Background: Although traditionally conceptualized as a language disorder, semantic variant primary progressive aphasia (svPPA) is often accompanied by significant behavioral and affective symptoms which considerably increase disease morbidity. Specifically, these neuropsychiatric symptoms are characterized by breaches in normative socioaffective function, for example, an inability to read social cues, excessive trusting of others, and decreased empathy. Our prior neuroimaging work identified 3 corticolimbic networks anchored in the amygdala, temporal pole, and frontoinsular cortex: an affiliation network, theorized to mediate social approach behavior; an aversion network, theorized to subserve the appraisal of social threat; and a perception network, theorized to mediate the detection of social cues. We hy-pothesized that degeneration of these networks could provide neuroanatomical substrates for socioaffective deficits in svPPA. Methods: We examined hypothesized relationships between subscores on the Social Impairment Rating Scale (SIRS) and atrophy in each of these 3 networks in a group of 16 svPPA patients (using matched cognitively normal controls as a reference). Results: Consistent with our predictions, the magnitude of atrophy in the affiliation network in svPPA patients correlated with the SIRS subscore of socioemotional detachment, while the magnitude of atrophy in the aversion network in svPPA patients correlated with the SIRS subscore of inappropriate trusting. We did not find the predicted association between perception network atrophy and the SIRS subscore of lack of attention to social cues. Conclusion: These findings highlight specific socioaffective deficits in svPPA and provide a neuroanatomical basis for these impairments by linking them to networks commonly targeted in this disorder.
Semantic variant primary progressive aphasia (svPPA) is a clinical syndrome heralded by progressive impairments in confrontation naming, single-word comprehension, object knowledge, and surface dyslexia/dysgraphia . svPPA typically arises as a result of fron-totemporal lobar degeneration TAR DNA binding protein-43 pathology. Despite its primary conceptualization as a language disorder, svPPA can be accompanied by clinically significant affective and behavioral changes [2-4]. While behavioral and affective symptoms are widely acknowledged in other forms of frontotemporolobar degeneration, they have received comparatively less attention in svPPA. Furthermore, in our clinical experience, these symptoms share a specific set of characteristics, some of which relate to a patient’s preoccupations and some of which center around how a patient relates to others. These socioaffective symptoms in svPPA can include an inability to understand social cues, interpret others’ affective states, or recognize people as familiar, a tendency to trust or inappropriately approach others, or loss of empathy. Despite substantial impact on quality of life, socioaffective neuropsychiatric symptoms in svPPA remain underinvestigated.
The goal of this study was to measure the types and severity of socioaffective symptoms in svPPA, as well as their neuroanatomical substrates. We previously defined 3 amygdalar-cortical networks important for socioaffective behavior in healthy adults based on extant neuroimaging and electrophysiological and neuropsychological lesion data [5, 6] (Fig. 1). The first network, a perception network, is posited to subserve the detection and interpretation of sensory cues, including those that communicate social information. This network includes the ventrolateral amygdala, lateral orbitofrontal cortex, fusiform gyrus, ventral temporal pole, and superior temporal sulcus. The second network, an affiliation network, is hypothesized to play a role in prosocial behavior (e.g., warmth and empathy). This network includes the ventromedial prefrontal, subgenual and rostral anterior cingulate cortices, dorsal temporal pole, hippocampus, parahippocampus, entorhinal cortex, and nucleus accumbens. The third network, an aversion network, is theorized to be involved in avoidant behaviors in social contexts and includes the anterior midcingulate cortex, anterior insula, and putamen. Notably, regions of this network closely overlap with key nodes of the salience network , which mediates the detection and processing of salient (including threatening) stimuli . We previously applied this network model to a clinically heterogeneous group of patients with frontotemporal dementia (FTD) and demonstrated that the magnitude of atrophy in these networks correlated with hypothesized socioaffective symptoms, as captured by the Social Impairment Rating Scale (SIRS) . The SIRS is a 6-domain scale designed by our laboratory to assess the clinical severity of socioaffective symptoms such as social affiliation, empathy, and threat perception.
