A joint history of the nature of genetic variation and the nature of schizophrenia

Abstract

This essay traces the history of concepts of genetic variation and schizophrenia from Darwin and Mendel to the present. For Darwin, the important form of genetic variation for evolution is continuous in nature and small in effect. Biometricians led by Pearson agreed and developed statistical genetic approaches utilizing trait correlations in relatives. Mendel studied discontinuous traits and subsequent Mendelians, led by Bateson, assumed that important genetic variation was large in effect producing discontinuous phenotypes. Although biometricians studied ‘insanity’, schizophrenia genetics under Kraepelin and Rüdin utilized Mendelian approaches congruent with their anatomical-clinical disease model of dementia praecox. Fisher showed, assuming many genes of small effect, Mendelian and Biometrical models were consilient. Echoing prior conflicts, psychiatric genetics since then has utilized both biometrical models, largely in twins, and Mendelian models, based on advancing molecular techniques. In 1968, Gottesman proposed a polygenic model for schizophrenia based on a threshold version of Fisher’s theory. Since then, rigorous studies of the schizophrenia spectrum suggest that genetic risk for schizophrenia is more likely continuous than categorical. The last 5 years has seen increasingly convincing evidence from genome-wide association study (GWAS) and sequencing that genetic risk for schizophrenia is largely polygenic, and congruent with Fisher’s and Gottesman’s models. The gap between biometrical and molecular Mendelian models for schizophrenia has largely closed. The efforts to ground a categorical biomedical model of schizophrenia in Mendelian genetics have failed. The genetic risk for schizophrenia is widely distributed in human populations so that we all carry some degree of risk.

…Variations, as Darwin and most breeders recognized, were of two types. There were sports, large discontinuous variations, relatively rare … [and] there were the less obvious but more pervasive and plentiful minor variations which occurred in every character of the organism.

(Provine,¹ p. 5)

This essay sketches a joint history of the concepts of genetic variation and schizophrenia from the time of Darwin and Mendel in the mid-19th century to the present. The nature of genetic variation can be thought of both at the individual level (Are most important differences in individuals the result of single genes of large effect or many genes of small effect?) and at the population level (Is the distribution of genetic liability normal or very bumpy with a small proportion of individuals at high risk and the rest at low risk?).

THE PARADIGM OF PSYCHIATRIC ILLNESS

The origin of our current anatomical-clinical concept of schizophrenia arose in Europe in the late 19th century.² Major influences included the discovery of microbial causes of disease and the associated move toward etiological models of illness,³ the rise of neuropathology and the ability to observe directly brain lesions, and the successful century long search for the cause of general paresis of the insane.⁴ Key features of this anatomo-clinical view of psychiatric illness included: (i) the rejection of unitary models of psychosis in favor of distinct diseases; (ii) an emphasis on symptom clusters, anatomical lesions and natural history; and (iii) the faith that despite a current inability to discern pathological lesions, the disorders were still fundamentally ‘organic’ and the ‘lesions’ might, in the meantime, be defined using physiological or psychological concepts (for example, ‘irritation’ or ‘negative symptoms’).² This definition was based on a promissory note that individual psychotic disorders reflected distinct diseases which would one day be traced to discrete pathological processes.

This approach remained dominant in the 20th century as illustrated by the following from the famous German psychiatrist, Kurt Schneider:

We do not know the disease processes underlying cyclothymia [his term for manic-depressive illness] and schizophrenia. However, diseases do underlie them. The frequent hereditary nature, the fact that most of these disease run in families, the general physical changes often present … [and] above all … they destroy cohesion, interrupt the meaningful order, the continuity of meaning which characterizes normal development of life.

(Moran and Smith,⁵ p. 377)

One of Schneider’s prominent justifications for this biomedical view of psychiatric illness is the hereditary/familial nature of the disorders.

THE BIOMETRICIANS

Darwin’s theory of natural selection assumed widespread small trait variations in biological populations that impacted on fitness and were transmitted across generations. A major debate emerged in the late 19th century about whether this key assumption was correct.¹ Did evolution act slowly on quantitative variation or was it saltative, based on large changes (called ‘sports’)?

