Molecular Psychiatry (2013), 1–9

& 2013 Macmillan Publishers Limited All rights reserved 1359-4184/13

www.nature.com/mp

EXPERT REVIEW
What psychiatric genetics has taught us about the nature of
psychiatric illness and what is left to learn
Kenneth S Kendler

Psychiatric genetics has taught us a great deal about the nature of psychiatric disorders. Traditional family, twin and adoption
studies have demonstrated the substantial role of genetic factors in their etiology, clarified the role of genetic factors in
comorbidity, elucidated development pathways, and documented the importance of gene–environment correlation and
interaction. We have also received some hard lessons when we were unable to detect replicable genes of large effect size and
found that our much-valued candidate genes did not live up to their expected promise. With more mature molecular and statistical
methods, we are entering now a different era. Statistical analyses of aggregate molecular signals are validating earlier heritability
estimates. Replicated findings from genome-wide association studies are beginning to emerge, as are discoveries of large-effect
size rare genomic variants. The number of such findings is likely to soon grow dramatically. The most pressing question facing the
field is what biological picture these results will reveal. I articulate four possible scenarios that reflect (i) no, (ii) minimal, (iii)
moderate and (iv) high biological coherence in the replicated molecular variant findings, which are soon likely to emerge. I discuss
the factors that will likely influence these patterns, including the problems of etiological heterogeneity and multiple realizability.
These findings could provide critical insights into the underlying biology of our psychiatric syndromes and potentially permit us to
perceive, ‘through a glass darkly,’ the levels of the mind–brain system that are disordered.

Molecular Psychiatry advance online publication, 30 April 2013; doi:10.1038/mp.2013.50

Keywords: GWAS; heritability; heterogeneity; levels of explanation; philosophy; psychiatric genetics

This essay reviews what psychiatric genetics has taught us about
the nature of psychiatric illness and what we may yet learn from
the new emerging findings. I organize this essay around a series of
propositions and observations with commentary. The literature is
so large that I can only refer selectively.
1. Family, twin and adoption studies (hereafter, FTA) have found
that all psychiatric disorders aggregate in families and are heritable
These results are unsurprising. Finding human traits or disorders
for which close relatives are uncorrelated and which are not
heritable is difficult.1 These findings teach us five lessons. First,
things that relatives share are important for the etiology of
psychiatric illness. Second, the radical environmentalist agenda for
psychiatric illness is misguided. Genes in aggregate have important
roles in the etiology of psychiatric disorders. Third, we are doing
something right in FTA research as the degree of replication is
usually good, probably because the effects are large.2–5 Fourth,
because genetic effects must be expressed through biological
processes, these findings suggest that psychiatric disorders are
‘biological.’ Finally, these findings justify efforts to localize
molecular genetic variants etiologic for psychiatric illness.6
These achievements are balanced by three significant limita-
tions to FTA methodology. First, classical FTA studies are
correlational and provide no insight into underlying mechanisms.
Demonstrating familial environmental effects does not clarify the
degree to which they arise from parental behavior, diet, peer
group influences or toxin exposure. Showing heritability gives no
insight into the molecular genetic mechanisms involved.
We should not feel too concerned about this. Many critical
medical findings began as correlations with unknown underlying
mechanisms, such as the associations between prior cow-pox
infections and reduction in small pox risk,7 drinking fouled water
and getting cholera,8 and cigarette smoking and lung cancer.9
Medical science often starts with correlations and the underlying
mechanisms are subsequently clarified through scientific
advances.
Second, the claim that FTA studies ‘prove’ that a disorder is
‘biological’ is weak. An astonishingly wide array of human
behaviors and traits, such as hours spent watching television,10
sports participation,11 church attendance,12 doing crossword
puzzles,13 age at first sexual intercourse14 and liking roller
coasters,13 are heritable. Turkheimer1 has suggested a first law
of behavior genetics: ‘All human behavioral traits are heritable’.
Knowing how much variation exists in the human genome, and
how intimately genetic factors are involved in the development of
the human brain, is it plausible that a human behavioral trait is
unrelated to genetic endowment? If the brain wired itself at
random with no reference to genomic instructions, this might be
defensible. But that is not how brains are made.
So if all behavior is instantiated in the brain, which is
constructed in part from a genetic program, then all behaviors
will have some genetic influence. But nothing new is learned by
claiming that a twin study showing that disorder X is heritable
now proves that it is ‘biological.’ The weakness of the FTA claim
can be clarified by contrasting it with how our science has
progressed with Mendelian disorders. Like FTA studies, initial

