Search

Supplementary Figure 3: Association of chemokine expression with a T cell-inflamed phenotype was validated in a second independent PDAC patient cohort

Supplementary Table 1: Top 99 genes associating with CD8+ TIL counts in primary resected cases

Purpose:The molecular drivers of antitumor immunity in pancreatic ductal adenocarcinoma (PDAC) are poorly understood, posing a major obstacle for the identification of patients potentially amenable for immune-checkpoint blockade or other novel strategies. Here, we explore the association of chemokine expression with effector T-cell infiltration in PDAC.Experimental Design:Discovery cohorts comprised 113 primary resected PDAC and 107 PDAC liver metastases. Validation cohorts comprised 182 PDAC from The Cancer Genome Atlas and 92 PDACs from the Australian International Cancer Genome Consortium. We explored associations between immune cell counts by immunohistochemistry, chemokine expression, and transcriptional hallmarks of antitumor immunity by RNA sequencing (RNA-seq), and mutational burden by whole-genome sequencing.Results:Among all known human chemokines, a coregulated set of four (CCL4, CCL5, CXCL9, and CXCL10) was strongly associated with CD8+ T-cell infiltration (P < 0.001). Expression of this “4-chemokine signature” positively correlated with transcriptional metrics of T-cell activation (ZAP70, ITK, and IL2RB), cytolytic activity (GZMA and PRF1), and immunosuppression (PDL1, PD1, CTLA4, TIM3, TIGIT, LAG3, FASLG, and IDO1). Furthermore, the 4-chemokine signature marked tumors with increased T-cell activation scores (MHC I presentation, T-cell/APC costimulation) and elevated expression of innate immune sensing pathways involved in T-cell priming (STING and NLRP3 inflammasome pathways, BATF3-driven dendritic cells). Importantly, expression of this 4-chemokine signature was consistently indicative of a T-cell–inflamed phenotype across primary PDAC and PDAC liver metastases.Conclusions:A conserved 4-chemokine signature marks resectable and metastatic PDAC tumors with an active antitumor phenotype. This could have implications for the appropriate selection of PDAC patients in immunotherapy trials.

Purpose:To determine the impact of basal-like and classical subtypes in advanced pancreatic ductal adenocarcinoma (PDAC) and to explore GATA6 expression as a surrogate biomarker.Experimental Design:Within the COMPASS trial, patients proceeding to chemotherapy for advanced PDAC undergo tumor biopsy for RNA-sequencing (RNA-seq). Overall response rate (ORR) and overall survival (OS) were stratified by subtypes and according to chemotherapy received. Correlation of GATA6 with the subtypes using gene expression profiling, in situ hybridization (ISH) was explored.Results:Between December 2015 and May 2019, 195 patients (95%) had enough tissue for RNA-seq; 39 (20%) were classified as basal-like and 156 (80%) as classical. RECIST response data were available for 157 patients; 29 basal-like and 128 classical where the ORR was 10% versus 33%, respectively (P = 0.02). In patients with basal-like tumors treated with modified FOLFIRINOX (n = 22), the progression rate was 60% compared with 15% in classical PDAC (P = 0.0002). Median OS in the intention-to-treat population (n = 195) was 9.3 months for classical versus 5.9 months for basal-like PDAC (HR, 0.47; 95% confidence interval, 0.32–0.69; P = 0.0001). GATA6 expression by RNA-seq highly correlated with the classifier (P < 0.001) and ISH predicted the subtypes with sensitivity of 89% and specificity of 83%. In a multivariate analysis, GATA6 expression was prognostic (P = 0.02). In exploratory analyses, basal-like tumors, could be identified by keratin 5, were more hypoxic and enriched for a T-cell–inflamed gene expression signature.Conclusions:The basal-like subtype is chemoresistant and can be distinguished from classical PDAC by GATA6 expression.See related commentary by Collisson, p. 4715

