Daily Sepsis Research Analysis
Three papers advance sepsis science across mechanisms and informatics: an AI-screened peptide that blocks gasdermin D pores delays pyroptosis and improves survival in murine sepsis; endothelial IL-6 signaling is shown to activate a noncanonical STAT1–cGAS–STING pathway driving IFN-like responses in endotoxemia; and NLP applied to ED triage free text markedly improves early sepsis prediction in a 134,266-patient cohort.
Summary
Three papers advance sepsis science across mechanisms and informatics: an AI-screened peptide that blocks gasdermin D pores delays pyroptosis and improves survival in murine sepsis; endothelial IL-6 signaling is shown to activate a noncanonical STAT1–cGAS–STING pathway driving IFN-like responses in endotoxemia; and NLP applied to ED triage free text markedly improves early sepsis prediction in a 134,266-patient cohort.
Research Themes
- Targeting pyroptosis via gasdermin D pore blockade as a sepsis therapy
- Endothelial IL-6 signaling through STAT1–cGAS–STING in shock pathophysiology
- NLP-enhanced early sepsis prediction from unstructured ED text
Selected Articles
1. Delaying pyroptosis with an AI-screened gasdermin D pore blocker mitigates inflammatory response.
An AI-screened peptide, SK56, directly blocks gasdermin D pores, delaying pyroptosis and curbing cytokine efflux. In murine LPS and CLP sepsis models, SK56 improved survival without altering IL-1β or GSDMD cleavage, and reduced mitochondrial damage and bystander activation in organoid and dendritic cell assays.
Impact: This work identifies a tractable, first-in-class pore-blocking strategy against pyroptosis with survival benefit in sepsis models, linking AI-guided peptide discovery to actionable immunotherapy.
Clinical Implications: Although preclinical, GSDMD pore blockade could inaugurate a therapeutic class targeting pyroptosis in sepsis and other hyperinflammatory states; next steps include safety, PK/PD, and translational studies.
Key Findings
- SK56 blocks GSDMD-NT pores and inhibits pyroptosis and cytokine release in macrophages.
- SK56 improves survival in LPS-induced endotoxemia and CLP polymicrobial sepsis in mice.
- SK56 does not affect IL-1β or GSDMD cleavage, indicating pore-level action.
- In organoid and immune co-cultures, SK56 reduces dendritic cell activation and prevents widespread death of human alveolar organoids; it also mitigates mitochondrial damage.
Methodological Strengths
- Multiple complementary systems: primary macrophages, dendritic cells, human alveolar organoids, and two in vivo sepsis models (LPS, CLP).
- Mechanistic specificity demonstrated by unchanged IL-1β and GSDMD cleavage with functional pore blockade.
Limitations
- Preclinical mouse models may not fully recapitulate human sepsis heterogeneity.
- Peptide delivery, stability, and immunogenicity require evaluation for clinical translation.
Future Directions: Define SK56 PK/PD, toxicity, and delivery; test efficacy in additional infection models; and explore combinations with antibiotics or immunomodulators.
The formation of membrane pores by cleaved N-terminal gasdermin D (GSDMD-NT) results in the release of cytokines and inflammatory cell death, known as pyroptosis. Blocking GSDMD-NT pores is an attractive and promising strategy for mitigating inflammation. Here we demonstrate that SK56, an artificial intelligence-screened peptide, effectively obstructs GSDMD-NT pores and inhibits pyroptosis and cytokine release in macrophages and human peripheral blood leukocyte-induced pyroptosis. SK56 prevents septic death induced by lipopolysaccharide or cecal ligation and puncture surgery in mice. SK56 does not influence cleavage of interleukin-1β or GSDMD. Instead, SK56 inhibits the release of cytokines from pyroptotic macrophages, mitigates the activation of primary mouse dendritic cells triggered by incubation with pyroptotic cytomembranes and prevents widespread cell death of human alveolar organoids in an organoid-macrophage coculture model. SK56 blocks GSDMD-NT pores on lipid-bilayer nanoparticles and enters pyroptotic macrophages to inhibit mitochondrial damage. SK56 presents new therapeutic possibilities for counteracting inflammation, which is implicated in numerous diseases.
2. Endothelial STING and STAT1 mediate IFN-independent effects of IL-6 in an endotoxemia-induced model of shock.
IL-6 elicits a transient, IFN-independent IFN-like gene program in endothelial cells via STAT1–cGAS–STING and IRFs, uncoupled from STAT3-mediated barrier regulation. Endothelial STING or global STAT1 deficiency attenuates endotoxemia responses and suppresses IFN-like signatures in vivo, revealing parallel IL-6 downstream pathways.
Impact: This study uncovers a noncanonical IL-6–STAT1–cGAS–STING axis in endothelium that drives IFN-like transcriptional responses in shock, redefining IL-6 signaling complexity and suggesting new intervention points.
Clinical Implications: Targeting endothelial DNA-sensing components (e.g., STING) or STAT1 may modulate IL-6–driven vascular inflammation in shock and sepsis, potentially complementing IL-6 or JAK pathway inhibitors.
Key Findings
- IL-6 induces a transient IFN-like gene program in human endothelial cells via a noncanonical, IFN-independent mechanism.
- This program requires STAT1, cGAS, STING, and IRF1/3/4 but is independent of STAT3.
- Endothelial STING knockout or global STAT1 knockout mice show attenuated responses to endotoxemia and lack the endotoxin-induced IFN-like signature.
- SOCS3 loss in endothelium enhances IFN-like responses in kidneys and brains during endotoxemia.
