Hemispheric contributions toward interoception and emotion recognition in left-vs right-semantic dementia

,


Introduction
Interoception refers to the ongoing perception and representation of the internal physiological condition of the body (Cameron, 2001;Craig, 2002Craig, , 2009Critchley and Harrison, 2013) and is a considered to be a bi-directional process facilitating brain-body communication (Chen et al., 2021). Recent theories of interoceptive predictive processing propose that the brain makes predictions or expectations about the body's state and compares these predictions/expectations with incoming physiological signals, with the aim of minimising prediction errors to save energy over time and balance allostatic load (Barrett, 2017;Barrett and Simmons, 2015;Kleckner et al., 2017;Migeot et al., 2022;Seth, 2013). Beyond contributing to allostatic regulation, interoception has been proposed to be involved in a range of processes including consciousness (Candia-Rivera, 2022;Damasio and Damasio, 2023) and meta-cognition (Nikolova et al., 2022). It is now increasingly recognised that interoception plays a role in social cognition. Interoception has been proposed to be involved in emotion regulation (Füstös et al., 2013;Klein et al., 2021), emotion recognition (Adolfi et al., 2017;Critchley and Garfinkel, 2017;Critchley and Harrison, 2013;Couto et al., 2015;Fittipaldi et al., 2020;Hazelton et al., 2023;Salamone et al., 2021;Terasawa et al., 2014), emotional arousal (Candia-Rivera et al., 2022), emotional sensitivity (Garfinkel et al., 2014), and self-perspective taking (Engelen et al., 2023). The hemispheric involvement of interoception and emotion recognition has been largely understudied and represents a unique opportunity to undercover structures involved in both processes.
Neuroanatomical models of interoception indicate that the anterior insula and anterior cingulate cortex are key structures involved in interoception (Craig, 2002(Craig, , 2005(Craig, , 2009. The insula integrates the internal physiological condition of the body with external environmental demands (Craig, 2002(Craig, , 2009Ibañez and Manes, 2012;Ibanez and Schulte, 2020;Menon and Uddin, 2010). Meta-analytic evidence from functional neuroimaging studies has reported activation in key regions such as the insula and thalamus bilaterally during interoception tasks (see for meta-analysis Adolfi et al., 2017;Schulz, 2016) with significant overlap between interoception, emotion recognition, and theory of mind in the anterior insula, anterior cingulate cortex, and temporal pole (Adolfi et al., 2017). Recently, Migeot et al. (2022) proposed an allostatic-interoceptive predictive coding framework that would explain the changes observed in behavioural-variant frontotemporal dementia. Allostasis is the process of the brain continually adjusts the conditions of the body (e.g., blood pressure, temperature, blood sugar) to anticipate or adapt to environmental changes (Guidi et al., 2021). This model posits that the allostatic-interoceptive network, which comprises hubs of the salience network (bilateral insula, anterior cingulate, thalamus, striatum, amygdala, hypothalamus) and of the default mode network (angular gyrus, precuneus, hippocampus and prefrontal cortices) would also contribute to cognitive processes such as memory, language, and emotion processing (Kleckner et al., 2017;Migeot et al., 2022;Seth, 2013). This framework suggests that interoceptive dysfunction may underlie behavioural and emotional difficulties in behaviour-variant frontotemporal dementia (Migeot et al., 2022). Importantly, this has not been applied to other dementia syndromes with similar patterns of neurodegeneration.
Whilst interoception likely involves inter-hemispheric functional connectivity (Adolfi et al., 2017;Schulz, 2016), evidence suggests that damage to key areas in right hemisphere structures in the interoceptive network may cause more severe interoceptive difficulties. For instance, Critchley et al. (2004) reported enhanced right anterior insula activity during interoceptive attention, as well as an association between the structural integrity of the right anterior insula and interoceptive accuracy. Using magnetoencephalography, Babo-Rebelo et al. (2016) found greater neural responses to heartbeats occurred when participants were engaging in more self-relevant thoughts compared to less self-relevant thoughts in the right dorsal and ventral anterior insula. Importantly, these changes in activation were not found in the homologous regions in the left hemisphere. Recent evidence from unilateral stroke patients has reported less sensitivity to interoceptive signals as rated by an interoceptive questionnaire in right-hemisphere stroke patients with personal neglect than left-hemisphere stroke patients and controls (Raimo et al., 2020), however, neuroimaging analyses were not conducted. These findings suggest a common neurobiological underpinning of interoception and emotion. To date, however, no study has investigated interoception and emotion in a lesion model characterised by predominantly left or right hemispheric atrophy. Here, we employed a lesion model approach to investigate hemispheric contributions toward interoception and emotion recognition.