Darwin’s cousin, Galton, had a complex role in this debate. He founded the field of human behavior genetics in his book ‘Hereditary Genius’ published 10 years after Darwin’s ‘Origin’.⁶ He laid the groundwork for the ‘Biometrical approach’ to genetics by collecting data in human relatives for a range of metrical traits and encouraging similar work in other organisms. He prompted his younger and better mathematically trained associate, Pearson, to examine these data. From these efforts emerged our modern methods of regression and correlation. However, Galton disagreed with his more famous cousin about the nature of evolution, favoring the theory of saltative evolution.

Pearson, however, who emerged as the intellectual leader of the ‘Biometrical Geneticists’ in the last decade of the 19th century, was a fervent Darwinian, convinced that continuous phenotypes were the material on which evolution worked, producing slow changes over long time intervals. The proper approach to genetics, he asserted, was to examine the association between continuous phenotypes in relatives—hence, the focus of the Biometricians on their new ‘high-tech’ genetic tool: the correlation coefficient. This statistical technique assumed that the trait (and implicitly the underlying genetic variation) was approximately normally distributed.

However, Pearson soon realized that this concept could be applied to a dichotomous phenotype, assuming that it reflected a threshold imposed on an underlying normal distribution. He described this method in a 1901 essay ‘Mathematical contributions to the theory of evolution VII, on the correlation of characters not quantitatively measurable’⁷ and called it a ‘tetrachoric correlation’. While it might seem common-sense to assume that discrete traits had to be due to discrete genes, nature did not have to work that way.

In the newly established Galton laboratories, directed by Pearson, a biometrical research group investigated, in the early decades of the 20th century, a wide range of continuous and discontinuous phenotypes, calculating correlations in relatives including stature, physique, oral temperature, mental defect, fertility, tuberculosis, deafness, eye color and insanity. In 1907, David Heron published ‘A First Study of the Statistics of Insanity and the Inheritance of the Insane Diathesis’,⁸ the first biometrical analysis of psychotic illness. Heron analyzed 331 pedigrees collected from patients at the James Murray Royal Asylum. Heron examined prevalence data by various matings (both sane, one insane and both insane) and concluded that ‘no argument for Mendelism seems possible on these figures’ (Heron,⁸ p. 3). Assuming population-based rates of insanity from 1 to 2%, Heron calculated that the tetrachoric correlations in sibling pairs for the ‘insane diathesis’ ranged from +0.44 to +0.51. (The predicted sibling correlation of liability for schizophrenia from the most recent meta-analysis is +0.51 (ref. 9)). In a follow-up monograph in 1914, Heron complained that the approaches in the collection and analysis of early psychiatric genetic studies were too influenced by ‘… the bias which arises from the hurried acceptance of dogmatic theories of heredity’ (Heron,⁸ p. 356). Very likely, he was referring to the work of the early Mendelians to whose history we now turn.

THE MENDELIANS

Mendel, in trying to understand the nature of hybridization, went to the unusual step of selecting for his experiments pure bred lines that differed qualitatively. That decision was critical to his success. He was also lucky the species he chose, Pisum sativum, had many chromosomes and a simple genetics.¹⁰

In 1900, Mendel’s 1866 monograph was ‘rediscovered’ and his theory widely publicized, largely, because it fit well into the evolutionary framework of the re-discoverers. We will focus on one of them, Bateson, who became the ardent champion of Mendelism and saltative views of evolution in the Anglophonic world.^11,12 Evolution, he asserted, arose not from gradual small shifts in quantitative characteristics but from large phenotypic changes in ‘one bound’.

Given his viewpoints, it makes sense that Bateson greeted the rediscovered Mendel with enthusiasm. Here was a biological basis for his saltative evolutionary view. The key concept of the Mendelians, captured in their first ‘law’, was that the gene had a ‘unit character’ and was transmitted intact from generation to generation. Furthermore, the effect of these unit characters was large, producing distinct changes. At this early stage of Mendelism, the Mendelians assumed a one-to-one relationship between what we would now call ‘genotype’ and ‘phenotype’. Each gene affected one phenotype. Each phenotype was controlled by one gene.