Virginia Institute of Psychiatric and Behavioral Genetics, and Departments of Psychiatry, and Human and Molecular Genetics, School of Medicine/Virginia Commonwealth
University, Richmond, VA, USA. Correspondence: KS Kendler, Department of Psychiatry, School of Medicine/Virginia Commonwealth University, Box 980126, Richmond,
VA 23298-0126, USA.
E-mail: kendler@vcu.edu
Received 8 December 2012; revised 19 February 2013; accepted 18 March 2013
 Psychiatric genetics and nature of psychiatric illness
KS Kendler
2
investigations of Mendelian disorders start with a statistical
concept such as a LOD score. However, they are now soon able
to learn the specific gene involved and the nature of the crippling
mutations. This is ‘strong’ biology that can lead to etiological and
therapeutic insights.
Individual psychiatric disorders are clinical-historical constructs
not pathophysiological entities. The historical traditions on which
we rely, although full of clinical wisdom, give no guarantee about
biological coherence—that underlying the clinical syndrome is a
single definable pathophysiology. The third important limitation
of FTA studies is that they provide no assurances about
neurobiological coherence. Let us define a new disorder called
LLR with three criteria: (i) left handedness, (ii) long nose and (iii)
red hair. Because handedness, nose size and hair color are all
independently influenced by genes, we would find that the LLR
syndrome is familial and heritable. We can then claim that LLR is
‘biological.’ However, LLR is not a coherent category and will have
no comprehensible unifying underlying biology.
2. Advanced genetic epidemiological methods allowed us to ask
important questions about the etiology of psychiatric disorders.
The field of FTA studies of psychiatric illness has remained
dynamic, moving from studying static heritabilities15–17 to
exploring the etiology of comorbidities, documenting
developmental changes in genetic and environmental
influences, showing how genes influence selection into high-risk
environments, and providing increasing insights into how genes
and environmental risk factors work together to produce
individuals at high or low risk to illness. These advances have
contributed substantially to our understanding of etiological
process in psychiatric illness and have not required elucidation
of the biological pathways through which risk genes operate. That
genes impact on risk for MD by influencing personality18 and on
risk for alcoholism through selection of deviant peers19 represents
important lessons even if we know nothing about the specific
genes involved.
3a. The first generation of molecular genetic studies of psychiatric
illness—linkage studies—were largely unsuccessful at producing
replicable findings.
3b. Candidate gene association studies taught us two important
lessons. First, if we do not properly control for multiple testing with
low prior probabilities, we can expect high rates of false positive
findings. Second, we knew next to nothing about the biological
mechanisms underlying psychiatric illness.
By the mid-1980s, FTA studies provided consistent evidence for
aggregate genetic effects for an increasing number of psychiatric
disorders. However, we knew nothing about how that risk was
distributed across the genome. While the patterns from FTA
studies were inconsistent with Mendelian transmission, a sub-
stantial proportion of the genetic signal could have been
concentrated in a few large-effect genes. This hypothesis was
tested by linkage studies. With a few possible exceptions (for
example, Prescott et al.,20 Lewis et al.21 and St Clair et al.22), other
findings ruled out this possibility.
Risk genes for psychiatric disorders could reside in the
neurotransmitter systems that served as the chief foci for the
early biological theories of the etiology of psychiatric disorders—
for example, dopamine for schizophrenia,23 serotonin and
norepinephrine for depression.24,25 Although still controversial,
with perhaps a few exceptions, this hope was also not realized.
These developments could be seen as nature giving the field of
psychiatric genetics two hard lessons.
4a. Early genome-wide association studies (GWAS) of psychiatric
illness were not successful at detecting replicable findings. However,
as the sample sizes began to increase, replicable findings began to
emerge.
Molecular Psychiatry (2013), 1 – 9
4b. Large deletions and duplications (aka copy number variants
(CNVs)) were beginning to be found and replicated that substantially
increased risk for psychiatric disorders, especially schizophrenia and
autism.25–27
4c. Two statistical methods were developed that assessed the
aggregate impact of variants ‘tagged’ by the common single-
nucleotide polymorphisms (SNPs) on GWA arrays.
The appeal of GWAS is its brute force nature—testing one-by-
one, with no prioritization, hundreds of thousands of common
variants. It is conceptually the opposite of candidate gene studies
that utilize pathophysiological theories to test a small number of
variants. GWAS assumes ignorance about etiology and treats every
common genomic variant the same. This allows the detection of
pathogenic variants never previously conceived of. The penalty is
its low power, due to the multiple testing burden. For psychiatry,
this bargain is beginning to pay off.
The earliest GWAS of psychiatric illness were under-powered.
Our ‘hard lessons’ continued, as our findings whittled away the
plausible parameter space wherein we might find variants
impacting on risk.28–30 One set of dramatic reproducible findings
emerged for smoking.31
In the past 2 years, replicable signals have begun to emerge
from the GWAS of some major psychiatric disorders. In the first
major report from the Psychiatric GWAS Consortium (PGC)
Schizophrenia group,32 81 of the most significant SNPs from the
primary sample (9394 cases and 12 462 controls) were typed in a
large replicate sample (8442 cases 29 839 controls). The P-value
distribution of these SNPs in the replicate sample was unlikely to
occur by chance (Po10  15), indicating that the original findings
were largely true positive results.
Similar findings emerged from the PGC bipolar group.33 Two
genome-wide significant findings emerged and replication of the 34
most significant SNPs in a large independent sample produced results
that were very unlikely to have arisen by chance (Po4  10  7).
Two statistical methods have been developed to assess
aggregate effects of DNA variation tagged by GWAS. These
methods can be best understood as molecular extensions of FTA
methodology. Instead of looking at informative relationships (for
example, monozygotic twins), they look at the aggregate effects
of genome-wide DNA variation. While FTA studies assess all
genomic variation, these methods only utilize the proportion of
variation ‘tagged’ by the available genotypes. But FTA methods
struggle to separate genetic from environmental influences on
resemblance. These newer methods have no such problem.
The first of these methods34 takes statistically independent
variants from a training case–control GWAS (NB75 000) and ranks
them from the smallest to the largest P-value. Each variant is then
weighted by the degree to which it differentiates cases from
controls. That information is then applied to a second target
sample where a risk allele at each location would be given a
positive and the protective allele a negative score. The method
was first applied to schizophrenia, generating a polygene score
algorithm from the International Schizophrenia Consortium
sample and applying it to the Molecular Genetics of Schizophrenia
study.34 When 50% of all the independent variants were used in
the polygene score, the cases from the MGS sample had a higher
score than the controls at a striking significance level:
P ¼ 2  10  28. Amidst these thousands of variants, a substantial
number of true positive findings must be providing this predictive
power. The polygene method has been used with similar results
from other samples with schizophrenia,32 with bipolar illness33,35
and between the two disorders.34,36
The second method uses a conceptual framework analogous to
FTA studies. Genome-wide complex trait analysis (GCTA)37 uses
the common genetic variants from GWAS to assess the degree of
remote genetic relationship between individuals. It then calculates
heritability on the basis of how closely these calculated genetic
& 2013 Macmillan Publishers Limited

