Introduction
Schizophrenia and bipolar disorder are severe psychiatric illnesses with high heritability but to date unknown aetiology (Sullivan et al. Reference Sullivan, Kendler and Neale2003; Nutt & Need, Reference Nutt and Need2014). Although these disorders are nosologically distinct, many patients with bipolar disorder experience psychotic symptoms similar to those found in schizophrenia, and studies indicate shared genetic vulnerability (Lichtenstein et al. Reference Lichtenstein, Yip, Björk, Pawitan, Cannon, Sullivan and Hultman2009). The ‘dysconnectivity hypothesis’, the idea that schizophrenia may result from abnormal neural connections, was formulated early in the history of psychiatry but has more recently been refined and extended to other disorders (Friston & Frith, Reference Friston and Frith1995; Pettersson-Yeo et al. Reference Pettersson-Yeo, Allen, Benetti, McGuire and Mechelli2011; Fornito & Bullmore, Reference Fornito and Bullmore2015). Brain-imaging studies have found evidence of altered connectivity (Moran et al. Reference Moran, Tagamets, Sampath, O'Donnell, Stein, Kochunov and Hong2013; Baker et al. Reference Baker, Holmes, Masters, Yeo, Krienen, Buckner and Ongur2014), but the neurobiological causes are yet to be determined.
Myelination is a key factor for efficient neural signalling in the brain. Myelin increases the speed and reliability of the nerve signal by allowing for saltatory axonal conduction, provides structural and trophic support and prevents aberrant sprouting of nerve connections (Baumann & Pham-Dinh, Reference Baumann and Pham-Dinh2001). Therefore, abnormal myelin repair or maintenance is a candidate mechanism for brain network dysfunction, and a potential cause of schizophrenia (Bartzokis, Reference Bartzokis2002; Davis et al. Reference Davis, Stewart, Friedman, Buchsbaum, Harvey, Hof, Buxbaum and Haroutunian2003). Indeed, evidence of myelin-related abnormalities has emerged from genetic association studies (Hakak et al. Reference Hakak, Walker, Li, Wong, Davis, Buxbaum, Haroutunian and Fienberg2001; Rietkerk et al. Reference Rietkerk, Boks, Sommer, de Jong, Kahn and Ophoff2009; Goudriaan et al. Reference Goudriaan, de Leeuw, Ripke, Hultman, Sklar, Sullivan, Smit, Posthuma and Verheijen2014; Yu et al. Reference Yu, Bi, Liu, Zhao, Zhang and Yue2014) and studies of post-mortem brain tissue from patients with schizophrenia and bipolar disorder (Tkachev et al. Reference Tkachev, Mimmack, Ryan, Wayland, Freeman, Jones, Starkey, Webster, Yolken and Bahn2003; Uranova et al. Reference Uranova, Vikhreva, Rachmanova and Orlovskaya2011).
Structural brain-imaging studies of myelin in schizophrenia have often focused on white-matter properties, including diffusion (Agartz et al. Reference Agartz, Andersson and Skare2001), magnetization transfer (Foong et al. Reference Foong, Maier, Barker, Brocklehurst, Miller and Ron2000; Du et al. Reference Du, Cooper, Thida, Shinn, Cohen and Ongur2013) and volume (Haijma et al. Reference Haijma, Van Haren, Cahn, Koolschijn, Hulshoff Pol and Kahn2013). Yet, the cerebral cortex is also myelinated. Intracortical myelin shows a consistent regional pattern with higher myelin content in primary sensory, motor, and early association cortices, and lower myelin content in higher level association cortices (Glasser & Van Essen, Reference Glasser and Van Essen2011). The degree of intracortical myelination has been associated with cognitive performance (Blackmon et al. Reference Blackmon, Halgren, Barr, Carlson, Devinsky, DuBois, Quinn, French, Kuzniecky and Thesen2011; Grydeland et al. Reference Grydeland, Walhovd, Tamnes, Westlye and Fjell2013, Reference Grydeland, Westlye, Walhovd and Fjell2015), and is considered to be critical for the optimization of brain function (Bartzokis, Reference Bartzokis2011).
In T1-weighted magnetic resonance (MR) images, signal intensity is derived from the longitudinal relaxation of hydrogen protons in the underlying tissue after excitation with a radio-frequency pulse. A major determinant of this intensity signal is the cholesterol in myelin (Koenig et al. Reference Koenig, Brown, Spiller and Lundbom1990; Koenig, Reference Koenig1991), and strong correlations are reported between T1 signal intensity and myelin content in healthy and lesioned white matter, and spinal cord tissue (Mottershead et al. Reference Mottershead, Schmierer, Clemence, Thornton, Scaravilli, Barker, Tofts, Newcombe, Cuzner, Ordidge, McDonald and Miller2003; Schmierer et al. Reference Schmierer, Scaravilli, Altmann, Barker and Miller2004, Reference Schmierer, Wheeler-Kingshott, Tozer, Boulby, Parkes, Yousry, Scaravilli, Barker, Tofts and Miller2008). Similarly, in cortical grey matter, the cross-sectional profile of T1 signal intensity values based on high-resolution MRI corresponds closely with histologically based myelin profiles (Eickhoff et al. Reference Eickhoff, Walters, Schleicher, Kril, Egan, Zilles, Watson and Amunts2005), and this signal is mainly due to myelin rather than iron content (Stuber et al. Reference Stuber, Morawski, Schafer, Labadie, Wahnert, Leuze, Streicher, Barapatre, Reimann, Geyer, Spemann and Turner2014).
