Neuron Perspective

New Approaches to Psychiatric Diagnostic Classification

Michael J. Owen1,* 1MRC Centre for Neuropsychiatric Genetics and Genomics, and the and Research Institute, Cardiff University, Hadyn Ellis Building, Maindy Road, Cardiff CF24 4HQ, UK *Correspondence: [email protected] http://dx.doi.org/10.1016/j.neuron.2014.10.028

Recent findings in have crystallized concerns that diagnostic categories used in the clinic map poorly onto the underlying biology. If we are to harness developments in genetics and neuro- science to understand disease mechanisms and develop new treatments, we need new approaches to patient stratification that recognize the complexity and continuous nature of psychiatric traits and that are not constrained by current categorical approaches. Recognizing this, the National Institute for Mental Health (NIMH) has developed a novel framework to encourage more research of this kind. The implications of these recent findings and funding policy developments for neuroscience research are considerable.

Introduction Psychiatric Diagnosis: Reliability at the Expense There has been no major advance in for of Validity over 40 years. Moreover, repeated and expensive failures have Current approaches to psychiatric diagnosis grew out of the resulted in many pharmaceutical companies scaling down appreciation in the early 1960s that the rate with which schizo- drug discovery in or abandoning it altogether (Abbott, phrenia was being diagnosed in the US was 5–20 times greater 2011). This is generally attributed to our poor understanding of than that in the UK (Cooper et al., 1972). Work over that decade disease mechanisms. The complexity and inaccessibility of the showed that this reflected diagnostic differences rather than real compared to other organs, and the difficulties inherent differences in prevalence and pointed to the need for diagnostic in modeling complex human systems and behaviors experimen- standardization. This led to the development of operationalized tally, certainly pose challenges. On the other hand, advances classifications that were designed to provide a reliable way of in genomics, stem cell biology, and neuroscience potentially assigning a patient with a particular constellation of signs and place us in an unprecedentedly strong position to tackle major symptoms to a diagnostic category (Allardyce et al., 2007). psychiatric disorders such as , , This works by defining a set of inclusion and exclusion criteria attention-deficit hyperactivity disorder (ADHD), and for each category. A ‘‘polythetic’’ approach is taken such that spectrum disorder (ASD) (McCarroll and Hyman, 2013). multiple positive features are identified but none is regarded as But researchers hoping to understand these disorders face essential; in order to make a diagnosis it is simply necessary to another challenge. This arises from the fact that, because we check the clinical features against the list of criteria. The main understand so little about disease mechanisms, our approach benefit of this ‘‘Chinese menu’’ approach is to greatly improve to diagnosis remains largely descriptive and syndromic; what diagnostic reliability. In other words, different clinicians should we recognize as disorders are actually syndromes: constella- make the same diagnosis of a given case assuming that they tions of signs and symptoms that tend to occur together. Recog- have been equally assiduous in eliciting the signs and symp- nition of these syndromes helps clinicians constrain the range toms. These approaches were developed to remedy a lack of of likely outcomes, choose treatments, and communicate diagnostic reliability and in this regard their introduction has with each other and with patients and their families. As far as had many benefits to clinical practice and to the collection of research is concerned, it has generally been assumed that data on prevalence and incidence (Kendell and Jablensky, current diagnoses, imperfect as we suspect them to be, are 2003). However, in the absence of a solid understanding of nevertheless the best basis we have for understanding etiology pathophysiology, the majority of diagnostic categories chosen and pathogenesis. As we acquire new knowledge from epidemi- were by necessity largely descriptive and syndromic in nature. ology, genetics, and neuroscience, so the argument goes, we They are in effect an operationalization of expert consensus of will be able to refine these categories and develop new treat- the best descriptors of the clinical syndromes recognized by ments, diagnostic tests, and biomarkers that will place psychia- ; they were not, nor were intended to be, valid de- try on a par with other branches of medicine. But recently scriptors of disease entities. The precise meaning of the concept another view has been gaining support. Perhaps our tendency of validity as applied to psychiatric diagnosis has been debated to view research findings through the primitive syndromic lens (Kendell and Jablensky, 2003), but for current purposes we can of current diagnoses is actually impeding progress. Perhaps consider it as referring to the degree to which a diagnostic we need to try new ways of classifying psychiatric disorders if construct delineates a group of cases that share common under- we are ever going to benefit from advances in neuroscience lying etiological and/or pathogenic processes (Allardyce et al., and develop more effective treatments. 2007). Applying this test, few diagnostic categories in psychiatry

