Kox et al. recently published a comprehensive review in Nature Reviews Nephrology, providing a timely and thorough summary of sepsis immunobiology^[1]. While Kox et al.^[1] provide an elegant synthesis of current concepts in sepsis immunobiology, the review also inadvertently underscores a persistent translational gap: despite substantial conceptual progress, the field still lacks robust evidence that these frameworks meaningfully improve patient outcomes in real-world clinical settings. Our specific perspective is that the next step is not another sepsis taxonomy, but an actionability test: a proposed endotype should demonstrate biological validity across time and compartments, prospectively identify differential response to a specified therapy, and return an interpretable result within the clinical decision window. This decision-focused standard frames the discussion that follows.

Sepsis is currently defined as life-threatening organ dysfunction caused by a dysregulated host response to infection. As the authors insightfully highlight, pro-inflammatory and anti-inflammatory pathways are often activated simultaneously rather than sequentially. Hyperinflammation and immunosuppression frequently coexist. The compartmentalization phenomenon further complicates the picture: sepsis involves multi-organ injury, and detection based solely on blood samples may fail to reflect tissue-specific responses. Metabolic reprogramming and long-term sequelae further exacerbate the complexity of the sepsis disease course.

The article emphasizes moving away from the traditional binary paradigm of “hyperinflammation versus immunosuppression” toward a more nuanced understanding of the dynamic, compartmentalized, and highly heterogeneous host response^[1]. This perspective is highly consistent with evidence accumulated from multiple independent cohort studies^[2-5] and explains why “one-size-fits-all” immunomodulatory trials have largely failed, while paving the way for predictive enrichment and personalized interventions^[6,7]. Nevertheless, these conceptual frameworks remain difficult to operationalize in clinical practice, and robust bedside tools to translate these theoretical constructs into real-time decision support are still lacking. Accordingly, reproducible prognostic separation should not be equated with treatment-predictive utility.

Kox et al. emphasize the balanced application of prognostic and predictive enrichment. Prognostic tools can identify high-risk patients but cannot guarantee response to specific therapies^[1]. Predictive enrichment has shown promising signals in secondary analyses: hyperinflammatory Acute Respiratory Distress Syndrome ‌(ARDS) subphenotypes benefit from simvastatin or fluid management, while the efficacy of glucocorticoids depends on sepsis response signatures (SRS) subtype or inflammatory status. These observations are hypothesis-generating and should not be presented as equivalent to prospectively validated precision therapy in sepsis. Several biomarker-guided therapeutic strategies have been proposed, aiming to translate immune heterogeneity into actionable interventions [Table 1]. These approaches span diverse biological states, including hyperinflammation, immunoparalysis, complement activation, and metabolic dysfunction. However, most remain supported by subgroup analyses, mechanistic studies, or early-phase trials, highlighting the substantial gap between biological plausibility and clinical implementation. Among these strategies, biomarkers such as mHLA-DR, suPAR, and ferritin have provided proof-of-concept for theragnostic approaches. Nevertheless, their clinical utility remains uncertain because most findings derive from post hoc analyses, and prospective randomized validation is still limited.

A systematic review of predictive enrichment using biomarkers in sepsis also supports this view^[64]. It identified 12 eligible randomized studies, all using single circulating protein or endotoxin biomarkers for predictive enrichment. Although five studies reported statistically significant primary outcomes, only two showed patient-important benefits. Only five used point-of-care assays, and none incorporated genetic, transcriptomic, or metabolomic markers, underscoring the narrow scope and limited clinical translation of current biomarker-guided strategies in sepsis trials. Beyond these limitations, challenges remain prominent: sepsis phenotypes exhibit significant dynamism, with one study showing that up to 46% of baseline phenotypes undergo substantial changes by days 3 and 5^[65], posing a major challenge for enrichment strategies reliant on single time-point assessments. Subphenotype studies also face a lack of standard assays with harmonized reliability and reproducibility; insufficient simultaneous assessment of multi-omics and non-omics responses across compartments^[66]; and, as noted in the review, marked compartmentalization of the host response, where blood-based biomarkers often fail to reflect the local immune status in target organs such as the lungs and kidneys, potentially leading to mis-stratification when relying solely on peripheral blood detection.

Most subphenotype and endotype studies still rely on retrospective cohorts, with insufficient prospective validation. Currently, there is also a lack of rapid, reliable point-of-care testing technologies in clinical practice, making it difficult to meet the urgent time-sensitive demands of sepsis diagnosis and treatment^[64]. As a pragmatic research framework, an initial state estimate could be obtained within 6 h, repeated at approximately 24-48 h, and reassessed at days 3-5 in patients with persistent or worsening organ dysfunction, with event-triggered testing for recurrent shock, secondary infection, or new organ injury. These intervals are proposed for prospective testing, not as a validated clinical standard. Any dynamic algorithm should prespecify how indeterminate results, blood-organ discordance, biomarker normalization, and safety-based stop or switch decisions will be handled.

