Cortical Folding Defects as Markers of Poor Treatment Response In First Episode Psychosis

Abstract

Context

At present, no reliable predictors exist to distinguish future responders from non-responders to treatment during the first episode of psychosis (FEP). Among potential neuroimaging predictors of treatment response, gyrification represents an important marker of the integrity of normal cortical development that may characterise, already at illness onset, a subgroup of patients with particularly poor outcome.

Objective

We test the hypothesis that patients with FEP who do not respond to 12 weeks of antipsychotic treatment have, already at illness onset, significant gyrification defects.

Design

Case-control study with 12-weeks longitudinal follow-up to determine treatment response.

Participants

126 subjects, including 80 patients presenting with FEP and 46 healthy controls. Patients were scanned at the outset and received various antipsychotic medications in a naturalistic clinical setting. They were followed up for 12-weeks and classified as Responders or Non-Responders if they reached criteria for symptom remission, evaluated with the Psychiatric and Personal History Schedule (PPHS).

Setting

Secondary psychiatric services in an inner city area (South London, United Kingdom).

Outcome

Cortical gyrification was assessed using local gyrification index (LGI) in a vertex-wise fashion across the entire cortical surface with correction for multiple testing using permutation analysis. Differences in LGI were assessed between Responders, Non-Responders, and healthy controls. The effect of diagnosis (affective vs. non-affective psychosis) on the LGI was also investigated in Responders and Non-Responders.

Results

Patients with FEP showed a significant reduction in gyrification (hypogyria) across multiple brain regions, compared to healthy controls. Interestingly, Non-Responders showed prominent hypogyria at bilateral insular, left frontal and right temporal regions when compared to Responders (all clusters significant at p<0.05). These effects were present for both affective and non-affective psychoses.

Conclusions

Gyrification appears to be a useful predictor of antipsychotic treatment response. Early neurodevelopmental aberrations may predict unfavourable prognosis in psychosis, irrespective of the existing diagnostic boundaries.

1. Introduction

At present, there is no reliable predictor of treatment response in first episode psychosis (FEP). Early treatment response is thought to be one of the strongest predictors of subsequent functional outcome in psychosis ^1,2; furthermore, early responders are also less likely to experience further psychotic episodes following illness onset ³. The potential of neuroimaging studies to provide measures of translational importance in predicting treatment response and reducing the duration of psychotic episodes has become paramount ⁴. Hence, it is essential to ascertain high-yield neuroimaging measures with a predictive potential by studying their association with prospective treatment response.

Several neuroimaging studies have investigated the relationship between treatment response and brain structure in psychosis. However, a pooled analysis of these studies showed no significant relationship between brain morphology and treatment response, possibly because of significant heterogeneity in illness duration, duration of untreated psychosis (DUP), age and length of treatment of the subject groups included in the analyses ⁵. More recent studies have found some support for the notion that diminished grey matter tissue in patients with psychosis is associated with poor treatment response ^6–8. We have previously shown that structural MRI can indeed contribute to personalised predictions of longitudinal outcome in FEP⁴, though discrepancies exist in the localisation and the direction of the changes ⁹. These inconsistencies are in part due to the varying levels of prior exposure to antipsychotics across studies ^7,9 and the use of Voxel Based Morphometry (VBM), a method that measures grey matter density or volume, which can be confounded by the presence of altered surface morphology ^10,11. Importantly, the presence of abnormal sulcal patterning in schizophrenia, putatively linked to aberrant neurodevelopment¹⁰, increases the likelihood of observing inconsistent results when using template-based registration for VBM. Recent methodological advances in structural imaging have allowed us to study the surface anatomy in greater detail ¹².

Such surface based approaches have been used successfully to study clinically important variables in psychosis, such as the risk of transition to overt psychosis from the prodromal state ¹³, the effects of antipsychotic treatment ¹⁴ and the persistence of symptom burden despite treatment ¹⁵.

