The non-neoplastic human pancreas and human pancreatic tumor tissues were obtained from the Cooperative Human Tissue Network Southern Division (University of Alabama at Birmingham, AL) (online supplemental tables 1 and 2). Serum specimens of healthy donors and human pancreatic cancer patients were provided by Georgia Cancer Center Biorepository.

OPN mRNA datasets of non-neoplastic human pancreas and human pancreatic tumor were extracted from TCGA database. For determination of patient survival time correlation with OPN expression level, TCGA pancreatic cancer patient survival and OPN expression datasets were extracted from OncoLnc database. The Kaplan-Meier survival curves were generated from the datasets with a cutoff-high 60% (n=105) and cutoff-low 40% (n=70) using the website’s default program. The significance was determined by log rank p value. The patient survival and OPN expression data are presented in online supplemental table 3.

Female and male C57BL/6 mice were purchased from the Jackson Laboratory (Bar Harbor, Maine, USA). All mice used in this study were between 2 and 3 months old at the start of the experiment.

The mouse pancreatic tumor UN-KC-6141 cell line was provided by Surinder K Batra at the University of Nebraska Medical Center (Omaha, NE). UN-KC-6141 cell line was characterized previously.24 PANC02-H7 cells were provided by Dr. Min Li at the University of Oklahoma Health Sciences Center (Oklahoma City, OK). PANC02-H7 cell line were characterized previously.25 J774M cells were sorted from J774 parent cells for CD11b⁺Gr1⁺ cells and characterized as previously described.26 Cell lines were tested bimonthly for mycoplasma contamination and are mycoplasma-free at the time of use.

PANC02-H7 and UN-KC-6141 cells (1×10⁴ cells/mouse) were surgically injected to the pancreas of C57BL/6 mice as previously described.10

WDR5-47 and WDR5-0102 were synthesized and characterized as previously described.27 Cpd23 was synthesized and characterized as previously described.28 In short, for the synthesis of WDR-47 and WDR5-0102, 2-amino-4-nitrofluorobenzene was reacted with the requisite benzoyl chloride in dichloromethane in the presence of pyridine, and the amide thus obtained was then reacted with N-methylpiperidine in DMF (dimethylformamide) in presence of anhydrous potassium carbonate. To synthesize Cpd23, 2-nitro-4-bromofluorobenzene was first reacted with N-methylpiperidine in DMF in presence of anhydrous potassium carbonate and the nitro group then converted to the amine by catalytic hydrogenation. The aromatic amine was then reacted with the appropriate substituted benzoyl chloride in dichloromethane in presence of pyridine. The aromatic bromide was finally coupled with four pyridylboronic acid in dioxane using Pd(PPh3)2Cl2 catalyst and cesium carbonate base. The purity of the compounds is determined by 1H-NMR spectroscopy as >98%. All three compounds were then validated for activity in inhibition of histone methyltransferase activity inhibition in Reaction Biology Corp (Malvern, Pennsylvania, USA).

Inhibition of MLL1 enzymatic activity by WDR5-47, WDR5-0102 and Cpd23 was determined in a 10-dose IC₅₀ mode with 3-fold serial dilution starting at 10 µM with HotSpot format at Reaction Biology Corp. MLL1-WDR5 protein complexes and nucleosome were used. Reactions were initiated by the addition of ³H-S-adenosyl-methionine (1 µM).

HEK293FT cells were cotransfected with pCMV-VSV-G (Addgene #8454), psPAX2 (Addgene #12260) and lentiCRISPRv2 (Genscript, Piscataway, NJ) plasmids containing scramble (GGAAGACTTAGTCGAATGAT) or Spp1-specific (GCAAATCACTGCCAATCTCA) sgRNA-coding sequence using Lipofectamine 2000 (Life Technologies) to produce CRISPR lentivirus.29 PANC02-H7 cells were transduced with the lentivirus particle. Cells were selected with puromycin to generate stable cell lines PANC02-H7.Scramble and PANC02-H7.Spp1 KO, respectively. Cells were cultured in 24-well plate (2.5×10⁵ cells/well in 1 mL medium) for 24 hours. Culture supernatants were collected for OPN protein measurement by ELISA as described below. PANC02-H7.Scramble and PANC02-H7.Spp1 KO cells (2×10⁴ cells/mouse) were surgically injected to pancreas of C57BL/6 mice as previously described.10 The tumor-bearing mice were treated 5 days after tumor cell injection with IgG or anti-PD-1 (clone RMP1-14, 200 µg/mouse, Bio X Cell, Lebanon, New Hampshire, USA) every 2 days via i.p. injection. Mice were sacrifice 20 days after tumor cell injection to analyze tumor.

