Functional multi-axis decomposition of the human gut microbiome: an operational definition of eubiosis and dysbiosis, and a clean-reference framework for disease stratification

For two decades, gut microbial eubiosis and dysbiosis have resisted operationalisation because diversity-, taxon- and ratio-based criteria fail cross-population transfer. We present TAGMOS, a pipeline-agnostic substrate-functional decomposition of the human gut consortium built on the terminal disposal of microbial hydrogen. Each metagenome is reduced to one primary thermodynamic Engine axis (the balance of eubiotic versus dysbiotic terminal hydrogen sinks), two co-primary syntrophy aggregator axes, eleven independent host-interface channel axes and a layer of calibrated composite indices, all computed from 151 bottleneck enzyme-commission (EC) entries against cutoffs frozen on a single 6,508-subject Italian real-world-evidence (RWE) cohort. The frozen cutoffs partition each sample into one of four ordinal tiers, T1_EUBIOTIC to T4_DYSBIOTIC, and a multidimensional cell state-space. The Engine axis transfers without recalibration across Japanese, traditional and Sardinian-longevity cohorts and is preserved in 2,000-year-old paleofecal material, while remaining stable across upstream profiling pipelines, establishing cross-population and cross-platform portability. Across an 18,138-metagenome external meta-analysis of 92 curatedMetagenomicData studies, the tiers then recover a monotonic disease-enrichment gradient that operationalises the proposed definition. Building on this, we show that conventional case-control microbiome comparisons are systematically diluted because roughly a third of nominal control subjects are themselves dysbiotic; restricting the reference to verified-eubiotic controls, and pairing it with a within-tier specificity test, sharpens disease discrimination where the signal is genuine and exposes it as a generic dysbiosis effect where it is not. Under this framework, colorectal cancer (ten-study replication, within-tier AUC 0.92), type-2 diabetes (clean-reference AUC 0.79) and atherosclerotic cardiovascular disease (a cross-continent concordant composite signature) emerge as robust, disease-specific functional signatures; inflammatory bowel disease yields a replicated dysbiosis signature that does not transfer as a single classifier; irritable bowel syndrome carries no binary signal but resolves into a methanogenesis-defined bowel-habit sub-phenotype; and in melanoma checkpoint immunotherapy, the baseline microbiome does not predict response while the on-treatment functional trajectory flags emerging immune-related toxicity, with sex dimorphism recurring across several of these phenotypes. Conventional alpha- and beta-diversity metrics underperform the decomposition at individual-level stratification, supporting the substrate-functional partition as a primary classification layer with direct clinical actionability.