A total of four AD patients (2 females, 2 males, mean age 78.5 $\pm$ 6.8) and four age-matched healthy controls (HC, 2 females, 2 males, mean age 71.8 $\pm$ 12.9) were included in the screening experiment. DNA methylation profiling applied the Infinium MethylationEPIC BeadChip 850K (15070022/WG-317-1002, Illumina, San Diego, CA, USA) on these 8 participants. Samples from 17 AD patients (8 females, 9 males, mean age 74.3 $\pm$ 7.9) and 17 age- and gender-matched HC (7 females, 10 males, mean age 76.1 $\pm$ 8.0) were included in the validation experiment between October 2019 and December 2022. Although the sex distribution between AD and HC in the validation study was not perfectly matched, our main goal was to ensure that both groups were matched on age, considering age is a well-recognized confounding factor in AD research. AD was diagnosed based on recommendations from the National Institute on Aging-Alzheimer’s Association workgroups [35]. HCs were contemporary outpatients for physical examination. Exclusion criteria include medical history of malignant tumors, cerebrovascular diseases including ischemic or hemorrhagic stroke, severe metabolic diseases, and severe liver or kidney dysfunction. These subjects were from Shanghai Jinshan Zhongren Aged Care Hospital between October 2019 and December 2022. The research protocol complied with the World Medical Association Declaration of Helsinki regarding ethical conduct of research involving human subjects.

Peripheral blood samples from all participants were drawn after fasting for at least 12 h. A total of 10 mL of whole blood was collected in EDTA tubes (Becton Dickinson, Franklin Lakes, NJ, USA). Genomic DNA was extracted from the blood samples using QIAamp DNA Mini Kit (# 51304, Qiagen, Germantown, MD, USA) according to the manufacturer’s instructions. The extracted DNA was aliquoted and stored at –80 °C until used.

Genomic DNA samples were bisulfite-converted by EpiTect Bisulfite Kit (#59104, Qiagen) according to the manufacturer’s instructions. Subsequently, DNA methylation profiles were depicted using Infinium 850K MethylationEPIC BeadChip (15070022/WG-317-1002, Illumina) on a HiScan system (SY-101-1001, Illumina), covering more than 850,000 different CpGs at single-nucleotide resolution. Methylation level at each CpG site was evaluated by a beta value that represents the normalized probe fluorescence intensity ratio of methylated bead signal to the sum of both methylated and unmethylated signals [beta = methylated intensity/(methylated intensity + unmethylated intensity)]. Resulting beta values vary from 0 (no methylation) to 1 (complete methylation). Raw data preprocessing was carried out, probes with a low detection rate (p $>$ 0.01) when compared with the background signal and probes with a bead count less than three in $\ge$5% of samples were filtered out. Before processing, quality control was performed using density plots of beta values, boxplots, and principal component analysis (PCA). The filtered raw data were normalized using the beta mixture quantile dilation method [36]. Methylation variable positions (MVPs) were identified using the Linear Models for Microarray Data (LIMMA) R Bioconductor package (version 3.52.2, http://www.bioconductor.org/) [37]. The criterion for selecting MVPs was p 0.01. The Benjamini & Hochberg false discovery rate was used for multiple test correction. To identify differentially methylated regions (DMRs), the Probe Lasso algorithm was run using the R package Chip Analysis Methylation Pipeline (ChAMP, version 2.8.1; http://www.r-project.org/) [38].

The differentially methylated candidate genes were validated using pyrosequencing on a cohort of blood samples (17 AD and 17 HC). Pyrosequencing assay was designed with the PyroMark Assay Design software (version 2.0, Qiagen). The primers are listed in Supplementary Table 1. The genomic DNA extracted from peripheral blood was bisulfite-converted. PCR was then performed with the KAPA2G Robust HotStart ReadyMix PCR Kit (KK5603, Kapa Biosystems, Wilmington, MA, USA) in a total volume of 25 µL per reaction, using the ABI Verity PCR System (Thermo Fisher Scientific, Waltham, MA, USA) according to the manufacturer’s instructions. Cycling conditions for PCR were 95 °C for 3 min, 35 cycles of 95 °C for 15 s, 55 °C for 15 s and 72 °C for 15 s followed by final extension at 72 °C for 3 min. The DNA methylation level was then quantified using PyroMark Q48 Advanced CpG Reagents on the Pyromark Q48 Autoprep system (Qiagen). The methylation status of CpG loci was quantified as the percentage of the relative light unit of the C peak (methylated cytosines)/[relative light unit of the C peak + T peak (unmethylated cytosines converted to thymines)].

