Effect of an antenatal diet and lifestyle intervention and maternal BMI on cord blood DNA methylation in infants of overweight and obese women: The LIMIT Randomised Controlled Trial.

BACKGROUND: To investigate the effect of an antenatal diet and lifestyle intervention, and maternal pre-pregnancy overweight or obesity, on infant cord blood DNA methylation.
METHODS: We measured DNA methylation in 645 cord blood samples from participants in the LIMIT study (an antenatal diet and lifestyle intervention for women with early pregnancy BMI ≥25.0 kg/m2) using the Illumina 450K BeadChip array, and tested for any differential methylation related to the intervention, and to maternal early pregnancy BMI. We also analysed differential methylation in relation to selected candidate genes.
RESULTS: No CpG sites were significantly differentially methylated in relation to either the diet and lifestyle intervention, or with maternal early pregnancy BMI. There was no significant differential methylation in any of the selected genes related to the intervention, or to maternal BMI.
CONCLUSION: We found no evidence of an effect of either antenatal diet and lifestyle, or of maternal early pregnancy BMI, on cord blood DNA methylation. CLINICAL TRIALS REGISTRATION: ACTRN12607000161426.

Introduction

There is a well recognised link between maternal overweight and obesity and the risk of overweight and obesity in children. Infants born to women who are overweight or obese in pregnancy have, on average, higher birthweight for gestational age [1, 2], and higher adiposity [2, 3]. They are also recognised to be at greater risk of childhood overweight and obesity [4], and its associated health consequences later in life [5-7]. These transgenerational effects most likely have multiple causes, including environmental exposures and genetic factors, although recent attention has been focused on peri-conceptional or in-utero exposures [8]. Such exposures include maternal overweight and obesity, gestational weight gain, antenatal nutrition and physical activity, pregnancy complications (including gestational diabetes (GDM) and hypertension), as well as social and behavioural factors. The mechanisms by which these exposures contribute to an increased susceptibility to child obesity are not fully understood. Current evidence suggests shared genetics explains only a small amount of the heritability of obesity [1, 9, 10]. Other postulated mechanisms include alterations to the maternal gut microbiome [1, 11], maternal hyperinsulinaemia and hyperglycaemia in pregnancy [4, 12]. While there is some evidence of paternal influences, these have been relatively under-studied [7, 12].

Among these possibilities, epigenetic mechanisms have been the focus of much recent investigation. Epigenetics is broadly taken to refer to changes in gene function which occur in the absence of changes to the underlying DNA [7, 11, 12], resulting in changes to regulation and expression of genes via mechanisms such as DNA methylation (DNAm), histone modification, and noncoding RNAs [4, 7]. DNA methylation is the most widely studied epigenetic mechanism, and involves the attachment of a methyl group to a CpG dinucleotide, which is then passed on in DNA replication and cell division [2, 4, 13, 14]. Methylation, particularly in gene promoter regions, is generally believed to contribute to gene silencing [7, 11] although this depends on a number of factors, including the methylation site (e.g. promoter region vs. gene body) [2, 14], and the interaction between different epigenetic mechanisms (e.g. between DNAm and histone modification) [2, 4].

Epigenetic mechanisms have been proposed as a potential means by which maternal obesity predisposes to obesity in offspring, both via the metabolic effects of obesity and via maternal diet in pregnancy (though these two effects are sometimes conflated). Evidence from non-human models demonstrates that maternal diet in pregnancy can alter DNAm profiles of offspring, and that this in turn influences offspring adiposity [2, 4, 7, 10, 11, 15]. The evidence from human studies to date is less robust. Studies of cohorts of women exposed to extreme undernutrition, either periconceptionally or in pregnancy [12], have demonstrated effects on DNAm (e.g. on the IGF2 imprinted gene) in offspring, along with a predisposition to adiposity, type 2 diabetes and other metabolic disorers in later life [2, 7, 12, 14, 16]. A range of studies have also reported differential methylation at various genomic loci in neonatal cord blood associated with pre-pregnancy overweight and obesity (S1 Table), and in children born to women following bariatric surgery [5].

The aims of this prespecified secondary study were to investigate DNAm in cord blood samples from 645 participants in the LIMIT randomised controlled trial of an antenatal diet and lifestyle intervention in women with body mass index (BMI) ≥25.0 kg/m2. We undertook an epigenome-wide analysis of differential methylation related to the diet and lifestyle intervention, and/or related to maternal early pregnancy BMI. We also investigated differential methylation in selected genes where previous research had found differential methylation associated with maternal BMI, and which were plausibly related to obesity, adiposity, metabolism, or growth, namely:

IGF2 on chromosome 11: a maternally expressed imprinted gene, expression of which has been found to relate to circulating IGF in cord blood. DNA methylation in the imprinting region for IGF2 and the associated paternally-expressed imprinted gene H19, have been found to be associated with adiposity [17];

RXRA on chromosome 9: differential methylation of this gene in cord blood has been found to be associated with childhood adiposity [10, 14];

PPARGC1A on chromosome 4: a gene which regulates genes involved in energy metabolism and has been found to be differentially methylated in adults with impaired glucose tolerance and in adults exposed to high-fat overfeeding [8]; some studies have found evidence of differential DNAm in cord blood associated with maternal obesity [18, 19];

MEST, a mostly paternally-expressed imprinted gene which may play a role in adipocyte differentiation, and which has been found to be differentially methylated in cord blood of women with obesity compared to normal BMI, and also of women with GDM [20].

