We performed this study according to the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments. The study procedure was approved by the Ethics Committee of the First Affiliated Hospital of Harbin Medical University. All participants voluntarily gave written informed consent.

In this study, a two-stage case-control DNA methylation study was designed, including discovery and replication analyses. The sample consisted of 192 IS patients and 192 age-(±3 years) and sex-matched controls from the First Affiliated Hospital of Harbin Medical University. Figure 1 demonstrates the study design and selection criteria for cases and controls. The control group was from healthy people who underwent physical examination or patients hospitalized from the First Affiliated Hospital of Harbin Medical University at the same period.

Data for subjects included questionnaire, laboratory investigations, and clinical characteristics. We collected data using a questionnaire related to demographic information (such as gender, age, and marital status), lifestyle habits (such as tobacco smoking and alcohol drinking), and past medical history (hypertension and type 2 diabetes mellitus) of the participants (Qin et al., 2019).

Laboratory investigations were done at the First Affiliated Hospital of Harbin Medical University. We collected blood samples in tubes with clot activator and gel separator from all participants in the morning after fasting and measured using an auto biochemical analyzer (Biobase BK-600; Shandong, China). The levels of total triglyceride (TG), total cholesterol (TC), high-density lipoprotein (HDL), and low-density lipoprotein (LDL) was collected from all participants.

Clinical characteristics included Trial of Org 10172 in Acute Stroke Treatment (TOAST) type and stroke location. Stratification analysis by TOAST criteria was performed for large-artery atherosclerosis (LAA), cardioembolism (CE), small vessel disease (SVD), other determined cause (ODC), and undetermined (UND) aetiologies. Stroke location was categorized as anterior, posterior, and both anterior and posterior circulation. Anterior circulation was further subdivided as left and right hemisphere based on computerized tomography or magnetic resonance imaging scans.

Two milliliters of blood samples were collected in EDTA tubes from all participants and stored at −80°C until DNA extraction.

Genomic DNA was extracted from the leukocytes in the peripheral blood using an DNA Methylation kit (Zymo, Irvine, CA), following the protocol (Genesky Biotechnologies Inc., Shanghai, China). Genome-wide DNA methylation was assessed using Infinium Human Methylation 850K BeadChip (Illumina Inc., San Diego, CA, United States)). More than 853,000 CpG sites were included in each chip. CpG sites containing documented single-nucleotide polymorphisms and mapping to X and Y chromosomes were removed to avoid potential confounding by single-nucleotide polymorphisms and gender.

To screen functional DNA methylation positions (DMPs), DMPs located in promoter regions (upstream or 5′untranslated region, 5′UTR), with |methylation difference|>0.1 and p < 0.05 were selected. We performed Gene ontology (GO, http://www.geneontology.org) and the Kyoto Encyclopedia of Genes and Genomes (KEGG, http://www.kegg.jp/) pathway enrichment analysis to clarify the function and biological pathways of differential methylation loci. Through GO enrichment analysis, the differentially methylated genes were classified according to cellular component, molecular function and biological process. In GO and KEGG enrichment analysis, p < 0.05 was significantly enriched by differential methylation loci-related genes.

Considering the p-values and biological functions of the genes, we selected 8 candidate genes from GO analytics top-ranked pathway including 10 CpG sites to validate 850 K BeadChips results using MethylTargetTM assay, which was a multi-targeted CpG methylation analysis method based on next-generation sequencing (Wan et al., 2021). As the 10 CpG sites in 850 K BeadChips results are not top-ranked loci and FDR p values of the 10 CpG sites are not statistically significant, we replicated 10 CpGs in a small replication sample (20 cases:20 matched controls, 8 samples from DNA methylation chip analysis and 32 samples from validation analysis). Moreover, we further qualified the methylation level of promoter regions in 8 candidate genes, containing 17 CpG islands and 308 CpG sites, using a large replication sample (188 cases:188 individually matched controls). The locations of the 17 CpG islands in 8 genes are shown in Figure 2. The primers were carefully designed (Additional File S1).

