Journal of Child Psychology and Psychiatry 46:10 (2005), pp 1030–1038 doi:10.1111/j.1469-7610.2005.01524.

Finding genes in child psychology and

psychiatry: When are we going to be there?
Robert Plomin
Social, Genetic and Developmental Psychiatry Centre, Institute of Psychiatry, London, UK

Background: The seven papers in this special section chart where we are in the quest for quantitative
trait loci (QTLs) in key areas of child psychology and psychiatry such as reading and hyperactivity. But
we are not there yet. Methods: This commentary considers some new developments that are likely to
accelerate the journey towards the identification of QTLs. Results: The single most important factor is
the need for very large samples to attain adequate power to detect and replicate QTLs of very small effect
size. Another important development is the microarray, which makes it possible to genotype hundreds
of thousands of SNPs simultaneously. Using microarrays in association studies allows SNPs across the
whole genome to be genotyped. Microarrays will boost power even more when they contain all functional
polymorphisms in the genome, including functional non-coding DNA. Conclusions: Once replicable
QTLs are identified in areas such as reading and hyperactivity, the real journey will begin. Future
studies will use sets of QTLs as genetic risk indicators in top-down behavioural genomic research,
leading to gene-based diagnoses, gene-based treatments tailored to the individual, and early warning
systems and interventions. These discoveries will eventually help to prevent or at least ameliorate
childhood disorders before they cast their long shadow over development. Keywords: QTLs, micro-
arrays, SNPs, whole-genome association, behavioural genomics, reading, speech-sound disorder,
hyperactivity, learning disabilities. Abbreviations: QTLs: quantitative trait loci; SNPs: single-nucleo-
tide polymorphisms.

This special section on molecular genetics comes at
an exciting time for child psychology and psychiatry.
The field will be transformed as we move from finding
genes to using them as genetic risk indicators in our
research and eventually in our clinics. This is
beginning to happen at the other end of the life span,
in the field of gerontology. The apolipoprotein E
(APOE) gene, the only known predictor of dementia,
is often used as an index of genetic risk in diverse
gerontological research (e.g., Laurin, Masaki, Foley,
White, & Launer, 2004; Mukamal et al., 2003;
Podewils et al., 2005). APOE is a quantitative trait
locus (QTL) in that it is neither necessary nor suffi-
cient for the development of dementia, although it is
one of the largest known QTL effects, accounting for
15% of the liability. Despite this large effect size, the
APOE gene is not yet used in clinics because no
intervention is available to prevent or treat the dis-
order.
No APOE-sized effect has been found for any other
common disorders. However, hope remains for some
large effects because the replicated QTL linkages for
reading, for example, described in two papers in this
special section, could only be found if the QTLs
responsible for the linkages have relatively large ef-
fects (for example, greater than 10% of the liability).
Nonetheless, it is now generally accepted that most
genetic influence on common disorders and complex
dimensions involve many more QTLs of much smaller
effect sizes than previously considered (Cardon &
Bell, 2001; Collins, Euyer, & Chakravarti, 1997;
Ó Association for Child and Adolescent Mental Health, 2005.
Published by Blackwell Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA
Lohmueller, Pearce, Pike, Lander, & Hirschhorn,
2003).
When are we going to be there? Being an optimist,
my response is ‘soon.’ But readers would be forgiven
for being sceptical because they have heard this be-
fore (for example, Plomin & Rutter, 1998). A small
personal example of impatience and embarrassment
about the slower-than-expected progress towards
identifying QTLs is that my co-authors and I decided
that we would not write the next edition of our
behavioural genetics textbook (Plomin, DeFries,
McClearn, & McGuffin, 2001) until we had some
solid DNA results to present. The reason for this
decision was that our 2001 edition had enthused
about the field being on the cusp of a new post-
genomic era in which DNA risk indicators would add
great value to behavioural research. We are still on
that cusp. The seven papers in this special section
are excellent representatives of where we are now in
the quest for QTLs.
How are we going to get there? In this comment-
ary about the special section, I discuss some new
developments that will accelerate the journey. The
good news is that there appear to be no roadblocks
that require major detours into unexplored territ-
ory. Explanations of genetic concepts (such as
linkage, association, quantitative trait loci, single-
nucleotide polymorphisms, introns, exons) can be
found in the introductory papers for this special
section (Eley & Craig, in press; Eley & Rijsdikj, in
press).

Where are we now?
Molecular genetic research on adults has largely
focused on schizophrenia and bipolar disorder
(Craddock, O’Donovan, & Owen, 2005) but research
on reading and hyperactivity has led the way in
disorders affecting children. Reading has primarily
been studied using linkage analyses. Indeed, the
first reported QTL linkage for a complex human
disorder involved reading (Cardon et al., 1994). The
paper in this special section by Gayán et al., in
addition to illustrating state-of-the-art QTL linkage
analysis, presents an important new development in
linkage analysis: multivariate analysis that identifies
QTLs contributing to co-morbidity between dis-
orders. Gayán et al. used bivariate linkage analysis
to examine co-morbidity between reading disability
and a quantitative trait hyperactivity score in a
sample of 182 sibling pairs in 95 families in which at
least one member of the pair met criteria for reading
disability. In addition to finding expected linkages
specific to reading and to hyperactivity, the most
exciting result is a new chromosomal region (14q32)
that is linked to both reading and hyperactivity but
that has not been reported previously in univariate
analyses focused on reading or hyperactivity alone.
Such co-morbid QTLs, suggesting pleiotropic effects,
will lead to new ways of thinking about the diagnosis
and origins of childhood disorders.
