Exploring Perinatal Biopsychosocial Factors and Epigenetic Age in 1-Year-Old Offspring

Abstract

Background: Little is known about the determinants of epigenetic aging in pediatric populations. Methods: Epigenetic age was estimated from 258 1-year-olds, using pediatric buccal epigenetic and Horvath clocks. We explored associations between epigenetic age and maternal indicators of mental and relational health, substance use and general physical health assessed during trimester three. Results: Higher anxiety and stress, BMI and higher parent–parent relationship quality were associated with pediatric buccal epigenetic clock differences. High blood pressure during pregnancy was associated with Horvath age acceleration. Third-trimester smoking and pre-pregnancy weight were associated with acceleration and deceleration respectively, and concordant across clocks. Conclusion: A broad range of maternal factors may shape epigenetic age in infancy; further research is needed to explore the possible effects on health and development.

Plain language summary

Molecules on our DNA, called DNA methylation, can be used in a laboratory test to estimate how old we are – also known as epigenetic age. In adults, a higher risk of age-related disease has been attributed to older epigenetic age. However, we know very little about epigenetic age in children. In this study, we look at the how measures of a mother’s health during pregnancy – such as using alcohol or tobacco, mental health (stress, anxiety and depression), or general health such as weight or high blood pressure – affect epigenetic age in children.

Keywords::

• aging
• DNA methylation
• epigenetic clock
• Horvath
• mental health
• PedBE
• pediatrics
• perinatal
• substance use

The ‘developmental origins of health and disease’ (DOHaD) hypothesis has long maintained that social and environmental exposures prior to conception, in utero and in early life can be causally related to later life health outcomes [1]. One mechanism driving risk is epigenetic programming of gene expression. Epigenetics is the study of structural and molecular mechanisms – primarily DNA methylation, histone modification and noncoding RNAs – that do not alter the DNA code but are involved in gene regulation [2]. Because epigenetic mechanisms are dynamic and respond to the environment, they have been of particular interest within the DOHaD framework [3,4].

A wide range of environmental exposures during pregnancy have now been associated with DNA methylation profiles in offspring. One of the most replicated examples of this is exposure to maternal smoking [5]. Current research also shows that DNA methylation differences, at least in part, may mediate smoking exposure-related outcomes such as low birthweight [6] and childhood asthma [7]. Maternal exposure to second-hand smoke can also affect the fetus’s epigenome, altering DNA methylation in genes associated with inflammation, carcinogens and neuronal functioning [8]. Further, exposure to less obvious environmental factors, such as maternal adverse childhood experiences years before conception [9] and poorer maternal mental health during pregnancy [10] has been shown to affect the offspring’s epigenome. Thus exploring the effects of factors such as maternal mental, social and general health on the offspring epigenetic profile is of great importance, given that they may be linked to negative health and developmental outcomes.

More recently, it has become clear that there are stable DNA methylation changes over a lifespan which track with age [11]. This gradual change, which reflects biological aging, regardless of chronological age, is known as epigenetic aging and is measured using epigenetic/DNA methylation ‘clocks’. There are now several clocks available, each measuring different aspects of aging [12]. The most popular clock, Horvath’s clock, was built by using methylation profiles of multiple tissues and finding methylation sites across the epigenome that correlate highly with age [13]. Another clock, by Hannum et al., was built similarly, but by using DNA methylation signatures from blood samples only [14]. There are also other clocks that measure ‘aging’ indirectly; for example, GrimAge is a measure of age reflected by time to death, derived using measures of morbidity and mortality that are associated with plasma-based proteins, as well as DNA methylation estimations of smoking pack-years [15]. A newly developed clock, DunedinPACE, estimates age based on methylation values associated with several other age-related biomarkers [16] rather than age itself, as Horvath’s clock does.

While measures such as these may be useful in predicting age-related disease and mortality risk [17], the majority of studies that develop or use these clocks focus on middle-aged to older adults. This has left a significant gap in pediatric populations and limited understanding of the developmental origins of biological aging. Most clocks are developed using blood, as well as using age ranges that are not particularly appropriate for the pediatric population. Indeed, a recently published review of aging clocks in pediatric populations has shown that few studies currently exist, and even then, seldom in the context of early development [18]. However, the recent development of the pediatric buccal epigenetic (PedBE) clock, which was built using similar methods to Horvath’s clock, now provides unique opportunities to extend work on adult biological aging to biological aging at the beginning of the life course. The PedBE clock has the added advantage of being derived from methylation signatures generated from easily collected buccal swabs, and only using pre-adulthood years (0–20), capturing pertinent developmental periods [19]. Indeed, a recent review of the use of epigenetic clocks in pediatric populations has highlighted that the PedBE clock is the best performer of age estimation when using DNA methylation generated from pediatric buccal samples [20].

