DNA methylation signatures associated with early-onset schizophrenia in Chinese patients

Abstract

Schizophrenia is a heterogeneous psychiatric disorder with diverse clinical manifestations and complex biological mechanisms, in which age-at-onset (AAO) critically influences disease trajectory. Patients with early-onset schizophrenia (EOS; AAO < 18 years) present with more pronounced neurodevelopmental deficits and poorer long-term outcomes compared to adult-onset (AOS) cases. Previous genetic research on AAO and EOS has primarily focused on candidate genes and genome-wide association studies (GWAS). DNA methylation, an epigenetic mechanism influenced by the interplay between environmental and genetic factors, remains understudied, especially in the Chinese population. Peripheral blood DNA from 120 schizophrenia patients (49 EOS, 71 AOS) was analyzed using the Infinium MethylationEPIC v2.0 array. Differential methylated analyses were conducted for both EOS-AOS dichotomous comparison and continuous AAO, with stringent adjustment for age, sex, smoking, and estimated cell proportions. At a suggestive significance threshold (p < 5 × 10⁻⁵), we identified 49 differentially methylated positions (DMPs) for EOS-AOS and 126 DMPs for AAO. Genes annotated to the identified DMPs included known schizophrenia and EOS-associated loci (such as ORMDL1, ANXA4, and TRRAP), as well as novel regions linked to cognitive function and neurodevelopment (such as AKAP8L, GPRC5C, and C4orf45). Enrichment analysis implicated key biological processes, including kinase signaling, cell cycle regulation, and microRNA pathways involved in apoptosis and oncogenesis. This study reveals novel differential DNA methylation patterns associated with EOS in the Chinese population and identifies key biological pathways potentially underlying its pathogenesis.

Introduction

Schizophrenia is a severe mental illness characterized by delusions, hallucinations, disorganized thinking, affective blunting, and impaired psychosocial functioning [1]. The global prevalence of schizophrenia is estimated at 0.32%, affecting 24 million individuals worldwide [2]. In Asian countries, schizophrenia accounts for around 13.8% of disability-adjusted life years (DALYS) related to mental disorders [3]. Moreover, projections indicate that the prevalence and disease burden of schizophrenia will rise steadily over the next decade. Age-at-onset (AAO) represents a critical clinical characteristic that strongly influences both disease presentation and prognosis in schizophrenia. Early-onset schizophrenia (EOS), defined by the emergence of psychotic symptoms before age 18 [4], represents a clinically distinct subgroup characterized by stronger neurodevelopmental contributions, greater genetic loading, more severe cognitive deficits, and poorer long-term outcomes compared to adult-onset schizophrenia (AOS) [5–7]. These clinical differences likely reflect unique neurobiological alterations associated with EOS’s developmental origins [8–10], emphasizing the critical need for targeted research into this severe disease subtype.

Schizophrenia is a highly heritable, polygenic disorder arising from complex gene-environment interactions [11]. While large-scale Genome-Wide Association Study (GWAS) by the Psychiatric Genomics Consortium (PGC) has recently identified 287 distinct risk loci contributing to schizophrenia [12], the genetic architecture underlying AAO and EOS remains largely unclear. SNP-based heritability estimates suggest AAO is moderately heritable (17% – 21%) [13], however, our recent review found minimal overlap among top GWAS hits for AAO, demonstrating limited reproducibility across studies [13]. This genetic complexity, particularly for EOS, which may represent a distinct etiological subtype, highlights critical gaps in our understanding of onset-related mechanisms that warrant further investigation.

Epigenetics offers critical insights into the complex interactions between genetic factors and environmental influences in human health and disease [14]. DNA methylation is an important and dynamic epigenetic mechanism, influencing gene expression through direct chemical modification of DNA cytosine sites without altering the underlying DNA sequence [15] and play a crucial role in biological processes including genomic imprinting, cellular differentiation, and X-chromosome inactivation. Epigenome-wide association studies (EWAS) have emerged as a powerful tool to investigate the role of epigenetic modifications, particularly DNA methylation in human health conditions. For instance, in major depressive disorder (MDD), EWAS meta-analyses reveal multiple differentially methylated positions (DMPs) and regions (DMRs) associated with MDD [16]. By using samples from two cohorts (298 MDD cases and 63 controls), researchers identified significant cytosine-phosphate-guanine (CpG) sites, including associations with TNNT3 [17]. Key findings include 127 DMRs with Sidak-corrected p-values < 0.05, and an enrichment of pathways related to neuronal synaptic plasticity, calcium signaling, and inflammation. In patients with attention-deficit/hyperactivity disorder (ADHD), EWAS have also successfully uncovered epigenetic markers and associated mechanisms [18]. A recent investigation of adults with ADHD identified differential methylation in genomic regions linked to autoimmune diseases, cancer, and neuroticism [19] Notably, this study also revealed a significant epigenetic overlap between ADHD and smoking-related phenotypes, including maternal smoking. These collective findings not only reinforce the role of DNA methylation in ADHD pathophysiology but also emphasize the critical need for larger-scale studies to elucidate these complex mechanisms.