564 Neuron 84, November 5, 2014 ª2014 Elsevier Inc. Neuron Perspective

would generally be accepted as valid. Many of these designate nize ‘‘interforms,’’ such as ‘‘schizoaffective disorder,’’ in which causes of intellectual disability or dementia such as Down syn- features of schizophrenia and severe mood disorder occur in drome, phenylketonuria, Huntington’s disease, and Jakob- the same individual (Heckers, 2009). A second approach is to Creutzfeldt disease (Kendell and Jablensky, 2003). In the case recognize ‘‘comorbidity,’’ whereby a patient is diagnosed with of the majority of psychiatric disorders, where etiology and path- more than one disorder. This may be clinically useful but raises ogenesis remain largely unknown, the validity of the syndromic questions about the validity of the diagnostic categories being concepts that form the basis of current diagnostic criteria is employed. Moreover, comorbidity is often obscured in research questionable (Kendell and Jablensky, 2003; Craddock and studies by the use of diagnostic hierarchies or exclusions. For Owen, 2005; Allardyce et al., 2007). example, until the most recent edition of DSM (DSM5) was pub- Most western clinicians use either the Diagnostic and Statisti- lished (American Psychiatric Association, 2013), it was not cal Manual of the American Psychiatric Association (DSM) possible to diagnose ADHD and autism in the same individual. (American Psychiatric Association, 2013) or the International Yet in population studies these two syndromes frequently co- Classification of Diseases (ICD) of the World Health Organization occur (Simonoff et al., 2008; Lee and Ousley, 2006). Finally, (International Statistical Classification of Diseases and Related not only are the boundaries between established diagnostic cat- Health Problems, 1992). These are updated every decade or egories indistinct at best, but so also are the boundaries between so but, while hopes are frequently expressed that diagnoses illness and wellness (Narrow and Kuhl, 2011). This might seem will be modified on the basis of new insights into etiology and obvious in instances such as depression and anxiety that are pathogenesis, the weight of evidence has largely been insuffi- widely recognized at different degrees of severity in many of cient to justify radical overhaul. Lack of validity was widely us. However, there is increasing realization that even psychotic recognized when these operationalized systems were devel- symptoms, such as auditory hallucinations and paranoid oped, but their convenience and reliability were such that this thinking, occur in attenuated form in the 5%–8% of the healthy significant shortcoming soon became overlooked. This has led population (van Os et al., 2009). to a process of ‘‘reification’’ whereby researchers, practitioners, and regulators have come to regard psychiatric diagnoses Impact of Recent Genetic Findings: Complexity as valid categories defining distinct illnesses with their own and Pleiotropy characteristic underlying etiology and pathogenesis (Kendell Genomic analysis of psychiatric disorders remains a work in and Jablensky, 2003; Hyman, 2010). progress; much genetic risk is still unexplained at the level of the genome, and progress has been greater for some disorders, Psychiatric Diagnoses Are Both Too Broad in particular autism and schizophrenia, than others (Sullivan and Too Narrow et al., 2012). However, empirical findings from those disorders The limitations of classifying psychiatric disorders using this with sufficient data now support a general framework for the categorical and syndromic approach have become increasingly genetic architecture of psychiatric disorders. As expected from apparent in recent years (Craddock and Owen, 2010). Indeed, genetic epidemiology and population genetics, it appears that many current diagnoses would appear to have the unusual prop- a spectrum of allelic risk underlies complex psychiatric traits erty of being simultaneously both too broad and too narrow. as for other common diseases (Sullivan et al., 2012). There are They are too broad in the sense that patients with the same contributions from alleles that are common in the population diagnosis can vary widely in symptoms, severity, course, and but whose effect sizes tend to be small due to the effects of outcome. For example, people with quite widely different symp- natural selection, as well as from rare alleles, some of which tom profiles can satisfy DSM5 criteria for schizophrenia (Amer- can have a large effect on disease risk pending their removal ican Psychiatric Association, 2013). Moreover, a diagnosis of from the population by selection (Sullivan et al., 2012; Gaugler schizophrenia is associated with a wide range of outcomes et al., 2014). The precise genetic architectures of different ranging from essentially full recovery to chronic symptoms and psychiatric disorders remains to be determined pending larger disability (van Os and Kapur, 2009). It is often assumed that, and more detailed studies. However, findings to date are suffi- with further research, it will be possible to resolve this heteroge- cient to yield two important implications for classification and neity into specific, valid disease subtypes. Yet, repeated at- for psychiatric neuroscience more widely. The first of these is tempts to do this convincingly using a variety of features have that the disorders are highly polygenic with hundreds of risk failed. Current diagnoses are too narrow because many diagno- variants involved at a population level (Sullivan et al., 2012). In ses have symptoms in common and the boundaries between schizophrenia, where the largest samples have been studied disorders are frequently indistinct and to a great extent arbitrary and our understanding is therefore greatest, genome-wide asso- as are the boundaries between disorder and wellness. For ciation studies (GWAS) have identified to date over 100 distinct example, psychotic symptoms, particularly delusions and hallu- genetic loci harboring relatively common alleles of small effect cinations, as well as episodes of affective disturbance, are at robust, ‘‘genome-wide’’ levels of statistical significance commonly seen in both schizophrenia and bipolar disorder (Schizophrenia Working Group of the Psychiatric Genomics (Craddock and Owen, 2005). The overlap of symptoms between Consortium, 2014). Furthermore, studies looking at the en masse diagnostic categories means that it is often difficult to assign an effects of common risk alleles that are not individually supported individual to a single specific category, and the diagnosis given at genome-wide levels of statistical significance have estimated to a particular patient may change over time (Craddock and that relatively common small-effect alleles account for at least Owen, 2005). One approach to this problem has been to recog- 25% of total liability to schizophrenia (about 33% of genetic