The aim of this study was to explore and identify perinatal exposures that may be associated with epigenetic age acceleration at 1 year of age, estimated with the PedBE and Horvath clocks. Specifically, the aims of the study were to examine the extent to which epigenetic aging in 12-month-old infants is associated with perinatal measures of: maternal mental and relational health status, including anxiety, stress, depression and intimate partner relationship quality; maternal substance use, including alcohol, tobacco and cannabis; and maternal physical health, including weight, blood pressure and BMI.

Materials & methods

Study participants

The Australian Temperament Project (ATP) is a multigenerational, longitudinal cohort, initially started as a population representative survey of social and emotional health [21]. For this study we used child participants from ATP generation 3 (G3), as well as data collected from their mothers from ATP generation 2 (97% born in Australia). Details of the cohort, informed consent and ethics have been described previously [22]. ATP G3 study protocols have been approved by the Royal Children’s Hospital Human Research Ethics Committee.

Biological samples, methylation data & epigenetic clock generation

Buccal swabs were collected from n = 281 G3 children, at approximately 1 year of age. DNA was extracted from buccal swabs using the salting-out extraction method. A subsample of 270 participants had an adequate amount of DNA for generation of methylation data and were subsequently used for this study. DNA methylation data were generated by the Australian Genome Research Facility, Melbourne, Australia (www.agrf.org.au/) using the Illumina Infinium HumanMethylationEPIC BeadChip. R version 3.6.1 was used to process all data [23]. Technical or biological replicates were not used in this study due to this being an exploratory study and the restrictive cost of generating epigenome-wide data. The minfi package [24] was used to carry out preprocessing and quality control. Methylation at one probe was missing for one participant and imputed using impute [25]. Methylation data were normalized using the β-mixture quantile method [26]. Data from 12 participants were removed due to poor quality of DNA samples, resulting in methylation data not passing quality control. This left 258 children (54.7% female; mean age = 62.77 weeks, standard deviation = 7.7; 48% first child) from 213 mothers (mean age = 32.5 years, standard deviation = 1.4) in the dataset.

The PedBE clock was generated following publicly available code from https://github.com/kobor-lab/Public-Scripts/blob/master/PedBE.Md. The Horvath clock was generated using the online portal https://dnamage.genetics.ucla.edu/new. Estimated biological ages from PedBE and Horvath clocks are reported in weeks rather than years. Measures of age acceleration residuals were derived from regressing estimated biological age onto chronological age (i.e., biological age adjusted for chronological age). As both clocks adjust for cell type, they may be considered measures of ‘intrinsic’ epigenetic age acceleration [27].

Maternal perinatal measures

Mothers responded to questionnaires during pregnancy (89.19% during the third trimester and 10.81% retrospectively about the third trimester) which covered a wide range of biopsychosocial factors [22]. Maternal anxiety and stress were measured using the 21-question short report Depression Anxiety Stress Scale (note that scores were not multiplied by two) [28]. Maternal depression was measured using the Edinburgh Postnatal Depression Scale [29]. Higher scores for anxiety, stress and depression indicate worse mental health. Relationship quality included in this study was measured using the dyadic adjustment scale adapted for the ATP study, which averages scores from seven items, ranked from 1 to 6 [30]. Social support was measured using the Maternity Social Support Scale [31]. Maternal–fetal attachment was measured using a maternal–fetal attachment scale adapted from the work of Cranley [32], using an average score from six items, ranked from 1 to 4, with each item multiplied by six. Higher scores represent higher relational quality.

Alcohol, tobacco and cannabis use were assessed at three time points: before pregnancy awareness (retrospectively), after pregnancy awareness (retrospectively) and during the third trimester (prospectively). For each time point and substance type, responses were recoded to indicate any use (none or any). For cannabis, due to the small number of mothers reporting use, responses for ‘during pregnancy’ were collapsed across ‘after pregnancy awareness’ and ‘during third trimester’.

Self-reported physical health measures used in this study included weight (kg), height (cm) and BMI leading up to pregnancy; a five-point scale of general physical health (poor, average, good, very good and excellent); gestational diabetes (no or yes); high blood pressure or preeclampsia (no or yes); and infections or fever (no or yes) during pregnancy.

Potential confounding factors

Common factors that confound epigenetic data were considered in statistical models. Proportions of epithelial and immune cells were estimated using epidish [33], and epithelial cell proportions were included as an adjustment in all models, as suggested by both Horvath and McEwen et al. [13,19]. Other adjustments included maternal age [34], parity (mother’s first pregnancy: no or yes), financial strain (scored from 0 [living comfortably] to 4 [finding it very difficult]), maternal smoking and alcohol use during any of the aforementioned time points [35,36], child sex and gestational age [37,38], child chronological age [39] and experimental batch (EPIC array data were collected from three separate experimental batches) [40].