The microarray data of AD from the Gene Expression Omnibus (GEO; https://www.ncbi.nlm.nih.gov/geo/) was used to compare the DMR-associated genes between AD patients and HCs. Raw data from datasets GSE66351 (106 AD and 84 controls), GSE48350 (80 AD and 173 controls), GSE5281 (84 AD and 74 controls), GSE118553 (167 AD and 100 controls) and GSE28146 (22 AD and 8 controls) were downloaded for bioinformatics analyses. GSE48350 gives sequencing data from entorhinal cortex (EC), hippocampus (H), postcentral gyrus (PCG) and superior frontal gyrus (SFG) of the brains. GSE5281 was from EC, H, medial temporal gyrus, posterior cingulate (PC), SFG, and primary visual cortex of the brain based on the GPL570 platform. GSE118553 included tissue samples from the cerebellum (C), EC, frontal cortex, and temporal cortex, which were obtained from the GPL10558 platform. GSE28146 data was from fresh frozen hippocampal tissue blocks containing both gray and white matter. The web tool GEO2R was utilized to identify differentially methylated genes and differentially expressed genes in the microarray data (p 0.05 and $|$t$|$ $>$2 as the cutoff criteria).

Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed using the EnrichR web platform (http://amp.pharm.mssm.edu/Enrichr). The criterion for selection of significant CpG loci in GO terms and KEGG pathways analyses was p 0.05.

Immune cell infiltration analysis was applied to identify the types and abundance of immune cells in the given tissues or samples. This analysis provides valuable insights into the roles of the immune system in AD. The immune cell infiltration scores for the GSE118553 and GSE48350 datasets were evaluated using the CIBERSORT algorithm (https://cibersort.stanford.edu/). This algorithm estimated the proportional analysis of twenty-two immune cell types and the infiltration levels of immune cells between AD and control groups. Additionally, it facilitated the correlation analysis of relationships between DMR-annotated genes and immunocyte subsets. Wilcoxon analysis by SPSS 19.0 software (IBM SPSS, Armonk, NY, USA) was applied to compare the differences between the two groups. Tests were considered statistically significant at p 0.05.

The nonparametric Mann-Whitney test was used to compare the methylation levels of candidate genes in AD patients and controls. A value of p 0.01 was considered statistically significant when determining MVPs. DMRs were determined with the “Probe-Lasso” algorithm of the ChAMP package (version 2.8.1; http://www.r-project.org/). The diagnostic performance of the candidates was evaluated by the area under the receiver operating characteristic (ROC) curve. All statistics were performed by SPSS 19.0 software (IBM SPSS, Armonk, NY, USA), GraphPad Prism 5.01 (GraphPad Software, San Diego, CA, USA), and MedCalc 12.1.4.0 software (MedCalc Software, Mariakerke, Belgium).

To explore the methylation profiling of AD, an Infinium MethylationEPIC BeadChip (850K) was used for eight participants (4 AD cases, and 4 HCs). Density plot depicted the averaged raw beta-values of all CpG sites (755743 probes) from AD patients and HCs (Fig. 1A). The beta mixture quantile dilation method was applied to standardize the original data. After normalization, the density plots of normalized beta-values were generated for each sample using 850K methylation data, where a red line represents the densities from the AD patient, and a black line for the HC (Fig. 1B). Normalized methylation beta values of all CpG probes in each sample were compared by box plot that reflected the mean value, the degree of dispersion and the maximum and minimum value which showed no significant intragroup diversity between the samples from AD and HCs (Fig. 1C). Principal component analysis (PCA) used normalized methylation beta values. A separation between the AD (red dots) and normal samples (green triangles) can be observed in the first PCA component (Fig. 1D). And 4335 MVPs were examined with two-way (samples and MVPs) hierarchical clustering. The two groups (AD and HC) separated into distinct clusters (Fig. 1E). The MVPs were further analyzed by volcano plot. The generated plot illustrated the MVPs between the HCs and AD patients. The red dots in the upper left corner represent the hypomethylated loci; and those at the upper right denote hypermethylated loci in AD patients when compared to HC (Fig. 1F).

Fig. 1.

