r Human Brain Mapping 00:00–00 (2017) r

Central and Non-central Networks, Cognition,

Clinical Symptoms, and Polygenic Risk Scores in
Schizophrenia

Clara Alloza ,1* Mark E. Bastin,2,3,5 Simon R. Cox ,2,4,5 Jude Gibson,1
Barbara Duff,1 Scott I. Semple,6 Heather C. Whalley,1 and
Stephen M. Lawrie1
1
Division of Psychiatry, University of Edinburgh, Edinburgh, United Kingdom
2
Centre for Cognitive Ageing and Cognitive Epidemiology, University of Edinburgh,
Edinburgh, United Kingdom
3
Centre for Clinical Brain Sciences, Western General Hospital, University of Edinburgh,
Edinburgh, United Kingdom
4
Department of Psychology, University of Edinburgh, Edinburgh, United Kingdom
5
Scottish Imaging Network: A Platform for Scientific Excellence (SINAPSE) Collaboration,
Edinburgh, United Kingdom
6
Clinical Research Imaging Centre, University of Edinburgh, Edinburgh,
United Kingdom

r r

Abstract: Schizophrenia is a complex disorder that may be the result of aberrant connections between
specific brain regions rather than focal brain abnormalities. Here, we investigate the relationships
between brain structural connectivity as described by network analysis, intelligence, symptoms, and
polygenic risk scores (PGRS) for schizophrenia in a group of patients with schizophrenia and a group
of healthy controls. Recently, researchers have shown an interest in the role of high centrality networks
in the disorder. However, the importance of non-central networks still remains unclear. Thus, we spe-
cifically examined network-averaged fractional anisotropy (mean edge weight) in central and non-
central subnetworks. Connections with the highest betweenness centrality within the average network
(>75% of centrality values) were selected to represent the central subnetwork. The remaining connec-
tions were assigned to the non-central subnetwork. Additionally, we calculated graph theory measures
from the average network (connections that occur in at least 2/3 of participants). Density, strength,
global efficiency, and clustering coefficient were significantly lower in patients compared with healthy
controls for the average network (pFDR < 0.05). All metrics across networks were significantly associ-
ated with intelligence (pFDR < 0.05). There was a tendency towards significance for a correlation
between intelligence and PGRS for schizophrenia (r 5 20.508, p 5 0.052) that was significantly medi-
ated by central and non-central mean edge weight and every graph metric from the average network.
These results are consistent with the hypothesis that intelligence deficits are associated with a genetic

Additional Supporting Information may be found in the online *Correspondence to: Clara Alloza; Royal Edinburgh Hospital,
version of this article. Morningside Park, University of Edinburgh, Edinburgh EH10
Contract grant sponsor: National Health Service (NHS) Research 5HF, UK. E-mail: c.alloza@sms.ed.ac.uk
Scotland, Scottish Mental Health Research Network; Contract Received for publication 10 March 2017; Revised 2 August 2017;
grant sponsor: Dr Mortimer and Theresa Sackler Foundation; Accepted 24 August 2017.
Contract grant sponsor: MRC; Contract grant number: MR/ DOI: 10.1002/hbm.23798
M013111/1 Published online 00 Month 2017 in Wiley Online Library
(wileyonlinelibrary.com).

C 2017 Wiley Periodicals, Inc.

V
 r Alloza et al. r

risk for schizophrenia, which is mediated via the disruption of distributed brain networks. Hum Brain
Mapp 00:000–000, 2017. C 2017 Wiley Periodicals, Inc.
V

Key words: schizophrenia; diffusion tensor MRI; connectivity; intelligence; genetics; symptoms

r r

less prominent role of high degree hub regions such as the

INTRODUCTION prefrontal and parietal cortex, while nonfrontal hubs
Schizophrenia is a neuropsychiatric disorder character- emerge more prominently (Bassett et al., 2008; Guusje Col-
ized by delusions, hallucinations, absence of function, and lin et al., 2013; Zhang et al., 2012). Rubinov et al. (2009)
cognitive impairments. It is increasingly seen as the result suggested that a characteristic of the disorder is a random-
of aberrant connections between specific brain regions ization of connections, an alteration of community struc-
rather than focal brain abnormalities (Friston, 1998; Friston ture which results in impaired integration and segregation,
and Frith, 1995; Stephan et al., 2006, 2009). The dysconnec- and reduced centrality of cortical hubs. Most brain imag-
tivity hypothesis of schizophrenia suggests that abnormal ing studies in schizophrenia focus on these effects in net-
brain integration may underlie the cognitive profile and works with high centrality while the remaining
symptoms found in the disorder. There is consistent evi- connections are overlooked (Collin et al., 2014; Schmidt
dence supporting reduced levels of overall structural con- et al., 2016). Owing to the apparent hierarchical disorgani-
nectivity in schizophrenia using diffusion tensor MRI (DT- zation of the brain in schizophrenia, the role of these cen-
MRI) with frontal, parietal, and temporal projections being tral nodes may be displaced to other brain regions or
the most consistently impaired in the disorder (Skudlarski networks. Thus, in this study, we address specifically net-
et al., 2010; van den Heuvel et al., 2010; van den Heuvel works based on centrality to investigate this hypothesis.
and Fornito, 2014; Zalesky et al., 2011). Additionally, more Even though the cognitive and symptomatic implications
specific white matter alterations in the uncinate fasciculus, of various network metrics have been addressed, there has
corpus callosum, cingulum, and arcuate fasciculus are con- been little discussion about the role of non-central net-
sistently described (reviewed in Burns et al., 2003; Ellison- works in the disorder.
Wright and Bullmore, 2009; McIntosh et al., 2005). Even Schizophrenia is associated with cognitive deficits; some
though a number of studies have discussed the importance correlations between intelligence and the brain’s function
of white matter impairments in schizophrenia, there is still and structure have been described in healthy participants.
no consensus on how to measure structural dysconnectiv- Although there are a small number of established associa-
ity in the disorder. One approach is to characterize how tions between intelligence and brain basic structural
impairments in white matter microstructure affect the parameters, such as fractional anisotropy (FA), the rela-
organization of the structural connectome using graph the- tionship between the observed white matter alterations in
ory, which conceives the brain as a network composed of schizophrenia and intelligence remains unclear. However,
nodes and the connections (edges) between them (Bull- graph theory metrics may be able to provide greater
more and Sporns, 2009). Graph theory segregation mea- explanatory power for these cognitive deficits in schizo-
sures, such as clustering coefficient and modularity, are phrenia than more traditional structural connectivity mea-
reportedly altered in schizophrenia (Alexander-Bloch sures, such as FA (Alloza et al., 2016). There is some
et al., 2010; van den Heuvel et al., 2013; van den Heuvel evidence that structural network metrics are related to
and Fornito, 2014; Zalesky et al., 2011) suggesting a more intelligence and that there is a degree of shared genetic
segregated pattern of network organization. In line with overlap between schizophrenia and these measures. For
this hypothesis, numerous authors have found longer path instance, Li et al. (2009) found significant correlations
lengths and reductions in communication efficiency, pro- between intelligence and network properties in a healthy
posing reduced communication between more segregated cohort of subjects. Specifically, higher intelligence scores
areas of the brain (reviewed in van den Heuvel and were associated with shorter path lengths and higher
Fornito, 2014). global efficiency. Yeo et al. (2016) showed that global mea-
Nodes and edges can be associated with peripheral or sures of increased characteristic path length and reduced
more central tasks, depending on their degree of connec- overall connectivity predicted lower general intelligence in
tivity and their position within or between modules a group of patients with schizophrenia, while van den
(Sporns, 2011). Nodes characterized by high degree and Heuvel et al. (2009) also found a strong negative correla-
high centrality are termed “hubs.” Several lines of investi- tion between characteristic path length and IQ suggesting
gation have suggested that topological organization of hub that more efficiently connected brains tend to show higher
nodes appear to be altered in schizophrenia. Both struc- levels of intelligence. Hence, graph theory metrics may
tural covariance and structural connectivity studies in provide an insight into the underlying brain structural
schizophrenia suggest a less hierarchical organization, a substrate for intelligence.