Among various anatomical measures that can be studied using structural MRI, cortical gyrification holds a distinct potential to investigate early neurodevelopmental disturbances that may contribute to the pathophysiology of psychosis ^12,16 and that may characterise, already at illness onset, a subgroup of patients with particularly poor outcome ^17,18. Defects in gyrification have been associated with exposure to obstetric complications, known to contribute to an elevated risk of schizophrenia ^19–21. Cortical gyrification has also been observed to be a structural marker of later neurodevelopmental outcome in preterm infants ²². Furthermore, and relevant to this study, regional abnormalities in gyrification seem associated with the expression of core psychotic symptoms that persist despite treatment^23,24. Also, given the proposed differences in the degree of neurodevelopmental aberrations in affective and non-affective psychosis ²⁵, the study of gyrification could potentially aid in predicting diagnostic differences in psychosis. Interestingly, early neuroimaging studies investigating the structural correlates of treatment response have observed an inverse relationship between prefrontal sulcal width and treatment response to clozapine in schizophrenia ^26–28. Although this indicates an association between folding patterns and treatment response, these studies were methodologically limited by the use of two-dimensional computed tomography based measurements from manual drawings on preselected brain regions, and by the evaluation of subjects with prolonged exposure to antipsychotics. To date, no MRI study has explored the relationship between the local gyrification pattern across the entire cortex, and subsequent treatment response, in FEP with minimal antipsychotic exposure.

In the present study, we investigated the gyrification pattern across the entire cortical surface using a vertex-wise mapping approach in patients with FEP, grouped on the basis of their subsequent response to treatment at 12 weeks. Given evidence that reduced grey matter is associated with poor treatment response ^6–8, we predicted that, already at illness onset, Non-Responders would show widespread abnormalities in gyrification when compared to Responders or healthy controls. Further, given evidence of differences in the degree of neurodevelopmental abnormalities in affective and non-affective psychosis ²⁵, we also performed an exploratory investigation of diagnostic differences in gyrification, and evaluated the effect of diagnosis on the relationship between gyrification and treatment response.

2. Methods

2.1. Participants

Patients with FEP were recruited from the South London and Maudsley (SLAM) National Health Service (NHS) Foundation Trust, South East London (United Kingdom). We defined FEP as the first ever presentation to secondary psychiatric services with evidence of any of the following: delusions, hallucinations, thought disorder or negative symptoms of schizophrenia, which would be scored 4 or more on the PANSS and that had lasted for at least 7 days (Nottingham Onset Schedule ²⁹) All patients with a functional psychotic illness (ICD10 F10-19, excluding coding F1x.0 for Acute intoxication; F20-29 and F30-39, psychosis codes) (World Health Organisation, 1992) were invited to participate. A sample of healthy controls similar to the patient group in age, gender, ethnicity, educational qualifications and employment status, was recruited from the same geographical area. Controls were administered the Psychosis Screening Questionnaire ³⁰, and excluded if they reported any psychotic symptom or had a history of any psychotic illness.

Exclusion criteria for all subjects were: history of head trauma or injury with loss of consciousness lasting longer than one hour; history of any serious medical or surgical illness; learning disabilities; history of current or past organic psychosis; lack of English fluency and known contraindications to conventional MRI. A posteriori exclusion criteria (after neuroimaging acquisition) included any evidence of neuro-radiological abnormality or problems with scan acquisition and motion artifacts. Ethical permission was obtained from the South London and Maudsley NHS Foundation Trust and Institute of Psychiatry research ethics committee. After complete description of the study to the subjects, written informed consent was obtained. A total of 80 patients and 46 healthy controls were included in this study.

Clinical assessment

Diagnosis was made using the Operational Criteria (OPCRIT) ³¹, according to the International Statistical Classification of Diseases and Related Health Problems - 10th Revision (ICD-10) criteria, using patient clinical notes for the month after their first contact with psychiatric services. All diagnoses were performed by qualified psychiatrists, subject to comprehensive training and achievement of good inter-rater reliability testing (kappa=0.91). Patients with a diagnosis of Bipolar disorder or Major Depression with psychotic symptoms were included in the affective psychosis group, while patients with Schizophrenia, Schizophreniform Disorder, Schizoaffective Disorder and Psychosis not Otherwise Specified formed the non-affective psychosis group.

