Subcortical Brain Volume Abnormalities in Individuals With an At-risk Mental State

Abstract

Previous structural magnetic resonance imaging studies of psychotic disorders have demonstrated volumetric alterations in subcortical (ie, the basal ganglia, thalamus) and temporolimbic structures, which are involved in high-order cognition and emotional regulation. However, it remains unclear whether individuals at high risk for psychotic disorders with minimal confounding effects of medication exhibit volumetric changes in these regions. This multicenter magnetic resonance imaging study assessed regional volumes of the thalamus, caudate, putamen, nucleus accumbens, globus pallidus, hippocampus, and amygdala, as well as lateral ventricular volume using FreeSurfer software in 107 individuals with an at-risk mental state (ARMS) (of whom 21 [19.6%] later developed psychosis during clinical follow-up [mean = 4.9 years, SD = 2.6 years]) and 104 age- and gender-matched healthy controls recruited at 4 different sites. ARMS individuals as a whole demonstrated significantly larger volumes for the left caudate and bilateral lateral ventricles as well as a smaller volume for the right accumbens compared with controls. In male subjects only, the left globus pallidus was significantly larger in ARMS individuals. The ARMS group was also characterized by left-greater-than-right asymmetries of the lateral ventricle and caudate nucleus. There was no significant difference in the regional volumes between ARMS groups with and without later psychosis onset. The present study suggested that significant volume expansion of the lateral ventricle, caudate, and globus pallidus, as well as volume reduction of the accumbens, in ARMS subjects, which could not be explained only by medication effects, might be related to general vulnerability to psychopathology.

Introduction

Subcortical structures consist of the basal ganglia and thalamus, while the temporolimbic structures (including the amygdala and hippocampus) are also sometimes considered as a part of subcortical structures. The basal ganglia and thalamus play a crucial role in motor control¹ and high-order cognition,² while the hippocampus and amygdala in integrating sensory information³ and emotional regulation.⁴ It is reported that volume changes in these structures reflect ventricular volume.^5–7 Previous magnetic resonance imaging (MRI) studies, including a meta-analysis⁸ and recent 2 large multicenter studies,^9,10 have demonstrated that chronically medicated patients with schizophrenia exhibit reduced volume in the thalamus and nucleus accumbens as well as volume expansion in the caudate, putamen, and globus pallidus compared with healthy subjects. These structural changes are, at least in part, related to the psychopathological symptoms¹¹ and cognitive impairment in the patients,¹² possibly reflecting abnormal cortico-striato-thalamic circuits.¹³ However, lack of these structural changes reported in unmedicated^14–19 or minimally medicated^20–24 first-episode patients, as well as progressive striatal expansion associated with antipsychotic medication,^25–27 indicates considerable effects of medication on the basal ganglia in schizophrenia. Volume reduction of the temporolimbic structures,^28–31 as well as ventricular enlargement,^32–34 have consistently been reported after the onset of schizophrenia, but there is also some evidence that antipsychotic medication could affect these structures.^8,35,36 Thus, further studies are required to clarify how the findings of subcortical and temporolimbic structures reported in schizophrenia reflect the disease process and medication effects.

The at-risk mental state (ARMS) construct was introduced to allow identification of individuals at clinically high risk for psychosis before the manifestation of florid psychotic symptoms.³⁷ They are defined by subthreshold psychotic symptoms and functional deterioration^38,39 and have an enhanced risk of developing psychosis within a relatively short period of time (approximately 30% at 2 years).⁴⁰ Increasing evidence from MRI studies in ARMS subjects, which have the advantage of minimizing the confounding effects of antipsychotic medication and disease progression on brain structure,⁴¹ has demonstrated that individuals with ARMS, especially those who later develop psychosis, exhibit gray matter reduction similar to those observed in schizophrenia (eg, frontal and temporolimbic atrophy).^42–44 However, compared to the cortical structures, the subcortical and temporolimbic structures have been understudied in this population; ARMS individuals likely exhibit normal volume for the ventral and dorsal striatum,^41,45–47 globus pallidus,^41,46 and lateral ventricles,^46,48,49 as well as reduced thalamic volume,⁴¹ while they are reported to have both normal or reduced amygdalar^{30,41,46,49,50} and hippocampal^{30,41,46,49–54} volume. Although these studies showed no difference between ARMS individuals with and without later psychosis onset,^45,47–53 these negative or partly inconsistent findings could be partly explained by the small sample size and/or clinical/biological heterogeneity of ARMS itself.^55,56 Thus, it remains unclear whether clinical high-risk subjects exhibit changes in subcortical and temporolimbic regions and whether such changes, if present, could be a predictive marker of later psychosis onset.

