Abstract Background Although multiple biomarkers have been linked to the future risk of Crohn’s disease (CD), the time trajectory of biological events during the preclinical phase remains poorly defined. Here, we leverage serum proteomics across three preclinical cohorts to map the molecular trajectory preceding CD onset and to develop a proteomic risk score for predicting disease onset. Methods We integrated three nested case-control preclinical cohorts: the GEM Project, PREDICTS, and UK Biobank, encompassing 2,034 serum samples (752 pre-CD and 1282 matched controls; Table 1), in which up to 5,416 proteins were quantified using Olink® HT. Conditional logistic regression was used to assess individual protein associations with future CD risk. Estimated trajectory analysis evaluated ‘dynamic change’ (the slope of protein level distribution over time in pre-CD vs controls) and ‘variation timepoint’ (when protein levels begin to diverge between pre-CD vs controls). Functional pathways were interrogated using Gene Ontology enrichment. A machine learning-based Proteomic Risk Score (PrRS) was developed using 70% of the GEM cohort (training) and validated in the remaining 30% (testing) and in the external PREDICTS cohort. Results In GEM, 73 proteins were associated with CD risk and 108 showed dynamic changes preceding diagnosis, with 97.3% and 75.2%, respectively, showing consistent directionality in ≥ 1 external cohort. These validated proteins were classified by variation timepoint prior to diagnosis: Near onset (2 years; n = 21), Early stage (2-4 years; n = 28), Very early stage (4 years; n = 12), or Parallel change (significant in association analysis but without dynamic change; n = 63). Pathway analysis revealed enrichment for host-microbe interaction pathways in Near onset proteins; extracellular matrix organization and barrier integrity in Early and Very early stage proteins; and innate immunity among Parallel change proteins. The PrRS, integrating both parallel-change and dynamic-change proteins (n = 11), achieved AUCs of 0.806 and 0.751 for predicting CD onset in the GEM testing and PREDICTS cohorts, respectively. PrRS values were associated with the time before diagnosis (i.e., higher the score, the closer to diagnosis) in both GEM testing (Kruskal-Wallis, P = 1.7 × 10-7; Figure 1) and PREDICTS (P 2.2 × 10-16). Conclusion Proteomic signatures with distinct time-trajectories reveal the molecular evolution of CD during its preclinical phase, with findings consistent across first-degree relatives and average-risk population. The PrRS derived from these proteins enables stratification of individuals along the pre-disease timeline, providing a framework for risk prediction and early intervention in CD. Conflict of interest: Chen, Rirong: No conflict of interest Turpin, Williams: No conflict of interest Bharali, Biju: No conflict of interest Patralia, Francesca: No conflict of interest K. Porter, Chad: No conflict of interest W. Lopes, Emily: No conflict of interest Ungaro, Ryan: No conflict of interest Khalili, Hamed: No conflict of interest Colombel, Jean-Frédéric: No conflict of interest Peter, Inga: No conflict of interest Croitoru, Kenneth: No conflict of interest Lee, Sun-Ho: No conflict of interest
Chen et al. (Thu,) studied this question.