Epigenome-wide association study of left-handedness for different tissues and ages


 We report the first epigenome-wide association study of left-handedness, a trait with low heritability for which epigenetic mechanisms have been proposed as an underlying etiological mechanism. A region-based meta-analysis of whole blood genome-wide DNA methylation data from two cohorts (3,914 adults) identified differentially methylated regions annotated to BLCAP (20q11.23), a negative regulator of cell growth involved in apoptosis, NNAT (20q11.23), involved in brain development, and IAH1 (2p25.1), which encodes an acyl esterase. CpGs located in proximity to the SNPs from the largest GWAS of handedness were more strongly associated with left-handedness than other differentially methylated positions. In longitudinal whole blood samples, cord blood, and buccal cells from children (N = 1,967), the association with handedness displayed moderate stability across age, but little consistency across cell types. These findings suggest new candidate loci associated with handedness.

Both cohorts include methylation data in children and adults. We excluded ambidextrous and mixed-handed persons, and treated handedness as a dichotomous trait (left or right-handed). First, we performed a meta-analysis of DNA methylation data from adults in these two large cohorts (total sample size=3,914) to identify differentially methylated positions (DMPs) and differentially methylated regions (DMRs) associated with left-handedness. Next, we performed additional analyses in which we 1) examined if the epigenetic signal for left-handedness was enriched near previously reported GWAS loci 13 ; 2) examined methylation differences between left and right-handed twins from discordant monozygotic (MZ) twin pairs; and 3) characterized the longitudinal and cross-tissue similarity of the genome-wide epigenetic signal associated with left-handedness using data from children. Finally, we created methylation scores and estimated their predictive value over and above polygenic scores (Fig.1)
The prevalence of left-handedness increased from 10% in individuals born before 1960's to 13.7% born after 1960's (see Table S3).
We tested 409,562 CpGs with adjustment for age, sex, smoking status, body mass index (BMI), measured or estimated cell proportions, and technical covariates. Genome-wide EWAS test statistics from each cohort separately, and from the meta-analysis showed no in ation (Tables S5-11). All the p-values were higher than the epigenome-wide Bonferroni-corrected threshold of 1.22 ´ 10 -7 and the False Discovery Rate (FDR) q-value of 0.05 (Fig.S1). The CpGs with lowest p-values in meta-analysis (p<1 ´ 10 -5 ) are shown in Table 3. Six of eight CpGs were located near transcription start sites on different chromosomes: in the LRCC2 gene on chromosome 3, in the ATP6V1B2 gene on chromosome 8, in the CKAP4, GALNTR6, and UNC1198 genes on chromosome 12, in the C13orf18 gene on chromosome 13, in the MBD2 gene on chromosome 18, and in the NTSR1 gene on chromosome 20.
The DMR meta-analysis detected two DMRs associated with left-handedness ( Fig.2a- Table 4): one on chromosome 20 (BLCAP; NNAT; 16 CpGs; p-value adjusted for multiple testing (p adj ) =0.00004), and one on chromosome 2 (IAH1; 7 CpGs; p adj =0.03). Both DMRs were within CpG islands, 15 of 16 CpGs in the rst DMR and 6 of 7 CpGs in the second DMR were hypomethylated in lefthanders. These DMRs were not detected in the individual cohorts. One DMR was detected in ALSPAC adults on chromosome 1 (AMPD2, 4 CpGs, p adj =0.031) that did not show associations in the meta-analysis. The follow-up of CpGs from DMRs associated with left-handedness in meta-analysis was done in previous EWAS conducted on other traits and reported in Table S12. GWAS follow-up We tested the overlap of our EWAS meta-analysis results with ndings from the most recent GWAS meta-analysis of handedness 13 .
CpGs located within 1 Mb window of SNPs associated with left-handedness (at p < 5 ´ 10 -8 ) were on average more strongly associated with left-handedness in the EWAS meta-analysis than the other tested CpGs (ß=0.027, p =0.04). The effect was weaker when less stringent GWAS p-value cut-offs were applied (i.e. SNPs with p<1 ´ 10 -6 , and SNPs with p<1 ´ 10 -5 ). Importantly, in a control analysis with a trait that is not associated with handedness, we did not see this stronger association for CpGs located near SNPs associated with type 2 diabetes in GWAS 54 (ß =0.005, p=0.265) (see Table S13, Fig. S2a-d).