In this study, we conducted a multicenter MRI study with sufficient statistical power to assess the subcortical and temporolimbic volumes of ARMS individuals and matched healthy controls recruited at 4 scanning sites. We aimed to elucidate regional volume changes as well as their relationships with clinical indices (eg, medication, symptomatology, and outcome) in ARMS subjects. Given the notion of leftward asymmetry of the striatal volume being related to psychosis,¹⁰ we also investigated possible alterations in the laterality of subcortical and temporolimbic structures in ARMS individuals.

Methods

Participants

One hundred and seven ARMS individuals were recruited from specialized clinical services for ARMS at Toyama University Hospital, Toho University Hospital, Tohoku University Hospital, and The University of Tokyo Hospital.^57,58 To determine whether participants satisfy the ARMS criteria (supplementary table 1), trained psychiatrists and/or psychologists interviewed the participants at intake using the Comprehensive Assessment of At-Risk Mental States (CAARMS)³⁸ (Toyama and Tohoku) or the Structured Interview for Prodromal Symptoms/the Scale of Prodromal Symptoms (SIPS/SOPS)³⁹ (Tokyo and Toho). On the basis of these interview data and other clinical information, clinical research meetings at each site reached a consensus on the diagnosis of ARMS for all cases. ARMS individuals were followed up prospectively (mean = 4.9 years, SD = 2.6 years) at each site and subdivided into 21 individuals (19.6%) who subsequently made the transition to psychotic disorders (ARMS-T), 72 individuals who did not transition to psychotic disorders during clinical follow-up of at least 2 years (ARMS-NT), and 14 individuals with unknown outcomes because of dropout from the study within 2 years (ARMS-UK). According to the CAARMS criteria (ie, at least one fully positive psychotic symptom several times per week for more than 1 week) or the SIPS criteria (ie, the presence of a positive symptom that has existed for more than 1 month or accompanying a serious disorganization or danger), transition to psychotic disorders was decided at each site. Based on the criteria of the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV),⁵⁹ the diagnoses of psychotic disorders in ARMS-T subjects were schizophrenia (n = 14), delusional disorder (n = 1), schizophreniform disorder (n = 1), brief psychotic disorder (n = 1), and psychotic disorder not otherwise specified (n = 4). According to the literature,⁶⁰ we also divided our ARMS cohort into good (ARMS-G, n = 33) and poor (ARMS-P, n = 57) outcome groups based on the median Global Assessment of Functioning ⁶¹ score at 1-year follow-up (the score was not available for 17 subjects) (supplementary table 2). Within the follow-up period (<2 years), at least 33 of the 72 ARMS-NT individuals had symptom amelioration and no longer fulfilled the ARMS criteria (ARMS-remitter, n = 33; ARMS-nonremitter, n = 39) (supplementary table 3). According to the exceptional conditions for the use of antipsychotic medication documented in the International Clinical Practice Guidelines for Early Psychosis,⁶² 43 of the 107 ARMS individuals (40.2%) were taking a low dosage of antipsychotics at the time of MRI scanning because of their relatively severe psychiatric symptoms (ie, rapid deterioration, suicidal risk, depression, and harmful risk). Twenty-five of the 107 individuals (23.4%) were also taking antidepressants at the time of MRI scanning. Gender- and age-matched healthy controls consisted of 104 healthy volunteers recruited from the community, hospital staff, and university students at each site. The exclusion criteria were as follows: (1) having a lifetime history of a serious head injury, neurological illness, or other serious physical disease, (2) fulfilling the criteria for substance abuse/dependence (including cannabis use), which was assessed based on verbal reports of the study participants, and (3) having previous psychotic episodes that met the DSM-IV criteria. Two hundred and eight of the 211 subjects were also included in our previous study that investigated brain gyrification patterns in ARMS individuals.⁶³ All participants were informed of the design and purpose of the study, which was approved by the Committees on Medical Ethics of each site, and gave their written consent.

MRI Data Acquisition

T1-weighted MRI images were acquired from 4 scanning sites. The scanner field strength was 1.5 T (Toyama, Toho, and Tohoku) for 3 sites and 3.0T for 1 site (Tokyo) (supplementary table 4).

