Abstract / Summary
Background: Metabolically healthy obesity and metabolically unhealthy obesity may occupy different positions along a continuum of adipose-tissue dysfunction, but stable molecular distinctions remain uncertain. Methods: We integrated three public human subcutaneous-adipose-tissue transcriptomic datasets while preserving participant and platform independence. GSE244118 was used for discovery, GSE152991 was restricted to 14 participants not overlapping with the discovery cohort, and GSE55200 provided cross-platform supportive evidence. We performed differential-expression analysis, correlation-adjusted Reactome testing, direction-aware cross-dataset integration, discovery-defined module scoring, and adipose-tissue deconvolution. Results: Fifty genes met conventional thresholds for metabolically unhealthy versus metabolically healthy obesity in the discovery dataset, but none passed the prespecified fold-change-aware test or strict cross-dataset replication. Among 1570 shared Reactome pathways, 409 reached false-discovery rate < 0.05 in signed-Z integration and 387 had concordant directions across all datasets. Metabolically unhealthy obesity consistently showed lower mitochondrial translation, respiratory-chain, and energy-metabolism programs, accompanied by secondary increases in extracellular-matrix remodeling. A discovery-defined mitochondrial module was lower in metabolically unhealthy obesity in synthesis restricted to non-discovery datasets under both rank and z-mean scoring. The 14-person holdout alone did not establish a significant module difference, and the extracellular-matrix module synthesis was nonsignificant. Ordered cross-sectional patterns and composition sensitivity described tissue-level associations. Conclusions: Metabolic obesity heterogeneity was therefore more reproducible at pathway and module levels than at the single-gene level. Unresolved clinical confounding and bulk tissue composition limit functional and causal interpretation.