Abstract Background Identification of immune signatures in pre-clinical IBD provides a unique window of therapeutic intervention with disease-modifying potential. This study aimed to define the antigen specificities, breadth, longitudinal stability, and predictive ability of pre-clinical serum antibody repertoires using a novel, high-throughput phage-display immunoprecipitation sequencing assay (PhIP-Seq). Methods Longitudinal serum samples from active component US military personnel collected at diagnosis and at ~10, ~4, and ~2 years pre-Crohn’s disease (CD, n=200), pre-ulcerative colitis (UC, n=200) and in age- and sex-matched healthy controls (HC, n=100) were analysed (Fig. 1A-B). Serum antibody repertoires were defined against 357,000 microbial-, viral-, food-, and immune-associated peptide antigens (Fig. 1C). Conditional logistic regression analyses were performed to identify differentially abundant antibodies per timepoint. Temporal stability of antibodies was assessed using point-biserial correlations. Longitudinal sample analyses were performed using generalized linear mixed-effects models. Finally, machine learning models, including tree-based ensemble methods, were used for prediction of disease. Results Across 2,000 samples, 123,980 unique antibodies were identified (Fig. 1D). Antibody repertoire diversity was significantly lower in pre-CD (P<0.05) and pre-UC (P<0.001) sera compared to HC sera (Fig. 1E). Up to ~4 years pre-diagnosis, pre-CD samples exhibited more fluctuations in antibody repertoires compared to pre-UC and HC samples (both P<0.001, Fig. 2A). Distinct antibody responses were evident up to 10 years pre-CD and pre-UC versus HCs (Fig. 2B-C). Pre-CD individuals demonstrated escalating trajectories of antibody responses against herpesviruses including Epstein-Barr virus (EBV), cytomegalovirus, and varicella zoster virus, and against a wide range of bacterial flagellins (Fig. 2D-E). Pre-UC individuals exhibited rising antibody responses against herpesviruses and autoantigens including MAP-kinase-activating death domain protein and reduced reactivity against encapsulated pathogens (e.g., Streptococcus pneumoniae- and Haemophilus influenzae) (Fig. 2F-G). Machine learning models identified antibody signatures highly predictive of CD (AUC=0.90) and UC (AUC=0.84) up to 10 years pre-diagnosis (Fig. 2H-K). Additionally, antibody repertoires differentiated complicated from non-complicated CD up to ~10 years pre-diagnosis with high accuracy (AUC=0.85). Conclusion Antibody profiling delineates a previously unappreciated landscape of serological responses against microbial and non-microbial antigens up to 10 years before disease onset in patients with CD and UC, allowing for novel insights on IBD pathogenesis and disease prediction.
Read full abstract