Genetic Variation in Schizophrenia Liability is Shared With Intellectual Ability and Brain Structure

Abstract

Background:

Alterations in intellectual ability and brain structure are important genetic markers for schizophrenia liability. How variations in these phenotypes interact with variance in schizophrenia liability due to genetic or environmental factors is an area of active investigation. Studying these genetic markers using a multivariate twin modeling approach can provide novel leads for (genetic) pathways of schizophrenia development.

Methods:

In a sample of 70 twins discordant for schizophrenia and 130 healthy control twins, structural equation modeling was applied to quantify unique contributions of genetic and environmental factors on human brain structure (cortical thickness, cortical surface and global white matter fractional anisotropy [FA]), intellectual ability and schizophrenia liability.

Results:

In total, up to 28.1% of the genetic variance (22.8% of total variance) in schizophrenia liability was shared with intelligence quotient (IQ), global-FA, cortical thickness, and cortical surface. The strongest contributor was IQ, sharing on average 16.4% of the genetic variance in schizophrenia liability, followed by cortical thickness (6.3%), global-FA (4.7%) and cortical surface (0.5%). Furthermore, we found that up to 57.4% of the variation due to environmental factors (4.6% of total variance) in schizophrenia was shared with IQ (34.2%) and cortical surface (13.4%).

Conclusions:

Intellectual ability, FA and cortical thickness show significant and independent shared genetic variance with schizophrenia liability. This suggests that measuring brain-imaging phenotypes helps explain genetic variance in schizophrenia liability that is not captured by variation in IQ.

Introduction

Schizophrenia is an etiologically complex disorder^1,2 with genetic variation^3–6 and environmental risk factors^7,8 playing an important role in its development. A central problem in schizophrenia research is that the clinical phenotype may reflect biologically heterogeneous subgroups. Identifying behavioral and biological markers that show a genetic overlap with schizophrenia liability (SZ) is a possible way forward in defining mechanisms for schizophrenia development. Indeed, various measures of cognitive ability and brain structure are reported to be important genetic markers for SZ.^9–15

Cognitive deficits constitute a core symptom of schizophrenia,¹⁶ with general intelligence already affected prior to onset of psychotic symptoms,¹⁷ declining after the first psychotic episode,¹⁸ and lacking normal gain in global cognitive abilities over time.^19,20 The intelligence quotient (IQ) is one of the most heritable cognitive phenotypes, with ~80% of the variance explained through additive genetics.²¹ Using a discordant twin design, it was shown that schizophrenia and IQ share a partially overlapping genetic background.^22,23 Furthermore, brain volume, cortical thickness (CT), and white matter connectivity are all reduced in schizophrenia.^24–26 These aspects of brain structure are highly heritable in the general population^27–36 and show genetic overlap with SZ.^37–42

Importantly, brain structure and intellectual ability are positively correlated^43,44 and this correlation is strongly influenced by genetic factors.⁴⁵ How variation in structural brain measures and intellectual ability genetically interacts with variance in SZ is an area of active investigation. Recently, a consortium of schizophrenia twin studies found that genetic variation in cognition explained ~25% of the total variance in SZ, and that variance in SZ explains ~4% of the variance in brain volume.¹⁴ However, brain volume represents only one component of brain structure and is likely influenced by several genetic factors. For example, CT and surface are under influence of independent genetic factors⁴⁶ and can be considered as separate measures in imaging genetics studies.⁴⁷ Furthermore, changes in CT and structural brain connectivity have been identified as independent genetic contributors to liability for schizophrenia.^15,40 This raises the question how these different aspects of brain structure genetically interact with SZ and whether this interaction occurs direct or through variation in intellectual ability.

In this study, we examined the interactions between intellectual ability (IQ), fractional anisotropy (FA), CT, cortical surface (CS), and SZ. The central research questions were: which amount of the genetic variance in SZ is shared with cognitive and brain phenotypes and is this shared variance overlapping or independent between these cognitive and brain phenotypes? A multivariate Cholesky-decomposition was applied to data from 70 twins discordant for schizophrenia and 130 control twins, to estimate the genetic associations of IQ, FA, CT, and CS with SZ.

