Dietary exposures, epigenetics and pubertal tempo.

Gene expression changes mediated by DNA methylation may play a role in pubertal tempo regulation, and availability of methyl donor nutrients affects these pathways. We examined first trimester maternal and adolescent diet patterns that may be associated with DNA methylation at long interspersed nucleotide (LINE-1) repetitive elements in adolescence using least absolute shrinkage and selection operator (LASSO) and calculated an 'Epigenetics-Associated Diet Score' (EADS) for each pattern; then tested the associations of these scores with pubertal tempo among adolescent boys and girls. The analytic sample included 118 boys and 132 girls aged 10-18 years. DNA methylation at LINE-1 repetitive elements was quantified. Typical maternal and adolescent nutrient intakes were estimated using food frequency questionnaires. Interval-censored time to event and ordinal regression models were used to examine associations EADS scores with pubertal tempo using physician-assessed Tanner stages and self-reported menarche, respectively, adjusted for confounders. We observed associations between maternal EADS and pubertal onset, but not pubertal progression. Each standard deviation (SD) greater maternal EADS was associated with 52% higher odds of having later onset of menarche in both cross-sectional and prospective analysis (P = 0.031 and 0.028, respectively). In contrast, we observed associations between adolescent EADS and pubertal progression, but not pubertal onset. Among boys, for each SD higher adolescent EADS, there was 13% increase in odds of slower genital progression (P = 0.050), as well as 26 and 27% increase in odds of slower left and right testicular development, respectively (P = 0.001). Epigenetic-associated diet influences pubertal tempo in a sex- and timing-specific manner.

Implications and Contribution

By examining first trimester maternal and adolescent diet patterns that may be associated with Long interspersed nucleotide (LINE-1) repetitive element DNA methylation in the adolescent, we observed different epigenetic-associated diet patterns in mothers and offspring. Most food items in the maternal diet selected in the least absolute shrinkage and selection operator (LASSO) model were rich in fat and protein, and were negatively associated with LINE-1 DNA methylation. However, selected food items of the adolescent diet included more high fiber vegetables that in turn had a positive association with DNA methylation levels. Our results also suggest a stronger effect of maternal diet with pubertal onset, while a statistically significant association was observed between adolescent diet and pubertal progression. We believe the findings will help to understand future scientific knowledge of the effect of epigenetic regulation on growth tempo and develop intervention strategies accommodating the health needs of adolescents.

Introduction

Early or late age at pubertal onset is an established risk factor for a number of reproductive tract cancers, insulin resistance, and adiposity in adulthood, as well as all-cause mortality [1-4]. Over recent decades, the risk factors for earlier or later puberty, including chemical exposures, unbalanced diet and abnormal hormone levels caused by diseases and psychological stress, have been widely studied [5-8].

Patterns of health, illness and disease risks are influenced and ‘programed’ at different stages of the life course by a combination of genetic, epigenetic and environmental factors, as articulated by the ‘Development Origins of Health and Disease’ concept [9]. Epigenetics is the study of mitotically heritable yet potentially reversible molecular modifications to DNA and chromatin without alteration to the underlying DNA sequence [10, 11]. The influence of epigenetic regulation, which includes DNA methylation, on pubertal onset has been considered in animal models and global changes in specific epigenetic factors appear to be key players in the regulation of the onset of puberty [12-14]. However, potential associations of DNA methylation with pubertal tempo (e.g. pubertal onset and progression through stages of maturation) in both sexes have not been considered in human or animal studies. In female rats, interference of DNA methylation was associated with delayed vaginal opening and compromised fecundity by inhibiting Kisspeptin (Kiss1) gene, whose product is involved in controlling expression of gonadotropin hormone-releasing hormone (GnRH) [12]. Another study found that elevated expression of neurokinin B and Kiss1 amplified GnRH secretion, triggering the onset of puberty in the mice [15].

The impact of diet on pubertal tempo has also been addressed in several epidemiological studies [16-19]. They observed that high total energy intake, as well as high animal (red meat) versus vegetable protein ratio, is associated with early menarche [16-19]. However, these studies focused primarily on macronutrients and relied on menarche as the sole puberty indicator [20, 21]. Previous animal and population-based studies provide strong evidence of diet and gene interactions [22-26]. Yet, our understanding of the impact of micronutrients on puberty as well as the underlying mechanisms is limited, especially, but not only, in boys.

