Pancreatic ductal adenocarcinoma (PDAC) is characterized by extensive desmoplasia, which challenges the molecular analyses of bulk tumor samples. Here we FACS-purified epithelial cells from human PDAC and normal pancreas and derived their genome-wide transcriptome and DNA methylome landscapes. Clustering based on DNA methylation revealed two distinct PDAC groups displaying different methylation patterns at regions encoding repeat elements. Methylation^low tumors are characterized by higher expression of endogenous retroviral transcripts and double-stranded RNA sensors, which lead to a cell-intrinsic activation of an interferon signature (IFNsign). This results in a protumorigenic microenvironment and poor patient outcome. Methylation^low/IFNsign^high and Methylation^high/IFNsign^low PDAC cells preserve lineage traits, respective of normal ductal or acinar pancreatic cells. Moreover, ductal-derived Kras^G12D/Trp53^−/− mouse PDACs show higher expression of IFNsign compared with acinar-derived counterparts. Collectively, our data point to two different origins and etiologies of human PDACs, with the aggressive Methylation^low/IFNsign^high subtype potentially targetable by agents blocking intrinsic IFN signaling. Significance: The mutational landscapes of PDAC alone cannot explain the observed interpatient heterogeneity. We identified two PDAC subtypes characterized by differential DNA methylation, preserving traits from normal ductal/acinar cells associated with IFN signaling. Our work suggests that epigenetic traits and the cell of origin contribute to PDAC heterogeneity. This article is highlighted in the In This Issue feature, p. 521

Artificial intelligence models using digital histopathology slides stained with hematoxylin and eosin offer promising, tissue-preserving diagnostic tools for patients with cancer. Despite their advantages, their clinical utility in real-world settings remains unproven. Assessing EGFR mutations in lung adenocarcinoma demands rapid, accurate and cost-effective tests that preserve tissue for genomic sequencing. PCR-based assays provide rapid results but with reduced accuracy compared with next-generation sequencing and require additional tissue. Computational biomarkers leveraging modern foundation models can address these limitations. Here we assembled a large international clinical dataset of digital lung adenocarcinoma slides (N = 8,461) to develop a computational EGFR biomarker. Our model fine-tunes an open-source foundation model, improving task-specific performance with out-of-center generalization and clinical-grade accuracy on primary and metastatic specimens (mean area under the curve: internal 0.847, external 0.870). To evaluate real-world clinical translation, we conducted a prospective silent trial of the biomarker on primary samples, achieving an area under the curve of 0.890. The artificial-intelligence-assisted workflow reduced the number of rapid molecular tests needed by up to 43% while maintaining the current clinical standard performance. Our retrospective and prospective analyses demonstrate the real-world clinical utility of a computational pathology biomarker.

Homologous repair deficiency (HRD) is present in many cancer types at variable prevalence and can indicate response to platinum-based chemotherapy and PARP inhibition. We developed a tumor classification system based on the loss of function of genes in the homologous recombination repair (HRR) pathway. To this end, somatic and germline alterations inBRCA1/2and 140 other HRR genes were included and assessed for the impact on gene function. Additionally, information on the allelic hit type and onBRCA1promoter hypermethylation was included. The HRDsum score including LOH, LST, and TAI was calculated for 8847 tumors of the TCGA cohort starting from genotyping data and for the subcohort of ovarian cancer also starting from WES data. Pan-cancer, deleteriousBRCA1/2alterations were detected in 4% of the tumors, while 18% of the tumors were HRD-positive (HRDsum ≥ 42). Across 33 cancer types, bothBRCA1/2alterations and HRD-positivity were most prevalent in ovarian cancer (20% and 69%). Pan-cancer, tumors with biallelic deleterious alterations inBRCA1/2were separated strongly from tumors without relevant alterations (AUC = 0.89), while separation for tumors with monoallelic deleteriousBRCA1/2alterations was weak (AUC = 0.53). Tumors with biallelic deleterious alterations in other HHR genes were separated moderately from tumors without relevant alterations (AUC = 0.63), while separation for tumors with such monoallelic alterations was weaker (AUC = 0.57). In ovarian cancer, HRDsum scores calculated from WES data correlated strongly with HRDsum scores calculated from genotyping data (R = 0.87) and were slightly (4%) higher. We comprehensively analyzed HRD scores and their association with mutations in HRR genes in common cancer types. Our study identifies important parameters influencing HRD measurement and argues for an integration of HRDsum score with specific mutational profiles.

PURPOSE: Rapidly expanding medical literature challenges oncologists seeking targeted cancer therapies. General-purpose large language models (LLMs) lack domain-specific knowledge, limiting their clinical utility. This study introduces the LLM system Medical Evidence Retrieval and Data Integration for Tailored Healthcare (MEREDITH), designed to support treatment recommendations in precision oncology. Built on Google's Gemini Pro LLM, MEREDITH uses retrieval-augmented generation and chain of thought . METHODS: We evaluated MEREDITH on 10 publicly available fictional oncology cases with iterative feedback from a molecular tumor board (MTB) at a major German cancer center. Initially limited to PubMed -indexed literature (draft system), MEREDITH was enhanced to incorporate clinical studies on drug response within the specific tumor type, trial databases, drug approval status, and oncologic guidelines. The MTB provided a benchmark with manually curated treatment recommendations and assessed the clinical relevance of LLM-generated options (qualitative assessment). We measured semantic cosine similarity between LLM suggestions and clinician responses (quantitative assessment). RESULTS: MEREDITH identified a broader range of treatment options (median 4) compared with MTB experts (median 2). These options included therapies on the basis of preclinical data and combination treatments, expanding the treatment possibilities for consideration by the MTB. This broader approach was achieved by incorporating a curated medical data set that contextualized molecular targetability. Mirroring the approach MTB experts use to evaluate MTB cases improved the LLM's ability to generate relevant suggestions. This is supported by high concordance between LLM suggestions and expert recommendations (94.7% for the enhanced system) and a significant increase in semantic similarity from the draft to the enhanced system (from 0.71 to 0.76, P = .01). CONCLUSION: Expert feedback and domain-specific data augment LLM performance. Future research should investigate responsible LLM integration into real-world clinical workflows.