Methods

The LIMIT Randomised Controlled Trial

The LIMIT randomised, controlled trial evaluated the effects of an antenatal diet and lifestyle intervention for women with early pregnancy BMI ≥25.0 kg/m2, with findings extensively reported elsewhere [21]. Women were eligible if they had early pregnancy BMI ≥25.0 kg/m2, a singleton pregnancy between 10+0 and 20+0 weeks’ gestation, and no previously existing diabetes. A total of 2212 women were randomised to receive either Lifestyle Advice (n = 1108), a comprehensive diet and lifestyle intervention, or Standard Care (n = 1104), in which antenatal care was delivered according to local guidelines (and did not include information on diet or physical activity). The primary outcome was birth of an infant large for gestational age (LGA). While there were no significant differences observed between the groups in relation to this outcome, a significantly lower incidence of birthweight >4kg was observed in the Lifestyle Advice group (Relative Risk (RR) 0.82 (95% Confidence Interval (CI): 0.68, 0.99, p = 0.04). Additionally, measures of diet quality and physical activity were improved in women in the Lifestyle Advice group compared with those in the Standard Care group [22].

Cord Blood DNA for a range of secondary studies was collected at the time of birth from consenting participants, and was frozen as whole blood preserved with EDTA. Funding was available to perform DNA methylation analysis for a total of 649 samples, which were randomly selected from the total number of available samples, balanced between the Lifestyle Advice and Standard Care groups. After DNA extraction, genome-wide DNA methylation was performed using the Illumina Infinium HumanMethylation 450K Bead-Chip array. Results were supplied as raw probe intensities (idats files).

Ethics approval and consent to participate

The study was reviewed by the ethics committee of each participating institution including the Women’s and Children’s Health Network Human Research Ethics Committee (1839 & 2051); The Central and Northern Adelaide Health Network Human Research Ethics Committee (2008033) and the Southern Adelaide Local Health Network Human Research Ethics Committee (formerly Flinders Clinical Research Ethics Committee) (128/08).

Informed, written consent was obtained for all participants to participate in the LIMIT study, and additional written consent was obtained to collect samples of umbilical cord blood at delivery for the purposes of gene expression research related to weight and to the diet and lifestyle intervention.

Data processing

Data processing and analysis was performed using R version 4.0 [23]. The minfi package [24] was used to read in the raw idats files, and to calculate detection p values (comparison of methylated (M) and unmethylated (U) intensities to background signal) both for all samples (across all probes) and all probes (across all samples). There were a total of 662 sets of results in the data, as 13 samples had been rerun due to chip failure. The initial results for these samples were identified and excluded. These were the only samples classified as ‘failed’ (detection p value ≥0.01). Of the 649 valid samples, four were excluded because of labelling errors where the correct study identifier could not be ascertained. A total of 645 samples were therefore retained for processing and analysis. The raw data for these 645 samples were converted to β values and normalisation (removal of technical variation due to, e.g. probe type differences or background signal) was undertaken using the Subset-Quantile Normalisation method [25, 26] as implemented in minfi. Following normalisation, failed probes (defined as detection p value ≥0.001 in 25% or more of the 645 samples) were filtered out. Finally, probes identified as cross-reactive [27], probes with an identified SNP within 3 nucleotides of the CpG site and minor allele frequency >1%, and probes on the X and Y chromosomes were filtered out using the DMRCate package [28]. This left 426,572 probes available for analysis. Batch effects were not removed at the processing stage [29] but were instead adjusted for in analyses. The estimateCellCounts function in minfi was used, with Cord Blood reference data, to estimate the proportions of B cells, CD4T, CD8T, granulocytes, monocytes, natural killer and nucleated red blood cells, and these estimated proportions were likewise used for adjustment in the analysis.

To investigate the sensitivity of the analysis results to choice of data processing methods, effects were also estimated using models run using a range of alternative analysis datasets. Firstly, raw data were also normalised using the Beta-Mixture Quantile (BMIQ) method [30] and the Subset-Within-Array Normalisation (SWAN) method [31]. Secondly, datasets were created in which batch effects were handled using the ComBat batch-effect-removal tool [32] implemented in the ChAMP package [33] instead of adjustment for batch effects in the models. The results of analyses using these datasets are reported in brief below, but are described in detail in a separate publication.

