Multimodal nonlinear correlates of behavioural symptoms in frontotemporal dementia

Background Studies exploring the brain correlates of behavioural symptoms in the frontotemporal dementia spectrum (FTD) have mainly searched for linear correlations with single modality neuroimaging data, either structural magnetic resonance imaging (MRI) or fluoro-deoxy-D-glucose positron emission tomography (FDG-PET). We aimed at studying the two imaging modalities in combination to identify nonlinear co-occurring patterns of atrophy and hypometabolism related to behavioural symptoms. Methods We analysed data from 93 FTD patients who underwent T1-weighted MRI, FDG-PET imaging, and neuropsychological assessment including the Neuropsychiatric Inventory, Frontal Systems Behaviour Scale, and Neurobehavioral Rating Scale. We used a data-driven approach to identify the principal components underlying behavioural variability, then related the identified components to brain variability using a newly developed method fusing maps of grey matter volume and FDG metabolism. Results A component representing apathy, executive dysfunction, and emotional withdrawal was associated with atrophy in bilateral anterior insula and putamen, and with hypometabolism in the right prefrontal cortex. Another component representing the disinhibition versus depression/mutism continuum was associated with atrophy in the right striatum and ventromedial prefrontal cortex for disinhibition, and hypometabolism in the left fronto-opercular region and sensorimotor cortices for depression/mutism. A component representing psychosis was associated with hypometabolism in the prefrontal cortex and hypermetabolism in auditory and visual cortices. Discussion Behavioural symptoms in FTD are associated with atrophy and altered metabolism of specific brain regions, especially located in the frontal lobes, in a hierarchical way: apathy and disinhibition are mostly associated with grey matter atrophy, whereas psychotic symptoms are mostly associated with hyper-/hypo-metabolism.


Introduction
The constellation of behavioural symptoms occurring in the syndromes of the FTD spectrum is extremely variable, and individual patients may present some behavioural symptoms without ever presenting others (Barker et al., 2022;Rascovsky et al., 2011).Studies have demonstrated clinical-anatomical correspondences by relating changes in a single neuroimaging modality with questionnaires measuring the severity of single behavioural symptoms in a linear way (Levenson et al., 2014;Rosen et al., 2005;Whitwell et al., 2007;Zamboni, 2016).Clinical experience, however, suggests that behavioural symptoms tend to cooccur in variable combinations across patients.In addition, the association between the severity of the symptoms and the brain may not be linear.The majority of previous studies on behavioural symptoms in the FTD spectrum have focussed on structural magnetic resonance imaging (MRI) with the assumption that behavioural variability can be fully explained by grey matter atrophy, a marker of neurodegeneration (Lansdall et al., 2017;Rosen et al., 2005;Zamboni et al., 2008).In parallel, other studies have independently tried to link behavioural symptoms to regional hypometabolism measured with ( 18 F)-2-uoro-deoxy-D-glucose positron emission tomography (FDG-PET), a marker of early synaptic dysfunction (Cerami et al., 2015;Ruby et al., 2007).The lines of research on the correlates of behavioural symptoms in FTD based on the two different imaging modalities (MRI and FDG-PET) have progressed independently.
In the present study we explored whether the variability of behavioural symptoms in FTD is better captured by changes in brain atrophy measured with structural MRI or brain hypometabolism measured with FDG-PET.We rst identi ed modes of variation (components) explaining the variability of behavioural symptoms in patients with FTD using several different behavioural questionnaires.We then studied how the identi ed components relate to changes in both brain structure (MRI) and metabolism (FDG-PET), using a novel multimodal decomposition technique (Arya, 2019;Arya et al., 2019) that allowed us to also identify nonlinear relationships.(Cummings et al., 1994), Frontal Systems Behavior Scale (FrSBe) (Grace & Malloy, 2001), and Neurobehavioral Rating Scale (NBRS) (Levin et al., 1987).All consecutive subjects with both MRI and FDG-PET data and behavioural assessment were included.

