Early-to-mid pregnancy sleep and circadian markers in relation to birth outcomes: An epigenetics pilot study

ABSTRACT Maternal sleep and circadian health during pregnancy are emerging as important predictors of pregnancy outcomes, but examination of potential epigenetic mechanisms is rare. We investigated links between maternal leukocyte DNA methylation of circadian genes and birth outcomes within a pregnancy cohort. Women (n = 96) completed a questionnaire and provided a blood sample at least once during early-to-mid pregnancy (average gestation weeks = 14.2). Leukocyte DNA was isolated and DNA methylation (average percent of methylation) at multiple CpG sites within BMAL1, PER1, and MTNR1B genes were quantified by pyrosequencing. Birth outcomes including gestational age at delivery, birthweight, and head circumference were abstracted from medical charts. Linear regression analyses were run between each CpG site with birth outcomes, adjusting for important confounders. Sleep duration and timing were assessed as secondary exposures. Higher methylation of a CpG site in PER1 was associated with smaller log-transformed head circumference (β=-0.02 with 95% CI −0.02 to 0.01; P, trend = 0.04). Higher methylation of MTNR1B (averaged across sites) was associated with lower log-transformed birthweight (−0.08 with 95% CI −0.16 to −0.01; P, trend = 0.0495). In addition, longer sleep duration was associated with higher birthweight (0.10 with 95% CI 0.02 to 0.18 comparing > 9 h to < 8 h; P, trend = 0.04). This pilot investigation revealed that higher methylation of PER1 and MTNR1B genes, and sleep duration measured in early-to-mid pregnancy were related to birth outcomes.


Introduction
Duration and timing of sleep during pregnancy are emerging as potentially important predictors of pregnancy outcomes.Shorter sleep duration during pregnancy has been related in some studies to pre-term birth (Loy et al. 2020) and to lower infant birthweight (Warland et al. 2018).Another study of >7000 US women showed that later sleep timing (as a marker of circadian misalignment, where sleep/wake cycles are out of sync with the underlying circadian rhythm), but not duration, was related to higher odds of pre-term delivery (Facco et al. 2019).Similarly, a recent study showed that later sleep timing during pregnancy was associated with higher adiposity among offspring at 2 years of age (Meng et al. 2022), whereas there were no associations with sleep duration or quality.Furthermore, short sleep duration and later sleep midpoint (median of bedtime and wake time) have each been associated with development of gestational diabetes (Facco et al. 2017(Facco et al. , 2018;;Reutrakul et al. 2018) which is known to affect risk of preterm birth, large-for-gestational-age size, hypoglycemia, and other birth complications (Tehrani et al. 2022).However, inconsistencies in the research remain, especially regarding whether sleep duration or sleep timing have truly independent effects on fetal outcomes.In addition, mechanistic studies, which are ultimately needed to understand the causal nature of these relationships, remain scarce.
One way to investigate potential mechanisms by which maternal sleep duration and timing during pregnancy could impact fetal outcomes is through an examination of epigenetic alterations as a mediator.Epigenetic alterations refer to mitotically heritable changes that regulate gene expression in the absence of underlying changes to the DNA sequence itself.One way this occurs is through DNA methylation (DNAm), the addition of a methyl group to a cytosineguanine dinucleotide (CpG) site.Studies in nonpregnant populations have shown that short and delayed sleep timing may impact DNA methylation.A night of full sleep deprivation has been shown to alter adipose and lean muscle DNA methylation across a wide range of genes (Cedernaes et al. 2015(Cedernaes et al. , 2018;;Nilsson et al. 2016).Regarding sleep timing, studies among nurses have shown differences in leukocyte DNA methylation of circadian genes including PER genes (1-3), BMAL1, and CSNK1E among those who work night shifts compared to day workers (Bhatti et al. 2015;Ritonja et al. 2022).Furthermore, a study we carried out in adolescents recruited from Mexico City showed that while sleep timing was related to DNAm of several circadian genes, sleep duration was not strongly linked to any of the 12 circadian genes investigated (Jansen et al. 2021), which supports the notion that sleep duration and sleep timing may affect the epigenome in different ways.Only a few studies have thus far been conducted among pregnant populations.A recent study showed that women with lower-quality sleep in the third trimester had higher placental DNA methylation of melatonin receptor 1B (MTNR1B) and lower placental protein expression of melatonin receptors (Lin et al. 2022).Another study conducted in China found that pregnancy sleep midpoint was associated with 45 differentially methylated probes in infant cord blood tissue, whereas there were no relationships with sleep duration or quality (Meng et al. 2022).Yet, neither of these studies investigated associations between pregnancy sleep-related DNA methylation and birth outcomes.Nonetheless, separate studies suggest that the maternal epigenome could affect birth outcomes (Das and Maitra 2021), including preterm birth (Jain et al. 2022), although the role of genes found to be related to sleep behavior has not been specifically examined.
In order to fill these research gaps, we conducted a pilot study within a Michigan (United States) pregnancy cohort to investigate (1) links between early-to -mid pregnancy sleep duration and timing with maternal leukocyte DNA methylation of the genes BMAL1, PER1, and MTNR1B and (2) associations between early-to-mid pregnancy sleep duration, timing, and circadian gene DNA methylation with birth outcomes (gestational age at delivery, birthweight, and head circumference).These specific genes were selected on the basis of their known functions in sleep/circadian rhythms (BMAL1 and PER1 are part of the core circadian loop governing the 24-h sleep/ wake cycle, and MTNR1B helps encode melatonin, an antioxidant that also aids in the regulation of sleep onset and duration) and for previous associations with perinatal outcomes (Lin et al. 2022) or cardiometabolic indicators (Samblas et al. 2016;Yu et al. 2019).

