This study identified novel neural networks associated with different symptom dimensions of FTD in a large cohort of individuals with schizophrenia (Fig. 1). While prior studies of FTD, including meta-analyses have identified abnormalities in superior temporal gyrus activation [43] and connectivity associated with FTD–the latter specifically for positive FTD [35]. This study suggests that these earlier findings are only one component of a more extensive network. The temporal lobe component of the structural network identified in this study overlaps with previously identified functional findings but connects to a wider range of regions, especially in the occipital lobe which is rarely studied in the context of schizophrenia.
Both positive and negative FTD were found to be related to fronto-occipital brain regions, namely the medial orbitofrontal cortex, anterior cingulate, lateral occipital cortex and negative FTD was also found to be related to the left amygdala. However, anatomical measures related differentially between the two FTD dimensions. The associations of our FTD core network provide further insight into a long-standing controversy in the field: whether FTD emerges from dysfunction in language processing networks (“dyssemantic hypothesis”) [44] or rather from deficits in higher-order cognitive processes (“dysexecutive hypothesis”) [38]. Of note, these core regions were outside canonical language-related circuits [45], but rather associated with cognitive and behavioral control (medial orbitofrontal [46]–[48] and anterior cingulate [49], [50]), affective processing (amygdala [51], [52]) which have been associated with schizophrenia in previous studies [18], [22], [53], and abstract thinking and imagination (lateral occipital cortex [54]–[59]), which has been less commonly associated with schizophrenia [57], [58], [60]. Our findings suggest a potential role for dysfunctional executive processing as a common feature shared across FTD domains. However, it should be noted that positive FTD was also associated with classical language-related regions, suggesting a role for impaired semantic functions especially in the case of positive FTD (Fig. 1B, 1E; Tables 1–3).
A closer look at differences between findings for distinct FTD symptom dimensions demonstrates a dissociation between the structural features of positive and negative FTD. In particular, positive and negative FTD showed opposing patterns of associations with cortical thickness in the orbitofrontal cortex and the rostral anterior cingulate (Fig. 1B-C, 1E-F). Negative FTD showed a positive correlation with cortical thickness in these frontal brain regions, however, it should be noted that these findings were indicative of a relative sparing from a fronto-temporal pattern of atrophy [18], [22] in individuals with schizophrenia, rather than an absolute increase, when compared to healthy controls (Fig. 1). When schizophrenia subjects were compared to healthy controls, widespread smaller cortical surface area, thinner cortex, and smaller volumes were observed (Supplementary Table 2). Different deficiency patterns in two central hubs involved in cognitive control may indicate that differential biological mechanisms or different cellular populations play a role in the emergence of these types of FTD. Additionally, positive FTD was the only symptom dimension that implicated brain regions in the temporal cortex, particularly in language-related areas (Fig. 1B, 1E; Table 2). A previous meta-analysis from our lab highlighted functional changes of the superior and medial temporal gyrus in FTD [43]; central hubs of the human language processing network [61]. The temporal pole, in turn, has been linked to a semantic network involved in creative thinking [62]. Importantly, connectome-based modeling with seeds in the superior and medial temporal cortex edpredicted positive FTD symptom severity, but not any other FTD dimension [21], which is well in line with our own findings. Together, these findings paint the picture of a role for language-related networks exclusively in positive FTD. The results of our study further support the idea of a fundamental neurobiological divergence between positive and negative symptom dimensions [1], [63] which has both been shown for general schizophrenia psychopathology and its neural correlates [21].
Due to its limited spatial resolution, MR imaging does not allow a direct link between macroscopic changes and underlying molecular or cellular pathologies, a key requirement for the development of new therapeutic approaches. Novel methods, however, allow at least indirect inference on these molecular processes. Our virtual histology approach [33], based on gene expression patterns provided by the Allen Human Brain Atlas [41], identified distinct transcriptomic fingerprints associated with each of the three symptom dimensions (Table 4). Common to both the positive and negative FTD dimensions was a transcriptomic signature associated with dendritic spine maintenance and astrocytes. Consistent with this finding, post mortem studies have reported lower dendritic spine density and impaired dendritic plasticity in the brains of individuals with schizophrenia [32], [64]. Mechanistically, loss of dendritic spines has been linked to altered function in human complex 4 (C4) [65]. Complex genetic variation in the C4 gene, in turn, has been linked to schizophrenia risk [66]. Our own finding that neuroanatomical variation associated with both FTD dimensions is situated in brain regions with a high demand for dendritic spine maintenance appears plausible in light of these prior findings. Besides their role in synapse formation during development [67], astrocytes are known to modulate glutamatergic signaling [29], [31]. Pharmacological antagonization of glutamate signaling, in turn, has been shown to induce both positive and negative FTD in healthy subjects [68]–[70]. Beyond these signatures, positive FTD was also found to be associated with brain regions enriched for another non-neuronal cellular fingerprint: microglia. As resident immune cells of the central nervous system, microglia cells are involved in synaptic pruning during development [42], [71].
Our study has several limitations. The large sample size of our study made possible by the ENIGMA consortium has enabled us to identify novel neural networks associated with FTD that earlier, smaller studies were unable to identify. However, pooling over multiple sites limited us to operationalization of FTD through the use of several PANSS items, as rating scales tailored more precisely for formal thought disorder were not available across all cohorts. Additionally, neuroanatomical abnormalities in schizophrenia have been shown to be progressive. Hence, a longitudinal approach might provide an even better link between brain structural alterations and FTD than a cross-sectional approach and will be an important aim for future studies. Finally, we used a recently established virtual histology approach to identify potential cellular contributions to FTD [33]. While we regard this approach as a unique option to generate observer-independent data-driven hypotheses about the cellular underpinnings of brain structural changes associated with FTD, it does not provide direct evidence of cellular pathologies in the brains of individuals with schizophrenia. Postmortem histological studies with ante-mortem FTD ratings are warranted to verify or falsify the virtual histology findings.
In sum, this study demonstrates a convergence between neuroimaging and cellular endophenotypes and is, to the best of our knowledge, the first to associate glial function with formal thought disorder specifically. The identification of a multi-scale associations between structural and transcriptomic networks associated with cellular function is of specific interest clinically, because it provides the basis for linking neuroimaging findings and clinically relevant molecular targets in a way that is not possible with either method in isolation.