Discordant MZ twins
In NTR, the DNA methylation datasets included 1,279 adult monozygotic (MZ) twins with blood samples and 710 MZ children with buccal samples. We found that 21% (N=133 pairs) of the adult MZ twin pairs and 24% (N=86 pairs) of the MZ twin children were discordant for handedness (Tables S1-S2). Characteristics of MZ discordant twins are presented in Table S4. In both groups, we performed an MZ discordant within-pair EWAS analysis, comparing right-and left-handed twins. Within-pair analyses of DNA methylation of left and right-handed twins did not nd association at DMPs or DMRs in blood or buccal samples at Bonferroni or FDR threshold ( Fig.S1q-t, Tables S14-17). We compared the methylation results obtained in discordant MZ twins to the EWAS metaanalysis results. To avoid sample overlap, we repeated the EWAS meta-analysis after exclusion of the MZ discordant twin pairs.

Longitudinal analysis
While handedness is a stable trait, DNA methylation can vary over time 55 . We analyzed DNA methylation in ALSPAC offspring measured in cord blood at birth, and in peripheral blood at 7, 17, and 24 years old (N~1021, Table 2 and Table S2) to examine the association between DNA methylation and left-handedness throughout childhood and adolescence. The correlations between the top 100 CpG effect sizes from EWASs of handedness performed at different time points were moderate to strong (r=0.355; p=0.0002 to r =0.578, p=1.2 ´ 10 -10 ), except for a weak correlation between top 100 CpGs effect sizes at 17 years and the same CpGs at age 24 years (r ALSPAC17-ALSPAC24 =0.079; p =0.435) (Fig. S3). There were no overlapping CpGs amongst the top 100 CpGs between analyses at different time points (Fig. S4). Correlations between top CpG effect sizes between ALSPAC adults (mothers and fathers) and offspring at birth were strong negative (r ALSPACadults-ALSPACatbirth =-0.68; p = 7.2 ´ 10 -15 ) (Fig.S5), and between ALSPAC adults and offspring at 7, 17, 24 years were weak (r from -0.006 to 0.141, p >0.0003). No CpGs passed a Bonferroni correction at any time point (Tables S18-25).

Sensitivity analyses
Above we reported the DNA methylation and left-handedness association study with adjustment for prenatal and postnatal factors that in uence DNA methylation as was shown in previous studies: smoking and BMI in adults 49,56 , and gestational age, birth weight and prenatal maternal smoking 57,58 in children. These characteristics were reported to be associated with handedness in studies of prenatal maternal smoking 59,60 , gestational age 61,62 , birth weight 63 , BMI 64 , and smoking 65 . We examined if the EWAS results for handedness differ without taking these factors into account. Across all analyses, the correlations between the effect sizes of the top 100 CpGs were strong between the models with and without adjustment for these factors (r ranged from 0.99 to 1), and overlaps of the top 100 CpGs were substantial (32 to 87 CpGs). Adjustment for the factors increased the number of DMRs associated with lefthandedness in meta-analysis (1 DMR without adjustment and 2 DMRs with adjustment), and in EWAS in children (2 DMRs without adjustment, and 4 with adjustment in buccal cells in NTR).

Handedness methylation scores
To examine if variation in handedness can be predicted by DNA methylation levels across multiple CpGs, methylation scores (MS) were created. These were based on EWAS summary statistics in NTR to predict into ALSPAC, and on ALSPAC summary statistics to predict into NTR given the following p-value thresholds for inclusion of CpGs: p < 1 ´ 10 -1 , p < 1 ´ 10 -3 , p < 1 ´ 10 -5 (Fig.1, Table S28, Fig.  S7). To estimate the variance explained by MS above genetic variants, polygenic scores (PGS) were created based on the summary statistics from the handedness GWAS of Cuellar-Partida et al. 13 with exclusion of NTR, ALSPAC and 23andMe cohorts (N GWAS =196,419). Since four scores were tested (3 methylation scores, one polygenic score), we applied Bonferroni correction for four tests (α=0.05/4=0.0125). The results are summarized in Tables S29-30. MS did not predict left-handedness in NTR and ALSPAC adults, or in children at 7-9 years old and did not explain variance over and above the variance explained by the PGS in the combined model (R 2 MS from -0.17 to 1.28%, R 2 PGS from 0 to 0.48%). The largest amount of explained variance was in ALSPAC at 7 years old for the MS based on CpGs with a p < 1 ´ 10 -5 (R 2 MS =1.28%, p=0.1, N CpGs =7).

