In the current large-scale cross-disorder mega-analysis study, we found shared and disease-specific alterations in subcortical volumes and their lateralization among SZ, BP, MDD, and ASD. Moreover, we revealed the ability of classification driven by subcortical volume data to account for diagnosis and cognitive/social functioning, resulting in the suggestion of a new four-biotype classification.
We demonstrated larger LV volume in SZ, BP, and MDD, smaller hippocampus volume in SZ and BP, and SZ-specific smaller amygdala, thalamus, and accumbens volumes and larger caudate, putamen, and pallidum volumes, using a conservative threshold of Bonferroni-corrected p < 0.05 (Fig. 1). In addition, we also found larger LV volume in ASD, smaller thalamus volume in BP, MDD, and ASD, smaller accumbens volume in BP, MDD, larger caudate and pallidum volumes in BP, and smaller hippocampus volume in MDD – although the results did not survive multiple corrections. Despite different numbers of participants, we were mostly successful in replicating the previous studies from ENIGMA WGs, in that group differences were similar between the current and previous studies (Fig. 2). The overall extent to which volumetric alterations occurred was the largest in SZ and this was followed by BP and MDD. ASD showed a tendency of fewer volumetric alterations compared to SZ and BP. This is in line with our previous diffusion tensor study.47 SZ-specific smaller accumbens volume was found, which could be related to impaired dopaminergic reward and learning processes and possible subsequent onset of psychotic symptoms in SZ.48 SZ-specific larger volumes were found in the caudate, putamen, and pallidum. Our previous study also reported larger volumes in the caudate, putamen, and pallidum in SZ,17 which the current study replicated with a larger sample. Prior mouse studies have revealed that behavioral, electrophysiological, and anatomical consequences of dopamine 2 receptor (D2R) perturbations are associated with striatal circuit function, and that D2Rs serve distinct physiological roles in different cell types and at different developmental time points, regulating motivated behaviors.49 Larger pallidum volumes in SZ compared to controls have been reported in other large-scale studies.18,50 The larger pallidum volumes may be accounted for by the effects of antipsychotic medications51 as well as by the chronicity of SZ.32 In the future, it will be necessary to explore distinct effects of antipsychotics and chronicity on the pallidum volume using a large-scale longitudinal dataset. Hippocampus volume was smaller in SZ and BP. Inflammatory cytokine levels are negatively correlated with hippocampus volume in SZ52 and BP.53 This may be a candidate common mechanism for hippocampal volumetric deficiencies in these disorders.
We observed a leftward alteration of lateralization for pallidum volume in SZ (Bonferroni-corrected p < 0.05) and BP (uncorrected p < 0.05), and a rightward alteration of lateralization for caudate volume in SZ (uncorrected p < 0.05) and for putamen volume in BP (uncorrected p < 0.05) (Fig. 3). Our previous study reported a leftward alteration of lateralization for pallidum volume in SZ,17 which the current study replicated with a larger sample size. In addition, prior studies have shown a leftward alteration of lateralization for pallidum volume in subjects with early-onset psychosis,24 subjects with ARMS,33 and even adolescents with subthreshold psychotic experiences who were not on antipsychotics.34 Thus, the leftward alteration of lateralization for pallidum volume may be a trait marker for SZ. Also, the leftward alteration of lateralization for pallidum volume was found in BP at a liberal significance threshold. While the mechanism is unknown, one possibility is that this may reflect a shared neural substrate between SZ and BP, possibly caused in part by such as common genetic factors.54,55 Another possibility is that antipsychotics, which are used not only for SZ but also for BP, have an influence on the leftward alteration of lateralization for pallidum volume. However, this possibility seems unlikely because, as noted, a leftward alteration of lateralization for pallidum volume was found even in adolescents with subthreshold psychotic experiences none of which are medicated with antipsychotics.34 We also found a rightward alteration of lateralization for caudate volume in SZ and for putamen volume in BP (uncorrected p < 0.05). A prior mega-analysis study reported increased right, but not left, putamen volume in BP.56 This is consistent with the current study’s findings.
