henhao
Sign in
GenomicsbioRxiv
Preprint

System-Specific Epigenetic Aging Signatures in Autistic Adults

Abigail Dickinson, Candace R. LewisiD, Daniel H. GeschwindiD, Judith CarrolliD, Catherine LordiD

Abstract

Background: Autism is a lifelong neurodevelopmental condition, but the biological processes shaping aging are poorly understood. DNA methylation-based measures index physiological decline and mortality risk independently of chronological age. We tested whether biological aging differs in autistic adults, whether differences are system-specific, and which factors predict within-autism variation. Methods: Saliva DNA methylation (Illumina EPIC) was assayed in 37 autistic adults (31 male, 6 female; mean age 29.8 years), followed from early childhood, spanning a broad ability range including intellectual disability, and 188 non-autistic adults from four public datasets, processed jointly. Fifteen measures were calculated: four global (Horvath, GrimAge, DunedinPACE, SystemsAge) and 11 system-specific SystemsAge subscores. Models adjusted for age, sex, and epithelial-cell proportion. Results: Global aging measures did not differ between groups (all q > .34). However, autistic adults had higher Brain ({beta} = 0.81), Blood ({beta} = 0.65), and Liver SystemsAge ({beta} = 0.58; all q < .01), with Brain and Blood robust across sensitivity analyses. Within autism, aging measures were unrelated to symptom severity, IQ, co-occurring diagnoses, or medication use. Instead, elevations were associated with psychosocial circumstances, including a lack of structured daytime activity, greater loneliness, and lower quality of life. Conclusions: Differences in aging biology in autism appeared system-specific rather than global, with no evidence of generalized epigenetic age acceleration. This vulnerability appeared more closely tied to psychosocial factors than core clinical features. System-specific DNAm measures may therefore provide a more sensitive framework for identifying biological heterogeneity and potentially modifiable contributors to healthy aging in autistic adulthood.

PDFDOI
Full-text digest ↓

02 /Specification & Classification

Click any tag to browse matching papers
Citations
1
OpenAlex Topics

In-depth summary

✦ Multi-agent verified full-text synthesis

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (18 Sept 2026) · CC BY 4.0

Saliva methylation from 37 autistic adults vs 188 public controls showed elevated Brain, Blood and Liver SystemsAge subscores, while global clocks (Horvath, GrimAge, DunedinPACE) showed no difference.

Little is known about biological aging in autistic adults, and earlier first- and second-generation epigenetic clock studies found little global age acceleration. This study adds a system-specific measure (SystemsAge) to global clocks in a cohort followed from early childhood that includes adults with intellectual disability, a group the authors say is missing from earlier epigenetic aging studies.

Key results

  1. ShownAutistic adults scored higher than controls on three of the 11 SystemsAge subscores (Brain, Blood, Liver), and each survived FDR correction across the 15-measure panel. A Metabolic elevation was nominal only.

    Autistic n = 37 vs control n = 188, linear regression adjusted for age, sex and epithelial-cell proportion, effects in SD units: Brain β = 0.81 (q < .001), Blood β = 0.65 (q = .002), Liver β = 0.58 (q = .008). Metabolic β = 0.38 (p = .035, q = .130). The other seven subscores: all q > .26. · 3.2. Group Differences

  2. ShownThere was no sign of global epigenetic age acceleration: none of the four composite measures differed between groups. The level-type clocks still tracked chronological age in the pooled sample.

    Horvath skin-and-blood clock, GrimAge, DunedinPACE and the SystemsAge composite: all q > .34 in the primary model, and nonsignificant in every sensitivity model. Correlation with chronological age in the combined sample: Horvath r = .94, GrimAge r = .87, SystemsAge composite r = .65 (all p < .001). · 3.1. Age Associations; 3.2. Group Differences; 3.3. Sensitivity Analyses

  3. ShownThe Brain and Blood elevations held up across sensitivity analyses. Liver was elevated in most specifications, but the Metabolic difference depended on which covariates and control datasets were used.

    Brain and Blood stayed FDR-significant in an age- and sex-restricted subsample, after excluding cross-reactive probes, and after adjusting for DNAm BMI and smoking. When each of the four GEO control datasets was left out in turn, Brain, Blood and Liver stayed positive and FDR-significant. Metabolic reached significance only under some covariate sets and varied across leave-one-out runs. · 2.5.3. Sensitivity Analyses; 3.3. Sensitivity Analyses

  4. SuggestedWithin autism, higher SystemsAge went with no structured daytime activity (Brain, Metabolic, Blood), greater loneliness (Blood) and lower quality-of-life and health ratings (Brain). Most held after adjustment for DNAm BMI and race/ethnicity; the Blood link to daytime activity weakened.