Statistical analysis

Statistical analyses of epigenome-wide data were conducted on M-values (logit-transformed β) using linear models, with adjustment of standard errors using Empirical Bayes methods, as implemented in the limma package [34]. The primary analysis model included intervention group (Lifestyle Advice vs Standard Care), BMI (as a continuous variable), their interaction, and the additional covariates parity (0 vs 1+), maternal age (continuous), smoking status, infant sex and study centre. Sample batch and estimated cell type proportions were also included as adjustment variables as described above. The number of probes differentially expressed between Lifestyle Advice and Standard Care groups (at different levels of maternal BMI), or corresponding to differences in maternal BMI (in each of the intervention groups) were determined using the decideTests function in limma, using Benjamini-Hochberg method (controlling for a false discovery rate of 5%) to adjust for multiple comparisons.

Secondary sensitivity analyses were also carried out, including unadjusted models (adjusted for only batch and cell type proportion) and models including interactions between intervention and sex, or between BMI and sex (as it has been hypothesised that effects of maternal obesity on gene expression may differ by infant sex [11]).

For candidate gene analyses, all probes at or near (±2000bp) each of the genes of interest were extracted, and linear models were fitted for each probe separately, and for the average M-value across all probes. As above, the models included intervention group, BMI and their interaction, as well as covariates (batch, cell type proportions, parity, maternal age, smoking status, infant sex, study centre, and quintile of relative socioeconomic disadvantage). The mean difference in M-values between Lifestyle Advice and Standard Care groups, and corresponding to a 5-unit increase in maternal BMI, was estimated, along with 95% Confidence Intervals.

Results

Baseline characteristics of participants whose data is included in this analysis are described in Table 1, and are similar to those of the full LIMIT cohort [21]. The median early pregnancy BMI was 31 kg/m2 (Interquartile Range (IQR) 28–37 kg/m2. A majority of women (60%) were in their second or subsequent pregnancy, and had a mean age of 29 years (SD 5 years). Most (85%) were nonsmokers, and almost all (91%) were of Caucasian ethnicity. Half of the women were from the highest two quintiles of relative socioeconomic disadvantage. Infant sex was evenly divided between males (51%) and females (49%).

Table 1

Baseline characteristics of participants.

Characteristic	Lifestyle Advice	Standard Care	Overall
Overall Numbers	n = 325	n = 320	n = 645
BMI (kg/m2): Median (IQR)	31.40 (28.10, 36.20)	31.45 (27.98, 36.90)	31.40 (28.00, 36.50)
BMI Category: (N%)
• 25.0–29.9	129 (39.69)	130 (40.62)	259 (40.16)
• 30.0–34.9	99 (30.46)	86 (26.88)	185 (28.68)
• 35.0–39.9	58 (17.85)	55 (17.19)	113 (17.52)
• ≥40.0	39 (12.00)	49 (15.31)	88 (13.64)
Height(cm): Mean (SD)	165.29 (6.66)	164.73 (6.48)	165.01 (6.57)
Weight(kg): Mean (SD)	89.81 (17.48)	89.75 (18.65)	89.78 (18.06)
Parity: N(%)
• 0	141 (43.38)	128 (40.00)	269 (41.71)
• 1+	184 (56.62)	192 (60.00)	376 (58.29)
Age at TE: Mean (SD)	29.28 (5.56)	29.63 (5.24)	29.45 (5.41)
Smoking: N(%)
• No	274 (84.31)	274 (85.62)	548 (84.96)
• Yes	47 (14.46)	37 (11.56)	84 (13.02)
• Missing	4 (1.23)	9 (2.81)	13 (2.02)
Ethnicity: N(%)
• Non-Caucasian	29 (8.92)	29 (9.06)	58 (8.99)
• Caucasian	294 (90.46)	291 (90.94)	585 (90.70)
• Missing	2 (0.62)	0 (0.00)	2 (0.31)
SEIFA IRSD^ Quintile: N(%)
• Q1	107 (32.92)	87 (27.19)	194 (30.08)
• Q2	61 (18.77)	83 (25.94)	144 (22.33)
• Q3	59 (18.15)	52 (16.25)	111 (17.21)
• Q4	46 (14.15)	52 (16.25)	98 (15.19)
• Q5	52 (16.00)	46 (14.37)	98 (15.19)
Infant Sex: N(%)
• Male	164 (50.46)	163 (50.94)	327 (50.70)
• Female	161 (49.54)	157 (49.06)	318 (49.30)
Study Site: N(%)
• WCH	135 (41.54)	136 (42.50)	271 (42.02)
• FMC	98 (30.15)	103 (32.19)	201 (31.16)
• LMH	92 (28.31)	81 (25.31)	173 (26.82)

SD = standard deviation

IQR = interquartile range

IRSD = Socioeconomic index as measured by SEIFA Index of Relative Socio-economic Disadvantage[35]

Epigenome-wide analyses

Results of tests for differential methylation associated with the intervention are shown in Table 2 and Fig 1 (top panels). Even using the less strict Benjamini-Hochberg method for Type I error control, there were no probes which were significantly differentially methylated between the Lifestyle Advice and Standard Care groups, and there was no evidence for effect modification by maternal BMI. The top 10 differentially methylated probes by p value were spread across the genome (with the exception of two probes on chr13 mapped to C13orf34/C13orf37), and effect sizes were small, (absolute log-FC between 0.1 and 0.15, corresponding to approximately 1.01x higher methylation). The top 10 differentially methylated probes by log-Fold Change (i.e. the probes where the magnitude of difference in methylation was greatest) did not overlap with the top 10 by p-value, and these effect sizes were relatively small (absolute log-FC all being between 0.3 and 0.4, corresponding 1.2 to 1.3 times higher methylation).