Semantic dementia (SD) is a progressive neurodegenerative illness characterised by fluent albeit empty speech, with loss of semantic knowledge observed across testing modalities (Gorno-Tempini et al., 2011;Hodges et al., 1992; see for review Landin-Romero et al., 2016). SD is characterised by brain atrophy to the anterior temporal lobe and anterior fusiform gyrus (Gorno-Tempini et al., 2011). Brain atrophy is typically left-lateralised (i.e., left-SD), however, in about 30% of cases, atrophy tends to be more pronounced in the right hemisphere early in the disease process (Chan et al., 2009). These patients presenting with right hemisphere atrophy have been variously referred to as right anterior temporal lobe frontotemporal dementia (Ulugut Erkoyun et al., 2020) and semantic behavioural variant frontotemporal dementia (Younes et al., 2022), although no agreed upon consensus criteria exist. Here, we use the term right-SD to reflect the patients with predominant right anterior temporal lobe atrophy. Unlike left-SD cases, in right-SD, the initial symptoms of the disease include difficulty recognising familiar people (i.e., reduced person-specific semantic knowledge), episodic memory impairments, and behavioural changes such as disinhibition, apathy, and reduced empathy (Ulugut Erkoyun et al., 2020). These two distinct syndromes provide a unique opportunity to understand interoception and emotion, given the predominantly lateralised patterns of atrophy in either the left or the right hemisphere.
Whilst right-SD patients present with non-language features early on, non-language features also emerge in left-SD over the disease course . In particular, difficulties in emotion recognition are increasingly recognised in both syndromes (see for review Fittipaldi et al., 2019; see also Hutchings et al., 2021;Gressie et al., 2023), with abnormal physiological responding to emotional stimuli reported in combined samples of left-SD and right-SD (Kumfor et al., 2019) and in right-SD (Marshall et al., 2018a(Marshall et al., ,2018b. To date, it is unclear whether these emotional difficulties in left-SD or right-SD are mediated by interoceptive processing disturbances. Importantly, in both left-and right-SD, damage to key areas in the proposed allostatic-interoceptive network occurs, with one hemisphere more predominantly impacted than the other. Research in interoception in these syndromes is currently scant. Only one study to date has to investigated interoception in left-SD, and reported impaired cardiac interoception in this group which was related to reduced carer-rated emotional sensitivity (Marshall et al., 2017). Impaired interoception may therefore represent a potential mechanism underlying emotion processing disturbances in left-SD. Of relevance here, impaired interoception in left-SD was associated with reduced integrity of the right insula, right amygdala, and right anterior cingulate cortex via regions of interest analyses (Marshall et al., 2017), highlighting the involvement of right-hemisphere structures in interoception. Whole brain imaging analyses are required to understand the neural mechanisms underlying interoception in these patients. In addition, no study to date has investigated interoception and the neural correlates underlying this ability in right-SD, despite predominant right hemisphere atrophy in these patients.
The first aim of this study was to compare interoception profiles in left-and right-SD patients. We hypothesised that both groups would show impaired interoceptive accuracy, but that this deficit would be greater in right-SD, due to the predominant right-lateralised atrophy in the anterior temporal lobe and insula in this patient group. Second, we aimed to examine emotion recognition in each patient group. Based on existing literature, we hypothesised that while both patient groups would show emotion recognition difficulties compared to controls, this deficit would also be greater in right-SD than left-SD. Third, we aimed to test whether interoception could be a possible mechanism mediating emotion recognition performance. We hypothesised that impaired interoception would predict worse emotion recognition regardless of clinical diagnosis. Finally, we aimed to identify the common neural correlates of both interoception and emotion recognition. We hypothesised that worse interoception and emotion recognition would be associated with reduced structural integrity of the insula as a common structure necessary for both. Moreover, we expected greater involvement of right hemisphere structures involved in interoception than left hemisphere structures.

