Journal of Child Psychology and Psychiatry 58:1 (2017), pp 19–27 doi:10.1111/jcpp.

12589

Prenatal unhealthy diet, insulin-like growth factor 2

gene (IGF2) methylation, and attention deficit
hyperactivity disorder symptoms in youth with early-
onset conduct problems
Jolien Rijlaarsdam,1,2 Charlotte A.M. Cecil,3 Esther Walton,3 Maurissa S.C. Mesirow,3
Caroline L. Relton,4 Tom R. Gaunt,4 Wendy McArdle,5 and Edward D. Barker3
1
Centre for Child and Family Studies, Leiden University, Leiden; 2Department of Child and Adolescent Psychiatry/
Psychology, Erasmus MC-University Medical Center Rotterdam, Rotterdam, The Netherlands; 3Department of
Psychology, Institute of Psychiatry, Psychology and Neuroscience, King’s College London, London; 4Medical Research
Council Integrative Epidemiology Unit, University of Bristol, Bristol; 5School of Social and Community Medicine,
University of Bristol, Bristol, UK

Background: Conduct problems (CP) and attention deficit hyperactivity disorder (ADHD) are often comorbid and
have each been linked to ‘unhealthy diet’. Early-life diet also associates with DNA methylation of the insulin-like
growth factor 2 gene (IGF2), involved in fetal and neural development. We investigated the degree to which prenatal
high-fat and -sugar diet might relate to ADHD symptoms via IGF2 DNA methylation for early-onset persistent (EOP)
versus low CP youth. Methods: Participants were 164 youth with EOP (n = 83) versus low (n = 81) CP drawn from the
Avon Longitudinal Study of Parents and Children. We assessed if the interrelationships between high-fat and -sugar
diet (prenatal, postnatal), IGF2 methylation (birth and age 7, collected from blood), and ADHD symptoms (age 7–13)
differed for EOP versus low CP youth. Results: Prenatal ‘unhealthy diet’ was positively associated with IGF2
methylation at birth for both the EOP and low CP youth. For EOP only: (a) higher IGF2 methylation predicted ADHD
symptoms; and (b) prenatal ‘unhealthy diet’ was associated with higher ADHD symptoms indirectly via higher IGF2
methylation. Conclusions: Preventing ‘unhealthy diet’ in pregnancy might reduce the risk of ADHD symptoms in
EOP youth via lower offspring IGF2 methylation. Keywords: DNA methylation; Avon Longitudinal Study of Parents
and Children; diet; conduct problems; attention deficit hyperactivity disorder; IGF2.

2004; Sonuga-Barke et al., 2013). A potential mech-

Introduction
anism that might help explain the link between
Conduct problems (CP) and attention deficit hyper-
‘unhealthy diet’ and CP and ADHD is the epigenetic
activity disorder (ADHD) commonly co-occur.
process of DNA methylation, which is highly respon-
Importantly, evidence from family (Faraone, 2000),
sive to the nutritional environment (Drake et al.,
twin (Thapar, Harrington, & McGuffin, 2001), and
2012), and also associates with CP-related pheno-
molecular genetic (Holmes et al., 2002) studies
types (Cecil et al., 2014) and ADHD (Schuch,
suggest that that this co-occurrence denotes a more
Utsumi, Costa, Kulikowski, & Muszkat, 2015; van
severe, familial, and heritable entity, compared to
Mil et al., 2014; Walton et al., 2016).
either CP or ADHD alone. Children with an early-
Diet has also been shown to influence methylation
onset and persistent pattern of CP represent a
of the insulin-like growth factor 2 gene (IGF2) (Hei-
particular at-risk group, as they often show the
jmans et al., 2008), an imprinted gene that lies close
highest rates of ADHD (Barker, Oliver, & Maughan,
to the insulin and tyrosine hydroxylase genes in a
2010), as well as the greatest levels of psychosocial
genomic region related to the metabolic regulation of
risk exposures in pregnancy (e.g. poverty, maternal
glucose homeostasis, cardiovascular functions, and
anxiety) and the early postnatal years (e.g. harsh
lipid metabolism (Faienza et al., 2010; Ukkola, Sun,
parenting, family discordance) (Barker & Maughan,
& Bouchard, 2001). IGF2 may be of interest to ADHD
2009).
as it is a major modulator of placental and fetal
One prenatal risk that is a correlate of these
growth (Constancia et al., 2002) and also plays an
psychosocial risks, yet has received far less atten-
integral role in brain development after birth (Pids-
tion, is diet. ‘Unhealthy diet’ (e.g. high fat/sugar) is of
ley, Dempster, Troakes, Al-Sarraj, & Mill, 2012).
particular interest as it has been reported to asso-
Animal and human studies report that IGF2 is
ciate with both CP and ADHD (Howard et al., 2011;
associated with developmental abnormalities in the
Jacka et al., 2013; Liu, Raine, Venables, & Mednick,
structure and/or function of the cerebellum (Pidsley
et al., 2012) and the hippocampus (Chen et al.,
Conflict of interest statement: No conflicts declared. 2011; Ouchi et al., 2013), both of which are
© 2016 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for Child and
Adolescent Mental Health.
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any
medium, provided the original work is properly cited.
20 Jolien Rijlaarsdam et al. J Child Psychol Psychiatr 2017; 58(1): 19–27

associated with ADHD (Castellanos et al., 2002;