Continuing the theme of multivariate linkage
analysis of reading disability, the paper by Smith
et al. investigates the hypothesis that chromosomal
regions previously linked to reading disability will
also be linked to sound-speech disorder, which is
often co-morbid with reading disability. Sound-
speech disorder has been called articulation disorder
or phonological disorder, as distinct from specific
language impairment, which involves deficits in
semantics and syntax. Smith et al. describe their
QTL linkage study of 86 sibling pairs from 65
families in which at least one sibling met criteria for
sound-speech disorder. Several DNA markers known
to be linked to reading disability were genotyped in
each of three chromosomal regions (1p36, 6p22, and
15q21). Significant linkage results emerged most
consistently between 15q21 markers and the Gold-
man–Fristoe Test of Articulation, suggesting co-
morbid QTLs for reading and sound-speech disorder.
In contrast to reading which has primarily been
the target of linkage analyses, hyperactivity has been
investigated primarily using association designs.
Three papers in this special section involve associ-
ation studies of hyperactivity; all three papers focus
on multivariate issues of heterogeneity and co-mor-
bidity. The paper by Todd et al. presents re-analyses
of association data combined across three studies for
two of the most widely studied candidate genes,
DRD4 and DAT. Although other meta-analyses sug-
gest associations between hyperactivity and these
candidate genes (Asherson & IMAGE Consortium,
When are we going to be there? 1031
2004), in their three studies, the authors had not
found associations for attention deficit hyper-
activity disorder (ADHD) as diagnosed using DSM-IV
criteria. However, associations were found for a
subtype of hyperactivity called severe combined
symptoms (N ¼ 332), which was derived from a
population-based latent-class analysis, but they did
not find associations for seven other subtypes nor for
DSM-IV inattentive (N ¼ 316), hyperactivity (N ¼ 54),
or for combined (N ¼ 368) subtypes. Although this
finding needs to be replicated, it is another example
of the possibility that QTLs will lead to new dia-
gnostic entities based on aetiology rather than
symptomatology.
The paper by van der Meulen et al. also reports a
failure to replicate associations between DSM-IV
diagnosed ADHD and DRD4 and DAT in a study of
82 children in 54 families. A novel result is that
parental ratings of response rate (degree of
improvement) to methylphenidate, the drug used to
treat hyperactivity, yielded a correlation of .56 for 28
sibling pairs. Moreover, this study provides an
example of the potential usefulness of responses to
therapeutic drugs as a tool for investigating genetic
heterogeneity: A trend towards an association with
DRD4 was found for methylphenidate response rate,
although other research linking DRD4 and DAT to
methylphenidate response has yielded mixed re-
sults. The fast-growing field of pharmacogenetics
aims to provide gene-based personalised medicine
(Goldstein, Tate, & Sisodiya, 2003; Kalow, 2004).
Gene-based diagnoses will arrive quickly in clinics if
genes can be identified that reliably predict response
to therapeutic drugs so that drugs are administered
to those who are most likely to benefit from them
(Arranz et al., 2000).
These two studies suggest genetic heterogeneity of
hyperactivity: Associations with dopamine genes
were found for subtypes of hyperactivity (severe
combined symptoms in the paper by Todd et al., and
hyperactivity that responds to methyphenidate in
the paper by van der Meulen et al.) but not for DSM-
IV diagnosed ADHD. The third paper, by Stevenson
et al., also points to genetic heterogeneity in that an
association was found for hyperactivity only when it
is co-morbid with reading disability, not for hyper-
activity itself. Rather than using a case–control
association design as in the other two studies, Ste-
venson et al. used a within-family design (parent–
child trios and duos) that compares transmitted and
non-transmitted alleles from parents to probands,
with the non-transmitted alleles serving as a within-
family control. For 152 children diagnosed with
ADHD, no association was found with a noradrener-
gic receptor gene (ADRA2A). However, for 82 of these
children who were also diagnosed with reading dis-
ability, a significant association was found with one
of the two noradrenergic receptor genes (ADRA2A).
Not all molecular genetic research relies on QTL
strategies. It is likely that, in addition to QTLs,
1032 Robert Plomin
common disorders also include a heterogeneous
collection of individually rare, highly penetrant
mutations. For example, although it is generally
accepted that most genetic influences on common
disorders such as mental retardation are primarily
caused by common polymorphisms of small effect in
the population, hundreds of rare mutations with ef-
fects on mental retardation have been identified
(Inlow & Restifo, 2004). The paper in this special
section by Milner et al. is a good example of the
benefits of studying rare genetic disorders. Prader-
Willi syndrome (PWS) has an incidence of about 1 in
15,000 children, with childhood obesity as its most
apparent symptom. Much genetic research has fo-
cused on PWS because it was one of the first human
examples of imprinted genes in which the expression
of genes is determined by the parent of origin. In a
large (1.5 Mb) region on the short arm of chromo-
some 15 (15q11-13), there are several imprinted
genes that are normally expressed only if they are on
the paternal chromosome. PWS is caused by loss of
expression of these genes. Although loss of expres-
sion most often occurs because of a deletion in this
chromosomal region on the paternal chromosome, it
can also be caused by a disomy in which two copies
of the maternal chromosome are inherited, called
maternal uniparental disomy (UPD). Although both
subtypes of PWS result in the absence of functional
copies of these paternally expressed genes, the
deletion subtype has only one copy of the genes
whereas UPD has the normal complement of two
copies. This genetic difference between the deletion
and UPD subtypes of PWS could result in phenotypic
differences between them. Some research has sug-
gested that the UPD subtype leads to a milder ver-
sion of PWS than the deletion subtype, but the
Milner et al. study points to a difference between the
two subtypes in which UPD causes more social
interaction problems than the deletion subtype. The
difference in the social interaction effect is moderate,
about half a standard deviation, and it is specific in
that differences were not found for other behaviour
problems or for IQ. This finding may have wider
significance because other evidence suggests that
these imprinted genes may not be limited just to
social interaction but may be more generally in-
volved in autistic spectrum disorder. Importantly,
the short arm of chromosome 15 has shown linkage
in studies of autism (Santangelo & Tsatsanis, 2005).