THE START OF PSYCHIATRIC GENETICS

Modern psychiatric genetics began with the work of Ernst Rüdin in the first decade of the 20th century. His daughter wrote: ‘Impressed by the recent rediscovery of the Mendelian laws, Rüdin hoped that by using the correct genetic methods, he would be able to uncover the proper Mendelian ratios for schizophrenia’. (Zebrin-Rudin and Kendler,¹³ p. 334) He began the first systematic family study of psychiatric illness in 1907 using admissions with definite schizophrenia to the Munich Psychiatric University Clinic.¹⁴ This study was undertaken in the same decade that the very first of human medical traits including brachydactylism, hemophilia, night-blindness and eye color were shown to have Mendelian segregation patterns.¹²

Rüdin’s study utilized then ‘cutting edge’ methods of proband and age corrections, but included no control group. This puzzles modern readers who assume that Rüdin’s goal was to quantify familial aggregation of schizophrenia. Again, in his daughter’s words: ‘Because his initial goal was to find the Mendelian mode of transmission of schizophrenia, rather than merely to document familial aggregation, a control group was not seen as necessary’. (Zebrin-Rudin and Kendler,¹³ p. 334) Rüdin found 86 cases of schizophrenia in the siblings of 721 schizophrenic probands, who together had 1590 lifetimes at risk.¹⁵ This produced a morbid risk of 5.4 ± 0.6%. Rüdin was disappointed that this did not approximate standard Mendelian ratios. However, he concluded the monograph, suggesting a two-locus recessive model that would predict a recurrence risk in siblings of ~ 0.25² = 0.0625, close to his estimate.

Manfred Bleuler, the son of Eugen, who coined the term ‘schizophrenia’, summarized this history:

… [the] assumption that the schizophrenias could be based essentially on one single, or on a few individual mutant genes … began to take shape under the enormously convincing impression of the new discovery that came into existence from the Mendelian doctrine at the beginning of this century [and it] persuaded even so eminent a hereditary scientist as Rüdin. He declared with absolute certainty that the frequency of psychoses in families was a result of the hereditary process and that the transmission of mental disorder was bound to proceed according to a Mendelian pattern.

(Kendler and Zebrin-Rudin,¹⁶ p. 429)

Rüdin’s study noted that other psychotic and non-psychotic disorders occurred at high rates in these siblings and theorized that the genetic risk for schizophrenia could, in some individuals, remain ‘mute’.¹⁵ Writing a few years later in the 8th edition of his textbook, his mentor, Kraepelin, for the first time noted the range of ‘eccentric’ personalities seen in siblings of schizophrenic patients, theorizing that they should be seen as ‘latent’ versions of illness.^17,18

In 1917, Rüdin joined Kraepelin’s newly established German Research Institute for Psychiatry in Munich as head of the ‘Genealogic-Demographic Department’ that trained many of the next generation of European psychiatric geneticists.¹³ In the 1930s, Rüdin became involved with the National Socialists and their racial genetic policies.¹⁹ We will not here pursue this important issue further.

Why did Rüdin, in his paradigm-setting study, utilize Mendelian and not biometrical models? The Mendelian genetic model fitted into the anatomical-clinical disease model for schizophrenia to which Rüdin and Kraepelin ascribed. The thrust of German medicine then was to ground medical diagnosis in etiologic processes for which the microbe was the paradigmatic example.³ This infectious disease approach had worked spectacularly for general paresis of the insane.⁴ But the Mendelian model could also provide a discrete etiologic agent—a gene. Trying to ground the disease concept of schizophrenia on a Mendelian basis was an important research goal for Kraepelin and Rüdin, particularly because Kraepelin’s other major effort to identify a discrete etiology—in Alzheimer’s neuropathology laboratory at Kraepelin’s institute—was not succeeding.²⁰ But Kraepelin’s and Rüdin’s efforts to ground their anatomo-clinical model of schizophrenia in Mendelian genes carried a contradiction. Mendel’s segregating peas had clear phenotypic discontinuities. As honest clinical observers, both Kraepelin and Rüdin noted that relatives of schizophrenic patients could not so easily be fitted into clean ‘affected’ and ‘unaffected’ categories.

THE UNIFICATION OF BIOMETRICAL AND MENDELIAN MODELS

The early decades of the 20th century saw a pitched intellectual battle between the Biometricians, led by Pearson, and the Mendelians, led by Bateson, about the workings of evolution, the nature of genetic variation, and the proper approach to human genetic research.^1,12 Should human geneticists calculate correlation coefficients in relatives for continuous traits, or detect Mendelian segregation ratios in human pedigrees? Pearson came to recognize that Mendelism applied for a few rare ‘sports’ in animal and human populations. But the majority of important traits demonstrated blending inheritance that could be better studied by statisticians than by biologists. Furthermore, the biometricians believed that the combination and recombination of large numbers of small genetic effects could produce the cumulative phenotypic effects being inappropriately attributed to Mendelian factors.