r 2 r
 r Networks, Cognition, and PGRS in Schizophrenia r

Differences in structural connectivity are useful for calculated from the average network (defined as networks
establishing brain topology abnormalities in schizophrenia in which connections that occur in at least two-thirds of
compared with healthy participants. However, as our aim participants are retained) in relation to intelligence, clinical
is to shed light on the clinical manifestation of schizophre- symptoms and PGRS for schizophrenia in patients with
nia, we therefore examine the extent to which clinical schizophrenia and healthy controls. We will focus on
symptoms are associated with brain extracted measures. graph theory metrics that have been consistently reported
What we know about brain connectivity and clinical to be impaired in schizophrenia, namely mean edge
symptoms is largely based on empirical studies that inves- weight, density, strength, clustering coefficient, and global
tigate the relationship between white matter and different efficiency in the average network. Due to the severely
symptom’s scales. For instance, FA of specific white matter affected hierarchical disorganization of the brain found in
tracts has been significantly associated with positive symp- schizophrenia, our aim is to investigate the roles central
toms in the disorder. These tracts include the internal cap- and non-central network mean edge weight play in this
sule, fronto-occipital fasciculus, superior longitudinal disorder. Thus, this is the first study where intelligence,
fasciculus, cingulum, and corpus callosum (Mitelman symptoms, and PGRS have been studied together in rela-
et al., 2007; Rotarska-Jagiela et al., 2008; Seok et al., 2007). tion to networks based on their centrality. Specifically, we
To date, several authors have examined the effects of hypothesized that impaired structural organization of the
graph theory metrics of connectivity on symptomatology networks (decreased mean edge weight, density, strength,
in schizophrenia. Positive symptom severity has been asso- clustering coefficient, and global efficiency) will be associ-
ciated with reduced overall connectivity, increases and ated with lower intelligence, higher genetic risk factor for
decreases in structural and functional coupling, strength of schizophrenia, and higher symptom score.
temporal and frontal regions, reduced network efficiency,
and reduced clustering (reviewed in van den Heuvel and
Fornito, 2014). Wang et al. (2012) found significant associa- METHODS
tions between global efficiency and positive, negative, and Participants
total symptoms. However, most studies focus on func-
tional connectivity determined using fMRI and thus, Information about participants has been reported in
uncertainty remains regarding the relationship between detail previously (Whalley et al., 2015). Participants were
structural connectivity measured in central, non-central, recruited across Scotland as part of the Scottish Family
and average networks and genetic risk factors. Mental Health Study. DT-MRI data were acquired from a
Graph theory analysis has shown that impairments pre- total of 28 individuals diagnosed with schizophrenia and
sent in patients with schizophrenia are also found in their 36 healthy controls. Diagnosis of schizophrenia was con-
relatives suggesting a genetic basis (Clemm von Hohen- firmed using the structured clinical interview for DSM IV
berg et al., 2014; Guusje Collin et al., 2014; Skudlarski (SCID) administered by one of two trained psychiatrists
et al., 2013). Moreover, topological network properties (First et al., 2002). Exclusion criteria included any major
have been found to be heritable (Thompson et al., 2013). medical or neurological conditions, or any personal history
For instance, in white matter FA, the variance explained of substance misuse in the last year. Additionally, subjects
by genetic factors has been reported to be between 75% were excluded if there were MRI safety considerations. A
and 90% in almost every white matter tract (Chiang et al., detailed description of the study was given to all recruited
2011). Moreover, in the same study, heritability of FA was individuals, and all participants provided written informed
associated with the level of IQ. Genome-wide association consent. The study was approved by the Multicentre
studies (GWAS) have indicated a polygenic component of Research Ethics Committee for Scotland (09/MRE00/81).
schizophrenia with hundreds of common alleles of small
effect at the population level having been reported (Inter- Scan Acquisition
national Schizophrenia Consortium et al., 2009; Schizo-
phrenia Working Group of the Psychiatric Genomics All imaging data were collected on an MAGNETOM
Consortium, 2014). Thus far, only a small number of stud- Verio 3T MRI scanner running Syngo MR B17 software
ies have analyzed the relationship between polygenic risk (Siemens Healthcare, Erlangen, Germany). For each sub-
scores (PGRS), neuroimaging biomarkers, and/or cogni- ject, whole-brain DT-MRI data were acquired using a pro-
tion (Birnbaum and Weinberger, 2013; McIntosh et al., totype single-shot spin-echo echo-planar (EP) imaging
2013; Whalley et al., 2015). Connectomic measures are, sequence with diffusion-encoding gradients applied in 56
potentially, possible intermediate phenotypes between directions (b 5 1,000 s/mm2) and six T2-weighted (b 5 0 s/
genetic liability and cognitive deficits in schizophrenia. mm2) baseline scans. Fifty-five 2.5-mm-thick axial slices
In this study, we investigate the relationships between were acquired with a field-of-view of 240 3 240 mm and
brain structural connectivity described by network- matrix 96 3 96 giving 2.5 mm isotropic voxels. In the
averaged FA (mean edge weight) measured in central and same session, a 3D T1-magnetization-prepared rapidly
non-central networks and by graph theory metrics acquired gradient-echo (MPRAGE) volume was acquired

r 3 r
 r Alloza et al. r

in the coronal plane with 160 contiguous slices and 1 mm tortuous streamlines. For instance, a streamline with a total
isotropic voxel resolution. path length 10 times longer than the distance between end
points was considered to be invalid. The values of the cur-
vature, anisotropy, and distance ratio metric constraints
Image Analysis
were set empirically and informed by visual assessment of
Image processing the resulting streamlines.