Severity of psychotic symptoms was evaluated on the day of MRI, and then again after 12 weeks, using the Positive and Negative Syndrome Scale (PANSS) ³². Duration of untreated psychosis (DUP) was quantified as time interval between first onset of psychotic symptoms and first contact with psychiatric services, using information from medical records, patient interview and significant others. Duration of Illness (DOI) was defined as including both the DUP and the time between contact with services and the MRI scan (time of treated illness). Because of their relevance to neuroimaging studies ³³, detailed information on antipsychotic drugs dose and length of exposure were collected during face-to-face interviews, from clinical notes and from interviews with the clinical team. Antipsychotic doses were converted to chlorpromazine equivalents, according to defined criteria ^34,35, to estimate the total daily chlorpromazine-equivalent dose, calculated by summing all daily doses from the first day of treatment with antipsychotics up to the day of MRI. The majority of the patients (n=62) were taking atypical antipsychotics (35 olanzapine, 18 risperidone, 4 quetiapine and 5 aripiprazole), except for 3 patients who were taking typical antipsychotics (one each on haloperidol, amisulpiride and flupentixol), and 15 who were medication naïve. Premorbid IQ was assessed using the New Adult Reading Test ³⁶. Handedness was assessed according to the Annett Hand Preference Questionnaire ³⁷.

2.2. Evaluation of treatment response

We evaluated response to treatment 12 weeks after MRI, because of the clinical recommendation that antipsychotic treatment with a specific drug should be continued for 6-8 weeks before switching to a different medication because of lack of efficacy or for adverse effects ³⁸. Hence, we considered that a 12-week interval would provide information on treatment response following at least one drug taken for an appropriate period of time (even by allowing for delay in starting or optimising the medication regime).

Response to treatment was our primary outcome measure and evaluated by using information obtained from clinical records, patient face-to-face interviews and reports from informants using the WHO Personal and Psychiatric History Schedule (PPHS), a standardized instrument to record symptoms presence and severity, which has been successfully used in WHO multi-centre studies of the incidence and outcome of schizophrenia ³⁹. Response was operationalized as a reduction in symptoms severity to the levels required by the Remission criteria of the Schizophrenia Working Group Consensus ⁴⁰. This Consensus established a set of criteria that provide an absolute threshold in severity of symptoms that should be reached for clinical improvement.

This approach was therefore preferred to symptom change cut-offs for this naturalistic study. In fact, cut-off points are often arbitrary, affected by variability in baseline symptom severity across studies, and are not understood intuitively by clinicians ^40,41. Instead, the remission criteria proposed by the Consensus are more suited for traditional concepts of remission in psychiatric disorders⁴⁰. According to these criteria, clinical improvement is reached when a simultaneous rating of mild or less (equivalent to 1,2 or 3) is given in the following items of the PANSS: Delusions (P1), Unusual thought content (G9), Hallucinatory behaviour (P3), Conceptual organization (P2), Mannerisms/posturing (G5), Blunted affect (N1), Social withdrawal (N4), Lack of spontaneity (N6). To this end, we considered the PPHS scores equivalent to the PANSS scores as follows: 0 equivalent to PANSS scores 1, 2 and 3; 1 equivalent to PANSS scores 5 and 6; and 2 equivalent to PANSS scores 7 and 8. Using this method, 40 patients were classified as Responders and 40 as Non-Responders.

2.3. MR image acquisition and processing:

MRI scans were obtained as soon as possible after first contact with the psychiatric services, whenever deemed appropriate by the treating clinician, to ensure minimal exposure to antipsychotic medications in patients. All MRI scans were acquired in a 3 Tesla GE (General Electric, Milwaukee) Signa HDx scanner at the Centre for Neuroimaging Sciences (CNS), Institute of Psychiatry, London. A sagittal three-dimensional MPRAGE volumetric scan was obtained from each subject. The MPRAGE image had an image matrix size of 256x256x166 voxels, with in-plane voxel size of 1.02x1.02mm and a slice thickness of 1.2 mm (echo time/repetition time/inversion time = 2.848/6.988/650 ms, excitation flip angle 20°, one data average). Full brain and skull coverage was required for the MRI datasets and detailed quality control carried out on all MR images according to previously published quality control criteria⁴².

2.4. Cortical gyrification analysis

Surface extraction was completed using FreeSurfer version 4.5.0 ⁴³. The preprocessing was performed according to the standard description given by Fischl et al ⁴⁴). Briefly, following skull-stripping and intensity correction, the grey–white matter boundary for each cortical hemisphere was determined using tissue intensity and neighbourhood constraints. The resulting surface boundary was tessellated to generate multiple vertices across the whole brain before inflating. The expansion of the grey-white boundary created the pial surface with a point-to-point correspondence. This was followed by spherical morphing and spherical registration using sulcogyral landmarks. Local gyrification indices (LGIs) were obtained using the method of Schaer ⁴⁵ in line with previous studies ^11,23,46. Schaer’s method is an automated vertex-wise extension of Zilles’ gyrification index, which computes a ratio of the inner folded contour to the outer perimeter of the cortex ⁴⁷ using images reconstructed through the Freesurfer pipeline ⁴⁸. It provides a local gyrification index for each of the thousands of vertices on the reconstructed cortical surface, which serves as a measure of the amount of cortex buried in the locality of each vertex. A LGI value of 1 assigned to a vertex suggests that the vertex is located on a flat pial surface with no sulcal ridges in its vicinity. We used a 25mm spherical region of interest around each vertex to compute the LGI.

