Cross-study reproducibility of skeletal-muscle transcriptomic remodelling differs between acute and late septic critical illness: implications for a brain-muscle-axis framework

Background Intensive care unit-acquired weakness (ICU-AW) is a heterogeneous syndrome in which critical illness polyneuropathy, critical illness myopathy, impaired central motor drive, neuromuscular transmission failure and skeletal-muscle remodelling may coexist. A brain-muscle-axis framework is useful for organising these interacting levels, but it requires matched neural, neuromuscular and functional measurements. Public transcriptomic datasets can define a reproducible muscle-side molecular layer for that future work. Whether the same muscle signal is reproducible at distinct stages of critical illness recovery remains uncertain. Methods We performed an exploratory secondary analysis of de-identified public skeletal-muscle transcriptomic datasets. GSE13205, containing 13 septic intensive-care samples and 8 controls, was analysed as a discovery cohort using an age- and sex-adjusted limma model. A fixed, prespecified discovery ranking was transferred to the external time-resolved GSE78929 cohort, which contained 8 controls, 14 day-7 samples and 10 month-6 samples. Directional concordance, Spearman correlation and an equal-size random-gene-set benchmark based on 100,000 permutations were evaluated separately at day 7 and month 6. A subject-blocked model was fitted for 10 verified day-7/month-6 pairs. The analysis estimated cross-study molecular concordance; it was not designed to estimate a causal brain-muscle effect or clinical diagnostic performance. Results Among 15,367 uniquely mapped genes, 2,583 met a Benjamini-Hochberg false-discovery rate below 0.05 in the adjusted discovery model, and 2,546 also exceeded an absolute log2 fold-change threshold of 0.5. Of 277 mappable genes from the fixed discovery top-300 list, 218 were directionally concordant at day 7 (78.7%; Spearman rho=0.563; empirical P=9.9999×10^-6), whereas 123 were concordant at month 6 (44.4%; rho=-0.096; empirical P=0.9229). Across 13,184 common genes, correlations were 0.165 and -0.083 at day 7 and month 6, respectively. Ten verified pairs were available for the secondary paired model; no gene reached a BH-adjusted P value below 0.05 for the month-6-minus-day-7 contrast. Conclusions Across the available public cohorts, skeletal-muscle transcriptomic remodelling was more reproducible at day 7 than at month 6. This time-specific muscle-side molecular layer can inform future matched EEG, corticospinal, muscle and functional studies within a brain-muscle-axis framework; it does not establish patient-level brain-muscle coupling, causality or diagnostic performance.