Table S1: Gene Sets; Table S2: Patient Clinical Information; Table S3: Clustering Results; Table S4: Differential expression analysis between clusters (full dataset); Table S5: GSEA results (full dataset); Table S6: Stromal highly variable genes; Table S7: DE analysis of NL-LGD subset; Table S8: GSEA results (NL-LGD subset); Table S9: Validation dataset clustering results

Figure S1: Hematoxylin and eosin (H&E) stained tissue images from patient p10 corresponding to Figure 1B; Figure S2: Quality control of DSP-RNA data; Figure S3: AOI Filtering Criteria; Figure S4: Gene Filtering Criteria; Figure S5: Normalization of DSP-RNA data; Figure S6: Dimension reduction analysis of DSP-RNA data; Figure S7: Immunofluorescent staining of epithelial marker genes; Figure S8: GSEA analysis comparing IPMN clusters to PDAC molecular subtyping studies; Figure S9: Leading edge analysis of GSEA results; Figure S10: Hallmark pathways associated with IPMN transcriptomic states; Figure S11: Analysis of adjacent stromal (PanCK-, CD3-, CD45-) AOIs; Figure S12: GSEA analyses of NL-LGD AOIs; Figure S13: Quality control of external DSP-RNA datasets incorporated for validation; Figure S14: PCA projection of validation datasets; Figure S15: GSEA and GSVA of the Carpenter and colleagues dataset

BackgroundArtificial intelligence (AI) models are being increasingly integrated into clinical care. Moreover, the availability of publicly accessible AI resources makes them attractive to patients seeking clinical information. Little is known regarding the use of large language models as patient resources for navigating major cancer diagnoses. ObjectiveThis study aimed to evaluate the content, readability, and safety of ChatGPT (OpenAI; GPT-4o)-generated responses to common perioperative queries about hepatic, pancreatic, and colon cancers. MethodsA 28-question survey was developed based on frequently asked surgical questions for select malignancies. Surgical oncologists rated ChatGPT-4o-generated responses on a 5-point Likert scale for accuracy, quality, and tangibility. Readability was assessed using the Flesch-Kincaid Reading Grade Level (FKRGL) and Flesch Reading Ease (FRE). Respondents provided free-text comments and reported their comfort with patients using ChatGPT. Survey completion implied consent. ResultsA total of 7 attending surgical oncologists with a median of 7 (IQR 4-13) years in practice completed the survey. Responses received mean scores of 3.5/5 (SD 0.28) for quality, 3.6/5 (SD 0.34) for accuracy, and 3.6/5 (SD 0.29) for tangibility. The responses had a median FKRGL score of 14.6 (IQR 13.3-15.6) and FRE score of 29.4 (IQR 20.5-36.3). On a post hoc analysis for select questions, the median FKRGL was 15.6 (IQR 14.4-16.7), decreasing to 7.1 (IQR 6.1-8.3) and 14.5 (IQR 13.2-15.4) with prompting and rephrasing, and the median FRE was 18.1 (IQR 14.6-24.7), increasing to 73.8 (IQR 66.6-79.3) and 32.0 (IQR 27.0-37.7) with prompting and rephrasing. Numerous inaccuracies and content gaps were reported, and approximately 43% (3/7) of providers did not report feeling “comfortable” in having patients consult publicly available AI for medical information. ConclusionsThis study provides cautionary, yet optimistic, findings regarding the value of publicly accessible ChatGPT as a patient resource for abdominal malignancies. Providers should be prepared to effectively counsel patients to identify their educational attainment level when using ChatGPT to mitigate readability challenges.