Since signal intensity in both grey and white matter is closely related to myelin content, the grey/white-matter contrast, although not a direct measure of myelin, could be sensitive to illness-related alterations of myelination. The grey/white-matter contrast is defined as the difference, or ratio, between T1 grey- and white-matter signal intensities. This can be computed as a local measure along the cortical surface. In neurodevelopmental stages prior to adulthood, grey- and white-matter signal intensities increase with age, with white matter showing a more protracted development in several regions (Westlye et al. Reference Westlye, Walhovd, Dale, Bjornerud, Due-Tonnessen, Engvig, Grydeland, Tamnes, Ostby and Fjell2010). In both normal ageing and neurological illness such as Alzheimer's disease the grey/white-matter contrast shows a decline which appears to be primarily driven by a decrease in white-matter intensity (Magnaldi et al. Reference Magnaldi, Ukmar, Vasciaveo, Longo and Pozzi-Mucelli1993; Salat et al. Reference Salat, Lee, van der Kouwe, Greve, Fischl and Rosas2009, Reference Salat, Chen, van der Kouwe, Greve, Fischl and Rosas2011; Westlye et al. Reference Westlye, Walhovd, Dale, Espeseth, Reinvang, Raz, Agartz, Greve, Fischl and Fjell2009). Few studies have examined this measure in psychiatric populations. Kong et al. (Reference Kong, Herold, Stieltjes, Essig, Seidl, Wolf, Wüstenberg, Lässer, Schmid, Schnell, Hirjak and Thomann2012) reported widespread decreases in T1 weighted grey/white-matter tissue contrast in patients with schizophrenia compared to age-matched healthy controls. These findings remained significant after cortical thickness was taken into account, and a follow-up study suggested the findings to be at least partly independent of gyrification and area (Kong et al. Reference Kong, Herold, Zöllner, Salat, Lässer, Schmid, Fellhauer, Thomann, Essig, Schad, Erickson and Schröder2015). Increased tissue contrast between whole-brain grey matter and white matter in schizophrenia and bipolar disorder has also been reported, but from a global contrast measure computed as the difference between averages of all voxel intensities in grey and white matter which therefore need not be comparable (Bansal et al. Reference Bansal, Hao, Liu, Xu, Liu and Peterson2013). Thus, it is not clear if and how the grey/white-matter contrast is altered in schizophrenia and bipolar disorder.
In the present study, we investigated differences in grey/white-matter contrast between patients with schizophrenia, bipolar disorder and healthy controls. Since regional cortical thickness, area, and gyrification previously have been found to differ between these groups (Rimol et al. Reference Rimol, Hartberg, Nesvag, Fennema-Notestine, Hagler, Pung, Jennings, Haukvik, Lange, Nakstad, Melle, Andreassen, Dale and Agartz2010, Reference Rimol, Nesvag, Hagler, Bergmann, Fennema-Notestine, Hartberg, Haukvik, Lange, Pung, Server, Melle, Andreassen, Agartz and Dale2012; Nesvag et al. Reference Nesvag, Schaer, Haukvik, Westlye, Rimol, Lange, Hartberg, Ottet, Melle, Andreassen, Jonsson, Agartz and Eliez2014) and may potentially influence the grey/white-matter contrast, we tested if the contrast was independent of these properties. To examine potential confounders which may influence brain structure, we tested for association with antipsychotic medication, cigarette smoking and alcohol use. We examined the post-hoc hypothesis that clusters which showed differences in contrast between patients and controls would be associated with hallucinations and delusional thinking. Finally, we tested if the grey/white-matter contrast differed between bipolar disorder patients with and without a history of psychosis.
Method and materials
Study sample
Participants were recruited between 2004 and 2012 as part of the Thematically Organized Psychosis (TOP) study, conducted by the Norwegian Centre for Mental Disorders Research (NORMENT). Patients with either a psychotic disorder or bipolar disorder were recruited consecutively from psychiatric in- and outpatient treatment facilities in the Oslo region in Norway. Inclusion criteria for the TOP study were age between 18 and 65 years, no history of neurological disorder or moderate to severe head injury, and IQ > 70. For the present study, patients with schizophrenia (n = 163), schizoaffective disorder (n = 29), and schizophreniform disorder (n = 22) were included in the schizophrenia group. Patients with bipolar I disorder (n = 112), bipolar II disorder (n = 63), or bipolar disorder NOS (n = 10) were included in the bipolar disorder group. Healthy controls were recruited from the same catchment area as the patients based on the Norwegian national population registry. In addition to the inclusion criteria listed for patients above, healthy controls were included if they had no current or previous psychiatric disorders, no family history of severe mental disorders, and did not meet criteria for alcohol or substance dependency, or any use of cannabis within the last 3 months prior to assessment. Demographic and clinical variables for the sample are found in Table 1.