In the final paper in this special section, Harlaar,
Butcher, Meaburn, Craig, and Plomin (in press) use
a set of five SNPs as an aggregate genotypic index to
conduct the first systematic behavioural genomic
analyses of multivariate, developmental and gene–
environment issues. The single-nucleotide poly-
morphisms (SNPs) were identified in previous
research to be associated with general cognitive
ability (‘g’) at 7 years. As an example of multivariate
behavioural genomic analysis, they report that the ‘g’
SNP set is also associated with reading. In terms of
development, the SNP set for ‘g’ at 7 years yields
significant correlations with ‘g’ as early as 2 years,
suggesting genetic continuity in development. This
finding foreshadows what will eventually be a major
clinical application of SNP sets or QTL sets more
generally: Predicting problems early in development
with a QTL set offers the hope of designing beha-
vioural and environmental interventions that will
prevent these problems before they occur. In relation
to gene–environment interplay, several examples of
gene–environment interaction and correlation were
identified by Harlaar et al.
More important than these specific results are the
general implications of the Harlaar et al. paper for
behavioural genomic research in developmental
psychopathology. If effect sizes are as small as .2% of
the variance, as reported in the Harlaar et al. paper,
sample sizes in the thousands are needed to achieve
reasonable power to use SNPs in behavioural ge-
nomic analyses. However, the authors show that
SNPs can be combined in a SNP set in which the
effects of the SNPs are aggregated, which will greatly
relieve the pressure for large samples. Although the
current SNP set for ‘g’ accounts for only about 1% of
the variance, which requires samples of about 1000
to reach 80% power (p <.01, one-tailed), this SNP set
was derived from a 10 K microarray which genotypes
10,000 SNPs simultaneously. The authors are cur-
rently using a 100 K microarray set which is ex-
pected to produce a SNP set that accounts for ten
times as much variance, about 10% of the total
variance; moreover, a 500 K microarray set is ex-
pected by the end of 2005, which could account for
the majority of the heritability of ‘g’. When QTL sets
of this magnitude are developed, they can be used in
any reasonably sized study as a measured index of
genetic risk in behavioural genomic research. I pre-
dict that QTL sets are the way in which DNA will
become integrated in top-down behavioural genomic
research in child psychology and psychiatry.
More powerful vehicles
In my view, the lack of power to detect QTLs of very
small effect size is by far the largest impediment to
progress in this area. As noted earlier, it is now gen-
erally accepted that most genetic influences on com-
mon disorders and complex dimensions involve many
more QTLs of much smaller effect sizes than previ-
ously considered. Type 1 diabetes, which has been the
focus of intense molecular genetic research, is an
instructive model for research on common disorders
and complex dimensions. Several monogenic forms of
Type 1 diabetes have been discovered but they ac-
count for fewer than 5% of diabetes cases (Florez,
Hirschhorn, & Altshuler, 2003). The hypothesis that
the same genes might harbour less deleterious (and
more common) polymorphisms that would contribute
to common forms of diabetes has not been supported.

As a recent review points out: ‘The genetic analysis of
diabetes provides under-appreciated support for a
model where rare, monogenic forms of diseases are
due to variants that are rare in the population and
have severe impact on the encoded proteins, and
common forms are due to variants that exist at signi-
ficant population frequencies and have modest ef-
fects’ (Florez et al., p. 278). Nonetheless, the genes
responsible for monogenic forms of diabetes have
been helpful in showing the diverse pathways that can
give rise to diabetes.
We do not know the distribution of QTL effect sizes
for any complex trait or common disorder but it is
beginning to look as if the biggest effects are not very
big – APOE seems to be the exception rather than the
rule. Moreover, we can be sure that the smallest ef-
fects are very small, so small that we will never have
the power to detect them. We need to break not just
the 1% QTL barrier (Plomin, DeFries, Craig, &
McGuffin, 2003) but perhaps even the .1% QTL
barrier. One percent QTLs require unselected sam-
ples on the order of 1000; .1% QTLs require samples
on the order of 10,000. Underpowered studies are
likely to be responsible for the widespread failure to
replicate linkages and associations for common dis-
orders, such as those reported in this special section
in relation to hyperactivity and the DRD4 and DAT
genes. The problem is not special to developmental
psychopathology; such failures to replicate occur in
research on all common disorders. As one of many
examples, the quest for QTLs for dementia other
than APOE has been intense. More than 10 such
studies were reported per month during 2003 but
a review indicated that none of these reports has
replicated consistently (Bertram & Tanzi, 2004).
Although there are many adjustments that can help
to improve the performance of our research vehicles,
reaching the power to detect QTLs of very small effect
size requires more than a tune-up. We need to trade up
for more powerful vehicles with bigger engines: Huge
samples of many thousands of individuals are needed
to detect QTLs of very small effect size.
Convoys
A potential benefit of the need for very large samples
is that research will require international collabora-
tion – we will need to travel in convoys. Several
international collaborations have emerged in autism
(e.g., International Molecular Genetic Study of Aut-
ism Consortium (IMGSAC), 2001); most recently, the
Autism Genome Project of the National Alliance for
Autism Research (2005) is a collaborative study of
1500 families with affected sibling pairs (National
Alliance for Autism Research 2005). For hyperactiv-
ity, a 12-site international collaborative project on
the genetics of hyperactivity (IMAGE) aims to collect
1400 sibling pairs (Asherson & IMAGE consortium,
2004). The paper in this section by Todd et al.
When are we going to be there? 1033
demonstrates the value of collaboration at the level
of analysis. Another example is a 15-site collabora-
tion on the genetics and neurobiology of reading
disability (NeuroDys), which has recently been fun-
ded by the European Framework VI to study 2000
children diagnosed as reading disabled and is led by
Gerd Schulte-Körne of Marburg, Germany.
