In this large sample of bvFTD patients, we investigated the brain structural correlates of mentalizing capabilities (i.e. ToM, a key aspect of SC) via a modified RMET and five different tests reflective of EF. Here we find better RMET performance to be positively associated with GMV within temporal cortical, insular and prefrontal brain regions. Analyses of CTH showed a similar pattern, however more pronounced for the right hemisphere. Associations of EF task performance, on the other hand, were mostly limited to the PFC including the mPFC and the dlPFC, as well as the insular cortex. Post-hoc analyses showed independent associations of RMET performance (i.e. after adjustment for EF performance) with GMV and CTH mainly within the bilateral temporal lobes (most prominently within the left temporal pole) and the bilateral insula. Independent associations of EF were only observed for H5PT and were largely limited to the PFC. For RMET and all tests of EF, we found GMV associations to overlap within the left insula and parts of the dlPFC, exploratory analyses at more liberal thresholds yielded additional overlapping regions, most prominently the left putamen. Source-based morphometry analyses of structural covariance networks yielded comparable results, with RMET performance being more strongly associated with temporal networks and EF performance being more strongly associated with prefrontal networks. Making use of the Neurosynth database, overlap of RMET with functional networks commonly associated with both SC and EF was again primarily located within the temporal and insular region and the dlPFC. Additionally, location based meta-analyses (also within the Neurosynth database) of independent RMET associations (i.e. after adjusting for EF) within the left temporal pole were found to be very strongly associated with terms related to SC and TOM in particular (i.e. “mind tom” “theory mind”) indicating a critical role of anterior temporal lobe atrophy in the development of social cognitive deficits in bvFTD.

Summarized these results indicate that bvFTD-related deficits of ToM and EF may be attributable to atrophy of distinct brain regions, with ToM and EF deficits being more related to temporal and prefrontal atrophy, respectively. Interestingly, deficits of ToM and EF appear to share the insula and distinct regions within the PFC (i.e. dlPFC) as common neural correlates.

Several previous neuroimaging studies have investigated brain structural correlates of SC in bvFTD. Although test batteries and corresponding aspects of SC (e.g. ToM, emotion processing, moral reasoning) differed across these studies, the overall pattern of associated brain regions is comparable to our results, pointing to a discrete network of medial, prefrontal, orbitofrontal, insular and anterior temporal cortical regions being involved in bvFTD related SC deficits, including ToM [8, 18,19,20,21,22,23,24,25,26]. Since bvFTD is a rare disease previous studies were done with comparably small sample sizes (i.e. all N < 30). For healthy humans on the other hand, an extensive number of neuroimaging studies have investigated the neurobiological underpinnings of SC and TOM, with meta-analyses across fMRI studies – including our Neurosynth analysis – corroboratively pointing to a network that includes part of the PFC, the anterior cingulate cortex, the TPJ and the temporal lobes [15,16,17]. In our study, RMET correlates were predominantly located within the temporal cortex – including both lateral and medial aspects and the temporal pole – a pattern that remained evident even after controlling for measures of EF. Note that our study represents a lesion approach, herewith validating results of functional studies in healthy subjects.

Important functions of the lateral temporal lobe include object and face recognition together with language and conceptual processing, all processes relevant in everyday social interactions. The medial temporal lobe with its prominent role in mnestic functions is implicated in memory retrieval of socially relevant information and experiences, while anterior parts of the temporal lobe and the temporal pole are thought to provide entire social semantic concepts [50, 51]. On the other hand, the amygdala, located within the anterior temporal lobe, has a prominent role in emotional processing, with human neuroimaging studies showing consistent activation in response to different kinds of facial expressions [50, 52,53,54,55].

We also observed positive associations of RMET within the bilateral dlPFC, however, these findings were strongly diminished after adjustments were made for EF. Nevertheless, considerable overlap of our results with functional neuroimaging studies was found within the bilateral dlPFC, when compared to Neurosynth based meta-analyses of SC and TOM. The dlPFC – as most parts of the PFC – is critically involved in higher cognitive processes, most prominently EF [56, 57]. Human behavior in its complexity depends on EF due to the necessity of constant evaluation, reevaluation, self-control and flexible adjustment, a concept that naturally also applies to complex human interactions [14, 58]. With respect to the dlPFC, growing evidence from both human and animal research indicates a prominent role for this region in tracking and cognitive evaluation of social interactions, its outcomes and subsequent adaptations [59,60,61] with cognitive perspective-taking, an aspect of SC closely related to ToM, being critically dependent on dlPFC function [62].

In our study, worse performance of EF tests was mainly related to prefrontal atrophy, including ventromedial, dorsomedial and dorsolateral aspects of the PFC. Although results differed to some degree between individual tests, the main pattern is in accordance with fMRI studies in healthy humans and studies in dementia based on glucose metabolism. Here, meta-analyses found EF to be dependent on a distinct network of prefrontal regions [3, 63], also highly converging with results from our Neurosynth meta-analysis. Nevertheless, the meta-analysis by Niendam et al. also highlighted the parietooccipital cortex, the insula and the basal ganglia [63], brain regions that only partially were related to EF in our sample. Of note, the relationship between brain structure and cognitive functions is complex and results of healthy participants may not be entirely comparable to diseased participants with cognitive deficits due to profound pathological brain changes, as in our sample of patients with bvFTD.

