Intake of mother's milk by very-low-birth-weight infants and variation in DNA methylation of genes involved in neurodevelopment at 5.5 years of age

ABSTRACT

Background

Mechanisms responsible for associations between intake of mother's milk in very-low-birth-weight (VLBW, <1500 g) infants and later neurodevelopment are poorly understood. It is proposed that early nutrition may affect neurodevelopmental pathways by altering gene expression through epigenetic modification. Variation in DNA methylation (DNAm) at cytosine-guanine dinucleotides (CpGs) is a commonly studied epigenetic modification.

Objectives

We aimed to assess whether early mother's milk intake by VLBW infants is associated with variations in DNAm at 5.5 y, and whether these variations correlate with neurodevelopmental phenotypes.

Methods

This cohort study was a 5.5-y follow-up (2016–2018) of VLBW infants born in Ontario, Canada who participated in the Donor Milk for Improved Neurodevelopmental Outcomes trial. We performed an epigenome-wide association study (EWAS) to test whether percentage mother's milk (not including supplemental donor milk) during hospitalization was associated with DNAm in buccal cells during early childhood (n = 143; mean ± SD age: 5.7 ± 0.2 y; birth weight: 1008 ± 517 g). DNAm was assessed with the Illumina Infinium MethylationEPIC array at 814,583 CpGs. In secondary analyses, we tested associations between top-ranked CpGs and measures of early childhood neurodevelopment, e.g., total surface area of the cerebral cortex (n = 41, MRI) and Full-Scale IQ (n = 133, Wechsler Preschool and Primary Scale of Intelligence-IV).

Results

EWAS analysis demonstrated percentage mother's milk intake by VLBW infants during hospitalization was associated with DNAm at 2 CpGs, cg03744440 [myosin XVB (MYO15B)] and cg00851389 [metallothionein 1A (MT1A)], at 5.5 y (P < 9E−08). Gene set enrichment analysis indicated that top-ranked CpGs (P < 0.001) were annotated to genes enriched in neurodevelopmental biological processes. Corroborating these findings, DNAm at several top identified CpGs from the EWAS was associated with cortical surface area and IQ at 5.5 y (P < 0.05).

Conclusions

In-hospital percentage mother's milk intake by VLBW infants was associated with variations in DNAm of neurodevelopmental genes at 5.5 y; some of these DNAm variations are associated with brain structure and IQ.

This trial was registered at isrctn.com as ISRCTN35317141 and at clinicaltrials.gov as NCT02759809.

Introduction

Breastfeeding supports optimal growth in infants, protects against childhood infections, and is associated with higher scores on neurodevelopmental assessments (1). The magnitude of improved cognitive development in human milk–fed compared with formula-fed infants appears greatest for those born preterm (2). Although the mechanisms are not fully elucidated, human milk compared with formula is better tolerated (3), allowing for faster progression from parenteral to enteral feeding in infants born very low birth weight (VLBW; <1500 g), hence providing an improved plane of nutrition. Furthermore, bioactive and immunomodulatory factors found in mother's milk may reduce the risk of sepsis and necrotizing enterocolitis in VLBW infants (3, 4). These morbidities have, in turn, been negatively associated with neurodevelopment (5, 6).

It has been proposed that early-life exposures, such as poor nutrition, adversely affect neurodevelopment by changes in gene expression that do not affect the underlying gene sequence—epigenetic modifications (7). DNA methylation (DNAm) at cytosine-guanine dinucleotides (CpGs) is the most commonly studied epigenetic modification in humans, playing a key role in regulating temporal and tissue-specific gene expression (8). Controlled studies using animal models suggest prenatal insults, including maternal malnutrition or uteroplacental insufficiency, can affect epigenetic programming that is associated with disease susceptibility (i.e., metabolic disease) in later life (9). In addition, studies in term-born infants suggest a relation between breastfeeding and DNAm (10). However, to our knowledge, no study has explored whether feeding mother's milk to infants, term or VLBW, is associated with DNAm and its further impact on brain development in early childhood (7, 8).

We hypothesized that intake of mother's milk by VLBW infants during initial hospitalization is associated with epigenetic variations that influence molecular pathways associated with neurodevelopment in early childhood. Using an epigenome-wide association study (EWAS), we assessed associations between intake of mother's milk as a percentage of total enteral feeds by VLBW infants (percentage mother's milk) during initial hospitalization and DNAm of their buccal cells at 5.5 y of age. Next, to facilitate biological interpretation of the EWAS results, we performed gene set enrichment analysis (GSEA) (11) to test if genes annotated to the EWAS-identified CpGs were enriched in any “biological process” (BP) gene sets (i.e., groups of genes that share a common biological function), as specified in the Gene Ontology (GO) project (12, 13). Finally, to corroborate that observed associations between percentage mother's milk and variations in DNAm of genes involved in molecular pathways related to neurodevelopment, we investigated whether percentage mother's milk was associated with neurodevelopmental phenotypes, and whether DNAm variations associated with percentage mother's milk correlated with these neurodevelopmental phenotypes. To our knowledge, this is the first study to investigate the association of DNAm with early intake of mother's milk by VLBW infants at an epigenome-wide scale, and its impact on neurodevelopment in children born VLBW, a population at high risk of suboptimal neurodevelopment.

