Purpose:

Precision medicine approaches in pancreatic ductal adenocarcinoma (PDAC) are imperative for improving disease outcomes. With molecular subtypes of PDAC gaining relevance in the context of therapeutic stratification, the ability to characterize heterogeneity of cancer-specific gene expression patterns is of great interest. In addition, understanding patterns of immune evasion within PDAC is of importance as novel immunotherapeutic strategies are developed.

Experimental Design:

Single-cell RNA sequencing (scRNA-seq) is readily applicable to limited biopsies from human primary and metastatic PDAC and identifies most cancers as being an admixture of previously described epithelial transcriptomic subtypes.

Results:

Integrative analyses of our data provide an in-depth characterization of the heterogeneity within the tumor microenvironment, including cancer-associated fibroblast subclasses, and predicts for a multitude of ligand-receptor interactions, revealing potential targets for immunotherapy approaches.

Conclusions:

Our analysis demonstrates that the use of de novo biopsies from patients with PDAC paired with scRNA-seq may facilitate therapeutic prediction from limited biopsy samples.

Translational Relevance

The application of single-cell RNA sequencing to diagnostic pancreatic cancer biopsies provides in-depth transcriptomic characterization of the tumor epithelium and microenvironment. This reveals the presence of subpopulations of cancer and stromal cells which may predict the emergence of chemoresistance and putative immunotherapeutic targets.

The majority of patients with pancreatic ductal adenocarcinoma (PDAC) present with locally advanced or metastatic disease, which precludes surgical resection of their cancer (1). The gold standard for diagnosis is a tissue biopsy prior to initiation of systemic therapy, which is obtained by either endoscopic ultrasound-guided fine-needle aspiration (EUS-FNA) of the primary lesion or through interventional radiology–guided percutaneous biopsy of a metastatic site. While typically adequate for histopathologic assessment, PDAC biopsies, especially those obtained via EUS-FNA without rapid onsite cytopathology assessment, can be limited by low neoplastic cellularity (2, 3). Nonetheless, such biopsies may be the only source of tissue from patients with locally advanced or metastatic PDAC who are likely to derive the most benefit from tailored approaches.

Multiple recent studies have demonstrated the feasibility of using next-generation sequencing (NGS) platforms for mutational analysis as applied to these limited “real-world” biospecimens obtained from patients with PDAC (4, 5). Notably, although targeted NGS panels can identify somatic or germline DNA mutations in a clinically meaningful timeframe, no more than a quarter of patients with PDAC, at best, harbor such actionable mutations (6). On the other hand, transcriptomic profiling of PDAC has revealed the presence of molecular subtypes that may guide therapeutic stratification as is seen in other cancer types (7). Single-cell RNA sequencing (scRNA-seq) has also aided in elucidating intratumoral heterogeneity with implications in evolution of cancer cells, resistance to therapy, and effects of stromal cells such as immune cells and fibroblasts on tumor behavior.

We explore the feasibility of using scRNA-seq on de novo limited biopsies obtained from patients with primary and metastatic PDAC. As recently described in studies using surgically resected PDAC samples, scRNA-seq provides an unprecedented level of insight into the architecture of the neoplastic cells and their tumor microenvironment (TME), including compartment specific heterogeneity (8–11). We demonstrate that scRNA-seq on limited fine-needle and core biopsies captured nearly all of the previously reported repertoire of cell types in surgical resections, including the tumor-stromal heterogeneity inherent to this disease, and revealed putative mechanisms for immune evasion within the TME. We believe that precision medicine approaches to a near universally lethal disease such as PDAC should deploy multiple orthogonal approaches, including NGS for actionable DNA alterations and scRNA-seq on contemporaneous de novo biopsies for assessment of the tumor-stromal architecture.

Sample acquisition

A total of 9 patients were recruited at the University of Texas MD Anderson Cancer Center (Houston, TX) through informed written consent following Institutional Review Board approval (Lab00-396 and PA15-0014). The study was conducted in accordance with Good Clinical Practices concerning medical research in humans per the Declaration of Helsinki. Five primary pancreatic cancers and four metastatic lesions in liver (LiM), lung (LuM), peritoneum (PM), and vaginal apex (VM) were used. P2–P5 primary samples came from treatment-naïve patients, while P1 was a second FNA from a patient who had been treated with gemcitabine/paclitaxel. For metastatic samples, the most recent therapies prior to sample acquisition were 5-fluorouracil/liposmal irinotecan, evofosfamide/ipilimumab, and capecitabine for VM, LuM, and PM, respectively. LiM was a treatment-naïve sample. Histologic confirmation of PDAC was performed by a pathologist.

Single-cell dissociation

Pancreatic cancer biopsies were collected in high-glucose DMEM supplemented with GlutaMAX, HEPES buffer, and 1% BSA (all Thermo Fisher Scientific). Samples were delivered to the laboratory within 2 hours after the procedure. Single-cell dissociation was performed as described previously (8). Briefly, samples were minced with sterile surgical scalpels to 0.5–1 mm fragments in approximately 1 mL of media. After centrifugation for 5 minutes at 125 ×g, cells were resuspended in DMEM with 0.5 mg/mL Liberase TH (Sigma-Aldrich) and 1% penicillin-streptomycin (Corning) and incubated in an orbital shaker for 15 minutes at 37°C. Liberase was quenched with equal volume of 1% BSA in DMEM followed by centrifugation for 5 minutes at 125 ×g. Cells were resuspended in 2 mL Accutase (Sigma-Aldrich) and incubated in an orbital shaker for 15 minutes at 37°C. Dissociated cells were passed through a 40-μm strainer and centrifuged for 5 minutes at 125 ×g. Isolated cells were resuspended in 0.04% BSA in PBS (Corning) for subsequent viability analysis and counting.

Single-cell RNA library preparation and sequencing

Single-cell transcriptomic amplification and library preparation was performed using the 5′ or 3′ Library Construction Kit (10× Genomics) following the manufacturer's recommendations. Final library quality and concentration were measured on TapeStation System using High Sensitivity D5000 ScreenTape (Agilent). Libraries were sequenced on NextSeq 500 (Illumina) according to the manufacturer's instructions.

Sequencing data processing and analysis

Raw base call files were demultiplexed and converted to FASTQ files, which were subsequently used to generate feature-barcode matrices using Cell Ranger RNA v3.1 (10× Genomics) and hg19 was used as a reference. Additional analytic methods are included in the Supplementary Data.

Data availability

The datasets generated during the current study are available in the Gene Expression Omnibus repository under accession number GSE156405.

scRNA-seq of limited de novo biopsies recapitulates the tumor-stromal heterogeneity of PDAC observed in surgical resections

We explored the feasibility of conducting scRNA-seq in a panel of limited “real-world” biopsies obtained from patients with primary and metastatic PDAC. Five primary samples (P1–P5) were obtained via EUS-FNAs of the pancreas, while four metastatic samples were obtained as core needle biopsies (liver, lung) or segmental biopsies (vaginal apex, peritoneal). Pooled analysis of all samples yielded a total of 31,720 cells organized into eight clusters (Fig. 1A), and highly expressed genes in each cluster were used to identify cell types (Fig. 1B). Epithelial cells were the most commonly represented cell type (47%), followed by T cells (28%) and myeloid cells (12%; Fig. 1A). Across primary and metastatic lesions, all cell types were represented and the average cell composition was similar (Supplementary Fig. S1A; Supplementary Table S1).

