Disrupted Coupling Between Cerebral Glucose Metabolism and Intrinsic Functional Connectivity: A Hybrid PET/fMRI Study on Frontotemporal Dementia

ABSTRACT

It is increasingly established that the organization of the brain into functional resting‐state networks allows efficient integration and processing of information. Functional hubs anchoring such networks are characterized by a high degree of communication, which relies on efficient utilization of glucose. Alzheimer's disease (AD) disrupts the balance between glucose metabolism and intrinsic functional connectivity (FC). We hypothesized that this critical coupling would also be weakened in frontotemporal dementia (FTD), particularly within the salience network, given its association with the disease. Towards this goal, behavioral variant FTD (bvFTD) patients (n = 21) and healthy participants (n = 18) underwent simultaneous FDG‐PET and functional MRI imaging in a hybrid PET/MR system, with an additional cohort completing the MRI component only. PET images were converted into standardized uptake value ratios (SUVr), and local FC was quantified using regional homogeneity (ReHo) and fractional amplitude of low‐frequency fluctuations (fALFF), two metrics that have been demonstrated to be related to FDG‐PET uptake. The interplay between FC and glucose metabolism was investigated within the salience and default mode networks. The bvFTD group showed network‐level functional breakdown and significantly weakened metabolism/FC coupling, especially in the dorsal anterior insula and posterior cingulate cortex. Importantly, reduced coupling in the posterior cingulate cortex was associated with cognitive and behavioral symptoms in patients. Though significant, the reduction in whole‐brain metabolic/FC coupling in bvFTD was not as strong as reported previously for AD. These results highlight the vulnerability of functional hubs to neurodegenerative disease. Aberrant regional disruptions in the coupling between metabolism and neuronal activity may drive network‐level dysfunction and contribute to functional impairments characteristic of the disease.

Using hybrid PET/MRI, we demonstrated disrupted coupling between glucose metabolism and functional communication in FTD, alongside impaired network‐level interactions. Functional‐metabolic decoupling in key salience (SN) and default mode network (DMN) hubs highlights the vulnerability of hub regions to neurodegeneration and its impact on large‐scale brain communication.

1. Introduction

Functional connectivity (FC), which refers to spatiotemporal patterns of covariance formed by intrinsic neuronal activity, can be mapped by temporally correlating low‐frequency blood oxygen level‐dependent (BOLD) signals detected by functional magnetic resonance imaging (fMRI) (Biswal 2012; Biswal et al. 1995; Fox and Raichle 2007). When applied without requiring participants to perform sensory or cognitive tasks, these patterns reveal the resting‐state (rs) networks (Damoiseaux et al. 2006; Fox et al. 2005). Combining rs‐fMRI with metabolic imaging using positron emission tomography (PET) and with the tracer [¹⁸F]‐fluorodeoxyglucose (FDG) indicates that up to 70% of cerebral energy consumption is dedicated to maintaining spontaneous neuronal activity, particularly to support synaptic transmission (Tomasi et al. 2013). Central hubs, which facilitate high volumes of communication within brain networks, have greater glucose demand (Tomasi et al. 2013) and may be more vulnerable to disruptions in energy production (Bullmore and Sporns 2012). As a result, these hubs could be key loci of neurodegenerative and neuropsychiatric disorders (Buckner et al. 2009; Seeley et al. 2009; Sporns 2014). Investigating the functional/metabolic relationship, particularly in neurodegenerative dementia marked by reduced cerebral energy metabolism, could provide valuable insights into how disease disrupts this critical coupling.

The relationship between regional cerebral metabolism and FC is frequently investigated using fMRI metrics developed to capture various aspects of local and network‐level communication. Commonly implemented metrics include regional homogeneity (ReHo), which represents the synchronization of local neural activity (Zang et al. 2004), and fractional amplitude of low‐frequency fluctuations (fALFF), which quantifies the relative contribution of low‐frequency oscillations to the total BOLD signal, reflecting the intensity of spontaneous neural activity (Zou et al. 2008). Both have been shown to correlate with glucose consumption at the voxel level, with ReHo consistently demonstrating the strongest correlation (Aiello et al. 2015; Bernier et al. 2017; Deng et al. 2022; Volpi et al. 2024). Notably, the strength of the correlation varies across brain regions and rs networks, likely due to factors such as varying degrees of neurovascular coupling and differences in activity among rs networks (Aiello et al. 2015; Liang et al. 2013).

