Pathogen identity is reshaping pulmonary sepsis biology
Pulmonary sepsis remains one of the deadliest forms of infection-driven critical illness, yet bedside care still begins with a definition built on organ dysfunction rather than biology. This review argues that approach is too blunt for a syndrome driven by highly variable pathogen-host interactions in the lung. In bacterial pneumonia, viral infection, and fungal invasion, the immune response can tilt toward explosive inflammation, adaptive immune exhaustion, or mixed states that shift over time. A single label of sepsis cannot capture those differences.
The lung is not a passive entry point. Its alveolar immune microenvironment, epithelial barrier, resident macrophages, and vascular interface all influence how pathogen-associated molecular patterns are sensed and amplified. Gram-negative organisms such as E. coli and K. pneumoniae engage LPS-driven signaling through TLR4, MD2, and CD14, and that pathway tends to push patients toward a hyperinflammatory state. Gram-positive bacteria use different ligands, including peptidoglycan, lipoteichoic acid, and exotoxins, and experimental models suggest a somewhat different immunopathology, including pneumolysin-linked alveolar epithelial pyroptosis through GSDMD and the NLRP3 inflammasome. Viral and fungal infections add yet another layer, with type I interferon surges, checkpoint upregulation, lymphocyte loss, and antigen-presenting cell dysfunction.
Why classic sepsis trials keep missing the mark
The review connects these biological differences to one of sepsis research’s oldest failures: immunomodulatory drugs tested in unselected populations. Trials of lipopolysaccharide blockade, TNF-α inhibition, corticosteroids, statins, and other immune-directed strategies have repeatedly produced neutral overall results. The problem is not simply that the drugs are weak. It is that they were often deployed against a syndrome that contains several different immune states, many of which would not be expected to respond in the same direction.
Conventional diagnostic scores identify the consequences of illness, not the immune program driving them. A patient in refractory septic shock from Gram-negative pneumonia may have a cytokine-dominant state with intense NF-κB activation, while another patient with the same organ failure score may already be in endotoxin tolerance, with HLA class II downregulation and T-cell exhaustion. Treating both with the same immune intervention invites failure. In that sense, the translational challenge is not only about finding better drugs; it is about finding the right biological target in the right patient at the right time.
Transcriptomic endotypes and clinical phenotypes do not line up neatly
The review gives substantial attention to the growing vocabulary of sepsis subgroups: SRS1/SRS2, Mars1–4, SENECA, hyperinflammatory/hypoinflammatory ARDS phenotypes, USE-1/2, and broader inflammatory, adaptive, or coagulopathic transcriptomic clusters. These systems are not interchangeable. They were derived from different cohorts, different data layers, and different analytic strategies, which explains why a patient can fit one framework but not another.
SRS1 is associated with endotoxin tolerance, T-cell exhaustion, and reduced HLA class II expression, while SRS2 preserves relatively more adaptive immune activity. Mars1 reflects immunometabolic dysregulation and carries a higher mortality signal. SENECA’s δ subtype is defined clinically by refractory hypotension, hyperlactatemia, and hepatic and cardiovascular failure. In ARDS cohorts, the hyperinflammatory phenotype has been linked to higher concentrations of IL-6, IL-8, and sTNFR1, along with lower bicarbonate, and it has shown different treatment responses than the hypoinflammatory phenotype in secondary analyses.
A baseline transcriptomic snapshot can be misleading.
Endotypes shift over time, and the review is explicit that this drift matters. A patient may move from an SRS1-like suppressed state toward a more permissive SRS2-like state, or follow the reverse trajectory depending on microbial control, source control, resuscitation, and organ support. This dynamic behavior means a one-time blood sample cannot be assumed to define the whole illness course. Repeated measurement is a biological necessity, not an academic luxury.
Pathogens appear to bias the immune program in different directions
The most useful part of the review is its pathogen-centered synthesis. Gram-negative bacteria are linked to the strongest evidence for a hyperinflammatory phenotype, with endotoxin activity tracking inflammatory intensity. Gram-positive organisms, particularly Staphylococcus aureus and Streptococcus pneumoniae, may produce a different clinical profile, including higher S100A12 and platelet counts in some cohorts. Respiratory viruses are associated with type I interferon signaling, dysregulated CD8+ T-cell differentiation, CD38/CD73-driven adenosine immunosuppression, and severe lymphopenia. Invasive fungi are tied to checkpoint upregulation, HLA class II loss, and APC exhaustion, a pattern that resembles immunosuppressed endotypes more than cytokine storm.
These are not yet causal maps. Much of the evidence comes from ARDS, general sepsis, severe pneumonia, or COVID-19 rather than pulmonary sepsis alone. Even so, the directionality is clinically persuasive. If a pathogen consistently pushes the host into a specific immune state, then pathogen identification should influence not only antimicrobial selection but also the logic of adjunctive therapy. The molecular trigger is part of the phenotype.
Treatment signals are strongest when they match the biology
Several interventions in the review show why phenotype matching matters. In the FACTT trial secondary analysis, a conservative fluid strategy appeared favorable in the hypoinflammatory subgroup but not in the hyperinflammatory subgroup, where mortality was higher with that approach. Hydrocortisone showed potentially divergent behavior as well: secondary analyses suggested possible harm in SRS2 and in a pulmonary-sepsis subgroup with an immunoadaptive phenotype, even as other datasets and a separate severe community-acquired pneumonia trial showed benefit in a different clinical setting. Simvastatin produced a survival signal only in the hyperinflammatory ARDS subgroup, not in the hypoinflammatory group.
The pattern is biologically plausible. Hyperinflammatory states may respond to therapies that temper endothelial leak, inflammatory amplification, or maladaptive cytokine signaling. Immunosuppressed states may need restoration of host defense rather than further suppression. Checkpoint inhibition, recombinant IL-7, and related immune-restorative therapies remain experimental, but their rationale rests on this same logic: rescue the immune system when it has been silenced, and avoid suppressing it when it is already failing to control pathogen burden.
Biomarker-guided care still needs prospective proof
The review is careful about the limits of the current evidence. Polymyxin B hemoperfusion was associated with lower adjusted mortality only in a post hoc endotoxin-enriched subgroup. Tocilizumab has a place in hypoxic COVID-19 with corticosteroids, but that evidence should not be carried over to bacterial pulmonary sepsis. Anti-PD-L1 therapy has shown acceptable short-term safety and pharmacodynamic activity in early work, yet clinical benefit remains unproven. Recombinant IL-7 raises lymphocyte counts, but consistent outcome benefit has not been established. GM-CSF and thymosin α1 are still confined to uncertain, conflicting, and context-specific data.
The central message is not that precision immunotherapy already works. It is that the biology is finally coherent enough to justify better trials. The next step is not another broad sepsis study that enrolls everyone with organ dysfunction. It is an adaptive design that enriches for pathogen-associated host-response states, measures them repeatedly, and tests whether intervention selection based on evolving phenotype can improve survival without causing harm. Until then, pulmonary sepsis remains a syndrome in which the microbe, the immune response, and the timing of measurement all have to be considered together.