Methods

Study design

This cohort study utilized demographic, feeding, and neurodevelopmental follow-up data from children born VLBW who originally participated in the Donor Milk for Improved Neurodevelopmental Outcomes randomized controlled trial (ISRCTN35317141), which evaluated the use of supplemental donor milk compared with preterm formula when mother's milk was unavailable during initial hospitalization. Neurodevelopment, growth, and body composition findings from the original trial have been published (14–16). DNAm findings obtained at 5.5 y are presented for the first time in this observational study. Briefly, enrollment of VLBW (birth weight <1500 g) infants into the trial occurred within 96 h of birth from 4 tertiary neonatal intensive care units in Southern Ontario, Canada. Infants with severe birth asphyxia or congenital/chromosomal anomalies potentially affecting neurodevelopment were ineligible. Daily enteral feed type (i.e., mother's milk, donor milk, formula) and volumes were collected prospectively from medical charts and milk preparation room records. Use of mother's milk in the present study was defined as “percentage of total enteral intake fed as mother's milk.” This definition does not include supplemental donor milk or preterm formula. Birth weight, sex, maternal age, household income, maternal education, and ethnicity were collected from medical charts or parental report.

Of 316 surviving trial participants, 158 (50%) children were enrolled in the 5.5-y follow-up study between June 2016 and July 2018 (Supplemental Figure 1) to assess neurodevelopment, growth, and body composition (NCT02759809). No differences in baseline characteristics were observed between participants and nonparticipants at follow-up, except maternal education which was higher among participants (15). The follow-up study was approved by The Hospital for Sick Children Research Ethics Board and guardian written consent and child verbal assent were obtained. We found that neurodevelopment at 18 mo (14) and body composition at 5.5 y (15) were comparable between the donor milk and preterm formula feeding groups. In addition, infants who achieved protein, fat, and energy enteral intake recommendations in the first postnatal month demonstrated improved brain white-matter microstructure at 5.5 y compared with those who did not achieve these recommendations (17). During the 5.5-y study visit, height and weight were measured and BMI (in kg/m²) was calculated as described previously (15). A subset of children within predefined in-hospital feeding groups (≥20% feeds as donor milk, ≥20% feeds as formula, exclusive mother's milk) were invited to undergo brain MRI. Children were excluded if they had metal implants, cerebral palsy, or vision or hearing impairments, which could all affect completion of scans.

DNAm of buccal cells

At the 5.5-y follow-up, buccal epithelial cells were collected using a sterile cytobrush by swabbing the inside of the cheek with the Gentra Puregene Buccal Cell Kit (Qiagen Inc.), and DNA was extracted per the manufacturer's instructions (18). Buccal cells, unlike blood cells, arise from the same germ-cell layer as the brain (19), thus they were chosen to study associations between early intake of mother's milk and epigenetic programming in the present study. It further avoided a blood draw, which, in our experience, is a significant disincentive for families contemplating participation in research not associated with clinical care.

DNAm was assessed with the Illumina Infinium MethylationEPIC BeadChip array, interrogating >850,000 CpG sites across the genome. Supplemental Table 1 summarizes citations for bioinformatic tools/resources and R statistical packages used in this study. DNAm data were processed, normalized, and evaluated following standard quality-control pipelines. Batch effects were assessed (Supplemental Figure 2). We excluded 1726 probes with a failure rate > 20% across samples (detection P > 0.01 set as the failure threshold). Subset-quantile within array normalization was applied. Finally, we filtered out CpGs within single nucleotide polymorphisms (SNPs) and on sex chromosomes. Further, genome-wide genotyping was conducted using the Infinium Omni2.5Exome-8 BeadChip array. SNPs with a genotype missing rate > 5% or minor allele frequency < 1% were filtered out. Principal component (PC) analysis was applied to the genotype data (Supplemental Figure 3), and the top 5 PCs were used as covariates in EWAS to control for potential confounding by population stratification (mixed ethnicity in our sample) (20). Furthermore, kinship coefficients (21) were estimated and used to adjust for family relatedness.

Brain imaging

T1-weighted anatomical images were collected [3D-MPRAGE, repetition time (TR)/echo time (TE) = 1870/3.14 ms, flip angle (FA) = 9°, field of view (FOV) = 240 × 256 mm, slices n = 192, resolution = 0.8 mm isotropic] on a 3T MRI scanner (MAGNETOM Siemens PrismaFIT) with a 20-channel head and neck coil. Images were filtered using a spatial adaptive nonlocal means (SANLM) denoising filter (22) and processed through the Corticometric Iterative Vertex-based Estimation of Thickness (23) (version 2.1.0) pipeline on the CBRAIN platform (24). Cortical thickness was measured in native space as the distance between gray- and white-matter surface boundaries for each vertex in the automated anatomical labeling atlas. Vertex-based cortical surface area was also calculated. Mean thickness and total surface area of the cerebral cortex were chosen as brain phenotypes herein, because they provide insight into neurodevelopment (25, 26). Whereas cortical surface area becomes stable after early childhood (∼2–4 y of age), cortical thickness continues to change (decrease) throughout childhood and adolescence (27). Thicker cortices and larger surface area at ages 1 and 2 y are positively associated with cognitive abilities (28). Only surface area (and not thickness) is associated (positively) with cognitive abilities at 10 y of age (29).

