The study addresses a stubborn blind spot in cardiometabolic research. Cholesterol fractions, particularly low-density lipoprotein cholesterol, have long dominated dyslipidemia science because of their established role in atherosclerosis, while triglycerides received comparatively little attention despite accumulating evidence that elevated levels mark subclinical atherosclerosis and vascular inflammation even in people with normal cholesterol, and that hypertriglyceridemia independently predicts adverse cardiovascular events. Current guidelines reserve triglyceride-lowering drugs for high-risk patients whose concentrations exceed 200 mg/dL, reflecting persistent uncertainty about the disorder’s pathophysiology — an uncertainty compounded by the inherent biological variability of triglyceride levels, which fluctuate by 23 to 40 percent. Meanwhile, dyslipidemia has become a substantial public health problem in China, where national studies reported a prevalence of 35.6 percent among adults as of 2018, with more than half of the population showing suboptimal lipid parameters. Although the gut microbiota’s influence on cholesterol metabolism has been extensively characterized, its specific role in systemic triglyceride homeostasis, particularly in non-Western populations, has remained largely undefined.

Between enrollment and final analysis, the investigators applied unusually stringent filters. They initially recruited 205 adults aged 18 or older, then excluded anyone unable to give written informed consent; anyone pregnant or lactating; anyone with inflammatory conditions such as irritable bowel syndrome or rheumatoid arthritis, active malignancy, or previous major gastrointestinal surgery; anyone with acute infections or severe cardiopulmonary, neurological or psychiatric disorders; anyone who had taken probiotics or prebiotics within six weeks or antimicrobials within six months; and anyone using medications capable of reshaping the microbiome or metabolic profiles, including lipid-lowering agents, immunosuppressants, acid suppressants and glucose regulators. The remaining 50 participants — 21 men and 29 women — were classified under the Chinese Guidelines for Lipid Management (2023): the hypertriglyceridemia group showed isolated fasting triglycerides of at least 1.7 mmol/L with all other lipid fractions within normal limits, while controls had triglycerides, total cholesterol and LDL cholesterol all within guideline-defined physiological ranges. All participants provided written informed consent under a protocol approved by the hospital’s ethics committee.

Each participant provided a fresh stool sample that reached the hospital laboratory within six hours and was stored at minus 80 degrees Celsius before being shipped on dry ice to a sequencing facility in Beijing. The team amplified the V3–V4 hypervariable regions of the bacterial 16S rRNA gene for broad taxonomic profiling and, in parallel, performed shotgun metagenomic sequencing on the same samples to capture high-resolution functional potential. Metagenomic reads were quality-filtered, stripped of human DNA, assembled, and compiled into a nonredundant gene catalog whose predicted genes were functionally annotated against the Kyoto Encyclopedia of Genes and Genomes. To distinguish patients from controls, the researchers trained two gradient-boosting classifiers — LightGBM, which grows trees leafwise, and XGBoost, which grows them depthwise with built-in regularization to guard against overfitting in high-dimensional microbial data — tuning hyperparameters through cross-validation. SHapley Additive exPlanations values quantified each taxon’s contribution to model decisions, while linear discriminant analysis effect size (LEfSe), Spearman rank correlations and principal component analysis provided complementary statistical perspectives.

The two groups were metabolically distinct before sequencing began. Compared with controls, hypertriglyceridemia participants had significantly greater body weight, body mass index, waist circumference and waist-to-hip ratio, while other measured parameters were comparable across groups, suggesting the cohorts diverged primarily in lipid status rather than demographic factors. Sequencing quality checks — species accumulation curves, rarefaction analysis and rank-abundance curves — confirmed adequate depth and broadly similar richness and evenness between groups. Taxonomic classification yielded 1,247 operational taxonomic units, with controls harboring more unique OTUs than patients. Yet standard alpha-diversity metrics, including the Shannon, Chao1, Simpson and Ace indices, showed no significant intergroup differences, and none of these diversity measures correlated with serum triglyceride concentrations.

