Participant recruitment protocols have been detailed elsewhere (Wang et al., 2007; Hu et al., 2018). Information and blood samples were collected from a total of 3,932 prospectively recruited PCa cases and 2,397 controls recruited through Mayo Clinic primary care clinics. Of these, 30 patients were identified and selected for this study. Of these 30 patients, 15 had been diagnosed with PCa and 15 without PCa had been diagnosed with recent type 2 diabetes.

The study protocol was reviewed and approved by the Mayo Clinic Institutional Review Board. All eligible individuals provided written informed consent to participate in the study. Information on demographic characteristics, body mass index (BMI), lifestyle, and comorbid conditions were collected using a self-administered questionnaire for both cases and controls.

Blood samples were collected from cases at the time of diagnosis and prior to receiving any treatment for PCa. Blood samples were collected from controls at the time of a routine medical visit. The Biospecimens, Accessioning, and Processing (BAP) core at Mayo Clinic extracted 25 ul of dsDNA at a concentration of 50 ng/ul. (Qiagen) The HumanMethylationEPIC Beadchip (Illumina, CA, United States) was used in this study to quantify methylation of approximately 850,000 methylation sites across the genome using standard protocols. To generate methylation β-values for all analyses, raw methylation data was normalized using negative control probes (Illumina GenomeStudio) and used the MethylationEPIC manifest for processing EPIC data. After standard quality control methods, 156,999 methylation markers remained for analysis.

Select descriptive demographics of the sampled population were compared using Fisher’s exact test and ANOVA F tests. Differential methylation analysis was conducted to identify disease-associated CpG sites and top methylation markers were characterized to identify disease-associated enrichment across the genome. Several logistic regression models were used to look for significant associations between CpGs and disease and all models adjusted for sex and age. Model 1 was between PCa and each CpG, model 2 was model 1 plus blood cell type adjustment, and model 3 was model 2 plus DM adjustment. Models 4–6 were the same as model 1–3 except involved DM as the primary disease instead of PCa. Genome-wide significance level was set at 9 × 10–8 for an EPIC array (Mansell et al., 2019). Exploratory analyses were conducted to propose importance of top CpG associated genes and visualized using boxplots. A methylation-based age predictor was also investigated for ability to distinguish disease associated disease groups from controls. Public databases, GTEx (GTEx Consortium, 2013) and GEO, (Barrett et al., 2013) were used to investigate methylation and gene expression data in other populations with larger sample sizes and as a way to provide biological context or functional importance of the most statistically significant methylation markers. The GTEx web-based visualization tool was used to generate plots while GEO expression data was downloaded and plotted using R. Additionally, R was used to perform and visualize all other analyses.

Selected characteristics of the study participants are described by disease status (Table 1). All characteristics are similar across the 4 disease groups with the only significant difference observed for smoking status. There are significantly more ever smokers in the PCa new onset DM group (86%) and significantly more never smokers in the PCa, no DM group (88%). When comparing these characteristics by PCa and non PCa disease groups in the larger Mayo Clinic Pancreatic Cancer resource, we observe that PCa cases are more likely to be male, have a higher usual BMI, more likely to have ever smoked and more likely to be diabetic (Appendix Table A1).

Overall, no significant CpG sites were associated with PCa or DM when setting genome-wide significance level to p-value < 0.05 × 10^–8. (Figure 1A). Multiple models were explored in an attempt to understand methylation marker associations in the context of different adjustment factors. The top 100 results of each model were evaluated to identify methylation markers which appeared across multiple models. There was little observed overlap in the sets of top 100 markers (Figure 1B) when comparing cancer to diabetes models.

We observed a similar hypermethylation/hypomethylation pattern with respect to genomic region. (Figure 2A).When visually comparing all ∼850,000 CpGs to the top 100 disease associated CpGs, (Figure 2B) Open Sea (i.e., CpG sites not classified as island, shore, or shelf) regions are enriched among PCa and there is slight under-representation among the Shore region (i.e., ∼2 kb from islands)..(Illumina, 2016).

The CpG marker, CpG04969764, observed in DM and PCa models, shows promise for discriminating patients with recent onset DM with and without PCa and for discriminating later stage PCa from those without PCa. (Figure 3). As illustrated by the boxplot, those participants with both PCa and DM had the highest average methylation of CpG04969764 while those with no PCa but DM had the lowest average methylation. The second boxplot shows that the mean CpG04969764 methylation of stage IV PCa is significantly different than the mean methuylation of the no PCa group.

One of the most commonly mentioned and understood biological functions of DNA methylation is related to gene expression. Depending on the gene and the methylation marker, increased methylation (i.e., increased beta value) is frequently either correlated with increased or decreased gene expression. The methylation site is controlling access to that local section of the DNA and therefore influencing the gene expression.

Methylation marker CpG04969764 is located within the LAMA5 gene region so LAMA5 gene expression was evaluated in public databases. Using the GTEx database, we can look at the RNA expression levels reported for LAMA5 across tissue type or specifically in the pancreas or tumor. (Figure 4). When looking at select tissues, we see high expression in adipose tissue, followed by the pancreas, with the lowest in whole blood. Within the pancreas, we see significantly higher expression of LAMA5 in tumor cells compared with adjacent normal pancreas followed by the lowest expression in stroma cells. It appears that a decrease in methylation leads to increase in LAMA5 expression. So there is a lower average methylation of cpg markers associated with LAMA5 and higher average expression of LAMA5 in pancreatic adenocarcinoma (PAAD) compared to normal tissue.

Using GEO public database, we evaluated another study for LAMA5 RNA expression and LAMA5-AS1 long non-coding RNA expression (Figure 5). An inverse relationship between methylation of cpg04969764 and LAMA5 expression is observed in this study. We also see a similar pattern for those with PCa and DM having the lowest expression and those with DM and no PCa having the highest expression. These observed differences are just based on trends with no statistical associations tested.

Each disease group is predicted to have younger methylation-based age estimates compared to their chronological age (Figure 6). We examined associations with age for the 354 CpGs included in the Horvath methylation age predictor (Horvath, 2013). The predicted age based on the calculator showed a strong correlation (r) with chronological age among all groups (no PCa, new onset DM r = 0.81, PCa, new onset DM r = 0.86, PCa, no DM r = 0.90, no PCa, no DM r = 0.93). The median predicted methylation-based age was younger for all disease groups compared to the chronological age and was 64.9 vs. 65.0 among no PCa, new onset DM, 64.7 vs. 70 among no PCa, no DM, 62.5 vs. 64.5 among PCa, new onset DM, and 59.9 vs. 62.5 among PCa, no DM. The heatmap of the 27/450 Horvath CpGs with a significant difference in methylation shows that there are visually discernible differences to the beta values across the marker set by disease status.