GATA6 Expression Distinguishes Classical and Basal-like Subtypes in Advanced Pancreatic Cancer

By 2030, pancreatic ductal adenocarcinoma (PDAC) will become the second leading cause of cancer-related mortality in North America (1). The majority of patients with PDAC present with advanced disease where the mainstay of treatment remains combination chemotherapy. Modified FOLFIRINOX (mFFX) and gemcitabine-nab-paclitaxel (GnP) are the most commonly used regimens resulting in median survival less than 1 year (2, 3). While the need to discover novel approaches is obvious, it is equally important to understand how to select the aforementioned regimens for current patients. There are no randomized data to show superiority of either combination and patient inclusion differences are evident in the two pivotal phase III trials (2, 3). The only molecular predictor of response is prior knowledge of a pathogenic germline variant in a homologous recombination repair gene, which may influence the regimen of choice (4). Other currently targetable genomic variants are uncommon in PDAC.

Gene expression profiling, primarily in resected pancreatic tumors, describes a number of subtypes with considerable overlap, yet presently these do not inform clinical practice (5–7). Collisson and colleagues documented three subtypes (classical, quasimesenchymal, and exocrine-like; ref. 6), Bailey and colleagues four subtypes (immunogenic, progenitor, ADEX, and squamous; ref. 5), and Moffitt and colleagues two subtypes (classical and basal-like; ref. 7). The squamous (Bailey), quasimesenchymal (Collisson), and basal-like (Moffitt) cohorts align well across the classifiers and all three are associated with a poor prognosis in these studies. Despite this, varying tumor cellularity and heterogeneity in clustering methodologies leaves uncertainty as to the most appropriate classifier and furthermore, the clinical application of these subtypes to advanced stage disease is unclear.

In an effort to reconcile and apply existing knowledge, we established the COMPASS trial [Comprehensive Molecular Characterization of Advanced Pancreatic Ductal Adenocarcinomas (PDAC) for Better Treatment Selection: A Prospective Study, NCT NCT02750657]. Unique to this prospective study is the acquisition of tissue prior to chemotherapy in the advanced setting, which then undergoes laser capture microdissection (LCM) to ensure high tumor cellularity. The primary endpoint of feasibility in obtaining a high-quality genome report within 8 weeks in the first 50 patients has been published (8). In this earlier analysis, we determined that a modified Moffitt RNA signature, optimized for use in advanced stage PDAC (classical vs basal-like; Supplementary Fig. S1) may have prognostic impact (8). Furthermore, we found that GATA6, a transcription factor required for normal pancreas development (9), which has been shown to align with the classical subtype, could represent a surrogate marker for classical PDAC (8). Here, we evaluated the modified Moffitt basal-like and classical subtypes together with GATA6 expression on outcomes in patients receiving mFFX or GnP regimens on the expanded COMPASS trial. We further explored specific clinical and pathologic characteristics of the subtypes and evaluated GATA6 as a surrogate biomarker and clinical tool. Post hoc exploratory analyses were performed to seek additional positive biomarkers for the basal-like subtype.

The COMPASS trial is a prospective multi-institutional Canadian cohort study. Patient eligibility for the study has been described previously (8). Briefly, patients require a radiologic or histologic diagnosis of locally advanced or metastatic PDAC, suitable for combination chemotherapy, and must consent to a fresh tumor biopsy prior to treatment start. Biopsies can be taken from the primary lesion or any metastatic sites. Patients must not have had prior treatment for advanced disease. Treatment decisions are at the discretion of their medical oncologist. Response to therapy is assessed using CT and measured using RECIST 1.1. Demographics and treatment details, including subsequent treatments, are prospectively collected using an electronic MEDIDATA database. This report includes all patients enrolled from December 2015 until May 2019 and follow-up censored on August 30, 2019. Patients on this study were enrolled at the Princess Margaret Cancer Centre (Toronto, Ontario, Canada), McGill University Health Centre (MUHC, Montreal, Quebec, Canada), and Kingston General Hospital (Kingston, Ontario, Canada) and the study was conducted in accordance with the Declaration of Helsinki. The COMPASS trial has been approved by participating site Institutional Review Board (University Health Network, Toronto, Ontario, Canada; MUHC Centre for Applied Ethics, Montreal, Quebec, Canada; and Queen's University Health Sciences and Affiliated Teaching Hospitals Research Ethics Board, Kingston, Ontario, Canada); each patient provided written informed consent prior to study entry.