Measures
Plessen et al., 2006), as well as other psychiatric
disorders such as depression and schizophrenia (Yu, DNA methylation data. Five hundred nanograms of
Shen, Zeng, Ma, & Hu, 2013). genomic DNA from cord blood (birth) or peripheral blood (age
7) was bisulfite-converted using the EZ-DNA methylation kit
Periconceptional risk exposure is associated with
(Zymo Research, Orange, CA). The protocol followed the
abnormal brain development (Jensen et al., 2015), manufacturer’s instructions using the recommended alterna-
with relevance to CP and ADHD (Fairchild et al., tive incubation conditions for use with Illumina Infinium
2011; Rubia, Smith, Brammer, Toone, & Taylor, arrays. Illumina HumanMethylation450 BeadChips (Illumina,
2005). Moreover, diet-induced IGF2 DNA methyla- San Diego, CA) were run following the manufacturer’s protocol
with no modifications, and arrays were scanned using an
tion modifications occur specifically during the peri-
Illumina iScan (software version 3.3.28). Initial quality control
conceptional period and may persist well in of data generated was conducted using GenomeStudio (Illu-
adulthood (Heijmans et al., 2008). Together, these mina; version 2011.1) to determine the status of staining,
findings suggest that the long-term impact of early- extension, hybridization, target removal, bisulfite conversion,
life dietary factors on CP and ADHD may, at least in specificity, nonpolymorphic, and negative controls. DNA
methylation data were only available for samples that passed
part, be explained by IGF2 DNA methylation. The
this stage. Samples were quantile normalized using the dasen
current study simultaneously examined, for early- function within the wateRmelon package (wateRmelon_1.0.3)
onset persistent (EOP) versus low CP youth, the (Pidsley et al., 2013) in R and batch-corrected using the
extent to which unhealthy prenatal and postnatal ComBat package (Johnson, Li, & Rabinovic, 2007).
diet (high fat, high sugar) is associated with ADHD We extracted 139 probes that are mapped to IGF2 or
overlapping regions adjacent to IGF2, including INS-IGF2 (i.e.
symptoms via DNA methylation of IGF2 (birth and
position 2150687 to 2183864). For each probe, methylation
age 7, collected from blood). levels were indexed by beta values (i.e. the ratio of methylated
signal relative to the sum of the methylated and unmethylated
signals). Factor analysis was used in the total sample to
establish the covariance structure among the 139 IGF2 probes
Methods in order to extract a smaller set of underlying factors, removing
Participants CpGs with low correlations as needed. A three-factor solution
showed the best fit to the data. Full details of the factor
The Avon Longitudinal Study of Parents and Children analysis procedure and results are provided as online sup-
(ALSPAC) is a prospective study of children born to 14,541 porting information (Appendix S1 and Table S1). We present
pregnant women residing in Avon, United Kingdom, with an findings relating specifically to factor 1 (37 probes) because
expected delivery date between April 1, 1991 and December factor 2 (11 probes) and 3 (5 probes) did not correlate with
31, 1992 (85% of eligible population (Fraser et al., 2013)). ‘unhealthy diet’ and ADHD symptoms for the EOP and low CP
When compared with 1991 national census data, the ALSPAC trajectory. See Appendix S2 for the location of the IGF2
sample was found to be similar to the UK population as a whole methylation probes included in this study, and how these are
(Boyd et al., 2013). Ethics approval for the study was obtained grouped into factors. See Appendix S3 for the correlations
from the ALSPAC Law and Ethics Committee as well as Local between the IGF2 methylation probes.
Research Committees. All participants provided informed
consent. The study website contains details of all the data High-fat and -sugar diet. The Food Frequency Ques-
that are available through a fully searchable data dictionary: tionnaire (FFQ; (Micali, Northstone, Emmett, Naumann, &
http://www.bris.ac.uk/alspac/researchers/data-access/data- Treasure, 2012) was used to assess (a) maternal dietary
dictionary/. patterns at 32 weeks of gestation, and (b) what the mother
This study uses a subsample (n = 346, 50% male) from a reported feeding to the child at 3, 4.5, and 7 years of age. The
larger study of DNA methylation in ALSPAC, the Accessible FFQ contains a set of questions about the frequency of
Resource for Integrated Epigenomics Studies (ARIES (Relton consumption of a wide variety of food and drink, with higher
et al., 2015), www.ariesepigenomics.org.uk), which follows scores indicating higher frequency of intake. Possible
previously established CP trajectories and has epigenetic data responses were: never or rarely; once in 2 weeks, 1–3 times
at birth and/or age 7. The CP trajectories, including (a) low per week; 4–7 times per week; and more than once daily.
(26.9%), (b) childhood-limited (25.4%), (c) adolescent-onset Prenatal and postnatal high-fat and -sugar diet scores, indi-
(19.7%), and (d) early-onset persistent (28.0%), have been cated by processed food (i.e. fried food, meat pies or pasties,
previously identified and validated (Barker & Maughan, 2009). chips) and confectionery (i.e. crisps, chocolate bars, cakes or
Specifically, general mixture modeling was used based on the buns, biscuits) had been previously created using latent
‘Conduct Problem’ subscale (4–13 years) of the Strengths and factors (Barker, Kirkham, Ng, & Jensen, 2013). For the current
Difficulties Questionnaire (SDQ; (Goodman, 2001). This ‘Epi- analyses, we combined the postnatal diet scores across time
genetic Pathways to Conduct Problems Study’ subsample is (age 3–7) by the use of a latent factor (range factor load-
comparable to the full trajectory sample (n = 7,218) in terms of ings = .74–.84).
environmental risk and psychiatric comorbidity (Barker et al.,
2010). DNA methylation data were available for 321 youth at
birth and 326 at age 7. Attention deficit hyperactivity disorder symp-
In light of the objective of this study, and to ensure a feasible toms. Attention deficit hyperactivity disorder (ADHD) symp-
set of statistical analyses, we included youth in the low and toms were repeatedly assessed (at age 7, 10, and 13) with the
early-onset CP trajectories, who had complete data for prenatal Development and Well-being Assessment (DAWBA; Goodman,
‘unhealthy diet’, IGF2 DNA methylation at birth and ADHD Ford, Richards, Gatward, & Meltzer, 2000), a validated
symptoms (ntotal = 164; nlow = 81; nEOP = 83). We did not semistructured interview. Parents completed open and closed
assess the childhood-limited or adolescent-onset as these questions about a range of symptoms relevant to youth
youth have different developmental risk pathways than the psychiatric disorders, including ADHD, oppositional defiant
early-onset and low CP youth (see Barker & Maughan, 2009; disorder (ODD), conduct disorder (CD), generalized anxiety
Moffitt et al., 2008). disorder (GAD), and major depressive disorder (MDD). For each