Neuron 84, November 5, 2014 ª2014 Elsevier Inc. 565 Neuron Perspective

liability) (Sullivan et al., 2012). Similar studies in autism also sup- the mounting evidence that large, rare CNVs are more prevalent port a substantial contribution of common small-effect alleles in schizophrenia than in bipolar disorder (Grozeva et al., 2010). It en masse (Lee et al., 2013). This implies that many specific loci seems likely, therefore, that, with more research, we can expect remain to be identified at robust levels of statistical significance to identify risk alleles with varying degrees of specificity. The key and implies that the so-called ‘‘missing heritability’’ may be question, as we shall see below, is not whether current diagno- merely hidden by the limitations of current technology (Sullivan ses define groups of cases that have more in common in patho- et al., 2012). This high level of genetic complexity means that genesis and pathophysiology than chance (it seems highly likely individuals, even those without a disorder, will carry a large that do), but whether these are the best phenotypes for research number of risk alleles and that levels of genetic heterogeneity aiming to chart the course between genetic risk and clinical among those affected will be high: individuals with the same outcome and best understand pathophysiology. disorder, even those who are closely related, will be likely to have a different complement of risk alleles. Where We Are Now The second important implication of recent studies is that We can draw two fairly clear conclusions about the limitations of genetic risk doesn’t map onto our current definitions of disease. current diagnostic approaches as tools for research. First, Indeed, the findings point to extensive pleiotropy with respect to recent genetic findings have crystalized longstanding concerns diagnosis. GWAS have found evidence for overlap of common (Kendell and Jablensky, 2003) about the lack of the validity of alleles at the level of individual single-nucleotide polymorphisms psychiatric classification. Current diagnoses were not designed (SNPs), genes, and the en masse effects of multiple common risk to group patients neatly into sets with the same pathogenic alleles (Sullivan et al., 2012). A recent study (Lee et al., 2013) mechanism or mechanisms, nor necessarily to distinguish found evidence for significant sharing of the relatively common groups in which different mechanisms are at play. Unfortunately, risk variants that are tagged in GWAS between schizophrenia their utility and widespread use in the clinic have come to and bipolar disorder, bipolar disorder and major depressive dis- obscure their lack of validity. The relationships between the syn- order, schizophrenia and major depressive disorder, ADHD and dromes observed in the clinic and underlying disturbances of major depressive disorder, and, to a lesser extent, schizophrenia brain function are far from simple, and we should not be sur- and ASD. prised that attempts to map current diagnostic categories Extensive pleiotropy has also been observed in the effects of onto specific disturbances in psychological and neurobiological rare risk alleles that individually confer much larger effects on function have largely failed. For example, despite extensive risk than common alleles. Chromosomal copy-number variants research over nearly two decades, has failed to (CNVs) have been found to confer risk of schizophrenia and a identify a single parameter of biomarker quality that can distin- range of childhood neurodevelopmental disorders such as guish patients with a particular mental disorder from controls ASD, intellectual disability (ID), and ADHD, as well as other phe- let alone distinguish between different mental disorders (Linden, notypes such as generalized (Owen et al., 2011; Malho- 2012). Second, the implication of very large numbers of tra and Sebat, 2012). There is also evidence for cross-disorder relatively common genetic risk alleles, many of which are genetic effects from family studies (Owen et al., 2011), as well pleiotropic, accords well with the fuzzy boundaries of current as from a recent study of small de novo mutations affecting diagnoses noted above. It suggests that we are dealing with one or a few nucleotides (Fromer et al., 2014). This latter study traits that are best conceived of as continuous rather than cat- demonstrated shared etiological overlap between schizo- egorical at both the phenotypic and genotypic levels. Attempts phrenia, ASD, and ID at the resolution not just of loci or even to identify mendelian forms of psychiatric disorders have been individual genes, but also of mutations with similar functional largely unsuccessful (Sullivan et al., 2012) and rather than impacts. It is worth noting at this point that the spectrum of dealing with a set of one-to-one relationships between genotype phenotypes associated with these rare mutations is remarkably and phenotype it would appear that we are dealing with multiple similar to the range of outcomes seen after obstetric compli- risk alleles, many of which will be associated with a variety cations and other factors such as maternal infection and poor of outcomes as determined by the combinatorial effects of prenatal nutrition, which are associated with early cerebral insult, other risk alleles, the environment, and doubtless chance. These and that have been consistently implicated as environmental risk observations lead to the conclusion that, as well as being factors for neurodevelopmental disorders (Pasamanick et al., insufficiently grounded in disease biology, a major weakness 1956). of our current approach to classification, at least as far as We have emphasized findings that suggest shared suscepti- much research is concerned, may be that it is categorical rather bility across traditional diagnostic categories. However, it is than dimensional. important to recognize that relatively nonspecific risk factors, which apply to a wide variety of cases, will be easier to identify How Can We Do Things Better? than those associated with more specific outcomes. There is Some might think it odd for a psychiatric geneticist to be calling evidence from both family and genomic studies for risk alleles for new approaches to diagnosis. After all, hasn’t genomics with differential effects on schizophrenia and bipolar disorder succeeded in identifying risk alleles despite, if not because, of (Craddock and Owen, 2010; Ruderfer et al., 2014) and also for current diagnosis? Genetics has succeeded to the extent it has alleles that have a degree of specificity to more specific clinical because its methods are highly scalable, because researchers phenotypes that are not recognized by current diagnostic cate- have collaborated to put together adequately powered studies, gories (Craddock et al., 2010). Another example comes from and because the field now demands high standards of statistical