svPPA is particularly suitable to studying brain-behavior relationships because, although in some ways it is a stereotyped clinico-anatomical-pathological syndrome, heterogeneity in these and other symptoms is present. For instance, some patients appear oblivious to socioemotional cues from others, such as facial expressions [10, 11]. Many patients with svPPA lose empathy [12, 13]. Some patients exhibit reduced fear expression , and our clinical experience includes examples of svPPA patients becoming vulnerable to social threats, including financial scams and putative burglars. Based on our functional--anatomic network model of socioaffective behaviors, the presence of one or more of these types of symptoms in some patients suggests that neurodegeneration is present in the amygdalar-cortical networks described above. These observations also fit well with the topography of svPPA pathology since atrophy is anatomically anchored around many critical nodes of these networks in the anteromedial temporal cortex and amygdala.
In this study, we sought to leverage individual differences to investigate whether the localization and magnitude of atrophy within the 3 amygdalar-cortical networks in svPPA correlated with symptoms in the 3 major socioaffective domains as indexed by the SIRS. We predicted that svPPA patients with greater atrophy in the perception network would exhibit a lack of attention/response to social cues, svPPA patients with greater atrophy in the affiliation network would exhibit greater socioemotional detachment, and svPPA patients with greater atrophy in the aversion network would demonstrate inappropriate trusting and approach behavior.
Materials and Methods
Sixteen right-handed participants (mean age = 64.1 years, SD = 7.6 years; 8 females) with a diagnosis of svPPA were recruited from a longitudinal study at the Massachusetts General Hospital FTD Unit’s PPA Program. The diagnosis of svPPA was rendered based on a comprehensive clinical evaluation as previously described  by a behavioral neurologist (B.C.D.) and speech-language pathologist (D.H. or M.Q.), as well as visual inspection of T1-weighted MRI for typical left anterior temporal lobar atrophy which was present in each case. That is, each case had an imaging-supported diagnosis according to contemporary clinical criteria . svPPA subject demographics and clinical characteristics are represented in Table 1. No svPPA patients harbored a pre-existing psychiatric disorder, other neurological disorders, or developmental cognitive disorder. Further clinical characterization was carried out with concomitant assessments through participation in measures in the National Alz-heimer’s Coordinating Center’s (NACC) Uniform Data Set 3.0 with FTLD Module. These assessments were performed as temporally close as possible to patients’ scan, often on the same day. Further neuropsychiatric patient characterization included the Neuropsychiatric Inventory Questionnaire (NPI-Q)  (see online suppl. Table 3; for all online suppl. material, see www.karger.com/doi/10.1159/000511431). A subset of participants (n = 6) were administered the Social Norms Questionnaire (SNQ22)  and averaged a score of 16.7/22. Nine of 16 patients completed the revised self-monitoring scale  and averaged a score of 21.8/65 (SD = 17). Participants were cognitively characterized with the global Clinical Dementia Rating (CDR) Scale, the Mini-Mental State Examination (MMSE), and tests of attention and executive function (Digit Symbol Task and Trail Making Task), visuospatial construction (Benson figure copy), and memory (Logical memory) (online suppl. Table 4). In addition, we administered the Progressive Aphasia Severity Scale (PASS) to specifically assess language impairment from a patient’s premorbid baseline . The PASS assesses 10 language domains (e.g., articulation, fluency, syntax, and grammar) based on a combination of quantitative testing and informant/patient interviews. Clinicians rate patients on a scale similar to that of the Clinical Dementia Rating (CDR) Scale, from 0 (no impairment) to 3.0 (severe impairment). Additional language assessments documented expected deficits in semantic processing (e.g., semantic associates test, confrontation naming, noun/verb naming, and semantic fluency) (online suppl. Table 5).
To assemble a sample of MRI scans from control subjects, age- and gender-matched normal participants (n = 30; mean age = 65.1 years, SD = 6.2; 14 females) were identified from the Massachusetts Alzheimer’s Disease Research Center Longitudinal Cohort. These participants were assessed as being cognitively normal (CDR = 0; MMSE ≥28; and no neurologic or psychiatric history). All participants gave written informed consent in accordance with guidelines established by the Massachusetts General Hospital/Partners Human Research Committee.