Human serum was analyzed for OPN protein level using the human OPN ELISA kit (Cat# DOST00, R&D Systems, Minneapolis, MN) according to the manufacturer’s instructions. Tumor cell culture supernatant and mouse serum was analyzed for OPN protein level using the mouse OPN ELISA kit (Cat # DY441, R&D System) according to the manufacturer’s instructions.

The orthotopic PANC02-H7 tumor-bearing mice were treated 5 days after tumor cell injection. The treatment groups include: solvent control (n=8), WDR5-47 (60 mg/kg body weight in PEG300), anti-PD-1 (clone RMP1-14, 200 µg/mouse). WDR5-47 was administered daily via i.p. injection. Anti-PD-1 was administered every 2 days via i.p. injection. Mice were sacrifice 20 days after tumor cell injection to analyze tumor.

Tissue sections were stained as previously described.14 The sections were stained with anti-human OPN antibody (Cat # AF1433. R&D System) and mounted in VectaMount Permanent Mounting Medium (Vector Lab, Burlingame, California, USA).

Tumor tissues were collected from orthotopic PANC02-H7 and UN-KC-6141 tumor-bearing mice. Normal whols pancreas were dissected from tumor-free C57BL/6 mice. The normal pancreas and tumor tissues were rinsed in PBS and used immediately for chromatin preparation using the SimpleChIP Plus Enzymatic Chromatin IP Kit (Cat# 9004. Cell Signaling Tech, Danvers, MA). ChIP was performed using anti-H3K4me3 (Cat# 9751. Cell Signaling Tech) and the SimpleChIP Plus Enzymatic Chromatin IP Kit according to the manufacturer’s instructions. The immunoprecipitated genomic DNA fragments were analyzed qPCR using PCR primers covering the region of Spp1 promoter regions (online supplemental table 4).

Chromatin fragments were prepared from pancreas of tumor-free C57BL/6 mice and tumors and immunoprecipitated as described above and used to construct the DNA library for next-generation sequencing (NGS) by NGS service provider Novogene Corp (Chula Vista, CA). The quality of raw sequencing reads was examined by FastQC and the adaptor and low quality sequences were trimmed and clean by Trim Galore. The cleaned reads were mapped to mouse reference genome (mm10) using Bowtie2. PCR duplicates were identified and removed by Picard tool. H3K4me3 enriched peaks were identified by MACS2 and annotated using ChIPseeker. Differential peaks between orthotopic pancreatic tumor and normal pancreas were identified by DiffBind. The analysis described above were performed in Galaxy server (use.galaxy.org). The entire dataset is deposited in GEO database (Accession # GSE178677).

Tumor tissues were collected and digested in collagenase solution (1 mg/mL collagenase, 0.1 mg/mL hyaluronidase, and 30 U/mL DNase I) and passed through a 100 µm cell filter. Cells were then lysed with red cell lysis buffer, stained with fluorescent dye-conjugated antibodies (online supplemental table 5) and analyzed in a LSRFortessa. All flow cytometry data were analyzed using FlowJo program.

RNA was isolated from tissues using GeneJET RNA Purification Kit (Cat# K0732. Thermo Fisher Scientific) according to manufacturer’s instructions. cDNA was synthesized from total RNA and used for analysis of gene expression by qPCR using gene-specific primers (online supplemental table 4) in a StepOne Plus Real-Time PCR System (Applied Biosystems, Foster City, California, USA).

Tumor tissues were collected from orthotopic PANC02-H7 and UN-KC-6141 tumor-bearing mice. Normal pancreas were dissected from tumor-free C57BL/6 mice. The normal pancreas and tumor tissues were rinsed in PBS and used for total RNA isolation using the GenJet RNA Isolation Kit (ThermalFisher Scientific). Total RNA was used to construct the cDNA library for high throughout DNA sequencing by Novogene. The cleaned reads were mapped to reference genome (mm10) using STAR aligner. The sequence counts for each gene were collected by featureCounts and annotated using annotateMyIDs function. The differential expression analysis between tumor and normal samples was performed using DESeq2. The above-mentioned analysis were performed in Galaxy server (use.galaxy.org). The heatmaps were generated using ComplexHeatmap in R V.3.6.3. KEGG pathway enrichment analysis was performed using clusterProfiler. The entire dataset is deposited in GEO database (Accession # GSE178677).