However, cracks began to appear in the confrontational edifice constructed by both sides in this debate. Relevant now to our story is the central limit theorem which postulates that when a number of independent variables sum in their effect, the distribution of their total effect rapidly approaches a normal distribution. Unless the risk variants are very rare, a normal distribution is quickly approximated by a small number of loci.²¹

In 1902, the statistician, Yule, noted that, as predicted by the central limit theorem, many Mendelian factors working together might approximate the patterns of resemblance in relatives sought by the Biometricians.²² In the same year, with more intuitive insight, Bateson wrote:

It must be recognized that in a typically continuous character (for example … stature …) there must certainly be on any hypothesis more than one pair of possible alleomorphs. There may be many such pairs and consequently the number of the different kinds of gametes is although together unlimited [which would] … give so near an approach to a continuous curve that the purity of the elements would be unsuspected and their detection practically impossible

(Bateson and Saunders,²³ p. 59–60).

By around 1910, one common position within the genetics community was that there were two types of heredity—categorical and blending—which required different genetic theories. Blending inheritance was best studied by biometrical-statistical methods while categorical traits were better investigated using experimental or Mendelian methods.¹² Kim summarized this position as articulated in a leading book in this era on human genetics as follows:

… while Mendelism was amply confirmed in many cases of alternative [aka categorical] inheritance, it nevertheless was less capable of dealing with blending inheritance, such as mental and moral traits

(Kim,¹² p. 111).

However, evidence began to accumulate that quantitative traits, which demonstrated blending inheritance, could result from multiple ‘Mendelian factors’,²⁴ one example being East’s 1910 paper on corn and wheat titled ‘A Mendelian Interpretation of Variation that is Apparently Continuous’.²⁵ Also important in this time period were results coming from the ‘fly room’ at Columbia, where a team headed by Morgan made major advances in understanding of gene action in Drosophila.²⁶ Their findings did not well support key concepts of the ‘saltation’ model of Bateson.^26,27 Most mutations had small and often subtle effects. Large-effect mutations were rare, maladaptive and showed no evidence of producing new species. The concept of the gene as a simple ‘unit character’ with a one-to-one relationship with a phenotype was disproven. Many mutations in different genes could affect one phenotype. Individual mutants could impact on multiple traits. The gene–phenotype relationship seemed best described as ‘many to many’.

The efforts to integrate Mendelian and Biometrician theory reach a statistical culmination in the efforts of Ronald Fisher. In his 1918 paper (‘On the Correlation Between Relatives on the Supposition of Mendelian Inheritance’),^28,5 Fisher wrote:

The simple hypothesis, and the one which we shall examine, is that such [metric] features as stature are determined by a large number of Mendelian factors and that the large variance among children of the same parents is due to the [genetic] segregation of those factors in respect to which the parents are heterozygous.

(Fisher,²⁸ p. 400)

He demonstrated, as predicted by the central limit theorem, that postulating a large number of genes of small effect, additive genetic risk would be accurately described by a normal distribution that would be correlated in first-degree relatives +0.50, a figure close to the empirical results for classic metric traits, such as stature, by Galton and Pearson. Referring to this data, Fisher wrote, ‘the hypothesis of cumulative Mendelian factors seems to fit the facts very accurately’. (Fisher,²⁸ p. 433) As Provine summarizes, ‘Fisher concluded that many continuously varying characters such as human stature were primarily determined by many Mendelian factors’. (Provine,¹ p. 147)

A statistical cross-walk was now established between the Biometricians and the Mendelians. However, this did not end the struggle between the two camps about the best way to think about gene action—as a single major (or Mendelian-like) locus (SML), or as distribution of liability from many genes of small effect.

By the middle third of the 20th century, biometrical genetics was a largely British discipline with two major schools in Edinburgh²⁹ and Birmingham.³⁰ They each developed increasingly sophisticated statistical approaches to the nature of gene action using biometrical methods applied to experimental plants and animals. However, in 1965, the then major figure in the Edinburgh school, Falconer, published a paper titled ‘The Inheritance of Liability to Certain Diseases, Estimated from the Incidence among Relatives’,³¹ in which he applied the biometrical model, using an approach very similar to that articulated earlier by Pearson,⁷ to human diseases. His term ‘correlation of liability’ replaced the tetrachoric correlation. This paper served as the basis for a reconnection, after a long hiatus, between psychiatric and biometrical genetics. This reconnection resulted in the important 1967 report by Gottesman and Shields (G&S) titled ‘A Polygenic Theory of Schizophrenia’.³² They wrote:

One of the possibilities not given sufficient attention involves positing a large proportion of cases as being polygenically determined. We should like to consider the merits of treating schizophrenia as a threshold character whose appearance is predictable from a diathesis-stress model… Let us suppose that the diathesis is polygenically determined and that what is inherited is a constitutional predisposition to developing schizophrenia.