PURPOSE:Although KRAS mutations represent the primary oncogenic driver in pancreatic ductal adenocarcinoma (PDAC), the association between codon-specific alterations and patient outcomes remains poorly elucidated, largely because of a lack of data sets coupling genomic profiling with rich clinical annotations across disease stages. MATERIALS AND METHODS:We used American Association for Cancer Research's GENIE Biopharma Consortium Pancreas v1.2 data set to test the association of codon-specific KRAS mutations with clinicogenomic features and patient outcomes in patients with PDAC diagnosed with localized (stages I to III) and advanced disease (stage IV). Overall survival (OS) was compared using Kaplan-Meier and multivariable Cox proportional hazards methods. RESULTS:Among 1,032 eligible patients, 949 (92%) exhibited mutant KRAS. These mutations were predominantly observed at G12D (n = 390, 41%), G12V (n = 305, 32%), and G12R (n = 149, 16%). In the group of patients who presented with localized disease, those with G12V mutation had notably longer survival compared with G12D mutation (P = .03). By contrast, patients with G12V mutation who presented with metastatic disease experienced shorter OS compared with those with G12R (P = .04) and G12D mutations (P = .04). Furthermore, no significant differences were observed in the frequencies of coaltered driver genes, including TP53, CDKN2A, and SMAD4, across the different KRAS mutations. CONCLUSION:These findings demonstrated that codon-specific KRAS mutations affect PDAC outcomes differently based on disease stage at diagnosis. As studies testing KRAS inhibitors continue to emerge and mature, the prognostic variability of individual KRAS mutations must be carefully considered to avoid confounding and ensure accurate evaluation of therapeutic efficacy in early-phase studies.

BACKGROUND:Hepatopancreatobiliary (HPB) malignancies are associated with high morbidity and frequent invasive procedures near the end of life. Although palliative care improves quality of life and goal-concordant care, its relationship with procedural burden in HPB cancers at the end of life remains poorly understood. We evaluated associations between inpatient palliative care and end-of-life procedural intensity in HPB malignancies. METHODS:We conducted a retrospective analysis of adult patients with HPB cancers admitted to a tertiary academic health system within 6 months of death between January 1, 2013, and December 31, 2024. Electronic health records and tumor registry data were analyzed through June 1, 2025. Patients were categorized by inpatient procedures and palliative care consultations. The primary outcome was the relationship between palliative care involvement and inpatient procedural burden. Multivariable logistic regression evaluated associations between palliative care timing and procedural characteristics. RESULTS:Among 2,848 patients, 1,373 (48.2%) underwent an inpatient procedure and 1,160 (40.7%) received palliative care consultation. From 2013 to 2024, palliative care utilization more than doubled, while end-of-life procedural rates remained constant. Palliative care involvement prior to any procedure was associated with lower odds of undergoing nonpalliative-intent procedures relative to palliative-intent procedures (odds ratio [OR] 0.71, 95% confidence interval [CI] 0.51-0.98; P = 0.038). CONCLUSIONS:End-of-life inpatient procedures were common in HPB malignancies and persisted despite increased palliative care utilization. Early palliative care involvement was associated with a greater proportion of palliative-type procedures, suggesting an association between earlier palliative care integration and procedural decision-making that may better align with patient quality of life and end-of-life priorities.

Background Gallbladder cancer (GBC) portends a poor prognosis, and guidelines recommend radical cholecystectomy for patients with T1b and greater disease. We evaluated prognostic factors for T2 and T3 GBC among a contemporary cohort. Methods The National Cancer Database (NCDB) was queried for patients with resected pathologic T2 and T3 GBC from 2004 to 2018. Patients were stratified by extent of resection: simple (SC) vs radical cholecystectomy (RC). Overall survival (OS) was compared with Kaplan-Meier and multivariable Cox proportional hazards methods. Results Altogether, 10,107 patients (6,426 [63.6%] T2; 3,681 [36.4%] T3) were identified: 2,203 (21.8%) underwent SC and 7,904 (78.2%) RC. Patients with SC were more likely to have T2 disease, negative nodes, and positive margins. Tumor grade, node positivity, margin positivity, and T3 stage were associated with greater mortality. Treatment at an academic center and receipt of adjuvant chemotherapy were associated with improved survival. Extent of surgical resection was not associated with OS in unadjusted and adjusted analyses (adjusted HR 0.98, 95% CI 0.92-1.05, p=0.62). Discussion Outcomes for patients with resected T2 and T3 GBC remain poor, though these retrospective data suggest RC is not associated with better OS. Multidisciplinary management and prospective investigation are needed to advance outcomes and facilitate selection of patients who may benefit from radical resection.