566 Neuron 84, November 5, 2014 ª2014 Elsevier Inc. Neuron Perspective

rigor and replication (Sullivan et al., 2012). The robust identifica- RDoC is a dimensional system, which spans the range from tion of risk alleles in the face of pleiotropy and extensive hetero- normal to abnormal. It has been conceptualized as a matrix geneity has been a triumph of brute force and collectivism over with four dimensions (Cuthbert and Insel, 2013; Cuthbert, diagnostic imprecision. However, as we have noted, it is easier 2014; http://www.nimh.nih.gov/research-priorities/rdoc/index. to identify alleles with a broad spectrum of risk compared to shtml): (1) domains of functioning (e.g., negative valence sys- those associated with more circumscribed outcomes. It is likely tems, positive valence systems, and cognitive systems), which that studies over the next 5–10 years using increasingly large are further subdivided into dimensional constructs (for example, samples and new genomic technologies will identify more risk cognitive systems are divided into attention, perception, working alleles at greater degrees of certainty and this will help us impli- memory, declarative memory, language behavior, and cognitive cate biological mechanisms (McCarroll and Hyman, 2013). control); (2) units of analysis (e.g., genes, molecules, cells, However, if we wish to understand what makes patients differ circuits, behavior); (3) developmental aspects (changes to the from each other in symptoms, course, and outcome, and how constructs over time); and (4) environmental aspects (how the genetic risk impacts on specific abnormalities of brain function environment affects and interacts with the constructs). The do- leading to symptoms of disorder, we will need new ways of mains of functioning and the dimensional constructs contained stratifying patients for research. within them have been selected based on current understanding It’s relatively easy to point out the shortcomings of current of neural circuitry. The current matrix is seen as a work in prog- approaches but more difficult to identify a better system and ress the structure and content of which will be developed and even more difficult to persuade researchers to abandon the refined as more evidence accumulates (Cuthbert, 2014). comfort of familiar diagnostic habits. So while many agree that Put simply, RDoC seeks to encourage researchers to begin new approaches are needed, the way forward is more conten- with current understandings of behavior-brain relationships tious (see Editorial by Maj, 2014 and associated articles). There and then link them to clinical phenomena, rather than starting is, however, broad agreement on three principles. First, there with an illness definition and seeking its neurobiological under- should be an increasing emphasis on studies that do not stratify pinnings (Figure 1). For example, patients could be chosen for solely by current diagnostic category as defined by DSM or ICD. study on the basis that they have a particular set of symptoms For example, these might select cases that share specific clinical such as delusions and no constraints placed on their DSM symptoms or risk factors but who have different categorical diagnosis. The study might then focus on the relationship of diagnoses or select those who form a subgroup of a current reward prediction error to strength of delusional symptoms. diagnostic category as defined by particular features. Second, Alternatively, patients might be selected because they have a in acknowledgment of the continuous nature of psychiatric particular risk factor such as a pathogenic CNV or a history of phenotypes, there should be an increasing focus on dimensional childhood abuse or trauma. The study might then focus upon measures of psychopathology as well as of underlying brain delineating the range of responses to a set of cognitive para- dysfunction. Third, in relating psychopathology to underlying digms as well as clinical symptoms. mechanisms, studies should seek intermediate phenotypes A number of concerns have been raised about RDoC (Cuth- that provide measures of underlying pathophysiology and, bert, 2014; Doherty and Owen, 2014; Maj, 2014), not least of where possible, these should access established brain-behavior which is that it appears to some to be too prescriptive given relationships. These intermediate phenotypes (also known as our currently limited understanding of brain-behavior relation- endophenotypes) might involve measures of neurocognition, ships. However, its authors see it as a conceptual framework neuroimaging, or electrophysiology. to encourage new thinking and capable of rapid evolution in This philosophy has been articulated, developed, and actioned the wake of new advances rather than a rigid new model. Initial in NIMH’s Research Domain Criteria (RDoC) project (Cuthbert anxieties that RDoC is intended to replace current DSM/ICD and Insel, 2013; http://www.nimh.nih.gov/research-priorities/ approaches in the clinic and that NIMH will immediately stop rdoc/index.shtml), which seems set to be a game changer at least funding research using traditional diagnostic approaches have for psychiatric research in the U.S. (Doherty and Owen, 2014). also largely dissipated (Doherty and Owen, 2014). But this Until recently it was virtually mandatory to use either DSM remains a bold and radical response by NIMH to the shortcom- or ICD criteria in research grants and papers. Now, by implement- ings of current classification for research and it will take a decade ing RDoC, NIMH (the major government funder of psychiatric or more before its success or otherwise can be judged (Insel research in the U.S.) is attempting to reorientate U.S. research et al., 2010). away from current diagnostic categories. RDoC is built around the principle that mental disorders are best viewed as disorders Emptying the Bath but Keeping the Baby involving brain circuits. It aims to define the basic dimensions This brings me on to what is, in my view, the most important of dysfunction that cut across, or subdivide, disorders as tradi- criticism of RDoC at least in its most extreme manifestation in tionally categorized in order to develop new ways of classifying which DSM/ICD categories are off limits. This is that, in attempt- psychopathology based on domains of observable behavior ing, laudably, to break the constraints of current diagnostic and their relationship to markers of potential underlying causes approaches and put neuroscience at the center of psychiatric and mechanisms. As such it is a framework for future research research, it has discarded what is known about the etiology on psychopathology. It is certainly not, nor claimed to be, a of psychiatric syndromes and the differing degree of clinical new diagnostic system that is nearly ready for introduction to similarity between them. For example, some disorders like ID the clinic (Cuthbert, 2014). are congenital, others like autism and ADHD tend to manifest