MRI Data Processing

Using FreeSurfer software (version 5.3; https://surfer.nmr.mgh.harvard.edu), the T1-weighted MR images were preprocessed in accordance with a standard automatic reconstruction algorithm, which consisted of removal of non-brain tissue, transformation to Talairach-like space, segmentation of gray/white matter tissue, triangular tessellation, and inspection of the white matter and pial surfaces.⁶⁴ The subcortical and temporolimbic segmentations of all reconstructed images were carefully inspected, and any errors were manually edited by 1 trained researcher (D.S.) without information of the subjects’ identities. Thus, we obtained images of subcortical and temporolimbic segmentation and then extracted the values of regional volumes (ie, bilateral lateral ventricles, thalamus, caudate, putamen, globus pallidus, hippocampus, amygdala, accumbens) as well as the intracranial volume (ICV).

Statistical Analysis

Demographic variables were compared among groups by one-way ANOVA or chi-square tests. For the comparison between the ARMS group as a whole and control groups, each regional volume was analyzed using repeated-measures ANCOVA with age, ICV, and scanning sites as covariates, with group and gender (male and female) as between-subject factors, and side (left and right) as a within-subject variable. For the comparison between the ARMS-T and ARMS-NT groups, gender was used as a covariate but not as a between-subject factor because of the small sample size, especially for male ARMS-T subjects. Other ARMS subgroup analyses (eg, ARMS-G vs ARMS-P, ARMS-remitter vs ARMS-nonremitter) were conducted in the same manner. Post-hoc Scheffe’s tests were used to follow-up significant main effects or interactions. Relationships of regional volumes with clinical variables (eg, antipsychotic medication dose [n = 43], duration between scanning and transition [n = 21], and positive symptoms subscale scores of the CAARMS [n = 55] or SOPS [n = 50]) were estimated by calculating a partial correlation coefficient controlling for gender, age, ICV, and scanning sites. Statistical significance level was set at P < .05 (two-sided). To prevent a possible type I error due to multiple tests, a Bonferroni correction was also applied for the group comparisons.

For the purpose of better management of inter-site variance,^9,10 we also calculated meta-analytic overall effect sizes (Cohen’s d) for a random effect model using Review Manager version 5.3 (the Nordic Cochrane Centre, Cochrane Collaboration, Copenhagen, Denmark) and Metasoft software.⁶⁵

Results

Demographic Background

Table 1 shows comparisons of the demographic and clinical characteristics among groups. The groups were matched for age and gender, while the healthy controls had attained a higher level of education compared with ARMS individuals. There was a significant difference in the distribution of right, mixed, and left-handedness between the control and ARMS groups. The ARMS-T and ARMS-NT groups did not differ in antipsychotic medication dose and frequency (n = 10/21 [47.6%] in ARMS-T and 30/72 [41.7%] in ARMS-NT groups, χ²(1) = 0.24, P = .63). The 2 groups did not differ either in antidepressant medication dose or frequency (n = 3/21 [14.3%] in ARMS-T and 20/72 [27.8%] in ARMS-NT groups, χ²(1) = 1.59, P = .21). There were no significant differences regarding the sum of the positive sections on the CAARMS and SOPS between the ARMS-T and ARMS-NT groups, while the ARMS-T group had higher scores compared with the ARMS-NT group in 2 SIPS/SOPS subscales (ie, unusual thought content/delusional ideas and perceptual abnormalities/hallucinations).

Table 1.