Table 1. Demographical and clinical data
WASI, Wechsler Abbreviated Scale of Intelligence; PANSS, Positive and Negative Syndrome Scale; n.a., not applicable; n.s., not significant.
Missing values: Handedness: n = 1 (SCZ: n = 1). Ambidextrous: n = 5 (SCZ: n = 0; BD: n = 2; HC: n = 2). Education: n = 79 (SCZ: n = 28; BD: n = 16; HC: n = 35). WASI: 91 (SCZ: n = 32; BD: n = 21; HC: n = 38). AOO and duration of illness: n = 12 (SCZ: n = 7; BD: n = 5). Medication status: n = 15 (SCZ: n = 7; BD: n = 8). Medication dose, AP: n = 4 (SCZ: 3; BD: 1). PANSS: n = 8 (SCZ: n = 6; BD: n = 2). Alcohol consumption: n = 12 (SCZ: n = 5; HC: n = 7). Illicit drug use: n = 10 (SCZ: n = 8; BD: n = 2). Cigarette smoking: n = 101 (SCZ: n = 8; BD: n = 12; HC: n = 81).
Age of onset was defined as the first occurrence of either psychotic, manic, or hypomanic symptoms. Information about alcohol consumption was recorded during clinical assessment interviews for the last 2 weeks prior to clinical assessment for patients and the last 2 weeks prior to cognitive testing for healthy controls. Illicit drug use refers to use of any illicit drug during the last two years prior to assessment. Data for cigarette smoking were obtained at clinical assessment for patients and by phone interviews for healthy controls.
All participants were informed about the study orally and in writing. The study physician or clinical psychologist evaluated if the information was fully understood before the patients gave their consent. The study was approved by the Regional Committee for Medical Research Ethics and the Norwegian Data Inspectorate.
Diagnostic and clinical assessment
Assessments were performed by trained physicians, psychiatrists or clinical psychologists. Diagnoses were verified using the Structured Clinical Interview for the DSM-IV Axis I disorders (SCID-IV; First et al. Reference First, Spitzer, Gibbon and Williams2002). Current symptoms at the time of assessment were rated using the Positive and Negative Syndrome Scales (PANSS; Kay et al. Reference Kay, Fiszbein and Opler1987). Information about medication use at the day of MRI scan was obtained by interview, and supplemented by chart review and information from clinical assessment. Current dose of medication at MRI day was converted to chlorpromazine equivalents (CPZ; Andreasen et al. Reference Andreasen, Pressler, Nopoulos, Miller and Ho2010). For details about assessment of alcohol use, drug use and smoking, see Table 1. The Wechsler Abbreviated Scale for Intelligence (WASI; Wechsler, Reference Wechsler2007) was used to measure general ability level. For details about assessment methods used in the TOP study, we refer to Engh et al. (Reference Engh, Friis, Birkenaes, Jonsdottir, Klungsoyr, Ringen, Simonsen, Vaskinn, Opjordsmoen and Andreassen2010).
MRI acquisition
MR images were acquired on a 1.5-T Siemens Sonata scanner (Siemens Medical Solutions, Germany) equipped with a standard head coil. After a conventional three-plane localizer, two sagittal T1-weighted magnetization-prepared rapid gradient echo volumes were acquired with the Siemens tfl3d1_ns pulse sequence (echo time = 3.93 ms, repetition time = 2730 ms, inversion time = 1000 ms, flip angle = 7°, field of view = 24 cm, voxel size = 1.33 × 0.94 × 1 mm, number of partitions = 160). Acquisition parameters were optimized for increased grey/white-matter contrast. There was no scanner hardware upgrade during the study period, but routine software updates were performed. Patients and controls were scanned interchangeably to avoid bias related to scanner drift. All scans were evaluated by a neuroradiologist and 12 (schizophrenia: n = 7; bipolar disorder: n = 2; healthy controls: n = 3) were excluded due to pathological findings.
