Clinical Proteomics: Crohn's Disease Risk 16 Years Early
- Jul 29
- 4 min read
Crohn's disease is usually caught after the bowel is already scarred, and no blood test reliably tells you who will develop it years ahead of symptoms. A clinical proteomics study in Nature Communications took aim at that gap, measuring 2,736 plasma proteins in 39,634 UK Biobank adults and tracking them for a median of 13.6 years. The question was blunt: can a protein pattern from one blood draw flag future Crohn's disease long before a gastroenterologist is ever involved?
Key Takeaways
A nine-protein plasma signature predicted Crohn's disease up to 16 years before diagnosis, reaching an AUC of 0.76 in an independent UK Biobank testing cohort.
Clinical proteomics can sort apparently healthy adults by future Crohn's risk; people in the top-risk group were 4.23 times more likely to be diagnosed.
The signature held across a separate European cohort and a small Chinese cohort, which argues the signal is not a quirk of one population.
Proteins beat clinical risk factors alone, yet the panel still needs prospective testing before it belongs in any screening decision.
How Clinical Proteomics Flagged Early Crohn's Risk
The measurements came from the UK Biobank Pharma Proteomics Project, an affinity-based panel rather than a mass spectrometry readout. Analysis paired Cox proportional-hazard models with machine-learning feature ranking. Three algorithms, LightGBM, XGBoost, and random forest, scored which of the 2,736 measured proteins carried the most predictive weight. That ranking trimmed the field to a compact nine-protein set, small enough to picture as a future targeted assay. Training happened in one slice of UK Biobank; every number below comes from cohorts the model never saw while it was being fit.
Key Findings
Forty-four proteins moved with future disease. Out of 2,736 measured, 44 plasma proteins were associated with incident Crohn's disease after adjustment.
A nine-protein model did the heavy lifting. CD274, CHI3L1, REG1B, ITGAV, PRSS8, ITGA11, GDF15, DEFA1/DEFA1B, and IL6 together reached AUC 0.76 in a geographically distinct UK Biobank cohort (n = 13,262).
External validation gave AUC 0.73 in EPIC-Norfolk (n = 2,944) and 0.79 in a cross-sectional Southern China cohort (n = 74).
Adding clinical variables pushed accuracy to AUC 0.78 up to 16 years before diagnosis, and high-risk individuals faced a 4.23-fold higher hazard.
Figure 1. Study design. Data from 52,896 UK Biobank participants (median 13.6-year follow-up) supplied 2,736 plasma proteins, 41 clinical predictors, and a Crohn's polygenic risk score. Cox models and machine-learning feature ranking selected the protein set, which was then tested in a separate UK Biobank cohort, the EPIC-Norfolk study, and a Southern China cohort. Adapted from Feng et al. (2025), Nature Communications.
What the Signature Points To
Several of the nine proteins read like a map of low-grade gut inflammation and tissue remodeling. GDF15 and IL6 track systemic inflammatory tone, CHI3L1 climbs with active intestinal injury, and REG1B ties to epithelial regeneration. The signal is biologically coherent, then, not a black-box correlation, which matters when you want a clinician or a regulator to trust it. Whether these proteins drive disease or simply mark its slow approach is a question the study leaves open.
Reading an Affinity Panel With Care
The platform reports relative protein abundance across a fixed panel, so it samples the plasma proteome differently than a discovery mass spectrometry run would, and absolute concentrations are not on offer. The Southern China cohort is tiny at 74 people, so the 0.79 figure there reads as encouraging rather than settled. Building on the broader UK Biobank proteomic effort from Sun and colleagues (2023), this work shows affinity panels can support proteome-wide risk prediction, though replication in prospective, multi-ethnic cohorts is the obvious next step.
Implications for Biomarker Programs
For biomarker teams, the durable lesson is that a predictive panel only earns trust once it survives a geographically separate cohort, which is exactly where this nine-protein model held its ground. We see the same thing in client work at Dalton, where a signature that looks clean in discovery often wobbles until it is re-measured on an orthogonal platform, one reason our MS-based proteomics services pair discovery panels with targeted confirmation. Groups planning that kind of program can start from our guide to multi-omics biomarker discovery before they commit a cohort.
Frequently Asked Questions
What is clinical proteomics? Clinical proteomics is the large-scale measurement of proteins in patient samples, usually blood, to find markers of disease risk, diagnosis, or treatment response. It spans affinity panels and mass spectrometry workflows. The aim is to turn protein patterns into decisions a clinician or drug developer can act on.
Can clinical proteomics predict Crohn's disease before symptoms appear? In this UK Biobank study, a nine-protein plasma signature separated future Crohn's patients from controls up to 16 years before diagnosis, with an AUC of 0.76 in an independent cohort. That is prediction, not proof of prevention. The panel still needs prospective testing before it guides screening.
What samples do plasma proteomics biomarker studies need? Most run on a single blood draw processed to plasma, then measured on an affinity panel or by mass spectrometry. Cohort size matters more than sample volume; the reliable models here drew on tens of thousands of participants with long follow-up. External validation cohorts are what separate a real signature from an overfit one.
Conclusion
A single blood draw carrying a nine-protein risk signal for Crohn's disease is now a believable proposition, at least in large European cohorts. What isn't settled is whether acting on that signal changes anything, since nobody has run the prospective screen-and-intervene trial the model invites. For now the practical value sits in enriching trials and surveillance cohorts with people genuinely likely to convert.
Related Reading
See how we run these analyses in one lab: Dalton's multi-omics CRO services.
Citation
Feng, J., Chen, S., Li, Q., Long, Y., Ma, Y., Zhang, L., Zeng, R., Luo, D., Meng, M., Yu, S., Chen, C., Wu, Y., Huang, W., Zhang, H., Li, L., Leung, F. W., Duan, C., Sha, W., & Chen, H. (2025). Plasma proteomic profiles identify biomarkers predicting Crohn's disease up to 16 years before onset. Nature Communications, 16(1), 11481. https://doi.org/10.1038/s41467-025-66483-4
Note
This blog post summarizes findings from the above-cited research. Figures are adapted from the original publication. For full details, please refer to the source article.
By Seungjun Yeo, CEO at Dalton Bioanalytics. Specializing in multi-omics mass spectrometry for drug discovery and biomarker research.