2.5. Statistical analysis

Baseline clinical and demographic variables were compared using ANOVA or chi-square test. Log transformation was undertaken to ensure normality for variables with a skewed distribution. Between-groups differences in cortical gyrification were estimated using Query-Estimate-Design-Contrast interface in the Freesurfer software. We used a general linear model controlling for the effect of age, gender and intracranial volume (ICV) to estimate differences in gyrification among Responders, Non-Responders and healthy controls at each vertex of the right and left hemispheric surfaces. In addition, when comparing Responders and Non-Responders, we explored the effect of diagnosis (affective vs. non-affective psychosis) and the interaction between diagnosis and categorical treatment response. Furthermore, we studied the effect of diagnostic status in the two treatment groups (Responders and Non-Responders) independently, controlling for the effect of age, gender and ICV. Monte-Carlo permutation approach implemented in the Freesurfer was used for statistical correction of multiple comparisons ⁴⁹. For the different clusters observed using a cluster-forming threshold p < 0.05, we estimated the probability of observing a cluster of equal (or greater) spatial extent from the null distribution of data across 10,000 permutations. Clusterwise probability values p<0.05 were considered as significant.

3. Results

3.1. Clinical variables

The clinical and demographic characteristics of the sample are shown in Table 1. There were no significant differences in the distribution of diagnostic categories between Responders and Non-Responders (proportion with non-affective psychosis 73% vs. 68%, X²=0.24, df=1, p=0.63) (Table 1). There were no significant differences between Responders and Non-Responders in terms of handedness, total intracranial volumes, baseline PANSS negative scores, median DUP, median duration of illness, average dose or duration of antipsychotic treatment or the proportion of antipsychotic-naive subjects at baseline (all p>0.05) (Table 1), nor in exposure to mood stabilizers, antidepressants or benzodiazepines (data not shown). Responders had lower baseline total PANSS and PANSS positive symptoms scores than Non-responders. Patient group was slightly older than the healthy controls (mean (SD) in patients =28.04(8.0), mean (SD) in controls = 24.65(5.63), F=6.39, p=0.013). Though there were no differences between groups in alcohol use, the healthy controls included a lower proportion of subjects with current or past use of other substances than either the Responders or the Non-responders, which in contrast included similar proportions of users (51%, 75% and 83% respectively, X²=26.2, df=2; p<0.001).

Table 1. Clinical and demographic variables.

	Non-responders (n=40)	Responders (n=40)	Healthy controls (n=46)	F/χ2 (*p<0.05)
Diagnosis (%)			-	χ2=3.6
Schizophrenia	64	55
Schizoaffective Disorder	8	8
Bipolar Disorder	10	18
Depressive Disorder	15	13
Other	3	5
Diagnostic groups (Affective/Non-Affective)	11/29	13/27	-	χ2=0.24
Age in years (SD)	28.1(7.9)	28.0(8.2)	24.7(5.6)	F=3.17*
Gender (females/males)	10/30	12/28	21/25	χ2=4.5
Education in years (SD)	13(3.6)	13.4(3.8)	14.7(3.2)	F=2.1 a
IQ (NART, SD)	93(11)	90(11)	95(9)	F=2.0 b
Handedness (left/right)	2/38	3/37	7/39	χ2=2.2
Intracranial volume in cm3 (SD)	1724(212.8)	1753(213.6)	1728 (189.4)	F=0.24
Baseline PANSS score (SD)	62.6(13.4)	55.6(12.5)	-	F=6.38*
Total
Positive	16.1(6.7)	13.4(4.6)	-	F=4.18*
Negative	16.6(6.3)	14.3(5.8)	-	F=2.85
No. of days on treatment at the time of scan (SD)	38.0(28.1)	42.6(32.7)	-	F=0.43
No. of treatment naïve subjects at the time of scan (Affective/non-affective)	1/6	2/6		χ2=0.08
Average dose in chlorpromazine equivalents at the time of scan (SD)	244.9(167.3)	223.1(136.4)	-	F=0.34
Median DUP in weeks (25th; 75th percentiles) a	9(3;74)	6(1;24)	-	Z=-1.52
Median time in weeks between contact with services and the scan (25th; 75th percentiles)	5(2;8)	5(3;10)	-	Z=-0.41
Median DOI in weeks at the time of scan (25th; 75th percentiles) c	18 (7;81)	13(5;26)	-	Z=-1.83