Methyl donor nutrients—including folate, choline/betaine, methionine, riboflavin (B2 vitamin), pyridoxine (B6 vitamin) and cobalamin (B12 vitamin)—play essential roles in the one-carbon metabolism cycle. DNA methylation, the most extensively studied epigenetic modification [27], provides a link to the one-carbon metabolism cycle through the generation of methyl donor, S-adenosylmethionine (SAM) [27]. DNA methyltransferases methylate the carbon-5 position of cytosine bases to methylated DNA using SAM [27]. Therefore, the functioning of the cycle and subsequent availability of SAM is important for the establishment and maintenance of DNA methylation and is in part dependent on the availability of methyl donor micronutrients.

To our knowledge, no previous study has examined the potential association between methyl donor-rich diet, DNA methylation and puberty tempo. To address these research gaps, this study examines the effect of maternal and adolescent diet on puberty tempo, and in particular, how foods rich in methyl donor nutrients influence puberty through epigenetic regulation. We hypothesize that methyl donor-rich diet may have an impact on pubertal tempo via altering DNA methylation. We utilized data from an ongoing cohort study in Mexico City to (i) examine which methyl donor-rich foods from maternal first trimester and adolescent diets are associated with methylation of the surrogate marker for global methylation, LINE-1 repeats, collected during adolescence (8–14 years of age), and calculate a DNA methylation-associated dietary scores from both maternal first trimester and adolescence, and (ii) examine the association of these scores with pubertal onset and progression in adolescent boys and girls.

Results

The analytical sample included 118 boys and 132 girls who attended the Visit 1, of whom 108 boys and 114 girls remained at Visit 2. The mean age for Visit 1 was 10.4 years in boys and 10.3 years in girls; the mean age was 13.7 years in boys and 13.5 for girls at Visit 2. We observed fewer children remaining at lower pubertal stages in Visit 2. Among boys, 79.7 and 48.3% were at Tanner stage 1 for pubic hair and genital development in Visit 1; and the number dropped to 25.0 and 6.5% in Visit 2. In terms of testicular volume, the percentage of boys in the pre-pubertal stage dropped from 15.3 to 0%. Among girls, 74.2% were classified Tanner stage 1 for pubic hair and 65.9% for breast development in Visit 1; and the percentage dropped to 7.9 and 4.4%, respectively, at the later visit (Table 1). Since some mothers were not recruited until the child was born, the study sample with maternal diet information was smaller (85 boys and 92 girls). Nevertheless, we observed similar changes in pubertal stages from Visit 1 to Visit 2 (Table 1).

Table 1:

Distributions of Tanner stages and other covariates among ELEMENT children at the early teen visit (Visit 1) and again at the late-teen visit (Visit 2) for children who continued follow-up, in both adolescent and maternal diet analysis samples