Discussion
We have performed an epigenome-wide association study of left-handedness including left and right-handed individuals from two population-based cohorts from the Netherlands and the UK. In the meta-analysis, combining the NTR and ALSPAC adult cohorts, two DMRs were associated with left-handedness. The rst DMR (genomic location: chr20q11.23, 36,148,679:36,149,022) is located within the 5'UTR of the BLCAP apoptosis inducing factor (BLCAP) gene and nearby the transcription start site (TSS1500) of the neuronatin (NNAT) gene. BLCAP encodes a protein that reduces cell growth by stimulating apoptosis. NNAT is involved in brain development and nervous system structure maturation and maintenance. CpGs from this region were previously associated with myalgic encephalomyelitis and chronic fatigue syndrome, preterm birth, obesity, metabolic parameters, and arm fat mass (DXA scan measurement). The second DMR (genomic location: chr2p25.1, 9,614,471:9,614,744) is located within the isoamyl acetate hydrolyzing esterase 1 (IAH1) gene. The IAH1 gene encodes an acyl esterase and is associated with neonatal in ammatory skin and bowel disease, and a disease with an inborn error of leucine metabolism (3 methylglutaconic aciduria type 1). CpGs from the region were previously associated with gestational age, bone mineral density, metabolic parameters, and schizophrenia. Some of these traits have been previously associated with handedness in epidemiological studies, e.g. BMI 64 and gestational age 61,62 , for which we adjusted in our analyses. Previous analysis of the genetic correlations between left-handedness and 1,349 complex traits using LDscore regression did not reveal any genetic correlations at FDR <5%, but suggestive positive correlations were observed with neurological and psychiatric traits including schizophrenia 13 .
Even though no DMPs were identi ed after correction for multiple testing, the high-ranking CpGs can be of potential interest. The second-ranking CpG cg09239756 (genomic location: chr12, 106,642,360) is located near the cytoskeleton associated protein 4 (CKAP4) gene. This gene mediates the anchoring of the endoplasmic reticulum to microtubules. Microtubules are an important cytoskeleton component that play a role in neuronal morphogenesis and migration, and axon transport 39 . Microtubules have been widely discussed in association with handedness 17,35 and brain anatomical asymmetry 40 , and genes involved in microtubule pathways were enriched in the GWAS of handedness 13 . Moreover, in our enrichment analysis, we found that CpGs located within a 1Mb window from SNPs associated with left-handedness from the GWAS meta-analysis by Cuellar-Partida et al. were more strongly associated with left-handedness in our meta-analysis compared to CpGs outside of this window.
Hand movements together with other lateralized movements and molecular signs of lateralization are observed at very early stages of human development in the uterus 51,66-68 . Therefore, DNA methylation differences associated with hand preference are expected to emerge early in development. While DNA methylation at some CpGs in the genome changes throughout the lifespan 55 , the DNA methylation signal associated with left-handedness was moderately consistent from birth throughout the lifespan: DMP effect sizes were correlated in ALSPAC individuals from birth to 24 years old, although genome-wide signi cance for DMPs was not reached. Consistency in DNA methylation signal associated with left-handedness at different time-points may indicate that the pattern for lefthandedness is conserved through the lifespan.
Several DMRs were detected in buccal cells in children around 9 years old (genomic locations: chr8, 22, 9, and 12) after correction for multiple testing. Annotation of these regions implicate the following protein coding genes: the plectin gene (PLEC1), the myoglobin gene (MB) gene, the elongator complex protein 1 gene (ELP1), the actin binding transcription modulator gene (ABITRAM), and the WNK lysine de cient protein kinase 1 gene (WNK1). The CpGs in the regions are mostly hypomethylated in left-handed individuals.
The genes encode for proteins that participate in cytoskeleton functions, chromatin organization, development of neurons, and metabolism. CpGs from DMRs in buccal cells have been previously associated with other phenotypes: CpGs from the DMR on chromosomes 8 were previously associated with myalgic encephalomyelitis, chronic fatigue syndrome, multiple sclerosis, and gestational age; CpGs from the DMR on chromosome 9 were reported in association with bone mineral density, tissue mass of the arm (DXA scans measurement), and multiple metabolic parameters. Interestingly, some of these phenotypes were also associated with CpGs from the blood meta-analysis.
