Hostname: page-component-74656f9884-npj9w Total loading time: 0 Render date: 2026-11-07T14:25:55.916Z Has data issue: false hasContentIssue false
  • English
  • Français

Recalibrating the epigenetic clock reveals divergent DNA methylation age signatures across schizophrenia, bipolar disorder and major depressive disorder

Published online by Cambridge University Press:  25 March 2026

Xiaohui Wu ,
Chenglin Lu ,
Huijun Kong ,
Renhao Chen ,
Zhongju Wang ,
Zhiying Deng ,
Wenbo Xiao ,
Zhanwang Huang ,
Qiyang Li  and
Diru Yao
...Show all authors
Xiaohui Wu
Affiliation:
Department of Medical Genetics, School of Basic Medical Sciences, Southern Medical University, Guangzhou, China Key Laboratory of Mental Health of the Ministry of Education, Southern Medical University, Guangzhou, China Guangdong-Hong Kong-Macao Greater Bay Area Center for Brain Science and Brain-Inspired Intelligence, Southern Medical University, Guangzhou, China Guangdong-Hong Kong Joint Laboratory for Psychiatric Disorders, Southern Medical University, Guangzhou, China Guangdong Province Key Laboratory of Psychiatric Disorders, Southern Medical University, Guangzhou, China Guangdong Basic Research Center of Excellence for Integrated Traditional and Western Medicine for Qingzhi Diseases, Southern Medical University, Guangzhou, China Guangdong Mental Health Center, Guangdong Provincial People’s Hospital (Guangdong Academy of Medical Sciences), Southern Medical University, Guangzhou, China
Chenglin Lu
Affiliation:
Department of Medical Genetics, School of Basic Medical Sciences, Southern Medical University, Guangzhou, China
Huijun Kong
Affiliation:
Department of Medical Genetics, School of Basic Medical Sciences, Southern Medical University, Guangzhou, China
Renhao Chen
Affiliation:
Department of Medical Genetics, School of Basic Medical Sciences, Southern Medical University, Guangzhou, China
Zhongju Wang
Affiliation:
Department of Medical Genetics, School of Basic Medical Sciences, Southern Medical University, Guangzhou, China
Zhiying Deng
Affiliation:
Department of Medical Genetics, School of Basic Medical Sciences, Southern Medical University, Guangzhou, China
Wenbo Xiao
Affiliation:
Department of Medical Genetics, School of Basic Medical Sciences, Southern Medical University, Guangzhou, China
Zhanwang Huang
Affiliation:
Department of Medical Genetics, School of Basic Medical Sciences, Southern Medical University, Guangzhou, China
Qiyang Li
Affiliation:
Department of Medical Genetics, School of Basic Medical Sciences, Southern Medical University, Guangzhou, China
Diru Yao
Affiliation:
Department of Medical Genetics, School of Basic Medical Sciences, Southern Medical University, Guangzhou, China
Xiaofan Ding
Affiliation:
Ministry of Education Frontiers Science Center for Precision Oncology, University of Macau, Taipa, Macao SAR, China
Cunyou Zhao*
Affiliation:
Department of Medical Genetics, School of Basic Medical Sciences, Southern Medical University, Guangzhou, China Key Laboratory of Mental Health of the Ministry of Education, Southern Medical University, Guangzhou, China Guangdong-Hong Kong-Macao Greater Bay Area Center for Brain Science and Brain-Inspired Intelligence, Southern Medical University, Guangzhou, China Guangdong-Hong Kong Joint Laboratory for Psychiatric Disorders, Southern Medical University, Guangzhou, China Guangdong Province Key Laboratory of Psychiatric Disorders, Southern Medical University, Guangzhou, China Guangdong Basic Research Center of Excellence for Integrated Traditional and Western Medicine for Qingzhi Diseases, Southern Medical University, Guangzhou, China Guangdong Mental Health Center, Guangdong Provincial People’s Hospital (Guangdong Academy of Medical Sciences), Southern Medical University, Guangzhou, China Experimental Education/Administration Center, School of Basic Medical Sciences, Southern Medical University, Guangzhou, China
*
Correspondence: Cunyou Zhao. Email: cyzhao@smu.edu.cn
Rights & Permissions [Opens in a new window]

Abstract

Background

Few studies have quantitatively characterised the shared and distinct features of the epigenetic age signature of schizophrenia, bipolar disorder and major depressive disorder.

Aims

To construct a multi-platform epigenetic clock tailored to human blood and brain tissues, and to characterise variations in epigenetic age acceleration across these three common psychiatric disorders.

Method

We integrated 31 publicly available DNA methylation data-sets generated on the platforms Illumina 27K, 450K and EPIC (850K) from patients with schizophrenia, bipolar disorder or major depressive disorder, and from matched controls. Using elastic net regression combined with sure independence screening, we developed the blood–brain clock and applied it to assess disorder-specific epigenetic age acceleration in blood and brain.

Results

The blood–brain clock achieved high accuracy across tissues and outperformed established predictors, particularly in brain samples. Epigenetic age acceleration was reduced in schizophrenia, increased in bipolar disorder and major depressive disorder and strongly elevated in Alzheimer’s disease (positive control). Alterations appeared earlier in blood than in brain. Meta-analysis confirmed that both reduced acceleration (schizophrenia) and increased acceleration (bipolar disorder, major depressive disorder, Alzheimer’s disease) were significantly associated with disease prevalence. Differential methylation analyses further revealed that the blood–brain clock probes captured disease-associated signals, with schizophrenia showing the greatest overlap with causal risk loci, and opposite methylation patterns distinguishing schizophrenia from bipolar disorder or major depressive disorder. A subset of blood DNA methylation probes enabled high-precision classification between schizophrenia and bipolar disorder or major depressive disorder.

Conclusions

This blood–brain clock reveals distinct patterns of epigenetic age acceleration across psychiatric disorders, reflecting disorder-specific and shared biological ageing signatures. The manifestation of these alterations in peripheral blood highlights its potential as a non-invasive biomarker for early detection, risk stratification and differential classification of schizophrenia, bipolar disorder and major depressive disorder.

Keywords

Neurodevelopmentepigenetic clockDNA methylationpsychiatric disordersdisease classification

Information

Type
Original Article
Information
The British Journal of Psychiatry , First View , pp. 1 - 9
Check for updates
Copyright
© The Author(s), 2026. Published by Cambridge University Press on behalf of Royal College of Psychiatrists