Cognitive assessment

The Wechsler Preschool and Primary Scale of Intelligence, 4th edition (WPPSI-IV) with Canadian norms (30) was administered by blinded assessors to measure the cognitive development of participants. The WPPSI-IV is a standardized assessment that yields scores with a mean of 100 and SD of 15 for Full-Scale IQ as well as the following index scores: Verbal Comprehension, Visual Spatial, Fluid Reasoning, Processing Speed, Working Memory, and Vocabulary Acquisition. Children unable to complete testing owing to disability were removed from the current analysis. Scores were dichotomized into <80 or ≥80, with scores <80 classified as borderline or extremely low as defined by the WPPSI-IV manual (30) and indicative of suboptimal development. In sensitivity analyses, a score of 49 was imputed for missing values caused by a child being unable to complete the WPPSI-IV owing to severe disability or performing below the threshold of the test (Supplemental Figure 1).

Statistical analyses

Primary analyses: EWAS and GSEA

Before statistical analysis, DNAm β-values were transformed to M-values, using the following equation (31):

(1)

For each of 814,583 CpG sites, a multivariable linear mixed-effects regression model was fitted to test the association between percentage mother's milk during hospitalization (exposure) and DNAm (outcome), using the “coxme” package (see Supplemental Table 1 for details on the statistical and bioinformatics packages used). Genetic relatedness was modeled by a random intercept with a covariance matrix based on kinship coefficients estimated from our genotype data. Models were adjusted for sex, population stratification, maternal education (university compared with no university), and percentages of fibroblast and immune cells estimated using the “EpiDISH” package (see Supplemental Table 1). Batches did not show systematic high or low signal intensities (Supplemental Figure 2); thus, no adjustment was made in the analyses for batch effects. To control for subtle population structure not captured, a calculated genomic inflation factor of 1.06 was used to adjust EWAS P values in subsequent analyses. A CpG with P < 9E−08 was considered statistically significant at the epigenome-wide level (32). To ensure in-hospital supplemental donor milk or formula feeding was not confounding percentage mother's milk–DNAm associations, we examined the effect size estimates and P values for the EWAS-significant CpGs from regression models, with and without additional adjustment for supplemental milk group. To further interrogate the possibility of confounding in examination of the associations between percentage mother's milk and DNAm, additional analyses were conducted on the top 20 CpGs from the EWAS. In these analyses, the following variables were added 1 at a time to statistical models to assess their impact on study findings: birth weight, gestational age at birth, small for gestational age at birth, maternal smoking, respiratory support, in-hospital nutrient intake, brain injury, morbidity during hospitalization, days of hospitalization, and diet quality (assessed using the Healthy Eating Index-2010) at 5.5 y (33) (Table 1).

TABLE 1.

Baseline characteristics of very-low-birth-weight infants and their families in the study sets¹

Characteristic	EWAS analysis (n = 143)	Corroboration (MRI) analysis (n = 41)
Sex
Male	73 (51.0)	18 (43.9)
Female	70 (49.0)	23 (56.1)
Birth gestational age, wk	27.8 ± 2.5	27.8 ± 1.8
Birth weight, g	1008 ± 264	1005 ± 262
Small for gestational age
Yes	19 (13.3)	6 (14.6)
No	124 (86.7)	35 (85.4)
Percentage mother's milk	72.0 ± 36.8	68.2 ± 37.8
Maternal education
University	78 (54.5)	23 (56.1)
No university	65 (45.5)	18 (43.9)
Family income
Below poverty line	28 (19.6)	6 (14.6)
Above poverty line	110 (76.9)	34 (82.9)
Missing	5 (3.5)	1 (2.4)
Duration of breastfeeding, d	228.0 ± 185.7	249.4 ± 186.5
Twin pairs, n	20	2
Mother's ethnicity
Asian	23 (16.1)	9 (22.0)
European	65 (45.5)	18 (43.9)
Middle Eastern	26 (18.2)	8 (19.5)
Other/mix	29 (20.3)	6 (14.6)
Maternal smoking
Yes	10 (7.0)	3 (7.3)
No	133 (93.0)	38 (92.7)
Brain injury2
Yes	22 (15.4)	5 (12.2)
No	121 (84.6)	36 (87.8)
Positive pressure ventilation3
Yes	96 (67.1)	29 (70.7)
No	47 (32.9)	12 (29.3)
Morbidity4
Yes	49 (34.3)	13 (31.7)
No	94 (65.7)	28 (68.3)
Length of hospitalization, d	84.3 ± 38.5	83.8 ± 36.4
Diet quality at 5.5 y5
Better	103 (72.0)	35 (85.4)
Poor	40 (28.0)	6 (14.6)
Macronutrient intake (1st week in hospital), g ⋅ kg−1 ⋅ d−1
Protein	3.0 ± 0.4	3.1 ± 0.4
Lipids	2.3 ± 0.8	2.4 ± 0.7
Energy	69.6 ± 11.2	71.6 ± 11.0

¹Values are mean ± SD or n (%) unless otherwise indicated. EWAS, epigenome-wide association study.

