Host autophagy-related transcriptional signatures in tuberculosis-diabetes comorbidity: an exploratory multi-dataset transcriptomic analysis

Tuberculosis-diabetes comorbidity is associated with altered host immunity and poorer clinical outcomes. Autophagy is relevant to intracellular mycobacterial control, but the transcriptional organisation of autophagy-related genes in tuberculosis-diabetes comorbidity remains incompletely defined. Public transcriptomic datasets related to tuberculosis-diabetes comorbidity were analysed, including the primary discovery set GSE114192 (TB-DM = 61, TB-only = 46, DM-only = 52, healthy controls = 36); drug-resistant-TB and lesion-tissue datasets were included only as complementary, cross-context resources rather than as a directly matched comorbidity or validation cohort. Autophagy-related genes were compiled from the Gene Ontology autophagy term and its descendants. A total of 310 autophagy-related genes were differentially expressed in the tuberculosis-diabetes versus healthy-control comparison (175 up-regulated, 135 down-regulated); 253 and 139 were altered in the TB-only and DM-only comparisons, and 78 genes were shared by all three. Protein interaction analysis of the prioritised genes identified a connected network (196 interactions among 60 genes) with MTOR, GABARAPL2, BCL2, RPTOR, ATG7 and ATG3 as the most central nodes, and pathway enrichment was dominated by canonical autophagy, mitophagy, lysosome and mTOR-related terms. A logistic model built on autophagy genes selected under nested cross-validation yielded an area under the receiver operating characteristic curve of 0.974 (95% CI 0.946–0.993). Direct comparisons and a formal TB x DM interaction test showed that none of the prioritised genes distinguished TB-DM from TB-only and that no gene showed a significant interaction after multiple-testing correction, indicating that the transcriptional changes are predominantly associated with tuberculosis status. The model showed no discriminative transfer when applied to an independent pulmonary-lesion dataset (AUC 0.558), confirming that it should not be regarded as validated across tissues. The findings provide a hypothesis-generating framework for host-directed investigations in diabetes-associated tuberculosis, and identify a coherent autophagy-related transcriptional programme that requires orthogonal experimental validation.