Psychiatric disorders remain among the most challenging diseases in medicine, with diagnoses based on symptoms and behaviour rather than direct biological measures. Despite decades of neurobiological research, causal mechanisms are unclear and debates continue as to whether these conditions represent distinct entities or overlapping syndromes. Reference Sullivan and Geschwind1 The neurodevelopmental model suggests that disruptions during brain development alter neurodevelopmental trajectories, Reference Sheffield, Karcher and Barch2–Reference Birnbaum and Weinberger4 whereas alternative hypotheses implicate premature ageing processes such as brain atrophy, telomere shortening and neurodegeneration. Reference Pousa, Souza, Melo, Correa, Mendonça and Simões-e-Silva5,Reference Van Assche, Morrens, Luyten, Van de Ven and Vandenbulcke6 It is evident that different psychiatric disorders manifest in disparate neurodevelopmental patterns. Reference Sheffield, Karcher and Barch2,Reference Patel, Parker, Shin, Howard, French and Thomopoulos3 Common psychiatric disorders such as schizophrenia, bipolar disorder and major depressive disorder show high comorbidity and overlapping symptoms, especially between bipolar disorder II and major depressive disorder, complicating diagnosis. Reference Anttila, Bulik-Sullivan, Finucane, Walters, Bras and Duncan7,Reference Phillips and Kupfer8 Increasing evidence suggests that epigenetic regulation, particular DNA methylation (DNAm), plays a critical role in these disorders by shaping neurodevelopmental programmes and long-term brain function. Reference Cosgrove and Suppes9–Reference Hemager, Plessen, Thorup, Christiani, Ellersgaard and Spang11 DNAm age, often referred to as the ‘epigenetic clock’, is a robust molecular estimator of biological age that closely tracks chronological age; Reference Nie, Li, Li, Yan, Zhang and Li12,Reference Petkovich, Podolskiy, Lobanov, Lee, Miller and Gladyshev13 deviations, known as epigenetic age acceleration, have been linked to disease risk. Such acceleration has been reported in both neurodevelopmental Reference Steg, Shireby, Imm, Davies, Franklin and Flynn14,Reference Shireby, Davies, Francis, Burrage, Walker and Neilson15 and psychiatric disorders, Reference Cole, Marioni, Harris and Deary16–Reference Han, Aghajani, Clark, Chan, Hattab and Shabalin21 including schizophrenia, bipolar disorder and major depressive disorder, although systematic cross-disorder comparisons remain limited. Several predictors have been developed to estimate DNAm age. Among these, Horvath’s ’multi-tissue predictor’, built on 353 cytosine guanine (CpG) sites across diverse tissues, Reference Horvath22 remains widely used but shows reduced accuracy in brain samples Reference Shireby, Davies, Francis, Burrage, Walker and Neilson15 and limited compatibility with the more recent Illumina EPIC (850K) array. Specialised clocks, such as the cortex-specific cortical clock, Reference Shireby, Davies, Francis, Burrage, Walker and Neilson15 Hannum’s multiple predictor Reference Zhang, Vallerga, Walker, Lin, Henders and Montgomery23 and the fetal brain clock, Reference Steg, Shireby, Imm, Davies, Franklin and Flynn14 show improved accuracy within certain tissues; however, no model to date has been optimised jointly for blood and brain. This gap limits cross-tissue studies, even though peripheral DNAm alterations are increasingly recognised as surrogates for brain biology. Reference Magwai, Shangase, Oginga, Chiliza, Mpofana and Xulu24

Here we developed a blood–brain clock, a DNAm-based age estimator trained on both blood and brain data, compatible with all major Illumina arrays and accurate across the lifespan. We applied this clock to large cohorts of schizophrenia, bipolar disorder and major depressive disorder, quantifying disorder-specific patterns of epigenetic age acceleration. We further examined differential methylation of clock CpG probes, providing insights into shared and distinct molecular signatures and their potential for diagnosis and classification of psychiatric disorders.

Method

We integrated 31 publicly available DNAm data-sets generated on the array platforms Illumina Infinium 27k, 450K and 850K Human DNA methylation Beadchip (Supplementary Table 1 available at https://doi.org/10.1192/bjp.2025.10508). Using a combination of elastic net regression Reference Horvath22,Reference Friedman, Hastie and Tibshirani25 and sure independence screening (SIS), Reference Saldana and Feng26 we developed a new epigenetic clock, the blood–brain clock, to quantify DNAm age in 2422 patients with schizophrenia, bipolar disorder, major depressive disorder or Alzheimer’s disease, and in 2480 non-psychiatric controls (R version 4.2.0, The R Project for Statistical Computing, Vienna, Austria; https://posit.co/download/rstudio-desktop/). To evaluate group differences, we calculated two measures of epigenetic age acceleration: DNAm chronological age (accelerationDiff) and residuals from linear regression of DNAm age on chronological age (accelerationResidual). The Kruskal–Wallis test was used to assess significance between disease groups and controls. A meta-analysis was then performed to estimate the impact of increases or decreases in epigenetic age acceleration on disease risk (R version 4.2.0). Summary data Mendelian randomisation (SMR) was used to test causal relationships between methylation sites from methylation quantitative trait loci (mQTL) data and psychiatric phenotypes. Differential methylation analysis was performed using the limma package in RStudio (R version 4.2.0). 33 Receiver operating characteristic (ROC) analysis was conducted with the multipleROC package, and plots were generated using ggplot2 (R version 4.2.0). Functional enrichment analyses were conducted with the Enrichr online tool (Ma’ayan Lab at the Center for Bioinformatics, Icahn School of Medicine at Mount Sinai, New York, USA; https://maayanlab.cloud/Enrichr/) Reference Kuleshov, Jones, Rouillard, Fernandez, Duan and Wang27,Reference Xie, Bailey, Kuleshov, Clarke, Evangelista and Jenkins28 and Cytoscape 3.10.3 Reference Bindea, Mlecnik, Hackl, Charoentong, Tosolini and Kirilovsky29 (for Windows, University of California, San Diego, USA; https://cytoscape.org/download.html) to identify enriched Gene Ontology terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways. Additional details are provided in the Supplementary materials.

Results

Construction and evaluation of the blood–brain clock across blood and brain tissues

To construct a DNAm age predictor applicable to both blood and brain, we assembled a large, publicly available DNAm data-set measured on the Illumina 27K, 450K and 850K arrays. The data-set included 3935 healthy individuals aged 0–108 years (Supplementary Table 2). The relatively balanced representation of blood (n = 1982) and brain (n = 1953) tissue helped minimise tissue-specific bias. We randomly assigned 75% of samples to the training set and 25% to the testing set, with 6 additional independent data-sets for external validation.

Using the training set, we developed an epigenetic age predictor, the blood–brain clock, constructed with the elastic net algorithm in combination with SIS. The final model incorporated 203 DNAm probes with non-zero coefficients, whose weighted sum provides the blood–brain clock age estimate (Supplementary Tables 2 and 3). Among these probes, 177 were novel (Fig. 1(a)) whereas 22 overlapped with Horvath’s multi-tissue predictor (353 probes), Reference Horvath22 7 with Hannum’s predictor (514 probes) Reference Zhang, Vallerga, Walker, Lin, Henders and Montgomery23 and 4 with the cortical clock (347 probes). Reference Shireby, Davies, Francis, Burrage, Walker and Neilson15 Genes located near the blood–brain clock probes were enriched in pathways related to nervous system development and neural signalling (Fig. 1(b) and Supplementary Table 4). Moreover, Epigenome-Wide Association Study trait enrichment revealed significant associations with ageing, Down syndrome, neurodevelopmental features and congenital anomalies (Fig. 1(c) and Supplementary Table 5), underscoring the relevance of these probes to neuronal function and disease.