Study population
The study population included pregnant women at Michigan Medicine enrolled in the BUMP (Biorepository for Understanding Maternal and Pediatric Health) study who gave birth between 19 August 2019 and 21 September 2020.All pregnant women who received care at Von Voigtlander Women's Hospital aged 18 years or older were eligible to join the study.Women were invited to participate in early prenatal visits, and those who consented completed a brief questionnaire that included information on lifestyle and sociodemographic characteristics.Women also provided blood and urine samples at enrollment, midpregnancy, and third trimester/delivery.Enrollment occurred at the first prenatal visit to the Von Voigtlander clinic.To be eligible for this pilot analysis, women had to have provided a blood sample in either the first or second trimester, to have given information on sleep habits during the same time frame, and to have had a singleton live delivery.Of the 112 samples analyzed for DNA methylation (randomly selected from eligible women with early pregnancy blood samples and sleep duration), 96 were included in the final analysis (see Figure 1 for flowchart).Informed consent was received from all participants.

Sleep duration and timing
The baseline questionnaire, completed for the majority of women in the 1 st or 2 nd trimester, had the following questions about typical bed and wake times: "What time do you usually go to sleep?" and "What time do you usually wake up?"These data were used to calculate the midpoint, the median between bed and wake time, as well as sleep duration.

DNA methylation
A whole-blood sample was taken by a phlebotomist up to three times during pregnancy, once during each trimester, and stored in Paxgene DNA preservation tubes at −80°C.This pilot investigation utilized only first trimester or second trimester samples (in the event that a woman did not have a stored first trimester sample).DNA was isolated from 112 blood leukocyte samples.Blood leukocyte DNA (500 ng) was treated with sodium bisulfite, which converts unmethylated cytosines to uracils leaving methylated cytosines unchanged, via an EpiTect bisulfite kit (Qiagen), and stored at −20°C until analysis.DNA methylation (percent of methylated cells) was then quantified at PER1 (eight sites), BMAL1 (three sites), and MTNR1B (eight sites) via pyrosequencing by a PyroMark ID Pyrosequencer (Qiagen) (Busato et al. 2018).Primers were adapted from published assays that amplify segments within or near the promoter region of each gene capturing multiple CpG sites (Supplemental Table S1) (Erdem et al. 2017;Nawrot et al. 2018).For each gene, sequences were amplified with HotStarTaq Master Mix (Qiagen) from approximately 40 ng bisulfiteconverted DNA.Samples were run in two batches (96-well plates) for each gene.Several controls were included to ensure quality in each batch (i.e. each 96-well plate): no-template PCR controls, 0% methylated human DNA, 50% methylated human DNA, and 100% methylated human DNA.A subset of samples (~40%) were duplicated, and results averaged when both replicates passed.Additional internal quality control checks were performed by the Pyro Q-CpG software to confirm proper bisulfite conversion, adequate signal above background noise, etc.Samples that failed quality control were excluded from analysis.Several sites within BMAL1 were not analyzed due to the extremely low variability in % methylation; analyses from CpG site 1 and CpG site 3 are shown.The second site of PER1 exhibited a range of methylation values, but 69% of the values were zero.Thus, this site was modeled categorically (0 vs. any other methylation value).

