Wilms Tumor 1–Expressing Stromal Cells Promote Pancreatic Cancer Progression

WT1-expressing mesenchymal cells in the normal pancreas can give rise to inflammatory cancer-associated fibroblasts in pancreatic cancer that promote cancer growth independent of T-cell responses, expanding understanding of cancer-associated fibroblast origins.

Abstract

Cancer-associated fibroblasts (CAF) are a prevalent cell population in the microenvironment of pancreatic cancer. The pancreas harbors diverse resident cell populations that can differentiate into CAFs, and the cell of origin might contribute to CAF heterogeneity. Expression of the transcription factor Wilms tumor 1 (WT1) marks mesothelial cells as well as a transcriptionally distinct population of fibroblasts in the normal pancreas. WT1 expression also identifies a population of CAFs in both human and mouse pancreatic cancers. In this study, we investigated the contribution of WT1⁺ mesenchymal cells to CAF populations and evaluated the functional role of WT1⁺ stromal cells in pancreatic cancer. Lineage tracing revealed that WT1⁺ cells expand in pancreatic cancer, giving rise to a population of inflammatory CAFs. Depletion of WT1⁺ stromal cells reduced orthotopic tumor growth, with increased immunosuppressive macrophage activation and reduced infiltration of CD8⁺ and FOXP3⁺ T cells. Notably, the reduction in tumor weight observed with WT1⁺ cell depletion was independent of CD8⁺ and CD4⁺ T cells. WT1⁺ CAFs expressed high levels of tumor-promoting ligands that likely interact directly with the tumor epithelium to drive tumor progression. Accordingly, WT1-expressing cell–depleted tumors had reduced epithelial MAPK activation. Together, these data show that WT1⁺ stromal cells represent a tumor-promoting CAF population. Although this population might constitute a potential therapeutic target, caution will be needed to avoid exacerbating immune suppression.

Significance:

WT1-expressing mesenchymal cells in the normal pancreas can give rise to inflammatory cancer-associated fibroblasts in pancreatic cancer that promote cancer growth independent of T-cell responses, expanding understanding of cancer-associated fibroblast origins.

Graphical Abstract

Introduction

Pancreatic ductal adenocarcinoma, the most prevalent form of pancreatic cancer (1), frequently arises from precursor lesions called pancreatic intraepithelial neoplasia (PanIN; ref. 2). Establishment of PanIN is accompanied by expansion of a fibroinflammatory stroma, rich in fibroblasts and immune cells (3–5). In malignant disease, cancer-associated fibroblasts (CAF) represent an abundant and poorly understood cellular component. Whereas CAFs were once thought to promote tumor progression in pancreatic ductal adenocarcinoma (PDAC; refs. 6–11), overwhelming evidence now suggests that CAFs play both tumor-promoting and tumor-restraining roles. These seemingly conflicting roles are potentially mediated by distinct fibroblast populations (12–21) and might explain why attempts to broadly target fibroblasts in clinical trials have had limited success (22).

The advent of new technologies, such as single-cell RNA sequencing (scRNA-seq), has revealed a previously unappreciated level of heterogeneity in pancreatic CAFs. CAFs are now classified by subtypes, including myofibroblastic CAFs (myCAF) and inflammatory CAFs (iCAF; ref. 12). myCAFs express high levels of extracellular matrix and matrix remodeling genes. In contrast, iCAFs express immunosuppressive cytokines and chemokines. Finally, antigen-presenting CAFs (apCAF) express components of the MHC-II antigen-presenting complex (13). Other CAF populations have been defined by expression of markers including fibroblast activation protein, endoglin (CD105), leucine-rich repeat–containing 15 (LRRC15), and complement cascade components (for review, see ref. 23). The heterogeneity of CAFs prompts questions about whether each transcriptional status reflects plasticity and response to tumor-derived signals or is determined by different cells of origin.

The healthy pancreas harbors heterogenous resident fibroblasts and other mesenchymal cells, which derive during embryogenesis from the lateral plate mesoderm (24). Resident fibroblasts include pancreatic stellate cells (PSC) containing vitamin A–rich lipid droplets. PSCs were long described as the cell of origin for all pancreatic CAFs (25). More recently, lineage tracing of PSCs using Fabp4 (a PSC and preadipocyte marker) showed that PSCs contribute only to about 10% of total CAFs, with some variability across individual tumors, and specifically form tumor-promoting CAFs (15). We have previously shown that fibroblasts expressing Gli1, a Hedgehog signaling transcription factor, contribute to up to 50% of the CAF population (16).

Mesothelial cells have been proposed as the cell of origin for apCAFs, capable of engaging CD4⁺ T cells and inducing differentiation into regulatory T cells (Treg; ref. 26). Mesothelial cells are specialized epithelial cells of mesodermal origin that form nonadhesive monolayers lining internal organs and serous cavities (27). Mesothelial cells selectively express Wilms tumor 1 (WT1)—a zinc-finger transcription factor necessary for normal pancreatic development (28) and homeostasis of the adult pancreas (29). Whereas Wt1 has been defined as a mesothelial cell–specific gene in the pancreas (29), it is also expressed in a small percentage of fibroblasts in the pancreas that lack other mesothelial markers (30).

Using a combination of human and mouse pancreatic cancer scRNA-seq data, we discovered that a fraction of CAFs express WT1/Wt1 but lack additional mesothelial markers and do not express apCAF markers. Further characterization revealed that WT1/Wt1⁺ CAFs represent a distinct iCAF population with a unique transcriptomic profile. We thus sought to determine whether Wt1⁺ CAFs arise from Wt1⁺ mesenchymal cells in the normal pancreas through lineage tracing approaches in mice. We determined that Wt1⁺ cells expand during carcinogenesis and largely remain WT1-positive as CAFs. Additionally, we depleted Wt1⁺ cells via a diphtheria toxin (DT)-based depletion model and discovered that Wt1⁺ CAFs promote tumor growth in an immune-independent manner. Wt1⁺ cells express high levels of tumor-promoting ligands; their depletion impairs MAPK signaling in oncogenic KRAS-expressing cancer cells, likely explaining their protumor effect. Together, our data expand our understanding of the origins of CAFs and uncover a broader role for Wt1⁺ CAFs. Furthermore, our data support a protumorigenic role for Wt1⁺ CAFs that is independent of immune regulation.

Materials and Methods

Human samples

Donor pancreata that were ineligible for transplant or had no eligible recipients were collected at the Gift of Life Michigan Donor Care Center as previously described (31). Utilization of donor pancreata for research purposes was approved by the Gift of Life research review group. Human PDAC samples were acquired from patients referred for Whipple or distal pancreatectomy according to Institutional Review Board protocol HUM000025339. Written informed consent forms were received from these patients. All PDAC samples were evaluated by a University of Michigan pathologist, and diagnosis was confirmed in patient charts. Human studies were approved by the Institutional Review Board of the University of Michigan Medical School, and research was conducted in accordance with recognized federal regulations (US Common Rule) and ethical guidelines (Belmont Report).

Mouse strains

All mice used in this study were housed in specific pathogen-free facilities in the Rogel Cancer Center, under the supervision of the University of Michigan Unit for Laboratory Animal Medicine. Age-matched male and female mice were used for experiments, and littermate controls were used whenever possible. Mouse experiments were approved by the University of Michigan Institutional Animal Care and Use Committee under the protocol #PRO00011612.

Wt1 ^CreERT2/+ (#010912, mixed background at the time of use, RRID: IMSR_JAX:010912) and R26tdTomato (#007914, C57BL/6J, RRID: I MSR_JAX:007914) mice were obtained from The Jackson Laboratory. For simplicity, homozygous R26td^{Tomato/tdTomato} and heterozygous R26td^Tomato/+ mice are all referred to as R26tdTomato as they behaved similarly across experiments. Notably, WT1^CreERT2/+ mice express CreER^T2/+ in Wt1⁺ cells (mesothelial cells and fibroblasts) because of the insertion of CreER^T2/+ cDNA into the endogenous WT1 locus. These mice were bred in University of Michigan animal facilities to generate Wt1^CreERT2/+;R26tdTomato mice. To induce expression of Wt1^CreERT2, adult Wt1^CreERT2/+;R26tdTomato mice (3–7 months old) were treated with tamoxifen (4 mg/day, #T5648, Sigma-Aldrich) dissolved in corn oil via oral gavage for 5 days. Relevant controls (R26tdTomato mice) were treated using the same regimen. Two days after the last dose of tamoxifen, mice were euthanized and pancreata were collected. The pancreata of age-matched, untreated Wt1^CreERT2/+;R26tdTomato mice were also collected.

Ptf1a ^FlpO/+;FSF-Kras^G12D/+ (KF) mice were donated by Dr. Howard Crawford. Wt1^CreERT2/+;R26tdTomato mice were bred with KF mice in University of Michigan animal facilities to generate Wt1^CreERT2/+;R26tdTomato; KF mice. At 5 weeks of age, these mice and controls (KF) were treated with tamoxifen dissolved in corn oil via oral gavage for 5 days. Mice were then aged for an additional 5 weeks before they were euthanized and pancreata were collected.