Neuron 84, November 5, 2014 ª2014 Elsevier Inc. 567 Neuron Perspective

Figure 1. Using the RDoC Framework Possible applications of the RDoC framework (see main text for further details). RDoC has been conceptualized as a matrix with four dimensions (Cuthbert and Insel 2013; Cuthbert 2014): (1) domains of functioning, which are further subdivided into dimensional constructs; (2) units of analysis; (3) developmental aspects, and (4) environmental aspects. The domains of functioning and the dimensional constructs contained within them have been selected based on current un- derstanding of neural circuitry. The figure shows two possible scenarios within the framework. In exemplar 1, patients are chosen for study on the basis that they have delusions and no constraints placed on their DSM diagnosis. The study might then focus on the relationship of reward prediction error to strength of delusional symptoms in an fMRI paradigm. In exemplar 2, patients are selected because they have a particular pathogenic CNV. The study then focuses upon delineating the range of responses to a set of cognitive paradigms as well as clinical symptoms. in childhood, and yet others like schizophrenia and bipolar occupy a gradient with the syndromes ordered by decreasing disorder tend to present in adolescence or adulthood. Severe relative contribution of genetically and environmentally induced cognitive impairment is seen more frequently in autism than neurodevelopmental impairment, with syndromic severity repre- schizophrenia and is even less common in bipolar disorder sented orthogonally (Figure 2). This approach accepts that (Craddock and Owen, 2005; Owen et al., 2011). Increased rates current diagnostic approaches have some utility in defining of de novo CNVs and disruptive point mutations are found in ID groups of cases that are more closely related than chance. and autism but are less prominent in schizophrenia (Kirov et al., A key feature is that it regards current categorical diagnoses 2012; Girirajan et al., 2011; Fromer et al., 2014) and even less so as arbitrary divisions of what is essentially a continuous land- in bipolar disorder (Grozeva et al., 2010; Georgieva et al., 2014). scape. This landscape is certainly multidimensional, but the Schizophrenia and bipolar disorder are frequently hard to distin- model posits that the degree of neurodevelopmental impairment guish clinically from each other and typically show episodic is a recognizable dimension. As shown in Figure 2, this model courses (Craddock and Owen, 2010) whereas autism and makes predictions about the relative extent of brain dysfunction ADHD frequently occur together. This is important information, (number of structures and circuits affected) in the various clinical which can be used to inform future neuroscience studies and syndromes and the relationships and likely similarities between help us generate hypotheses and prioritize work across diag- disorders. As such, it may provide a valuable heuristic around nostic categories. It is possible to represent the relationship which clinical neuroscience and genomic studies, including between current disorder categories in a way that makes sense those using the RDoC framework, can be designed. of available etiological, clinical, and neuroscience data and that This model is certainly an oversimplification but it has already can act as a heuristic for future neuroscience-based research proved useful in predicting the relative burden of large patho- across diagnostic approaches as espoused in RDoC. This model genic CNVs in childhood neurodevelopmental and psychiatric (Craddock and Owen, 2010; Owen et al., 2011) integrates data disorders (Grozeva et al., 2010; Girirajan et al., 2011). More from a number of sources to propose that psychiatric syndromes recently, it predicted that the genes hit by de novo mutations