Covariates
Covariates included gestational age at early pregnancy assessment (i.e., when sleep and other lifestyle information was collected), self-identified race/ethnicity, education, total household income, marital status, smoking status, number of household members, fastfood consumption, fresh vegetable consumption, stress, snoring, pre-pregnancy chronic metabolic conditions, parity, BMI from first prenatal visit, sex assigned at birth of baby, and maternal age at estimated date of delivery.Gestational age was clinically determined based on the last menstrual period and the first accurate ultrasound according to the American College of Obstetricians and Gynecologists' guidelines (Methods for Estimating the Due Date, 2017) and classified into quartiles, from 6.57 to 8.57 weeks, 8.58 to 9.71 weeks, 9.72 to 18.00 weeks, and 18.01 to 37.71 weeks.Self-identified race was classified as White and non-White.Education was classified into four categories; high-school diploma or below, some college/ trade training/Associate's degree, Bachelor's degree, and Master's degree or above.Income was classified as $49,999 or less, $50,000 to $99,999, $100,000 to $149,999, and $150,000 or more.Marital status was categorized as married/domestic partnership vs other.Smoking status was defined as never, former, or current.The number of household members was grouped into 1 or 2, 3, and 4 or more.Fast-food consumption was assessed with the question "Do you eat at fast food restaurants (at least once a week)?" and considered dichotomously.The frequency of fresh vegetable consumption was assessed with the questions "Do you eat fresh vegetables (at least once a week)" and "If yes, how often?"For analysis, these questions were combined into a dichotomous variable for 1 time/week or less vs 2 or 3 times.Feelings of stress were assessed with the question "Do you feel stressed?"and considered dichotomously.Snoring was assessed with the question "Do you currently snore 3 or more times a week?" and considered as yes, no, or don't know.Chronic metabolic conditions before pregnancy (as a dichotomous variables) included Type 1 diabetes, Type 2 diabetes, or hypertension, and were extracted from the medical chart.Parity was classified as 0, 1, and 2 or more children.Sex of the baby assigned at birth was abstracted from the medical chart.BMI at the first prenatal visit was abstracted from the medical record and classified as overweight/obese (BMI ≥ 25) or non-overweight.Maternal age at the estimated date of delivery was calculated based on the due date and the mother's birth date and was categorized as <25 years, 25-34 years, and ≥35 years.

Statistical analysis
Univariate analyses were first conducted to assess normality and to check for potential outliers in all variables.Log transformations were completed for outcomes that were not normally distributed.We also conducted Pearson and Spearman correlations (depending on normality) between CpG sites of the same genes in order to determine if the individual CpG sites should be considered separately or averaged across sites.DNA methylation at the eight CpG sites in MTNR1B was highly correlated; as such, the values across all sites were averaged for primary analysis.In bivariate analysis, we examined associations between lifestyle and demographic characteristics with sleep characteristics and with birth outcomes, whereby means (SD) or medians (IQR) of sleep and birth outcomes were computed according to categories of lifestyle and demographic characteristics.The statistical significance of these associations was ascertained with Kruskal-Wallis tests.We also tested whether DNA methylation values differed in samples collected in the first versus second trimester.To evaluate our first aim, we calculated means (SD) or medians (IQR) of % methylation of individual or averaged CpG sites by categories of sleep duration and midpoint.We conducted secondary analyses using percent DNA methylation from each individual CpG site.We then performed a linear regression analysis that accounted for maternal age and education.To evaluate our second aim, we conducted a linear regression analysis such that gestational age at delivery, birthweight, and head circumference were the continuous outcomes and sleep characteristics and individual or averaged CpG sites were the predictors (each run in separate models).To account for potential confounding, we adjusted for variables that are known determinants of infant birth outcomes and were found to be associated with both the exposure and the outcome in our bivariate analysis.For gestational age at delivery, we accounted for parity, maternal age, and education.For birthweight, we accounted for gestational age, sex of baby, parity, and smoking.For head circumference, we accounted for gestational age, sex of baby, and education.