^*group differences significant at p<0.05,

^abased on 63 patients and 29 healthy controls;

^bbased on 63 patients and 33 healthy controls;

^cMann-Whitney U tests. NART – National Adult Reading Test, PANSS – Positive and Negative Symptoms Scale, DUP – duration of untreated psychosis, DOI – total duration of illness (both treated + untreated).

3.2. Group differences in gyrification

3.2.1. All patients vs. controls

When all FEP patients were compared with the healthy controls, significant reductions in gyrification were observed in middle / inferior frontal gyrus, precentral gyrus and precuneus in the left hemisphere; and middle frontal gyrus and inferior parietal (angular gyrus) region in the right hemisphere (Table 2: Figure 1DS). There were no regions with increased gyrification in patients.

Table 2. Group differences in gyrification among Responders, Non-responders and healthy controls [threshold for inclusion in a cluster p<0.05].

Cortical Region	Talairach coordinates of the peak	Cluster Size (mm2)	Clusterwise probability
Healthy Controls > All patients
Left posterior cingulate/precuneus	-8,-40,41	8304	0.0001
Left precentral	-41,-12,48	3589	0.0001
Left middle/inferior frontal	-35,36,12	1167	0.0440
Right middle frontal cortex	33,23,43	1304	0.0003
Right inferior parietal	45,-58,36	755	0.0364
Responders > Non-Responders
Left insula/central sulcus	-57,-8,10	3697	0.0001
Right insula/ lateral temporal	37,-13,2	6703	0.0001
Left superior frontal/mid-cingulate	-9,32,27	5080	0.0001
Left middle frontal gyrus	-35,51,-1	2581	0.0001
Healthy Controls > Non-Responders
Left posterior cingulate/precuneus	-9,-49,26	7219	0.0001
Left precentral	-37,7,37	3775	0.0001
Right superior/middle frontal	20,33,34	3415	0.0001
Right superior temporal	53,-30,3	1811	0.0001
Right cuneus / lingual gyrus	10,-58,2	4382	0.0001
Right insula/lateral temporal	37,-13,2	4502	0.0001
Left middle frontal	-38,43,5	2330	0.0001
Right angular gyrus	45,-59,36	1415	0.0020
Left inferiortemporal	-53,-28,-24	792	0.0244
Left orbitofrontal	-33,20,-17	1077	0.0030
Healthy Controls > Responders
Left lingual	-4,-82,1	862	0.0151
Non-affective psychosis < Affective psychosis
Right insula/superior temporal	37,-12,2	5778	0.0001
Right inferior temporal	51,-27,-22	7245	0.0001
Left insula/inferior frontal	-42,19,7	5870	0.0001
Left middle frontal	-22,44,20	1037	0.0049
Right subgenual anterior cingulate	7,15,-17	1245	0.0007
Right posterior cingulate/precuneus	41,-81,6	1809	0.0001
Left lateral occipitotemporal	-49,-62,-2	4116	0.0001
Left posterior cingulate	-4,-15,31	1310	0.0004

3.2.2. Responders vs Non-Responders

Non-Responders showed reduced cortical folding (hypogyric regions) in several regions across the cortical surface when compared to Responders, with no regions of increased folding (hypergyria). The hypogyria was predominantly observed in the insula, superior frontal and rostral middle frontal regions in the left hemisphere and inferior and superior temporal cortex extending to the insula and the temporal pole in the right hemisphere (Figure 1; Table 2).

Figure 1. Clusters showing differences in gyrification among Responders, Non-Responders and controls.

