Integrative transcriptome-based drug repurposing in tuberculosis

Abstract

Tuberculosis (TB) remains the second leading cause of infectious disease mortality worldwide, killing over one million people annually. Rising antibiotic resistance has created an urgent need for host-directed therapeutics (HDTs) - preferably by repurposing existing approved drugs - that modulate host immune responses rather than directly targeting the pathogen. Repurposed therapeutics have been successfully identified for cancer and COVID-19 by finding drugs that reverse disease gene expression patterns (an approach called 'connectivity scoring'), but this approach remains largely unexplored for bacterial infections like TB. The application of transcriptome-based methods to TB faces significant challenges, including dataset heterogeneity across transcriptomics platforms and biological conditions, uncertainty about optimal scoring methods, and lack of systematic approaches to identify robust disease signatures. Here, we developed an integrative computational workflow combining multiple connectivity scoring methods with consensus disease signature construction and used it to systematically identify FDA-approved drugs as promising TB host-directed therapeutics. Our framework 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. Our approach prioritized 140 high-confidence drug candidates that consistently reverse TB-associated gene expression changes, successfully recovering known HDTs, including statins (atorvastatin, lovastatin, fluvastatin) and vitamin D receptor agonists (calcitriol). We identified promising novel candidates such as niclosamide and tamoxifen, both recently validated in experimental TB models, and revealed enrichment for therapeutically relevant mechanisms, e.g., cholesterol metabolism inhibition and immune modulation pathways. Network analysis of disease-drug interactions identified 10 key bridging genes (including MYD88, RELA, and CXCR2) that represent potential novel druggable targets for TB host-directed therapy. This work establishes transcriptome-based connectivity mapping as a viable approach for systematic HDT discovery in bacterial infections and provides a robust computational framework applicable to other infectious diseases. Our findings offer immediate opportunities for experimental validation of prioritized drug candidates and mechanistic investigation of identified druggable targets in TB pathogenesis.

Figure 1.. Overview of disease signature datasets and the integrative drug repurposing workflow.

(a) Sankey diagram summarizing the biological and technical heterogeneity across the TB transcriptomic signatures. These signatures were constructed from multiple TB datasets spanning diverse biological sources, including different tissue and cell types, sample origins (circulating blood vs. lung), sample types (primary vs. cell line), TB types (PTB: pulmonary TB vs. MTB: active TB cases without explicit “pulmonary” specification), and profiling technologies (microarray and RNA-seq). (b) Overview of our integrative drug repurposing workflow implementing (i) individual and (ii) aggregated signatures compared against drug signatures using six connectivity scoring methods: CMAP 1.0 and LINCS scores (WCS, NCS, Tau) for enrichment-based, and Extreme Pearson (XCor) and Spearman (XSpe) for correlation-based approaches. (c) Drug prioritization pipeline applied to microarray and RNA-seq signatures, including both individual and aggregated versions. Drugs reversing ≥50% of the disease signature with scores in the top 10% most negative values were selected, and only those supported by ≥2 of 3 metric subgroups (CMAP 1.0, LINCS, correlation-based) were retained. Final drug rankings were generated using Bayesian Latent Variable Approach for partial rank aggregation; “BiG” method [44], and high-confidence predictions were obtained by intersecting results across signature types.

Figure 2.. Aggregated signatures capture highly variable pathways across individual signatures.

(a) Heatmap showing highly variable GO biological processes (GO:BP) across the individual TB transcriptomic signatures that are enriched in the aggregated upregulated TB signature. Many of these pathways are consistently enriched across studies, reflecting host metabolic adaptation and mitochondrial stress during TB infection. (b) Heatmap showing highly variable GO:BP terms across individual TB signatures that are enriched in the aggregated downregulated TB signature. These pathways include key immune-related processes such as T-helper cell differentiation, CD4-positive T-cell activation, and MAP kinase signaling. Signature annotations (profiling technology, tissue/cell type, TB sample annotation (PTB: pulmonary vs. MTB: non-specified), and origin type) are indicated in the annotation bars, highlighting the biological and technical diversity across datasets.

Figure 3.. Disease-drug-target network identifies key linking genes and prioritized drug candidates.

Network visualization showing the shortest-path connections between TB disease genes (purple), known drug targets (pink), and prioritized drugs (red) through intermediate in-path genes (yellow) within the STRING genome-wide host protein-protein interaction network. This analysis highlights 10 key in-path genes—including MYD88, RELA, CXCR2, UBC, GRB2, CBL, TIMP1, TRAF6, IL8, and TP53—that serve as mechanistic bridges between TB-perturbed genes in our aggregated signatures and known drug targets. Prioritized drug candidates (red nodes) connected through these paths include both known TB-relevant HDTs such as simvastatin, fulvestrant, and novel predictions such as fipronil, phenazopyridine, tretinoin, dactinomycin, noscapine, piretanide, and everolimus. These network-level insights support the biological relevance of our drug predictions and suggest new avenues for host-directed TB therapeutics.

Figure 4.. Synergy and antagonism of pairs of mechanisms of action (MOA) associated with our 140 high-confidence drug candidates.

The heatmap shows combined synergy Z-scores for all pairwise combinations of MOAs derived from our predicted TB drug candidates. Synergy scores were computed by integrating four synergy metrics (ZIP, Bliss, Loewe, and HSA) from DrugCombDB and summarized using combined Z-scores (blue = synergy; red = antagonism).