Fig. 1

Performance assessment of the blood–brain clock. (a) Venn diagram comparing cytosine guanine probes used in the blood–brain clock with three established epigenetic clocks (multi-tissue predictor, multiple predictor and cortical clock). (b) Functional enrichment analysis of the 203 blood–brain clock probe-annotated genes, including Gene Ontology biological process (GO-BP), cellular component (GO-CC) and molecular function (GO-MF), and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway. (c) Epigenome-wide association study trait enrichment of the 203 blood–brain clock probe-annotated genes. (d, e) Scatter plots of chronological age (x axis) versus DNA methylation (DNAm) age (y axis) estimated by the blood–brain clock in blood (red circle) and brain (blue triangle) tissues from the training cohort (d) and testing cohort (e) of healthy controls. The dashed line indicates equality between DNAm and chronological age; solid lines indicate regression fits. Model performance metrics (prediction error (err), mean absolute error (MAE), root mean square error (RMSE) and Pearson’s correlation (R) are shown). (f, g) Scatter plots comparing chronological age and DNAm age in (f) blood and (g) brain from independent healthy controls, assessed using the blood–brain clock, multiple predictor, cortical clock and multi-tissue predictor.

We evaluated prediction accuracy using two complementary parameters: age correlation (cor), defined as the Pearson correlation coefficient between predicted DNAm age and chronological age, and error (err), defined as the median absolute difference between the two. Reference Horvath22 To further benchmark performance, we also calculated mean absolute error and root mean square error. We applied the blood–brain clock to healthy blood (n = 1982) and brain (n = 1953) samples and demonstrated robust performance across both, achieving high accuracy in both the training set (err 2.4 years, cor 0.99 years; Fig. 1(d)) and testing set (err 2.6 years, cor 0.99 years; Fig. 1(e)). Moreover, the model maintained stable predictive accuracy across the entire age spectrum (0–20, 20–40, 40–60 and 60–100 years; Supplementary Fig. 1). To further validate model robustness and cross-platform compatibility, we evaluated the blood–brain clock using 6 independent data-sets comprising 637 blood and 155 brain samples profiled on the Illumina 450K and 850K arrays (Supplementary Table 6). Its accuracy was then compared with that of three established predictors: (a) Horvath’s multi-tissue predictor, Reference Horvath22 which performs broadly on 27K and 450K arrays but shows reduced utility on 850K; (b) Hannum’s multiple predictor, Reference Zhang, Vallerga, Walker, Lin, Henders and Montgomery23 developed from blood and saliva and compatible with the 850K array but less accurate for brain; and (c) the cortical clock, Reference Shireby, Davies, Francis, Burrage, Walker and Neilson15 optimised for cortical data but less reliable for blood. Although all predictors exhibited strong correlations with chronological age, the blood–brain clock consistently achieved the highest overall stability and accuracy across tissues, with particularly strong performance for brain samples (Fig. 1(f, g)).

Together, these findings establish the blood–brain clock as a robust and versatile predictor of DNAm age across both blood and brain, making it a valuable tool for epigenetic age profiling in neurodevelopmental and neuropsychiatric research.

Divergent blood–brain DNAm age acceleration among schizophrenia, bipolar disorder and major depressive disorder

We next applied the blood–brain clock to quantification of DNAm age in schizophrenia (n = 1000 blood, 370 brain), bipolar disorder (n = 38 blood, 101 brain) and major depressive disorder (n = 98 blood, 52 brain), and in Alzheimer’s disease (n = 75 blood, 717 brain), to characterise disorder-specific epigenetic age acceleration patterns across blood and brain tissues. Alzheimer’s disease was included as a positive control owing to its well-established association with epigenetic age acceleration. For each disorder, age-matched healthy controls were randomly selected to match the chronological age distribution of cases (Supplementary Table 7).

Across all cohorts, DNAm age from blood (Fig. 2(a1–d1), left) and brain (Fig. 2(a2–d2) left) showed strong correlation with chronological age and consistently low error. We then evaluated epigenetic age acceleration using accelerationDiff (DNAm age–chronological age) and accelerationResidual (residuals from regressing DNAm age on chronological age). In schizophrenia, both blood (accelerationDiff: P = 2.4 × 10−5; accelerationResidual: P = 9 × 10−6; Fig. 2(a1), right) and brain (P = 1.7 × 10−6 and P = 2.5 × 10−6; Fig. 2(a2), right) exhibited significantly decreased epigenetic age acceleration relative to controls, consistent with prior reports. Reference Wu, Ye, Wang and Zhao17 In contrast, bipolar disorder showed significantly increased epigenetic age acceleration in blood (P = 2.7 × 10−4 and P = 1.9 × 10−4; Fig. 2(b1), right), and a non-significant increase in brain (P = 0.13 and P = 0.066; Fig. 2(b2), right). Major depressive disorder patients displayed significantly elevated epigenetic age acceleration in both blood (P = 0.045, P = 0.031; Fig. 2(c1), right) and brain (P = 0.011 and P = 0.008; Fig. 2(c2), right). As expected, Alzheimer’s disease patients showed pronounced acceleration in both blood (P = 1.4 × 10−5 and P = 1.3× 10−3; Fig. 2(d1), right) and brain (P = 5.3 × 10−8 and P = 2 × 10−9; Fig. 2(d2), right), consistent with previous studies. Reference Levine, Lu, Bennett and Horvath30,Reference Pellegrini, Pirazzini, Sala, Sambati, Yusipov and Kalyakulina31 Moreover, the observed age accelerations remained robust for schizophrenia and Alzheimer’s disease in both blood and brain, and for major depressive disorder in brain, even after adjusting for cell type composition (Supplementary Table 7). Comprehensive statistical metrics and power analyses for sample size, gender, age and DNAm age and epigenetic age acceleration across all comparisons are provided in Supplementary Table 8.

Fig. 2

Prediction and comparison of epigenetic age in four neuropsychiatric disorders versus controls using the blood–brain clock. Scatter plots show chronological age (x axis) versus DNA methylation (DNAm) age (y axis) in blood (a1–d1, left) and brain (a2–d2, left), alongside comparison of epigenetic age acceleration (accelerationDiff) and residuals from linear regression of DNAm age on chronological age (accelerationResidual) (a1–d1, right; a2–d2, right). Disorders include schizophrenia (SZ; a1, a2), bipolar disorder (BD; b1, b2), major depressive disorder (MD; c1, c2) and Alzheimer’s disease (AD; d1, d2). Con, controls. In scatter plots, the dashed line represents equality between DNAm and chronological age; solid lines indicate regression fits. Model performance metrics (prediction error (err), correlation coefficient (cor) and Pearson’s P) are displayed in each panel. The Kruskal–Wallis test was used to evaluate differences in accelerationDiff and accelerationResidual between disease groups and controls.