The contrast for Responders > Non responders is shown in red; Controls > Non-responders is shown in yellow; Controls > Responders is shown in purple. No clusters showed significant results in the opposite directions. All clusters are displayed on a reconstructed average white matter surface (fsaverage) and survived multiple testing using Monte-Carlo simulation with a cluster inclusion criterion of p=0.05. Left hemisphere is on the left side and right hemisphere on the right side of the image. The exact values of clusterwise probability for the clusters are presented in Table 2.

3.2.3. Responders/Non-Responders vs healthy controls

Non-Responders showed reduced gyrification (hypogyric regions) in several brain regions when compared with healthy controls, including the bilateral middle frontal gyrus (extending to precentral gyrus), and superior / inferior temporal cortex, angular gyrus and medial occipital cortex on the right, posterior cingulate and precuneus extending to the medial occipital cortex on the left. Both Responders and Non-Responders showed a significant hypogyria of the left lingual gyrus when compared to controls (Figure 1; Table 2). Responders did not have any other regional differences when compared with controls.

Given the baseline difference in total PANSS score between Responders and Non-Responders, we repeated the analysis first including Total PANSS score, and then PANSS posiitve scores as covariates, and the above results remained unaltered. Furthermore, vertex-wise correlation analysis between total PANSS score or PANSS positive score and gyrification did not reveal any significant clusters of positive or negative correlations.

3.3. Gyrification in affective and non-affective psychosis

On the whole, patients with non-affective psychosis showed several regions of reduced gyrification when compared to those with an affective psychosis (Table 2 & Web Supplement Figure1DS). There was no significant diagnosisXresponse interaction on gyrification. However, among Non-Responders, there were no significant differences in the gyrification between the two diagnostic groups. In contrast, among Responders, patients with a non-affective psychosis had a significant reduction in the gyrification of bilateral insula including a portion of the Broca’s area, and medial orbitofrontal, dorsolateral prefrontal and superior temporal sulcus in the left hemisphere (Web Supplement Figure1DS).

3.4. Spatial overlap between the effect of diagnosis and treatment response

We created two binary masks, one including all significant clusters (thresholded at clusterwise significance of p<0.05) from the analysis of Responders vs. Non-Responders and the other including all significant clusters from the affective vs. non-affective psychosis comparison. We computed the proportion of spatial overlap (or intersection) between the two masks, and derived an inclusive mask that contained the vertices that were present in both masks (Figure2DS). Overall this inclusive mask contained only 13.7% of all vertices that showed either an effect of diagnosis or treatment response, while 86.3% of vertices were non-overlapping. The spatial overlap was noted in bilateral insula, right anterior lateral prefrontal and right medial orbitofrontal regions.

4. Discussion

In the first study using unbiased whole brain estimate of three dimensional gyrification to predict treatment response in FEP, we have shown that, already at illness onset, patients with FEP who subsequently do not respond to treatment have significant cortical folding defects compared to patients who subsequently respond and to healthy controls, while those who go on to respond are virtually indistinguishable from the controls. Furthermore, we have shown that patients with non-affective psychosis show a significant reduction in gyrification in line with the notion of greater neurodevelopmental deficits in schizophrenia than affective psychosis.

Non-Responders had significant hypogyria of several frontotemporal regions and the insula when compared to Responders, and more widespread deficits in gyrification extending to the precuneus, angular gyrus and lingual gyrus when compared to healthy controls. Responders had notable hypogyria of the left lingual gyrus compared to healthy controls. These differences were not explained by the effect of age, gender or diagnosis, though a small degree of overlap was noted between the spatial distribution of hypogyria related to treatment response and the hypogyria related to a diagnosis of non-affective psychosis. The gyrification abnormalities that characterize poor treatment response in FEP seem largely distinct from the gyrification defects that characterise schizophrenia, though reduced insular, orbitofrontal and lateral prefrontal gyrification may be shared by both schizophrenia and poor treatment response.

To our knowledge, no previous study has investigated the relationship between local gyrification indices of the cortex and subsequent treatment response, making comparisons difficult. Though voxel-based (VBM) and surface-based morphometric approaches are not directly comparable, our results may be interpreted as an extension of previous studies suggesting that diminished grey matter tissue in frontal and temporal cortex is associated with poor treatment response ^6–8,50. However, our finding of an inverse relationship between insular gyrification and treatment response is somewhat contrary to Molina and colleagues’ finding of a direct association between the degree of insular volume deficit estimated using a VBM approach and treatment response ⁹. Considering the inherent nature of VBM methods ⁵¹, it is likely that patients with a differential insular folding may be classified as showing a reduction in the probabilistic grey matter volume. In fact, a significant portion of the variance in insular grey matter measured using VBM in schizophrenia is not accounted for by surface anatomical measures such as thickness or surface area ¹¹. Nevertheless, our observation of prefrontal and insular hypogyria in association with non-response in FEP is consistent with previous observations of significant hypogyria in frontoinsular cortex in medicated patients with chronic schizophrenia and significant symptom burden despite clinical stability ¹⁶. This consistency highlights the strength of using surface based morphometry over other approaches to predict prognosis and treatment response.