© 2016 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
doi:10.1111/jcpp.12589 Prenatal unhealthy diet, IGF2 methylation, and ADHD 21

disorder, an ordered categorical measure was generated using Model fit was first established using the chi-square statistic.
computer algorithms (Goodman, Heiervang, Collishaw, & Missing data were handled through full information maximum
Goodman, 2011), comprising six categories indicating the likelihood. Youth with scores > 3.29 standard deviation from
likelihood of each youth having the disorder from level 0 up the mean on any study variable were treated as outliers (n = 3)
to level 5. For the current analyses, we created factor scores and their scores winsorized (i.e. transformed to match next
across time (age 7–13) for ADHD (range factor load- highest or lowest value).
ings = .73–.81), ODD (range factor loadings = .58–.82), GAD
(range factor loadings = .58–.59), and MDD (range factor
loadings = .37–.72).
Results
Control variables. We included two types of control Descriptive statistics
variables; repeated measures of cumulative risk and cell type
distribution. First, cumulative risk variables were summated Table 1 contains the correlations and descriptive
into indices spanning two developmental periods (pregnancy statistics of the study variables. These statistics are
and early-childhood [birth–age 7]) and regressed on all endoge-
presented separately for the two CP trajectories. Five
nous study variables. For each developmental period, a
cumulative risk index had been previously created using latent results are highlighted. First, in line with previous
factor analyses (Cecil et al., 2014), based on maternal reports, research, EOP children showed higher levels of
covering five risk domains: (a) life events (e.g. death in family, ADHD symptoms compared to low CP children.
accident, illness), (b) contextual risks (e.g. poor housing However, means and variances for ADHD differed
conditions, financial problems), (c) parental risks (e.g. parental
from zero for the two groups (EOP and low CP youth;
psychopathology, criminal involvement and substance use), (d)
interpersonal risks (e.g. intimate partner violence, family all p-values < .001). Second, we found that in EOP
conflict), and (e) direct victimization (e.g. child bullied by peers but not in low CP youth (a) factor 1 IGF2 mean DNA
or physically hurt; available postnatally). We also assessed methylation at birth was positively correlated with
maternal smoking during pregnancy, which was measured ADHD symptoms and (b) factor 1 IGF2 DNA methy-
during the first trimester of pregnancy via maternal ratings,
lation at age 7 was negatively correlated with post-
using a yes (n = 29)/no (n = 135) binary variable. However,
this variable did not correlate with IGF2 DNA methylation (see natal cumulative risk. Third, for the EOP youth,
Table 1) and, hence, was not added as a covariate. ‘unhealthy diet’ correlated at a trend level with IGF2
Second, we controlled for cell type heterogeneity to estimate DNA methylation at birth (r[83] = .20, p = .06) and
cell proportions using DNA methylation data (Houseman et al., ADHD (r[83] = .18, p = .10). Fourth, for EOP and low
2012). Specifically, IGF2 DNA methylation scores were residu-
CP youth alike, prenatal cumulative risk was highly
alized for estimated proportions of cells in whole blood (propor-
tion of CD8+ T cells, CD4+ T cells, natural killer [NK] cells, B cells, correlated with postnatal cumulative risk, but for
and monocytes). Granulocytes were removed because the cell EOP only, higher postnatal cumulative risk was
type proportions add up to approximately 100%. significantly associated with lower IGF2 DNA methy-
lation at age 7. Fifth, early-onset CP youth signifi-
cantly differed from the low CP in ODD, GAD, and
Data analysis MDD (p < .05); for EOP and low CP youth alike,
The analysis proceeded in three main steps. In the first step, we prenatal and postnatal IFG2 DNA methylation was
tested for developmental interrelationships between ‘unhealthy not correlated with ODD, GAD, or MDD.
nutrition’ and ADHD using a multiple group autoregressive
cross-lagged (ARCL) model. We did so by testing the degree to
which, for EOP versus low CP, the relationship between high-fat
Step 1: Autoregressive cross-lagged (ARCL) model
and -sugar diet and IGF2 DNA methylation differed. CP trajec-
tory differences and sex differences were tested in nested model predicting ADHD symptoms
comparisons using chi-square difference tests. Two models were
The unadjusted and adjusted models did not differ in
estimated in step 1. The first was an unadjusted model, where we
did not control for cumulative risks, and the second was an terms of significant path coefficients or model com-
adjusted model, where cumulative risks were regressed on parisons. Therefore, we present the adjusted model
endogenous study variables. In the second step, we tested, for only (see Figure S1 for the unadjusted model). Figure 1
EOP versus low CP youth, the degree to which prenatal high-fat depicts the adjusted ARCL model for ‘unhealthy diet’,
and -sugar diet might indirectly relate to higher levels of ADHD
IGF2 DNA methylation, and youth ADHD symptoms.
symptoms via IGF2 DNA methylation at birth. This indirect
pathway was programmed in a model constraint statement. We tested a series of nested model comparisons, where
Difference between the EOP and low CP were tested by a we assessed differences between EOP and low CP
bootstrapped (see below) difference between the respective youth in: (a) the auto-regressions; (b) the cross-lagged
indirect pathways (i.e. EOP – Low CP). In the third step, we associations; (c) the prenatal predictions to ADHD
examined the extent to which high-fat and -sugar diet and IGF2
symptoms; and (d) the postnatal predictions to ADHD
DNA methylation are specific risk factors to the development of
ADHD symptoms as opposed to other externalizing (i.e. ODD) or symptoms of ‘unhealthy diet’ and IGF2 DNA methyla-
internalizing disorders (i.e. GAD, MDD). Cell type was controlled tion. The freely estimated model, which served as the
in all models across steps 1, 2, and 3. comparison model for all nested tests presented below,
Analyses were performed in Mplus version 7.11 (Muth
en & showed acceptable fit to the data (v2[4] = 6.24,
Muth
en, 1998–2013) using maximum likelihood estimation.
p = .18). Because sex differences were not identified
Given the small sample size, we used bootstrapped with bias-
corrected 95% confidence intervals (10,000 bootstraps) to across the parameters in EOP versus low CP youth
derive variance from the empirical distribution of the observed (Dv2EOP [9] = 4.38, p = .88; Dv2low [9] = 9.04, p = .43), we
data. report the results for males and females together.