	Adolescent diet sample				Maternal diet sample
Boys	Visit 1 (n = 118)		Visit 2 (n = 108)		Visit 1 (n = 102)		Visit 2 (n = 94)
	n	%	n	%	n	%	n	%
Pubic Haira
1	94	79.66	27	25.00	83	81.37	27	28.72
2	17	14.41	16	14.81	13	12.75	16	17.02
3	3	2.54	30	27.78	2	1.96	25	26.60
4	1	0.85	18	16.67	1	0.98	11	11.70
5	0	0.00	14	12.96	0	0.00	12	12.77
Missing	3	2.54	3	2.78	3	2.94	3	3.19
Genital development
1	57	48.31	7	6.48	53	51.96	7	7.45
2	43	36.44	17	15.74	35	34.31	17	18.09
3	10	8.47	26	24.07	8	7.84	22	23.40
4	5	4.24	37	34.26	3	2.94	31	32.98
5	0	0.00	18	16.67	0	0.00	14	14.89
Missing	3	2.54	3	2.78	3	2.94	3	3.19
Testicular development (L)
1–3 ml	18	15.25	0	0.00	17	16.66	0	0.00
3–11 ml	75	63.56	16	14.81	65	63.74	16	17.02
>11 ml	22	18.65	89	82.41	17	16.66	75	79.79
Missing	3	2.54	3	2.78	3	2.94	3	3.19
Testicular development (R)
1–3 ml	18	15.25	0	0.00	17	16.66	0	0.00
3–11 ml	76	64.41	16	14.81	65	63.74	16	17.02
>11 ml	20	16.95	89	82.41	16	15.68	75	79.79
Missing	4	3.39	3	2.78	4	3.92	3	3.19
Age	10.35 ± 1.61		13.72 ± 1.75		10.24 ± 1.60		13.51 ± 1.73
BMI	19.06 ± 3.14		20.43 ± 3.68		18.96 ± 3.19		20.33 ± 3.88
Household SES: quartile	n = 100				n = 98		n = 90
1	24	24.00			23	23.47	21	23.33
2	27	27.00			26	26.53	24	26.67
3	24	24.00			22	22.45	22	24.44
4	25	25.00			27	27.55	23	25.56
Girls	Visit 1 (n = 132)		Visit 2 (n = 114)		Visit 1 (n = 117)		Visit 2 (n = 103)
	n	%	n	%	n	%	n	%
Pubic hair
1	98	74.24	9	7.90	92	78.63	9	8.74
2	22	16.67	39	34.21	15	12.82	38	36.89
3	9	6.82	29	25.44	8	6.84	26	25.24
4	2	1.52	21	18.42	1	0.85	17	16.50
5	1	0.76	14	12.28	1	0.85	12	11.65
Missing	0	0.0	2	1.75	0	0.00	1	0.97
Breast development
1	87	65.90	5	4.39	82	70.09	5	4.85
2	20	15.15	12	10.53	18	15.38	12	11.65
3	18	13.63	46	40.35	13	11.11	45	43.69
4	7	5.30	31	27.19	4	3.42	24	23.30
5	0	0.00	18	15.79	0	0.00	16	15.53
Missing	0	0.0	2	1.75	0	0.00	1	0.97
Menarche
Yes	30	22.73	90	78.95	22	18.80	79	76.70
No	102	77.27	23	20.17	95	81.20	23	22.33
Missing	0	0.00	1	0.88	0	0.00	1	0.88
Age	10.30 ± 1.72		13.54 ± 1.75		10.12 ± 1.66		13.37 ± 1.72
BMI	19.66 ± 3.95		21.61 ± 4.07		19.63 ± 3.87		21.80 ± 4.16
Household SES: quartile	n = 102				n = 108		n = 94
1	25	24.51			30	27.78	23	24.47
2	29	28.43			28	25.93	26	27.66
3	24	23.53			26	24.07	24	25.53
4	24	23.53			24	22.22	21	22.34

aDistribution were listed across different Tanner Stages. Also applies for Genital Development and Breast Development.

As analyzed in our previous published results, LINE-1 methylation was higher among boys compared with girls in our study cohort [Averaged LINE-1 methylation across four CpG sites (mean/SD): 78.98/2.39 in boys, 78.07/2.15 in girls]. Unpaired t-test suggested a significant sex-difference of LINE-1 methylation (P value = 0.002) while actual methylation differences were minimum. The food groups that contributed to the Epigenetics-Associated Diet Score (EADS) were different for maternal diet and adolescent diet (Table 2). LASSO-selected items that contributed to the maternal EADS included: high-protein, high-fat and high-carbohydrate food items, such as high-fat dairy, yogurt, beef, chicken, potato, refined grains and whole grains. The contributions from some items were positive while others negative. High-protein and high-fat food items were negatively associated with LINE-1 methylation, and thus diets with a high frequency of these items resulted in lower maternal EADS. Overall, whole grains were positively associated with LINE-1 methylation levels, and thus diets rich in these diets had higher EADS.

Table 2:

Patterns of diets related to child’s LINE-1 methylation at Visit 1 using LASSO feature selection

Maternal diet pattern
Food item	Estimatea	Average intake (serving/day) in study sample: mean (± SD)
High-fat dairy	−22.46	1.00 ± 0.76
Yogurt	−18.88	0.45 ± 0.44
Beef	−37.73	0.26 ± 0.20
Potato	30.12	0.31 ± 0.23
Refined grain	−2.41	2.42 ± 1.15
Chicken	−9.41	0.35 ± 0.25
Whole grain	2.77	0.37 ± 0.61
Adolescent diet pattern
Food Item	Estimate	average intake (g/day) in study sample: mean (± SD)
Tomato	1.55	6.34 ± 12.12
Yogurt	0.20	91.04 ± 73.97
Fish	−0.78	10.35 ± 10.57
Egg	−0.33	30.49 ± 24.76
Cruciferous vegetables	0.55	5.44 ± 7.27
Leafy greens	0.05	42.58 ± 49.32
Pork	0.11	7.95 ± 11.82
Other vegetables	−0.04	84.40 ± 68.41

aEstimate = β * 100.