MRI processing
FreeSurfer 5.3.0 (http://surfer.nmr.mgh.harvard.edu/) was used to automatically create three-dimensional models of the cortical surfaces (Dale & Sereno, Reference Dale and Sereno1993; Dale et al. Reference Dale, Fischl and Sereno1999; Fischl & Dale, Reference Fischl and Dale2000; Segonne et al. Reference Segonne, Dale, Busa, Glessner, Salat, Hahn and Fischl2004; Reuter et al. Reference Reuter, Rosas and Fischl2010). In summary, the procedure consists of within-subject motion correction and averaging, intensity corrections to account for field inhomogeneities, skull stripping and tissue classification. A triangular tessellation is then overlaid and smoothly deformed to create accurate and robust representations of the grey/white and pial surfaces. Cortical thickness is measured as the shortest distance from the white to the pial surface at each vertex. Importantly, since spatial intensity gradients are used to place the surfaces, measurements of cortical thickness, area or gyrification, derived from the local gyrification index (LGI; Schaer et al. Reference Schaer, Cuadra, Tamarit, Lazeyras, Eliez and Thiran2008), do not rely on absolute signal intensity. Further description of the FreeSurfer processing steps can be found in Fischl (Reference Fischl2012). All surfaces were visually inspected and minor manual editing was performed according to standard FreeSurfer quality control procedures. Since we expected the grey/white-matter contrast to be sensitive to image quality, we computed the signal-to-noise ratio (SNR) of each image using the wm_anat_snr tool and re-inspected every image with SNR values in the lower 5th percentile. Nine scans were excluded due to minor artefacts or poor image quality.
Sampling of signal intensity and calculation of grey/white contrast
For each subject, signal intensities in grey and white matter were sampled from the non-uniform intensity normalized volume (nu.mgz) at each vertex along the grey/white boundary using mri_vol2surf. The percentage difference between white- and grey-matter intensity values at each vertex was computed as
$$100 \,\ast\,\left({\rm white} -{\rm grey}\right)/\left[ {\left( {{\rm white} + {\rm grey}} \right)/2} \right],$$
Before our final intensity sampling method was decided on, we performed a series of tests. Using correlation maps and preliminary analyses covarying for thickness, we determined which of three different sampling parameters yielded results least dependent on cortical thickness. First, we sampled voxel intensities at a fractional distance of 30% into the cortical ribbon from the grey/white boundary and a fixed distance of 1 mm into subcortical white matter (default settings in the FreeSurfer pctsurfcon script). Second, we sampled intensity values from both grey and white matter at a fixed distance of 1 mm from the grey/white boundary. Third, we averaged voxel intensities between the grey/white surface to 60% into the cortical ribbon and 1.5 mm into subcortical white matter (Fig. 1). These last parameters were found to be the most independent of cortical thickness and were therefore chosen for the final analysis. For maps displaying within-group means and results of using different sampling methods (see Supplementary Figs S1–S3). Before statistical analyses, all surface maps were smoothed using a Gaussian kernel of 20 mm full width at half maximum (FWHM).
Fig. 1. Illustration of the intensity sampling method used in the present study: The yellow and red lines represent the white and pial surfaces, respectively, obtained from the standard FreeSurfer processing procedure (recon-all). Blue lines represent distances from the grey/white surface (yellow line) where intensity values were sampled at each vertex. Cortical intensity values were not sampled from the whole cortical ribbon, but at points ranging from 0% to 60% from the grey/white surface and into the cortical ribbon. This was done to avoid contamination of voxels containing cerebrospinal fluid. At each vertex, intensity values sampled at different distances from the grey/white surface were averaged to obtain a single measure of grey-matter intensity. Similarly, blue lines in white matter represent distances from 0 to 1.5 mm from the grey/white surface. Intensity values at different distances from the grey/white surface were averaged to obtain a single measure of white-matter intensity. Note that the image shows a 2D slice where 3D surfaces are overlaid. Therefore, blue lines may appear to be unevenly distanced whereas the sampling points are not.
Statistical analysis
Group differences in demographical and clinical variables were tested using t or χ2 statistics. In the main analyses, General Linear Models (GLMs) with age and sex as covariates were fitted at each vertex along the cortical surface. Since age and sex differed significantly, main analyses were also performed without adjusting for these variables, yielding highly similar results (for details, see Supplementary Fig. S4). The percentage contrast between grey and white matter, described above, was used as dependent variable. Group differences were tested using F tests where these contrasts were selected: schizophrenia v. control group, bipolar disorder v. control group, and schizophrenia v. bipolar disorder. To correct for multiple testing, we used false discovery rate (FDR) set at 5%. We also tested age × group interaction contrasts. To account for differences in cortical thickness, area or gyrification, we fitted additional models where these measures were included as per-vertex regressors. We also generated surface maps of the vertex-wise correlation between cortical contrast and thickness, area and gyrification. To examine the influence of antipsychotic medication, we entered CPZ for current dose of antipsychotic medication and dose of second-generation antipsychotics into a GLM that was fitted within the patient sample (n = 380). Similar to the previous analyses, we adjusted for age and sex, but since patients with bipolar disorder and schizophrenia were grouped together, we also adjusted for diagnosis in these analyses. Subgroup analyses within bipolar disorder were done by contrasting patients with lifetime psychotic symptoms to patients with no history of psychosis and patients with bipolar I and II disorder using GLM adjusted for age and sex. We also contrasted patients with bipolar I disorder and healthy controls. To examine potential confounders, we added alcohol use (mean alcohol units/week) and cigarette smoking (dichotomous variable) as additional covariates in the linear models testing for group differences. For post-hoc analyses of association with symptoms, we selected the PANSS P1 (‘delusional thinking’) and P3 (‘hallucinatory behaviour’) items. Ratings were grouped into symptom absent (1), present (4–7) or subthreshold (2–3). Contrasts between groups were examined using ANCOVA adjusted for age, sex, and diagnosis.
