Section 1.4: Consensus Framework for Sepsis Transcriptomics

Sepsis is a life-threatening organ dysfunction syndrome caused by a dysregulated host response to infection. For a long time, the core clinical dilemma has been the high heterogeneity of host responses, which has repeatedly thwarted clinical trials relying on uniform diagnostic criteria and “one-size-fits-all” therapeutic strategies. Over the past decade, scientists have proposed multiple subtyping methods based on blood transcriptomic data from sepsis patients, such as the MARS subtypes (Mars1-4), SRS subtypes (SRS1, SRS2), and Stanford subtypes (hyperinflammatory, adaptive, coagulopathic). These approaches have opened new avenues for understanding the immunopathological mechanisms of sepsis. However, these subtyping systems lack unified standards in terms of the number of subtypes, nomenclature, and biological interpretation. Furthermore, differences in cohorts, technical platforms, and analytical methods make direct comparison and reproducible validation difficult, severely limiting their clinical translational potential. To overcome this bottleneck, Professor Tom van der Poll’s team conducted an international collaborative study, recently published in Nature Medicine. This research aims to integrate existing subtypes to construct a standardized, reproducible, and biologically grounded consensus transcriptomic subtyping framework to advance precision diagnosis and treatment of sepsis. This article will provide an in-depth interpretation and discussion of this milestone study.

I. Construction of the Sepsis Transcriptomic Consensus Framework

This study integrated prospective cohorts from the Dutch MARS project and the UK GAinS study, encompassing blood transcriptomic data from 1,122 ICU patients with sepsis. Through network analysis and unsupervised clustering algorithms, the researchers fused and compared three previously published independent molecular subtyping systems, ultimately establishing three stable consensus transcriptomic subtypes (CTS) with clear biological connotations. By combining gene set enrichment analysis and single-cell RNA sequencing, the study performed multidimensional profiling of the molecular and cellular characteristics of each subtype.

CTS1 is characterized by the activation of typical pro-inflammatory pathways, including interleukin-6 (IL-6) signaling, reactive oxygen species (ROS) generation, glycolysis, lipogenesis, oxidative phosphorylation, and mTORC1 signaling. Plasma levels of inflammatory cytokines (IL-6, IL-8, IL-10), granulocyte activation markers (MMP-8, NGAL), and endothelial injury indicators (soluble E-selectin, Ang-2/Ang-1 ratio) were significantly elevated. At the single-cell level, its transcriptional signature was enriched in immature neutrophils, particularly the IL1R2⁺ and PADI4⁺ cell subsets, collectively reflecting a robust innate immune response and endothelial injury.

CTS2 is primarily characterized by significant activation of heme metabolism, myogenesis, and estrogen response pathways. Patients exhibited fibrinolytic system dysregulation, with elevated tissue plasminogen activator (tPA) and plasminogen activator inhibitor-1 (PAI-1) levels, and the lowest plasma protein C levels. Its unique transcriptional signals were most prominent in platelets and eosinophils. These findings suggest that this subtype suffers from severe red blood cell destruction or metabolic abnormalities, leading to increased free heme, iron metabolism disorders, and oxidative stress, which may underlie its highest APACHE IV scores and ICU mortality rates.

CTS3 is marked by the activation of interferon signaling, allograft rejection, Wnt/β-catenin, MYC, and DNA damage repair pathways. Patients had relatively higher plasma protein C levels, indicating preserved anticoagulant function. Its transcriptional features were closely associated with increased populations of all lymphocyte subsets (B cells, T cells, NK cells) and non-classical monocytes, suggesting a more active adaptive immune response with milder inflammatory and coagulopathic disturbances. Compared to existing subtypes, SRS1, Mars2, and the hyperinflammatory subtype clustered into CTS1; portions of SRS2, Mars1, and the coagulopathic subtype clustered into CTS2; while portions of SRS2, Mars3, and the adaptive subtype clustered into CTS3. Mars4 patients did not form a distinct cluster in this study, but hypergeometric testing revealed a significant association with the adaptive subtype, likely due to the low prevalence of Mars4.

Based on network analysis, the authors identified a core set of samples representing each CTS subtype, which exhibited highly consistent subtype assignments across different classification methods. Using the gene expression data from these core samples, they constructed a predictive classifier for CTS types. This 18-gene classifier achieved an out-of-bag error rate of only 2.2%. These 18 genes are not randomly selected but precisely reflect the core immunopathological processes of each subtype. Genes highly expressed in CTS1 include ACER3, SERPINB1, HIK3, TDRD9, NLRC4, PGD, UBE2H, METTL9, STOM, SNX3, and GADD45A. Together, they form a molecular map of systemic inflammatory burst and tissue stress. For example, NLRC4 encodes a key inflammasome component; its high expression indicates activation of the core pyroptosis pathway, a critical step triggering systemic inflammation and tissue injury. GADD45A and STOM reflect DNA damage stress in tissue cells under inflammatory conditions and endothelial dysfunction/barrier disruption, respectively, revealing the widespread damage caused by inflammation to host tissues. Genes highly expressed in CTS2 include BPGM, CA1, SLC4A1, EPB42, FECH, and GLRX5. All six are closely related to red blood cell physiology and heme metabolism. BPGM, CA1, SLC4A1, and EPB42 are key molecules for gas transport, ion exchange, and cytoskeletal stability in mature erythrocytes; their high expression indicates active erythrocyte-related transcriptional programs. FECH and GLRX5 are directly involved in heme synthesis and iron-sulfur cluster assembly, providing direct evidence of heme metabolism pathway activation. The high expression of this gene set strongly suggests that CTS2 patients may experience severe red blood cell destruction and ineffective erythropoiesis, leading to excess free heme and iron ions. BTN3A3 is highly expressed exclusively in CTS3. This gene plays a crucial role in adaptive immune responses, directly participating in T cell costimulation and γδ T cell antigen recognition, with its expression strongly induced by interferons. Its high expression indicates a shift in the host immune response toward an adaptive immune phase centered on T cells and accompanied by active interferon responses. This state may correlate with better viral clearance, immune regulation, and relatively preserved anticoagulant function. The authors further independently validated the 18-gene classifier using prospective cohort data from a Ugandan referral hospital. In standardized RNA sequencing data, the classifier successfully assigned patients to CTS1-3, demonstrating good extrapolability across populations and geographies. Additionally, the researchers tested the classifier’s sensitivity across different hospital-to-ICU admission intervals (representing the progression of organ dysfunction). Results showed that regardless of whether patients were transferred to the ICU within 1 day, after 8 days, or even after 28 days, the CTS distribution showed no significant differences, with 98%–99% concordance with the original classification. Silhouette analysis and confusion matrices further confirmed that the classifier maintained robust performance across different admission times and study sites. These sensitivity analyses collectively indicate that the CTS classifier’s performance is unaffected by hospital-to-ICU transfer intervals or geographical variations.

