Laughter as a Paradigm of Socio-emotional Signal Processing in Dementia

Background: Laughter is a fundamental communicative signal in our relations with other people and is used to convey a diverse repertoire of social and emotional information. It is therefore potentially a useful probe of impaired socio-emotional signal processing in neurodegenerative diseases. Here we investigated the cognitive and affective processing of laughter systematically in patients representing all major syndromes of frontotemporal dementia, a disease spectrum characterised by severe socio-emotional dysfunction, as well as typical amnestic Alzheimer’s disease in relation to healthy age-matched individuals. Methods: We assessed cognitive labelling and valence rating of samples of spontaneous (mirthful and hostile) and volitional (posed) laughter versus two auditory control conditions (a synthetic laughter-like stimulus and spoken numbers) in 47 patients with frontotemporal dementia (22 with behavioural variant frontotemporal dementia, 12 with semantic variant primary progressive aphasia and 13 with nonuent-agrammatic primary progressive aphasia), 15 patients with typical amnestic Alzheimer’s disease and 20 healthy age-matched individuals. Neuroanatomical associations of laughter processing were assessed using voxel-based morphometry of patients’ brain MR images. Results: While all dementia syndromes were associated with impaired identication of laughter subtypes relative to healthy controls, this was signicantly more severe overall in frontotemporal dementia than in Alzheimer’s disease and particularly in the behavioural and semantic variants, which also showed abnormal affective evaluation of laughter. Certain striking syndromic signatures emerged, including enhanced liking for hostile laughter in behavioural variant frontotemporal dementia, impaired processing of synthetic laughter in the nonuent-agrammatic variant (consistent with a generic complex auditory perceptual decit) and ‘numerophilia’ in the semantic variant. Across the patient cohort, overall laughter identication accuracy correlated with regional grey matter in a core network encompassing inferior frontal and cingulo-insular cortices; and more specic correlates of laughter identication accuracy were delineated in cortical regions mediating affective disambiguation (identication of hostile and posed laughter in orbitofrontal cortex) and authenticity (social intent) decoding (identication of mirthful Signicant regional grey matter associations of overall laughter identication accuracy and unbiased hit rates (see text) for each laughter condition over the combined patient cohort, based on voxel-based morphometry. All clusters with extent larger than 10 voxels are shown. Coordinates of local maxima are in standard Montreal Neurological Institute space. P values were all signicant (< 0.05) after family-wise error (FWE) correction for multiple voxel-wise comparisons over the whole brain. Across the combined patient cohort, overall laughter identication accuracy was signicantly positively associated (P FWE <0.05 over the whole brain) with grey matter volume in the left pars orbitalis of inferior frontal gyrus, anteromedial prefrontal cortex, medio-dorsal thalamus and posterior insula and in bilateral anterior insula. Examining the neuroanatomical correlates of accurate identication of particular laughter subtypes in the combined patient cohort, unbiased hit rates for mirthful, hostile and posed laughter were all signicantly positively associated (P FWE <0.05 over the whole brain) with grey matter volume in the left pars orbitalis of inferior frontal gyrus and anterior insula. In addition, hit rates for mirthful laughter were signicantly positively associated with grey matter in left dorsal anterior cingulate cortex and posterior middle temporal gyrus; hit rates for hostile laughter were signicantly positively associated with grey matter in right posterior insula and left orbitofrontal cortex; while hit rates for posed laughter were signicantly positively associated with grey matter in left anteromedial prefrontal, anterior cingulate and orbitofrontal cortices. neural networks (46): a ‘sensory’ network mediating analysis of auditory features and patterns, here represented by the thalamic, posterior insular and inferior frontal correlates of general laughter identication; a ‘limbic’ (mesial temporal - ventromedial prefrontal) network mediating affective disambiguation of stimuli, here represented by the orbitofrontal correlates of hostile and posed laughter identication; an ‘intermediate’ (cingulo-insular) network integrating salient environmental and bodily states, here represented across laughter subtypes; and a ‘higher associative’ (temporo-parietal dorsomedial prefrontal) network engaged in decoding mental states, here represented by the identication of mirthful and posed laughter.