Characteristics of the Study Participants

	HC	Whole ARMS	ARMS-T	ARMS-NT	HC vs Whole ARMS		ARMS-T vs ARMS-NT
					Test Statistic	P value	Test Statistic	P value
Number of subjects (total), n	104	107	21	72
Scanning site 1 (Toyama)	52	22	5	11
Scanning site 2 (Toho)	4	19	4	15
Scanning site 3 (Tohoku)	17	35	9	25
Scanning site 4 (Tokyo)	31	31	3	21
Follow-up period, years, mean (SD)		4.9 (2.6)	4.9 (2.1)	5.7 (2.1)
Gender, male/female, n	52/52	49/58	7/14	36/36	χ 2(1) = 0.37	.54	χ 2(1) = 1.82	.18
Age, years, mean (SD)	22.6 (4.0)	21.3 (5.4)	20.4 (4.4)	21.7 (5.8)	F(1,210) = 3.69	.06	F(1,92) = 0.92	.34
Education, years, mean (SD)a	15.0 (2.3)	12.4 (2.5)	12.6 (2.3)	12.4 (2.6)	F(1,208) = 59.88	<.001	F(1,90) = 0.07	.79
Parental education, years, mean (SD)b	14.0 (2.3)	13.8 (2.5)	13.1 (3.7)	13.9 (2.2)	F(1,177) = 0.35	.56	F(1,78) = 1.44	.23
Handedness, right/both/left, nc	88/0/1	83/6/16	13/2/4	58/4/10	χ 2(2) = 18.19	<.001	χ 2(2) = 1.35	.51
Duration between scanning and transition, months, mean (SD)			10.9 (9.3)
Antipsychotic medication dose, Chlorpromazine equivalent, mg/day, mean (SD)d		181.5 (143.3) [n = 43]	191.4 (123.2) [n = 10]	155.2 (104.8) [n = 30]			F(1,39) = 0.82	.37
Antipsychotic medication type, typical/atypical/mixed, ne		5/36/2	0/10/0	5/24/1
Antidepressant medication dose, imipramine equivalent, mg/day, mean (SD)f		88.1 (45.5) [n = 25]	125.0 (43.3) [n = 3]	78.8 (41.3) [n = 20]			F(1,22) = 3.24	.09
CAARMS unusual thought global rating scale, mean (SD)g		3.6 (1.3)	4.1 (1.0)	3.4 (1.4)			F(1,47) = 2.40	.13
CAARMS unusual thought frequency scale, mean (SD)g		4.4 (1.5)	4.7 (0.8)	4.3 (1.7)			F(1,47) = 0.61	.44
CAARMS perceptual abnormalities global rating scale, mean (SD)g		2.9 (1.6)	3.1 (1.7)	2.8 (1.5)			F(1,47) = 0.29	.59
CAARMS perceptual abnormalities frequency scale, mean (SD)g		2.9 (1.7)	3.2 (1.8)	2.8 (1.6)			F(1,47) = 0.49	.49
CAARMS disorganized speech global rating scale, mean (SD)g		2.0 (1.2)	2.2 (1.3)	2.0 (1.3)			F(1,47) = 0.38	.54
CAARMS disorganized speech frequency scale, mean (SD)g		3.9 (2.2)	3.8 (2.2)	3.9 (2.3)			F(1,47) = 0.02	.90
CAARMS sum of the positive sections, mean (SD)g		19.7 (5.7)	21.0 (5.4)	19.1 (5.7)			F(1,47) = 1.08	.30
SIPS/SOPS unusual thought content/delusional ideas, mean (SD)		3.5 (1.8)	5.0 (1.2)	3.4 (1.8)			F(1,42) = 4.97	.03
SIPS/SOPS suspiciousness/persecutory ideas, mean (SD)		3.3 (1.5)	4.0 (1.2)	3.4 (1.4)			F(1,42) = 1.36	.25
SIPS/SOPS grandiose ideas, mean (SD)		1.0 (1.3)	1.3 (1.4)	1.0 (1.4)			F(1,42) = 0.26	.61
SIPS/SOPS perceptual abnormalities/hallucinations, mean (SD)		3.2 (1.9)	4.9 (1.2)	3.1 (1.9)			F(1,42) = 5.79	.02
SIPS/SOPS disorganized communication, mean (SD)		2.3 (1.9)	2.1 (2.3)	2.6 (1.8)			F(1,42) = 0.36	.55
SIPS/SOPS sum of the positive sections, mean (SD)		13.4 (5.9)	17.3 (4.8)	13.5 (5.8)			F(1,42) = 2.70	.11
Intracranial volume, cm3, mean (SD)	1548.5 (139.4)	1537.6 (162.4)	1531.8 (146.5)	1542.1 (173.9)	F(1,210) = 0.43	.51h	F(1,92) = 0.15	.70h

Note: HC, healthy controls; ARMS, at-risk mental state; T, transition; NT, non-transition; CAARMS, comprehensive assessment of at-risk mental states; SIPS/SOPS, the Structured Interview for Prodromal Symptoms/the Scale of Prodromal Symptoms.

^aData missing for 2 individuals.

^bData missing for 33 individuals.

^cData missing for 17 individuals.

^dDifferent typical and atypical antipsychotic dosages were converted into Chlorpromazine equivalents using the guideline by Inada and Inagaki.⁶⁶

^eForty-three individuals received antipsychotic medication therapy.

^fDifferent antidepressant dosages were converted into Imipramine equivalents using the guideline by Inada and Inagaki.⁶⁶

^gData missing for 2 individuals.

^hANCOVA with age as a covariate was used for group comparison.