(Gottesman and Shields,³² p. 199–200)

And then added: Although schizophrenia is necessarily viewed as an all-or-none character for recordkeeping purposes, clinical contact with preschizophrenics or ‘recovered’ cases shows clearly the artificiality of such a dichotomy Falconer has now made available a model and methods for the handling of disease incidences in the relatives of probands with threshold characters, i.e., diseases that appear to have an all-or-none manifestation but are in fact determined by an underlying graduation of some attribute really causing the disease. The latter attribute has been termed liability…

(Gottesman and Shields,³² p. 200)

G&S noted the connection between genetic and diagnostic models for schizophrenia. At the level of genetic liability, Mendelian models required unambiguous boundaries between the affected and unaffected. In polygene models, by contrast, the differences are quantitative and can be quite modest. While spectrum concepts can be imposed statistically on Mendelian models, they are conceptually more consonant with polygenic models where all individuals have a quantifiable disease liability.

The rise of structural equation modeling in psychiatric genetics traces its roots to Falconer’s model that stimulated G&S, and parallel efforts from the Birmingham school. Heritabilities calculated for schizophrenia from the ACE twin models (for example,^9,33–36) assume the liability threshold model applied by Heron and Falconer, advocated by G&S and based on the mathematical foundations of Fisher.

A year later saw publication of the first rigorous demonstration of a genetic relationship between classical schizophrenia and ‘schizophrenia-like’ personality disorders from the Danish Adoption Study in 1968 (ref. 37) verifying the descriptions of the odd schizophrenia-like features seen in excess in relatives of schizophrenics from the time of Kraepelin and Bleuler.¹⁸ Genetic risk for schizophrenia was not, it appeared, confined to the classic psychotic poor-outcome disorder.

THE RISE OF MOLECULAR GENETICS

Rüdin was not alone in his Mendelian models for schizophrenia. Kallmann proposed an even simpler one-locus recessive model for the ‘schizophrenia disease complex’.³⁸ Others later proposed their own versions of single-gene models for schizophrenia including the Swedish investigator Book,³⁹ the eminent British psychiatric geneticists Slater⁴⁰ and the influential clinical psychologist Meehl.^41,42

By the 1970s, psychiatric genetics had two genetic theories for schizophrenia: a polygene model and a Mendelian SML model. The first empirical efforts to discriminate these models utilized purely statistical methods: model fitting to risk figures in relatives (for example,^43–46) and segregation analysis (for example,^47–50). Neither method was able to provide resolution of this question, although the polygene model did better.

The next method to be widely utilized—linkage analysis—the first based on molecular techniques—assumed an SML model. In the heyday of linkage analysis in schizophrenia, the talk was of finding ‘the gene for’ schizophrenia. Practically, as power analyses made clear,⁵¹ the only kind of genetic variants detectable by linkage analysis would have large effects, although not necessarily a strictly Mendelian gene.

The story of linkage analysis for schizophrenia is probably best remembered for the early high profile report on 5q in 1988 (ref. 52) followed by non-replications.^53–55 Some individual findings were replicated such as DISC1,^56,57 neuregulin^58,59 and dysbindin,^60,61 and a few signals were detected by combined samples or more formal meta-analyses.^62,63 However, almost none of these results have carried forward using the more refined methods described below.

I witnessed these times. The enthusiasm for linkage, especially when applied, as it often was to small pedigree samples, substantially exceeded what was then supportable by a dispassionate review of the available evidence.⁶⁴ This enthusiasm had many sources including excitement about the new molecular methods. However, some of this zealousness—the desire to find ‘the gene for schizophrenia’—was driven by the same biomedical disease model of schizophrenia that propelled Rüdin on his quest for Mendelian ratios 80 years earlier. This time saw an emerging division in psychiatric genetics between those applying biometrical models, typically to twin samples, and those using molecular methods, on pedigrees, to find SML. This division resembled in several ways the older debate between Pearson and Bateson. The molecular geneticists argued that their work reflected underlying biology, while those applying biometrical methods were ‘just interested in statistics, and not real genes’.