(svPPA) and non-uent/agrammatic primary progressive aphasia (nfvPPA). Whilst de cits in emotion recognition, empathy and social understanding are de ning features of bvFTD (4), they are also well documented in svPPA and nfvPPA (5)(6)(7)(8). Changes in social and emotional cognition are also increasingly recognised in Alzheimer's disease (AD) (9). However, despite their signi cant impact, these changes are poorly understood and challenging to assess objectively. This is attributable both to the inherently complex and multifaceted nature of emotional and social behaviour and a lack of tractable models and instruments with which to measure these phenomena.
To date, studies of emotional and social signal processing in dementia have focussed largely on recognition and categorisation of facial expressions, characterising impairments across the FTD spectrum that particularly impact recognition of negative expressions and interpreting the expressions of other people (10)(11)(12)(13)(14)(15). However, socio-emotional de cits in FTD extend to other sensory channels, in particular auditory signals (16)(17)(18)(19). Vocal paralinguistic affective signalling ampli es, contextualises or may even override verbal messages (as exempli ed in affective prosody and sarcasm (20)(21)(22)(23)). Processing of such signals is impaired in bvFTD and PPA syndromes (20,(24)(25)(26)(27). Impaired processing of emotional prosody has also been described in typical AD; here (as in nfvPPA) perhaps re ecting a more elementary de cit of auditory pattern analysis (23,25,28,29). Nonverbal emotional vocalisations represent another essential component of social communication, enabling emotional signals to be broadcast rapidly even under conditions that would hinder visual signalling.
Among the cardinal nonverbal vocalisations that we use as humans, arguably the richest, most universal and most socially resonant is laughter. Laughter is phylogenetically ancient (30); in primates it serves to signal positive affect and a liation, primarily during play and social grooming (31). It develops in human infants before speech (32) and is trans-cultural and socially ubiquitous (33,34). However, we seldom laugh alone and laughter is extensively modulated by social context (35)(36)(37)(38): besides conveying mirth or conviviality, laughter may be used to taunt an opponent, express delight in another's misfortune (schadenfreude) or cover embarrassment. Even more frequently, laughter is voluntarily generated, or 'posed': unlike spontaneous laughter that is stimulus driven and emotionally tuned, voluntary laughter is not necessarily associated with any strong emotional experience but may rather facilitate a liation or polite agreement (35). These laughter types are distinguished by their acoustic signatures (39,40) and under experimental conditions, healthy subjects can reliably classify laughter even when non-auditory cues are removed (36). However, because laughter can express wide variation in affect and authenticity within the frame of a single acoustic carrier, it is an intrinsically ambiguous stimulus: a characteristic that is mirrored in the multi-dimensionality of natural social scenarios more generally. Not surprisingly, the neural apparatus responsible for decoding and evaluating such a complex signal is elaborate. fMRI studies of laughter processing in the healthy brain have implicated distributed cerebral networks, encompassing superior temporal and inferior frontal cortices engaged in decoding auditory sequences, mesial temporal and insular regions mediating sensory-affective integration and emotional reactivity, and anteromedial prefrontal and orbitofrontal circuitry that appraises and evaluates affective sensory signals (38,(41)(42)(43)(44)(45). Together these neural networks reconstitute much of the recently de ned human social brain connectome (46). As a neuropsychological tool, laughter is well equipped to expose subtle degrees of socio-emotional dysfunction in people with FTD, who typically struggle to resolve ambiguity and context in social situations, even while still performing relatively well on standard neuropsychological tests of emotion recognition (Snowden, Austin et al. 2008, Downey, Mahoney et al. 2015, Clark, Nicholas et al. 2017). Moreover, the neural substrates of laughter processing in the healthy brain are affected early and prominently in the course of major dementias, particularly FTD (47,48), suggesting that laughter may constitute a neuroanatomically pertinent probe of socio-emotional processing in dementia. Indeed, neurodegenerative diseases have been linked to abnormalities of laughter behaviour in daily life. In the context of punctuating conversation, patients with bvFTD (and also AD) laugh less whereas patients with nfvPPA may laugh more than their healthy caregivers (49); while patients with bvFTD and svPPA often laugh inappropriately, for example in response to others' misfortune (50). However, laughter processing has not been studied systematically in neurodegenerative disease.
Here we addressed this issue in patients representing canonical FTD syndromes and AD, referenced to a group of healthy older individuals. We created a novel battery of stimuli, representing genuine mirthful and hostile laughter along with posed (volitional) laughter together with a synthetic, perceptually complex laughter-like stimulus. These stimulus conditions were designed to allow us to distinguish a hierarchy of potential cognitive de cits, ranging from primary perceptual (laughter-like signals versus natural laughter) through semantic emotional categorisations (mirthful versus hostile laughter) to social cognitive categorisation (posed versus spontaneous laughter). We assessed explicit identi cation (perceptual cognitive categorisation) of the laughter subtypes represented, alongside affective evaluation (valence rating of laughter subtypes) and in relation to daily life measures of socio-emotional reactivity.
