylococcal secretory antigen (SsaA). D. Shared epitope network graph depicting S. aureus immunoprevalent ToxScan peptides and linear IEDB antibody epitopes. E. Reactivities to E. coli: alpha-hemolysin (HlyA), hemolysin (Hly), and translocated intimin receptor (Tir). F. Shared epitope network graph depicting E. coli immunoprevalent ToxScan peptides and linear IEDB antibody epitopes. See also Figure 2, Table S1, and Figure S3.

Figure 4.. ToxScan reactivities reach a stable diversity set point in adulthood.

A. Total number of reactive ToxScan peptides versus age (VRC cohort) reveals a decline of 0.89 reactivities per year (one-way ANCOVA, p = 2.8×10⁻⁶) after adjusting for covariates (sex, age, race, CMV status). B. The diversity of ToxScan reactivities over ~5 years in older adults (BLSA cohort, n = 47). The first time point is indicated in red, the second in blue. ToxScan (C) and VirScan (D) Longitudinal correlations of public peptide reactivities over ~5 years in older adults (BLSA cohort, n = 47). See also Table S1.

Figure 5.. GWAS analysis of ToxScan reactivities in the VRC cohort links the MHC-II locus with selection of bacterial antibody epitopes.

A. Three Staphylococcus aureus peptides (two overlapping peptides from clumping factor A protein and one from Staphylococcal secretory antigen) associated with the MHC-II locus on chromosome 6, and one peptide associated with nucleoredoxin gene (NXN) on chromosome 17. All variants with a log₁₀ transformed Bayes factor of ≥ 6 (dashed red lines) were considered significant. B. Zoom plot for association with reactivity targeting two overlapping peptides from S. aureus clumping factor A (ClfA). Region defined by credible sets analysis shaded in pink. C. Similar to B but for S. aureus Staphylococcal secretory antigen A1 (ssaA1). See also Figure S4 and Table S3.

Figure 6.. ToxScan profiling associates anti-flagellin antibodies with autoinflammatory diseases.

A. Volcano plot depicting differences in prevalence of antibody reactivities among Crohn’s disease patients versus the VRC cohort. Each point denotes a flagellin peptide. Significance of association was corrected for multiple testing using the Benjamini-Hochberg (BH) procedure. The 14 starred peptides are significantly associated with Crohn’s; these peptides share common epitopes (Figure S6). The data labels indicate the percentage of the cohorts which are reactive (Crohn’s %, VRC %). B. Comparison of strengths of reactivity between the Crohn’s and VRC cohorts for the most differentially reactive peptide (Helicobacter mustelae, Mann-Whitney U Test, p = 0.003). C. Correlation of ToxScan reactivities in 3 different myositis cohorts versus the VRC cohort (Fisher’s exact test with BH corrected p-values). D. E. coli flagellin FliC MSD ratio is shown for 22 JDM cases and their age-matched controls (matched pairs are connected by lines). Lines connect matched cases and controls. See also Figure S6, Table S4, Table S5, and Table S6.