Overlap between structural correlates of RMET and EF was consistently observed for the anterior insular cortex and adjacent frontal regions including parts of the inferior frontal cortex, as well as distinct regions within the PFC, partially corresponding to the dlPFC. As detailed above, a role for the dlPFC in both SC and EF is well established. The insula on the other hand serves a number of roles with processing internal signals regarding the physiological state of the body (i.e. interoception) being one of its main functions. Due to a large body of evidence, it has been postulated that the central representation of bodily states (e.g. hunger, thirst, pain) within the insula may be the neurobiological foundation of emotions and its awareness [64,65,66].

While the posterior insula receives as primary homeostatic cortex directly information on bodily states, the anterior insula is supposed to be primarily involved in its emotional processing. Interestingly, neuroimaging studies showed identical anterior insular activation patterns when directly experiencing bodily and emotional distress (e.g. disgust and pain) as compared to seeing facial expressions of other people making these experiences. Based on these observations, the insular cortex has been proposed to process one’s own emotional state but also to predict the emotional states of others [67]. Of note, large clusters of von Economo neurons (i.e. “Mirror Neurons”) can be found in high density within the frontoinsular cortex, a special population of neurons which is thought to be critically involved in socioemotional functions and vicarious learning (i.e. via observing others), a crucial aspect of human social behavior [67,68,69,70,71]. Remarkably, bvFTD seems to affect von Economo neurons in particular [2].

Based on extensive evidence from human neuroimaging studies and its connectivity profile, the insula, and its anterior aspects in particular, has been conceptualized as a hub for multimodal information with an overarching function as switch gate for a number of different brain networks such as salience, default mode and executive networks [72] thus enabling the integration of emotional and cognitive information. With this perspective, our findings of consistent overlap for ToM and EF deficits within the insula seem plausible within a simplified framework of the insula functioning as common linking structure between social cognitive processing on the level of perceptive, and affective processing – primarily located within the temporal lobe – and its cognitive evaluation and integration in behavioral patterns, which in turn largely depends on the prefrontal cortex, including the dlPFC.

Finally, we want to place our results into a conceptual framework of bvFTD and its symptoms as recently developed with the MARS approach [2, 29, 30]. Our results underline the importance of atrophy in frontolateral and frontomedian structures for deficits in EF in bvFTD. On the other hand, results support the importance of atrophy in anterior insula, basal ganglia and frontal structures for deficits in SC and emotion processing, but indicate also important contributions of the temporal lobes to those deficits. Moreover, our results support the hypothesis that deficits in EF and SC in bvFTD share associated and involve dissociated brain structures at the same time.

However, several limitations need to be considered, thus our findings should be kept in perspective. First, we did include patients with possible bvFTD, next to patients with both probable and definite bvFTD. Although all patients with possible bvFTD fulfilled clinical diagnostic criteria of bvFTD and were diagnosed after multidisciplinary discussion of each case, we cannot entirely rule out the possibility of bvFTD phenocopies in our sample, such as primary psychiatric disorder or frontal variant Alzheimer’s disease (AD). With respect to the latter, only limited data regarding CSF based AD pathology were available. However, these data did not indicate a systematic bias (data not shown), while comparable frequencies of co-pathology could be observed as previously published [73]. Moreover, longitudinal stability of (possible and probable) bvFTD diagnosis in our cohort prevent a systematic bias of phenocopy cases. Next, the modified RMET as used in this study only captures one aspect of SC (i.e. ToM). In this context, it also has to be noted that other tests reflective of ToM (e.g. Faux Pas Test, Sarcasm Detection) may provide more accurate information with respect to mentalizing deficits in bvFTD [74]. This is a multi-centric study with neuroimaging data having been acquired on different scanners. Although imaging protocols have been harmonized across participating centers via standard operating procedures, scanner effects can still affect results. Due to aforementioned reasons we did choose not to control for scanner site, in order to prevent excessive loss of information, and assuming high statistical power and validity due to a very large cohort. Of note, our analyses involved mainly correlation analyses, where different scanners and protocols are not assumed to systematically bias results. Data were not equally available for all cognitive tests, resulting in differences with respect to the number of subjects analyzed for each test, which may have had an impact on our results and the level of significance. Since bvFTD is a rare disease and the acquisition of good quality data is cumbersome, we chose not to exclude data for this particular reason. Next, some discrepancies can be noted, when comparing results of VBM and CTH analyses, although the directionalities and spatial patterns (i.e. temporoinsular pronunciation for RMET analyses and prefrontal pronunciation for EF analyses) were similar for both analyses. VBM captures a number of different brain structural aspects including CTH, whereas analyses of CTH are exclusive of other structural variables, thus highlighting CTH only. While both modalities provide information regarding structural aspects of the brain, they should be considered complimentary but not equivalent [75, 76].