A difference in handedness preference in MZ twin pairs has always fascinated parents of twins and twins themselves: how can children with almost identical genes differ for such a prominent trait? Handedness discordance in identical twins was described a long time ago 18-21 , and the percentage of MZ discordant twins were reported as 20% of 3,486 MZ twins in East Flanders 22 , and 19% of 1,724 MZ twins from a London twin study 37 . We observed that 21% of adult twins and 24% of young MZ twins were discordant for handedness. Different hypotheses have been proposed for handedness discordance: 1) strongly signi cant unique environmental effects and low heritability 10-12 ; 2) adverse prenatal factors 69,70 ; 3) atypical brain lateralization 43,71 ; 4) epigenetic mechanisms (e.g., genomic imprinting) 43 ; 5) late-split twinning 69 , and 6) mirror twinning 72 . However, we did not detect strong methylation differences between discordant MZ twins. Our discordant MZ twin analysis may be underpowered to detect small DNA methylation differences 73 , as it included only 133 MZ discordant adult twin pairs and 86 child twin pairs.
There is a growing interest to improve the prediction of traits with use of other omics data than SNPs, like DNA methylation 46 . Even though single CpGs did not individually reach statistical signi cance in our EWAS, combining information across multiple CpGs into an overall methylation score can be a more powerful approach to capture variation in handedness. Given the low heritability of handedness (~25% 10,11 ), it is expected that non-genetic factors could play role. We calculated methylation scores as weighted sums of the individual's methylation loci beta values of a pre-selected number of CpG sites. However, the predictive value of polygenic and methylation scores for handedness was low, which likely re ects that current GWAS and EWAS analyses for handedness are still underpowered. Future larger GWAS and EWAS studies of handedness are expected to result in more powerful scores.
Our multi-cohort epigenome-wide association study can be summarized in several key steps presented in Figure 3. We examined DNA methylation data in different tissues (whole blood, cord blood, buccal cells) and ages (from birth tto adulthood). The limitations of the study are related to available tissues, handedness measurements, differences in platforms used for DNA methylation (Illumina 450k, EPIC), and study power. Speci cally, the primary tissues of interest for handedness would be brain 2,35 , spinal cord 51 , and arm muscle tissues 4 that were not available in our cohorts. The difference in left-handedness rates among children born before and after 1960 may be due a move away from forcing right handedness prior to 1960 74 . We accounted for this trend by including age (which correlates almost perfectly with birth year in these samples) as a covariate in the analyses, however, it should be noted that the forced use of the right hand in older generations may render the phenotype de nition of handedness less precise.
We reported an EWAS of left-handedness in large population-based cohorts, and examined performance of methylation scores and polygenic scores. Despite the plausible rationale of multiple genetic and non-genetic factors that may be acting via epigenetic pathways to in uence the development of handedness, only two DMRs on chr2p25.1 and chr20q11.23 were associated with lefthandedness in our meta-analysis, but we did not nd prediction of left-handedness by methylation scores. Future EWAS studies, DNA methylation in other tissues related to motor activity and central nervous system would be a valuable addition to our ndings.

Methods And Subjects
Overview. The primary epigenome-wide association study (EWAS) of handedness was performed in two cohorts with DNA methylation data in whole blood (Illumina, 450k): NTR adults (N=2,682 individuals including twins, mean age at methylation 36.5, standard deviation (SD) 12.7), and ALSPAC adults (N=1,232, mean age at methylation 48.98, SD 5.55). EWAS analyses were performed in each dataset separately, and summary statistics were combined in the meta-analysis (N=3,914) testing 409,563 CpGs. Further, we tested whether EWAS signal was enriched nearby loci detected in the previous GWAS on handedness 13 . We carried out within-pair twin analysis in MZ twins discordant for handedness (N adults = 133 twin pairs, N children = 86 twin pairs). Secondary analyses were performed in different tissues: in cord blood and peripheral blood in ALSPAC children (N=1,021 with DNA methylation data at birth, at 7, 17 years old, Illumina 450k chip, and/or at 24 years old, Illumina EPIC array) and in buccal cells in NTR children (N=946 twins, mean age 9.5, SD 1.85, Illumina EPIC array). We performed EWAS analyses in each dataset and examined correlations between the effect sizes of top CpGs (de ned as CpGs with the lowest p-value) in each analysis. Finally, we created and tested polygenic and DNA methylation scores for left-handedness. The study methodology and design are presented in Fig. 1 and Fig. 3.