relationships map onto phenotypic resemblance. These
heritabilities again reflect lower bound estimates because they
capture only the genetic relationships associated with the
common SNPs, whereas FTA methods capture the genetic
relationships based on the full frequency spectrum of causal
variants. Using this method, the following levels of ‘SNP-
heritability’ from a range of samples have been reported:
Schizophrenia, 23%;38,39 bipolar illness, 37–40%,40 25%;39 major
depression, 32%,41 21%;39 autism spectrum disorder, 17%,39
40–60%;42 and attention deficit hyperactivity disorder, 28%.39
Results from these two approaches have three important
implications. First, they give insight into the genetic architecture
of psychiatric disorders, suggesting that a substantial proportion of
genetic variation results from large numbers of small effect variants.
Second, as noted by Turkheimer,43 they bring closure to the long
debate about the validity of prior evidence for genetic influences on
risks for psychiatric disorders. For years, critics have claimed that
twin studies of psychiatric illness are uninterpretable because excess
resemblance of MZ vs DZ twins could arise from environmental
sources, and adoption studies were deeply flawed because
adoption agencies do not place children into adoptive families at
random.44,45 Now that we have methods to determine heritability
that rely on entirely different assumptions, the plausibility of these
objections, already frequently addressed (for example, Kendler and
Prescott17 and Kendler46) are further eroded. Third, FTA methods
are cumbersome and rely on rare relationships to infer genetic risk.
We now have the tools to assess genetic variants directly in
individuals and from that data calculate aggregate risk.
However, like FTA studies, the demonstration of genetic
variance from these two methods yields no insight into biology.43
Like FTA studies, they only tell us that, somewhere on the
genome, variants exist, which impact on disease risk. These
methods also provide no better assurance about diagnostic
coherence than FTA studies. Our LLR syndrome would have
demonstrable heritability by these methods.
INSIGHTS INTO THE ETIOLOGY OF PSYCHIATRIC DISORDERS
FROM GWAS FINDINGS
Sample sizes for GWAS, particularly those achieved through
the PGC,47 are increasing. Both statistical logic and evidence
for other complex disorders predict that we will soon have for
several major psychiatric disorders6 dozens of genome-wide
significant common variants. The first clear sign of this trend is
in results recently presented from the second phase of the
PGC schizophrenia analyses where, with 25 785 cases and 28 441
controls, 62 genome-wide significant sites containing schizo-
phrenia risk variants were found (Stephan Ripke for Psychiatric
Genetics Consortium Schizophrenia (PGCSCZ), oral presentation
20th World Congress of Psychiatric Genetics, October 14–18 2012,
Hamburg, Germany).
Pathogenic CNVs reflect major genomic events and are harder
to detect and more complex biologically than SNP variants.
Individually, they are rare and often associated with a spectrum of
neurodevelopmental anomalies. Their causal relationship with at
least schizophrenia and autism is now well established.6,48
Furthermore, sequencing studies are rapidly increasing in
sample size, and it is likely49 but not certain,50 that verified rare
risk variants will emerge from these methods. (For example, in
both inflammatory bowel and cardiovascular disease, a proportion
of genes that contain common risk variants of small effect also
contain rare variants of larger effect).51,52
In aggregate, the advances in these methods coupled with new
methods of gene network analyses may soon present the field of
psychiatry with a moment of truth. We will be able to peel back our
layers of ignorance and get our first look at the architecture of
variants in the human genome that impact on risk for the historical-
clinical syndromes around which we have organized our discipline.
& 2013 Macmillan Publishers Limited
Psychiatric genetics and nature of psychiatric illness
KS Kendler
3
After failed efforts to uncover a consistent set of neuropatho-
logical changes associated with major psychiatric disorders in the
latter half of the 19th century,53 the next influential effort to
develop biological theories for psychiatric illnesses, beginning in
the 1960s, were based on the recently discovered catecholamine
and indoleamine neurotransmitter systems.23 Repeated attempts
have been made to verify these theories including studying
candidate genes in the synthetic, degradative or receptor
components of these systems. When directly tested, the
empirical track records of these theories have, in general, been
poor.23 Both of these methods were ‘top—down,’ seeking to go
from neuropathological or neurochemical findings to etiology. We
now see early fruits of a ‘bottom–up’ approach to clarifying the
etiology of psychiatric disorders beginning with DNA variants.
Many factors will influence the pattern of results that will
emerge. Two are likely to be particularly important: the degree of
etiological heterogeneity and the biological level at which the
disorders predominantly arises.54 Heterogeneity is the more
straightforward. At one extreme, there may be dozens of
biologically distinct pathways to illness with little or no sharing
between them. At the other extreme—etiologic homogeneity—
just one pathway to illness awaits discovery.
By level of illness, I mean where the ‘causal action’ predomi-
nantly occurs that leads to disease54,55(I have previously argued
that multiple levels of etiological factors typically contribute to
psychiatric disorders.56,57 In this essay, I focus on attempts to
clarify the biological level at which sets of genetic variants
predominantly impact on illness. While it is indeed likely that
genes act at multiple levels, that complexity is largely ignored
here as are the levels of effects from environmental risk factors or
from key mediating mental processes.). Consider four hypothetical
levels: (i) biochemical, (ii) cellular, (iii) neuronal network and (iv)
brain circuit. Genes act at a biochemical level. It will be easier to
find simple and direct networks of genes containing common or
rare variants or knocked out by pathogenic CNVs if the etiology of
psychiatric disorders is biochemical (Figure 1a). If the pathophy-
siology of psychiatric illness arises through cellular processes—
such as synaptic functioning—the pathways from genes will be
more complex and interactive (Figure 1b). Detecting a clear
network from GWAS and/or sequencing data would likely be more
challenging. If the etiology of psychiatric disorders arises within
small neuronal networks, the pathways from genes to disorder will
be further complicated with more intervening steps and
possibilities for interaction (Figure 1c). Detection of coherent
gene sets will be more difficult. Finally, what if psychiatric
disorders result from dysfunctions of complex higher order brain
circuitry as postulated by Panksepp and Biven58? For example, he
suggests that major depression arises from a disordered ‘Grief’
system comprising the anterior cingulate, dorsomedial thalamus,
periaqueductal gray, ancient parts of the cerebellum and probably
the ventral septal area, dorsal preoptic area and bed nucleus of
the stria terminalis and all the inter-connecting fiber pathways.58 It
might then be very difficult to ‘see’ coherent gene networks
because of the numerous intervening steps and interactions that
occur between gene function and casual effects (Figure 1d).
Think of it as the focus on a microscope. If the etiology of illness
is in the same focal plane as our method (here GWAS and/or
sequencing), we can likely see crisp etiological pathways emerge.
But what if psychiatric illnesses arise at a cellular, network or
circuit levels? Then our image will become increasingly out of
focus. The biological coherence will be harder and harder to see
and, at some point, might disappear altogether.
POSSIBLE PATTERNS OF FINDINGS
With this background, I outline four possible scenarios of GWAS
and/or sequencing results (meant to be illustrative and not
comprehensive), which differ in the degree of biological
Molecular Psychiatry (2013), 1 – 9
 Psychiatric genetics and nature of psychiatric illness
KS Kendler
4
Figure 1. (a) The simple and direct relationship between three risk
genes found to be significantly associated with a disease in a
genome-wide association or sequencing study and the biochemical
system that causes the disease. (b) The moderately direct relation-
ship with a few intervening steps between three risk genes found to
be significantly associated with a disease in a genome-wide
association or sequencing study and the disruption in cellular
function that causes the disease. (c) The indirect relationship with a
number of intervening steps between three risk genes found to be
significantly associated with a disease in a genome-wide association
or sequencing study and the disruption in the neuronal network
that causes the disease. (d) The very indirect relationship with many
intervening steps between three risk genes found to be significantly
associated with a disease in a genome-wide association or
sequencing study and the disruption in the brain circuit that causes
the disease.
coherence that emerges. To reiterate, by coherence, I mean the
genes whose altered expression or structure is indexed by the
detected common or rare genetic variants or CNVs tell a sensible
biological story. One of the main ways that they will do that is by
forming networks of connected risk genes whose function is
disrupted by sequence changes or CNVs. While the technical
details are outside my purview, the connectivity of genes can be
defined in several ways, including (i) being in the same metabolic
pathway, (ii) acting at different stages in a biological system (for
example, a neurotransmission system that would include synthetic
and degradative enzymes, receptors and second messengers), (iii)
having coordinated expression in developmental time, particular
tissues and/or in response to specific stressors and (iv) shown to
physically interact with each other.
I first describe the ‘book-ends’ of scenarios 1 and 4 that
represent the two extremes. Then, I shall sketch out two possible
outcomes in between as scenarios 2 and 3.
Scenario 1—no coherence
As depicted in Figure 2a, where  s, circles and triangles reflect,
respectively, risk genes with common variants, rare variants and
those knocked out (or duplicated) by CNVs, this scenario predicts
that as the number of confidently identified risk genes increases
for psychiatric disorders, efforts to integrate these genes into a
coherent picture of the underlying biology of the disorder will fail.
Molecular Psychiatry (2013), 1 – 9
What we will find will be a ‘mess’ lacking even ‘islands’ of connec-
tivity reflecting subsets of variants working together to influence
disease risk. Such an outcome could arise as a result of what
philosophers have called the problem of ‘multiple realizability.’
At a biochemical level, there are lots of ways to make a neuron
hypofunction. For example, it could have under-developed dendrites,
too few or dysfunctional receptors, downregulated second-
messenger systems or deficient transport mechanisms. There are
even more biochemical pathways to make a complex neural circuit
dysfunction that involves hundreds of thousands of neurons and
many different cell types with a complex developmental history.
This scenario is similar to the one depicted in Figure 1d.
Box 1 provides two philosophical approaches to the extreme
heterogeneity that may exist for psychiatric disorders at a
biochemical level. The first, by Fodor,59 exemplifies the problems
of multiple realizability. The second, by Mackie,60 outlines an
approach to causes that is highly contextual, being neither
necessary nor sufficient.
Scenario 1 would be likely if we had, for psychiatric illness,
hundreds of distinct biochemical changes that individually
contributed to illness (while being neither necessary nor sufficient)
with independent pathways to the phenotype. Perhaps there are
too many ways for the human brain to produce the symptoms
and signs of psychiatric disorders (for example, sad mood,
auditory hallucinations, grandiosity) for us to have any chance
for biologically coherent pathways to emerge from the hundreds
or thousands of risk variants that make small contributions to risk.
This concern is reinforced by studies showing that when large sets
of gene knockouts are screened in the mouse, 20% of them
impact on a single measure of behavior: the open field test for
‘anxiety’.61 In the first 93 widely available gene knock-outs strains
in mice, 60% of them altered alcohol self-administration.62 Could
psychiatric disorders arise at such a high level within the mind–
brain system that we have no way to integrate GWAS or sequencing
findings? If so, then a genetic mess is a plausible outcome.
This story reillustrates the point made above. We cannot
assume that anything that is ‘heritable’ will be biologically
coherent. Complex behavioral traits may exist that are heritable
in FTA studies and for which we would find polygene signals from
GWAS (and perhaps sequencing) and some true positive individual
variants when our samples get big enough. But these projects will
not disclose any coherent biological pathways because discrete
biochemical or neural systems that impact on these traits do not
exist. The individual small effects scattered throughout the genome
tell no meaningful biological story. The philosophical point here
would be that genes are the ‘wrong level’ for trying to understand
the biological mechanisms that cause such a trait.
Scenario 4—high coherence
In this most optimistic scenario, depicted in Figure 2d, most or all
of the verified risk genes identified through GWAS, sequencing
and/or CNV analyses will map to a single coherent inter-connected
biological pathway. This will occur only if the genetic under-
pinnings of the disorder reflect a high degree of etiological
homogeneity. Put in another way, the individual genes would
reflect a system with a high degree of equifinality—all pointing to
a single disease process.
This could arise because psychiatric disorders are truly
biochemical disorders as illustrated in Figure 1a. Or, they could
result from disorders at a cellular or network level but in such a
way as to make their biological connections easily detectable with
our current methodology. That is, we might have a robust
equifinal model in which we can detect multiple causal routes to a
final higher level cause.
This result would be the gene network equivalent of discover-
ing a Mendelian disorder. What makes a Mendelian disorder so
potentially scientifically tractable is the ability to localize the
& 2013 Macmillan Publishers Limited
 Psychiatric genetics and nature of psychiatric illness
KS Kendler
5