²Brain injury is defined as echodense intraparenchymal lesions, periventricular leukomalacia, porencephalic cysts, or ventriculomegaly with or without intraventricular hemorrhage.

³Positive pressure ventilation is yes if the number of days with respiratory support (mechanical ventilation or continuous positive airway pressure) is over half of the days in hospital.

⁴Morbidity includes late-onset sepsis (positive culture in blood or cerebrospinal fluid at ≥5 d of life), chronic lung disease (requirement for oxygen support at 36 wk), necrotizing enterocolitis (Bell Stage ≥ II), and retinopathy of prematurity (International stage 4/5, laser or intravascular injection).

⁵Healthy Eating Index-2010 score >50 means better diet quality; otherwise, poor quality.

GSEA was carried out to evaluate potential biological pathways or functions that could be potentially epigenetically modified by mother's milk, by characterizing the top-ranked CpGs with P < 0.001 identified from our EWAS. To do this, we first obtained the set of genes annotated to the top-ranked CpGs (annotation for Illumina's EPIC array, see Supplemental Table 1). Using the Database for Annotation, Visualization, and Integrated Discovery (DAVID) (see Supplemental Table 1), the identified gene set was then tested for whether any BP categories (i.e., groups of genes sharing common biological pathways and functions) were enriched (over-represented) in the set. For biological categories, we considered the BP terms curated by the GO project (12, 13). The GO-BP terms with false discovery rate <0.05 were considered to be statistically enriched.

To help understand possible regulatory roles of DNAm in the identified biological pathways, we examined the genomic annotation of the EWAS-significant CpGs using the UCSC Genome Browser (https://genome.ucsc.edu/) (see Supplemental Table 1), a graphical viewer for exploring and displaying various biological data available on the genome published in the previous literature.

Secondary corroboration analyses

To corroborate associations between percentage mother's milk and DNAm at CpGs in genes involved in neurodevelopment, we performed the following analyses.

First, we tested if percentage mother's milk was associated with neurodevelopmental phenotypes, namely cortical surface area (linear regression), mean cortical thickness (linear regression), and Full-Scale IQ (logistic regression). In basic models, we adjusted for sex and maternal education. To see whether these associations were confounded by any other variables, we also examined associations adjusted for the additional covariates listed in Table 1 using stepwise regression. Next, we tested if DNAm of the top 20 CpGs from the EWAS of percentage mother's milk was associated with neurodevelopmental outcomes. In this step, we only used phenotypes that showed significant association with percentage mother's milk (P < 0.05) in the previous step. We used a linear mixed-effects model for brain structure phenotypes (“coxme” package, see Supplemental Table 1) and a generalized linear mixed-effects model for Full-Scale IQ (“GMMAT” package, see Supplemental Table 1) with adjustments as described in the EWAS analysis. Owing to the exploratory nature of the analyses, we used the nominal significance level of 0.05 (i.e., without multiple-testing correction) for declaring statistical significance.

Results

Of 158 participants in this 5.5-y follow-up, 13 families declined buccal cell collection and 2 had maternal education missing (Supplemental Figure 1). Thus, 143 participants were included in the EWAS analysis. These participants had a mean ± SD birth weight of 1008 ± 264 g, gestational age of 27.8 ± 2.5 wk, and included 20 pairs of twins. On average, children were (mean ± SD) 5.7 ± 0.2 y of age and had healthy body weights with a BMI z score of −0.27 ± 1.21. There were no statistically significant differences in baseline characteristics between those included in the EWAS (n = 143) and corroboration (MRI) analyses (n = 41) (Table 1, Supplemental Figure 4). The majority of the 143 participants had WPPSI-IV scores available; missing scores were due to the child's inability to complete the assessment owing to disability, child refusal, or tester error (Full-Scale IQ, n = 10; Verbal Comprehension, n = 8; Visual Spatial, n = 9; Fluid Reasoning, n = 7; Processing Speed, n = 12; Working Memory, n = 6; Vocabulary Acquisition, n = 4).

