Multiple Highly Methylated CpG Sites as Potential Epigenetic Markers for the Diagnosis of Prostate Cancer

1. Introduction

Prostate cancer (PCa) ranks among the five most common malignancies worldwide [1]. The incidence of PCa is increasing sharply, due to the aging of the population and the improvement of diagnostic methods such as Magnetic Resonance Imagery (MRI). The measurement in serum of the Prostate-Specific Antigen (PSA), a protein produced almost exclusively by prostate epithelial cells, is widely used for PCa screening [2,3]. Because of its lack of specificity, its measurement alone is disputed, as it can lead to over-diagnosis and unnecessary treatment. Indeed, serum PSA levels can be increased in cases of inflammation of the prostate (prostatitis), benign prostatic hypertrophy or urinary tract infection, when there is no cancer [4]. False positive results can expose patients to invasive and costly examinations (MRI and prostate biopsies), increasing their anxiety levels and exposing them to complications of treatments (surgery, radiotherapy), including urinary incontinence and impaired erectile function [5]. To date, no health agency in the world has given a favourable opinion on the introduction of a systematic, organised prostate cancer screening programme based on serum PSA dosage. In this context, replacing this dosage with a new highly sensitive and specific blood test remains a real medical need. Aberrant DNA methylation is frequently observed in cancers [6,7], including prostate cancer [8,9,10].

By employing the EM-seq method, we were able to identify the eight most relevant differentially methylated regions (DMRs) in prostate cancer tissues. In total, 66 methylated CpG sites were selected in the following genes: Claudin-5 (CLDN5), Glutathione S-transferase Pi 1 (GSTP1), Neurobeachin like 2 (NBEAL2), Prickle planar cell polarity protein 2 (PRICKLE2), Spalt-like transcription factor 3 (SALL3), Traffick-ing regulator and scaffold protein tamalin (TAMALIN), Tight junction protein 2/zonula occluden-2 (TJP2) and Transmembrane protein 106A (TMEM106A). The CpG sites identified make it possible to distinguish prostate cancer tissue from non-tumour adjacent tissue with great precision.

2. Materials and Methods

2.1. Tissue Sample Collection and Patient Information

The 15 patients selected for this study were recruited and nested in the PROGENE study (FWA00006032). They all provided written informed consent to participate in this study that complied with the Declaration of Helsinki and was approved by the CCP Ile de France IV (IRB: 00003835). Archival tissues were provided from fresh frozen tissues of radical prostatectomies. All specimens (n= 30) were reviewed by an expert urological pathologist, who selected, for molecular analyses, a first area in the tumour (n=15) and a second area in the non-tumour adjacent tissue (n= 15). Both areas were macro-dissected, and histologically verified to confirm the presence of more than 70% of tumour cells in the tumour area and, the absence of tumour cells in the non-tumour adjacent tissue. Table 1 shows the clinical and pathological information of patients.

2.2. DNA Isolation

DNA was extracted using the QIAamp DNA mini kit (Qiagen) in accordance to the manufacturer’s protocol. DNA was quantified by fluorimetry using the Qubit^TM dsDNA HS assay (Invitrogen).

2.3. DNA Methylation Sequencing Analysis

200 ng of genomic DNA were used for preparing librairies following manufacturer’s recommendations with the Enzymatic Methyl-seq Kit (New England Biolabs). Librairies were sequenced on Novaseq 6000 ILLUMINA with S4-200 cycles cartridge (2x10000 Millions of 100 base reads), corresponding to 2x333 Millions of reads per sample after demultiplexing. Fastq files were processed on hg19 assembly using nf-core/methyleq v2.3.0 with methyldackel to generate methylation calls as bedgraph. Differentially methylated loci and regions were extracted using DSS v2.46.0 and annotated with ChIPpeakAnno v3.32.0. Identification of DMR and selection of highly methylated CpG sites were performed from the exported statistic tables. Code to generate custom tables and figures is available at https://github.com/guillaumecharbonnier/mw-oncodiag