© 2016 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
 22

Table 1 Correlations and descriptive statistics of the variables by low conduct problem youth (above the diagonal, n = 81) and early-onset persistent conduct problem youth (below the
diagonal, n = 83)

1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11.
Jolien Rijlaarsdam et al.

1
1. Unhealthy diet – .55* .16 .02 .03 .08 .01 .01 .20 .12 .17
prenatal
2. Unhealthy diet age .54* – .20 .02 .01 .08 .07 .02 .13 .15 .15
3–7 years
3. Factor 1 IGF2 .201 .04 – .19 .09 .05 .13 .01 .04 .03 .11
methylation
at birth (mean)
4. Factor 1 IGF2 .03 .06 .14 – .13 .13 .09 .07 .06 .11 .03
methylation
age 7 (mean)
5. ADHD age 7–13 years .181 .07 .27* .14 – .35* .211 .37* .06 .10 .18
6. ODD age 7–13 years .13 .13 .211 .15 .60* – .16 .31* .12 .05 .12
7. MDD age 7–13 years .07 .08 .09 .16 .30* .40* – .40* .004 .14 .22*
8. GAD age 7–13 years .07 .15 .04 .07 .221 .23* .45* – .26* .14 .10
9. Cumulative risk .11 .08 .01 .21 .03 .04 .10 .07 – .55* .12
prenatal
10. Cumulative risk .11 .10 .03 .30* .05 .02 .28* .15 .61* – .28*
birth age 7
11. Prenatal smoking .16 .201 .16 .06 .14 .01 .04 .02 .17 .22* –
(yes = 1, no = 0)
Low CP youth
a a a
Median (interquartile 0.04 (0.91) 0.11 (0.85) 0.16 (0.02) 0.17 (0.02) 0.75a (0.56) .63 (1.03) .39 (0.95) .22 (1.16) 0.17 (0.50) 2.34a (5.86) 11.1%b
range)
EOP CP youth
Median (interquartile 0.09 (0.99) 0.06 (1.45) 0.16 (0.02) 0.17 (0.02) 0.32 (1.20) .44 (1.23) .09 (0.95) .31 (0.73) 0.11 (0.57) 0.32 (7.66) 24.1%b
range)

EOP, early-onset persistent; CP, conduct problems.

1
p-value = <.10; *p-value < .05.
a
Scores were significantly higher for EOP versus low CP youth (p < .05). p-values are derived from Mann–Whitney–Wilcoxon tests.
b
Measured as frequency (%) of prenatal smoking for EOP (n = 20) versus low CP (n = 9) youth (v2 [1] = 3.37, p = .067).

© 2016 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
J Child Psychol Psychiatr 2017; 58(1): 19–27

Child and Adolescent Mental Health.

doi:10.1111/jcpp.12589 Prenatal unhealthy diet, IGF2 methylation, and ADHD 23

Figure 1 Prospective interrelationships between unhealthy diet, IGF2 methylation and ADHD for youth with early-onset persistent
(n = 83) versus low (n = 81) conduct problems. Multiple group path analysis. Solid arrowed lines indicate standardized path coefficients
that survived bootstrap-corrected confidence intervals (i.e. significant paths) for EOP versus low CP youth or averaged across all youth.
ADHD, attention deficit hyperactivity disorder; EOP, early-onset persistent; CP, conduct problems.

symptoms did not significantly differ between EOP

Auto-regressions. The omnibus auto-regressions
and low CP youth (Δv2[1] = 0.40, p = .53) and was
nested chi-square test constrained two parameters
not significant when averaged across all youth (see
(i.e. stability in diet and IGF2 DNA methylation,
Figure 1).
respectively). We found temporal stability in diet but
not in IGF2 DNA methylation, and these estimates
Postnatal predictions to ADHD symptoms. The
did not significantly vary between EOP and low CP
omnibus nested model chi-square difference test
youth (Δv2[2] = 5.96, p = .05), indicating a similar
constrained the parameters of postnatal diet and
pattern of auto-regressions across trajectories. This
IGF2 DNA methylation at age 7 predicting ADHD
result is statistically significant.
symptoms. These associations did not significantly
vary between EOP and low CP youth (Δv2[2] = 1.70,
Cross-lagged associations. The omnibus nested p = .43) and were not significant when averaged
chi-square test for the cross-lagged associations across all youth.
constrained the parameters of diet influencing IGF2
DNA methylation and IGF2 DNA methylation influ-
encing diet. Prenatal diet was associated with IGF2 Step 2: Indirect pathway
DNA methylation at birth for both EOP and low For EOP youth, prenatal ‘unhealthy diet’ was indi-
CP youth, and the strength of this association did rectly associated with ADHD symptoms via higher
not significantly differ between the trajectories IGF2 DNA methylation at birth. The bias-corrected
(Δv2[3] = 1.98, p = .58). confidence interval (via 10,000 bootstraps) for the
indirect pathway of prenatal ‘unhealthy diet’ to
Prenatal predictions to ADHD symptoms. The ADHD symptoms via IGF2 DNA methylation at birth
omnibus nested model chi-square difference test did not cross zero (b = .069; 95% CI .003, .206). For
for the prenatal predictions to ADHD symptoms low CP youth, the 95% CI of this indirect pathway via
constrained two parameters (i.e. predictions from IGF2 DNA methylation did cross zero (b = .015;
prenatal diet and IGF2 DNA methylation at birth to 95% CI .086, .019). The indirect pathway was
ADHD symptoms) and these varied significantly different between the EOP and low CP youth: the
between EOP and low CP youth (Δv2[2] = 6.49, 95% CI of the difference of the indirect pathways did
p = .04). Follow-up difference tests showed an inter- not cross zero (b = .084; 95% CI .224, .005).
action, where the association between IGF2 DNA
methylation at birth and ADHD symptoms at age 7–
Step 3: Other disorders
13 years was significantly higher for EOP versus low
CP youth (Δv2[1] = 5.58, p = .02) (see Figure 1). This In light of the findings above, we repeated the ARCL
is noteworthy given that (a) we had previously shown model to examine the extent to which the association
no DNA methylation difference between the EOP and of IGF2 DNA methylation was specific to ADHD
low CP youth (see Table 1) and that (b) the associ- symptoms as opposed to other psychiatric disorders
ation between prenatal ‘unhealthy diet’ and ADHD (ODD, GAD, and MDD) for EOP versus low CP youth.