Host transcriptomic endotypes have revealed differences in adaptive immunity, innate inflammation, endothelial activation, coagulation, and metabolism^[3-5]. A recent consensus analysis reconciled major blood transcriptomic systems into three subtypes using the Molecular Diagnosis and Risk Stratification of Sepsis (MARS) and the Genomic Advances in Sepsis (GAinS) cohorts, including longitudinal samples^[67]. Host genetic variation can also shape the sepsis transcriptomic response: an analysis of 638 adults identified 16,049 independent expression quantitative trait loci and genotype-by-endotype interactions^[68]. Beyond transcriptomic regulation, genetic causal-inference approaches may help prioritize candidate biomarkers. For example, an observational and two-sample Mendelian randomization study reported an association between higher serum alkaline phosphatase levels and sepsis susceptibility, although its utility for immune endotyping or treatment prediction remains unestablished^[69]. Together, these findings support mechanism discovery and biomarker prioritization, but neither molecular association nor Mendelian randomization evidence alone establishes an actionable endotype or predicts treatment response.

Longitudinal integration of transcriptomics, proteomics, metabolomics, and functional immune assays may better capture transitions than a single baseline measurement^[70]. However, additional molecular layers are useful only if they improve a prespecified decision beyond routine clinical data, can be measured within the treatment window, and retain calibration across populations.

The review highlights several repurposed and novel agents: Janus Kinase-Signal Transducer and Activator of Transcription (JAK-STAT) inhibitors, metformin, selective Complement Component 5a (C5a) inhibitors, Interleukin-7 (IL-7), anti- Programmed Death (Ligand)-1 [PD(L)1] monoclonal antibodies, and β-glucan-induced trained immunity. These agents target specific nodes in the inflammatory process and align with conceptual frameworks such as resistance, tolerance, resolution, and repair. Combinations (e.g., C5a + IL-6 blockade) merit exploration given pathway crosstalk.

Sepsis drug development has faced long-standing difficulties, with most randomized controlled trials ending in failure. The primary reasons include extreme patient heterogeneity, limitations of single-target interventions, lack of reliable predictive biomarkers, and the inability of traditional fixed-design trials to address the dynamic disease course^[71,72]. On ClinicalTrials.gov, three randomized controlled trials on precision immunotherapy for sepsis are currently registered: ImmunoSep, titrated administration of IgM-enriched preparation, and Adaptive Platform Trial for Personnalisation of Sepsis Treatment in Children and Adults (PALETTE). The results of the ImmunoSep trial were published in JAMA this year, demonstrating the importance of precision immunotherapy: researchers personalized treatment with anakinra or recombinant interferon-γ according to patients’ immune status, and the precision immunotherapy group showed significantly better improvement in organ function and infection clearance compared with the placebo group. Its prespecified primary endpoint-a decrease of at least 1.4 points in mean Sequential Organ Failure Assessment (SOFA) score from baseline through day 9-was achieved in 35.1% of patients assigned to immune-status-guided therapy and 17.9% assigned to placebo {difference, 17.2% [95%CI, 6.8% to 27.2%]; P = 0.002}. Mortality at 28 days was not statistically significantly different between groups^[73]. Results from the other two trials have not yet been published. Independent replication, assay standardization, and external validation are required before routine implementation.

Most current evidence supporting sepsis phenotyping and precision immunotherapy is derived from European and North American cohorts. However, substantial inter-population variability in genetic background, comorbidity patterns, pathogen distribution, and healthcare systems may limit the direct generalizability of these findings to Asian populations. In addition, differences in baseline immune status and treatment accessibility may further influence biomarker performance and enrichment strategy effectiveness. Therefore, region-specific validation in Asian critically ill populations is urgently required to ensure the global applicability of precision medicine approaches. We briefly note the CMAISE (Chinese Multi-omics Advances in Sepsis) database as an emerging longitudinal multi-omics resource that may support future mechanistic and translational studies in Asian critically ill populations^[74]. Across 43 Chinese centers, multi-omics profiling has been completed for 1,284 patients meeting Sepsis-3 criteria, generating more than 4,500 samples at principal time points on days 1, 3, and 5 and integrating transcriptomic, proteomic, metabolomic, single-cell, and dense clinical data. Multi-omics initiatives such as CMAISE may help address this gap by enabling population-specific model development and validation.

The core future directions for sepsis treatment include: establishing dynamic, multidimensional stratification systems that integrate clinical phenotypes, molecular endotypes, and treatable traits to achieve real-time precision intervention; developing point-of-care rapid multi-omics detection combined with artificial intelligence-assisted decision-making systems to overcome the limitations of traditional single-time-point detection and static stratification; and promoting adaptive platform trials and embedded study designs to accelerate clinical validation of multi-target combined immune modulation strategies.

Kox et al.’s^[1] article clearly declares that sepsis immunology has moved beyond the era of simple binary classification. The realization of precision medicine depends on three pillars: first, deepening the understanding of core mechanisms such as immune resistance, disease tolerance, resilience, resolution and repair; second, using multi-omics technologies and artificial intelligence to identify reliable, dynamic, and point-of-care detectable “treatable traits”; and third, adopting flexible and efficient clinical trial designs, such as adaptive platform trials and Bayesian methods, to simultaneously evaluate multiple therapies targeting specific endotypes.

The main future challenge lies in how to systematically integrate these insights into clinical practice and develop “combination” precision treatment regimens that can act at different time points, target different individuals, and affect different tissues and organs. For critical care physicians and researchers, this is a pivotal moment to abandon “one-size-fits-all” thinking and embrace the complexity and precision of the disease. Correction of immune dysregulation alone is unlikely to restore established multiorgan dysfunction; antimicrobial therapy, source control, lung-protective ventilation, hemodynamic optimization, renal support, nutrition, rehabilitation, and avoidance of iatrogenic injury must proceed in parallel. Precision immunotherapy should therefore be judged by reproducible patient-important outcomes, not biomarker normalization alone.