SIRS Structured Clinical Interview and Scoring Method
The Social Impairment Ratings Scale (SIRS) (online suppl. Table 1) is a structured clinical interview performed with the patient’s informant (usually a spouse or adult child), patterned after the Clinical Dementia Rating (CDR) Scale and described in our prior work . The SIRS assesses socioaffective function across 6 domains: (1) lack of attention/response to social cues, (2) inappropriate trusting or approach behavior, (3) lack of adherence to social norms, (4) person recognition difficulty, (5) social withdrawal, and (6) socioemotional detachment. For the purposes of this analysis, we focused on lack of attention/response to social cues, inappropriate trusting or approach behavior, and socioemotional detachment and report on the other domains in supplementary material. Our prior work has shown that these 3 SIRS domains map onto dysfunction of one of the amygdalar-cortical networks described above. More specifically, lack of attention to social cues is associated with dysfunction of the perception network, social detachment with aberrant processing in the affiliation network, and inappropriate trusting or approach behavior with the integrity of the aversion network. Like the CDR, scoring of the SIRS ranges from 0 (completely normal) to 3.0 (severely impaired). A trained clinician (in this study, B.C.D.) assigns ratings based on his/her interpretation of the informant’s description of changes from the patients’ baseline; reliability and validity data have been previously published . The SIRS sum of boxes is calculated as the sum of all component scores.
Structural MRI Data Acquisition and Analysis
MRI data were acquired with a Siemens Trio 3.0 T scanner (Siemens Medical Systems, Erlangen, Germany). Sequences acquired included a high-resolution T1-weighted MPRAGE (TR = 2.3 s, TE = 2.98 ms, FOV = 256 mm, flip angle = 7°, 192 sagittal 1-mm-thick slices, and matrix 240 × 256). A FLAIR sequence was also included for the purpose of ruling out nondegenerative neurological diseases which may have been contributing to patients’ clinical syndromes.
The methods for MRI data analysis were performed exactly as previously described [6, 9]. Each subject’s T1-weighted MPRAGE MRI scan was analyzed using Freesurfer 5.3.0 to estimate cortical thickness and subcortical volume (http://surfer.nmr.mgh.harvard.edu). Each subject’s scan was manually checked for errors in segmentation leading to errors in white matter or pial surface identification, and manual edits were made (this was primarily an issue in the anterior temporal cortex of svPPA patients).
Based on our prior work, the topography of the perception, affiliation, and aversion networks was derived from seed-based analyses of resting-state functional connectivity data in an independent sample of young adults [5, 18]. Specifically, seeds were placed in 3 amygdala subregions, with resultant functional connectivity maps defining each of the 3 networks: a ventrolateral amygdala seed generated a map of the perception network, a medial amygdala subregion seed generated a map of the affiliation network, and a dorsal amygdala seed generated a map of the aversion network. We used a combination of gyral topography and these functional network maps to manually label 16 ROIs per hemisphere (32 total) on the Freesurfer fsaverage cortical surface template. For example, portions of the fusiform gyrus and the lateral oribitofrontal cortex were labeled within the perception network map, a portion of the ventromedial prefrontal cortex was labeled within the affiliation network, and a portion of the caudal anterior cingulate cortex and frontoinsula were labeled within the aversion network. In this study, each svPPA patient’s cortical surface was mapped to the fsaverage template, and these ROIs were used to measure cortical thickness and subcortical volume in each ROI. The measurement from each ROI in each subject was then compared to the age-matched control group to generate a z score. The network atrophy score was calculated as the average of z scores across all ROIs within that network. To assist in interpretation, z scores were converted into percent atrophy scores relative to controls, as previously published .
