Cdh11^-/- (https://www.jax.org/strain/023494) (25) and KPC (p48-Cre;LSL-Kras^G12D/+;LSL-Trp53^R172H/+ ) mice were bred to each other to generate cohorts of KPC-Cdh11^+/+ , and KPC-Cdh11^+/- as previously described (23). Mice were housed in a pathogen-free environment under standard conditions at Georgetown University (GU) and Lawrence Livermore National Laboratory (LLNL). All animal work was conducted under approved Institutional Animal Care and Use Committee (IACUC) protocols at GU and LLNL and conformed to the National Institute of Health (NIH) guide for the care and use of laboratory animals.

KPC-Cdh11^+/+ , KPC-Cdh11^+/- and Cdh11^+/+ cohorts (n≥4, mixed sex) were euthanized at 4-5 months of age (23). Pancreases were excised and single-cell suspensions were prepared using a collagenase digestion solution (3 mg/ml Collagenase I (ThermoFisher Scientific Catalog #17018029), 100 µg/mL DNase I (Roche Catalog #11284932001), Dispase II (Roche Catalog #4942078001) in RPMI1640 media supplemented with 10% FBS (ThermoFisher Scientific, Catalog #11875101) for 1 hour with shaking at 37°C. Digests were filtered through a 100 µm cell strainer and depleted of red blood cells using ACK lysis buffer, per manufacturer guidelines (Gibco, Catalog #A1049201). Remaining cells were analyzed for viability and CD45 expression. CD45^-/7-AAD^- (non-immune) and CD45⁺/7-AAD^- (immune) cells populations were sorted via fluorescently activated cell sorting (FACS) and collected for scRNA-seq preparation.

Cells from digested pancreases were stained in a 1:100 dilution of APC/Cyanine7 anti-mouse CD45 (Biolegend, Catalog #103116) in PBS with 1% FBS and incubated for 20 minutes in the dark. Cells were resuspended in 7-AAD Viability Staining Solution (BioLegend, Catalog #420404) and analyzed on a BD FACSMelody cytometer, sorting CD45⁺/7-AAD^- and CD45⁺/7-AAD^- populations for scRNA-seq analysis.

Single-cell libraries were prepared from sorted cell populations from KPC-Cdh11^+/+ , KPC-Cdh11^+/- and Cdh11^+/- mice using the Chromium Single Cell 3′ GEM, Library & Gel Bead Kit v3 (10x Genomics, Pleasanton, CA, USA; catalog no. 1000075) on a 10x Genomics Chromium Controller following manufacturers protocol and sequenced using an Illumina (San Diego, CA, USA) NextSeq 500 sequencer as described before (26). The scRNA-seq data was demultiplexed and aligned against mouse reference genome mm10 using Cell Ranger Single-Cell Software Suite (10x Genomics, Pleasanton, CA, USA) to obtain Unique Molecular Identifier (UMI) counts. Subsequent analysis was performed in Seurat (27) as described before (26). Briefly, after pre-processing, normalization, feature selection and data scaling, data from various experimental groups were integrated to generate an integrated dataset. Subsequently, dimensional reduction by principal component analysis (PCA), clustering, Uniform Manifold Approximation and Projection (UMAP) reduction, and visualization of clusters were performed in Seurat as described before (26). Genes differentially expressed between clusters were identified using ‘FindMarkers’ function implemented in Seurat. For each experimental group, the cell type proportions were estimated as a ratio of the number of cells in each cell cluster relative to the total number of cells sequenced. Any CD45⁺ clusters found in non-immune scRNA-seq data or CD45^- clusters found in immune data were excluded from the analysis.

