Sepsis · Patient stratification
Sepsis split into two host responses — and the immunosuppressed one died more often
Save this development or follow its topic to track what changes.
Short answer
Two reproducible sepsis response signatures differ in immune phenotype and 14-day mortality, with the immunosuppressed signature carrying roughly 2.4 to 2.8 times the hazard.
What happened
In the GAinS study, Davenport and colleagues profiled peripheral-blood leukocyte gene expression in 265 UK ICU adults with community-acquired pneumonia sepsis and organ dysfunction, then validated in 106 more. Unsupervised clustering produced two signatures. SRS1, about 41% of the discovery cohort, showed endotoxin tolerance, T-cell exhaustion and HLA class II downregulation, with higher 14-day mortality (discovery HR 2.4; validation HR 2.8). A small predictive gene set reproduced the assignment.
Why it matters
Sepsis trials have repeatedly failed against a single average patient. Two reproducible host-response states with different mortality is the kind of finding that explains those failures — and the validation cohort is what separates it from the clustering results that do not survive scrutiny.
Evidence
- Study type
- Prospective cohort with unsupervised transcriptomic clustering and an independent validation cohort
- Sample
- 265 discovery patients and 106 validation patients in UK ICUs
- Journal
- The Lancet Respiratory Medicine · peer reviewed
- Replication
- The two-signature structure and its mortality association reproduced in the validation cohort
- Limitations
- Pneumonia-driven ICU sepsis in UK adults may not represent all sepsis. Prognostic stratification is not evidence that treating by signature changes outcomes.
What this connects to
Sources
The 2 studies this explanation is built from, by the role each plays. Every source links to PaperFren’s explanation of it and to the original paper.
Primary study
- Two blood gene signatures track sepsis outcomes
In UK ICU adults with pneumonia sepsis, leukocyte transcriptomes split into SRS1 (immunosuppressed, ~41%) and SRS2; SRS1 had roughly 2–3× higher 14-day mortality, and a 7-gene set classified the groups.
What it does not showLimitations
Does not prove that assigning SRS at the bedside and treating differently improves survival in a randomized trial. The discovery setting is pneumonia-driven ICU sepsis in UK adults—not every infection source or age group. Genetic eQTL findings explain expression variation; they are not prescriptions for IL-6 drugs.
PaperFren explanationStudy with cards and a quizOriginal paper (DOI)cc by
Supporting evidence
- A 22-gene classifier flags early metastasis after prostatectomy
In Mayo radical-prostatectomy men, a 22-marker genomic classifier from primary-tumour expression predicted early clinical metastasis after PSA rise better than clinical variables (validation AUC 0.75).
What it does not showLimitations
Improving metastasis prediction is not the same as proving that treating by GC score improves survival in an RCT. The design enriches for PSA-rise/metastasis cases from a surgical registry — not every newly diagnosed man. AUC 0.75 is useful discrimination, not perfect triage.
PaperFren explanationStudy with cards and a quizOriginal paper (DOI)cc by
Before
Sepsis was enrolled and treated as one syndrome defined by clinical criteria, with the heterogeneity of the host response acknowledged but not measurable at the bedside.
Now
A transcriptomic stratification replicates across cohorts and predicts mortality. It does not show that assigning patients to a signature and treating them differently improves survival — that requires a trial nobody has run here.