Figure 1.

PDAC biopsies contain diverse cell types. A, UMAP plot of single cells from nine biopsy samples (left) and bar plot showing proportions of cell types in pooled primary and metastatic samples (right). B, Bubble plot showing highly expressed marker genes in each cell type, with cell types in rows and genes in columns. Size of each bubble represents percent of cells expressing marker and color represents the level of expression. C, UMAP plot of epithelial cells reclustered from A (left) and bar plot showing proportions of epithelial cell subclusters in pooled primary and metastatic samples (right). D, Violin plots showing relative expression levels of selected marker genes in all epithelial subclusters.

Figure 1.

PDAC biopsies contain diverse cell types. A, UMAP plot of single cells from nine biopsy samples (left) and bar plot showing proportions of cell types in pooled primary and metastatic samples (right). B, Bubble plot showing highly expressed marker genes in each cell type, with cell types in rows and genes in columns. Size of each bubble represents percent of cells expressing marker and color represents the level of expression. C, UMAP plot of epithelial cells reclustered from A (left) and bar plot showing proportions of epithelial cell subclusters in pooled primary and metastatic samples (right). D, Violin plots showing relative expression levels of selected marker genes in all epithelial subclusters.

Close modal

To assess the heterogeneity specific to cancer cells, we extracted the epithelial cluster for further analysis, which revealed six distinct transcriptomic subclusters (Fig. 1C; Supplementary Table S1). Epithelial 1 subcluster was characterized by enrichment of antigen presentation, type I IFN response, and response to IFNγ pathways (Supplementary Fig. S1B). Along the same line, epithelial 2 subcluster was enriched for response to reactive oxygen species and intrinsic apoptosis, while epithelial subclusters 3 and 4 were enriched for epigenetic process and DNA replication/cell cycle, respectively, altogether demonstrating diverse transcriptomic programs present in PDAC tumor cells. A previous scRNA-seq analysis of PDAC demonstrated lipid processing and secretory cells within ductal epithelial cells (9), which were also present in our samples. Specifically, subcluster 5 expressed lipid processing cell markers APOA1, FABP1, and DNASE1, while subcluster 6 expressed secretory cell markers CFTR, SOD3, and COL18A1 (Fig. 1D). All epithelial subclusters were present in both primary and metastatic tissues, but lipid processing and secretory subclusters were proportionately less common in metastases (Supplementary Fig. S1C; Supplementary Table S2). Furthermore, signature genes for ductal and acinar cells were applied to the lipid processing and secretory subclusters (12). This analysis demonstrated that the former was enriched in acinar cells while the latter was enriched in ductal cells (Supplementary Table S3).

Copy-number inference from scRNA-seq showed numerous alterations in subclusters 1–4, whereas lipid processing and secretory subclusters did not contain matching alterations (Supplementary Fig. S1D), suggesting that the latter were derived from normal epithelium, as postulated previously (9). In addition, this combined inferred copy-number analysis demonstrated how limited biopsies can readily capture tumor heterogeneity of primary and metastatic PDAC.

Molecular subtyping of single cells reveals intratumoral subtype heterogeneity

Previous studies have classified PDAC into molecular subtypes based on their bulk transcriptome. These classifications include Bailey subtypes [pancreatic progenitor, squamous, aberrantly differentiated endocrine exocrine (ADEX), and immunogenic], Collisson subtypes (classical, quasimesenchymal or QM, and exocrine-like), and Moffitt subtypes (classical and basal-like; refs. 13–15). A subsequent study demonstrated two common epithelial subtypes, namely classical (or pancreatic progenitor) and basal-like (or squamous/QM), whereas immunogenic and ADEX or exocrine-like subtypes were associated with low tumor purity (16). To precisely delineate the cellular contributions to bulk molecular subtypes, we used nearest template prediction (17) to classify all single cells into a subtype based on the three classifiers (Supplementary Fig. S2A–S2D; Supplementary Table S4). Aligning the cell type to molecular subtypes confirmed that immunogenic, ADEX, and exocrine-like subtypes were mostly composed of non-epithelial cells (Supplementary Fig. S2E), consistent with the previous finding of low tumor purity in these subtypes (16).

To understand the molecular subtypes in the context of epithelial cells, we reapplied the classifiers on the epithelial subset (Supplementary Fig. S2A–S2D and S2F; Supplementary Table S5). Of note, while subtype prediction from pseudobulk projection yielded one molecular subtype for each biopsy sample (Supplementary Fig. S2G), all tumor samples were composed of more than one subtype at single-cell level regardless of the classifier (Fig. 2A–C; Supplementary Fig. S2H–S2J), consistent with recent reports that Moffitt classical and basal-like subtypes coexist in a single tumor (18, 19)

Figure 2.

PDAC molecular subtypes at single-cell resolution. A–C, Bar plots showing proportions of epithelial cells classifying into Bailey (A), Collisson (B), and Moffitt (C) subtypes with FDR < 0.2 in pooled primary and metastatic samples (left) and individual samples (right). D, Pseudotime trajectory of epithelial cells from all biopsy samples labeled with Moffitt molecular subtypes or Raghavan hybrid subtype (top) and bar plot representing the proportions of cells in each pseudotime bin (50 bins total) classifying into one of three subtypes (bottom). E, Pseudotime trajectory plot from D labeled with sample identification separated into primary lesions (top) and metastatic lesions (bottom). F, Branched heatmap showing dynamic gene expression changes in epithelial cells along the pseudotime trajectory, with pseudotime progressing from left to right. Enriched Hallmark or Gene Ontology biological process (GO-BP) terms for each gene cluster are listed on the right. LiM, liver metastasis; LuM, lung metastasis; PM, peritoneal metastasis; P1–5, primary 1–5; VM, vaginal apex metastasis.

Figure 2.

PDAC molecular subtypes at single-cell resolution. A–C, Bar plots showing proportions of epithelial cells classifying into Bailey (A), Collisson (B), and Moffitt (C) subtypes with FDR < 0.2 in pooled primary and metastatic samples (left) and individual samples (right). D, Pseudotime trajectory of epithelial cells from all biopsy samples labeled with Moffitt molecular subtypes or Raghavan hybrid subtype (top) and bar plot representing the proportions of cells in each pseudotime bin (50 bins total) classifying into one of three subtypes (bottom). E, Pseudotime trajectory plot from D labeled with sample identification separated into primary lesions (top) and metastatic lesions (bottom). F, Branched heatmap showing dynamic gene expression changes in epithelial cells along the pseudotime trajectory, with pseudotime progressing from left to right. Enriched Hallmark or Gene Ontology biological process (GO-BP) terms for each gene cluster are listed on the right. LiM, liver metastasis; LuM, lung metastasis; PM, peritoneal metastasis; P1–5, primary 1–5; VM, vaginal apex metastasis.

