Summary of findings
In the present study, we used a connectome gradient mapping approach to investigate functional network organisation in a large sample of patients affected by FTD. Healthy control subjects presented a prominent expected hierarchy and differentiation of functional networks’ connectivity patterns. The principal gradient captured a progressive hierarchy between sensorimotor (external processes) and transmodal association regions (internal processes), whilst the secondary gradient underpinned a sharp distinction between the SN (predicted processes) and visual cortex (observed processes). Together, these principles of network organisation explained 48% of the variance within the data. Such macroscale functional network organisation captures a healthy topography enabling the transition from concrete perception to abstract cognitive functions, the basis of the evolutionary transition from apes to humans and of higher-order cognitive and behavioural functions28. Though such gradients of network organisation were broadly maintained in patients with FTD, we found evidence of extreme-end networks shifting towards the centre of these axes, leading to an overall FC space constriction in all patient groups, which was more severe in bvFTD, and confirming our first hypothesis of a breakdown of this organisation in FTD. Importantly, connectome gradient mapping enabled us to identify changes at the macroscale level. We found that vital hierarchies of network organisation were constricted in all three FTD patient groups, with bvFTD patients’ pathology particularly impacting the axis dissociating predicted vs observed processes captured by the secondary gradient. Moreover, nfvPPA and svPPA patients showed specific sensorimotor and limbic network behavioural changes. Finally, the visual network showed an opposite pattern of modification in relation to other networks in all FTD patient groups. The DMN and visual networks demonstrated contrasting correlations with social cognition performances measured by the miniSEA test.
DMN and SN dedifferentiation in bvFTD
Previous work in healthy individuals has described a principal gradient spanning from the DMN to the sensorimotor network, with intermediate networks positioned in between10. This axis differentiates patterns of connectivity of regions involved in internal processes versus those responsible for external processes. In bvFTD patients, we found a noteworthy shift in DMN embedding values towards the center, contracting the principal gradient axis. Altered DMN FC in bvFTD patients have been found by several studies using classical methods. Some studies find increased FC in patients with bvFTD compared to controls (i.e. hyperconnectivity) 18,29–33 and others have found decreased (i.e. hypoconnectivity) or a mix of directional changes 19,34–37. In response to the accumulation of brain pathology, the recruitment of neural resources not usually involved may be compensatory to sustain normal cognition 38. However, it remains unclear whether hyperconnectivity represents compensatory processes or aberrant excitation due to early damage. In bvFTD patients, we found that the DMN, originally anchored at the internal processing end of the principal gradient spectrum, shifts towards regions associated with sensory/external processing. This alteration suggests a decrease in the specificity of DMN activation patterns, leading to neural dedifferentiation whereby the DMN may activate in situations in which it would not normally be activated in or vice versa. This phenomenon may explain the mixed findings of hypo- and hyper- connectivity observed in previous studies.
Additionally, we found a correlation between the DMN principal gradient embedding values and scores of social cognition test (miniSEA) in bvFTD patients. A shift of the DMN towards the center, constraining the axis, correlated with poorer miniSEA scores. This highlights the significance of maintaining an axis that distinguishes FC patterns of regions involved in external (sensorimotor network) versus internal (DMN) processes for better social cognition. Taken together, these results suggest that the DMN is losing its network differentiation, and this loss of functional specificity relates to the behavioural and cognitive deficits seen in bvFTD patients. Thus, by examining the global network organisation, our work sidesteps questions about the direction of change and highlights core features at play in bvFTD patients.
There is a growing body of evidence indicating significant decreases in FC of the SN in bvFTD patients 9,18–20,29–32,36,37,39–46. The frontoinsula and anterior cingulate cortex, hubs within the SN, are home to a concentrated number of von Economo neurons and fork cells. Such cells are believed to be particularly vulnerable to misfolded FTD pathological protein accumulation (Seeley et al. 2007, 2012). This may explain why the SN is found to be particularly disrupted in bvFTD patients. SN disruption has been related to symptoms in bvFTD patients, with previous research highlighting the interplay of FC changes involving both the DMN and SN 18,29,47,48. These findings support a model in which the SN and DMN are anticorrelated and exert an inhibitory influence on each other which seems crucial for responding to prevailing goals and conditions 49. In the current study, the SN, located at the external processing extreme of the principal gradient, exhibited a significant shift towards the DMN/internal processing extreme compared to controls. Moreover, the SN’s change along the secondary gradient, which placed this network at the opposite end from the visual network, was much the same. This reduced differentiation of FC activity patterns between the SN and DMN, observed along both the principal and secondary gradients, is novel and may contribute to other networks’ FC patterns and, consequently, symptomatology in bvFTD patients. These findings emphasise the intricate relationship between the DMN, SN, and broader network dynamics in neurodegeneration.
Visual network functional compensation?