Each 3D T1-weighted MPRAGE volume was parcellated

Network construction
into 85 (Desikan-Killiany atlas; Desikan et al., 2006) and 165
(Destrieux atlas) regions-of-interest (ROIs) using FreeSurfer FA-weighted networks were constructed by recording
(http://surfer.nmr.mgh.harvard.edu). The results of the the mean FA value along streamlines connecting all ROI
segmentation procedure were then used to construct grey (network node) pairs. The endpoint of a streamline was
and white matter masks for use in network construction considered to be the first grey-matter ROI encountered
and to constrain the tractography output as described when tracking from the seed location.
below. Using tools provided by the FDT package in FSL In this study, we assume the existence of a central sub-
(http://fsl.fmrib.ox.ac.uk/fsl), the DT-MRI data were pre- network that is shared across participants (Reijmer et al.,
processed to reduce systematic imaging distortions and 2016). To identify this central subnetwork, the average brain
bulk subject motion artifacts by affine registration of all network across both patients and controls was determined
subsequent EP volumes to the first T2-weighted EP volume by including those connections which occurred in more
(Jenkinson and Smith, 2001). Skull stripping and brain than two-thirds of the participants (de Reus and van den
extraction were performed on the registered T2-weighted Heuvel, 2013). Connections with the highest centrality (the
EP volumes and applied to the FA volume calculated by fraction of all shortest paths in the network that contain a
DTIFIT in each subject (Basser and Pierpaoli, 1996; Smith, given connection, also referred as “edge betweenness cen-
2002). The neuroanatomical ROIs determined by Freesurfer trality”) within this average network (>threshold value of
were then aligned from 3D T1-weighted volume to diffu- 75%) were selected and used to create a mask representing
sion space using a cross-modal nonlinear registration the central subnetwork. The remaining connections were
method. As a first step, linear registration was used to ini- assigned to the non-central subnetwork mask. Therefore,
tialize the alignment of each brain-extracted FA volume to connections with high values of centrality are involved in a
the corresponding FreeSurfer extracted 3D T1-weighted large number of shortest paths and as a consequence con-
brain volume using a mutual information cost function and tribute to the global efficiency of the network. These masks
an affine transform with 12 degrees of freedom (Jenkinson were then used as templates and applied to each partici-
and Smith, 2001). Following this initialization, a nonlinear pant’s connectivity matrix to select central and non-central
deformation field based method (FNIRT) was used to refine subnetworks. As the threshold value of 75% is arbitrary,
local alignment (Andersson et al., 2007). FreeSurfer segmen- analyses were repeated for thresholds of 25% and 50% of
tations and anatomical labels were then aligned to diffusion connections with highest centrality.
space using nearest neighbour interpolation. Organizational properties of the different networks were
then obtained using the brain connectivity toolbox (www.
Tractography brain-connectivity-toolbox.net). For each FA-weighted con-
nectivity matrix for the average network, five global net-
Whole-brain probabilistic tractography was performed work measures were computed, namely, mean edge
using FSL’s BedpostX/ProbTrackX algorithm (Behrens weight (mean value of FA across the network), density
et al., 2007). Probability density functions, which describe (the fraction of present connections to possible connec-
the uncertainty in the principal directions of diffusion, were tions), strength (the average sum of weights per node),
computed with a two-fiber model per voxel (Behrens et al., clustering coefficient (fraction of triangles around a node),
2007). Streamlines were then constructed by sampling from and global efficiency (the average of the inverse shortest
these distributions during tracking using 100 Markov Chain path length). As a result of possible alterations in topology
Monte Carlo iterations with a fixed step size of 0.5 mm when extracting central and non-central networks, only
between successive points. Tracking was initiated from all mean edge weight was computed for these subnetworks
white matter voxels (Buchanan et al., 2014) and streamlines (Reijmer et al., 2016).
were constructed in two collinear directions until termi-
nated by the following stopping criteria designed to mini- Polygenic Risk Score Calculation
mize the amount of anatomically implausible streamlines:
(i) exceeding a curvature threshold of 708; (ii) entering a PGRS is a method to aggregate the small effects that con-
voxel with FA below 0.1 (Verstraete et al., 2011); (iii) enter- tribute to the liability of schizophrenia on predicting the dis-
ing an extracerebral voxel; (iv) exceeding 200 mm in length; order. The capacity to predict onset of schizophrenia has
and (v) exceeding a distance ratio metric of 10. The distance been established and has been reported to explain up to 7%
ratio metric (Bullitt et al., 2003), excludes implausibly of additive genetic liability for the disorder (Schizophrenia

r 4 r
 r Networks, Cognition, and PGRS in Schizophrenia r

TABLE I. Demographic details for healthy controls and patients with schizophrenia

HC SZ P value

Age in years (SD) 37.22 (14.99) 38.04 (10.34) 0.807

Gender, M/F (%) 53/47 57/43 0.733
IQ (SD) 116.11 (10.75) 105.09 (15.89) 0.003
PANSS positive (SD) [range] 12.30 (5.19) [7, 28]
PANSS negative (SD) [range] 13 (7.05) [7, 35]
PANSS total (SD) [range] 51.64 (17.33) [34, 91]
Age of onset in years (SD) 25.25 (9.89)
Duration of illness in years (SD) 13.58 (10.30)
CPZ (SD) 434.97 (371.90)

HC, healthy controls; SZ, schizophrenia; CPZ, chlorpromazine equivalents; SD, standard deviation.
Bold typeface indicates significant group difference (P < 0.05).

Working Group of the Psychiatric Genomics Consortium, mean edge weight for central, non-central, and connectiv-
2014). PGRS for schizophrenia were created for all individu- ity metrics for the average networks separately. Age, sex,
als with suitable genotype data; only genotypes passing strin- diagnosis, and the interaction between diagnosis and sex
gent quality control were used in analyses. PGRS for were entered as predictors. FA was added as additional
schizophrenia were estimated using summary data from an predictor in the average network analysis. Owing to small
independent GWAS of schizophrenia in 150,064 individuals sample size, effect sizes were then calculated using
(36,989 cases and 113,075 controls), conducted by the Psychi- Hedges’ g and based on the p value of the individual anal-
atric Genomics Consortium (Schizophrenia Working Group ysis of covariance (ANCOVAS). Using the whole sample,
of the Psychiatric Genomics Consortium, 2014). PGRS were regression analyses were then performed separately for
estimated using the PRSice software package according to central, non-central, and average metrics and IQ. Due to
previously described protocols (Euesden et al., 2015), with the distribution of data, PANSS positive, negative, and
linkage disequilibrium and distance thresholds for clumping total symptom scores were only analyzed in the patient
of r2 5 0.2 and within a 300 kb window. Five scores were cre- sample. For both models, age, gender, and CPZ were used
ated for each individual using single-nucleotide polymor- as covariates. p values (a 5 0.05) were corrected for multi-
phisms (SNPs) selected according to the significance of their ple comparisons using false discovery rate (FDR; pFDR)
association with the phenotype at nominal p value thresholds (Benjamini and Hochberg, 1995). Analyses were repeated
of 0.01, 0.05, 0.1, 0.5, and 1.0 (all SNPs). For the analysis, we for varying threshold values to define the number of cen-
used the threshold of 0.5 which explained the most variance tral connections (25%, 50%, and 75%). Analyses were also
in our data and has been reported to maximally capture repeated for different Freesurfer brain atlases (Desikan
schizophrenia liability (International Schizophrenia Consor- and Destrieux). Regression models were then applied to
tium et al., 2009). The four multidimensional scaling factors investigate the association between risk score and case-
were entered as additional “nuisance” covariates to control control status in the whole sample. Connectivity metrics
for population stratification, along with age. were dependent variables and principal components for
population stratification, PGRS, age, gender, and diagnosis
were used as predictors. All statistical analyses were per-
Cognitive Testing and Medication formed with R version 3.2.3 (https://www.r-project.org).
Participants underwent cognitive assessment using tests Mediation analysis was subsequently used to examine
from the Wechsler Adult Intelligence Scale (WASI; Wechs- the hypothesis that higher PGRS is related to poorer intel-
ler, 1955) using standard administration and scoring pro- ligence via reduced structural connectivity. We employed
cedures. Symptom severity was assessed using the the PROCESS macro in SPSS 22.0 (Hayes and Rockwood,
Positive and Negative Symptoms Scale (PANSS) (Kay 2016) to formally quantify mediation effects using 5000
et al., 1987). Full-scale IQ was derived from four subtests bootstrapped samples. Due to our clear directional hypoth-
of the WASI: vocabulary, block design, similarities, and esis, a one-tailed test of mediation was conducted (http://
matrix reasoning. Participants also provided information www.afhayes.com). Mediation effects were considered sig-
on antipsychotic medication which was transformed into nificant if the confidence interval (CI) did not include zero
chlorpromazine equivalents (CPZ) (Woods, 2003). (Preacher and Hayes, 2008).

