Systematic Comparison of Pancreatic Ductal Adenocarcinoma Models Identifies a Conserved Highly Plastic Basal Cell State

Abstract

Intratumoral heterogeneity and cellular plasticity have emerged as hallmarks of cancer, including pancreatic ductal adenocarcinoma (PDAC). As PDAC portends a dire prognosis, a better understanding of the mechanisms underpinning cellular diversity in PDAC is crucial. Here, we investigated the cellular heterogeneity of PDAC cancer cells across a range of in vitro and in vivo growth conditions using single-cell genomics. Heterogeneity contracted significantly in two-dimensional and three-dimensional cell culture models but was restored upon orthotopic transplantation. Orthotopic transplants reproducibly acquired cell states identified in autochthonous PDAC tumors, including a basal state exhibiting coexpression and coaccessibility of epithelial and mesenchymal genes. Lineage tracing combined with single-cell transcriptomics revealed that basal cells display high plasticity in situ. This work defines the impact of cellular growth conditions on phenotypic diversity and uncovers a highly plastic cell state with the capacity to facilitate state transitions and promote intratumoral heterogeneity in PDAC.

Significance: This work provides important insights into how different model systems of pancreatic ductal adenocarcinoma mold the phenotypic space of cancer cells, highlighting the power of in vivo models.

Figure 1.. Single-cell profiling of autochthonous murine PDAC reveals a continuum of transcriptional cell states along the epithelial-mesenchymal transitional axis.

(A) Experimental workflow. (B). Uniform Manifold Approximation and Projection (UMAP) embedding of scRNA-seq profiles of 14,392 cells from 15 independent tumors, classified as classical (purple), basal (green) or mesenchymal cells (orange). (C) UMAP projection of previously described classical and basal expression profiles of scRNA-seq data. Scale represents gene signature score, with color scale from 1st to 99th percentile. (D) Ternary plot depicting the cell state classification probability of each cell, calculated by a Markov absorption classifier trained on a small subset of classical, basal and mesenchymal cells, colored by subtype (purple – classical; green – basal; orange – mesenchymal). (E) Diffusion-based pseudotemporal ordering of cells along the classical-basal-mesenchymal axis. (F) Expression of specific classical, basal, and mesenchymal gene signatures along the pseudotemporal axis. Columns represent the smoothed (sliding window smoother with n=250 cells) expression of individual genes along the pseudotemporal axis. The panel on the leftmost side shows the fractions of classical (purple), basal (green), and mesenchymal (orange) cells along the same pseudotime axis using the same smoothing approach. (G) Left: Representative UMAP embedding displaying the expression of candidate marker genes, Lgals4 (classical), Krt17 (basal), and Vim (mesenchymal). Scale represents size factor normalized log2-transformed UMI counts, with color scale from 1st to 99th percentile. Right: 400x Representative immunofluorescence images from KPCT PDAC tumors showing galectin-4, CK17, and vimentin in PDAC cells (green in respective panels, from left to right); tdTomato⁺ cancer cells are red. (H) Schematic depicting model of axis of heterogeneity in PDAC.

Figure 2.. The fidelity of cancer cell states is dependent on the model system.