Subjects. Primary analysis. The adult NTR Biobank cohort 75  In all other inconsistent cases, handedness was coded as missing. The YNTR surveys include parent reports as well as self-reports (from age 14 onward), and included questions on hand preference at different ages. In the remaining surveys, handedness was assessed using a single item with 3 categories: left-handed, right-handed, and both hands. Exception was the assessment at age 5 when handedness was assessed based on 7 items (drinking from a cup; eating; throwing a ball; drawing on paper; picking up a coin; combing hair; and thumb sucking while asleep) which were then classi ed in the same 3 categories as other measurements. The reports at age 2, 5, 14, 16, and 18 (all with categories right-handed, left-handed, and both hands) were checked for consistency across time to obtain handedness based on YNTR surveys. Non-survey projects assessed handedness with 3 categories (left-handed, righthanded, and both hands), except for one project where 7 items were used which were rated on a 5-point scale (writing, throwing, using scissors, toothbrushing, using fork, using spoon, lighting a match). These 7 items were recoded into a single 3 category item using an algorithm similar to the one used for the age 5 questionnaire. Finally, ANTR surveys, YNTR surveys, and projects were combined. The ANTR measurement was used, if not inconsistent with project handedness. If this was not available, YNTR handedness was used, if not inconsistent with project handedness. The nal variable was coded left-handed, right-handed, and both hands (that included ambidextrous and mix-handed individuals).
ALSPAC. Child handedness was assessed at 42 months by questionnaire in which the mother was asked which hand the child used to draw, throw a ball, color, hold a toothbrush, cut with a knife, and hit things (6 questions). Responses were scored -1, 0 or 1 for left, either or right, respectively. Those with score sums from -6 to -4 were labelled left-handed and those with sums from 4 to 6 were labelled right-handed. Mothers and fathers were similarly asked which hand they used to write, draw, throw, hold a racket or bat, brush their teeth, cut with a knife, hammer a nail, strike a match, rub out a mark, deal from a pack of cards or thread a needle (11 questions).
Responses were scored -1, 0 or 1 for left, either or right, respectively. Those with score sums from -11 to -7 were labeled left-handed and those with sums from 7 to 11 were labeled right-handed. Individuals with scores outside these ranges were labeled ambidextrous or mixed-handed and excluded from this study. Handedness was coded as 1 for left-handed or 0 for right-handed in both cohorts.
DNA methylation. NTR adults. The NTR blood DNA methylation data was generated as part of the Biobank-based Integrative Omics Study (BIOS) consortium 76,85 . Blood collection procedures were described previously 75  Hardy-Weinberg equilibrium p-value was < 1×10 -6 , and/or if the call rate was < 0.95. Subsequently, for each platform, the genotype data was aligned with the 1000 Genomes reference panel using the HRC and 1000 Genomes checking tool, which tests and lters for SNPs with allele frequency differences larger than 0.20 as compared to the CEU population, palindromic SNPs and DNA strand issues. The data of the six platforms was then merged into a single dataset, keeping all quality-controlled SNPs of each platform. For each individual, one platform was chosen. Based on the ~10.8k SNPs that all platforms have in common, DNA Identity By Descent state was estimated for all individual pairs using the Plink and King programs. CEU population outliers, based on per platform 1000 Genomes PC projection with the Smartpca software 94 , were removed from the data. Data were phased per platform using Eagle, and then imputed to 1000 Genomes and Topmed using Minimac following the Michigan imputation server protocols 95 . For the polygenic scoring the imputed data were converted to best guess data, and were ltered to include only ACGT SNPs, SNPs with MAF > 0.01, HWE p > 10 -5 and a genotype call rate > 0.98, and to exclude SNPs with more than 2 alleles. All Mendelian errors were set to missing. 20 PCs were calculated with Smartpca using LD-pruned 1000 Genomes-imputed SNPs that were also genotyped on at least one platform, had MAF > 0.05 and were not present in the long-range LD regions.