To generate an inducible model of Wt1⁺ cell depletion, Wt1^CreERT2/+;R26tdTomato mice were bred with C57BL/6J R26DTR mice (#007900, RRID: IMSR_JAX:007900) to generate Wt1^CreERT2/+;R26tdTomato/DTR mice. At 2 to 4 months of age, Wt1^CreERT2/+;R26tdTomato/DTR mice and controls had orthotopic tumors implanted in their pancreata (see Materials and Methods: Orthotopic experiments). Four days after orthotopic injection, mice were treated with five daily doses of tamoxifen (4 mg/day by oral gavage) to induce expression of tdTomato and DTR in Wt1-expressing cells. After the last dose of tamoxifen gavage, mice were put on tamoxifen chow and concurrently administered DT (25 μg/kg) every 4 to 7 days until the end of the experiment (3 weeks after orthotopic injection) to maintain efficacious depletion. Controls included a mixture of R26tdTomato/DTR mice treated with both tamoxifen and DT and Wt1^CreERT2/+;R26tdTomato/DTR mice treated with tamoxifen but no DT.

For arginase inhibition experiments, cohoused R26tdTomato and Wt1^CreERT2/+;R26tdTomato mice were randomly assigned to receive 75 mg/kg of CB-1158 or vehicle control (sterile H₂O). These mice received orthotopic tumors and were treated with the same regimen of tamoxifen and DT described above. On the same day as the initial DT injection, mice started receiving CB-1158 or vehicle twice daily (8 hours apart) by oral gavage for the remaining duration of the experiment.

For CD8/CD4 depletion experiments, cohoused R26tdTomato and Wt1^CreERT2/+;R26tdTomato mice were randomly assigned to receive 200 μg each of anti-CD8 (Bio X Cell, BE0061) and anti-CD4 (Bio X Cell, BE0003) or 400 μg of IgG control (Bio X Cell, BE0090). Again, these mice received orthotopic tumors and were treated with the same regimen of tamoxifen and DT described above. On the same day as the initial DT injection, mice received anti-CD8/CD4 or isotype controls by intraperitoneal injection every 4 days until the end of the experiment.

Cell line authentication

Cancer cells used in this study (7940b and 9805) were previously genotyped for anticipated mutations. Likewise, fibroblasts (CD1 WT1) were genotyped to confirm a lack of mutations commonly found in pancreatic cancer cell lines. Cells were then frozen and subsequently maintained at low passage. Experiments in this study were performed with low-passage cells. All cells used in this study were passaged at least one time after thawing and tested for Mycoplasma prior to orthotopic injection/cell culture experiments using the MycoAlert Plus Mycoplasma Detection Kit (LT07-710, Lonza). More information about the origin of the cell lines used in this study can be found in the methods subsections in which they were applied.

Orthotopic experiments

For orthotopic injections, 5 × 10⁴ cells [cell line 7940b (32), C57BL/6J strain Pdx1^cre/+;LSL-Kras^G12D/+;LSL-Trp53^R172H (KPC)] were resuspended in 50 μL of 1:1 ratio of DMEM (#11965, Gibco) supplemented with 10% heat-inactivated FBS (#A3840202, Gibco) and Matrigel matrix basement membrane (#354234, Corning) and injected into the pancreas of mice backcrossed into a C57BL/6J background (RRID: IMSR_JAX:000664). 7940b cells were previously derived from a male KPC mouse and generously provided by Dr. Gregory Beatty (32). In all experiments, tumors were allowed to grow for 3 weeks prior to takedown.

scRNA-seq

For sequencing of murine tissue, normal pancreata or orthotopic tumors were dissected from mice. Tissues were then minced and incubated in collagenase V (1 mg/mL in DMEM) at 37°C for 30 minutes to enzymatically digest the tissue. To obtain a single-cell solution, tissues were then subsequently filtered through 500-, 100-, and 40-μm mesh filters. Dead cell removal was conducted using the MACS Dead Cell Removal Kit (Miltenyi Biotec). The University of Michigan Advanced Genomics Core (RRID: SCR_025788) prepared single-cell cDNA libraries and sequenced samples using the 10× Genomics Platform. Samples were run using 50-cycle paired-end reads on the NovaSeq 6000 (Illumina, RRID: SCR_016387) to a read depth of 100,000. For depletion experiments, Cell Ranger count version 6.0.1 was used, with an expected cell count of 10,000. For sequencing of untreated normal mouse and orthotopic mouse pancreata, Cell Ranger count version 6.1.2 (10x Genomics) was used. R version 4.4.0, RStudio version 2024.12.0+467, and R package Seurat version 5.2.1 (33, 34) were used for scRNA-seq data analysis. Data were filtered to include cells that expressed at least 100 genes; data were also filtered for genes that were expressed in more than three cells. Normalization was then performed using the NormalizeData function with the LogNormalize and a scale factor of 10,000. Data were then further filtered to include cells with RNA counts between 1,000 and 60,000 and <15% mitochondrial genes. Variable genes were identified using the FindVariableFeatures function. Data were then scaled using the ScaleData function and batch-corrected using linear regression on the RNA counts. Principal component analysis (PCA) was then conducted using the RunPCA function, and the PCA results were used to identify dimensions that account for approximately 90% of the data variance. Cells were clustered using the FindNeighbors and FindClusters functions, using the dimensions defined by PCA. Uniform Manifold Approximation and Projection (UMAP) clustering was performed using the RunUMAP function.

Using the FindMarkers function and user-defined criteria, clusters were identified and assigned cell type names. To perform differential gene expression analysis between populations, we used the FindMarkers function and model-based analysis of single-cell transcriptomics (MAST; ref. 35). For calculation of module scores associated with Pi16⁺ and Col15a1⁺ fibroblasts, we used the AddModuleScore function and a list of genes identified by Buechler and colleagues (36) that were significantly differentially expressed between the two populations (fold change >1 or < −1, adjusted P value equal to or less than 0.05). For Pi16⁺ fibroblasts, these genes included Pi16, Dpp4, Ly6c1, Fn1, Ly6a, Igfbp6, Emilin2, Cd248, Sema3c, Igfbp5, Has1, and Cxcl13; for Col15a1⁺ fibroblasts, these genes included Col15a1, Hspg2, Col4a1, Sparcl1, Bgn, Igfbp7, Mgp, Cygb, Smoc2, Apoe, Mfap4, and Cxcl14. Significance of module scores was evaluated between groups using the Wilcoxon rank-sum test with Bonferroni correction.

Interactome

Interactome analysis was performed as described by Velez-Delgado and colleagues (5). In short, a curated list of ligand–receptor pairs was used to query scRNA-seq data from Wt1⁺ cell–depleted tumors and nondepleted controls. Average expression of each ligand and receptor was calculated for each cell type in each condition. Ligands and receptors expressed below the calculated average median expression were filtered out of the analysis. Additionally, ligands or receptors not expressed in both experimental groups were filtered out. Differences in ligand and receptor expression were determined by MAST testing (35) using the Seurat FindMarkers function. Differences in ligand and receptor expression between groups were considered significant at P < 0.05. The resulting data were visualized using Circlize (version 0.4.16) to implement Circos in R. Heatmaps within the Circos plots display the average expression of the given ligand or receptor in the given cell type.

Histopathology and IHC

Tissues were fixed in 10% neutral-buffered formalin overnight and stored in 70% ethanol prior to paraffin embedding. Tissues were then dehydrated, paraffin-embedded, and sectioned onto slides. Hematoxylin and eosin staining was conducted according to the manufacturer’s guidelines. IHC was conducted for cleaved caspase-3 (CC3) and FOXP3. These sections were deparaffinized and rehydrated using two serial xylene washes, two serial 100% ethanol washes, and two serial 95% ethanol washes. Slides were then washed with running water for 5 minutes. For antigen retrieval, slides were placed in Antigen Retrieval Citra Solution (BioGenex, HK0860823) and microwaved for 8 minutes. Once cooled to room temperature, slides were then washed with distilled water followed by PBS. Slides were blocked with 1% BSA in PBS for 30 minutes. Primary antibodies were then incubated on slides at 4°C overnight. The next day, slides were washed with PBS and secondary antibodies were applied (1:300) to sections for 45 minutes at room temperature. Sections were then incubated with ABC reagent from Vectastain Elite ABC Kit (Vector Labs, PK-6100) for 30 minutes, followed by DAB (Vector Labs, SK-4100).

Coimmunofluorescence

Tissue sections were deparaffinized and rehydrated using two serial xylene washes, two serial 100% ethanol washes, and two serial 95% ethanol washes. Slides were then washed with running water for 5 minutes. For antigen retrieval, slides were placed in Antigen Retrieval Citra Solution (BioGenex, HK0860823) and microwaved for 8 minutes. Once cooled to room temperature, slides were then washed with distilled water followed by PBS. Slides were blocked with 1% BSA in PBS for 30 minutes. Primary antibodies were then incubated on slides at 4°C overnight. The next day, slides were washed with PBS and secondary antibodies were applied (1:300) to sections for 45 minutes at room temperature. Slides were then incubated with DAPI (1:30,000) for 5 minutes, washed three times with PBS, and mounted with Prolong Diamond Antifade mounting solution (Invitrogen, P36930).

