The effect of season of birth on brain epigenome-wide DNA methylation of older adults

Abstract Perinatal light exposure predisposes towards health and behaviour in adulthood. Season of birth is associated with psychiatric, allergic, cardiovascular and metabolic problems. It has been proposed that early-life environmental light disrupts the development of biological rhythms which, in turn, influence later-life health. However, the mechanisms linking perinatal seasonal light to later-life biological rhythm and health in humans are unknown. In this study, we investigated the association between season of birth and epigenome-wide DNA methylation of two postmortem human brain regions (16 hypothalamus, 14 temporal cortex). We did not find statistically significant differences at the whole epigenome level, either because we lacked statistical power or that no association exists. However, when we examined 24 CpG sites that had the highest significance or differential methylation, we identified regions which may be associated with circadian rhythm entrainment, cholinergic neurotransmission and neural development. Amongst methylation of the core clock genes, we identified that hypothalamus Neuronal PAS Domain Protein 2 (NPAS2) gene has hypermethylated regions in long photoperiod-born individuals. In addition, we found nominal associations between season of birth and genes linked to chronotype and narcolepsy. Season of birth-related brain DNA methylation profile was different than a previously reported blood methylation profile, suggesting a tissue-specific mechanism of perinatal light programming. Overall, we are the first to analyse the relationship between season of birth and human brain DNA methylation. Further studies with larger sample sizes are required to confirm an imprinting effect of perinatal light on the circadian clock.


Introduction
The concept of The Developmental Origins of Health and Disease states that early-life environmental exposures may influence an individual's behaviours and health in adulthood. 1 For example, maternal care and malnutrition are well-established as being associated with later-life mental health and metabolism. 2,3 Early-life light exposure has also been suggested to predispose towards health issues in adulthood. Epidemiological studies have found that season of birth is associated with psychiatric conditions, 4 allergies, 5 cardiovascular diseases, 6 cancer, 7 metabolic problems 8 and mortality. 9 The Perinatal Light Imprinting of Circadian Clocks and Systems (PLICCS) hypothesis suggests that environmental light exposure may disrupt the development of biological rhythms which in turn influence later-life health. 10 In line with this, day length at birth was associated with chronotype, which is the timing preferences of daily behaviours and which represents the biological rhythm of an individual. 11,12 Moreover, animal models have shown that seasonal daylight duration can modify developing circadian clocks. 13 If mice are reared under a summer solstice-like (16 h light:8 h dark) photoperiod environment, the period of their circadian rhythm of gene expression, the period of their free-running activity rhythm in constant darkness and the period of individual neurons of their principal pacemaker, the suprachiasmatic nucleus (SCN) in the brain, all become shorter in adulthood compared to animals reared under a winter solstice-like (8 h light:16 h dark) environment. 13 Epigenetic regulations are highly dynamic during embryonic and postnatal development, and it has been hypothesised that early-life programming is due to lasting epigenetic remodelling during development. 14 Season of birth effects may also include early-life epigenetic programming. In the hippocampus and olfactory bulb of mice, long day length was associated with lower concentrations of total 5-Methylcytosine and DNA methyltransferase (DNMT) transcription, and higher DNA hydroxymethylation and Ten-eleven translocation (TET2) transcription, which is an indicator of DNA demethylation. 15 SCN genome-wide DNA methylation and transcription were compared between 22-and 26-hour day length entrained mice. 16 Differently methylated DNA regions were enriched in pathways of synapse formation, neuroendocrine s ystem, axon pathfinding and neurophysiology. 16 A methyltransferase inhibitor drug prevented adaptation to day length cycle. 16 This study is important as it provides evidence supporting circadian epigenetic programming in the mammalian SCN during entrainment to new period length. However, the effect of seasonal day length on epigenetics of the SCN is still largely unknown. In the only human study, blood samples from 367 individuals (age 18) were analysed for global methylation variations in relation to their season of birth. 17 Allergic diseases were found associated with autumn-born participants in this cohort, and 92 seasonal differentially methylated CpG sites were identified, two of which were causally associated with allergy. 17 Involvement of seasonal epigenetic programming in PLICCS has not been characterised in the human hypothalamus.
In this study, we investigate epigenome-wide DNA methylation differences associated with seasonal photoperiod at birth in the human hypothalamus, which includes biological rhythm and sleep centres, and the inferior temporal gyrus (ITG), which has previously been reported as a candidate brain region for season of birth effect in a brain imaging study. 18