© 2016 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
24 Jolien Rijlaarsdam et al. J Child Psychol Psychiatr 2017; 58(1): 19–27

For EOP and low CP youth, IGF2 DNA methylation through stress response pathways, such as
was unrelated to all other disorders (i.e. associations increased cortisol activity, which has been found to
did not survive bootstrapped confidence intervals). be associated with lower IGF2 methylation (Vangeel
et al., 2015). Given that maternal (or child) stress
can co-occur with different dietary patterns (Heij-
Discussion mans et al., 2008), the effects of cortisol and inflam-
In the present study, we used a longitudinal design mation (Thorburn, Macia, & Mackay, 2014) on IGF2
to investigate prospective associations between ‘un- methylation may be promising avenues for future
healthy diet,’ IGF2 DNA methylation, and ADHD research.
symptoms in EOP versus low CP youth. Our results Several limitations should be considered when
showed that prenatal ‘unhealthy diet’ was positively interpreting the present results. First, this research
associated with IGF2 DNA methylation at birth is correlational in nature; hence, causality cannot be
across both EOP and low CP youth. However, only inferred. However, the present research is based on a
for EOP youth, (a) higher IGF2 DNA methylation at longitudinal design, which does allow prospective
birth predicted ADHD symptoms; and (b) prenatal assessment of prenatal and postnatal effects on DNA
‘unhealthy diet’ was associated with higher ADHD methylation and can facilitate the use of methods
symptoms indirectly via higher IGF2 DNA methyla- that can strengthen causal inference (e.g. Mendelian
tion at birth. randomization) (Pingault, Cecil, Murray, Munaf o, &
The present findings showed a statistical interac- Viding, 2016; Relton & Smith, 2012). Second, the
tion, whereby although DNA methylation levels did magnitude of the observed associations was not
not differ between the early-onset and low CP youth, large, and necessitates replication in larger epidemi-
higher levels of DNA methylation were associated ological samples. Third, all measures except DNA
with higher symptoms of ADHD for the early-onset methylation were based on maternal reports. Hence,
but not for the low CP youth. What could be the the temporal stability of diet may be overestimated.
reason for this? One could posit that the reason However, it is unlikely that the magnitude of the
could lie in symptoms of ADHD – the same associ- pathways of interest (i.e. prenatal ‘unhealthy diet’ to
ation would manifest for low CP youth if their levels IGF2 DNA methylation; IGF2 DNA methylation to
of ADHD were the same as those of the early-onset ADHD symptoms) is artificially inflated by shared
youth. Another potential explanation could lie in the method variance. The use of in-depth interviews for
biological vulnerability of CP with ADHD (Beau- the assessment of ADHD symptoms adds to the
chaine, Hinshaw, & Pang, 2010). Indeed, we identi- robustness of our findings. Fourth, the present
fied an indirect effect where higher prenatal intake of study did not identify sex differences in the associ-
unhealthy fats/sugars associated with increased ation between IFG2 methylation and ADHD. Sex
ADHD via higher IGF2 DNA methylation, which differences are nevertheless a promising avenue for
may suggest a developmental risk pathway for the future investigation given that boys are generally
early-onset youth alone. It is important to mention higher in externalizing problems (such as CP and
that this biological vulnerability may be tapped by ADHD). Fifth, the IGF2 locus is a complex genomic
other measures of diet (metabolomics) and biology region that produces multiple transcripts from alter-
(HPA axis) that are sensitive to stress (Jones, Park, & native promoters, serves different biological func-
Ziegler, 2012; Reynolds, Godfrey, Barker, Osmond, tions, and is differentially expressed in different
& Phillips, 2007). tissues and at different developmental periods. This
In this study, we found that for early-onset youth, gene may also shift from monoallelic to biallelic IGF2
higher prenatal ‘unhealthy diet’ was correlated with promoter methylation during development (Issa,
higher IGF2 DNA methylation at birth, but higher Vertino, Boehm, Newsham, & Baylin, 1996); can
postnatal cumulative risk exposure was correlated show sex differences in monoallelic tissue-specific
with lower IGF2 DNA methylation at age 7. Why expression via parent of origin genetic effects (Pidsley
might DNA methylation associate with the environ- et al., 2012); can show loss of imprinting due to diet
ment differently at birth versus age 7? Findings may (Waterland, Lin, Smith, & Jirtle, 2006); and has
reflect two distinct types of risk exposure (i.e. diet vs. important functional genomic relationships (Gonza-
cumulative risk), which may differentially influence lez-Rodriguez et al., 2016). Therefore, it will be
IGF2 function. On one hand, diet (or prenatal nutri- important to establish the extent to which the
tion) has been found to directly affect the metabolic present results can be replicated and extended with
functions of the gene (Ukkola et al., 2001), showing – the addition of these molecular and epigenetic
in the present findings – higher offspring methylation mediators and moderators. Finally, it is important
with higher maternal caloric intake, and also lower to note that at the bivariate level, the association
methylation in the case of maternal caloric depriva- between prenatal ‘unhealthy diet’, IFG2 methylation,
tion (Dutch Hunger Winter; Heijmans et al., 2008). and ADHD was not significant (all p ≤ .10), but
On the other hand, more distal influences, such as became significant in the overall autoregressive
the cumulative risks examined here (e.g. poverty, cross-lag model (i.e. when controlling for all other
family discord), may affect IGF2 DNA methylation variables in the model). While such a difference can

© 2016 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
doi:10.1111/jcpp.12589 Prenatal unhealthy diet, IGF2 methylation, and ADHD 25

arise when examining bivariate associations (show- methylation patterns at birth.