Multiplex IF (tyramide superboost amplification)

Tissue sections were deparaffinized and rehydrated using two serial xylene washes, two serial 100% ethanol washes, and two serial 95% ethanol washes. Slides were then washed with running water for 5 minutes. For antigen retrieval, slides were placed in Antigen Retrieval Citra Solution (BioGenex, HK0860823) and microwaved for 8 minutes. For pSTAT3 staining, 1x EDTA Unmasking Solution (14747S, Cell Signaling Technology) was used for antigen retrieval instead. Once cooled to room temperature, slides were then washed with distilled water followed by PBS. Next, slides were treated with 3% hydrogen peroxidase to quench endogenous peroxidase activity. Slides were blocked with 10% goat serum for 30 minutes. Primary antibodies intended for tyramide superboost amplification were incubated on the slides at 4°C overnight. The Alexa Fluor 488 Tyramide Superboost Kit (Invitrogen, B40922) was then used according to the manufacturer’s instructions. In short, slides were incubated in poly-horseradish peroxidase–conjugated secondary antibody for 1 hour at room temperature. Excess antibody was washed away with three washes of PBS. A tyramide working solution comprised of the tyramide stock solution, hydrogen peroxide, and proprietary reaction buffer were applied to the slides for 10 minutes at room temperature. Then, a 1:11 dilution of the kit’s stop reagent in PBS was added to the slides for an additional 5 minutes at room temperature. Antigen retrieval was repeated as previously described to remove bound primary and secondary antibodies, and slides were blocked with 1% BSA for 30 minutes at room temperature. Remaining primary antibodies were then incubated on slides at 4°C overnight. The next day, slides were washed with PBS and secondary antibodies were applied (1:300) to sections for 45 minutes at room temperature. Slides were then incubated with DAPI (1:30,000) for 5 minutes, washed three times with PBS, and mounted with Prolong Diamond Antifade mounting solution (Invitrogen, P36930).

Imaging and quantification

Hematoxylin and eosin and IHC slides were imaged at 20× magnification using the Olympus BX53F microscope (RRID: SCR_022568) with Olympus DP80 digital camera and CellSens software (Olympus, RRID: SCR_014551). Coimmunofluorescence (co-IF) of CD8 and CD4 were also imaged at 20× with the same microscope and software, using the X-Cite Xylis LED fluorescent light source (Excelitas Technologies). All other co-IF slides were imaged using the Leica Stellaris 5 inverted confocal microscope (RRID: SCR_024663) at ×40 magnification using LAS X software (Leica Microsystems, RRID: SCR_013673). Due to the small size of some of our tumor samples, 2 to 3 images were taken per slide for quantification. When imaging the periphery and center of orthotopic tumors, 2 to 3 images were taken of both regions (4–6 total images). Periphery was defined as images in which the edge of the tumor was visible.

Imaging and quantification of images was done with investigators double-blinded to the experimental condition of the slides to reduce bias. Quantification of absolute cell counts measured by IHC or co-IF were manually counted. Counts were averaged within each slide and reported with each data point representing a mouse. Quantification of percentage of positive staining in co-IF images was done using ImageJ (37). To quantify % area overlap between markers (e.g. %αSMA⁺ area that is tdTomato⁺), the positive area of the reference channel (e.g. tdTomato) was selected and copied onto the channel of interest (e.g. αSMA). The % positive area within the selected area was then measured. Quantification of percent positivity or mean fluorescent intensity (MFI) in select regions of interest (ROI) was done using CellProfiler Image Analysis Software version 4.2.8 (RRID: SCR_007358; e.g. % of nuclei or fibroblasts that are WT1⁺ and/or pSTAT3⁺, MFI of pERK in E-cadherin⁺ cells; ref. 38). In short, nuclei were identified by DAPI and related to cytoplasmic staining to define individual cells. To calculate percent positivity of WT1 and/or pSTAT3, these stains were thresholded using the Otsu method (two classes) and related to nuclear staining. We then counted the total number of nuclei and nuclei that were single-positive and double-positive for each stain. For MFI of E-cadherin⁺ areas, E-cadherin⁺ staining was thresholded using the minimum cross-entropy method and used to identify epithelial cells. The MFI of pERK and pEGFR was measured in CellProfiler using the MeasureImageIntensity function in the defined E-cadherin⁺ ROI.

Flow cytometry

Single-cell suspensions of pancreatic tumors were prepared as previously described (39). Cells were then stained with surface markers CD45, CD326 (EpCAM) and CD140α (PDGFRα) using fluorescently conjugated mAbs listed in Supplementary Table S1. Flow cytometric analysis was performed on a ZE5 Cell Analyzer (Bio-Rad, RRID: SCR_019712) immediately after the staining, and data were analyzed with FlowJo version 10 software (RRID: SCR_008520).

Conditioned media generation

Conditioned media were generated by plating 150,000 cancer cells (top well) and/or 300,000 fibroblasts (bottom well) in a 6-well 0.4 μm polyester membrane transwell. Cells were plated in DMEM with 10% heat-inactivated FBS and 1% penicillin–streptomycin (PS; #15070063, Gibco). The cancer cell line used was the previously described 7940b line used in our in vivo experiments. The fibroblast line, CD1 WT, was previously isolated from a healthy murine pancreas (10). The following day, cells were washed with PBS and switched to low-serum media (DMEM + 1% FBS + 1% PS). After an additional 48 hours, media were collected and centrifuged at 300 RCF for 5 minutes to pellet contaminating cells. The supernatant was then collected and stored at −80°C until media were ready for use.

Treatment of KRAS^G12D cancer cells with conditioned media

In 6-well plates, we plated the 9805 iKRAS cell line (150,000 cells/well) in DMEM + 10% FBS + 1% PS + doxycycline (1 μg/mL, #9891, Sigma-Aldrich). The 9805 cell line was previously generated from a mixed background male iKras^G12D, Trp53^R172H/+ (Ptf1a^Cre/+;TetO-Kras^G12D;Rosa26^rtTa/+; Trp53^R172H/+, RRID: MGI:5569005) mouse (3). The Ptf1a^Cre/+ allele was provided by Dr. Chris Wright (40). After 24 hours, cells were washed with PBS and cultured for an additional 24 hours in DMEM + 10% FBS + 1% PS with (iKRAS “ON”) or without doxycycline (iKRAS “OFF”). The following day, cells were washed with PBS and cultured in low-serum media (DMEM + 1% FBS + 1% PS) or a 1:1 mixture of low-serum media and conditioned media (cancer cell, fibroblast, or cancer cell–fibroblast conditioned media). iKRAS “ON” cells were again supplemented with doxycycline to maintain oncogenic KRAS expression. Cells were collected for protein after 24 hours in conditioned media.

Western blotting

To isolate protein, RIPA buffer (#R0278, Sigma-Aldrich) supplemented with a protease and phosphatase inhibitor cocktail (1:100; #87785, Sigma-Aldrich) was added to cells on tissue culture plates. Cells were incubated in buffer for 20 minutes on ice and manually scraped. Protein concentrations were determined using the Pierce BSA Protein Assay Kit (#23225, Thermo Fisher Scientific) according to the manufacturer’s instructions. Protein was brought to the same concentration, separated using SDS-PAGE, and transferred to a polyvinylidene difluoride membrane using a wet tank (Bio-Rad). Membranes were washed with water and allowed to dry completely before reactivation with methanol. The membranes were then blocked with 5% milk in TBST [1x TBS (#1706435, Bio-Rad) and 0.1% Tween-20 (#P7949, Sigma-Aldrich)] for 1 hour at room temperature. After blocking, membranes were incubated with primary antibody in either 5% BSA in TBST (pERK) or 5% milk (ERK and vinculin) in TBST overnight at 4°C. The next day, primary antibodies were washed off with TBST and replaced with horseradish peroxidase–conjugated secondary antibodies in 5% milk in TBST (2 hours at room temperature). Membranes were washed once more with TBST prior to protein visualization using Clarity Western ECL (#1705060, Bio-Rad) and the ChemiDoc Imaging System (Bio-Rad, RRID: SCR_019037). Quantification of Western blots was completed using ImageJ (RRID: SCR_003070).

Statistical analysis

GraphPad Prism 10 (RRID: SCR_002798) was used for statistical analysis outside of scRNA-seq. All data were presented as means ± SEM. Groups were compared using an unpaired two-tailed Student t test or two-way ANOVA with Tukey multiple comparisons test as described in the figure legends. Comparisons made within the same images were done using a paired two-tailed Student t test. Select multiple comparison results were reported when the interaction between the variables in the two-way ANOVA test were significant (P < 0.05). Correlation plots were analyzed using a Pearson correlation (two-tailed), and data were fit to a simple linear regression model. R² and P values are reported on the correlation plots.

Differential gene expression analysis between groups from scRNA-seq was done using the Seurat FindMarkers function using MAST (35). Adjusted P values can be found within violin plots presented in the figures.

Results

WT1 is expressed by both mesothelial cells and a population of pancreatic fibroblasts

To characterize expression of WT1 in the normal mouse pancreas, we performed co-IF with podoplanin (PDPN), expressed by fibroblasts and mesothelial cells, and PDGFRα/β, encompassing fibroblasts and pericytes. We observed WT1 staining in PDPN⁺, PDGFRα/β⁻ cells at the edge of the normal mouse pancreas, consistent with the mesothelial layer (Fig. 1A). We also observed WT1 staining in a small number of PDGFRα/β⁺ fibroblast-like cells near the outer edges of the tissue but away from the mesothelial layer. To validate this unexpected finding, we analyzed scRNA-seq data of normal mouse pancreas. We identified several clusters of pancreatic cell populations, including a large acinar population, ductal cells, immune cells, and stromal cells, expressing known lineage markers (Fig. 1B; Supplementary Fig. S1A). Relevant to our study, we identified both a fibroblast population expressing Pdpn, Dcn, and Pdgfra and a mesothelial population expressing Dcn, Pdpn, and Krt8, as well as mesothelin (Msln) and Wt1 (Fig. 1C; Supplementary Fig. S1A). Consistent with our co-IF, we also detected a population of tissue-resident fibroblasts (Msln⁻, Dcn⁺, and Fap⁺) expressing Wt1, albeit at lower levels relative to mesothelial cells, and negative for the mesothelial markers Msln and Upk3b (Supplementary Fig. S1B). Whereas most mesothelial cells expressed detectable levels of Wt1, approximately 28.5% of fibroblasts in the normal mouse pancreas were also Wt1⁺ (Supplementary Fig. S1C). Through differential expression analysis, we determined that Wt1⁺ fibroblasts are distinct from both mesothelial cells and Wt1⁻ fibroblasts. Relative to Wt1⁻ fibroblasts, Wt1⁺ fibroblasts had higher expression of Has1, Il33, and Thy1 and lower expression of the chemokine Cxcl14 (Fig. 1D; Supplementary Fig. S1D), all functionally implicated and/or highly expressed in pancreatic cancer (13, 41).