Close modal

We performed pseudotime analysis on the epithelial cells to elucidate potential evolutionary trajectories of PDAC tumor cells, which revealed two major branches (Fig. 2D). Interestingly, cells from the liver metastasis formed their own branch in the trajectory (cell fate 1; Fig. 2E; Supplementary Fig. S3A) and showed upregulation of epithelial–mesenchymal transition and hypoxia pathways (Fig. 2F), which may contribute to their more aggressive phenotype. Cell fate 2 mostly consisted of lung and vaginal metastases, with lung metastatic cells lying towards the end of the trajectory (Fig. 2E; Supplementary Fig. S3A). Lung metastatic cells were enriched for aerobic metabolism pathways, such as oxidative phosphorylation and electron transport chain (Fig. 2F), which may be expected given the presence of ample oxygen in the microenvironment. Of note, given that the liver- and lung-metastatic cells form diverging trajectories with enrichment of distinct gene sets even though cells in both sites mostly classified as basal-like (Fig. 2C and D), our results suggest that there may be heterogeneity even within a single subtype that may be responsive to the metastatic site. In addition, we included a third molecular “hybrid” subtype which has recently been reported by Raghavan and colleagues (20) and found it to be present in every case studied (Supplementary Fig. S3B).

scRNA-seq on PDAC biopsies capture cancer-associated fibroblast diversity

In the PDAC TME, functionally and transcriptomically distinct subtypes of cancer-associated fibroblasts (CAF) are present (21, 22). Inflammatory CAFs (iCAF) secrete cytokines, whereas myofibroblastic CAFs (myCAF) are characterized by the expression of extracellular matrix (ECM) components (21, 23). A third subtype of CAFs, named antigen-presenting CAFs or apCAFs, express class II MHC genes (9). Initial analysis of the CAF population from our scRNA-seq revealed two transcriptomically distinct subclusters, which were designated as iCAFs and myCAFs based on the expression of PDGFRA, CXCL12, ACTA2, and TAGLN as described previously (Supplementary Fig. S4A and S4B). To assess the presence of apCAFs, we looked for coexpression of the invariant chain of MHC II molecule CD74 with HLA-DPA1, HLA-DRA, and HLA-DRB1, which was present in a subset of CAFs (Fig. 3A). Utilizing nearest template prediction (17), we categorized each fibroblast into one of the three subtypes (Fig. 3B and C) and found no significant difference in the proportions of the CAF subtypes in primary and metastatic tumors (Supplementary Fig. S4C; Supplementary Table S6). In our samples, apCAFs did not coexpress PTPRC or MSLN (Supplementary Fig. S4D), confirming that these cells are not immune or mesothelial cells, as has been suggested in a recent study using murine PDAC (24). While apCAFs express MHC II molecules and can present antigens to CD4 T cells, they lack the costimulatory molecules required to activate an immune response. Therefore, it was hypothesized that apCAFs would contribute to the immunosuppressive environment in PDAC by leading to anergy or differentiation of CD4 T cells into regulatory T cells (Treg; ref. 9). Indeed, the proportion of apCAFs in biopsy samples was negatively correlated with the ratio of memory and effector CD8 T cells to Tregs (Fig. 3D). Our results recapitulate a potential mechanism by which stromal cells within the PDAC microenvironment may lead to immune suppression.

Figure 3.

Stromal heterogeneity of PDAC is captured by biopsies. A, Coexpression of selected apCAF marker genes projected onto UMAP plot of CAFs. CD74 expression is represented by black (low) to red (high), whereas the expression of HLA genes is represented by black (low) to green (high). High expression of both CD74 and variant MHC II molecule is shown as yellow. B, CAF subtypes from nearest template prediction projected onto UMAP plot. C, Heatmap showing scaled expression of apCAF, iCAF, and myCAF marker genes ordered by subtype. D, Correlation plot between apCAF abundance in each sample and the ratio of CD8 T cells (memory and effector) to Treg cells.

Figure 3.

Stromal heterogeneity of PDAC is captured by biopsies. A, Coexpression of selected apCAF marker genes projected onto UMAP plot of CAFs. CD74 expression is represented by black (low) to red (high), whereas the expression of HLA genes is represented by black (low) to green (high). High expression of both CD74 and variant MHC II molecule is shown as yellow. B, CAF subtypes from nearest template prediction projected onto UMAP plot. C, Heatmap showing scaled expression of apCAF, iCAF, and myCAF marker genes ordered by subtype. D, Correlation plot between apCAF abundance in each sample and the ratio of CD8 T cells (memory and effector) to Treg cells.

Close modal

PDAC-infiltrating immune cells acquire an immune suppressive signature

PDAC is characterized by immunosuppressive microenvironment (1). To understand how immune cells change from the circulation to the PDAC TME, we performed scRNA-seq on peripheral blood mononuclear cells (PBMC) from 8 of 9 patients obtained at or around the time of their tissue biopsies. Global comparison of all cell types revealed that there was a subpopulation of myeloid cells that were found only in the PDAC but not in the periphery (Fig. 4A; Supplementary Fig. S5A), and subclustering analysis revealed this unique population to be macrophages (Fig. 4B; Supplementary Fig. S5B and S5C). Given the potential polarization of tumor-associated macrophages into classical M1 and alternative M2 phenotypes (25), we performed trajectory inference analysis on monocytes and macrophages from PBMCs and tumors, which revealed a single unbranched trajectory representing the differentiation from monocytes to macrophages (Fig. 4C). The expression of immune suppressive genes such as MARCO and TREM2 increased with pseudotime (Fig. 4D). Global gene expression changes along the pseudotime trajectory revealed upregulation of macrophage activation, acute and chronic inflammation, and negative regulation of myeloid-mediated immunity pathways (Fig. 4E). Furthermore, macrophages were characterized by an upregulation of angiogenesis and hypoxia pathways (Fig. 4E), which are associated with the M2 phenotype (25), confirming the immune suppressive role of tumor-infiltrating macrophages.

Figure 4.