Although the molecular hallmarks of neurodegenerative diseases are misfolded proteins (i.e., amyloid‐β plaques and tau protein tangles), these biomarkers only correlate moderately with cognitive function, as roughly 30% of older people with Alzheimer's pathology are cognitively unimpaired (Elobeid et al. 2016). This observation suggests that the underlying mechanisms of neurodegeneration are multifaceted, and there is growing evidence that metabolic dysfunction plays a role (Muddapu et al. 2020). It is well‐known that different neurodegenerative dementias can be differentiated by specific spatial patterns of hypometabolism (Ardanaz et al. 2022; Garrett and Niccoli 2022; Salmon et al. 2024; Zilberter and Zilberter 2017). In terms of Alzheimer's disease (AD), which is the most prevalent form of neurodegeneration, regional changes in cerebral metabolism are evident before cognitive deficits manifest (Raut et al. 2023). Furthermore, greater glucose metabolism—particularly in the anterior cingulate cortex and temporal pole—has been linked to preserved cognitive function, even in the presence of significant amyloid load (Arenaza‐Urquijo et al. 2019). Given the dependence of neuronal activity on energy availability, such regional metabolic patterns underscore the need to investigate how local metabolism supports functional communication in neurodegenerative diseases. Recent hybrid FDG‐PET/rsfMRI studies have examined the impact of Alzheimer's on functional/metabolic coupling. In patients with mild cognitive impairment (MCI) and AD, voxel‐wise correlations between glucose consumption and FC were found to be significantly disrupted across the whole brain (Marchitelli et al. 2018) and within the default mode network (DMN) (Ding et al. 2021; Scherr et al. 2018, 2021). Beyond the one‐to‐one correspondence between energy metabolism and intrinsic connectivity, regional metabolic activity has also been shown to significantly contribute to connectivity within and between networks (Riedl et al. 2014). This has prompted further multimodal investigations on AD, revealing significant disruptions in the relationship between glucose consumption and inter‐regional FC (Balajoo et al. 2022), as well as network segregation—a measure of functional specialization of brain networks (Zhang et al. 2022). Moreover, multimodal measures of functional/metabolic dissociation provide a strong explanation of AD‐related cognitive deficits and may even outperform unimodal measures from PET or fMRI alone (Scherr et al. 2018, 2021).

The growing interest in examining functional/metabolic associations in aging and disease has predominantly focused on AD pathology, leaving other neurodegenerative disorders unexplored. The complexity of multimodal data requirements and the relative rarity of non‐AD dementias contribute to this research gap. This study addresses this issue by investigating potential disruptions to metabolic/functional coupling in frontotemporal dementia (FTD), which is a leading cause of early onset dementia (age < 65 years) (Knopman and Roberts 2011). FTD encompasses multiple syndromes characterized by distinct and progressive degeneration of neurons in the frontal and temporal lobes. Among the clinical sub‐types, the most common is behavioral variant FTD (bvFTD), which accounts for half of all cases (Johnson et al. 2005). Behavioral‐variant FTD manifests primarily as progressive changes in behavior, including disinhibition, apathy, repetitive behaviors, and impulsivity (Finger 2016; Laforce 2013; Piguet and Hodges 2013).

FDG‐PET studies show that FTD is associated with hypometabolism in the frontal and temporal regions extending to the insula, cingulate cortex, and subcortical regions (Garrett and Niccoli 2022; Liu et al. 2023). Resting‐state fMRI studies have highlighted reduced connectivity predominantly within the salience network (SN) (Ferreira et al. 2022; Rus et al. 2023), a critical network responsible for regulating behavioral responses to salient stimuli (Menon and Uddin 2010; Seeley et al. 2007). In addition, bvFTD is frequently associated with elevated connectivity within the DMN (Ferreira et al. 2022). The current study aimed to investigate the association between alterations in energy metabolism and intrinsic FC in bvFTD patients, focusing on the SN and DMN. By combining simultaneously acquired FDG‐PET and rs‐fMRI data, we aimed to examine dysfunctional communication at the network level and regional metabolic changes. Local connectivity was assessed using ReHo and fALFF since they are most consistently related to FDG (Aiello et al. 2015; Deng et al. 2022). Finally, voxel‐wise correlation was performed to explore the link between metabolism and intrinsic neuronal activity within the SN and DMN hubs. We hypothesized that correlations would be weakened in bvFTD, particularly in the SN, suggesting that impaired coupling between energy consumption and neural activity is a common feature of neurodegenerative dementias.