    Not occupied vs occupied: Brain β = 1.76 (p = .003), Metabolic β = 1.15 (p < .001), Blood β = 0.85 (p = .011). After adding DNAm BMI and race/ethnicity, Brain β = 1.68 and Metabolic β = 0.83 stayed significant, while Blood fell to β = 0.54 (p = .092). Loneliness and Blood: β = 0.54 (p = .006), adjusted β = 0.59 (p = .030). Quality of life and health vs Brain: β = −0.44 (p = .024) and β = −0.43 (p = .003). · 3.4.4. Psychosocial Characteristics

  5. SuggestedThe elevated subscores were unrelated to autism symptom severity, IQ, verbal-IQ group, medication exposure and most co-occurring conditions. Body mass index (BMI), whether measured or methylation-derived, did relate to several subscores.

    No association with ADOS-2 calibrated severity scores or full-scale, verbal or nonverbal IQ; VIQ ≥ 70 and < 70 did not differ. A physical health condition and Metabolic SystemsAge: β = 0.69 (p = .020), then β = 0.36 (p = .210) with DNAm BMI added. Measured BMI correlated with DNAm BMI (r = .72), with Metabolic SystemsAge (partial r = .42, p = .013) and with Brain SystemsAge (partial r = .41, p = .017). · 3.4. Within Autism Correlates of SystemsAge; 3.4.2. Co-occurring Conditions and Medication…; 3.4.3. Autism Symptomatology and Cognitive…

Methods, data and reproducibility

Saliva DNA methylation (Illumina EPIC v1.0) from 37 autistic adults in a longitudinal early-diagnosis cohort was processed jointly with 188 saliva EPIC controls from four GEO datasets. The authors compared 15 DNAm aging measures by linear regression adjusted for age, sex and EpiDISH-estimated epithelial proportion, then ran sensitivity analyses and exploratory within-autism correlates.

  • Early Diagnosis Study (EDX) autistic adults, referred before age 3 and followed into adulthood; saliva from the age-30 wave, plus clinical, cognitive and psychosocial phenotypes · human · n = 37 (31 male, 6 female; mean age 29.8 years, range 27.6–32.8)
  • Non-autistic control saliva DNA methylation (EPIC array), raw IDAT files from four GEO datasets selected by prespecified criteria · human · n = 188 after QC (204 eligible) · GSE151485, GSE232332, GSE111631, GSE232891
Methods & tools
Illumina HumanMethylationEPIC v1.0 BeadChip, Oragene DNA self-collection kits, EpiDISH, minfi, meffil, Horvath skin-and-blood clock, GrimAge, DunedinPACE, SystemsAge, ADOS-2, WHOQOL-BREF, Asher Loneliness and Social Dissatisfaction Scale
Reagents & environment
Analyses run in R 4.5.1. Preprocessing thresholds, software implementation and clock probe coverage are given in supplementary Methods S4 and Tables S2-S3.
Reproducibility gaps
The paper gives no code repository. Control saliva EPIC data are public in GEO (GSE151485, GSE232332, GSE111631, GSE232891; accession numbers only, no URLs printed). No access route is given for the autistic cohort's methylation or phenotype data.
Limitations (6)
  • The autistic sample is modest and spans a narrow age range (27.6–32.8 years). This limits power to detect age-dependent effects within autism and rules out conclusions about longitudinal aging trajectories. · stated by the authors
  • Controls come from independent public datasets, so diagnostic group is confounded with data source. Joint preprocessing, matched tissue and array, technical sensitivity models and leave-one-dataset-out analyses reduce this risk but do not remove it. · stated by the authors
  • Saliva is a heterogeneous peripheral tissue. SystemsAge subscores should not be read as direct measures of aging in specific organs, nor as inherited genetic risk. · stated by the authors
  • The within-autism analyses were exploratory: several measures were available only for a subset (e.g. self- and caregiver-report quality-of-life data for 28 participants), some categorical groups were small and uneven, and cross-sectional associations cannot establish directionality. · stated by the authors
  • Within-autism associations are reported as nominal p-values across many predictors and four subscores, with n of 37 or fewer. Single findings such as loneliness with Blood SystemsAge could therefore be chance results. · noted in this summary
  • The group models adjust for age, sex and epithelial proportion (plus DNAm-derived proxies in sensitivity runs). Race/ethnicity, which was linked to Brain and Liver SystemsAge within autism, is not listed as a group-level covariate. · noted in this summary

Who should read:Read this if you study epigenetic aging in neurodevelopmental or psychiatric conditions and want to see system-specific DNAm measures (SystemsAge) compared with global clocks, using public saliva EPIC data as controls.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISTollenaar '21Rasero '26Eyring '25Dickinson '21Bruellman '26Moyakhe '24
66% semantic matchGenetics
Marieke S. Tollenaar, Roseriet Beijers, Elika Garg+8 more·Biological Psychology·Jan 18, 2021· 23 citations