Polygenic risks for joint developmental trajectories of
internalizing and externalizing problems: findings
from the ALSPAC cohort
Lydia Gabriela Speyer,1,2 Samuel Neaves,3 Hildigunnur Anna Hall,1 Gibran Hemani,3
Michael Vincent Lombardo,4,5 Aja Louise Murray,1 Bonnie Auyeung,1,5† and
Michelle Luciano1†
1Department of Psychology, University of Edinburgh, Edinburgh, UK; 2Department of Psychology, University of
Cambridge, Cambridge, UK; 3Bristol Medical School, University of Bristol, Bristol, UK; 4Laboratory for Autism and
Neurodevelopmental Disorders, Center for Neuroscience and Cognitive Systems @UniTn, Istituto Italiano di
Tecnologia, Rovereto, Italy; 5Autism Research Centre, Department of Psychiatry, University of Cambridge,
Cambridge, UK
Background: Joint developmental trajectories of internalizing and externalizing problems show considerable
heterogeneity; however, this can be parsed into a small number of meaningful subgroups. Doing so offered insights
into risk factors that lead to different patterns of internalizing/externalizing trajectories. However, despite both
domains of problems showing strong heritability, no study has yet considered genetic risks as predictors of joint
internalizing/externalizing problem trajectories. Methods: Using parallel process latent class growth analysis, we
estimated joint developmental trajectories of internalizing and externalizing difficulties assessed across ages 4 to 16
using the Strengths and Difficulties Questionnaire. Multinomial logistic regression was used to evaluate a range of
demographic, perinatal, maternal mental health, and child and maternal polygenic predictors of group membership.
Participants included 11,049 children taking part in the Avon Longitudinal Study of Parents and Children. Polygenic
data were available for 7,127 children and 6,836 mothers. Results: A 5-class model was judged optimal: Unaffected,
Moderate Externalizing Symptoms, High Externalizing Symptoms,Moderate Internalizing and Externalizing Symptoms
and High Internalizing and Externalizing Symptoms. Male sex, lower maternal age, maternal mental health problems,
maternal smoking during pregnancy, higher child polygenic risk scores for ADHD and lower polygenic scores for IQ
distinguished affected classes from the unaffected class. Conclusions: While affected classes could be relatively well
separated from the unaffected class, phenotypic and polygenic predictors were limited in their ability to distinguish
between different affected classes. Results thus add to existing evidence that internalizing and externalizing
problems have mostly shared risk factors. Keywords: Joint mental health trajectories; internalizing; externalizing;
polygenic risk; ALSPAC.
Introduction
Developmental mental health trajectories in child-
hood and adolescence are complex, correlated across
disorders and characterized by considerable hetero-
geneity (Murray, Eisner, Eisner, Nagin, & Ribeaud,
2020; Patalay, Moulton, Goodman, & Ploubidis,
2017). This developmental heterogeneity can be
summarized in terms of a small number of develop-
mental subtypes to help illuminate aetiology, mech-
anisms and consequences of specific patterns of
mental health development. This is typically done in
relation to one disorder at a time for instance, for
conduct problems (e.g. Barker & Maughan, 2009),
ADHD (e.g. Riglin et al., 2016) or depression (Rice
et al., 2019). Internalizing and externalizing difficul-
ties commonly co-occur, with up to 45% of children
and adolescents showing clinical levels of internal-
izing symptoms also suffering from high levels of
externalizing symptoms and vice versa (Patalay
et al., 2017). These domains also have correlated
developmental trajectories such that children follow-
ing a particular trajectory of internalizing problems
likely follow a similar trajectory for externalizing
difficulties (Nivard et al., 2017). Whether such sub-
groups of joint internalizing and externalizing prob-
lems can be identified and differentiated on the basis
of genetic and phenotypic risks, such as maternal
age or prematurity, is what we test in the current
study. This novel approach can illuminate their co-
occurrence, providing new insights into when and
why internalizing and externalizing problems occur
together and in isolation.
Using methods such as parallel process latent
class analysis or growth mixture modelling, a few
studies have investigated the joint development of
internalizing and externalizing difficulties, categoriz-
ing them into potentially meaningful subgroups (e.g.
Murray, Eisner, et al., 2020; Patalay et al., 2017).
Patalay et al. (2017) analysed the joint trajectories of
emotional and behavioural problems in 3- to 11-
year-old children from the UK Millennium Cohort
Study, identifying five distinct classes characterized
as ‘low symptoms’, ‘moderate behavioural’, ‘moder-
ate emotional’, ‘high emotional and moderate
†These authors contributed equally to this paper.
Conflict of interest statement: No conflicts declared.
© 2021 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.
Journal of Child Psychology and Psychiatry **:* (2021), pp **–** doi:10.1111/jcpp.13549
PFI_12mmX178mm.pdf + eps format
behavioural’ and ‘high behavioural and moderate
emotional’. Predictors considered as potential differ-
entiators of subtypes only distinguished well
between high and low symptom groups but not
between subtypes with similar symptom levels but
with different predominant symptoms (i.e. ‘high
emotional and moderate behavioural’ vs ‘moderate
emotional and high behavioural’). Murray, Eisner,
et al. (2020) estimated the joint trajectories of
internalizing and externalizing difficulties and ADHD
symptoms across ages 7–15, finding that risk factors
mostly distinguished unaffected from affected
groups but were again limited in their ability to
differentiate between affected trajectory groups with
different predominant symptoms.