In our study, the secondary gradient maximally differentiated the SN from the visual network in healthy subjects, underlying a functional dissociation between predicted and observed processes. In patients, both bvFTD and nfvPPA patients exhibited a significant shift of the SN towards the center. Moreover, the visual network displayed an opposite change with a shift outwardly, expanding the gradient's range in all three FTD variants. Alterations in FC within the visual network have been reported in previous FTD studies 9,50. Some work has suggested changes in the relationship between visual network and others such as the SN 20,29 and dorsal attentional network 51. The collapse of the SN in bvFTD and nfvPPA, whether pathological or functional, may prompt changes in the visual network’s behaviour.
The shift of visual network outwardly along the secondary gradient could be interpreted as a compensatory mechanism, preserving the neural differentiation between regions involved in predicted and observed processes. The dissociation of such mechanisms is vital for sustaining cognitive abilities. This hypothesis gains support from the observed correlation between visual network embedding values along the secondary gradient and scores on the miniSEA in bvFTD patients. Enhanced differentiation of the visual network from other networks correlated with better miniSEA scores, reinforcing the behavioural relevance of this possible compensatory response. The relationship between social cognition and the visual system is supported by parallels with studies in autism spectrum disorder patients, where visual gaze patterns are linked to social cognition 52. Moreover, a recent behavioural study found that patients with bvFTD showed increased fixations to the eyes of emotional faces compared to controls, which may enable the allocation of attention to emotionally relevant cues to compensate for a core deficit in contextualising emotional information 53. However, further investigation is needed to establish a direct link between visual behavioural changes and social cognition in bvFTD patients.
Another hypothesis is that visual network may show overexcitation due to lack of inhibitory control processes because of neurodegeneration. Functional rearrangements during neurodegeneration can manifest as a prevalent and progressive increase of FC in earlier stages of the disease and a subsequent decrease as neurodegeneration progresses towards a more severe stage. Similar FC phenomenon in visual networks have also been observed in other focal neurodegenerative diseases 54. In the framework of a loss of inhibitory control, previous work has suggested that the SN, for example, plays a role in modulating the activity of large-scale networks, relating to aberrant judgement and interpersonal behaviour in patients with bvFTD 47. The current findings of significant shifts in all networks along the secondary gradient in bvFTD patients imply a large-scale disinhibited system lacking neural specificity which supports the modulation role of SN. Nevertheless, if this hypothesis were to be confirmed, the reasons behind changes in visual network activity in the present svPPA patients remain unclear due to not finding SN modifications in these patients 55.
While the study could not explore this functional relevance in the PPA subgroups due to sample size constraints, previous work has reported svPPA patients to frequently exhibit frank changes in their visual attention allocation 56,57, aberrant gaze patterns 58 and disproportionate impairments with naming visual objects 59 and visual attribute reporting relative to other modalities 60,61. Despite these functions not exclusively relying on the visual network, evidence suggests visual behavioural changes across all FTD variants.
Sensorimotor and limbic network alterations in PPA variants
Though PPA variants exhibited certain network changes akin to those observed in bvFTD patients, these patients manifested distinctive alterations within sensorimotor and limbic networks. While nfvPPA patients showed a constriction of the principal gradient at the DMN/internal processing end of the axis, they also showed large constrictions at the sensorimotor network/external processing end which was less differentiated from other networks. Other work highlights the involvement of sensorimotor regions by showing reduced grey matter volume 62,63 hypometabolism 64,65 or disrupted white-matter tracts 63,66,67 within motor cortices in nfvPPA patients. Work investigating FC changes in these patients is sparse and has only suggested local FC reductions involving frontotemporal cortex and subcortical structures 68,69. Using a priori regions of interest, a previous study found that right supplementary motor area showed lower FC with speech and language regions and this correlated with their articulatory error score 70. Using whole-brain connectome gradient mapping, without limiting our analyses to specific networks, we highlight intrinsic sensorimotor network changes at a whole network level particularly in nfvPPA patients.
Moreover, the changes of sensorimotor network along the SN/predicted versus visual/observed axis enables us to conjecture further. In our controls, this network demonstrated a bimodal distribution including two distinct modes along this secondary gradient thus dissociating two groups of regions with differing patterns of connectivity. nfvPPA patients showed a striking smoothing of this distribution. Previous studies using gradient mapping may have also uncovered a bimodal distribution of sensorimotor network in controls 71 but the authors do not comment upon this limiting our interpretation of what the single mode distribution may mean in nfvPPA patients. A hypothesis could be that this bimodal distribution dissociates sets of body part representations. The 17-network parcellation described by Yeo and colleagues divided the sensorimotor strip into dorsal and ventral subnetworks and the boundary between these was roughly positioned between the hand and tongue representations 27. Thus, the high kurtosis found in nfvPPA patients could represent a blurring of this boundary, which may relate to the specific oro-facial symptoms characteristic of this patient group 17.