Cells were cultured in the presence of various concentrations of WDR5-47, WDR5-0102 and Cpd23 for 24 hours. Cells were then lysed, blotted, and probed with anti-mouse OPN antibody (Cat# AF808, R&D Systems and Cat#691302, Biolegend) using procedures as previously described.26

Cell viability assays were performed using CellTiter 96 Aqueous One Solution Cell Proliferation Assay Kit (Cat# G3582, Promega, Madison WI) according to the manufacturer’s instructions.

A two-factor analysis of variance was used to examine whether there was a synergistic effect of WDR5-47 and OPN (WDR5-47/Control or Scramble/Spp1 KO) with anti-PD-1 immunotherapy (anti-PD-1 vs control) for tumor growth control (tumor size and weight). Each model contained fixed effects of WDR5-47 and OPN, anti-PD-1 status, and the two-factor interaction between these two effects. Of statistical interest was the F-test for the two-factor interaction between WDR5-47/OPN and anti-PD-1. If this F-test is statistically significant, this will indicate that a synergistic effect is evident. A Tukey-Kramer multiple comparison procedure was used to examine post hoc pairwise differences between groups.

To profile H3K4me3 deposition change in pancreatic carcinoma genome, we surgically transplanted UN-KC-6141 and PANC02-H7 cells to mouse pancreas to establish orthotopic pancreatic tumor mouse models (figure 1A). Pancreas of UN-KC-6141 and PANC02-H7 tumors were collected and subjected to ChIP and high throughout sequencing. H3K4me3 deposition is enriched throughout the entire genome in both UN-KC-6141 and PANC02-H7 tumors as compared with the normal pancreas (figure 1B,C). As expected, the promoter region of Cd274, the gene encoding PD-L1, has increased H3K4me3 deposition (figure 1B,C). Of interest is the finding that Spp1 and its receptor Cd44 have increased H3K4me3 deposition in their promoter regions (figure 1B). H3K4me3 deposition is enriched in a region flanking the Spp1 transcription start site (figure 1D). Individual ChIP analysis validated that H3K4me3 is enriched around the Spp1 transcription start site in mouse pancreatic carcinoma in vivo (figure 1E).

Consistent with increased H3K4me3 deposition, RNA-Seq analysis determined that Spp1 and Cd44 expression levels are increased in both UN-KC-6141 and PANC02-H7 tumors (figure 2A). Increased Spp1 and Cd44 expression in both pancreatic mouse tumor was validated by qPCR analysis (figure 2B,C). Pathway analysis revealed multiple pathways, including PI3K-Akt, MAPK signaling, proteoglycans in cancer, focal adhesion, phagosome, pancreatic secretion, ECM (Extracellular matrix)-receptor interaction, gap junction are enriched in the tumors (online supplemental figure 1). Among these enriched pathways, PI3K-Akt pathway has a large set of genes with increased expression (online supplemental figure 2). Spp1 and Cd44 are enriched in the ECM-receptor and focal adhesion pathways (online supplemental figure 2). Furthermore, several integrins that function as OPN receptors are also enriched in the ECM-receptor pathways in both tumors in vivo (online supplemental figure 3).

To determine the human relevance of the above finding, we analyzed OPN protein level in no-neoplastic human pancreas and pancreatic tumors. OPN protein level is higher in pancreatic tumors than in non-neoplastic pancreas in all five pancreatic patients analyzed (figure 3A–C). Analysis of OPN mRNA datasets from TCGA (The Cancer Genome Atlas) database also determined that OPN expression level is significantly higher in human pancreatic tumor than in normal pancreas (figure 3D). OPN exists as a secreted protein, analysis of serum OPN determined that OPN protein level is significantly higher in serum from pancreatic cancer patients as compared with serum from healthy donors (figure 3E). Furthermore, OPN expression level is inversely correlated with pancreatic cancer patient survival time (figure 3F).

OPN protein is present in both tumor cells and tumor-infiltrating immune cells in human pancreatic carcinoma (figure 3A). Myeloid cells are major populations of immune cells in pancreatic carcinoma that act as potent immune suppressors. We then sought to determine OPN expression profiles in tumor-infiltrating myeloid cells. A subset of CD11b⁺Gr1⁺ myeloid-derived suppressor cells (MDSCs) are OPN⁺ in both UN-KC-6141 and PANC02-H7 tumors (figure 4B,C). Further analysis of MDSCs into monocytic MDSCs (M-MDSCs) and polymorphonuclear MDSCs (PMN-MDSCs) indicates that tumor-infiltrating OPN⁺ MDSCs are primarily M-MDSCs in both UN-KC-6141 and PANC02-H7 tumors in vivo (figure 4B,C).