Participants
Fourteen individuals diagnosed with SD (8 left-SD, 6 right-SD) and 21 healthy controls were recruited through FRONTIER, the youngeronset dementia research clinic based in Sydney, Australia. All participants underwent a neuropsychological examination and had an MRI scan. All patients were assessed and diagnosed by an experienced behavioural neurologist. Diagnosis of left-SD was based on the current diagnostic criteria (Gorno-Tempini et al., 2011), whereas diagnosis of right-SD was based on presenting clinical symptoms (e.g., difficulty recognising familiar people, reduced empathy, social disinhibition, increased mental rigidity and/or compulsive behaviours, and/or hyper-religiosity) as no current diagnostic criteria for right-SD exists. In addition, each patient's clinical diagnosis of left-or right-SD was supported by a predominant side of atrophy (i.e., greater left-or right-lateralised anterior temporal lobe atrophy on MRIdetermined by the experienced behavioural neurologist). All control participants scored >88/100 on the Addenbrooke's Cognitive Examination-III . For all participants, exclusion criteria included: current or prior history of psychiatric illness, significant head injury, alcohol or substance abuse disorder, presence of another neurological disorder, or limited proficiency in English.

Disease duration and severity
The Frontotemporal dementia Rating Scale (FRS) was used as a measure of disease severity and reflects changes in functional abilities and behaviours in everyday life (Mioshi et al., 2010). The FRS provides an index of disease severity (very mild, mild, moderate, severe, very severe, profound) and associated Rasch scores ranging between 4.12 and − 4.99, where higher scores reflect higher functional capabilities.
Disease duration was calculated as the number of years between the assessment date and estimated date of symptom onset.

Neuropsychological examination
Addenbrookes Cognitive Examination (ACE-III) was used to measure global cognition . This measure assesses cognitive domains including memory, attention, language, fluency, and visuospatial ability. A total score out of 100 was calculated, with higher scores reflecting better cognitive ability.
Sydney Language Battery (SYDBAT) was used to measure language ability (Savage et al., 2013). This test measures four aspects of language: naming, word comprehension, semantic association, and word repetition. Each subscale is out of 30, with higher scores reflecting greater language ability.
Trail Making Task (TMT) was used to assess attention and cognitive flexibility (Tombaugh, 2004). This measure has two parts, where time and errors are recorded. In part A, participants are required to connect circles numbered 1 to 25 as quickly as possible. In part B, participants are required to alternate between numbers and letters in order, again as quickly as possible. Time difference between time to complete Part B and time to complete Part A was used as a measure of cognitive flexibility.
Rey Figure Complex (RCF) was used to measure visuospatial construction skills and non-verbal episodic memory (Rey, 1941). Participants are first instructed to copy a complex figure, with copying time recorded. After a delay of 3 min, participants are required to recall the figure from memory. Higher copy scores represent greater detail copied or recalled.

Emotion recognition
The Facial Affect Selection task (FAST) was used as a measure of emotion recognition (Kumfor et al., 2014a,b;Miller et al., 2012). Here, participants viewed an array of seven faces which displayed the six basic emotions (happiness, sadness, fear, anger, disgust, and surprise) and a neutral expression. Participants were required to point to the face corresponding to the emotional label spoken by the examiner, (e.g., 'Point to the happy face'). This task was chosen to reduced verbal loading. Responding for this task was untimed with no feedback provided. Non-morphed images from the NimStim database were used (www.macbrain.org), which were converted to grayscale and cropped to remove non-facial information (e.g., hair) (see Kumfor et al., 2014a,b for an example of stimuli used). Total scores were then converted to a percentage correct for analysis.