Methylation profiling and analysis of AD and control samples using Infinium MethylationEPIC BeadChip. (A) Density plot of raw beta-values from 755,743 CpG sites across eight participants (4 AD cases, 4 controls). (B) Post-normalization density plots show standardized beta-values for each individual, with AD patients (red) and healthy controls (black). (C) Box plot of normalized methylation beta values for each sample. (D) PCA scatter plot of the normalized methylation beta values. (E) Two-way hierarchical clustering of the 4335 MVPs identified, showing distinct clusters formed by AD and control groups. (F) Volcano plot of the MVPs comparing HC and AD patients. The horizontal axis depicts the delta normalized beta values and the vertical axis reflects the –log₁₀ (p-value). AD, Alzheimer’s disease; CpG, cytosine-phosphate-guanine; PCA, principal component analysis; MVPs, methylation variable positions; HC, healthy control.

The array analysis by LIMMA indicated that 4335 MVPs showed DNA methylation differences (p 0.01) in AD patients relative to HCs. Of these, the methylation levels of 3826 MVPs and 509 MVPs, respectively increased and decreased. The relevant information of the detected MVPs, including probe ID, methylation level, fold change, p-value, genes and features, is summarized in Supplementary Table 2. The most significant MVP annotated to unique gene was located in the TSS200 promotor region of the SEC31B gene and showed hypermethylation in AD patients (probe cg26705561, p = 3.91 $\times$ 10^-6). The most variable loci of hypermethylation (probe cg08291996) annotated to the ZNF8 gene body, while the most hypomethylated loci (cg26023019) annotated to GRIK1 (Supplementary Table 2).

Functional annotation analysis (Gene Ontology, GO) was performed for genes that were associated with the identified MVPs (Fig. 2A). GO annotations, including biological process, cellular component, and molecular function categories, were analyzed according to the MVP-annotated genes. The biological process category emphasized the processes of nervous system development, generations of neurons, and neurogenesis. The cellular component category highlighted neuron projection, dendrite, and synaptic membrane. The molecular function category was differentially enriched in calcium ion binding, Ras/Rho guanyl-nucleotide exchange factor activity, phospholipid binding, GTPase/nucleoside-triphosphatase regulator activity, cytoskeletal protein binding and other related functions (Supplementary Table 3). KEGG pathway enrichment of the targeted genes showed that glutamatergic synapse, neuroactive ligand-receptor interaction, retrograde endocannabinoid signaling, γ-Aminobutyric acid (GABA)ergic synapse, circadian entrainment and Hippo signaling pathway were highly connected to AD (Fig. 2B, Supplementary Table 3). The KEGG pathway enrichment bubble plot of MVPs is shown in Fig. 2C.

Fig. 2.

Functional annotation and pathway analysis of genes linked to MVPs in AD. (A) GO categories for MVP-associated genes, highlighting roles in nervous system development and synaptic structures. (B,C) KEGG pathway enrichment analysis, showing significant pathways (e.g., glutamatergic synapse and Hippo signaling) relevant to AD pathophysiology. GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes.

DMRs represent discrete genomic sequences with a distinct CpG methylation status that distinguishes one phenotype from another reducing the genome-wide scale to easy-to-understand regions. The identification and utility of genome-wide DMRs have led to a better understanding of the pathogenesis of various diseases and have introduced the development of effective assays for clinical diagnosis. DMR hunting is run by the algorithm “Probe-Lasso”. Probes which are located in the promoter regions (including TSS200 or/and TSS1500) and the other regions (including 5^′-untranslated region (UTR), 1st Exon, gene body, 3′-UTR and intergenic region) are stratified by epigenomic CpG-island-relation region (island, shore, shelf, and open sea). Considering that the probe racing is not uniform in the methylation tests, Probe Lasso defines dynamic boundaries for each probe depending upon its genomic feature type (Fig. 3A). Probe Lasso uses dynamic windows to identify a clear DMR boundary for spanning multiple neighboring MVPs. Probe Lasso identified 68 altered DMRs annotated to 10 genes in this study (see Table 1 and Supplementary Table 4 for detailed annotation of the regions). As shown in Table 1, the DMRs showing hypermethylation in AD samples annotated to ANKRD33, MIR487B (miRNA cluster also contains MIR381, MIR539, MIR381HG and MIR889), SLC43A2, TMEM100, GDF7, SALL4, KRBOX1-AS1 (KRBOX1), VWA5B2 (MIR1224), CTBP1 (C4orf42) and PCDHB11. The methylation levels of the DMRs from different annotated genes were estimated by the average $\beta$-values in AD and HCs (Fig. 3B). DMRs located in the transcription start site (TSS) with most significant difference were especially annotated to the CTBP1 gene (chr4: 1243964 – 1244047, p = 7.66 $\times$ 10^-7) and contained four identified probes with DMR size 84 bp (Table 1).