Figure 2. (a) A schematic gene network analysis of risk genes found by replicated GWAS or sequencing to contain, for a given psychiatric
disorder, common risk variants (  ), rare risk variants (o) and or genes knocked out or duplicated by a CNV (D). In this figure, reflecting
scenario 1 (no coherence), no meaningful connections are observed between the genes. (b) A schematic gene network analysis of risk genes
found by replicated GWAS or sequencing to contain, for a given psychiatric disorder, common risk variants (  ), rare risk variants (o) and or
genes knocked out or duplicated by a CNV (D). In this figure, reflecting scenario 2 (minimal coherence), pockets of meaningful connections are
observed between small numbers of genes. (c) A schematic gene network analysis of risk genes found by replicated GWAS or sequencing to
contain, for a given psychiatric disorder, common risk variants (  ), rare risk variants (o) and or genes knocked out or duplicated by a CNV (D).
In this figure, reflecting scenario 3 (moderate coherence), meaningful connections are observed between two large groups of genes but these
networks are not inter-connected. (d) A schematic gene network analysis of risk genes found by replicated GWAS or sequencing to contain,
for a given psychiatric disorder, common risk variants (  ), rare risk variants (o) and or genes knocked out or duplicated by a CNV (D). In these
figures, reflecting scenario 4 (high coherence), meaningful connections are observed between nearly all the genes, which largely combined
into a single large network.

Scenario 2—minimal coherence

Box 1
Fodor59 makes a famous argument about multiple realizability
using as an example the concept of ‘money.’ Fodor argues that
there is no underlying physical essence of money. Rather,
money is multiply realized in coins of many different metallic
compounds, various kinds of paper bills, beads and electronic
signals on a hard drive in your credit card account. These are all
money but share no physical property or essence in
common.To put this into a genetics context, let us imagine a
GWAS-like effort to collect thousands of samples of money
from all over the world and submit them to physical analysis,
seeking to understand the fundamental physical essence of
money. We will fail because money is not defined in physical
terms. Increasing the grant budget and the sample size will not
help.Another philosophical approach to multiple causation was
taken by Mackie60 and its application to psychiatric illness
developed by Meehl.86 Mackie60 argued that there can be
multiple packages of sufficient causes, in which each element
in any given package is necessary, but only within that
package. This is what he called an inus notion of cause, where
inus is an acronym standing for ‘an insufficient but necessary
part of a condition that is itself unnecessary but sufficient for
the result’ (Mackie60, page 245). From this perspective,
psychiatric disorders would be seen as having many, many
inus causes.
biological abnormality to a single gene product. These results
would reflect a broadly parallel process in a disease but
happening at a higher biological level—that of a gene network.
Scenarios 1 and 4 define the extremes of possible GWAS and
sequencing results for psychiatric disorders. Most likely, results will
be somewhere in between. Scenario 2, depicted in Figure 2b,
predicts that these analyses will reveal minimal coherence with
small pockets of connectivity. The genes identified by GWAS,
sequencing and CNV analyses would form modest-sized inter-
related sets but with no meaningful connections between them.
They would not connect up to reveal major pathways to illness.
This might emerge in two particularly plausible ways. First, the
results would arise from an LLR-like syndrome. In a large case
control study of the LLR syndrome, we will expect to find clusters
of variants that underlie the three traits of being left-handed,
having a long nose and having red hair. But these clusters won’t
cohere as there is no underlying LLR biology.
Some psychiatric disorders could be like the LLR syndrome—an
arbitrary concatenation of etiologically distinct symptoms that
happen to occur within the same individual. Recent twin structural
modeling analyses of the individual DSM-IV criteria have
consistently found evidence for more than one underlying
dimension of genetic liability in four psychiatric disorders.63–66
Second, the psychiatric disorders might have no coherent
biological center but have some independent sets of risk factors
each of which are then influenced by moderately inter-connected
genes. Consider the heritable trait of ‘liking roller coasters’ that
might be influenced by three high-level traits: risk for nausea,
hedonic effects of rapid acceleration changes and thrill-seeking.
Each of these traits is in turn influenced by a moderately coherent
set of genes. But, because roller coasters are a human-constructed

