Metabolic terrain as the keystone: lactate is a stratified, stroma-independent metabolic-immune axis in pancreatic cancer - a multi-omic analysis of 5,132 tumours

Background. Pancreatic ductal adenocarcinoma (PDAC) is refractory to immune-checkpoint blockade, and the metabolic determinants of its immune coldness remain incompletely defined. Tumour-derived lactate suppresses cytotoxic lymphocytes in vitro, but whether this metabolism–immunity antagonism is a universal property of cancers, whether it is independent of the desmoplastic stroma, and which cells carry the machinery, have not been established across human cohorts. Methods. We analysed bulk transcriptomes and overall survival from ten TCGA cohorts (n=5,132) and an independent CPTAC-PAAD cohort (n=134); single-cell RNA-seq from CZ CELLxGENE Census (lung and colorectal carcinoma, ~424,000 cells) and pancreatic tumours (GSE155698; six tumours, 9,731 cells); spatial transcriptomics of PDAC (GSE111672; eight sections, 2,495 spots); and PubMed bibliometrics. Fermentation (glycolysis; lactate LDHA/SLC16A3) and CD8 cytotoxic signatures were compared by Spearman correlation. Survival was modelled by Cox proportional hazards, with adjustment for proliferation (MKI67), immune content (PTPRC) and a cancer-associated-fibroblast (CAF) signature. Results. The metabolism–immunity antagonism was context-specific rather than universal (pooled Spearman r ≈ 0); significant negative associations were confined to immune-cold tumours led by PDAC (r = −0.30). In TCGA-PAAD (n=176) and CPTAC-PAAD, a lactate signature was a reproducible adverse prognostic factor (HR 1.69 and 1.48), whereas the CD8 signature was non-prognostic in both (HR 1.09 and 1.00). A fermentation×CD8 four-group stratification did not improve separation (four-group log-rank p=2.8×10⁻³ vs fermentation alone p=1.6×10⁻³) and the Cox interaction term was null (p=0.88); a cut-point sweep moved the p-value from 0.0004 to 0.40. The lactate effect was independent of desmoplasia: adjusting for a CAF signature left the hazard essentially unchanged (HR 1.69→1.66) while the CAF signature itself was non-prognostic (HR 1.09, p=0.49), and the lactate–CD8 partial correlation strengthened (r −0.26→−0.33). At single-cell resolution the lactate exporter MCT4 was restricted to malignant cells (3.9–8× over CD8 T cells) with tumour-associated macrophages as the second handler, while LDHA was broadly expressed; spatial data confirmed that fermentation co-localises with tumour/ductal regions (mean r=+0.24). Bibliometric mapping showed the metabolism×reversion×dormancy intersection is nearly empty (triple overlap = 2 records). Conclusions. Lactate is a stratified, stroma-independent, tumour-cell-intrinsic metabolic–immune axis in pancreatic cancer. Metabolic normalisation should be pursued as a biomarker-selected sensitising strategy (MCT4/HK2 score) rather than as a universal or CD8-composite approach, and is a rational backbone for a sequenced terrain–reversion–immunity programme anchored at minimal residual disease.