Table 2

Top 10 differentially methylated probes (lifestyle advice vs standard care).

Rank	Top 10 Probes by p-Value					Top 10 Probes by log-Fold Change
	chr	Name	UCSC RefGene Name	logFC (95% CI)^	adj P.Val*	chr	Name	UCSC RefGene Name	logFC (95% CI)^	adj P.Val*
Intervention at Mean BMIa
1	chr19	cg03057840		-0.10 (-0.14, -0.06)	0.26	chr17	cg08103988		0.50 (0.16, 0.84)	>0.99
2	chr4	cg14712262	ZFYVE28	-0.06 (-0.08, -0.03)	>0.99	chr17	cg24686902		0.44 (0.13, 0.75)	>0.99
3	chr13	cg20260570	C13orf34;C13orf37	-0.08 (-0.12, -0.04)	>0.99	chr17	cg21358336		0.43 (0.14, 0.73)	>0.99
4	chr21	cg01233397		0.11 (0.06, 0.16)	>0.99	chr1	cg04798314	SMYD3	-0.30 (-0.67, 0.07)	>0.99
5	chr12	cg09636302	HAL	-0.11 (-0.16, -0.06)	>0.99	chr17	cg08750459		0.29 (0.09, 0.49)	>0.99
6	chr18	cg17242353		0.14 (0.08, 0.21)	>0.99	chr1	cg06928484	VANGL2	-0.28 (-0.53, -0.04)	>0.99
7	chr11	cg13932624	TBRG1	-0.06 (-0.08, -0.03)	>0.99	chr2	cg04131969	MYADML	-0.28 (-0.70, 0.15)	>0.99
8	chr4	cg16269431	GLRB	-0.06 (-0.09, -0.03)	>0.99	chr1	cg08477332	S100A14	-0.28 (-0.52, -0.03)	>0.99
9	chr12	cg11551902	FOXM1;C12orf32	-0.14 (-0.21, -0.07)	>0.99	chr10	cg02113055		0.27 (-0.11, 0.64)	>0.99
10	chr8	cg24258108	WHSC1L1	0.10 (0.05, 0.15)	>0.99	chr19	cg25755428	MRI1	0.26 (0.00, 052)	>0.99
Intervention at +5 BMIb
1	chr19	cg03057840		-0.09 (-0.13, -0.05)	>0.99	chr17	cg08103988		0.51 (0.17, 0.85)	>0.99
2	chr12	cg09636302	HAL	-0.12 (-0.17, -0.07)	>0.99	chr17	cg24686902		0.45 (0.14, 0.76)	>0.99
3	chr21	cg01233397		0.11 (0.06, 0.16)	>0.99	chr17	cg21358336		0.44 (0.14, 0.74)	>0.99
4	chr4	cg14712262	ZFYVE28	-0.06 (-0.08, -0.03)	>0.99	chr1	cg08477332	S100A14	-0.31 (-0.57, -0.06)	>0.99
5	chr13	cg20260570	C13orf34;C13orf37	-0.08 (-0.11, -0.04)	>0.99	chr1	cg04798314	SMYD3	-0.30 (-0.67, 0.07)	>0.99
6	chr18	cg17242353		0.14 (0.8, 0.21)	>0.99	chr17	cg08750459		0.30 (0.09, 0.50)	>0.99
7	chr16	cg06730286	IFT140	-0.08 (-0.12, -0.04)	> 0.99	chr8	cg24634471	JRK	-0.29 (-0.57, 0.00)	>0.99
8	chr20	cg11336672	RBL1	0.14 (0.07, 0.21)	>0.99	chr1	cg06928484	VANGL2	-0.28 (-0.53, -0.03)	>0.99
9	chr11	cg13932624	TBRG1	-0.06 (-0.08, -0.03)	>0.99	chr10	cg02113055		0.28 (-0.10, 0.66)	>0.99
10	chr17	cg04435975	LOC404266;HOXB6	0.08 (0.04, 0.11)	>0.99	chr2	cg04131969	MYADML	-0.28 (-0.71, 0.16)	>0.99

* Adjusted for multiple comparisons using Benjamini-Hochberg method

^ Model included as covariates parity (0 vs 1+), age (continuous), smoking status, quintile of socioeconomic disadvantage, study centre, infant sex, array batch and estimated cell type proportions (BCell, CD4T, CD8T, Granulocytes, Monocytes, NK, nRBC).

a Effect of intervention estimated at the mean BMI for the cohort (approx. 30 kg/m2)

b Effect of intervention estimated at mean + 5 kg/m2 BMI (approx 35 kg/m2)

Fig 1

Volcano plots (-log10 p value vs log2 fold change) for intervention effects (at mean BMI and at +5 kg/m2 BMI) and BMI effects (in lifestyle advice and standard care groups).