Figure 1.

WT1 is expressed by both mesothelial cells and a population of pancreatic fibroblasts. A, Co-IF staining of healthy mouse pancreata (M1–M5 represent different mice). Blue, DAPI; green, WT1; magenta, podoplanin; white, PDGFRα/β. B, UMAP visualization of scRNA-seq of healthy mouse pancreas (n = 2,15,463 cells). C, Violin plots depicting expression of stromal markers. RBC, red blood cell. D, Heatmap of differentially expressed genes. E, Co-IF staining of healthy donor pancreata obtained through Gift of Life Michigan. Blue, DAPI; green, WT1; magenta, PDGFRα/β. F, UMAP visualization of scRNA-seq data from healthy donor pancreata. G, Violin plots depicting expression of stromal markers WT1 and MSLN in scRNA-seq from healthy donor pancreata. H, Dot plot of WT1 and MSLN expression in healthy donor pancreata. I, Heatmap of differentially expressed genes.

We then sought to validate the mouse data in healthy human pancreata obtained through a collaboration with Gift of Life Michigan (31). Co-IF staining for WT1 and PDGFRα/β in four distinct human pancreata revealed a parenchymal WT1⁺ stromal population (Fig. 1E) that morphologically and geographically did not overlap with mesothelial cells. Using scRNA-seq data from normal human pancreata (n = 6; Fig. 1F; Supplementary Fig. S1E), we observed WT1 expression in a subset of fibroblasts (Fig. 1G and H; Supplementary Fig. S1F). These fibroblasts were negative for mesothelial markers MSLN, LRRN4, UPK3B, and CALB2 (Fig. 1G and H; Supplementary Fig. S1F) and represented 8.5% of total pancreas-resident fibroblasts. Due to the nature of the sampling of human organs, mesothelial cells were not included in tissue blocks or sequencing data. Notably, WT1⁺ fibroblasts from human donor pancreata paralleled the expression profile of Wt1⁺ fibroblasts in the healthy mouse pancreas (Fig. 1I, * denotes differentially expressed genes in both mouse and human WT1⁺ fibroblasts compared with WT1⁻ fibroblasts). Specifically, human WT1⁺ fibroblasts had elevated expression of HAS1, encoding for a hyaluronic acid synthase, and IL33, a tumor-promoting cytokine, (Fig. 1I; Supplementary Fig. S1G), mirroring expression in mouse Wt1⁺ fibroblasts. We also observed species-specific differences, such as the chemokine CXCL14/Cxcl14, which is upregulated in human and downregulated in mouse fibroblasts.

Other studies have defined the heterogeneity of fibroblasts in the normal pancreas based on expression of Hedgehog signaling components (Gli1-3 and Ptch1; refs. 16, 42), CD105 (Eng; ref. 14), or stellate cell markers (Fabp4; ref. 15). These markers did not distinguish Wt1⁺ and Wt1⁻ fibroblasts in mouse pancreata (Supplementary Fig. S2A). We also examined expression of Lrrc15, identified as a unique marker of CAFs (21); we detected no expression, as expected, given these samples were of the normal pancreas. As Wt1⁺ fibroblasts had high expression of the universal fibroblast marker Pi16 (36) in both mouse and human pancreata (Fig. 1D and I), we sought to determine the extent to which Wt1⁺ fibroblasts overlap with this previously defined transcriptional profile and the mutually exclusive Col15a1⁺ fibroblast population (36). Wt1 expression overlapped partially, but incompletely, with Pi16 expression (Supplementary Fig. S2A) and with the Pi16⁺ module score, whereas Wt1⁻ fibroblasts were enriched in the Col15a1⁺ module score (Supplementary Fig. S2B). In both cases, the overlap was only partial. As a PI16⁺ fibroblast signature has yet to be defined in normal human pancreas, we plotted PI16 and COL15A1 expression, together with other fibroblast markers, in the normal human pancreas data (Supplementary Fig. S2C). Consistent with the mouse normal pancreas, we observed partial overlap of PI16⁺ and WT1⁺ fibroblasts and no notable overlap with other previously defined markers.

Collectively, these results show that WT1 expression defines both mesothelial cells and a small proportion of transcriptionally distinct resident fibroblasts in the healthy mouse and human pancreas.

Wt1 ⁺ cells are a source of CAFs

Next, we implemented a lineage tracing system to ultimately determine whether Wt1⁺ cells give rise to CAFs in PDAC. We crossed Wt1^CreERT2/+ mice with mice conditionally expressing the tdTomato reporter from the Rosa26 locus (R26tdTomato). In the resulting Wt1^CreERT2/+;R26tdTomato mice, administration of tamoxifen induces Cre recombination of the stop cassette in the Rosa26 locus to drive expression of tdTomato in the recombined cells and their descendants (Fig. 2A).

Figure 2.

Wt1 ⁺ cells are a source of CAFs. A, Genetic scheme of Wt1^CreERT2/+;R26tdTomato mice and experimental design of Wt1⁺ cell lineage tracing in healthy pancreas. d, day. B, Co-IF of mouse pancreata from R26tdTomato controls treated with tamoxifen and Wt1^CreERT2/+;R26tdTomato (Wt1‐CreER;tdTom) treated with or without tamoxifen. Blue, DAPI; red, tdTomato; green, WT1. C, Scheme of Wt1⁺ cell lineage tracing in a pancreatic orthotopic tumor model. D, Co-IF of R26tdTomato (tdTom) and Wt1^CreERT2/+;R26tdTomato orthotopic tumors. Images were taken at the edge (periphery) and center of these tumors. Blue, DAPI; red, tdTomato; green, αSMA; white, WT1. E, Percentage of tdTomato+ area; each dot represents an animal. F, Percentage of αSMA⁺;tdTomato⁺ area at the periphery and center of Wt1^CreERT2/+;R26tdTomato orthotopic tumors. G, Summary image. H, UMAP visualization of scRNA-seq data from orthotopic tumors. RBC, red blood cell. I, Violin plots depicting expression of stromal markers. J, Heatmap of differentially expressed genes. Bar plots show the means ± SEM, compared using a two-tailed Student t test.

We first evaluated the efficacy and specificity of labeling Wt1⁺ cells in the normal pancreas. Adult Wt1^CreERT2/+;R26tdTomato mice received tamoxifen by oral gavage over the course of 5 days, and then we harvested pancreata at day 7 (Fig. 2A). By co-IF, we detected tdTomato expression in cells with WT1⁺ nuclei in the mesothelial layer surrounding the pancreas (Fig. 2B). Although we observed expression of WT1 in a population of parenchymal fibroblasts, we did not see expression of tdTomato in this population (Supplementary Fig. S3A), possibly explained by lower levels of Wt1, and thus Cre^ERT2, expression that fail to drive recombination (Supplementary Fig. S1C and S1D). Notably, control mice did not have expression of tdTomato, indicating specificity of the staining and no leakiness of the Cre^ERT2 line.

To evaluate tdTomato labeling over an extended period of time, we treated adult Wt1^CreERT2/+;R26tdTomato mice with tamoxifen for 5 days and then aged them for 45 days after the final dose (Supplementary Fig. S3B). Whereas tdTomato expression was largely restricted to the mesothelium, we identified rare, sporadic expression of tdTomato in PDGFRα/β⁺ fibroblasts (Supplementary Fig. S3C and S3D) throughout the tissues, possibly indicating that the mesothelium contributes to the resident fibroblast pool during pancreas homeostasis.

We then designed a lineage tracing experiment to follow the fate of Wt1⁺ cells during cancer growth. For this purpose, we administered tamoxifen to Wt1^CreERT2/+;R26tdTomato mice; 3 days after the last dose of tamoxifen, we implanted KPC pancreatic cancer cells 7940b orthotopically into the pancreas of syngeneic mice (Fig. 2C). Orthotopic tumors grew for 3 weeks before mice were taken down for histology. We observed a substantial population of tdTomato-positive cells that seemed to grow out from the mesothelial layer and further give rise to fibroblasts at the center of the tumor (Fig. 2D). We then costained tissue with the pan-fibroblast marker PDGFRα/β and αSMA, an activated fibroblast marker. Expression of tdTomato partially overlapped with PDGFRα/β and αSMA, with tdTomato⁺ cells constituting a larger percentage of CAFs in the tumor periphery (15% of the PDGFRα/β⁺ area) relative to the center (6% of the PDGFRα/β⁺ area; Fig. 2E and F; Supplementary Fig. S3E and S3F). Notably, the center of the healthy mouse pancreas showed similar prevalence of TdTomato⁺ fibroblasts, suggesting that tdTomato⁺ fibroblasts primarily derive from Wt1⁺ mesothelial cells. Additionally, we observed persistent expression of WT1 in tdTomato⁺ fibroblasts, with approximately 35% of tdTomato⁺, PDGFRα/β⁺ fibroblasts also expressing WT1 (Supplementary Fig. S3G and S3H). However, we also observed some de novo expression of WT1 in fibroblasts of different origin, with approximately 9% of tdTomato⁻, PDGFRα/β⁺ fibroblasts expressing WT1. Thus, whereas WT1 expression in orthotopic tumors preferentially labels fibroblasts derived from Wt1⁺ cells, WT1 expression can also be induced de novo in CAFs. These findings are summarized in Fig. 2G.