Methods and Materials

Sample Description

The U-TWIN cohort¹⁵ consists of 12 monozygotic (MZ) discordant twin pairs (9 male pairs/ 3 female pairs), 2 MZ concordant twin pairs (2 male pairs), 2 MZ co-twins (non-affected member of a twin pair; 2 female), 19 discordant dizygotic (DZ) twin pairs (10 male pairs, 4 female pairs, 5 opposite sex pairs), 2 DZ twin patients (2 male), 72 MZ control twins (17 male pairs/ 19 female pairs), and 58 DZ control twins (9 male pairs / 13 female pairs / 5 opposite sex pairs / 4 male single twins). For a complete overview of the sample demographics see table 1 and supplementary methods.

Table 1.

Sample Demographics

Sample Demographics
	MZ Twins			DZ Twins
Sample (N = 200)	Discordant MZ	Concordant MZ	Control MZ	Discordant DZ	Concordant DZ	Control DZ
Complete pairs	12	2	36	19	0	27
Incomplete pairs	2	0	0	2	0	4
Characteristics per group	Patient	Co-twin	Control	Patient	Co-twin	Control
N	16	14	72	21	19	58
Sex M/F (p male)	13/3 (0.81)	9/5 (0.64)	34/38 (0.47)	13/8 (0.62)	14/5 (0.74)	27/31 (0.47)
Age (SD)	35.3 (12.9)	36.1 (14.0)	36.7 (14.1)	37.6 (11.2)	37.9 (12.2)	35.4 (13.4)
Handedness right/non-right (p right)	14/2 (0.87)	12/2 (0.86)	66/6 (0.92)	20/1 (0.95)	16/3 (0.84)	11/47 (0.81)
Years of education (SD)	14.0 (2.6)	12.8 (2.8)	14.5 (2.0)	12.5 (2.6)	14.2 (1.6)	14.4 (2.3)
IQ	92.5 (13.2)	95.8 (13.0)	105.1 (12.1)	91.1 (13.9)	110.4 (14.5)	104.7 (14.4)
Age at onset (SD)	22.5 (5.9)			19.6 (4.7)
Duration of illness (SD)	12.8 (10.2)			18.2 (11.6)
Type AP (typical/atypical/ both/none)	(1/14/0/1)			(2/14/4/1)
Chlorpromazine equivalent (mg/d)
Typical	600 (—)			450 (212)
Atypical	528 (398)			489 (360)
Both	—			818 (311)
PANSS	50.9 (22.2)			45.4 (13.3)

Note: AP, antipsychotic; MZ, monozygotic; DZ, dizygotic; PANSS, Positive and Negative Syndrome Scale; IQ, intelligence quotient. The following variables are expressed as proportion (p): Sex: p male, Handedness: p right. Continuous variables are followed by the SD.

Measurement of IQ

An evaluation of intellectual ability was obtained by means of a shortened version of the WAIS III general intelligence test, consisting of 5 subtests: Digit Symbol-Coding, Block Design, Arithmetic, Digit Span, and Information. All 5 subtests were used to calculate a proxy measure for the full-scale IQ, which will be referred to as IQ from now on. In 16 subjects, we were unable to obtain an IQ score (2 MZ control twins, 3 MZ patients, 2 MZ co-twins, 4 DZ control twins, 1 DZ patient, 4 DZ co-twins).

Brain Image Acquisition and Processing

Magnetic resonance imaging (MRI) scans were acquired on a Philips Achieva scanner operating at 3 Tesla, using an 8-channel SENSE head-coil. A T1-weighted 3D fast-field echo scan was acquired from each participant. Scan acquisition was performed using the following parameters: 220 0.8mm contiguous slices; echo time (TE) 4.6ms; repetition time (TR) 10ms; flip angle 8°; in-plane voxel size 0.75×0.75mm.² CS and thickness of the whole brain were calculated using the automated segmentation pipeline of the Freesurfer 5.1.0 package.⁴⁸ In 1 MZ co-twin the MRI scans had to be discarded as an incidental neurological finding was reported. Two patients (DZ) and 2 healthy controls (DZ) did not complete the MRI part of the experiment.