Results of tests for differential methylation associated with maternal BMI are shown in Table 3 and Fig 1 (lower panels). There were no probes which demonstrated significant differential methylation according to maternal BMI. As with the intervention effects, the top 10 probes were spread across the genome, with quite small effect sizes. The 10 probes with greatest estimated log-FC did not overlap with those 10 probes with smallest p values.

Table 3

Top 10 differentially methylated probes (5 kg/m2 increase in BMI).

Rank	Top 10 Probes by p-Value				Top 10 Probes by log-Fold Change
	chr	Name	UCSC RefGene Name	logFC (95% CI)^	adj P.Val	chr	Name	UCSC RefGene Name	logFC (95% CI)^	adj P.Val
BMI in Standard Care Group
1	chr3	cg25821785	CACNA2D2	-0.05 (-0.07, -0.03)	0.22	chr6	cg06864789		0.03 (0.00, 0.06)	0.69
2	chr10	cg21348752	C10orf114;MIR1915	0.04 (0.02, 0.05)	0.22	chr17	cg03226844	RPH3AL	-0.03 (-0.05, -0.01)	0.57
3	chr3	cg01919208	LAMB2	-0.07 (-0.10, -0.04)	0.23	chr6	cg18136963		0.03 (0.01, 0.06)	0.71
4	chr10	cg18646207	VAX1	0.09 (0.05, 0.12)	0.23	chr8	cg03547562		0.03 (0.01, 0.05)	0.76
5	chr9	cg01263574	TMEM8C	0.04 (0.02, 0.05)	0.23	chr21	cg11287055	DSCR3	-0.03 (-0.05, -0.01)	0.66
6	chr10	cg16310045	TCF7L2	-0.05 (-0.06, -0.03)	0.23	chr9	cg13558371	CRB2	-0.03 (-0.05, -0.01)	0.61
7	chr2	cg16639766	HJURP	0.06 (0.04, 0.09)	0.23	chr3	cg03329597	MYH15	-0.03 (-0.05, 0.00)	0.58
8	chr7	cg22005393	DNAJC2	-0.05 (-0.06, -0.03)	0.23	chr1	cg01072550		-0.03 (-0.04, -0.01)	0.72
9	chr2	cg05223061	NGEF	0.08 (0.05, 0.12)	0.23	chr11	cg24851651	CCS	0.03 (0.00, 0.05)	0.69
10	chr6	cg27244242	LY6G5C	0.03 (0.02, 0.05)	0.23	chr13	cg20293942		0.03 (0.00, 0.05)	0.60
BMI in Lifestyle Advice Group
1	chr11	cg07823293	TBRG1	0.09 (0.06, 0.13)	0.27	chr6	cg06864789		0.28 (0.08, 0.47)	>0.99
2	chr4	cg12630714		0.07 (0.04, 0.10)	0.63	chr6	cg18136963		0.26 (0.09, 0.43)	>0.99
3	chr3	cg11118235	GNAI2	0.05 (0.03, 0.08)	0.63	chr8	cg21847720	MYOM2	-0.23 (-0.40, -0.06)	>0.99
4	chr14	cg12154261	TDRD9	0.06 (0.03, 0.09)	0.63	chr8	cg10596483	JRK	-0.22 (-0.38, -0.05)	>0.99
5	chr2	cg06695611	ZNF385B;MIR1258	-0.11 (-0.16, -0.06)	0.63	chr13	cg20293942		0.21 (0.03, 0.38)	>0.99
6	chr9	cg15850063		0.04 (0.02, 0.06)	0.63	chr1	cg08477332	S100A14	-0.19 (-0.33, -0.04)	>0.99
7	chr3	cg17241937	C3orf26;FILIP1L	0.05 (0.03, 0.08)	0.63	chr8	cg24634471	JRK	-0.18 (-0.35, -0.02)	>0.99
8	chr1	cg08867825	OLFM3	0.08 (0.05, 0.12)	0.63	chr21	cg00159953	COL6A2	0.18 (0.03., 0.34)	>0.99
9	chr8	cg16903025	FBXO32	-0.07 (-0.11, -0.04)	0.63	chr6	cg07185983		0.18 (0.07, 0.29)	>0.99
10	chr1	cg16274353	TROVE2	0.06 (0.03, 0.09)	0.63	chr6	cg25399239		0.18 (0.06, 0.29)	>0.99

* Adjusted for multiple comparisons using Benjamini-Hochberg method

^ Model included as covariates maternal BMI (continuous), parity (0 vs 1+), age (continuous), smoking status, quintile of socioeconomic disadvantage, study centre, infant sex, and array batch.