Better maps and navigational instruments
Although the Human Genome Project has produced
an excellent map of our 3 billion DNA sequences and
amazing bioinformatics instruments to use this map,
these are not enough in themselves to find QTLs for
complex traits. Better navigational instruments are
coming into use such as whole-genome association
analysis, microarrays, and direct association strat-
egies, as explained below.
Linkage and association
Linkage is like using a telescope to scan the horizon
systematically for major features such as distant
mountains (large QTL effects). However, the tele-
scope goes out of focus when trying to detect nearby
hills (small QTL effects). Allelic association offers a
complementary instrument that shows nearby hills
but cannot chart distant mountains. The past dec-
ade has seen a shift from linkage to association de-
signs because of their greater power to detect QTLs of
small effect size. Although linkage scans have been
extremely successful in identifying genes responsible
for rare monogenic disorders, they have met with
limited success for common disorders and quanti-
tative traits. Again, Type 1 diabetes can serve as an
example. Four whole-genome linkage scans for Type
1 diabetes have been reported involving thousands of
probands, but results are mixed other than for the
HLA locus and no new gene has been identified
based solely on its initial localization in a whole-
genome scan for linkage (Florez et al., 2003). Linkage
designs dominated research in the first areas of child
psychopathology that were investigated using
molecular genetic techniques: reading disability and
autism. In contrast, areas such as hyperactivity,
which have been investigated more recently, have
relied on association designs. In this special section,
the two papers on reading (Gayán et al., in press;
Smith, Pennington, Boada, & Shriberg, in press) in-
volve linkage, whereas the three papers on hyper-
activity (Stevenson et al., Todd et al., van der Meulen
et al.) involve association.
As mentioned, linkage is systematic but not
powerful; association is powerful but not systematic
(Zondervan & Cardon, 2004). For this reason, link-
age studies can systematically scan the genomic
landscape with only a few hundred DNA markers. If
linkage is a telescope, association is a microscope.
Because association is myopic, hundreds of
1034 Robert Plomin
thousands of DNA markers – a huge fine-scale map –
are needed if association were to be systematic like
linkage (Ke et al., 2004). Because such a whole-
genome scan for association has not been possible
until recently, association studies have focused on a
few candidate genes, especially dopamine and sero-
tonin genes, which are of known relevance to
psychopathology.
Microarrays and DNA pooling
An important new development is that it is now
possible to conduct systematic whole-genome
association studies (Hirschhorn & Daly, 2005).
Large-scale maps with millions of SNPs are available
(Hinds et al., 2005), and a new technology called
microarrays makes it possible to use a single primer
to genotype tens of thousands of SNPs rather than
needing a different primer for each SNP (Matsuzaki
et al., 2004). The problem is that microarrays are
expensive and can be used only once; it would be
very expensive to conduct a microarray study of the
very large samples needed to detect QTLs of small
effect size. However, as described in the Harlaar et al.
paper in this section, a novel strategy is to use SNP
microarrays to genotype DNA pooled across indi-
viduals in extreme groups such as cases versus
controls or low versus high groups on a quantitative
trait. In other words, instead of using thousands of
microarrays to genotype large samples, we can use
just one microarray to estimate average SNP allelic
frequencies for a group’s pooled DNA (Norton, Wil-
liams, O’Donovan, & Owen, 2004; Sham, Bader,
Craig, O’Donovan, & Owen, 2002). Because micro-
arrays can be used to genotype hundreds of thou-
sands of SNPs, replication designs are needed to
weed out false positive results – with an alpha level of
1%, thousands of significant results are expected by
chance for a microarray with hundreds of thousands
of SNPs.
Direct versus indirect association
SNPs used on commercially available microarrays
are evenly distributed across the genome in order to
facilitate whole-genome scans. However, this ap-
proach relies on indirect association between the
SNPs and the QTLs that actually affect the quantit-
ative trait: The SNP is just a marker for a nearby
QTL. Indirect association greatly reduces power (as a
function of the distance between the SNP and the
QTL, called linkage disequilibrium) as compared to
direct association in which the SNP is assumed to be
the QTL itself. Power is reduced as a function of the
distance between the SNP and the QTL, which is why
hundreds of thousands of SNPs are needed for a
whole-genome association scan based on indirect
association.
The greater power of direct association is the
reason why association studies have focused on
functional polymorphisms in candidate genes. Power
of whole-genome association scans would be greatly
increased if the scans were based on direct associ-
ation in which DNA polymorphisms were functional
QTLs rather than just markers of QTLs (indirect
association). The next major advance in whole-
genome association research will be microarrays
with functional SNPs, which will permit direct
association.
Selecting SNPS for direct association
The first step in selecting SNPs for whole-genome
studies testing direct association will involve SNPs
in exons that result in a change in amino acid
sequence, called non-synonymous SNPs, which are
likely to be functional. However, there are only a
few tens of thousands of non-synonymous SNPs
because fewer than half of all SNPs are in gene
regions broadly defined, only 1% of these are in
exons, and only half of these are non-synonymous
(Hinds et al., 2005). All exons in the human gen-
ome are currently being re-sequenced on a sample
of 48 individuals, which is expected to yield as
many as 50,000 non-synonymous SNPs (Sanger
Centre, 2005). Microarrays with these non-
synonymous SNPs will be available in the next year
or two and will lead to more powerful whole-
genome association scans that will permit direct
association analyses.
Non-coding DNA
Although non-synonymous SNPs are the source of
many rare single-gene disorders, there are other
sources of functional polymorphisms. Polymorph-
isms other than SNPs are known to be functional,
such as repetitive DNA sequences in exons. More-
over, DNA polymorphisms do not need to be in exons
in order to be functional. RNA from introns and from
regions upstream from genes called promoters is well
known to have regulatory effects on genes and gene
products.