The diffusion-weighted scan consisted of a single shot Echo Planar Imaging- Diffusion Tensor Imaging (EPI-DTI) with 30 diffusion-weighted volumes (b = 1000s/mm²) with non-collinear gradient directions and 5 diffusion-unweighted volumes (b = 0s/mm²), TR/TE = 7035/68ms, field of view (FOV) 240mm, matrix 128/128, 75 slices at 2mm thickness, no gap, SENSE factor 3, no cardiac gating. Two DWI datasets were acquired in the transverse plane per subject using the same parameters but with reverse k-space readout, allowing for correction of weighted imaging artifacts and increasing signal to noise ratio.

Preprocessing of the diffusion-weighted scans was performed with the diffusion toolbox of Andersson et al^49,50 and in-house developed software.⁵¹ First, susceptibility artifacts were corrected by calculating a distortion map based on the 2 b=0 images acquired with reversed k-space readout. Subsequently it was applied to the 2 sets of 30 direction-weighted images. This resulted in a corrected DWI set consisting of a single b=0 image and 30 corrected weighted images, thereby avoiding the need for nonlinear registration approaches to the T1-weighted images.⁵⁰ The diffusion-weighted set was corrected for Eddy-current distortions and small head movements by realigning all scans to the diffusion-unweighted image.⁴⁹

A measure of global FA was obtained by averaging the FA-values over all voxels within the white matter skeleton (FA > 0.2), as obtained by TBSS analysis (FSL v4.1.7). In 1 MZ patient and 1 MZ co-twin we were unable to obtain a full DTI dataset, 2 MZ control twins were excluded from the FA analysis as outliers (average FA > 4 SD from mean).¹⁵

Preparing for Model Fitting

Regression analysis was done to control for the effects of sex, handedness, and age on the imaging data, in R.⁵² The unstandardized residuals of the regressions were then used as input for the genetic model. All input variables were normalized to have a mean of 0 and a SD of 1. To prepare for genetic model fitting, the dichotomous variable “disease status” was assumed to represent an underlying continuous liability with a mean of 0 and SD of 1. A patient will have a high value on the liability scale, thereby crossing the disease threshold (patient status = 1). All other individuals will have lower liability scores and will not cross the threshold (discordant co-twin of patient or control twin pairs, patient status = 0). Using OpenMx software⁵³ as implemented in R,⁵² the ordinal and continuous variables could be combined within the same model. The critical threshold and heritability for the underlying liability for schizophrenia was not based on this sample because participant inclusion was not population based. Instead, the prevalence and heritability of schizophrenia were fixed to population values; prevalence was set to 1%, and heritability was set to 81%. Influences of shared environment on SZ were set to 11% and influences of unique environment were set to 8%.^3,41,54

Genetic Model Fitting

Genetic model fitting consisted of a multivariate Cholesky decomposition with additive genetic (A) and unique environmental (E) influences for IQ and brain measures. Shared environmental influences (C) were only entered as a fixed value for SZ (see figure 1 for full model). For the other measures, univariate analyses confirmed that shared environmental influences did not play a significant role. The twin model was implemented using structural equation modeling (SEM) applied in OpenMx,⁵³ in which the contributions of A and E to trait variation and covariation are estimated by maximum likelihood. SEM was implemented in OpenMx software in R.

Fig. 1.

Multivariate model representation. Rectangular boxes represent the observed variables: cortical surface (CS), intelligence quotient (IQ), fractional anisotropy (FA), cortical thickness (CT) and schizophrenia liability (SZ). Circles in shades of red represent latent genetic factors, the arrows (a₁₁…a₅₅) represent the genetic factor loadings on the observed variables. Circles in shades of blue represent the latent unique environmental factors, the arrows (e₁₁…e₅₅) represent the factor loadings on the observed variables. The green circle represents the latent common environmental factor loading on SZ. The heritability of SZ was fixed, thus squared factor loadings $a_{15}^2\dots a_{55}^2$ add up to 81%. Squared factor loadings $e_{15}^2\dots e_{55}^2$ add up to 8% and $c_{55}^2$ was constrained at 11%. Fat arrows indicate significant paths, dashed arrows represent paths that were not significant. The numbers next to the paths represent the factor loadings.