The absence of alpha-diversity differences is itself informative. In microbiome research, alpha diversity reflects the number of species present and how evenly individuals are distributed among them, and reduced diversity is frequently reported in metabolic diseases ranging from obesity to type 2 diabetes. Its preservation here suggests that hypertriglyceridemia, at least in this isolated form, may reshape the relative abundances of specific taxa rather than collapsing the overall architecture of the community. That pattern matters for interpretation: it implies the dyslipidemic signal resides in the identities and functional capacities of particular organisms, not in a generalized degradation of the ecosystem, and it helps explain why differential-abundance and machine learning approaches — which are sensitive to compositional shifts — succeeded where diversity indices did not.

At the taxon level, the findings were strikingly consistent. Faecalibacterium, one of the most abundant genera in the healthy human gut and a major producer of the short-chain fatty acid butyrate, was enriched in the hypertriglyceridemia group and tracked positively with serum triglyceride concentrations. Bacteroides coprocola, a less-studied member of the Bacteroides genus, showed the same positive relationship. Moving in the opposite direction, Bifidobacterium pseudocatenulatum and Lactobacillus salivarius — both genera with long histories of association with metabolic health and widespread use as probiotics — were depleted in patients and inversely correlated with triglyceride levels. The directionality of these associations is biologically plausible: butyrate-producing organisms influence host energy harvest and lipid handling through multiple pathways, while Bifidobacterium and Lactobacillus species have been shown in experimental systems to modulate bile acid pools and reduce intestinal lipid absorption. The authors are careful, however, to frame these as correlations; cross-sectional data cannot establish whether the microbial shifts drive the lipid disturbance, respond to it, or reflect shared upstream factors such as diet.

The functional layer of the analysis added a dimension that 16S sequencing alone could not provide. Shotgun metagenomics of the same samples revealed that microbial genes involved in starch and sucrose metabolism were significantly upregulated in the hypertriglyceridemia cohort. This finding dovetails with the well-established link between refined carbohydrate intake and elevated triglycerides: excess dietary carbohydrate that cannot be oxidized or stored as glycogen is converted in the liver to fat, a process that raises circulating triglycerides. A gut community primed for more efficient carbohydrate degradation could, in principle, increase the caloric yield extracted from the diet and thereby contribute to the substrate burden the liver must process. The KEGG-based annotation allowed the team to move beyond a shopping list of species toward a hypothesis about what the altered community is actually doing — a shift in emphasis that mirrors the broader maturation of microbiome science from taxonomy toward function.

The machine learning component represented the study’s most methodologically ambitious element. Rather than relying on a single classifier, the team trained LightGBM and XGBoost independently and looked for taxa that both algorithms flagged as important. This ensemble strategy is a safeguard against the idiosyncrasies of individual models: gradient-boosting methods can latch onto different features depending on their tree-growing strategy and regularization settings, so features that survive both approaches carry more credibility. The two algorithms converged on five consensus biomarkers, and three of these — consistent with the correlation and differential-abundance analyses — were independently validated by conventional statistical testing. SHapley Additive exPlanations values, borrowed from game theory, allowed the researchers to attribute each prediction to specific taxa in a mathematically principled way, offering a transparent account of why the models classified individuals as they did. The result is a preliminary, five-marker microbial signature for isolated hypertriglyceridemia — exploratory in nature, but grounded in convergent evidence rather than a single statistical lens.

The study’s design choices deserve emphasis because they address many of the confounders that have muddied earlier microbiome-disease associations. By requiring isolated hypertriglyceridemia — with cholesterol fractions within normal limits — the investigators separated triglyceride metabolism from the cholesterol-centric pathology that usually accompanies dyslipidemia, sharpening the specificity of any microbial signal. By excluding probiotic, prebiotic, antimicrobial, lipid-lowering, immunosuppressive, acid-suppressive and glucose-regulating medications within defined windows, they reduced the risk that pharmaceutical agents, rather than the lipid disturbance itself, explained the microbial differences. The six-hour stool processing window and cold-chain handling minimized post-collection artifacts that can distort community profiles. And the restriction to a single hospital population in northwestern China, while limiting generalizability, provided a relatively homogeneous environmental and dietary context in which to detect associations.