Subjects and sample preparation
Subjects in this study were participants of The University of Manchester Longitudinal Study of Cognition in Normal Healthy Old Age (UMLCHA) cohort. 19 Starting in 1983, this study recruited 6375 healthy individuals without dementia aged between 40 and 90 years. The participants were followed for up to 35 years for changes in self-reported health, cognition, demographics and lifestyle data including diurnal sleeping habits 11 under the approval of The University of Manchester research ethics committee with written consent from all participants. Participants were approached in 2003 for consent to brain donation and 312 individuals agreed to donate their brain after death. 20 One hemisphere of the postmortem brains was fresh frozen at −80°C and is stored by The Manchester Brain Bank (https://www.bmh.manchester.ac.uk/ research/domains/neuroscience-mental-health/manchester-brainbank), which has ethical approval to distribute tissue and data to researchers. The present study was approved by the Manchester Brain Bank Management Committee (REC reference 19/NE/ 0242; application number TR83).
Amongst the donated brains, only 16 samples had intact hypothalamus and ITG regions because of tissue sectioning and storing procedures, and usage of the tissues in other studies. These 16 individuals were unrelated with 10 being female. The demographic and brain pathology data of subjects are summarised in Table S1. Hypothalamic tissue blocks were cut into sections with 300 μm thickness using a cryostat (Leica CM 1950). ITG tissue blocks and hypothalamic sections were stored in a −80°C freezer until use. The hypothalamus region was identified by means of a human brain atlas, 21 and 10 punches (1 mm diameter) per participant were obtained. From the ITG tissue blocks, 30-50 mg of tissue was cut. Genomic DNA was isolated with a PureLink Genomic DNA mini kit (Invitrogen, USA). Agarose gel electrophoresis was performed to confirm that DNA was not degraded. For all participants, quantity of extracted DNA was verified to be >500 ng. Bisulphite conversion of DNA was performed using the EZ DNA Methylation™ Kit, following the manufacturer's protocol (Zymo Research, USA). Epigenome-wide methylation analysis was conducted using the Infinium MethylationEPIC BeadChip using the protocol provided by the manufacturer (Illumina Inc., USA). The arrays were scanned using the iScan System, and signal intensities were recorded as IDAT files by GenomeStudio (Illumina Inc., USA).

Data filtering
Quality control and analyses were carried out in R version 4.0.2 (R Foundation for Statistical Computing, Vienna, Austria). IDAT files were imported and preprocessed using the R package minfi (version 1.34.0). 22 Two ITG samples were excluded because their mean detection P-values were above 0.01. Sex chromosomes were removed from the analysis. Probes that had a mean detection P-value above 0.01 were also excluded. Probes which were cross reactive, or which contained a single nucleotide polymorphism (SNP) on the CpG of interest, were removed. 23 Genomic locations were reported in Genome Reference Consortium Human Build 37 (GRCh37). After quality control, 771,382 CpG probes in 30 samples remained. β-values, the ratio of methylated intensity to total intensity, and M-values, the logit transformed β, were calculated and normalised using SWAN. 24 Density plots of β-values were controlled for bimodal distributions (Fig. S1). Principal component analysis was performed using the R package pcaMethods (version 1.80.0) to identify potential confounders such as batch effect and tissue type. 25 Cell type heterogeneity was inspected using the R packages RefFreeEWAS (version 2.2.0) 26 and FlowSorted. DLPFC.450k (version 1.24.0). 27