Subjects
Behavioural data were analysed using IBM SPSS Statistics 23 for Mac.We fed all the neurobehavioral data (including the 3 scores from the FrSBe, the 27 items of the NBRS, the 10 items of the NPI) by means of a Principal Component Analysis (PCA), with direct oblimin rotation.

Imaging analysis
Imaging acquisition and preprocessing steps are reported in Supplemental Methods.Unimodal Voxelbased Morphometry (VBM) linear analyses were used to identify regions of signi cant linear correlation between grey matter density and each of the three components obtained from the PCA, by applying permutation-based non-parametric inference (Nichols & Holmes, 2002).The model included age, sex, and the Mattis-DRS total score (as a measure of global dementia severity) as covariates of no interest.Unimodal PET linear analyses were used to identify regions of signi cant linear correlation between regional metabolism and the three components, with the same covariates of no interest.In addition, PET unimodal analyses were also controlled for local atrophy by including in the model the grey matter images from the VBM as an additional voxelwise covariate of no interest.
Multimodal VBM and PET non-linear analyses were performed with an new technique, that can be considered as an exploratory tool, where the aim is to identify voxel "trajectories", i.e. the rates of change of the voxel values with respect to a considered variable, in order to reveal clusters of voxels that change in the same way across the modalities of interest (Arya, 2019;Arya et al., 2019) (Supplementary Material).This method is explicitly informed by a variable of interest (in our case the components obtained from the PCA, one at a time) and is then applied to multimodal data (MRI and FDG-PET images).

Behavioural results
The PCA with direct oblimin rotation was computed on the scores of the neurobehavioral tests (NPI, FrSBe, NBRS) and gave good indicators of factorability (Kaiser-Meyer-Olkin Measure of Sampling Adequacy: 0.56, Bartlett's Test of Sphericity: Approx.Chi Square 1907.707,df 780, Sig.0.000), and the residuals indicate that the solution was a good one.Three components with an eigenvalue of greater than 3.0 were found (Supplementary Fig. 1 and Supplementary Table 1).The following three components were identi ed: Component 1 loaded 'decreased motivation', 'emotional withdrawal', and 'blunted affect' of the NBRS, and 'apathy' of the FrSBe and NPI.It also loaded 'inattention', 'conceptual disorganisation' and 'poor planning' of the NBRS, and 'executive dysfunction' of the FrSBe.We labelled it as Apathy.
Component 2 loaded, on its negative end, 'disinhibition' from the FrSBe and NPI, and on its positive end, 'expressive de cit', 'speech articulation defect' and 'depressive mood' of the NBRS.We labelled it as Disinhibition versus depression/mutism.
The last component loaded 'hallucinations' and 'delusions' of the NPI, as well as 'hallucinatory behaviour', 'unusual thought content', and 'suspiciousness' of the NBRS.We labelled it as Psychosis.
Sensitivity analyses performed on the bvFTD sample only (i.e., by removing the 20 patients with PPA) yielded overlapping results.

Unimodal imaging results
Preliminary VBM linear analyses showed that Component 1 (Apathy) was associated with atrophy in the medial prefrontal cortex (cingulate and orbitofrontal cortices) especially on the right side, and right posterior middle frontal gyrus (Fig. 1A).Component 2 (Disinhibition versus depression/mutism) was associated with atrophy in the posterior insula bilaterally and middle-inferior temporal gyrus, especially in the right hemisphere on the end of increasing disinhibition and euphoria (Fig. 1B), whereas the opposite end did not lead to signi cant results.Component 3 (Psychosis) was not associated with signi cant unilinear atrophy.
PET correlational analyses showed that increasing values on Component 1 were associated with hypometabolism in two discrete regions in the right anterior cingulate cortex and right lateral orbitofrontal cortex (Fig. 1C).Decreasing values on Component 2, indicating greater disinhibition, were associated with hypometabolism in the right anterior temporal pole (Fig. 1D).Increasing values on Component 3 were associated with hypometabolism in the right medial posterior frontal cortex (Fig. 1E).