An exciting new discovery is that DNA well away
from protein-coding genes may be functional. Much
intergenic DNA is now known to be transcribed into
RNA and it has been suggested that this intergenic
non-coding RNA regulates the expression of other
genes and gene products (Mattick, 2004). One
finding in support of this theory is that the human
species has the greatest ratio of non-coding DNA to
protein-coding DNA – more than 98% of the human
genome consists of non-coding DNA. It seems hard
to believe that 98% of the genome is just non-
functional evolutionary debris, but this has been
the orthodox view among molecular geneticists. One
reason for this orthodox view is that most non-
coding DNA is not conserved across species, espe-
cially as compared to protein-coding DNA, which is
highly conserved. Although conserved DNA regions

are likely to be functional, it does not necessarily
follow that non-conserved DNA regions are not
functional. To the contrary, it is possible that un-
iquely human DNA variation represents especially
good candidates for subtle QTL effects on human
behaviour.
I think that there is much more to be learned about
how non-coding DNA affects behaviour. For this
reason, I prefer to focus my gene-finding efforts on
the whole genome rather than limiting it to genes
that code for proteins. Also for this reason, I prefer to
refer to ‘QTLs’ rather than ‘genes’ because the word
‘QTL’ is neutral as to whether or not the functional
DNA polymorphism is in a traditional protein-coding
gene.
In summary, future maps for the QTL journey are
likely to be on a much finer scale, identifying small
but salient features of the DNA landscape not seen
on maps today. Thanks to microarrays, these large-
scale maps will be able to cover the entire genome
and eventually they will include all functional poly-
morphisms, whether in non-coding DNA or protein-
coding DNA (Botstein & Risch, 2003).
Where is ‘there’?
Our current maps do not tell us what we will find at
the end of the journey. If at the end of this long
journey all we find are such tiny QTLs, will it have
been worth the trip? The answer is definitely yes. The
basic science goal is to find as many of the QTLs as
possible that are responsible for the heritability of
mental health and illness. It would have been easier
if most of the heritability of complex traits and
common disorders were largely due to one or two
APOE-sized QTLs but that does not seem to be the
case.
Even if ‘there’ means identifying many QTLs of
small effect, they can be used together as ‘QTL sets’
that account for substantial amounts of variance
even if the constituent QTLs have very small effects,
as explained in the paper by Harlaar et al. Because
genotyping has become so inexpensive (and will be
even less expensive as trait-specific QTLs are avail-
able on microarrays), it is quite feasible to genotype
100 SNPs that together account for 25% of the vari-
ance of a trait and thus only require a few dozen
subjects to achieve reasonable power when used as a
genetic risk index. QTL sets will also eventually have
clinical implications for diagnosis, treatment and
prediction.
QTL sets will not be as useful for molecular genetic
research because each of the many QTLs will have a
different molecular mechanism. Nonetheless, each
QTL opens a tiny window on pathways between
genes and behaviour, although very large samples
will be needed to see anything through those tiny
windows. Moreover, molecular genetic research
might profit from considering QTL sets as a system of
When are we going to be there? 1035
molecular processes that have common functional
effects on behaviour.
When we say that effect sizes are small, this is a
statement about the average effect of QTLs in a
population; it is possible that some QTLs have larger
effects on certain individuals but that their effects
are diluted when averaged across the population. It
is also possible that small effects of QTLs are due to
gene–gene interactions (epistasis) or gene–environ-
ment interactions, which could make QTLs ex-
tremely difficult to detect. However, quantitative
genetic research suggests that this is not generally
the case, although quantitative genetic data do not
provide definitive evidence for additive effects be-
cause other genetic processes such as assortative
mating and environmental factors such as sibling
interaction could mask epistatic effects. Nonethe-
less, even if QTL effects are additive on average,
epistatic interactions can occur for specific QTLs,
although even greater care must be taken to avoid
false positive results (Maier et al., 2005; Marchini,
Donnelly, & Cardon, 2005).
Gene–environment interaction (GxE) would also
reduce the effect size of QTLs and would make it
difficult to identify QTL associations because the
QTLs would need to be identified as they interact
with specific measures of the environment (Hunter,
2005). QTLs would not be seen at all as main effects
if GxE were disordinal (a cross-over interaction in
which the QTLs have effects in the opposite direction
in low and high risk environments). However, it is
much more likely that interactions are ordinal which
means that the QTLs involved in GxE will also show
some main effect that can be detected independent of
the environment even though the main effect of the
QTLs would be diluted by the interaction. Although
quantitative genetic research does not show much
evidence for GxE (e.g., Asbury, Wachs, & Plomin, in
press), this may be because human quantitative
genetic designs have little power to detect GxE
(Wahlsten, 1990). However, GxE has also been dif-
ficult to document in animal studies in which both
genes and environments can be manipulated, greatly
increasing the power to detect GxE (e.g., Henderson,
1972). Nonetheless, some interactions between spe-
cific genes and specific environmental measures
have been reported (e.g., Caspi et al., 2003) and
replicated (Eley et al., 2004).
Finally, we might end up finding QTL effects on
components and constellations of behaviour that
look very different from the disorders and dimen-
sions we currently study. Genetic effects might be
narrower or broader than the traits assessed by our
current measures or diagnostic schemes, which are
based on phenotypic rather than genetic analyses.
Multivariate quantitative genetic research is needed
to chart the genetic structure of psychopathology.
For example, the subtypes used in the paper in this
special section by Todd et al. are based on a multi-
variate genetic analysis, which suggests that the
1036 Robert Plomin
subtypes are largely genetically independent
(Rasmussen et al., 2004). The molecular genetic test
of the hypothesis of genetic heterogeneity is that
different QTLs are associated with the different
subtypes, an hypothesis supported in part by the
results of the Todd et al. analysis. Similarly, Ste-
venson et al. report that a particular QTL association
was found only when hyperactivity was co-morbid
with reading disability.