& 2013 Macmillan Publishers Limited Molecular Psychiatry (2013), 1 – 9

 Psychiatric genetics and nature of psychiatric illness
KS Kendler
6
situation, there is no comprehensible underlying biology. Or, the
construct of loving roller coasters exists at such a high level within
the mind–brain system that it is currently out of reach for our
science. In large GWAS or sequencing studies of the love of roller
coasters, we would see these three gene sets turn up. But they
would not be connected, nor with our current methods, would
they point us to a coherent biological cause for love of roller
coasters. Model 2 can illustrate the implausibility of Fodor’s model
for psychiatric illness. Consider the phenotype of loving roller
coasters that we know from twin studies is moderately heritable.
So, with big enough sample sizes, we will detect some variants in
individual genes that have a phenotypic impact. But we have
strong reason to believe, from basic biology, that the pathway
from A to phenotype must involve other genes. Surely A will have
a biochemical neighbor whose function, if altered by a genetic
variant, would also contribute to the phenotype. Thus, it is
possible that we might disclose only many, many little bundles of
related genes as predicted by scenario 2, but not the complete
mess predicted by Fodor’s money model in which no gene is
related to any other (scenario 1, Box 1).
Scenario 3—moderate coherence
As seen in Figure 2c, this scenario predicts that our bioinformatic
analysis of the risk genes detected in GWAS, CNV and sequencing
Box 2
Imagine that one key mind–brain circuit (for example, the
‘grief’ system) is like a glass. The glass is constructed from many
cells, under partial genetic guidance, wiring themselves over
development, while responding to environment perturbations.
Genes plus experience construct the needed set of nested
feedback loops. Imagine that the total strength of the glass—
the resilience of the system—reflects an emergent property of
many, many genes acting together that constructs a glass that
is either stress sensitive or resistant. (This might reflect the
strong ‘polygene’ background signal we see in GWAS of most
psychiatric disorders). Further imagine that we have a machine
to deliver a precise hammer-blow (aka a critical life stressor) to
the glass such that a set percentage of them break (aka
develop illness). We will focus on those that break. Scenario 1
would arise when each glass that breaks shatters in its own
unique way. Hardly ever would two individuals ‘break down’
along the same fault lines. Recalling that because each part of
the glass would be structured by sets of genes working over
development, this will produce a ‘mess’ in a large-scale GWAS.
If this were true, we would have a very hard time getting a
single reliable set of genes, variants that relate to disease risk,
because of the high degree of multiple realizability. There are
just too many different ways for the glass to break. Scenario 4
would occur if, in all humans, the glass developed with one
deep notch in it—one weak point constructed by a small
coordinated gene network. Each time the glass was struck, if it
broke, it would (nearly) always break along the notch. In this
case, our large-scale GWAS and sequencing studies would get
a consistent strong signal from those sets of genes which
knitted the glass around the notch—that represented the
‘weak link’ in the system. Scenarios 2 and 3 would reflect
intermediate models. In scenario 2, there might be a few,
moderately stable small notches over parts of the glass.
However, most of the glass would shatter in a random manner.
In scenario 3, there would be a modest number of possible
notches with some people being weak at none, some with one
and some more than one. This would produce sufficient across-
individual consistency in a large sample study to detect the
gene networks responsible for the various notches with some
reproducibility.
Molecular Psychiatry (2013), 1 – 9
analyses would reveal a number of pockets of biological
coherence that would reflect relatively discrete and substantial
pathways contributing to disease risk. But they would not connect
up into one grand pathway.
The most likely way in which this pattern could arise is
classical biochemical genetic heterogeneity. There might be
several independent genetically influenced pathways to our major
psychiatric syndromes. Each of these pathways would produce
‘clouds’ of associated variants but they will not link up. Perhaps
these pathways produce syndromes with subtly different clinical
features but efforts to extract subtypes have failed because of too
much noise between the genetic and phenotypic levels.
A slightly different scenario is possible. Imagine that the
disorder arises from dysfunction at a high-level thalamo-cortical
circuit that is contributed to by abnormalities in any one of
multiple neuronal cell types with distinct neurotransmitter
systems and key glial support cells. These genetic systems
‘interact’ but their interaction occurs so many steps away from
the pathogenic genes that the interaction is muted and highly
variable across individuals because of stochastic developmental
process and variable environmental exposures. Some connectivity
between these pathways is there, but it is patchy and unstable
enough to be statistically unreliable.
Alternative metaphors for these four scenarios of GWAS findings
for psychiatric illness
In Figure 2a–d and (more indirectly) Figure 1a–d, I have tried to
convey diagrammatically four possible scenarios for outcomes of
GWAS, CNV and sequencing analysis for psychiatric disorders. In
Box 2 and Box 3, I take a more intuitive/metaphorical approach to
this same problem that might aid some readers.
EXAMPLES FROM OTHER COMPLEX DISORDERS
Our speculations about outcomes of GWAS signals from
psychiatric disorders can be informed by reviewing what has
been found for other complex disorders.67
Box 3
Imagine you are a senior mechanic undergoing final testing on
a Boeing 747 jet airliner. Your job is to ‘diagnose’ the problems
with the plane. You can assess the functioning of every
part on the plane but you only know that the pilot
complained that ‘the plane was not flying well.’ A 747
has six million parts, about the number of common SNPs in
the human genome (http://www.boeing.com/commercial/
747family/pf/pf_facts.html). In scenario 1, you find that after
your detailed screen uncovers 50 broken parts, these parts all
come from entirely different systems of the aircraft. You cannot
see any pattern. The broken parts do not all contain the same
kind of components, come from the same supplier or originate
from the same kind of system. You are stumped. In scenario 4,
you review the 50 broken parts. Although some reflect
electrical and other mechanical parts of the plane, it becomes
immediately clear that although spread out over several areas
of the plane and reflecting different subsystems, all of them
have an important role in one system in the airplane—the
functioning of the wing flaps. Scenarios 2 and 3 would reflect
intermediate models. In scenario 3, you might find that after
much searching, that 40 of the 50 broken parts come from
three rather distinct systems in the airplane that do not
seemed linked together. In scenario 2, you might find four
small groups of broken parts—3 to 5 parts each—that reflect
individual sub-systems. But most of the parts seem random
and cannot be meaningfully linked together in any useful way.
& 2013 Macmillan Publishers Limited