(A) Orthotopic modeling workflow. Note that in addition to orthotopic tumors, 2D and 3D cell cultures were also processed independently for scRNA-seq directly from culture conditions (red dashed box). (B) UMAP embedding of PDAC cells from autochthonous and orthotopic models after projection into the autochthonous gene expression space. Cells are colored according to subtype (purple – classical; green – basal; orange – mesenchymal) based on Markov adsorption classification. (C) Representative cellular heterogeneity in each PDAC model system. First row: PDAC tissues stained with hematoxylin and eosin at 200x magnification. Scale bar: 100 μm. Second row: UMAP embedding of each model system (colored dots) onto the full data set (grey dots) depicting the distribution of individual cells throughout phenotypic space. Third row: Ternary plots of predicted subtype probability per cell, calculated by Markov absorption colored by subtype (purple – classical; green – basal; orange – mesenchymal). Note that all models recapitulate all three cell states to some degree. Fourth row: Ternary plots depicting the overall subtype classification of each individual tumor sample (dots). Note some inherent sample-to-sample (inter-tumoral) heterogeneity in each model system. (D) Box plot of phenotypic volume based on unbiased global transcriptional heterogeneity of each model system calculated. Vertical black line represents the mean phenotypic volume of the autochthonous samples as reference. Higher values are associated with increased spread of cells over phenotypic space. Each dot represents an individual sample. (E) Box plot of the compositional heterogeneity based the relative frequency of classical, basal, and mesenchymal cells in each tumor. Values represent the Shannon entropy of the fractions of classical, basal, and mesenchymal cells. Higher values indicate a more balanced composition of cell states. Vertical black line represents the mean phenotypic volume of the autochthonous samples as reference. (F) Paired phenotypic heterogeneity in 2D and 3D organoid cell culture and orthotopic transplant. Left panel: Ternary plots of individual cells isolated from 2D (top) and 3D organoid cell culture (bottom). Right panel: Ternary plots of cells isolated from orthotopic tumors derived from the same cell culture conditions displayed on the left. Within each panel, the top plots display the cell state classification of all cells per model and their distribution in combined UMAP space. Note the in vitro enrichment of epithelial phenotypes, and the robust re-establishment of all cell states after orthotopic transplantation. (G) Box plots of (top) the phenotypic volume and (bottom) compositional entropy of cell in in vitro cell culture conditions and their matched orthotopic transplants.

Figure 3.. Lineage-tracing of a subset of basal state cells demonstrates functional phenotypic plasticity.

(A) Pseudotime analysis of cell states demonstrating mixed classical and mesenchymal expression profiles within the basal cell phenotype. Colored lines represent the average model-fitted expression of all pseudotime-dependent genes belonging to either the classical (purple), basal (green) and mesenchymal I (orange) gene expression clusters established in Fig. S2C. Shading represents the area between the 5^th and 95^th percentile of genes. The panel on the top shows the fraction of classical (purple), basal (green) and mesenchymal (orange) along the same pseudotime axis divided into ten bins. (B) Single-cell assay for transposase-accessible chromatin using sequencing (scATAC-seq) of autochthonous KPCT tumors at 7-8 weeks of age (n = 4). (C) Total chromatin accessibility, calculated as number of fragments per cell. (D) Accessibility of representative loci in each cell state. Note co-accessibility at classical (Lgals4) and mesenchymal (Vim) marker genes in the basal state. (E) Lgr5 overlaps with basal cell phenotype in the autochthonous (left) and fragment-based (right) tumor models. UMAPs displaying Lgr5 expression. Scale represents size factor normalized log2-transformed UMI counts, with color scale from 1st to 99th percentile. (F) Schematic of KPF; Rosa26^mTmG/+; Lgr5^CreER/CreER lineage-tracing system. (G) Lineage-tracing of fragment derived KPF; Rosa26^mTmG/+; Lgr5^CreER/CreER tumors. Top: Day 3 and Bottom: Day 10 post-tamoxifen labeling. Left; average expression of Lgr5 in tdTomato⁺ vs. GFP⁺ cells at day 3 and day 10. Note the initial enrichment of Lgr5 gene expression at day 3, which is no longer present at day 10, reflecting transition of the basal cells to other cell states. Right: UMAP embedding displaying the position of GFP⁺ traced cells either as single cells. Note how at day 3 the cells occupy a relatively restricted phenotypic space and diffuse over time. (H) Phenotypic volume of fragment derived GFP⁺ traced cells over time. (I) Immunofluorescence of the classical marker galectin-4 and mesenchymal marker vimentin in the Lgr5-traced cells. At initial labeling, Lgr5-traced cells express GFP but do not express galectin-4 or vimentin. Second insert shows traced cells at day 10, with white arrows indicating co-expression of GFP with galectin-4 (top) or vimentin (bottom) consistent with the acquisition of classical and mesenchymal phenotypes, respectively. (J) Graphical summary of lineage-tracing system demonstrating plasticity of the basal cell population.