Results
Women were on average 30.3 ± 5.3 years of age.The majority were White (81%), married (73%), and had at least some post-high school education (74%).The average self-reported sleep duration was 8.6 ± 1.6 h and average sleep midpoint was 2:46 AM ±79 min.Self-reported feelings of stress and higher parity were related to shorter sleep duration (Table 1).Lower levels of education, lower income, non-married status, and younger maternal age were associated with later sleep midpoint.Average gestational age at the time of delivery was 268 ± 16 days (38.2 ± 2 weeks), average birthweight was 3182 ± 611 grams, and average head circumference was 33.9 ± 2.1 cm.Higher maternal education, being married/partnered, and not having pre-pregnancy diabetes or hypertension was associated with longer gestation (Table 2).Higher parity was associated with smaller birthweight and smaller head circumference, while higher maternal education was related to larger head circumference.
Overall, there were no statistically significant associations between sleep duration or timing with DNAm (Table 3).There was one marginally significant positive association (P = 0.08) between sleep duration and PER1, but this association did not hold after adjustment for maternal age and education (P = 0.57).DNA methylation values did not differ significantly in samples collected in the first versus the second trimester (Supplemental Table S2).When comparing DNA methylation at each individual CpG site assessed across sleep categories, the results were similar to the dichotomous analyses for BMAL1 and PER1 (Supplemental Tables S3 and S4).In site-specific rather than aggregate analysis for MTNR1B, one site had higher methylation in the longest sleep duration tertile in unadjusted analyses (p = 0.04, Supplemental Table S4).
Higher methylation of CpG site 1 in PER1 was associated with smaller log-transformed head circumference (β=-0.02with 95% CI −0.02 to 0.01; P, trend = 0.04; Table 4) after accounting for gestation age, sex of infant, and maternal education.Higher methylation of MTNR1B (averaged across all CpG sites) was associated with lower log-transformed birthweight (−0.08 with 95% CI −0.16 to −0.01; P, trend = 0.0495; Table 4) after accounting for gestation age, infant sex, parity, and maternal smoking.In addition, longer sleep duration was associated with higher birthweight (0.10 with 95% CI 0.02 to 0.18 comparing >9 h to <8 h; P, trend = 0.04), accounting for the same confounders.