Type 2 diabetes
In a recent large-scale genomic analysis,68 the authors ask whether
‘as the number of (identified) susceptibility loci increases, there is
evidence that the pathophysiological mechanisms y coalesce
around a limited set of core pathways and networks.’ They answer
‘Our data suggest that this may be the case,’ and point to three
major disease mechanisms for type 2 diabetes emerging from
their analysis: cell-cycle regulation, adipocytokine signaling and
transcription related to the co-activator protein CREBBP.
Crohn’s disease
Searching across gene categories identified from GWAS, Holmans
et al.69 found significant enrichment of 30 gene categories, many
related to the major histocompatibility locus, immune response
and antigen processing. An analysis of individual variants
identified in Crohn’s shows many to be involved in autophagy
and innate immunity.70 An examination of GWAS data for Crohn’s
disease found a highly significant concentration of genetic
variants in 20 genes from the IL12/IL23 pathway.71
Parkinson’s disease
Both Mendelian forms and the recent variants detected from
large-scale GWAS of sporadic cases may point to a single,
complex, pathophysiological pathway involving oxidative stress
and mitochondrial dysfunction that predispose to death of critical
dopamine neurons.72
Multiple sclerosis
Of the total of 57 confirmed risk variants from GWAS, ‘the majority
of associated genes have immune-related functions y especially
cell-mediated immunity’ (Baranzini and Nickles73, page 240).
Alzheimer’s disease
Three Mendelian genes have been identified for the early onset
form and considerable evidence suggests that these three genes
all impact on the synthesis and/or degradation of amyloid
b-protein (Ab).74 Bertram and Tanzi conclude that six common
risk variants for Alzheimer’s disease can now be considered
established, with all but one (APOE) having been discovered by
GWAS.74 They write
‘y nearly all of the newly reported loci have been proposed to
be linked to Ab metabolism in one or more ways. These
metabolic effects include Ab aggregation and clearance y
(and) effects on the immune response to Ab-related toxicity
(page 90).’
Height
An analysis of 180 markers significantly associated with height in a
GWAS of over 130 000 revealed a range of detectable patterns.75
Variants in or close to genes involved in abnormal skeletal growth
were highly over-represented. Using both text-mining and gene-
set enrichment analyses, these variants were more closely
clustered together than expected by chance. GWAS variants for
height were also highly significantly concentrated in genes
expressed in the mammalian growth plate, a structure critical to
bone elongation.76
However, these disorders may be more genetically tractable
than psychiatric illnesses. Psychiatric disorders may be more
heterogeneous than other complex disorders, most of which have
at least partially understood pathophysiologies. Psychiatric
disorders remain historical-clinical syndromes of unknown etiol-
ogy. Furthermore, the mind–brain system is considerably more
complex genetically than other tissues. Brain gene expression
& 2013 Macmillan Publishers Limited
Psychiatric genetics and nature of psychiatric illness
KS Kendler
7
changes dramatically over development, and a much higher
proportion of the human genome is expressed in the brain than in
other tissues.77 The greater genetic complexity of brain-based
phenotypes is also suggested by evidence from mouse knock-out
lines that find the percentage of knockouts which influence a
single behavioral test (open field) is twice as great as those
influencing classical traits such as plasma cholesterol and blood
urea nitrogen, and four times as many as influence blood glucose
or hemoglobin levels.61 The greater complexity of the mind–brain
system may instantiate highly emergent traits that are more remote
from individual gene effects than those seen in other tissues.
We also ought to briefly consider what hints are available from
gene network analyses of psychiatric disorders. These are sketchy
because the number of replicated variants yet published is so
small and the methods so new and uncertain that risk of false
positive findings is high. For example, in the PGC GWAS analysis of
schizophrenia, five of the seven genome-wide significant findings
were potentially related to micro-RNA 137).32 Jia et al.78 utilized an
integrative analysis of multiple GWAS data sets in which they
overlaid association signals onto a protein–protein interaction
database. These identified potential risk genes, which they
replicated in an independent sample and then submitted to
network analyses. Ten potential pathways were identified that
were enriched for known neuronal-relevant functions, including
CREB and calcium signaling, and synaptic long-term depression.
Using convergent functional genomics applied to GWAS data for
schizophrenia as well as human and rodent gene expression data
bases, Ayalew et al.79 identified another set of potential pathways,
including cell-adhesion, myelin-related, glutamate receptor and
G-coupled receptor signaling. O’Dushlaine et al.80 used a
molecular pathway analysis applied to nominally significant
variants in GWAS data from schizophrenia and found evidence
for a cell adhesion molecular pathway. In an analysis of GWAS data
from bipolar illness, Holmans et al.69 found over-representation of
variants in a variety of gene ontology pathways, in particular those
reflecting hormonal activity, RNA splicing and macroautophagy.
Meta-analyses of GWAS data from autism reported that many of
the identified variants were in genes involved in transcriptional
regulation, specifically chromatin-related proteins active during
brain development81 and from major depression supports the
etiological role of glutamatergic transmission.82 Pathway analysis
applied to GWAS data from schizophrenia, bipolar illness, major
depression, autism spectrum disorder, and attention deficit
hyperactivity disorder suggested a cross-disorder etiological role
for calcium channel signaling variants.83 Network and pathway
analyses are proceeding in psychiatry and may reveal coherent
biological processes contributing to our illnesses but the evidence
is limited to date.
Finally, other tools are emerging that might provide insight into
disease pathways from GWAS data. For example, Maurano et al.84
showed selective enrichment of GWAS disease-associated variants
within DNA hypersensitivity sites of specific cell types. Enriched
signals were seen from immune cells for both Crohn’s disease and
multiple sclerosis. We can also now examine spatiotemporal
correlations in gene expression in the human brain77,85 as yet
another way to understand how diverse genes might work
together in the development and function of neural systems.
COMMENT ON THE FOUR SCENARIOS FOR GWAS AND
SEQUENCING RESULTS
In the context of these findings, what useful can be said about our
four scenarios for the outcome of GWAS and sequencing findings
from psychiatric disorders? These scenarios have very different
implications for the underlying nature of our major psychiatric
disorders and the availability of targets for drug development.
Although advocated by Turkheimer43 in a thoughtful essay, I
consider scenario 1 to be improbable. Although psychiatric
Molecular Psychiatry (2013), 1 – 9
 Psychiatric genetics and nature of psychiatric illness
KS Kendler
8
disorders are likely etiological heterogeneous, and some of the
coherence may be emergent at levels in the mind–brain system
too high for us to now study, the chances that no meaningfully
connectivity between variants will emerge seems low. Scenario 4
is also improbable. As insights into the etiology of our psychiatric
syndromes have, over their history, been so difficult to come by, it
seems unlikely that risk genes for these conditions would display a
high level of coherence at a network level. A scenario 4 outcome
would be a boon for drug development, as it should yield a rich
harvest of potentially drugable targets that might prove to have