Frozen biospecimens underwent LCM for tumor enrichment. RNA-sequencing (RNA-seq) analysis was performed at the Ontario Institute of Cancer Research (Ontario, Canada) as described previously (10). Briefly, reads were aligned to the human reference genome (hg38) and transcriptome (Ensembl v84) using STAR v.2.5.2a (11). Duplicated reads were marked using Picard v. 1.121 (https://github.com/broadinstitute/picard). Gene expression was calculated in fragments per kilobase of exon per million reads mapped using the cufflinks package v. 2.2.1 (12). A modified Moffitt classification (classical vs basal-like) was applied to each sample with sufficient RNA for analysis (Supplementary Fig. S1). Cut-off threshold levels for GATA6 expression were determined using the maximal χ² method on RECIST response and dichotomized GATA6 expression.

Given our early results, the COMPASS trial was amended (February 1, 2017) to include GATA6 staining using an RNAscope In Situ Hybridization (ISH) Assay (Advanced Cell Diagnostics Inc.). A semiquantitative score was used by the study pathologist (S.E. Fischer; Supplementary Fig. S2A) as reported previously (8). Scoring was applied blinded to results of the modified Moffitt classifier.

To provide more widely applicable diagnostic tests for PDAC subtypes, we optimized a protocol for GATA6 IHC (Supplementary Materials and Methods) using a polyclonal anti-GATA6 antibody from R&D (catalog number AF1700), and secondary antibody from Vector (catalog number VECTABA5000). DAB+ (3,3-diaminobenzidine tetrahydrochloride plus, DAKO, catalog number K3468) was used as chromogen and nuclei were counterstained with Mayer hematoxylin (Supplementary Fig. S2B). To assess the pattern of GATA6 staining across larger tumor regions, we used whole sections (n = 30) from a previously described resection cohort with matched RNA-seq data (10) together with biopsies (n = 41) from the advanced cohort.

In an exploratory analysis, we sought additional clinical markers to aid subtype identification; cytokeratins associated with GATA6 expression were identified from RNA-seq data and further explored by IHC (Supplementary Materials and Methods).

To control for potential bias of manual scorings of both ISH or IHC, we performed image analysis on preannotated tumor sections using image analysis software QuPath v0.1.3 (13). Detection parameters were established on unequivocal GATA6-high versus -low versus -absent tumors and confirmed by the study pathologist. Semiquantitative scores were also predicted from image analysis data using the maximal χ² method.

Qualitative variables were compared by Fisher exact test, and quantitative variables by Wilcoxon rank-sum test for pairwise comparison and the Kruskal–Wallis test for multiple group comparison. All patients receiving at least one cycle of chemotherapy were included in the analysis of overall response rate (ORR). Survival curves were plotted using the Kaplan–Meier method and HRs were calculated using Cox proportional hazard regressions with P values calculated using the Wald statistic. All tests were two-sided. Multiple tests P values were adjusted using Benjamini and Hochberg method (14) for independent tests or Benjamini and Yekutieli method (15) for dependent tests, respectively. Statistical significance was set at P = 0.05. All analyses were conducted in R version 3.2 (16). Spearman correlation coefficients were ascertained for evaluating gene expression. Sensitivity, specificity, and accuracy scores were computed to assess prediction quality.

Between December 30, 2015 and May 30,2019, 250 patients were enrolled and 206 were eligible (Fig. 1, consort). Of these, 195 patients (95%) had enough tissue for RNA analysis and are included in this report. Table 1 shows baseline clinical and pathologic characteristics of those patients. Using the modified Moffitt classifier, 39 (20%) baseline tumor samples were basal-like and 156 (80%) classical. Locally advanced disease at diagnosis was present in 24 (12%) and these cases, in this small subset, were all identified as classical (P = 0.005). Liver metastases were present in 97% of basal-like tumors compared with 69% of classical tumors (which includes the locally advanced cases; P < 0.0001). Although basal-like tumors were more frequent in male patients (P = 0.02) the overall sample size was small. Other characteristics were similar between the groups (Table 1).

