Introduction
A substantial body of evidence supports the hypothesis that schizophrenia is a developmental disorder in which the cerebral connectivity and morphology are disturbed (Rapoport et al.
2012). Investigation of the cortical morphology is potentially informative about pathological deviations in neurodevelopment (Gay et al.
2012). In particular, neuroimaging and post-mortem studies report abnormal cortical folding in schizophrenia (White et al.
2003; Harris
2004; Wheeler and Harper
2007; Bonnici et al.
2007; Cachia et al.
2008; Schultz et al.
2010,
2011). Whole brain vertex wise localization studies note reduced gyrification in several brain regions including insula, parieto-temporal region, precuneus, lateral prefrontal cortex and precentral region, and increased gyrification in anterior aspect of prefrontal cortex (Nesvåg et al.
2014; Palaniyappan and Liddle
2012). Further, the longitudinal trajectory of regional gyrification deviates from that of age-matched peers without schizophrenia (Palaniyappan et al.
2013a). This suggests that the cross-sectional observations of altered regional gyrification in schizophrenia can be linked to maturational disturbances (White and Hilgetag
2011).
While comparing diagnostic groups with mass univariate whole brain analysis reveal localized regional changes in a ‘lesional’ sense, this approach fails to quantify the relationship between concomitant changes in different brain areas. Crucial information about abnormalities in the integrated development of the brain as a connected system can be gathered by studying the covariance of morphology. Graph-based approaches provide a powerful mode of finding subtle differences in brain organization (Bullmore and Sporns
2009). In particular, morphological networks based on anatomical covariance among brain regions capture an important aspect of developmental maturation crucial for understanding the pathophysiology of psychotic disorders (Alexander-Bloch et al.
2013a; Evans
2013). Direct evidence linking anatomical covariance to coordinated brain development is beginning to emerge in recent times (Raznahan et al.
2011; Alexander-Bloch et al.
2013a,
b). Graph theory offers a powerful technique for investigating the organization of the pairwise connections between nodes of networks. Application of graph theory to neuroimaging data reveals that in the normal human brain, regions tend to be connected in manner that creates an efficient ‘small world’ network in which long-range connections link or ‘integrate’ key local hubs that in turn connect to multiple nearby brain regions in a modular or segregated fashion. In patients with schizophrenia, the pattern of connections reveals a more segregated, less integrated and inefficient system (Bassett et al.
2008; Alexander-Bloch et al.
2010; Fornito et al.
2012; Zhang et al.
2012; van den Heuvel et al.
2013).
Among various morphological properties of the brain, cortical folding appears especially relevant to the development of brain as a connected system (Mota and Herculano-Houzel
2012; Chen et al.
2012). Experimental disruption of cortical connections during early stages of primate development produces alterations in folding patterns both proximal and distal to the induced lesions (Goldman-Rakic
1980; Goldman-Rakic and Rakic
1984). Thus the connections between regions can exert a strong influence on the cortical folding of the connected regions, and it might be expected that the correlations between folding patterns in different brain regions would be informative of the development of cerebral connectivity (Neal et al.
2007; Takahashi et al.
2011). In light of the time locked patterns of fetal sulcation and gyrification (Dubois et al.
2008b; Nishikuni and Ribas
2013; Zhang et al.
2013), investigating disturbance of structural covariance patterns of cortical folding in adults could provide insights into neurodevelopmental aberrations.
To our knowledge, the structural covariance patterns of cortical gyrification are yet to be investigated in schizophrenia. In the present study, we applied graph theory to analyze the pattern of regional correlations in gyrification in patients with schizophrenia and in healthy controls to test the hypothesis that patients with schizophrenia would exhibit a greater degree of segregated architecture affecting key regional nodes such as the insula and the lateral prefrontal cortex, previously shown to have localizable cortical folding defects in patients (Palaniyappan and Liddle
2012). We also anticipated that patients with more severe illness would show a pronounced aberration in the connectomic architecture of cortical folding in these regions, implying a neurodevelopmental pathway to illness severity.
Discussion
To our knowledge, we report the presence of robust small-world properties in the gyrification-based network for the first time in both healthy controls and in schizophrenia. Presence of small-worldness in gyrification-based network suggests that even in the absence of a direct covarying relationship in folding patterns between some brain regions, a small number of other regions mediate the overall interrelatedness, with the folding pattern of the entire brain showing a complex relationship typical of several evolutionarily advantageous biological networks (Bullmore and Sporns
2012). Importantly, this also suggests that while most of the structural covariance in cortical folding is limited to proximal nodes (clustering), gyrification patterns of distal regions are also strongly interrelated. Despite the absence of prominent alterations in the global efficiency, we note significant alterations in the regional topological properties, with increased segregation of right anterior insula and right middle frontal (dorsolateral prefrontal) gyrus along with a reduction in the segregation of structures around the central sulcus and lateral occipital cortex. In addition, the prominent centrality of cingulate structures that was observed in healthy controls was not present in patients.