SummaryRecent advances in digital pathology have led to an explosion in high-content multidimensional imaging approaches. Yet, our ability to gainfully process, visualize, integrate and mine the resulting mass of bioimaging data remains a challenge. We have developed Hourglass, an open access user-friendly software that streamlines complex biology-driven post-processing and visualization of multiparametric data. Directed at datasets derived from tissue microarrays or imaging methods that analyze multiple regions of interest per patient specimen, Hourglass systematically organizes observations across spatial and global levels as well as within patient subgroups. Application of Hourglass to our large and complex pancreatic cancer bioimaging dataset (540,617 datapoints derived from 26 bioimaging analyses applied to 596 specimens from 165 patients) consolidated a breadth of known IL-6 functions in a well-annotated human pancreatic cancer cohort and uncovered new unprecedented insights into a sex-linked Interleukin-6 (IL-6) association with immune phenotypes. Specifically, regional effects of IL-6 on the intratumoral T cell response were restricted to male patients only. In conclusion, Hourglass facilitates multi-layered knowledge extraction from complex multiparametric bioimaging datasets and provides tailored analytical means to productively harness heterogeneity at the sample and patient level.

Purpose: To determine the impact of basal-like and classical subtypes in advanced pancreatic ductal adenocarcinoma (PDAC) and to explore GATA6 expression as a surrogate biomarker. Experimental Design: Within the COMPASS trial, patients proceeding to chemotherapy for advanced PDAC undergo tumor biopsy for RNA-sequencing (RNA-seq). Overall response rate (ORR) and overall survival (OS) were stratified by subtypes and according to chemotherapy received. Correlation of GATA6 with the subtypes using gene expression profiling, in situ hybridization (ISH) was explored. Results: Between December 2015 and May 2019, 195 patients (95%) had enough tissue for RNA-seq; 39 (20%) were classified as basal-like and 156 (80%) as classical. RECIST response data were available for 157 patients; 29 basal-like and 128 classical where the ORR was 10% versus 33%, respectively (P = 0.02). In patients with basal-like tumors treated with modified FOLFIRINOX (n = 22), the progression rate was 60% compared with 15% in classical PDAC (P = 0.0002). Median OS in the intention-to-treat population (n = 195) was 9.3 months for classical versus 5.9 months for basal-like PDAC (HR, 0.47; 95% confidence interval, 0.32–0.69; P = 0.0001). GATA6 expression by RNA-seq highly correlated with the classifier (P < 0.001) and ISH predicted the subtypes with sensitivity of 89% and specificity of 83%. In a multivariate analysis, GATA6 expression was prognostic (P = 0.02). In exploratory analyses, basal-like tumors, could be identified by keratin 5, were more hypoxic and enriched for a T-cell–inflamed gene expression signature. Conclusions: The basal-like subtype is chemoresistant and can be distinguished from classical PDAC by GATA6 expression. See related commentary by Collisson, p. 4715

SummaryPancreatic ductal adenocarcinoma (PDAC) remains resistant to most treatments and demonstrates a complex pathobiology. Here, we deconvolute regional heterogeneity in the human PDAC tumor microenvironment (TME), a long-standing obstacle, to define precise stromal contributions to PDAC progression. Large scale integration of histology-guided multiOMICs with clinical data sets and functional in vitro models uncovers two microenvironmental programs in PDAC that were anchored in fibroblast differentiation states. These sub-tumor microenvironments (subTMEs) co-occurred intratumorally and were spatially confined, producing patient-specific cellular and molecular heterogeneity associated with shortened patient survival. Each subTME was uniquely structured to support discrete aspects of tumor biology: reactive regions rich in activated fibroblast communities were immune-hot and promoted aggressive tumor progression while deserted regions enriched in extracellular matrix supported tumor differentiation yet were markedly chemoprotective. In conclusion, PDAC regional heterogeneity derives from biologically distinct reactive and protective TME elements with a defined, active role in PDAC progression.Graphical Abstract & Key findingsPDAC regional heterogeneity originates in sub-tumor microenvironments (subTMEs)SubTMEs exhibit distinct immune phenotypes and CAF differentiation statesDifferent subTMEs are either tumor-promoting or chemoprotectiveIntratumoral subTME co-occurrence links stromal heterogeneity to patient outcome