Of the 195 patients, 14 (7%) did not receive any chemotherapy and were considered nonevaluable (NE). A further 23 patients (12%) died as a result of rapid functional decline prior to their first scan, of which 19 received only one cycle of chemotherapy; five of these 23 had basal-like PDAC. One patient receiving mFFX did not have measurable disease at enrollment. Accordingly, RECIST response data were available for 157 patients (81%) including 29 patients with basal-like tumors and 128 with classical tumors (Fig. 2A). The ORR in classical PDAC was 33% versus 10% in basal-like PDAC (P = 0.02). The rates of progression by RECIST criteria at first CT image were much higher in basal-like versus classical PDAC (52% vs 16%; P < 0.0001). Figure 2A shows the percentage change in target lesions, demonstrating chemoresistance of the basal-like subtype. In patients treated with mFFX and evaluable for response (n = 91), progression was evident in 60% of basal-like versus 15% of classical PDAC (P = 0.0002; Fig. 2B). The ORR was 29.6% versus 10% in classical versus basal-like PDAC (P = 0.09). One patient in the latter group with a partial response had a germline BRCA2 pathogenic variant and displayed genomic characteristics of homologous recombination deficiency. The numbers treated with GnP regimens and available for response were small, progression of disease was seen in three of nine (33%) patients with basal-like versus eight of 54 (15%) with classical tumors (P = 0.18; Fig. 2C). The ORR was 39% versus 11% in classical versus basal-like PDAC, respectively (P = 0.14). Of note, 20 of 63 (32%) received additional experimental agents in this group.

Overall survival (OS) in the intention-to-treat population (n = 195) is shown in Fig. 3A. Median follow-up was 7.17 months. Median OS according to receipt of chemotherapy is shown in Fig. 3B. In patients receiving mFFX (n = 103), where performance status was less likely to confound results, median OS was 6.5 months in basal-like versus 10.6 months in classical subgroups [HR, 0.33; 95% confidence interval (CI), 0.19–0.60; P = 0.0001). These observations suggest favorable impact of mFFX in classical PDAC but little impact of mFFX in the basal-like population (Fig. 3C). In contrast, there was no difference between subgroups when treated with GnP regimens, where median OS, was 8.12 months in basal-like versus 8.19 months in classical groups, respectively (HR, 0.80; 95% CI, 0.40–1.60; P = 0.53; Fig. 3D). In a multivariate Cox proportional hazard regression analysis, the Moffitt subtype remained highly prognostic (P = 0.018). Substituting GATA6 expression for the Moffit subtype also demonstrated the prognostic impact of GATA6 in the model, again supporting its use as a biomarker of the subtypes. Of note, stage (locally advanced vs metastatic) or chemotherapy type had no impact in this observational cohort study (Supplementary Fig. S3). To further explore whether there was a significant interaction between mFFX or GnP and the subtypes, we performed an interaction analysis. There was no statistically significant difference to suggest one chemotherapy regimen for one particular subtype, although basal-like tumors trended toward improved survival with GnP (P = 0.08). Of note, the modified Moffitt classifier used in this study outperforms the previously published Moffitt classifier in identifying the poor prognostic basal-like subtype (Supplementary Fig. S1B).