BACKGROUND:The benefit of adjuvant chemotherapy (AC) for ampullary adenocarcinoma is unclear. The Hidden Genome model classifies prognostic subtypes with greater accuracy than standard histologic classification (intestinal [INT] vs pancreatobiliary [PB]), but its predictive capacity to guide the use of AC remains unstudied. METHODS:We applied the Hidden Genome model to an international cohort of 183 patients with resected ampullary adenocarcinoma who underwent genomic sequencing. The model quantified the predicted cell of origin (colorectal vs pancreas/distal bile duct) in all specimens. Overall survival (OS) was compared using Kaplan-Meier estimates, stratified by AC use versus surgery alone (SA). RESULTS:Most patients (n=128; 69.9%) received AC, which was not associated with a significant improvement in OS (median, 50.9 months [95% CI, 36.5-76.9] vs 53.8 months [95% CI, 32.4-119.0]; P=.816). Histologic subtype was neither associated with prognosis (P=.241) nor predictive of chemotherapy efficacy for INT-subtype (P=.379) or PB-subtype (P=.544) tumors. When stratified by genomic subtype, the colorectal group had a favorable prognosis regardless of AC use (median OS, 74.4 months [95% CI, 33.8-97.8] for AC vs 98.7 months [95% CI, 32.4-140.9] for SA; P=.889). Among patients with pancreas/distal bile duct tumors, those who received AC had longer survival compared with those who underwent SA (78.2 months [9.8-not reached] vs 22.7 months [2.3-not reached], respectively; hazard ratio, 0.17 [95% CI, 0.04-0.80]; P=.024). CONCLUSIONS:AC regimens were not associated with improved survival in histologically defined INT- or PB-subtype ampullary adenocarcinoma. However, genomic classification better stratified risk groups and identified patients more likely to benefit from AC.

772 Background: The therapeutic resistance of pancreatic ductal adenocarcinoma (PDAC) to conventional cytotoxic therapy underscores the need for advances in targeted molecular therapeutics. Previous studies have reported approximately 2% of PDAC patients with HER2 gene amplification. While anti- HER2 cancer therapy has yielded success in other disease sites, it has had minimal reported benefit in PDAC. Multiple HER2 isoforms exist that are generated either by loss of exon 16 ( d16 HER2 ), or through N-terminal truncations ( p95 HER2 ). These isoforms are implicated in differential tumor behavior and response to anti-HER2 therapy in the breast but have not been investigated in PDAC. Herein, we hypothesize that these isoforms may underlie the observed resistance to HER2 therapy in PDAC. Methods: To test for the presence of HER2 structural variants within human PDAC, a tissue microarray was constructed of fifty-one primary tumor samples and two metastatic samples. Multiplex immunohistochemistry (mIHC) was performed for staining of the extracellular N-terminal and intracellular C-terminal HER2 domains. We then utilized our previously described HER2 + cancer rainbow (Crainbow) mouse model in which human wild-type HER2 ( WT HER2 ), d16 HER2 , and p95 HER2 were expressed in the same mouse along with fluorescent protein reporters. HER2 Crainbow mice were crossed with PDX1-Cre recombinase to initiate recombination and expression of each fluorescently labelled HER2 isoform during pancreas development. PDX1-Cre / HER2 Crainbow mice were followed for up to one year and sacrificed to lineage trace each isoform and register HER2 lineages with histopathology. Results: In our human data, four of fifty-three patient samples (three primary and one metastatic) had moderate-to-strong HER2 staining. All samples identified on mIHC as HER2 positive stained primarily for the p95 variant of HER2 , with absence of the N-terminal domain and preservation of the intracellular domain. In our transgenic mice, 65% of the mouse cohort developed pre-malignant low grade pancreatic intraepithelial neoplasia (PanIN) by eight weeks. At one year, 58.3% of mice demonstrated advanced high grade PanIN, and 25% developed invasive disease. Lineage tracing revealed that while early low grade PanIN began as polyclonal lesions, comprised of WT- , d16- , and p95 HER2 expressing cells, advanced PanIN and invasive disease were unanimously monoclonal with one dominant isoform arising. WT HER2 was represented abundantly in low grade disease, but high grade and invasive disease were predominantly driven by d16- and p95-HER2. Conclusions: Our data implicates the oncogenic d16- and p95 HER2 variants, rather than WT HER2 , as the primary drivers of HER2+ PDAC tumorigenesis. In breast p95 HER2 has been described as a known cause of Trastuzumab resistance. Therefore, future efforts will be aimed at characterizing the mechanisms driving therapeutic resistance in p95-HER2 PDAC tumors.