We next sought to determine PD-1 and PD-L1 expression profiles in pancreatic tumor. CD4⁺ and CD8⁺ T cells are present in both UN-KC-6141 and PANC02-H7 tumors (figure 5A–C, 5G-I). Approximately 39%–49% tumor-infiltrating CD4⁺ T cells are PD-1⁺ and 50%–61% CD8⁺ T cells express PD-1 (figure 5D,J). More than 50% tumor-infiltrating MDSCs express PD-L1. About 41%–44% M-MDSCs express PD-L1 and almost all PMN-MDSCs are PD-L1⁺ (figure 5E,F,K,L). About 83%–93% of tumor-associated macrophages are PD-L1⁺ cells (figure 5E,F,K,L). Furthermore, all pancreatic tumor cells express high level of PD-L1 in vivo (figure 5E,F,K,L).

The mixed lineage leukemia family of proteins (Mll1-4/Kmt2a-d, SET1A and SET1B, Set1, and Smyd) selectively methylate histone 3 Lys4 (H3K4). The enzyme activity of these histone methyltransferases requires adaptor proteins WDR5, RBBP5 and ASH2L.30 Histone methyltransferase Smyd2/Kmt3c has been shown to be up-regulated in pancreatic cancer to render resistance to chemotherapy.31 RNA-seq analysis revealed a large set of histone lysine methyltransferases, including H3K4me3-selective methyltransferase, are highly upregulated in both UN-KC-6141 and PANC02-H7 tumors in vivo as compared with normal pancreas (figure 6A). WDR5 forms a protein complex with these methyltransferases and is essential for enzyme activity of all H3K4me3-specific histone methyltransferases.27 28 We, therefore, pursued WDR5 inhibition to suppress H3K4me3 in pancreatic tumors. WDR5-47, WDR5-0102, and cpd23 are well-characterized small molecule WDR5 inhibitors for H3K4 trimethylation.27 28 Analysis of MLL1-WDR5 histone methyltransferase activity in the presence of WDR-47, WDR5-0102, and cpd23 determined that all these three compounds inhibit MLL1-WDR5 enzymatic activity in a dose-dependent manner (figure 6B). We then cultured pancreatic tumor cells in the presence of these inhibitors and analyzed OPN protein level. All these three WDR5-selective inhibitors decrease OPN protein level in pancreatic tumor cells in vitro (figure 6C). WDR5-47 and WDR5-0102 exhibit minimal cytotoxicity to pancreatic tumor cells, whereas cpd23 showed significant cytotoxicity to pancreatic tumor cells (figure 6D). We next sought to determine whether these WDR5-selective inhibitors inhibit OPN expression in MDSCs. J774M is a CD11b⁺Gr1⁺ MDSC-like cell line established from the parent J774 monocytic leukemia cell lines. J774M cells have potent inhibitory activity against T cell activation in vitro and thus mimic MDSCs in vitro.26 We treated J774M cells with these WDR5-selective inhibitors and analyzed OPN protein level. All three inhibitors decrease OPN protein level in J774M cells in a dose-dependent manner (figure 6F). Analysis of cell viability determined that WDR5-47 has no cytotoxicity to J774M, and WDR5-0102 and cpd23 exhibit a dose-dependent cytotoxicity to J774M cells (figure 6E).

We then reasoned that targeting pancreatic tumor with a WDR5 inhibitor should be effective in suppression of pancreatic tumor growth in vivo. To test this hypothesis, we made use of the orthotopic pancreatic tumor mouse models. Both UN-KC-6141 and PANC02-H7 tumor-infiltrating T cells express FasL and granzyme B (online supplemental figure 4A-C), two essential effectors of the host T cells.32 Analysis of tumor cells revealed that PANC02-H7 express the death receptor Fas while UN-KC-6141 tumor cells lack Fas expression on its surface in vivo (online supplemental figure 4D). We therefore made use of PANC02-H7 tumor model in this proof-of-principle study. WDR5-47 was used as the WDR5 inhibitor due to its minimal direct cytotoxicity and low IC₅₀. Tumor-bearing mice were treated with WDR5-47 and anti-PD-1 either alone or in combination. OPN is secreted protein. Consistent with the in vitro observation that WDR5-47 treatment decreases OPN protein level in tumor cells and MDSCs (figure 6C,F). Treatment of tumor-bearing mice with WDR5-47 significantly decreased OPN protein level in the peripheral blood (figure 7A).