Epigenome-wide association analysis
Linear models were calculated to examine differentially methylated probes (DMPs) between hypothalamus and ITG using the R package limma (version 3.44.3). 28 Since different tissue samples were taken from the same individual, a paired analysis was performed to adjust within-individual differences. P-values were adjusted for multiple testing using the Benjamini and Hochberg method to control the false discovery rate (FDR). Significant hypothalamusspecific methylation profiles (FDR < 0.05) were compared with human lateral hypothalamus DNA methylation, 29 human hypothalamus gene expression in The Genotype-Tissue Expression (GTEx) database, 30 and mice suprachiasmatic nucleus gene expression 31 using Spearman's correlation (Stata Statistical Software: Release 14. College Station, TX: StataCorp LP.).
Season of birth was categorised as long photoperiod (LP), March 21-September 22, and short photoperiod (SP), September 23-March 20. Postmortem brains from 8 SP-born and 8 LP-born subjects were used in this study. Linear models were fitted to find DMPs in season of birth groups. Two separate models were fitted for hypothalamus and ITG. Sex and age at death were included as covariates in the models. DMPs with adjusted P-value < 10 −5 were accepted as a suggestive association. Power calculations for DNA methylation studies using the Illumina EPIC microarray were performed using the University of Essex EPIC Array Power Calculations. 32 In addition, probes that had a

Season of birth-associated CpGs and chronotype
With season of birth being correlated with later-life midsleep time, 11 we investigated whether nominally CpGs associated with season of birth predict later-life sleep time. Self-reported sleep timing data were available for 15 subjects. Midsleep time, a measure of chronotype, was used as an outcome for the linear regression model. Candidate CpGs (P < 10 −5 or P < 0.05, logFC < −3, logFC > 3) were included in the regression models as the main predictor, adjusted for age and sex. P-value below 0.0017 was accepted as significant with Bonferroni correction of multiple testing.

Pathway analysis and correlations with previous findings
The top 1000 CpGs in terms of logFC from the season of birth epigenome-wide results were tested for potential enrichment in biological pathways, molecular function or cellular components using the Gene Ontology database. 34 The pathway analysis was performed using the R package missMethyl (version 1.22.0) with a P-value threshold <10 −5 used to determine statistical significance. 35 Hypothalamus and ITG methylation profiles that were related to season of birth in this study were compared with previously reported season of birth DNA methylation in blood 17 using Spearman's correlation (Stata Statistical Software: Release 14. College Station, TX: StataCorp LP.). In addition, overlapping methylated CpGs between our season of birth methylation (P < 0.05) and season of birth-associated disorders (Frontal cortex DNA methylation in schizophrenia 36 and Lateral Hypothalamus DNA methylation in narcolepsy 29 ) were investigated using a chi-squared test.
Candidate probes were further investigated for their association with other phenotypes using the EWAS Atlas. 37 Our candidate probes were also compared with primate brain diurnal or seasonal gene expression and DNA methylation, 38,39 fetal and early-life DNA methylation in blood and brain, 36,40,41 and season of birthassociated placenta gene expression. 42

MeQTL analysis
Blood samples of the UMLCHA cohort were previously genotyped and imputed to The Haplotype Reference Consortium. 43 A total of 15 subjects had both DNA methylation and genotype data. Using PLINK v2, SNPs located within 1 million bases around the CpGs that had P-value <10 −5 were extracted. Cis-acting methylation quantitative trait loci (meQTL) were investigated using the R package MatrixEQTL (version 2.3.0) with a P-value threshold <10 −5 to determine statistical significance. 44 In addition, previously reported meQTLs in the developing human brain were investigated for our candidate loci. 45 Technical replication using pyrosequencing Amongst the strongest associated CpG regions, four hypothalamus probes (cg12323699, cg18445760, cg12833267 and cg25628433) and four ITG probes (cg01839373, cg03370588, cg16402757 and cg14859874) were selected as target regions for technical replication using pyrosequencing. Predesigned PyroMark CpG Assays for validating results from Infinium MethylationEPIC BeadChip arrays were used (Qiagen, Germany). Before pyrosequencing, PCR amplification was performed using 10 ng of bisulphiteconverted DNA and the PCR product was cleaned using Streptavidin Sepharose High Performance Beads (GE Healthcare, UK). Pyrosequencing was performed using PyroMark Q24 (Qiagen, Germany). β-values from epigenome-wide arrays were compared with the methylation percentage from pyrosequencing using Spearman's correlation.

