TRAMP C2 cells were a kind gift from Dr Barbara Foster17; they were maintained in Dulbecco's Modified Eagle Medium (DMEM) with 10% fetal bovine serum (FBS) supplemented with 1 nM Dihydrotestosterone (DHT), 0.008 mg/mL insulin and penstrep. C42 cells were maintained in RPMI1640 with 10% FBS and P/S. C42 cells were validated via microsatellite PCR at the Roswell Genomics core. Both cell lines were mycoplasma negative.

Short hairpin RNA (shRNA) knockdown of WHSC1 in C42 cells was prepared by the Roswell Park Gene Editing shared resource. Briefly, stable knockdown cells were generated using lentiviral-based shRNA (containing shRNA-pGIPZ plasmid or pGIPZ non-silencing control plasmid) produced in HEK293T cells as per manufacturer instructions. These lentiviral-based shRNAs express green fluorescent protein (GFP) and have a puromycin selectable marker. For stable transduction of target cells, media from 70% confluent C42 cells were removed, and virus supernatant plus 4 μg/mL polybrene (total 1 mL for six-well plate) was directly added on top of the cells and the plate was sealed and centrifuged at 1800 rpm for 45 min at room temperature (RT). Next, the plate was incubated for 3–6 hours at 37°C/5% CO₂ and the media were replaced with normal target cell media and incubated again overnight at 37°C/5% CO₂. The next day, cells were split 1:5 and were allowed to attach for 1 hour, and the media were replaced with puromycin selection media every 3–4 days (normal target cell media+puromycin (2 μg/mL). The resistant transduced C42 cells cultured for at least four to five passages in puromycin selection media and GFP expression were observed by microscopy before being using for further experiments.

Transient knockdown experiments in TRAMP C2 cells were made using SiWHSC1 (4390771, ThermoFischer Scientific) or SiCTR (4390843, ThermoFischer Scientific), and those in DU145 cells using siWHSC1 (SR305101) or siCTR (SR30004) from Origene.

For siRNA knockdown, 200×10³ C42 cells or TRAMP C2 cells in 200 µL optiMEM were plated in a 24-well plate overnight and the next day were transfected with or without siWHSC1 (Invitrogen) using lipofectamine RNAimax at 37C, 5% CO₂. After 6 hours, media were gently aspirated and replenished with fresh RPMI1640 or DMEM supplemented with DHT and further incubated at 37°C, 5% CO₂ for 48 hours. Cells were harvested after 48 hours for RNA isolation using TRIzol (Invitrogen, 15596026) for quantitative PCR (qPCR) or for protein isolation by radioimmunoprecipitation assay (RIPA) buffer for western blot. The concentration of RNA was measured by nanospectrometer, and for RT-qPCR, 2 ng/µL of RNA was used for cDNA synthesis using iscript cDNA synthesis kit (Biorad, 170–8891). SYBR Green/Rox qPCR master mix (ThermoFischer Scientific, K0221) was used to analyze the expression of WHSC1, PDL-1, DNMT1 and GAPDH. The primer sequence is for these genes is provided in online supplemental table S2.

Soluble OVA gene, which was amplified from pCI-neo-sOVA (a gift from Maria Castro, Addgene plasmid 25 09818), and monomeric enhanced green fluorescent protein (mEGFP) gene, which was amplified from mEGFP-N1 (a kind gift from Michael Davidson, Addgene plasmid 54767), were genetically fused via P2A translational skipping sequence and were cloned in the Sleeping Beauty transposon plasmid with the human elongation factor 1α promoter.19 This plasmid, pT2-EF-OVA-mEGFP, was electroporated together with the Sleeping Beauty Transposase plasmid, pCMV(CAT)T7-SB100 (a gift from Zsuzsanna Izsvak; Addgene plasmid #34 87920), into TRAMP-C2 by Nucleofector 4D instrument. The electroporated cells were kept in maintenance medium (DMEM supplemented with 10% FBS, 0.005 mg/mL bovine insulin, 1 nM DHT and cell sorted based on mEGFP expression using FACS Aria I cell sorter. The expression of OVA on sorted cells were confirmed by western blotting using rabbit polyclonal OVA antibody (ab186717) at 1:4000 dilution and flow cytometry using PE anti-mouse H-2K^b bound to SIINFEKL (BioLegend) before incubating with or without MCTP-39 for 48 hours. The expression of OVA was analyzed after 48 hours by flow cytometry using BDLSRIIA cytometer, and data were analyzed by FCS Express V.7 Research Edition.