Most multivariate molecular genetic research,
however, suggests broader rather than narrower
genetic effects. This can be seen in molecular genetic
analyses reported in this special section that show
co-morbid QTLs for reading and hyperactivity (Gay-
án et al.), for reading and sound-speech disorder
(Smith et al.), and for reading and ‘g’ (Harlaar et al.).
More multivariate results of this sort can be expected
because multivariate quantitative genetic analyses
indicate that genetic effects are broad across inter-
nalising disorders and across externalising disorders
(Kendler, Prescott, Myers, & Neale, 2003; Krueger et
al., 2002), across learning disabilities (Plomin &
Kovas, in press), and between hyperactivity and
reading disability (Willcutt et al., 2003). More genetic
research is also needed to investigate the relation-
ship between quantitative dimensions and qualitat-
ive disorders in order to explore the QTL hypothesis
that common disorders are the quantitative extreme
of the same genetic and environmental influences
responsible for variation throughout the distribu-
tion. At the most fundamental level, more work is
always welcome on phenotypic measurement issues
such as multiple sources of information (not just
self-report questionnaires and interviews), multiple
situations (school and friends as well as family), and
multiple ages (not just a single measurement occa-
sion) (Freimer & Sabatti, 2003; Stevenson et al.,
2005). Endophenotypes, especially measures of the
brain and mind, are currently in vogue as a simpler
level of analysis for studying complex traits (Bear-
den, Reus, & Freimer, 2004; Gottesman & Gould,
2003). Although all levels of analysis from genes to
behaviour are important to study in their own right
and in terms of understanding pathways between
genes and behaviour, I am not convinced that en-
dophenotypes will prove to be useful for finding QTLs
for what are quintessentially behavioural disorders
such as autism, hyperactivity and reading disability.
For childhood disorders, most research on en-
dophenotypes has been done in the area of hyper-
activity; recent reviews do not find that
endophenotypes have as yet advanced the quest for
QTLs, although larger studies and other biological
endophenotypes may be more successful (Asherson
& IMAGE Consortium, 2004; Stevenson et al., 2005).
In summary, questions about the final destination in
the quest for QTLs include the distribution of QTL
effect sizes and the extent to which QTL main effects
are diluted by epistasis, GxE and measurement
problems.
Not the end
The end of this journey to identify QTLs for mental
health and illness in childhood is just the beginning
of a much more interesting trip through the country
lanes that cut across our fields. This will be a trip for
everyone, not just for a few QTL-hunters: Once QTLs
are identified, they can be used as genetic risk indic-
ators to enrich any behavioural research. Con-
versely, behavioural research will also contribute
importantly to our understanding of how genes work
for complex traits and common disorders. The
phrase functional genomics has become synonymous
with a bottom-up molecular biological approach to
understand how genes work beginning with gene
products in cells. However, it is important to
understand how genes work at all levels of analysis
from genes to brain to behaviour. As described in the
paper by Harlaar et al., a top-down behavioural ge-
nomics level of analysis seems likely to pay off more
quickly in terms of gene-based diagnoses, gene-
based treatments tailored to the individual, and
especially in terms of early warning systems and
interventions that prevent or at least ameliorate
childhood disorders before they cast their long
shadow over development.
Acknowledgements
Preparation of this paper was supported in part by a
programme grant from the UK Medical Research
Council (G9424799), a grant from the Wellcome
Trust (GR075492), and a grant from the US National
Institute of Child Health and Human Development
(HD044454).
Correspondence to
Robert Plomin, Box Number P080, Social, Genetic
and Developmental Psychiatry Centre, Institute of
Psychiatry, King’s College London, De Crespigny
Park, London, United Kingdom SE5 8AF, UK; Tel:
0044 20 7848 0893; Fax: 0044 20 7848 0895;
Email: r.plomin@iop.kcl.ac.uk
References
Arranz, M.J., Munro, J., Birkett, J., Bolonna, A.,
Mancama, D., Sodhi, M., Lesch, K.P., Meyer, J.F.,
Sham, P., Collier, D.A., Murray, R.M., & Kerwin, R.W.
(2000). Pharmacogenetic prediction of clozapine
response. Lancet, 355, 1615–1616.
Asbury, K., Wachs, T., & Plomin, R. (in press).
Environmental moderators of genetic influence on
verbal and nonverbal abilities in early childhood.
Intelligence.
Asherson, P. and IMAGE Consortium. (2004). Atten-
tion-Deficit Hyperactivity Disorder in the post-geno-
mic era. European Child and Adolescent Psychiatry,
131, I50–I70.

Bearden, C.E., Reus, V.I., & Freimer, N.B. (2004). Why
genetic investigation of psychiatric disorders is so
difficult. Current Opinion in Genetic Development, 14,
280–286.
Bertram, L., & Tanzi, R.E. (2004). Alzheimer’s disease:
One disorder, too many genes? Human Molecular
Genetics, 13, R135–R141.
Botstein, D., & Risch, N. (2003). Discovering genotypes
underlying human phenotypes: Past successes for
Mendelian disease, future approaches for complex
disease. Nature Genetics, 33, 228–237.
Cardon, L.R., & Bell, J. (2001). Association study
designs for complex diseases. Nature Genetics, 2,
91–99.
Cardon, L.R., Smith, S.D., Fulker, D.W., Kimberling,
W.J., Pennington, B.F., & DeFries, J.C. (1994).
Quantitative trait locus for reading disability on
chromosome 6. Science, 266, 276–279.
Caspi, A., Sugden, K., Moffitt, T.E., Taylor, A., Craig,
I.W., Harrington, H., McClay, J., Mill, J., Martin, J.,
Braithwaite, A., & Poulton, R. (2003). Influence of life
stress on depression: Moderation by a polymorphism
in the 5-HTT gene. Science, 301, 386–389.
Collins, F.S., Euyer, M.S., & Chakravarti, A. (1997).