Cellular heterogeneity
Since brain samples include different cell types, association analysis may be confounded. Therefore, we used a reference-free deconvolution method to predict cell type proportions using methylation profiles. The estimation revealed that there were two main cell types in the tissue. A proportion of neural cells was subsequently estimated using publicly available data, FlowSorted.DLPFC.450k (version 1.24.0). 27 One of the two estimated cell types clustered with neural cells, while the other type clustered with non-neural cells. We found the estimated proportion of neurons to be higher in ITG (26%) compared to hypothalamus (18%) (P = 5.4 × 10 −3 ). The higher glia:neuron ratio of the hypothalamus compared to cortex has been previously shown in primates. 46 The proportion of neurons did not significantly differ between SP and LP season of birth groups (Fig. 1) in the hypothalamus (P = 0.33) and ITG (P = 0.78).
A total of 75 CpGs in our analysis have been previously reported as season of birth-related blood DNA methylation markers. 17 Neither the hypothalamus (Spearman's Rho = −0.12; P = 0.30) nor the ITG (Spearman's Rho = 0.15; P = 0.19) results were correlated with blood (Fig. S5). In addition, schizophrenia frontal cortex DMPs 36 were not enriched in our hypothalamus and ITG results. Narcolepsy lateral hypothalamus DMPs 29 were not enriched in our ITG results, but they were associated with our hypothalamus season of birth DMPs (OR = 1.53; chi2(1) = 102.66; P = 4.0 × 10 −24 ). 6458 DMPs (P < 0.05 and |Δβ| > 0.05) have been previously reported. Amongst these, 6143 CpGs were available in our data set, in which 630 probes had P-value lower than 0.05 in season of birth analysis. In addition, GNAS is one of the top ten DMRs for narcolepsy (Table S5).

Season of birth-associated CpGs and later-life chronotype
The UMLCHA cohort has data available for sleep timing (mean age 58.76 (SD = 4.14) years). The average midsleep time was 3:30 am (SD = 26 min). After Bonferroni correction for multiple testing (0.05/29 = 0.0017), P-values did not provide enough evidence of association. Methylation of cg14859874 within the UBAP2L gene in both tissues showed nominal significance to estimate later-life chronotype (Fig. 3). Increased DNA methylation in the hypothalamus was associated with a 30 min earlier chronotype (B = −0.09; SE = 0.04; p = 0.04) and in the ITG, it was associated with a 45 min earlier chronotype (B = −0.13; SE = 0.05; P = 0.02).

Pathway analysis results
We found no significant enrichment for any biological pathways. The most significant P-value for hypothalamus season of birthrelated genes belonged to the biological function of "Specification of animal organ identity" (GO:0010092, P = 4.6 × 10 −4 ). The top

MeQTL analysis results
Methylation β-values of the DMPs that had a P-value lower then 10 −5 were interrogated for cis-meQTL separately for each tissue. No SNPs significantly estimated DNA methylation in our study.
We searched our target CpGs (P < 10 −5 or |Δβ| > 3) in imputed Bonferroni significant meQTLs in fetal brain. We found that methylation of cg18445760, which is located within the SLC44A4 gene, was associated with 39 linked SNPs. The strongest associated variant rs660550 was found to be associated with ITG methylation in our data after adjusting for age and sex (B = −0.018; P = 6.0 × 10 −5 ).