The adolescent EADS comprised fresh high-fiber vegetables, including tomato, cruciferous vegetables, leafy greens, other vegetables, as well as lean protein sources, such as yogurt, fish, egg and pork. Similarly, both positive and negative directions were observed between adolescent food items and LINE-1 methylation. Fresh vegetables were positively associated with LINE-1 methylation level, whereas lean protein food items were negatively associated, and thus resulted in higher or lower adolescent EADS, respectively (Table 2).

Maternal and adolescent EADS had different strengths of associations with pubertal onset and progression. We observed significant associations between maternal, but not adolescent EADS, on pubertal onset in girls (Table 3). In the adjusted analysis, each SD increase of the maternal EADS was associated with a 76% lower probability of having menarche at Visit 1 (P = 0.059), and 33% lower probability of having menarche at Visit 2 (P = 0.028). Each SD increase of the maternal EADS was associated with approximately half a year increase of the age at menarche (P = 0.031). However, statistically significant associations were not observed between adolescent EADS and most indicators of pubertal onset.

Table 3:

Associations between predicted ‘maternal’ and ‘adolescent’ EADS and Visit 1 (early teen) as well as Visit 2 (late-teen) pubertal onset, accounted for MR,

Maternal diet
Boys (n = 85)	Pubic hair		Genital development			Testicular volume (L)		Testicular volume (R)
	Hazard ratio (CI)		Hazard ratio (CI)			Hazard ratio (CI)		Hazard ratio (CI)
	Visit 1	Visit 2	Visit 1	Visit2	Visit 1		Visit 2	Visit 1	Visit 2
	Adjusted	Adjusted	Adjusted	Adjusted	Adjusted		Adjusted	Adjusted	Adjusted
	0.92 (0.30. 2.85)	0.91 (0.63, 1.31)	0.83 (0.54, 1.27)	1.17 (0.88, 1.56)	1.19 (0.82, 1.72)		1.28 (0.86, 1.91)	1.14 (0.81, 1.62)	1.20 (0.84, 1.73)
Girls (n = 92)	Pubic hair		Breast development			Menarche (Y/N)		Menarche age
	Hazard ratio (CI)		Hazard ratio (CI)			Hazard ratio (CI)		Hazard ratio (CI)
	Adjusted	Adjusted	Adjusted	Adjusted	Adjusted		Adjusted	Adjusted	Adjusted
	1.17 (0.47, 2.96)	0.76 (0.50, 1.15)	1.42 (0.81, 2.49)	1.24 (0.84, 1.84)	0.24 (0.06, 1.06)		0.67 (0.47, 0.96)	0.48 (0.25, 0.93)	0.85 (0.68, 1.06)
Adolescent diet
Boys (n = 118)	Pubic hair		Genital development			Testicular volume (L)		Testicular volume (R)
	Hazard ratio (CI)		Hazard ratio (CI)			Hazard ratio (CI)		Hazard ratio (CI)
	Visit 1	Visit 2	Visit 1	Visit2	Visit 1		Visit 2	Visit 1	Visit 2
	Adjusted	Adjusted	Adjusted	Adjusted	Adjusted		Adjusted	Adjusted	Adjusted
	0.53 (0.21. 1.33)	0.88 (0.64, 1.21)	0.99 (0.70, 1.38)	0.80 (0.61, 1.06)	1.31 (0.82, 2.07)		1.32 (0.80, 2.15)	1.45 (0.89, 2.35)	1.46 (0.86, 2.47)
Girls (n = 132)	Pubic hair		Breast development			Menarche (Y/N)		Menarche age
	Hazard ratio (CI)		Hazard ratio (CI)			Hazard ratio (CI)		Hazard ratio (CI)
	Adjusted	Adjusted	Adjusted	Adjusted	Adjusted		Adjusted	Adjusted	Adjusted
	1.17 (0.67, 2.05)	0.71 (0.47, 1.07)	0.95 (0.61, 1.49)	0.78 (0.53, 1.15)	0.97 (0.65, 1.43)		0.88 (0.66, 1.19)	0.98 (0.71, 1.36)	0.93 (0.74, 1.16)

aAdjusted for age, BMI and SES status at Visit 1 (early teen).

bBolded value indicates the association is significant with a P value < 0.1.

cModel 2: β0 + β1*EADS + β2*MR + β3*Covariates.