Variations on a theme: Cataloguing human DNA
sequence variation. Science, 278, 1580–1581.
Craddock, N., O’Donovan, M.C., & Owen, M.J. (2005).
The genetics of schizophrenia and bipolar disorder:
Dissecting psychosis. Journal of Medical Genetics, 42,
193–204.
Eley, T.C. & Craig, I.W. (in press). Introductory guide to
the language of molecular genetics. Journal of Child
Psychology and Psychiatry [in this issue].
Eley, T.C. & Rijsdikj, F. (in press). Introductory guide to
the statistics of molecular genetics. Journal of Child
Psychology and Psychiatry [in this issue].
Eley, T.C., Sugden, K., Gregory, A.M., Sterne, A., Plomin,
R., & Craig, I.W. (2004). Gene–environment interaction
analysis of serotonin system markers with adolescent
depression. Molecular Psychiatry, 9, 908–915.
Florez, J.C., Hirschhorn, J., & Altshuler, D. (2003). The
inherited basis of diabetes mellitus: Implications for
the genetic analysis of complex traits. Annual Review
of Genomics and Human Genetics, 4, 257–291.
Freimer, N. & Sabatti, C. (2003). The human phenome
project. Nature Genetics, 34, 15–21.
Gayán, J., Willcutt, E.G., Fisher, S.E., Francks, C.,
Cardon, L.R., Olson, R.K., Pennington, B.F., Smith,
S.D., Monaco, A.P. & DeFries, J.C. (in press). Bivari-
ate linkage scan for reading disability and attention-
deficit/hyperactivity disorder localises pleiotropic
loci. Journal of Child Psychology and Psychiatry [In
this issue].
Goldstein, D.B., Tate, S.K., & Sisodiya, S.M. (2003).
Pharmacogenetics goes genomic. Nature Review
Genetics, 4, 937–947.
Gottesman, I.I. & Gould, T.D. (2003). The endopheno-
type concept in psychiatry: Etymology and strategic
intentions. American Journal of Psychiatry, 160, 636–
645.
Harlaar, N., Butcher, L., Meaburn, E., Craig, I.W., &
Plomin, R. (in press). A behavioural genomic analysis
of DNA markers associated with general cognitive
ability in 7-year-olds. Journal of Child Psychology and
Psychiatry.
When are we going to be there? 1037
Henderson, N.D. (1972). Relative effects of early rearing
environment on discrimination learning in house-
mice. Journal of Comparative and Psychological
Psychology, 72, 505–511.
Hinds, D.A., Stuve, L.L., Nilsen, G.B., Halperin, E.,
Eskin, E., Ballinger, D.G., Frazer, K.A., & Cox, D.R.
(2005). Whole-genome patterns of common DNA
variation in three human populations. Science, 307,
1072–1079.
Hirschhorn, J.N., & Daly, M.J. (2005). Genome-wide
association studies for common diseases and com-
plex traits. Nature Reviews Genetics, 6, 108.
Hunter, D.J. (2005). Gene–environment interactions in
human diseases. Nature Reviews Genetics, 6, 287–
298.
Inlow, J.K., & Restifo, L.L. (2004). Molecular and
comparative genetics of mental retardation. Genetics,
166, 835–881.
International Molecular Genetic Study of Autism Con-
sortium (IMGSAC). (2001). A genomewide screen for
autism: Strong evidence for linkage to chromosomes
2q, 7q, and 16p. American Journal of Human Genet-
ics, 69, 570–581.
Kalow, W. (2004). Human pharmacogenomics: The
development of a science. Human Genomics, 1, 375–
380.
Ke, X., Hunt, S., Tapper, W., Lawrence, R., Stavrides,
G., Ghori, J., Whittaker, P., Collins, A., Morris, A.P.,
Bentley, D., Cardon, L.R., & Deloukas, P. (2004). The
impact of SNP density on fine-scale patterns of
linkage disequilibrium. Human Molecular Genetics,
13, 577–588.
Kendler, K.S., Prescott, C.A., Myers, J., & Neale, M.C.
(2003). The structure of genetic and environmental
risk factors for common psychiatric and substance
use disorders in men and women. Archives of General
Psychiatry, 60, 929–937.
Krueger, R.F., Hicks, B.M., Patrick, C.J., Carlson, S.R.,
Iacono, W.G., & McGue, M. (2002). Etiologic connec-
tions among substance dependence, antisocial beha-
vior, and personality: Modeling the externalizing
spectrum. Journal of Abnormal Psychology, 111,
411–424.
Laurin, D., Masaki, K.H., Foley, D.J., White, L.R., &
Launer, L.J. (2004). Midlife dietary intake of anti-
oxidants and risk of late-life incident dementia: The
Honolulu–Asia Aging Study. American Journal of
Epidemiology, 159, 959–967.
Lohmueller, K.E., Pearce, C.L., Pike, M., Lander, E.S., &
Hirschhorn, J.N. (2003). Meta-analysis of genetic
association studies supports a contribution of com-
mon variants to susceptibility to common disease.
Nature Genetics, 33, 177–182.
Maier, L.M., Chapman, J., Howson, J.M., Clayton,
D.G., Pask, R., Strachan, D.P., McArdle, W.L., Twells,
R.C. & Todd, J.A. (2005). No evidence of association
or interaction between the IL4RA, IL4 and IL13 genes
in type 1 diabetes. American Journal of Human
Genetics, 76, 517–521.
Marchini, J., Donnelly, P., & Cardon, L.R. (2005).
Genome-wide strategies for detecting multiple loci
that influence complex diseases. Nature Genetics, 37,
413–417.
Matsuzaki, H., Dong, S., Loi, H., Di, X., Liu, G.,
Hubbell, E., Law, J., Berntsen, T., Chadha, M., Hui,
1038 Robert Plomin
H., Yang, G., Kennedy, G.C., Webster, T.A., Cawley,
S., Walsh, P.S., Jones, K.W., Fodor, S.P., & Mei, R.