Intratumoral heterogeneity is a critical frontier in understanding how the tumor microenvironment (TME) propels malignant progression. We recently deconvoluted regional heterogeneity in the human PDAC stroma to assess its role in disease progression and discovered two types of ‘sub-tumor microenvironments’ (subTMEs), called ‘reactive’ and ‘deserted’. These histologically definable tissue states exhibit strong regional relationships with tumor immunity, subtypes, differentiation, and treatment response. Here, we set out to define their cell biological underpinnings through a combination of subTME-specific cancer-associated fibroblast (CAF) models, integrative histopathology, quantitative image analysis, multiOMICs, scRNAseq, and controlled functional assays. Remarkably, the growth patterns of CAF cultures closely recapitulated the characteristic histomorphology of their originating subTMEs, and these distinct phenotypes were accompanied by behavioral differences. Unsupervised graph-based clustering of scRNAseq profiles showed that CAFs largely grouped by their originating subTME yet comprised up to 10 individual clusters. The subTME-specific multi-subpopulation CAF communities self-organized into distinct ‘coordinated states’, represented by cluster-overarching functional profiles and distinct morpho-histological and behavioral phenotypes. These differences originated in cellular differentiation trajectories, with an ‘intermediate’ transitory state evident both in single cell transcriptomics and in situ. Noticeably, this CAF differentiation potential was associated with distinct tumor-related functions and, similar to stem cells, was marked by RNA diversity and pluripotency markers. Therefore, regional TME programs in PDAC appear to result largely from transitions between subpopulation-overarching fibroblast differentiation states that guide multifaceted CAF and immune cell communities into recurrent tissue self-organizational units. Citation Format: Barbara T. Grünwald, Curtis McCloskey, Antoine Devisme, Foram Vyas, Geoffroy Andrieux, Kazeera Aliar, Faiyaz Notta, Grainne O’Kane, Julie Wilson, Jennifer Knox, Sandra Fischer, Thomas Kislinger, Melanie Boerries, Steven Gallinger, Rama Khokha. Fibroblast differentiation trajectories elicit regional tissue states in pancreatic cancer [abstract]. In: Proceedings of the AACR Virtual Special Conference on Pancreatic Cancer; 2021 Sep 29-30. Philadelphia (PA): AACR; Cancer Res 2021;81(22 Suppl):Abstract nr PO-107.

Intratumoral heterogeneity is a critical frontier in understanding how the tumor microenvironment (TME) propels malignant progression. Here, we deconvolute the human pancreatic TME through large-scale integration of histology-guided regional multiOMICs with clinical data and patient-derived preclinical models. We discover "subTMEs," histologically definable tissue states anchored in fibroblast plasticity, with regional relationships to tumor immunity, subtypes, differentiation, and treatment response. "Reactive" subTMEs rich in complex but functionally coordinated fibroblast communities were immune hot and inhabited by aggressive tumor cell phenotypes. The matrix-rich "deserted" subTMEs harbored fewer activated fibroblasts and tumor-suppressive features yet were markedly chemoprotective and enriched upon chemotherapy. SubTMEs originated in fibroblast differentiation trajectories, and transitory states were notable both in single-cell transcriptomics and in situ. The intratumoral co-occurrence of subTMEs produced patient-specific phenotypic and computationally predictable heterogeneity tightly linked to malignant biology. Therefore, heterogeneity within the plentiful, notorious pancreatic TME is not random but marks fundamental tissue organizational units.