Statistical analysis

Stata v. 17 (StataCorp, LLC, TX, USA) [41] was used to carry out all statistical models. Correlations between age estimated by PedBE and Horvath clocks and chronological age of children were examined. Age acceleration residuals were used to explore relationships between biological age and maternal biopsychosocial factors. Specifically, linear regression models regressed age acceleration residuals onto maternal perinatal measures, separately, using a robust standard error estimator to account for within-family clustering. Models were run adjusting for only cell proportions (model 1) and then additionally adjusting for all other potential confounding factors, specifically maternal age, parity, financial strain, maternal alcohol use and smoking during pregnancy, child sex, chronological age and gestational age, epithelial cell proportion and experimental batch (model 2), unless the covariate was the exposure of interest (i.e., we did not adjust for smoking if smoking was the primary exposure in the model). For inferential analyses, continuous exposure variables are standardized (z-score). Results are reported as regression coefficients representing divergence from chronological age in weeks (β), with 95% CIs and findings considered significant using a two-sided test p-value of < 0.05. Where data were missing on any maternal participant from the 258 mother–child dyads, they were excluded from the specific analytical test, and the number within each test was tabulated in the results.

Results

Participant characteristics

Child and maternal characteristics can be seen in Table 1. On average, children were 1.2 years old when swabs were collected, with slightly more female participants (54.7%) than males. Epithelial cells contributed to the majority of estimated cell types within buccal swabs collected (mean = 80% of all cells across samples collected). Mothers were on average 32.1 years old, and a minority experienced greater than mild depression (15% ≥10), stress (8% ≥8) or anxiety (6% ≥4). Most women did not smoke during pregnancy; however, most women did consume alcohol before awareness of pregnancy (84%). Cannabis was only used during pregnancy by a very small proportion of women (1%).

Table 1. Participant characteristics.

Infant characteristics at 1 year of age	n
Gestational age in weeks^†, mean (SD)	258	38.8 (2.2)
Chronological age in weeks, mean (SD)	258	62.7 (7.7)
PedBE age in weeks, mean (SD)	258	100.2 (32.1)
Horvath age in weeks, mean (SD)	258	130.7 (28.1)
Estimated epithelial cell proportion in buccal sample, mean (SD)	258	80% (12%)
Maternal characteristics assessed during third trimester
Maternal mental and relational health
Depression^‡, mean (SD)	220	5.4 (3.9)
Stress^§, mean (SD)	220	3.9 (2.9)
Anxiety^§, mean (SD)	220	1.2 (1.6)
Dyadic adjustment^¶, mean (SD)	216	5.4 (0.3)
Maternity social support scale^#, mean (SD)	219	4.6 (0.4)
Maternal–fetal attachment scale^††, mean (SD)	220	15.5 (2.6)
Maternal general health
Self-reported pre-pregnancy weight (kg), mean (SD)	221	67.9 (14.7)
Self-reported pre-pregnancy height (cm), mean (SD)	220	165.6 (13.6)
Pre-pregnancy BMI^‡‡, mean (SD)	220	24.6 (5.1)
Self-rated physical health during pregnancy^§§, mean (SD)	221	3.7 (1)
High blood pressure/preeclampsia during pregnancy, n (%)	221	11 (5%)
Blood sugar problems during pregnancy, n (%)	231	11 (5%)
Infection during pregnancy, n (%)	220	38 (17%)
Maternal substance use
Smoking
Before pregnancy awareness, n (%)	221	17 (8%)
Just after pregnancy awareness, n (%)	219	15 (7%)
During third trimester, n (%)	213	7 (3%)
Alcohol
Before pregnancy awareness, n (%)	221	185 (84%)
Just after pregnancy awareness, n (%)	221	22 (10%)
During third trimester, n (%)	221	33 (15%)
Cannabis
Any use before pregnancy, n (%)	221	6 (3%)
Any use during pregnancy, n (%)	221	3 (1%)

† Calculated based on reported child date of birth.

‡ Edinburgh Postnatal Depression Scale [29], ranked from 0 to 30. Higher scores indicate more depressive symptoms.

§ Depression Anxiety Stress Scale [28], ranked from 0 to 21. Higher scores indicate more anxiety and stress symptoms.

¶ Adapted from the Dyadic Adjustment Scale [30], average score from seven items ranked 1 to 6. Higher scores indicate better relational quality with partner of significance.

# Maternity Social Support Scale [31], average score from six items ranked 1 to 5. Higher scores indicate higher social support.

†† Adapted from the Maternal–Fetal Attachment Scale [32], average score from six items ranked 1 to 4, each item multiplied by six. Higher scores indicate more higher attachment to fetus.

‡‡ BMI calculated based on self-reported pre-pregnancy weight and height.

§§ Self rated physical health ranked from 1 (poor) to 5 (excellent).

