Multicohort Transcriptomic Profiling Identifies LCN2‐Associated Myeloid Biomarker Signatures Relevant to Nutritional Immunity in Sepsis
Jie Gao, Zhiping Huang, Zhifeng Lu, Weiwei Pan, Shuangjie Ren, Weina SunBackground
Sepsis is an inflammation‐associated syndrome with substantial myeloid heterogeneity and limited clinically transferable biomarkers. LCN2 encodes neutrophil gelatinase‐associated lipocalin (NGAL), an iron‐binding protein not only involved in antibacterial nutritional immunity but also influenced by acute kidney injury and systemic inflammation. We investigated the transcriptomic context of LCN2 expression and whether associated immune states could inform biomarker discovery and molecular stratification in sepsis.
Methods
Bulk transcriptomic datasets GSE57065, GSE65682, GSE66099, and GSE95233 and a single‐cell RNA sequencing cohort were analyzed. Circadian gene‐set scoring, differential expression, weighted gene coexpression network analysis, functional enrichment, machine learning, Shapley additive explanations, immune‐cell deconvolution, and non‐negative matrix factorization clustering were integrated. Candidate‐feature generation and model development were restricted to the GSE65682/GSE95233 training cohort, whereas GSE57065 and GSE66099 were reserved for locked retrospective external evaluation. The classification task separated sepsis or septic shock from healthy controls and was not designed as a clinical differential‐diagnosis study.
Results
Single‐cell analysis identified 10 computational clusters that were consolidated into seven major immune lineages. Monocytes had the highest circadian gene‐set scores, and LCN2, S100P, and CD177 were upregulated in sepsis‐associated monocyte states. Bulk deconvolution independently indicated altered monocyte‐ and neutrophil‐associated abundance scores. Integrated analyses prioritized 23 candidate genes enriched in mitochondrial organization, oxidative phosphorylation, and inflammatory signaling. A descriptive molecular classifier achieved an apparent training area under the receiver operating characteristic curve of 0.997. Retrospective external evaluation values were 1.000 in GSE57065 and 0.979 in GSE66099. Shapley additive explanation analysis prioritized RBX1, UCP2, CASP4, ARL4A, and SFPQ, while non‐negative matrix factorization identified two molecular subtypes with distinct estimated immune‐cell abundance and pathway profiles.
Conclusions
Multicohort transcriptomic integration identified sepsis‐associated LCN2 expression and myeloid biomarker signatures that provide a human transcriptomic framework for studying NGAL‐mediated nutritional immunity. These findings may inform mechanistic evaluation of NGAL against iron‐dependent pathogens, including Vibrio vulnificus. However, the cohorts were not pathogen‐stratified and did not measure NGAL protein, bacterial growth, siderophore binding, fur/FUR responses, or renal‐adjusted effects. No V. vulnificus–specific or causal antibacterial mechanism can therefore be inferred. Prospective clinical and experimental validation is required before biomarker‐guided or pathogen‐specific application.