Identifying developmental subtypes of symptom
trajectories is only beneficial if these subtypes also
represent meaningful clinical groups that, for exam-
ple, differ in their aetiology and thus might respond
differently to treatment. Whereas demographic, peri-
natal, child dispositional, and family and broader
social environmental predictors of joint internaliz-
ing/externalizing developmental trajectories have
been studied, the role of genetic predisposition to
mental health problems has only been indirectly
considered through parental history of psychiatric
illness. Yet, the heritability for both internalizing and
externalizing difficulties is estimated to be above
30% (Nikstat & Riemann, 2020) and for externalizing
difficulties as high as 80% (Hicks, Krueger, Iacono,
McGue, & Patrick, 2004). The genetic architecture of
internalizing and externalizing difficulties is complex
with hundreds to thousands of genes contributing to
their heritability and individual risk alleles typically
only having very small effects. Thus, composite
measures, such as polygenic risk scores (PRS), are
usually more useful indicators of genetic risk than
single genetic risk variants (Ronald, 2020). PRS
measure an individual’s genetic risk by summing
up all trait-associated alleles weighted by associa-
tion study allele effect size. PRS for mental health
disorders and cognitive abilities predict internalizing
and externalizing symptoms in clinical as well as
general populations (Jansen et al., 2018). To date,
only a few studies have considered PRS to disentan-
gle the aetiology of different subtypes of mental
health trajectories (Agnew-Blais et al., 2021). Rice
et al. (2019) investigated the associations of PRS for
major depressive disorder, schizophrenia and ADHD
with developmental trajectories of depression, find-
ing that a later-adolescence–onset class was associ-
ated with higher PRS for major depressive disorder,
whereas an early-adolescence–onset class was asso-
ciated with higher PRS for schizophrenia and ADHD.
Riglin et al. (2016) explored polygenic antecedents of
ADHD symptom subtype trajectories in the general
population. They found that higher PRS for ADHD
distinguished individuals with persistent ADHD
symptom trajectories from individuals following
other ADHD trajectories. These studies provide
initial evidence that PRS might be useful in differen-
tiating subgroups of individuals following distinct
mental health trajectories. In addition, PRS for some
mental health problems are more strongly related to
co-occurring presentation of disorders than prob-
lems occurring in isolation; thus, PRS might be
particularly relevant in the aetiology of joint trajec-
tories (Hamshere et al., 2013).
The extent to which polygenic scores capture
purely genetic effects or whether gene-environment
correlations drive the effects identified by PRS (since
a child’s PRS profile is directly related to their
parent’s genome) should also be considered. Any
direct environmental links from parent to child are
likely to be confounded by shared genetics while
direct effects of the child’s genotype could still be
mediated through the parental genome, for instance
through parental genetic effects on the intrauterine
and rearing environments. This effect of ‘genetic
nurture’ might have effects over and above the
transmission of alleles that might be independently
associated with certain risks (Taylor & Polderman,
2020). The effect of genetic nurture on children’s
development has not been widely investigated, but
some preliminary evidence suggests that up to 14%
of variance in childhood depressive symptoms can be
explained through genetic nurture (Cheesman et al.,
2020), hence, highlighting the need for also consid-
ering parental genetic effects when investigating
children’s socio-emotional development.
The aims of the current study are twofold: first,
using Parallel Process Latent Class Growth Analysis
(PP-LCGA), we examine whether joint developmental
trajectories of internalizing and externalizing diffi-
culties across childhood and adolescence (ages 4–16)
can be classified into meaningful subtypes of symp-
tom trajectories in a large UK-based longitudinal
birth cohort (N = 11,049). Second, we investigate
whether a number of candidate predictors differen-
tiate children who belong to an affected class from
those who belong to an unaffected class, with a focus
on PRS for mental health problems. We included PRS
for the child and for the mother, the latter as a
measure of genetic nurture. We additionally consider
demographic risk factors, perinatal risk factors and
maternal mental health as these have previously
been shown to differentiate between symptom
classes. We include polygenic scores for intelligence
and years of schooling as genetic indicators of
cognitive abilities.
Methods
Participants
Participants were part of the Avon Longitudinal Study of
Parents and Children (ALSPAC), a longitudinal birth cohort
study. Pregnant women resident in Avon, UK, with expected
dates of delivery 1 April 1991 to 31 December 1992 were
invited to take part in the study. The initial number of
pregnancies enrolled is 14,541 (for these at least one
© 2021 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
2 Lydia Gabriela Speyer et al.
questionnaire has been returned or a ‘Children in Focus’ clinic
had been attended by 19/07/99). Of these initial pregnancies,
there was a total of 14,676 foetuses, resulting in 14,062 live
births and 13,988 children who were alive at 1 year of age
(Boyd et al., 2013; Fraser et al., 2013). The study website
contains details of available data through a fully searchable
data dictionary and variable search tool (http://www.bristol.
ac.uk/alspac/researchers/our-data/). For families with mul-
tiple births, we only included the oldest child in order to guard
against biases due to overlapping genotypes and to allow for
the inclusion of as many time points as possible. The current
study included all children who had data on the Strengths and
Difficulties Questionnaire, the study’s outcome measure, at
least at one time-point (N = 11,049). Polygenic data were
available for 7,127 children and 6,836 mothers.