Relationship between SIRS Scores and Network Atrophy Measures
Analyses were conducted using PASW Statistics 18 (SPSS, Inc., Chicago, IL, USA; alpha = 0.05, 1-tailed). Our 3 main hypotheses were that (1) the severity of lack of attention to social cues would correlate with the magnitude of atrophy of the perception network, (2) the severity of socioemotional detachment would cor-relate with the magnitude of atrophy in the affiliation network, and (3) the severity of inappropriate trusting and approach behavior would correlate with the magnitude of atrophy in the aversion network. To these 3 hypotheses, we ran Pearson correlation analyses. Because we did not have a priori hypotheses concerning laterality effects, this was done separately for the left and right hemispheres of each of the 3 networks.
To examine the specificity of these relationships, we then ran correlation analyses between each of these 3 SIRS subscores and percent atrophy in networks not hypothesized to be correlated with them (e.g., socioemotional detachment and perception network atrophy). Furthermore, we ran correlation analyses between each of the 3 SIRS subscores and 2 “control” networks subserving other aspects of social behavior. These included a mentalizing network, hypothesized to be involved in assessing the mental state of others and consisting of the dorsomedial prefrontal cortex, ventral precuneus and posterior cingulate cortex, and the angular gyrus, and a mirror network, involved in simulating others’ behavior and comprises the premotor cortex, posterior superior temporal sulcus, and the intraparietal sulcus. Although several other cortical and subcortical regions (including other portions of the striatum and cerebellum) are implicated in social cognition, we opted to focus on the above regions as they represent key nodes of the respective networks . To examine whether demographic or clinical variables played any role in these relationships, we ran Pearson correlation analyses between each of the 3 SIRS subscores and the following nuisance variables: age, gender, illness duration, years of education, MMSE scores, global CDR score, and 3 PASS scores (global language, single-word comprehension, and word retrieval).
For any univariate relationships that were found between these variables of less interest and the 3 SIRS subscores, we performed a hierarchical linear regression analysis to ensure that the said variable did not add additional variance beyond that conferred by the hypothesized network atrophy variable. For these analyses, the hypothesized variable was entered into the first block, and the nuisance variable(s) with a univariate relationship with the SIRS score was entered in the second block. Two separate regression models were run for each of the 2 SIRS subscores found to correlate with network atrophy measures (left affiliation/socioemotional detachment and right aversion/inappropriate trust).
Patient Clinical Characteristics
Patients did not differ from controls with respect to age, gender, or years of education (Table 1). Patients’ overall cognitive functional status ranged from fully functionally independent (n = 2) to mild cognitive impairment (n = 10), mild dementia (n = 3), or moderate dementia (n = 1). Subscores on the PASS indicated overall mild impairment in the single-word comprehension (mean 1.0, SD 0.5) domain (i.e., able to carry on a brief conversation but with several instances of word comprehension difficulties) and mild overall language impairment.
Relationships between SIRS Domain Scores and Network Atrophy
Consistent with our predictions, 2 of the 3 SIRS domain scores showed first-order correlations with cortical atrophy in the network hypothesized to subserve them. Specifically, the magnitude of atrophy in the left affiliation network correlated with the severity of socioemotional detachment (r = 0.526, p = 0.037). This SIRS domain did not correlate with the atrophy in the right affiliation network, nor with atrophy in either hemisphere of the perception or aversion networks (all p > 0.05). Similarly, the magnitude of atrophy in the right aversion network correlated with the severity of inappropriate trusting and approach behavior (r = 0.597, p = 0.015). No such relationships were found between this SIRS domain and atrophy in the left aversion network, nor with either hemisphere of the affiliation or perception networks (all p > 0.05) (Fig. 2). Inconsistent with our hypotheses, the magnitude of atrophy in either the left or right perception network did not correlate with the severity of impairment measured by SIRS ratings assessing lack of attention to social cues (both p > 0.05). The correlation matrices for all 6 SIRS domains and atrophy of the affiliation, aversion, and perception networks are shown in online suppl. Table 3. Correlation matrices between SIRS subscores are shown in online suppl. Table 4.
