Public-data multi-omic study maps pancreatic cancer–stroke gene overlap

A multi-omic analysis of public human datasets found that pancreatic ductal adenocarcinoma (PDAC) and ischemic stroke share a directionally aligned transcriptional signal, but the overlap is organized by cellular compartment rather than forming a uniform blood-based disease signature. The most reproducible component centered on the epithelial genes CEACAM6, ANXA3 and IL1RAP, while TREM1 and IL1R2 marked a more variable myeloid program and F5 represented a context-dependent coagulation-associated component. The study did not recruit participants or perform new laboratory experiments; it integrated transcriptomic, proteomic, spatial, clinical and human genetic data to test the reproducibility, cellular origins and interpretive limits of the apparent overlap.

Researchers first modeled three ischemic stroke cohorts and two paired pancreatic cancer datasets separately before combining standardized effects through random-effects meta-analysis. The stroke comparisons included blood sampled at different stages, while the pancreatic cancer cohorts compared patient-matched tumor and normal tissue. Among 15,221 genes evaluated in both disease-level analyses, 230 were significant in both, 126 changed in the same direction, and 46 passed subsequent replication and effect-size filters.

Six genes were selected as sentinels because they represented distinct epithelial, myeloid and coagulation-associated biology and different patterns of supporting evidence, rather than because they formed a statistically exclusive top-six ranking. This staged design preserved the different tissue sources, sampling times and cohort structures instead of directly merging expression matrices generated on different platforms. The strongest cross-disease module contained CEACAM6, ANXA3 and IL1RAP.

It was the only predefined module to be significant in both diseases, was positive in all five discovery comparisons and remained significant after adjustment for inferred cell abundance. By contrast, the myeloid or NETosis module and the complement-coagulation module were prominent in stroke but did not reach within-disease false-discovery-rate significance in pancreatic cancer. The distinction matters because it separates a comparatively stable epithelial state from downstream immune and coagulation signals that appear to vary with sampling time, tissue compartment and disease context.

Pathway-level integration also implicated inflammatory and NF-κB signaling, hypoxia and vascular responses, complement-coagulation processes and tissue remodeling, but the cellular analyses showed that these broad responses did not arise from a uniform set of cell types. Single-cell RNA sequencing localized much of the shared signal to distinct cell populations. In one resource containing 142,353 initially filtered cells from pancreatic cancer tissue, adjacent pancreas and blood, quality control retained 112,587 cells.

CEACAM6 was concentrated predominantly in pancreatic cancer epithelial cells, whereas TREM1 and IL1R2 were associated with myeloid and neutrophil populations. IL1RAP and ANXA3 appeared in both epithelial and granulocytic compartments, while F5 was detected in blood monocytes and tissue neutrophils in that dataset. An independent reanalysis of 57,423 cells from 24 pancreatic cancer and 11 normal donors reproduced ductal localization of CEACAM6 and ANXA3, myeloid localization of TREM1 and the mixed distribution of IL1RAP.

These results establish where candidate genes were expressed, but no candidate remained significant after correction across all tested cell subtypes, so they do not constitute a patient-level differential-expression signal on their own. Cell-composition analyses further separated stable and variable components. Detection of the principal epithelial, lymphoid, monocyte or macrophage and neutrophil compartments was distributed across multiple patients rather than being confined to a few samples.

A patient-paired analysis of malignant and non-malignant epithelial cells included six patients with enough cells of both types to qualify. ANXA3, CEACAM6, IL1RAP and IL1R2 generally had higher detection fractions in malignant cells, but the leading candidate-level false-discovery rate was 0.0518, just above the study’s significance threshold. The authors accordingly treated this pattern as exploratory evidence rather than statistically significant malignant-cell enrichment.

Computational interaction analysis also identified expression-compatible CEACAM and IL-1 routes, including epithelial CEACAM interfaces and myeloid or neutrophil interactions involving IL1R1–IL1RAP and IL1R2 complexes. These scores did not demonstrate ligand secretion, receptor activation, downstream signaling or directional communication, and formal ligand-target inference could not be evaluated because the required malignant-cell receiver gene set was absent. Independent transcript and protein measurements prioritized four of the six sentinels.

An external pancreatic cancer cohort containing 145 primary tumors and 46 normal pancreata significantly reproduced the tumor-associated direction of all six genes. The log2 fold changes were 3.22 for CEACAM6, 0.56 for IL1RAP, 0.95 for ANXA3, 0.90 for TREM1, 0.81 for IL1R2 and 0.77 for F5. These values could not be compared directly with the discovery results because the discovery analysis reported standardized effects while the external study reported log2 fold changes.