GATA6 expression remained strongly associated with the modified Moffitt transcriptomic classifier (P < 0.001) in this expanded cohort (Fig. 4A, left). In addition, the proposed RNA ISH semiquantitative score was highly associated with GATA6 gene expression (RNA-seq; P < 0.001; Fig. 4A, right). Matched RNA-seq and ISH results were available in 106 patients (23 with basal-like and 83 with classical subtypes). Semiquantitative scoring of GATA6 ISH confirmed higher GATA6 expression (74/83 score 2–4) in classical versus basal-like PDAC (19/23 score 0–1). Furthermore, GATA6 ISH correlated with modified Moffitt with a sensitivity of 89%, specificity of 83%, and accuracy of 88%. Both manual scores and modified Moffitt calls could be predicted from image analysis data with concordance of 92% and 81%, respectively, confirming reproducibility of semiquantitative assessment (n = 106). The modified Moffitt signature remained prognostic in this staining subcohort (Fig. 4B); both GATA6 ISH SQ scoring (Fig. 4C) and subtyping inferred from image analysis of GATA6 ISH (Fig. 4D) predicted outcome in a similar manner.

Matched IHC and ISH results were available in only 78 advanced PDAC cases. GATA6 levels by IHC and ISH were well correlated using quantitative assessment (Supplementary Fig. S4A) and semiquantitative scoring (concordance 88%), indicating that GATA6 protein levels mirror RNA expression and could aid subtype identification when RNA detection is not feasible. Indeed, IHC-based semiquantitative scoring identified most patients with classical subtype tumors by strong and moderate GATA6 staining (52/63 with scores 2–4), while basal-like subtype patients mostly exhibited no or weak GATA6 staining (9/15 with scores 0–1), so that GATA6 protein detection by IHC was associated with modified Moffitt subtypes in advanced PDAC with a sensitivity of 83%, specificity of 60%, and accuracy of 78%. Once more, this was confirmed by quantitative assessment (Supplementary Fig. S4B) and the concordance between prediction of GATA6 scoring from image analysis to manual scoring of GATA6 was 90%.

Recent data are emerging that basal and classical subtypes can coexist in PDAC (17, 18). We therefore explored potential variation in GATA6 expression patterns. We used whole sections from selected resected cases (n = 30) in addition to needle biopsies (n = 41). Although early stage tumors may not necessarily reflect the biology of advanced disease, adequate tumor content is available for more complete evaluation. GATA6 staining (IHC) in resected specimens that were basal-like (n = 14) and classical (n = 16) also associated with the Moffitt subtypes: extensive immunopositivity for GATA6 (>50% of tumor cells with score 2 or higher) was found in nine of 16 (56%) classical tumors versus one of 14 (7%) basal-like tumors (Supplementary Table S1). Interestingly, variable GATA6 immunopositivity (<50% of tumor cells with score 2 or higher) was present in four of 16 (25%) and four of 14 (28%) of classical and basal-like tumors, respectively, documenting a group where these subtypes may coexist. This was furthermore observed in a number of advanced PDAC biopsies, which also exhibited variable GATA6 expression by ISH and IHC (Supplementary Fig. S5), demonstrating that regional GATA6 heterogeneity can exist in resectable and advanced-stage tumors. These differences were also observed at the cellular level by image analysis (Supplementary Fig. S6). In sum, GATA6 staining patterns were widely comparable across whole sections of 22 of 30 (73%) resection cases. Variable GATA6 immunopositivity was present in a subset of both, classical and basal-like subtypes, in resectable and advanced disease, which may point at the presence of classical and basal regions in the same tumor.

GATA6 positively identifies classical PDAC, but markers for the basal subtype are lacking.