568 Neuron 84, November 5, 2014 ª2014 Elsevier Inc. Neuron Perspective

Figure 2. Hypothesized Relationship between Current Diagnostic Categories, Extent of Neurodevelopmental Impairment and Associated Cognitive Dysfunction, Symptoms, and Various Risk Factors This simple model integrates data from a number of sources (Pasamanick et al., 1956; Craddock and Owen, 2010; Owen et al., 2011; Kirov et al., 2012; Girirajan et al., 2011; Fromer et al., 2014; Grozeva et al., 2010; Girirajan et al., 2011) to propose that psychiatric syndromes as currently classified occupy a gradient with the syndromes ordered by decreasing relative contribution of genetically and/or environmentally induced neurodevelopmental impairment. This indexes the number of structures and circuits that are affected, which in turn is manifest by the extent and degree of associated cognitive impairment. This approach accepts that current diagnostic approaches have some utility in defining groups of cases that are more closely related than chance. A key feature is that it regards current categorical diagnoses as arbitrary divisions of what is essentially a continuous landscape. This model makes predictions about the relative extent of brain dysfunction (number of structures and circuits affected) in the various clinical syndromes and the relationships and likely similarities between disorders. In the interests of clarity, this two-dimensional representation does not show the severity of individual syndromes. These are conceived as being orthogonal and as reflecting the severity, rather than the extent, of damage to structures and circuits. Further details and references are given in the text. and the mutation sites themselves show the highest degree of and methods that are scalable and standardized across centers evolutionary conservation (a proxy measure of functional impor- as well as the assembly of data from deeply phenotyped cohorts tance) in ID, then ASD, with mutations in schizophrenia least to which many researchers can gain access (Insel et al., 2010). conserved (Fromer et al., 2014). It also predicted genetic findings The high degree of polygenicity and pleiotropy seen in psychi- for overlap between disorders at the level of biological systems atric phenotypes also poses challenges for basic neuroscien- (Grant, 2012; Fromer et al., 2014). tists seeking to study genetic risk in experimental systems. Complex genetic signals from a particular disorder, or group Conclusions and Implications for Neuroscience of disorders, can potentially be interpreted through their conver- There is increasing recognition that current diagnostic ap- gence on pathways or molecular networks (McCarroll and Hy- proaches, while retaining their utility in the clinic, do not map man, 2013), and there are encouraging signs of success for onto underlying biology and are at odds with the continuous both autism (Berg and Geschwind, 2012) and schizophrenia nature of psychiatric phenotypes. It seems clear that we cannot (Hall et al., 2014). Studies of this sort can implicate new targets continue to rely on them as the basis for understanding the for translational studies at network and pathway levels rather pathophysiology of these disorders or for developing and testing than at the level of individual genes (McCarroll and Hyman, new treatments. There is, therefore, an urgent need for research 2013), but the scope of this approach is limited by the fact that is not constrained by dogmatic adherence to DSM/ICD that the functional annotation of much of the genome and pro- categories and that brings modern neuroscience to bear on in- teome is less advanced for the brain than other tissues. We termediate phenotypes that index the pathophysiology underly- need shared resources that will provide information about tran- ing psychiatric symptoms. The degree of etiological complexity scriptomes, proteomes, and epigenomes for different brain re- revealed by psychiatric genetics suggests that the application gions and at different stages of development, as well as about of these new, neuroscience-based approaches will require large protein interactions, if we are to close this ‘‘annotation gap.’’ studies, which in many instances will require both collaboration Large-scale enterprises such as the Human Brain Project