Growing evidence also suggests the involvement of epigenetic variation in the molecular etiology of schizophrenia [20–22]. For instance, differential DNA methylation in schizophrenia has been reported for the GABBR1 gene [23], which encodes a receptor for gamma-aminobutyric acid (GABA). Importantly, SNPs within GABBR1 have been linked to treatment-resistant schizophrenia [24]. To our knowledge, only three previous studies have been conducted to examine the methylation variations for EOS or AAO in European populations [25–27], and there is a dearth of research on this topic in East Asians.

Taken together, epigenetic research in schizophrenia shows promise to provide insights into neurobiological mechanisms. EOS, a severe form of schizophrenia characterized by significant neurodevelopmental abnormalities, has been associated with limited and inconsistent GWAS findings. Whilst epigenetics may be involved in EOS and may influence AAO in schizophrenia, research in this area remains limited and confined to the European populations. To address these important knowledge gaps, we conducted an epigenetic study on a sample of Chinese schizophrenia patients, aiming to identify DNA methylation signatures associated with AAO by comparing EOS and AOS patients.

Methods

Study participants and diagnostic criteria

A sample of 120 Chinese schizophrenia patients (64 females and 56 males) was selected from a larger pool of first-episode schizophrenia patients who participated in an endophenotype project of first-episode psychosis conducted at Castle Peak Hospital, Hong Kong [28]. The earlier endophenotype project involved the collection of peripheral blood DNA from first-episode schizophrenia patients in an early psychosis intervention program in 2009–2013. The diagnosis of schizophrenia was ascertained by qualified psychiatrist, according to the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) [29]. The inclusion criteria included 1) Chinese ethnicity; 2) DSM-IV diagnosis of schizophrenia in 2009–2013; and 3) end-point best-estimate diagnosis of schizophrenia in 2020. The exclusion criteria included 1) IQ < 70; 2) history of neurological disorders; 3) history of head injury with loss of consciousness for more than 30 min; 4) history of electroconvulsive therapy in the past 6 months before DNA collection; 5) comorbid substance abuse; and 6) comorbid alcohol abuse. The AAO was determined by the treating-psychiatrists during the first structured clinical interview. In this study, EOS participants were defined as having AAO < 18 years of age, whilst AOS participants had AAO ≥ 18 years of age [4]. We further reviewed the documented information regarding AAO on participants’ medical records.

DNA methylation data

Genomic DNA was extracted from whole blood using the high-salt method with chloroform, and the DNA was subsequently processed with sodium bisulfite conversion using the Zymo EZ DNA methylation Kit (Zymo Research, Irvine, California, United States). Then, we analyzed the “bisulfite-converted DNA” using the Infinium Methylation EPIC v2.0 Beadchip Kit provided in the Illumina iScan system (Illumina, San Diego, California, United States). Stringent quality control was performed using the R package minfi [30], which involved the removal of probes with high detection p-values (>0.05), low bead counts ( < 3 beads), no signal, or more than 5% missing data, as well as cross-reactive probes. Quantile normalization was conducted using the “PreprocessQuantile” function to minimize variance and bias.

Differential methylated position (DMP) analysis

We examined the association of differential DNA methylation positions in peripheral blood with onset-age status (EOS vs AOS) in 49 EOS and 71 AOS participants, after controlling for the participants’ chronological age, sex, smoking score (estimated by the methylation level of cg05575921, AHRR) [31], and relevant cell type proportion estimated by Horvath’s epigenetic clock. The clock predicts biological age based on DNA methylation levels at specific CpG sites and utilizes the projectCellType function from the minfi package to accurately estimate cell type proportions [30, 32]. Cell type compositions were selected based on their association with the targeted phenotype; only variables that demonstrated a significant association with EOS were included as covariates in the analysis. The differentially methylated position (DMP) analysis was conducted using the limma package in R [33]. The reference group was defined as individuals with AOS, the log2 fold change (logFC) represents the differential methylation between EOS and AOS, with positive values indicating higher methylation in EOS compared to AOS. Genomic feature of probes was annotated by the manifest file provided by Illumina. After rigorous quality control filtering, epigenome-wide analysis of 876,984 CpG sites employed two significance thresholds: a stringent Bonferroni-corrected threshold p < 5.7 × 10⁻⁸ (as calculated by 0.05/876984), and a suggestive ‘discovery’ significance threshold of p < 5 × 10⁻⁵, consistent with thresholds commonly used in previous EWAS studies [21, 34, 35].