None of the SIRS subscores correlated with atrophy measures in the mentalizing or mirror control networks (all p values >0.05). With respect to the nuisance variables, socioemotional detachment did correlate with PASS global language (r = 0.613, p = 0.012) and PASS single-word comprehension (r = 0.574, p = 0.02), but not with any other nuisance variable. The other 2 SIRS subscores did not correlate with any nuisance variable. We therefore performed a hierarchical linear regression analysis in which percent atrophy in the left affiliation network was added to the first block, and the 2 PASS scores were entered into the second and third blocks. PASS global language impairment showed a trend-level effect, potentially explaining an additional 15% of the variance in socioemotional detachment scores (F[1, 13] = 3.41, R2 = 0.426, p = 0.088). We also entered all nuisance variables into a primary hierarchical linear regression model, with the SIRS score being the dependent variable, percent atrophy in the network hypothesized to be correlated with this as the first block independent variable, and nuisance variables entered as the second block independent variables. In both cases, the nuisance variables did not explain additional variance (p values >0.1).
In a group of svPPA patients, we found that dimensional ratings of impaired socioaffective function, as measured by the SIRS, correlated with the magnitude of atrophy in specific amygdalar networks theorized to instantiate the processing of social approach and social avoidance behaviors. More specifically, the degree of atrophy in the affiliation network, which is hypothesized to mediate prosocial behaviors, correlated with the level of socioemotional detachment on the SIRS subscale. In contrast, the magnitude of atrophy in the aversion (cingulo-insular) network, which we theorize to subserve the processing of social threat, correlated with the level of inappropriate trust behaviors. We did not find support for our hypothesis that atrophy in the perception network would relate to the SIRS domain of lack of attention to social cues. This was likely due to the fact that our sample of svPPA patients did not have significant impairment in this domain, limiting the variance needed to establish an association between this SIRS score and perception network atrophy. These findings situate svPPA in our neural network-behavioral model of socioaffective dysfunction and suggest that specific disturbances in social approach/avoidance behaviors and threat processing in svPPA map onto structural atrophy within discrete cortical circuits.
This study further contributes to a growing literature highlighting socioaffective deficits in svPPA, which are often apparent in clinical settings. For example, previous studies have documented breaches in empathy and warmth as well as egocentrism in svPPA [4, 13, 19, 20]. Moreover, svPPA patients tend to overestimate their own level of empathy . With respect to social perception deficits, a number of studies have documented impairments in the recognition of basic emotions, emotional valence, or the affective states of others in svPPA [4, 10, 11, 22-28], even when confounds of general facial recognition deficits are removed [22, 24, 29, 30]. We previously showed that deficits in facial emotion recognition in svPPA are rooted in diminished concept knowledge of specific emotions . svPPA patients also have difficulty interpreting sarcasm [23, 31] and exhibit deficits in affective Theory of Mind processing . Regarding aversive processing, svPPA patients demonstrate alterations in pain perception , temperature regulation, and the interpretation of interoceptive stimuli (e.g., satiety after meals) . Our study is among the first to highlight social aversion deficits in svPPA, although Mendez et al.  presented case reports of patients with temporal variant FTD who exhibited abnormally heightened social approach and extroversion.
Structural neuroimaging studies support a relationship between these socioaffective deficits and frontal and anteromedial temporal regions commonly atrophied in svPPA. For example, empathic deficits in svPPA have been correlated with the atrophy of the temporal pole [13, 19] and the amygdala . Notably, these studies emphasized structures lateralized to the right hemisphere, in contrast to our findings that the left affiliation network correlated with impairments in social approach. Emotion recognition impairments have been mapped onto the right anteromedial temporal pole, consistent with data supporting social and person knowledge being represented in the anterior temporal lobe . With respect to social aversion, extensive lesion/neuropsychological, electrophysiological, and functional imaging evidence supports a role for the frontoinsular cortex in the perception of aversive stimuli generally (e.g., pain) and socially aversive stimuli specifically , and this region is almost uniformly atrophied with disease progression in svPPA . Similarly, the amygdala is heavily implicated in threat detection , and this is also commonly atrophied in svPPA [37, 38]. While we are not aware of many studies which have probed the neural basis of social threat perception, there is evidence that the right hemisphere (particularly the right amygdala and right extrastriate visual regions) is specialized for detecting negative (e.g., fearful) facial expressions [39, 40].