Acquisition of immune-suppressive phenotype by tumor-infiltrating immune cells. A, UMAP plot of immune cells from PBMCs and tumors colored by cell type. B, UMAP of myeloid cells reclustered from A colored by cell type (left) and bar plot showing proportions of each myeloid cell type as percent of all cells in PBMC, primary, and metastatic samples (right). C, Pseudotime trajectory of monocytes and macrophages. D, Dot plot representing the expression of MARCO (top) and TREM2 (bottom) along the pseudotime trajectory from D colored by cell type. Solid black line indicates the mean expression at a given pseudotime. E, Heatmap showing dynamic gene expression changes through pseudotime (left) and enrichment of Gene Ontology biological process (GO-BP) terms for selected gene cluster (right). F, UMAP of myeloid cells reclustered from Fig. 1A (left) and bar plot showing proportions of myeloid cell types in pooled primary and metastatic samples (right). DC, dendritic cell; mac, macrophage; mono, monocyte; pDC, plasmacytoid DC. G, Heatmap showing scaled expression of select marker genes in each myeloid subcluster. H, UMAP of T and NK cells reclustered from Fig. 1A (left) and bar plot showing proportions of T- and NK-cell types in pooled primary and metastatic samples (right). IFN, interferon; MAIT, mucosal-associated invariant T; Treg, regulatory T; Trm, tissue resident memory. I, Heatmap showing scaled expression of select marker genes in each T and NK subcluster.

Figure 4.

Acquisition of immune-suppressive phenotype by tumor-infiltrating immune cells. A, UMAP plot of immune cells from PBMCs and tumors colored by cell type. B, UMAP of myeloid cells reclustered from A colored by cell type (left) and bar plot showing proportions of each myeloid cell type as percent of all cells in PBMC, primary, and metastatic samples (right). C, Pseudotime trajectory of monocytes and macrophages. D, Dot plot representing the expression of MARCO (top) and TREM2 (bottom) along the pseudotime trajectory from D colored by cell type. Solid black line indicates the mean expression at a given pseudotime. E, Heatmap showing dynamic gene expression changes through pseudotime (left) and enrichment of Gene Ontology biological process (GO-BP) terms for selected gene cluster (right). F, UMAP of myeloid cells reclustered from Fig. 1A (left) and bar plot showing proportions of myeloid cell types in pooled primary and metastatic samples (right). DC, dendritic cell; mac, macrophage; mono, monocyte; pDC, plasmacytoid DC. G, Heatmap showing scaled expression of select marker genes in each myeloid subcluster. H, UMAP of T and NK cells reclustered from Fig. 1A (left) and bar plot showing proportions of T- and NK-cell types in pooled primary and metastatic samples (right). IFN, interferon; MAIT, mucosal-associated invariant T; Treg, regulatory T; Trm, tissue resident memory. I, Heatmap showing scaled expression of select marker genes in each T and NK subcluster.

Close modal

To elucidate the different types of myeloid cells present in the PDAC TME, we performed subclustering analysis of tumor-infiltrating myeloid cells and found eight distinct cell types (Fig. 4F; Supplementary Table S7). Using previously published subtyping of monocytes and dendritic cells (DC) based on scRNA-seq profiles (26), we classified these subclusters as monocyte 1 (equivalent to CD14high cells), monocyte 2 (equivalent to CD16high cells), macrophage 1 (expressing APOE and SPP1), macrophage 2 (expressing complement molecules), DC1 (CLEC9A), DC2/3 (CD1C), plasmacytoid (p)DCs (TCF4, IRF7), and mast cells (TPSAB1, CPA3; Fig. 4G; Supplementary Fig. S5D). All subclusters were represented in primary tumors and metastases without significant differences in their proportions (Supplementary Fig. S5E). In addition to MARCO and TREM2, tumor-associated macrophages expressed PPARG and IL6 transcripts (Supplementary Fig. S5F), the products of which are also implicated in immune suppression (27). Of note, a recent study comparing the immune compartments of PDAC and lung cancer revealed reduced numbers of DC infiltration in PDAC that led to the attenuation of antitumor T-cell activity (28). Consistent with these findings, DCs represented smallest subsets of myeloid cells in our samples. Moreover, given that the DC1 subset expressed IDO1 (Fig. 4G), which inhibits T-cell proliferation (9, 29), the DC1 subset may contribute to the immune suppressive phenotype in PDAC.

Analysis of tumor-infiltrating T and natural killer (NK) cells revealed 12 distinct subclusters (Fig. 4H; Supplementary Table S8). On the basis of known markers for T and NK cells (30, 31), we identified memory (ANXA1) and naïve (SELL, CCR7) CD4 T cells, memory and effector (CX3CR1) CD8 T cells, naïve CD8 T cells, mucosal-associated invariant T (MAIT) cells (SLC4A10), tissue resident memory CD8 T (Trm) cells (ZNF683), Tregs (FOXP3, CTLA4), CD56 NK cells, CD16 NK cells, IFN-responsive T cells (IFI44L, ISG15), and NK T cells (TYMS, MKI67; Fig. 4I; Supplementary Fig. S6A). Many T cells expressed inhibitory receptors PDCD1, LAG3, and BTLA (Supplementary Fig. S6B), suggesting that T cells are present in the PDAC TME but their functions may be inhibited. Next, we compared T and NK cells in PBMCs and tumors (Supplementary Fig. S6C–S6E). There was a trend of naïve CD4 T cells decreasing in proportion from periphery to primary to metastatic tumors, whereas memory CD4 and CD8 T cells showed a reverse trend (Supplementary Fig. S6F), likely reflecting antigen exposure and creation of memory T-cell populations predating metastatic spread. Gene set enrichment analysis of tumor-infiltrating T and NK cells revealed the enrichment of hypoxia and negative regulation of immune response pathways (Supplementary Fig. S6G), demonstrating the acquisition of immune suppressive phenotype similar to tumor-associated macrophages.

Prediction of the PDAC ligand-receptor interactome reveals multiple immune regulatory pathways and potentially actionable nodes

scRNA-seq has been used to predict potential ligand-receptor (LR) interactions, which we utilized to reveal potential mechanisms by which the cross-talk between tumor cells and immune cells may contribute to immunosuppression in PDAC. We applied an average expression-based LR interaction prediction (32, 33) to assess potential relationships between major cell types of two primary (P3, P4) and two metastatic (liver, lung) tumors, as they contained the highest cell numbers and represented the most cellular diversity. We did not combine multiple samples for analysis to rule out potentially false interactions (e.g., T cells from liver metastasis to epithelial cells in a primary tumor). Global overview of the interactomes revealed that myeloid and DCs were the most connected, and epithelial cells interacted closely with these two cell types (Fig. 5A and B; Supplementary Fig. S7A and S7B). One of the most significant LR pairs responsible for these connections was predicted to be between macrophage migration inhibitory factor (MIF) and the transcript for the invariant HLA-DR allele, CD74 (Fig. 5C). This interaction has been implicated in promoting myeloid-derived suppressor cell–mediated immunosuppression in melanoma and breast cancer (34, 35), while tumor-cell MIF expression has been linked to a more aggressive phenotype in PDAC (36). Another prevalent interaction between tumor epithelial cells and myeloid/DCs involved amyloid precursor protein (APP) and CD74; of interest, we have previously described that APP overexpression in PDAC cells upregulates proliferation (37). The potential role of neoplastic APP in subverting antigen presentation via CD74 interaction on immune cells in the PDAC TME is uncharted. In addition to myeloid cells, CD74 expression in B cells has been linked to promote their proliferation (38) and recent work has demonstrated the role of tumor-infiltrating B cells in the initiation and proliferation of PDAC (39–41). Our LR analysis predicted a putative interaction between CD74 receptor on B cells and its cognate ligand, MIF, expressed by neoplastic epithelial cells, respectively. We were able to validate this putative interaction in a panel of archival PDAC samples by multiplex immunofluorescence, confirming the proximity of the two cell types expressing the corresponding receptor-ligand pair (Supplementary Fig. S8).

