TL;DR
This paper presents a novel computational framework to identify existing FDA-approved drugs for repurposing in ulcerative colitis (UC). By integrating multi-cohort transcriptomic data with large-scale clinical records, they identified atorvastatin as a promising candidate, showing a significant reduction in colectomy risk in UC patients.
What Problem Does This Paper Address?
Ulcerative colitis (UC) is a chronic inflammatory bowel disease affecting nearly a million people in the US, with limited effective long-term treatment options. Many patients don't respond to standard therapies like corticosteroids or anti-TNF drugs, leading to a need for colectomy, a major surgery with significant side effects. Drug repositioning, finding new uses for existing drugs, offers a faster and cheaper path to new treatments. However, previous attempts to identify drugs for UC using transcriptomic data have been limited by relying on single-cohort studies, which fail to capture the real-world heterogeneity of the disease. This lack of robust molecular signatures and validation in clinical data has hindered the translation of computational findings into actual therapies.
Methods
The researchers employed a multi-pronged approach:
- Multi-Cohort Transcriptomic Analysis: They integrated data from 11 independent gene expression datasets (totaling 272 colon biopsy samples) representing diverse patient populations and experimental conditions. This meta-analysis, using the
MetaIntegratorR package, aimed to identify a robust UC disease gene signature that reflects real-world heterogeneity. They used a "leave-one-dataset-out" approach to ensure the signature's robustness. - Drug-Disease Signature Correlation: They compared the identified UC gene signature against the transcriptomic profiles of 781 FDA-approved small-molecule drugs from the Library of Integrated Network-Based Cellular Signatures (LINCS) L1000 platform. This was done by calculating Pearson correlations to find drugs whose gene expression effects were inversely correlated with the UC signature, suggesting they might reverse the disease state.
- Retrospective Clinical Validation: To validate their top drug candidates, they conducted a retrospective cohort study using two large, independent patient datasets: the Stanford Research Repository (STARR) and the Optum Clinformatics DataMart. They modeled this analysis after a "target trial" to minimize biases common in observational studies, comparing colectomy risk in UC patients prescribed atorvastatin versus those on comparator cardiovascular or lipid-modulating drugs.
Datasets
- Gene Expression Datasets: 11 publicly available gene expression datasets from the NCBI Gene Expression Omnibus (GEO) were used, containing a total of 272 colon biopsy samples from UC patients (N=171) and healthy controls (N=101). These datasets were chosen for their biological, clinical, and technical heterogeneity.
- Drug Transcriptomic Data: The LINCS L1000 platform provided transcriptomic profiles for 781 FDA-approved small-molecule drugs.
- Clinical Datasets:
- Stanford Research Repository (STARR): Electronic health records (EHRs) of 1.8 million adult and pediatric patients from Stanford University Medical Center (Jan 1, 2008 - Dec 31, 2015).
- Optum Clinformatics DataMart: A national insurance claims database of 63 million US residents (Jan 1, 2004 - Dec 31, 2016).
Key Insights and Ideas
- The Power of Heterogeneity: Combining multiple, diverse gene expression datasets is crucial for identifying a robust disease signature that generalizes better to real-world patient populations than single-cohort studies.
- Computational Prediction Meets Clinical Validation: The study highlights the critical need to validate computationally predicted drug candidates using real-world clinical data to bridge the gap between in-silico findings and clinical utility.
- Atorvastatin's Anti-inflammatory Potential: Beyond its cholesterol-lowering effects, atorvastatin exhibits significant anti-inflammatory properties that may be beneficial in treating UC, as suggested by its inverse correlation with the UC gene signature and its association with reduced colectomy risk.
- A Framework for Drug Repurposing: The paper proposes a systematic framework for integrating large-scale molecular and clinical data to identify novel drug repurposing opportunities for various diseases.
Key Results
- A robust UC gene signature was identified, comprising 2306 differentially expressed genes.
- Among 781 FDA-approved drugs, atorvastatin showed the highest inverse correlation with the UC gene signature among non-oncolytic therapies.
- In both the STARR (n=827) and Optum (n=7821) cohorts, atorvastatin use was significantly associated with a decreased risk of colectomy (a marker of treatment-refractory disease):
- STARR: Hazard Ratio (HR) = 0.47, P = 0.03
- Optum: HR = 0.66, P = 0.03
- This association remained significant even after adjusting for confounders and using propensity score matching.
- Long-term atorvastatin use (over 6 months) showed a greater protective effect on colectomy risk compared to short-term use.
- Topiramate, previously suggested as a potential UC therapy, did not show a significant association with reduced colectomy risk in this study.
Limitations and Future Work
- Cell Line Limitations: The LINCS database primarily uses immortalized cell lines, which may not fully represent the complexity of in vivo cellular responses and could bias drug effect predictions.
- Administrative Data Constraints: The use of administrative data (Optum) limits the ability to account for factors like disease phenotype, smoking history, or non-prescription drug use.
- Colectomy as a Marker: While colectomy is a significant outcome, it may not perfectly capture all aspects of disease severity.
- Future Work: The authors emphasize the need for prospective, controlled clinical trials to confirm the therapeutic benefits of atorvastatin in patients with UC. They also suggest further investigation into the specific mechanisms by which atorvastatin exerts its anti-inflammatory effects in UC, potentially by exploring its impact on pathways like TNF, CXCL10, and ICAM1. The developed framework can be applied to identify drug repurposing candidates for other diseases, particularly those with limited treatment options.
Summary updated