The dataset GSE22255 was acquired from the GEO database (https://www.ncbi.nlm.nih.gov/geo/) and was utilized as the validation sample. Gene expression profiling was performed in peripheral blood of 20 IS patients and 20 sex- and age-matched controls. From this dataset, the downloaded data format was MINIML. Box plots of expression were drawn by the R software package ggplot2, and principal component analysis graphs were drawn by the R software package ggord. The Wilcoxon rank-sum test was used to detect whether the mRNA expression of differentially methylated genes was different between stroke patients and controls.

Quantitative variables were reported as the mean and standard deviation, while qualitative variables were expressed as frequencies and percentages (%). The normality of quantitative variables was assessed using the Shapiro–Wilk test.

First, we compared the differences of baseline demographic characteristics, lifestyle habits, medical history, and lipid levels between IS cases and controls. The differences of the categorical variables were measured with a chi-square test. We performed a paired t test for the normal distributed continuous variables. Non-parametric test was used for the skewed distributed continuous variables.

Second, the methylation status was compared between IS cases and marched controls using conditional logistic regression analysis. Smoking, drinking, medical history of diabetes mellitus and hypertension, and lipid levels were added as adjusted factors. The association of the methylation level of genes was assessed using Pearson’s correlation. All subjects were classified into four groups (Q1, Q2, Q3, and Q4) based on the quartile value of methylation level. Q1 group having the lowest methylation level is regarded as the reference group, and all other groups (Q2 having 25–50% of the values, Q3 having 50–75%, and Q4 having 75–100%) were compared with Q1 group (Qin et al., 2019).

The methylation level of each CpG site was measured as the percentage of the methylated cytosines over total tested cytosines. The average methylation level of all CpG sites was worked out as the methylation level of the DNA segments and genes. Results were considered statistically significant when the p values were less than 0.05. All statistical analyses were performed with the SAS version 9.4 for Windows (SAS Institute, Inc., Cary, NC).

The distribution of characteristics for the participants are presented in Table 1. Compared with controls, IS patients in validation stage had a higher prevalence of smoking, drinking and hypertension (p < 0.0001), and higher level of triglyceride (p = 0.0288).

In the discovery analysis, 2656 DMPs exhibited differences (p < 0.05, |methylation difference|>0.1) in DNA methylation between the two groups. The results are shown in another unpublished article (Xu J, et al.). After screening, a total of 462 functional DMPs (located in the promoter region, |methylation difference| > 0.1 and p < 0.05) corresponding to 373 genes were selected (Additional File S2). Among them, the majority of the DMPs were in CpG island (26.4%) and N-shore (28.6%). In addition, 87.7% (405 of 462) CpG cites corresponding to 327 genes were found to be hypomethylated in IS cases compared to controls.

We screened and finally included 12 KEGG pathways (p < 0.05, Figure 3). A total of 315 significant GO terms (p < 0.05) were selected. The top 10 GO items were shown in Table 2. Among them, the “calcium-dependent cell-cell adhesion via plasma membrane cell adhesion molecules” item showed the most significant fold enrichment. There are 8 genes in this item including CDH2/PCDHB10/PCDHB11/PCDHB14/PCDHB16/PCDHB3/PCDHB6/PCDHB9, and the 8 genes were selected for replication.

In the replication stage, 5 of 10 CpG sites in 8 candidate genes from 850K BeadChip results were verified to be statistically different (p < 0.05). The trend of the methylation level of the rest 5 CpG sites with no significant differences is consistent with the 850k BeadChip results (Additional File S3), suggesting that the 850k results are reliable. The genes screened by GO analysis are worthy of further verification.