Figure 5.

LR interaction predictions between major cell types. A, Network plots of a primary PDAC (P3, left) and liver metastasis (right) demonstrating potential LR interactions. Each node represents a cell type and size reflects relative number of cells. Each edge represents the number of significant interactions between each cell-type pair and its thickness is proportional to the number of interactions. B, Heatmaps of log-transformed number of significant interactions between cell-type pairs in a primary PDAC (P3, left) and liver metastasis (right). C, Bubble plots representing top significant LR interactions between different cell-type pairs involving epithelial cells in a primary PDAC (P3, left) and liver metastasis (right). Size of each bubble represents P value and color represents the mean expression of ligand and receptor genes.

Figure 5.

LR interaction predictions between major cell types. A, Network plots of a primary PDAC (P3, left) and liver metastasis (right) demonstrating potential LR interactions. Each node represents a cell type and size reflects relative number of cells. Each edge represents the number of significant interactions between each cell-type pair and its thickness is proportional to the number of interactions. B, Heatmaps of log-transformed number of significant interactions between cell-type pairs in a primary PDAC (P3, left) and liver metastasis (right). C, Bubble plots representing top significant LR interactions between different cell-type pairs involving epithelial cells in a primary PDAC (P3, left) and liver metastasis (right). Size of each bubble represents P value and color represents the mean expression of ligand and receptor genes.

Close modal

Given that scRNA-seq data can be biased by gene dropout, thereby diminishing the average expression of genes in each cell type, we next predicted LR interactions at single-cell level (42). When looking at predicted interactions involving ligands from cells in the TME and cognate receptors on epithelial cells, we found that epithelial cells were more strongly connected to fibroblasts and myeloid cells, whereas there were fewer connections with T and B cells (Fig. 6A; Supplementary Fig. S9A and S9B). Many predicted interactions between epithelial cells and fibroblasts involved components of the ECM (Fig. 6B), which is expected given that one of the major functions of fibroblasts is the production of ECM proteins. There were many pro-inflammatory interactions such as IFNG from T/NK cells and IFNGR1/2 on epithelial cells (Fig. 6B), once again pointing to inflammatory signals present in the PDAC microenvironment.

Figure 6.

LR interaction predictions between single cells. A, Network plots of a primary PDAC (P3, left) and liver metastasis (right) demonstrating potential LR interactions in which epithelial cells express the receptor. Each node represents a single cell and the edge represents the number of LR pairs between two cells. B, Heatmap showing top interaction scores (normalized number of LR interactions) for each cell-type pair where epithelial cells express the receptor. Sample origin, stage, and ligand-expressing cells are annotated above the heatmap. See bottom of figure for legend. C, Network plots of a primary PDAC (P3, left) and liver metastasis (right) demonstrating potential LR interactions in which epithelial cells express the ligand. D, Heatmap showing top interaction scores for each cell-type pair where epithelial cells express the ligand. Sample origin, stage, and receptor-expressing cells are annotated above the heatmap. LR interaction class is annotated to the left. LiM, liver metastasis; LuM, lung metastasis; PM, peritoneal metastasis; P1–5, primary 1–5; VM, vaginal apex metastasis.

Figure 6.

LR interaction predictions between single cells. A, Network plots of a primary PDAC (P3, left) and liver metastasis (right) demonstrating potential LR interactions in which epithelial cells express the receptor. Each node represents a single cell and the edge represents the number of LR pairs between two cells. B, Heatmap showing top interaction scores (normalized number of LR interactions) for each cell-type pair where epithelial cells express the receptor. Sample origin, stage, and ligand-expressing cells are annotated above the heatmap. See bottom of figure for legend. C, Network plots of a primary PDAC (P3, left) and liver metastasis (right) demonstrating potential LR interactions in which epithelial cells express the ligand. D, Heatmap showing top interaction scores for each cell-type pair where epithelial cells express the ligand. Sample origin, stage, and receptor-expressing cells are annotated above the heatmap. LR interaction class is annotated to the left. LiM, liver metastasis; LuM, lung metastasis; PM, peritoneal metastasis; P1–5, primary 1–5; VM, vaginal apex metastasis.

Close modal

We also explored the reverse interactions in which ligands were expressed on epithelial cells and receptors on TME cells, which again revealed more connections with fibroblasts and myeloid cells (Fig. 6C; Supplementary Fig. S9C and S9D). Most prominent interactions with fibroblasts were growth factors, and interactions with immune cells involved cytokine and checkpoint genes (Fig. 6D). Of note, epithelial cells interacted with T/NK cells via IL18-IL18R1 and LGALS9-HAVCR2. IL18 has been shown to promote IFNγ production in T/NK cells (43), and the interaction between LGALS9 and HAVCR2 can suppress T cell–mediated immunity (44). Additional immunosuppressive interactions between epithelial and T cells included PVR-TIGIT and CD274-PDCD1. Interactions with myeloid and DCs included CD24-SIGLEC10, which promotes immune evasion and is a potential target in cancer immunotherapy (45). One of the most common interactions between B and epithelial cells was BTLA-TNFRSF14 (also known as herpesvirus entry mediator or HVEM), which can suppress both T cell–mediated and B cell–mediated immunity (46). Overall, these results suggest that PDAC tumor cells can drive an immunosuppressive microenvironment by affecting multiple immune-cell types.

We systematically analyzed scRNA-seq from “real-world” biopsies of primary and metastatic PDAC, demonstrating that even these scant samples recapitulate significant cellular heterogeneity. Recently, there have been multiple studies utilizing single-cell technologies in human and murine PDAC demonstrating heterogeneity in both tumor epithelial and TME compartments (8–11, 24, 47–49). Our data showed higher proportion of epithelial cells compared with recent scRNA-seq studies on surgically resected PDAC tumors (9, 10), while the most common immune-cell types were myeloid and T cells, consistent with these studies. We performed comprehensive analyses on our dataset including transcriptomics subtyping at single-cell resolution, which revealed two subtypes specific to tumor epithelial cells, namely classical (or pancreatic progenitor) and basal-like (or QM/squamous), confirming the results from a previous analysis of bulk tumors (16). Furthermore, metastatic lesions were more likely to harbor a higher proportion of aggressive basal-like subtype of cells, which was also previously reported in a separate PDAC cohort using bulk transcriptome (50). Notably, we show that although projected bulk transcriptome of each tumor sample classifies PDAC dichotomously as one of two subtypes, at single-cell resolution, even limited PDAC biopsies are comprised of more than one subtype. This finding may be of clinical importance in light of the COMPASS trial, which showed that classical PDAC is more likely to respond to first-line chemotherapy, while basal-like tumors tend to be chemoresistant (51). The existence of basal-like cells within an “apparently classical” tumor could lead to preferential enrichment of basal-like cells and emergence of resistance. Thus, future iterations of trials such as COMPASS that rely on subtyping using core biopsies may need to provide a quantitative measure of cellular subtypes present rather than a qualitative dichotomous readout. The coexistence of multiple subtypes within each tumor also raises the possibility of transcriptomic plasticity between epithelial and mesenchymal phenotypes with disease progression or response to therapy, as demonstrated previously (52).