Calculation of phenotypic associations (r _ph) was based on within-twin/cross-trait correlations. Calculation of heritability (h ²) of cognitive and brain measures was based on between-twin/within-trait correlations within MZ and DZ groups. Genetic (r _g) and environmental (r _e) correlations between connectivity and disease liability were disentangled based on the polychoric correlations cross-twin/cross-trait within MZ and DZ groups. The extent to which the association between schizophrenia and other phenotypes was mediated through genes was expressed by calculating relative proportions of the r _g and r _e to the phenotypic correlation. For IQ, this results in: $r_{ph-a}=\surd \left(h_{SZ}^2\right)*r_g*\surd \left(h_{IQ}^2\right)$ and $r_{ph-e}=\surd \left(h_{SZ}^2\right)*r_e*\surd \left(h_{IQ}^2\right)$, and likewise for the other measures. The r _ph-a/ r _ph-e are very informative measures as they take into account the genetic correlation between traits in proportion to the heritability of the trait. A high genetic/ environmental correlation can be observed even when heritability of traits is very low, but in that case it will only explain a small amount of variance.^15,23

The amount of variance independently shared between each cognitive/brain phenotype with SZ was estimated using the squared genetic pathways (a_15…a₅₅; figure 1). As they were constrained to add up to a fixed heritability for schizophrenia, their relative contributions can be expressed as percentages of genetic variance. The same was done for the environmental pathways (e_15…e₅₅; figure 1).

Around all a_ij and e_ij paths, as well as the r _ph, r _ph-a, and r _ph-e, 95% CIs were fitted to calculate significance. As the Cholesky decomposition model is not symmetrical, the order in which the variables are inserted can have an effect on the path estimates. Therefore, we permuted the multivariate analysis with every possible ordering of the IQ, FA, CT, and CS variables in positions 1 through 4. SZ was always kept at position 5 so that the fixed heritability could be “partitioned” into 5 shares, one unique to the trait and 4 with each brain and cognitive phenotype. The results of shared genetic/environmental variance are thus expressing that eg, an x% of the heritability of SZ is shared with genes for IQ, not that genes for IQ explain the heritability of SZ.

Results

Variance in SZ Shared With Brain and Cognitive Phenotypes

The model ordering of CS, IQ, FA, CT, and SZ provided the largest proportion of shared genetic variance with SZ at 28.1%. In this model, 57.4% of variance due to environmental factors was shared with variation in cognitive and brain measures.

Given the fixed parameter estimates on schizophrenia, this means that respectively 22.8% (genetic component) and 4.6% (environmental component) of the total variance in SZ were shared with these 4 factors. For heritability estimates, as well as the phenotypic, genetic and environmental correlations between all variables see table 2 and figure 1.

Table 2.