Ethical considerations
Ethical approval for the study was obtained from the ALSPAC
Ethics and Law Committee and the Local Research Ethics
Committees. Informed consent for the use of data collected via
questionnaires and clinics was obtained from participants
following the recommendations of the ALSPAC Ethics and Law
Committee at the time. Consent for biological samples has
been collected in accordance with the Human Tissue Act
(2004). For further information, see http://www.bristol.ac.uk/
alspac/researchers/research-ethics.
Measures
Internalizing and externalizing difficulties. Data on
internalizing and externalizing difficulties were collected when
children were median-aged 4, 7, 8, 9, 11, 13 and 16, using
parent-reported Strengths and Difficulties Questionnaires
(SDQ) (Goodman, 1997). The SDQ is a behavioural screening
tool measuring children’s socio-emotional development in five
domains of psychosocial functioning: prosocial behaviour,
emotional problems, peer problems, conduct problems and
hyperactivity/inattention. Each domain is assessed using five
items scored on a 3-point Likert scale (‘not true’, ‘somewhat
true’, ‘certainly true’). Internalizing difficulties are calculated
by summing up responses to peer and emotional problems
(range 0–20), whereas externalizing difficulties are calculated
as the sum score of conduct problems and hyperactivity/
inattention items (range 0–20). Higher scores represent more
difficulties. Psychometric analyses of the SDQ have found
support for structural, discriminative and convergent validity
(Kersten et al., 2016) as well as configural, metric and scalar
gender and longitudinal invariance over ages 5–14 (Murray,
Speyer, et al., 2020). The SDQ has been shown to have good
predictive validity for psychiatric disorders (Goodman, Ren-
frew, & Mullick, 2000).
Phenotypic predictors. Phenotypic predictors included
child’s sex, maternal age at birth, maternal education, area-
based deprivation, prematurity, maternal prenatal smoking,
ever breastfed, maternal prenatal depression, maternal post-
natal depression and maternal history of psychiatric illness.
For details regarding assessment and coding of phenotypic
predictors, see Appendix S1.
Polygenic predictors. Polygenic risk scores (PRS) were
calculated for the following mental health problems: Autism
Spectrum Disorder (ASD), Attention Deficit and Hyperactivity
Disorder (ADHD), Conduct Disorder, Anxiety Disorder, Major
Depressive Disorder (MDD) and Schizophrenia. In addition,
polygenic scores (PS) for Intelligence (IQ) and Years of Schooling
were included as genetic indicators of cognitive abilities. PRS
were calculated for mothers and children (insufficient geno-
typed fathers were available). For details regarding the
calculation of PRS, the discovery samples that PRS were based
on and selected p-value thresholds for each of the included
PRS, see Appendix S2.
Statistical analysis
Parallel process latent class growth analysis. To
identify meaningful subtypes of joint externalizing and inter-
nalizing difficulties trajectories, a Parallel Process Latent Class
Growth Analysis (PP-LCGA) was fitted using Mplus 8.5
(Muth
en & Muth
en, 2018). PP-LCGA estimates latent classes
based on growth factors from parallel estimated growth curves.
In contrast to growth mixture modelling, LCGA does not
assume that these classes represent ‘true’ subtypes but rather
assumes that these subtypes are a convenient summary of a
continuous distribution and therefore do not allow for within
class variation. This is reflected in the model by fixing factor
variances, and by extension, covariances to zero (Nagin &
Odgers, 2010). Growth for internalizing and externalizing
difficulties was defined by intercepts and linear and quadratic
slopes since previous research has found that children and
adolescents’ mental health trajectories follow a curvilinear
trend (Murray, Eisner, et al., 2020). Models with one to eight
classes were estimated and compared using Lo-Mendell-Rubin
(LMR) tests, Akaike information criterion (AIC), Bayesian
information criterion (BIC), sample size adjusted BIC (saBIC),
model entropy and smallest class size. To select the optimal
number of classes, the LMR test was used as the main
selection criterion with other criteria reported to provide
supplementary information to help arbitrate if the LMR test
yielded ambiguous results. The LMR test quantifies whether a
k-class model performs better than a k
1 class model and was
assessed at a < .001. Lower values on AIC, BIC and saBIC
indicate better model fit and higher model entropy values
indicate better distinguishability of classes (though the latter is
not recommended for use in model selection; Henson, Reise, &
Kim, 2007). The decision to stop at eight classes was made a
priori to ensure that smaller classes would have large enough
sample sizes to allow for further analysis. Models were fit using
a robust maximum likelihood estimator with full information
maximum likelihood estimation thus used to handle missing
data (Enders, 2001).
Prediction of class membership. Upon selection of the
optimal latent class model, following the three-step approach
which accounts for class membership uncertainty (Asparou-
hov & Muth
en, 2014), multiple multinomial logistic regression
with the largest group as reference group was used to examine
the associations between class membership and all candidate
predictors at once. The analysis was adjusted for the first five
principal components of ancestry. Missing predictor data were
dealt with using multiple imputation (n = 20) with chained
equations following the ‘H0’ approach which, using a Bayes
estimator and incorporates the main analysis model in the
imputation model. Since a substantial number of children did
not have complete data for child and maternal polygenic (risk)
scores, a sensitivity analysis with only children with complete
polygenic data was carried out for the prediction step
(N = 5,008).