Interoception
Interoception was measured with two 3-min behavioural conditions, where attention was directed to either internal or external cues (Fittipaldi et al., 2020). In each condition, participants were instructed to respond via button press each time they: 1) felt their own heartbeat, without physically measuring or checking their pulse (cardiac interoception); or 2) heard an audio recording of a heartbeat, based on a variable rate equivalent to an average heartbeat of 60 BPM (exteroception, control). In the exteroception condition, participants were instructed to respond immediately after hearing the sound, without anticipating the next beat.

Physiological recording
Electrocardiographic (ECG) signals were recorded using an 8/35 Powerlab Data Acquisition System (ADinstruments, Castle Hill, Australia) with Labchart Pro for Windows (v. 8). ECG was measured via a 3-lead placement using 3 Ag/Ag-Cl electrodes and ECG signals were digitised at sampling rate of 1000 Hz/s. R-wave values from the ECG signal were identified using the peak detection module in LabChart Pro and were verified manually.

Interoceptive accuracy
Performance accuracy for both cardiac interoception and exteroception was calculated using the mean distance index (see supplementary materials for details; de la Fuente et al., Fittipaldi et al., 2020). In brief, a comparison was made between the frequency of the event in the condition (i.e., actual heartbeat or audio cue) and the frequency of the participant's button press responses in that condition. A mean distance index of 0 reflects an exact match between frequencies, whereas higher scores indicate larger distances between frequencies, and therefore, worse performance. A coefficient of variation was calculated to assess the regularity of participants' responses and to account for any potential lapses in attention. This coefficient was calculated as the standard deviation divided by the mean response frequency over the duration of each condition (segmented into 60-s overlapping windows). Response windows where the coefficient of variation was >0.5 did not meet the threshold of regularity and were not included in the final mean distance index calculation. Two control participants were identified as significant outliers on the exteroception task (>3 x interquartile range) and were removed from subsequent analyses.

Neuroimaging analyses 2.4.1. Data acquisition
Whole-brain structural MRI data were obtained using a GE 3 T Scanner, with 256 × 256 x 200 slices, echo time/repetition time = 2.5/ 6.7 ms with 1 mm 3 isotropic resolution and a flip angle α = 8 • . MRI scans were available for 21 controls, 8 left-SD, and 5 right-SD. One right-SD scan was used for diagnostic purposes only due to movement artifact.

Pre-processing
MRI data were analysed using the FSL suite (http://fsl.fmrib.ox.ac. uk) (Ashburner and Friston, 2000;Mechelli et al., 2005;Smith et al., 2004;Woolrich et al., 2009). Structural images were first extracted using BET, then tissue segmentation was conducted via automatic segmentation (Zhang et al., 2001). Then, grey matter partial volume maps were aligned to Montreal Neurological Institute (MNI) standard space (MNI152) using non-linear registration (FNIRT) which uses a b-spline representation of the registration warp field (Rueckert et al., 1999). A study specific template was then created using equal numbers of participants from each diagnostic group and the native grey matter images were non-linearly re-registered. Modulation of the registered partial volume maps were carried out by dividing them by the Jacobian of the warp field. Finally, the modulated, segmented images were smoothed with an isotropic Gaussian kernel with a sigma of 3 mm.

Statistical analyses 2.5.1. Behavioural analyses
Clinical and behavioural data were analysed using IBM SPSS version 28. One-way ANOVAs were conducted for continuous variables (e.g., age, education, cognition, emotion recognition) and chi-square analyses for categorical variables (e.g., sex).
Both interoception and exteroception mean distance indexes were square root transformed to meet normality assumptions (Tabachnick et al., 2007). The square root transformed scores for each index are used throughout the manuscript. The analyses were re-run on the raw data and yielded a similar pattern of results.
A repeated measures ANOVA was used to determine group differences in interoceptive accuracy with condition (interoception vs. exteroception) as the within-subjects variable and diagnostic group (left-SD vs. right-SD vs. Controls) as the between-subjects variable. Post hoc tests using Sidak correction were used to investigate interactions while correcting for multiple comparisons. Next, a multiple linear regression was conducted to investigate significant predictors of emotion recognition performance. Predictors included demographic variables (e.g., age, sex, education), diagnosis dummy coded in reference to controls, and interoceptive accuracy. All predictors were entered simultaneously using the enter method. Finally, partial correlations were conducted controlling for disease duration and severity, to investigate whether the relationship between interoception and emotion was influenced by disease. Supplementary analyses investigated within-group associations between cognition, emotion, and interoception (Supplementary Table 2). Statistical significance was set to p < .05, unless otherwise stated.