In terms of pubertal progression, we observed statistically significant associations between adolescent EADS, but not for the associations with maternal EADS (Table 4). Among boys, for each SD higher in the adolescent EADS, there was 13% increased odds of slower genital progression (P = 0.050), as well as 26 and 27% increased odds of slower testicular developments (Left and right: P = 0.001 and 0.001). Among girls, each standard deviation higher in the adolescent EADS score was associated with 16% increased odds of faster pubic hair progression (P = 0.082) only.

Table 4:

Associations between predicted ‘maternal’ and ‘adolescent’ EADS and pubertal progression from Visit 1 (early teen) to Visit 2 (late-teen), in adjusted multivariate regression model,b,c

Maternal diet
Boys (n = 85)	Pubic hair		Genital development		Testicular volume (L)		Testicular volume (R)
	Odds ratio (CI)		Odds ratio (CI)		Odds ratio (CI)		Odds ratio (CI)
	Main effect	Score by time	Main effect	Score by time	Main effect	Score by time	Main effect	Score by time
	0.83 (0.38, 1.79)	0.98 (0.74, 1.28)	0.77 (0.46, 1.28)	1.17 (0.91, 1.51)	1.28 (0.72, 2.27)	0.90 (0.67, 1.21)	1.27 (0.76, 2.14)	0.99 (0.74, 1.31)
Girls (n = 92)	Pubic hair		Breast development		Menarche (Y/N)
	Odds ratio (CI)		Odds ratio (CI)		Odds ratio (CI)
	Main effect	Score by time	Main effect	Score by time	Main effect	Score by time
	0.64 (0.32, 1.26)	0.86 (0.67, 1.10)	1.08 (0.68, 1.71)	0.92 (0.77, 1.10)	2.59 (1.05, 6.35)	1.19 (0.87, 1.62)
Adolescent diet
Boys (n = 118)	Pubic Hair		Genital Development		Testicular Volume (L)		Testicular Volume (R)
	Odds ratio (CI)		Odds ratio (CI)		Odds ratio (CI)		Odds ratio (CI)
	Main Effect	Score by Time	Main Effect	Score by Time	Main Effect	Score by Time	Main Effect	Score by Time
	0.89 (0.57, 1.39)	0.93 (0.79, 1.10)	1.03 (0.77, 1.38)	0.87 (0.75, 1.00)	1.94 (1.15, 3.27)	0.74 (0.61, 0.88)	2.15 (1.32, 3.51)	0.73 (0.61, 0.87)
Girls (n = 132)	Pubic Hair		Breast Development		Menarche (Y/N)
	Odds ratio (CI)		Odds ratio (CI)		Odds ratio (CI)
Main effect	Score by Time	Main effect	Score by Time	Main effect	Score by Time
	1.28 (0.88, 1.85)	1.16 (0.98, 1.36)	0.97 (0.62, 1.51)	0.98 (0.83, 1.14)	1.10 (0.55, 2.17)	0.88 (0.69, 1.12)

Odds ratios are shown from adjusted models for the main effect of methylation donor score along with the interaction between the score and time interval between visits.

aAdjusted for age, BMI and SES status at Visit 1 (early teen).

bBolded value indicates the association is significant with a P value < 0.1.

cModel: logit (Y) =: logit (Y) = β0 + β1* Age + β2*Time + β3* EADS + β4*EADS*Time + β5*MR + β6*MR*Time + β7*Age*Time + β8*covariates.

Discussion

Among participants in this Mexico City cohort, we observed different epigenetic-associated diet patterns in mothers and offspring. Most food items in the maternal diet selected in the LASSO model were rich in fat and protein, and were negatively associated with LINE-1 DNA methylation. However, selected components of the adolescent diet included more high fiber vegetables that in turn had a positive association with DNA methylation levels. We also found evidence suggesting time- and gender-specific associations between EADS and pubertal tempo. Specifically, we observed significant associations between maternal EADS and pubertal onset in girls only, and significant associations between adolescent EADS and pubertal progression predominantly in boys.

The environment during development is emerging as a strong predictor of phenotype and disease in later life. Major environmental influences on developmental plasticity, including nutrition, behavior, stress and toxicants, can act though different mechanisms and result in an array of changes to the epigenome, including DNA methylation [28]. Based on previous in vivo and in vitro studies, periods of DNA methylation lability occur in three waves: primordial germ cell methylation reprograming, post-fertilization zygotic methylation reprograming, and somatic cell differentiation methylation reprograming of adults [28]. Other than the perinatal period, growth and the hormonally active puberty as well as senescence are also vulnerable periods to a variety of external insults [29]. Previous population- and animal-based research has highlighted the importance of maternal methyl donor dietary intake on fetal development [29-31], childhood respiratory health [32, 33], high childhood cognition scores [34], healthy weight status [29, 35–37] and lower cancer risks [38]. However, its impact on timing and stages of puberty is not well understood. This study expands our understanding of both maternal first trimester diet and adolescent diet on LINE-1 DNA methylation and subsequently pubertal tempo, as well as the change in the strength of the associations with timing of the dietary exposure.