Percentage mother's milk and DNAm: EWAS and GSEA

The EWAS identified 2 CpGs, cg03744440 [myosin XVB (MYO15B)] and cg00851389 [metallothionein 1A (MT1A)], associated with percentage mother's milk at the EWAS-significant level (P = 2.09E−08 and 4.67E−08, respectively) (Figure 1). All top 20 CpGs were less methylated in children who consumed a higher percentage of mother's milk (Table 2, Supplemental Figure 5). For both EWAS-significant CpGs, further adjustment for supplemental milk type did not change the results (Supplemental Table 2). Additional analyses on the top 20 CpGs of the EWAS showed that inclusion of potentially confounding variables did not affect study findings (Supplemental Figure 6). Confounding variables assessed included infant birth weight, gestational age at birth, size at birth, maternal smoking, respiratory support, macronutrient intake in hospital, days of hospitalization, and diet quality at 5 y of age. Inclusion of breastfeeding duration over the first year as a covariate was not considered because it was collinear with in-hospital percentage mother's milk. Similarly, family income was not included, because it was highly correlated with maternal education, which was already included in the base regression model (correlation P = 1.6E−05).

FIGURE 1.

Manhattan (A) and Volcano (B) plot of 814,583 CpGs based on the EWAS analysis (n = 143). (A) Chromosome positions are displayed along the horizontal axis, with negative logarithm of association P values for each CpG site presented on the vertical axis. (B) Effect size estimates of mother's milk on DNA methylation against negative log P values are displayed. In both plots, the solid horizontal line represents the EWAS significance threshold P = 9E−08, and the dashed line indicates the threshold level of P = 0.001, which was used to select the subset of CpGs for gene set enrichment analysis and secondary analyses. CpG, cytosine-guanine dinucleotide; EWAS, epigenome-wide association study; MT1A, metallothionein 1A; MYO15B, myosin XVB.

TABLE 2.

Top-ranked CpGs in epigenome-wide association study analysis of percentage mother's milk (primary analysis) and their associations with total surface area of the cerebral cortex and full-scale IQ (secondary analyses)¹

CpG	Chromosome	Position	Gene symbol	Gene name	Primary		Secondary		Secondary
					% Mother's milk2		Total surface area3		Full-Scale IQ4
					Effect size	P value	Effect size	P value	Effect size	P value
cg03744440	17	73,584,108	MYO15B	Myosin XVB	−0.70	2.09E−08	−61.15	0.117	−1.69	0.050
cg00851389	16	56,669,681	MT1A	Metallothionein 1A	−0.39	4.67E−08	48.27	0.460	−2.79	0.053
cg05406347	5	133,773,142	—	—	−0.33	1.36E−07	−48.12	0.571	−3.04	0.063
cg11298446	10	119,749,726	—	—	−0.48	2.34E−07	−88.28	0.086	−0.82	0.452
cg10997248	11	117,699,019	FXYD2	FXYD domain containing ion transport regulator 2	−0.42	3.79E−07	−61.23	0.249	−1.16	0.342
cg27176927	11	120,386,965	GRIK4	Glutamate ionotropic receptor kainate type subunit 4	−0.36	3.81E−07	−127.4	0.050	−1.16	0.373
cg13424029	10	101,297,508	—	—	−0.34	4.31E−07	−36.68	0.565	−3.08	0.045
cg20565065	9	130,542,853	—	—	−0.29	6.10E−07	−52.17	0.572	−1.62	0.317
cg23615741	10	101,297,642	—	—	−0.48	1.64E−06	−54.86	0.167	−1.90	0.070
cg07398661	15	37,111,929	CSNK1A1P	Casein kinase 1 α 1 pseudogene 1	−0.28	2.18E−06	−177.4	0.012	−2.73	0.152
cg01765653	6	29,599,160	GABBR1	γ-Aminobutyric acid type B receptor subunit 1	−0.22	2.63E−06	−204.8	0.012	−0.33	0.877
cg13313522	12	107,024,230	RFX4	Regulatory factor X4	−0.33	2.71E−06	−152.2	0.018	−2.15	0.169
cg14053590	7	100,321,217	EPO	Erythropoietin	−0.22	3.67E−06	−173.5	0.036	−0.17	0.932
cg09941581	4	124,220,074	SPATA5	Spermatogenesis associated 5	−0.49	3.69E−06	−18.64	0.718	−1.86	0.096
cg08145067	19	3,688,176	PIP5K1C	Phosphatidylinositol-4-phosphate 5-kinase type 1 γ	−0.26	3.95E−06	−166.5	0.016	−0.38	0.859
cg08639339	1	205,648,336	SLC45A3	Solute carrier family 45 member 3	−0.23	4.39E−06	−59.54	0.482	−2.79	0.239
cg11724691	11	72,475,966	STARD10	StAR related lipid transfer domain containing 10	−0.22	4.70E−06	−139.4	0.163	−1.31	0.487
cg19619387	17	1,881,036	RTN4RL1	Reticulon 4 receptor like 1	−0.48	4.75E−06	−34.09	0.558	−1.49	0.121
cg01883425	6	41,606,770	MDFI	MyoD family inhibitor	−0.51	4.80E−06	−40.83	0.252	−0.73	0.432
cg23575219	9	115,250,537	KIAA1958	KIAA1958	−0.31	4.82E−06	19.23	0.760	−3.10	0.069

¹CpG, cytosine-guanine dinucleotide; DNAm, DNA methylation; PC, principal component.