Neuroimaging analyses
Voxel-wise general linear models (GLM) were created to investigate grey matter intensity differences across the whole brain between groups using t-tests with permutation-based, non-parametric tests, with 5000 permutations per contrast (Nichols and Holmes, 2002). Neural correlates of interoception and emotion recognition were investigated via separate GLMs in all patients and controls combined to increase variance in scores (Sollberger et al., 2009). Inclusive masking analyses were conducted to identify common brain regions associated with interoception and emotion. The statistical threshold was set to p < .001 uncorrected for multiple comparisons, with a conservative cluster extent threshold of 100 voxels, unless otherwise specified. This threshold was chosen to balance type I and type II errors (Lieberman and Cunningham, 2009;Noble et al., 2020). Anatomical locations of significant results were overlaid on the MNI standard brain and anatomical labels were determined with reference to the Harvard-Oxford probabilistic cortical and subcortical atlases.

Demographics
Demographic characteristics of the study participants are displayed in Table 1. In brief, no differences were observed in age or sex across diagnostic groups. Differences in years of education were observed, with control participants reporting more years of education than both patient groups (left-SD: p = .03; right-SD: p = .02). Right-SD patients had a longer disease duration (p = .03) and greater disease severity (p = .03) than patients with left-SD.

Patterns of brain atrophy
Patterns of brain changes characteristic of each SD syndrome were observed ( Fig. 2A and B, Supplementary Table 1). In brief, compared to controls left-SD showed reduced grey matter density in predominantly left-lateralised temporal structures, including the temporal fusiform cortex, inferior temporal gyrus, middle and superior temporal gyrus as well as the left frontal pole and left insula. Bilateral atrophy was also observed in the temporal pole, hippocampus, amygdala, and orbitofrontal cortex in left-SD. In right-SD, compared to controls, reduced grey matter density was observed in the bilateral temporal pole, insula, inferior temporal gyrus, temporal fusiform cortex, hippocampus, amygdala, as well as the right orbitofrontal cortex. Direct comparisons between patient groups showed lower grey matter density in the left frontal pole in left-SD compared to right-SD (Fig. 2C, Supplementary  Table 1). Whereas right-SD showed reduced grey matter density in the right sided structures only, including the right temporal pole, inferior temporal gyrus, middle temporal gyrus, superior temporal gyrus, hippocampus, thalamus, and insula than left-SD (Fig. 2D, Supplementary  Table 1).

Neural correlates of interoception and emotion recognition
Worse interoception was associated with reduced structural integrity of right sided structures, including the right insula, temporal gyrus (superior, and middle), thalamus, hippocampus, temporal pole, temporal fusiform cortex, planum polare, and angular gyrus (Table 3, Fig. 3A).
Worse emotion recognition was associated with reduced structural integrity of bilateral regions including the orbitofrontal, temporal pole, middle temporal gyrus, superior temporal gyrus, inferior temporal gyrus, temporal fusiform gyrus, hippocampus, amygdala, cerebellum as well as the right insula and right frontal pole (Table 3, Fig. 3B).
Overlapping brain regions involved in both interoception and emotion recognition are shown in Table 3 and Fig. 3C. In brief, worse interoception and emotion recognition were associated with reduced structural integrity of the right-lateralised regions including the insula, temporal pole, middle and superior temporal gyrus, temporal fusiform cortex, hippocampus, and thalamus.