Brain Measures

For the comparison between the ARMS group as a whole and the controls, ANCOVA (figure 1, table 2) of the lateral ventricle revealed significant main effects for group and side, as well as a significant group-by-side interaction. Post-hoc analyses demonstrated that both left (P < .001) and right (P = .023) lateral ventricles were significantly larger in the ARMS individuals. In addition, the lateral ventricle had a left-greater-than-right asymmetry (P < .001) only for the ARMS group. For the caudate, ANCOVA revealed a significant main effect for side and a significant group-by-side interaction (table 2). Post-hoc analyses showed that the ARMS individuals had a significantly larger caudate on the left hemisphere than the controls (P < .001) and that only the ARMS subjects were characterized by leftward asymmetry (P < .001). ANCOVA for the globus pallidus revealed a significant main effect for gender [F(1, 204) = 5.53, P = .020], a significant group-by-side interaction (table 2), and a significant gender-by-group-by-side interaction [F(1, 207) = 4.48, P = .036], where the male ARMS individuals (mean = 1577.45 mm³, SD = 290.48 mm³) had a significantly larger globus pallidus in the left hemisphere than did the male control subjects (mean = 1461.94 mm³, SD = 290.97 mm³) (post-hoc test, P = .033). For the accumbens, ANCOVA revealed a significant group-by-side interaction (table 2), where the ARMS individuals had a significantly smaller accumbens of the right hemisphere than the controls (post-hoc test, P < .001). For the thalamus, putamen, hippocampus, and amygdala, ANCOVA showed no main effect or interactions involving group. The group difference for the left lateral ventricle, left caudate, and right accumbens remained significant even after the Bonferroni correction for multiple comparisons (P < .003125 [8 regions by 2 groups; 0.05/16 comparisons]). Furthermore, these volume alterations, except for the right lateral ventricle and left globus pallidus, were significant even when we examined only a drug-naive ARMS subsample (n = 64) (supplementary table 5). When we included handedness as a covariate, the results of group comparisons of regional volumes remained essentially the same.

Fig. 1.

Subcortical and temporolimbic volumes between the ARMS and control subjects. The mean and SD values for each regional volume are illustrated for the (a) left and (b) right hemispheres. *All P < .05. “+” denotes ARMS > healthy controls only for the male subjects.

Table 2.

Group Comparison of Subcortical and Temporolimbic Volumes Among the ARMS and Control Subjects

Region of Interest (mm3)	HC (n = 104)		Whole ARMS (n = 107)		ARMS-T (n = 21)		ARMS-NT (n = 72)		ANCOVA (HC vs Whole ARMS)a						ANCOVA (ARMS-T vs ARMS-NT)
	(Male 52, Female 52)		(Male 49, Female 58)		(Male 7, Female 14)		(Male 36, Female 36)		Group		Side		Group × Side		Group		Side		Group × Side
	Mean	SD	Mean	SD	Mean	SD	Mean	SD	F (1, 204)	P	F (1, 207)	P	F (1, 207)	P	F (1, 87)	P	F (1, 91)	P	F (1, 91)	P
Lateral ventricle									8.68	<.01	30.80	<.01	13.18	<.01	0.85	.36	35.31	<.01	0.27	.60
Left*	5914.21	2823.67	7390.82	4038.49	6526.80	2465.04	7736.84	4537.98
Right	5654.85	2589.30	6158.30	3503.07	5413.73	2173.72	6408.35	3981.88
Thalamus									0.21	.64	381.80	<.01	1.27	.26	0.26	.61	123.85	<.01	0.08	.78
Left	8278.51	1068.27	8638.19	1113.50	8693.42	927.63	8705.56	1173.35
Right	7350.58	857.18	7601.55	901.03	7615.03	912.99	7680.58	913.91
Caudate									3.45	.06	21.95	<.01	14.39	<.01	0.82	.37	13.40	<.01	1.64	.20
Left*	3467.52	462.10	3709.75	490.16	3742.64	440.69	3711.52	520.91
Right	3442.19	597.03	3474.60	578.41	3615.46	577.26	3447.66	586.91
Putamen									1.28	.26	20.60	<.01	0.0007	.98	0.58	.45	1.83	.18	1.93	.17
Left	5314.96	692.80	5280.88	773.18	5229.44	843.41	5255.63	759.31
Right	5192.84	655.20	5160.77	745.03	5231.12	806.70	5123.35	747.01
Globus pallidus									1.47	.23	1.62	.20	5.63	.02	1.22	.27	0.95	.33	2.95	.09
Left	1403.06	287.10	1487.86	286.51	1487.00	284.64	1481.63	275.42
Right	1455.32	270.44	1476.42	268.95	1556.31	270.56	1462.52	260.96
Hippocampus									0.78	.38	4.21	.04	0.01	.92	0.79	.38	4.60	.03	0.71	.40
Left	4289.62	658.55	4301.40	587.57	4295.50	498.87	4323.11	604.64
Right	4379.30	605.04	4400.75	554.39	4486.88	483.35	4406.82	554.47
Amygdala									0.26	.61	49.96	<.01	0.54	.46	1.63	.20	36.44	<.01	0.01	.92
Left	1423.11	233.49	1407.76	218.07	1432.67	217.41	1401.40	211.66
Right	1513.16	234.61	1518.83	210.23	1545.86	253.06	1518.43	206.54
Accumbens									3.14	.08	0.18	.67	9.33	<.01	0.03	.87	1.19	.28	1.23	.27
Left	550.31	94.58	521.00	107.01	507.79	114.04	521.00	105.91
Right*	575.74	125.52	502.20	105.83	508.05	144.90	491.07	94.43

