Neurodegenerative disorders are among the top leading cause of death and disabilities worldwide [
1]. Among people aged more than 65 years old, Alzheimer’s disease (AD) is the most common form of neurodegeneration, while frontotemporal dementia (FTD) represents the first cause of cognitive impairment in younger individuals. In AD, memory is typically the earliest sign of cognitive deterioration. FTD serves as an umbrella term for several clinical syndromes, including the behavioral variant FTD (bvFTD), usually characterized by behavioral disturbances in the earliest stages [
2,
3]. To date, no cure is available for these diseases. Recent significant advancement in the pharmacological field has been done, although findings are still far from being conclusive [
4‐
6]. A better understanding of the pathophysiological mechanisms underlying these cognitive/behavioral symptoms might pave the way to novel treatments and rehabilitation options [
7].
Resting-state functional magnetic resonance imaging (rs-fMRI) is widely used to assess the putative functional architecture of the brain at rest. This technique can investigate in vivo brain oscillations in the blood oxygen level-dependent (BOLD) signal between different brain regions. Brain areas showing temporal BOLD synchronization are assumed to be functionally grouped into neural networks. Functional connectivity (FC) exhibits a low-dimensional spatiotemporal pattern [
8,
9]. This functional scaffold might have a representational role for cognitive abilities [
10,
11]. The default mode network (DMN) is associated with episodic memory performance and shows a gradual shrinking with aging, in line with the natural decline of memory performance in the elderly [
12]. Similarly, a group of “attentional networks” is linked with executive, language, and attentional abilities, that is the frontoparietal (FPN), the dorsal attention (DAN), and the ventral attention (VAN) networks.
In typical AD, breakdown of DMN is linked with core symptoms, i.e., impaired episodic memory [
13]. By contrast, bvFTD manifests reduced FC of the VAN (also referred to as salience network) that is linked with clinical severity [
14]. Other cognitive functions and networks are involved during disease progression, such as attentional networks/functions in both conditions [
15,
16]. However, a simple 1:1 relationship between (lower) FC and (impaired) cognition is too simplistic to explain the complex pattern of cognitive and brain changes. Large-scale networks are closely interconnected and alterations in one network can have effects on other networks and undermine the balance of this functional scaffold. The triple network theory states that aberrant dynamic cross-network interactions of the VAN, FPN, and DMN underlie a wide range of cognitive/behavioral disturbances [
17]. This theory posits that VAN integrates external information acting as an interface between DMN and FPN, regulating their competing inter-network activity and promoting appropriate behavioral response. Similarly, the VAN acts as a circuit breaker when attention is reoriented to relevant environmental stimuli, interrupting ongoing activity in the DAN, which in turn shifts attention to the new source of information [
18,
19]. These studies suggested a general role in switching between networks supporting cognitive functions, which may explain previous evidence of between-network alterations in neurological disorders. Brain stroke lesions increase connectivity between networks commonly anti-correlated, such as the DMN and the DAN, with detrimental consequences on cognitive abilities [
20]. In neurodegenerative disorders, the pivotal study of Zhou et al. [
14] reported a divergent connectivity pattern in AD and bvFTD, whereby reduced connectivity of the DMN in AD was accompanied by hyper-connectivity of the salience network, while the opposite was seen in bvFTD. More recently, the same group observed that AD and bvFTD show divergent abnormalities in the topological organization of functional brain networks extending into subcortical and inter-network connections [
21]. These studies pointed out a complex pattern of network connectivity alterations within the connectivity gradient. However, the relationship between this divergent functional pattern in AD and bvFTD and cognition is still unclear. A “classical” cognitive-network approach, which considers the relationship between a single test score (or a composite score across apriori defined domains) with FC might mask some latent relationships, considering also that cognitive scores are highly correlated. Here, we aimed at identifying the latent cognitive space in the continuum from normal to pathological aging. A low latent behavioral space was previously reported in stroke patients, supporting the validity of this approach in neurological diseases. These studies showed that three main cognitive components explained the large majority of variance in cognitive performance [
22,
23], linked with a specific spatiotemporal trajectory of functional networks [
24]. Based on these premises, in the present study, we sought to investigate whether each cognitive “motif” may capture a specific spatiotemporal pattern of neural cortical connectivity across different neurodegenerative diseases. To this aim, we used (i) a principal component analysis (PCA) to identify low-dimensional representations of cognition and (ii) a connectome analysis to assess FC patterns in core cognitive networks. We then assessed convergent and divergent cognitive-connectivity relationships using univariate and multivariate analyses. We hypothesize to find, within a latent low-dimensional space, both shared and divergent cognitive-FC patterns.