We additionally conducted DMP analysis by treating AAO as a continuous variable in linear regression, using identical covariates (age, sex, smoking score, and cell proportion) to identify methylation loci associated with disease onset timing.

Differential methylated region (DMR) analysis

We utilized the DMRcate Bioconductor R package to identify differentially methylated regions (DMRs) associated with EOS and AAO [36], with adjustment for the same covariates used in the DMP analysis [37]. DMRs were calculated using a parameter setting of lambda = 1000, and kernel adjustment C = 2. Statistically significant DMRs were defined to have at least two distinct probes and meet the harmonic mean of the individual component FDRs (HMFDR) of < 0.05 [38].

Functional enrichment analysis

To characterize the functional properties of differentially methylated positions (DMPs), we performed gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses using ‘g:Profiler’ web tool [39]. Genes corresponding to the top-ranked EWAS probes (p < 1 × 10⁻⁴) were selected for analysis, a conventional cutoff employed in exploratory EWAS that balances discovery potential with statistical stringency [40, 41]. These genes were identified by mapping significant probes to their annotated gene targets using the illumina manifest file. The enrichment analysis was conducted using default g:Profiler parameters for Homo sapiens, to ensure biologically meaningful interpretation while avoiding overly broad or narrow terms, we constrained term size to 10–1000 and implemented false discovery rate (FDR) correction with a threshold of p < 0.05.

Sensitivity analyses

Given that the original age-based categorization (using a cut-off of 18 years) may be inadequate to capture neurodevelopmental or epigenetic heterogeneity, we conducted sensitivity analyses by defining more extreme subgroups based on AAO. Specifically, we compared a subgroup with AAO < 16 years to a subgroup with AAO > 20 years.

Moreover, to address potential concerns regarding collinearity between AAO and chronological age, we conducted a sensitivity analysis using an alternative statistical approach. We first generated residualized AAO values by regressing AAO on chronological age. We employed this residualized measure as the primary predictor in a replicate of the secondary analysis model, from which the chronological age covariate had been excluded. The consistency of effect estimates between this model and the primary model would be interpreted as evidence for the robustness of the associations with AAO.

This study was approved by ethics committees of The University of Hong Kong and Castle Peak Hospital Hong Kong (Approval number: NTWC/REC/823/10; NTWC/CREC/1293/14 and UW14–325). All participants provided written informed consent. All methods were performed in accordance with the relevant guidelines and regulations.

Results

Our cohort comprised 49 EOS patients and 71 AOS patients of Chinese ancestry. From these 120 participants, high-quality DNA methylation data encompassing 876,984 probes passed rigorous quality control and were retained for downstream association analyses. EOS participants were significantly younger at the time of assessment (19 ± 4.0 vs. 39.4 ± 9.0 years, p < 2.2e-16) and had earlier mean AAO (14.7 ± 1.2 vs. 36.0 ± 8.9 years, p < 2.2e-16). The AOS group had a slightly higher proportion of females (62% vs. 41%, p = 0.036). While 51% of EOS participants and 35% of AOS participants received clozapine at the time of assessments, the clozapine prescription rate (a proxy index for treatment resistance) did not differ between the two AAO groups (p = 0.124), as shown in Table 1.

Table 1.

Sample characteristics.

	Early-onset schizophrenia patients (n = 49)	Adult-onset schizophrenia patients (n = 71)	P-value
Gender ratio, female (%)	40.81%	61.97%	0.036
Age (mean ± sd)	19 ± 4.00	39.44 ± 9.02	<2.2e-16
Age At Onset (AAO)
Min	12	24
Max	17	55
Mean ± sd	14.65 ± 1.18	36 ± 8.92	<2.2e-16
Clozapine prescription ratio (%)	51.02%	35.21%	0.124

After controlling for covariates, we identified 49 differentially methylated positions (DMPs) associated with EOS that reached a suggestive significance level (p < 5 × 10⁻⁵), as depicted in the Manhattan plot (Fig. 1A). Among them, 19 EOS-associated DMPs were hypermethylated and 30 were hypomethylated. Figure 1B shows the QQ plot of the DMP analysis. Table 2 shows the top 20 DMPs, and a complete list of all DMPs surpassed the suggestive significance level can be found in Supplementary Table S1. The most significant signal was detected at probe cg23276760 at chr18: 13222181, with a p-value of 2.07 × 10⁻⁷. This site is not annotated to any known gene in the current genomic map. Probe cg14588779 was mapped to the shore area of the gene AKAP8L (p = 7.54 × 10⁻⁷). This probe was found to have higher methylation in the EOS group than the AOS group (Fig. 2). Probe cg07577294 was mapped to the open sea area of the gene ORMDL1 (p = 3.31 × 10⁻⁶), cg19627145 was annotated to the shore of gene MIR6893 (p = 4.99 × 10⁻⁶), and cg25830048 was mapped to the shore of gene HES7 on chromosome 17 (p = 6.67 × 10⁻⁶). Additional suggestive associations were identified in genes C4orf45, LINC01357, FGD6, TRIM16, ARHGAP15, ANXA4, MAFHS1, C16orf87, and TRRAP, as well as intergenic regions. In general, most DMPs we found were located in open sea areas, while a minority were distributed in island and shore areas.