Pancreatic ductal adenocarcinoma (PDAC) is characterized by exceedingly high rates of metastatic progression, with the liver representing the most common site of distant spread. Here, we established a platform for multisite immune profiling of human PDAC encompassing the tumor, peripheral circulation, and premetastatic liver, to more comprehensively study how various immune subsets might contribute to patient outcomes. Tumor, liver, and blood samples were obtained from patients undergoing resection for non-metastatic PDAC. Derived immune cells underwent paired single-cell RNA and TCR sequencing. Immune composition, cell-type functional profiles, and T cell clonal expansion patterns were evaluated across tissue sites. In total, 106,539 immune cells were sequenced, of which 85,748 met criteria for analysis. We identified 32 cell populations, of which seven demonstrated significant enrichment within a particular tissue, highlighting that this workflow possesses the granularity needed for identifying potential future biomarkers. Functional profiling revealed tissue-specific differences in cell phenotypes. This included terminally differentiated exhausted CD8 T cells within the tumor, highly active Tregs within the premetastatic liver and tumor, and M1 versus M2 polarization of liver and tumor macrophage populations, respectively. Within the tumor, expanded Treg clones were uniquely abundant, and while expanded clones could be tracked to the blood and premetastatic liver, many of these mapped back to known viral antigens. Leveraging previously validated gene sets, we show how these can be applied to predict the tumor reactivity of intratumoral T cells using transcriptional signatures. We demonstrated a high degree of concordance between multiple independent signatures and tracked high-priority TCRs within the blood and liver. This study demonstrates the feasibility of a platform, which has already been implemented into ongoing clinical protocols, for immune profiling of human PDAC across the sites most relevant to metastatic progression. Future applications of this work can monitor immune populations throughout metastatic progression to build a temporal database of immune phenotypes and track association with clinical outcomes.

PURPOSE:Intraductal papillary mucinous neoplasms (IPMN) occur in 5% to 10% of the population, but only a small minority progress to pancreatic ductal adenocarcinoma (PDAC). The lack of accurate predictors of high-risk disease leads to both unnecessary operations for indolent neoplasms and missed diagnoses of PDAC. Digital spatial RNA profiling (DSP-RNA) provides an opportunity to define and associate transcriptomic states with cancer risk. EXPERIMENTAL DESIGN:We performed whole-transcriptome DSP-RNA profiling on 10 IPMN specimens encompassing the spectrum of dysplastic changes from normal duct to cancer. Epithelial regions within each tissue were annotated as normal duct, low-grade dysplasia, high-grade dysplasia, or invasive carcinoma. The resulting digital gene expression data were analyzed with R/Bioconductor. RESULTS:Our analysis uncovered three distinct epithelial transcriptomic states-"normal-like" (cNL), "low risk" (cLR), and "high risk" (cHR)-which were significantly associated with pathologic grade. Furthermore, the three states were significantly correlated with the exocrine, classical, and basal-like molecular subtypes described in PDAC. Specifically, exocrine function diminished in cHR, classical activation distinguished neoplasia (cLR and cHR) from cNL, and basal-like genes were specifically upregulated in cHR. Intriguingly, markers of cHR were detected in normal duct and low-grade dysplasia regions from specimens with PDAC but not from specimens containing only low-grade IPMN. CONCLUSIONS:DSP-RNA of IPMN revealed low-risk (indolent) and high-risk (malignant) expression programs that correlated with the activity of exocrine and basal-like PDAC signatures, respectively, and distinguished pathologically low-grade specimens from malignant specimens. These findings contextualize IPMN pathogenesis and have the potential to improve risk stratification.

