Alterations in the methylome of the stromal tumour microenvironment signal the presence and severity of prostate cancer

Background Prostate cancer changes the phenotype of cells within the stromal microenvironment, including fibroblasts, which in turn promote tumour progression. Functional changes in prostate cancer-associated fibroblasts (CAFs) coincide with alterations in DNA methylation levels at loci-specific regulatory regions. Yet, it is not clear how these methylation changes compare across CAFs from different patients. Therefore, we examined the consistency and prognostic significance of genome-wide DNA methylation profiles between CAFs from patients with different grades of primary prostate cancer. Results We used Infinium MethylationEPIC BeadChips to evaluate genome-wide DNA methylation profiles from 18 matched CAFs and non-malignant prostate tissue fibroblasts (NPFs) from men with moderate to high grade prostate cancer, as well as five unmatched benign prostate tissue fibroblasts (BPFs) from men with benign prostatic hyperplasia. We identified two sets of differentially methylated regions (DMRs) in patient CAFs. One set of DMRs reproducibly differed between CAFs and fibroblasts from non-malignant tissue (NPFs and BPFs). Indeed, more than 1200 DMRs consistently changed in CAFs from every patient, regardless of tumour grade. The second set of DMRs varied between CAFs according to the severity of the tumour. Notably, hypomethylation of the EDARADD promoter occurred specifically in CAFs from high-grade tumours and correlated with increased transcript abundance and increased EDARADD staining in patient tissue. Across multiple cohorts, tumours with low EDARADD DNA methylation and high EDARADD mRNA expression were consistently associated with adverse clinical features and shorter recurrence free survival. Conclusions We identified a large set of DMRs that are commonly shared across CAFs regardless of tumour grade and outcome, demonstrating highly consistent epigenome changes in the prostate tumour microenvironment. Additionally, we found that CAFs from aggressive prostate cancers have discrete methylation differences compared to CAFs from moderate risk prostate cancer. Together, our data demonstrates that the methylome of the tumour microenvironment reflects both the presence and the severity of the prostate cancer and, therefore, may provide diagnostic and prognostic potential.


Figure S1
Comparison of methylation values on the EPIC platform with WGBS. a Scatter plots of DNA methylation between cross-platform replicates of WGBS and EPIC arrays at CpG sites interrogated by both platforms. r = Pearson correlation coefficient. All samples had a correlation P value <2.2x10 -16 . b-c MDS plots showing clear separation of CAFs and NPFs based on the 1000 most variably methylated CpGs in (b) EPIC array and (c) WGBS data. d Cross-platform validation of EPIC and WGBS methylation data for the 45% of WGBS CAF-DMRs overlapped by probes on the EPIC array. Each dot represents the mean difference in DNA methylation (n=3 pairs) averaged across each DMR. Pearson's r =0.87, P <2. 2x10 -16 . e EPIC and WGBS data for the TBX3 gene for each NPF (blue) and CAF (red). The average difference in DNA methylation in CAFs compared to NPFs is shown in purple. The height of each vertical line represents the percentage of DNA methylation at each CpG site. The location of EPIC probes is shown in grey. The purple boxes highlight large regions of CAF hypermethylation measured on both platforms.

Figure S2
Concordance between in vitro functional assays and DNA methylation profiles. Patient 10 was used as a representative positive control, with co-cultures displaying the expected morphological changes, unlike Patient 18. a Representative images of cocultures between matched CAFs and NPFs from Patients 10 and 18 and RWPE-1 prostate epithelial cells, which are labelled in green. RWPE-1 cells adopt more elongated, spindle-shaped morphology when co-cultured with CAF10 versus NPF10, but this trend is not apparent in CAF18 versus NPF18. Scale bar equals 200 µm. b Quantitative image analysis of RWPE-1 morphology in co-cultures with CAFs and NPFs. There is a significant decrease in the average shape factor and a significant increase in the average cell length of RWPE-1 cells in CAF10 versus NPF10 cocultures, indicative of a more migratory phenotype (**P<0.01, T test, n=8 fields of view per co-culture). There is no significant change in these parameters between cocultures with CAF18 versus NPF18. Schematics in green next to each graph indicate how changes in cell morphology (shape, spread area and length) are measured.

Figure S3
The correlation between methylation and mRNA abundance of CAF-DMRs. Plots show relative gene expression measured using qRT-PCR versus the percentage of DNA methylation measured using EPIC arrays for (a) PITX2 and (b) ESR1. Both genes had three CAF-DMRs covered by probes on EPIC arrays. The co-ordinates of each CAF-DMR are shown and methylation values were averaged across EPIC probes within each CAF-DMR. Gene expression values below the detection limit are not shown. P and rho values are from Spearman correlations.

Figure S4
Controls and epithelial staining for EDARADD immunohistochemistry. a Images of EDARADD and rIgG immunoreactivity in cell pellets of HMC1 cells (EDARADD high) and PC3 cells (EDARADD low). b Representative images of immunoreactivity for rIgG in benign and tumour samples matching those stained for EDARADD in Figure 4c. c Plot of stromal IHC scores for benign and tumour samples stained with EDARADD and rabbit IgG control (rIgG). There was lower immunoreactivity with rIgG, including for the tumour samples with high levels of EDARADD. d Plot of average IHC scores (± SEM) for EDARADD staining in the epithelium of benign (blue) and tumour (red) tissue. There were no significant differences between any of the patient groups (One Way ANOVA with Tukey post-hoc analysis). All scale bars are 50 μm.

Figure S6
EDARADD hypomethylation is associated with age in non-malignant prostate tissue, but does not signify a more widespread DNA methylation aging phenotype in cancer. a-d Scatter plots comparing cg09809672 DNA methylation versus patient age for (a) NPFs, (b) non-malignant tissue from TCGA, (c) CAFs, and (d) tumour tissue from TCGA. Samples from patients with GG≤3 tumours are shown in light blue, while samples from patients with GG≥4 are in orange, RP = Radical Prostatectomy. Rho and P values are from Spearman's correlations. e Plots showing the average (± SEM) calculated DNA methylation age based on the Horvath signature for NPFs versus CAFs and GG≤3 CAFs versus GG≥4 CAFs. There are no significant differences between groups (paired T test for NPF vs CAF, unpaired T test for GG≤3 CAFs versus GG≥4 CAFs). f Scatter plot of cg09809672 DNA methylation in CAFs and NPFs showing no significant correlation with the calculated DNA methylation age from the Horvath signature (Spearman's correlation).

Figure S7
EDARADD methylation, expression and IHC staining is associated with poor relapse free survival. Kaplan Meier plot of relapse free survival for patients in the lowest quartile of EDARADD methylation and top quartile of expression (orange) versus the rest of the cohort (grey). The Kaplan Meier plots for methylation, expression and IHC data are the same, because the same patients were in the lowest quartile of EDARADD methylation and the highest quartile of EDARADD expression or stromal EDARADD staining. For methylation, the Cox HR = 0.96 (0.931-0.998), P=0.040. For expression, the Cox HR = 5.49 (0.29-47.93), P=0.123. For stromal EDARADD staining, the Cox HR = 1.26 (1.021-1.544), P=0.031.