Furthermore, the overall gene expression profile and cell-type proportions revealed several putative mechanisms of immune suppression in PDAC, such as the expression of immune-suppressive genes by myeloid cells and inhibitory receptors in T cells, polarization of macrophages toward M2 phenotype, as well as the presence of apCAFs that may inhibit T cell–mediated immune response. A recent study demonstrated the expression of immune checkpoint receptors such as TIGIT and HAVCR2 in T and NK cells, as well as the expression of their corresponding ligands PVR and LGALS9 in cancer cells, demonstrating alterations of the immune environment of PDAC (11). Prediction of the prevalent LR “interactome” in our data also revealed multiple mechanisms by which tumor–TME interactions may contribute to inflammation and immune evasion including PVR-TIGIT and LGALS9-HAVCR2. Furthermore, CAFs have been shown to drive PDAC plasticity and heterogeneity via TGFβ1 upregulation of MAPK/STAT3 (53). Accordingly, our LR interactome predicted TGFB-TGFBR axis to be upregulated between fibroblasts and cancer cells.

Through integrative analysis of scRNA-seq data from “real-world” biopsies, we uncovered implicit heterogeneity of low-input biopsy samples in PDAC that demonstrates potential mechanisms for immune evasion and provide high-resolution data correlating cell types to clinically relevant molecular subtypes. Deep characterization using a contemporaneous biopsy may provide a valid orthogonal approach to precision medicine approaches when paired with further precision strategies such as NGS for actionable DNA alterations and drug sensitivity testing on patient-derived ex vivo cultures such as organoids.

J.J. Lee reports grants from NIH/NCI during the conduct of the study. V. Bernard reports grants from NIH during the conduct of the study. A. Semaan reports grants from German Research Foundation (DFG; SE-2616/2-1) during the conduct of the study. C.L. Haymaker reports personal fees from Briacell Therapeutics and Nanobiotix, as well as other support from Mesothelioma Applied Research Foundation outside the submitted work. C. Bernatchez reports personal fees from Myst Therapeutics and grants from Iovance Biotherapeutics outside the submitted work. C.M. Taniguchi reports grants from NIHR01CA227517-01, Reaumond Family Foundation, V Foundation, Mark Foundation, and Childress Family Foundation during the conduct of the study, as well as support from Xerient Pharmaceuticals outside the submitted work. A. Maitra reports other support from Cosmos Wisdom Biotechnology and Thrive Earlier Detection, an Exact Sciences Company outside the submitted work. No disclosures were reported by the other authors.

J.J. Lee: Conceptualization, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. V. Bernard: Conceptualization, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. A. Semaan: Conceptualization, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. M.E. Monberg: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. J. Huang: Data curation, formal analysis, investigation, visualization, methodology, writing–review and editing. B.M. Stephens: Data curation, validation, investigation, methodology, writing–review and editing. D. Lin: Data curation, validation, investigation, methodology, writing–review and editing. K.I. Rajapakshe: Methodology. B.R. Weston: Conceptualization, resources, data curation, supervision, validation, investigation, writing–review and editing. M.S. Bhutani: Conceptualization, resources, data curation, supervision, validation, investigation, writing–review and editing. C.L. Haymaker: Data curation, formal analysis, validation, methodology, writing–review and editing. C. Bernatchez: Data curation, formal analysis, validation, methodology, writing–review and editing. C.M. Taniguchi: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. A. Maitra: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. P.A. Guerrero: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

We thank Mark W. Hurd (The University of Texas MD Anderson Cancer Center) for assistance with sample acquisition. A. Maitra is supported by the MD Anderson Pancreatic Cancer Moon Shot Program, the Khalifa Bin Zayed Al-Nahyan Foundation, and the NIH (U01CA196403, U01CA200468, P50CA221707). C.M. Taniguchi is supported by the NIH (R01CA227518-01A1), Cancer Prevention and Research Institute of Texas (CPRIT RR140012), Mark Foundation, V Foundation (V2015-22), Kimmel Foundation, McNair Foundation, and Reaumond Family Foundation. A. Semaan is supported by the German Research Foundation (SE-2616/2-1). V. Bernard is supported by the NIH (U54CA096300, U54CA096297, T32CA217789). J.J. Lee is supported by the NIH (T32CA009599).

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.

