Computational repurposing finds FDA drugs for tuberculosis

Tuberculosis remains one of the world’s deadliest infectious diseases, killing over one million people annually and becoming harder to treat as antibiotic resistance rises. Instead of searching only for new antibiotics, researchers are exploring host-directed therapeutics (HDTs) that change how the human immune system responds to infection. Janani Ravi and colleagues adapted a strategy proven in cancer and COVID-19 — using patterns of gene activity to find drugs that reverse disease-related changes — to the much less-explored area of bacterial infection. That strategy, often called connectivity scoring, compares the gene expression fingerprints of disease to the fingerprints created when cells are treated with drugs, seeking drugs that produce the opposite pattern. Applying this idea to tuberculosis is not straightforward: datasets come from different transcriptomics platforms, like microarray and RNA-seq, cover diverse cell types and infection conditions, and researchers have disagreed about how best to score and combine results. To overcome these challenges, Janani Ravi’s team built an integrative computational workflow that combines multiple scoring methods and creates consensus disease signatures, aiming to systematically identify FDA-approved drugs that could be repurposed as TB host-directed therapeutics.

The team developed a transcriptome-based connectivity mapping framework that integrates six complementary connectivity methods and constructs weighted consensus signatures from 21 TB gene expression datasets spanning microarray and RNA-seq platforms, diverse cell types, and infection conditions. By bringing these datasets together and using multiple scoring approaches, the workflow prioritized 140 high-confidence drug candidates that consistently reverse TB-associated gene expression changes. The approach recovered known HDTs, including statins (atorvastatin, lovastatin, fluvastatin) and vitamin D receptor agonists (calcitriol), showing that the method can rediscover drugs already thought useful. It also highlighted promising novel candidates such as niclosamide and tamoxifen, both of which have been recently validated in experimental TB models. Analysis showed enrichment for therapeutically relevant mechanisms, for example cholesterol metabolism inhibition and immune modulation pathways. A network analysis of disease-drug interactions further identified 10 key bridging genes, including MYD88, RELA, and CXCR2, that link disease signatures and drug effects and may represent novel druggable targets for TB host-directed therapy.

This work demonstrates that transcriptome-based approaches can be systematically applied to bacterial infections like tuberculosis and not only to cancer or viral diseases. By combining multiple connectivity scoring methods, creating weighted consensus disease signatures, and drawing on 21 diverse datasets, the framework addresses problems of dataset heterogeneity and scoring uncertainty. The practical result is a ranked list of 140 FDA-approved candidates ready for follow-up: some, like atorvastatin and calcitriol, validate existing hypotheses about host modulation in TB, while others, like niclosamide and tamoxifen, point to newly supported directions. The identification of bridging genes such as MYD88, RELA, and CXCR2 provides concrete targets for mechanistic study and potential combination strategies. Because the drugs are already FDA-approved, the path to experimental validation and clinical testing could be faster than for novel compounds. The authors present a robust computational pipeline that can be applied to other infectious diseases, offering a scalable way to discover host-directed therapies using public gene expression data.