²Values are the coefficient and corresponding P value of percentage mother's milk in the regression, DNAm = % mother's milk + sex + mother's education + cell type + PCs (n = 143).

³Values are the coefficient and corresponding P value of DNAm in the regression, total surface area (cm²) = DNAm + sex + mother's education + cell type + PCs (n = 41).

⁴Values are the coefficient and corresponding P value of DNAm in the regression, Full-Scale IQ = DNAm + sex + mother's education + cell type + PCs (n = 133).

In total, 788 genes mapped to the top 1346 differentially methylated CpGs (P < 0.001) were used in the GSEA. The majority of enriched gene sets (i.e., GO terms) were related to the “nervous system development” pathway (parent) and more specialized (child) terms, i.e., “generation of neurons,” “neurogenesis,” and “neuron differentiation,” in the GO hierarchy (Table 3). This suggests that the biological mechanisms of DNAm associated with percentage mother's milk may be involved in the neurodevelopment of children born VLBW.

TABLE 3.

Enriched biological pathways identified based on 788 genes mapped to the CpGs from the EWAS analysis¹

Term	Term description	P value	FDR
GO:0007399	nervous system development2	6.96E−09	1.35E−05
GO:0051094	positive regulation of developmental process	2.68E−08	5.20E−05
GO:0030029	actin filament-based process	6.83E−08	1.33E−04
GO:0048699	generation of neurons2	9.96E−08	1.93E−04
GO:0035295	tube development	1.25E−07	2.43E−04
GO:0045595	regulation of cell differentiation	1.26E−07	2.45E−04
GO:0022008	neurogenesis2	1.38E−07	2.69E−04
GO:0051240	positive regulation of multicellular organismal process	1.59E−07	3.08E−04
GO:0048468	cell development	1.73E−07	3.36E−04
GO:0030182	neuron differentiation2	2.43E−07	4.72E−04
GO:0006928	movement of cell or subcellular component	2.60E−07	5.04E−04
GO:0045597	positive regulation of cell differentiation	4.70E−07	9.11E−04
GO:2000026	regulation of multicellular organismal development	6.12E−07	1.19E−03
GO:0009790	embryo development	9.80E−07	1.90E−03
GO:0051128	regulation of cellular component organization	2.02E−06	3.92E−03
GO:0051960	regulation of nervous system development2	2.37E−06	4.60E−03
GO:0035239	tube morphogenesis	2.40E−06	4.65E−03
GO:0060429	epithelium development	2.66E−06	5.17E−03
GO:0007389	pattern specification process	3.26E−06	6.33E−03
GO:0045944	positive regulation of transcription from RNA polymerase II promoter	5.06E−06	9.82E−03
GO:0050767	regulation of neurogenesis2	6.37E−06	1.24E−02
GO:0006366	transcription from RNA polymerase II promoter	8.03E−06	1.56E−02
GO:0045664	regulation of neuron differentiation2	8.45E−06	1.64E−02
GO:0052697	xenobiotic glucuronidation	9.80E−06	1.90E−02
GO:0010557	positive regulation of macromolecule biosynthetic process	1.07E−05	2.08E−02
GO:1902680	positive regulation of RNA biosynthetic process	1.09E−05	2.11E−02
GO:0072358	cardiovascular system development	1.09E−05	2.12E−02
GO:0072359	circulatory system development	1.09E−05	2.12E−02
GO:0045893	positive regulation of transcription, DNA-templated	1.11E−05	2.16E−02
GO:1903508	positive regulation of nucleic acid–templated transcription	1.11E−05	2.16E−02
GO:0031328	positive regulation of cellular biosynthetic process	1.18E−05	2.28E−02
GO:0060562	epithelial tube morphogenesis	1.19E−05	2.30E−02
GO:0060284	regulation of cell development	1.25E−05	2.42E−02
GO:0048646	anatomical structure formation involved in morphogenesis	1.32E−05	2.56E−02
GO:0009891	positive regulation of biosynthetic process	1.51E−05	2.94E−02
GO:0021915	neural tube development2	1.69E−05	3.28E−02
GO:0051130	positive regulation of cellular component organization	1.87E−05	3.63E−02
GO:0006357	regulation of transcription from RNA polymerase II promoter	1.92E−05	3.73E−02
GO:0050769	positive regulation of neurogenesis2	2.24E−05	4.34E−02
GO:0010628	positive regulation of gene expression	2.28E−05	4.43E−02

¹Gene set enrichment analysis was conducted based on 788 genes that were mapped to the top differentially methylated CpGs (P < 0.001 in the EWAS analysis). The top identified gene ontology (GO) terms are classes of genes that were overly represented in the 788 genes. The P value and false discovery rate (FDR) of enrichment test for each GO term are presented.

²Biological pathways related to nervous system development and its child terms in the GO hierarchy that reached statistical significance (FDR<0.05).