Discussion
In this pilot investigation of a Michigan pregnancy cohort, higher DNA methylation of PER1 and MTNR1B genes measured in early-to-mid pregnancy was related to smaller head circumference and birthweight of offspring, respectively.Although there was no strong evidence that self-reported sleep duration and timing were related to DNAm of these circadian genes, longer maternal sleep duration was independently associated with higher birthweight.Overall, these findings add to the growing evidence base that maternal circadian mechanisms may be involved in birth outcomes, and highlight the need for future studies that utilize objective sleep assessments and investigate a greater number of genes.
Sleep duration and timing were not statistically significantly related to DNAm of these genes.There are very few existing studies within pregnant populations with which we can compare our present findings, although studies on shift workers have generally found that misaligned sleep times are associated with the genes we investigated and other circadian genes (e.g., PER2, CRY2) (Bukowska-Damska et al. 2017;Reszka et al. 2018).Further, one of the prior studies conducted in pregnant women found that those with poor sleep quality late in pregnancy had higher placental DNA methylation of MTNR1B, coupled with lower expression of melatonin receptor proteins (MTNR1B and MTNR1A) (Lin et al. 2022).It is possible that we did not see associations because we do not have extreme variations in sleep timings such as shift work, and the blood epigenome may be a less sensitive tissue to sleep disturbances during pregnancy compared with placenta at these genes.The small sample size and self-reported nature of the sleep measures were also very likely factors in the inability to detect an association.Nonetheless, we observed a nonstatistically significant but suggestive crude association between sleep duration and one CpG site of PER1.Interestingly, polymorphisms in PER1 have been related to various aspects of sleep, including selfreported habitual wake time (Lee et al. 2015) and shift work disorder, which is characterized by excessive sleepiness while working at night and insomnia during the day (Taniyama et al. 2015).Although we did not have strong evidence for associations for self-reported sleep and DNA methylation of our selected genes, we did find two associations between DNAm and birth outcomes.First, higher methylation of PER1 CpG site 1 was related to a smaller head circumference.Although this finding is novel within the extant literature, PER1 has been previously identified for possible associations with pregnancy complications.Specifically, a study of 114 Dutch women showed that placental PER1 DNA methylation levels differed in women with preeclampsia compared to uncomplicated controls (van den Berg et al. 2017).We also found that higher methylation of MTNR1B, one of the genes that codes for melatonin, was associated with lower birthweight.Melatonin plays an important role in circadian rhythms, and it is also an antioxidant.Within a sample of 32 pregnant women with intrauterine growth restriction, expression of MTNR1A and MTNR1B in placental tissue was lower compared to 30 control women with uncomplicated pregnancies matched for gestation age (Berbets et al. 2021).In the UK Biobank, maternal SNPs at MTNR1B are related to offspring birthweight (Beaumont et al. 2018).One potential mechanism by which MTNR1B could be related to birthweight is through maternal glycemia.SNPs in the MTNR1B gene have been implicated in type 2 diabetes in general populations (Bouatia-Naji et al. 2009), and more recently with gestational diabetes in pregnant populations (Jia et al. 2020;Tarnowski et al. 2017).However, in the present sample, we were underpowered to examine the possible role of gestational diabetes, since only seven women had this diagnosis.
We also found a dose-response relation between sleep duration and infant birthweight, after accounting for fetal sex, gestational age, parity, and maternal smoking.A positive association between pregnancy sleep duration and birthweight has been reported in a few other studies, although not consistently (Warland et al. 2018).A meta-analysis of studies completed before July 2021 reported no significant association between either short sleep duration or long sleep duration with low birthweight (Wang et al. 2022), although it is worth noting that the study considered low birthweight dichotomously rather than examining birthweight as a continuous outcome.In contrast, a recent study that utilized Mendelian Randomization to investigate the causal role of sleep duration on offspring birthweight found that both short sleep duration and excessively long sleep duration were related to higher odds of offspring with low birthweight (Yang et al. 2022).We were not powered to investigate low birthweight as a dichotomous outcome, since the majority of the infants in our sample was in the normal range according to their gestational age.Nonetheless, a positive relationship between sleep duration and birthweight is plausible, given that shorter sleep duration during pregnancy has been associated in prior studies with higher risk of pregnancy complications (Facco et al. 2017;Larson et al. 2022;Reutrakul et al. 2018), and generally that low birthweight is indicative of more unfavorable intrauterine conditions as well as long-term consequences for the infant (Finken et al. 2018).
This pilot investigation has both strengths and limitations.A notable strength is the novelty of the research question, and the ability to examine prospective relationships between sleep and epigenetic markers relatively early in pregnancy in relation to birth outcomes, rather than the assessment of sleep and epigenetics at birth or late in the third trimester.We also had access to a large number of known or potential confounders, allowing us to conduct multivariable analysis.One of the main limitations was the small sample size, which was based on sample availability and funding constraints.As mentioned previously, this precluded us from conducting certain analyses  (i.e., examining gestational diabetes and offspring outcomes including preterm birth, and large-and smallfor-gestational age), and likely limited our power to detect associations, especially between sleep characteristics and epigenetic markers.We also examined only a small selection of candidate genes.Effect sizes were small, and the biological relevance of these associations is uncertain.We did not have information on morning versus evening preferences in order to assess chronotype.Finally, all lifestyle and sociodemographic characteristics were self-reported, including sleep habits, which are notoriously difficult to self-assess.However, one study of >7,000 pregnant women showed high concordance between self-reported sleep measures and objectively assessed sleep (94% for late sleep midpoint and 77% for short sleep duration) (Facco et al. 2018), suggesting that the selfreporting of sleep, particularly sleep timing, may not be highly inaccurate during pregnancy.Furthermore, any measurement error would be most problematic if it was differential with respect to the study outcomes (e.g., if women with preterm deliveries were more likely to underestimate sleep duration than women with full-term deliveries); however, we have no reason to believe this is the case.Future studies that include both objective sleep measurements and a wider array of circadian and sleep-related genes are warranted.

Figure 1 .
Figure 1.Flow chart of study sample.

Table 1 .
Early-to-mid pregnancy sleep duration and timing in relation to sociodemographic and lifestyle characteristics.1

Table 1 .
(Continued).Sleep duration and midpoint were calculated based on the questionnaire asking about typical bed and wake times.P values were obtained from Kruskal-Wallis test.

Table 2 .
Gestational age at delivery and birthweight in relation to maternal pregnancy sociodemographic and lifestyle characteristics.
1 P values were obtained from Kruskal-Wallis test.2Feeling stressed comes from question "Do you feel stressed?." 3 Chronic metabolic conditions include Type 1 Diabetes, Type 2 Diabetes, or hypertension.

Table 3 .
Sleep duration and timing in relation to DNA methylation of selected circadian genes during pregnancy.P values of BMAL CpG1 and CpG2 and PER1 CpG1 and CpG3 were obtained from a Kruskal-Wallis test.P values of PER CpG2 were obtained from a Chi-square test.P values of MTNR1B were obtained from an ANOVA Test. 2 Due to strong positive correlations, 8 individual sites were averaged prior to analysis.

Table 4 .
Early-to-mid pregnancy sleep and circadian markers in relation to gestational age at delivery and birthweight.