Protein concentration was measured using bicinchoninic acid (BCA) kit and 30 µg of protein was loaded into sodium dodecyl sulphate-polyacrylamide gel electrophoresis (SDS-PAGE) gel from either transfected or untransfected C42 or TRAMP C2 cells. The protein from gel was transferred into PVDF and further incubated with human (anti-WHSC1, Abcam, ab225625) or mouse (anti-WHSC1, Abcam, ab75359) primary antibody to WHSC1 using 1:2000 dilution for C42 lysate and 1:1000 for TRAMP C2 protein lysate. The primary antibody was further detected using either goat anti-rabbit (Abcam, dilution 1:10,000) or goat anti-mouse (Abcam, dilution 1:5000), while the house keeping gene, GAPDH, in both C42 and TRAMP C2 protein lysate was detected using anti-GAPDH antibody at 1:50 000 dilution (Abcam, ab181602).

Knockdown of WHSC1 in C42 and TRAMP C2 cells is discussed in previous sections. Cells were counted at 48, 96 and 144 hours. For pharmacological inhibition, cells either C42 or TRAMP C2 were seeded overnight at 4×10³ cells per 100 µL media in a 96-well plate; the following day, cells were treated with vehicle control or different concentrations of MCTP-39 (0–10 µM) for 48 hours. After 48 hours, C42 or TRAMP C2 cells were either counted or used for staining with MHCI/II antibodies. Briefly, cells were detached using trypsin and washed with fluorescence-activated cell sorting (FACS) buffer for 5 min at 300 g. The pellet was resuspended in FACS buffer and stained with MHC-I/II antibodies for 20 min at 4°C. After incubation, cells were washed with FACS buffer two times and resuspended in 200 µL of FACS buffer before acquiring the data on a flow cytometer.

Male C57B/6J mice (6–8 weeks of age) and NSG mice (10–12 weeks of age) were obtained from the Roswell Park’s Center For Immunotherapy breeding colonies. All in vivo experiments were made following institutional and Institutional Animal Care and Use Committee (IACUC) regulations. In vivo experiments were not blinded. TRAMP C2 cells (1 × 10⁶ cells/100 µL) were injected subcutaneously into the right flank of male C57BL/6J mice (n=6/group) or NSG mice (n=5/group) using a 27 G needle. Sample size was chosen based on pilot studies and chosen to include the potential loss of mice prior treatment. Tumor volume was monitored weekly with an electronic caliper and calculated as V=(W²×L)/2, where V is tumor volume, W is tumor width and L is tumor length. When tumors reached 100 mm³, mice were randomized prior to treatment with either MCTP-39 (10 mg/kg 5×/week/4 weeks) or vehicle control. Mice were euthanized either when tumors reached 2000 mm in any dimension, as per institutional IACUC regulations, when mice showed signs of advanced disease or after 4 weeks of treatment. Tumors were harvested and weighted; single cell suspension was prepared using the tumor dissociation kit (Miltenyi Biotech) as per manufacturing instructions prior to flow cytometry analysis.