Results
Demographic and clinical characteristics
Demographic and clinical characteristics of the sample are found in Table 1.
Increased contrast in schizophrenia compared to healthy controls
Fig. 2 (thresholded p values) show group differences in grey/white contrast between patients with schizophrenia and healthy controls (top row), bipolar disorder v. healthy controls (middle row), and schizophrenia v. bipolar disorder (bottom row). Yellow and red areas represent increased contrast among patients compared to controls, indicating greater difference between grey- and white-matter intensity values in these regions. Effect size maps showing mean difference between the groups are shown in Supplementary Fig. S5.
Fig. 2. Statistical maps of group differences: The top row (1) displays the difference between patients with schizophrenia and healthy controls. The middle row (2) displays the difference between patients with bipolar disorder and healthy controls. The bottom row (3) displays the difference between patients with schizophrenia and patients with bipolar disorder. Yellow/red areas represent increased contrast in patients compared to controls, or in the bottom row, increased contrast in schizophrenia compared to bipolar disorder. The maps were generated from vertex-wise general linear models where the grey/white-matter contrast was the dependent variable and age and gender were included as covariates. Note that correction for multiple testing is performed using false discovery rate (FDR) set at 5% for the top and middle row, while for the bottom row, uncorrected maps (p < 0.01) are shown (for this contrast, no region showed significant differences after FDR correction). SCZ, Schizophrenia, HC, healthy controls; BD, bipolar disorder.
Three bilateral regions showed significantly (p < 0.05, FDR corrected) increased contrast in schizophrenia: (1) Pre- and post-central gyri, including the paracentral lobule. (2) Transverse temporal gyri and posterior regions of the insular cortices. (3) Parieto-occipital regions including the bilateral cunei, lingual gyri as well as the superior parietal and lateral occipital lobules. A small right orbitofrontal region showed reduced contrast in patients with schizophrenia.
In bipolar disorder, increased contrast compared to healthy controls was found in the left central sulcus and small regions in the transverse temporal gyrus and occipital lobe, overlapping with findings in schizophrenia.
Cluster statistics are found in Table 2. Descriptive statistics for intensity values in the significant clusters are found in Supplementary Table S1.
Table 2. Clusters showing altered grey/white-matter contrast ranked by significance value a
a Clusters <150 mm2 are not shown in the table.
b Desikan–Kiliany atlas label name and Talairach coordinates are given for the vertex showing highest significance in each cluster.
There was no significant difference between schizophrenia and bipolar disorder after FDR correction. Uncorrected p-maps (p < 0.01) are shown in the bottom row of Fig. 2. Increased contrast in schizophrenia compared to bipolar disorder was found in small regions in the left post-central gyrus and the left and right parietal lobe.
Effects of including thickness, area and gyrification as vertex-wise regressors
Correlations between the grey/white-matter contrast and area, thickness, and gyrification are shown in Fig. 3. Fig. 4 shows differences between patients with schizophrenia and healthy controls with area, thickness, and gyrification included as vertex-wise regressors. To allow for comparison, all results are shown with a p value based on FDR set conjointly and results from the main analysis are displayed in the top row. Similarly, differences between bipolar disorder patients and healthy controls with area, thickness and gyrification included as vertex-wise regressors are shown in Fig. 5. As can be seen, including area, thickness and gyrification as vertex-wise covariates only marginally affected the main findings. Effect sizes for these contrasts are found in Supplementary Figs S6 and S7.
Fig. 3. Maps displaying vertex-wise correlations between the grey/white-matter contrast and area (top row), thickness (middle row) and local gyrification index (bottom row). These maps were not smoothed before calculations. Red/yellow areas represent positive correlations, while blue/light blue areas represent negative correlations.
Fig. 4. Statistical p-maps illustrating the effect of including area, thickness, or local gyrification index (LGI) as vertex-wise covariates in the general linear models: The top row (1) displays group differences between patients with schizophrenia (SCZ) and healthy controls (HC) before adding vertex-wise covariates. The model is corrected for age and sex. The second row (2) shows the model with cortical area has been included as a vertex-wise covariate. The third row (3) shows the model with cortical thickness included as a vertex-wise covariate. The bottom row (4) shows the model with LGI included as a vertex-wise covariate. The false discovery rate (FDR) was set based on all four models conjointly (using p values from four statistical p-maps) in each hemisphere, to allow for comparison between models.
Fig. 5. Statistical p-maps illustrating the effect of including area, thickness, or local gyrification index (LGI) as vertex-wise covariates in the general linear models. Similar to Fig. 4, the top row (1) displays group differences between patients with bipolar disorder (BD) and healthy controls (HC) before adding vertex-wise covariates. The model is corrected for age and sex. The second row (2) shows the model with cortical area has been included as a vertex-wise covariate. The third row (3) shows the model with cortical thickness included as a vertex-wise covariate. The bottom row (4) shows the model with local gyrification index (LGI) included as a vertex-wise covariate. The false discovery rate (FDR) was set based on all four models conjointly (using p values from four statistical p-maps) in each hemisphere, to allow for comparison between models. Since no finding survived FDR correction in the right hemisphere for this contrast, only the left hemisphere is shown.
