Background Sepsis is a heterogeneous syndrome in which patients with the same clinical diagnosis may harbour different host immune responses. Existing transcriptomic endotyping frameworks differ in gene selection, clustering, and cohort composition, limiting comparability. Few studies have distinguished endotype-intrinsic prognostic signal from baseline severity, and most prior frameworks relied on curated immune-gene panels or single-cohort discovery. Methods We performed consensus clustering on 1002 sepsis patients from 5 discovery cohorts after quantile normalisation, ComBat batch correction, and data-driven feature selection (122 genes with median absolute deviation <InlineEquation ID="IEq1"><EquationSource Format="TEX">\(\ge 1.0\)</EquationSource></InlineEquation> from a 2456-gene set common to nine platforms), using PAM clustering with Pearson correlation distance (<InlineEquation ID="IEq2"><EquationSource Format="TEX">\(k = 2\)</EquationSource></InlineEquation>–10, 1000 bootstrap iterations) and composite metric-based k selection. Endotype–mortality associations were assessed by logistic regression and, where severity scores were available, by SOFA-adjusted models; a LASSO multinomial classifier evaluated generalisability across 4 external cohorts. Results Three endotypes were resolved at optimal <InlineEquation ID="IEq3"><EquationSource Format="TEX">\(k = 3\)</EquationSource></InlineEquation>: C1 (Immune Activation, <InlineEquation ID="IEq4"><EquationSource Format="TEX">\(n = 306\)</EquationSource></InlineEquation>, 30.5%), C2 (Interferon Response, <InlineEquation ID="IEq5"><EquationSource Format="TEX">\(n = 447\)</EquationSource></InlineEquation>, 44.6%), and C3 (Erythroid Dysregulation, <InlineEquation ID="IEq6"><EquationSource Format="TEX">\(n = 249\)</EquationSource></InlineEquation>, 24.9%). Mortality differed across endotypes (logistic regression likelihood ratio <InlineEquation ID="IEq7"><EquationSource Format="TEX">\(p = 2.99 \times 10^{-4}\)</EquationSource></InlineEquation>), with C2 showing lower 28-day mortality than C1 (15.4% vs. 26.1%; OR <InlineEquation ID="IEq8"><EquationSource Format="TEX">\(= 0.52\)</EquationSource></InlineEquation>, 95% CI <InlineEquation ID="IEq9"><EquationSource Format="TEX">\(0.36\text {--}0.74\)</EquationSource></InlineEquation>, <InlineEquation ID="IEq10"><EquationSource Format="TEX">\(p = 3.33 \times 10^{-4}\)</EquationSource></InlineEquation>). Crucially, SOFA-adjusted analysis in GSE185263 (<InlineEquation ID="IEq11"><EquationSource Format="TEX">\(n = 342\)</EquationSource></InlineEquation>, the only discovery cohort with SOFA scores) fully attenuated the C2 advantage (adjusted OR <InlineEquation ID="IEq12"><EquationSource Format="TEX">\(= 0.99\)</EquationSource></InlineEquation>, <InlineEquation ID="IEq13"><EquationSource Format="TEX">\(p = 0.98\)</EquationSource></InlineEquation>), indicating that in this cohort endotype-mortality associations are primarily explained by baseline severity. A LASSO-regularised multinomial classifier achieved >98% per-class sensitivity and generalised to 4 external cohorts (<InlineEquation ID="IEq14"><EquationSource Format="TEX">\(n = 412\)</EquationSource></InlineEquation>) with 100% gene coverage and mean classification confidence 0.86−0.94. Longitudinal analysis of 129 patients across four multi-timepoint datasets demonstrated 66.7% endotype stability (Cohen’s <InlineEquation ID="IEq15"><EquationSource Format="TEX">\(\kappa = 0.49\)</EquationSource></InlineEquation>). Conclusions Three severity-aligned transcriptomic patterns—immune activation, interferon response, and an underrecognised erythroid/metabolic axis—recur across heterogeneous sepsis cohorts. In the one cohort with organ-dysfunction scores, the C2 mortality advantage was fully attenuated after SOFA adjustment, indicating that—in this cohort—endotype–mortality associations are largely explained by baseline severity rather than independent prognostic biology; whether this generalises requires cohorts with harmonised severity data. Future endotyping studies should incorporate systematic severity adjustment before drawing prognostic or treatment-stratification conclusions.

Multi-cohort consensus clustering identifies three distinct transcriptomic endotypes in sepsis

期刊论文
出版模式：

开放获取
发表日期：

2026年7月17日

查看全文

Naixun Chi,
Siyu Mu,
Xin Jin,
Gangyu Zhang,
Min Gao,
Yurou Guo,
Yuelong Chen,
Yong Qi

摘要

Background

Sepsis is a heterogeneous syndrome in which patients with the same clinical diagnosis may harbour different host immune responses. Existing transcriptomic endotyping frameworks differ in gene selection, clustering, and cohort composition, limiting comparability. Few studies have distinguished endotype-intrinsic prognostic signal from baseline severity, and most prior frameworks relied on curated immune-gene panels or single-cohort discovery.

Methods

We performed consensus clustering on 1002 sepsis patients from 5 discovery cohorts after quantile normalisation, ComBat batch correction, and data-driven feature selection (122 genes with median absolute deviation \(\ge 1.0\) from a 2456-gene set common to nine platforms), using PAM clustering with Pearson correlation distance (\(k = 2\)–10, 1000 bootstrap iterations) and composite metric-based k selection. Endotype–mortality associations were assessed by logistic regression and, where severity scores were available, by SOFA-adjusted models; a LASSO multinomial classifier evaluated generalisability across 4 external cohorts.

Results

Three endotypes were resolved at optimal \(k = 3\): C1 (Immune Activation, \(n = 306\), 30.5%), C2 (Interferon Response, \(n = 447\), 44.6%), and C3 (Erythroid Dysregulation, \(n = 249\), 24.9%). Mortality differed across endotypes (logistic regression likelihood ratio \(p = 2.99 \times 10^{-4}\)), with C2 showing lower 28-day mortality than C1 (15.4% vs. 26.1%; OR \(= 0.52\), 95% CI \(0.36\text {--}0.74\), \(p = 3.33 \times 10^{-4}\)). Crucially, SOFA-adjusted analysis in GSE185263 (\(n = 342\), the only discovery cohort with SOFA scores) fully attenuated the C2 advantage (adjusted OR \(= 0.99\), \(p = 0.98\)), indicating that in this cohort endotype-mortality associations are primarily explained by baseline severity. A LASSO-regularised multinomial classifier achieved >98% per-class sensitivity and generalised to 4 external cohorts (\(n = 412\)) with 100% gene coverage and mean classification confidence 0.86−0.94. Longitudinal analysis of 129 patients across four multi-timepoint datasets demonstrated 66.7% endotype stability (Cohen’s \(\kappa = 0.49\)).

Conclusions

Three severity-aligned transcriptomic patterns—immune activation, interferon response, and an underrecognised erythroid/metabolic axis—recur across heterogeneous sepsis cohorts. In the one cohort with organ-dysfunction scores, the C2 mortality advantage was fully attenuated after SOFA adjustment, indicating that—in this cohort—endotype–mortality associations are largely explained by baseline severity rather than independent prognostic biology; whether this generalises requires cohorts with harmonised severity data. Future endotyping studies should incorporate systematic severity adjustment before drawing prognostic or treatment-stratification conclusions.