Pancreatic ductal adenocarcinoma (PDAC) has the highest mortality rate among major cancers, with over 50% of patients having liver metastases at diagnosis. Given that conventional T-cell based immunotherapies have been ineffective against PDAC liver metastases, it is vital to elucidate the mechanisms regulating tumor immunity within the hepatic microenvironment to enhance anti-tumor T cell. Notably, PDAC cells exhibit exceptional plasticity and intrinsic heterogeneity, contributing to the evolution of immunosuppressive primary and early liver PDAC metastases. Our study leverages an array of congenic PDAC clones with varying degrees of immunogenicity to delineate the signaling pathways driving primary and liver immunosuppression. We investigate the growth kinetics of PDAC clones both in vitro and in vivo, alongside their transcriptomic variations. Using model antigens (chicken ovalbumin), we assess how different clones modulate the immune microenvironment to elicit antigen-specific adaptive anti-tumor responses. Transcriptomic profiling informs CRISPR-based knockout and overexpression strategies to uncover intrinsic gene contributions to primary anti-tumor immunity in vivo. Additionally, we have developed stable PDAC liver metastasis models through portal vein injection to evaluate each clone’s metastatic efficiency and associated immune landscape within the hepatic microenvironment. In vivo engraftment of PDAC clones revealed differential infiltration of CD4+/CD8+ T cells and neutrophils. While all OVA+ PDAC clones triggered antigen-specific immune responses, only immune-sensitive tumors were eradicated, whereas immune-insensitive tumors persisted. Despite OVA expression, all clones successfully established liver metastases following intraportal injection. RNA-seq analysis identified upregulation of key pancreatic development regulators, including PDX1 and RBPJ, in immune-insensitive clones. CRISPR knockout of these transcription factors in vivo demonstrated altered immune cell composition and reduced tumor growth in immunocompetent mice expressing OVA. Using diverse PDAC models, we created tumors with immune heterogeneity and altered transcriptional profiles. We identified validated immune-altering cytokines (CXCL1 and CSF3) and immunomodulatory roles of PDX1 and RBPJ. Immune-insensitive clones evaded antigen-specific immunity in both subcutaneous and liver metastasis models, while immune-sensitive clones were eradicated subcutaneously but gained protection in the liver. Ongoing studies confirm the phenotypes of CRISPR-KO clones and delineate the immunomodulatory roles of these transcription factors in liver metastasis. Bushangqing Liu, Ethan Agritelley, Junping Wei, Gangjun Lei, Tao Wang, Jason McBane, Daniel Nussbaum, Zachary Hartman. Decoding the role of intrinsic pancreatic cellular signaling in shaping the immunosuppressive tumor microenvironment [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 908.

Approximately 50% of colorectal cancer patients develop liver metastases. Hepatic metastases represent the most common cause of colorectal cancer-related mortality. Metastasectomy, if possible, represents the most effective treatment strategy; 20% of patients will be cured and more than 50% survive at least 5 years. Nuances to treatment planning hinge on whether patients present with resectable disease upfront, whether the future liver remnant is adequate, and whether the primary tumor, if present, is colon versus rectal in origin. This article discusses considerations impacting our approach to patients with colorectal liver metastases and the role for various multimodal treatment options.