Results
Parallel process latent class growth analysis
Results of the PP-LCGA showed that a 5-class model
best described parallel trajectories of internalizing
and externalizing difficulties. The LMR test indicated
that a 5-class model was significantly better than a
4-class model while adding a 6th class did not
© 2021 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
Polygenic risks for joint trajectories 3
improve model fit. Classes were reasonably well
separated (entropy = 0.773), hence, allowing for fur-
ther analyses. For model fit indices of all estimated
models and classification probabilities as well as
growth parameters for the different classes in the 5-
class model, see Table S1–S3. As visualized in Fig-
ure 1, the chosen model included the following
classes: Unaffected (40.6%), Moderate Externalizing
Symptoms (29.8%), High Externalizing Symptoms
(11.4%), Moderate Internalizing and Externalizing
Symptoms (13.3%) and High Internalizing and Exter-
nalizing Symptoms (4.86%).
Prediction of class membership
Full results of the multinomial logistic regression of
class membership for phenotypic and polygenic
predictors with the Unaffected class as reference
group summarized in Figure 2 and presented in the
Table S4. Sex was a strong predictor, with children
in the High Internalizing and Externalizing Symp-
toms, the Moderate Externalizing Symptoms and the
High Externalizing Symptoms classes being more
likely to be male compared to the Unaffected class.
Lower maternal age was also found to increase the
chance of falling into the High Internalizing and
Externalizing Symptoms, the Moderate Externalizing
Symptoms and the High Externalizing Symptoms
classes, whereas living in a more deprived area was
only associated with a higher likelihood of being in
the High Internalizing and Externalizing Symptoms
class. Perinatal factors did not distinguish well
between classes with only maternal smoking during
pregnancy being a significant class membership
predictor. In particular, children born to mothers
who smoked during pregnancy were more likely to be
in the High Internalizing and Externalizing Symp-
toms, the High Externalizing Symptoms and the
Moderate Externalizing Symptoms classes. Com-
pared to the Unaffected class, children in all other
classes were more likely to have a mother who
suffered from pre- or postnatal depression. Having
a mother with a history of psychiatric illness further
increased the chances of being in the High Internal-
izing and Externalizing Symptoms, the Moderate
Externalizing and Internalizing Symptoms and the
High Externalizing Symptoms classes. Higher PRS for
ADHD and Schizophrenia as well as lower PS for IQ
were found to increase the likelihood of being in the
High Internalizing and Externalizing Symptoms class.
Higher PRS for ADHD as well as lower PS for IQ and
years of schooling, increased the chances of being in
the Moderate Externalizing Symptoms and the High
Externalizing Symptoms classes. Maternal PRS were
not associated with class membership. Since it is
possible that some of the included phenotypic risk
factors, such as breastfeeding or maternal mental
health problems, lie on the pathway from maternal
Figure 1 Trajectories of Internalizing and Externalizing Difficulties for the selected 5-class model
© 2021 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
4 Lydia Gabriela Speyer et al.
PRS to children’s internalizing and externalizing
problems, we additionally estimated a model that
did not include any environmental risk factors. This
model still indicated that maternal PRS were not
associated with class membership (see Table S5),
thus making it unlikely that there could be a
potential mediation pathway from maternal PRS
through environmental risk factors.
The multinomial regression model was also fit with
the High Internalizing and Externalizing Symptoms
class and the High Externalizing Symptoms class as
the reference group in order to investigate whether
any of the candidate predictors separated the most
affected classes from the other affected classes (see
Table S4). In these comparisons, worse maternal
mental health emerged as the strongest indicator of
class membership, increasing the likelihood of
belonging to the High Internalizing and Externalizing
Symptoms class. The sensitivity analysis only
including children with complete polygenic data
yielded very similar results to the main analysis
(see Table S6).
Discussion
Parallel-process latent class growth analysis indi-
cated that internalizing and externalizing trajectories
across childhood and adolescence can be summa-
rized in terms of five distinct groups: Unaffected,
Moderate Externalizing Symptoms, High Externaliz-
ing Symptoms, Moderate Internalizing and External-
izing Symptoms and High Internalizing and
Externalizing Symptoms; with most children Unaf-
fected (40.6%). A number of predictors differentiated
membership in all other groups compared to the
unaffected group, including male sex, lower
Figure 2 Predictor profiles for all classes with the Unaffected class as reference group. Estimates represent odds ratio. C = child, M =
maternal. Risk factors in red represent phenotypic factors, risk factors in green represent polygenic scores for cognitive abilities and risk
factors in blue represent polygenic risk scores for mental health problems
© 2021 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
Polygenic risks for joint trajectories 5
maternal age, maternal mental health problems,
maternal smoking during pregnancy, higher poly-
genic risk scores for ADHD as well as lower polygenic
scores for IQ, while pure externalizing trajectories
were further differentiated from the unaffected class
through lower polygenic scores for years of school-
ing. Maternal mental health problems emerged as
the strongest distinguishing factors between the
most affected co-occurring issues class (High Inter-
nalizing and Externalizing Symptoms) and the other
classes. Genetic nurture, via mothers, did not pre-
dict class membership.