Furthermore, various groups have linked the widening of sulcal width, which could lead to a hypogyric effect, to treatment response with clozapine in schizophrenia ^26–28. Unlike the inverse relationship observed with prefrontal sulcal width^26–28, larger width of the perisylvian cerebrospinal fluid space (adjacent to the temporo-insular regions showing hypogyria in the present study) has been associated with better response to clozapine⁵², suggesting that some structural indicators of poor response to early treatment can indeed predict better response to clozapine.

The insula is an integral part of a cognitive-control/salience network comprised of anterior cingulate and connected to limbic subcortical structures, cerebellum and dorsolateral prefrontal cortex ^53,54. This network has been proposed to play a cardinal role in the pathophysiology of psychosis^55,56. Given our current observation of fronto-temporo-insular hypogyria in non-responders, other indices of abnormal cortical connectivity involving these structures might also help predict treatment response. Indeed, this has been previously shown using fMRI and cognitive behavioral therapy in medicated patients with schizophrenia⁵⁷. Despite this apparent consistency, it is important to bear in mind that the integrity of other cortical/sub-cortical networks also contributes to treatment response⁵⁸.

The comparison between the non-affective and affective groups suggests that there is a widespread hypogyria in patients with a non-affective psychosis. These changes were predominantly observed in regions with multimodal functions such as lateral prefrontal and lateral temporal regions, in addition to the paralimbic regions (insula, anterior and posterior cingulate). This is coherent with conventional VBM studies showing less extensive grey matter reduction in first episode bipolar disorder than in schizophrenia in comparison to healthy controls^59–63, and even in high-risk individuals who subsequently develop affective rather than schizophrenia⁶⁴. Direct comparisons of patients with schizophrenia and bipolar disorder confirm the presence of more distributed GM deficits in schizophrenia, especially in the posterior cingulate cortex ⁶⁵, inferior, middle and superior frontal regions ⁶⁶. Our observations suggest that at least in part, these GM deficits may be driven by the presence of hypogyria in non-affective psychoses.

Though the exact mechanisms behind formation of cortical folds remain unclear, the adult pattern of gyrification seems tightly linked to the integrity of corticocortical and cortico-subcortical connectivity in the developing brain ¹⁸. In this context, it is likely that a higher loading of neurodevelopmental aberrations is present in the pathophysiology of psychosis in subjects who do not respond to treatment. Relevant to this notion is the observation that neurological soft signs, often considered to be an index of neurodevelopmental disturbances in schizophrenia⁶⁷, are related to both reduced gyrification⁶⁸ and poor response to antipsychotics in psychosis⁶⁹. Moreover, unlike Responders, the Non-Responders showed no diagnostic differences in gyrification. The presence of marked hypogyria, alongside the lack of diagnosis related differences in gyrification in the Non-Responders suggests that as a group, Non-Responders are likely to have a more homogenous pathophysiological process underlying psychosis than the Responders. Consistent with this finding, Penttilä and colleagues ⁷⁰ observed an association between reduced cortical folding and treatment resistance in bipolar disorder and concluded that in affective disorders, poor treatment response may be driven by neurodevelopmental anomalies. Reduced gyrification may also result from an atrophic process affecting the grey matter, at least in certain frontal clusters in the non-affective group, in which progressive grey matter reduction has been shown to be maximal in schizophrenia⁷¹. This issue warrants further investigation of longitudinal changes in gyrification in psychosis.