Statistical Analysis RESULTS

Group differences were analyzed using a multivariate Table I shows the demographic data for both healthy
general linear model (GLM). Dependent variables were controls and schizophrenia patients.

r 5 r
 r Alloza et al. r

TABLE II. Mean 6 standard deviation (SD) values of con- Age, Antipsychotic Medication, and Illness
nectivity metrics in the average network for healthy Duration
controls and patients with schizophrenia
There were positive significant associations between age,
Average mean edge weight (r 5 20.290, p 5 0.02) and clustering
Metric HC SZ PFDR coefficient (r 5 20.269, p 5 0.03) for the average network.
However, these associations did not survive multiple com-
Mean edge weight 0.44 6 0.02 0.43 6 0.02 0.08 parison correction (pFDR > 0.05). Antipsychotic medication
Density 33.15 6 0.92 32.56 6 1.25 0.04 show a significant effect on mean edge weight (r 5 20.262,
Strength 12.31 6 0.57 11.88 6 0.73 <0.001 pFDR 5 0.048), strength (r 5 20.287, pFDR 5 0.048), global
Global efficiency 0.30 6 0.01 0.30 6 0.01 <0.001 efficiency (r 5 20.263, pFDR 5 0.048), and clustering
Clustering coefficient 0.30 6 0.01 0.30 6 0.01 <0.001

HC, healthy controls; SZ, schizophrenia.

Bold typeface indicates significant group difference (PFDR < 0.05).

Average Network
Diagnosis (F (5, 54) 5 703.1, p < 0.001, partial eta
squared 5 0.080), age (F (5, 54) 5 137.64, p < 0.001, partial
eta squared 5 0.030), gender (F (5, 54) 5 19.80, p < 0.001,
partial eta squared 5 0.032), and mean edge weight (FA) (F
(5, 54) 5 15263.7, p < 0.001, partial eta squared 5 0.001)
effects were significant for the average network graph the-
ory metrics.
As indicated in Table II, there were significant differ-
ences in network density (Hedges’ g 5 0.54 (0.03, 1.05),
pFDR 5 0.04), strength (Hedges’ g 5 1.08 (0.54, 1.62),
pFDR < 0.001), global efficiency (Hedges’ g 5 1.95 (1.34,
2.56), pFDR < 0.001), and clustering coefficient (Hedges’
g 5 1.94 (1.33, 2.55), pFDR < 0.001) between groups. Mean
edge weight showed a tendency towards significance
(Hedges’ g 5 0.43 (20.07, 0.93), pFDR 5 0.08). All metrics
were reduced in patients compared to healthy controls.
Boxplots for group differences can be found in Supporting
Information, Figure 1.

Central Subnetwork
Figure 1 shows the network maps for the central (>75%
of centrality values) and non-central subnetworks across
all participants. There was no significant difference in cen-
tral subnetwork mean edge weight between patients with
schizophrenia and healthy controls (mean HC 5 0.45, SD
5 6 0.02; mean SZ 5 0.44, SD 5 6 0.02) (Hedges’ g 5 0.36
95% CI (20.14, 0.86), p > 0.05). Central mean edge weight
was reduced in patients compared with healthy controls. Figure 1.
Medium view of (A) central (>75% of centrality values) and (B)
Non-central Subnetwork non-central subnetworks for all participants indicating node loca-
tion and edge (FA) strength. The nodes which are connected by
There was a tendency towards significance for a differ- edges with the highest weights (FA > 0.5) in the central subnet-
ence in mean edge weight between patients with schizo- work are brainstem, left hemisphere precuneus cortex, thala-
phrenia and healthy controls (mean HC 5 0.44, SD mus, caudate, ventral diencephalon and superior frontal gyrus,
5 6 0.02; mean SZ 5 0.43, SD 5 6 0.02) (Hedges’ g 5 0.45 bilateral caudal anterior division of the cingulate cortex, and
95% CI (20.06, 0.95), p 5 0.07). Non-central mean edge isthmus division of the cingulate gyrus. Nodes are color-coded
weight was reduced in patients compared with healthy to indicate the lobe in which they are situated. [Color figure can
controls. be viewed at wileyonlinelibrary.com]

r 6 r
 r Networks, Cognition, and PGRS in Schizophrenia r

TABLE III. Correlation matrix for IQ and connectivity theory measures and PGRS. None of the connectivity met-
metrics for the average network rics was significantly associated with PGRS across networks
(pFDR > 0.05). Regression analysis between IQ and PGRS at
Metric R PFDR
a threshold of p  0.5 showed a tendency towards signifi-
cance (r 5 20.742, p 5 0.052). There was no significant corre-
Mean edge weight 0.343 0.016
Density 0.284 0.045 lation between PGRS and symptoms (pFDR > 0.05).
Strength 0.471 0.004
Global efficiency 0.394 0.007
Clustering coefficient 0.434 0.004 Mediation Analysis
Bold type indicates significant associations (PFDR < 0.05). We aimed to identify mediation candidates that were
consistent with the hypothesis that a greater genetic pre-
coefficient (r 5 20.270, pFDR 5 0.048) for the average disposition for schizophrenia is partly related to lower
network. Neither antipsychotic medication nor age had a intelligence through the disruption of brain connectivity.
significant effect on central mean edge weight. However, As indicated by the bivariate association, the correlation
age (r 5 20.313, p 5 0.012) and CPZ (r 5 20.271, p 5 0.033) between IQ and PGRS (r 5 20.742, p 5 0.052) showed a
showed a significant effect on non-central mean edge tendency towards significance. The negative correlation
weight. There were no significant associations between between PGRS and IQ suggests a genetic liability to intelli-
network metrics and illness duration for any of the metrics gence; mediation analysis allows us to quantify the role of
(pFDR > 0.05). topological network measures in this relationship. Given
the substantial effect sizes, and the need to consider medi-
IQ ation in terms of zero and nonzero rather than using p val-
ues in isolation (Hayes, 2009), we tested whether the direct
Regression coefficients between IQ and the average net- effect of PGRS and IQ was significantly mediated by mean
work graph theory metrics are shown in Table III. All met- edge weight and average network metrics (i.e., magnitude
rics were significantly associated with IQ (r range of change from path c to path c0 ; Fig. 2A). The results are
0.284–0.471). For central network, mean edge weight was shown in Figure 2B,C. A bias-corrected bootstrap CI for
significantly associated with IQ (r 5 0.344, p 5 0.010). For the indirect effect based on 5,000 bootstrap samples served
non-central network, mean edge weight was also signifi- as a formal statistical test of the degree to which mean
cantly associated with IQ (r 5 0.338, p 5 0.014). Medication, edge weight mediated the relationship between PGRS and
as CPZ equivalents, did not show any significant effect in IQ. The 30.52% reduction in magnitude (b 5 20.154 to
central, non-central, and average networks. Scatterplots b 5 20.107) identified central mean edge weight as a sig-
with the associations between metrics and IQ can be found nificant partial mediator (CI not containing zero; 20.363 to
in Supporting Information, Figures 2 and 3. 20.055). For non-central mean edge weight (Fig. 2C), the
reduction in magnitude was 46.62% (b 5 20.474 to
Clinical Symptoms b 5 20.253) identifying also non-central mean edge weight
as a significant partial mediator (CI 20.673 to 20.050). The
Table IV shows the regression coefficients for positive, model was corrected for age and population stratification
negative, and total symptom scores and central, non-cen- components. Additionally, Table V shows the mediation
tral mean edge weight, and average network connectivity results for the metrics of the average network.
metrics. Central network mean edge weight showed a ten-
dency towards significance in relation to total symptoms TABLE IV. Correlation matrix for PANSS and connectivity
(r 5 20.348, p 5 0.073). The addition of medication as a metrics for central (>75% of centrality values), non-central,
covariate in the model made the associations weaker and and average networks
non-significant (p > 0.05). However, medication did not
have any significant effect in the regression model. Metric Positive Negative Total