ing the degree of relationship between variables in a Figure S1. Unadjusted model not controlling for pre-
pairwise fashion) versus multivariate associations natal and postnatal cumulative risk factors.
(i.e. which examine a system of predictions; e.g.
Tabachnick & Fidell, 2006), the present results
should be considered hypothesis-generating and Acknowledgements
are in need of replication. The UK Medical Research Council and the Wellcome
In summary, this study is the first to examine IGF2 Trust (Grant ref: 102215/2/13/2) and the University of
Bristol provided core support to ALSPAC. This publica-
DNA methylation as a potential intermediary biolog-
tion is the work of the authors who will serve as
ical mechanism in the association between prenatal guarantors for the contents of this paper. This research
diet and ADHD symptoms, for early-onset conduct was specifically supported by the National Institute of
youth. That we did not find continuity in IGF2 DNA Child and Human Development grant to E.D.B.,
methylation between birth and age 7 may support (R01HD068437). ARIES was funded by the BBSRC
ideas focusing on the prenatal maternal health as an (BBI025751/1 and BB/I025263/1). C.R. and T.G. are
important risk for postnatal disease vulnerability members of the Medical Research Council Integrative
(Barker, 1995). For example, a prenatal maternal Epidemiology Unit at the University of Bristol
high-fat and -sugar diet may alter the DNA methy- (MC_UU_12013/2 and MC_UU_12013/8). CC was sup-
lation status of the IGF2 gene at birth, which in turn, ported by the Economic and Social Research Council
may increase risk for a range of psychiatric and (ESRC, grant no. ES/N001273/1). J.R. was supported
by a grant from the Sophia Foundation for Scientific
health disorders. The present study highlights preg-
Research (SSWO, grant no. 201310). The authors are
nancy as being a promising window of opportunity grateful to all the families who took part in this study,
for reducing the risk of ADHD symptoms associated the midwives for their help in recruiting them, and the
with the nutritional environment and IGF2 DNA whole ALSPAC team, which includes interviewers,
methylation. This is encouraging, given the poten- computer and laboratory technicians, clerical workers,
tially modifiable nature of nutritional and epigenetic research scientists, volunteers, managers, reception-
risk factors. ists, and nurses. With regard to the ALSPAC DNA
methylation, we thank all involved, particularly the
laboratory scientists and bioinformaticians who con-
Supporting information tributed considerable time and expertise to the data in
Additional Supporting Information may be found in the this paper. The authors declare that they have no
online version of this article: competing or potential conflicts of interest.
Appendix S1. Factor analysis procedure for reducing
IGF2 methylation data and results.
Appendix S2. Location of IGF2 methylation probes Correspondence
included in the study. Edward D. Barker, Department of Psychology, King’s
Appendix S3. Intercorrelations between the IGF2 DNA College London, Institute of Psychiatry, Psychology and
methylation probes at birth. Neuroscience, De Crespigny Park, London SE5 8AF,
Table S1. Confirmatory Factor Model of IGF2 UK; Email: [email protected]

Key points
• This population-based study used a longitudinal design to investigate, in youth with early-onset persistent
(EOP) versus low conduct problems (CP), the interrelations between unhealthy diet and IGF2 DNA methylation
in the prediction of attention deficit hyperactivity disorder (ADHD) symptoms.
• Prenatal unhealthy diet was positively associated with IGF2 methylation at birth for both the EOP and low CP
youth.
• Only for EOP youth, higher IGF2 methylation predicted ADHD symptoms.
• Only for EOP youth, prenatal unhealthy diet was associated with higher ADHD symptoms indirectly via higher
IGF2 methylation.

References Barker, E.D., & Maughan, B. (2009). Differentiating early-

Barker, D.J. (1995). The fetal and infant origins of disease. onset persistent versus childhood-limited conduct problem
European Journal of Clinical Investigation, 25, 457–463. youth. American Journal of Psychiatry, 166, 900–908.
Barker, E.D., Kirkham, N., Ng, J., & Jensen, S.K. (2013). Barker, E.D., Oliver, B.R., & Maughan, B. (2010). Co-occurring
Prenatal maternal depression symptoms and nutrition, and problems of early onset persistent, childhood limited, and
child cognitive function. British Journal of Psychiatry, 203, adolescent onset conduct problem youth. Journal of Child
417–421. Psychology and Psychiatry, 51, 1217–1226.

© 2016 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
26 Jolien Rijlaarsdam et al. J Child Psychol Psychiatr 2017; 58(1): 19–27