PedBE: Pediatric buccal epigenetic clock; SD: Standard deviation.

Clock estimate correlations

PedBE clock age estimates were more highly correlated with child chronological age compared with Horvath age estimates (r = 0.49; p < 0.001; and r = 0.22; p < 0.001, respectively) (Figure 1A & B).

Figure 1. Correlation between estimated biological age measures and chronological age in weeks.

Correlation line shown. (A) PedBE and chronological age (r = 0.49; p < 0.001). (B) Horvath and chronological age (r = 0.22; p < 0.001).

PedBE: Pediatric buccal epigenetic clock.

Age acceleration

Associations between age acceleration residuals and maternal perinatal biopsychosocial factors are shown in Table 2. Supplementary analyses between age acceleration residuals and any potential confounding birth-related factors (adjusted for epithelial cell proportion, batch and child age) can be seen in Supplementary Table 1.

Table 2. Associations between infant epigenetic age acceleration residuals and perinatal maternal characteristics.

	Model 1^†							Model 2^‡
		Horvath AAR			PedBE AAR				Horvath AAR			PedBE AAR
	n	β	95% CI	p-value	β	95% CI	p-value	n	β	95% CI	p-value	β	95% CI	p-value
Maternal mental and relational health
Depression	220	‐1.72	(-5.47 to 2.02)	0.366	-2.30	(-4.54 to -0.05)	0.045	220	‐1.15	(-5.13 to 2.84)	0.572	‐1.91	(-4.27 to 0.44)	0.111
Stress	220	‐1.92	(-6.10 to 2.26)	0.367	‐2.35	(-4.75 to 0.05)	0.055	220	‐1.97	(-6.18 to 2.23)	0.355	-2.32	(-4.49 to -0.16)	0.036
Anxiety	220	‐3.29	(-7.45 to 0.87)	0.120	-3.79	(-6.34 to -1.23)	0.004	220	‐2.70	(-7.19 to 1.79)	0.237	-3.60	(-5.92 to -1.28)	0.003
Dyadic adjustment	216	0.21	(-3.38 to 3.79)	0.909	2.80	(0.64 to 4.96)	0.011	216	0.47	(-3.21 to 4.16)	0.800	2.24	(0.21 to 4.26)	0.030
Maternity Social Support Scale	219	‐1.07	(-4.51 to 2.36)	0.539	1.19	(-0.78 to 3.16)	0.236	219	‐0.94	(-4.57 to 2.68)	0.609	0.63	(-1.34 to 2.60)	0.527
Maternal–Fetal Attachment Scale	220	‐0.21	(-3.65 to 3.22)	0.904	0.27	(-1.90 to 2.44)	0.808	220	‐0.55	(-4.22 to 3.11)	0.766	0.39	(-1.69 to 2.47)	0.711
Maternal substance use
Alcohol before pregnancy awareness^§	221	‐0.92	(-9.66 to 7.81)	0.835	1.76	(-3.24 to 6.77)	0.487	221	‐0.83	(-9.82 to 8.15)	0.855	‐0.80	(-5.68 to 4.07)	0.746
Alcohol just after pregnancy awareness^§	221	11.68	(0.98 to 22.38)	0.033	3.67	(-1.82 to 9.17)	0.188	221	10.64	(-0.23 to 21.51)	0.055	1.56	(-4.26 to 7.38)	0.598
Alcohol during third trimester^§	221	7.14	(-2.73 to 17.01)	0.155	3.91	(-0.74 to 8.56)	0.099	221	6.02	(-4.32 to 16.36)	0.252	2.45	(-2.74 to 7.63)	0.353
Smoking before pregnancy awareness^¶	221	3.64	(-8.23 to 15.51)	0.546	0.39	(-6.41 to 7.18)	0.911	221	2.76	(-8.97 to 14.49)	0.644	2.49	(-5.89 to 10.87)	0.558
Smoking just after pregnancy awareness^¶	219	9.08	(-0.34 to 18.50)	0.059	3.80	(-1.91 to 9.52)	0.191	219	8.67	(-1.36 to 18.70)	0.090	5.98	(-1.77 to 13.73)	0.129
Smoking during third trimester^¶	213	15.84	(5.47 to 26.20)	0.003	9.00	(0.35 to 17.64)	0.041	213	14.94	(3.56 to 26.32)	0.010	10.52	(2.07 to 18.98)	0.015
Cannabis use before pregnancy	221	‐0.53	(-12.32 to 11.25)	0.929	2.92	(-0.14 to 5.98)	0.061	221	1.80	(-15.66 to 19.26)	0.839	2.57	(-8.51 to 13.65)	0.648
Cannabis use during pregnancy	221	3.21	(-1.26 to 7.69)	0.159	3.58	(1.17 to 5.98)	0.004	221	9.63	(-7.80 to 27.06)	0.277	11.48	(-3.74 to 26.71)	0.138
Maternal general health
Self-rated physical health in pregnancy	221	3.01	(-0.69 to 6.72)	0.111	1.65	(-0.53 to 3.82)	0.136	221	3.46	(-0.33 to 7.25)	0.073	1.56	(-0.39 to 3.51)	0.116
Weight	221	-3.38	(-6.20 to -0.56)	0.019	‐1.70	(-4.17 to 0.77)	0.176	221	-3.49	(-6.48 to -0.51)	0.022	-2.20	(-4.33 to -0.08)	0.042
Height	220	‐0.10	(-3.21 to 3.01)	0.949	‐1.03	(-2.26 to 0.21)	0.103	220	‐0.54	(-3.54 to 2.46)	0.723	‐1.01	(-2.24 to 0.22)	0.106
BMI	220	‐2.59	(-5.61 to 0.42)	0.092	‐1.53	(-3.71 to 0.65)	0.167	220	‐2.60	(-5.80 to 0.59)	0.109	-1.87	(-3.70 to -0.04)	0.045
Gestational diabetes	213	‐2.61	(-25.68 to 20.46)	0.823	‐2.16	(-15.09 to 10.77)	0.742	213	‐2.37	(-26.12 to 21.39)	0.844	‐0.56	(-15.28 to 14.17)	0.941
High blood pressure/preeclampsia	221	-13.68	(-26.32 to -1.03)	0.034	‐6.47	(-17.36 to 4.42)	0.243	221	-15.50	(-27.95 to -3.05)	0.015	‐5.32	(-14.87 to 4.22)	0.273
Infection or fever	220	2.68	(-7.71 to 13.08)	0.611	‐3.31	(-9.60 to 2.98)	0.300	220	3.49	(-7.13 to 14.11)	0.517	1.13	(-5.18 to 7.45)	0.723