Our results showed a stronger effect of maternal diet with pubertal onset, while a statistically significant association was observed between adolescent diet and pubertal progression. The findings suggest a long-term effect of maternal diet and a short-term health impact of concurrent diet on pubertal tempo. Previous animal-based studies found evidence that maternal high-fat [39] diet or high-methyl donor [26, 40] diet, or undernutrition [41] would alter the epigenomic profile, including mRNA expression and fatty acid synthase promoter methylation of the developing offspring and result in alterations in fetal gene expression. In terms of the effect of adolescent diet, Tomizawa et al. [42] found that a 3-week folate-, methionine- and choline-deficient during the developmental phase was associated with decreased glutamate receptor 1 gene expression in the mouse hippocampus, affecting learning and memory. Previous studies have suggested that the need for methyl donors might be much greater in pregnancy when DNA methylation patterns in the developing zygote are reprogramed [30]. However, methyl donors in adolescence may contribute to the maintenance and the stability of DNA methylation, which may potentially withstand more fluctuations in availability of these nutrients.

We also observed sex-specific differences between the effects of EADS and pubertal tempo. For instance, high maternal EADS was statistically associated with later menarche onset among girls, whereas higher adolescent EADS was associated with slower progression of genital and testicular development among boys. These were potential interesting findings since they highlighted the differences in the female and male reproductive system development. According to biological evidence, females are considered the ‘fundamental sex’, in which without chemical prompting, all fertilized eggs would develop into females [43]. Gonadal differentiation occurs before the end of the embryonic period, approximately the 7th week of gestation; both the reproductive ducts, external genitalia and sex differentiates occur around the 10th week of gestation [43]. After birth, maternal and placental estrogens no longer suppress the hypogonadal production of GnRH and pituitary gonadotropin. This results in the second major surge of hormone production in female and male development. During puberty, initiated by hormonal signals from the brain to the gonads (the ovaries in a girl, the testes in a boy), transformation of the nervous, muscular and reproductive systems are promoted, height and weight growth are accelerated. However, studies have found sexual dimorphism in the growth of the hippocampus in adolescence, which was associated with a more pronounced pubertal growth in males [44, 45]. These and previous studies suggest that the embryonic period is more sensitive for the female reproductive system, while adolescence may be a more efficient time for the male reproductive system to mature. Such differences may make children more prone to the impact of diet at specific developmental periods, which was supported by our EADS findings.

It is possible that the associations between LINE-1 methylation and diet were not induced solely by methyl-donor nutrients. Previous studies showed that chronic maternal high-fat diet could modify gene expression through epigenetic changes [46-48]. In terms of health impact, animals that were exposed to high-fat diet in utero had been associated with higher susceptibility to Type 2 diabetes, overweight/obesity and non-alcoholic fatty liver disease [49]. Animal studies also provided evidence on the causal association between maternal protein restriction and alterations in DNA methylation [50, 51]. Among micronutrients, Gaedicket et al. [52] found that vitamin E deficiency resulted in reduced expression on microRNA.

This analysis has some limitations, including a relatively small sample size. Considering the sparsity and bias of the LASSO selection in high-dimensional linear regression, we applied cross-validation to estimate prediction error [53]. However, every method of statistical inference depends on a complex set of assumptions, so interpretation of these analyses should be done cautiously. In terms of data collection, maternal diet, adolescent diet and menarche information were self-reported and may not be accurate due to recall bias. We would argue that this systematic within-person error applies to all subjects equally since methyl donor-rich dietary patterns are not known in the general population. Thus, it should not distort measures of EADS and the associations between EADS and pubertal tempo [54]. Moreover, since the epigenetic programing varies by genomic loci and by cell and tissue type, we need to consider the limitation of including the LINE-1 DNA methylation from blood leukocytes as the sole indicator for individual’s global methylation status.