In an exploratory analysis, we evaluated keratin markers associated with GATA6 expression. In-line with their use as basal markers in other tumor types (19, 20), keratins 15, 5/6, 23, and 14 were inversely correlated with GATA6 expression and thus the classical subtype (Supplementary Fig. S7A). In this post hoc analysis, none of the identified cytokeratins were superior to GATA6 in their association with modified Moffitt subgroups, including keratin 17, a prognostic marker in PDAC (ref. 21; Supplementary Fig. S7B). Among these, keratin 5 (CK5) demonstrated the strongest expression differences between basal-like and classical tumors and was found to be complementary to GATA6 expression in our cohort (Supplementary 8). Furthermore, GATA6 and keratin 5 often demonstrated complementary staining pattern by IHC in PDAC tissues, including in 41 COMPASS biopsies and 30 resected PDAC whole sections (Fig. 5; Supplementary Table S1). From these specimens, we observed the presence of both GATA6 and CK5 staining in a subset of cases (Fig. 5, bottom). Indeed, the intratumoral staining pattern of the two markers was predominantly inversely correlated in 149 individual tumor regions from the 30 resected cases (Fig. 6A). Of note, this analysis revealed a small number of regions that contained considerable number of both CK5⁺ and GATA6⁺ cells (Fig. 6A). Double immunostaining confirmed distinct GATA6⁺/CK5⁻ and GATA6⁻/CK5⁺ regions within the same tumor (Fig. 6B) and in individual ducts (Fig. 6C), which further support the notion that basal-like and classical programs can coexist in the same tumor. Overall, many basal-like cases of advanced PDAC showed CK5 positivity (10/19, 53%), whereas most classical tumors (22/23, 96%) exhibited scant (<10%) or negative CK5 staining. Keratin 5 was thus highly specific and also showed remarkable intratumoral complementarity to GATA6 staining suggesting a clinically relevant biomarker of the basal-like subtype.

Among the 195 eligible COMPASS patients, all eight (4%) with adenosquamous histology were basal-like and stained positive for keratin 5 by IHC, with negligible GATA6 expression by RNA ISH. We have previously shown that the basal-like subgroup is enriched in a hypoxia-associated gene signature by gene set enrichment analysis (22) and this observation is retained in this expanded dataset (P = 0.0003). In addition, we found higher PD-L1 expression in the basal-like cohort (P < 0.001), higher PD-1 expression (P < 0.001), and enrichment of a T-cell–inflamed signature reported previously (refs. 23, 24; P = 0.007; Supplementary Fig. S9). Tumor mutational burden was not different between groups (2.02 mutations/Mb vs 1.96 mutations/Mb) and was consistent with that seen in an unselected PDAC cohort (25).

Combination chemotherapy is used in the treatment of most patients with advanced PDAC, yet the field is lacking robust biomarkers of outcome to guide regimen selection. Here, we show that patients with tumors of a modified “basal-like” phenotype, or those with low GATA6 RNA expression, have inferior outcomes compared with those with the “classical” phenotype. The latter are accurately identified by high GATA6 expression and positive GATA6 staining by ISH. Our data also suggests that basal-like tumors are particularly resistant to mFFX, warranting further investigation.

Both the PRODIGE4/ACCORD 11 and the MPACT PDAC trials of metastatic disease demonstrated an improvement in survival with FOLFIRINOX and GnP across all subcohorts compared with gemcitabine alone (2, 3), yet they provide little insight into which subgroups might benefit the most. Notably, our study shows no superiority in either regimen in an unselected population with regard to survival. In the aforementioned trials, histologic groups were not documented in either study, which is not unusual because many patients have a diagnosis made from very small samples or brushings. In contrast, the histologic classification in resected specimens can be more easily reported and the PRODIGE24/CCTG PA6 trial of mFFX in the adjuvant setting documented the prognostic impact of tumor grade in multivariate analysis (26) In patients receiving mFFX, those with well-differentiated tumors benefited the most (HR, 0.52; 95% CI, 0.34–0.81), whereas the impact in poorly differentiated tumors was not significant. Although limitations to the three-tiered histologic classification (poor, moderate, and well-differentiated) in PDAC have been noted (27), well-differentiated tumors highly express the classical program and GATA6 (28).

