Longitudinal ecological disruption and metabolite associations of the gut microbiome during acute myeloid leukemia therapy

Acute myeloid leukemia therapy is accompanied by antibiotic exposure and marked gut microbiome disruption. We analyzed 508 longitudinal stool 16S rRNA gene profiles from 67 patients in PRJNA451154. We summarized the depletion of putative short-chain fatty acid (SCFA)-associated genera, pathobiont expansion, and domination with an ecological composite score (ECS). Disruption peaked descriptively during days 15-28 (D15-D28). In a patient-clustered generalized estimating equation model jointly including treatment window and antibiotic exposure during the preceding 0-14 days, the D15-D28 ECS contrast was attenuated (coefficient, 0.251; 95% confidence interval [CI], -0.372 to 0.874), whereas recent antibiotic exposure remained associated with higher ECS (0.722 per 7 antibiotic-days; 95% CI, 0.529 to 0.915). The D1-D7 to D15-D28 increase persisted with patient-level fivefold cross-fitted score construction (paired difference, 1.279; bootstrap 95% CI, 0.861 to 1.758). Treatment-window-adjusted models retained associations of serum butyrate/isobutyrate with the putative SCFA-associated genus signal (0.476; 95% CI, 0.309 to 0.643) and fixed ECS (-0.372; 95% CI, -0.516 to -0.227). ECS did not add stable information beyond Shannon loss and maximum pathobiont abundance for clinical events during the subsequent 7 or 14 days. Frozen ecological formulas produced similar rankings across two external data sets but did not provide clinical validation. Thus, D15-D28 was a descriptive window strongly coupled to recent antibiotic exposure, and ECS functioned as an integrative ecological summary rather than a validated predictor.IMPORTANCEAcute myeloid leukemia therapy creates a clinically important perturbation of gut bacterial communities, yet microbiome injury is often summarized by diversity alone or by isolated taxa. We integrated the depletion of genera associated with short-chain fatty acid production and expansion of potential pathogens into a longitudinal ecological score. The apparent disruption peak during days 15-28 was largely attenuated after accounting for recent antibiotic exposure, whereas patient-level resampling supported the stability of the ecological contrast. Associations with serum metabolites persisted after adjustment for treatment period, but amplicon sequencing cannot establish microbial function or metabolite production. The composite score did not add stable information beyond diversity loss and maximum potential-pathogen abundance for near-term clinical events. These findings define a reproducible framework and a high-yield sampling window for prospective studies while avoiding causal or predictive claims.