The overall small-world architecture of the gyrification network is preserved in schizophrenia, suggesting that abnormalities in the folding patterns seen in patients are subtle and do not affect the basic organizing principles of cortical folding. Nevertheless, patients with schizophrenia showed significant changes in the regional topological properties. Right anterior insula and right dorsolateral prefrontal region were highly segregated with more localized covariance in patients than controls. These two regions belong to two distinguishable large-scale networks that form an integrated information processing system (Seeley et al.
2007; Menon and Uddin
2010). Several neuroimaging studies have repeatedly implicated the importance of these two regions in a myriad of cognitive processing tasks, highlighting the relative importance of these two regions in enabling efficient coordination with the rest of the brain (Critchley et al.
2004; Bressler and Menon
2010). Abnormalities in the functional connectivity pertaining to these regions have been repeatedly observed in patients with schizophrenia (Moran et al.
2013; Manoliu et al.
2013; Palaniyappan et al.
2013b,
c). Reduced gyrification of insula and dorsolateral frontal cortex has also been previously reported in schizophrenia (Bonnici et al.
2007; Nesvåg et al.
2014). Abnormally increased segregation of these regions suggest that in patients, the developmental trajectory of the right anterior insula and the middle frontal gyrus has relatively less influence on distributed brain regions, but more influence on anatomically constrained proximal regions. Altered localized covariance or cliquishness could be due to an aberrant developmental process that affects all of the neighboring brain regions that are highly connected to these structures. Alternatively, processes affecting plasticity such as learning or training in association with repeated and excessive recruitment can bring about an increased covariance within the clustered regions, though such effects have not been directly demonstrated so far. In this context, the increased segregation of these regions can be also interpreted as a compensatory process (Griffa et al.
2013).
Patients had a reduction in segregation and local efficiency in primary sensory regions such as the structures around the central sulcus and occipital cortex. These findings are somewhat unexpected given that whole brain univariate approaches have hitherto not identified prominent folding deficits in these regions. Structural covariance studies in adolescents report a developmental reduction in the local efficiency of primary sensory regions (Alexander-Bloch et al.
2013a) In general, primary sensory (and motor) regions show much more segregated developmental pattern than association cortices in healthy controls (Raznahan et al.
2011; Li et al.
2013). Notably, patients with early onset schizophrenia show accelerated grey matter loss around the central sulcus (Gogtay et al.
2004). These changes have been ascribed to disturbances in synaptic pruning in schizophrenia, though no direct evidence exists to date to confirm or refute this notion. In-so-far as the tension related to neuronal connections determines cortical folding (Essen
1997; Hilgetag and Barbas
2006), excessive pruning of such connections, if it indeed occurs in schizophrenia, can also alter the anatomical covariance patterns in gyrification.
In the present sample, healthy controls had a high-degree centrality involving several cingulate regions. This suggests that the gyrification of the cingulate cortex correlates with a large number of other brain regions in healthy state, but not in schizophrenia. Consistent with these observations, alterations in cingulate morphology have been reported previously in schizophrenia (Wheeler and Harper
2007; Baiano et al.
2007). Furthermore, the visual inspection of the modularity and degree distributions patterns (Fig.
3) reveals that various midline structures show a reduction in their degree of covariance in patients. In addition, the inferior temporal and superior parietal sulcus also show reduced degree, while there were no regions showing increased degree in patients. This implies that the pathophysiological process that characterizes gyrification defects in schizophrenia predominantly reduces the overall structural covariance patterns.
Patients with greater illness severity display a more segregated pattern of covariance especially for the right anterior insula extending to include the right superior temporal gyrus and inferior frontal gyrus (pars triangularis). These structures are right homologues of critical language-processing regions that are repeatedly implicated in the generation of psychotic symptoms (Jardri et al.
2011; Li et al.
2012; Modinos et al.
2013). This is consistent with a recent observation indicating that abnormal fronto-temporo-insular gyrification may predict poor outcome in psychosis (Palaniyappan et al.
2013b). Taken together, these results support a speculation that in the presence of a well-coordinated development of the peri-sylvian regions, especially the anterior insula, the longitudinal course of schizophrenia could turn out to be more favorable. Understanding the factors that influence the maturation of these brain regions in health and disease states could provide opportunities to modify illness trajectories in future. To our knowledge, this is the first time that the topological properties of the structural graph networks from patients with schizophrenia are shown to be related to severity of clinical symptoms (van den Heuvel et al.
2010; Fornito et al.
2012), highlighting the utility of studying gyrification patterns in this illness.
Our study has a number of strengths. We adopted a whole brain approach to study structural covariance, instead of seed-based or subset-based approaches. This data-driven approach obviates the need for generating region-based hypotheses that are likely to be tenuous given the heterogeneity of results from previous studies in schizophrenia (White and Gottesman
2012). We defined nodes based on parcellations derived from a sulcogyral atlas that is based on anatomical boundaries of consistent sulci and gyri (Destrieux et al.