Multimodal imaging results
Component 1, Apathy, could be explained by 3 multimodal trajectories, representing clusters of voxels changing in the same way across the two modalities, VBM and PET, in relation to the behavioural nuances captured by the component (see the dendrogram obtained with hierarchical clustering in Supplementary Fig. 2A).Subject weights were plotted against Component 1 to visualize the extracted trajectories (Fig. 2, Supplementary Table 2).The multimodal results were visualised using spatial maps both for VBM and PET of the various trajectories.In these maps the highlighted regions represent clusters of voxels that " t best" with the variability along Component 1, Apathy.Two trajectories (T2 and T3) captured clusters of voxels in which volume and metabolism decrease with increasing values of Apathy: these voxels were mainly located in the right prefrontal cortex for both modalities.More precisely, T2 showed that the higher the Apathy component value, the lower the grey matter volume in the anterior insula bilaterally and in the right anterior cingulate cortex (from VBM), paired to decreased metabolism in bilateral frontal poles and right thalamus (from PET).T3 showed that the higher the Apathy component value, the lower the volume in the right cingulate and right putamen, paired to decreased metabolism in the whole right prefrontal cortex (Fig. 3).The rst trajectory (T1) was associated mainly with PET and captured voxels in which metabolism increased with increasing values of Apathy: these voxels were located in the left temporal and temporo-parietal regions.
Component 2, Disinhibition versus depression/mutism, could be explained by 3 trajectories representing clusters of voxels changing in tandem across the two modalities (Supplementary Fig. 2B).Subject weights were plotted against Component 2 to visualize the extracted trajectories (Fig. 3).The rst trajectory (T1) captured brain voxels in which metabolism decreased with increasing values of depressive mood, stillness, and mutism: it highlighted larger clusters of voxels in the PET rather than in the VBM, mainly located in the left fronto-opercular region and in the sensory-motor cortex bilaterally.Another trajectory (T3) captured clusters of voxels in which volume and metabolism decreased with increasing disinhibition.These clusters were larger in the VBM than in the PET and were located for VBM only in the insular cortices bilaterally and right anterior cingulate, and for PET only in the right temporal pole.
Component 3, Psychosis, could also be explained by 3 trajectories (Supplementary Fig. 2C).These trajectories represent clusters of voxels changing in the same way across the two imaging modalities in relation to the variability of Component 3. Subject weights were plotted against Component 3 to visualize the extracted trajectories (Fig. 4).A rst trajectory (T1) almost had a U-shape and depicted clusters of voxels in which metabolism rapidly increased for high scores on hallucinations and delusions, but also for low values on the opposite end of the same Component.Clusters emerged mainly from PET and were in the occipital and auditory cortices.Another trajectory (T2) captured clusters in which volume and metabolism decreased for high values of hallucination and delusions.It captured clusters of increasing atrophy in the basal ganglia bilaterally and left lingual gyrus from the VBM and clusters of hypometabolism in the prefrontal cortex bilaterally in the PET.T3 captured clusters in the occipitotemporal gyrus, in the insula bilaterally and in the right cingulate cortex in the VBM, and in the basal ganglia (especially left thalamus and caudate) in the PET.