Effects of antipsychotic medication, alcohol, and drug use
There was no significant association between antipsychotic medication dose and grey/white-matter contrast. Similarly, there was no significant association between grey/white-matter contrast and alcohol use or cigarette smoking after FDR correction, and main effects of group remained when including these variables as covariates.
Analysis of interaction with age
In the whole sample, the grey/white-matter contrast decreased with age in large parts of the cortical surface. The age slope was not significantly different in the schizophrenia or bipolar disorder groups compared to healthy controls after FDR correction (see Supplementary Fig. S8 for uncorrected findings, p < 0.01).
Post-hoc analysis of bipolar disorder subgroups
No grey/white-matter contrast difference was found between patients with bipolar disorder and a history of psychosis compared to those with no history of psychosis. Similarly, no difference was found between patients with bipolar I and bipolar II disorder. When restricted to those with bipolar I disorder, no significant differences compared to healthy controls survived FDR correction.
Post-hoc analysis of association with hallucinations and delusional thinking
Patients with hallucinations at symptom level showed higher contrast compared to the symptom absent group in the transverse temporal cortices bilaterally (Table 3). Patients with subthreshold symptoms also showed increased contrast in the left transverse temporal cortex and left occipital lobe clusters. No other cluster was significantly associated with hallucinations, and no clusters showed association with delusions (Supplementary Table S2).
Table 3. Test of association between mean grey/white-matter contrast in significant clusters and PANSS item P3 (‘hallucinatory behaviour’) scores
a Estimated marginal means (s.e.), adjusted for age, sex, and diagnosis.
b Desikan-Killiany atlas label for the vertex showing highest significance in the schizophrenia v. healthy control analysis for each cluster.
c The left precentral gyrus cluster showing difference between bipolar disorder and controls overlaps with this cluster and is therefore not included in the table.
Discussion
The main finding of the present study was increased grey/white-matter contrast in three bilateral sensory and motor regions of the cortex in schizophrenia compared to healthy controls. In bipolar disorder, significantly increased contrast compared to healthy controls was observed primarily in the left central sulcus, but no significant differences were found between schizophrenia and bipolar disorder. Regions showing increased contrast did not overlap substantially with reductions in cortical thickness and area previously demonstrated in schizophrenia and bipolar disorder. Group differences showing increased contrast were not affected by covarying for cortical area, thickness and gyrification, and we found no association with antipsychotic medication, smoking or alcohol use. Therefore, regional alterations in the cortical grey/white-matter contrast represent an independent signal from previously studied morphological measures.
Intriguingly, regions displaying increased contrast in schizophrenia comprised primary or secondary sensorimotor regions, related to the auditory, visual, tactile, and motor senses. Alterations of perceptual processes in schizophrenia were recognized early in the history of psychiatry (Uhlhaas & Mishara, Reference Uhlhaas and Mishara2007). Already in 1961, detailed descriptions of experiences from patients with early schizophrenia led researchers McGhie and Chapman to formulate the hypothesis that ‘schizophrenic symptoms are a result of a disturbance in the selective-inhibitory mechanism of attention’ (McGhie & Chapman, Reference McGhie and Chapman1961). Since then, deficits in early sensory processing in schizophrenia have been found using several neurophysiological paradigms, including pre-pulse inhibition, sensory gating and mismatch negativity (Javitt & Freedman, Reference Javitt and Freedman2015). It has been shown both using electrophysiological and phenomenological methods that perceptual alterations may be present before the onset of positive symptoms (Nelson et al. Reference Nelson, Thompson and Yung2012; Bodatsch et al. Reference Bodatsch, Brockhaus-Dumke, Klosterkotter and Ruhrmann2014). Interestingly, two resting state fMRI studies have reported altered connectivity specifically between sensory and motor regions and the thalamus in these disorders: Anticevic et al. (Reference Anticevic, Cole, Repovs, Murray, Brumbaugh, Winkler, Savic, Krystal, Pearlson and Glahn2014) reported increased connectivity between sensory and motor regions and the thalamus in schizophrenia and bipolar disorder, a finding that was recently replicated in a high-risk sample (Anticevic et al. Reference Anticevic, Haut, Murray, Repovs, Yang, Diehl, McEwen, Bearden, Addington, Goodyear, Cadenhead, Mirzakhanian, Cornblatt, Olvet, Mathalon, McGlashan, Perkins, Belger, Seidman, Tsuang, van Erp, Walker, Hamann, Woods, Qiu and Cannon2015). In a similar study focusing on schizophrenia, both increased and decreased connectivity between sensory and motor nodes and the thalamus were reported (Kaufmann et al. Reference Kaufmann, Skatun, Alnaes, Doan, Duff, Tonnesen, Roussos, Ueland, Aminoff, Lagerberg, Agartz, Melle, Smith, Andreassen and Westlye2015). Taken together, these findings are consistent with the possibility of a cortical disinhibition of sensory regions in schizophrenia, although neuronal mechanisms are not yet clarified (Anticevic et al. Reference Anticevic, Cole, Repovs, Murray, Brumbaugh, Winkler, Savic, Krystal, Pearlson and Glahn2014).