Consistent with the decreased OPN protein level by WDR5-47 treatment in tumor-bearing mice, a statistically significant interaction between WDR5-47 and anti-PD-1 was detected for tumor weight (F(1,28)=6.97, p=0.0134) (online supplemental table 6, figure 7B). The control group had the highest tumor weight, followed by WDR5-47, anti-PD-1 alone, and then WDR5−47+ anti-PD-1 with the lowest tumor weight (online supplemental table 6). The Tukey-Kramer multiple comparison test on the interaction effect indicated that tumor weights were significantly lower for the WDR5−47+ anti-PD1 when compared with the control (p<0.0001) and anti-PD-1 (p=0.0010). The control group had significantly higher tumor weights than all other groups (p<0.0001 for each comparison to anti-PD-1, WDR5-047, and WDR5+ anti-PD-1). There were no statistically significant differences between anti-PD-1 and WDR-47 (p=0.3161) nor WDR-47 compared with WDR5−47+anti-PD-1 (p=0.0756). A statistically significant interaction between WDR5-47 and anti-PD-1 was also detected for tumor size (F(1,28)=14.29, p=0.0008) (online supplemental table 6, figure 7B). The control group had the largest tumor size, followed by anti-PD-1, WDR5-47, and then WDR5−47+ anti-PD-1 with the smallest tumor size. The Tukey-Kramer multiple comparison test on the interaction effect indicated that the control group had significantly greater tumor size than all other groups (p<0.0001 for each comparison to anti-PD-1, WDR5-47, and WDR5−47+ anti-PD-1. There were no other statistically significant differences between groups (anti-PD-1 vs WDR5-47: p=0.4640; anti-PD-1 vs WDR5−47+anti-PD-1: p=0.0504; WDR5-47 vs anti-PD-1: p=0.6016). Taken together, these findings indicate that combined WDR5-47 and anti-PD-1 therapy has a significantly higher efficacy in suppression of the orthotopic PANC02-H7 tumor growth than either WDR5-47 or anti-PD-1 therapy alone.

H3K4me3 also regulates PD-L1 expression.14 To determine the direct effect of OPN on pancreatic tumor growth and response to anti-PD-1 immunotherapy, we knocked out Spp1, the gene that encodes OPN (figure 7C), in PANC02-H7 cells. ELISA analysis confirmed that Spp1 is knocked out (figure 7C). PANC02-H7.Scramble and PANC02-H7.Spp1 KO cells were then surgically transplanted to C57BL/6 mice to establish orthotopic tumor. A higher tumor cell dose was used in this model due to the increased immunogenicity of the tumor cells after lentiviral vector transduction. The OPN WT and KO tumor-bearing mice were then treated with anti-PD-1 mAb. A statistically significant interaction between OPN status and anti-PD-1 was detected for tumor weight (F(1,21)=5.28, p=0.0319) (online supplemental table 7, figure 7D). OPN WT (Scramble) had the highest tumor weight, followed by WT +anti-PD1, Spp1 KO, and then Spp1 KO +anti-PD1 (the lowest tumor weight). The Tukey-Kramer multiple comparison test on the interaction effect indicated that tumor weights were significantly higher for the WT tumor when compared with all other groups (Spp1 KO: p<0.0001, Spp1 KO+ anti-PD1: p=0.0010, WT +anti-PD1: p=0.0041). There were no statistically significant differences between Spp1 KO and Spp1 KO+ anti-PD1 (p=0.6576), Spp1 KO and WT +anti-PD1 (p=0.7371), or Spp1 KO+ anti-PD1 and WT+ anti-PD1 (p=0.2115).

Unlike the patterns in tumor weight, there was no statistically significant interaction between OPN status and anti-PD-1 in terms of tumor size (online supplemental table 7, figure 7D). However, the Tukey-Kramer multiple comparison test on the interaction effect indicated that WT tumor had significantly higher tumor size when compared with all other groups (Spp1 KO: p<0.0001, Spp1 KO+ anti-PD1: p=<0.0001, WT+ anti-PD1: p=0.0031). Spp1 KO+ anti-PD-1 had significantly lower tumor size when compared with WT+ anti-PD1 (p=0.0017). There were no statistically significant differences in tumor size for Spp1 KO and Spp1 KO+ anti-PD1 (p=0.1092) or Spp1 KO and WT +anti-PD1 (p=0.1478). Taken together, these findings indicate that knocking out OPN significantly reduced both tumor size and weight and there is a statistically significant interaction between OPN status and anti-PD-1 efficacy in terms of tumor weight in tumor-bearing mice.