Results of sensitivity analyses generally confirmed the results of the main analyses. No differentially methylated probes corresponding to intervention or BMI effects were detected in any of the alternative models fitted. There was no evidence of any effects in the unadjusted model (Table 1 in S2 Table), or of effect modification between infant sex and either intervention or BMI.

In data normalised using different methods, the overall results were generally similar to the main analysis. In data normalised using the SWAN method. there were no significantly differentially methylated probes corresponding to intervention effects, BMI effects, or their interaction (Table 2 in S2 Table). In BMIQ-normalised data, no differentially methylated probes were found for intervention effects, or for the effect of BMI in the Lifestyle Advice group; 5 probes were significantly differentially methylated for the effect of BMI in the Standard Care group (Table 3 in S2 Table). Where a supervised ComBat algorithm (specifying Intervention, BMI and their interaction as effects of interest) was used (in the SQN normalised data) instead of correction for batch in the analysis model, several probes were found to be differentially methylated for most of the effects (Table 4 in S2 Table). However, none of the probes which were significantly differentially methylated in one analysis were replicated in another; the 5 significant probes in BMIQ-normalised data did not even appear in the top-ranked probes in SQN or SWAN-normalised data.

Candidate gene analysis

The results of candidate gene analyses are presented in Figs 1–4 in S1 File. There was no evidence of differential methylation of probes mapped to PPARGC1A, IGF2, RXRA, or MEST, related to either intervention or maternal BMI. For all genes, the pattern of methylation across all probes was similar between the Standard Care and Lifestyle Advice group, and between different maternal BMI values. Estimated effects did not have a consistent direction for either intervention or BMI, with a combination of positive and negative effect estimates across probes. While there were a few individual probes where effects were statistically significant (p<0.05), these p values were not adjusted for multiple comparisons, and it is doubtful that they are meaningful in the context of a large number of other probes in which no effects were evident.

Discussion

In our investigation of DNA methylation related to an antenatal diet and lifestyle intervention, and overweight and obesity in pregnancy, we have found no evidence of any effect of these factors on DNA methylation in cord blood. In both the main analysis model and a range of sensitivity analyses, we consistently found no differentially methylated probes even with a less strict method of Type I error control. Moreover, observed effects were small in magnitude and not consistent in direction.

While a few statistically significant differentially methylated probes were found with data processed using ComBat, and in data normalised using BMIQ, there are reasons to doubt these results. Firstly, the logFC estimates for these probes were extremely small, and (as noted) the significant probes in the BMIQ-normalised data did not appear in ‘top 10’ probe lists in data normalised using SQN or SWAN. Secondly, implementation of the ComBat procedure allows the user to specify the factors of interest (which in this case were given as intervention group, BMI category and their interaction). Nygaard et al. [29] caution that this may produce spurious effects in situations where the groups are not evenly spread across batches, as was the case in these data. Nevertheless, the discrepancies resulting from different data-processing choices are concerning, and are discussed further in a companion paper (submitted for publication) in which they are investigated more systematically.

Strengths and limitations

This study has a number of strengths, including its moderately large sample size (645 samples) giving substantial statistical power to detect meaningful differences. Further, these data are from a randomised study with BMI category (25.0–29.9 vs ≥30.0 kg/m2) as a stratification variable which was reliably measured in early pregnancy by research staff (rather than self-reported), and are therefore less subject to measurement error or reporting bias. Additionally, participants all had early pregnancy BMI ≥25.0 kg/m2, providing greater power to investigate effects of higher BMI, which is often underrepresented in random samples of the population.

The limitations of the study include the study population, the use of cord blood to assess DNA methylation, and the limited coverage of the Illumina 450K array. As the LIMIT study recruited only women with early pregnancy BMI ≥25.0 kg/m2, we did not capture the entire BMI range and in particular do not have DNA methylation levels for women of ‘normal’ BMI. It is possible that there is a nonlinear effect of BMI on DNA methylation, such that the main differences are between women with ‘normal’ BMI and those with higher-than-‘normal’ BMI. However, it seems unlikely that there would be substantial differences between women with BMI <25.0 and women with BMI ≥25.0, but none between women with BMI 25.0–29.9 and women with BMI ≥30.0. We are currently investigating DNA methylation in infants born to participants in the OPTIMISE study (a randomised controlled trial of an antenatal diet and lifestyle intervention for women with early pregnancy BMI 18.5–24.9 kg/m2) to further evaluate the effect of maternal BMI.

Secondly, DNA methylation in cord blood may not be a reliable proxy for the DNA methlation status of infant tissues. Cord blood is commonly used for DNA methylation studies, as it can be obtained non-invasively and in larger quantities [13]. Further, DNA methylation in cord blood is considered to be a good indicator of DNA methylation in infant blood and other tissues [2, 6, 16]. Additionally, cord blood contains different cell types, which may be present in differing proportions in different samples, potentially confounding the effects of interest [36]. While all analyses were adjusted for estimated cell type proportions, the true cell type proportions in the samples are unknown.