Discussion
This study is the first to investigate interoception and emotion in individuals diagnosed with left and right semantic dementia. Our analyses identified worse interoception in right-SD patients than in both left-SD patients and controls. Moreover, whilst all patients showed worse emotion recognition than controls, worse interoception predicted deficits in emotion recognition. This finding provides further evidence for the role of interoception in emotion processing. In addition, our neuroimaging analyses identified a biological underpinning of both interoception and emotion recognition. Whilst emotion recognition relied on widespread bilateral frontal and temporal brain structures, interoception primarily relied on the integrity of right-lateralised structures including the right insula, temporal pole, hippocampus, thalamus, middle, and superior temporal gyrus. In the following sections, we discuss how these findings advance our current understanding of left-and right-SD and symptoms experienced beyond the domain of language impairment, as well as our theoretical understanding of the hemispheric contributions toward interoception and its relation to emotion recognition.
Interoceptive deficits have been identified in neurological conditions involving damage to fronto-temporo-insular brain regions (Adolfi et al., 2017;García-Cordero et al., 2016;Hazelton et al., 2023;Raimo et al., 2020;Salamone et al., 2018Salamone et al., , 2021. Here, we observed a unique interoceptive deficit in right-SD, with no observable differences on an exteroception, control task. Previous research has shown that abnormal autonomic processing occurs during emotional contexts in combined samples of left-and right-SD cases (Kumfor et al., 2019;Marshall et al., 2018aMarshall et al., ,2018b and right-SD (Marshall et al., 2018a(Marshall et al., ,2018b. Moreover, individuals with either left-and right-SD also experience difficulties in emotion recognition Hutchings et al., 2021). Taken together with our current findings, it is plausible that people with right-SD are no longer able to detect and interpret signals arising from within their own bodies and use this information to inform and update their behaviours (Baez et al., 2016;Birba et al., 2022;Ibañez and Manes, 2012). This finding also fits with theories of interoceptive predictive coding and may suggest that allostatic overload occurs in right-SD,  where an increased mismatch between interoceptive predictions and interoceptive signals is taking place (Barrett and Simmons, 2015;Barrett, 2017;Birba et al., 2022;Migeot et al., 2022;Seth, 2013). This deficit in interoception in right-SD is likely to have a widespread impact beyond emotion recognition and may also contribute to the behavioural and somatic symptoms reported in this syndrome (Ulugut Erkoyun et al., 2020). In our neuroimaging analyses, we found that worse interoception was associated with reduced structural integrity of key interoceptive regions predominantly in the right hemisphere. Our findings further support the role of the right insula in interoception (Craig, 2002(Craig, , 2009Cameron, 2001;Critchley et al., 2004), with damage to this area previously associated with worse interoception in behavioural-variant frontotemporal dementia, frontoinsular stroke, and left-SD patients (García-Cordero et al., 2016;Salamone et al., 2021;Marshall et al., 2017). Most research to date has examined the insula as a single homogenous region, whereas this brain region can be parcellated into anywhere from three (posterior, mid, anterior) to thirteen subregions (Uddin et al., 2017). Moreover, worse interoception was associated with (B) Exteroception compared between groups; (C) Emotion recognition compared between groups; (D) Scatterplot of interoception vs emotion recognition. In (A) and (B), scores displayed are square root transformed, with higher scores represent worse performance. (C) Higher scores represent better performance. In (D), line represents overall line of best fit. *p < .05; **p < .01; ***p < .001; Circles = controls; Triangles = left-SD; Squares = right-SD; SD = Semantic Dementia. Missing data: Emotion recognition %: 3 controls. Note. SE = Standard Error of b; SD = Semantic Dementia.
J.L. Hazelton et al. reduced integrity of the right thalamus, which has also been previously related to interoception (Berntson and Khalsa, 2021). From a network perspective, the insula and thalamus have both been proposed to be involved in the allostatic-interoceptive network (Kleckner et al., 2017;Migeot et al., 2022) and support autonomic functioning (Beissner et al., 2013), with the thalamus relaying information about the body to the insula as a key hub for interoceptive processing (Chen et al., 2021).