Neuroanatomical associations of laughter identi cation in the patient cohort were assessed using voxelbased morphometry. Based on available evidence (20,23,(25)(26)(27)(49)(50)(51), we hypothesised that impairments of laughter processing would be widespread across FTD and AD but would show dissociated patterns of de cits in different syndromes. We predicted more severe de cits in FTD syndromes than in AD, with a more elementary de cit of perceptual analysis in nfvPPA and more severe social and emotional processing de cits in svPPA and bvFTD. We further hypothesised that laughter identi cation in these diseases would have neuroanatomical correlates in distributed cerebral networks previously implicated in laughter processing in the healthy socio-emotional brain (41,42,(44)(45)(46)52), with partially separable correlates for different laughter subtypes and hub zones for signal salience, affective and mental state decoding in insula, orbitofrontal and medial prefrontal cortices, respectively.

Participants
Forty-seven patients with a syndrome of FTD (22 with bvFTD, 12 with svPPA, 13 with nfvPPA) and 15 patients with typical amnestic AD were recruited from a specialist cognitive disorders clinic. All patients ful lled consensus criteria for the relevant syndromic diagnosis (4, 53, 54), of mild to moderate severity. Twenty healthy older individuals with no history of neurological or psychiatric illness also participated.
No participant had a history of signi cant hearing loss; peripheral hearing function was assessed using pure tone audiometry following a previously described procedure (55) (details in Supplementary Material online) and composite hearing scores were included as covariates in behavioural, physiological and anatomical analyses. General neuropsychological assessment and brain MRI corroborated the syndromic diagnosis in all patients; no participant had radiological evidence of signi cant cerebrovascular damage.
To assess the relations between laughter processing and impairments of daily life emotional and social behaviour, the Modi ed Interpersonal Reactivity Index (mIRI) (56) and Revised Self-Monitoring Scale (RSMS) (57) were completed by primary caregivers of patients with FTD syndromes. Whilst there is no standardised measure of social and emotional behaviour in dementia, the mIRI is a validated, widely used measure of cognitive and emotional empathy that has been administered previously to people with dementia (58) whilst the RSMS is a measure of sensitivity and responsiveness to others' emotional expressions and behaviour that has been used in previous studies of both healthy and clinical populations (59,60).

Creation of experimental stimuli
We created sound stimuli to represent each of the three major natural laughter categories of interest: mirthful (spontaneously reactive, involuntary laughter induced by an intrinsically amusing situation), hostile (spontaneous laughter in response to others' misfortune or discom ture, with the effect of taunting or deriding them) and posed (laughter produced volitionally with a more intentional, communicative purpose, generally in response to social cues and disproportionate to any felt amusement). Short samples of mirthful and posed laughter were derived from a previously published battery (45); additional examples of mirthful, posed and hostile laughter were derived from video clips publicly available on youtube.com. Highly identi able examples of each laughter condition were selected based on an initial pilot experiment in healthy young adults.
In addition to these natural laughter conditions, we created two control stimulus conditions to allow us to interpret the affective response elicited by laughter stimuli. The rst control condition was intended to calibrate for the effect of hearing a human voice, by establishing a baseline vocal condition that did not express any clear emotion: trials in this condition each comprised one of two male voices reading aloud a three-digit number with neutral intonation (this verbal carrier was chosen because nonverbal vocalisations that are not intended to convey emotion -e.g. yawning -often tend to have affective connotations). The second control condition was intended to calibrate for the effect of hearing an affectively arousing, laughter-like signal: trials in this condition comprised samples of spectrally inverted laughter, synthesised digitally from raw recordings representing each of the laughter conditions using a previously described algorithm (61) (this stimulus retains the spectrotemporal complexity of laughter but is normally perceived as 'alien' and aversive). Stimuli in each of the control conditions were edited digitally to have the same general acoustic parameters as the laughter stimuli.