Central mean edge weight 20.282 20.184 20.348

Polygenic Risk Score Non-central mean edge weight 20.206 20.163 20.268
The association between genetic risk score at a threshold Average Mean edge weight 20.101 20.118 20.178
of p  0.5 and case–control status in the total sample was Density 20.195 20.041 20.092
Strength 20.201 20.114 20.193
significant (p < 0.05). The regression estimate of the genetic
Global efficiency 20.132 20.133 20.195
risk score at the threshold p  0.5 was 0.44 (adjusted R- Clustering coefficient 20.114 20.119 20.178
square 5 0.057; p 5 0.029).
Next, we studied the association between central and This table shows the associations between symptoms and metrics
non-central mean edge weight and average network graph using CPZ as a covariate.

r 7 r
 r Alloza et al. r

Figure 2.
(A) Schematic representation of relationships where an indepen- study, where X 5 polygenic risk score for schizophrenia (PGRS
dent variable (X) and an outcome (Y) are hypothesized to be at P  0.5), Y 5 IQ and M 5 mean edge weight (central). (C)
explained by a mediator (M). The direct effect of X on M is a, X 5 polygenic risk score for schizophrenia (PGRS at P  0.5),
the effect of M on Y is b, and the effect of X on Y is c. c0 Y 5 IQ and M 5 mean edge weight (non-central). Asterisks rep-
denotes the effect of X on Y when M is taking into account in resent statistically significant partial correlations.
the model. (B) Representation of the variables analyzed in this

Additional Analyses: Network Density showed larger differences between groups and stronger
associations with intelligence. Results were in the expected
Using network density as a covariate did not signifi- direction. Nevertheless, analyses showed comparable
cantly affect the results of any of the regression models results across both atlases for central and non-central
described above. mean edge weight and metrics from the average network
(data not shown).
Additional Analyses: Thresholds
Results for the different thresholds of centrality (25% DISCUSSION
and 50%) showed that associations between intelligence, This study was set out to assess the ability of graph the-
symptoms, and mean edge weight were comparable across ory metrics in schizophrenia to build a coherent model
thresholds (data not shown). from brain structure, cognition, and genetics. This is, to our
knowledge, the first study reporting results for both high
and low centrality networks in schizophrenia, and provides
Additional Analyses: Destrieux Atlas
much-needed structural MRI perspective on links between
Group differences using the Destrieux atlas (165 regions) brain connectivity and intelligence in this population. We
as a parcellation scheme showed no significant differences sought to investigate the evolving hypothesis that schizo-
between patients and controls for mean edge weight for phrenia is a hub disease in which central connections are
both central and non-central mean edge weight (p > 0.05). more severely affected in contrast to non-central connec-
Graph theory metric results from the average network tions. Our data indicate that this may not be the case.

TABLE V. Mediation analysis for the average network

b Mediation model
%
X Y M c c0 Attenuation F(df) Lower CI Upper CI

PGRS IQ Mean edge weight 20.474 20.253 46.62 8.20 (2, 43) 20.657 20.048
PGRS IQ Density 20.527 20.168 62.76 4.63 (2, 43) 20.431 20.002
PGRS IQ Strength 20.288 20.439 252.43 14.20 (2, 43) 20.893 20.142
PGRS IQ Global efficiency 20.376 20.348 7.44 9.85 (2, 43) 20.765 20.117
PGRS IQ Clustering coefficient 20.405 20.326 19.50 10.66 (2, 43) 20.705 20.090

X: independent variable; Y: outcome variable; M: mediator; c: path from X to Y; c0 : path from X to Y accounting for M.
Bold type face indicates significant mediation effect (confidence intervals do not include 0; Preacher and Hayes, 2008) and after FDR
correction (PFDR < 0.05).
All tests of mediation are one-tailed and bias-corrected.

r 8 r
 r Networks, Cognition, and PGRS in Schizophrenia r

Instead, schizophrenia may be a disorder characterized by thought disorders, while negative symptoms comprise
the disruption of distributed brain regions affecting the blunted affect, alogia, anhedonia, asociality, and avolition.
whole brain, rather than exclusively affecting hubs. Our These processes are likely to comprise distant and unique
study supports the conceptualization of schizophrenia as a regions (i.e., visual hallucinations could be associated with
disorder characterized by impaired integration between visual processing) and therefore, may not be captured by
brain regions rather than local brain abnormalities. an average network or by networks based on centrality.
The network analysis reported here shows that struc- Thus far, a number of functional studies have investigated
tural connectivity abnormalities are present in the schizo- the effects of graph theory metrics on symptomatology
phrenia patient group. Specifically, most graph theory (Bassett et al., 2012; Skudlarski et al., 2010). One study
metrics from the average network were significantly reported that higher levels of positive and negative symp-
reduced in the patient sample compared with healthy con- toms were associated with reduced clustering coefficient
trols. These results are consistent with previous findings and increased path lengths (Shim et al., 2014). A further
(van den Heuvel and Fornito, 2014), in particular, density, study found that local connectome organization relates to
strength, global efficiency, and clustering coefficient were longitudinal increases in overall PANSS, in particular, these
significantly reduced in the patient group compared with associations were driven by clustering coefficient (Collin
controls. The central subnetwork was principally com- et al., 2016). Previous studies have found negative correla-
posed of subcortical areas and regions located in the fron- tions between FA (using DT-MRI) and positive, negative,
tal and parietal lobes. Mean edge weight (FA) for central and total PANSS score (Michael et al., 2008; Skelly et al.,
and non-central subnetworks, was not significantly differ- 2008). For instance, negative correlations between FA and
ent between patients and healthy controls. Taken together, negative symptoms in specific white matter tracts, such as
these results suggest that in schizophrenia, the structural the corpus callosum, have been reported (Nakamura et al.,
connectome is characterized by weaker connections being 2012). However, the inconsistency of the findings may be
less segregated and less integrated compared with healthy the result of different methodological techniques, use of
controls. Thus, here we have shown that differences medication, and heterogeneity of the disease.
between patients and controls can be found in the average In the central and non-central subnetworks, comparable
network, suggesting the presence of more extensive associations were found between intelligence and mean
impairments that are seemingly not limited to central edge weights across all thresholds. Stronger associations
connections. were found for symptoms with non-central mean edge
We also found that every graph theory metric across the weight when considering the top 75% of network connec-
different networks was significantly associated with IQ. tions based on their centrality. A lower centrality threshold
These results are likely to reflect the integrative nature of (25%–50% central connections) showed weaker correla-
intelligence, involving distributed brain networks that tions, probably because of a reduced specificity of the sub-
comprise a wide variety of cognitive functions (Colom network and exclusion of some important connections.
et al., 2010). The absence of an interaction between graph There is an overlap between the genetic risk factor for
theory metrics and group indicates that the same effect schizophrenia and intelligence (Glahn et al., 2007; McIn-
occurs in healthy participants and schizophrenia patients. tosh et al., 2013; Toulopoulou et al., 2007) and thus, brain
These results are consistent with those of Li et al. (2009) structure may be an intermediate phenotype between
who reported that IQ was positively correlated with global genetics and intelligence. In this study we have shown
efficiency and negatively with path length. Central and that central and non-central mean edge weight signifi-
non-central mean edge weight (FA) were positively associ- cantly mediated the relationship between genetics and
ated with IQ, this is in accordance with numerous investi- intelligence between 30% and 47%, respectively. Moreover,
gations assessing, for instance, relationships between every graph theory metric from the average network sig-
intelligence and general factors of FA (Alloza et al., 2016; nificantly mediated this relationship. Thus, we propose
Chiang et al., 2009; Deary et al., 2006; Penke et al., 2010; that structural brain topology measures are potential inter-
Yu et al., 2008). Thus, in this study, we have been able to mediate phenotypes in this model. Although metrics were
establish robust associations between intelligence and the not significantly associated with PGRS, statistical signifi-
structural connectome in schizophrenia. cance of all paths is not a prerequisite to determining a
The dysconnection hypothesis proposes that altered mediation model (Hayes and Rockwood, 2016). The
topological connectivity and abnormal integration between approach taken here detected moderate effect sizes and
distinct brain regions may underlie the symptomatology had the ability to formally quantify the degree and signifi-
found in the disorder (Stephan et al., 2006, 2009). In this cance of the mediation. However, better-powered studies
study, none of the graph metrics were significantly associ- are needed to confirm this.
ated with positive, negative, or total symptoms. These These findings suggest that prominent associations and
results suggest that symptoms may be specifically based disruptions occur also in average and non-central net-
on deficiencies in distinctive networks. For instance, posi- works, which are not driven by medication effects and are
tive symptoms include hallucinations, delusions, and present across different brain parcellation schemes. We

