Introduction
Traditional biomarker discovery in ulcerative colitis (UC) often relies on correlations with clinical endpoints, potentially overlooking novel disease endotypes and therapy responses. Here, we introduce an unsupervised, data-driven trajectory framework that leverages longitudinal multi-omics data from patients with UC to uncover early circulating molecular features predictive of disease course in UC.
Aims & Methods
We analyzed transcriptomic (579 samples) and proteomic (386 samples) data from 193 UC patients in the VARSITY trial at baseline, week 14, and week 52. K-nearest neighbors clustering was used to classify transcriptomic profiles into initial disease states, with cluster stability assessed across varying K values and granularities. Patient disease state trajectories over time were reconstructed using a multi-commodity flow algorithm [1]. For each trajectory, we performed differential protein expression and pathway enrichment analysis between baseline and week 14 to identify early serum biomarkers (false discovery rate ≤ 0.05).
Results
Eight distinct, dynamic disease states were identified (Table), each associated with specific clinical outcomes and characteristic biological pathways (e.g., immune regulation, inflammation, tissue remodeling, gut barrier function). Notably, four distinct molecular states were enriched with baseline samples, indicating the existence of molecular endotypes despite comparable clinical indices (Mayo score, Robarts Histopathology Score, Geboes index). Among the 42 patients who transitioned to a different state by week 14, 39 patients followed trajectories of clinical and or histology improvement. Importantly, these early improvement trajectories were associated with significant changes in serum protein levels, including previously reported UC-related proteins such as OSM, MMP10, and CXCL9 [2], as well as 22 novel candidate prognostic biomarkers.
Conclusion
An unsupervised multi-omics trajectory analysis successfully delineated distinct UC patient disease states and their corresponding trajectory-specific serum biomarkers, independent of pre-defined clinical criteria. The identification of state-specific proteins and pathways linked to disease progression and treatment response highlights new candidate serum biomarkers for patient stratification and monitoring in UC. This data-driven approach underscores the value of unbiased methods in biomarker discovery for complex inflammatory diseases.
| Table. Disease State Clusters and Baseline Clinical Characteristics, Mean ± Standard Error |
Disease State Cluster
| Total Mayo Score | Geboes Score | Robarts Histopathology Index | C-Reactive Protein | Fecal Calprotectin |
| 0 | 4 ± 0.34
| 8 ± 0.65 | 9 ± 1.08
| 3.81 ± 11.46
| 640 ± 532.78
|
| 1 | 7.5 ± 0.73
| 9 ± 0.95
| 8 ± 1.29
| 0 ± 3.55
| 325.5 ± 597.26
|
| 2 | 1 ± 0.18
| 3 ± 0.32
| 1 ± 0.46
| 7.62 ± 7.81
| 36 ± 42.01
|
| 3 | 1 ± 0.23
| 3 ± 0.37
| 1 ± 0.54
| 6.67 ± 18.59
| 79 ± 113.95
|
| 4 | 7 ± 0.51
| 16.5 ± 0.96
| 20.5 ± 1.7
| 0 ± 29.89
| 1082 ± 428.04
|
| 5 | 8 ± 0.22 | 15 ± 0.41
| 18 ± 0.74
| 0 ± 7.67
| 1262.5 ± 251.76
|
| 6 | 7 ± 0.29 | 15 ±0.55
| 18 ± 0.91
| 0 ± 6.65
| 1544 ± 395.42
|
| 7 | 8 ± 0.29
| 19 ± 0.65
| 25 ± 1.19
| 0 ± 8.07
| 1268.5 ± 289.78
|
References
1. Hasanaj E, et al. Bioinformatics. 2024;40(Suppl 1):i151–9. doi:10.1093/bioinformatics/btae241
2. Salomon B et al. J Crohns Colitis. 2025;19(5):jjae169. doi:10.1093/ecco-jcc/jjae169
Disclosure
Funding: This study was sponsored by Takeda Development Center Americas, Inc.
Jay Tang, Pratik Panchal, Christian Agboton, Sachin Kumar, Mukesh Bansal, Vinayagam Arunachalam, Olusegun Onabajo, Haakan Wennbo are employees of Takeda Pharmaceuticals and hold stock/options in Takeda Pharmaceutical Company Limited. Clara Domingo Sabugo, Veronica Miro, Pablo Latorre are employees of Clarivate Analytics, which received funding from Takeda related to this work. Stefan Schreiber reports consultancy and personal fees from AbbVie, Arena, BMS, Biogen, Celltrion, Celgene, Dr Falk, Ferring, Fresenius, Gilead, HIKMA, IMAB, Janssen, MSD, Mylan, Pfizer, Protagonist, Provention Bio, Sandoz, Takeda, and Ventyx.