Experimental paradigm
Participants were rst familiarised with the experimental set-up and practice trials were delivered (using stimuli not subsequently used in the experiment proper) to ensure they understood the procedure and were able to comply. All auditory stimuli were delivered in randomised order at an individually comfortable, xed listening level (approximately 70dB) via AudioTechnica® ATH-M50X headphones from a notebook computer running Eyelink Experiment Builder software (SR Research, Ottawa, Canada).
In a rst experimental session, all stimulus conditions were presented and the task on completion of each sound was to rate its valence on a modi ed 5-point Likert scale (1, very unpleasant; 5, very pleasant). In a second, separate experimental session, the laughter conditions were presented and the task on each trial was to decide if the sound represented mirthful ('happy'), hostile ('nasty'), posed ('faked') or spectrally inverted ('computer') laughter (the spoken number condition was not presented during this session). The separation of sessions was intended both to avoid the cognitive demands of dual tasks administered in a single session and to minimise any mutual priming between affective rating and identi cation of laughter conditions. No feedback about performance was given and no time limits were imposed.
Statistical analysis of general phenotypic and experimental behavioural data Data were analysed using Stata14® (StataCorp, College Station, TX, USA). Between-group comparisons of continuous demographic and neuropsychological data were performed using analysis of variance (ANOVA) whilst analogous comparisons for categorical data (e.g. gender, handedness) were carried out using chi-squared tests.
Sound classi cation was a multiple-choice task and therefore unbiased hit rates (Hu) were computed for each laughter condition, to yield a measure of perceptual sensitivity taking into account both the hit rate and false alarm rate. The Hu measure was devised for use in category judgement experiments (62), calculated as: Hu =(Ai/Bi) x (Ai/Ci), where Ai = frequency of hits, Bi = number of trials where i is target and Ci = frequency of i responses (hits and false alarms). This was converted to a percentage with a score of zero denoting chance performance.
These unbiased hit rates were compared amongst groups and conditions using a linear regression model, with diagnostic group, condition and their interactions, along with age, WASI Matrices score (an index of overall disease severity) and composite audiometry score as predictor variables. The non-independence of the repeated responses (across conditions) was accounted for by using robust (Huber-White) standard errors (63,64) that allowed for correlated responses by participant for construction of con dence intervals and hypothesis tests.
Bonferroni-corrected post hoc t-tests were carried out where a joint test of the group or condition effects or their interaction was statistically signi cant. So, for example, when making pairwise comparisons between ve groups for a particular condition, the p-values were multiplied by ten. Similar modi cations were made to 95% con dence intervals. An analogous approach to that for unbiased hit rates was used to compare valence judgements by group and condition.
For laughter identi cation, numbers of each type of error (out of 16) were analysed to look for any evidence of systematic bias or difference between groups. Separate logistic regression models were tted for each type of error. These models included age, WASI matrices score and composite audiometry score as well as group as predictor variables. Since the distribution of the number of errors might not be binomial, robust Huber-White standard errors were used as above. In cases where the omnibus test of comparisons amongst groups was statistically signi cant, pairwise comparisons were made with Bonferroni adjustment for multiple comparisons as above.
Associations between total laughter identi cation accuracy and the two questionnaire-based measures of social behaviour were assessed using Pearson correlation coe cients.

Brain image acquisition and analysis
Each patient had a sagittal 3-D magnetisation-prepared rapid-gradient-echo T1-weighted volumetric brain MR sequence (echo time/repetition time/inversion time 2.9/2200/900 msec, dimensions 256 256 208, voxel size 1.1 × 1.1 × 1.1 mm), acquired on a Siemens Prisma 3T MRI scanner using a 32-channel phasedarray head-coil. Pre-processing of brain images was performed using the New Segment and DARTEL toolboxes of SPM12 (www. l.ion.ucl.ac.uk.spm), following an optimised protocol (65). Normalisation, segmentation and modulation of grey and white matter images were performed using default parameter settings and grey matter images were smoothed using a 6 mm full-width-at-half-maximum Gaussian kernel. A study-speci c template mean brain image was created by warping all bias-corrected native space brain images to the nal DARTEL template and calculating the average of the warped brain images. Total intracranial volume was calculated for each patient by summing grey matter, white matter and cerebrospinal uid volumes after segmentation of tissue classes.