r 9 r
 r Alloza et al. r

hypothesize that the construction of subnetworks in Boards (Grampian, Tayside, Lothian, and Greater Glasgow
schizophrenia may be affected by its inherent reduced cen- and Clyde), Scottish Enterprise, and Pfizer, who have
trality and thus, central networks may include less central reviewed and approved the manuscript. The authors would
connections. This is in line with a recent publication where also like to thank Dr Thorsten Feiweier from Siemens
the authors propose that schizophrenia may not be Healthcare for providing the prototype diffusion sequence
entirely, nor specifically, a hub disease (Griffa et al., 2015). used in this study.
Based on previous literature and the limitation of our own
study, we propose that schizophrenia is a disorder charac-
REFERENCES
terized by the disruption of distributed brain regions
affecting the whole brain rather than hubs exclusively.
Our study therefore supports the conceptualization of Alexander-Bloch AF, Gogtay N, Meunier D, Birn R, Clasen L,
schizophrenia as a disorder characterized by impaired Lalonde F, Lenroot R, Giedd J, Bullmore ET (2010): Disrupted
integration between brain regions rather than local brain modularity and local connectivity of brain functional networks
abnormalities. in childhood-onset schizophrenia. Front Syst Neurosci 4:147.
Alloza C, Cox SR, Duff B, Semple SI, Bastin ME, Whalley HC,
Lawrie SM (2016): Information processing speed mediates the
LIMITATIONS relationship between white matter and general intelligence in
schizophrenia. Psychiatry Res 254:26–33.
Our findings are limited by the intrinsic nature of the Andersson JL, Jenkinson M, Smith S (2007): Non-Linear Registration,
methodology implemented. For example, limitations associ- aka Spatial Normalisation FMRIB Technical Report TR07JA2,
ated with DT-MRI—a technique that relies on water diffu- Vol. 2. FMRIB Analysis Group of the University of Oxford.
sion as an indirect marker for white matter microstructure Basser PJ, Pierpaoli C (1996): Microstructural and physiological
which has not yet been able to resolve complex fiber archi- features of tissues elucidated by quantitative-diffusion-tensor
MRI. J Magn Reson Ser B 111:209–219.
tecture (Jones et al., 2013)—need to be acknowledged.
Bassett DS, Bullmore E, Verchinski BA, Mattay VS, Weinberger
Other limitations include the fact that most of the
DR, Meyer-Lindenberg A (2008): Hierarchical organization of
patients in this study used antipsychotic medication, human cortical networks in health and schizophrenia.
which may affect structural brain connectivity (Szeszko J Neurosci 28:9239–9248.
et al., 2014). Nonetheless, it should be noted that impaired Bassett DS, Nelson BG, Mueller BA, Camchong J, Lim KO (2012):
white matter connectivity has also been shown in never- Altered resting state complexity in schizophrenia. NeuroImage
medicated patients (Cheung et al., 2008; Mandl et al., 59:2196–2207.
2013). Additionally, the patients were recruited from out- Behrens TEJ, Berg HJ, Jbabdi S, Rushworth MFS, Woolrich MW
patient clinics, thus generalizability of the results may be (2007): Probabilistic diffusion tractography with multiple fibre
orientations: What can we gain? NeuroImage 34:144–155.
less applicable to more severely affected populations.
Benjamini Y, Hochberg Y (1995): Controlling the false discovery
Moreover, the sample size used is small by contemporary rate: A practical and powerful approach to multiple testing.
standards raising the possibility of Type II errors. Thus J R Stat Soc Ser B (Methodol) 57:289–300.
interpretations of our novel but preliminary results should Birnbaum R, Weinberger DR (2013): Functional neuroimaging and
be taken cautiously. To further validate the results pre- schizophrenia: A view towards effective connectivity modeling
sented here, replication of this study using larger datasets and polygenic risk. Dialog Clin Neurosci 15:279–289.
is needed. Buchanan CR, Pernet CR, Gorgolewski KJ, Storkey AJ, Bastin ME
(2014): Test-retest reliability of structural brain networks from
diffusion MRI. NeuroImage 86:231–243.
ACKNOWLEDGMENTS Bullitt E, Gerig G, Pizer SM, Lin W, Aylward SR (2003): Measur-
The authors would like to thank all the participants who ing tortuosity of the intracerebral vasculature from MRA
images. IEEE Trans Med Imag 22:1163–1171.
took part in this study and the radiographers who acquired
Bullmore E, Sporns O (2009): Complex brain networks: Graph the-
the MRI data at the Clinical Research Imaging Centre, Uni-
oretical analysis of structural and functional systems. Nat Rev
versity of Edinburgh (http://www.cric.ed.ac.uk). They are Neurosci 10:186–198.
also grateful to Professor Edwin van Beek for assistance Burns J, Job D, Bastin ME, Whalley H, Macgillivray T, Johnstone
with overall co-ordination of the study and for radiological EC, Lawrie SM (2003): Structural disconnectivity in schizophre-
reporting of the structural MRI scans. The investigators also nia: A diffusion tensor magnetic resonance imaging study. Br J
acknowledge the financial support of National Health Ser- Psychiatry 182:439–443.
vice (NHS) Research Scotland, through the Scottish Mental Cheung V, Cheung C, McAlonan GM, Deng Y, Wong JG, Yip L,
Health Research Network (http://www.smhrn.org.uk) who Tai KS, Khong PL, Sham P, Chua SE (2008): A diffusion tensor
imaging study of structural dysconnectivity in never-
provided assistance with subject recruitment and cognitive
medicated, first-episode schizophrenia. Psychol Med 38:
assessments. This work was supported by an award from 877–885.
the Translational Medicine Research Collaboration—a con- Chiang M-C, Barysheva M, Shattuck DW, Lee AD, Madsen SK,
sortium made up of the Universities of Aberdeen, Dundee, Avedissian C, Klunder AD, Toga AW, McMahon KL,
Edinburgh and Glasgow, the four associated NHS Health Zubicaray GI, de Wright MJ, Srivastava A, Thompson PM