The genomic annotation on the Genome Browser indicated that the EWAS-significant CpG cg03744440, and its highly correlated CpGs, cg12184886 (r = 0.85; EWAS-P = 5.33E−06) and cg22287064 (r = 0.82; EWAS-P = 3.20E−05), are located within a 200-bp block upstream of the transcription start site of MYO15B (Supplemental Figure 7). The pattern of histone modifications [e.g., high H3K4me1 signal (34)], location within a DNase-sensitive area, and presence of binding sites for many transcription factors suggest that DNAm variations at these 3 CpGs may affect transcription of MYO15B. The genomic annotations of the other significant CpG cg00851389 (MT1A) and its correlated CpG cg00509108 (r = 0.86; EWAS-P = 1.07E−05) indicate that the CpGs are located in a region containing binding sites for multiple transcription factors and DNA-regulatory elements in embryonic stem cells (Supplemental Figure 7). Thus, DNAm variations at these 2 CpGs may influence expression of MT1A during development.

DNAm and neurodevelopmental phenotypes: corroboration studies

Associations between percentage mother's milk consumed by infants during hospitalization and both cortical surface area and Full-Scale IQ were significant such that, after adjustment for infant sex and maternal education, children fed with 100% mother's milk (compared with children fed with 0% mother's milk) had larger cortical surface area by 173.1 cm² (95% CI: 73.6, 272.5 cm²; P = 0.002), and were almost 6 times more likely to have a Full-Scale IQ ≥ 80 (OR: 5.8; 95% CI: 1.3, 26.9; P = 0.02); but associations were not significant for cortical thickness (P = 0.079). Further, when adjusted for the additional variables selected by stepwise regression (brain injury for surface area; brain injury and energy intake for Full-Scale IQ), these associations remained significant (P = 0.0006 for surface area and P = 0.002 for IQ). Therefore, we focused the following corroboration analyses on cortical surface area and Full-Scale IQ only. Screening the top 20 CpGs from the EWAS of percentage mother's milk, 6 CpGs were associated with cortical surface area and 1 CpG was associated with Full-Scale IQ (P < 0.05) (Table 2). Notably, all these associations were such that higher percentage mother's milk intake was associated with lower DNAm, and lower DNAm was associated with larger cortical surface area and higher Full-Scale IQ. Supplemental Table 3 provides the results for the IQ subscales, i.e., Verbal Comprehension, Processing Speed, and Fluid Reasoning, as well as the sensitivity analyses using imputed values for all cognitive scores.

Discussion

In this study, the intake of mother's milk as a percentage of total enteral feeding by VLBW infants was associated with DNAm at 5.5 y. This result was supported by our findings that the top-ranked CpGs associated with percentage mother's milk (P < 0.001) were annotated to genes enriched in biological pathways related to neurodevelopment (Table 3). In addition, in our corroboration analysis, several of the top 20 CpGs from the EWAS were associated with neurodevelopmental outcomes at 5.5 y (i.e., total surface area of the cerebral cortex and Full-Scale IQ) that were correlated with percentage mother's milk (P < 0.05). All these data provide another argument for preferential provision of mother's milk for feeding VLBW infants during initial hospitalization.

Our EWAS revealed that higher percentage mother's milk was associated (at the EWAS-wide significance level) with lower DNAm levels at 2 CpGs, cg03744440 (MYO15B) and cg00851389 (MT1A). MYO15B encodes an unconventional myosin, myosin XVB, a member of the myosin superfamily of actin-based molecular motor heavy-chain proteins that facilitate muscle contraction and other cellular motility processes. Although our understanding of the role of MYO15B in neurodevelopment is rudimentary, the predicted protein structure is similar to that encoded by myosin XVA (MYO15A), which is associated with the normal development of hearing and hereditary deafness in humans and mice (35). Other myosins that are known to be related to neurosensory development include myosin II, which plays a role in migrating neurons and growth cones, and in organization of actin bundles in postsynaptic spines (36). MT1A encodes for metallothionein 1A. Metallothioneins are cysteine-rich proteins that bind heavy metals and thereby protect cells from oxidative stress (37). Preterm infants are susceptible to oxidative stress owing to immaturity of their oxidative stress–defense mechanisms and higher prevalence of hypoxia, ischemia, and infection than in healthy infants born at term. Oxidative stress and associated morbidities (e.g., chronic lung disease, retinopathy of prematurity, necrotizing enterocolitis) are known contributors to suboptimal neurodevelopment (38).

The complex nutrient and bioactive (e.g., antibodies, human milk oligosaccharides, lactoferrin) composition of human milk explains many of the immediate benefits of breastfeeding, such as the reduction in infection (1). The mechanisms of how mother's milk extends its benefits to health and neurodevelopment into early childhood and beyond are largely unknown. It is postulated that epigenetic processes may play a role under the developmental origin of health and disease hypothesis (39).