Human PDAC scRNA-seq data from early and metastatic tumors (28) were obtained from the Gene Expression Omnibus database (GSE205013). The raw barcode, feature, and matrix were downloaded and analyzed using Seurat (29), as described above to identify the cell types and cell type-specific gene expression as outlined in the original analysis (28). Correlation between CDH11 expression and immune cell infiltration was determined using TIMER (http://timer.cistrome.org/). Correlation between the expression of CDH11 and other genes in human tumors was determined using TNMplot (https://tnmplot.com/). UALCAN (https://ualcan.path.uab.edu) was used to determine protein expression of CDH11 in PDAC patient tumors, using data from Clinical Proteomic Tumor Analysis Consortium (CPTAC).

Tissues were embedded and sectioned as previously described (23). Formalin-Fixed Paraffin-Embedded (FFPE) 5µm thick sections were deparaffinized with xylene and rehydrated through a series of decreasing ethanol concentrations. Heat induced antigen retrieval method (HIER) using either sodium citrate (pH 6) at 65 °C or universal antigen retrieval solution (Unitrieve) at 65°C for 45 minutes was used to expose antigens for immunohistochemical staining. Slides were counterstained with DAPI and stained for proteins of interest using the following primary and secondary antibodies: Pdgfra (Abcam, ab203491), Cdh11 (ThermoFisher 32-1700), Il33 (ThermoFisher, PA5-47007), CD8a (Abcam, ab21769), F4/80 (Abcam, ab111101) and Ly6g (Abcam, ab238132), anti-rabbit IgG Alexa Fluor 488 (Catalog #A21441, ThermoFisher Scientific), anti-rat IgG2a Alexa Fluor 488 (Catalog #A11006, ThermoFisher Scientific) and anti-rabbit IgG Alexa Fluor 594 (Catalog #A11037, ThermoFisher Scientific). IL33, F4/80 and Ly6g IHC staining was quantified using ImageJ (https://imagej.nih.gov/ij/). Total cell numbers were estimated based on DAPI nuclear staining and cell types of interest were identified as cells with positive signal for IL33, F4/80 or Ly6g that colocalized with the nuclear DAPI stain. Cells exhibiting positive signal were then quantitated relative to the total number of cells within an image. CD8a expression was quantified with a Vectra quantitative pathology imaging system and Inform software.

Sera and pancreases collected at the time of euthanasia from KPC-Cdh11^+/+ and KPC-Cdh11^+/- mice were analyzed for cytokine expression using Mouse XL Cytokine Array (R&D Systems). Equal amount of protein lysates or serum belonging to the same experimental group were pooled together before analysis. This data has been previously described (23).

Two-proportion Z-test was used to identify statistically significant differences between cell proportions. A p value <0.05 was considered as significant.

To investigate the role of Cdh11 in PDAC, we isolated non-immune (CD45^-) and immune (CD45⁺) cells from pancreases of KPC-Cdh11^+/+ and KPC-Cdh11^+/- mice and both fractions were analyzed separately using scRNA-seq ( Figure 1A ). First, non-immune scRNA-seq data was computationally analyzed to determine Cdh11 deficiency-induced changes in cancer and stromal cells. CD45^- cells were clustered using an unbiased clustering approach which identified eleven clusters with distinct gene expression profiles ( Figures 1B-D ). The identity of each cluster was determined based on the expression of previously established cell type-specific markers ( Figures 1B, C ) (23, 26, 30, 31).