The regions showing differences in schizophrenia overlap with heavily myelinated regions of the cerebral cortex as evidenced by both MRI-based myelin maps and post-mortem studies (Glasser et al. Reference Glasser, Goyal, Preuss, Raichle and Van Essen2014). Therefore, increased tissue contrast in patients could be a result of lower grey-matter signal intensity reflecting reduced intracortical myelin. In other cortical regions, such as the prefrontal cortex and parietal regions, reduced oligodendrocyte density and ultrastructural alterations of myelin sheaths have been reported in schizophrenia. To our knowledge, no histological study has investigated myelination of sensorimotor regions in schizophrenia. Using MRI-based methods, only one previous study has examined a grey/white-matter contrast measure similar to the one used in the present study in schizophrenia (Kong et al. Reference Kong, Herold, Stieltjes, Essig, Seidl, Wolf, Wüstenberg, Lässer, Schmid, Schnell, Hirjak and Thomann2012). Contrary to our findings, this study reported decreased grey/white-matter contrast in different frontal, temporal, and parietal regions compared to those where we found increased contrast. However, the Kong et al. study recruited older inpatients with longer duration of illness (25 years), and sample size differed. While we found little evidence of differential age trajectories between patients and controls, the effect of illness duration is less easily studied. Regarding methodology, scanner type and field strength, pulse sequence, and sampling method also differed between the studies. Sampling methods alone did not explain the divergent findings, but the effect of scanner-related properties is difficult to exclude.
Unlike patients with schizophrenia, patients with bipolar disorder showed only limited alterations of the grey/white-matter contrast compared to controls. At the same time, effect size maps revealed trends in the same direction as in the schizophrenia group, and the direct comparison between schizophrenia and bipolar disorder showed limited differences. Results did not indicate that altered contrast in bipolar disorder was exclusively driven by the subgroup with lifetime psychosis or bipolar I disorder. Based on previous literature, it is possible that similar myelin deficits are present in bipolar disorder and schizophrenia (Tkachev et al. Reference Tkachev, Mimmack, Ryan, Wayland, Freeman, Jones, Starkey, Webster, Yolken and Bahn2003; Yu et al. Reference Yu, Bi, Liu, Zhao, Zhang and Yue2014). One interpretation is that similar but attenuated alterations exist in bipolar disorder, but it is difficult to draw firm conclusions based on the present data.
Maturation of intracortical myelin is an ongoing process throughout adolescence and early adulthood (Westlye et al. Reference Westlye, Walhovd, Dale, Bjornerud, Due-Tonnessen, Engvig, Grydeland, Tamnes, Ostby and Fjell2010), coinciding with the period of increased risk of developing schizophrenia and bipolar disorder (Castle et al. Reference Castle, Sham and Murray1998). Through regulating speed of neural conductance, myelin serves important functions in optimizing functional neural networks (Bartzokis, Reference Bartzokis2011). Dysfunction of high frequency neural oscillations have been shown in schizophrenia (Uhlhaas & Singer, Reference Uhlhaas and Singer2015), and correlated with impaired sensory processing (Sun et al. Reference Sun, Castellanos, Grützner, Koethe, Rivolta, Wibral, Kranaster, Singer, Leweke and Uhlhaas2013). A key feature of neural sensory processing is to differentiate between self-generated and external sensory input, since peripheral sensory stimulations are by themselves indistinguishable. If disrupted, self-generated stimuli could be hypothesized to be perceived as externally caused phenomena (Feinberg, Reference Feinberg1978). The responsible ‘corollary discharge’ mechanism can be mediated by inhibitory interneurons (Poulet & Hedwig, Reference Poulet and Hedwig2006; Crapse & Sommer, Reference Crapse and Sommer2008). Indeed, movement-related suppression of the auditory cortex in mice was shown to be initiated by direct excitatory connections from neurons in the motor cortex but mainly mediated by local circuits involving parvalbumin-positive (PV+) interneurons (Schneider et al. Reference Schneider, Nelson and Mooney2014). Gamma-band oscillations, which are related to the interplay between PV+ interneurons and pyramidal neurons, have been associated with corollary discharges related to self-generated speech (Chen et al. Reference Chen, Mathalon, Roach, Cavus, Spencer and Ford2010). Imprecision of corollary discharge responses has been shown to correlate with auditory hallucinations in schizophrenia both using verbal tasks and EEG paradigms (Heinks-Maldonado et al. Reference Heinks-Maldonado, Mathalon, Houde, Grey, Faustman and Ford2007) and, intriguingly, motor tasks and fMRI paradigms, where less attenuation of secondary motor cortex activation was found during self-generated movement in schizophrenia (Shergill et al. Reference Shergill, White, Joyce, Bays, Wolpert and Frith2014). A hypothesis linking delayed corollary discharge and abnormal frontal white-matter myelination in schizophrenia has been proposed (Whitford et al. Reference Whitford, Ford, Mathalon, Kubicki and Shenton2012). However, intracortical myelin is also late-maturing, although less studied, and reduced cortical myelination may affect inhibitory activation (Merkler et al. Reference Merkler, Klinker, Jurgens, Glaser, Paulus, Brinkmann, Sereda, Stadelmann-Nessler, Guedes, Bruck and Liebetanz2009).