Multivariate Modeling

Multivariate Modeling
Heritability estimates
	CS	IQ	FA	CT
h 2	0.85 (0.77 to 0.90)	0.82 (0.72 to 0.89)	0.61 (0.41 to 0.75)	0.75 (0.62 to 0.84)
e 2	0.15 (0.23 to 0.10)	0.18 (0.11 to 0.28)	0.39 (0.25 to 0.59)	0.25 (0.16 to 0.38)
Phenotypic associations between variables
	CS	IQ	FA	CT	SZ
CS		—	—	—	—
IQ	0.13 (−0.03 to 0.29)		—	—	—
FA	0.18 (0.02 to 0.32)	0.10 (−0.07 to 0.25)		—	—
CT	−0.11 (−0.26 to 0.05)	0.07 (−0.09 to 0.23)	0.25 (0.10 to 0.39)		—
SZ	−0.12 (−0.26 to 0.03)	−0.42 (−0.54 to −0.27)	−0.24 (−0.38 to −0.09)	−0.25 (−0.39 to −0.11)	—
Genetic/ environmental associations between variables
	CS	IQ	FA	CT	SZ
CS	—	0.01 (−0.04 to 0.06)	0.01 (−0.06 to 0.09)	0.01 (−0.05 to 0.06)	−0.04 (−0.08 to −0.004)
IQ	0.13 (−0.04 to 0.29)	—	0.05 (−0.02 to 0.14)	0.03 (−0.02 to 0.11)	−0.07 (−0.12 to −0.02)
FA	0.17 (0.003 to 0.32)	0.05 (−0.13 to 0.22)	—	0.13 (0.05 to 0.24)	−0.06 (−0.13 to 0.02)
CT	−0.12 (−0.27 to 0.05)	0.04 (−0.13 to 0.21)	0.11 (−0.06 to 0.28)	—	−0.04 (−0.09 to 0.01)
SZ	−0.07 (−0.22 to 0.08)	−0.34 (−0.48 to −0.20)	−0.18 (−0.33 to −0.02)	−0.22 (−0.35 to −0.06)	—
Path estimates
Genetic paths	CS (ai,1)	IQ (ai,2)	FA (ai,3)	CT (ai,4)	SZ (ai,5)
CS (a1,j)	0.90 (0.78 to 1.03)	—	—	—	—
IQ (a2,j)	0.13 (−0.04 to 0.31)	0.87 (0.75 to 1.01)	—	—	—
FA (a3,j)	0.18 (0.01 to 0.35)	0.02 (−0.17 to 0.21)	0.75 (0.58 to 0.90)	—	—
CT (a4,j)	−0.12 (−0.30 to 0.06)	0.06 (−0.13 to 0.24)	0.17 (−0.04 to 0.38)	0.81 (0.69 to 0.95)	—
SZ (a5,j)	−0.08 (−0.24 to 0.08)	−0.37 (−0.52 to −0.21)	−0.21 (−0.41 to −0.004)	−0.20 (−0.37 to −0.01)	0.76 (0.65 to 0.84)
Environmental paths	CS (ei,1)	IQ (ei,2)	FA (ei,3)	CT (ei,4)	SZ (ei,5)
CS (e1,j)	0.38 (0.32 to 0.46)	—	—	—	—
IQ (e2,j)	0.02 (−0.10 to 0.13)	0.41 (0.34 to 0.51)	—	—	—
FA (e3,j)	0.03 (−0.15 to 0.21)	0.12 (−0.06 to 0.30)	0.60 (0.49 to 0.73)	—	—
CT (e4,j)	0.02 (−0.11 to 0.15)	0.08 (−0.06 to 0.22)	0.19 (0.07 to 0.33)	0.43 (0.36 to 0.53)	—
SZ (e5,j)	−0.11 (−0.23 to −0.01)	−0.17 (−0.25 to −0.04)	−0.06 (−0.17 to 0.06)	−0.02 (−0.12 to 0.09)	0.18 (0.10 to 0.26)

Note: This is the model output when the variables are ordered as follows: cortical surface (CS), intelligence quotient (IQ), fractional anisotropy (FA), cortical thickness (CT), schizophrenia liability (SZ), other permutations of the model may yield slightly different estimates. The upper panel shows the heritability estimates for the cognitive and brain measures (95% CI). The second panel shows the phenotypic (r _ph) correlations between variables, significant correlations are in bold. The third panel shows the genetic (r _ph-a, lower triangle) and environmental (r _ph-e, upper triangle) correlations. The bottom panel shows the genetic (a_i,j) and environmental (e_i,j) path estimates.

Model Permutations

After averaging the outcome of all permutations, IQ, FA and CT and CS together, 27.8% (ranging from 27.7% to 28.1%) of the genetic variance and 57.4% (ranging from 57.2% to 57.5%) of the variance due to environmental components in SZ was shared with these phenotypes.

The shared “genetic” variance in SZ was similar for all 24 permutations of the variables (range): IQ = 16.4% (15.1%–17.7%), FA = 4.7% (3.1%–6.5%), CT = 6.3% (4.8%–6.5%), and CS = 0.5% (0.3%–1.7%). The IQ and CT factor remained significant in all possible permutations, the FA factor remained significant in 12/24 permutations, and the CS factor never reached significance.

The shared “environmental” variance in SZ was also similar for all 24 permutations of the variables (range): IQ = 34.2% (30.2%–39.5%), FA = 6.1% (0.6%–11.5%), CT = 3.6% (0.6%–7.5%), and CS = 13.4% (12.2%–15.0%). The IQ and CS factor remained significant in all permutations, the FA and CT factor never reached significance.