Several limitations nonetheless temper the conclusions. The cross-sectional design captures a snapshot and cannot disentangle cause from effect. The sample size of 50, though respectable for a metagenomic study with stringent exclusion criteria, is modest by the standards of machine learning, and the five-biomarker signature will require validation in independent, ideally prospective cohorts before it can be considered robust. Residual confounding by diet, physical activity and body composition remains possible despite the matching of demographic parameters, particularly given the significant differences in adiposity measures between groups. The authors themselves frame the biomarker panel as exploratory, a hypothesis-generating step rather than a diagnostic tool.

Even so, the work fills a genuine gap. Most microbiome-dyslipidemia studies have concentrated on Western cohorts and on mixed lipid phenotypes, leaving the triglyceride-specific axis underexplored in Asian populations whose diets, genetics and microbial compositions differ in ways that can alter disease associations. By demonstrating that isolated hypertriglyceridemia carries a reproducible taxonomic and functional fingerprint — enrichment of Faecalibacterium and Bacteroides coprocola, depletion of two probiotic-associated species, and upregulated carbohydrate metabolism — and by showing that machine learning can detect that fingerprint, the Honghui Hospital team has laid groundwork for two lines of future inquiry. The first is mechanistic: animal models and longitudinal human studies could test whether the enriched or depleted taxa causally modulate triglyceride levels, potentially through short-chain fatty acid production, bile acid transformation or carbohydrate harvest. The second is translational: if the signature holds up under validation, microbial markers might eventually complement lipid panels in identifying individuals at risk, or guide microbiome-targeted interventions — from dietary modification to defined probiotic formulations — aimed at the roughly one in three Chinese adults whose lipid profiles now fall outside healthy ranges.

This study, published in Nature Microbiology, pivots on the discovery of deoxycholic acid 3-sulfate (DCA-3S), a sulfated bile acid variant whose levels correlate strongly with the severity and progression of sepsis in paediatric patients. Using state-of-the-art bile acid-targeted metabolomics combined with deep gut metagenomic sequencing, the researchers identified DCA-3S as a robust biomarker associated with sepsis stages, suggesting it could serve both diagnostic and prognostic purposes.

One of the most surprising revelations of this investigation is the identification of Enterococcus raffinosus, a commensal bacterial species within the human gut flora, as the primary producer of DCA-3S. This discovery challenges the long-established dogma that bile acid sulfation—a detoxification process—occurs exclusively in the liver via hepatocytes. Instead, the research highlights a previously unrecognized microbial pathway of bile acid modification occurring directly in the gut lumen, contributing over 80% of DCA-3S production.

Sepsis, characterized by an overwhelming immune response to infection leading to organ dysfunction, has long been associated with disruptions in the gut microbiota and bile acid profiles. However, the causative links and therapeutic targets have remained elusive. Through rigorous in vitro culture experiments and in vivo mouse models, the study rigorously demonstrated that E. raffinosus synthesizes DCA-3S via sulfotransferase enzymes encoded within its genome—enzymes traditionally thought to be exclusive to liver tissues in mammalian hosts.

Further functional assays employing murine models of sepsis showed that administration of purified or synthetic DCA-3S markedly improved survival rates. Mechanistically, DCA-3S enhances intestinal barrier integrity by upregulating tight junction proteins and reducing epithelial permeability, which is critical in preventing bacterial translocation and systemic inflammation—hallmarks of sepsis pathology.

Interestingly, beyond fortifying epithelial barriers, DCA-3S also exhibited potent anti-inflammatory effects. Its presence dampened systemic cytokine storms by modulating key signaling pathways in immune cells, including NF-κB and MAP kinase pathways, which are central to inflammatory amplification in sepsis. This dual action—barrier protection coupled with immunomodulation—positions DCA-3S as a multifaceted therapeutic agent.