CAFs were one of the most abundant cell types identified in pancreatic cancer TME and robustly expressed markers such as Pdgfra, Col3a1 and Dcn ( Figures 1C, F ). Based on scRNA-seq data, KPC-Cdh11^+/+ mice had ~22.6% more CAFs in their TME compared to KPC-Cdh11^+/- mice (Two-proportion z-test; p value <0.00001). Consistent with Cdh11 known expression in CAFs, Cdh11 expression was primarily observed in CAF clusters, which also had robust expression of CAF marker Pdgfra ( Figure 1F ). Further IHC analysis showed co-expression of Pdgfra and Cdh11 in KPC-Cdh11^+/+ mice, confirming that CAFs in PDAC robustly express Cdh11 ( Figure 1G , S1 ). Analysis of pancreases from wildtype mice without tumors (Cdh11^+/+ ) revealed that normal fibroblasts also express Cdh11, suggesting that Cdh11 may play a significant role in healthy resident fibroblasts as well ( Figure S2 ). Low level of Cdh11 expression was also observed in cells undergoing epithelial-to-mesenchymal transition (EMT) in PDAC ( Figure S3A ). Interestingly, KPC-Cdh11^+/+ mice had significantly more EMT cells than KPC-Cdh11^+/- mice ( Figure 1E ). We also analyzed previously published human ‘early’ and ‘metastatic’ PDAC scRNA-seq data (28) and found that CDH2⁺ EMT cells expressed low levels of CDH11 ( Figure S3B-C ). However, in human tumors also CAFs were the predominant CDH11-expressing cells ( Figure S3C ).

To further understand Cdh11 deficiency-induced changes in CAFs, we extracted cells from CAF clusters (cluster 0, 1 and 7; Figure 1B ) and analyzed them in more detail. This analysis identified three CAF subtypes including Acta2 ^high myofibroblasts (myCAFs), Il33 ^high inflammatory CAFs (iCAFs), and a Thy1 ^high fibroblast cluster ( Figures 2A-C ). Consistent with our previous findings (23), CAFs from KPC-Cdh11^+/+ mice expressed higher levels of Acta2 than KPC-Cdh11^+/- mice ( Figure 2D ). Expression of other myofibroblast markers such as Tagln and Myl9 were also elevated in KPC-Cdh11^+/+ mice ( Figure 2D ). We also identified several other genes differentially expressed between KPC-Cdh11^+/+ and KPC-Cdh11^+/- derived CAFs. These genes included cystatin C (Cst3), inflammatory cytokines Ccl11, Il33, Il11 and Il6, insulin growth factor 1 (Igf1) and heparin binding growth factors midkine (Mdk) and pleiotrophin (Ptn), all of which had higher expression in KPC-Cdh11^+/+ mice. Proteases Adamts9 and Mmp13 and Wnt inhibitor Dkk2 were found to be upregulated in KPC-Cdh11^+/- mice ( Figure 2D ). Furthermore, a cytokine array analysis showed increased Il33 and Cst3 expression in both tumor and serum from KPC-Cdh11^+/+ mice compared to KPC-Cdh11^+/- mice, while Il11 was highly enriched in KPC-Cdh11^+/+ serum alone and Ccl11 was enriched in tumor alone ( Figure 2E ). Il33 is a member of the IL-1 family of cytokines, secreted by a variety of cells including epithelial, endothelial and fibroblast-like cells (32). It has been suggested that the increased expression of IL33 in CAFs promotes tumor growth and metastasis via modulation of the immune system (32). Consistent with the cytokine data, IHC confirmed a higher number of Il33-expressing CAFs in KPC-Cdh11^+/+ mice ( Figure 3 , S4 ). Il11, Il6, Ccl11, Mdk and Ptn have also been shown to play key roles in cancer progression and immune modulation (33–38), suggesting that altered expression of these genes in KPC-Cdh11^+/- mice may contribute to changes in tumor immune profile. Igf1 signaling has also been implicated in tumor growth, metastasis, drug resistance in PDAC (39).

Next, we analyzed publicly available human pancreatic cancer data (data from CPTAC) and found that CDH11 protein expression was significantly elevated in tumor samples compared to normal ( Figure 2F ). Additionally, we observed a strong positive correlation between the expressions of CDH11 and ACTA2 genes in human PDAC ( Figure 2G ). Consistent with our scRNA-seq data, genes such as IL33 and IGF1 also showed a positive correlation with CDH11 expression while PDGFA was negatively correlated ( Figure S5 ). Together, these findings further confirm the role of Cdh11 in cancer progression and suggest that Cdh11 deficiency in stromal cells may significantly alter the TME.