To verify these findings, we conducted a meta-analysis of epigenetic age acceleration and estimated prevalence risk, defining thresholds exclusively from the combined control groups to ensure a disease-independent reference. Specifically, the first quartile (blood, −2.77; brain, −2.98) of control accelerationDiff was applied to schizophrenia to represent reduced acceleration, whereas the third quartile (blood, 3.01; brain, 3.20) was used for bipolar disorder, major depressive disorder and Alzheimer’s disease to indicate increased acceleration. In schizophrenia, blood accelerationDiff values below −2.77 were associated with a 1.93-fold higher prevalence (odds ratio 1.93, P < 0.0001, 95% CI 1.55 to 2.39; Fig. 3(a), left), whereas brain values below −2.98 increased prevalence 2.30-fold (odds ratio 2.30, P < 0.0001, 95% CI 1.62 to 3.28; Fig. 3(a), right). In bipolar disorder, blood accelerationDiff values above 3.01 corresponded to a 9.93-fold higher prevalence (odds ratio 9.93, P = 0.0454, 95% CI 1.05 to 93.97; Fig. 3(b), left), whereas brain accelerationDiff values exceeding 3.20 showed no significant association (odds ratio 1.36, P = 0.3464, 95% CI 0.72 to 2.57; Fig. 3(b), right). In major depressive disorder, blood accelerationDiff greater than 3.01 corresponded to a marginal 1.87-fold increase in prevalence (odds ratio 1.87, P = 0.0952, 95% CI 0.90 to 3.91; Fig. 3(c), left), whereas brain values above 3.20 were associated with a 6.94-fold higher prevalence (odds ratio 6.94, P = 0.0049, 95% CI 1.80 to 26.81; Fig. 3(c) right). In Alzheimer’s disease, blood accelerationDiff exceeding 3.01 increased prevalence 2.42-fold (odds ratio 2.42, P < 0.0001, 95% CI 1.07 to 5.43; Fig. 3(d), left), whereas brain values above 3.20 doubled prevalence (odds ratio 2.13, P < 0.0001, 95% CI 1.67 to 2.72; Fig. 3(d), right).

Fig. 3

Meta-analyses of epigenetic age acceleration and disorder prevalence. Forest plots show associations between epigenetic age acceleration and prevalence of (a) schizophrenia (SZ), (b) bipolar disorder (BD), (c) major depressive disorder (MD) and (d) Alzheimer’s disease (AD; positive control). Thresholds were defined as the first quartile of the combined control epigenetic age acceleration (accelerationDiff) for SZ (–2.77 in blood, –2.98 in brain) and the third quartile for BD, MD and AD (3.01 in blood, 3.20 in brain). For each disorder, events represent subjects with accelerationDiff values below the first quartile (SZ) or above the third quartile (BD, MD, AD) of the control distribution. Odds ratios (ORs) with 95% confidence intervals are shown for individual data-sets and the overall pooled effect. Tests for heterogeneity were assessed using I 2 and τ 2 statistics.

Given the age dependence of epigenetic age acceleration, analyses were stratified by age into youth (18–40 years) and middle-aged/elderly (40–90 years) cohorts. In schizophrenia, reduced acceleration was consistently observed for blood across all ages but reached significance in brain only after age 40 (Fig. 4(a)). In bipolar disorder, blood acceleration was elevated across age groups whereas brain acceleration became significant only after age 40 (Fig. 4(b)). In major depressive disorder, increased blood acceleration was evident before age 40, whereas brain acceleration became significant only after age 40 (Fig. 4(c)). Patterns in Alzheimer’s disease could not be assessed owing to the restricted patient age range (60–90 years).

Fig. 4

Age-stratified changes in epigenetic age acceleration across three disorders. Epigenetic age acceleration (accelerationDiff) in blood (left panels) and brain (right panels) is shown for (a) schizophrenia (SZ) (yellow [upper] line), (b) bipolar disorder (BD) (purple [upper] line) and (c) major depressive disorder (MD) (blue [upper] line) compared with age-matched controls (green [lower] line) across different chronological age categories. Solid dots indicate group means, with error bars representing standard deviations. Significant case–control differences based on Kruskal–Wallis tests are marked by arrows (↑ increase, ↓ decrease), with significance thresholds indicated as ***P < 0.001, **P < 0.01, *P < 0.05.

Taken together, schizophrenia is characterised by reduced acceleration in both tissues, bipolar disorder by modest blood-specific increases, major depressive disorder by brain-dominant increases and Alzheimer’s disease by strong acceleration in both blood and brain. Across the three psychiatric disorders, alterations consistently appeared earlier in blood than in brain, suggesting that blood may serve as a sensitive peripheral marker of disease-related epigenomic changes.

DNAm probes from the blood–brain clock distinguish schizophrenia, bipolar disorder and major depressive disorderr

To investigate shared biological ageing mechanisms embedded in the blood–brain clock and their potential causal roles in psychiatric disorders, we performed SMR analysis using Brain-mMeta mQTL data from European populations (1000 Genomes Project reference) and Genome-Wide Association Study (GWAS) outcomes for schizophrenia, bipolar disorder and major depressive disorder. Reference Zhu, Zhang, Hu, Bakshi, Robinson and Powell32 We identified 29 probes in the blood–brain clock causally associated with schizophrenia (15 in females, 21 in males), 19 associated with bipolar disorder and 7 associated with major depressive disorder (Fig. 5(a), Supplementary Fig. 1 and Supplementary Table 9). Notably, cg26825412, located near SOX18, was the only probe shared across all 3 disorders. Functional enrichment analysis of the genes corresponding to these 42 probes revealed significant associations with neurotrophin production, neuropeptide activity, neurotransmitter receptor internalisation, ceramide metabolic proces, and embryonic development (Fig. 5(b) and Supplementary Table 10), underscoring their neurodevelopmental relevance.

Fig. 5

Changes in methylation levels of blood–brain clock DNA methylation (DNAm) probes in three disorders. (a) Venn diagram of summary data Mendelian randomisation (SMR)-identified probes associated with schizophrenia (SZ), bipolar disorder (BD) and major depressive disorder (MD). (b) Gene Ontology (GO) enrichment analysis of genes corresponding to SMR identified 42 probes, highlighting pathways related to neurodevelopment and neuronal signalling. (c) Venn diagrams showing overlap between SMR probes and disease-associated differential methylation probes (DMPs) in blood and brain for SZ, BD and MD. (d) Blood-based classification: heatmap of representative DMPs (left) and receiver operating characteristic (ROC) analyses (right) demonstrating high accuracy in distinguishing SZ from BD (area under the receiver operating characteristic curve (AUC) = 0.963, 95% CI = 0.929–0.986), and SZ from MD (AUC = 0.842, 95% CI = 0.765–0.894). Sensitivity (sens), specificity (spec), negative predictive value (NPV), positive predictive value (PPV) and optimal fitted value (Ir.eta) are indicated. (e) Brain-based classification: heatmap of representative DMPs (left) and ROC analyses (right) showing robust performance in differentiating SZ from BD (AUC = 0.979, 95% CI = 0.953–0.992), and SZ from MD (AUC = 0.941, 95% CI = 0.892–0.969).