r 10 r
 r Networks, Cognition, and PGRS in Schizophrenia r

(2009): Genetics of brain fiber architecture and intellectual per- Observations, recommendations, and implementation. Behav
formance. J Neurosci 29:2212–2224. Res Therapy. doi: 10.1016/j.brat.2016.11.001.
Chiang M-C, McMahon KL, de Zubicaray GI, Martin NG, Hickie International Schizophrenia Consortium, Purcell SM, Wray NR,
I, Toga AW, Wright MJ, Thompson PM (2011): Genetics of Stone JL, Visscher PM, O’Donovan MC, Sullivan PF, Sklar P
white matter development: A DTI study of 705 twins and their (2009): Common polygenic variation contributes to risk of
siblings aged 12 to 29. NeuroImage 54:2308–2317. schizophrenia and bipolar disorder. Nature 460:748–752.
Clemm von Hohenberg C, Pasternak O, Kubicki M, Ballinger T, Jenkinson M, Smith S (2001): A global optimisation method for
Vu M-A, Swisher T, Green K, Giwerc M, Dahlben B, Goldstein robust affine registration of brain images. Med Image Anal 5:
JM, Woo T-UW, Petryshen TL, Mesholam-Gately RI, 143–156.
Woodberry KA, Thermenos HW, Mulert C, McCarley RW, Jones DK, Kn€ osche TR, Turner R (2013): White matter integrity,
Seidman LJ, Shenton ME (2014): White matter microstructure fiber count, and other fallacies: The do’s and don’ts of diffu-
in individuals at clinical high risk of psychosis: A whole-brain sion MRI. NeuroImage 73:239–254.
diffusion tensor imaging study. Schizophr Bull 40:895–903. Kay SR, Fiszbein A, Opler LA (1987): The positive and negative
Collin G, de Nijs J, Hulshoff Pol HE, Cahn W, van den Heuvel syndrome scale (PANSS) for schizophrenia. Schizophr Bull 13:
MP (2016): Connectome organization is related to longitudinal 261–276.
changes in general functioning, symptoms and IQ in chronic Li Y, Liu Y, Li J, Qin W, Li K, Yu C, Jiang T (2009): Brain anatom-
schizophrenia. Schizophr Res 173:166–173. ical network and intelligence. PLoS Comput Biol 5:e1000395.
Collin G, de Reus MA, Cahn W, Hulshoff Pol HE, Kahn RS, van Mandl RCW, Rais M, van Baal GCM, van Haren NEM, Cahn W,
den Heuvel MP (2013): Disturbed grey matter coupling in Kahn RS, Pol HEH (2013): Altered white matter connectivity in
schizophrenia. Eur Neuropsychopharmacol 23:46–54. never-medicated patients with schizophrenia. Hum Brain
Collin G, Kahn RS, Reus D, A, M, Cahn W, Heuvel VD, P, M Mapp 34:2353–2365.
(2014): Impaired rich club connectivity in unaffected siblings McIntosh AM, Gow A, Luciano M, Davies G, Liewald DC, Harris
SE, Corley J, Hall J, Starr JM, Porteous DJ, Tenesa A, Visscher
of schizophrenia patients. Schizophr Bull 40:438–448.
PM, Deary IJ (2013): Polygenic risk for schizophrenia is associ-
Colom R, Karama S, Jung RE, Haier RJ (2010): Human intelligence
ated with cognitive change between childhood and old age.
and brain networks. Dialog Clin Neurosci 12:489–501.
Biol Psychiatry 73:938–943.
de Reus MA, van den Heuvel MP (2013): Estimating false posi-
McIntosh AM, Job DE, Moorhead TWJ, Harrison LK, Lawrie SM,
tives and negatives in brain networks. NeuroImage 70:402–409.
Johnstone EC (2005): White matter density in patients with
Deary IJ, Bastin ME, Pattie A, Clayden JD, Whalley LJ, Starr JM,
schizophrenia, bipolar disorder and their unaffected relatives.
Wardlaw JM (2006): White matter integrity and cognition in
Biol Psychiatry 58:254–257.
childhood and old age. Neurology 66:505–512.
Michael AM, Calhoun VD, Pearlson GD, Baum SA, Caprihan A
Desikan RS, Segonne F, Fischl B, Quinn BT, Dickerson BC, Blacker
(2008): Correlations of diffusion tensor imaging values and
D, Buckner RL, Dale AM, Maguire RP, Hyman BT, Albert MS,
symptom scores in patients with schizophrenia. Conference
Killiany RJ (2006): An automated labeling system for subdivid-
proceedings: Annual International Conference of the IEEE
ing the human cerebral cortex on MRI scans into gyral based
Engineering in Medicine and Biology Society. Annual Confer-
regions of interest. NeuroImage 31:968–980.
ence. IEEE Eng Med Biol Soc 2008:5494–5497.
Ellison-Wright I, Bullmore E (2009): Meta-analysis of diffusion Mitelman SA, Torosjan Y, Newmark RE, Schneiderman JS, Chu K-
tensor imaging studies in schizophrenia. Schizophr Res 108: W, Brickman AM, Haznedar MM, Hazlett EA, Tang CY,
3–10. Shihabuddin L, Buchsbaum MS (2007): Internal capsule, cor-
Euesden J, Lewis CM, O’Reilly PF (2015): PRSice: Polygenic risk pus callosum and long associative fibers in good and poor out-
score software. Bioinformatics (Oxford, England) 31:1466–1468. come schizophrenia: A diffusion tensor imaging survey.
First MB, Spitzer RL, Gibbon M, Williams JBW (2002): Structured Schizophr Res 92:211–224.
Clinical Interview for DSM-IV-TR Axis I Disorders, Research Nakamura K, Kawasaki Y, Takahashi T, Furuichi A, Noguchi K,
Version, Patient Edition. (SCID-I/P) New York: Biometrics Seto H, Suzuki M (2012): Reduced white matter fractional
Research. State Psychiatric Institute, New York. anisotropy and clinical symptoms in schizophrenia: A voxel-
Friston KJ (1998): The disconnection hypothesis. Schizophr Res 30: based diffusion tensor imaging study. Psychiatry Res Neuroi-
115–125. mag 202:233–238.
Friston KJ, Frith CD (1995): Schizophrenia: A disconnection syn- Penke L, Maniega SM, Murray C, Gow AJ, Hern andez MCV,
drome? Clin Neurosci (New York, N.Y.) 3:89–97. Clayden JD, Starr JM, Wardlaw JM, Bastin ME, Deary IJ
Glahn DC, Almasy L, Blangero J, Burk GM, Estrada J, Peralta JM, (2010): A general factor of brain white matter integrity predicts
Meyenberg N, Castro MP, Barrett J, Nicolini H, Ravent os H, information processing speed in healthy older people.
Escamilla MA (2007): Adjudicating neurocognitive endopheno- J Neurosci 30:7569–7574.
types for schizophrenia. Am J Med Genet B Neuropsychiatr Preacher KJ, Hayes AF (2008): Asymptotic and resampling strate-
Genet 144B:242–249. gies for assessing and comparing indirect effects in multiple
Griffa A, Baumann PS, Ferrari C, Do KQ, Conus P, Thiran J-P, mediator models. Behav Res Methods 40:879–891.
Hagmann P (2015): Characterizing the connectome in schizo- Reijmer YD, Fotiadis P, Piantoni G, Boulouis G, Kelly KE, Gurol
phrenia with diffusion spectrum imaging. Hum Brain Mapp ME, Leemans A, O’Sullivan MJ, Greenberg SM, Viswanathan
36:354–366. A (2016): Small vessel disease and cognitive impairment: The
Hayes AF (2009): Beyond Baron and Kenny: Statistical mediation relevance of central network connections. Hum Brain Mapp
analysis in the new millennium. Commun Monogr 76:408–420. 37:2446–2454.
Hayes AF, Rockwood NJ (2016): Regression-based statistical medi- Rotarska-Jagiela A, Sch€ onmeyer R, Oertel V, Haenschel C,
ation and moderation analysis in clinical research: Vogeley K, Linden DEJ (2008): The corpus callosum in