Fig. 1. Manhattan plot and QQ plot of the differentially methylated position (DMP) analysis of early-onset schizophrenia, with adjustment for sex, age, smoking, and blood cell composition.

A The Manhattan plot, with the x-axis showing the chromosomal positions; the y-axis showing p-values of the DMP on a -log10 scale; and the red horizontal line indicates the threshold for suggestive sites (p = 5 × 10^-5). B The QQ plot, with the x-axis showing the expected p-values on a -log10 scale; and the y-axis showing observed p-values on a -log10 scale.

Table 2.

Top 20 differentially methylated probes associated with early-onset schizophrenia in models adjusted for sex, age, smoking, and blood cell-type composition.

Chromosome	Position	Probe	Relation to Island	Gene annotation	LogFC	P-value	FDR
chr18	13222181	cg23276760	Shelf		0.334	2.07E-07	0.096
chr9	94779964	cg11128804	OpenSea		−0.279	2.19E-07	0.096
chr19	15417719	cg14588779	Shore	AKAP8L	0.264	7.54E-07	0.221
chr2	189775366	cg07577294	OpenSea	ORMDL1	−0.525	3.25E-06	0.522
chr11	71330645	cg16586747	OpenSea		0.376	4.11E-06	0.522
chr8	144436727	cg19627145	Shore	MIR6893	−0.23	4.99E-06	0.522
chr9	95382434	cg15183380	OpenSea		−0.319	5.01E-06	0.522
chr16	15661053	cg13534698	OpenSea		−0.21	5.48E-06	0.522
chr17	8124142	cg25830048	Shore	HES7	−0.511	6.67E-06	0.522
chr3	108586951	cg07986500	Shelf	CIP2A	−0.292	7.20E-06	0.522
chr3	106999913	cg03787282	OpenSea		−0.473	7.57E-06	0.522
chr2	179825042	cg04625976	OpenSea		0.319	7.87E-06	0.522
chr19	4620811	cg27363092	OpenSea	RPS10	−0.214	8.12E-06	0.522
chr7	108488998	cg14121150	OpenSea		0.249	8.87E-06	0.522
chr4	159035112	cg02753742	OpenSea	C4orf45	−0.424	8.92E-06	0.522
chr1	112849958	cg27077219	Island	LINC01356; LINC01357	−0.507	1.00E-05	0.546
chr12	95091938	cg06809729	OpenSea	FGD6	−0.322	1.06E-05	0.546
chr17	15683810	cg01440105	OpenSea	TRIM16	−0.397	1.20E-05	0.582
chr2	143519200	cg12657374	OpenSea	ARHGAP15	−0.358	1.26E-05	0.582
chr2	69741872	cg08311172	Island	ANXA4	0.423	1.37E-05	0.599

FC fold change, FDR false discovery rate.

Fig. 2. Top differentially methylated probes associated with early-onset schizophrenia.

Note. The x-axis shows the group status (adult-onset, early-onset) of participants. The y-axis shows the beta-value, which quantifies the percentage of methylation at the differentially methylated probes (ranging from 0 to 1).

While the DMR analysis did not identify any differentially methylated regions significantly associated with EOS in our sample, exploratory enrichment analysis of the 98 selected CpGs (mapping to 66 distinct genes, Supplementary Table S1) revealed several enriched functional categories. These included ten biological process (BP) terms, five molecular function (MF) terms, and four cellular component (CC) terms in Gene Ontology (GO) analysis, along with three significant KEGG pathways As detailed in Supplementary Table S2, our most significant findings suggested the involvement of mitotic cell cycle (p < 0.001), kinase activity (p = 0.004), mitotic cell cycle processes (p = 0.005), and pathways of proteoglycans in various types of cancer (p = 0.009), leukocyte transendothelial migration (p = 0.033), and MicroRNAs in cancer (p = 0.043).

Taking AAO as a continuous variable, we identified 10 significant DMPs with p < 5.7 × 10⁻⁸, and 126 reached the suggestive threshold (p < 5 × 10⁻⁵), as shown in Table 3 and Supplementary Figure S1. The strongest signal was observed at chr11: 64765403, on probe cg05061107 with a p-value of 2.6 × 10⁻⁹, and was mapped to gene SF1. The other highlighted genes included cg01571491 on GPRC5C (p = 6.3 × 10⁻⁹), cg01048098 on PCBP3 (p = 3.4 × 10⁻⁸), cg19331686 on LINC02541 (p = 4.4 × 10⁻⁸), and cg00592383 on ZNF573 (p = 5.8 × 10⁻⁸). A complete list of all DMPs associated with AAO at the suggestive significance level is provided in Supplementary Table S3.