To specifically highlight disease-associated methylation alterations within the blood–brain clock probes, we next performed differential methylation probe (DMP) analysis between patients and controls using the data-sets shown in Fig. 2, adjusting for batch effects, gender and cell type composition (Supplementary Table 11). In schizophrenia, 67 blood DMPs and 66 brain DMPs overlapped with the blood–brain clock (Fig. 5(c), left), of which 9 blood DMPs and 10 brain DMPs also overlapped with schizophrenia risk-associated probes identified by SMR. In bipolar disorder, ten blood DMPs and seven brain DMPs overlapped with the blood–brain clock, including two blood DMPs that corresponded to bipolar disorder risk-associated loci (Fig. 5(c), middle). In major depressive disorder, 12 blood DMPs and 4 brain DMPs overlapped, with 1 blood DMP identified as major depressive disorder risk-associated (Fig. 5(c), right). Functional enrichment analysis of the 23 overlapping schizophrenia-associated DMPs between blood and brain revealed significant enrichments in terms related to axonal extension, ErbB signalling pathways, dopaminergic neuron differentiation, neural crest cell development and immune cell proliferation (Supplementary Fig. 3), underscoring the neurodevelopmental and neuroimmune relevance of these shared methylation signatures.

We next investigated whether disease-associated methylation patterns could distinguish between the disorders. In blood, 11 disease-associated DMPs (9 from schizophrenia, 2 from bipolar disorder and 1 from major depressive disorder, including one overlapping probe) were analysed using logistic regression with automatic stepwise selection, followed by multiple ROC curve analysis (Fig. 5(d)). This approach demonstrated that specific subsets of blood DMPs enabled high-precision differentiation between schizophrenia and bipolar disorder (eight probes: cg09835085, cg1129964, cg14315198, cg22736354, cg05836154, cg11285843, cg07924575 and cg1899097; showing opposite alterations in schizophrenia versus bipolar disorder), and between schizophrenia and major depressive disorder (seven probes: cg20935106, cg20951444, cg03547797, cg05836154, cg11285843, cg07924575 and cg18919097; showing opposite alterations in schizophrenia versus major depressive disorder).

In brain tissue, ten disease-associated DMPs (all from schizophrenia) were analysed (Fig. 5(e)). Of these, six probes (cg22736354, cg05404787, cg10301967, cg04739485, cg20935106 and cg24871743) differentiated schizophrenia from bipolar disorder, showing opposite alterations, whereas eight probes (cg01731685, cg14315198, cg20951444, cg26377677, cg22736354, cg05404787, cg10301967 and cg20935106) facilitated accurate classification between schizophrenia and bipolar disorder, again showing opposite alterations.

Taken together, these findings suggest that methylation levels of disease-associated DNAm probes in blood, particularly those incorporated in the blood–brain clock, can robustly differentiate schizophrenia from bipolar disorder or major depressive disorder, while also highlighting shared molecular pathways implicated in psychiatric disease aetiology.

Discussion

The pathogenic mechanisms of psychiatric disorders remain poorly understood; uncertainties persist regarding their nosological boundaries – whether they represent distinct entities, overlapping syndromes or variable manifestations of shared genetic risks. 33 Exploring neurodevelopmental signatures through genetic and epigenetic factors is therefore critical for the advancement of psychiatric nosology, improving risk prediction and identifying therapeutic targets. Although large-scale genomic and transcriptomic studies have provided important insights, 33–36 systematic analyses of genome-wide DNA methylation, particularly quantitative comparison of epigenetic age across disorders, have been lacking. In this study we developed the blood–brain clock, an epigenetic clock optimised for blood and brain tissues, enabling robust assessment of epigenetic age acceleration in schizophrenia, bipolar disorder and major depressive disorder.

Our findings revealed distinct trends in epigenetic age acceleration across disorders, reflecting divergent epigenetic age signatures. Schizophrenia was consistently characterised by reduced acceleration in both blood and brain, in line with previous studies suggesting delayed neurodevelopmental processes. Reference Andreazza and Nierenberg37–Reference Walker, Curtis, Shaw and Murray39 By contrast, bipolar disorder and major depressive disorder showed increased acceleration, with bipolar disorder exhibiting more prominent changes in blood and major depressive disorder showing stronger effects in brain. These results parallel neuroanatomical observations: cognitive and structural abnormalities are evident at the onset of schizophrenia but typically emerge later in bipolar disorder and major depressive disorder. Reference Demjaha, MacCabe and Murray40,Reference Gotlib and Ordaz41 Moreover, the similarity between bipolar disorder and major depressive disorder in epigenetic age profiles is consistent with their shared depressive symptomatology, and with genetic and imaging evidence of overlapping biological substrates. Reference Patel, Parker, Shin, Howard, French and Thomopoulos3,Reference Gandal, Haney, Parikshak, Leppa, Ramaswami and Hartl42,Reference Park, Kebets, Larivière, Hettwer, Paquola and van Rooij43 Large-scale Psychiatric Genomics Consortium cross-disorder analyses have shown that bipolar disorder and major depressive disorder share substantial proportions of common genetic risk loci and polygenic architecture, particularly involving genes implicated in mood regulation, synaptic signalling and circadian pathways. Reference Stahl, Breen, Forstner, McQuillin, Ripke and Trubetskoy44 Neuroimaging studies further reveal convergent alterations in frontolimbic circuitry and cortical thickness abnormalities across both disorders, paralleling shared depressive symptom dimensions. Reference Fears, Schür, Sjouwerman, Service, Araya and Araya45,Reference Warden, Selimbeyoglu, Mirzabekov, Lo, Thompson and Kim46 Clinically, diagnostic ambiguity between bipolar disorder II and recurrent major depression has long been recognised, reflecting overlapping affective and cognitive features that may confound diagnostic boundaries. Reference Phillips and Kupfer8 These findings collectively support the interpretation that similar epigenetic ageing profiles in bipolar disorder and major depressive disorder reflect shared neurobiological mechanisms. Nonetheless, the heterogeneity within bipolar disorder, including differences between bipolar disorders I and II, warrants caution in interpretation and emphasises the need for future subtype-specific analyses. Importantly, alterations in all three disorders emerged earlier in blood than in brain, suggesting that peripheral DNAm signatures may serve as sensitive, minimally invasive biomarkers for early detection.

The meta-analysis revealed that deviations in epigenetic age acceleration were not only statistically significant but also predictive of disease prevalence, highlighting the potential of DNAm age measures as tools for epidemiological and clinical risk stratification. At the molecular level, disease-associated DMPs within blood–brain clock probes overlapped with SMR-identified causal loci, particularly in schizophrenia. Most shared DMPs showed opposite methylation patterns between schizophrenia and bipolar disorder or major depressive disorder, but similar patterns between bipolar disorder and major depressive disorder, indicating distinct yet related epigenetic signatures. Overlapping blood–brain DMPs probably represent systemic ageing markers whereas non-overlapping DMPs capture tissue-specific effects. Given their cross-tissue relevance, these loci may reflect shared biological ageing mechanisms with potential causal roles in psychiatric risk. Functional enrichment of shared DMPs implicates axon extension, ErbB/ERK signalling, dopaminergic neuron differentiation, neural crest development and immune regulation, suggesting coordinated neurodevelopmental and neuroimmune involvement. Finally, logistic regression and ROC analyses demonstrated that a small subset of blood DMPs achieved high-precision differentiation among schizophrenia, bipolar disorder and major depressive disorder, highlighting their promise as diagnostic biomarkers.