Neuron 84, November 5, 2014 ª2014 Elsevier Inc. 569 Neuron Perspective

(https://www.humanbrainproject.eu/en_GB) offer medium-term Craddock, N., and Owen, M.J. (2010). The Kraepelinian dichotomy - going, solutions to this problem, but in the short term it should be going... but still not gone. Br. J. Psychiatry 196, 92–95. possible to assemble such data from several hundred Craddock, N., Jones, L., Jones, I.R., Kirov, G., Green, E.K., Grozeva, D., in multiple brain regions over different developmental periods. Moskvina, V., Nikolov, I., Hamshere, M.L., Vukcevic, D., et al.; Wellcome Trust Case Control Consortium (WTCCC) (2010). Strong genetic evidence for a Current translational paradigms aiming to understand the selective influence of GABAA receptors on a component of the bipolar disorder functional consequences of disease risk alleles were developed phenotype. Mol. Psychiatry 15, 146–153. to study mendelian disorders, and we are going to need to Cuthbert, B.N. (2014). The RDoC framework: facilitating transition from ICD/ find new, higher-throughput approaches if we are to translate DSM to dimensional approaches that integrate neuroscience and psychopa- findings from complex, polygenic disorders into biological and thology. World Psychiatry 13, 28–35. therapeutic insights (McCarroll and Hyman, 2013). While there Cuthbert, B.N., and Insel, T.R. (2013). Toward the future of psychiatric are few, if any, fully penetrant mendelian mutations identified diagnosis: the seven pillars of RDoC. BMC Med. 11, 126. for psychiatric disorders, the discovery of rare, highly penetrant Doherty, J.L., and Owen, M.J. (2014). The Research Domain Criteria: moving mutations in autism and schizophrenia (Sullivan et al., 2012; Mal- the goalposts to change the game. Br. J. Psychiatry 204, 171–173. hotra and Sebat, 2012; Kirov et al., 2012; Fromer et al., 2014) Fromer, M., Pocklington, A.J., Kavanagh, D.H., Williams, H.J., Dwyer, S., arguably offers the most tractable opportunities, but even here Gormley, P., Georgieva, L., Rees, E., Palta, P., Ruderfer, D.M., et al. (2014). the range of associated outcomes in humans is wide and likely De novo mutations in schizophrenia implicate synaptic networks. Nature dependent on genetic background, which may be highly poly- 506, 179–184. genic, as well as environmental factors. In the future, therefore, Gaugler, T., Klei, L., Sanders, S.J., Bodea, C.A., Goldberg, A.P., Lee, A.B., it will be necessary to study a range of variables that impact on Mahajan, M., Manaa, D., Pawitan, Y., Reichert, J., et al. (2014). Most genetic risk for autism resides with common variation. Nat. Genet. 46, 881–885. outcomes in cellular and animal models of these mutations. Moreover, it will not be possible to conclude that any specific Georgieva, L., Rees, E., Moran, J.L., Chambert, K.D., Milanova, V., Crad- dock, N., Purcell, S., Sklar, P., McCarroll, S., Holmans, P., et al. (2014). De neurobiological or behavioral outcome maps onto a specific novo CNVs in bipolar affective disorder and schizophrenia. Hum. Mol. Genet. categorical disorder or human intermediate phenotype. Thus, Published online July 23, 2014. http://dx.doi.org/10.1093/hmg/ddu379. in order to gain mechanistic insights into psychiatric disorders Girirajan, S., Brkanac, Z., Coe, B.P., Baker, C., Vives, L., Vu, T.H., Shafer, N., from even the relatively simple reagents provided by single highly Bernier, R., Ferrero, G.B., Silengo, M., et al. (2011). Relative burden of large penetrant mutations, it will be necessary to relate findings from CNVs on a range of neurodevelopmental phenotypes. PLoS Genet. 7, animal and cellular systems directly to comparable findings e1002334, http://dx.doi.org/10.1371/journal.pgen.1002334. from clinical neuroscience. Therefore, both clinical and basic Grant, S.G. (2012). Synaptopathies: diseases of the synaptome. Curr. Opin. will increasingly need to adopt translational Neurobiol. 22, 522–529. measures of brain dysfunction and, ideally, work closely together Grozeva, D., Kirov, G., Ivanov, D., Jones, I.R., Jones, L., Green, E.K., St Clair, to directly compare the effects of genetic and other risk variables D.M., Young, A.H., Ferrier, N., Farmer, A.E., et al.; Wellcome Trust Case Control Consortium (2010). Rare copy number variants: a point of rarity in on neurobiological function across levels of complexity. genetic risk for bipolar disorder and schizophrenia. Arch. Gen. Psychiatry 67, 318–327.

ACKNOWLEDGMENTS Hall, J., Trent, S., Thomas, K.L., O’Donovan, M.C., and Owen, M.J. (2014). Genetic risk for schizophrenia: convergence on synaptic pathways involved The author’s work is supported by MRC Centre (G0800509) and MRC in plasticity. Biol. Psychiatry. Published online July 18, 2014. http://dx.doi. org/10.1016/j.biopsych.2014.07.011. Programme (G0801418) Grants, a Wellcome Trust Strategic Award (503147), the European Community’s Seventh Framework Programme (HEALTH-F2- Heckers, S. (2009). Is schizoaffective disorder a useful diagnosis? Curr. 2010-241909 (Project EU-GEI)), and the European Union Seventh Framework Psychiatry Rep. 11, 332–337. Programme (FP7/2007-2013) under grant agreement n 279227. The author declares no conflict of interest related to this work. Hyman, S.E. (2010). The diagnosis of mental disorders: the problem of reifica- tion. Annu. Rev. Clin. Psychol. 6, 155–179.

REFERENCES Insel, T., Cuthbert, B., Garvey, M., Heinssen, R., Pine, D.S., Quinn, K., Sanislow, C., and Wang, P. (2010). Research domain criteria (RDoC): toward a new classification framework for research on mental disorders. Am. J. Abbott, A. (2011). Novartis to shut brain research facility. Nature 480, 161–162. Psychiatry 167, 748–751.

