Distinct pattern of brain structural deficits in subsyndromes of schizophrenia delineated by psychopathology

Abstract

Brain morphological changes are among the best-studied potential endophenotypes in schizophrenia and linked to genetic liability and expression of disease phenotype. Yet, there is considerable heterogeneity across individual subjects making its use as a disease-specific marker difficult. In this study we consider psychopathological variability of disease phenotype to delineate subsyndromes of schizophrenia, link them to distinct brain morphological patterns, and use a classification approach to test specificity of achieved discrimination. We first applied voxel-based morphometry (VBM) to compare 99 patients with DSM-IV schizophrenia (stable psychopathology and antipsychotic medication) with 113 matched healthy controls, then delineated three subgroups within the patient cohort based on psychopathology pattern and compared differential patterns of grey matter abnormalities. Finally, we tested accuracy of assigning any individual MRI scan to either the control group or any of the three patient subgroups. While VBM analysis showed overlap of brain structural deficits mostly in prefrontal areas, the disorganised subsyndrome showed stronger deficits in medial temporal and cerebellar regions, the paranoid/hallucinatory subsyndrome showed additional effects in the superior temporal cortex, and the negative subsyndrome showed stronger deficits in the thalamus. Using an automated algorithm, we achieved 95.8% accuracy classifying any given scan to one of the subgroups. Patterns of psychopathology are meaningful parameters in reducing heterogeneity of brain morphological endophenotypes in schizophrenia.

Introduction

The clinical heterogeneity of schizophrenia is a major obstacle in identifying both phenotypes suitable as potential disease markers as well as studying the complex genetics of this disorder. Among the most robust biological markers of pathology in schizophrenia are alterations in brain structure as detected with magnetic resonance imaging (MRI). Grey matter reductions are already observed at the onset (Steen et al., 2006) of the disorder and possibly at prodromal stages (Pantelis et al., 2003) while showing modest disease-related subsequent progression (Weinberger and McClure, 2002). Studies in siblings (Honea et al., 2008) and in twins (Hulshoff Pol et al., 2004, Hulshoff Pol et al., 2006) furthermore demonstrate genetic influence on grey matter reductions. The most consistently reported regional abnormalities beside the enlargement of the lateral ventricles are grey matter reductions in the medial temporal lobe (hippocampus and amygdala), thalamus, prefrontal cortex, superior temporal cortex (Honea et al., 2005, Wright et al., 2000), and more recently also the cerebellum (Andreasen and Pierson, 2008). While this underlines the potential usefulness as an endophenotype for this disorder (Goldman et al., 2008), there are also limits to its use as a biological marker for schizophrenia. Using conventional volumetric or morphometric approaches, none of the single regional alterations on its own is either sensitive or specific enough to distinguish patients from control subjects.

Strategies aimed at overcoming these difficulties have included studies within the schizophrenia spectrum disorders (Hazlett et al., 2008, Takahashi et al., 2006) as well as comparison to other disorders sharing symptoms or clinical features of schizophrenia, such as (psychotic) bipolar disorder (Kasai et al., 2003, McDonald et al., 2006, McIntosh et al., 2006). Hippocampal volume reductions, for example, have been hypothesised to be present in schizophrenia but less so in bipolar disorder (McDonald et al., 2006). These studies have demonstrated some overlap both in the regional distribution as well as extent of local grey matter changes. This limitation might, however, be overcome by investigating the pattern of regional changes. More recent studies have shown that pattern classification techniques might be useful to enhance specificity of morphometric findings, making use of the set of changes across the entire brain (Davatzikos et al., 2005, Kawasaki et al., 2007, Soriano-Mas et al., 2007, Yushkevich et al., 2005). The advantage of this approach would be the use of multi-regional information – not limited to one single area – in developing a more specific brain structural signature of schizophrenia.

While the mentioned studies in schizophrenia and related disorders have produced important data, they are limited through their use of categorical diagnostic approaches. Thus, inclusion of additional information related to phenotypic variation within a sample might be useful to form subgroups, which are more homogeneous and could then be subject to classification analysis to prove or disprove sufficient specificity for brain structure as a biological marker.

Only a few structural MRI studies have addressed the issue of subgroups or subsyndromes within schizophrenia, even despite the availability of the subtyping according to DSM-IIIR or DSM-IV diagnostic criteria. For example, abnormalities in cortical folding have been shown to be more pronounced in the disorganised subtype of schizophrenia (Sallet et al., 2003a), although there was no indication of subgroups within patients in another study on cortical thickness (Lawyer et al., 2008), while yet another study dividing patients in paranoid vs. non-paranoid schizophrenia has given evidence for the negative symptom dimension to be related to increased rightward structural asymmetry (Sallet et al., 2003b).

In this study, we implemented such an approach by applying a classification analysis (based on voxel-based morphometric analysis of whole-brain MRI) on both a comparison of schizophrenia patients and healthy controls as well as subdividing the schizophrenia sample into subgroups based on symptoms. We tested whether the pattern of brain structural changes (i.e. combination of regionally distributed changes) would provide sufficiently accurate classification of a given brain scan from the cohort to be assigned to either schizophrenia or healthy control groups, and (for patients) to one of three schizophrenia subgroups. These three subgroups were formed on the basis of factor analysis applied to cross-sectional psychopathology. The rationale for this approach to forming subgroups was based on previous studies on robustness of three-factor models in schizophrenia (Cuesta and Peralta, 1995b, Peralta et al., 1997) as well as the fact that this grouping most closely matches the already established clinical subtypes within the DSM-IV and ICD-10 diagnostic systems. Secondly, we challenged the notion that there is a core pattern of changes independent of clinical or phenotypic variability, assuming that the most consistently reported structures in schizophrenia pathophysiology (hippocampus, thalamus, and dorsolateral prefrontal cortex) would be altered in all subgroups, while other structural changes (e.g. in the superior temporal cortex, cerebellum), which have been linked to particular psychopathological phenomena (Gaser et al., 2004), might be altered in only one of the subgroups expressing this particular disease phenotype.