2.4. Quantification of Methylation Levels by MM-SPCR

Twenty-five nanograms of DNA were bisulfite converted using EZ DNA Methylation Kit (Zymo Research). Universal methylated human DNA standard (Zymo Research) was used as positive control. Bisulfite-converted DNA was eluted in 20 µl of elution buffer. One M-SPCR to amplify GSTP1 (singleplex) and four MM-SPCR to co-amplify Vic-ALB and 6Fam-TMEM106A (duplex1), 6Fam-CLDN5 and Vic-TAMALIN (duplex2), 6Fam-SALL3 and Ned-TJP2 (duplex3) and 6Fam-NBEAL2 and Ned-PRICKLE2 (duplex4), respectively, were developed. To normalize the amount of DNA loaded per well, Vic-ALBUMIN (ALB) containing no CpG sites was used. Four µl of bisulfite-converted DNA (≈ 5 ng) were used as template for each duplex for a total volume of 20 µl per reaction. Each reaction contained 1x QuantiNova Multiplex Kit (Qiagen), ROX reference dye diluted 1:20 (Qiagen), 400 nM primers (Eurogentec) and 250 nM TaqMan-MGB probes (Life Technologies). The sequences of the primers and TaqMan MGB probes are shown in Table 2. MM-SPCR was performed using the QuantStudio 5 Real-Time PCR system (Life Technologies). PCR cycling parameters were as follows: initial denaturation of 3 min at 95°C, followed by 40 cycles consisting of 5 sec at 95°C (denaturation) and 30 sec at 60°C (annealing/extension and data collection). The percentage of methylated reference (PMR) was determined by the “2^-∆∆Ct” method were ΔΔCt = (Ct marker - Ct ALB)Control - (Ct marker - Ct ALB)Sample. The diagnostic accuracy of each marker was assessed by Receiver Operating Characteristic (ROC) curves using CombiROC software (htpp://CombiRoc.eu).

2.5. Array RNA Expression Validation

Normalized expressions of genes identified by the methylation study were analyzed with 15 genes referenced in the literature as hypermethylated in PCa. Analyse was performed on 170 samples according to their prostate tissue origin (68 non-tumoral samples, 47 tumor ISUP-1, 35 tumors ISUP-2 or 3 and 20 tumours ISUP 4 or 5) obtained from radical prostatectomy specimens. The transcriptomic dataset using Affymetrix HG U133 arrays was obtain thanks the PROGENE cohort previously characterize (E-MTAB-6128) [16]. Comparisons were performed using the XLSTAT statistical and data analysis solution 2024 (Lumivero, https://www.xlstat.com/fr) with either the Mann-Whitney test or Kruskal-Wallis test depending on the number of groups compared. Significance was based on Bonferroni corrected p-value <0.05.

2.6. Gene Expression Validation Using Spatial Transcriptomics

Spatial transcriptomics data were obtained from our previously published dataset obtained from radical prostatectomy tissue of a patient with multifocal prostate cancer [11]. The analysis included eight distinct tissue sections, encompassing a total of 32,156 spatial transcriptomics spots. To ensure data quality, spots with fewer than 500 Unique Molecular Identifier (UMI) counts were excluded. Raw FASTQ files were processed using 10X Genomics® Visium Spaceranger software. Libraries were normalised using spaceranger aggr (v2.0) function to correct for batch effects, ensuring comparable gene expression data across samples and reducing variance. Gene expression data were extracted from the barcode ID using R (v4.4.0), and visualization was performed with LoupeBrowser (v8.0.0). Pathological annotation followed a consensus approach, with two independent pathologists reviewing each spatial transcriptomics spot. Violin plots and statistical analyses were generated using GraphPad Prism (v10.4.0).

3. Results

3.1. Analysis of Cytosine Methylation Using EM-seq

Methylation analysis was performed from 15 paired tumors (T) and non-tumour adjacent (NT) prostate tissues by Enzymatic Methyl-seq Kit (EM-seq). Different patterns of global methylation were observed between normal and tumor samples, where normal samples exhibited more completely hypomethylated loci (paired t-test p-value = 5.6e-5), while tumor samples had more loci with average methylation (paired t-test p-value = 7.3e-4, Figure 1A-1B). We noticed a mean methylation slightly up in normal samples relative to tumoral ones in CpG context, and the opposite trend in CHG and CHH contexts (Figure S1). The main component of the variance in the methylation landscape was strongly associated with the normal/tumoral variable (Figure 1C). This analysis led to the identification of more than 4-million loci that were statistically differentially methylated (p-value < 0.001) between T and NT samples (Figure 1D).

The methylome analysis identified more than 350,000 DMRs. Among the most differentially hypermethylated regions in tumor (mean methylation difference tumor vs. normal > 0.25 for regions with more than 20 sites), we selected eight that are associated with the following genes: CLDN5, GSTP1, NBEAL2, PRICKLE2, SALL3, TAMALIN, TJP2 and TMEM106A (Figure 2A). MM-SPCR has been developed for the accurate and cost-effective detection and quantification of the eight candidate genes. In order to ensure reliable and robust amplification, the following criteria have been established: a gene fragment of around 100 nucleotides in size, a minimum of seven hypermethylated CpGs (with a minimum of one in each primer and probe), and an annealing and elongation temperature of 60°C. Heat map analysis was performed for each of the eight genes in accordance with the aforementioned criteria, with a comparison between T and NT (Figure 2B). Mean methylation levels of selected CpG sites in T and NT are provided in Table S1.