In 188 paired samples, 203 of 308 CpG sites from 17 CpG islands in the promoter regions of 8 candidate genes were significantly different (p < 0.05) between the IS and control group (Additional File S4). Consistent with the 850k BeadChip results, the methylation level of all 203 CpG sites was lower in the IS group than in the control group. 13 of 17 CpG islands were differentially methylated between IS patients and controls (p < 0.05, Additional File S5). 7 genes (CDH2/PCDHB10/PCDHB11/PCDHB14/PCDHB16/PCDHB3/PCDHB9) showed differences (p < 0.05) between the two groups even if adjusting for smoking, drinking, history of hypertension and diabetes, and blood lipid levels, and adjust FDR p values of 6 genes (CDH2/PCDHB10/PCDHB11/PCDHB16/PCDHB3/PCDHB9) were statistically significant (p < 0.05, Table 3). The methylation levels of the 7 genes were highly correlated in both the case group and the control group (p < 0.05, Additional File S6). Besides, we further split the samples by sex (male or female). There were 7 differentially methylated genes (CDH2/PCDHB10/PCDHB11/PCDHB14/PCDHB16/PCDHB3/PCDHB9) in the male subgroup and 4 (CDH2/PCDHB10/PCDHB11/PCDHB9) in the female subgroup, which is shown in Table 4.

We split the case samples by IS subtypes (TOAST classification), stratified by LAA (n = 67), CE (n = 13), SVD (n = 64), ODC (n = 5), and UND (n = 39). However, no analysis was conducted in the CE and ODC groups because of the small sample size. We also did not perform analysis in the UND group as it is hard to identify the cause of IS. There were 8 differentially methylated genes (CDH2/PCDHB10/PCDHB11/PCDHB14/PCDHB16/PCDHB3/PCDHB6/PCDHB9) in the LAA subgroup and 1 (CDH2) in the SVD subgroup after adjusting for smoking, drinking, history of hypertension and diabetes, and blood lipid levels (p < 0.05, Table 5). Besides, the case samples were seperated according to the location of the lesion (anterior circulation-left hemisphere, n = 72; anterior circulation-right hemisphere, n = 63; posterior circulation, n = 48; and both anterior and posterior circulation, n = 5). We also did not perform analysis in the “both anterior and posterior circulation” group because of the small sample size. Four genes (CDH2/PCDHB10/PCDHB11/PCDHB9) in the anterior circulation-left hemisphere subgroup, 8 genes (CDH2/PCDHB10/PCDHB11/PCDHB14/PCDHB16/PCDHB3/PCDHB6/PCDHB9) in the anterior circulation-right hemisphere subgroup and 4 genes (CDH2/PCDHB10/PCDHB11/PCDHB9) in the posterior circulation subgroup showed differences after adjusting for smoking, drinking, history of hypertension and diabetes, and blood lipid levels (p < 0.05, Table 6).

The methylation level of each gene was grouped into quartiles. The DNA methylation level of the 7 differentially methylated genes in Q4 groups was associated with higher risk of IS than Q1 groups (p < 0.05), which are listed in Table 7.

We tested the biomarker potential of the 7 differentially methylated genes using receiver operating characteristic (ROC) analysis. The sensitivity, specificity, accuracy and area under the curve (AUC) in CDH2/PCDHB10/PCDHB11/PCDHB14/PCDHB16/PCDHB3/PCDHB9 genes was determined based on methylation values of replication stage (Table 8, Figure 4A). To enhance the diagnostic ability of these differentially methylated genes in IS, we tested the performance of the polygenic methylation model. The sensitivity, specificity, accuracy and AUC of polygenic methylation model was shown in Table 8 and Figure 4B using ROC analysis. The polygenic methylation model performed better than each of the individual genes, especially with higher specificity (0.8723), sensitivity (0.883), accuracy (0.8777), and AUC (0.9384).

In the replication stage, there were 7 differentially hypomethylated genes (CDH2/PCDHB10/PCDHB11/PCDHB14/PCDHB16/PCDHB3/PCDHB9) in the IS group. Generally, a decrease in DNA methylation leads to abnormal gene transcription, resulting in the upregulation of gene expression. We validated the expression levels of these 7 genes in the GSE22255 dataset. The expression of two genes (PCDHB9 and PCDHB11) increased, while the methylation level of the same 2 genes decreased in IS patients compared to controls (p < 0.05, Figure 5). The expression levels of the other three genes (CDH2/PCDHB14/PCDHB16) also increased in the IS group, but there were no significant differences between the two groups (p > 0.05, Additional Files S7–11).