Beauchaine, T.P., Hinshaw, S.P., & Pang, K.L. (2010). Comor- British and Norwegian samples. Social Psychiatry and
bidity of attention-deficit/hyperactivity disorder and early- Psychiatric Epidemiology, 46, 521–532.
onset conduct disorder: Biological, environmental, and Heijmans, B.T., Tobi, E.W., Stein, A.D., Putter, H., Blauw, G.J.,
developmental mechanisms. Clinical Psychology-Science Susser, E.S., . . . & Lumey, L.H. (2008). Persistent epigenetic
and Practice, 17, 327–336. differences associated with prenatal exposure to famine in
Boyd, A., Golding, J., Macleod, J., Lawlor, D.A., Fraser, A., humans. Proceedings of the National Academy of Sciences of
Henderson, J., . . . & Smith, G.D. (2013). Cohort Profile: The the United States of America, 105, 17046–17049.
‘Children of the 90s’-the index offspring of the Avon Longi- Holmes, J., Payton, A., Barrett, J., Harrington, R., McGuffin,
tudinal Study of Parents and Children. International Journal P., Owen, M., . . . & Thapar, A. (2002). Association of
of Epidemiology, 42, 111–127. DRD4 in children with ADHD and comorbid conduct
Castellanos, F.X., Lee, P.P., Sharp, W., Jeffries, N.O., Green- problems. American Journal of Medical Genetics, 114,
stein, D.K., Clasen, L.S., . . . & Rapoport, J.L. (2002). 150–153.
Developmental trajectories of brain volume abnormalities Houseman, E.A., Accomando, W.P., Koestler, D.C., Chris-
in children and adolescents with attention-deficit/hyperac- tensen, B.C., Marsit, C.J., Nelson, H.H., . . . & Kelsey, K.T.
tivity disorder. JAMA, 288, 1740–1748. (2012). DNA methylation arrays as surrogate measures of
Cecil, C.A., Lysenko, L.J., Jaffee, S.R., Pingault, J.B., Smith, cell mixture distribution. BMC Bioinformatics, 13, 86.
R.G., Relton, C.L., . . . & Barker, E.D. (2014). Environmental Howard, A.L., Robinson, M., Smith, G.J., Ambrosini, G.L.,
risk, Oxytocin Receptor Gene (OXTR) methylation and youth Piek, J.P., & Oddy, W.H. (2011). ADHD is associated with a
callous-unemotional traits: A 13-year longitudinal study. ‘Western’ dietary pattern in adolescents. Journal of Attention
Molecular Psychiatry, 19, 1071–1077. Disorders, 15, 403–411.
Chen, D.Y., Stern, S.A., Garcia-Osta, A., Saunier-Rebori, B., Issa, J.P.J., Vertino, P.M., Boehm, C.D., Newsham, I.F., &
Pollonini, G., Bambah-Mukku, D., . . . & Alberini, C.M. Baylin, S.B. (1996). Switch from monoallelic to biallelic
(2011). A critical role for IGF-II in memory consolidation human IGF2 promoter methylation during aging and car-
and enhancement. Nature, 469, 491–497. cinogenesis. Proceedings of the National Academy of
Constancia, M., Hemberger, M., Hughes, J., Dean, W., Fergu- Sciences of the United States of America, 93, 11757–11762.
son-Smith, A., Fundele, R., . . . & . Reik, W. (2002). Placental- Jacka, F.N., Ystrom, E., Brantsaeter, A.L., Karevold, E., Roth,
specific IGF-II is a major modulator of placental and fetal C., Haugen, M., . . . & Berk, M. (2013). Maternal and early
growth. Nature, 417, 945–948. postnatal nutrition and mental health of offspring by age
Drake, A.J., McPherson, R.C., Godfrey, K.M., Cooper, C., 5 years: A prospective cohort study. Journal of the Ameri-
Lillycrop, K.A., Hanson, M.A., . . . & Reynolds, R.M. (2012). can Academy of Child and Adolescent Psychiatry, 52,
An unbalanced maternal diet in pregnancy associates with 1038–1047.
offspring epigenetic changes in genes controlling glucocor- Jensen, S.K., Dickie, E.W., Schwartz, D.H., Evans, C.J.,
ticoid action and foetal growth. Clinical Endocrinology, 77, Dumontheil, I., Paus, T., & Barker, E.D. (2015). Effect of early
808–815. adversity and childhood internalizing symptoms on brain
Faienza, M.F., Santoro, N., Lauciello, R., Calabro, R., Giordani, structure in young men. JAMA Pediatrics, 169, 938–946.
L., Di Salvo, G., . . . & Cavallo, L. (2010). IGF2 gene variants Johnson, W.E., Li, C., & Rabinovic, A. (2007). Adjusting batch
and risk of hypertension in obese children and adolescents. effects in microarray expression data using empirical Bayes
Pediatric Research, 67, 340–344. methods. Biostatistics, 8, 118–127.
Fairchild, G., Passamonti, L., Hurford, G., Hagan, C.C., von Jones, D.P., Park, Y., & Ziegler, T.R. (2012). Nutritional
dem Hagen, E.A., van Goozen, S.H., . . . & Calder, A.J. (2011). metabolomics: Progress in addressing complexity in diet
Brain structure abnormalities in early-onset and adoles- and health. Annual Review of Nutrition, 32, 183–202.
cent-onset conduct disorder. American Journal of Psychia- Liu, J.H., Raine, A., Venables, P.H., & Mednick, S.A. (2004).
try, 168, 624–633. Malnutrition at age 3 years and externalizing behavior
Faraone, S.V. (2000). Genetics of childhood disorders: XX. problems at ages 8, 11, and 17 years. American Journal of
ADHD, Part 4: Is ADHD genetically heterogeneous? Journal Psychiatry, 161, 2005–2013.
of the American Academy of Child and Adolescent Psychia- Micali, N., Northstone, K., Emmett, P., Naumann, U., &
try, 39, 1455–1457. Treasure, J.L. (2012). Nutritional intake and dietary pat-
Fraser, A., Macdonald-Wallis, C., Tilling, K., Boyd, A., Golding, terns in pregnancy: A longitudinal study of women with
J., Davey Smith, G., . . . & Lawlor, D.A. (2013). Cohort Profile: lifetime eating disorders. British Journal of Nutrition, 108,
The Avon Longitudinal Study of Parents and Children: 2093–2099.
ALSPAC mothers cohort. International Journal of Epidemiol- Moffitt, T.E., Areseneault, L., Jaffee, S.R., Kim-Cohen, J.,
ogy, 42, 97–110. Koenen, K.C., Odgers, C.L., . . . & Viding, E. (2008). Research
Gonzalez-Rodriguez, P., Cantu, J., O’Neil, D., Seferovic, M.D., review: DSM-V conduct disorder: Research needs for an
Goodspeed, D.M., Suter, M.A., & Aagaard, K.M. (2016). evidence base. Journal of Child Psychology and Psychiatry,
Alterations in expression of imprinted genes from the H19/ 49, 3–33.
IGF2 loci in a multigenerational model of intrauterine Muth
en, L.K., & Muth
en, B.O. (1998–2013). Mplus. Statistical
growth restriction (IUGR). American Journal of Obstetrics analyses with latent variables. User’s guide. Los Angeles:
and Gynecology, 214, 625.e1–625.e11. Muth
en & Muth
en.
Goodman, R. (2001). Psychometric properties of the strengths Ouchi, Y., Banno, Y., Shimizu, Y., Ando, S., Hasegawa, H.,
and difficulties questionnaire. Journal of the American Adachi, K., & Iwamoto, T. (2013). Reduced adult hippocam-
Academy of Child and Adolescent Psychiatry, 40, 1337–1345. pal neurogenesis and working memory deficits in the Dgcr8-
Goodman, R., Ford, T., Richards, H., Gatward, R., & Meltzer, deficient mouse model of 22q11.2 deletion-associated
H. (2000). The Development and Well-Being Assessment: schizophrenia can be rescued by IGF2. Journal of Neuro-
Description and initial validation of an integrated assess- science, 33, 9408–9419.
ment of child and adolescent psychopathology. Journal of Pidsley, R., Dempster, E., Troakes, C., Al-Sarraj, S., & Mill, J.
Child Psychology and Psychiatry and Allied Disciplines, 41, (2012). Epigenetic and genetic variation at the IGF2/H19
645–655. imprinting control region on 11p15.5 is associated with
Goodman, A., Heiervang, E., Collishaw, S., & Goodman, R. cerebellum weight. Epigenetics, 7, 155–163.
(2011). The ‘DAWBA bands’ as an ordered-categorical mea- Pidsley, R., Wong, C.C.Y., Volta, M., Lunnon, K., Mill, J., &
sure of child mental health: Description and validation in Schalkwyk, L.C. (2013). A data-driven approach to