3.3. Evaluation of Diagnostic Performance with MM-SPCR

PCR fragments were designed allowing methylation quantification of the identified hypermethylated CpG sites by MM-SPCR and determining the Percentage of Methylated Reference (PMR) of each gene in the different tissues (Figure 3A). The mean cumulative PMR for each gene was significantly different between the T and NT groups. PMR values were as follows with 50.4 ± 34.1% vs. 1.2 ± 2.0% for CLDN5 (p = 3.4e-05), 103.4 ± 34.1% vs. 5.0 ± 6.1 for GSTP1 (p = 2.3e-08), 135.6 ± 68.7% vs. 9.6 ± 5.7% for NBEAL2 (p = 1.4e-05), 66.1 ± 28.2% vs. 0.4 ± 1.5% for PRICKLE2 (p = 2.8e-07), 48.2 ± 26. 8% vs. 4.3 ± 2.3% for SALL3 (p = 1.0e-05), 104.2 ± 37.2% vs. 3.7 ± 4.5% for TAMALIN (p = 2.3e-08), 100.7 ± 39.6 vs. 8.3 ± 9.9% for TJP2 (p = 9.3e-08) and 81.0 ± 27.1% vs. 7.0 ± 7.3% for TMEM106A genes (p = 5.4e-09) (Figure 3B). Considering the eight genes altogether, the mean cumulative PMR values were significantly different between T and NT tissues (86.2 ± 30.3% vs. 5.0 ± 3.2%, p < 0.0002) (Figure 3C). Using, CombiROC software, the diagnostic performance for each gene was evaluated using the area under the ROC curve (AUC). With AUCs ranging from 0.987 to 1.0, each gene has a very high diagnostic accuracy to discriminate non tumoural from tumour prostate tissues (data not shown).

3.4. Tissue RNA Expression of Candidate Genes Using Affimetryx Arrays Data Set and In Situ Expression using Spatial Transcriptomics Visium 10X

3.4.1. Tissue RNA Expression of Candidate Genes Using Affimetryx Arrays Data Set

A Volcano plot (Figure 4) was used to visualize the differential expression between non tumoural and matched tumour tissues for the eight candidate genes and compared it to those of other genes previously reported to be methylated in prostate cancer [12,13,14], and to the mean differential expression levels of genes associated with two prostate cancer aggressiveness signatures (proliferation [15] and S1 signature [16]). Significance of differential expression is reported in Supplement Table S2. For the eight candidate genes, a threshold of significance was obtained for CLDN5 (p-value =0.001), GSTP1 (p-value <0.0001), PRICKLE2 (p-value <0.0001), SALL3 (p-value <0.0001, TAMALIN (p-value <0.0001), and TMEM106A (p-value <0.0001), but not for NBEAL2 and TJP2.

Expression of genes (CLDN5; NBEAL2; PRICKLE2; SALL3; TJP2; TMEM106A; GRASP/TAMALIN) was compared between non tumoral samples and tumours samples according to their aggressiveness defined by ISUP score (Figure S2a,b,c,d,e). Levels of expression were significantly different between non tumoural and cancer with ISUP-1 for CLDN5 (p-value <0.0001), PRICKLE2 (p-value <0.0001), SALL3 (p-value <0.0001), TMEM106A (p-value <0.0001) and GRASP/TAMALIN (p-value <0.0001). Regarding aggressiveness, only the expression of the gene SALL3 was significantly different between ISUP-1 vs ISUP 4 or 5 (p-value <0.006).

3.4.2. In Situ RNA Expression Using Spatial Transcriptomics Visium 10X

Spatial transcriptomics was used to analyse, alongside histology, the gene expression of the eight candidate genes in three tumour sections containing non-tumoural glands and different grades of aggressiveness (Figure 5). Statistical results are presented in supplement Table S3 and Figure S3.

4. Discussion

So far, serum PSA levels have constituted the most commonly employed measure for the purpose of prostate cancer screening. The main limitation of the serum PSA test is its lack of specificity in the detection of benign prostate disease and prostate cancer. In this context, the replacement of serum PSA with a new accurate blood test remains an unmet need. Alterations in DNA methylation have been associated to many types of cancer. In order to identify novel epigenetic markers with high accuracy and reliability, we conducted a comparative analysis between prostate cancer tissue and adjacent non-tumour tissue samples using the EM-seq. This method was selected as an alternative to bisulfite sequencing and the high-density DNA methylation array (Infinium Human Methylation EPIC/450K array, Illumina). Compared with the other two methods mentioned above, it enables complete and highly accurate detection of CpG sites with single-nucleotide resolution without damaging the DNA [17,18]. EM-seq has been shown to identify more CpG sites and DMRs than the EPIC/450K array [19].