Discussion
The aim of the present work was to improve our understanding of the structural and functional basis of the constellation of behavioural symptoms of FTD, by studying them with data-driven approaches, and relating them to two different imaging modalities (structural MRI and FDG-PET) in combination and nonlinearly.
We found that the variability of behavioural and psychological symptoms in an FTD cohort was best captured by three components, which we labelled as (i) Apathy, (ii) Disinhibition versus depression/mutism, and (iii) Psychosis.
The fact that ratings for apathy and disinhibition from behavioural questionnaires (such as the FrSBe and the NPI) loaded on two different components contributes to the ongoing discussion on whether apathy and impulsivity represent opposite ends of a one-dimensional continuum or rather they tend to cooccur.Earlier clinical-anatomical studies aimed at capturing the variability of the constellation of behavioural and psychological symptoms in the FTD spectrum had identi ed two presentations with distinct neural correlates: one predominantly characterized by disinhibition and impulsivity, and the other predominantly characterized by apathy and inertia (Le Ber et al., 2006;Snowden et al., 2001;Zamboni et al., 2008).Whereas these early studies assumed that such "disinhibited" and "apathetic" pro les were the opposite ends of a behavioural continuum (i.e., a patient could have one or the other presentation), it has been now demonstrated that disinhibition and apathy usually co-occur in the same patient with FTD (Kok et al., 2021;Lansdall et al., 2017;Peters et al., 2006), and often coexist in cognitively healthy young individuals (Petitet et al., 2021).Our ndings support the hypothesis that apathy and impulsivity may coexist to variable degrees but remain independent constructs with separate neuroanatomical correlates.They suggest that there may be patients who are both apathetic and disinhibited, as well as patients who are apathetic and depressed.
The second Component, labelled as Disinhibition versus depression/mutism, was the only component that loaded speci c behavioural disturbances on both its negative and positive ends.More precisely, it contrasted 'disinhibition' with 'depressive mood', highlighting aspects of disinhibition related to mania and abnormally elevated, expansive mood.But it also contrasted 'disinhibition' with 'expressive de cit' and 'speech articulation defect', highlighting aspects of disinhibition related to the prepotent verbal response and excessive garrulous chatter that FTD patients may present.Thus, we may assume that Component 2 captured several aspects of the multifaceted phenomenon associated with the broad term 'disinhibition', including those re ecting enhanced impulsivity or hyperactivation of the processes that generate the impulse, as well as those related to the loss of the knowledge of social rules or impairments in the suppression of prepotent responses and resistance to distractor interference (Magrath Guimet et al., 2021; Migliaccio et al., 2020).Importantly, this component did not change when we excluded patients that had started with language disturbances from the PCA, suggesting that it was not simply driven by their aphasia but rather captured behavioural variability across the different presenting phenotypes.
The third Component, labelled as Psychosis, remained stable and distinct even when increasing the number of extracted components in the PCA.This is consistent with ndings in several previous studies (Aalten et al., 2008) and with the hypothesis that psychotic symptoms identify a speci c phenotype in dementia (Ballard et al., 2020;Murray et al., 2014).
The second aim of the present study was to examine how the identi ed components of behavioural variability relate to changes in brain structure (MRI) and metabolism (FDG-PET).We preliminarly studied each modality separately with regression models exploring linear correlations: the unimodal VBM results were consistent with previous studies that had performed VBM correlational analyses of single behavioural questionnaires (Rosen et al., 2005;Sheelakumari et al., 2020;Zamboni et al., 2008).Interestingly, there were no regions of signi cant correlation between grey matter volume and Psychosis (Component 3).
By using a newly developed fusion analysis we then studied, for the rst time, how the identi ed components of behavioural variability relate to the two imaging modalities in conjunction, i.e., whether they are mainly associated with changes in structure (MRI), metabolism (FDG-PET), or both.In fact, it would be reasonable to think that some symptoms may mainly derive from alterations in the metabolism and not be associated with detectable atrophy, which takes longer to occur.Some other symptoms, instead, may be a direct consequence of the neurodegenerative process, which causes cell death and synapsis loss, seen as focal grey matter atrophy.In addition, our multimodal decomposition technique allowed us to also uncover nonlinear relationships, as depicted by the trajectory plots often showing relationships that were at for some portion and then changed or were even U-shaped, whereas previous studies had mainly searched for linear relationships.
The fusion analysis of MRI and PET data showed that voxels in which grey matter volume and metabolism decreased with increasing values of Apathy were mainly located in the anterior insula and anterior cingulate cortex, regions known to be part of the salience network (SN), and with hypometabolism in the right prefrontal cortex.The SN is speci cally thought to be involved in detecting and processing salient information (Seeley et al., 2007).Another trajectory showed that increasing values on the Apathy component were associated with decreasing volume in the right cingulate and bilateral putamen, paired with largely decreased metabolism in the right prefrontal cortex.These two trajectories of multimodal covariation seem to capture what has been indicated as the motivational and cognitive components of apathy, respectively (Ducharme et al., 2018).Interestingly, in both trajectories a decreasing volume for bilateral subcortical structures was associated with hypometabolism of the right prefrontal cortex.
The fusion analysis on the Disinhibition versus depression/mutism component identi ed a trajectory with large clusters of hypometabolism, more than for atrophy, associated with increasing depression, mutism, and stillness in the left prefrontal cortex and in the sensory-motor cortex bilaterally.These regions have been associated, respectively, with language production, motor control, and depression (Davis et al., 2010;Ray et al., 2021).Another trajectory of the same Component captured instead clusters of atrophy, which was predominant for this trajectory, associated with increasing disinhibition, which was localised in the anterior insula bilaterally and right anterior cingulate.In addition, hypometabolism also involved the temporal poles.According to one functional interpretation of frontal-subcortical circuits (Tekin & Cummings, 2002), temporo-limbic structures are part of the orbitofrontal circuit, whose dysfunction is characterized by disinhibition syndromes including irritability, impulsivity, and undue familiarity.This has been interpreted both as primarily frontal, i.e., due to the loss of inhibition by the frontal monitoring system on the limbic system responsible for instinctual behaviors (Cummings, 1995), but also as primarily subcortical, i.e., due to the impaired risk perception mechanisms (Ghika, 2000).
Lastly, the fusion analysis on the Psychosis component mainly showed results from PET rather than VBM, suggesting that the symptoms described by Component 3 have greater functional rather than structural substrates.Among the trajectories associated with increasing scores of 'psychosis', one showed small clusters of atrophy in the basal ganglia (striatum) from the VBM and larger clusters of hypometabolisms in the prefrontal cortex bilaterally from the PET.This component may capture the mesolimbic dopaminergic pathway, the dysfunction of which has been associated with positive symptoms in schizophrenia (