To our knowledge, this is the first longitudinal study examining the association between methyl donor-rich diet, DNA methylation and pubertal tempo. Our findings suggested timing- and sex-specific differences between the effects of methyl donor associated diets and pubertal tempo. Our observations may suggest the potential to develop dietary recommendations for mothers in the first trimester or for adolescents that may influence pubertal tempo. Future work in this field should consider examining other ‘developmentally plastic’ phases and include epigenome-wide DNA methylation to consider whether the same findings apply.

Materials and Methods

Study Population

The study population comprised a subset of participants from the Early Life Exposure in Mexico to ENvironmental Toxicants (ELEMENT) project, a longitudinal epidemiological study consisting of three sequentially enrolled birth cohorts [55]. The mother-child pairs were recruited at three maternity hospitals representing low- to moderate-income populations in Mexico City from 1997 to 2005. The subjects in this project were a subset of children from the second and third birth cohorts (n = 646 pairs at baseline). At the research visit after the child was born, mothers provided household and demographic information, including age, education, and previous numbers of pregnancies (Fig. 1). Their newborns were followed from birth until 4 years of age. Starting in 2010, a subset of offspring were re-contacted (n = 250; henceforth referred to as Visit 1) based on availability of prenatal and neonatal biospecimens [56]. One more peri-pubertal visit (Visit 2) was completed approximately five years later (n = 549, with 223 having also participated in the 2010 Visit 1). Fasting blood, pubertal status, anthropometry and household socioeconomic status (SES) were collected at both teen visits [57].

Figure 1:

Timeline and selection of ELEMENT subjects for the study

Laboratory Measurements and Outcomes

DNA Methylation

Blood samples were obtained at Visit 1 and collected in PAXGene tubes by trained staff following standard protocols. High-molecular weight DNA was extracted from blood leukocytes with the PAXgene Blood DNA kit (PreAnalytix, Switzerland). DNA samples were treated with sodium bisulfite using kits from Zymo or Qiagen [58]. Percent of methylated cells was then quantitatively analyzed in a consensus region of repetitive elements from the LINE-1 family. DNA methylation was quantified via pyrosequencing using a PyroMark MD at four CpG sites (Supplementary Table S1). Full details on primers, quality control (which included running unmethylated and fully methylated human DNA controls with each batch and duplicating > 10% of the samples), and analysis methods have been previously published [59]. DNA methylation levels exhibited batch effects and as such were standardized to controls included on each experimental batch (96-well plate), as previously described [59].

Dietary Intake

Diets of pregnant women were assessed during the first trimester using an interviewer-administered semi-quantitative food frequency questionnaire (FFQ) designed to allow recall of dietary intake over the previous month [60]. The list of 104 food items was built from the items that proved most representative of local consumption based on the 1983 Dietary Survey of the Mexican National Institute of Nutrition [61].

Usual dietary intake of adolescents over the past week was collected using a 116-item interviewer-administered semi-quantitative FFQ adapted from the 2006 Mexican Health and Nutrition Survey [61] at Visit 1. The questionnaire asked participants to recall how often they typically consumed one serving of a standard portion size of each food item (in g or ml); response options ranged from never to ≥6 times per day.

Pubertal Outcomes

Pubertal outcomes were obtained at both Visit 1 and Visit 2. Trained physicians assessed Tanner stages of breast and pubic hair growth in girls as well as Tanner stages of genitalia and pubic hair growth in boys using standardized methods at both visits [62, 63]. Outcomes were recorded with a range from Stage 1 indicating pre-puberty to Stage 5 indicating full maturation [63, 64]. Testicular volumes were measured by trained physicians using orchidometers (range from 1 to 25 ml). Occurrence and age of menarche were gathered from a self-reported questionnaire [64-66].

Covariates

Based on a priori knowledge and preliminary correlation tests of predictors, outcomes and potential confounders, covariates included in the final models were household SES and body mass index (BMI) of the child, obtained at Visit 1. The SES measure included material wellbeing (for instance, number of bedrooms in the home) and education level of the head of the household. Combined information was ranked from A to E based on the scale developed by Mexican Association of Market Research and Public Opinion [67]. Weight and height of the child were measured by trained nurses, following standardized protocols, as previously described [68]; BMI was calculated as weight over height squared (kg/m2) [68]. Children’s age was collected at both visits.