Note: ARMS, at-risk mental state; HC, healthy controls; T, transition; NT, non-transition; ICV, intracranial volume. Age, ICV, scanning sites, and (gender) were entered as covariates.

^aGender × group × side interaction was significant only in globus pallidus.

*The group difference (HC vs whole ARMS) remained significant even after the Bonferroni correction for multiple comparisons.

As a corroborative analysis, we performed a meta-analysis of group differences in regional volumes (supplementary figures 1 and 2a–e and supplementary table 6). Compared with controls, individuals with ARMS demonstrated significantly larger left caudate (mean difference from control mean: +6.99%) and bilateral lateral ventricles (+24.97% for left side and +8.90% for right side). The biggest group contrast (ARMS-HC) effect sizes were observed for larger left ventricle volume (d = 0.51, p = 2.75 × 10^–4), followed by larger left caudate volume (d = 0.50, p = 5.92 × 10^–5). Each of these region of interests, except right lateral ventricle volume (p = 1.27 × 10^–2), demonstrated group differences even at the Bonferroni-corrected threshold. Group differences for right accumbens (d = −0.48, p = 0.16) and left globus pallidus (only for the male subjects) (d = 0.35, p = 0.49) volumes were not detectably different from zero. Heterogeneity of effect sizes, which was measured by the I² static, was absence to extreme for all brain regions (range: 0%–87%).

There were no significant differences in the regional volumes between the ARMS-T and ARMS-NT groups (table 2).

Other ARMS subgroup analyses showed that (1) temporolimbic structures, except for right hippocampus, were smaller in ARMS-nonremitter compared with ARMS-remitter and that (2) both left and right lateral ventricles were larger in ARMS-P compared with ARMS-G subjects, while these differences were not statistically significant (supplementary tables 7 and 8).

Brain Measures and Clinical Variables in ARMS Individuals

There were no significant correlations between regional volumes and clinical variables (antipsychotic medication dose, duration between scanning and transition, and positive symptoms subscale scores of the CAARMS or the SOPS).

Discussion

This is to our knowledge the first multicenter MRI study to comprehensively examine subcortical and temporolimbic volumes and their laterality in a relatively large high-risk population for psychosis. ANCOVAs controlling for different scanning sites demonstrated that the ARMS individuals had larger volumes of the bilateral lateral ventricles, left caudate, and left globus pallidus, as well as left-greater-than-right asymmetries for these structures, but a smaller volume of the right accumbens, compared with healthy controls. These findings were corroborated using a meta-analytic approach, which could better manage the inter-site variance, for the caudate and lateral ventricles. There were no significant differences in the volumes or lateralities of these structures between the ARMS-T and ARMS-NT populations. Notably, we found no significant effects of antipsychotic medication on the present results. These findings suggest that subcortical changes in psychosis may predate the onset as a general vulnerability factor.