Fig. 3.

Probe Lasso algorithm calling AD-associated DMRs and functional analysis of DMR-related genes. (A) Visualization of lasso quantile and radius across various genomic features stratified by CpG island (CGI) relations, showcasing the methodology for identifying AD-associated DMRs. The vertical axis represents the lasso radius and the horizontal axis represents the 28 genetic/epigenetic categories (7 genomic features $\times$ 4 CGI relations). Lasso quantile = 0.59; mean lasso radius = 375 bp. (B) Comparison of average methylation levels in DMR-annotated genes between AD patients. (C) GO enrichment analysis for DMR-related genes. (D) KEGG pathway analysis for these genes, showing enriched pathways such as Notch signaling and Hippo signaling pathways related to AD pathology. DMRs, differentially methylated regions.

GO annotation and KEGG pathway enrichment analyses were carried out based on the DMRs annotated genes. The biological process category emphasized the enriched GO terms related to nervous system development, including neural tube formation and development. The cellular component category highlighted heterochromatin and transcriptional repressor complex. The molecular function category was differentially enriched in RNA polymerase II transcription corepressor, repressing transcription factor binding, protein homodimerization activity, and NAD binding (Fig. 3C). KEGG pathway enrichment highlighted the roles for the Notch, TGF-β, Wnt, and Hippo signaling pathways in AD (Fig. 3D). This suggests that DMRs annotated genes in AD may influence the disease’s progression by modulating immune processes.

Based on the enrichment analysis results, the aim was to understand the alterations in immune cell dynamics within AD. Consequently, immune cell infiltration in AD was further investigated using CIBERSORT. The immune cell infiltration analysis (Fig. 4) compares the distribution of immune cells in AD patients and HCs using data from public GEO datasets. Fig. 4A,B show bar graphs representing the immune cell distributions in datasets GSE48350 and GSE118553, respectively, depicting the overall distribution of various immune cells. Fig. 4C,D provide box plots that further detail these distributions, offering insights into the variability and median levels of immune cells across both conditions. In the GSE48350 dataset, monocytes (p 0.001) and T cells CD4 memory resting (p 0.001) exhibited significantly elevated levels in AD patients, whereas natural killer (NK) cells activated (p 0.05), plasma cells (p 0.05), and T cells follicular helper (p 0.05) were significantly reduced. In the GSE118553 dataset, eosinophils (p 0.01) were significantly elevated in AD, while macrophages M0 (p 0.05) and T cells CD4 memory resting (p 0.001) showed significantly decreased levels. Additionally, data from both datasets were pooled to analyze correlations between the expressions of genes associated with DMRs and immune cell distributions, revealing potential links between these epigenetic markers and immune profiles (Fig. 4E). Among them, CTBP1 exhibited strong correlations with immune cell types. It was significantly negatively correlated with memory B cells (p 0.001), M2 macrophages (p 0.001), and eosinophils (p 0.001). Conversely, significant positive correlations were observed between CTBP1 and plasma cells (p 0.001), follicular helper T cells (p 0.001), and M0 macrophages (p 0.001). Correlations between various immune cell types and AD were assessed, identifying key immune cells that may play key roles in the pathology of AD (Fig. 4F,G). In both datasets, T cells follicular helper demonstrated a significant correlation with AD (p 0.001). This analysis not only improves our understanding of the immunological changes in AD but also highlights the critical role of DMRs-related genes in influencing these immune variations. Furthermore, it emphasizes the potential of immune cells as biomarkers in the diagnosis and progression of this disease, offering insights into both biological mechanisms and therapeutic targets.

Fig. 4.

Analysis of immune cell infiltration in AD using CIBERSORT. (A,B) Bar graphs depicting the distribution of immune cells across AD patients and healthy controls, as analyzed from datasets GSE48350 and GSE118553, respectively. (C,D) Box plots providing a detailed view of the immune cell distributions within the GSE48350 and GSE118553 datasets. (E) Heatmap showing the correlation between the expression of genes associated with DMRs and the distribution of immune cells. (F,G) Correlation plots illustrating the relationship between various immune cell types and the pathology of AD, with significance levels marked. * p 0.05, ** p 0.01. *** p 0.001.