Table 3.

Top 20 differentially methylated probes associated with AAO (continuous variable) in models adjusted for sex, age, smoking, and blood cell-type composition.

Chromosome	Position	Probe	Relation to Island	Gene annotation	Log FC	P-value	FDR
chr11	64765403	cg05061107	Shore	SF1	0.028	2.60E-09	0.001
chr2	213037492	cg00185482	OpenSea		0.058	3.67E-09	0.001
chr20	44631887	cg09512837	OpenSea		0.034	3.91E-09	0.001
chr17	74443911	cg01571491	Shelf	GPRC5C	0.035	6.31E-09	0.001
chr22	25154757	cg24569239	OpenSea		0.037	1.01E-08	0.002
chr9	135931658	cg00549706	Island		0.030	1.30E-08	0.002
chr12	96327090	cg17951958	OpenSea		−0.060	2.48E-08	0.003
chr21	45900497	cg01048098	Shore	PCBP3	0.048	3.39E-08	0.004
chr6	113650684	cg19331686	OpenSea	LINC02541	0.035	4.36E-08	0.004
chr7	133400193	cg18321423	OpenSea		0.050	5.20E-08	0.005
chr19	37740796	cg00592383	OpenSea	ZNF573	0.031	5.82E-08	0.005
chr15	66307840	cg12542836	OpenSea		0.042	7.02E-08	0.005
chr2	70161334	cg19642252	OpenSea		0.036	7.79E-08	0.005
chr6	31961826	cg20942691	Shelf		−0.032	1.21E-07	0.008
chr17	82088342	cg09799307	Island		0.037	1.33E-07	0.008
chr13	97363876	cg05041591	OpenSea	RNA5SP37	0.142	2.00E-07	0.010
chr13	99590260	cg10309230	OpenSea	ENSG00000287746	0.037	2.11E-07	0.010
chr11	125257881	cg00634783	OpenSea		0.030	2.11E-07	0.010
chr16	47499917	cg11918427	OpenSea		0.040	2.12E-07	0.010
chr6	168322190	cg17511017	Shelf		0.049	2.83E-07	0.012

FC fold change, FDR false discovery rate.

Moreover, we identified a DMR with an HMFDR < 0.05 at chr11:64765367-64765875 for AAO, which overlapped with the SF1 gene again. The second-ranked DMR overlapped with the gene LINC00539, with an HMFDR of 0.47. Detailed results are provided in Supplementary Table S4. Enrichment analysis for EWAS of AAO did not identify any GO terms or KEGG pathways that surpassed the significance threshold, suggesting no statistically significant enrichment in this exploratory test.

We compared the top-ranked DMPs (p < 5 × 10⁻⁵) between the linear and logistic regression models and did not observe any overlapping probes. However, upon further investigation of proximal probes within a 500 kb window, we identified three pairs of spatially clustered DMPs, namely (1) cg22991768 and cg03646823 on chromosome 17 (175 kb apart), (2) cg18490648 and cg12657374 on chromosome 2 (209 kb apart), and (3) cg19642252 and cg08311172 on chromosome 2 (419 kb apart).

A sensitivity analysis was conducted on refined earlier-onset (<16 years, n = 35) and later-onset (>20 years, n = 71) subgroups to test the robustness of our findings to this alternative definition of onset age group (Supplementary Table S5). Despite the reduced statistical power from the smaller sample size, the top differentially methylated positions from the primary analysis showed consistent directionality of effect. Notably, the association for our primary top signal, cg23276760, remained suggestive significant and retained a consistent effect size, with the expected attenuation in statistical significance (sensitivity analysis p = 5.76 × 10⁻⁷ vs. primary analysis p = 2.07 × 10⁻⁷). This consistency supports the robustness of the primary finding.

The sensitivity analysis using a residualized AAO measure yielded results that were highly consistent with those from our primary model. Among the original 126 top-ranked CpG sites that met the suggestive significance threshold, 122 (96.8%) remained associated with AAO at the same level (see Supplementary Table S6). Furthermore, the strongest association, probe cg05061107, maintained a comparable level of statistical significance (p = 3.02 × 10⁻⁹ in the sensitivity analysis versus p = 2.60 × 10⁻⁹ in the primary analysis).