† Model 1: adjusted for epithelial cell proportion.

‡ Model 2: adjusted for maternal age, parity, financial strain, maternal alcohol use and smoking during pregnancy, child sex, biological age and gestational age, epithelial cell proportion and batch.

§ Using model 2 (full adjustment), excluded alcohol use, as alcohol use is key exposure in model.

¶ Using model 2 (full adjustment), excluded tobacco use, as tobacco use is key exposure in model.

Significant (p < 0.05) results in bold. Continuous exposure variables are z-scored.

AAR: Age acceleration residual; p: p-value for specific adjustment model; PedBE: Pediatric buccal epigenetic clock.

Maternal perinatal mental & relational health

When adjusting for only epithelial cell proportions (model 1), higher maternal depression and anxiety were associated with decelerated PedBE clock age (β = -2.30; 95% CI: -4.54 to -0.05; p = 0.045; and β = -3.79; 95% CI: -6.34 to -1.23; p = 0.004, respectively). In contrast, higher parent–parent relational quality was associated with accelerated PedBE clock age (β = 2.80; 95% CI: 0.64 to 4.96; p = 0.011). When further accounting for potential confounding factors (model 2), the associations with maternal anxiety and decelerated PedBE clock age strengthened (β = -3.60; 95% CI: -5.92 to -1.28; p = 0.003; Figure 2A) and there was further indication that maternal stress may also be associated with decelerated PedBE clock age (β = -2.32; 95% CI: -4.49 to -0.16; p = 0.036; Figure 2B). However, the association with depression was attenuated (β = -1.91; 95% CI: -4.27 to 0.44; p = 0.111). The association between higher parent–parent relationship quality and accelerated PedBE clock age remained significant (β = 2.24; 95% CI: 0.24 to 4.26; p = 0.03; Figure 2C). No measures of maternal mental health were associated with measures of Horvath clock age acceleration or deceleration.

Figure 2. Maternal factors associated with differences in child pediatric buccal epigenetic clock age residuals (in weeks).

Continuous measures are displayed as standardized z-scores, regression line shown. (A) Prenatal anxiety: β = -3.60; 95% CI: -5.92 to -1.28; p = 0.003. (B) Prenatal stress: β = -2.32; 95% CI: -4.49 to -0.16; p = 0.036. (C) Dyadic adjustment: β = 2.24; 95% CI: 0.24 to 4.26; p = 0.03. (D) Maternal smoking during the third trimester: β = 10.52; 95% CI: 2.07 to 18.98; p = 0.015. (E) Pre-pregnancy weight: β = -2.20; 95% CI: -4.33 to -0.08; p = 0.042. (F) Pre-pregnancy BMI: β = -1.87; 95% CI: -3.70 to -0.04; p = 0.045.