Statistical Analysis

Descriptive statistics were obtained for all variables. We created an ‘EADS’ as a weighted average of methyl donor-rich food items for maternal and adolescent diets, respectively, with weights based on the association of the food items with the mean of methylation of four CpG sites of LINE-1. The weights were obtained by using the LASSO regression with LINE-1 methylation as the dependent variable, and intake of methyl donor-rich food items as the predictors. The LASSO method is a powerful method in feature selection, which can efficiently reduce the number of variables included in the model. LASSO applies a shrinking process that penalizes the coefficients of the regression variables shrinking some to zero, while variables that still have a non-zero coefficient after the shrinking are selected to be part of the model [53]. Methyl donor-rich food items used as input for the LASSO regression were those that are known to have high content of at least one of the following six major micronutrients: folate, choline/betaine, methionine, B2/riboflavin, B6/pyridoxine and B12/cobalamin. To examine the food items used in the regression we identified, for each of these six ‘methyl donor’ nutrients, up to 20 food items with highest specific nutrient content per serving based on information gathered from National Institute of Health Dietary Supplement Fact Sheets datasets [69]. We cross-referenced the up to 20 items for each micronutrient, with food items included on the FFQ used in the study. For the ELEMENT study, some food items have been grouped into categories based on nutritional similarity and cultural relevance [66]. Because many food items are rich in more than one of the six micronutrients, in total, there was a possible of 21 unique food groups that served as predictors in the LASSO regression model (Supplementary Table S2).

We retained seven food items (high-fat dairy, yogurt, beef, potato, refined grain, chicken and whole grain) from the maternal diet and eight food items (tomato, yogurt, fish, egg, cruciferous vegetables, leafy greens, pork and other vegetables) from the adolescent diet based on cross-validated results from LASSO selections. Based on the output from the LASSO algorithm, we calculated predicted LINE-1 methylation values and named them EADS. EADS was calculated separately for maternal diet and adolescent diet. From these regression models, we then calculated residuals by taking the observed LINE-1 methylation value and subtracting the EADS. We term these residuals as ‘methylation residuals’ (MRs), and note that they represent DNA methylation variation not explained by the dietary score (Spearman correlation between EADS and MR = −0.206). To improve interpretation of model results, these scores were transformed into z-scores; i.e. coefficients from outcome models (below) can be interpreted as differences in outcome per one SD higher EADS.

Second, we examined means ± SD for the selected food groups according to categories of maternal and adolescent characteristics to identify potential confounders. We conducted a linear trend test for ordinal characteristics (age, BMI) and a Type-III Wald test for nominal characteristics (SES).

To examine the association between EADS and onset of each pubertal outcome, we performed time-to-event analysis using interval-censored regression models. Within the interval-censored time to event model, age at Visit 1 was the ‘time in follow-up’ and attainment of each of the following Tanner stages and menarche was the ‘event’. Children were classified as having experienced the ‘pubertal onset event’ of interest if Tanner stages were > 1 [63-65] for pubic hair, genital development, breast development characteristics or answered ‘Yes’ on self-reported menarche questionnaire at the visit time. Testicular volume ≤ 3 ml indicates pre-pubertal stage; testicular volume > 3 ml but ≤ 11 ml indicates pubertal onset; and > 11 ml indicates sexual maturity [63-65]. Models were adjusted for BMI and SES measured at Visit 1. Then, we repeated the model further adjusted for the MR term to test the association while controlling the effect of methylation not due to diet on puberty.

Next we assessed the association between EADS and pubertal progression, using the following ordinal regression model: logit (Y) = β0 + β1* Agei + β2*Time + β3* EADS + β4*EADS*Time + β5*MR + β6*MR*Time + β7*Age*Time + β8*covariates to analyze the potential, gender-specific associations between EADS and pubertal progression between Visit 1 and Visit 2. Agei represents the age at Visit 1. Time represents the difference between visits.

Funding

The authors acknowledge the research staff at participating hospitals and the American British Cowdray Hospital in Mexico City for providing research facilities. We thank the mothers and children for participating in the study. This work was supported by US Environmental Protection Agency (US EPA) grants RD834800 and RD83543601 and National Institute for Environmental Health Sciences (NIEHS) grants P20 ES018171, P01 ES02284401, R01 ES007821, R01 ES014930, R01 ES013744 and P30 ES017885. This study was also supported and partially funded by the National Institute of Public Health/Ministry of Health of Mexico. The contents of this publication are solely the responsibility of the grantee and do not necessarily represent the official views of the US EPA or the National Institute of Health. Further, the US EPA does not endorse the purchase of any commercial products or services mentioned in the publication.

Conflict of interest statement. None declared.

Click here for additional data file.

Click here for additional data file.