r 11 r
 r Alloza et al. r

schizophrenia-volume and connectivity changes affect specific Thompson PM, Ge T, Glahn DC, Jahanshad N, Nichols TE (2013):
regions. NeuroImage 39:1522–1532. Genetics of the connectome. NeuroImage 80:475–488.
Rubinov M, Knock SA, Stam CJ, Micheloyannis S, Harris AWF, Toulopoulou T, Picchioni M, Rijsdijk F, Hua-Hall M, Ettinger U,
Williams LM, Breakspear M (2009): Small-world properties of Sham P, Murray R (2007): Substantial genetic overlap between
nonlinear brain activity in schizophrenia. Hum Brain Mapp 30: neurocognition and schizophrenia: Genetic modeling in twin
403–416. samples. Arch Gen Psychiatry 64:1348–1355.
Schizophrenia Working Group of the Psychiatric Genomics Con- van den Heuvel MP, van den Mandl RCW, Stam CJ, Kahn RS, Pol
sortium (2014): Biological insights from 108 schizophrenia- HEH (2010): Aberrant frontal and temporal complex network
associated genetic loci. Nature 511:421–427. structure in schizophrenia: A graph theoretical analysis.
Schmidt A, Crossley NA, Harrisberger F, Smieskova R, Lenz C, J Neurosci 30:15915–15926.
Riecher-R€ossler A, Lang UE, McGuire P, Fusar-Poli P, van den Heuvel MP, Fornito A (2014): Brain networks in schizo-
Borgwardt S (2016): Structural network disorganization in sub- phrenia. Neuropsychol Rev 24:32–48.
jects at clinical high risk for psychosis. Schizophr Bull 43: van den Heuvel MP, Sporns O, Collin G, Scheewe T, Mandl
583–591. RCW, Cahn W, Go~ ni J, Hulshoff Pol HE, Kahn RS (2013):
Seok J-H, Park H-J, Chun J-W, Lee S-K, Cho HS, Kwon JS, Kim Abnormal rich club organization and functional brain dynam-
J-J (2007): White matter abnormalities associated with audi- ics in schizophrenia. JAMA Psychiatry 70:783–792.
tory hallucinations in schizophrenia: A combined study of van den Heuvel MP, Stam CJ, Kahn RS, Hulshoff Pol HE (2009):
voxel-based analyses of diffusion tensor imaging and struc- Efficiency of functional brain networks and intellectual perfor-
tural magnetic resonance imaging. Psychiatry Res 156: mance. J Neurosci 29:7619–7624.
93–104. Verstraete E, Veldink JH, Mandl RCW, van den Berg LH, van den
Shim M, Kim D-W, Lee S-H, Im C-H (2014): Disruptions in small-
Heuvel MP (2011): Impaired structural motor connectome in
world cortical functional connectivity network during an audi-
amyotrophic lateral sclerosis. PLoS One 6:e24239.
tory oddball paradigm task in patients with schizophrenia.
Wang Q, Su T-P, Zhou Y, Chou K-H, Chen I-Y, Jiang T, Lin C-P
Schizophr Res 156:197–203.
(2012): Anatomical insights into disrupted small-world net-
Skelly LR, Calhoun V, Meda SA, Kim J, Mathalon DH, Pearlson
works in schizophrenia. NeuroImage 59:1085–1093.
GD (2008): Diffusion tensor imaging in schizophrenia: Rela-
Wechsler D (1955): Manual for the Wechsler Adult Intelligence
tionship to symptoms. Schizophr Res 98:157–162.
Scale, Vol. vi. Oxford, England: Psychological Corp.
Skudlarski P, Jagannathan K, Anderson K, Stevens MC, Calhoun
Whalley HC, Hall L, Romaniuk L, Macdonald A, Lawrie SM,
VD, Skudlarska BA, Pearlson G (2010): Brain connectivity is
Sussmann JE, McIntosh AM (2015): Impact of cross-disorder
not only lower but different in schizophrenia: A combined
polygenic risk on frontal brain activation with specific effect of
anatomical and functional approach. Biol Psychiatry 68:61–69.
Skudlarski P, Schretlen DJ, Thaker GK, Stevens MC, Keshavan schizophrenia risk. Schizophr Res 161:484–489.
MS, Sweeney JA, Tamminga CA, Clementz BA, O’Neil K, Woods SW (2003): Chlorpromazine equivalent doses for the newer
Pearlson GD (2013): Diffusion tensor imaging white matter atypical antipsychotics. J Clin Psychiatry 64:663–667.
endophenotypes in patients with schizophrenia or psychotic Yeo RA, Ryman SG, van den Heuvel MP, de Reus MA, Jung RE,
bipolar disorder and their relatives. Am J Psychiatry 170: Pommy J, Mayer AR, Ehrlich S, Schulz SC, Morrow EM,
886–898. Manoach D, Ho B-C, Sponheim SR, Calhoun VD (2016): Graph
Smith SM (2002): Fast robust automated brain extraction. Hum metrics of structural brain networks in individuals with
Brain Mapp 17:143–155. schizophrenia and healthy controls: Group differences, rela-
Sporns O (2011): The non-random brain: Efficiency, economy, and tionships with intelligence, and genetics. J Int Neuropsychol
complex dynamics. Front Comput Neurosci 5, 5. Soc 22:240–249.
Stephan KE, Baldeweg T, Friston KJ (2006): Synaptic plasticity and Yu C, Li J, Liu Y, Qin W, Li Y, Shu N, Jiang T, Li K (2008): White
dysconnection in schizophrenia. Biol Psychiatry 59:929–939. matter tract integrity and intelligence in patients with mental
Stephan KE, Friston KJ, Frith CD (2009): Dysconnection in schizo- retardation and healthy adults. NeuroImage 40:1533–1541.
phrenia: From abnormal synaptic plasticity to failures of self- Zalesky A, Fornito A, Seal ML, Cocchi L, Westin C-F, Bullmore
monitoring. Schizophr Bull 35:509–527. ET, Egan GF, Pantelis C (2011): Disrupted axonal fiber connec-
Szeszko PR, Robinson DG, Ikuta T, Peters BD, Gallego JA, Kane J, tivity in schizophrenia. Biol Psychiatry 69:80–89.
Malhotra AK (2014): White matter changes associated with Zhang Y, Lin L, Lin C-P, Zhou Y, Chou K-H, Lo C-Y, Su T-P,
antipsychotic treatment in first-episode psychosis. Neuropsy- Jiang T (2012): Abnormal topological organization of structural
chopharmacology 39:1324–1331. brain networks in schizophrenia. Schizophr Res 141:109–118.

r 12 r