To further verify the hypermethylation of the CTBP1 TSS1500 promoter, the methylation levels of DMRs covering four CpG sites (cg26479374, cg16399632, cg25897951, and cg08948841) were validated by pyrosequencing on a larger cohort of blood samples. Analysis revealed that all four sites had a similar differential methylation pattern when compared with the 850K array data. CpG island sites cg26479374, cg25897951, cg08948841 in the CTBP1 promoter region were significantly hypermethylated in AD when compared with HCs. Although, the CpG island site cg16399632 showed no significant difference in the methylation between the two groups, when the methylation percentages of the four probes in CTBP1 promoter were averaged, the mean methylation levels were also significantly increased in AD cases relative to HCs (Fig. 5A). The ROC curve analysis was applied to evaluate the diagnostic performance of CTBP1 promoter hypermethylation for AD. The area under curve (AUC) was 0.779 (95% CI: 0.604–0.90, p 0.001) (Fig. 5B), exhibiting promising potential as a biomarker candidate that distinguishes AD from HCs.

Fig. 5.

Validation and diagnostic analysis of CTBP1 promoter methylation in AD. (A) Scatter plots illustrating methylation levels at four specific CpG sites within the CTBP1 promoter region (cg26479374, cg16399632, cg25897951, cg08948841) in AD and HC. n = 34 (HC: 17, AD: 17), n.s., no significance (* p 0.05, ** p 0.01). (B) ROC curve demonstrating the diagnostic performance of CTBP1 promoter hypermethylation in distinguishing AD from HCs, with an AUC of 0.779, indicating a high diagnostic potential. (C) Box plots representing the distribution of methylation levels for eight differentially methylated loci related to genes identified in the study (GDF7, SALL4, VWA5B2) across AD patients and HCs. n = 190 (Control: 84, AD: 106, *** p 0.001, **** p 0.0001). CTBP1, C-terminal binding protein1; ROC, receiver operating characteristic; AUC, area under curve.

Previously, Song et al. [39] obtained 10054 aberrantly methylated genes via methylation microarray in Gasparoni DNAm data (GSE66351), including 5002 hypermethylated and 5052 hypomethylated loci. To further verify the status of the 68 altered DMRs, eight aberrant methylated loci were found annotated to the genes in the 850K microarray data, including four probes annotated to GDF7 (cg14780466, cg12334488, cg10553204, cg10687131), three probes annotated to SALL4 (cg06303238, cg14160518, cg24332577), and one probe annotated to VWA5B2 (cg03981074; Fig. 5C and Table 2).

Regions of aberrant methylation were initially identified in the DNA in peripheral blood from AD patients and these regions were linked to key genes. To investigate whether the expression of these genes in the brain undergoes changes that might influence the progression of AD, four datasets from the GEO public database were analyzed: GSE5281, GSE118553, GSE48350, and GSE28146. Results indicated that CTBP1 expression changes were the most significant and consistent across these datasets (Fig. 6A–D), However, the expression of CTBP1 varied significantly across these datasets, showing both downregulation (Fig. 6B) and upregulation (Fig. 6C,D) in AD when compared to HCs. This variability in CTBP1 expression was attributable to its distinct expression specificity in different brain regions, such as decreased CTBP1 in the cerebellum (GSE118553) and increased CTBP1 in the hippocampus (GSE48350, GSE28146), implying a regional involvement in AD development

Fig. 6.

Expression analysis and diagnostic evaluation of DMR-annotated genes across multiple GEO datasets. (A–D) Box plots show the expression levels of DMR-annotated genes in AD patients versus healthy controls across four different GEO datasets: GSE5281, GSE118553, GSE48350, and GSE28146. Plots illustrate varying levels of CTBP1 expression, with notable downregulation and upregulation in certain datasets, reflecting the gene’s complex role in AD progression. (E–H) ROC curves evaluating the diagnostic potential of CTBP1 and other gene expressions as biomarker for AD. Each panel represents one of the four datasets. Curves show AUC values, highlighting the elevated and stable diagnostic potential of CTBP1 across different datasets (* p 0.05, ** p 0.01, *** p 0.001, **** p 0.0001, ns, not significant).

Further analysis was conducted using ROC curve analysis to evaluate the potential of these gene expressions as biomarkers for AD in the brain. CTBP1 showed the highest and most stable AUC values (GSE5281, AUC = 0.583; GSE118553, AUC = 0.627; GSE48350, AUC = 0.701; GSE28146, AUC = 0.795), reinforcing the significance of its role in AD progression (Fig. 6E–H). However, when compared to findings presented above, measuring the methylation level of the CTBP1 promoter in peripheral blood is more suitable and convenient for use as a biomarker for AD. Our study emphasizes the relevance of CTBP1 in AD research as a biomarker. Additionally, it highlighted that the expression of CTBP1 likely has a crucial regulatory role in the progression of AD within the brain.