Thirdly, the Illumina 450k array analyses around 485,577 sites in the human genome, with a focus on areas of epigenetic interest, i.e., genes and CpG islands [37]. However, this array covers only approximately 2% of CpG sites in the human genome [38]. It is therefore possible that diet and lifestyle in pregnancy, or early pregnancy BMI, have effects on DNA methylation in areas of the genome not covered by the 450K array. Additionally, it is possible that other epigenetic effects may exist, and may interact with DNA methylation. For example, it has been noted that histone modifications may play a part in adipogenesis and hence in susceptibility to obesity [14].

Finally, a larger sample size may be required to reliably detect differences in DNAm due to antenatal interventions, or maternal early pregnancy BMI; the lack of statistically significant findings in the present study may reflect insufficient sample size. However, while a larger sample size would allow detection of smaller differences in DNAm, it is not clear that very small differences would be clinically meaningful.

Consistency with the existing literature

Our findings may seem at odds with the existing literature, in which numerous studies have found associations between DNA methylation in cord blood, and maternal early pregnancy BMI / obesity. A range of genes and/or loci found to be differentially methylated in relation to maternal obesity and/or BMI have been summarised in S1 Table. These loci include the promoter region of PPARGC1A [19]; sites on ESM1 and MS4A3 [16]; 86 CpG sites found by the PACE consortium [39]; 28 CpG sites found in the ALSPAC cohort [38]; multiple CpGs mapped to TAPBP [39]; a single CpG site mapped to ZCCHC10 [40]; sites mapped to FLJ41941 and an unnamed gene [41]; DMRs related to imprinted genes PLAGL1 and MEG3 [42]; sites on MEST [20]; and 2 CpGs mapped to RXRA [10]. Related findings include differential methylation in cord blood in genes ATP5A1, MFAP4, PRKCH, SLC17A4 related to Gestational Diabetes (GDM) [43]; and hypermethylation of the LEP gene promoter associated with maternal obesity on the fetal side of the placenta [44].

However, as indicated by the diversity of this list, the findings from different studies are not consistent, with each study discovering a different set of differentially methylated sites. Where studies have found a range of differentially methylated loci, these are often single CpG sites located on diverse regions of the genome with no known connection to adiposity, obesity, or growth [38, 39]. Moreover, explicit attempts to replicate the findings of other studies have not thus far succeeded [39, 42, 45], and where evidence of differential methylation is found, it is often reported that the actual effect sizes are both of small magnitude, and uncertain clinical significance [18, 39, 41, 43]. Differential methylation may also be found for one analysis approach but not another, e.g. significant findings may become non-significant when analysing BMI as a continuous variable rather than as categories [38]; when adjusting for multiple comparisons [18]; or when adjusting for potential confounders [43]. This lack of consistent, robust evidence has already led others to conclude that DNA methylation is likely not a major causal pathway linking maternal and child obesity [39, 41], with which our findings are in agreement.

Even if reliable evidence of differential DNA methylation in neonates related to maternal obesity / BMI were discovered, it would still remain to be shown that cord blood DNA methylation is causally linked to childhood adiposity, obesity or cardiometabolic health. Some evidence exists that cord blood DNA methylation is associated with child or adult BMI [6, 10, 17, 46]. However, others have found at best weak associations and remain skeptical [14, 38, 47].

Conclusions

Our study found no evidence of any differentially methylated sites associated with an antenatal lifestyle intervention, or maternal early pregnancy BMI, in cord blood. Moreover, we were unable to find evidence of differential methylation associated with the intervention, or with BMI, for selected candidate genes. The lack of association persisted for different analysis approaches (adjusting for confounders vs not adjusting; using categorical vs continuous BMI; including interaction terms) and for data processed using different methods. Together with the lack of consistent findings from other studies, our results suggest that other causal pathways are primarily responsible for the link between maternal and child obesity.

Genes/loci reported as differentially methylated in cord blood in previous studies.

(DOCX)

Click here for additional data file.

Results of sensitivity analyses.

(DOCX)

Click here for additional data file.

Results of candidate gene analyses.

(DOCX)

Click here for additional data file.

Transfer Alert

This paper was transferred from another journal. As a result, its full editorial history (including decision letters, peer reviews and author responses) may not be present.

14 Feb 2022

21 Apr 2022

Response to Reviewers

1. Ensure manuscript meets PLOS ONE’s style requirements.

We have updated the manuscript to conform to PLOS ONE style requirements.

2. Please provide additional details regarding participant consent.

Written informed consent to participate in the LIMIT study was obtained from all participants (who were all 18 years of age). Additional written consent was obtained to collect samples of umbilical cord blood at delivery, and participants were informed that this would be used for gene expression research related to the diet and lifestyle intervention, and to weight.

This information has been added to the ethics statement, which (per (4) below) has been moved to the Methods section, lines 135-145).

3. Information in ‘Funding Information’ and ‘Financial Disclosure’ sections do not match.

We have expanded the ‘Financial Disclosure’ statement to match the information in the original manuscript. To conform with PLOS ONE style requirements, we have removed this information from the manuscript itself.