Indeed, the insula and thalamus have previously been shown to be functionally (Cauda et al., 2011) and structurally via white matter pathways (Ghaziri et al., 2018) connected. Therefore, it is likely that the pathological changes in these brain regions and their associated white matter tracts as a result of neurodegenerative processes may disrupt effective interoceptive processing. Our neuroimaging analyses also identify the brain regions that support both interoception and emotion processing, demonstrating their biological coupling. In line with our current findings, the right insula, thalamus, temporal pole, superior temporal gyrus and temporal fusiform cortex have been previously implicated in both interoception and emotion, albeit in separate functional imaging studies (see for metaanalysis Adolfi et al., 2017). As detailed above, the insula plays a key role in interoception and emotion, by integrating the internal and external environments into a coherent picture of the current bodily state (Chen et al., 2021). In relation to emotion, the temporal fusiform cortex, or 'fusiform face area' has been previously involved in face-processing of both static and dynamic expressive faces (Ganel et al., 2005;Kawasaki et al., 2012) and the superior temporal gyrus is thought to be involved in understanding emotional prosody and changeable face cues (Engell and Haxby, 2007;Hutchings et al., 2021;Mitchell et al., 2003;Narumoto et al., 2001). Moreover, we found evidence for reduced structural integrity of the hippocampus, which is involved in emotional memory (Kumfor et al., 2014a,b;Phelps, 2004), and of the parahippocampal region, which is involved in deciphering social contexts (Baumann and Mattingley, 2016). Taken together, our neuroimaging results support the hypothesis that when atrophy spreads to these brain regions, particularly in the right-hemisphere, patients are no longer able to integrate internal interoceptive information with external social cognitive information, resulting in the impaired emotion recognition observed here. Unlike in previous studies (Marshall et al., 2017), left-SD patients performed here similarly to controls on the interoception task. This discrepancy may reflects differences in interoception measurements and their associated cognitive processes needed during the tasks, for example language and working memory as in the Marshall et al. (2017)'s study. These cognitive processes may be thought of as domain-general functions that support interoception and emotion. Here, by minimising the verbal and working memory demands of the task, we sought to disentangle this complex interaction and found that interoception involved the integrity of right hemisphere brain structures, including the insula and thalamus. It remains plausible that processes that enable the cognitive appraisal of the interoceptive signal (e.g., what does it mean when one's heart beats faster, or when one's stomach grumbles?) may be more reliant on left-hemisphere structures. This loss of meaning of interoceptive signals, rather than a loss of interpreting the raw visceral interoceptive cue, fits together with previous reports of alexisomia, or reduced awareness of somatic sensations, in left-SD (Gan et al., 2016). Future research modulating the intensity of the interoceptive signal (e.g., active vs. passive conditions) combined with interoceptive awareness measures (e.g., interoceptive sensitivity and magnitude (Khalsa et al., 2018)) may help determine whether there is a deficit in sensing the physiological change of the body and/or the cognitive appraisal of this physiological change. Another consideration is the disease stage of the left-SD patients included in the current study. Longitudinal studies in SD have shown that overtime, brain atrophy will progressively involve the homologous regions in the right hemisphere . Therefore, over time, patients with left-SD may develop disturbances associated with processing the raw visceral signal itself, rather than having difficulties with associating meaning with the interoceptive cue. Whilst left-SD patients in the current study showed reduced integrity in some right-hemispheric structures, these patients did not exhibit damage to the right insula, which may explain their relatively intact interoceptive performance.
Although our study represents the first investigation of lateralisation of interoception and emotion in two rare dementia syndromes, several caveats should be noted. Firstly, we had a relatively small number of participants in each patient group, therefore our results must be interpreted with some caution. Previous literature in these patient groups has also yielded a similar number of patients per group (Hutchings et al., 2021;Marshall et al., 2018aMarshall et al., ,2018b, however, future studies will benefit from recruiting larger samples of well-characterised patients to further understand how interoceptive cues inform and shape social interactions. Indeed, larger patient cohorts will also allow for multimodal neuroimaging (i.e., structural and functional) to further replicate and extend upon our findings. Structural neuroimaging modalities allow us to understand brain regions necessary for behaviours (Adolphs, 2016;Irish and van Kesteren, 2018), whereas functional imaging modalities allow us to identify brain activity correlated with task performance (Lindquist et al., 2012). Importantly, findings from lesion model approaches also localise to specific brain networks that are functionally connected (Darby et al., 2018). Together, both imaging modalities provide crucial information needed to understand brain-behaviour relationships. Future studies that utilise