The joint trajectories were largely characterized by
steady levels of internalizing difficulties and slightly
decreasing levels of externalizing difficulties during
adolescence. Two co-occurring issues trajectories
emerged, one with moderate symptom levels of
internalizing and externalizing problems and one
with high symptom levels in both domains. Internal-
izing symptoms in the co-occurring issues trajectory
classes were characterized by a slight increase in
symptoms before adolescence. This increase is in
line with research showing that the median age of
onset for internalizing disorders such as anxiety is
around age 11 (Kessler et al., 2005). We identified
two pure externalizing difficulties trajectories, one at
moderate levels and one at high levels, which were
not present in previous analyses taking similar
approaches. Patalay et al. (2017) who investigated
the joint trajectories of emotional and behavioural
difficulties in a similar UK-based population, found
that externalizing difficulties were always accompa-
nied by internalizing difficulties, either at higher or
lower severity but never at the same severity. We
found that internalizing and externalizing problems
either occurred at approximately the same level of
severity, or externalizing difficulties occurred in
isolation. Murray, Eisner, et al. (2020) did not
identify a pure externalizing group and also found
a pure internalizing trajectory group, but like us,
they did identify trajectories that were characterized
by similar levels of both externalizing and internal-
izing difficulties in Swiss youths. Our study covered
a wider age range of 4–16 years, whereas Patalay
et al. only investigated trajectories from ages 3 to 11,
and Murray et al. included ages 7–15. This will have
affected our trajectory parameters and, thus, the
identification of classes; further research is needed
to replicate our findings in a similar age group.
Examining whether individuals in the affected
classes had a different set of risk profiles than
individuals in the unaffected class, our findings of
phenotypic predictors were in line with existing
literature (Patalay et al., 2017). Sex, maternal age,
maternal smoking during pregnancy and maternal
mental health problems were all found to be rela-
tively strong distinguishing factors. We further found
that some child polygenic scores distinguished
between trajectory groups. In particular, higher
PRS for ADHD and as well as lower PS for IQ
emerged as a distinguishing factor between affected
classes and the unaffected class. The two co-
occurring issues trajectories were characterized by
a similar set of risk factors but were distinguished by
the severity of these risk factors. While children in
both co-occurring issues classes had lower PS for IQ
compared to the unaffected class, lower PS for IQ
also made it more likely to be in the high co-
occurring issues class compared to the moderate
co-occurring issues symptom class. This pattern was
also true for younger maternal age and maternal
mental health problems. Comparing the pure exter-
nalizing trajectories with the co-occurring issues
trajectories no strong distinguishing factors
emerged, with only PRS for schizophrenia signifi-
cantly distinguishing between the high co-occurring
issues and the moderate externalizing symptoms
class. This is in line with previous literature hypoth-
esizing that psychotic disorders are the most severe
manifestation of a general risk for psychopathology
(Stochl et al., 2015). However, both pure externaliz-
ing trajectories were characterized by lower PS for
years of schooling compared to the unaffected class,
suggesting that genetic factors underlying noncog-
nitive education-related traits such as self-efficacy or
conscientiousness may be particularly important in
relation to the development of externalizing difficul-
ties.
While some of the PRS and PS distinguished
between different trajectory groups, overall our find-
ings indicated that, similarly to phenotypic risks,
genetic risks mainly separate affected from unaf-
fected classes, thus supporting the hypothesis that
most mental health difficulties share common aeti-
ological factors (Caspi & Moffitt, 2018). Several
recent studies have found evidence in line with this
‘p-factor’ model of psychopathology with many phe-
notypic risk factors for specific mental health prob-
lems having been shown to be significantly
associated with the p-factor (Deutz et al., 2020).
There is also evidence for a p-factor at the genetic
level, with most mental health disorders having been
found to share genetic risks (Brikell et al., 2020).
Hence, genes involved in mental health disorders
potentially operate through pleiotropic mechanisms,
consequently enhancing the risk for developing any
or multiple mental health problems rather than one
specific disorder. This is in line with the generalist
genes hypothesis, which suggests that there is a
general genetic risk for all mental health disorders
with environmental factors determining specific
manifestations (Marceau & Neiderhiser, 2020). How-
ever, the environmental risk factors considered in
the current study did not differentiate trajectories
well either; thus, future studies into which factors
might distinguish trajectories are needed, including
tests of interactions between polygenic scores and
environmental factors.
That PRS for ADHD emerged as a distinguishing
factor between affected and unaffected trajectories is
© 2021 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
6 Lydia Gabriela Speyer et al.
also in line with previous research that has consis-
tently identified ADHD PRS to be associated with
internalizing and externalizing problems (Brikell
et al., 2020), whereas evidence for association of
PRS of other mental health problems with internal-
izing and externalizing problems has so far been
more mixed (Riglin et al., 2020). Interestingly, while
polygenic scores for IQ differentiated all affected
classes from the unaffected class, years of schooling
only distinguished the pure externalizing trajectories
from the unaffected class. This is consistent with
previous research finding stronger links for an
association of polygenic scores for educational
attainment with externalizing problems than with
internalizing problems (Ensink et al., 2020). One
potential explanation for this finding is that, in the
pure externalizing symptoms class, externalizing
behaviours could be more of an overt reaction to
struggles in keeping up with school, caused by
differences in traits that have been associated with
staying in school such as conscientiousness, rather
than the result of a more general genetic liability to
mental health problems that manifests itself in
multiple domains. This is in line with findings that
academic underachievement often precedes exter-
nalizing problems (Zimmermann, Sch€utte, Taskinen,
& K€oller, 2013). Finally, we did not identify any
significant effects of maternal polygenic scores; thus,
we found no evidence for an effect of genetic nurture
on joint developmental trajectories of internalizing
and externalizing problems. It could, however, be
that this effect is of a magnitude such that larger
samples are needed to detect any effects; hence,
future larger studies need to be conducted to confirm
these null findings ideally using Bayesian
approaches.