(2004). Genotyping over 100,000 SNPs on a pair of
oligonucleotide arrays. Nature Methods, 1, 109–111.
Mattick, J.S. (2004). RNA regulation: A new genetics?
Nature Reviews Genetics, 5, 316–323.
McArdle, W.L., Twells, R.C., & Todd, J.A. (2005). No
evidence of association or interaction between the
IL4RA, IL4, and IL13 genes in type 1 diabetes.
American Journal of Human Genetics, 76, 517–521.
Milner, K.M., Craig, E.E., Thompson, R.J., Veltman,
M.W.M., Thomas, N.S., Roberts, S., Bellamy, M.,
Curran, S.R., Sporikou, C.M.J., & Bolton, P.F. (in
press). Prader-Willi Syndrome: Intellectual abilities
and behavioural features by genetic subtype. Journal
of Child Psychology and Psychiatry [in this issue].
Mukamal, K.J., Kuller, L.H., Fitzpatrick, A.L., Longst-
reth, W.T., Jr., Mittleman, M.A., & Siscovick, D.S.
(2003). Prospective study of alcohol consumption and
risk of dementia in older adults. Journal of the
American Medical Association, 289, 1405–1413.
National Alliance for Autism Research. (2005). http://
www.naar.org/news/render_pr.asp?intNewsItemID¼
176.
Norton, N., Williams, N.M., O’Donovan, M.C., & Owen,
M.J. (2004). DNA pooling as a tool for large-scale
association studies in complex traits. Annals of
Medicine, 36, 146–152.
Plomin, R., DeFries, J.C., Craig, I.W., & McGuffin, P.
(2003). Behavioral genomics. In R. Plomin, J.C.
DeFries, I.W. Craig, & P. McGuffin (Eds.), Behavioral
genetics in the postgenomic era (pp. 531–540). Wash-
ington, DC: American Psychological Association.
Plomin, R., DeFries, J.C., McClearn, G.E., & McGuffin,
P. (2001). Behavioral genetics (4th edn). New York:
Worth Publishers.
Plomin, R., & Kovas, Y. (in press). Generalist genes and
learning disabilities. Psychological Bulletin.
Plomin, R., & Rutter, M. (1998). Child development,
molecular genetics, and what to do with genes once
they are found. Child Development, 69, 1223–1242.
Podewils, L.J., Guallar, E., Kuller, L.H., Fried, L.P.,
Lopez, O.L., Carlson, M., & Lyketsos, C.G. (2005).
Physical activity, APOE genotype, and dementia risk:
Findings from the Cardiovascular Health Cognition
Study. American Journal of Epidemiology, 161, 639–
651.
Rasmussen, E.R., Neuman, R.J., Heath, A.C., Levy, F.,
Hay, D.A., & Todd, R.D. (2004). Familial clustering of
latent class and DSM-IV defined attention-deficit/
hyperactivity disorder (ADHD) subtypes. Journal of
Child Psychology and Psychiatry, 45, 589–598.
Santangelo, S.L., & Tsatsanis, K. (2005). What is known
about autism: Genes, brain, and behavior. American
Journal of Pharmacogenomics, 5, 71–92.
Sanger Centre. (2005). http://www.sanger.ac.uk/
genetics/exon/.
Sham, P., Bader, J.S., Craig, I., O’Donovan, M., &
Owen, M. (2002). DNA pooling: A tool for large-scale
association studies. Nature Review Genetics, 3, 862–
871.
Smith, S.D., Pennington, B.F., Boada, R., & Shriberg,
L.D. (in press). Linkage and speech sound disorder to
reading disability loci. Journal of Child Psychology
and Psychiatry [In this issue].
Stevenson, J., Asherson, P., Hay, D., Levy, F., Swanson,
J., Thapar, A., & Willcutt, E. (2005). Characterizing
the ADHD phenotype for genetic studies. Develop-
mental Science, 8, 115–121.
Stevenson, J., Langley, K., Pay, H., Turic, D., Payton,
A., Worthington, J., Ollier, W., Williams, J., Owen,
M.J., O’Donovan, M.C., & Thapar, A. (in press).
Attention deficit hyperactivity disorder with and
without reading disabilities: Testing for the involve-
ment of noradrenergic pathway genes. Journal of
Child Psychology and Psychiatry [in this issue].
Todd, R.D., Huang, H., Smalley, S.L., Nelson, S.F.,
Willcutt, E.G., Pennington, B., et al. (in press). Colla-
borative analysis of DRD4 and DAT genotypes in
population-defined ADHD subtypes. Journal of Child
Psychology and Psychiatry [in this issue].
van der Meulen, E.M., Bakker, S.C., Pauls, D.L.,
Oteman, N., Kruitwagen, C.L.J.J., Pearson, P.L.,
et al. (in press). High sibling correlation on methyl-
phenidate response but no association with DAT1-
10R homozygosity in Dutch sibpairs with ADHD.
Journal of Child Psychology and Psychiatry [in this
issue].
Wahlsten, D. (1990). Insensitivity of the analysis of
variance to heredity–environment interaction. Be-
havioral and Brain Sciences, 13, 109–161.
Willcutt, E.G., DeFries, J.C., Pennington, B.F., Smith,
S.D., Cardon, L.R., & Olson, R.K. (2003). Genetic
etiology of comorbid reading difficulties and ADHD. In
R. Plomin, J.C. DeFries, I.W. Craig, & P. MuGuffin
(Eds.), Behavioral genetics in the postgenomic era
(pp. 227–246). Washington, DC: American Psycholo-
gical Association.
Zondervan, K.T., & Cardon, L.R. (2004). The complex
interplay among factors that influence allelic associa-
tion. Nature Reviews Genetics, 5, 89–100.
Manuscript accepted 3 May 2005