4. Ethics statement should only appear in Methods section

The ethics statement has been moved to the methods section and deleted from ‘Declarations’ (lines 135-145).

5. Upload a copy of Figure 2 or remove reference to it within the text

Apologies for the error; this should refer to the bottom panels of figure 1. We have amended the manuscript accordingly.

6. Include captions for Supporting Information files at the end of the manuscript and update in-text citations to match accordingly

Supporting Information is now divided into three files (S1 Table, S2 Tables and S3 Figures), described in lines 581-587 after the references, and the references to this information in the main text have been updated throughout.

7. Review reference list to ensure that it is complete and correct and does not include retracted papers

We have checked the list of references to ensure correctness and completeness. No references have been retracted.

Reviewer 1

1. Error on Table 1 (baseline characteristics) reports 85% of participants were smokers.

Thank you for pointing out this error; it has been corrected. (Line 212 Table 1)

Reviewer 2

2. Study may not have sufficient statistical power to detect the small effect size. The power and sample size were not calculated for this secondary analysis. This raises the question whether the results are really negative or the sample size is not enough to detect it.

Insufficient statistical power is always a potential explanation where differences are not statistically significant; since statistical significance is a function of sample size, a large enough sample would allow detection of statistically significant differences between two groups even though these may not be clinically meaningful.

Our sample size is larger than many other studies of cord blood DNAm relating to antenatal interventions or maternal overweight/obesity, including those which have found statistically significant differences. It provides 80% power to detect differences as small as 0.2 standard deviations between groups; while in the context of high-dimensional data (and resulting multiple-comparisons issues) the question of statistical power is more complicated, we believe that we were adequately powered to detect robust and clinicdally meaningful differences in DNAm.

We have added a sentence to the discussion of limitations (lines 334-337) to note that statistically significant differences may have been found with a larger sample size, but that these differences would have been of uncertain clinical significance.

3. Table 1 did you compare the characteristics between the two arms What tests were used and what are the results? All non-significant?

We did not perform statistical tests to compare baseline characteristics at baseline; as noted in the CONSORT (Consolidated Standards of Reporting Trials) Statement Explanation and Elaboration) standards, such tests are not recommended. As the groups were randomised, it is already known that any differences between the groups are due to chance. A statistical test, which estimates the probability that differences as large as those observed would arise by chance, is therefore inappropriate.

Additionally, even outside of a randomised setting, statistical significance is not a valid indicator of the presence of confounding (nor is non-significance a valid indicator of absence of counfounding). A statistically significant difference between groups is not necessarily a confounder of the effect of interest, and a non-statistically-significant difference can still confound the effect of interest.

4. Tables 2 and 3 are not clearly presented. Any significant interaction? There is no mention of interaction results. If there is no significant interaction, you can show main effects rather than simple effects shown in Table 2 and 3.

What does it mean ‘intervention at Mean BMI’? Mean BMI of overall?

‘Intervention at +5 BMI’ is also confusing. Mean BMI + 5 units?

The interaction between intervention and maternal BMI was not statistically significant – we have clarified this in the manuscript (lines 222-223).

However, we disagree that a non-statistically-significant interaction implies that ‘main effects’ can or should be estimated instead. Firstly, the analysis model including the interaction term was the prespecified analysis for this study, and should therefore be reported rather than the results of an analysis undertaken after viewing the results of the main analysis. Secondly, the sample size required to detect interaction effects is usually many times larger than that required to detect the individual effects, and tests of interactions are therefore usually underpowered. If the interaction term is dropped (due to non-significance in an underpowered test) and ‘main effects’ are estimated instead, there is potential for the estimates of effects of interest to be biased. As can be seen from Tables 2 and 3, the ‘top 10’ probes differ substantially for effect of intervention at different values of maternal BMI (and for effect of maternal BMI in the different intervention groups).

The interaction term was between intervention (2 groups) and maternal BMI (as a continuous variable). It was therefore necessary, when presenting results, to state the value of maternal BMI at which intervention effects were estimated. ‘Intervention at Mean BMI’ is the effect of intervention estimated at the mean BMI of the cohort; ‘Intervention at +5 BMI’ is the effect of intervention estimated at mean + 5 kg/m2 BMI. We have added some explanation to the Table 2 notes to clarify this.

Submitted filename: Response to Reviewers.docx

Click here for additional data file.

27 May 2022

Effect of an antenatal diet and lifestyle intervention and maternal BMI on cord blood DNA methylation in infants of overweight and obese women: the LIMIT Randomised Controlled Trial

PONE-D-21-39849R1

Dear Dr. Louise,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Diane Farrar

Academic Editor

PLOS ONE

14 Jun 2022

PONE-D-21-39849R1

Effect of an antenatal diet and lifestyle intervention and maternal BMI on cord blood DNA methylation in infants of overweight and obese women: the LIMIT Randomised Controlled Trial

Dear Dr. Louise:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Diane Farrar

Academic Editor

PLOS ONE