The resistance of the basal-like subtype to mFFX is supported by a recent collaborative study by Tiriac and colleagues (29) demonstrating that patient-derived organoid chemotherapy signatures may predict treatment response. The signatures indicative of individual cytotoxic agents were applied to our COMPASS cohort suggesting that the basal-like cohort subgroup was most likely to have a nonoxaliplatin-sensitive signature (29). We furthermore hypothesize that basal-like tumors may have limited sensitivity to 5-fluorouracil (5-FU). Martinelli and colleagues demonstrated GATA6 loss in resected PDAC with a basal-like phenotype in the ESPAC-3 trial, and shorter survival in these patients when treated with adjuvant 5-FU. This study also showed that GATA6-low cell lines derived from patient-derived xenografts were particularly resistant to 5-FU but not gemcitabine (30). Notably oxaliplatin was not evaluated. In search of treatment alternatives, we report here that basal-like tumors had higher hypoxia scores, and higher PD-1 and PD-L1 expression with enrichment of a T-cell–inflamed signature (24), which may be predictive of immunotherapy 6 in this chemoresistant group. Similarly, triple-negative breast cancers, although associated with worse outcomes, have higher levels of tumor-infiltrating lymphocytes compared with hormone receptor–positive, HER2⁻ tumors. The impact of immune populations within subtypes in PDAC will require further investigation (31).

Clinical applicability of RNA-seq and tumor enrichment by LCM is currently limited given tissue acquisition, cost, and time to reporting. GATA6 detection from formalin-fixed, paraffin-embedded needle biopsies at diagnosis is therefore an attractive surrogate for transcriptomic classifiers. We demonstrate concordance of GATA6 ISH with the subtypes with sensitivity and specificity of over 80% in our tumor-enriched samples. Of note, the GATA6 gene is not part of the original Moffitt subtype signatures but rather the Bailey squamous classifier, which largely overlaps with Moffitt calls in high purity samples (5). The number of tissue specimens available for matched ISH, IHC, and RNA-seq was low in our study (n = 78, 40%). Therefore, although specificity was much lower for IHC compared with ISH, a prospective study with adequate tissue for matched analysis is needed. Recognizing that the identification of the basal-like subtype is critical and that GATA6 is a negative marker, we sought additional positive keratin biomarkers that may be more feasible for the practicing clinician. Of these, keratin 5 predicted outcomes best after GATA6 expression and was found to exhibit high complementarity to GATA6 staining pattern and RNA expression levels. Moreover, combined keratin 5 and GATA6 stainings on serial sections and by double immune-staining have consistently suggested that basal-like and classical elements can coexist in a subset of PDAC cases, which strongly reinforces the need for a positive basal-like biomarker and has major implications for rationalizing subtype-specific treatments. We are currently evaluating combined staining of GATA6 and keratin 5 on the COMPASS trial.

Notably microdissected tissue, although impractical in laboratory medicine practice, most accurately detects tumor gene expression, with comparatively less exocrine and immune compartments compared with TCGA datasets, as recently described (32). This therefore implies that more reliable biomarkers can be determined from highly cellular specimens. CA-19.9 is the only approved biomarker for monitoring disease in the advanced setting (33) and the POLO trial has now documented a benefit for maintaining PARPi in patients with germline BRCA mutations (4). Robust subtyping of pancreatic cancer will be critical to advancing the field, GATA6 as a single biomarker and highly correlated with the Moffitt classifier will now be evaluated in a prospective trial.

This study is limited by few progression biopsies to understand the stability of the subtypes under selective pressure during chemotherapy. This is especially interesting in light of the coexistence of basal-like and classical elements, documented here and elsewhere (17, 18). In addition, the numbers of basal-like tumors treated with GnP regimens is low and the GnP group is potentially confounded by performance status. The interaction term for chemotherapy type and subtype was not significant in this study although numbers were low. We therefore cannot conclude whether GnP is a better strategy in the basal-like cohort, rather our data suggests alternative therapies are urgently needed, and clinical trials to evaluate this particular group are warranted. With mFFX as current treatment of choice in the adjuvant setting, understanding chemotherapy response to subtypes has increasing importance. It should also be noted that the response rates and survival between those receiving mFFX and GnP were not statistically different in this analysis. This is supported by the recent HALO trial 109–321 study where response rates and OS are comparable with historic outcomes with mFFX (2, 34). This furthermore supports the need to understand which populations can benefit most from these regimens and a prospective trial has now been planned.