Maternal perinatal substance use

Maternal smoking during the third trimester (β = 9.00; 95% CI: 0.35 to 17.64; p = 0.041) and cannabis use during pregnancy (β = 3.58; 95% CI: 1.17 to 5.98; p = 0.004) were associated with accelerated PedBE clock age when adjusting for cell proportions (model 1). Associations remained for smoking exposure after adjusting for other potential confounding factors (β = 10.52; 95% CI: 2.07 to 18.98; p = 0.015) (Figure 2D), but not cannabis (β = 11.48; 95% CI: -3.74 to 26.71; p = 0.138). Tobacco smoking during the third trimester and alcohol use shortly after pregnancy knowledge were both associated with increased Horvath clock age acceleration (β = 15.84; 95% CI: 5.47 to 26.20; p = 0.003; and β = 11.68; 95% CI: 0.98 to 22.38; p = 0.033, respectively), of which smoking exposure remained significant after full adjustment (β = 14.94; 95% CI: 3.56 to 26.32; p = 0.01; Figure 3A), but the trend was not as strong with exposure to alcohol (β = 10.64; 95% CI: -0.23 to 21.51; p = 0.055).

Figure 3. Maternal factors associated with differences in child Horvath clock age residuals (in weeks).

Continuous measures are displayed as standardized z-scores, regression line shown. (A) Maternal smoking during third trimester: β = 14.94; 95% CI: 3.56 to 26.32; p = 0.01. (B) Pre-pregnancy weight: β = -3.49; 95% CI: -6.48 to -0.51; p = 0.022. (C) Reported high blood pressure, including preeclampsia, during pregnancy: β = -15.50; 95% CI: -27.95 to -3.05; p = 0.015.

Maternal perinatal physical health measures

PedBE age acceleration was not associated with any self-reported perinatal health measures before adjustments. However, after adjustment, both weight (β = -2.20; 95% CI: -4.33 to -0.08; p = 0.042; Figure 2E) and BMI (β = -1.87; 95% CI: -3.70 to -0.04; p = 0.045; Figure 2F) were separately associated with a deceleration of the PedBE clock age. As well, higher reported weight (but not BMI) before pregnancy was associated with decelerated Horvath age (β = -3.38; 95% CI: -6.20 to -0.56; p = 0.019), as was having reported issues with high blood pressure during pregnancy (β = -13.68; 95% CI: -26.32 to -1.03; p = 0.034). Both of these associations remained after full adjustment models (β = -3.49; 95% CI: -6.48 to -0.51; p = 0.022; and β = -15.50; 95% CI: -27.95 to -3.05; p = 0.015) (Figure 3B & C).

Discussion

Here we report findings from an exploratory study, using a nested epigenetic sample from infants paired with maternal biopsychosocial data, drawn from a larger population-based cohort study. These findings suggest that offspring epigenetic aging may be shaped by a range of maternal biopsychosocial perinatal factors. In studies of older adults, exposures that negatively impact health are generally associated with an acceleration of epigenetic aging [17,42]. In our early-life study, findings were more variable, potentially reflecting the more dynamic nature of methylation events in early than later life. Specifically, while maternal substance use was associated with accelerated epigenetic age, maternal anxiety, stress and high blood pressure were associated with the opposite; that is, decelerated epigenetic age. Furthermore, dyadic adjustment representing low-conflict parent–parent relationships was associated with accelerated epigenetic aging. Small sample sizes necessitate cautious interpretation of the findings; however, if replicated, the results may demarcate early childhood as a unique developmental period within which the nature of biological aging follows a more variable, and not necessarily contiguous, course than biological aging later in the life course.

Increased maternal smoking, alcohol and cannabis use showed an association with accelerated infant age. Specifically, maternal smoking during the third trimester had the strongest association with accelerated biological age in infant offspring (PedBE age acceleration: +10.5 weeks; Horvath acceleration: +14.9 weeks). These findings are consistent with previous work using Horvath’s ‘skin and blood’ epigenetic clock [43] measured in blood samples of 1173 6- to 11-year-olds (average age: 7.2 years) [44]. Our findings should, however, be interpreted with caution due to the particularly low level of individuals reporting substance use. Of note, concordant findings across PedBE and Horvath clocks (in this study, smoking in the third trimester and pre-pregnancy weight) are of particular interest, as the two clocks are based on different methylation sites, with none overlapping (PedBE: 94 methylation sites; Horvath: 353 methylation sites). Together, these findings imply that substance use exposure in utero puts environmental pressure on the developing offspring to age.