The results of linkage analysis of schizophrenia also laid to rest a different kind of genetic theory of schizophrenia—that the disorder was a common pathway for a large number of rare quasi-Mendelian disorders. Had this been the case, linkage analysis should have revealed many Mendelian-like pedigrees with linkage to various genomic regions. This, however, was not the observed pattern.

During this period, increasing evidence for a familial genetic schizophrenia spectrum emerged in new family^65–67 and adoption analyses^68–70 validating earlier clinical notions. These results undermined the idea that, from a genetic perspective, schizophrenia was a qualitatively distinct syndrome. In the words of Manfred Bleuler, this increasing acceptance of a range of milder schizophrenia associated disorders

… suggested the idea of recognizing something in the phenomenon of psychoses that was generally human [and] helped to establish … a sympathetic meeting with the mentally ill patient that was unencumbered by the hard, cold dogmas depicting the mental patient as something different, inaccessible, and beyond the reach of human empathy.

(Kendler and Zebrin-Rudin,¹⁶ p. 434)

POSITIVE EVIDENCE FOR POLYGENIC TRANSMISSION OF SCHIZOPHRENIA

While the lack of success of linkage studies could rule out common SML for schizophrenia, they provided no positive evidence for polygenes. This was to emerge from GWAS and sequencing studies.

The first positive evidence of schizophrenia polygenes appeared in a 2009 article with the striking title: ‘Common Polygenic Variation Contributes to Risk of Schizophrenia and Bipolar Disorder’.⁷¹ The abstract contained the following: ‘We provide molecular genetic evidence for a substantial polygenic component to the risk of schizophrenia involving thousands of common alleles of very small effect’. Using ~ 75 000 independent single-nucleotide polymorphisms (SNPs) from the International Schizophrenia Consortium from 3322 European individuals with schizophrenia and 3587 controls, summed scores for the top 10, 20, 30, 40 and 50% of these markers robustly distinguished schizophrenic cases from controls in several independent samples.⁷¹ Extensive simulations showed that these results were not likely artifactual and that common SNP variants tagged by the GWAS accounted for a minimum of ~ 33% of risk liability for schizophrenia (what is now called ‘SNP heritability’)—about half of the total genetic variance predicted by twin studies.^9,36

In 2012, Lee et al ⁷² applied a new statistical tool (Genome-Wide Complex Trait Analysis⁷³) to GWAS data on schizophrenia from five studies, the largest of which contained 9087 cases and 12 171 controls. Genome-Wide Complex Trait Analysis directly infers the presence of polygenes by first using the common genetic variants from GWAS to assess the degree of remote genetic relationship between individuals and then associating these calculated genetic relationships with phenotypic resemblance. Across the five studies, the SNP heritability estimates ranged from 23 to 31%. More critically, the contribution of variants to schizophrenia risk on each chromosome was strongly correlated with the length of the chromosome (r = +0.89), a pattern strongly supportive of a highly polygenic model.⁷⁴ The largest GWAS of schizophrenia published to date, utilizing 16 245 cases and 31 829 controls, applied approximate Bayesian polygenic analysis to their results. This method estimated that 8300 independent, mostly common SNPs contribute to risk for schizophrenia and these SNPs accounted for ~ 50% of the total variance in liability.⁷⁵

These studies detected statistical signals of polygenes from molecular data that derive from many variants, some of which are true and some false positive findings. However, the sample size of cases and controls contributed to the Schizophrenia group of the Psychiatric Genomics Consortium has grown dramatically over the last 2 years. In 2012, with 25 785 cases and 28 441 controls, 62 genome wide significant sites containing schizophrenia risk variants were found (Stephan Ripke for PGC SCZ, oral presentation WCPG 2012). The precise locations of these variants are known and several lines of evidence suggest that most of them are true positives. These findings address a key criticism of the statistical versions of the polygene model articulated by Fisher, Falconer and G&S—that it was not biologically actionable. Active efforts are already underway to apply various network methods to polygenes to clarify the biological pathways involved in disease risk.⁷⁶