We identified more than 4-million differentially methylated between tumour tissue and non-tumour tissue samples CpG sites in EM-Seq. We validated 66 CpG of them by MM-SPCR, which are associated with CLDN5, GSTP1, NBEAL2, PRICKLE2, SALL3, TAMALIN/GRASP, TJP2/ZO-2 or TMEM106A genes.

Of the genes identified, GSTP1 is the most frequently one referenced in prostate cancer research articles [20,21]. This gene has been identified on chromosome 11q13, a region known to be affected by alterations in numerous cancers [22], and is subject to hypermethylation in different types of cancer, including prostate, breast, lung and hepatocellular carcinoma [23]. Furthermore, Casadio et al. established a causal relationship between the methylation of the GSTP1 promoter and the downregulation of protein expression in prostate cancer [24]. A low level of expression of the GSPT1 protein may be a contributing factor in the progression of prostate cancer [25]. Our study confirms that methylation of the GSPT1 gene promoter is an accurate epigenetic marker for the detection of prostate cancer from tissue samples. In the same way, other studies, including ours, had identified hypermethylated sequences in various genes at different stages of prostate cancer and have been able to detect some of them in circulating cells [13] or from free DNA in plasma [12]. CLDN5 and TJP2/ZO-2 belong to the family of tight junction proteins (TJP). This family of proteins plays an important role in cell-cell adhesion [26], which is lost during malignant transformation [27,28,29,30]. Down-regulation of CLDN5 expression was previously reported in several malignancies, including breast, colorectal, esophageal, lung and brain cancers [31,32,33,34,35]. In prostate cancer, a lower expression of CLDN5 was associated with a higher Gleason score (>7) and high serum PSA, two bad prognosis factors [36,37]. The tissue profile of RNA expression as a function of disease aggressiveness is closely linked to the GSTP1 profile. TJP2/ZO-2 protein expression is downregulated in pancreatic, breast [38], and bladder cancers [39]. We observed that TJP2/ZO-2 RNA expression in tissue decrease only with aggressive disease (ISUP-4 and 5). NBEAL2 and PRICKLE2 genes are located on the short arm of chromosome 3 (3p), a region which is well known for the presence of tumour suppressor genes [40]. Senchenko and colleagues reported DNA hypermethylation of NBEAL2 and PRICKLE2 genes in cervical and ovarian cancers [41,42]. However, NBEAL2 tissue RNA expression did not show significant expression variation in our study. To our knowledge, our work is the first to report hypermethylation of PRICKLE2 in prostate cancer. The gene SALL3 is mapped to chromosome 18q, a region frequently deleted in prostate cancer [43]. Down-regulation of the SALL3 protein, due to hypermethylation of the gene, was observed in most human cancers, such as bladder, cervical, head and neck, liver, oral, pancreatic, and thymic carcinoma [44,45,46,47,48,49,50,51,52]. Interestingly, we observed that SALL3 expression decreases significantly with aggressiveness of tumoural tissue defined by ISUP score.

Bjerre et al. using a high-density DNA methylation array (Infinium Human Methylation EPIC/450K Array, Illumina) found a correlation between high methylation level of the promoter of the TAMALIN/GRASP gene and the CAPRA score, which predicts biochemical recurrence after radical prostatectomy [53]. Other studies reported that TAMALIN/GRASP is also hypermethylated in colon [54] and liver [55,56] cancer tissues. Similarly, using the Infinium HumanMethylation450 array, Govorun and colleagues published that the promoter of the TMEM106A gene showed hypermethylation in prostate cancer [57]. Other studies reported low levels of the TMEM106A protein in different cancers, including prostate, gastric, liver, lung and kidney cancer [58,59,60,61,62]. The observed down-regulation of the expression of this protein could be attributed to methylation of the promoter of the TMEM106A gene, notably in prostate cancer.

In this study, we found a DNA hypermethylation of CLDN5, PRICKLE2, GSTP1, SALL3, TAMALIN, TJP2 and TMEM106A genes in prostate cancer tissue compared to non-tumour adjacent tissue. These markers have the potential to be employed in the development of new non-invasive diagnostic test.

5. Conclusions

Our whole methylome sequencing analysis has revealed epigenetic markers that can accurately discriminate between normal and prostate cancer tissues. To improve the diagnostic accuracy of prostate cancer, it is crucial to evaluate the clinical value of these epigenetic markers, both individually and in conjunction with PSA test, in liquid biopsy samples.