Significant variation in the performance of DNA methylation predictors across data preprocessing and normalization strategies

Meta-analysis of epigenetic aging in schizophrenia reveals multifaceted relationships with age, sex, illness duration, and polygenic risk

BACKGROUND

The study of biological age acceleration may help identify at-risk individuals and reduce the rising global burden of age-related diseases. Using DNA methylation (DNAm) clocks, we investigated biological aging in schizophrenia (SCZ), a mental illness that is associated with an increased prevalence of age-related disabilities and morbidities. In a whole blood DNAm sample of 1090 SCZ cases and 1206 controls across four European cohorts, we performed a meta-analysis of differential aging using three DNAm clocks (i.e., Hannum, Horvath, and Levine). To dissect how DNAm aging contributes to SCZ, we integrated information on duration of illness and SCZ polygenic risk, as well as stratified our analyses by chronological age and biological sex.

RESULTS

We found that blood-based DNAm aging is significantly altered in SCZ independent from duration of the illness since onset. We observed sex-specific and nonlinear age effects that differed between clocks and point to possible distinct age windows of altered aging in SCZ. Most notably, intrinsic cellular age (Horvath clock) is decelerated in SCZ cases in young adulthood, while phenotypic age (Levine clock) is accelerated in later adulthood compared to controls. Accelerated phenotypic aging was most pronounced in women with SCZ carrying a high polygenic burden with an age acceleration of + 3.82 years (CI 2.02-5.61, P = 1.1E-03). Phenotypic aging and SCZ polygenic risk contributed additively to the illness and together explained up to 14.38% of the variance in disease status.

CONCLUSIONS

Our study contributes to the growing body of evidence of altered DNAm aging in SCZ and points to intrinsic age deceleration in younger adulthood and phenotypic age acceleration in later adulthood in SCZ. Since increased phenotypic age is associated with increased risk of all-cause mortality, our findings indicate that specific and identifiable patient groups are at increased mortality risk as measured by the Levine clock. Our study did not find that DNAm aging could be explained by the duration of illness of patients, but we did observe age- and sex-specific effects that warrant further investigation. Finally, our results show that combining genetic and epigenetic predictors can improve predictions of disease outcomes and may help with disease management in schizophrenia.

Characterisation and reproducibility of the HumanMethylationEPIC v2.0 BeadChip for DNA methylation profiling

BACKGROUND

The Illumina family of Infinium Methylation BeadChip microarrays has been widely used over the last 15 years for genome-wide DNA methylation profiling, including large-scale and population-based studies, due to their ease of use and cost effectiveness. Succeeding the popular HumanMethylationEPIC BeadChip (EPICv1), the recently released Infinium MethylationEPIC v2.0 BeadChip (EPICv2) claims to extend genomic coverage to more than 935,000 CpG sites. Here, we comprehensively characterise the reproducibility, reliability and annotation of the EPICv2 array, based on bioinformatic analysis of both manifest data and new EPICv2 data from diverse biological samples.

RESULTS

We find a high degree of reproducibility with EPICv1, evidenced by comparable sensitivity and precision from empirical cross-platform comparison incorporating whole genome bisulphite sequencing (WGBS), and high correlation between technical sample replicates, including between samples with DNA input levels below the manufacturer's recommendation. We provide a full assessment of probe content, evaluating genomic distribution and changes from previous array versions. We characterise EPICv2's new feature of replicated probes and provide recommendations as to the superior probes. In silico analysis of probe sequences demonstrates that probe cross-hybridisation remains a significant problem in EPICv2. By mapping the off-target sites at single nucleotide resolution and comparing with WGBS we show empirical evidence for preferential off-target binding.

CONCLUSIONS

Overall, we find EPICv2 a worthy successor to the previous Infinium methylation microarrays, however some technical issues remain. To support optimal EPICv2 data analysis we provide an expanded version of the EPICv2 manifest to aid researchers in understanding probe design, data processing, choosing appropriate probes for analysis and for integration with methylation datasets from previous versions of the Infinium Methylation BeadChip.

Circulating Leukocyte Subsets Before and After a Breast Cancer Diagnosis and Therapy

Importance

Changes in leukocyte composition often precede chronic disease onset. Patients with a history of breast cancer (hereinafter referred to as breast cancer survivors) are at increased risk for subsequent chronic diseases, but the long-term changes in peripheral leukocyte composition following a breast cancer diagnosis and treatment remain unknown.

Objective

To examine longitudinal changes in peripheral leukocyte composition in women who did and did not develop breast cancer and identify whether differences in breast cancer survivors were associated with specific treatments.

Design, Setting, and Participants

In this prospective cohort study, paired blood samples were collected from 2315 women enrolled in The Sister Study, a US-nationwide prospective cohort study of 50 884 women, at baseline (July 2003 to March 2009) and follow-up (October 2013 to March 2015) home visits, with a mean (SD) follow-up interval of 7.6 (1.4) years. By design, approximately half of the included women had been diagnosed and treated for breast cancer after enrollment and before the second blood draw. A total of 410 women were included in the present study, including 185 breast cancer survivors and 225 who remained free of breast cancer over a comparable follow-up period. Data were analyzed from April 21 to September 9, 2022.

Exposures

Breast cancer status and, among breast cancer survivors, cancer treatment type (chemotherapy, radiotherapy, endocrine therapy, or surgery).