Our findings are consistent with large-scale neuroimaging studies showing shared and distinct brain alterations across schizophrenia, bipolar disorder and major depressive disorder, Reference Patel, Parker, Shin, Howard, French and Thomopoulos3 suggesting that peripheral DNAm age acceleration may reflect underlying neurobiological differences. However, several limitations should be acknowledged. First, the reliance on post-mortem and cross-sectional data limits inference on temporal dynamics, underscoring the need for longitudinal and living-tissue validation. Second, although the blood–brain clock was trained across the lifespan, the absence of stratified age modelling may restrict interpretability across developmental stages. Third, modest sample sizes – particularly for bipolar disorder blood data-sets – and minor age mismatches between tissues may have reduced robustness. Additionally, merging data-sets across Illumina platforms reduced methylome coverage to ∼24 000 CpGs; however, this trade-off increased sample size and improved cross-platform generalisability. Finally, the meta-analysis of epigenetic age acceleration was performed without inclusion of covariates, and used quartile-based thresholds to emphasise biologically meaningful extremes. Although this approach facilitates consistent cross-disease interpretation, it may also introduce residual confounding and limit sensitivity to intermediate effects.

In conclusion, this study provides a large-scale, cross-disorder epigenetic age analysis in schizophrenia, bipolar disorder and major depressive disorder. The blood–brain clock revealed disorder-specific alterations in epigenetic age in both blood and brain. These findings advance our understanding of the divergent epigenetic age signatures underlying psychiatric disorders, and highlight the utility of peripheral epigenetic age measures as biomarkers for early diagnosis, risk stratification and differential classification.

Acknowledgements

We thank the many researchers who made available their DNA methylation data-sets publicly and responded to our email requests. This study would not have been possible without valuable data from the NCBI Gene Expression Omnibus database. We thank the National Natural Science Foundation of China (grant nos 32000419, 82201655 and 82471527), the Guangdong Basic and Applied Basic Research Foundation (grant nos 2024A1515012896 and 2023A1515012374) and Guangdong–Hong Kong Joint Laboratory for Psychiatric Disorders (no. 2023B1212120004) for providing financial support.

Supplementary material

The supplementary material is available online at https://doi.org/10.1192/bjp.2025.10508

Author contributions

X.W. and C.Z. conceived and designed the experiments; X.W., C.L., H.K., Z.D., R.C., Z.H., Q.L., D.Y., W.X., Z.W., X.D. and C.Z. analysed the data; X.W., C.L. and C.Z. wrote the paper.

Declaration of interest

None.

Analytic code availability

The blood–brain clock is available online at https://github.com/melon908/The-Blood_Brain-Clock-Epigenetic-Age-Predictor-