To further characterize Wt1-expressing cells in pancreatic cancer, we performed scRNA-seq on orthotopic tumors 3 weeks following implantation. UMAP visualization of cell populations showed expected cell types and, notably, a substantial population of fibroblasts and mesothelial cells (Fig. 2H; Supplementary Fig. S4A). Consistent with the observation that mesothelial cells give rise to fibroblasts in response to cancer, mesothelial cells from orthotopic tumors were significantly enriched in the hallmark epithelial to mesenchymal transition gene set (Supplementary Fig. S4B). Genes that contributed most significantly to this signature enrichment included the noncanonical WNT agonist Cthrc1, which promotes epithelial to mesenchymal transition via activation of WNT, TGFβ, and MAPK signaling (Supplementary Fig. S4C).

Similar to the normal pancreas, we observed expression of Wt1 in both fibroblasts (Msln⁻, Dcn⁺, and Fap⁺) and mesothelial cells (Msln⁺, Dcn⁺, and Fap⁻; Fig. 2I). Notably, the mesothelial cell marker Msln was also expressed in epithelial cells, consistent with its designation as a tumor antigen in PDAC (43). Wt1⁺ CAFs had enriched expression of genes that were also high in Wt1⁺ fibroblasts in the normal pancreas, including Has1, Clec3b, and C3 (Fig. 2J). These data indicate that Wt1 may label a subset of iCAFs and complement-secreting CAFs (12, 44). Wt1⁺ CAFs also expressed high levels of noncanonical WNT agonists, including Wnt4 and Cthrc1, and immunomodulatory cytokines, including Ccl7, Cxcl1, and Ccl2, all of which have known immune-suppressive and tumor-promoting roles in cancer (5, 45–48).

As orthotopic surgeries introduce injury that could confound data interpretation, we sought to validate our lineage tracing results in a spontaneous model of pancreatic cancer development. We bred Wt1^CreERT2/+;R26tdTomato mice with KF mice—a genetically engineered model of PanIN and PDAC (16, 49). The resulting Wt1^CreERT2/+;R26tdTomato; KF mice were treated with five daily doses of tamoxifen at 5 weeks old to label Wt1⁺ cells prior to lesion formation (50). Mice were then taken down at 11 to 12 weeks of age when abundant PanIN lesions were expected, based on known progression in this model (Supplementary Fig. S4D; ref. 49).

Notably, we observed that Wt1^CreERT2/+;R26tdTomato;KF mice had large cystic lesions in the pancreas, surrounded by smaller PanIN lesions (Supplementary Fig. S4E). Co-IF staining for tdTomato, αSMA, and PDGFRα/β revealed that Wt1⁺ stromal cells give rise to fibroblasts surrounding lesions in this spontaneous model of pancreatic cancer (Supplementary Fig. S4E). As expected, tdTomato expression was absent in KF controls.

To assess the relevance of WT1⁺ fibroblasts in human disease, we stained human PDAC samples for WT1, αSMA, and PDGFRα/β by co-IF. Across four different patients, we identified WT1⁺ fibroblasts (yellow arrows) and WT1⁻ fibroblasts (red arrows; Fig. 3A). Additionally, we analyzed scRNA-seq data of human PDAC samples (Fig. 3B) and identified WT1⁺ fibroblasts (MSLN⁻, DCN⁺, and FAP⁺; Fig. 3C). We then performed differential gene expression analysis between WT1⁺ and WT1⁻ fibroblasts. WT1⁺ fibroblasts expressed markers of immunomodulation, including S100A10, LY6E, C3, and HAS1, growth factors IGF1 and FGF18, and the noncanonical WNT agonist CTHRC1 (Fig. 3D–F), consistent with an ability to promote MAPK activation (48, 51, 52). Taken together, Wt1⁺ cells give rise to fibroblasts in different models of mouse PDAC and identify Wt1⁺ CAFs as a transcriptionally distinct subset of CAFs in both mouse and human pancreatic cancers.

Figure 3.

WT1 is expressed by CAFs. A, Co-IF staining of human PDAC samples from different patients. Blue, DAPI; green, WT1; magenta, αSMA; white, PDGFRα/β. Yellow arrows, WT1⁺ fibroblasts; red arrows, WT1⁻ fibroblasts. B, UMAP visualization of human PDAC scRNA-seq data. C, Violin plots depicting expression of stromal markers. D, Dot plot of WT1 and MSLN expression. E, Heatmap of significantly differentially expressed genes. F, Violin plots of select genes.

Wt1 ⁺ stromal cells promote pancreatic cancer progression

To directly assess the functional role of Wt1⁺ mesenchymal cells in pancreatic cancer, we used a genetically engineered model of Cre-inducible DT receptor (DTR) expression. We crossed mice homozygous for the DTR allele with Wt1^CreERT2/+;R26tdTomato mice to generate Wt1^CreERT2/+;R26tdTomato/DTR mice. Upon administration of tamoxifen, we obtained constitutive expression of tdTomato and DTR in cells expressing Wt1. By administering DT, DTR-expressing cells can be specifically ablated, as DT blocks protein translation and consequently induces cell death by apoptosis (53).

We implanted syngeneic orthotopic tumors in Wt1^CreERT2/+;R26tdTomato/DTR mice and controls. Of note, controls were chosen to account both for potential off-target effects of DTR and genotype-specific changes in tumor growth (e.g., due to the partial loss of endogenous Wt1 alleles). Starting 4 days after tumor implantation, mice were treated with five daily doses of tamoxifen to induce expression of tdTomato and DTR in Wt1⁺ stromal cells.

Following tamoxifen gavage, mice were put on tamoxifen chow and administered intraperitoneal injections of DT every 4 to 7 days for the duration of the experiment (Fig. 4A). This protocol leads to continued depletion of Wt1⁺ fibroblasts and their progeny as well as cells activating Wt1 de novo in tumors. Mice were taken down 16 or 24 days after orthotopic implantation. Interestingly, we observed a significant decrease in tumor weight in Wt1^CreERT2/+;R26tdTomato/DTR mice with DT administration relative to controls both at day 16 (controls = R26tdTomato/DTR, +DT, n = 5) and day 24 (controls = R26tdTomato/DTR, +DT, n = 7; Wt1^CreERT2/+;R26tdTomato/DTR, −DT, n = 3; Fig. 4B). To understand the reasons underlying smaller tumors, we performed immunostaining for CC3, a marker of apoptosis, and Ki67, a proliferation marker. We observed no change in CC3 (Supplementary Fig. S5A). Consistent with the reduction in tumor weight, orthotopic tumors from depletion mice at day 16 had reduced expression of Ki67, mostly affecting E-cadherin⁺ tumor cells (Fig. 4C). Because three of seven Wt1-depleted mice lacked detectable tumors at day 16, all succeeding analyses were done for the day-24 time point.

Figure 4.

Wt1 ⁺ stromal cells promote pancreatic cancer progression. A, Genetic scheme of Wt1^CreERT2/+;R26tdTomato/DTR murine model and experimental design of Wt1⁺ cell depletion in orthotopic PDAC. d, day. B, Absolute tumor weights (mg) from control and Wt1^CreERT2/+;R26tdTomato/DTR orthotopic tumors treated with tamoxifen and DT (depletion) at days 16 and 24 after implantation. Day 16 controls include R26tdTomato/DTR mice treated with tamoxifen and DT. Day 24 controls include R26tdTomato/DTR mice treated with tamoxifen and DT (empty points, n = 7) and Wt1^CreERT2/+;R26tdTomato/DTR mice treated with tamoxifen (filled points, n = 3). C, Co-IF staining of orthotopic tumors from day 16 R26tdTomato/DTR and Wt1^CreERT2/+;R26tdTomato/DTR mice. Blue, DAPI; white, E-cadherin; green, Ki67. Quantification shows %Ki67⁺ area; each dot represents an animal. D, Co-IF staining of orthotopic tumors from day 24 Wt1^CreERT2/+;R26tdTomato/DTR mice that did or did not receive DT. Blue, DAPI; red, tdTomato; green, PDGFRα/β. Quantification shows %tdTomato⁺ area in the periphery and center of orthotopic tumors. E, Quantification of flow cytometry data from Wt1^CreERT2/+;R26tdTomato/DTR mice that did (depletion) or did not (control) receive DT. F, UMAP representation of scRNA-seq data of orthotopic tumors from R26tdTomato/DTR (tdTom/DTR) and Wt1^CreERT2/+;R26tdTomato/DTR (Wt1-CreER;tdTom/DTR) mice treated with tamoxifen and DT. RBC, red blood cell. G, Stacked bar graph of relative cell type abundance represented in scRNA-seq data of orthotopic tumors from R26tdTomato/DTR and Wt1^CreERT2/+;R26tdTomato/DTR mice. H, UMAP representation of fibroblasts from R26tdTomato/DTR and Wt1^CreERT2/+;R26tdTomato/DTR orthotopic scRNA-seq dataset. I, Stacked bar graph of fibroblast populations.

Co-IF for tdTomato and PDGFRα/β revealed a reduction in tdTomato⁺ fibroblasts in tumors from Wt1^CreERT2/+;R26tdTomato/DTR mice with depletion of Wt1⁺ cells (+DT) compared with those that retained Wt1⁺ cells (−DT), consistent with successful depletion of Wt1⁺ cells (Fig. 4D). Wt1⁺ mesenchymal cell depletion with DT was further confirmed via flow cytometry, measured as percentage of tdTomato⁺;PDGFRα⁺ cells of total and of CD45⁻;EpCAM⁻ stromal cells in Wt1^CreERT2/+;R26tdTomato/DTR tumors (Fig. 4D and E; Supplementary Fig. S5B). Notably, there was minimal difference in the percentage of total PDGFRα⁺ cells, indicating that the reduction in tumor weight is not simply a consequence of fewer fibroblasts in Wt1⁺ cell–depleted tumors.