experimental designs whilst patients are in the scanner will elucidate the link between structural correlates and functional connectivity. Moreover, it should be noted that our right-SD patient group had a longer disease duration and severity than our left-SD group, which may have had some influence on our findings. Encouragingly, we observed a similar relationship between interoception and emotion recognition when controlling for disease specific measures in patients. These results, however, should be interpreted with caution due to the small sample size and further reiterate the need for future research in these patient groups. In the current study, although all right-SD cases included presented with initial features of the disease (e.g., difficulty with recognising familiar people, reduced empathy, increased mental rigidity and/or compulsive behaviours), the language profiles were similar between left-and right-SD cases, suggesting that the right-SD cases were at a more advanced disease stage where language also was impaired. This pattern of longer disease duration and/or disease severity has also been observed in other direct comparisons of left-and right-SD groups (Gressie et al., 2023;Kumfor et al., 2016;Marshall et al., 2018aMarshall et al., , 2018bYounes et al., 2022). This delay in diagnosis may reflect the complex clinical presentation of right-SD, where the behavioural symptoms in right-SD may not be identified as early as language features in left-SD. The establishment of clinical diagnostic criteria for right-SD will be an essential to aid in earlier disease detection of this disease entity. Additionally, our study was cross-sectional. Longitudinal studies of interoception in dementia syndromes are required to understand when interoceptive deficits occur in the disease course. Indeed, whether deficits in interoceptive processing may represent a biomarker of disease progression will be useful for pharmacological interventions. Moreover, whilst we aimed to minimise confounds that may have impacted our interoception task, such as by using an adaptive measure of interoceptive accuracy and a control task, more methodological work is needed in this area. For instance, future research is needed to consider how physiological factors such as stroke volume, breathing rate, and hemodynamic measures may be involved in interoception in dementia. Next, we measured emotion recognition using static images. Although our emotion recognition task has been used previously in this population and has reduced verbal demands (Hutchings et al., 2021;Kumfor et al., 2016), dynamic and ecologically valid emotion recognition tests will be useful to further understand the interplay between interoception and emotion. Finally, we focused on cardiac interoception in the current study. Other measures of interoception, such as respiration (Farb et al., 2013) and gastric perception (Herbert et al., 2012), and cross-modal measures, such as affective touch (Kirsch et al., 2020;Burleson and Quigley, 2021) rely on similar brain regions. Indeed, future research is needed to fully understand brain-body communication in dementia syndromes. These future directions will benefit from combining fine grained neuroimaging techniques, such as advanced segmentation methods and white matter tractography, with experimental interoceptive measures across different axes of interoception. Broadening our understanding of interoceptive alterations beyond cardiac interoception will be useful to understand the full experience of the person living with dementia and will further inform our theoretical understanding of interoception.

Conclusion
Our study is the first to show interoceptive deficits in right-SD patients, providing further evidence for impairment in this dementia syndrome beyond the language domain. Whilst impaired social cognition is increasingly recognised in both left-and right-SD, we provide evidence for disturbed interoceptive processing as a candidate mechanism underlying these symptoms in right-SD. Impaired interoception may represent a potential biomarker for disease progression and may be useful for future clinical trials investigating pharmacological interventions. By taking a lesion model approach, our findings demonstrate that key right-lateralised brain regions, such as the right insula and temporal lobe, are involved in interoception and emotion recognition. Our study advances our theoretical understanding of interoception as a necessary component of emotion, highlighting the importance of internal perception for successful social navigation.

Funding
This work was supported in part by funding to ForeFront, a collaborative research group dedicated to the study of frontotemporal dementia and motor neuron disease, from the National Health and Medical Research Council (NHMRC) (GNT1037746). JLH is supported by a National Health and Medical Research Council (NHMRC) Postgraduate scholarship (GNT1168597). OP is supported by a NHMRC Leadership Fellowship (GNT2008020). FK is supported by a NHMRC Career Development Fellowship (GNT1158762).

Declaration of competing interest
The authors report no competing interests.

Data availability
The data that has been used is confidential.