CD45-positive and CD45-negative fraction from single cell suspension was stained with the antibodies listed in online supplemental table S3. Gating strategies are described in online supplemental figures S4–S6). For C42 cells and TRAMP C2, cells were trypsinized and washed with FACS buffer two times for 5 min at 300 g. The pellet of C42 or TRAMP C2 cells were resuspended in FACS buffer and C42 cells were surface stained with HLA-B7, HLA-F, HLA-E, and HLA-DQ, while TRAMP C2 cells were stained with a cocktail of I-A/I-E and H-2Kb antibodies for 20 min at 4°C. To detect HLA-DM, C42 cells were first surface stained followed by fixation and permeabilization followed by staining with anti HLA-DM antibody for 30 min at RT. Unstained C42 cells or TRAMP C2 cells were used as negative control. After incubation, cells were washed with FACS buffer two times and resuspended in 200 μL of FACS buffer before acquiring data on BDLSRIIA cytometer. For mice, single cell suspension from tumors were incubated with anti-mouse CD16/32 antibody (FcR blocker) for 10 min at RT and subsequently stained with Zombie aqua for 15 min at RT followed by staining with anti-CD45, CD8a, CD3 and H-2Kb for 20 min at 4°C. After staining, cells were washed and fixed with fixation buffer for 15 min at 4°C followed by washing two times with FACS buffer. Cells were resuspended in 200 μL of FACS buffer before acquiring data on flow cytometer. For compensation, the Ultracomp beads was used to stain for individual fluorochrome-conjugated antibodies as single-color compensation controls, and each drop of bead contained a positive population that captures fluorochrome-conjugated antibody and a negative population that does not react with antibody. This bimodal distribution was used a single-color compensation control in multiparameter analyses. For the live/dead cell compensation, live cells were divided into two aliquots and saturating volume of methanol was added to one of the aliquots, vortexed and incubated at 4°C for 5 min. After 5 min, cells were washed three times with FACS buffer and mixed with original aliquot of live cells to get the heterogenous population of live/dead cells that was used for compensation in flow cytometry. Data were analyzed using FCS express V.7 Research Edition.

Publicly available PCa datasets were retrieved from cBioportal (https://www.cbioportal.org/). Detailed statistical methods are explained further in the appropriate section. GSEA analysis on TCGA data was done using the package DOSE,21 clusterprofiler22 and enrichplot (https://github.com/GuangchuangYu/enrichplot) using Reactome gene signature downloaded from the GSEA website (https://www.gsea-msigdb.org/gsea/index.jsp). To create the antigen processing and presentation signature, the following gene sets were used: REACTOME ANTIGEN PROCESSING CROSS PRESENTATION, REACTOME ANTIGEN PROCESSING UBIQUITINATION PROTEASOME DEGRADATION, REACTOME MHC CLASS II ANTIGEN PRESENTATION, REACTOME ANTIGEN PRESENTATION FOLDING ASSEMBLY AND PEPTIDE LOADING OF CLASS I MHC, REACTOME CLASS I MHC MEDIATED ANTIGEN PROCESSING PRESENTATION. ChIPSeq data were downloaded from Cistrome (http://cistrome.org/) with the following accession numbers: GSM1527830, GSM1679107, GSM353624, GSM353627, GSM875813, GSM875814, GSM1679108, GSM2252903, ENCSR849APH_1, and ENCSR849APH_2. Predicted target genes from Cistrome that scored 0 were removed, leaving a total of 13 878 unique genes (online supplemental file S2). Differential gene expression analysis from TCGA data comparing patients with highversus low WHSC1 transcript levels was performed using the limma R package at a significance level of FDR<0.05 and LogFC 0.25.23 Precomputed TCGA immune infiltration data were downloaded from xCell.24

RNA extraction and library preparation were performed by the Roswell Park Genomics Shared Resources. RNA libraries were constructed using the KAPA mRNA HyperPrep Kit (Roche Sequencing Solutions), and the libraries were sequenced on the Illumina NextSeq 500 sequencer with 2×75 cycle sequencing. Raw reads were compiled into fastq files, mapped onto the human hg38 reference genome using STAR25 and quantified at the gene level using the tximport R package.26 Genes differentially expressed between conditions were identified using limma.23 GSEA analysis was performed as described previously. APM signature was generated using HLA genes and genes involved in the antigen processing and presentation.

ATACSeq was run on C42 cells with stable knockdown of WHSC1 via shRNA and control. The ATACSeq libraries were sequenced using NextSeq 500 sequencer at 2×75 cycle sequencing. Raw data were processed with MACS227 28 and further processed using ChIPSeeker29 to annotate identify the genes within the genomic regions within ATAC peaks. To calculate the fold changes between the two conditions, we created a consensus list of genomic regions covered by both conditions (shCTR and shWHSC1) using the soGGi R Bioconductor package.30 Reads spanning over these regions were then quantified using Rsubread,31 summarized at the gene level using peaks that are within 1000 bp to the nearest TSS and used to calculate the log2FC between WHSC1 knockdown and control. Results were merged with the RNASeq DEGs to identify genes that positively correlate with RNASeq data, hence higher ATAC signal, higher gene expression. The ATACSeq log2FC was then used to rank genes used as input for GSEA analysis using the GO signature. GSEA analysis was run in R using the clusterprofiler package and a p value cut-off of 0.05.