To investigate potential causes of tumor growth reduction, we performed scRNA-seq on orthotopic tumors from R26tdTomato/DTR controls and Wt1^CreERT2/+;R26tdTomato/DTR mice that were treated with both tamoxifen and DT (Fig. 4F; Supplementary Fig. S5C). Gross cell population percentages suggested a slight reduction in fibroblasts and T cells (Fig. 4G), although single-cell data are not an accurate readout of cell composition.

To understand how depletion of Wt1⁺ cells transcriptionally affects the stromal compartment, we computationally isolated the fibroblast and mesothelial populations and visualized them via UMAP (Fig. 4H).

We identified six distinct populations of fibroblasts. In the depletion condition (Wt1^CreERT2/+;R26tdTomato/DTR), we observed a loss of population FB4, which is characterized by high expression of Wt1 as well as genes associated with the iCAF phenotype, including the tumor-promoting cytokine Cxcl1 (45, 54) and Has1. (Fig. 4I; Supplementary Fig. S5D and S5E). We also identified the top expressed genes in the other fibroblast subsets (Supplementary Fig. S5E). FB1 had a gene expression profile consistent with myCAFs, including high expression of Acta2, which encodes for αSMA, and Thy1. FB2 was defined by expression of complement pathway component C3, and FB3 was defined by Mki67, suggesting that these clusters represent complement-secreting CAFs and proliferating CAFs, respectively. FB5 expressed high levels of the PSC marker Fabp4 (15) and Saa3, a protumorigenic and immunosuppressive signaling molecule (55). FB6 expressed Pi16 (36), a suspected pan-tissue marker of adventitial fibroblasts. When clustered using conventional PDAC CAF nomenclature (56), FB2, FB4, FB5, and FB6 resembled iCAFs, with low Acta2 and high expression of inflammatory markers (Supplementary Fig. S5F). FB1 and FB3 were myCAFs and proliferating CAFs, respectively. None of the fibroblast clusters showed specific, high expression of canonical apCAF markers, including Cd74 and Slpi (13).

We then performed differential gene expression analysis between fibroblasts in control and Wt1⁺ cell–depleted tumors. Whereas we observed no significant change in expression of the myCAF marker Acta2, we observed a significant decrease in the iCAF marker Has1 (Supplementary Fig. S5G). This is consistent with FB4—the fibroblast population lost upon Wt1⁺ cell depletion—representing a subset of iCAFs. Interestingly, the marker Saa3, which has been shown to mark both iCAFs and apCAFs, increased following Wt1⁺ cell depletion. This is likely because subcluster FB5 increases in relative size following Wt1⁺ cell depletion and is defined by high Saa3 expression (Fig. 4I; Supplementary Fig. S5E). Whereas we did not have a clearly definable apCAF cluster in our scRNA-seq analysis, we observed a significant decrease in the apCAF marker Cd74, consistent with work done by Huang and colleagues (26). Notably, we identified an increase in Fabp4 expression in Wt1⁺ cell–depleted tumors, likely driven by the relative increase in FB5 fibroblasts in these tumors (Supplementary Fig. S5G).

Collectively, these data show that Wt1⁺ cells promote pancreatic cancer progression and resemble a subset of iCAFs; their subsequent deletion correlates with a reduction in tumor cell proliferation.

WT1⁺ cells include a subset of pSTAT3⁺ CAFs

Whereas most markers of iCAFs have been defined on a transcriptional level, the iCAF phenotype has also been defined by activation of JAK/STAT3 signaling (57). To further explore whether our WT1⁺ CAFs represent a subset of iCAFs, we costained for WT1, pSTAT3, and tdTomato in orthotopic tumors from Wt1^CreERT2/+;R26tdTomato/DTR mice treated with DT and controls (R26tdTomato/DTR, +DT, n = 2; Wt1^CreERT2/+;R26tdTomato/DTR, −DT, n = 6). We found that tdTomato⁺ cells partially colocalized with pSTAT3⁺ nuclei in Wt1^CreERT2/+;R26tdTomato/DTR mice (Fig. 5A), consistently accounting for approximately half of the pSTAT3⁺ nuclei in these tumors (n = 6 mice; Fig. 5B). Tumors depleted of Wt1⁺ cells had significantly fewer pSTAT3⁺ nuclei (Fig. 5C). These findings are consistent with our scRNA-seq data, which demonstrate loss of approximately half of the iCAF population upon Wt1⁺ cell depletion (Supplementary Fig. S5G).

Figure 5.

Partial overlap between WT1 and pSTAT3 in mouse and human pancreatic CAFs. A, Co-IF staining of orthotopic tumors from control and Wt1⁺ cell–depleted tumors. Blue, DAPI; red, WT1; green, pSTAT3; gray, tdTomato. Filled-in arrows, WT1⁺, pSTAT3⁻ cells; single pronged arrows, WT1⁻, pSTAT3⁺ cells; double pronged arrows, WT1⁺, pSTAT3⁺ cells. B, Quantification of the relative abundance of nuclei that were single- or double-positive for WT1 and pSTAT3 in orthotopic tumors (n = 6 mice, M1–M6), excluding cells that were double-negative. Statistical analysis was performed using one-way ANOVA. C, Relative abundance of nuclei that were single- or double-positive for WT1 and pSTAT3 averaged across orthotopic tumor samples. D, Percentage of pSTAT3⁺ nuclei in control and Wt1⁺ cell–depleted tumors. Each point represents an average of three fields of view from a single mouse. Statistical analysis was performed using unpaired t test. E, Co-IF staining of human PDAC samples in which PDAC #1–4 represent tumors from different patients. Blue, DAPI; red, WT1; green, pSTAT3; white, PDGFRα/β. Filled-in arrows, WT1⁺, pSTAT3⁻ cells; single pronged arrows, WT1⁻, pSTAT3⁺ cells; double pronged arrows, WT1+, pSTAT3⁺ cells; F, Quantification of the relative abundance of PDGFRα/β⁺ cells that are single-positive, double-positive, or double-negative for WT1 and pSTAT3 in human PDAC samples (n = 4, PDAC #1–4). G, Quantification of the relative abundance of αSMA⁺ cells that are single-positive, double-positive, or double-negative for WT1 and pSTAT3 in human PDAC samples (n = 4, PDAC #1–4).

We then sought to determine whether these findings were relevant to human disease. To accomplish this, we costained human PDAC samples (n = 5) for WT1, pSTAT3, and PDGFRα/β (Fig. 5D). Similar to mouse, our data show partial colocalization of WT1 and pSTAT3 in PDGFRα/β⁺ cells (Fig. 5E and F). The human samples showed more variable prevalence of each fibroblast type, consistent with the more complex genomic landscape and tumor evolution when compared with mouse models. We also stained human samples for αSMA, together with WT1 and pSTAT3, and identified overlap of WT1 and pSTAT3 within αSMA⁺ cells (Supplementary Fig. S6A), consistent with previous evidence of dual-positive pSTAT3⁺, αSMA⁺ fibroblasts (58). This colocalization was more consistent between samples and mirrored the relative abundance of WT1⁺, pSTAT3⁺ fibroblasts observed in mouse tumors (Fig. 5G).

As mesothelial cells were previously identified as the cell of origin to apCAFs in PDAC (26), we sought to determine whether components of the antigen-presenting machinery were enriched in CAFs of Wt1⁺ origin. We thus stained orthotopic tumors from tamoxifen-treated Wt1^CreERT2/+;R26tdTomato/DTR mice for tdTomato, PDGFRα/β, and the MHC-II chaperone CD74 (Supplementary Fig. S6B). CD74 primarily stained small, circular cells, consistent with antigen-presenting immune cells. We observed limited overlap between PDGFRα/β and CD74. When we stratified PDGFRα/β ROI into areas that were tdTomato⁻ or tdTomato⁺, we found that less than 2% of tdTomato⁺, PDGFRα/β⁺ area overlapped with CD74 (Supplementary Fig. S6C). Interestingly, we observed significantly more overlap between tdTomato−, PDGFRα/β⁺ areas with CD74. Overall, our data show that Wt1⁺ stromal cells do not primarily give rise to apCAFs but rather to a large subset of pSTAT3⁺ iCAFs.

Depletion of Wt1⁺ stromal cells enhances the immunosuppressive tumor

Consistent with the partial colocalization of WT1 and pSTAT3, we observed loss of multiple transcriptional iCAF markers in Wt1⁺ cell–depleted tumors relative to controls, including Clec3b and Lyc1 (Fig. 6A; ref. 13). In addition, we observed reduced expression of Cxcl9 and Ccl7 and increased expression of Cxcl12, Cxcl13, Cxcl14, and Ccl11. Considering the inferred immunomodulatory functional role of iCAFs, we queried the effect of Wt1⁺ fibroblast depletion on the tumor-immune microenvironment.

Figure 6.