ALSPAC. Genetic data were collected from the blood samples obtained in clinic visits. Genotyping was conducted with the Illumina HumanHap550 quad chip for children and the Illumina human660W-quad array for mothers. Quality control measures were carried out and haplotypes estimated using ShapeIT. A phased version of the 1000 genomes reference panel from the Impute2 reference data repository was used, and imputation of the target data was performed with this, using all reference haplotypes. Following imputation, variants were retained only given info scores > 0.8 and minor allele frequency > 0.01. Retained variants were then converted to bestguess genotype calls. To avoid potential confounding due to relatedness, closely related individuals were removed using GCTA with a GRM cutoff of 0.05.

Statistical analysis
Intergroup differences. We tested if there were differences in characteristics that were included in EWAS models (such as age at biosample collection, sex, BMI, smoking status at blood collection for adults, and gestational age, maternal smoking during pregnancy, birth weight for children, cell proportions/percentages in buccal swabs and in blood samples) between left-handed and right-handed individuals by generalized estimating equations (GEE) to accommodate the relatedness among the twins in NTR, and by standard logistic regression in ALSPAC. The R package 'gee' was used with the following speci cations: binomial (for ordinal data) link function, 100 iterations, and the 'exchangeable' option to account for the correlation structure within families and within persons.
Right-and lefthanded MZ discordant twins were compared with paired t-test for the traits that were not identical in twins (birth weight, BMI, smoking, cell percentages).
Epigenome-Wide Association Analyses. Primary analyses. The association between DNA methylation levels and left-handedness was tested for each site under a linear model (ALSPAC) or generalized estimating equation (GEE) model accounting for relatedness of twins (NTR). DNA methylation β-value was the dependent variable, and the following predictors were included in the basic model: handedness (coded as 0=right-handed and 1=left-handed), sex, age at blood sampling, percentage of blood cells (monocytes, eosinophils, neutrophils) for blood samples, sample plate and array row in NTR. In ALSPAC, the predictors were handedness (coded the same way), sex, age at blood sampling, percentage of blood cells, and surrogate variables (n=20) were included as predictors. An adjusted model was tted to account for BMI and smoking status at blood draw in both NTR and ALSPAC adult cohorts, because BMI and smoking are known to have large effects on DNA methylation in adults 49,56 and were associated with handedness in some studies 64,65 . The primary results reported in the paper are based on the fully adjusted model. The models are described in Appendix 2.
Secondary analyses. The same basic models were tted to the data from ALSPAC and NTR children. For DNA methylation in buccal cells, cell proportions (epithelial cells, natural killer cells) for buccal samples were included instead of percentage of blood cells. As several characteristics showed association with handedness in previous studies 64,65 and effect on DNA methylation 57,58 , they were included in adjusted model in children (gestational age and birthweight, see Appendix 2).
In the within-pair analysis of discordant MZ twins, paired t-tests were employed to test for methylation differences between the lefthanded and the right-handed twins. Paired t-tests were performed in R on residual methylation levels, which were obtained by adjusting the DNA methylation β-values for sample plate, array row, cell proportions in buccal samples (epithelial cells, natural killer cells) in children and sample plate, array row, and percentages of blood cells (monocytes, eosinophils, neutrophils) in adults.
Additional covariates, birth weight in children and BMI and smoking status in adults, were added in adjusted model. Age, sex, maternal smoking, and gestational age were not included because these variables are identical in MZ twins.
To account for multiple testing, we considered Bonferroni correction and a False Discovery Rate (FDR) of 5%. The Bonferroni corrected p-value threshold was calculated by dividing 0.05 by the number of genome-wide CpGs tested, and false discovery rate (FDR) q-values were computed with the R package 'qvalue' with default settings. The Bayesian in ation factor (λ) was calculated with the R package Bacon 96 (see Table S5).  Differentially methylated regions. We used the R dmrff library for R 45 to identify regions where multiple correlated methylation sites showed evidence for association with handedness. Dmrff identi es DMRs by combining EWAS summary statistics from nearby CpG sites with methylation datasets to compute correlations between CpGs (https://github.com/perishky/dmrff). Dmrff was applied in each cohort separately. For the meta-analysis, we calculated CpG site correlations separately for NTR adults and ALSPAC adults (parents) cohorts, and performed DMR meta-analysis. We applied an adjusted p-value that was a p-value multiplied by the total number of tests performed with the number of tests equal to the number of regions for which DMR statistics are calculated. We report signi cant regions (p adj < .05) with at least two methylation sites within a 500bp window. We plotted the DMRs with the coMET R Bioconductor package 98 to graphically display additional information on physical location of CpGs, correlation between sites, statistical signi cance, and functional annotation (annotation tracks included genes Ensembl, CpG islands (UCSC), regulation Ensembl).