2010). There is no consensus on the choice of nodes for connectomic studies; the absolute values of small-world properties have been reported to vary significantly according to the size of the nodes (Zalesky et al.
2010; Fornito et al.
2013). Nevertheless, the use of a common spatial scale for group comparison has been shown to provide valid results (Evans
2013). In line with other anatomical covariance studies, we used population-level variability to determine topological properties for each group; as a result, topological measures at an individual level were not available to relate to symptom burden or cognitive scores. Nevertheless, we have used a median-split approach to study subgroups with varying clinical severity to establish the relationship with symptom burden. We studied a sample of medicated patients; antipsychotic use is reported to be associated with structural changes in schizophrenia (Ho et al.
2011), though at present there is no evidence that suggests that cortical folding patterns are affected by the use of antipsychotics. Our investigation of the linear association between antipsychotic dose and topological properties (Supplementary Material) suggested that though antipsychotics have some influence on the covariance pattern, this is not sufficient to affect the topological properties of the gyrification-based network. However, our findings must be interpreted cautiously until replicated in a sample of untreated patients.
Abnormalities in the covariance patterns involving the frontal cortex are of particular interest for the study of schizophrenia. Phylogenetic variations in sulcal patterns predominantly involve the frontal cortex, suggesting a link between cognitive/linguistic evolution and cortical folding (Zilles et al.
2013). To our knowledge, there are no comparative studies on the structural covariance of gyrification across species. Within-species variations in gyrification appear to be influenced by factors that exert region-specific effects on the brain (Kochunov et al.
2010). These factors influence the ontogeny of cerebral gyrification and may relate to the structural covariance. Broadly, we can group the factors influencing gyrification as fetal events, early infantile events and later developmental events (including the maturational changes related to puberty). Distribution of regional axonal tension (in fetal, infantile or later life periods), differential rates of surface expansion (especially in fetal and infantile period) and genetic variations in cellular proliferation and migration (occurring in fetal life) could influence the establishment of specific patterns of structural covariance in cortical folding (Zilles et al.
2013).
Interestingly, a substantial portion of within-species variance in gyrification patterns appears to be non-genetic (Rogers et al.
2010; Zilles et al.
2013), highlighting the role of non-heritable, possibly late maturational events. Synchronized recruitment of brain regions can
induce structural covariance through use-dependent synaptogenesis even in the absence of direct axonal connectivity (Evans
2013). Mindfulness meditators, who repeatedly use the interoceptive brain regions such as the insula and the salience network structures, show increased insular folding in direct relationship to the duration of their meditative practice (Luders et al.
2012). Our cross-sectional design precludes further parsing of the factors operating at different stages of life to influence the structural covariance. Nevertheless, it is important to note that the largest transformation in cortical folding patterns in one’s lifespan occurs during fetal life (White et al.
2010). During this time, a well-coordinated spatial relationship in sulcogyral development is seen across the brain (Habas et al.
2012). This suggests that a substantial amount of the structural covariance observed in later life is determined during this period. Events that disrupt fetal neurodevelopment significantly affect cortical folding (Dubois et al.
2008a;
b), and its relationship with white matter connectivity (Lodygensky et al.
2010; Melbourne et al.
2014), affecting later functional capacity in adult life (Kesler et al.
2006; Dubois et al.
2008a). This supports our interpretation that the prominent reshuffling of the gyrification covariance (Fig.
2) is an early developmental aberration in brain connectivity in patients. To conclusively differentiate the early developmental influences from the later life events, a prospective study of prenatal or newborn cohorts followed up till late adult life, ideally even after the onset of schizophrenia, is required.
The use of graph theoretical approach has revealed novel insights about the complex connected system of relationship among different brain regions in health and its aberration in disease states (Johansen-Berg
2013). Several years of neuroimaging research in schizophrenia with ‘lesional’ approaches has uncovered some, but most of the pathophysiological processes associated with the clinical picture of schizophrenia remains yet to be discovered (Carpenter et al.
1993; Tandon et al.
2008). This approach, when used to study cortical folding patterns for the first time, reveals significantly altered covariance, suggesting abnormalities in the developmental synchrony of connected brain regions in patients. In particular, our study has identified specific regions, whose coordinated maturation with the rest of the brain may be specifically altered in schizophrenia, contributing to greater severity. Longitudinal and interventional (e.g. motivated cognitive training) studies of cortical gyrification that specifically focuses on these brain regions are likely to uncover a more complete picture of the structural substrate of dysconnectivity that characterizes schizophrenia. Given the emerging evidence implicating the importance of cortical folding defects in treatment response future studies utilizing gyrification-based covariance approaches could aid in further characterizing poor-outcome phenotype in psychotic disorders.