In the major tumor types of lung and colorectal cancer, factors such as histologic subtype, molecular profile, and PD-L1 status can influence the choice of upfront systemic treatment in advanced disease and have resulted in survival gains (35–37). Because PDAC will soon become the second leading cause of cancer-related mortality, it behooves the oncology community to invest in biomarkers helpful for selecting standard chemotherapy. In this study, we confirm the prognostic impact of the modified Moffitt subtypes and demonstrate that basal-like PDAC responds poorly to mFFX. The basal-like cohort can be accurately identified by GATA6 RNA expression, providing a putative single important biomarker in clinical trial design.

G.M. O’Kane reports receiving speakers bureau honoraria from Roche and Eisai. J.M.S. Bartlett reports receiving commercial research grants from Thermo Fisher Scientific, Genoptix, Agendia, NanoString Technologies, Inc., Stratifyer GmbH, and Biotheranostics, Inc., speakers bureau honoraria from NanoString Technologies, Inc., Oncology Education, and Biotheranostics, Inc., and is an advisory board member/unpaid consultant for Insight Genetics, Inc., BioNTech AG, Biotheranostics, Inc., Pfizer, Rna Diagnostics, and oncoXchange. J.J. Knox is an employee/paid consultant for Merck, and reports receiving commercial research grants from Astra Zeneca, Ibsen, Roche, and Merck. No potential conflicts of interest were disclosed by the other authors.

Conception and design: G.M. O'Kane, M. Masoomian, J.J. Knox, S.E. Fischer

Development of methodology: G.M. O'Kane, B.T. Grünwald, J.J. Knox, S.E. Fischer

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G.M. O'Kane, B.T. Grünwald, G.-H. Jang, M. Masoomian, S. Picardo, Y. Wang, J.K. Miller, B. Lam, P.M. Krzyzanowski, I.M. Lungu, J.M.S. Bartlett, F. Vyas, R. Khokha, J. Biagi, D. Chadwick, S. Ramotar, S. Hutchinson, A. Dodd, J.M. Wilson, G. Zogopoulos, S. Gallinger, J.J. Knox, S.E. Fischer

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G.M. O'Kane, B.T. Grünwald, G.-H. Jang, R.C. Grant, R.E. Denroche, Y. Wang, F. Vyas, F. Notta, S. Gallinger, J.J. Knox

Writing, review, and/or revision of the manuscript: G.M. O'Kane, B.T. Grünwald, G.-H. Jang, R.C. Grant, Y. Wang, J.K. Miller, J. Biagi, G. Zogopoulos, S. Gallinger, J.J. Knox, S.E. Fischer

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): G.M. O'Kane, S. Picardo, R.E. Denroche, A. Zhang, J.K. Miller, I.M. Lungu, J.M.S. Bartlett, M. Peralta, R. Khokha, S. Ramotar, S. Hutchinson, A. Dodd, J.M. Wilson, G. Zogopoulos, S. Gallinger, J.J. Knox, S.E. Fischer

Study supervision: G.M. O'Kane, P.M. Krzyzanowski, J.J. Knox, S.E. Fischer

Other (collection of data): M. Masoomian

Other (primary study coordinator): S. Ramotar

This study was conducted with support of the Ontario Institute for Cancer Research (PanCuRx Translational Research Initiative) through funding provided by the Government of Ontario, the Wallace McCain Centre for Pancreatic Cancer supported by the Princess Margaret Cancer Foundation, the Terry Fox Research Institute, the Canadian Cancer Society Research Institute, and the Pancreatic Cancer Canada Foundation. The study was also supported by charitable donations from the Canadian Friends of the Hebrew University (Alex U. Soyka). J.J. Knox is the recipient of the Lewitt Chair in Pancreatic Cancer. G.M. O'Kane is supported by the Lewitt fellowship. S. Gallinger is the recipient of an Investigator Award from OICR. G. Zogopoulos is a clinical research scholar of the Fonds de recherche du Québec–Santé. B.T. Grünwald was supported by the Princess Margaret Cancer Foundation, EMBO (ALTF 116-2018), and the Alexander-von-Humboldt Foundation (DEU 1199182 FLF-P). We acknowledge the contributions of team members at OICR within the Diagnostic Development platform and the Genomics & Bioinformatics platform (genomics.oicr.on.ca).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.