Meta-analysis. A meta-analysis was performed in METAL
GWAS follow-up. GWAS follow-up analyses were performed to examine whether CpGs within a 1 Mb window of loci detected by the GWAS for left-handedness 13 , on average, showed a stronger association with left-handedness than other genome-wide methylation sites (In nium HumanMethylation450 BeadChip). We obtained a SNP list based on the GWAS meta-analysis without NTR, ALSPAC, and 23andMe by Cuellar-Partida et al 13  bootstraps with the R-package "simpleboot". Statistical signi cance was assessed at α = 0.05. As control analysis the same follow-up was performed using GWAS summary statistics on a trait that is unrelated to handedness -type 2 diabetes in UK Biobank cohort (N=244,890) 54 . GWAS summary statistics were downloaded from GWASAtlas (https://atlas.ctglab.nl/traitDB/3686; 41204_E11_logistic.EUR.sumstats.MAC lt.txt; access on February 1 2021).
Polygenic and methylation scores. Polygenic scores (PGS) for handedness were calculated based on the GWAS meta-analysis without 23andMe by Cuellar-Partida et al 13 . To avoid overlap between the discovery and target samples, summary statistics without NTR and ALSPAC were requested (196,419 individuals, N SNPs = 13,550,404). The linkage disequilibrium (LD) weighted betas were calculated using a LD pruning window of 250 KB, with the fraction of causal SNPs set at 0.50by LDpred 101 . We randomly selected 2500 2nd degree unrelated individuals from each cohort as a reference population to calculate the LD patterns. The resulting betas were used to calculate the PGSs in each dataset using the PLINK 1.9 software. All PGSs were standardized (mean of 0 and standard deviation of 1).
Methylation scores (MS) were calculated in NTR based on EWAS summary statistics obtained from ALSPAC, and vice versa, as previously done to create methylation scores for BMI, height, and prenatal smoking 102,103 . We calculated same-tissue same-age DNAmethylation scores based on methylation data from NTR adults (blood) and ALSPAC parents (blood), and cross-tissue DNAmethylation scores based on data from NTR and ALSPAC children, with DNA methylation measured in buccal cells, and blood, respectively (see Fig.3). For each individual, a weighted score sum was calculated for left-handedness by multiplying the methylation value at a given CpG by the effect size of the CpG (β), and then summing these values over all CpGs: DNA methylation score (i) = β 1 *CpG 1i + β 2 *CpG 2i …. + β n *CpG ni , where CpG n is the methylation level at CpG site n in participant i, and β n is the regression coe cient at CpG n taken from summary statistics of the EWAS analysis. All methylation scores were standardized (mean of 0 and standard deviation of 1).
Testing predictive value. We analysed the predictive value of the left-handedness polygenic scores and methylation scores in NTR and ALSPAC adult and child cohorts from our EWAS study. We are extremely grateful to all the families who took part in this study, the midwives for their help in recruiting them, and the whole NTR and ALSPAC teams, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists and nurses.

Author information
These     Values are presented as mean (SD) or n(%).
Current smokers in ALSPAC were de ned as those with cg05575921 methylation below 0.82 (see Methods).    Figure 1 Flowchart of epigenome-wide association study of left-handedness Differentially methylated regions associated with left-handedness in meta-analysis. The top panel of each plot shows the EWAS pvalues for all CpGs in the window, with the most strongly associated CpG highlighted. The middle panel shows the genomic coordinates (genome build GRCh37/hg19) and the functional annotation of the region: the ENSEMBL Genes track shows the genes in the genomic region (orange); the CpG Island track shows the location of CpG islands (green); the Regulation ENSEMBL track shows regulatory regions (blue). CpGs from DMR associated with handedness are indicated with red lines above the correlation heatmap.