Discussion
We investigated associations between photoperiod at birth and human hypothalamus and ITG epigenome-wide DNA methylation patterns. This study is the first of its kind that presents prospective results in the field of early-life seasonal programming via human brain epigenetic modifications. Small sample size may have limited our analyses and results. Nevertheless, suggestive weak associations in this study insinuated that season of birth may generate lasting epigenetic changes, with distinct mechanisms, in different tissues. These results are also in line with the PLICCS hypothesis, since season of birth was weakly associated with hypothalamus DNA methylation, which includes the principal biological clock, and the suggestive significant genes we found may be involved in circadian rhythm pathways. Moreover, reported candidate CpG regions were also enriched in narcolepsy-related DNA methylation, which suggests a relationship between season of birth and narcolepsy in adulthood. Despite limitations, this study is the first to examine the imprinting effects of perinatal light on the human brain and presents unique data that can be used in future meta-analyses and an in-depth evaluation studies of early-life seasonal programming. Season of birth is associated with traits and behaviours such as height, pubertal timing, educational attainment, biological rhythms and lifetime disease risk. 5,9,12,48 Day length sunlight exposure is the main seasonal cue, but there are multiple seasonal factors that may be involved in season of birth effect such as temperature, exposure to infectious agents and nutrition. [49][50][51][52] Seasonal factors may convey their effects via lasting epigenetic modifications, but the exact mechanism of how seasonal cues regulate specific targets is unknown. The strongest candidates are melatonin, thyroid hormone and vitamin-D pathways, all of which are seasonal and have been shown to regulate epigenetic modifier enzymes. [53][54][55][56] Depressive behaviour in mice is associated with postnatal short day length, but when Melatonin receptor 1 was silenced, the mutant animals did not show any postnatal programming in their adulthood depressive behaviour, or midbrain concentration of serotonin and norepinephrine. 57 In humans, blood epigenomewide DNA methylation was shown to be postnatally modified by season of birth. 17 Specifically, season of birth-associated differentially methylated CpG sites were enriched in development, the cell cycle and apoptosis pathways and were causally related to allergic diseases. 17 Neither hypothalamus nor ITG DNA methylation in our study was correlated with season of birth blood DNA methylation patterns. Taken together, these results indicate that early-life seasonal factors can influence epigenetic regulation but there is still little evidence about the exact mechanisms. Furthermore, perinatal programming mechanisms may vary in different tissues. The circadian biological clock is an endogenous pacemaker that regulates 24-hour rhythms in physiology and behaviour and has the ability to synchronise to the outer rhythmic cues such as light. 58 A negative feedback loop of the molecular clock, located within the neurons of the hypothalamic suprachiasmatic nucleus in mammals, maintains the self-oscillation. 58 Photic input from the eyes entrains the circadian pacemaker and sets the phase of the oscillation. 58 The circadian system is not completely mature at birth; therefore, environmental exposures may programme the later-life function of the biological clock. 59 The PLICCS hypothesis suggests that the predisposition of the biological rhythm to be programmed by perinatal factors may contribute towards the association between season of birth and later-life health problems. 10 In line with this, epidemiological studies have shown associations between season of birth and later-life circadian rhythms, and circadian rhythms and health. Narcolepsy is associated with spring season of birth, 60 insomnia with spring and summer season of birth, 18 and evening chronotype with summer season of birth. 12 Furthermore, evening-oriented chronotype is associated with psychiatric and metabolic diseases. 11 Similarly, disruption of biological rhythms and sleep is associated with negative health outcomes and mortality. 61,62 Therefore, to examine the perinatal seasonal programming of biological rhythms, we investigated the human hypothalamus.  We found that the NPAS2 gene was the clock gene in the hypothalamus to be most affected by season of birth, with nine CpGs being hypermethylated in LP-born individuals, which points towards a downregulated gene expression in individual born in the summer/spring. The NPAS2 protein is a paralogue of CLOCK, and after dimerisation with BMAL1, it triggers transcription of the PER and CRY clock genes in brain. 63 Knock-out of NPAS2 in mice results in a shortened circadian period, altered response to light and sleep disturbances. 63 In line with this, LPreared mice show shortened periods of behavioural and SCN PER1 gene expression rhythms compared to SP-reared animals. 13 Another factor that influences circadian rhythm period is stability of clock proteins due to degradation. 64 A suggestive significant gene in our analysis was CUL2, which targets proteins for degradation by ubiquitination. 65 Cullins are scaffold proteins in the ubiquitin ligases complex and have been shown to work in circadian regulation. 64,66 These findings support the idea that the earlylife photoperiod may influence the period of the biological rhythms in adulthood via epigenetic regulations.