Purpose: The molecular drivers of antitumor immunity in pancreatic ductal adenocarcinoma (PDAC) are poorly understood, posing a major obstacle for the identification of patients potentially amenable for immune-checkpoint blockade or other novel strategies. Here, we explore the association of chemokine expression with effector T-cell infiltration in PDAC. Experimental Design: Discovery cohorts comprised 113 primary resected PDAC and 107 PDAC liver metastases. Validation cohorts comprised 182 PDAC from The Cancer Genome Atlas and 92 PDACs from the Australian International Cancer Genome Consortium. We explored associations between immune cell counts by immunohistochemistry, chemokine expression, and transcriptional hallmarks of antitumor immunity by RNA sequencing (RNA-seq), and mutational burden by whole-genome sequencing. Results: Among all known human chemokines, a coregulated set of four (CCL4, CCL5, CXCL9, and CXCL10) was strongly associated with CD8+ T-cell infiltration (P < 0.001). Expression of this “4-chemokine signature” positively correlated with transcriptional metrics of T-cell activation (ZAP70, ITK, and IL2RB), cytolytic activity (GZMA and PRF1), and immunosuppression (PDL1, PD1, CTLA4, TIM3, TIGIT, LAG3, FASLG, and IDO1). Furthermore, the 4-chemokine signature marked tumors with increased T-cell activation scores (MHC I presentation, T-cell/APC costimulation) and elevated expression of innate immune sensing pathways involved in T-cell priming (STING and NLRP3 inflammasome pathways, BATF3-driven dendritic cells). Importantly, expression of this 4-chemokine signature was consistently indicative of a T-cell–inflamed phenotype across primary PDAC and PDAC liver metastases. Conclusions: A conserved 4-chemokine signature marks resectable and metastatic PDAC tumors with an active antitumor phenotype. This could have implications for the appropriate selection of PDAC patients in immunotherapy trials.

Pancreatic ductal adenocarcinoma (PDAC) remains mostly untreatable while its incidence is on the rise. PDAC stroma should be a reservoir for novel therapeutic targets. Yet, stromal cellular complexity and an extensive intratumoral heterogeneity in human patients have left the functions of PDAC stroma poorly understood and likely hamper its targeting. Here, we set out to deconvolute regional heterogeneity in human PDAC stroma and assess its role in disease progression. Using resected & advanced biospecimens with known patient outcome, we mapped the stromal and epithelial landscapes of PDAC by combining integrative histopathology, quantitative image analysis of a 30-marker IHC panel, genomics and proteomics, machine learning, scRNAseq of PDAC-derived fibroblasts, and controlled functional assays. This histology-guided multidimensional approach revealed two overarching types of sub-tumormicroenvironments (subTMEs) in PDAC, present across stages and primary and metastatic sites. These subTMEs co-occurred intratumourally but were spatially confined, producing patient-specific cellular and molecular heterogeneity in PDAC stroma. SubTME-resolved IHC characterization of PDAC specimens and multiplexed functional analyses with subTME-specific patient-derived fibroblasts and organoids then identified how both microenvironmental programs were uniquely structured to support crucial aspects of tumour biology. Forming a complementary stromal support system, reactive subTME regions rich in fibroblasts and immune cells were vascularized and promoted features of aggressive tumour progression, while protective subTME regions enriched in ECM supported tumour differentiation yet were markedly chemoprotective and translated into poor treatment response during 1st line chemotherapy. Consequently, tumours benefited from concomitant presence of both subTMEs and patients with phenotypically heterogenous stroma displayed shortened survival. Our patient-matched multidimensional data were then used to train a Random Forest model for prediction of subTME presence from bulk RNAseq samples. This again stratified PDAC patients in the independent TCGA cohort into distinct clinical outcomes by stromal phenotypic heterogeneity. In conclusion, we have delineated human PDAC stroma to subtype its heterogeneity into phenotypically distinct reactive and protective elements. This defines the active role of two overarching sub-microenvironment types in PDAC progression and sets a path towards developing novel stroma-instructed therapies. Citation Format: Barbara T. Grünwald, Antoine Devisme, Foram Vyas, Geoffroy Andrieux, Gun Ho Jang, Kazeera Aliar, Curtis McCloskey, Andrew Macklin, Joan Miguel Romero, Laura Tamblyn, Nikolina Radulovich, Peter Bronsert, Grainne O’Kane, Julie Wilson, Jennifer Knox, Sandra Fischer, Thomas Kislinger, Melanie Boerries, Steven Gallinger, Rama Khokha. Stromal phenotypic heterogeneity fosters pancreatic cancer progression [abstract]. In: Proceedings of the AACR Virtual Special Conference on Pancreatic Cancer; 2020 Sep 29-30. Philadelphia (PA): AACR; Cancer Res 2020;80(22 Suppl):Abstract nr PR-002.