We revealed that clustering-classification results driven by subcortical volumes could possibly account, to some extent, for diagnosis (Fig. 4a). The most frequent diagnostic group in Clusters E, F, and G was HC. Cluster E was characterized by the volumes close to the average of HCs. Clusters F and G were characterized by large hippocampus, amygdala, thalamus, and accumbens volumes and small LV volumes. Smaller LV volumes and larger hippocampus, amygdala, thalamus, and accumbens volumes may be an indicator for being psychiatrically healthy. The most frequent diagnostic group in Clusters A, B, and C was SZ. Cluster C was characterized by large caudate, putamen, and pallidum volumes, and Clusters A and B were characterized by large LV volume and small hippocampus, amygdala, thalamus, and accumbens volumes. This is in line with the theory of two distinct neuroanatomical subtypes of SZ, in which one subtype has larger basal ganglia volumes and the other has smaller gray matter volumes, especially in the thalamus and accumbens.57 The most frequent diagnostic group in Cluster D was MDD and the least frequent was SZ. Cluster D was characterized by small caudate, putamen, and pallidum volumes as well as moderately small hippocampus, amygdala, thalamus, and accumbens volumes. This is in line with the theory of inflammation-related volumetric deficiencies in MDD.58 Overall, clustering classification based on subcortical volumes may be a useful biomarker to assist diagnosis. Moreover, in the future, it may be possible to reconstruct a new diagnostic system based on subcortical volumes. In the current study, z-scores of regional volumes were calculated according to the distribution of HCs in each MRI protocol. Thus, one assumption of our method is that HCs’ data are available for each MRI protocol.
We also revealed that, across the current diagnostic categories, clustering-based classification results driven by subcortical volumes can possibly account for some of the variance in cognitive/social functioning in the subcohorts (Fig. 4b, Supplementary Fig. 3). This finding is in line with those of prior studies.38–40 Clustering classification based on subcortical volumes may be a predictive biomarker for cognitive/social functioning. In addition, by combining some clusters with normal cognitive/social functioning into one group, a total of four brain biotypes (BB1, extremely smaller limbic regions; BB2, moderately smaller limbic regions; BB3, larger basal ganglia; and BB4, normal subcortical volumes) were obtained (Fig. 5). From a clinical standpoint, subjects who will be classified as belonging to BB1, BB2, or BB3 may possibly need psychiatric treatment or support from others, given their impaired functioning. Regarding this, it should be noted that a few of HCs were categorized not only in BB2 and BB3 but also in BB1. To our knowledge, the current study is the first large-scale study to report this finding. It is suggested that, although these subjects are clinically healthy now, they might be possibly vulnerable given a slight psychological burden. This point is important and may be the first step toward psychiatric prevention using a biological data-driven approach. Next, some subjects diagnosed as having a psychiatric disorder belonged to BB4. To our knowledge, the current study is the first large-scale study to report this finding. It is implied that normal subcortical volumes may be a biomarker of a better prognosis including higher treatment sensitivity and possibilities of recovery even after being diagnosed as having a psychiatric disorder. Overall, in the current study, we expanded the two-type neuroanatomical theory for SZ, developed by Chand et al,57 to a four-type theory for multiple psychiatric disorders and clinically healthy subjects. Notably, we suggest that our current findings could lead to novel classification criteria for psychiatric disorders based on subcortical volumes. It may be possible in the future to reconstruct a new diagnostic system, based on multi-layer information including subcortical volumes and cognitive/social functioning, in accordance with the Research Domain Criteria framework.59 In the future, it is necessary to explore how and when the anatomical differences occur and how these differences are associated with different clinical and cognitive/social outcomes – not only through multimodal human research but also translational research across species.60 These strategies are expected to lead to the development of novel treatment recommendations for psychiatric disorders.
The current study has some limitations. First, the current mega-analysis study is cross-sectional in nature; thus, the volumetric alterations over time in each psychiatric disorder were not examined. Some previous studies, most of which were not large-scale, reported differences in volumetric alterations between first episode and chronic stage. The collection and analysis of a large-scale longitudinal MRI data across psychiatric disorders would be of great value in the future. Second, the subject number ratio of diagnostic groups in the current study was different from that in the real world. Thus, our clustering analysis results should be carefully interpreted, especially if it is applied for practical use in the future. Moreover, a population-based cohort study may be necessary to strengthen our current results. Third, we did not directly compare any of two psychiatric disorders, because we did not have a sufficient number of MRI protocols. For example, we only had three protocols in which both subjects with BP and ASD were scanned. Fourth, by examining four different disorders together, the ability to relate imaging measures to clinical symptoms/severity was almost lost because most symptom assessment scales would be different by diagnosis. Finally, the medication effects on subcortical brain volumes were not explored as they were beyond our scope in this study. As we have already discussed above, it will be necessary in the future to explore distinct effects of antipsychotics and chronicity on the subcortical volume using a large-scale longitudinal dataset.
In the current large-scale cross-disorder mega-analysis study, we found shared and disease-specific alterations in subcortical volumes and their lateralization among SZ, BP, MDD, and ASD. Moreover, we revealed the ability of classification driven by subcortical volume data to account for diagnosis and cognitive/social functioning, resulting in the suggestion of a new four-biotype classification. Our results will contribute to the future creation of novel biological data-driven psychiatry diagnostic criteria, which is expected to support appropriate treatment selection.