Overall, our findings resonate with previous neuroanatomical correlates of socioaffective deficits in svPPA but extend these findings by positing a network-based substrate for these deficits in the disorder. More specifically, we propose that atrophy in cortical networks we previously described , anchored in the amygdala and incorporating anteromedial and frontal paralimbic regions commonly degenerated in svPPA, gives rise to these symptoms. This conceptual framework echoes others which have linked symptom profiles in neurodegenerative disorders to atrophy in distributed networks. For example, converging evidence suggests that neurodegenerative diseases topographically target established functional networks [41-43], and that specific symptoms correlate with atrophy in the specific network undergoing degeneration; for example, an amnestic presentation correlates with posterior default network (hippocampal-parietal subnetwork) degeneration, a language presentation correlates with language network degeneration, and a visuospatial presentation correlates with higher-order visual network degeneration .
As with our previous work, this study is limited by the absence of performance-based assessments of social approach and avoidance. For example, we did not perform neuropsychological testing of socioaffective deficits (e.g., testing patients’ performance on tests of the trustworthiness of others) . While we are exploring this topic, we have identified a number of challenges with regard to patients’ ability to perform these kinds of tests in a valid manner, given their conceptual and linguistic impairments. Our study is also limited by a lack of more comprehensive neuropsychological measures (e.g., measures of executive function) which could affect socioaffective functioning in this patient sample. Future studies could also utilize neurophysiologic assessments such as measuring autonomic activity during psychometric testing of socioaffective function. In support of this, Sturm and colleagues  recently used respiratory sinus arrhythmia (a measure of parasympathetic tone) and galvanic skin responses (a measure of sympathetic tone) to show that in FTD patients autonomic outflow irregularities are dependent on which anteromedial temporal lobe is atrophied (with right amygdala atrophy correlating with impaired sympathetic tone and left ventral anterior insula correlating with impaired parasympathetic tone). Nonetheless, despite their subjectivity, there are advantages to using informant-based ratings to characterize socioaffective function. One such advantage is that informants are uniquely able to assess subtle clinical changes from a patient’s baseline . Therefore, ideally, both subjective and objective assessments should be integrated in order to derive brain-behavior assessments in these types of studies. Our study is also limited by its cross-sectional nature. Future efforts which correlate SIRS ratings with longitudinal changes in network atrophy are likely to offer further insights into the neural underpinnings of socioaffective function in svPPA by linking their development to atrophy progression.
Finally, our results raise the possibility of using structural network metrics to improve socioaffective symptom characterization in svPPA and in other neurodegenerative conditions more broadly. Detection of atrophy in vulnerable regions could identify biomarkers predicting the emergence of specific symptoms and provide a neurobiological explanation of aberrant behavior for caregivers. There is also the opportunity for svPPA to inform about circuits promoting socioaffective functions in healthy individuals. The relatively confined nature of atrophy in svPPA (early in the disease course) could provide a type of “lesion” model of normative socioaffective function, and the relatively indolent nature of svPPA progression could allow for longitudinal tracking of these functions as other network nodes become affected.
We thank our patients and family members for their participation, without which this work would not have been possible.
Statement of Ethics
The protocol used for research presented in this study was approved by the Massachusetts General Hospital/Partners Human Research Committee, Protocol No. 2016 P001421. All participants gave written informed consent in accordance with guidelines established by the Massachusetts General Hospital/Partners Human Research Committee.
Conflict of Interest Statement
Dr. Dickerson has served as a consultant for Biogen, Merck, Wave LifeSciences, Arkuda, Novartis, and Axovant. The other authors have no competing interests.
This study was supported by grants from the US National Institute of Deafness and Communication Disorders (R01 DC014296) and the National Institute on Aging (P30 AG062241). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
BC.D., M.C.E., D.L.P., and L.F.B. designed the study. M.C.E. and B.C.D. wrote the manuscript. D.H., M.Q., and B.C.D. collected the data. L.F.B., A.T. D.L.P., and M.C.E. analyzed the data. All authors critically reviewed, edited, and approved the final manuscript.