Based on our novel findings, we hypothesize that abnormal intracortical myelination of low-level sensory regions may reduce inhibition of sensory input, through desynchronization of local or long-range neural circuitry, causing trait-like alterations of perceptual processes which may give rise to more transient positive psychotic symptoms. We stress that further evidence is required to verify this hypothesis, as neither reduced intracortical myelin nor a causal relationship with symptoms was conclusively demonstrated by the findings in the present study. Post-hoc analyses aimed to examine association with symptoms using the hallucination (P3) and delusion (P1) items from the PANSS scale. It is intriguing that the clusters showing association with hallucinations included the temporal cortices bilaterally, given that auditory hallucinations are the most common type (Goodwin & Rosenthal, Reference Goodwin and Rosenthal1971). However, these findings would not survive strict correction for multiple testing (e.g. Bonferroni correction). Conclusions based on these analyses are further limited by the temporal fluctuations of positive symptoms. Future studies should examine both lifetime symptoms and perceptual alterations (Javitt & Freedman, Reference Javitt and Freedman2015), or adopt a longitudinal design.
Another limitation of the present study is that we have not distinguished if contrast differences were driven by grey- or white-matter intensities. Theoretically, increased contrast could be caused by either decrease in grey matter intensities (e.g. reduced intracortical myelin) or increase in white-matter intensities (e.g. reduced interstitial white-matter neuron density), or a combination of the two. To distinguish between these alternatives, it is necessary to consider the MRI signal in grey and white matter separately. Unfortunately, this is made difficult by effects of radio-frequency field bias, which may occur due to differences in positioning in the scanner, head size and shape, shielding effects or other factors. Importantly, field inhomogeneities are not identical across subjects and their effects are not limited to random noise but may also confer systematic bias. When a local ratio is computed these effects are substantially reduced. Due to the smoothness of the bias field, inhomogeneities will have similar influence on sampling points located close in space (~2 mm), and will therefore be largely canceled out when computing a local ratio (Salat et al. Reference Salat, Lee, van der Kouwe, Greve, Fischl and Rosas2009). However, we acknowledge that we cannot completely rule out residual effects of field bias in the ratio. There is no such intensity ‘normalization’ when grey- or white-matter intensities are sampled separately. This renders direct comparisons of tissue intensities notoriously susceptible to spurious field bias effects, and post-hoc correction for field bias with sufficient precision to allow T1 mapping is difficult (Glasser et al. Reference Glasser, Goyal, Preuss, Raichle and Van Essen2014). In future studies, using alternative measures sensitive to myelin, such as the T1/T2 ratio (where field bias is also to a large extent cancelled out since they are correlated) or R1 (Glasser & Van Essen, Reference Glasser and Van Essen2011; Lutti et al. Reference Lutti, Dick, Sereno and Weiskopf2014), and obtaining field maps during scan acquisitions to allow for optimal correction could alleviate this problem.
Other biases that should be considered include motion artifacts and water content of the brain tissue, but since we observed increased contrast rather than decreased in specific regions, both these factors seem unlikely to have influenced the results. We addressed confounding effects of alcohol and cigarette smoking by including these variables as covariates, and we examined associations with antipsychotic medication within the patient sample. Still, we emphasize the inherent limitations shared by all naturalistic and cross-sectional clinical studies. An additional limitation is that PANSS interviews were not conducted on the day of the MRI scan.
To conclude, cortical grey/white-matter contrast is increased in sensory and motor regions in schizophrenia. This may indicate abnormal intracortical or white-matter myelination in these regions. The hypothesis that this abnormality may cause perceptual alterations leading to psychotic symptoms, which is both consistent with the phenomenology of the illness and leads to testable predictions, should be examined in future research.
Supplementary material
For supplementary material accompanying this paper visit http://dx.doi.org/10.1017/S0033291716000593.
Acknowledgements
The study was supported by grants from the Research Council of Norway (grant nos. 190311/V50, 167153/V50, 223273), the South-Eastern Norway Regional Health Authority (grant nos. 2012100, 2011092, 2011096) and the K. G. Jebsen Foundation.
Declaration of Interest
O.A.A. has received speaker's honorarium from pharmaceutical companies Osaka, GSK, Lundbeck. The remaining authors report no conflict of interest.