Following quality control of the pre-processed brain images, scans from 60 patients (13 AD, 22 bvFTD, 13 svPPA and 12 nfvPPA) were entered into the VBM analysis. A regression model was used to assess associations of regional grey matter volume (indexed as voxel intensity) with overall laughter identi cation score (percentage of all laughter trials accurately identi ed) for the combined patient cohort. In addition, grey matter associations with unbiased hit rates for each laughter condition were assessed in separate models across the combined patient cohort. Age, total intracranial volume and WASI Matrices score (a measure of nonverbal executive function and overall disease severity) were incorporated as covariates of no interest in all models. Statistical parametric maps of regional grey matter associations were assessed at threshold p < 0.05 after family-wise error (FWE) correction for multiple voxel-wise comparisons over the whole brain.

General characteristics of participant groups
Demographic, clinical and neuropsychological characteristics of the participant groups are summarised in Table 1 Mirthful laughter was more accurately identi ed by both the healthy control and AD groups than by all three FTD syndromic groups (all p bonf <0.001); there was no signi cant difference between the control and AD groups (p bonf =0.213). Hostile and posed laughter were more accurately identi ed by the healthy control group than by all patient groups (all p bonf <0.001), and more accurately identi ed by the AD group than by all three FTD groups (all p bonf <0.05). Comparing between FTD syndromic groups, the svPPA group was less accurate identifying posed laughter than both the nfvPPA and bvFTD groups (both p bonf <0.001) whilst the bvFTD group were less accurate identifying hostile laughter than both the nfvPPA and svPPA groups (both p bonf ≤0.01). For identi cation of spectrally inverted laughter, the healthy control group performed at ceiling; the nfvPPA group was less accurate than all other groups (all p bonf <0.04) and the AD group less accurate than both the control and bvFTD groups (both p bonf <0.03).
Within-group pro les comparing identi cation of different laughter conditions are detailed in Supplementary Table 2. In summary, all participant groups were more accurate identifying spectrally inverted laughter than all other laughter-subtypes (all p bonf <0.001). In addition, the AD group was more accurate identifying mirthful laughter than hostile or posed laughter (both p bonf <0.001); while the svPPA group was more accurate identifying mirthful laughter than posed laughter and the bvFTD group was less accurate identifying hostile laughter than all other laughter subtypes (all p bonf <0.001).
Pro les of laughter identi cation and misidenti cation are presented in Fig. 3 and raw data on laughter confusion errors with odds ratios are presented in Supplementary Table 3. Most saliently, the bvFTD group confused mirthful with hostile laughter more often than did all other participant groups; while the svPPA group confused mirthful and hostile laughter with posed laughter more often than did all other participant groups.
Associations between laughter identi cation accuracy and questionnaire measures of social and emotional behaviour

Valence ratings of auditory stimuli
Perceived valence of auditory stimuli by participant group and sound condition are presented in Fig. 1 and Supplementary Table 4 whilst difference in valence scores between the patient groups and the healthy control group are illustrated in Fig. 2  Compared with the healthy control group, each of the FTD syndromic groups found mirthful laughter signi cantly less pleasant and hostile laughter signi cantly more pleasant (all p bonf ≤0.001; there were no signi cant differences between the healthy control and AD groups (all p bonf >0.3). The AD group found mirthful laughter signi cantly more pleasant than did each of the FTD syndromic groups and hostile laughter less pleasant than did the svPPA and bvFTD groups (all p bonf ≤0.001). Comparing between FTD syndromic groups, the nfvPPA group found mirthful laughter signi cantly more pleasant and hostile laughter less pleasant than did the svPPA and bvFTD groups, while the bvFTD group also found hostile laughter signi cantly more pleasant than did the svPPA group (all p bonf ≤0.001). There were no statistically signi cant differences in valence ratings for posed or spectrally inverted laughter between the groups (all p bonf >0.3). However, the svPPA group found the spoken number control condition signi cantly more pleasant than did all other groups (all p bonf ≤0.001).
Within-group pro les are detailed in Supplementary Table 2. In summary, the healthy control, AD and nfvPPA groups found mirthful laughter signi cantly more pleasant than posed laughter and hostile laughter less pleasant than both mirthful and posed laughter. Conversely, the bvFTD group found hostile laughter signi cantly more pleasant than posed or mirthful laughter. All groups found spectrally inverted laughter the least pleasant sound. The svPPA group found spoken numbers signi cantly more pleasant than all other sounds apart from mirthful laughter (although this also approached statistical signi cance [p bonf =0.058]).