Methylation analysis was run on C42 cells with stable knockdown of WHSC1 via shRNA and control using the Illumina Infinium MethylationEPIC BeadChip Kit (Illumina). Raw files were processed using the Champ Bioconductor R package32 using default parameters. Methylation probes that coincided with known SNPs were removed. Probe IDs from the differentially methylated probe (DMP) list were merged with RNASeq DEG data, aggregated using mean intensity values at the gene level and correlated with RNASeq log2FC results to identify genes with reduced methylation and increased gene expression, or vice versa, on WHSC1 knockdown.

Survival analysis was performed dividing patients based on the upper/lower 25% of the expression levels of WHSC1; significance, HR and CIs were calculated using Cox proportional hazard model available in the R package survival.33 The AUC analysis to evaluate WHSC1 as predictor for biochemical recurrence was done using the R package survivalROC,34 censoring for biochemical recurrence. Since TCGA does not offer the history of PSA testing per each patient over time, we used a threshold of PSA>0.4 ng/mL as described in Brockman et al.35 Significance for the AUC analysis was calculated by simulating 10 000 AUCs using randomly selected genes in the RNASeq dataset. The empirical p value was calculated by dividing the number of expected/simulated AUCs higher than our observed value by 10 000 (number of simulations).36 Significance when comparing two groups was calculated via two-tailed Student’s t-test at a significance threshold of 0.05. When more than two groups were compared, one-way analysis of variance with Tukey’s post hoc correction was used at a significance threshold of 0.05. In both cases, bar plots indicated the mean and SE of at least three biological replicates unless specified otherwise. The correlation between WHSC1 and AR gene expression was calculated using Spearman correlation using the log2 expression values from TCGA RNASeq data. Growth curves in mice were compared using permutation test with 10 000 simulations via the statmod R package using the compareGrowthCurves function.37 38

We first evaluated WHSC1 expression in 489 PCa samples from The Cancer Genome Atlas (TCGA) collected following radical prostatectomy39 (online supplemental table S1) and its association with patients disease-free survival (DFS). This analysis revealed that patients with elevated WHSC1 expression (top 25%, online supplemental file S3) have significantly shorter DFS than patients with low WHSC1 levels (bottom 25%) (figure 1A and online supplemental figure S1A) (Cox p=0.000121, Cox HR=3.2221, 95% CI 1.801 to 6.127). Raising prostate-specific antigen (PSA) concentration post-therapy in PCa are often indicative of disease recurrence,35 and we found that WHSC1 gene expression levels are a modest, but significant, predictor for biochemical recurrence with an area under the curve (AUC) of 0.742 (online supplemental figure S1B,C; empirical p=0.0046). Moreover, PCa is an androgen-driven disease, and we found a positive correlation between WHSC1 and AR expression (online supplemental figure S1D) (Spearman’s coefficient=0.397, p<2.2e-16), corroborating previously published data on WHSC1 function as a transcriptional regulator of AR.13 We then investigated whether WHSC1 expression levels correlate with the presence of specific immune populations and found that patients with elevated WHSC1 have a highly immunosuppressive tumor microenvironment composed of high Th2 and Treg and low Th1 cells, NKT cells and M1 macrophages (figure 1B) (p value<0.001, Student’s t-test). Next, we interrogated TCGA RNA Sequencing (RNASeq) data to evaluate whether different WHSC1 levels correlate with altered HLA expression in the tumor, which would undermine the tumor’s ability to present tumor antigens to the immune system. Indeed, patients with elevated WHSC1 were found with consistently low levels of HLA class I and class II transcripts (figure 1C). We then investigated which transcriptional pathways were altered in patients with high versus low expression of WHSC1. Gene set enrichment analysis (GSEA) revealed that patients with elevated WHSC1 have increased expression of genes involved with cell proliferation pathways such as cell division, DNA replication and DNA repair (figure 1D), corroborating the results from previous studies.9 10 13 However, we also identified a downregulation of pathways involved with response to interferon gamma (IFN-γ), antigen processing and presentation and cytokine secretion (figure 1E), suggesting that high levels of WHSC1 negatively correlate with the status of both tumor-resident immune pathways and the antigen processing and presentation machinery (APM), which we previously demonstrated as a determinant of T cells’ ability to recognize neoantigens.40