In their recent review of the literature, Mallisetty et al. (10) identified 10 studies examining the association between breastfeeding and DNAm in term-born children; all suggested some relation between breastfeeding metrics and DNAm. The Avon Longitudinal Study of Parents and Children (ALSPAC), for example, reported that breastfeeding (ever compared with never) was associated with DNAm at cg11414913 (blood) with the same direction of effect at 7 and 15–17 y of age (40). A more recent analysis from the ALSPAC child cohort reported exclusive breastfeeding elicited more substantial DNAm variation, and more extreme hypo- and hyper-methylated CpG sites, during infancy than other stages of childhood (41). This variation, in turn, mediated associations between exclusive breastfeeding and children's BMI. In a smaller cohort study, term-born infants exclusively breastfed for 4 mo (compared with not exclusively breastfed for 4 mo) had lower DNAm at several CpGs in the glucocorticoid receptor gene (buccal cells) (42). Lower DNAm at 1 CpG site in this gene was inversely associated with cortisol reactivity in 5-mo-old infants. Whereas the aforementioned studies in term-born infants demonstrate associations between breast milk intake and DNAm, none to our knowledge have explored the associations between breastfeeding metrics, DNAm, and neurodevelopment (10).

The components in human milk that elicit changes in the expression of genes are not understood (43). There is very preliminary evidence that long-chain fatty acids (omega-3) or micronutrients (e.g., folate, choline, betaine, and other B-vitamins) in milk may affect DNAm or that changes in DNAm are mediated by the infant's gut microbiome whose composition, in turn, is strongly influenced by breastfeeding. The gut microbiota produce large quantities of epigenetically active metabolites, such as folate and SCFAs (butyrate and acetate). The rich source of microRNA in human milk (contained in exosomes) is also hypothesized to affect gene transcription and regulation of cellular events.

DNAm findings in healthy term-born infants may differ from those in VLBW infants. First, DNAm is significantly associated with gestational age and birth weight (8). Second, VLBW infants are differentially exposed to other factors known to be associated with DNAm, including maternal stress, excess inflammatory cytokine concentrations, and medications (8). Finally, because the advantage of breastfeeding to cognition is greater for preterm than term infants, the opportunity for mother's milk to play a role in DNAm of CpGs regulating neurodevelopmental pathways may be greater (2).

Strengths of the study include examination of the exposure (mother's milk) as percentage enteral feeds compared with a dichotomous variable (ever or never mother's milk), and corroboration of EWAS “neurodevelopmental” findings using brain MRI measurements of total surface area of the cerebral cortex (44) and cognitive assessment using the WPPSI-IV. The main study limitations are sample size and unmeasured potential confounders (e.g., mother's genetic information and maternal presence at the infant's bedside). It would be valuable to assess causal relations between mother's milk and neurodevelopmental phenotypes and mediation effects of DNAm on these relations in future studies with larger sample sizes, access to additional confounding variables, and perhaps a cluster-randomized clinical trial design where mother's milk feeding is promoted and supported in 1 arm of the study. In addition, we acknowledge we have not adjusted for multiple testing in our DNAm and neurodevelopmental phenotype corroboration analysis (secondary analyses).

In summary, we found that percentage mother's milk intake by infants during initial hospitalization was associated with variations in DNAm of genes enriched in neurodevelopmental pathways in early childhood. Our corroboration studies found that top-ranked CpGs from the present EWAS of percentage mother's milk were also associated with variations in brain structure and IQ. Given the high prevalence of suboptimal neurodevelopment in VLBW infants, future studies with a larger sample size are warranted to confirm these findings and to explore further the role of epigenetic programming in the association between mother's milk and neurodevelopment of VLBW infants.

Notes

Supported by Canadian Institutes of Health Research (CIHR) grants #FHG 129919 (to DLO) and #FDN 143233 (to DLO). CIHR had no role in the design or conduct of the study, including the collection, analysis, and interpretation of the data, in writing the manuscript, or in the decision to submit the manuscript for publication.

Author disclosures: The authors report no conflicts of interest.

Supplemental Figures 1–7 and Supplemental Tables 1–3 are available from the “Supplementary data” link in the online posting of the article and from the same link in the online table of contents at https://academic.oup.com/ajcn/.

Abbreviations used: ALSPAC, Avon Longitudinal Study of Parents and Children; BP, biological process; CpG, cytosine-guanine dinucleotide; DNAm, DNA methylation; EWAS, epigenome-wide association study; GO, Gene Ontology; GSEA, gene set enrichment analysis; MT1A, metallothionein 1A; MYO15B, myosin XVB; PC, principal component; SNP, single nucleotide polymorphism; VLBW, very low birth weight; WPPSI-IV, Wechsler Preschool and Primary Scale of Intelligence, 4th edition.

Data availability

Data described in the article, code book, and analytical code will not be made available in order to protect the privacy and confidentiality of our participants; we do not have consent from participant families to share their anonymized data, nor do we have permission from the research ethics boards of participating hospitals. For access to summary statistics for the EWAS analysis, please contact one of the corresponding authors.