References

Sullivan, PF, Geschwind, DH. Defining the genetic, genomic, cellular, and diagnostic architectures of psychiatric disorders. Cell 2019; 177: 162–83.10.1016/j.cell.2019.01.015CrossRefGoogle ScholarPubMed
Sheffield, JM, Karcher, NR, Barch, DM. Cognitive deficits in psychotic disorders: a lifespan perspective. Neuropsychol Rev 2018; 28: 509–33.10.1007/s11065-018-9388-2CrossRefGoogle ScholarPubMed
Patel, Y, Parker, N, Shin, J, Howard, D, French, L, Thomopoulos, SI, et al. Virtual histology of cortical thickness and shared neurobiology in 6 psychiatric disorders. JAMA Psychiatry 2021; 78: 47–63.Google ScholarPubMed
Birnbaum, R, Weinberger, DR. Genetic insights into the neurodevelopmental origins of schizophrenia. Nat Rev Neurosci 2017; 18: 727–40.10.1038/nrn.2017.125CrossRefGoogle ScholarPubMed
Pousa, PA, Souza, RM, Melo, PHM, Correa, BHM, Mendonça, TSC, Simões-e-Silva, AC, et al. Telomere shortening and psychiatric disorders: a systematic review. Cells 2021; 10: 1423.10.3390/cells10061423CrossRefGoogle ScholarPubMed
Van Assche, L, Morrens, M, Luyten, P, Van de Ven, L, Vandenbulcke, M. The neuropsychology and neurobiology of late-onset schizophrenia and very-late-onset schizophrenia-like psychosis: a critical review. Neurosci Biobehav Rev 2017; 83: 604–21.10.1016/j.neubiorev.2017.08.024CrossRefGoogle ScholarPubMed
Anttila, V, Bulik-Sullivan, B, Finucane, HK, Walters, RK, Bras, J, Duncan, L, et al. Analysis of shared heritability in common disorders of the brain. Science 2018; 360: eaap8757.Google ScholarPubMed
Phillips, ML, Kupfer, DJ. Bipolar disorder diagnosis: challenges and future directions. Lancet 2013; 381: 1663–71.10.1016/S0140-6736(13)60989-7CrossRefGoogle ScholarPubMed
Cosgrove, VE, Suppes, T. Informing DSM-5: biological boundaries between bipolar I disorder, schizoaffective disorder, and schizophrenia. BMC Med 2013; 11: 127.10.1186/1741-7015-11-127CrossRefGoogle Scholar
Pai, S, Li, P, Killinger, B, Marshall, L, Jia, P, Liao, J, et al. Differential methylation of enhancer at IGF2 is associated with abnormal dopamine synthesis in major psychosis. Nat Commun 2019; 10: 2046.10.1038/s41467-019-09786-7CrossRefGoogle ScholarPubMed
Hemager, N, Plessen, KJ, Thorup, A, Christiani, C, Ellersgaard, D, Spang, KS, et al. Assessment of neurocognitive functions in 7-year-old children at familial high risk for schizophrenia or bipolar disorder: the danish high risk and resilience study VIA 7. JAMA Psychiatry 2018; 75: 844–52.10.1001/jamapsychiatry.2018.1415CrossRefGoogle ScholarPubMed
Nie, C, Li, Y, Li, R, Yan, Y, Zhang, D, Li, T, et al. Distinct biological ages of organs and systems identified from a multi-omics study. Cell Rep 2022; 38: 110459.10.1016/j.celrep.2022.110459CrossRefGoogle ScholarPubMed
Petkovich, DA, Podolskiy, DI, Lobanov, AV, Lee, SG, Miller, RA, Gladyshev, VN. Using DNA methylation profiling to evaluate biological age and longevity interventions. Cell Metab 2017; 25: 954–60.e6.10.1016/j.cmet.2017.03.016CrossRefGoogle ScholarPubMed
Steg, LC, Shireby, GL, Imm, J, Davies, JP, Franklin, A, Flynn, R, et al. Novel epigenetic clock for fetal brain development predicts prenatal age for cellular stem cell models and derived neurons. Mol Brain 2021; 14: 98.10.1186/s13041-021-00810-wCrossRefGoogle ScholarPubMed
Shireby, GL, Davies, JP, Francis, PT, Burrage, J, Walker, EM, Neilson, GWA, et al. Recalibrating the epigenetic clock: implications for assessing biological age in the human cortex. Brain J Neurol 2020; 143: 3763–75.10.1093/brain/awaa334CrossRefGoogle ScholarPubMed
Cole, JH, Marioni, RE, Harris, SE, Deary, IJ. Brain age and other bodily ‘ages’: implications for neuropsychiatry. Mol Psychiatry 2019; 24: 266–81.10.1038/s41380-018-0098-1CrossRefGoogle ScholarPubMed
Wu, X, Ye, J, Wang, Z, Zhao, C. Epigenetic age acceleration was delayed in schizophrenia. Schizophr Bull 2021; 47: 803–11.10.1093/schbul/sbaa164CrossRefGoogle ScholarPubMed
Fries, GR, Bauer, IE, Scaini, G, Valvassori, SS, Walss-Bass, C, Soares, JC, et al. Accelerated hippocampal biological aging in bipolar disorder. Bipolar Disord 2020; 22: 498–507.10.1111/bdi.12876CrossRefGoogle ScholarPubMed
Starnawska, A, Tan, Q, Soerensen, M, McGue, M, Mors, O, Børglum, AD, et al. Epigenome-wide association study of depression symptomatology in elderly monozygotic twins. Transl Psychiatry 2019; 9: 214.10.1038/s41398-019-0548-9CrossRefGoogle ScholarPubMed
Wolf, EJ, Logue, MW, Morrison, FG, Wilcox, ES, Stone, A, Schichman, SA, et al. Posttraumatic psychopathology and the pace of the epigenetic clock: a longitudinal investigation. Psychol Med 2019; 49: 791–800.10.1017/S0033291718001411CrossRefGoogle Scholar
Han, LKM, Aghajani, M, Clark, SL, Chan, RF, Hattab, MW, Shabalin, AA, et al. Epigenetic aging in major depressive disorder. Am J Psychiatry 2018; 175: 774–82.10.1176/appi.ajp.2018.17060595CrossRefGoogle ScholarPubMed
Horvath, S. DNA methylation age of human tissues and cell types. Genome Biol 2013; 14: R115.10.1186/gb-2013-14-10-r115CrossRefGoogle ScholarPubMed
Zhang, Q, Vallerga, CL, Walker, RM, Lin, T, Henders, AK, Montgomery, GW, et al. Improved precision of epigenetic clock estimates across tissues and its implication for biological ageing. Genome Med 2019; 11: 54.10.1186/s13073-019-0667-1CrossRefGoogle ScholarPubMed
Magwai, T, Shangase, KB, Oginga, FO, Chiliza, B, Mpofana, T, Xulu, KR. DNA methylation and schizophrenia: current literature and future perspective. Cells 2021; 10: 2890.10.3390/cells10112890CrossRefGoogle ScholarPubMed
Friedman, J, Hastie, T, Tibshirani, R. Regularization paths for generalized linear models via coordinate descent. J Stat Softw 2010; 33: 1–22.10.18637/jss.v033.i01CrossRefGoogle ScholarPubMed
Saldana, DF, Feng, Y. SIS: an R package for sure independence screening in ultrahigh-dimensional statistical models. J Stat Softw 2018; 83: 1–25.10.18637/jss.v083.i02CrossRefGoogle Scholar
Kuleshov, MV, Jones, MR, Rouillard, AD, Fernandez, NF, Duan, Q, Wang, Z, et al. Enrichr: a comprehensive gene set enrichment analysis web server 2016 update. Nucleic Acids Res 2016; 44: W90–7.10.1093/nar/gkw377CrossRefGoogle ScholarPubMed
Xie, Z, Bailey, A, Kuleshov, MV, Clarke, DJB, Evangelista, JE, Jenkins, SL, et al. Gene set knowledge discovery with enrichr. Curr Protoc 2021; 1: e90.10.1002/cpz1.90CrossRefGoogle ScholarPubMed
Bindea, G, Mlecnik, B, Hackl, H, Charoentong, P, Tosolini, M, Kirilovsky, A, et al. ClueGO: a Cytoscape plug-in to decipher functionally grouped gene ontology and pathway annotation networks. Bioinformatics 2009; 25: 1091–3.10.1093/bioinformatics/btp101CrossRefGoogle ScholarPubMed
Levine, ME, Lu, AT, Bennett, DA, Horvath, S. Epigenetic age of the pre-frontal cortex is associated with neuritic plaques, amyloid load, and Alzheimer’s disease related cognitive functioning. Aging 2015; 7: 1198–211.10.18632/aging.100864CrossRefGoogle ScholarPubMed
Pellegrini, C, Pirazzini, C, Sala, C, Sambati, L, Yusipov, I, Kalyakulina, A, et al. A meta-analysis of brain DNA methylation across sex, age, and Alzheimer’s disease points for accelerated epigenetic aging in neurodegeneration. Front Aging Neurosci 2021; 13: 639428.10.3389/fnagi.2021.639428CrossRefGoogle ScholarPubMed
Zhu, Z, Zhang, F, Hu, H, Bakshi, A, Robinson, MR, Powell, JE, et al. Integration of summary data from GWAS and eQTL studies predicts complex trait gene targets. Nat Genet 2016; 48: 481–7.10.1038/ng.3538CrossRefGoogle ScholarPubMed
Cross-Disorder Group of the Psychiatric Genomics Consortium. Identification of risk loci with shared effects on five major psychiatric disorders: a genome-wide analysis. Lancet 2013; 381: 1371–9.10.1016/S0140-6736(12)62129-1CrossRefGoogle Scholar
Lee, SH, Ripke, S, Neale, BM, Faraone, SV, Purcell, SM, Perlis, RH, et al. Genetic relationship between five psychiatric disorders estimated from genome-wide SNPs. Nat Genet 2013; 45: 984–94.Google ScholarPubMed
Gandal, MJ, Zhang, P, Hadjimichael, E, Walker, RL, Chen, C, Liu, S, et al. Transcriptome-wide isoform-level dysregulation in ASD, schizophrenia, and bipolar disorder. Science 2018; 362: eaat8127.10.1126/science.aat8127CrossRefGoogle ScholarPubMed
Consortium C-DGotPG. Genomic relationships, novel loci, and pleiotropic mechanisms across eight psychiatric disorders. Cell 2019; 179: 1469–82.e11.10.1016/j.cell.2019.11.020CrossRefGoogle Scholar
Andreazza, AC, Nierenberg, AA. Mitochondrial dysfunction: at the core of psychiatric disorders? Biol Psychiatry 2018; 83: 718–19.10.1016/j.biopsych.2018.03.004CrossRefGoogle ScholarPubMed
Young, KD, Bodurka, J, Drevets, WC. Differential neural correlates of autobiographical memory recall in bipolar and unipolar depression. Bipolar Disord 2016; 18: 571–82.10.1111/bdi.12441CrossRefGoogle ScholarPubMed
Walker, J, Curtis, V, Shaw, P, Murray, RM. Schizophrenia and bipolar disorder are distinguished mainly by differences in neurodevelopment. Neurotoxicity Res 2002; 4: 427–36.10.1080/1029842021000022070CrossRefGoogle ScholarPubMed
Demjaha, A, MacCabe, JH, Murray, RM. How genes and environmental factors determine the different neurodevelopmental trajectories of schizophrenia and bipolar disorder. Schizophr Bull 2012; 38: 209–14.10.1093/schbul/sbr100CrossRefGoogle ScholarPubMed
Gotlib, IH, Ordaz, SJ. The importance of assessing neural trajectories in pediatric depression. JAMA Psychiatry 2016; 73: 9–10.10.1001/jamapsychiatry.2015.2453CrossRefGoogle ScholarPubMed
Gandal, MJ, Haney, JR, Parikshak, NN, Leppa, V, Ramaswami, G, Hartl, C, et al. Shared molecular neuropathology across major psychiatric disorders parallels polygenic overlap. Science 2018; 359: 693–7.10.1126/science.aad6469CrossRefGoogle ScholarPubMed
Park, BY, Kebets, V, Larivière, S, Hettwer, MD, Paquola, C, van Rooij, D, et al. Multiscale neural gradients reflect transdiagnostic effects of major psychiatric conditions on cortical morphology. Commun Biol 2022; 5: 1024.10.1038/s42003-022-03963-zCrossRefGoogle ScholarPubMed
Stahl, EA, Breen, G, Forstner, AJ, McQuillin, A, Ripke, S, Trubetskoy, V, et al. Genome-wide association study identifies 30 loci associated with bipolar disorder. Nat Genet 2019; 51: 793–803.10.1038/s41588-019-0397-8CrossRefGoogle ScholarPubMed
Fears, SC, Schür, R, Sjouwerman, R, Service, SK, Araya, C, Araya, X. Brain structure-function associations in multi-generational families genetically enriched for bipolar disorder. Brain J Neurol 2015; 138: 2087–102.10.1093/brain/awv106CrossRefGoogle ScholarPubMed
Warden, MR, Selimbeyoglu, A, Mirzabekov, JJ, Lo, M, Thompson, KR, Kim, SY, et al. A prefrontal cortex-brainstem neuronal projection that controls response to behavioural challenge. Nature 2012; 492: 428–32.10.1038/nature11617CrossRefGoogle ScholarPubMed
Figure 0