Discussion

To our knowledge, this study is the first investigation of DNA methylation patterns of EOS and AAO in schizophrenia in the Chinese context. The recruitment of clinical samples from a well-established early psychosis intervention clinic, longitudinal follow-up for ascertaining the best-estimate end-point diagnosis, and inclusion of sex, age, smoking and cell type heterogeneity as covariates had further enhanced the robustness of our findings. While the extant literature has primarily focused on the common genetic variants associated with the AAO in schizophrenia, this study explored potential epigenetic signals for EOS patients and AAO. By leveraging cutting-edge methods, our preliminary findings provided insights to the methylation markers for EOS, and the explorative results using enrichment analysis further unveiled possible biological mechanisms for this subtype.

Several probes associated with EOS at the suggestive significance level were identified, and appeared to bear biological implications. Among the top-ranked DMPs, several genes for EOS identified in this study have been reported to be associated with the risk of developing schizophrenia previously, and these convergent findings might suggest potential overlaps in the epigenetic mechanisms underlying the disease status and AAO of schizophrenia. For example, the gene ORMDL1 (cg07577294, p = 3.25 × 10⁻⁶), a protein-coding gene involved in the ceramide metabolic process, has been reported as a risk gene for schizophrenia in a prior genetic study and a multi-omics study [42, 43]. Moreover, a variant, rs721689-T, near its paralog gene ORMDL3 was found to be associated with the AAO and the severity of asthma [44], though its biological functions remained unclear. The ANXA4 gene (cg08311172, p = 1.37 × 10⁻⁵) is a member of annexin family which has been found to play a role in immune responses [45, 46], and was also associated with rheumatoid arthritis in a prior EWAS [47]. A recent systematic review reported that ANXA4 was expressed differentially between schizophrenia patients and healthy controls from three different studies, though the directions of effect were inconsistent [48]. The gene C16orf87 (cg05862027, p = 1.83 × 10⁻⁵) was found to be a risk gene for schizophrenia in a previous EWAS conducted in the Chinese population [49]. ARHGAP15 (cg12657374, p = 1.26 × 10⁻⁵) is an important gene involved in the Rho GTPases and RAC2 GTPase cycle, which are key cellular processes in brain development. Recent studies have revealed the effect of ARHGAP15 on intellectual disability [50, 51]. Another gene in the same family, ARHGAP10, has been reported as a novel gene for schizophrenia risk recently [52]. Additionally, gene TRRAP has been implicated in a range of neurodevelopmental disorders, including developmental delay, autism spectrum disorder, and autosomal dominant non-syndromic hearing loss [53–55]. Importantly, a de novo variant of TRRAP was identified to link to very-early onset psychosis in a previous case report [56]; in our study, the CpG on this gene was found to be hypomethylated in EOS participants compared to AOS participants.

Another gene TGOLN2 (cg21513352, p = 3.06 × 10⁻⁵) is involved in the membrane trafficking. Prior research on the peripheral blood gene-expression profiles of schizophrenia patients suggested that TGOLN2 was downregulated in schizophrenia patients and their unaffected first-degree siblings, relative to healthy controls [57]. Additionally, probe cg27077219, located in the CpG island of LINC01357, was identified in a previous epigenetic study, as being significantly associated with suicidal ideation in patients with schizophrenia [58].

Besides the genes related to schizophrenia, AKAP8L (cg14588779, p = 7.54 × 10⁻⁷) was reported to be associated with education attainment [59], similar to the potential function found for cg02753742, which was mapped to the gene C4orf45, with a p-value of 8.92 × 10⁻⁶ [59]. This gene was associated with cognitive performance [60] and sex hormone-binding globulin measurement [61], and was identified as a shared gene for sleep phenotypes and Alzheimer’s disease [62]. HES7 encodes a protein crucial for nervous system development [63]. In mouse models, its expression is regulated by the Notch signaling pathway, and recent studies have suggested its potential involvement in the pathogenesis of autism spectrum disorder [64, 65].

Our enrichment analysis suggested that microRNAs (miRNAs) and proteoglycan in cancer may be associated with EOS. The miRNAs exhibit either oncogenic or tumor-suppressive functions in cancer, with nearly 70% of miRNAs being expressed in the central nervous system (CNS) [66]. Dysregulations of these miRNAs may contribute to schizophrenia by altering cellular pathways involved in gene expression, a possibility supported by large-scale genomic studies from the PGC [67]. Meanwhile, the extant literature has suggested that several miRNAs are implicated in both disorders, such as let-7 and miR-98 [68]. These miRNAs may influence schizophrenia risk via regulating genes responsible for brain development and function. Intriguingly, schizophrenia is characterized by genetic activity that promotes apoptosis and suppresses cellular proliferation, while cancer arises from uncontrolled cell proliferation. It has therefore been hypothesized that schizophrenia patients may have a reduced risk of developing cancer [69, 70], an interesting topic that warrants further investigation.