Main Outcomes and Measures

Blood DNA methylation data were generated in 2019 using a genome-wide methylation screening tool and deconvolved to estimate percentages of 12 circulating leukocyte subsets.

Results

Of the 410 women included in the analysis, the mean (SD) age at enrollment was 56 (9) years. Compared with breast cancer-free women, breast cancer survivors had decreased percentages of circulating eosinophils (-0.45% [95% CI, -0.87% to -0.03%]; P = .03), total CD4+ helper T cells (-1.50% [95% CI, -2.56% to -0.44%]; P = .01), and memory B cells (-0.22% [95% CI, -0.34% to -0.09%]; P = .001) and increased percentages of circulating naive B cells (0.46% [95% CI, 0.17%-0.75%]; P = .002). In breast cancer survivor-only analyses, radiotherapy was associated with decreases in total CD4+ T cell levels, whereas chemotherapy was associated with increases in naive B cell levels. Surgery and endocrine therapy were not meaningfully associated with leukocyte changes.

Conclusions and Relevance

In this cohort study of 410 women, breast cancer survivors experienced lasting changes in peripheral leukocyte composition compared with women who remained free of breast cancer. These changes may be related to treatment with chemotherapy or radiotherapy and could influence future chronic disease risk.

Integration of datasets for individual prediction of DNA methylation-based biomarkers

BACKGROUND

Epigenetic scores (EpiScores) can provide biomarkers of lifestyle and disease risk. Projecting new datasets onto a reference panel is challenging due to separation of technical and biological variation with array data. Normalisation can standardise data distributions but may also remove population-level biological variation.

RESULTS

We compare two birth cohorts (Lothian Birth Cohorts of 1921 and 1936 - nLBC1921 = 387 and nLBC1936 = 498) with blood-based DNA methylation assessed at the same chronological age (79 years) and processed in the same lab but in different years and experimental batches. We examine the effect of 16 normalisation methods on a novel BMI EpiScore (trained in an external cohort, n = 18,413), and Horvath's pan-tissue DNA methylation age, when the cohorts are normalised separately and together. The BMI EpiScore explains a maximum variance of R2=24.5% in BMI in LBC1936 (SWAN normalisation). Although there are cross-cohort R2 differences, the normalisation method makes a minimal difference to within-cohort estimates. Conversely, a range of absolute differences are seen for individual-level EpiScore estimates for BMI and age when cohorts are normalised separately versus together. While within-array methods result in identical EpiScores whether a cohort is normalised on its own or together with the second dataset, a range of differences is observed for between-array methods.

CONCLUSIONS

Normalisation methods returning similar EpiScores, whether cohorts are analysed separately or together, will minimise technical variation when projecting new data onto a reference panel. These methods are important for cases where raw data is unavailable and joint normalisation of cohorts is computationally expensive.

Changes in methylation-based aging in women who do and do not develop breast cancer

BACKGROUND

Breast cancer survivors have increased incidence of age-related diseases, suggesting that some survivors may experience faster biological aging.

METHODS

Among 417 women enrolled in the prospective Sister Study cohort, DNA methylation data were generated on paired blood samples collected an average of 7.7 years apart and used to calculate 3 epigenetic metrics of biological aging (PhenoAgeAccel, GrimAgeAccel, and Dunedin Pace of Aging Calculated from the Epigenome [DunedinPACE]). Approximately half (n = 190) the women sampled were diagnosed and treated for breast cancer between blood draws, whereas the other half (n = 227) remained breast cancer-free. Breast tumor characteristics and treatment information were abstracted from medical records.

RESULTS

Among women who developed breast cancer, diagnoses occurred an average of 3.5 years after the initial blood draw and 4 years before the second draw. After accounting for covariates and biological aging metrics measured at baseline, women diagnosed and treated for breast cancer had higher biological aging at the second blood draw than women who remained cancer-free as measured by PhenoAgeAccel (standardized mean difference [β] = 0.13, 95% confidence interval [CI) = 0.00 to 0.26), GrimAgeAccel (β = 0.14, 95% CI = 0.03 to 0.25), and DunedinPACE (β = 0.37, 95% CI = 0.24 to 0.50). In case-only analyses assessing associations with different breast cancer therapies, radiation had strong positive associations with biological aging (PhenoAgeAccel: β = 0.39, 95% CI = 0.19 to 0.59; GrimAgeAccel: β = 0.29, 95% CI = 0.10 to 0.47; DunedinPACE: β = 0.25, 95% CI = 0.02 to 0.48).

CONCLUSIONS

Biological aging is accelerated following a breast cancer diagnosis and treatment. Breast cancer treatment modalities appear to differentially contribute to biological aging.

An overview of DNA methylation-derived trait score methods and applications

Microarray technology has been used to measure genome-wide DNA methylation in thousands of individuals. These studies typically test the associations between individual DNA methylation sites ("probes") and complex traits or diseases. The results can be used to generate methylation profile scores (MPS) to predict outcomes in independent data sets. Although there are many parallels between MPS and polygenic (risk) scores (PGS), there are key differences. Here, we review motivations, methods, and applications of DNA methylation-based trait prediction, with a focus on common diseases. We contrast MPS with PGS, highlighting where assumptions made in genetic modeling may not hold in epigenetic data.