The present study demonstrated that enlarged volumes of the caudate and globus pallidus, which had been reported in schizophrenia,^9,10 exist also in high-risk subjects regardless of later psychosis onset. Leftward lateralization seen in these structures is also congruent with previous studies of schizophrenia.¹⁰ Furthermore, these findings in the globus pallidus are consistent not only with structural studies that reported an enlarged volume in genetic high-risk subjects,^67,68 but also with functional studies that showed greater left-sided activation in schizophrenia.^69,70 A recent study also reported leftward asymmetries in pallidal volume in early adolescents with subclinical psychotic experiences (SPEs), which may suggest a premorbid predisposition of psychosis.⁷¹ In the present study, the globus pallidus volume was increased only in male ARMS subjects, which may have relevance to the notion of sexually divergent effects of genotypic variation associated with schizophrenia on subcortical volumes.⁷² Our caudate finding of a left-greater-than right asymmetry is in line with increased amplitude of low-frequency fluctuations of the blood oxygen level-dependent signal⁷³ and increased dopamine D₂ receptor density⁷⁴ centered upon the left caudate in a genetic high-risk population, while such a leftward asymmetry is not always evident in first-episode^11,19 or chronic¹⁰ schizophrenia patients. While previous studies in clinical high-risk subjects^41,45,47 have not shown such striatal volume changes, a large sample size and minimization of the confounding effects of cannabis⁷⁵ in this cohort might enable us to detect subtle changes in the striatum. Furthermore, as antidepressants could cause striatal atrophy,⁷⁶ a lesser use of antidepressant use in our cohort (table 1) compared to other large ARMS cohorts in previous studies (n = 32/91 [35.2%],⁴¹ 114/274 [41.6%],⁴⁶ or 37/69 [53.6%]⁴⁹) that exhibited a normal caudate volume might partly explain our finding of expanded caudate volume. Handedness could also significantly affect brain morphology especially in the basal ganglia,⁷⁷ but group comparisons controlling for handedness replicated our main findings. Nevertheless, it should be noted that our findings of the pallidum were not evident in the meta-analytic approach, suggesting that changes in pallidal volume are rather small and not detectable constantly in individuals with ARMS, which is a complex and heterogeneous group.^55,56 Although the mechanisms of enlarged basal ganglia volumes in schizophrenia remain unclear, they may reflect abnormal structural plasticity, which could be due to failure of normal synaptic pruning⁷⁸ or neuroinflammation related to endothelial cell activation.⁷⁹

In contrast to the findings of the caudate and globus pallidus, our ARMS cohort had a reduced right accumbens volume. Previous MRI studies have demonstrated reduced accumbens volume in both medicated^24,80 and drug-naive⁸¹ schizophrenia patients but not in genetic⁸² or clinical⁴¹ high-risk subjects, suggesting that volume reduction of the accumbens occurs closer to the time of psychosis onset.⁸² Such a presumably progressive volume reduction in the nucleus accumbens over time should be tested in future using a longitudinal design. Given an inverse correlation between the accumbens volume and positive symptomatology in schizophrenia patients,⁸³ our novel finding of right accumbens atrophy in the ARMS group might be partly explained by more severe subthreshold positive symptoms in our cohort (table 1), compared with previous high-risk studies (sum of the positive sections on the CAARMS = 16.3 ± 7.4 in whole ARMS [n = 69],⁴⁹ sum of the positive sections on the SOPS = 13.5 ± 3.1 in ARMS-T [n = 35] and 11.9 ± 4.1 in ARMS-NT [n = 239])⁴⁶ that did not find volumetric differences in the accumbens.

Although not consistently,⁸⁴ several MRI studies have reported significant volume expansion and altered laterality of the basal ganglia following antipsychotic treatment of schizophrenia.^35,85,86 The persistent blockade of dopamine D₂ receptors induced by antipsychotics might stimulate the proliferation of these receptors, which contribute to the enlargement of the basal ganglia via incrementation of the neuronal size of dendritic trees.⁸⁷ Although several ARMS subjects in this study were taking antipsychotics, our main findings were not related to medication dosage and were also evident in the drug-naive subsample. Indeed, there are several studies that have demonstrated increased basal ganglia volumes in treatment-naive patients with schizophrenia.^26,88 It may be thus suggested that our basal ganglia findings predominantly reflect brain changes intrinsic to the pathology of psychosis, but that such changes could also be affected by disease processes¹⁷ and persistent exposure to dopamine D₂ receptor antagonists.⁸⁹ Likewise, given the possibility that antidepressants also cause thalamic atrophy,⁷⁶ the effect of antidepressants is another consideration for the discrepant thalamic findings.⁴¹

The absence of amygdalar and hippocampal volume changes in our ARMS group was consistent with previous ARMS studies,^{30,46,49–52} supporting longitudinal atrophy in these regions during the course of psychosis.^52,90 While a few high-risk studies have reported amygdalar⁴¹ and hippocampal^41,53,54 atrophy before psychosis onset, this apparent discrepancy may be partly explained by the notion that hippocampal reduction occurs only in specific subregions.^50,53 In addition, although previous longitudinal studies in the ARMS individuals have generally found no relation between hippocampal atrophy and psychosis onset,^48,90–94 it has been reported that only an ARMS subsample with persistent prodromal psychopathology exhibited deterioration of hippocampal volume.⁵² The present study also suggests that the ARMS individuals with persistent prodromal symptoms at follow-up likely exhibit reduced temporolimbic volumes even in early initial prodromal states. Given that ARMS is a heterogeneous concept for its clinical course (ie, later psychosis onset, functional outcome, and other factors),^56,95,96 stratification across ARMS subgroups and examination of detailed subregion effects will be needed to clarify brain morphologic characteristics before psychosis onset.