broad efficacy. My guess is that results will point to the space
occupied by hypotheses 2 and 3, and the shades of gray in between.
Of course, different psychiatric disorders might fall at different places
along this continuum. Furthermore, as science advances, our ability
to detect coherence will improve and may move us, for particular
disorders, from lower to higher numbered scenarios.
CONCLUSION
Psychiatric genetics has taught us a great deal about the nature of
psychiatric disorders but provided painful lessons. We know that
familial and genetic factors make an important contribution to the
etiology of nearly all psychiatric disorders. Yet, despite our wishing
so, individual gene variants of large effect appear to have a small
to non-existent role in the etiology of major psychiatric disorders.
We have clarified the role of genetic factors in comorbidity,
elucidated development pathways and documented the impor-
tance of gene—environment correlation and interaction but our
valued candidate genes produced quite limited insights into the
nature of genetic risk for illness. While important in a number of
ways, our heritability estimates provide no insight into underlying
biological process and no guarantee that the syndrome is
biologically coherent.
With more mature molecular and statistical methods, we are
entering a new era now. Statistical tools applied to aggregate
genetic variants from GWAS have shown that prior estimates of
heritability from FTA studies could not have resulted from flaws in
their methodology. Most excitingly, replicated GWAS signals,
which we expect to increase substantially in coming years, have
the potential, along with CNVs, to provide insights into the
underlying biology of our disorders, their level of heterogeneity,
and their biological coherence. We even have a chance to
perceive, ‘through a glass darkly,’ the levels of the mind–brain
system that are disordered in our syndromes.
What should we expect to see? The most pessimistic prediction
that we will observe only a mess is unlikely. But discovering a
highly coherent single pathway to illness also seems improbable.
We can hope that the heterogeneity is not too great and the real
level of illness not too hidden in the upper reaches of the mind–
brain system. If this is true, important insights into the nature of
psychiatric illness are likely to await our efforts.
CONFLICT OF INTEREST
The authors declare no conflict of interest.
ACKNOWLEDGEMENTS
This paper benefited greatly from discussions of this material with and/or comments
on early drafts from Eric Turkheimer, Mark Reimers, John Campbell, Naomi Wray,
Jordan Smoller and Jonathan Flint.
REFERENCES
1 Turkheimer E. Three Laws of Behavior Genetics and what they mean. Curr Dir
Psychol Sci 2001; 9: 160–164.
2 Sullivan PF, Neale MC, Kendler KS. Genetic epidemiology of major depression:
review and meta-analysis. Am J Psychiatry 2000; 157: 1552–1562.
Molecular Psychiatry (2013), 1 – 9
3 Sullivan PF, Kendler KS, Neale MC. Schizophrenia as a complex trait: evidence
from a meta-analysis of twin studies. Arch Gen Psychiatry 2003; 60: 1187–1192.
4 Verweij KJ, Zietsch BP, Lynskey MT, Medland SE, Neale MC, Martin NG et al.
Genetic and environmental influences on cannabis use initiation and problematic
use: a meta-analysis of twin studies. Addiction 2010; 105: 417–430.
5 Prescott CA. The genetic epidemiology of alcoholism: sex differences and future
directions. In: Agarwal DP, Seitz HK (eds) Alcohol in Health and Disease. Marcel
Dekker, Inc: New York, pp 125–149, 2001.
6 Sullivan PF, Daly MJ, O’Donovan M. Genetic architectures of psychiatric disorders:
the emerging picture and its implications. Nat Rev Genet 2012; 13: 537–551.
7 Jenner E. An Inquiry into the Causes and Effects of the Variolae Vaccinae, a Disease
Discovered in Some of the Western Counties of England Particularly Gloucestershire,
and Known by the Name of the Cow Pox. S. Low: London, UK, 1798.
8 Snow J. On The Mode of Communication of Cholera. John Churchill, New Bur-
lington Street, England: London, UK, 1855.
9 Doll R, HILL AB. Smoking and carcinoma of the lung; preliminary report. Br Med J
1950; 2: 739–748.
10 Plomin R, Corley R, DeFries JC, Fulker DW. Individual differences in television
viewing in early childhood: nature as well as nurture. Psychol Sci 1990; 1: 371–377.
11 Maia JA, Thomis M, Beunen G. Genetic factors in physical activity levels: a twin
study. Am J Prev Med 2002; 23(2 Suppl): 87–91.
12 Kendler KS, Myers JA. Developmental twin study of church attendance and
alcohol and nicotine consumption: a model for analyzing the changing impact of
genes and environment. Am J Psychiatry 2009; 166: 1150–1155.
13 Olson JM, Vernon PA, Harris JA, Jang KL. The heritability of attitudes: a study of
twins. J Person Soc Psychol 2001; 80: 845–860.
14 Waldron M, Heath AC, Turkheimer E, Emery R, Bucholz KK, Madden PAF et al. Age
at first sexual intercourse and teenage pregnancy in Australian female twins. Twin
Res Hum Genet 2007; 10: 440–449.
15 Kendler KS. Twin studies of psychiatric illness. Current status and future directions.
Arch Gen Psychiatry 1993; 50: 905–915.
16 Kendler KS. Twin studies of psychiatric illness: an update. Arch Gen Psychiatry
2001; 58: 1005–1014.
17 Kendler KS, Prescott CA. Genes, Environment, and Psychopathology: Understanding
the Causes of Psychiatric and Substance Use Disorders. 1st edn (Guilford Press:
New York, 2006).
18 Kendler KS, Gatz M, Gardner C, Pedersen NL. Personality and major depression: a
Swedish longitudinal, population-based twin study. Arch Gen Psychiatry 2006; 63:
1113–1120.
19 Kendler KS, Gardner CO, Prescott CA. Toward a comprehensive developmental
model for alcohol use disorders in men. Twin Res Hum Genet 2011; 14: 1–15.
20 Prescott CA, Sullivan PF, Kuo PH, Webb BT, Vittum J, Patterson DG et al. Geno-
mewide linkage study in the Irish affected sib pair study of alcohol dependence:
evidence for a susceptibility region for symptoms of alcohol dependence on
chromosome 4. Mol Psychiatry 2006; 11: 603–611.
21 Lewis CM, Levinson DF, Wise LH, DeLisi LE, Straub RE, Hovatta I et al. Genome scan
meta-analysis of schizophrenia and bipolar disorder, part II: Schizophrenia. Am J
Hum Genet 2003; 73: 34–48.
22 St Clair D, Blackwood D, Muir W, Carothers A, Walker M, Spowart G et al. Asso-
ciation within a family of a balanced autosomal translocation with major mental
illness. Lancet 1990; 336: 13–16.
23 Kendler KS, Schaffner KF. The Dopamine hypothesis of Schizophrenia: an histor-
ical and philosophical analysis. Philos Psychiatry Psychol 2011; 18: 41–63.
24 Hirschfeld RMA. History and evolution of the monoamine hypothesis of depres-
sion. J Clin Psychiatry 2000; 61(suppl 6): 4–6.
25 Walsh T, Mcclellan JM, Mccarthy SE, Addington AM, Pierce SB, Cooper GM et al.
Rare structural variants disrupt multiple genes in neurodevelopmental pathways
in schizophrenia. Science 2008; 320: 539–543.
26 Stone JL, O’Donovan MC, Gurling H, Kirov GK, Blackwood DHR, Corvin A et al. Rare
chromosomal deletions and duplications increase risk of schizophrenia. Nature
2008; 455: 237–241.
27 Marshall CR, Scherer SW. Detection and characterization of copy number variation
in autism spectrum disorder. Methods Mol Biol 2012; 838: 115–135.
28 Lencz T, Morgan TV, Athanasiou M, Dain B, Reed CR, Kane JM et al. Converging
evidence for a pseudoautosomal cytokine receptor gene locus in schizophrenia.
Mol Psychiatry 2007; 12: 572–580.
29 Shifman S, Johannesson M, Bronstein M, Chen SX, Collier DA, Craddock NJ et al.
Genome-wide association identifies a common variant in the reelin gene that
increases the risk of schizophrenia only in women. PLoS Genet 2008; 4: e28.
30 Kirov G, Zaharieva I, Georgieva L, Moskvina V, Nikolov I, Cichon S et al. A genome-
wide association study in 574 schizophrenia trios using DNA pooling. Mol Psy-
chiatry 2009; 14: 796–803.
31 Tobacco and Genetics Consortium. Genome-wide meta-analyses identify multiple
loci associated with smoking behavior. Nat Genet 2010; 42: 441–447.
& 2013 Macmillan Publishers Limited