Conclusions
Our results show that there is a hierarchy in the way two imaging modalities (MRI and FDG-PET) relate to behavioural disturbances in FTD.Some components of behavioural variability such as Apathy and Disinhibition versus depression/mutism appear to be predominantly associated with changes in brain atrophy.Others, such as Psychosis, with changes in brain metabolism.If replicated these results have clinical implications in the choice of the neuroimaging investigation that should be prioritised in patients with different behavioural disturbances.

Declarations
Figures  Figure 4 We enrolled patients seen at the Cognitive Neuroscience Section of the National Institute of Neurological Disorders and Stroke, NIH, between 2002 and 2009.In order to be included they needed to have a diagnosis of FTD according to the criteria available at the time (Neary et al., 1998) but also according to subsequent criteria for FTD spectrum disorders, which include PPA and behavioural variant of FTD (Gorno-Tempini et al., 2011; Rascovsky et al., 2011).During a single 1-week visit at the NIH, patients underwent brain MRI and FDG-PET scanning, and extensive neuropsychological evaluation including Neuropsychiatric Inventory (NPI)

Table 1
FrSBe, Frontal Systems Behavior Scale; NBRS, Neurobehavioral Rating Scale; NPI, Neuropsychiatric Inventory Zmigrod et al., 2016)18;McCutcheon et al., 2019).Another multimodal trajectory showed that increasing scores of psychosis are also associated with increasing metabolism in visual and auditory cortices, in line with the hypothesis that psychotic productive symptoms derive from aberrant hyperfunctioning primary sensory areas (Alderson-Day et al., 2016;Zmigrod et al., 2016).