In paired CPTAC proteomics, 36 of the 46 transcript-selected candidates were quantifiable and 34 showed concordant transcript and protein directions. Twenty-six candidates reached proteome-wide false-discovery-rate significance. CEACAM6, ANXA3, IL1RAP and TREM1 had significant positive protein effects, making them the four candidates with the most consistent multi-omic support; F5 was positive but not significant, and IL1R2 was not quantifiable.

Spatial data offered partial rather than complete confirmation of the proposed architecture. GeoMx profiling covered 543 quality-controlled areas of illumination from 19 patients and separated epithelial, immune and cancer-associated fibroblast compartments. The panel included ANXA3 but not CEACAM6 or IL1RAP, so it could not test the entire epithelial bridge.

ANXA3-associated epithelial scores were higher in epithelial regions than in immune or fibroblast regions, while a 14-gene NETosis module was enriched in immune compartments. These differences persisted in 13 untreated patients. A second spatial dataset supplied gene-level summaries from 30 sections but lacked a reconstructable spot-level matrix and enough metadata to recover full patient-level independence.

It showed greater spatial organization of the NETosis module in primary pancreatic cancer, although mean module expression did not differ significantly from normal pancreas. Neither dataset established direct epithelial-myeloid proximity or functional communication. Transportability to an independent stroke setting was limited.

A small external dataset included five acute cases, five subacute cases and five controls. CEACAM6 was the only sentinel with a concordant acute effect, while the other five genes were reversed in the acute comparison and all six were reversed in the subacute comparison. Several reversed effects were statistically significant, but the groups were very small and showed substantial within-group variation.

The result therefore indicates that the six-gene subset is not a universal peripheral-blood classifier for stroke. Differences in sampling time, blood-cell composition, stroke characteristics and systemic inflammatory state may all contribute, but the study did not establish which factor caused the discordance. Clinical and genetic analyses did not extend the shared expression signal into prognostic or causal claims.

In 178 primary pancreatic tumors from TCGA-PAAD, CEACAM6 and ANXA3 expression correlated positively with tumor purity, supporting their epithelial localization but leaving open whether the signal reflected cellular state or cell proportion. None of the six sentinels was associated with pathological stage after multiple-testing correction, and none reached false-discovery-rate significance in survival models adjusted for age, sex, stage and purity. A preliminary protein-quantitative-trait-locus screen suggested an IL1R2 signal, but it did not replicate with an independent plasma measurement.

Regional colocalization across 5,026 shared variants strongly favored a protein-quantitative-trait-locus-only explanation rather than a shared stroke-associated variant, and no sentinel met the study’s threshold for genetically supported causality. The investigators also noted that protein tractability annotations did not establish that inhibiting a candidate would prevent stroke or benefit pancreatic cancer. The resulting model is therefore a hierarchy rather than a universal thrombo-inflammatory pathway.

CEACAM6, ANXA3 and IL1RAP define a reproducible epithelial component, TREM1 anchors a more variable myeloid component, and IL1R2 and F5 remain more dependent on dataset and modality. The analysis compared pancreatic tumor tissue with predominantly peripheral-blood stroke profiles, did not include a cohort following pancreatic cancer patients to adjudicated stroke, and had modest samples in the malignant-cell and external stroke analyses. Communication and spatial analyses were computational and localization-oriented rather than functional demonstrations, and no prospective cohort established clinical relevance.

The authors therefore characterize CEACAM6, ANXA3, IL1RAP and TREM1 as experimentally prioritized components of a shared compartment-resolved program, not as established causes of stroke or proven therapeutic targets for either disease. The study’s next step would require functional perturbation and prospective clinical cohorts with adjudicated cerebrovascular outcomes.

Subject of Research: Cross-disease multi-omic analysis of epithelial, myeloid and coagulation-associated gene programs in pancreatic ductal adenocarcinoma and ischemic stroke

Article Title: Compartment-resolved epithelial and myeloid programs are shared by pancreatic ductal adenocarcinoma and ischemic stroke

Article References: Lu, H., Qi, X., Zhang, M., Yang, H., & Zou, M. (2026). Compartment-resolved epithelial and myeloid programs are shared by pancreatic ductal adenocarcinoma and ischemic stroke. Frontiers in Cell and Developmental Biology, 14, Article 1969947. https://doi.org/10.3389/fcell.2026.1969947

Image Credits: AI Generated

DOI: 10.3389/fcell.2026.1969947

Keywords: pancreatic ductal adenocarcinoma, ischemic stroke, multi-omics, tumor microenvironment, single-cell RNA sequencing, spatial transcriptomics, epithelial genes, myeloid inflammation, coagulation, CEACAM6, Compartment-resolved, epithelial