Fig. 1 Performance assessment of the blood–brain clock. (a) Venn diagram comparing cytosine guanine probes used in the blood–brain clock with three established epigenetic clocks (multi-tissue predictor, multiple predictor and cortical clock). (b) Functional enrichment analysis of the 203 blood–brain clock probe-annotated genes, including Gene Ontology biological process (GO-BP), cellular component (GO-CC) and molecular function (GO-MF), and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway. (c) Epigenome-wide association study trait enrichment of the 203 blood–brain clock probe-annotated genes. (d, e) Scatter plots of chronological age (x axis) versus DNA methylation (DNAm) age (y axis) estimated by the blood–brain clock in blood (red circle) and brain (blue triangle) tissues from the training cohort (d) and testing cohort (e) of healthy controls. The dashed line indicates equality between DNAm and chronological age; solid lines indicate regression fits. Model performance metrics (prediction error (err), mean absolute error (MAE), root mean square error (RMSE) and Pearson’s correlation (R) are shown). (f, g) Scatter plots comparing chronological age and DNAm age in (f) blood and (g) brain from independent healthy controls, assessed using the blood–brain clock, multiple predictor, cortical clock and multi-tissue predictor.

Figure 1

Fig. 2 Prediction and comparison of epigenetic age in four neuropsychiatric disorders versus controls using the blood–brain clock. Scatter plots show chronological age (x axis) versus DNA methylation (DNAm) age (y axis) in blood (a1–d1, left) and brain (a2–d2, left), alongside comparison of epigenetic age acceleration (accelerationDiff) and residuals from linear regression of DNAm age on chronological age (accelerationResidual) (a1–d1, right; a2–d2, right). Disorders include schizophrenia (SZ; a1, a2), bipolar disorder (BD; b1, b2), major depressive disorder (MD; c1, c2) and Alzheimer’s disease (AD; d1, d2). Con, controls. In scatter plots, the dashed line represents equality between DNAm and chronological age; solid lines indicate regression fits. Model performance metrics (prediction error (err), correlation coefficient (cor) and Pearson’s P) are displayed in each panel. The Kruskal–Wallis test was used to evaluate differences in accelerationDiff and accelerationResidual between disease groups and controls.

Figure 2

Fig. 3 Meta-analyses of epigenetic age acceleration and disorder prevalence. Forest plots show associations between epigenetic age acceleration and prevalence of (a) schizophrenia (SZ), (b) bipolar disorder (BD), (c) major depressive disorder (MD) and (d) Alzheimer’s disease (AD; positive control). Thresholds were defined as the first quartile of the combined control epigenetic age acceleration (accelerationDiff) for SZ (–2.77 in blood, –2.98 in brain) and the third quartile for BD, MD and AD (3.01 in blood, 3.20 in brain). For each disorder, events represent subjects with accelerationDiff values below the first quartile (SZ) or above the third quartile (BD, MD, AD) of the control distribution. Odds ratios (ORs) with 95% confidence intervals are shown for individual data-sets and the overall pooled effect. Tests for heterogeneity were assessed using I2 and τ2 statistics.

Figure 3

Fig. 4 Age-stratified changes in epigenetic age acceleration across three disorders. Epigenetic age acceleration (accelerationDiff) in blood (left panels) and brain (right panels) is shown for (a) schizophrenia (SZ) (yellow [upper] line), (b) bipolar disorder (BD) (purple [upper] line) and (c) major depressive disorder (MD) (blue [upper] line) compared with age-matched controls (green [lower] line) across different chronological age categories. Solid dots indicate group means, with error bars representing standard deviations. Significant case–control differences based on Kruskal–Wallis tests are marked by arrows (↑ increase, ↓ decrease), with significance thresholds indicated as ***P < 0.001, **P < 0.01, *P < 0.05.

Figure 4

Fig. 5 Changes in methylation levels of blood–brain clock DNA methylation (DNAm) probes in three disorders. (a) Venn diagram of summary data Mendelian randomisation (SMR)-identified probes associated with schizophrenia (SZ), bipolar disorder (BD) and major depressive disorder (MD). (b) Gene Ontology (GO) enrichment analysis of genes corresponding to SMR identified 42 probes, highlighting pathways related to neurodevelopment and neuronal signalling. (c) Venn diagrams showing overlap between SMR probes and disease-associated differential methylation probes (DMPs) in blood and brain for SZ, BD and MD. (d) Blood-based classification: heatmap of representative DMPs (left) and receiver operating characteristic (ROC) analyses (right) demonstrating high accuracy in distinguishing SZ from BD (area under the receiver operating characteristic curve (AUC) = 0.963, 95% CI = 0.929–0.986), and SZ from MD (AUC = 0.842, 95% CI = 0.765–0.894). Sensitivity (sens), specificity (spec), negative predictive value (NPV), positive predictive value (PPV) and optimal fitted value (Ir.eta) are indicated. (e) Brain-based classification: heatmap of representative DMPs (left) and ROC analyses (right) showing robust performance in differentiating SZ from BD (AUC = 0.979, 95% CI = 0.953–0.992), and SZ from MD (AUC = 0.941, 95% CI = 0.892–0.969).

Supplementary material: File

Wu et al. supplementary material 1

Wu et al. supplementary material
Download Wu et al. supplementary material 1(File)
File 1.1 MB
Supplementary material: File

Wu et al. supplementary material 2

Wu et al. supplementary material
Download Wu et al. supplementary material 2(File)
File 2.4 MB

This journal is not currently accepting new eletters.

eLetters

No eLetters have been published for this article.