Cortical Thickness and Connectivity in Schizophrenia

What does the brain look like in schizophrenia? This apparently simple question has been the focus of intense research effort since the advent of modern neuroimaging, with the introduction of computerized tomography (CT), and then MRI, in the 1980s.

It is salutary to recall that schizophrenia was generally regarded as a functional psychosis, rather than an organic psychosis like dementia, at that time, about 30 years ago. One implication of this diagnostic dogma was that the brain was not expected to look structurally (organically) abnormal in schizophrenia. The first CT studies generated intense controversy by providing disruptive evidence for significant enlargement of the ventricles (1). It is a measure of the theoretical impact of neuroimaging in psychiatry that the prior concept of functional psychosis has been largely abandoned in the face of overwhelming evidence of structural brain imaging abnormalities in schizophrenia. It is now beyond reasonable doubt that the brain does not look structurally normal in schizophrenia, but it remains an open question how best to characterize and interpret the abnormalities disclosed by contemporary MRI research (2).

Advances in statistical theory and computational processes have made it routine to estimate any regional MRI parameter, like cortical thickness, at hundreds of well-defined cortical areas and subcortical nuclei. Whole-brain maps of MRI abnormalities associated with schizophrenia have been published in a standard anatomical format and can be quantitatively meta-analyzed (3). It is clear that there are anatomically widespread or spatially distributed abnormalities in macrostructural MRI markers, like volume and thickness, measured over multiple voxels representing each region. There is also increasing evidence that microstructural MRI markers, like magnetization transfer ratio, can provide diagnostically relevant information about the tissue composition of each voxel (4), allowing a finer-grained characterization of brain differences between case and control subjects (5).

The scale and richness of contemporary MRI experiments have in turn driven an increasing theoretical focus on understanding the brain as a network. The concept of psychosis as a brain network disorder was first and vaguely formulated by Carl Wernicke in the 19th century, then eclipsed by other theoretical accounts of psychosis until the 1990s, when it resurfaced in the form of functional connectivity analysis of human positron emission tomography and MRI data (6). Whole-brain MRI mapping of schizophrenia as a dysconnectivity syndrome, or brain network disorder, has now largely supplanted the prior concept of schizophrenia as a cortical lesion syndrome, attributable to abnormality of one or a few regions of interest (7).

In this context, the study by Wannan et al. (8), published in this issue of the Journal, makes a timely and important contribution to the field for two main reasons. First, it is based on a large, clinically heterogeneous sample. The University of Melbourne has a strong track record of schizophrenia research, and this study reports MRI data on 270 patients with clinically, well-characterized psychotic disorder—categorized as first-episode psychosis, chronic schizophrenia, and treatment-resistant schizophrenia—as well as data on 279 healthy control subjects. This allowed the investigators to assess the replicability of case-control differences between the three schizophrenia cohorts (and carefully matched control subjects) and to describe variation in the degree of case-control differences between these cohorts.

Wannan et al. found widespread case-control differences in cortical thickness in all three patient cohorts. Thickness was consistently reduced in all cortical areas but especially in the frontal, temporal, insular, and cingulate cortex. Descriptively, and as expected (9), more cortical regions were significantly thinner in patients with treatment-resistant schizophrenia compared with healthy control subjects than in patients with chronic schizophrenia or first-episode psychosis compared with control subjects. Unfortunately, differences in MRI scanners used to collect data on the three patient cohorts precluded direct quantitative comparison of case-control effect sizes between the cohorts.

A second strength of this study is that the MRI data were rigorously analyzed to address the relationship between cortical thickness, a regional measure of brain structure, and structural covariance, a network measure of the anatomical connectome. The principal finding was that areas where cortical thickness was reduced in schizophrenia had high structural covariance in both patients and control subjects. For example, regions of temporal and frontal cortex that were thinner in schizophrenia patients had stronger interregional correlation of cortical thickness (putatively stronger anatomical connectivity) in healthy control subjects. This is the basis for the authors’ plausible claim that the topography of cortical thickness reductions in schizophrenia “is shaped by [normal] network topology” (8).

However, the exact causal mechanism that links the anatomical pattern of cortical thickness changes in schizophrenia to normal brain network organization remains unclear. In MRI studies of neurodegenerative disorders, distributed local changes in cortical structure have been linked to network topology by various biologically plausible mechanisms, including trans-synaptic propagation of a neurotoxic agent from one area to other areas (10). Wannan et al. hypothesize, likewise, that “cortical networks may propagate local pathological processes involved in schizophrenia (e.g., aberrant neuronal signaling, disruption to neurotransmitter systems)” (8). However, as they acknowledge, this interpretation comes with many caveats. Most fundamentally, it is a limitation of the study design that there are no longitudinal or repeated measures on the participants over time. Such a longitudinal data set would provide more compelling evidence that progressive cortical thinning in schizophrenia was indeed shaped or constrained by prior development of network topology (11).

One negative set of results is also worth highlighting. There was no significant case-control difference in structural covariance for the treatment-resistant or chronic schizophrenia group. This could be regarded as a true negative, indicating that brain network connectivity is not abnormal in (more severe) schizophrenia, or a false negative, indicating some limitation in the statistical power of the study. I suspect that this is a false negative, reflecting the limitations of structural covariance analysis as a method for modeling schizophrenia-related abnormalities of the anatomical connectome (12).

Structural covariance was simply estimated by the interregional correlation of a single MRI parameter (e.g., thickness) that was measured in a pair of regions of the brain in multiple individuals. Thus, for each schizophrenia cohort (and the healthy control group), there was only one structural covariance matrix, representing brain anatomical connectivity “on average” over all individuals. This provided no opportunity to find individual differences in network topology that might be related to individual differences in clinical or psychological status. It is also likely suboptimal that structural covariance analysis was based on the single parameter of cortical thickness. The biological interpretation of cortical thickness, measured through the prism of MRI, is unsettled. Does loss of thickness on an MRI scan represent a loss of cortical neurons or synaptic pruning? Or could it be a partial-volume effect representing denser myelination of deeper cortical layers in schizophrenia? These questions about the cellular substrates of the regional signals (cortical thickness) make it difficult to be confident about the cellular substrates of the network signal (covariance) derived from them. As Wannan et al. point out, it may be important in the future to explore the potential of methods that use one or more microstructural MRI parameters as the basis for assessing similarity and connectivity between cortical areas, rather than relying exclusively on interregional correlation of cortical thickness (13, 14).

Neuroimaging research studies of this scale and rigor are important in continuing to deepen our understanding of what the brain looks like in schizophrenia and, intriguingly, how the MRI fingerprint of schizophrenia may be related to configuration of the human connectome.