Proteoglycans in cancer exert diverse effects through multiple mechanisms: while some (e.g., perlecan) display pro- or anti-angiogenic activities, others (e.g., syndecans, glypicans) could directly regulate tumor growth by modulating critical signaling pathways [71]. Evidence suggests that proteoglycans are expressed differentially in the olfactory epithelium tissue of schizophrenia patients relative to healthy controls [72]. Our findings thus propose a novel mechanism by which differential proteoglycan expression may contribute to the impaired sensory function prevalent in schizophrenia [73].

Leukocyte transendothelial migration was also suggested to have impacts on AAO in schizophrenia, potentially via the putative mechanisms of neurovascular dysfunction and immune system interaction. Furthermore, multiple lines of evidence suggest that the kinase activity and mitotic cell cycle processes might contribute to EOS through the impacts on neurodevelopment, signal transduction, and cellular dysfunction of the brain [74–77]. Together, enrichment analyses of these GO and KEGG terms suggest biological pathways involved in the development of EOS.

Meanwhile, the analyses of AAO also provided some insights into the methylation patterns link to the onset age of schizophrenia. The SF1 (Splicing Factor 1) gene, which has been implicated by both DMP and DMR, encodes a protein involved in the regulation of RNA splicing which is a critical process for gene expression and cellular function. Recently, researchers found that the long non-coding RNA Gomafu (MIAT and RNCR2) was associated with alternative splicing through its interaction with SF1, and this process is involved in the etiology of schizophrenia [78]. Another notable finding is the identification of GPRC5C, a member of the type 3 G protein-coupled receptor (GPCR) family, as associated with schizophrenia AAO. Given the central role of GPCRs in neurotransmitter signaling and intracellular cascades, processes critical for cognitive and behavioral function [79], we provide growing evidence that atypical GPCRs may modulate neurodevelopmental trajectories and represent potential therapeutic targets. However, further mechanistic studies are required to delineate the precise contribution of GPRC5C to AAO variability. PCBP3, a member of poly(rC)-binding protein (PCBP) family, has been demonstrated to contribute to being a pathogenic factor in neurodegenerative diseases (NDs) through the iron dysregulation process [80]. Additionally, the gene ZNF573, with a predicted role in transcriptional regulation [63], may contribute to the AAO in schizophrenia through this important biological process.

Additionally, our findings on the composition of peripheral blood cells in the EOS and AOS groups support the role of immune functioning in schizophrenia. Specifically, to reduce heterogeneity, we selected cell composition which are different between the EOS and AOS, and revealed that only the NK cell remained significantly associated with EOS (p = 0.0014, which is less than the Bonferroni-corrected threshold of 0.0083), and was therefore included as a covariate in the analyses (Supplementary Figure S2). NK cells are recognized as a type of lymphocyte and an important component of the innate immune system [81]. Our findings aligns with genetic studies that highlight a specific role for lymphocytes (the class of immune cells that includes NK cells) in schizophrenia [82]. Besides, we observed a positive correlation between the CD4T/CD8T ratio and the continuous variable of AAO (p = 0.018) in our schizophrenia sample. The CD4T/CD8T ratio is a well-recognized biomarker of immune system function, with low or inverted ratios (normal range: 1.5–2.5) indicating altered immune function in various diseases [83–85].

Notably, the sensitivity analysis supported the robustness of the conventional cutoff value for defining early-onset schizophrenia (EOS), as comparisons using more stringent onset-age subgroups (<16 vs. >21 years) yielded consistent epigenetic patterns. Moreover, the high concordance of results from the model utilizing residualized AAO strongly suggests that the methylation signatures we identified are specifically related to the disease process underlying AAO and not confounded by the general aging effect.

To our knowledge, only two previous studies had examined DNA methylation patterns associated with EOS or AAO, with one additional study reporting the correlation between methylation variability and AAO. The earliest study, published in 2022, employed the HumanMethylation450 BeadChip array to analyze methylation differences in 138 schizophrenia patients of European ancestry (58 EOS vs. 80 AOS). It identified four intergenic methylation sites on chromosome 2 linked to EOS, with cg10392614 as the strongest signal [25]. However, no probe-level overlap was observed between their findings and ours. Such discrepancy may stem from methodological differences, including the broader EOS definition (AAO < 21 years) and narrower AAO range (18–21 years) in the EOS group in the previous study than ours, as well as the population differences (European vs. Asian ancestry). Another study in 2023, performed an EWAS for multiple phenotypes in 381 Australian patients with schizophrenia, using the EPIC (850 K) beadchip, and focused on continuous AAO (range: 11–51 years) [26]. Although this study reported clozapine usage rates and cognitive status, their associations with AAO were unknown. They identified 60 DMPs associated with AAO at a suggestive threshold (p < 6.72 × 10⁻⁵), with top signals including cg09788791, cg10120056, and cg25112641. A more recent study in 2024, investigated blood DNA methylation variance in schizophrenia across 1,036 patients and 954 controls of European ancestry using the 450 K BeadChip [27]. This work revealed that variably methylated positions (VMPs) in schizophrenia correlated with AAO (Pearson r = 0.31), suggesting that epigenetic variability may contribute to heterogeneity in schizophrenia onset. Compared to these previous studies, our work was the first to address this question in a Chinese / Asian population. Moreover, we expanded the analytical scope by including a broader AAO range (12–55 years); examining both dichotomous (EOS vs. AOS) and continuous AAO variables; and conducting sensitivity analyses incorporating different analytical approaches.