Another finding of this study is a significant bilateral ventricular enlargement in ARMS individuals. Previous longitudinal studies have demonstrated progressive ventricular enlargement during the transition period from high-risk state to psychosis^46,48,97 and after the onset of schizophrenia, especially in patients with poor outcome.^60,98 In contrast to our findings, several cross-sectional comparisons of ARMS have found no ventricular enlargement at the baseline.^46,48,49 Preliminary comparison of ARMS-G and ARMS-P groups suggests that the ARMS individuals with poor functional outcome exhibit ventricular enlargement at earlier stages (even at intake) compared to those with a good outcome, while such outcome data are not available in previous studies. Presumably, our novel ventricular finding might be attributable to a relatively high ratio of the ARMS population with poor functional outcome. This leftward ventricular enlargement prior to the illness onset might be partly underlain by a genetic mechanism that controls the development of cerebral asymmetry in schizophrenia.⁹⁹

Some limitations of this study should be taken into account. First, this multicenter study used 4 different MRI scanners with different magnetic field strengths that could have confounded the results,¹⁰⁰ although we corrected for the intensity nonuniformity in our MRI data¹⁰¹ and also statistically controlled for the effects of the scanning site. While a corroborative analysis by a meta-analytic method using a random effect model, which could better control for inter-scanner variations, has largely replicated our main findings (supplementary figures 1 and 2a–e and supplementary table 6), such a method using summary measures for each site could not address brain changes in each subject and their relation to clinical variables. Thus, our findings need replication using better calibration approach¹⁰² in future multicenter studies. Second, different criteria (ie, CAARMS or SIPS/SOPS) for ARMS diagnosis at each site might have affected our results. Although the CAARMS and SIPS differ in several psychopathological definitions (eg, time and frequency criteria, functional decline criterion), they provide excellent agreement in the identification of ARMS subjects¹⁰³ and deliver comparable predictive values over follow-up time.⁴⁰ Third, we could not divide the ARMS-NT individuals who showed significant symptom amelioration into full and partial remissioners¹⁰⁴ owing to the lack of relevant follow-up information. Finally, several ARMS subjects dropped out during follow-up (ie, ARMS-UK sample) and the ARMS-T sample was rather small. While this study suggested that subcortical volumes likely reflect general psychopathology, their role as a biological predictor of psychosis onset should be further tested in a larger ARMS-T cohort.

In conclusion, this multicenter ARMS study demonstrated significant volume expansion of the caudate, globus pallidus, and lateral ventricles as well as volume reduction of the accumbens in high-risk subjects regardless of later psychosis onset, which could not be fully explained by the effects of antipsychotic medication. Given the possibility of a relationship between the brain morphology and several clinical factors (eg, functional outcome, prodromal symptomatology, and antidepressant treatment), our results might also support the biological/clinical heterogeneity of ARMS. Finally, our findings suggested that leftward asymmetry in these expanded structures, which is qualitatively similar to that observed in patients with chronic schizophrenia¹⁰ as well as adolescents with SPEs,⁷¹ may represent a general vulnerability to psychopathology.

Funding

This work was supported by Japanese Society for the Promotion of Science (grant number 18K15509 to D.S., 26461738 and 18K07549 to Y.T., 18K07550 to T.T., 19H03579 to S.K., 24390281 to M.S., 16H06395, 16H06399, 16K21720, and 16H06280 to K.K.); the SENSHIN Medical Research Foundation to Y.T. and D.S.; and by the Health and Labour Sciences Research Grants for Comprehensive Research on Persons with Disabilities (grant number 16dk0307029h0003 to M.S., K.M., and M.M) from the Japan Agency for Medical Research and Development (AMED). The study was also supported in part by AMED (grant number JP18dm0307001, JP18dm0307004); UTokyo Center for Integrative Science of Human Behavior (CiSHuB) and by World Premier International Research Center Initiative (WPI) at the University of Tokyo Institutes for Advanced Study (UTIAS), the Ministry of Education, Culture, Sports, Science and Technology (MEXT), Japan to K.K.