Next, we investigated how the lack of Cdh11 in the stroma impacts the tumor immune microenvironment. CD45-expressing cells purified and quantified by FACS were found in higher proportion in the KPC-Cdh11^+/- than in the KPC-Cdh11^+/+ pancreas ( Figure S6A ), suggesting an increase in immune infiltration when the Cdh11 is absent from the PDAC stroma. scRNA-seq analyses of these CD45⁺ cells from KPC-Cdh11^+/+ and KPC-Cdh11^+/- mice identified twelve distinct immune cell subtypes including macrophages, neutrophils, dendritic cells (DCs), plasmacytoid dendritic cells (pDCs), B cells and T cells ( Figure 4A ). Cells in cluster 0 had high expression levels for members of the B cell receptor (BCR) signaling complex: Cd79a and Cd79b (40). This cluster also expressed Ms4a1 (CD20) and was identified as CD20^hi B cells ( Figure 4B ). Cluster 8 also expressed moderate levels of Cd79a and Cd79b, in addition to showing enrichment for Jun and Mef2c. This cluster was identified as Jun^hi B cells. Cluster 1 had high transcript levels of monocyte/macrophage genes Cd14 and Csf1r, designating this grouping of cells as monocytes/macrophages (Mono-Mac) (41) ( Figure 4B ). Genes critical to T cell signaling including Thy1, Cd3e and Cd3d were highly expressed in clusters 3, 4, 5 and 10 (42). Cluster 5 had additional markers including Foxp3 and Rora and was identified as Foxp3 T cells while cluster 4 was identified as Cd8 T cells based on the expression of cytotoxic T cell markers Cd8a and Nkg7 ( Figure 4B ). Cluster 10 represented a small proportion of cells expressing high levels of Il22 and Ccr6 and was identified as Il22^hi T cells. Cluster 2 was identified as neutrophils based on high expression of S100a8 and S100a9 and cluster 6 was annotated as proliferating cells based on the expression of Mki67 and Top2a ( Figures 4A, B ). Cluster 7 expressed DC markers while cluster 11 expressed markers of pDCs. Jchain and Igkc, two genes highly expressed in plasma cells were enriched in cluster 9 (43–46).

Consistent with our previous finding (23), T cells and Cd8 T cells were found in high abundance in scRNA-seq data from KPC-Cdh11^+/- tumors (Two-proportion z-test; p<0.05). ( Figure 4C ). IHC analysis further confirmed increased Cd8 T cell infiltration in KPC-Cdh11^+/- tumors ( Figure 4D , S6B-C ). We also found that KPC-Cdh1^+/+ tumors had a slightly higher proportion of FOXP3+ T cells compared to KPC-Cdh11^+/- (P value: 0.18). In addition, we observed elevated expression of FoxP3, interleukin 2 receptor gamma (Il2rg), Il4ra, Il7r, cytotoxic T-lymphocyte-associated protein 4 (Ctla4), and glucocorticoid-induced tumor necrosis factor receptor-related protein (Tnfrsf18) in KPC-Cdh1^+/+ tumors ( Figure 4E ), many of which have been previously suggested to play critical roles in Treg differentiation or function (23, 47–49). We also observed a small (not statistically significant) increase in the proportion of Cd20^hi B cells in KPC-Cdh11^+/- mice while Jun^hi B cells were significantly reduced.