32 Ripke S, Sanders AR, Kendler KS, Levinson DF, Sklar P, Holmans PA et al. Genome-wide
association study identifies five new schizophrenia loci. Nat Genet 2011; 43: 969–976.
33 Psychiatric GWAS Consortium Bipolar Disorder Working Group. Large-scale
genome-wide association analysis of bipolar disorder identifies a new
susceptibility locus near ODZ4. Nat Genet 2011; 43: 977–983.
34 Purcell SM, Wray NR, Stone JL, Visscher PM, O’Donovan MC, Sullivan PF et al.
Common polygenic variation contributes to risk of schizophrenia and bipolar
disorder. Nature 2009; 460: 748–752.
35 Whalley HC, Papmeyer M, Sprooten E, Romaniuk L, Blackwood DH, Glahn DC et al.
The influence of polygenic risk for bipolar disorder on neural activation assessed
using fMRI. Transl Psychiatry 2012; 2: e130.
36 Hamshere ML, O’Donovan MC, Jones IR, Jones L, Kirov G, Green EK et al. Polygenic
dissection of the bipolar phenotype. Br J Psychiatry 2011; 198: 284–288.
37 Yang J, Lee SH, Goddard ME, Visscher PM. GCTA: a tool for genome-wide complex
trait analysis. Am J Human Genet 2011; 88: 76–82.
38 Lee SH, DeCandia TR, Ripke S, Yang J, Sullivan PF, Goddard ME et al. Estimating
the proportion of variation in susceptibility to schizophrenia captured by com-
mon SNPs. Nat Genet 2012; 44: 247–250.
39 Lee SH, Ripke S, Neale BM, Faraone SV, Purcell SM et al. Cross-Disorder Group of
the Psychiatric Genomics Consortium(PGC-CDG). Genetic relationship between
five psychiatric disorders estimated from genome-wide SNPs. Nature Genetics
2013 (under review).
40 Lee SH, Wray NR, Goddard ME, Visscher PM. Estimating missing heritability
for disease from genome-wide association studies. Am J Human Genet 2011; 88:
294–305.
41 Lubke GH, Hottenga JJ, Walters R, Laurin C, de Geus EJ, Willemsen G et al.
Estimating the genetic variance of major depressive disorder due to all single
nucleotide polymorphisms. Biol Psychiatry 2012; 72: 707–709.
42 Klei L, Sanders SJ, Murtha MT, Hus V, Lowe JK, Willsey AJ et al. Common genetic
variants, acting additively, are a major source of risk for autism. Mol Autism 2012;
3: 9.
43 Turkheimer E. Genome wide association studies of behavior are social science. In:
Plaisance KS, Reydon TAC (eds). Philosophy of Behavioral Biology: Boston Studies in the
Philosophy of Science. Springer Science & Busines Media B.V, 2012, pp 43–64.
44 Lewontin RC, Rose S, Kamin LJ. Not in Our Genes: Biology, Ideology, and Human
Nature. Pantheon, 1985.
45 Joseph J. The Missing Gene: Psychiatry, Heredity, And the Fruitless Search for Genes.
Algora Publishing, 2006.
46 Kendler KS, Sundquist K, Ohlsson H, Palmer K, Maes H, Winkleby MA et al. Genetic
and familial-environmental influences on risk for drug abuse: A National Swedish
Adoption Study. Arch Gen Psychiatry 2012; 69: 690–697.
47 Cichon S, Craddock N, Daly M, Faraone SV, Gejman PV, Kelsoe J et al. Genomewide
association studies: history, rationale, and prospects for psychiatric disorders. Am J
Psychiatry 2009; 166: 540–556.
48 Levinson DF, Duan J, Oh S, Wang K, Sanders AR, Shi J et al. Copy number variants
in schizophrenia: confirmation of five previous findings and new evidence for
3q29 microdeletions and VIPR2 duplications. Am J Psychiatry 2011; 168: 302–316.
49 Xu B, Roos JL, Dexheimer P, Boone B, Plummer B, Levy S et al. Exome sequencing
supports a de novo mutational paradigm for schizophrenia. Nat Genet 2011; 43:
864–868.
50 Need AC, McEvoy JP, Gennarelli M, Heinzen EL, Ge D, Maia JM et al. Exome
sequencing followed by large-scale genotyping suggests a limited role for
moderately rare risk factors of strong effect in schizophrenia. Am J Human Genet
2012; 91: 303–312.
51 Johansen CT, Wang J, Lanktree MB, Cao H, McIntyre AD, Ban MR et al. Excess of
rare variants in genes identified by genome-wide association study of hyper-
triglyceridemia. Nat Genet 2010; 42: 684–687.
52 Rivas MA, Beaudoin M, Gardet A, Stevens C, Sharma Y, Zhang CK et al. Deep
resequencing of GWAS loci identifies independent rare variants associated with
inflammatory bowel disease. Nat Genet 2011; 43: 1066–1073.
53 Engstrom EJ. Clinical Psychiatry in Imperial Germany: A History of Psychiatric
Practice. Cornell University Press: Ithaca, NY, 2003.
54 Carandini M. From circuits to behavior: a bridge too far? Nat Neurosci 2012; 15:
507–509.
55 Anderson PW. More is different. Science 1972; 177: 393–396.
56 Kendler KS. Levels of explanation in psychiatric and substance use disorders:
implications for the development of an etiologically based nosology. Mol
Psychiatry 2012; 17: 11–21.
57 Kendler KS. The dappled nature of causes of psychiatric illness: replacing the
organic-functional/hardware-software dichotomy with empirically based plural-
ism. Mol Psychiatry 2012; 17: 377–388.
58 Panksepp J, Biven L. The Archaeology of Mind: Neuroevolutionary Origins of Human
Emotions. First ed. W.W. Norton & Company, Inc: New York, NY, 2012.
& 2013 Macmillan Publishers Limited
Psychiatric genetics and nature of psychiatric illness
KS Kendler
9
59 Fodor JA. Special sciences (OR: The disunity of science as a working hypothesis).
Synthese 1974; 28: 97–115.
60 Mackie JL. I. Causes and Conditions. Am Phil Q 1965; 2: 245–264.
61 Flint J, Mott R. Applying mouse complex-trait resources to behavioural genetics.
Nature 2008; 456: 724–727.
62 Crabbe JC, Phillips TJ, Harris RA, Arends MA, Koob GF. Alcohol-related genes:
contributions from studies with genetically engineered mice. Addict Biol 2006; 11:
195–269.
63 Kendler KS, Aggen SH, Patrick CJ. A multivariate twin study of the DSM-IV criteria
for antisocial personality disorder. Biol Psychiatry 2012; 71: 247–253.
64 Kendler KS, Aggen SH, Prescott CA, Crabbe J, Neale MC. Evidence for multiple
genetic factors underlying the DSM-IV criteria for alcohol dependence. Mol
Psychiatry 2011; 17: 1306–1315.
65 Kendler KS, Aggen SH, Patrick CJ. Familial influences on conduct disorder reflect
two genetic and one shared environmental factor. Arch Gen Psychiatry 2012; 70:
78–86.
66 Kendler KS, Aggen SH, Neale MC. Evidence for multiple genetic factors
underlying DSM-IV criteria for major depression. Arch Gen Psychiatry 2012; 17:
1306–1315.
67 Wang K, Li M, Hakonarson H. Analysing biological pathways in genome-wide
association studies. Nat Rev Genet 2010; 11: 843–854.
68 Morris AP, Voight BF, Teslovich TM, Ferreira T, Segre AV, Steinthorsdottir V et al.
Large-scale association analysis provides insights into the genetic architecture
and pathophysiology of type 2 diabetes. Nat Genet 2012; 44: 981–990.
69 Holmans P, Green EK, Pahwa JS, Ferreira MA, Purcell SM, Sklar P et al. Gene
ontology analysis of GWA study data sets provides insights into the biology of
bipolar disorder. Am J Human Genet 2009; 85: 13–24.
70 Parkes M. Evidence from genetics for a role of autophagy and innate immunity in
IBD pathogenesis. Digest Dis 2012; 30: 330–333.
71 Wang K, Zhang H, Kugathasan S, Annese V, Bradfield JP, Russell RK et al. Diverse
genome-wide association studies associate the IL12/IL23 pathway with Crohn
Disease. Am J Human Genet 2009; 84: 399–405.
72 Martin I, Dawson VL, Dawson TM. Recent advances in the genetics of Parkinson’s
disease. Ann Rev Genom Human Genet 2011; 12: 301–325.
73 Baranzini SE, Nickles D. Genetics of multiple sclerosis: swimming in an ocean of
data. Curr Opin Neurol 2012; 25: 239–245.
74 Bertram L, Tanzi RE. The genetics of Alzheimer’s disease. Progr Mol Biol Transl Sci
2012; 107: 79–100.
75 Lango AH, Estrada K, Lettre G, Berndt SI, Weedon MN, Rivadeneira F et al.
Hundreds of variants clustered in genomic loci and biological pathways affect
human height. Nature 2010; 467: 832–838.
76 Lui JC, Nilsson O, Chan Y, Palmer CD, Andrade AC, Hirschhorn JN et al.
Synthesizing genome-wide association studies and expression microarray reveals
novel genes that act in the human growth plate to modulate height. Hum Mol
Genet 2012; 21: 5193–5201.
77 Kang HJ, Kawasawa YI, Cheng F, Zhu Y, Xu XM, Li MF et al. Spatio-temporal
transcriptome of the human brain. Nature 2011; 478: 483–489.
78 Jia P, Wang L, Fanous AH, Pato CN, Edwards TL, Zhao Z. Network-assisted
investigation of combined causal signals from genome-wide association studies
in schizophrenia. PLoS Comput Biol 2012; 8: e1002587.
79 Ayalew M, Le-Niculescu H, Levey DF, Jain N, Changala B, Patel SD et al. Con-
vergent functional genomics of schizophrenia: from comprehensive under-
standing to genetic risk prediction. Mol Psychiatry 2012; 17: 887–905.
80 O’Dushlaine C, Kenny E, Heron E, Donohoe G, Gill M, Morris D et al. Molecular
pathways involved in neuronal cell adhesion and membrane scaffolding con-
tribute to schizophrenia and bipolar disorder susceptibility. Mol Psychiatry 2011;
16: 286–292.
81 Ben-David E, Shifman S. Combined analysis of exome sequencing points toward a
major role for transcription regulation during brain development in autism. Mol
Psychiatry advance online publication, 13 November 2012.
82 Lee PH, Perlis RH, Jung JY, Byrne EM, Rueckert E, Siburian R et al. Multi-locus
genome-wide association analysis supports the role of glutamatergic synaptic
transmission in the etiology of major depressive disorder. Transl Psychiatry 2012;
2: e184.
83 Cross-Disorder Group of the Psychiatric GWAS Consortium. Genome-wide analysis
identifies loci with shared effects on five major psychiatric disorders. Lancet
advance online publication, 28 February 2013 (in press).
84 Maurano MT, Humbert R, Rynes E, Thurman RE, Haugen E, Wang H et al.
Systematic localization of common disease-associated variation in regulatory
DNA. Science 2012; 337: 1190–1195.
85 State MW, Sestan N. The emerging biology of autism spectrum disorders. Science
2012; 337: 1301–1303.
86 Meehl PE. Specific etiology and other forms of strong influence: some quantita-
tive meanings. J Med Philos 1977; 2: 33–55.
Molecular Psychiatry (2013), 1 – 9