We sought to pinpoint potential mechanistic events that allow WHSC1 to regulate the expression of APM genes. To this end, we downloaded publicly available ChIPSeq data for PCa cell lines from the Cistrome database41 42 and generated a consensus list of genes residing in H3K36me2 and H3K36me3 loci as proxy for WHSC1 targets, exploiting the fact that WHSC1 can deposit both H3K36me2 and H3K36me3 marks.7 We ran differential gene expression analysis for this signature in the TCGA PCa dataset comparing patients with high versus low WHSC1 levels. A total of 5788 genes were differentially expressed, including ubiquitinases, proteases and HLAs, involved with protein and antigen processing, DNA methylatransferases and immune genes (figure 2). Notably, patients with low WHSC1 tend to present with high expression of HLA-C and HLA-F, low CD276/B7H3 and CD274/PDL1, and low DNMT1, DNMT3A and DNMT3B levels (figure 2). To validate our computational observations, we stably knocked down WHSC1 in C42 cells and noted a significant reduction in cell proliferation (online supplemental figure S2A), also observed on transient knockdown of WHSC1 in DU145 cells (online supplemental figure S2B). On WHSC1 knockdown, 3783 genes were differentially expressed (1834 down and 1949 up, false discovery rate (FDR)<0.05) (figure 3A). We first confirmed the downregulation of AR and its downstream target KLK2, indicating a transcriptional suppression of the androgen signaling (figure 3B). GSEA analysis revealed a downregulation of TGF-β signaling and upregulation of IFN-γ and tumor necrosis factor signaling (figure 3C). The APM signature, curated from the Reactome database, was also upregulated on WHSC1 knockdown (figure 3D). At the gene level, we noticed an increase in HLAs, proteinases and ubiquitinase genes, parallel to a downregulation of DNMT1 and CD276 (figure 3E), consistent with the initial computational observations. A number of differentially expressed genes (DEGs) were common with those identified integrating ChIPSeq and TCGA data, including UBL5, UBXN1, UBA52, PSME1, UBE2N, PSMC5, UBL7, UBE2E3, UBR7, UBR3 and UBE4A, suggesting that protein processing and degradation could be directly regulated by WHSC1 via methylation of the H3K36 histone mark. The degree of dissimilarity at the gene level between the predicted H3K36me2-H3K36me3 Cistrome and the DEGs following WHSC1 knockdown is likely caused by the different biological sources from which data were originated (Cistrome/cell lines vs TCGA/patient biopsies) and by the fact that multiple enzymes can methylate H3K36.43–48 These results support a mechanistic role for WHSC1 in regulating the expression of APM genes in PCa.

Our computational and transcriptional analyses indicate that one of the mechanisms by which elevated WHSC1 regulates target gene expression is by upregulating DNMT1, corroborating previous studies indicating a link between H3K26me2 and DNA methylation.49 To evaluate the link between DNA methylation and the expression of genes in immune pathways or APM, we performed methylomic analyses in C42 cells following WHSC1 knockdown (figure 4A,B).

A total of 2209 DEGs contained differentially methylated probes, and in 651 genes, changes in transcript levels negatively correlated with changes in DNA methylation on WHSC1 knockdown (figure 4C,D). In accordance with the data presented earlier, the results includes six genes involved in peptide proteosomal degradation (UBE2E1, UBE2E6, UBE2L6, UBE4A, RNF135 and PSMD8); RUNX1, which is involved in promoting class I Major Histocompatibility complex (MHC) expression50; SMAD7, which negatively regulates the immunosuppressive TGF-β signaling (reviewed in Stolfi et al51); a number of membrane trafficking proteins (SRL, RIMS1, STXBP6, and NAPB); interleukin (IL)-6R, increased during PCa development52 and in breast cancer53; INHBB, which belongs to the TGF-β family and is increased in PCa54; IL-5, which was shown to promote cancer metastases modulating the TME55 56; PARP10, which limits cell proliferation and metastases57; and CD276, which is elevated in several cancers and is a potential immunotherapeutic target.58–60 Lastly, we noticed that probes associated with the DNMT1 gene had increased methylation with a parallel reduction in DNMT1 gene expression (figure 4D). These results suggest that WHSC1 can regulate tumor-resident immune pathways and APM by altering the DNA methylation landscape in PCa.