1.
Kleeff
J
,
Korc
M
,
Apte
M
,
Vecchia
CL
,
Johnson
CD
,
Biankin
AV
, et al
Pancreatic cancer
.
Nat Rev Dis Primers
2016
;
2
:
16022
.
2.
Conrad
R
,
Castelino-Prabhu
S
,
Cobb
C
,
Raza
A
. 
Cytopathology of the pancreatobiliary tract-the agony, and sometimes, the ease of it
.
J Gastrointest Oncol
2013
;
4
:
210
9
.
3.
VanderLaan
PA
. 
Fine-needle aspiration and core needle biopsy: an update on 2 common minimally invasive tissue sampling modalities
.
Cancer Cytopathol
2016
;
124
:
862
70
.
4.
Chantrill
LA
,
Nagrial
AM
,
Watson
C
,
Johns
AL
,
Martyn-Smith
M
,
Simpson
S
, et al
Precision medicine for advanced pancreas cancer: the individualized molecular pancreatic cancer therapy (IMPaCT) trial
.
Clin Cancer Res
2015
;
21
:
2029
37
.
5.
Aguirre
AJ
,
Nowak
JA
,
Camarda
ND
,
Moffitt
RA
,
Ghazani
AA
,
Hazar-Rethinam
M
, et al
Real-time genomic characterization of advanced pancreatic cancer to enable precision medicine
.
Cancer Discov
2018
;
8
:
1096
111
.
6.
Dreyer
SB
,
Chang
DK
,
Bailey
P
,
Biankin
AV
. 
Pancreatic cancer genomes: implications for clinical management and therapeutic development
.
Clin Cancer Res
2017
;
23
:
1638
46
.
7.
Collisson
EA
,
Bailey
P
,
Chang
DK
,
Biankin
AV
. 
Molecular subtypes of pancreatic cancer
.
Nat Rev Gastroentero
2019
;
16
:
207
20
.
8.
Bernard
V
,
Semaan
A
,
Huang
J
,
Lucas
FAS
,
Mulu
FC
,
Stephens
BM
, et al
Single cell transcriptomics of pancreatic cancer precursors demonstrates epithelial and microenvironmental heterogeneity as an early event in neoplastic progression
.
Clin Cancer Res
2019
;
25
:
2194
2205
.
9.
Elyada
E
,
Bolisetty
M
,
Laise
P
,
Flynn
WF
,
Courtois
ET
,
Burkhart
RA
, et al
Cross-species single-cell analysis of pancreatic ductal adenocarcinoma reveals antigen-presenting cancer-associated fibroblasts
.
Cancer Discov
2019
;
9
:
1102
23
.
10.
Peng
J
,
Sun
B-F
,
Chen
C-Y
,
Zhou
J-Y
,
Chen
Y-S
,
Chen
H
, et al
Single-cell RNA-seq highlights intra-tumoral heterogeneity and malignant progression in pancreatic ductal adenocarcinoma
.
Cell Res
2019
;
29
:
725
38
.
11.
Steele
NG
,
Carpenter
ES
,
Kemp
SB
,
Sirihorachai
VR
,
The
S
,
Delrosario
L
, et al
Multimodal mapping of the tumor and peripheral blood immune landscape in human pancreatic cancer
.
Nat Cancer
2020
;
1
:
1097
112
.
12.
Muraro
MJ
,
Dharmadhikari
G
,
Grün
D
,
Groen
N
,
Dielen
T
,
Jansen
E
, et al
A single-cell transcriptome atlas of the human pancreas
.
Cell Syst
2016
;
3
:
385
94
.
13.
Initiative
APCG
,
Bailey
P
,
Chang
DK
,
Nones
K
,
Johns
AL
,
Patch
A-M
, et al
Genomic analyses identify molecular subtypes of pancreatic cancer
.
Nature
2016
;
531
:
47
52
.
14.
Collisson
EA
,
Sadanandam
A
,
Olson
P
,
Gibb
WJ
,
Truitt
M
,
Gu
S
, et al
Subtypes of pancreatic ductal adenocarcinoma and their differing responses to therapy
.
Nat Med
2011
;
17
:
500
3
.
15.
Moffitt
RA
,
Marayati
R
,
Flate
EL
,
Volmar
KE
,
Loeza
SGH
,
Hoadley
KA
, et al
Virtual microdissection identifies distinct tumor- and stroma-specific subtypes of pancreatic ductal adenocarcinoma
.
Nat Genet
2015
;
47
:
1168
78
.
16.
Raphael
BJ
,
Hruban
RH
,
Aguirre
AJ
,
Moffitt
RA
,
Yeh
JJ
,
Stewart
C
, et al
Integrated genomic characterization of pancreatic ductal adenocarcinoma
.
Cancer Cell
2017
;
32
:
185
203.e13
.
17.
Hoshida
Y
. 
Nearest template prediction: a single-sample-based flexible class prediction with confidence assessment
.
PLoS One
2010
;
5
:
e15543
.
18.
Chan-Seng-Yue
M
,
Kim
JC
,
Wilson
GW
,
Ng
K
,
Figueroa
EF
,
O'Kane
GM
, et al
Transcription phenotypes of pancreatic cancer are driven by genomic events during tumor evolution
.
Nat Genet
2020
;
52
:
231
40
.
19.
Hayashi
A
,
Fan
J
,
Chen
R
,
Ho
Y
,
Makohon-Moore
AP
,
Lecomte
N
, et al
A unifying paradigm for transcriptional heterogeneity and squamous features in pancreatic ductal adenocarcinoma
.
Nat Cancer
2020
;
1
:
59
74
.
20.
Raghavan
S
,
Winter
PS
,
Navia
AW
,
Williams
HL
,
DenAdel
A
,
Kalekar
RL
, et al
The tumor microenvironment drives transcriptional phenotypes and their plasticity in metastatic pancreatic cancer
.
Biorxiv
2021
.
21.
Öhlund
D
,
Handly-Santana
A
,
Biffi
G
,
Elyada
E
,
Almeida
AS
,
Ponz-Sarvise
M
, et al
Distinct populations of inflammatory fibroblasts and myofibroblasts in pancreatic cancer
.
J Exp Med
2017
;
214
:
579
96
.
22.
Biffi
G
,
Oni
TE
,
Spielman
B
,
Hao
Y
,
Elyada
E
,
Park
Y
, et al
IL-1-induced JAK/STAT signaling is antagonized by TGF-beta to shape CAF heterogeneity in pancreatic ductal adenocarcinoma
.
Cancer Discov
2018
;
9
:
282
301
.
23.
Hosein
AN
,
Brekken
RA
,
Maitra
A
. 
Pancreatic cancer stroma: an update on therapeutic targeting strategies
.
Nat Rev Gastroenterol Hepatol
2020
;
17
:
487
505
.
24.
Hosein
AN
,
Huang
H
,
Wang
Z
,
Parmar
K
,
Du
W
,
Huang
J
, et al
Cellular heterogeneity during mouse pancreatic ductal adenocarcinoma progression at single-cell resolution
.
JCI Insight
2019
;
5
:
e129212
.
25.
Henze
A-T
,
Mazzone
M
. 
The impact of hypoxia on tumor-associated macrophages
.
J Clin Invest
2016
;
126
:
3672
9
.
26.
Villani
A-C
,
Satija
R
,
Reynolds
G
,
Sarkizova
S
,
Shekhar
K
,
Fletcher
J
, et al
Single-cell RNA-seq reveals new types of human blood dendritic cells, monocytes, and progenitors
.
Science
2017
;
356
:
eaah4573
.
27.
Lavin
Y
,
Kobayashi
S
,
Leader
A
,
Amir
ED
,
Elefant
N
,
Bigenwald
C
, et al
Innate immune landscape in early lung adenocarcinoma by paired single-cell analyses
.
Cell
2017
;
169
:
750
65
.
e17
.
28.
Hegde
S
,
Krisnawan
VE
,
Herzog
BH
,
Zuo
C
,
Breden
MA
,
Knolhoff
BL
, et al
