Gut bacteriophage diversity is inversely associated with organ dysfunction severity in critically ill patients: a two-cohort metagenomic study

The gut microbiota is profoundly disrupted during critical illness, yet the bacteriophage component has received almost no systematic study. Bacteriophages actively shape bacterial community composition through lytic and lysogenic interactions, but whether gut phage diversity tracks organ dysfunction severity in critically ill patients—and whether any such relationship is reproducible—is unknown. We analysed shotgun metagenomes from two independent ICU cohorts: a discovery cohort of 196 medical ICU patients (BioProject PRJNA1134172 Chicago Cohort, US) and an external exploratory replication cohort of 22 ARDS patients from a geographically distinct centre Qingdao cohort, China. A catalogue of 1,815 non-redundant viral operational taxonomic units (vOTUs) was constructed using geNomad and CheckV and clustered at 95% average nucleotide identity over 85% alignment fraction. Phage alpha diversity was correlated with the Sequential Organ Failure Assessment score by Spearman rank correlation, with partial correlation controlling for sequencing depth, comorbidity, sampling time and age. Bacterial hosts were predicted with iPHoP; phage–bacteria co-occurrence was assessed with FastSpar. To test whether the diversity–SOFA association was robust to catalogue choice, a cohort-specific de novo viral catalogue (644 vOTUs) was additionally constructed from the replication cohort. Phage richness and Shannon diversity were each inversely correlated with SOFA score in the discovery cohort (richness: r = − 0.308, p = 1.1 × 10⁻ 5 ; Shannon: r = − 0.259, p = 2.5 × 10⁻ 4 ) and remained significant after full adjustment (richness partial r = − 0.255; Shannon partial r = − 0.224). The association was directionally consistent in the external exploratory replication cohort, where both alpha-diversity metrics showed significant inverse associations (richness: r = − 0.442, p = 0.040; Shannon: r = − 0.430, p = 0.046). Co-occurrence analysis identified thirteen high-confidence phage–bacteria pairs (r = 0.339–0.636) targeting Enterococcus faecium and Lacticaseibacillus rhamnosus ; the quadrant of high phage detection with undetected host was entirely empty for both hosts, while the co-detection quadrant showed the highest illness severity. Individual vOTUs were largely cohort-specific: none of the thirteen focal pairs was detected in more than one replication-cohort sample, only 12.1% (78/644) of replication cohort de novo vOTUs matched the discovery catalogue, and median catalogue mapping rates were 3.06% versus 0.44% across cohorts. In contrast, community-level diversity estimates were robust to catalogue choice (per-sample richness r = 0.82, Shannon r = 0.77 between catalogues). Gut bacteriophage diversity is cross-sectionally associated with organ dysfunction severity in critically ill patients, and this association was directionally consistent across two independent cohorts. These exploratory, association-based findings are hypothesis-generating and motivate confirmation in larger prospective cohorts with dedicated virome sequencing.