Discussion

In this study we disentangled the shared genetic and environmental variance between structural brain, cognitive phenotypes and SZ. We report that about 28% of the genetic variance (given the fixed schizophrenia heritability of 81%, this constitutes 22.8% of the total variance) in schizophrenia was shared with sources from cognition and brain structure. On average 16.4% of the genetic variance in SZ was shared with IQ. On average 11.5% in total was shared with brain structure, which could be separated in CT (6.3%), global-FA (4.7%), and CS (0.5%). Variation in FA and CT was directly associated with SZ independent of variation in IQ. Furthermore, we found that up to 57.4% of the variation due to environmental factors in schizophrenia was shared with imaging and cognitive phenotypes (given the fixed environmental component at 8%, this constitutes 4.6% of the total variance): significant factors were IQ (34.2%) and total CS area (13.4%).

The shared genetic variance between IQ and SZ was the largest compared to other phenotypes measured; fitting with the hypothesis that cognitive deficit is a core symptom of schizophrenia.¹⁶ Furthermore, 34% of the environmental variance in SZ was shared with IQ. This finding is consistent with reports that IQ is a heritable trait,²¹ and with intellectual ability declining after the onset of psychosis.¹⁸ However, due to the relatively long duration of illness in our patient sample we cannot speculate on the timing of this environmentally mediated decline of IQ after psychosis onset. As recent evidence shows cognitive decline may not occur during the first psychotic episode,⁵⁵ our finding could be related to prolonged disease duration, as well as early environmental factors influencing IQ and schizophrenia risk.

We previously showed in this dataset that the association between global-FA and SZ is almost entirely explained though shared genes.¹⁵ Also, that a substantial part of the genetic variance was accounted for by CT, independent to the variance accounted for by global-FA. These findings suggest that different aspects of brain structure may be markers for independent genetic pathways for schizophrenia development.^15,56 Possibly, separate genetic factors contribute via independent pathways to risk at the neuronal level (CT)⁴⁷ vs risk at the level of supportive tissue (white matter mean FA).⁵⁷ Our current results suggest that structural brain imaging phenotypes share genetic variance with SZ that is not accounted for by variation in intellectual ability only.

We did not find a significant overlap between variation in IQ and variation in structural brain measures. This suggests that the genetic variance shared between SZ and brain measures is largely independent of that shared with IQ in this twin sample discordant for schizophrenia. This adds to the recent finding that the largest proportion of shared genetic variance between brain volume and SZ is through a direct interaction and not through an indirect interaction shared with intellectual ability based on reciprocal causation modeling.¹⁴ Testing hypotheses about the underlying direction of causation between global white matter FA, CT, intellectual ability, and SZ may be a promising way forward as the Cholesky-model is not suitable for making assumptions about causation.⁵⁸

CS and SZ did not show a significant genetic correlation. However a substantial proportion of variance was shared through unique environmental factors. This may indicate that disease-specific factors account for the observed reduction in CS in schizophrenia. CS is genetically closer to volumetric measurements, compared to CT.⁴⁷ This result may thus be comparable to earlier findings that grey matter volume is reduced in schizophrenia patients, compared to both their co-twins and healthy controls, suggesting that disease related factors influence this decrease.³⁸

Recently, it was shown that common genetic variants are implicated in subcortical and intracranial volumes.^59,60 Our results suggest that CT and FA are also strong phenotypes for genetic association studies that attempt to link common variants to brain structure. Based on our results, it could be predicted that each phenotype is related to a different set of common genetic variants, and that this may result in independent overlap with genetic loci for schizophrenia.^5,6 Our results further suggest that the genetic overlap between schizophrenia and intelligence^14,23,61 is not significantly overlapping with brain imaging phenotypes currently studied. Such differential overlap with SZ may provide important clues regarding the modeling of heterogeneous biological pathways of schizophrenia development.^15,56

Aside from associations with SZ, brain, and cognitive phenotypes also showed several associations among each other. We show that global-FA and CS are phenotypically correlated irrespective of disease, and that this correlation is likely to be driven by shared genes. Indeed, in the healthy individuals of this cohort, global-FA and white matter volume are correlated.³⁶ However, it was estimated that both traits contribute independently to SZ, suggesting that their overlap with SZ is due to independent pathways.