To further narrow the mechanistic role of WHSC1 in modulating APM genes, we performed ATACSeq analysis of C42 cells following knockdown of WHSC1. We first created a consensus peak list, compared the peak intensity between the two conditions and noticed that the biggest differences were, as expected, in the peaks for mononucleosomal or dinucleosomal regions with no differences in larger peaks (figure 5A,B). After annotating genes to the peaks, we kept those with open chromatin and increased gene expression, or vice versa, as indicated by a positive correlation between the log fold changes in ATACSeq and RNASeq (figure 5C), and performed GSEA analysis. Within the results (online supplemental file S1), we identified upregulation of genes in immune signaling and protein ubiquitination pathways (figure 5D,E). Next, we evaluated the presence of peaks at the gene level for the core enrichment of the upregulated pathways and demonstrated that increased peak magnitude at the transcriptional start site (TSS) is associated with increased gene expression (figure 5F). Interestingly, while the results between the H3K36 Cistrome, ATACSeq and DNA methylation have a modest overlap, genes consistently belong to immune and antigen processing pathways. This suggests that WHSC1 plays complementary direct and indirect roles in regulating cellular behavior by modifying chromatin accessibility and DNA methylation in cancer.

Following the results from computational and in vitro experiments, we sought to evaluate in vivo whether the pattern of expression of WHSC1 and companion markers was conserved. Prostate samples were isolated from TRAMP mice61 with tumors at 20/25 weeks of age, with aggressive palpable tumors (~35 weeks) and healthy wild type (WT) mice. The protein levels of WHSC1, DNMT1, H3K36me2 and CD274/PD-L1 were evaluated via western immunoblotting. CD274 was included since it is elevated in patients from TCGA with high WHSC1 levels (figure 2) and is a downstream target of IFN-γ,62 63 which is elevated on WHSC1 knockdown (figure 3C). Furthermore, CD274 has potential translational relevance as target for checkpoint inhibitor therapy. Results show that healthy WT prostates have no detectable expression of any of the aforementioned proteins while higher levels of DNMT1, WHSC1, CD274/PD-L1 and H3K36me2 are detected in the tumor samples via western blot (figure 6A). We then tested whether the effect of WHSC1 knockdown in human C42 cells was reproducible in the murine TRAMP C2 cells,17 as proxy for potential in vivo effects. Knockdown of Whsc1 via siRNA led to a significant reduction in cell proliferation (online supplemental figure S3A,B) and increased transcript levels of DNMT1 and CD274 (figure 6B).

Since we detected increased MHC transcript levels following WHSC1 knockdown, we investigated whether this observation can be replicated at the protein level on pharmacological inhibition of WHSC1 (with MCTP-399) in both human (C42) and murine (TRAMP C2) cell lines. The combined treatment with MCTP-39 and IFN-γ had an additive effect on upregulating HLA-B7, HLA-F, HLA-E, HLA-DQ, HLA-DM (figure 7A) and murine H2Kb I-A/I-E (figure 7B), as measured via flow cytometry analysis. We then tested whether antigen-bound MHC was also elevated on WHSC1 inhibition using ovalbumin (OVA)-overexpressing TRAMP C2 cells, and demonstrated that treatement with MCTP39 increased the OVA-bound H2Kb fraction (figure 7C,D). These results suggest that pharmacolocycal inhibition of WHSC1 increases PCa cells’ ability to present processed antigens on the cell’s surface via upregulation of MHC molecules.

Hypothesizing that higher tumor antigen presentation would affect tumor growth and the levels of infiltrating immune cells in the tumor, we grafted C57B/6 mice with TRAMP C2 cells, administered MCTP39 (10 mg/kg) for 4 weeks intraperitoneally and evaluated tumor size weekly and T-cell infiltration at endpoint. Tumor growth was significantly reduced on treatment with MCTP39 (figure 8A) (p=0.0023), while this inhibitory effect was not observed when we grafted the same number of TRAMP C2 cells in immunocompromised NSG mice (figure 8B). Moreover, the reduction in tumor growth in C57B/6 mice was accompanied by increased H2Kb expression, increased CD8⁺ T-cell infiltration and reduced tumor weight (figure 8C). These results indicate that that the antitumor effect observed following Whsc1 inhibition requires the presence of a functional immune system for optimal tumor control.