Depletion of Wt1⁺ cells enhances the immunosuppressive tumor microenvironment. A, Heatmap depicting genes differentially expressed by fibroblasts in orthotopic tumors from tdTom/DTR and Wt1-CreER;tdTom/DTR mice. B, Co-IF staining of macrophages from orthotopic tumors (periphery and center) from Wt1^CreERT2/+;R26tdTomato/DTR mice with and without DT administration. blue, DAPI; red, ARG1; green, F4/80; white, E-cadherin. C, Quantification of F4/80+ area; each dot represents an animal. Controls include R26tdTomato/DTR mice treated with tamoxifen and DT (empty points, n = 2) and Wt1^CreERT2/+;R26tdTomato/DTR treated with tamoxifen (filled points, n = 10). D, Quantification of %ARG1⁺ area of F4/80⁺ area at the periphery and center of Wt1^CreERT2/+;R26tdTomato orthotopic tumors in which each dot represents an animal. E, Heatmap depiction of differentially expressed genes. F, Co-IF staining of CD8⁺ T cells from orthotopic tumors (periphery and center) from Wt1^CreERT2/+;R26tdTomato/DTR mice with and without DT administration. blue, DAPI; red, granzyme B (GZMB); green, CD8α; white, E-cadherin. G and H, Quantification of CD8⁺ cells and of CD8⁺, granzyme B⁺ cells; each dot represents an animal. I, Quantification of FOXP3⁺ cells. J, Experimental design. K, Relative tumor weights. L, Representative co-IF images of CD8⁺ and CD4⁺ cells. blue, DAPI; red, CD8α; green, CD4. Quantification of experiments are means ± SEM. Experiments with two conditions are compared using a two-tailed Student t test, whereas experiments with four conditions are compared using a two-way ANOVA. When two-way ANOVA interactions were not significant, multiple comparisons P values were not plotted.

We first stained for macrophages in Wt1⁺ cell–depleted and control tumors. Tumor-associated macrophages (TAM) are key mediators of immunosuppression in pancreatic cancer, in part due to their high expression of arginase 1 (ARG1; refs. 49, 59). Total TAMs remained similar in both groups (Fig. 6B and C), but we observed significantly more ARG1⁺ TAMs in Wt1⁺ cell–depleted tumors (Fig. 6D). The percentage of ARG1⁺ TAMs in each sample did not correlate with tumor size (Supplementary Fig. S7A and S7B), suggesting that the increase in ARG1⁺ TAMs was not secondary to the reduction in tumor size observed in Wt1⁺ cell–depleted tumors. Increased Arg1 expression in TAMs was also observed in Wt1⁺ cell–depleted tumors from our scRNA-seq data (Fig. 6E). We then stained our orthotopic tumors for CD8a and granzyme B by co-IF. Consistent with the increase in ARG1⁺ TAMs, we observed significantly reduced CD8⁺ T-cell infiltration upon Wt1⁺ stromal cell depletion (Fig. 6F and G) but no difference in granzyme B positivity (Fig. 6H). Unlike the ARG1⁺ TAMs, the number of CD8⁺ T cells per field significantly correlated with tumor size (Supplementary Fig. S7C). This finding is unsurprising given the drastic reduction of CD8⁺ T cells observed upon Wt1⁺ cell depletion. Nonetheless, it may suggest that CD8⁺ T-cell reduction is a consequence of tumor size reduction as opposed to the cause. Additionally, we observed a trending decrease in the number of FOXP3⁺ Tregs by IHC that did not correlate with tumor size (Fig. 6I; Supplementary Fig. S7D and S7E).

To interrogate changes to the tumor-immune microenvironment in Wt1⁺ cell–depleted mice, we computationally isolated and subclustered myeloid and select lymphoid populations (T and NK cells) using previously defined markers of differentiation and function, such as Cd68, S100a8, and Batf3 for myeloid cells and Cd3e and Nkg7 for lymphocytes (Supplementary Fig. S7F–S7J). In myeloid cells, we identified neutrophils and diverse populations of dendritic cells and macrophages (Supplementary Fig. S7F and S7G). These macrophages included Lyve-high tissue-resident macrophages (Lyve-HI TRM) defined by high expression of Lyve, Cd163, and Folr2. In PDAC, these cells primarily produce and remodel the extracellular matrix (60). We also identified monocytic macrophages (Cd14-high, Ccr2-high, Adgre1-low, and Mrc1-low) and nonclassically activated macrophages (Apoe-high, Fcgr3-high, and Lilra5⁺). We saw a slight increase in neutrophils and monocytic macrophages, both associated with immune suppression (61–63), with Wt1⁺ cell depletion. Conversely, we saw a decrease in dendritic cells, which are required for antitumor immunity (Supplementary Fig. S7H; ref. 64).

We then analyzed lymphoid cells and identified multiple populations of CD8⁺ and CD4⁺ T cells (Supplementary Fig. S7I and S7J). Among CD8⁺ T cells, we identified naïve (Tcf7-high, Il7r-high, and Ifng-low) and exhausted (Pdcd1-high, Tox-high, Lag3-high, and Gzmk-high) populations. For CD4⁺ T cells, we identified regulatory (Treg; Foxp3⁺), T follicular helper (Cxcr5⁺, Bcl6⁺, Tox-high, Pdcd1-high, Stat3-high, and Tbc1d4-high), Th17 (Il17a⁺ and Rorc⁺), naïve (Lef1-high, Tcf7-high, Ccr7⁺, and Cd69-low), and effector memory (CD4⁺ T effector memory; Sell⁻ and Ccr7⁻) T cells. We also identified a small population of type 2 innate lymphoid cells (Gata3⁺, Il17rb⁺, and Areg⁺). Notably, there were fewer T cells in the Wt1⁺ cell–depleted tumors, consistent with immunostaining data. However, we observed a relative increase in naïve CD4⁺ T cells in the depleted samples (Supplementary Fig. S7K). Collectively, these data indicate that depletion of Wt1⁺ cells results in significantly smaller tumors and a complex shift in the tumor-immune microenvironment.

Given the unexpected increase in ARG1⁺ TAMs and decrease in CD8⁺ T cells with Wt1⁺ cell depletion, we next sought to determine whether changes in these cell populations had any role in regulating tumor weight upon Wt1⁺ cell depletion. We first sought to target the immunosuppressive macrophage population. To do this, we used the arginase inhibitor CB-1158, which we previously showed can induce CD8⁺ T-cell infiltration (49). As such, we anticipated that arginase inhibition would further reduce tumor weights in combination with Wt1⁺ cell depletion. Mice were implanted with syngeneic orthotopic tumors and treated as shown in Supplementary Fig. S8A. Contrary to our expectations, arginase inhibition had no effect on relative tumor weights (Supplementary Fig. S8B) or absolute tumor weights (Supplementary Fig. S8C).

To determine whether T cells played any role in regulating tumor progression in the context of Wt1⁺ cell depletion, we repeated our previously described schema for depleting Wt1⁺ cells in combination with dosage of anti-CD8/CD4 or isotype controls as in Fig. 6J. Anti-CD8/CD4 had no effect on orthotopic tumor weights in either control or Wt1⁺ cell–depleted mice (Fig. 6K and L; Supplementary Fig. S8D–S8F). Overall, these experiments suggest that the reduction in tumor weights upon Wt1⁺ cell depletion is not dependent on induction of an antitumor immune response.

Wt1 ⁺ stromal cells express protumorigenic ligands

Based on the lack of immune dependency in reducing tumor weights with Wt1⁺ cell depletion, we then queried our scRNA-seq data for signaling changes using a curated list of known ligands–receptors (4, 5, 65). Differences in ligand and receptor expression between experimental groups were determined by MAST testing (35). We plotted interactions predicted to be downregulated with Wt1⁺ cell depletion. From this, we identified numerous decreased interactions between fibroblast-derived ligands and their corresponding receptors expressed by epithelial cells, upon Wt1⁺ cell depletion (Fig. 7A). These ligands include the noncanonical WNT agonists Wnt4 and Cthrc1 and growth factors Fgf18 and Tgfa (Fig. 7B). WNT receptors Fzd2 and Fzd5 and growth factor receptors Fgfr1 and Egfr were expressed most highly in epithelial cells and fibroblasts (Fig. 7C). This suggests that these Wt1⁺ cell derived ligands induce both paracrine epithelial and autocrine fibroblast activation of these pathways, potentially accounting for the tumor-promoting role of Wt1⁺ cells in PDAC.

Figure 7.

Wt1 ⁺ stromal cells express protumorigenic ligands. A, Circos plot of predicted fibroblast–cancer cell interactions. Ligands shown are lost in Wt1^CreERT2/+;R26tdTomato/DTR tumors. Adjusted P < 0.001, and a log-fold change ≥1. B, Feature plots of differentially expressed fibroblast-derived ligands. C, Violin plots depicting expression of WNT receptors (Fzd2 and Fzd5) and growth factor receptors (Fgfr1 and Tgfa). D, Co-IF staining of orthotopic tumors from Wt1^CreERT2/+;R26tdTomato/DTR mice with and without DT administration. Blue, DAPI; red, pERK; green, pEGFR; white, E-cadherin. E, Quantification of pERK MFI in E-cadherin⁺ cells between control and Wt1^CreERT2/+;R26tdTomato/DTR orthotopic tumors treated with tamoxifen and DT (depletion). Bar plots are means ± SEM and compared using a two-tailed Student t test unless otherwise stated. F, Quantification of pEGFR MFI in E-cadherin⁺ cells. Control tumors include R26tdTomato/DTR mice treated with tamoxifen and DT (empty points, n = 2) and Wt1^CreERT2/+;R26tdTomato/DTR treated with tamoxifen (filled points, n = 3–4). G, Experimental scheme. Conditioned media was generated from cancer cells alone (CCM), fibroblasts alone (FCM), or a combination of both (C-FCM). Conditioned media was then applied to iKRAS^G12D cancer cells in the presence (iKRAS^G12D “ON”) or absence (iKRAS^G12D “OFF”) of doxycycline (dox). Cells were collected for protein after 24 hours in conditioned media. H, Representative Western blot of iKRAS^G12D lysates. I, Quantification of pERK normalized to total ERK and vinculin, relative to the DMEM-treated iKRAS^G12D “ON” condition. Statistical analysis was performed using ordinary one-way ANOVA.