Strengths and limitations
The key strength of this study is that we were able to
investigate the role of genetic risks for joint develop-
mental trajectories of internalizing and externalizing
difficulties in addition to phenotypic risks over the
whole period of childhood and adolescence in a well-
powered sample. We could also explore the role of
genetic nurture in joint developmental trajectories.
With regard to limitations, ALSPAC is a longitudinal
cohort with non-random attrition; children with
more behavioural difficulties are more likely to
drop-out (Wolke et al., 2009), which likely underes-
timates our effect sizes. Due to insufficient numbers,
we could not examine the effect of paternal non-
transmitted alleles and hence had a weaker measure
of genetic nurture (Bates et al., 2018). Also, some of
the PRS were based on GWAS with little power as
they only included limited numbers of cases (partic-
ularly Conduct and Anxiety Disorder). These PRS
may not be as informative as, for example, PRS for
ADHD which were based on a better-powered GWAS.
In addition, the ADHD GWAS was based on a sample
that included a large number of children while most
of the other GWAS were based on adult samples.
Considering that the importance of genetic effects
may vary across development, this may also explain
why ADHD PRS were a particularly strong predictor
of children’s internalizing and externalizing prob-
lems. Replications of this study using PRS solely
based on well-powered GWAS that ideally also
include children would therefore be highly valuable
and could potentially lead to different results. Fur-
ther, PRS-threshold selection was based on class
membership prediction within the same dataset
rather than on predictions within an independent
dataset. This may have inflated results of the multi-
nomial logistic regression (Choi, Mak, & O’Reilly,
2020). Finally, LCGA requires modelling choices (e.g.
stopping criterion for number of classes, estimation
of within-group variances) that can affect class
estimation and therefore class identification.
Conclusion
Affected subgroups from joint trajectory modelling of
internalizing and externalizing difficulties could be
separated from the unaffected class by polygenic risk
scores for ADHD and polygenic scores for IQ, and by
various phenotypic predictors. However, these risk
factors were limited in differentiating specific
affected trajectory types. Future work focusing on
the interaction of genetic and environmental predic-
tors, or on a larger range of predictors may uncover
aetiological factors related to these developmental
trajectories.
Supporting information
Additional supporting information may be found online
in the Supporting Information section at the end of the
article:
Appendix S1. Phenotypic predictors.
Appendix S2. Polygenic predictors.
Table S1. Model fit indices for the parallel process-
latent class growth analysis.
Table S2. Classification probabilities for the selected 5-
class model.
Table S3. Growth parameters for the selected 5-class
model.
Table S4. Regression parameters with different classes
as reference group.
Table S5. Regression parameters for analysis excluding
environmental risk factors with the Unaffected class as
reference group.
Table S6. Regression parameters with the Unaffected
class as reference group (using complete cases on child
and maternal polygenic scores, N = 5008).
Acknowledgements
The authors are extremely grateful to all the families
who took part in this study, the midwives for their help
© 2021 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
Polygenic risks for joint trajectories 7
in recruiting them and the whole ALSPAC team, which
includes interviewers, computer and laboratory techni-
cians, clerical workers, research scientists, volunteers,
managers, receptionists and nurses.
The UK Medical Research Council and Wellcome
(Grant ref: 217065/Z/19/Z) and the University of
Bristol provide core support for ALSPAC. GWAS data
were generated by Sample Logistics and Genotyping
Facilities at Wellcome Sanger Institute and LabCorp
(Laboratory Corporation of America) using support from
23andMe. This publication is the work of the authors,
and L.G.S., S.N., H.A.H., G.H., M.V.L., A.L.M., B.A. and
M.L. will serve as guarantors for the contents of this
paper. A comprehensive list of grants funding (PDF,
459KB) is available on the ALSPAC website. Information
on how to access the ALSPAC data can be found here:
http://www.bristol.ac.uk/alspac/researchers/access/
L.G.S. was funded by the University of Edinburgh
through a Principal’s Careers Development Scholar-
ship. H.A.H. was funded by the UK Economic and
Social Research Council (ES/R500938/1). B.A. was
supported by the European Union’s Horizon 2020
research and innovation programme under the Marie
Skłodowska-Curie grant agreement (No.813546), the
Baily Thomas Charitable Fund (TRUST/VC/AC/SG/
469207686), the Data Driven Innovation Initiative and
the UK Economic and Social Research Council (ES/
N018877/1) during the course of this work. S.N., G.H.,
M.V.L.,A.L.M.andM.L.havenofundingtodeclare for this
project. The study sponsors had no part in the design,
data analysis and interpretation of this study, in the
writing of themanuscript or in the decision to submit the
paper for publication, and the authors’ work was
independent of their funders. The authors have declared
that they have no competing or potential conflicts of
interest.
Correspondence
Lydia Gabriela Speyer, Department of Psychology,
University of Cambridge, Downing Site, Cambridge
CB2 3EA, UK; Email: ls945@cam.ac.uk
Key points
 Previous research indicated that internalizing and externalizing developmental trajectories can be
summarized into distinct subgroups, giving unique insights into risk factors that lead to different patterns
of internalizing/externalizing trajectories.