Neuroanatomical associations of laughter identi cation
Signi cant grey matter associations of overall laughter identi cation accuracy and unbiased hit rates for each laughter condition, across the entire patient cohort are summarised in Table 2; statistical parametric maps are presented in Fig. 4. Signi cant regional grey matter associations of overall laughter identi cation accuracy and unbiased hit rates (see text) for each laughter condition over the combined patient cohort, based on voxel-based morphometry. All clusters with extent larger than 10 voxels are shown. Coordinates of local maxima are in standard Montreal Neurological Institute space. P values were all signi cant (< 0.05) after family-wise error (FWE) correction for multiple voxel-wise comparisons over the whole brain.
Across the combined patient cohort, overall laughter identi cation accuracy was signi cantly positively associated (P FWE <0.05 over the whole brain) with grey matter volume in the left pars orbitalis of inferior

Discussion
Relative to healthy older individuals, patients with major syndromes of FTD and AD exhibit richly differentiated pro les of impaired cognitive and affective processing of laughter. These pro les are summarised graphically in Fig. 5. While all dementia syndromes demonstrated impaired identi cation of laughter subtypes, this was more severe overall in FTD syndromes (particularly svPPA and bvFTD) than in AD. A qualitatively similar differentiation was found for the affective evaluation of laughter: this was normal in AD but severely affected in bvFTD and svPPA. Dementia syndromes were further strati ed based on the processing of particular laughter subtypes. Impaired processing of mirthful and hostile laughter was a hallmark of FTD syndromes compared with both healthy controls and patients with AD.
The bvFTD group in particular frequently confused mirthful and hostile laughter and demonstrated an abnormal liking for unpleasant (hostile) laughter. Impaired processing of synthetic (spectrally inverted) laughter-like signals was a hallmark of nfvPPA relative to other participant groups; while enhanced liking for a non-affective vocal signal (spoken numbers) over laughter was a striking feature of svPPA. Impaired processing of laughter in the patient cohort was underpinned by regional grey matter atrophy in distributed cerebral networks encompassing inferior and orbitofrontal, cingulate, insular, posterior temporal and anteromedial prefrontal cortices.
The panoply of 'laughter phenotypes' in different dementia syndromes is consistent with the diverse behavioural ends that laughter serves in everyday communication and with other signal processing de cits previously described in these canonical dementias. The severely impaired comprehension of laughter as an emotional and social signal in the bvFTD and svPPA groups here accords with the welldocumented di culty these patients have with understanding and responding appropriately to many kinds of social and emotional signals, including elementary emotional expressions, sarcasm and mental state attribution (10)(11)(12)(13)(14)(15)(16)(17)(18)25). Indeed, the capacity to understand laughter as a socio-emotional signal may predict daily-life socio-emotional reactivity, as evidenced by the strong positive correlation of laughter identi cation accuracy with mIRI and RSMS scores in the svPPA and bvFTD groups here.
Whereas impaired processing of 'negatively' valenced emotions has been emphasised in previous neuropsychological studies of FTD (17,18,(66)(67)(68), this might re ect a bias inherent in standard instruments such as the Ekman faces, which comprise four negative elementary emotions but only a single prototypical positive emotion (happiness). Our ndings suggest that the cognitive differentiation of perceptually related emotional signals (rather than their valence per se) challenges dysfunctional mechanisms of emotion decoding in FTD.
It is noteworthy that the identi cation of posed laughter here was abnormal across the dementia syndromes, and not restricted to those groups with more severe di culty judging authenticity of others' emotions in daily life (namely svPPA and bvFTD). However, judgements about laughter authenticity are likely to depend sensitively on accurate perceptual encoding as well as social cognitive decoding, and the mechanism of impairment is likely to have varied between the syndromes. The severe de cit in the svPPA group here is consistent with other evidence for impaired mentalising, affective semantic and social conceptual decoding in this syndrome (69)(70)(71)(72), ampli ed in situations that call for resolution of ambiguity or con ict. By contrast, in nfvPPA, our ndings suggest that the perception of complex spectrotemporal signals is fundamentally abnormal, building on emerging evidence for a generic disorder of acoustic analysis in this syndrome (25,55,61,73,74). A fundamental impairment of vocal perceptual analysis would potentially also account for the frequent confusion of mirthful and hostile laughter by the nfvPPA group, as these laughter conditions here were acoustically rather similar (see Supplementary  Table 5). Moreover, natural laughter is usually accompanied by various other contextual cues that patients with nfvPPA may be able to exploit in their daily lives, perhaps accounting for the lack of correlation between laughter labelling and measures of everyday social and emotional behaviour in this syndromic group.