© 2016 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
doi:10.1111/jcpp.12589 Prenatal unhealthy diet, IGF2 methylation, and ADHD 27

preprocessing Illumina 450K methylation array data. Bmc and psychological treatments. American Journal of Psychia-
Genomics, 14, 293. try, 170, 275–289.
Pingault, J.-B., Cecil, C.A.M., Murray, J., Munaf o, M., & Tabachnick, B.G., & Fidell, L.S. (2006). Using multivariate
Viding, E. (2016). Causal inference in psychopathology: A statistics (5th edn). Boston: Pearson.
systematic review of Mendelian randomisation studies Thapar, A., Harrington, R., & McGuffin, P. (2001). Examining
aiming to identify environmental risk factors for psycho- the comorbidity of ADHD-related behaviours and conduct
pathology. Psychopathology Review. Advanced online problems using a twin study design. British Journal of
publication. doi:10.5127/pr.038115. Psychiatry, 179, 224–229.
Plessen, K.J., Bansal, R., Zhu, H., Whiteman, R., Amat, J., Thorburn, A.N., Macia, L., & Mackay, C.R. (2014). Diet,
Quackenbush, G.A., . . . & Peterson, B.S. (2006). Hippocam- metabolites, and ‘western-lifestyle’ inflammatory diseases.
pus and amygdala morphology in attention-deficit/hyper- Immunity, 40, 833–842.
activity disorder. Archives of General Psychiatry, 63, 795– Ukkola, O., Sun, G., & Bouchard, C. (2001). Insulin-like
807. growth factor 2 (IGF2) and IGF-binding protein 1 (IGFBP1)
Relton, C.L., Gaunt, T., McArdle, W., Ho, K., Duggirala, A., gene variants are associated with overfeeding-induced
Shihab, H., . . . & Davey Smith, G. (2015). Data resource metabolic changes. Diabetologia, 44, 2231–2236.
profile: Accessible resource for integrated epigenomic stud- van Mil, N.H., Steegers-Theunissen, R.P., Bouwland-Both, M.I.,
ies (ARIES). International Journal of Epidemiology, 44, 1181– Verbiest, M.M., Rijlaarsdam, J., Hofman, A., . . . & Tiemeier, H.
1190. (2014). DNA methylation profiles at birth and child ADHD
Relton, C.L., & Smith, G.D. (2012). Two-step epigenetic Men- symptoms. Journal of Psychiatric Research, 49, 51–59.
delian randomization: A strategy for establishing the causal Vangeel, E.B., Izzi, B., Hompes, T., Vansteelandt, K., Lam-
role of epigenetic processes in pathways to disease. Interna- brechts, D., Freson, K., & Claes, S. (2015). DNA methylation
tional Journal of Epidemiology, 41, 161–176. in imprinted genes IGF2 and GNASXL is associated with
Reynolds, R.M., Godfrey, K.M., Barker, M., Osmond, C., & prenatal maternal stress. Psychoneuroendocrinology, 61, 16.
Phillips, D.I.W. (2007). Stress responsiveness in adult life: Walton, E., Pingault, J.B., Cecil, C.A., Gaunt, T.R., Relton,
Influence of mother’s diet in late pregnancy. Journal of C.L., Mill, J., & Barker, E.D. (2016). Epigenetic profiling of
Clinical Endocrinology and Metabolism, 92, 2208–2210. ADHD symptoms trajectories: A prospective, methylome-
Rubia, K., Smith, A.B., Brammer, M.J., Toone, B., & Taylor, E. wide study. Molecular Psychiatry. Advanced online publica-
(2005). Abnormal brain activation during inhibition and tion. doi:10.1038/mp.2016.85.
error detection in medication-naive adolescents with ADHD. Waterland, R.A., Lin, J.-R., Smith, C.A., & Jirtle, R.L. (2006).
American Journal of Psychiatry, 162, 1067–1075. Post-weaning diet affects genomic imprinting at the insulin-
Schuch, V., Utsumi, D.A., Costa, T.V., Kulikowski, L.D., & like growth factor 2 (Igf2) locus. Human Molecular Genetics,
Muszkat, M. (2015). Attention deficit hyperactivity disorder 15, 705–716.
in the light of the epigenetic paradigm. Frontiers in Psychi- Yu, Y., Shen, H., Zeng, L.L., Ma, Q., & Hu, D. (2013).
atry, 6, 126. Convergent and divergent functional connectivity patterns
Sonuga-Barke, E.J., Brandeis, D., Cortese, S., Daley, D., in schizophrenia and depression. PLoS One, 8, e68250.
Ferrin, M., Holtmann, M., . . . & Sergeant, J. (2013). Non-
pharmacological interventions for ADHD: Systematic review Accepted for publication: 10 May 2016
and meta-analyses of randomized controlled trials of dietary First published online: 18 August 2016

© 2016 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.