 Even though heritability estimates are very high, the role of genetic predisposition in joint internalizing/
externalizing trajectories has not been explored yet.
 Our findings indicated that higher child polygenic risk scores for ADHD, lower polygenic scores for IQ and a
number of phenotypic predictors distinguished affected from the unaffected group, but were limited in
distinguishing between different affected groups.
 Importantly, our results show that developmental patterns of mental health difficulties mostly share
common genetic and other etiological factors.
 Future work needs to focus on gene-by-environment interaction and a broader range of predictors.
References
Agnew-Blais, J.C., Belsky, D.W., Caspi, A., Danese, A., Moffitt,
T.E., Polanczyk, G.V., . . . & Arseneault, L. (2021). Polygenic
risk and the course of attention-deficit/hyperactivity disor-
der from childhood to young adulthood: Findings from a
nationally-representative cohort. Journal of the American
Academy of Child & Adolescent Psychiatry, 60, 1147–1156.
Asparouhov, T., & Muth
en, B. (2014). Auxiliary variables in
mixture modeling: Three-step approaches using Mplus.
Structural Equation Modeling, 21, 329–341.
Barker, E.D., & Maughan, B. (2009). Differentiating early-
onset persistent versus childhood-limited conduct problem
youth. American Journal of Psychiatry, 166, 900–908.
Bates, T.C., Maher, B.S., Medland, S.E., McAloney, K., Wright,
M.J., Hansell, N.K., . . .&Gillespie, N.A. (2018). The nature of
nurture: Using a virtual-parent design to test parenting
effects on children’s educational attainment in genotyped
families. Twin Research and Human Genetics, 21, 73–83.
Boyd, A., Golding, J., Macleod, J., Lawlor, D.A., Fraser, A.,
Henderson, J., . . . & Smith, G.D. (2013). Cohort profile: The
’Children of the 90s’-The index offspring of the Avon longi-
tudinal study of parents and children. International Journal
of Epidemiology, 42, 111–127.
Brikell, I., Larsson, H., Lu, Y., Pettersson, E., Chen, Q., Kuja-
Halkola, R., . . . & Martin, J. (2020). The contribution of
common genetic risk variants for ADHD to a general factor of
childhood psychopathology.Molecular Psychiatry, 25, 1809–
1821.
Caspi, A., & Moffitt, T.E. (2018). All for one and one for all:
Mental disorders in one dimension. American Journal of
Psychiatry, 175, 831–844.
Cheesman, R., Eilertsen, E.M., Ahmadzadeh, Y.I., Gjerde, L.C.,
Hannigan, L.J., Havdahl, A., . . . & McAdams, T.A. (2020).
How important are parents in the development of child
anxiety and depression? A genomic analysis of parent-
offspring trios in the Norwegian Mother Father and Child
Cohort Study (MoBa). BMC Medicine, 18, 284.
Choi, S.W., Mak, T.S.H., & O’Reilly, P.F. (2020). Tutorial: A
guide to performing polygenic risk score analyses. Nature
Protocols, 15, 2759–2772.
Deutz, M.H.F., Geeraerts, S.B., Belsky, J., Dekovi
c, M., van
Baar, A.L., Prinzie, P., & Patalay, P. (2020). General
psychopathology and dysregulation profile in a longitudinal
community sample: Stability, antecedents and outcomes.
Child Psychiatry and Human Development, 51, 114–126.
Enders, C.K. (2001). The performance of the full information
maximum likelihood estimator in multiple regression mod-
els with missing data. Educational and Psychological Mea-
surement, 61, 713–740.
© 2021 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
8 Lydia Gabriela Speyer et al.
Ensink, J.B.M., Moor, M.H.M., Zafarmand, M.H., Laat, S.,
Uitterlinden, A., Vrijkotte, T.G.M., . . . & Middeldorp, C.M.
(2020). Maternal environmental risk factors and the devel-
opment of internalizing and externalizing problems in
childhood: The complex role of genetic factors. American
Journal of Medical Genetics Part B: Neuropsychiatric Genet-
ics, 183, 17–25.
Fraser, A., Macdonald-Wallis, C., Tilling, K., Boyd, A., Golding,
J., Davey Smith, G., . . . & Lawlor, D.A. (2013). Cohort profile:
The Avon longitudinal study of parents and children:
ALSPAC mothers cohort. International Journal of Epidemiol-
ogy, 42, 97–110.
Goodman, R. (1997). The strengths and difficulties question-
naire: A research note. Journal of Child Psychology and
Psychiatry, 38, 581–586.
Goodman, R., Renfrew, D., & Mullick, M. (2000). Predicting
type of psychiatric disorder from Strengths and Difficulties
Questionnaire (SDQ) scores in child mental health clinics in
London and Dhaka. European Child and Adolescent Psychi-
atry, 9, 129–134.
Hamshere, M.L., Langley, K., Martin, J., Agha, S.S., Stergiak-
ouli, E., Anney, R.J.L., . . . & Thapar, A. (2013). High loading
of polygenic risk for ADHD in children with comorbid
aggression. American Journal of Psychiatry, 170, 909–916.
Henson, J.M., Reise, S.P., & Kim, K.H. (2007). Detecting
mixtures from structural model differences using latent
variable mixture modeling: A comparison of relative model fit
statistics. Structural Equation Modeling: A Multidisciplinary
Journal, 14, 202–226.
