Blood transcriptomics separates infection signals from cell-composition effects in sepsis
Whole-blood host-response testing keeps moving closer to the bedside, but this study asks a more exacting question than diagnosis alone. Can a fixed transcriptomic score do more than separate sepsis from non-infectious systemic inflammation? Can it also tell clinicians which signals are being driven by neutrophil-rich blood and which are tied to monocyte or macrophage biology?
The answer in this analysis is partly yes, but with important boundaries. Investigators built a 12-gene framework from four infection-related contrasts drawn across three public datasets, then froze that framework before external testing. In the closest target-context cohort, GSE134347, the fixed score reached an AUROC of 0.827 (95% CI 0.764–0.889) for sepsis versus non-infectious critical illness. Performance dropped in pediatric day-1 SIRS comparisons, reinforcing a central point in host-response diagnostics: a blood signature is not a universal object, and its behavior changes with age, comparator group, sample type, and clinical setting.
The strongest signal came from a framework that was deliberately designed to be auditable. Rather than collapsing all blood RNA variation into one black-box classifier, the work separated candidate genes into expression contexts, then tested whether those candidates tracked granulocyte-associated biology or a more monocyte/macrophage-linked immunometabolic program.
A fixed 12-gene framework with context-specific performance
The discovery set started from 9,486 genes measured in all four contrasts. From there, the team narrowed the field to genes that were differentially expressed in multiple contrasts and directionally consistent across studies. The resulting framework included six immunometabolism-associated candidates — SMPDL3A, PLA2G7, GSTO1, SLC39A8, MTHFD2, and TXNIP — plus HLA-DRB1 and a set of diagnostic immune markers.
That design matters because it does not pretend every marker contributes in the same way. Some genes appear to follow cell abundance, especially in neutrophil-heavy samples. Others may reflect activation programs inside a particular lineage. The study therefore used multiple layers of validation: external cohort scoring, transcript-score sensitivity analyses, donor-level whole-blood single-cell pseudobulk profiling, cell-line qRT-PCR, and exploratory genetic analyses.
In the external cohort most closely aligned to the intended clinical use case, the score performed reasonably well. In pediatric cohorts, especially those comparing sepsis with active SIRS on day 1, performance was lower. The same pattern argues against overgeneralizing from one dataset or one age group to another. A score that performs well in adult ICU sepsis versus non-infectious critical illness may still struggle when the comparator is an inflammatory syndrome that already resembles infection at the bedside.
Neutrophil-associated markers and PLA2G7 did not behave the same way
The most clinically interesting result may be the separation of candidate biology into distinct axes. Adjustment for alternative neutrophil-associated transcript scores produced marker-dependent attenuation. Among the six-gene immunometabolic family, SMPDL3A was among the most attenuated candidates, while PLA2G7 was among the least attenuated. No candidate remained FDR-significant in the primary six-gene family after that adjustment.
That result suggests the immunometabolic panel is not a single coherent module. Some of the signal is clearly entangled with neutrophil-associated transcriptional shifts. Some of it appears less dependent on that axis, especially PLA2G7. The study’s donor-level single-cell work and qRT-PCR point in the same direction: PLA2G7 localizes to a monocyte/macrophage context, while the diagnostic immune markers map more closely to granulocyte-like biology.
This is the kind of distinction that matters in blood-based infection testing. A patient with severe bacterial infection may show a strong granulocyte signature because neutrophils dominate the sample. A different patient may show changes in monocyte/macrophage-associated pathways that are more relevant to immune regulation, antigen presentation, or lipid-mediated inflammatory signaling. The transcriptome can capture both, but only if the analysis keeps cell context in view.
Cell-level localization gives the score biological meaning
The whole-blood single-cell component helps anchor the bulk findings. Donor-level pseudobulk analysis in sepsis, compared with cardiac-surgery and healthy controls, supported a granulocyte-like localization for the diagnostic immune markers and a monocyte/macrophage context for PLA2G7. Cell-line qRT-PCR strengthened that interpretation in a controlled experimental setting, using THP-1 macrophage-like cells, dHL60 cells, and a mixed coculture system exposed to control conditions, LPS, or heat-killed E. coli.
This does not turn the score into a mechanistic explanation of sepsis, but it does move the biology closer to interpretation. If a transcript rises because neutrophils flood the blood, that signal has a different meaning than a transcript that rises because a macrophage program is switched on. The difference affects how clinicians and translational researchers think about validation, biomarker portability, and therapeutic hypotheses.
Genetic follow-up was exploratory, not confirmatory
The investigators also used cis-eQTL Mendelian randomisation and colocalisation to ask whether any of the immunometabolism-linked candidates might show a susceptibility effect or share a common genetic signal with disease-associated loci. The exploratory analyses did not support a susceptibility effect for the testable immunometabolism-associated candidates. The HLA-DRB1 regional signal also lacked evidence for a shared variant under single-variant colocalisation assumptions.
Those results should not be overread. They do not rule out a functional role for these genes in host defense or inflammatory remodeling. They do, however, caution against treating the transcriptomic associations as direct proof of inherited causal risk. In practical terms, the genetic analysis trims the list of candidates for deeper work and keeps the focus on expression context rather than causal claims.
What this framework can and cannot do at the bedside
The value of this study is not that it produces another standalone diagnostic test. Its value lies in showing how a fixed whole-blood signature can be dissected into cell-informed components. The framework can help separate granulocyte-associated diagnostics from a PLA2G7-linked monocyte/macrophage axis and may guide which genes deserve follow-up in future sepsis studies.
The limits are equally clear. Performance changed across cohorts. The pediatric day-1 comparisons were less favorable than the adult ICU setting. The immunometabolism-associated genes were not uniformly independent of neutrophil-related signal. And the genetic analyses were exploratory rather than definitive.
That leaves a practical research agenda rather than a finished clinical product. Future studies will need prospectively sampled cohorts, more tightly matched comparators, and direct testing of whether the PLA2G7-associated axis adds information beyond neutrophil-rich inflammation. The present framework is best read as a biologically annotated map of the infection response, not as a final answer on how to triage every patient with systemic inflammation.