In contrast, higher maternal stress and anxiety were associated with decelerated epigenetic age in infant offspring. These findings were unexpected and not concordant with the few prior studies in the field. For example, one recent study measured both Horvath and PedBE epigenetic age acceleration from buccal swabs in two cohorts, collected at different time points (n = 165 at 6 and 10 years; n = 340 at 3, 9 and 48 months) [45]. After adjusting for genetic principal component scores, sex, time point, gestational age, maternal education and buccal cell heterogeneity, the authors found, in contrast to our observations, that prenatal anxiety was associated with a small but significant PedBE clock age acceleration between time points in both cohorts. Similarly, previous work using the Knight et al. gestational epigenetic age clock [46] from the cord blood of 303 neonates found a weak positive correlation between prenatal maternal depression and gestational epigenetic age (although the association did not withstand adjustment for selective serotonin-reuptake inhibitor use), but no similar association for stress [47]. Similar to maternal anxiety and stress, maternal high blood pressure (which is also associated with infant offspring developmental problems [48,49]) was associated with decelerated aging, an unexpected direction, and the largest decelerated aging effect (Horvath age deceleration: -15.5 weeks). Yet another study of 1801 neonates, which used the Knight et al. gestational epigenetic age clock, found an association in the same direction as in this study, in which preeclampsia (but not gestational hypertension) was associated with decelerated gestational epigenetic age [50]. This association remained after adjusting for maternal ethnicity, race, pre-pregnancy BMI, age, prenatal smoking, educational level, child sex and cord blood cell composition estimates.

One explanation of this finding is that high blood pressure may be an indicator of factors involved with disrupted growth in the womb (e.g., intrauterine growth restriction, preterm birth, lower birth weight and size), with age deceleration simply marking ‘lagging age’. Indeed, prenatal anxiety, stress and hypertension are associated with intrauterine growth restriction [51,52]. It is important to note, however, that substance use is also well documented to cause intrauterine growth restriction [53–55] and did not exhibit the same pattern of findings.

Taken together, epigenetic age in infants appears dynamic in association with different exposures, and further research in early life is needed to resolve the nature of relationships between maternal perinatal exposures (risk and protective) and infant offspring biological aging. Additionally, the stability and continuity of patterns of infant epigenetic aging into later life ages and stages remains unknown. For example, it could be that any divergence (either ‘older’ or ‘younger’) from the linear relationship between biological and chronological age before development is completed (i.e., before adulthood) may result in possible downstream health or developmental effects. Even so, there must be a transition point in development where accelerated aging becomes more consistently associated with adverse health and developmental exposures and outcomes. For example, when examining environmental exposures during childhood, rather than perinatal factors mediated through the mother, most studies show results that are more in line with adult aging studies, in that risk exposures associate with accelerated aging across domains such as maltreatment and internalizing [56,57]. Longitudinal studies that assess biological aging during development, as well as following up in early, middle and late adulthood, are needed to understand whether changes to aging trajectories in early life affect trajectories thereafter, and where these divergences affect health or development.

Strengths, limitations & future directions

One of the primary strengths of this study is the broad range of biopsychosocial factors that were able to be explored in association with age acceleration. This also gave us the ability to adjust for a wide range of possible confounding factors, bringing robustness to our statistical models. Further, our epigenetic clock estimates for PedBE are more highly correlated (r = 0.49) compared with Horvath’s estimate (r = 0.22). This is most likely due to the fact that we used the same specific tissue (buccal) and were within a closer range of ages (0–20 years, vs Horvath’s 0–100 years) that were used to build the PedBE clock. This increased strength of association may also reflect an increased sensitivity in PedBE versus Horvath, in finding associations between maternal exposures and differences in aging in the offspring. This may explain why associations with maternal mental health were found in association with PedBE but not Horvath clock estimates.

A limitation of this study is that many of our groups – for example, cannabis use (n = 3) versus no cannabis use (n = 218) during pregnancy – are very small, which results in imprecise estimates (wide confidence intervals) of effects due to low power. These types of findings need to be interpreted with caution, with replication sought in other studies in the future. A further limitation of this study is that all maternal data collected were self-reported, and thus social desirability bias may be an issue. While 97% of the study population were born in Australia, we were unable to adjust for specific ethnicities, which is an important consideration in DNA methylation studies [58]. The two clock algorithms used in this study were selected because of their suitability for use with DNA methylation data extracted from buccal cells. We did not apply clocks developed on DNA methylation data extracted from blood cells (e.g., the Wu et al. epigenetic clock [59]). Future studies with access to both blood and buccal DNA will be needed to compare the performance of clocks based on different tissue types. Another limitation is that we were not able to discern whether the differences found in aging were apparent at birth, or whether they persisted into early childhood. This of course is an important consideration, due to the fact that specific methylation differences track steadily with age [13]. However, this will be a major point of interest in future longitudinal epigenetic studies of the ATP cohort, given that 8-week and 4-year buccal samples are available. Further, epigenetic aging at these time points will also be examined in the context of anthropometric (height, weight, BMI), socio-emotional and developmental outcomes of the offspring. This will be a major step in determining whether biological age divergences in the early years have positive or negative impacts when associated with outcomes.