Allardyce, J., Gaebel, W., Zielasek, J., and van Os, J. (2007). Deconstructing International Statistical Classification of Diseases and Related Health Prob- conference February 2006: the validity of schizophrenia and alter- lems (1992). 10th Revision. (Geneva: WHO). native approaches to the classification of psychosis. Schizophr. Bull. 33, 863–867. Kendell, R., and Jablensky, A. (2003). Distinguishing between the validity and utility of psychiatric diagnoses. Am. J. Psychiatry 160, 4–12. American Psychiatric Association (2013). Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition. (Arlington, VA: American Psychiatric Kirov, G., Pocklington, A.J., Holmans, P., Ivanov, D., Ikeda, M., Ruderfer, D., Publishing). Moran, J., Chambert, K., Toncheva, D., Georgieva, L., et al. (2012). De novo CNV analysis implicates specific abnormalities of postsynaptic signalling com- Berg, J.M., and Geschwind, D.H. (2012). Autism genetics: searching for plexes in the pathogenesis of schizophrenia. Mol. Psychiatry 17, 142–153. specificity and convergence. Genome Biol. 13, 247. Lee, D.O., and Ousley, O.Y. (2006). Attention-deficit hyperactivity disorder Cooper, J.E., Kendell, R.E., Gurland, B.J., Sharpe, L., Copeland, J.R.M., and symptoms in a clinic sample of children and adolescents with pervasive Simon, R. (1972). Psychiatric Diagnosis in New York and London. Maudsley developmental disorders. J. Child Adolesc. Psychopharmacol. 16, 737–746. Monograph 20. (Oxford: Oxford University Press). Lee, S.H., Ripke, S., Neale, B.M., Faraone, S.V., Purcell, S.M., Perlis, R.H., Craddock, N., and Owen, M.J. (2005). The beginning of the end for the Mowry, B.J., Thapar, A., Goddard, M.E., Witte, J.S., et al.; Cross-Disorder Kraepelinian dichotomy. Br. J. Psychiatry 186, 364–366. Group of the Psychiatric Genomics Consortium; International Inflammatory

570 Neuron 84, November 5, 2014 ª2014 Elsevier Inc. Neuron Perspective

Bowel Disease Genetics Consortium (IIBDGC) (2013). Genetic relationship Ruderfer, D.M., Fanous, A.H., Ripke, S., McQuillin, A., Amdur, R.L., Gejman, between five psychiatric disorders estimated from genome-wide SNPs. Nat. P.V., O’Donovan, M.C., Andreassen, O.A., Djurovic, S., Hultman, C.M., et al.; Genet. 45, 984–994. Schizophrenia Working Group of the Psychiatric Genomics Consortium; Bipo- lar Disorder Working Group of the Psychiatric Genomics Consortium; Linden, D.E.J. (2012). The challenges and promise of neuroimaging in psychi- Cross-Disorder Working Group of the Psychiatric Genomics Consortium atry. Neuron 73, 8–22. (2014). Polygenic dissection of diagnosis and clinical dimensions of bipolar disorder and schizophrenia. Mol. Psychiatry 19, 1017–1024. Maj, M. (2014). Keeping an open attitude towards the RDoC project. World Psychiatry 13, 1–3. Schizophrenia Working Group of the Psychiatric Genomics Consortium (2014). Biological insights from 108 schizophrenia-associated genetic loci. Nature Malhotra, D., and Sebat, J. (2012). CNVs: harbingers of a rare variant revolution 511, 421–427. in psychiatric genetics. Cell 148, 1223–1241. Simonoff, E., Pickles, A., Charman, T., Chandler, S., Loucas, T., and Baird, G. McCarroll, S.A., and Hyman, S.E. (2013). Progress in the genetics of polygenic (2008). Psychiatric disorders in children with autism spectrum disorders: brain disorders: significant new challenges for neurobiology. Neuron 80, prevalence, comorbidity, and associated factors in a population-derived 578–587. sample. J. Am. Acad. Child Adolesc. Psychiatry 47, 921–929. Narrow, W.H., and Kuhl, E.A. (2011). Clinical significance of disorder thresholds in DSM5: the role of disability and distress. In The Conceptual Sullivan, P.F., Daly, M.J., and O’Donovan, M. (2012). Genetic architectures Evolution of DSM5, D.A. Regier, W.E. Narrow, E.A. Kuhl, and D.J. Kupfer, of psychiatric disorders: the emerging picture and its implications. Nat. Rev. eds. (Arlington, VA: American Psychiatric Publishing), pp. 147–162. Genet. 13, 537–551.

Owen, M.J., O’Donovan, M.C., Thapar, A., and Craddock, N. (2011). Neurode- van Os, J., and Kapur, S. (2009). Schizophrenia. Lancet 374, 635–645. velopmental hypothesis of schizophrenia. Br. J. Psychiatry 198, 173–175. van Os, J., Linscott, R.J., Myin-Germeys, I., Delespaul, P., and Krabbendam, Pasamanick, B., Rogers, M.E., and Lilienfeld, A.M. (1956). Pregnancy experi- L. (2009). A systematic review and meta-analysis of the psychosis continuum: ence and the development of behavior disorders in children. Am. J. Psychiatry evidence for a psychosis proneness-persistence-impairment model of psy- 112, 613–618. chotic disorder. Psychol. Med. 39, 179–195.

Neuron 84, November 5, 2014 ª2014 Elsevier Inc. 571