Integrative Mendelian randomization reveals the mediating role of immune cell traits in the causal pathway from gene expression to neonatal bacterial sepsis

Neonatal bacterial sepsis (NBS) remains a major global burden. Although immune dysfunction is central to its pathogenesis, the causal contribution of specific immune cell phenotypes and their upstream genetic regulation is unclear. This study aimed to test whether genetically predicted gene expression influences NBS risk through immune cell traits using an integrative two-step Mendelian randomization (MR) mediation framework. We combined eQTLGen whole-blood expression quantitative trait loci (n ≈ 30,000), genome-wide association study of 731 immune traits (Sardinian cohort, n ≈ 3757), and FinnGen R12 NBS summary statistics (P16_BACTERIAL_SEPSIS_NEWBO; 203 cases, 499,933 controls). Instruments met genome-wide significance, stringent linkage disequilibrium clumping, and F-statistic >10. Primary analyses used inverse-variance weighted MR, with MR-Egger and weighted median as sensitivity analyses. Mediation was estimated as the product of effects. Colocalization (coloc; summary-data-based Mendelian randomization-heterogeneity in dependent instruments) was used to evaluate shared causal variants. Heterogeneity (Cochran Q), MR-Egger intercepts, leave-one-out, and reverse MR were performed. We identified 147 gene expressions and 23 immune traits associated with NBS after false discovery rate correction. Two-step MR yielded multiple gene-immune-NBS mediation chains, with CD28-related T-cell phenotypes consistently prominent. The TRPM4 → CD28 on double-negative T cells → NBS pathway showed the largest mediation proportion (~40.54%) and remained directionally consistent across sensitivity tests. Several colocalized signals (e.g., DBF4B, SNCA, SYCE1L) supported shared variants with immune traits; however, not all corresponding mediation paths passed heterogeneity/sensitivity checks, and are therefore interpreted as suggestive rather than definitive. Associations involving CD64 on monocyte subsets were not robust across sensitivity analyses and were not retained as primary findings. Reverse MR did not indicate feedback from NBS liability to the immune traits prioritized in forward analyses. Genetically informed immune phenotypes (particularly CD28-related T-cell features and the broader CD8bright/regulatory T cell axis) appear to partly mediate the effect of gene expression on NBS susceptibility. Colocalized mediation signals provide supportive, hypothesis-generating evidence but require cautious interpretation given sensitivity/heterogeneity findings. These results motivate neonatal-specific validation and functional studies to refine mechanistic targets for risk stratification.