CS showed the highest correlation of all brain measures with intellectual ability, irrespective of disease (r _ph = .13 [−0.04 to 0.31]). Although this association was not significant, 94% was due to shared genes. This is consistent with previous reports that intellectual ability and brain volume are genetically correlated.⁴⁵ Despite other reports,^62–64 we observed no significant associations between intellectual ability and global-FA or CT. This may be because we only studied global measures in a cross-sectional design here, which possibly occluded local effects and/or effects over time.

The mean IQ score of the DZ co-twins showed a pattern that was unexpected. DZ co-twins showed slightly higher average IQ scores than control twins, while MZ co-twins were more similar to patients. This pattern was also observed in the average factional anisotropy measures even though IQ and global-FA were not correlated. These observations may indicate a protective mechanism that has been hypothesized earlier in siblings of schizophrenia patients.⁶⁵ The observation that the MZ co-twins showed high rates of nonpsychotic psychiatric diagnoses may indicate that these phenotypes are only protective with regard to SZ, and perhaps not for more generalized psychiatric conditions.

The high rate of nonpsychotic psychiatric disorders among MZ co-twins can be expected under a liability threshold model, where the ordinal variable of disease status is hypothesized to be an underlying continuous trait.⁶⁶ Due to the high schizophrenia heritability, MZ co-twins have on average a higher liability then DZ co-twins. The higher occurrence of nonpsychotic disorders with higher genetic liability for schizophrenia could indicate a genetic overlap between both traits.⁶⁷ However, post hoc t tests did not reveal significant differences between MZ or DZ co-twins with and without psychiatric diagnoses on any of the modeled phenotypes, suggesting that the findings are specific to SZ.

Some limitations should be taken into consideration when interpreting the findings of this study. Firstly, we did not exhaustively test the model fit for all genetic and environmental paths separately. Performing such a test might have revealed that certain paths could be dropped from the model without significantly worsening the fit. However, with 5 phenotypes included, there are 25 paths to be fitted to the model (all paths in figure 1 can be present or not, except for e_ii for i = 1…5, which are present by default). This means that 2²⁵ ≥ 33 million possible submodels should be tested. As it is not feasible to fit this many models, we chose to rely on CIs for all paths to determine the relevance of their contribution. Secondly, since the Cholesky-decomposition is not a symmetrical model, the ordering of the variables can influence the path estimates. While the estimation of heritability and genetic correlations is not affected by reordering the variables, the estimation of shared genetic variance between variables can vary. It has been noted that a rationally defined order of variables is required when applying a multivariate Cholesky-decomposition.⁵⁸ As there is no rationally predefined order in the variables currently studied, we decided to run all possible permutations and test this effect. The model that provided the largest proportion of shared genetic variance between SZ and cognitive/imaging phenotypes was presented. However, all models produced similar results, which may be explained by overall lack of strong genetic interactions among imaging and cognitive phenotypes irrespective of disease. Third, the estimated heritability of schizophrenia varies between studies.^3,4 Therefore, we chose focus on the contributors to genetic vs environmental variance separately and not so much the total amount of variance. However, after rerunning the analysis with a more conservative schizophrenia heritability estimate of 64% the explained genetic variance shared between liability for schizophrenia and intellectual ability, FA and CT remained detectable. Last, although the sample size is not small for a twin-study, we cannot exclude the possibility of power limitations. With a sample size of n = 200, this study was adequately powered to detect traits with substantial heritability⁶⁸ and the multivariate design further increases power to detect genetic effects.⁶⁹ Power to detect environmental effects was lower¹⁵ and these results should thus be considered carefully.

In summary, genetic variance in SZ is significantly overlapping with genetic variance in intellectual ability, FA, and CT. Together, about 28% of the genetic variance in SZ is shared with these phenotypes, with about 16% due to variation in IQ and about 12% due to variation in brain structure. This indicates that brain-imaging phenotypes share genetic variance with SZ that is not captured by variation in IQ.

Supplementary Material

Supplementary material is available at http://schizophreniabulletin.oxfordjournals.org.

Funding

H.E.H.P. was supported by VIDI grant NWO-MW 917.46.370 from the Netherlands Organisation for Scientific Research.