Dendritic cell paucity leads to dysfunctional immune surveillance in pancreatic cancer
.
Cancer Cell
2020
;
37
:
289
307
.
e9
.
29.
Munn
DH
,
Sharma
MD
,
Lee
JR
,
Jhaver
KG
,
Johnson
TS
,
Keskin
DB
, et al
Potential regulatory function of human dendritic cells expressing indoleamine 2,3-dioxygenase
.
Science
2002
;
297
:
1867
70
.
30.
Guo
X
,
Zhang
Y
,
Zheng
L
,
Zheng
C
,
Song
J
,
Zhang
Q
, et al
Global characterization of T cells in non-small-cell lung cancer by single-cell sequencing
.
Nat Med
2018
;
24
:
978
85
.
31.
Crinier
A
,
Milpied
P
,
Escalière
B
,
Piperoglou
C
,
Galluso
J
,
Balsamo
A
, et al
High-dimensional single-cell analysis identifies organ-specific signatures and conserved NK cell subsets in humans and mice
.
Immunity
2018
;
49
:
971
986.e5
.
32.
Vento-Tormo
R
,
Efremova
M
,
Botting
RA
,
Turco
MY
,
Vento-Tormo
M
,
Meyer
KB
, et al
Single-cell reconstruction of the early maternal–fetal interface in humans
.
Nature
2018
;
563
:
347
53
.
33.
Efremova
M
,
Vento-Tormo
M
,
Teichmann
SA
,
Vento-Tormo
R
. 
CellPhoneDB: inferring cell–cell communication from combined expression of multi-subunit ligand–receptor complexes
.
Nat Protoc
2020
;
15
:
1484
506
.
34.
Yaddanapudi
K
,
Rendon
BE
,
Lamont
G
,
Kim
EJ
,
Rayyan
NA
,
Richie
J
, et al
MIF is necessary for late-stage melanoma patient MDSC immune suppression and differentiation
.
Cancer Immunol Res
2016
;
4
:
101
12
.
35.
Simpson
KD
,
Templeton
DJ
,
Cross
JV
. 
Macrophage migration inhibitory factor promotes tumor growth and metastasis by inducing myeloid-derived suppressor cells in the tumor microenvironment
.
J Immunol
2012
;
189
:
5533
40
.
36.
Funamizu
N
,
Hu
C
,
Lacy
C
,
Schetter
A
,
Zhang
G
,
He
P
, et al
Macrophage migration inhibitory factor induces epithelial to mesenchymal transition, enhances tumor aggressiveness and predicts clinical outcome in resected pancreatic ductal adenocarcinoma
.
Int J Cancer
2013
;
132
:
785
94
.
37.
Hansel
DE
,
Rahman
A
,
Wehner
S
,
Herzog
V
,
Yeo
CJ
,
Maitra
A
. 
Increased expression and processing of the alzheimer amyloid precursor protein in pancreatic cancer may influence cellular proliferation
.
Cancer Res
2003
;
63
:
7032
7
.
38.
Klasen
C
,
Ziehm
T
,
Huber
M
,
Asare
Y
,
Kapurniotu
A
,
Shachar
I
, et al
LPS-mediated cell surface expression of CD74 promotes the proliferation of B cells in response to MIF
.
Cell Signal
2018
;
46
:
32
42
.
39.
Pylayeva-Gupta
Y
,
Das
S
,
Handler
JS
,
Hajdu
CH
,
Coffre
M
,
Koralov
SB
, et al
IL35-producing B cells promote the development of pancreatic neoplasia
.
Cancer Discov
2016
;
6
:
247
55
.
40.
Lee
KE
,
Spata
M
,
Bayne
LJ
,
Buza
EL
,
Durham
AC
,
Allman
D
, et al
Hif1a deletion reveals pro-neoplastic function of B cells in pancreatic neoplasia
.
Cancer Discov
2016
;
6
:
256
69
.
41.
Gunderson
AJ
,
Kaneda
MM
,
Tsujikawa
T
,
Nguyen
AV
,
Affara
NI
,
Ruffell
B
, et al
Bruton tyrosine kinase–dependent immune cell cross-talk drives pancreas cancer
.
Cancer Discov
2016
;
6
:
270
85
.
42.
Camp
JG
,
Sekine
K
,
Gerber
T
,
Loeffler-Wirth
H
,
Binder
H
,
Gac
M
, et al
Multilineage communication regulates human liver bud development from pluripotency
.
Nature
2017
;
546
:
533
8
.
43.
Nakanishi
K
,
Yoshimoto
T
,
Tsutsui
H
,
Okamura
H
. 
Interleukin-18 regulates both Th1 and Th2 responses
.
Annu Rev Immunol
2001
;
19
:
423
74
.
44.
Rabinovich
GA
,
Toscano
MA
. 
Turning “sweet” on immunity: galectin–glycan interactions in immune tolerance and inflammation
.
Nat Rev Immunol
2009
;
9
:
338
52
.
45.
Barkal
AA
,
Brewer
RE
,
Markovic
M
,
Kowarsky
M
,
Barkal
SA
,
Zaro
BW
, et al
CD24 signalling through macrophage Siglec-10 is a target for cancer immunotherapy
.
Nature
2019
;
572
:
392
6
.
46.
Vendel
AC
,
Calemine-Fenaux
J
,
Izrael-Tomasevic
A
,
Chauhan
V
,
Arnott
D
,
Eaton
DL
. 
B and T lymphocyte attenuator regulates B cell receptor signaling by targeting Syk and BLNK
.
J Immunol
2009
;
182
:
1509
17
.
47.
Lin
W
,
Noel
P
,
Borazanci
EH
,
Lee
J
,
Amini
A
,
Han
IW
, et al
Single-cell transcriptome analysis of tumor and stromal compartments of pancreatic ductal adenocarcinoma primary tumors and metastatic lesions
.
Genome Med
2020
;
12
:
80
.
48.
Schlesinger
Y
,
Yosefov-Levi
O
,
Kolodkin-Gal
D
,
Granit
RZ
,
Peters
L
,
Kalifa
R
, et al
Single-cell transcriptomes of pancreatic preinvasive lesions and cancer reveal acinar metaplastic cells' heterogeneity
.
Nat Commun
2020
;
11
:
4516
.
49.
Dominguez
CX
,
Müller
S
,
Keerthivasan
S
,
Koeppen
H
,
Hung
J
,
Gierke
S
, et al
Single-cell RNA sequencing reveals stromal evolution into LRRC15+ Myofibroblasts as a determinant of patient response to cancer immunotherapy
.
Cancer Discov
2020
;
10
:
232
53
.
50.
Connor
AA
,
Denroche
RE
,
Jang
GH
,
Lemire
M
,
Zhang
A
,
Chan-Seng-Yue
M
, et al
Integration of genomic and transcriptional features in pancreatic cancer reveals increased cell cycle progression in metastases
.
Cancer Cell
2019
;
35
:
267
82.e7
.
51.
Aung
KL
,
Fischer
SE
,
Denroche
RE
,
Jang
G-H
,
Dodd
A
,
Creighton
S
, et al
Genomics-driven precision medicine for advanced pancreatic cancer - early results from the COMPASS Trial
.
Clin Cancer Res
2017
;
24
:
1344
54
.
52.
Porter
RL
,
Magnus
NKC
,
Thapar
V
,
Morris
R
,
Szabolcs
A
,
Neyaz
A
, et al
Epithelial to mesenchymal plasticity and differential response to therapies in pancreatic ductal adenocarcinoma
.
Proc National Acad Sci U S A
2019
;
116
:
26835
45
.
53.
Ligorio
M
,
Sil
S
,
Malagon-Lopez
J
,
Nieman
LT
,
Misale
S
,
Pilato
MD
, et al
Stromal microenvironment shapes the intratumoral architecture of pancreatic cancer
.
Cell
2019
;
178
:
160
75.e27
.