The monocyte/macrophage population was dramatically reduced in KPC-Cdh11^+/- mice compared to KPC-Cdh11^+/+ mice (Two-proportion z-test; p<0.00001) ( Figure 4C ). To further analyze differences in monocyte and macrophage subpopulations between KPC-Cdh11^+/+ and KPC-Cdh11^+/- , all cells from the Mono-Mac cluster (cluster 1; Figure 4A ) were extracted and re-analyzed using Seurat which resulted in five different groups with distinct transcriptional profiles ( Figure 5A ). Cluster 3 expressed monocyte markers such as Plac8 and Ly6c2 while all other clusters highly expressed tumor associated macrophage (TAM) marker Mrc1 and Trem2 ( Figure 5B , S7A ). However, these macrophage-like clusters had distinct gene expression profiles. Cluster 0 showed enrichment for Apoe and several cytokines including Ccl7, Ccl8 and Ccl12 whereas cluster 1 highly expressed markers of M2-like polarization including Arg1 and Chil3 ( Figure 5C ). Cluster 2 showed enrichment for Il1r2, Mmp12 and Dcstamp while cluster 4 expressed osteoclasts markers including Acp5, Ctsk and Mmp9 and likely represent osteoclast-like giant-cells (50). We also observed that monocytes and macrophages from KPC-Cdh11^+/+ mice expressed higher levels of chemokines including Ccl2, Ccl4, Ccl6, Ccl7, Ccl8 and Osm compared to KPC-Cdh11^+/- mice ( Figure 5D ). Increased expression of TAM marker Trem2 (51) was also observed in KPC-Cdh11^+/+ mice ( Figure 5D ). Furthermore, IHC analysis showed that KPC-Cdh11^+/+ mice had more F4/80 expressing TAMs than KPC-Cdh11^+/- mice ( Figure 5E , S7B-C ).

Neutrophils were another major immune cell type identified in the PDAC TME and the proportion of neutrophils was reduced in KPC-Cdh11^+/- mice compared to KPC-Cdh11^+/+ mice (Two-proportion z-test; p <0.01) ( Figure 4C , 6A , S8A-B ). Compared to other immune cell clusters, neutrophils showed significant enrichment for inflammatory genes Il1b, Tnf, Ptgs2, Cxcl2, Cxcl3, and Csf1, a cytokine involved in macrophage recruitment and differentiation ( Figure 6B ). Neutrophils also expressed additional cytokines and chemokines including oncostatin M (Osm), Ccl6 and Cxcl1 and proteases such as Cathepsin A (Ctsa), Ctsb, Ctsc and Ctsc, although Mono-Mac cluster had the highest expression values for these genes ( Figure 6B ). We observed that neutrophils from KPC-Cdh11^+/- mice had lower levels of Csf1 compared to KPC-Cdh11^+/+ mice, which may have contributed to decreased number of TAMs in these mice ( Figure 6C ). In addition, cytokines and chemokines including Osm, Cxcl2, Cxcl3, Ccl3, Ccl4, Ccl6 and Il1b were also downregulated in neutrophils from KPC-Cdh11^+/- mice ( Figure 6C ). Cathepsins including Ctsa, Ctsb and Ctsd were also significantly downregulated in neutrophils from Cdh11-deficient mice ( Figure 6D ). In addition to Mono-Macs and neutrophils, proportion of DCs was also reduced in KPC-Cdh11^+/- mice ( Figure 4C ). Consistent with these results, we found a strong positive correlation between CDH11 expression and neutrophil and myeloid DC infiltration in human PDAC (Rho > 65, P <0.0001; Figure S8C ), suggesting that reduced CDH11 expression in PDAC may contribute to reduced myeloid infiltration to the TME.

A survey of The Cancer Genome Atlas (TCGA) suggests that CDH11 expression is also elevated in many other cancers including breast cancer (BRCA), cholangiocarcinoma (CHOL), colon adenocarcinoma (COAD), esophageal carcinoma (ESCA), head and neck cancer (HNSC), stomach adenocarcinoma (STAD) and glioblastoma (GBM) ( Figure S9 ). It has previously been suggested that increased CDH11 expression indicates a poor prognosis in advanced gastric cancer (52) and triple negative breast cancer (TNBC) (22). These findings suggest that Cdh11 plays a role in multiple cancer types and Cdh11 inhibition may promote survival in these cancers.