Similarly, svPPA patients showed pronounced changes within the limbic network along the principal gradient. Limbic network shifted away from the DMN/internal processing end and towards the sensorimotor/external processing end in these patients. Additionally, controls also showed a bimodal distribution of the limbic network along the secondary gradient, which was smoothed in all FTD groups, but particularly in svPPA patients. Very little work has assessed limbic network FC changes in svPPA, with some studies mentioning changes to limbic structures 31 or specific disconnections of anterior and inferior temporal lobe from other brain regions 34,72. A recent study found that svPPA patients as well as bvFTD patients showed lower mean FC within the limbic network compared to controls 55,73. Our results are consistent with these recent suggestions.
However, the resting-state functional networks described in this work 27 do not include subcortical structures or the basal ganglia, making our results limited to the network at the cortical level only. Furthermore, Yeo’s 17-network parcellation also dissociates the limbic network into two subnetworks, orbitofrontal cortex versus temporal pole, which may be the reason for the bimodal distribution we found in controls. Our FC changes may be blurred by the fact that atrophy in svPPA is particularly pronounced in temporal pole and less so in orbitofrontal cortex. Finally, limbic network interpretation is limited by orbitofrontal and temporal lobe regions being particularly affected by susceptibility gradients causing image distortions and signal loss 74.
Overall, most previous work in PPA has highlighted changes in DMN or SN 31,51,75 or disconnections of specific brain regions with others 34,68,69,72 rather than investigating global functional network changes. Our findings of a specific FC fingerprint involving sensorimotor and limbic network changes in nfvPPA and svPPA patients respectively using a whole-brain functional connectome approach are therefore novel and warrant further investigation to understand the specific behavioural patterns of the networks, their relationship to function and their potential future use as a biomarker in these PPA variants.
Relationship between Atrophy and FC
Atrophy and FC are undeniably closely related 4. Findings of DMN and SN changes are fully consistent with grey matter hubs of atrophy in bvFTD 76 and we found up to 20–30% reductions in grey matter volumes in regions within these networks in bvFTD patients compared to controls. Similarly, findings of changes within sensorimotor and limbic network in nfvPPA and svPPA respectively are consistent with their atrophy patterns within sensorimotor cortices or orbitofrontal cortex and temporal poles 77. Although related, our study highlights that atrophy patterns do not perfectly overlap with functional network changes 78. Thus, functional network changes, including reorganisation, do not seem to be fully explained by atrophy alone in any FTD subtype 79.
However, recent work has suggested that longitudinal spread of atrophy can be predicted using an individual’s functional connectome 80. In this previous work, shortest path length (in FC space) to the epicenter, combined with quantifying atrophy within a given region’s network neighbours, accurately estimated the spatial pattern of subsequent atrophy in patients with bvFTD and svPPA. Such work highlights the use of FC profiles in atrophy pattern progression prediction. The onset, evolution and relationship of pathological structural and functional changes in FTD is still not fully understood, even at the group level, as studies have lacked multimodal longitudinal data 81.
Concluding remarks and perspectives
Gradient mapping offers a lens through which to characterise large-scale brain network relationships, the building blocks of brain function. Though there may be specific network changes, this method enabled us to highlight that FC changes involve a widespread disruption of the evolutionarily derived global network topography in each FTD variant. We found important FC pattern dissociations between internal/external processes, as well as between predicted/observed processes to be highly diminished in these patients. This was found to be behaviourally relevant in bvFTD patients for whom the changes were the most pronounced and widespread.
It has been proposed that FTD disease is a ‘molecular nexopathy’ whereby a specific conjunction of pathogenetic protein and neural circuit characteristics lead to the specific form the disease takes on 82,83. Nexopathies target particular types of network connections and therefore transcend canonical macro-network boundaries. Connectome gradient mapping is therefore an appropriate approach to identify such FC profiles. There is a need to correlate FC profiles with underlying molecular pathologies which is thought to drive much of this clinico-anatomical heterogeneity found amongst bvFTD patients. Fine-tuning these findings will enable us to identify the specific role FC can play for disease detection or staging, differential diagnosis and measurement of disease progression in FTD.
Our study also highlighted alterations in the activity of the visual network as a potential outcome of the effect of pathology on other networks. It is not clear if this is compensatory or purely dysfunctional. These results contribute to the perspective that FTD symptomatology is linked to a dysfunctional interplay of networks, particularly along an axis differentiating observed from predicted processes. The FC of the visual network along this axis could also serve as a biomarker for disease staging or classifying patients with FTD.
Finally, if FC changes are found to precede brain atrophy as some studies suggest 37, these measures may be appropriate for stratifying patient enrolment and providing sensitive markers for evaluating the effects and efficacy of disease-modifying therapies, particularly in a framework of dynamic changes of biomarkers 8. As up to 30% of cases of FTD are due to inheriting an autosomal dominant genetic mutation 84, there is a community of individuals carrying FTD-causing mutations but who have not yet developed symptoms. These individuals offer a unique window of opportunity to study FTD at the presymptomatic stage. More work in such presymptomatic cases as well as multimodal longitudinal data are needed to clarify the role FC may have as a biomarker in FTD.