Hicks, B.M., Krueger, R.F., Iacono, W.G., McGue, M., &
Patrick, C.J. (2004). Family transmission and heritability
of externalizing disorders: A Twin-Family Study. Archives of
General Psychiatry, 61, 922–928.
Jansen, P.R., Polderman, T.J.C., Bolhuis, K., van der Ende, J.,
Jaddoe, V.W.V., Verhulst, F.C., . . . & Tiemeier, H. (2018).
Polygenic scores for schizophrenia and educational attain-
ment are associated with behavioural problems in early
childhood in the general population. Journal of Child
Psychology and Psychiatry and Allied Disciplines, 59, 39–47.
Kersten, P., Czuba, K., McPherson, K., Dudley, M., Elder, H.,
Tauroa, R., & Vandal, A. (2016). A systematic review of
evidence for the psychometric properties of the Strengths
and Difficulties Questionnaire. International Journal of
Behavioral Development, 40, 64–75.
Kessler, R.C., Berglund, P., Demler, O., Jin, R., Merikangas,
K.R., & Walters, E.E. (2005). Lifetime prevalence and age-of-
onset distributions of DSM-IV disorders in the national
comorbidity survey replication. Archives of General Psychi-
atry, 62, 593–602.
Marceau, K., & Neiderhiser, J. (2020). Generalist genes and
specialist environments for adolescent internalizing and
externalizing problems: A test of severity and directionality.
Development and Psychopathology, 1–8. https://doi.org/10.
1017/s0954579420001108
Murray, A.L., Eisner, M., Nagin, D., & Ribeaud, D. (2020). A
multi-trajectory analysis of commonly co-occurring mental
health issues across childhood and adolescence. European
Child and Adolescent Psychiatry, 1, 3.
Murray, A.L., Speyer, L.G., Hall, H.A., Valdebenito, S., &
Hughes, C. (2020). SDQ developmental invariance. https://
doi.org/10.31234/osf.io/zs6q5.
Muth
en, L.K., & Muth
en, B. (2018). Mplus. The comprehensive
modelling program for applied researchers: User’s guide. Los
Angeles, CA: Author.
Nagin, D.S., & Odgers, C.L. (2010). Group-based trajectory
modeling in clinical research. Annual Review of Clinical
Psychology, 6, 109–138.
Nikstat, A., & Riemann, R. (2020). On the etiology of internal-
izing and externalizing problem behavior: A twin-family
study. PLoS One, 15, e0230626.
Nivard, M.G., Lubke, G.H., Dolan, C.V., Evans, D.M., St
Pourcain, B., Munafo, M.R., & Middeldorp, C.M. (2017).
Joint developmental trajectories of internalizing and exter-
nalizing disorders between childhood and adolescence.
Development and Psychopathology, 29, 919–928.
Patalay, P., Moulton, V., Goodman, A., & Ploubidis, G.B.
(2017). Cross-domain symptom development typologies and
their antecedents: Results from the UK millennium cohort
study. Journal of the American Academy of Child and
Adolescent Psychiatry, 56, 765–776.
Rice, F., Riglin, L., Thapar, A.K.A., Heron, J., Anney, R.,
O’Donovan, M.C., & Thapar, A.K.A. (2019). Characterizing
developmental trajectories and the role of neuropsychiatric
genetic risk variants in early-onset depression. JAMA Psy-
chiatry, 76, 306–313.
Riglin, L., Collishaw, S., Thapar, A.K., Dalsgaard, S., Langley,
K., Smith, G.D., . . . & Thapar, A. (2016). Association of
genetic risk variants with attention-deficit/hyperactivity
disorder trajectories in the general population. JAMA Psy-
chiatry, 73, 1285–1292.
Riglin, L., Thapar, A.K., Leppert, B., Martin, J., Richards, A.,
Anney, R., . . . & Thapar, A. (2020). Using genetics to examine
a general liability to childhood psychopathology. Behavior
Genetics, 50, 213–220.
Ronald, A. (2020). Editorial: Polygenic scores in child and
adolescent psychiatry – strengths, weaknesses, opportuni-
ties and threats. Journal of Child Psychology and Psychiatry,
61, 519–521.
Stochl, J., Khandaker, G.M., Lewis, G., Perez, J., Goodyer,
I.M., Zammit, S., . . . & Jones, P.B. (2015). Mood, anxiety and
psychotic phenomena measure a common psychopatholog-
ical factor. Psychological Medicine, 45, 1483–1493.
Taylor, M.J., & Polderman, T.J.C. (2020). Introduction to the
Special Issue on ‘The genetic architecture of neurodevelop-
mental disorders’. Behavior Genetics, 50, 185–190.
Wolke, D., Waylen, A., Samara, M., Steer, C., Goodman, R.,
Ford, T., & Lamberts, K. (2009). Selective drop-out in
longitudinal studies and non-biased prediction of behaviour
disorders. British Journal of Psychiatry, 195, 249–256.
Zimmermann, F., Sch€utte, K., Taskinen, P., & K€oller, O.
(2013). Reciprocal effects between adolescent externalizing
problems and measures of achievement. Journal of Educa-
tional Psychology, 105, 747–761.
Accepted for publication: 22 October 2021
© 2021 The Authors. Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for
Child and Adolescent Mental Health.
Polygenic risks for joint trajectories 9