Intraductal papillary mucinous neoplasms (IPMN) are precancerous cysts representing the only radiographically identifiable precursor to pancreatic ductal adenocarcinoma (PDAC). Successful interception of PDAC requires identifying high-risk disease, which accounts for approximately 15-20% of IPMN. Currently, this is only achieved by detecting high-grade dysplasia (HGD) in the epithelial lining of the cyst, which is challenging due to the heterogeneity of the disease. Minute foci of HGD can reside adjacent to abundant low-grade dysplasia (LGD), making it nearly impossible to detect HGD without histopathological review of a surgically resected lesion. Additionally, IPMN-associated PDAC is often detected only after surgical removal. The heterogeneity of these lesions has hindered efforts to accurately profile pure HGD or invasive regions, leading to unsuccessful attempts to identify diagnostic biomarkers. The predominant pathology in many IPMN lesions is LGD. Therefore, we hypothesized that if there are differences between the low-grade regions of low-risk, high-risk, and invasive IPMN, we could identify a diagnostic biomarker without the need to detect HGD or PDAC specifically. While single-cell RNA-sequencing (scRNA-seq) presents an attractive approach for characterizing IPMNs, it has been limited by the inability to sequence IPMN samples with confirmed pathology. Recently, 10X Genomics introduced a scRNA-seq method for formalin-fixed, paraffin-embedded (FFPE) tissues, mitigating the challenge of sequencing these highly heterogeneous samples prior to histological annotation. We employed this scRNA-seq assay on 13 FFPE tissue blocks confirmed to harbor LGD epithelium only. These blocks were derived from five patients with only LGD within the entire specimen (low-risk), five patients who had focal HGD detected in other blocks from the same specimen (high-risk), and four patients who had invasive cancer detected in other blocks. We also included two samples containing LGD, HGD, and PDAC within the same specimen for comparative analysis. Overall, we profiled 90,950 cells, the largest known IPMN single-cell sequencing effort to date. We identified a cell population indicative of acinar-to-ductal metaplasia (ADM) in low-risk samples that disappeared through disease progression. Moreover, we discovered that populations of immune cells differed between the low-grade lesions of low-risk, high-risk, and invasive IPMN. For instance, low-risk samples had the highest proportion of effector memory CD8+ T cells, high-risk samples had the highest proportion of pro-inflammatory macrophages, and invasive samples had the highest proportion of B cells. Furthermore, we analyzed inferred cell–cell interactions and identified differences in signaling strength between distinct ductal, stromal, and immune cell populations across these sample groups. These results demonstrate the first instance of resolving and comparing the microenvironment surrounding LGD lesions in low-risk, high-risk, and invasive IPMN. Ashley A. Fletcher, Erika J. Crosby, Daniel P. Nussbaum, Margaret O'Connor, Elishama N. Kanu, Peter J. Allen. scRNA-seq from FFPE tissue resolves the tumor microenvironment of intraductal papillary mucinous neoplasms [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Advances in Pancreatic Cancer Research—Emerging Science Driving Transformative Solutions; Boston, MA; 2025 Sep 28-Oct 1; Boston, MA. Philadelphia (PA): AACR; Cancer Res 2025;85(18_Suppl_3):Abstract nr B074.

BACKGROUND:Veterans Affairs (VA) patients participate within the largest USA equal-access healthcare system, though still experience cancer mortality disparities by social determinants. The Social Vulnerability Index (SVI)-measuring community poverty, minority status, housing, and transportation-has delineated gradients in cancer care utilization and outcomes. We therefore assessed relationships between SVI and mortality in veterans with pancreatic ductal adenocarcinoma (PDAC). PATIENTS AND METHODS:Records of patients with stage I-III PDAC diagnosed January 2005-December 2018 were abstracted from the VA Corporate Data Warehouse. Addresses were geocoded to SVI. Mann-Whitney U and chi-squared tests compared baseline characteristics. Overall survival (OS) was compared using Kaplan-Meier and multivariable Cox proportional hazards methods. RESULTS:A total of 1778 patients were evaluated, of which 570 (32%) underwent surgical resection. A total of 516 (29%) patients originated from census tracts in the highest SVI quartile. These patients were more likely Black (38% vs 13%), unemployed (49% vs 35%), and metropolitan-based (87% vs 76%; all p < 0.001). Over a median follow-up of 8.7 months, median OS was 9.5 months versus 10.2 months for high and low SVI groups, respectively (log-rank p = 0.03). In the resected cohort, high and low SVI groups reflected a median OS of 18.2 months versus 23.1 months, respectively (log-rank p < 0.01). High SVI remained independently associated with mortality for resected patients only (hazard ratio [HR] 1.4, p < 0.01). CONCLUSIONS:High SVI was associated with poorer prognosis after PDAC resection. Acknowledging a need to identify mechanisms, SVI has potential use in triaging PDAC survivorship interventions in veterans.