Another characteristic of the circadian clock is its ability to entrain to environmental light cycles. However, responses of individuals to the same environmental input vary. Interindividual variations in sensitivity and response to environmental light have been described in humans, where evening light exposure suppressed melatonin expression differently between individuals. 67 Furthermore, phase differences of circadian genes significantly predicted chronotype. 68 In addition, LP-reared mice had invariant rhythm across different seasons compared to SP-reared animals. 13 Therefore, if early-life light experiences imprint future sensitivity of synchronisation with natural cues, this may explain the association between season of birth and later-life circadian rhythms. Photic entrainment of the circadian clock starts with the retinohypothalamic track conveying light input from the eyes to the SCN through glutamate and PACAP neurotransmission. 69 Conveyed light input activates Ca 2þ influx and cAMP signalling cascades, in turn light-induced response genes, carrying cAMP response elements (CRE) in their promoters, undergo expression and provide the phase shift. 69 One of the differentially methylated regions in our analysis was the GNAS gene, which was generally hypermethylated in LP-born individuals and is known for its high expression in seasonality-related tissues such as thyroid, pituitary and hypothalamus. GNAS is one of the major regulators of adenylyl cyclases that produce cAMP. 70 Through Ca 2þ /cAMP-dependent circadian entrainment pathways, several protein kinases are involved and another nominal significant gene in our analysis is PKIB, which is a peptide that can inhibits cAMP-dependent protein kinase function. 69,71 Taken together, lasting epigenetic modifications during postnatal period may alter GNAS and PKIB expression in adulthood and may influence later-life sensitivity to light.
Narcolepsy is a sleep disorder which is characterised by a loss of orexin neurons, likely due to an autoimmune process. 29,60 Genome-wide associations support the involvement of the immune system, and epigenetic analyses indicate differential methylation in myelin production. 29 Narcolepsy is a season of birth-associated disease. 60 Season of birth is associated with blood DNA methylation, immune responses and allergies 17 ; therefore, it is possible that narcolepsy may also be imprinted via seasonal epigenetic programming. We found that previously described narcolepsy-associated DNA methylation in the lateral hypothalamus 29 has a significant overlap with our season of birth methylation analysis in the hypothalamus. These findings point towards a causative role of season of birth methylation in narcolepsy. In addition, one of the candidate CpGs in our analyses, cg14859874 within the UBAP2L gene, was associated with both season of birth and later-life chronotype. This may indicate a role of UBAP2L gene methylation by season of birth on later-life circadian behaviour. UBAP2L protein is responsible for the formation of stress granules, which are cytoplasmic dense membraneless structures of mRNA and proteins formed after translation inhibition due to a stress exposure. 72 Stress granules are associated with Alzheimer's disease pathology and show circadian rhythmic characteristics. 73 This may also be a link between season of birth and neurodegenerative disorders. Taken together, these findings suggest that early-life seasonal epigenetic programming may have a role in mechanisms linking season of birth and disease.
Early-life light exposure is associated with neurotransmitter levels in adulthood, such as serotonin, norepinephrine and dopamine. 57,74,75 This differential neurotransmitter expression may be due to epigenetic imprinting and may alter health and behaviours in adulthood. In mice, different day length exposure during development resulted in hyper or hypomethylation of genes enriched in synapse formation, neuroendocrine system, axon pathfinding and neurophysiology in the SCN. 16 In line with this, a candidate differentially methylated gene in our analyses was SLC44A4, which expresses a choline transporter protein, and was hypermethylated in LP-born individuals. The cholinergic neurotransmission is involved in modulation of wakefulness and REM sleep. 76 Even if SLC44A4 has low transcription in brain tissue, it may have a role in brain development and neural tissue. 77,78 Season of birthassociated CpG at the SLC44A4 gene was shown to be significantly more methylated in postnatal human brain compared to fetal. 36 The SLC44A4 is also differentially expressed in the placenta in different seasons of birth. 42 Taken together, season of birth may programme neurotransmitter levels in adulthood.