Common polygenes are not the only molecular variants contributing to genetic risk for schizophrenia, as rare copy number and single-nucleotide variants also have a role.^77–80 The most recent evidence from exome sequencing in schizophrenia suggests many rare variants of modest or moderate effect size as indicated by the title ‘A Polygenic Burden of Rare Disruptive Mutations in Schizophrenia’.⁸⁰ These authors estimated the relative effects on risk to schizophrenia from common polygenes, rare copy number variants and rare polygenes which, scaled to 100%, equaled 91, 3 and 6%. Because they were not considering the entire genome, the authors noted that their estimate for the contribution of rare coding variation represents ‘a conservative lower bound. ‘We do not yet precisely know the relative contributions of these different sources, but current estimates suggest that at least 50%,^71,73 and perhaps as high as 90% of the genetic risk for schizophrenia comes from common polygenes.⁸⁰

CONCLUSIONS

The historical effort to ground the categorical nature of schizophrenia in genetic theory has failed. While much remains to be learned, it is now relatively clear that a large proportion of the genetic liability to schizophrenia results from hundreds if not thousands of individual risk variants both common and rare. The population distribution of genetic risk for schizophrenia is likely to be approximately normal (with some small bumps in the high liability tail). G&S were largely correct when they postulated that the model of continuous variation assumed by Darwin for his evolutionary theory, subsequently modeled and studied as correlations by the Biometricians (Pearson and Heron), and converted into a formal polygenic model by Fisher and applied to disorders by Falconer, was appropriate for schizophrenia. The early view of the biometricians that traits could be usefully divided into categorical and blending—and that the mode of genetic transmission of these two classes was distinct (that is, single major loci versus numerous polygenes)—has turned out to be relatively accurate.

Important parallels exist between the history of our ideas about the nature of genetic variation and schizophrenia. Vigorous arguments arose over a hundred years ago about whether discontinuous and continuous models of genetic variation were correct. In recent decades, we have seen a recapitulation of this debate between those who favored liability threshold versus SML models for schizophrenia. In the background has been a simmering dispute about the underlying nature of schizophrenia itself. Is it a classical organically based biomedical disorder with clean boundaries due to the effects of a single etiologic agent, or is it the severe end of a spectrum of syndromes that aggregate together in families? From a genetic perspective, is the clinical distinction between well and ill the result of an SML or an unfavorable accumulation of multiple modest genetic risk factors?

Fisher provided the intellectual resolution of the Mendelian-Biometrical debate by showing that a model of many segregating genes of small effect could explain the pattern of continuous variation and correlation in relatives seen by the Biometricians. Recent GWASs of schizophrenia have shown that for schizophrenia, Fisher’s model is largely correct. So, in a second round of consilience—a bringing together of divergent scientific perspectives—psychiatric genetics now confronts a situation where for schizophrenia we need to merge our Mendelian and Biometrical models.⁷³ That is, in the older language, the ‘statisticians’ and the ‘biologists’ meet in molecularly verified polygene models. Perhaps the debates about who are the ‘real’ geneticists will diminish.

The early clinical impressions of Kraepelin and Bleuler¹⁸ that schizophrenia spectrum disorders represented milder conditions on the same continuum of genetic risk as schizophrenia are likely correct. Indeed, this intuitive model translated into the statistical formalism of a ‘multiple threshold model’ well explains the pattern of schizophrenia spectrum disorders in families.^45,46 Utilizing GWAS-derived polygene scores, schizophrenia spectrum cases have values in between those seen for schizophrenic cases and controls.⁸¹

Our best current evidence suggests that in the large majority of instances, those with schizophrenia differ from unaffected individuals quantitatively and not qualitatively. Reductionist biomedical researchers have wanted to understand schizophrenia as arising from a single aberrant cause because it made more tractable the laudable goals of helping to clarify etiology and assist in prevention and treatment. Others, for less admirable reasons, have taken the same model and concluded that ‘the insane’ are fundamentally different from ‘us’. Using simulations, Agarwala et al⁸² recently showed that the differences in the expected number of risk alleles in affected and unaffected individuals for another complex human disorder, type II diabetes, are surprisingly modest. A large proportion of the genetic variants that form the basis of schizophrenia are common variants which constitute the gene pool of our species. Nearly, all realistic genetic models suggest that we all carry many risk variants for this disorder. Perhaps, this knowledge will help us all to deal with schizophrenia individuals ‘unencumbered by the hard, cold dogmas depicting the mental patient as something different, inaccessible…’