Whereas less is known about the functional role of WNT4 in pancreatic cancer, CTHRC1 has been shown to drive PDAC progression and migration through activation of multiple downstream signaling pathways in KRAS-mutant tumor cells, including further activation of MAPK signaling (48). KRAS-driven PDAC development is also dependent on intact EGFR signaling (66, 67). To test whether Wt1⁺ cell depletion limits activation of tumor-promoting pathways in tumor cells, we stained for activated ERK (pERK) and EGFR (pEGFR) in Wt1⁺ cell–depleted tumors and controls (Fig. 7D). Tumor cells were stained for E-cadherin via co-IF. We observed a significant reduction in pERK staining and pEGFR (Fig. 7E and F) in E-cadherin⁺ tumor cells. This result suggests that Wt1⁺ cells promote tumor progression by acting as a source of tumor-promoting ligands for cancer cells.

To further evaluate the ability of fibroblasts to enhance cancer cell MAPK activation, we treated our doxycycline-inducible iKRAS pancreatic cancer cells (cell line: 9805; Ptf1a^Cre/+;TetO-Kras^G12D;Rosa26^rtTa/+; Trp53^R172H/+) with conditioned media from cancer cells (7940b), fibroblasts (CD1 WT), or cancer cell–fibroblast coculture in transwells (Fig. 7G). Relative to plain media, iKRAS cells exposed to cancer-conditioned media had elevated pERK (Fig. 7H and I) with doxycycline treatment (iKRAS “ON”), contributing to previous evidence of a role for autocrine growth factor signaling in promoting cancer progression (67). Similar results were observed with iKRAS cells exposed to conditioned media from cancer cell–fibroblast coculture. Notably, there was no difference in pERK levels when iKRAS cells were cultured without doxycycline (iKRAS “OFF”), indicating that the ability of cancer cell or fibroblast conditioned media to activate MAPK signaling is contingent on cancer cell expression of oncogenic KRAS. Collectively, these data suggest that CAFs can directly promote cancer progression by producing secreted factors that activate core pathways implicated in tumorigenesis.

Discussion

Pancreatic cancer is characterized by the establishment of a fibroinflammatory microenvironment comprised largely of CAFs (3–5). Both CAFs and normal fibroblasts are heterogeneous in terms of gene expression profiles and function. The contribution of normal pancreas cell populations to CAFs and the role of these distinct CAF populations in tumors are areas of active investigation (56, 68). Notably, during embryonic development, mesothelial cells give rise to tissue-resident stromal cells via mesothelial-to-mesenchymal transition (27–29)—a process that can be coopted by mesothelial cells in adult tissues in response to tissue injury or carcinogenesis (27). Whereas previous research did not identify a contribution by mesothelial cells to the maintenance of pancreas-resident fibroblasts during pancreas homeostasis (28), in this study we observed a small population of pancreas-resident fibroblasts derived from Wt1⁺ cells in the normal mouse pancreas. This difference might be attributable to the extended duration of our lineage tracing experiments or the advanced age of our mice (we used 3- to 7-month-old mice whereas previous experiments utilized 4-week-old mice).

Additionally, we show that Wt1⁺ cells (mesothelial cells and fibroblasts) predominantly give rise to a subset of iCAFs in pancreatic cancer. Recent work by Huang and colleagues (26) identified mesothelial cells as giving rise to apCAFs in PDAC, based on transcriptional similarities between the two cell populations. The similarity in gene expression between mesothelial cells and apCAFs is supported by other studies (69), some suggesting that apCAFs actually represent cancer-associated mesothelial cells that have been described in ovarian cancer (70). Our data show that Wt1 expression is maintained in a subset of mouse and human CAFs that is characterized by high expression of immunomodulatory signaling molecules and growth factors but lacks antigen-presenting components. We also confirmed a lack of antigen presentation via staining. Thus, these CAFs constitute a subpopulation of iCAF-like cells that are distinct from apCAFs. Accordingly, depletion of Wt1⁺ cells resulted in loss of these pSTAT3⁺ CAFs. We also demonstrate that these Wt1⁺ cells promote PDAC progression, likely by acting as a source of growth-promoting ligands for cancer cells.

Previous work targeting mesothelial cells and mesothelial cell–derived CAFs with a MSLN blocking antibody to prevent mesothelial-to-mesenchymal transition in vivo resulted in smaller tumor weights and a loss of Tregs (26), consistent with our observation using Wt1⁺ cell depletion. However, whereas treatment with anti-MSLN caused an increase in CD8⁺ T cells (26), our data targeting Wt1⁺ cells through DTR-driven depletion showed an increase in immunosuppressive macrophages and a decrease in CD8⁺ T-cell infiltration. Two scenarios could explain this discrepancy: (i) our model targets Wt1⁺ CAFs that do not express MSLN but instead targets CAFs throughout tumor progression; (ii) MSLN is highly expressed by cancer cells (43), and therefore treatment with anti-MSLN could directly target those cancer cells, resulting in decreased tumor size and increased T-cell infiltration.

Whereas we initially expected the decrease in size of Wt1⁺ cell–depleted tumors to be immune cell–driven (26), we ultimately attributed this change to loss of direct fibroblast–cancer cell cross-talk. Specifically, when Wt1⁺ cells were depleted, we observed loss of fibroblasts expressing high levels of noncanonical WNT agonists (Wnt4 and Cthrc1). WNT4 activates β-catenin–dependent pathways (47) that are known to play an important role in PDAC progression (71) and immunosuppression (39). Interestingly, our laboratory previously showed that WNT signaling in pancreatic cancer promotes epithelial MAPK signaling (10). Additionally, CTHRC1 signaling has also been shown to activate MAPK signaling in PDAC (48). Future experiments aim to identify the specific, targetable fibroblast-derived ligands that drive epithelial MAPK activation in pancreatic cancer. Overall, our data suggest that Wt1⁺ cells are a critical source of tumor-promoting ligands whose loss causes decreased MAPK activation and decreased tumor growth. Our work complements previous understanding of the origin and function of CAFs and highlights interactions between CAFs and cancer cells that might be worth targeting therapeutically.

Supplementary Material

Supplementary Table 1

Antibody information (target antigen, supplier, catalogue number, application, and dilution)

Supplementary Figure 1

Characterization of Wt1/WT1+ fibroblasts in mouse and human normal pancreata by single-cell RNA-sequencing.

Supplementary Figure 2

Comparison of Wt1/WT1 expression in normal pancreata to expression of previously defined mesenchymal lineage markers.

Supplementary Figure 3

Lineage tracing Wt1+ cells in mouse normal pancreata and orthotopic tumors.

Supplementary Figure 4

Characterization of orthotopic single-cell RNA-sequencing object and lineage-tracing in a genetic model of early PDAC progression.

Supplementary Figure 5

Characterization of Wt1-CreER;tdTom/DTR mice by staining, flow cytometry, and single-cell RNA-sequencing.

Supplementary Figure 6

Co-staining of WT1 and pSTAT3 in human PDAC and CD74 in lineage-traced orthotopics.

Supplementary Figure 7

Characterization of immune microenvironmental changes, by staining and single-cell RNA-sequencing, observed with Wt1+ cell depletion.

Supplementary Figure 8

Combinatorial depletion of Wt1+ cells with arginase inhibition and anti-CD8/CD4 treatment.

Data Availability

The complete R script used for analysis and visualization is publicly available on GitHub (https://github.com/PascaDiMagliano-Lab/), and scRNA-seq datasets generated in this study are publicly available at NCBI Gene Expression Omnibus accession number GSE288830. Human scRNA-seq data were previously published by Carpenter and colleagues (NIH dbGaP database accession #phs003229.v1.p1; ref. 31). All other raw data generated in this study are available upon request from the corresponding author.

Authors’ Disclosures

A.C. Bischoff reports grants from the NCI/NIH during the conduct of the study. J. Shi reports grants from the NIH during the conduct of the study and outside the submitted work. No disclosures were reported by the other authors.

Authors’ Contributions

A.C. Bischoff: Conceptualization, formal analysis, funding acquisition, validation, investigation, visualization, methodology, writing–original draft. K. Brown: Investigation, writing–review and editing. E.L. Lasse Opsahl: Investigation, visualization, writing–review and editing. H.R. Watkoske: Investigation, visualization, methodology, writing–review and editing. C.E. Espinoza: Investigation, visualization, writing–review and editing. J.O. Okoye: Investigation, methodology, writing–review and editing. A.C. Olivei: Investigation, writing–review and editing. L.M. Green: Investigation, writing–review and editing. R. Rai: Resources, investigation, writing–review and editing. S. The: Resources, methodology, writing–review and editing. W. Yan: Conceptualization, investigation, methodology, writing–review and editing. A.D. denDekker: Conceptualization, investigation, visualization, writing–review and editing. E.S. Carpenter: Conceptualization, resources, funding acquisition, investigation, methodology, writing–review and editing. J. Shi: Resources, writing–review and editing. F. Bednar: Conceptualization, investigation, methodology, writing–review and editing. T.L. Frankel: Conceptualization, investigation, visualization, writing–review and editing. Y. Zhang: Conceptualization, funding acquisition, investigation, methodology, writing–review and editing. M. Pasca di Magliano: Conceptualization, supervision, funding acquisition, investigation, visualization, methodology, project administration, writing–review and editing.