Some of the candidate genes in the hypothalamus have been previously reported to have roles in brain development. MYO5A is a motor protein that is abundant in neurons and carries synaptic neurotransmitter vesicles, and deficiency of it is associated with developmental brain abnormalities. 79,80 The Ventral Anterior Homeobox 1 gene is responsible for the development of the anterior ventral forebrain and visual system. 81,82 The Homeobox A3 gene has a role in the development of the thyroid gland. 83 These studies suggest a role of season of birth epigenetic modulation in brain development.
Season of birth has previously been associated with grey matter volume of the left superior temporal gyrus in adult human males. 84 The temporal fusiform cortex and inferior temporal gyrus were also shown as suggestively significant with season of birth, with a summer peak. 18 A reduction in the grey matter volume of the temporal gyri has been linked to schizophrenia and narcolepsy, 85,86 both of which are associated with season of birth. 4,60 Therefore, it is possible that season of birth modifies the brain structure in these regions and in turn influences later-life health. We hypothesised that the epigenome of the ITG may be influenced by photoperiodic history and associated with later-life health. We found that neither schizophrenia nor narcolepsy-associated DNA methylation was correlated with ITG season of birth methylation levels.
One of the nominal significances of biological pathways in the ITG was 'Retina development in camera-type eye'. The ITG has a role in visual processing of faces, colours and shape. 87 This may indicate that season of birth epigenetic programming may influence later-life role of the ITG. In addition, when we examined the probes that had the highest fold change, we observed an overlap between ITG and hypothalamus. We found that the CpGs at CUL2,
MYO5A and UBAP2L genes had the highest methylation differences in both ITG and hypothalamus, which indicates a possible common mechanism for these regions. There are some limitations in our study. First, the small sample size limited the statistical power. This study has over 80% power to detect a mean difference of 10% in only 33% of tested sites using a 10 −5 P-value threshold. 32 Season of birth effect has a small effect size which may make it difficult to detect in a small sample. 49,88 Even if we adjusted our analyses for age and sex, other sociocultural, genetic, environmental and latitudinal factors may confound the analyses. 7,43,49,88 We categorised the photoperiod of birth as LP and SP, but using real seasonal day length would provide better insight to postnatal light exposure. Our hypothalamus DNA methylation is correlated with hypothalamus gene expression and DNA methylation in the literature, which support the validity of our tissue blocks. However, the human brain tissue used in the study was slow-frozen, which made it difficult to identify hypothalamus subregions. The hypothalamus comprises many nuclei with distinct functions, cellular and molecular compositions. Therefore, to understand the real association between circadian rhythms and DNA methylation, SCN-specific methylation should be investigated. Furthermore, cellular heterogeneity is one of the major limitations when analysing whole tissue. Nevertheless, we show that neuron and non-neuron proportions were similar in season of birth groups. Finally, DNA methylation may be daily or seasonally rhythmic. 39 However, none of the reported suggestive significant genes or CpG probe showed circadian or seasonal characteristics in the diurnal transcriptome atlas of baboon SCN 38 and human prefrontal cortex. 39 In summary, perinatal light exposure may predispose physical and mental health in adulthood. Despite the limitations of the present study, our results suggest that season of birth may programme later-life health and behaviour via brain DNA methylation. In addition, different mechanisms appear to underlie season of birth effects in different tissues and health disorders. Importantly, the photoperiod around birth may influence the development of the circadian rhythm system, which is essential for health. Further studies are required to fully understand the mechanisms behind early-life programming, as a key first step to preventing, diagnosing and treating future health problems in predisposed individuals.
Supplementary material. To view supplementary material for this article, please visit https://doi.org/10.1017/S2040174421000453