While impaired cognitive labelling of laughter subtypes was accompanied by alterations in affective evaluation across the patient cohort, these two dimensions of laughter processing did not correlate simply within particular dementia syndromes; rather, there was evidence for substantial dissociation.
Patients with AD showed normal affective evaluation of all laughter subtypes and even patients representing FTD syndromes showed normal affective evaluation of posed and inverted laughter, despite de cient cognitive labelling. Altered hedonic behaviours in response to environmental sounds and music in daily life are frequently reported in FTD syndromes (in particular bvFTD and svPPA) as well as AD (75) and abnormal affective evaluation of music has been described in bvFTD and svPPA (76), while another study found that explicit affective valuation of environmental sounds may be normal in these syndromes (51). The strikingly abnormal affective preference for hostile over mirthful laughter in the bvFTD group here is in keeping with other evidence that these patients may nd humour in frankly inappropriate or unpleasant situations (50,77), and may have contributed to the frequent confusion between hostile and mirthful laughter in this group. The svPPA group showed a unique preference for spoken numbers over laughter -this 'numerophilia' may re ect a shift in hedonic drive toward inanimate stimuli akin to the behavioural repertoire of sometimes obsessive, impersonal preoccupations and interests exhibited by patients with syndromes of focal temporal lobe atrophy, which often includes mathematical puzzles (47,75,(78)(79)(80). Taken together, this evidence paints a complex picture of dissociable linkages between different dimensions of complex auditory signal analysis in canonical dementias.
The neuroanatomical substrates for overall accuracy identifying laughter in our patient cohort centred on a common, distributed fronto-cingulo-insular network previously implicated in processing and resolving ambiguity in emotional sounds including human socio-emotional signals and more particularly, laughter (41,81) Fronto-cingulo-insular circuitry appraises the salience of sensory stimuli and prepares contextually appropriate behavioural responses (82). The anterior insula hosts an interface between sensory, affective and cognitive brain systems that process emotional sounds (83)(84)(85)(86)(87)(88). Within the inferior frontal cortex, pars orbitalis acts as a hub zone for the cognitive and affective decoding of auditory signals (89), particularly where these constitute patterns bound by 'rules' and expectancies. Besides its well-known role in linguistic grammar processing, this region is involved in processing musical syntax (90) and affective evaluation of harmonic progressions in melodies (76). Anteromedial prefrontal and anterior cingulate cortices behave as an integrated functional 'hub' in appraising the social value of heard laughter and disambiguating its social intent (44,45,91) and programming adaptive output behaviours, including own laughter and the subjective experience of mirth (92)(93)(94)(95)(96). Further correlates of overall laugher identi cation accuracy were identi ed here in closely structurally and functionally interconnected regions that are likely to be obligatorily engaged in appraising and responding to laughter: posterior insula, essential for integrating interoceptive information (97) and key acoustic cues that convey emotional content (98-101) during behavioural preparation; and mediodorsal thalamus, implicated in cognitive set shifting to meet changing behavioural contingencies (102).
In line with its core role in the analysis of salient auditory signals, anterior insular and inferior frontal circuitry was correlated here with accuracy identifying all laughter subtypes when these were examined separately. Additionally, more speci c cortical associations were delineated for the identi cation of particular laughter subtypes. Accurate identi cation of mirthful laughter was additionally linked to posterior middle temporal gyrus, a region previously implicated in the processing of sensory 'templates' for humour (77). Identi cation of hostile laughter was additionally linked to posterior insula (as anticipated for a sensory signal with powerful homeostatic resonance) and orbitofrontal cortex, integral to the resolution of con ict and ambiguity in social signals based on hedonic and behavioural cues (69,76,(103)(104)(105)(106). Identi cation of posed laughter -a paradigmatic 'socially ambiguous' vocalisation -was additionally linked both to orbitofrontal cortex and an anteromedial prefrontal cortical region previously proposed to engage in obligatory mentalising during the evaluation of laughter authenticity and intent (45). These condition-speci c associations illustrate the potency of laughter as a probe of social brain mechanisms. The human social brain connectome principally comprises four hierarchically interlocking neural networks (46): a 'sensory' network mediating analysis of auditory features and patterns, here represented by the thalamic, posterior insular and inferior frontal correlates of general laughter identi cation; a 'limbic' (mesial temporal -ventromedial prefrontal) network mediating affective disambiguation of stimuli, here represented by the orbitofrontal correlates of hostile and posed laughter identi cation; an 'intermediate' (cingulo-insular) network integrating salient environmental and bodily states, here represented across laughter subtypes; and a 'higher associative' (temporo-parietaldorsomedial prefrontal) network engaged in decoding mental states, here represented by the identi cation of mirthful and posed laughter.