The translational potential of these findings is profound. Current sepsis management relies heavily on broad-spectrum antibiotics and supportive care, often with limited specificity and significant side effects such as microbiome dysbiosis. The prospect of harnessing a microbial metabolite to restore gut homeostasis and mitigate inflammatory injury opens new avenues for precision medicine in critical care settings, especially for vulnerable paediatric populations.

Complementing animal studies, experiments utilizing human intestinal organoids recapitulated DCA-3S’s beneficial effects in vitro, underscoring its potential applicability in human therapeutics. These organoid models confirmed enhancement of epithelial barrier function and attenuated inflammatory signaling upon DCA-3S treatment, validating its role as a bioactive metabolite directly influencing gut mucosal health.

The researchers also explored the dynamics of E. raffinosus colonization in sepsis patients, discovering that abundances of this bacterium—and consequently levels of DCA-3S—were inversely correlated with disease severity. This suggests a protective microbiota signature that could be harnessed prognostically or through microbiota-targeted therapies such as probiotics or fecal microbiota transplantation.

Moreover, the sulfation process mediated by E. raffinosus represents a paradigm shift in bile acid biology, expanding the concept of microbial co-metabolism in the gut-liver axis. It highlights microbes as active contributors, not just passive recipients, in host xenobiotic metabolism. This insight may prompt a reevaluation of bile acid biochemistry, with implications extending beyond sepsis to other inflammatory and metabolic disorders influenced by bile acid signaling.

This study exemplifies the power of integrative omics approaches, combining metabolomic profiling with high-resolution metagenomics to unravel complex host-microbe interactions. It underscores the importance of microbial functional capacity—not merely microbial presence—in dictating health outcomes, thereby refining our approach to microbiome research from descriptive to mechanistic.

The discovery of DCA-3S and its microbial origin adds a new dimension to sepsis biology, bridging gaps in our knowledge about how gut microbial metabolites modulate systemic immunity and organ function. It invites future exploration into microbial enzymes mediating bile acid transformations and their regulation by diet, antibiotics, and other environmental factors.

Future clinical translation will require careful assessments of dosing, safety, and delivery mechanisms for DCA-3S-based therapies. Moreover, longitudinal studies in diverse populations are needed to consolidate the role of E. raffinosus and DCA-3S as biomarkers, potentially leading to tailored microbiome-informed interventions that improve patient outcomes.

This innovative research not only enriches our understanding of gut microbiota’s contributions to sepsis but also epitomizes the emergent field of microbial metabolite therapeutics. It paves the way for exploiting naturally occurring molecules synthesized by commensal bacteria to modulate human disease, heralding a new era where microbiome-derived compounds transition from bench to bedside.

As paediatric sepsis continues to challenge healthcare systems globally, discoveries like DCA-3S offer hope for more effective and less invasive treatments. By elucidating how a single bacterial species can profoundly influence host bile acid metabolism and immune responses, this work elevates the human microbiome from a mysterious entity to a tangible partner in combating critical illnesses.

In summary, the identification of Enterococcus raffinosus as the microbial architect of the sulfated bile acid DCA-3S ushers in a novel conceptual framework for sepsis diagnosis and treatment. This microbial metabolite emerges as a promising biomarker and therapeutic agent, capable of restoring gut barrier function and modulating the immune landscape in pediatric sepsis—a breakthrough that could reshape future strategies to tackle this devastating condition.

Subject of Research: The role of microbial bile acid sulfation in paediatric sepsis progression and treatment

Article Title: Sulfated bile acid produced by a human gut commensal alleviates paediatric sepsis in mice

Article References:
Liu, X., Zhang, H., Wang, YZ. et al. Sulfated bile acid produced by a human gut commensal alleviates paediatric sepsis in mice. Nat Microbiol (2026). https://doi.org/10.1038/s41564-026-02351-1

Image Credits: AI Generated

DOI: https://doi.org/10.1038/s41564-026-02351-1