Several mechanisms may underlie our observation of extensive gyrification defects in Non-Responders. On one hand, Non-Responders may represent a distinct group of patients in whom a specific pathophysiological substrate underlies the development of psychosis. Cortical folding shows the highest activity during intrauterine growth and early infancy ¹⁸. Non-Responders may have experienced a pathophysiological insult at an earlier phase of the neurodevelopmental trajectory than the Responders, leading to extensive hypogyria in adult life. Psychosis in Non-Responders may be associated with an aberration in the molecular and genetic factors that determine axonal integrity, which in turn influences cortical gyrification ⁷². It is also possible that Non-Responders represent a subset of patients with a more severe form of the same underlying disease process. Neverthelss, the brain regions showing gyrification defects in FEP patients as a whole (when compared to healthy controls) were distinct from the regions showing gyrification abnormalities in Non-Responders (when compared to Responders). This suggests the presence of a distinct set of cortical neurodevelopmental abnormalities that contribute to poor prognosis in FEP.

Our study has a number of strengths. Unlike the curvature based methods used in some studies ^73,74, the Schaer’s method we used provides a composite index that captures both the curvature or depth and the spatial frequency or density of sulcogyral patterns on the cortical surface. Hence, the hypogyric regions we observed could either reflect a reduced number of sulcal formations or an increased width and reduced depth of the sulci themselves. Future investigations focusing on sulcal frequency as well as the gyrification index can clarify the exact nature of the gyrification defects in the non-responders. A number of previous studies seeking to identify structural predictors of treatment response included extensively medicated subjects^5,9,52, while our sample consisted of patients who were either unmedicated or minimally medicated before scanning. Further, the results of our post-hoc analysis comparing medication-naïve subjects with a subset of medication-exposed subjects (Supplementary Material) suggest that the changes we observed are unlikely to be explained by the use of antipsychotics.

In terms of limitations, the surface based morphometric approach that we employed is limited to the cortical mantle and does not provide information on deep grey matter structures such as the striatum, which have been previously associated with treatment response in psychosis ⁹. Also, the cross-sectional nature of our study precludes firm conclusions as to the timing of the gyrification defects, and the assumption regarding their neurodevelopmental origin must be considered with caution. The use of dichotomous treatment response variable can also be seen as a limitation, but this allows our results to be considered in line with antipsychotic clinical trials and neuroimaging studies investigating treatment response. By investigating the relationship between vertexwise gyrification and treatment response for the first time, we have established that the hypothesized relationship is indeed present in several brain regions in FEP. In future, the application of statistical discriminant approaches such as support vector machine ⁷⁵ could confirm whether an optimal combination of these distributed surface anatomical changes could satisfactorily classify the prospective responders from the non-responders, using an independent set of test data.

Our study provides crucial evidence of neuroimaging markers that can be used early in psychosis to predict prognosis in clinical settings. Identifying putative poor responders at the outset could assist in stratified treatment plans at an individual level and with appropriate resource allocation at the service level. Furthermore, the identification of cortical surface morphology and connectivity markers that characterize poor response opens the question as to whether risk factors that affect cortical surface morphology and integrity could be used to improve the proportion of patients responding to antipsychotic treatment at first episode.

The effect of baseline treatment exposure on gyrification

To address whether the exposure to antipsychotics at baseline affects the cortical gyrification observed in patients, we undertook an additional analysis comparing cortical gyrification of the treatment-naïve group (n=15) and of a matched sample of patients exposed to treatment at baseline (treatment-exposed) from the original sample reported in the manuscript. We identified 15 patients (7 responders, 8 non-responders) exposed to antipsychotics at baseline, and pair them with the antipsychotic-naïve group on the basis of age, gender and diagnostic category. The clinical and demographic details of this sample are given below in Table 1S.

Using ICV as covariate, we compared these two groups using Monte-Carlo permutations (n=10000) in line with the statistical analysis that was used to test our primary hypothesis comparing Responders and Non-Responders (cluster forming threshold of p<0.05, and clusterwise probability of p<0.05). We did not find any region of increased or decreased gyrification in the treatment-exposed group compared to the treatment-naïve group. We repeated the analysis with a more lenient cluster forming threshold of p=0.1 (one-tailed), but continued to observe no difference between the two groups. This null observation suggests that short-term exposure to antipsychotics does not alter the cortical folding patterns measured using the local gyrification index in patients with psychosis.

Follow-up PANSS scores at 12 weeks

We performed an additional analysis using only the 38 patients who had 12-week follow-up PANSS ratings. As expected, there was a significant Group X Time interaction in a repeated measure ANOVA (mean difference in Responders = 10.8 points, mean difference in Non-Responders = 2.0 points, F [1,36]=4.29, p=0.046) for total PANSS score.