Methodological Strengths
- Integrated in vivo genetic models (endothelial STING-KO, global STAT1-KO) with primary human endothelial cell mechanistic assays.
- Clear dissection of pathway specificity distinguishing STAT1–cGAS–STING from STAT3-mediated functions.
Limitations
- Endotoxemia models may not capture the complexity of human polymicrobial sepsis.
- Therapeutic targeting feasibility and safety of endothelial STING/STAT1 modulation remain to be tested.
Future Directions: Evaluate endothelial-targeted STING/STAT1 modulators in polymicrobial sepsis models and assess synergy with IL-6/JAK pathway inhibitors.
Severe systemic inflammatory reactions, including sepsis, often lead to shock, organ failure, and death, in part through an acute release of cytokines that promote vascular dysfunction. However, little is known about the vascular endothelial signaling pathways regulating the transcriptional profile in failing organs. Our work focused on signaling downstream of IL-6, due to its clinical importance as a biomarker for disease severity and predictor of mortality. Here, we show that loss of endothelial expression of the IL-6 pathway inhibitor SOCS3 promoted a type I IFN-like (IFNI-like) gene signature in response to endotoxemia in mouse kidneys and brains. In cultured primary human endothelial cells, IL-6 induced transient IFNI-like gene expression in a noncanonical, IFN-independent fashion. We further show that STAT3, which we had previously demonstrated to control IL-6-driven endothelial barrier function, was dispensable for this activity. Instead, IL-6 promoted a transient increase in cytosolic mitochondrial DNA and required STAT1, cGAS, STING, and IRF1, -3, and -4. Inhibition of this pathway in endothelial cell-specific STING-KO mice or global STAT1-KO mice led to reduced the severity of the response to acute endotoxemic challenge and prevented expression of an endotoxin-induced IFNI-like gene signature. These results suggest that permeability and DNA-sensing responses are driven by parallel pathways downstream of this cytokine, provide potential insights into the complex response to acute inflammatory responses, and offer the possibility of novel therapeutic strategies for independently controlling the intracellular responses to IL-6 in order to tailor the inflammatory response.
3. Automated sepsis prediction from unstructured electronic health records using natural language processing: a retrospective cohort study.
In a 134,266-patient ED cohort, adding triage free text to ML models significantly improved sepsis prediction: random forest achieved AUPRC 0.789 and AUC 0.80, while BERT using raw text reached AUPRC 0.7542 and AUC 0.7735. Key operational and demographic variables further contributed to performance.
Impact: Demonstrates scalable, real-world NLP of triage text to enhance early sepsis detection at the point of ED intake, a critical window for outcome improvement.
Clinical Implications: Integrating free-text NLP into ED triage decision support could flag high-risk patients earlier, enabling timely antibiotics and resuscitation while reducing missed sepsis cases.
Key Findings
- Random forest using free-text triage features achieved AUPRC 0.789 (95% CI 0.7668–0.8018) and AUC 0.80 (95% CI 0.7842–0.8173).
- BERT on raw text achieved AUPRC 0.7542 (95% CI 0.7418–0.7741) and AUC 0.7735 (95% CI 0.7628–0.8017).
- Key predictors included ED treatment time, age, arrival-to-treatment time, Australasian Triage Scale, and visit type; incorporating free text improved detection and identified missed cases.
Methodological Strengths
- Very large retrospective cohort (N=134,266) with both structured variables and unstructured free-text triage comments.
- Comparative evaluation across multiple algorithms including BERT, with performance assessed by AUPRC and AUC.
Limitations
- Retrospective single-system data may limit generalizability; prospective external validation is lacking.
- Sepsis labeling and documentation quality in EHRs may introduce bias despite model performance.
Future Directions: Prospective deployment with clinician-in-the-loop evaluation, cross-site external validation, and assessment of time-to-antibiotics and outcome impact.
OBJECTIVE: Artificial intelligence (AI) holds promise for predicting sepsis. However, challenges remain in integrating AI, natural language processing (NLP) and free text data to enhance sepsis diagnosis at emergency department (ED) triage. This study aimed to evaluate the effectiveness of AI in improving sepsis diagnosis. METHODS: This retrospective cohort study analysed data from 134 266 patients admitted to the ED and subsequently hospitalised between 1 January 2016 and 31 December 2021. The data set comprised 10 variables and free-text triage comments, which underwent tokenisation and processing using a bag-of-words model. We evaluated four traditional NLP classifier models, including logistic regression, LightGBM, random forest and neural network. We also evaluated the performance of the BERT classifier. We used area under precision-recall curve (AUPRC) and area under the curve (AUC) as performance metrics. RESULTS: Random forest exhibited superior predictive performance with an AUPRC of 0.789 (95% CI: 0.7668 to 0.8018) and an AUC of 0.80 (95% CI: 0.7842 to 0.8173). Using raw text, the BERT model achieved an AUPRC of 0.7542 (95% CI: 0.7418 to 0.7741) and AUC of 0.7735 (95% CI: 0.7628 to 0.8017) for sepsis prediction. Key variables included ED treatment time, patient age, arrival-to-treatment time, Australasian Triage Scale and visit type. DISCUSSION: This study demonstrates AI, particularly random forest and BERT classifiers, for early sepsis detection in EDs using free-text patient concerns. CONCLUSION: Incorporating free text into machine learning improved diagnosis and identified missed cases, enhancing sepsis prediction in the ED with an AI-powered clinical decision support system. Large, prospective studies are needed to validate these findings.