Nevertheless, our study has several notable limitations. First, given the relatively low prevalence of EOS among first-episode schizophrenia patients, we could only recruit a small sample size of EOS patients, limiting the ability to detect robust signals within the epigenome. Second, although we controlled for major demographic and lifestyle variables (age, sex, smoking score), other important factors such as childhood adversities, perinatal insults, drug abuse, and environmental stress were not measured in this cohort. From a confounding perspective, factors such as childhood trauma may independently contribute to both epigenetic modifications and earlier schizophrenia onset, thereby distorting the observed association (i.e., Supplementary Figure S3 - Condition 1), future epigenetic studies should consider adjusting for such factors. Regarding mediation, two distinct pathways warrant consideration. First, the onset of schizophrenia may subsequently alter methylation patterns through downstream biological consequences, such as illness-related changes in immune cell composition (i.e., Supplementary Figure S3 - Condition 2). Second, pre-existing methylation differences could themselves influence AAO, potentially through mechanisms like stress-responsive epigenetic programming of neurodevelopmental pathways (i.e., Supplementary Figure S3 - Condition 3).

Another notable limitation of our study is the potential confounding effect of antipsychotic medications on DNA methylation patterns. Antipsychotics, such as haloperidol, risperidone, and clozapine, have been shown to induce both global and site-specific DNA methylation changes. Such alterations may influence gene expression and potentially confound the results of methylation analyses. Nevertheless, we have attempted to circumvent this limitation by examining the proportion of clozapine prescription. Future studies should address this limitation by using the medication-naïve cohorts at the baseline, or incorporating stratified statistical modeling to rigorously account for variations in medication type, duration, and dosage. Furthermore, methylation profiles are highly tissue specific, our findings were derived from peripheral blood samples, which may not fully reflect the methylation status in the brain tissues. Therefore, caution should be taken when extrapolating our results to neural tissues. Future research may explore methylation patterns in brain tissues, to better understand the epigenetic mechanisms underlying schizophrenia. Lastly, DNA methylation is dynamic and reversible. Future research may adopt a longitudinal design to track the changes of methylated DNA patterns over time, from early childhood (pre-psychotic) and schizophrenia onset to established schizophrenia. It is plausible that EOS patients and AOS patients may differ in the progression and changes of DNA methylation in the critical period of onset of schizophrenia.

Collectively, our methylation findings suggest shared biological mechanisms between EOS and the genetic architecture of schizophrenia more broadly. The identified DMPs and subsequent functional enrichment analyses implicate dysregulation across several key pathways in EOS pathogenesis, including altered kinase activity and cell cycle control, immune system dysfunction, and neurodevelopmental processes.

Author contributions

NZ: Conceptualization, Formal analysis, Investigation, Visualization, Writing – original draft, Writing – review & editing. PBML: Validation, Writing – review & editing. YXZ: Writing – review & editing. KCYW: Writing – review & editing. TCKH: Investigation, Resources. HCS: Conceptualization, Supervision, Writing – review & editing, Funding acquisition. PCS: Conceptualization, Supervision, Funding acquisition, Writing – review & editing. CCYW: Methodology, Supervision, Validation, Writing – review & editing. SSYL: Conceptualization, Supervision, Funding acquisition, Writing – review & editing.

Funding

This work was funded by HKU Seed Fund for Basic Research for New Staff (202009185071) and the HKU Enhanced Start-up Fund for New Staff granted to Simon SY Lui; Suen Chi-Sun Foundation for their support for the Suen Chi-Sun Endowed Professorship in Clinical Science to Pak C Sham; This work was also partially supported by the Health and Medical Research Fund (HC So, grant number 07180376), the Young Collaborative Research Grant (C4003-23Y), the KIZ-CUHK Joint Laboratory of Bioresources and Molecular Research of Common Diseases, the Hong Kong Branch of the Chinese Academy of Sciences Center for Excellence in Animal Evolution and Genetics, and the Lo-Kwee Seong Biomedical Research Fund from The Chinese University of Hong Kong granted to Hon-Cheong So.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Competing interests

The authors declare no competing interests.

Ethics

We confirm that all methods were performed in accordance with the relevant guidelines and regulations.

Supplementary information

The online version contains supplementary material available at 10.1038/s41398-026-03869-y.