The researchers also analyzed the impact of corticosteroid therapy on outcomes across different CTS subtypes. In the MARS cohort, the proportion of patients receiving corticosteroids was significantly higher in CTS1 and CTS2 subtypes than in CTS3 (61.5%, 62.5%, and 38.1%, respectively). To further investigate the effect of glucocorticoid therapy on mortality across CTS groups, the authors used propensity score matching to balance confounding factors. After matching, baseline characteristics were comparable across CTS1, CTS2, and CTS3 groups, with minor differences only in abdominal sepsis diagnosis, pneumonia diagnosis, and hospital-to-ICU interval. Logistic regression analysis showed that corticosteroid use significantly increased the 28-day mortality risk (OR=3.83, 95% CI 2.38–6.28, P=5.2×10⁻⁸). Stratified analysis revealed that among untreated patients, CTS2 patients already had a significantly higher baseline mortality risk than other subtypes (OR=2.14, 95% CI 1.22–3.80, P=0.00085). More critically, corticosteroid therapy demonstrated clear harm to CTS1 and CTS2 patients: compared to untreated counterparts, their 28-day mortality rates increased by 33.7% and 34%, respectively. Conversely, in CTS3 patients, corticosteroid therapy may have a protective effect, significantly reducing mortality risk compared to CTS1 patients. In the independent VANISH RCT dataset, although no significant inter-subtype differences in corticosteroid efficacy were found, the CTS2 subtype was again confirmed as a strong predictor of mortality (OR=5.76, 95% CI 1.31–25.40, P=0.021). These results collectively suggest that immune status assessment based on CTS subtyping may be highly valuable for guiding corticosteroid use in sepsis patients, particularly warning against the potential risks of corticosteroid administration in the CTS2 subtype.

II. Clinical Significance and Limitations

This study did not propose a novel subtyping system from scratch but innovatively constructed a standardized, consensus-based framework by comprehensively integrating classic cohorts from previous studies, laying a foundation for more targeted clinical trials. Simultaneously, it preliminarily revealed potential associations between sepsis molecular subtypes and treatment responses. This sounds an alarm for “one-size-fits-all” immunomodulatory strategies and points toward a viable direction for individualized treatment based on host response subtyping. However, the clinical translation of this consensus framework requires attention to its limitations. First, although validated in a Ugandan cohort, the primary data still originate from ICU patients in high-income countries. Its applicability across different healthcare resource levels, pathogen spectra, ethnicities, and pediatric populations remains to be confirmed. Second, the subtyping is primarily based on a single timepoint at ICU admission. Although the study mentions that subtypes may dynamically shift during disease progression, the patterns, driving factors, and impacts on treatment decisions remain unknown. Finally, current findings regarding corticosteroid efficacy stem from post hoc analyses and urgently require rigorous validation in prospective RCTs to establish their reliability for guiding therapy.

III. Future Prospects and Implications

This research charts a new direction for precision medicine in sepsis. Clinically, simplified molecular subtyping tools, such as the 18-gene classifier, can be used in the future to early identify high-risk CTS subtype patients, avoiding potentially harmful corticosteroids and instead strengthening organ support and targeted interventions. Simultaneously, CTS3 patients who may benefit from immunomodulation can be identified, enabling individualized therapeutic strategies. In research and trial design, novel mechanism-targeted therapies can be tested in specific subtype populations. Looking ahead, research in this field should focus on prospective interventional studies to validate the therapeutic guidance value of subtyping; elucidate the evolutionary trajectory of host responses through multi-timepoint dynamic monitoring; integrate multi-omics data to deeply dissect the driving mechanisms of each subtype; and accelerate the translation and application of rapid, simple bedside molecular subtyping diagnostic technologies, truly ushering sepsis diagnosis and treatment into a new era of precision stratification.

In conclusion, by constructing a standardized, reproducible consensus blood transcriptomic subtyping framework, this study successfully categorizes the complex heterogeneity of sepsis into three host response subtypes with distinct biological features and clinical significance. This lays a crucial theoretical and methodological foundation for the era of precision subtyping and individualized treatment in sepsis. As more prospective clinical trials are conducted, we can reasonably anticipate a new era of sepsis diagnosis and treatment based on host response characteristics, truly achieving “precision targeting.”

(Pei Shuaijie, Xie Keliang; General Hospital of Tianjin Medical University)