Conclusion

Findings from this study suggest that several maternal perinatal factors may impact the overall rate of biological aging measured in 1-year-old infant offspring. PedBE clock measures are likely more appropriate than Horvath clock measures in a pediatric setting, given that the clock is more highly specific to younger ages. This potentially highlights the need to split up aging clocks before adulthood for use in specific age ranges (neonatal, early childhood, puberty etc.), as ‘aging’ during development may have a different, more variable, natural history to the more linear aging in later years. Our findings emphasize the importance of studies of epigenetic aging in pediatric populations, with future studies needing to include different time points and to explore the relationship between age acceleration and developmental outcomes. Finally, due to the varying results of acceleration and deceleration of biological age, studies in this field should consider that it may not be that aging faster is bad, or slower is good, but that a divergence from the biological aging norms may be negatively affecting a foundation of age-related epigenetic integrity established during development, and that may lead to an instability which increases the risk of later-life accelerated aging and disease.

Summary points

• Maternal perinatal factors may influence the offspring epigenetic profile, but less is known about how epigenetic aging is influenced.

• The pediatric buccal epigenetic (PedBE) clock and Horvath’s multi-tissue clock were generated from buccal swabs from 258 1-year-olds.

• A series of maternal perinatal biopsychosocial factors were used in covariate adjusted multivariate models to assess their associations with offspring epigenetic age.

• Higher maternal anxiety and stress, as well as maternal pre-pregnancy BMI, were associated with a slight deceleration of PedBE clock aging, while higher parent–parent relationship quality was associated with accelerated PedBE clock aging.

• High maternal blood pressure was associated with a large deceleration of Horvath epigenetic age (β = -15.50 weeks).

• Concordant findings between the PedBE and Horvath clocks include acceleration in epigenetic age associated with smoking in the third trimester and deceleration in association with higher pre-pregnancy weight.

• Further research is needed to determine whether the findings can be replicated and to ascertain whether early differences in biological age remain over time and to what extent early differences in biological age are predictive of health and development.

Author contributions

P Fransquet, C Greenwood and C Olsson contributed to the concept of this work. P Fransquet oversaw bioinformatics, epigenetic age generation and primary authorship of the article. C Greenwood oversaw statistical analysis. P Fransquet, C Greenwood and C Olsson interpreted the data for discussion. P Fransquet, J Macdonald, J Ryan, C Greenwood and C Olsson all contributed to drafting and revising the manuscript.

Writing disclosure

No writing assistance was utilized in the production of this manuscript.

Ethical conduct of research

This study involves human participants, and the main ATP study was approved by the Human Research Ethics Committee (HREC) at the Royal Children’s Hospital (RCH) from 1983 to 1993; La Trobe University HREC from 1994 to 1995; RCH HREC from 1996 to 1997; University of Melbourne HREC from 1998 to 2000; and the Australian Institute of Family Studies Ethics Committee from 2000. ATPG3 protocols have been approved by RCH HREC and have been ratified by Deakin University and The University of Melbourne. Parents of the ATP G3 sample have given written informed consent.

The authors state that they have obtained appropriate institutional review board approval or have followed the principles outlined in the Declaration of Helsinki for all human or animal experimental investigations. In addition, for investigations involving human subjects, informed consent has been obtained from the participants involved.

Acknowledgments

The Australian Temperament Project (ATP) and Generation 3 (ATPG3) studies are located at The Royal Children’s Hospital Melbourne and are a collaboration between Deakin University, The University of Melbourne, La Trobe University, The Australian Institute of Family Studies, The University of New South Wales, The University of Otago (New Zealand) and the Royal Children’s Hospital; further information is available at https://www.melbournechildrens.com/atp/. The views expressed in this paper are those of the authors and may not reflect those of their organizational affiliations, nor of other collaborating individuals or organizations. The authors acknowledge all collaborators who have contributed to the ATP, especially A Sanson, M Prior, F Oberklaid and D Smart. They are grateful to all study research team members involved in data collection and management. They would also like to sincerely thank the participating families for their time and invaluable contribution to the study.

Supplementary data

To view the supplementary data that accompany this paper please visit the journal website at: www.tandfonline.com/doi/suppl/10.2217/epi-2023-0284

Financial disclosure

Funding for this work was supported by grants from the Australian Research Council (DP130101459; DP160103160; DP180102447), the National Health and Medical Research Council of Australia (APP1082406; APP1175086) and Financial Markets Foundation for Children. Data collection for the ATP study was supported primarily through Australian grants from the Melbourne Royal Children’s Hospital Research Foundation, National Health and Medical Research Council, Australian Research Council and the Australian Institute of Family Studies. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

Competing interests disclosure

The authors have no competing interests or relevant affiliations with any organization or entity with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

Data sharing statement

Inquires about collaboration are possible through our institutional data access protocol: https://lifecourse.melbournechildrens.com/data-access/. Data from this study are available upon reasonable request.