Limitations
This study has several limitations that should direct future work. Larger patient cohorts with histopathological and molecular correlation and autonomic, electrophysiological and dynamic neuroimaging techniques that can capture functional changes in neural networks will ultimately be required to de ne fully the pathophysiological phenotypes delineated here (107). As an intrinsically ambiguous stimulus (108), laughter is likely to tax neural perceptual and socio-emotional processing mechanisms and therefore might constitute a 'stress test' or biomarker for early detection and tracking of reduced processing delity in neurodegenerative proteinopathies: however, this will only be con rmed with longitudinal studies, ideally including presymptomatic mutation carriers. From a more practical standpoint, multi-centre, international studies addressing social cognition in dementia populations could exploit the non-linguistic and trans-cultural status of laughter (109). In everyday life, however, laughter does not occur in the disembodied form presented here but embedded in a social context: neuropsychological de cits of laughter processing will need to be assessed in relation to such contextual factors as well as behavioural symptoms, in order to fully evaluate laughter as an index of socialemotional dysfunction in dementia. Our ndings (in particular, the indexing of daily life socio-emotional competence by laughter identi cation accuracy) present a strong prima facie case for further studies incorporating additional measures of social cognition.

Conclusions
This study demonstrates how laughter can be used as an ecological and versatile probe of socioemotional signal processing in neurodegenerative disorders. Distinct signatures of impaired laughter processing have been revealed that add to our understanding of mechanisms that underlie disordered socio-emotional behaviour and help account for clinical features recognised in the major dementia syndromes. Furthermore, brain regions identi ed in this study as critical for distinct aspects of laughter processing, expand our understanding of vocal signal processing in both neurodegeneration as well as the healthy brain. The study was approved by the University College London institutional ethics committee and all participants gave informed consent in accordance with the Declaration of Helsinki.

Consent for Publication
Not applicable

Availability of data
The data that support the ndings of this study are available on request from the corresponding author. The data are not publicly available as they include information that could compromise the privacy of          parametric maps (SPMs) of regional grey matter volume positively associated with overall laughter identi cation accuracy and accuracy of identi cation of particular laughter subtypes (derived from a voxel-based morphometric analysis) are shown for the combined patient cohort (see also Table 2). SPMs are overlaid on representative sections of the normalised study-speci c T1-weighted group mean brain MR image, thresholded at p<0.05FWE corrected for multiple voxel-wise comparisons over the whole brain.
The MNI coordinate (mm) of the plane of each section is indicated and the left cerebral hemisphere is shown on the left for coronal sections and at the top for axial sections; the colour bars code T values for each SPM. Figure 4 parametric maps (SPMs) of regional grey matter volume positively associated with overall laughter identi cation accuracy and accuracy of identi cation of particular laughter subtypes (derived from a voxel-based morphometric analysis) are shown for the combined patient cohort (see also Table 2). SPMs are overlaid on representative sections of the normalised study-speci c T1-weighted group mean brain MR image, thresholded at p<0.05FWE corrected for multiple voxel-wise comparisons over the whole brain.
The MNI coordinate (mm) of the plane of each section is indicated and the left cerebral hemisphere is shown on the left for coronal sections and at the top for axial sections; the colour bars code T values for each SPM. Figure 4 parametric maps (SPMs) of regional grey matter volume positively associated with overall laughter identi cation accuracy and accuracy of identi cation of particular laughter subtypes (derived from a voxel-based morphometric analysis) are shown for the combined patient cohort (see also Table 2). SPMs are overlaid on representative sections of the normalised study-speci c T1-weighted group mean brain MR image, thresholded at p<0.05FWE corrected for multiple voxel-wise comparisons over the whole brain.
The MNI coordinate (mm) of the plane of each section is indicated and the left cerebral hemisphere is shown on the left for coronal sections and at the top for axial sections; the colour bars code T values for each SPM.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. Laughterdementiasupplementary.docx Laughterdementiasupplementary.docx Laughterdementiasupplementary.docx