2. Methods

The study involved two research institutes: Lawson Research Institute (LRI) in London, Canada, and IRCCS SYNLAB SDN in Naples, Italy. Lawson: The study was approved by the Western University Health Sciences Research Ethics Board. IRCCS SYNLAB SDN: The study was approved by the local Institutional Review Board of the IRCCS SYNLAB SDN. All participants provided written informed consent, and both studies were conducted in accordance with the Declaration of Helsinki ethical standards.

2.1. Participants

The Lawson dataset included 41 healthy controls and 22 clinically diagnosed bvFTD patients. The IRCCS SDN dataset consisted of nine bvFTD patients, resulting in a total of 31 bvFTD patients. Details of the complete dataset are provided in Figure 1. Diagnosis was based on the international consensus criteria for bvFTD (Rascovsky et al. 2011), and the full dataset consisted of 28 probable bvFTD patients, two possible bvFTD patients, and one patient with progressive non‐fluent aphasia with behavioral variant features. Exclusion criteria included significant neurological or psychiatric disorders other than bvFTD, any major systemic illnesses, and MRI incompatibility. To assess various cognitive domains, standard neuropsychological tests including the Mini‐Mental State Examination (MMSE) and the Frontal Behavioral Inventory (FBI) questionnaire were completed. The FBI was developed to differentiate between FTD and other dementias by evaluating the presence and severity of behavioral symptoms (Kertesz et al. 1997). FBI scores consist of two subscales: scores of positive (impulsivity, irritability, etc.) and negative behaviors (apathy, indifference, etc.). Group comparisons of demographic and clinical variables between patients and controls (Table 1) and across patient sites (Table 2) were performed using independent samples t‐tests for normally distributed variables, Mann–Whitney U tests for non‐normally distributed variables, and Fisher's exact test for categorical variables.

FIGURE 1.

Study summary.

TABLE 1.

Demographic and clinical characteristics: Controls vs. All bvFTD patients.

Variables	Controls	bvFTD	p
	(n = 41)	(n = 31)
Sex (M:F)	19:22	17:14	0.63
Age	61.3 ± 8.3	64.6 ± 8.6	0.11
MMSE	29.5 ± 0.7	22.9 ± 6.5	< 0.0001 a
FBI total score	—	28.8 ± 10.9	—
FBI score (positive symptoms)	—	10.4 ± 5.8	—
FBI score (negative symptoms)	—	18.4 ± 7.6	—

Note: Test scores were available for n = 27 bvFTD patients.

^a Significant difference between controls and patients.

TABLE 2.

Site‐wise comparison of bvFTD patients (Lawson vs. IRCCS).

Variables	Lawson	IRCCS	p
	(n = 22)	(n = 9)
Sex (M:F)	11:11	6:3	0.45
Age	64.4 ± 8.8	65.1 ± 8.5	0.84
MMSE	21.4 ± 7.1	26 ± 4.1	0.10
FBI total score	32.2 ± 10.6	21 ± 8.5	0.01 a
FBI score (positive symptoms)	12.4 ± 5.1	5 ± 3.9	0.001 a
FBI score (negative symptoms)	19.7 ± 7.9	16 ± 6.6	0.25

Note: Test scores were available for n = 27 patients (Lawson: n = 19, IRCCS: n = 8).

^a Significant difference between sites.

2.2. Image Acquisition

Imaging at both centers was performed on a hybrid 3T PET/MR system (Biograph mMR, Siemens Healthcare GmbH, Erlangen, Germany) using a 12‐channel PET‐compatible head coil. All participants were required to fast for at least 6 h prior to imaging.

2.2.1. Lawson Research Institute

Patients were recruited from the Cognitive Neurology and Aging Brain clinics at Parkwood Hospital (St. Joseph's Health Care London), while age‐matched control participants were drawn from the clinic's pool of volunteers. Of the 41 controls and 22 patients, 18 controls and 12 patients underwent both FDG‐PET and MR imaging, while the rest (23 controls and 10 patients) only completed the MRI component. These additional MRI datasets, obtained from a concurrent study using the same fMRI protocol, were included to enhance the robustness of the fMRI‐specific analyses. The imaging protocol also included arterial spin labeling perfusion images, which have been previously published (Anazodo et al. 2018).

Acquisition of FDG‐PET data in list mode was started immediately after an intravenous bolus injection of [¹⁸F]‐FDG (~250 MBq). PET data from 30 to 45 min were reconstructed into one image volume (344 × 344 × 127 matrix, voxel size: 0.83 × 0.83 × 2.031 mm) using an ordered subset expectation maximum (OSEM) algorithm (3 iterations, 21 subsets). Images were corrected for scattering, dead time, and decay. Attenuation correction was performed using the vendor‐provided ultra‐short echo time (UTE) sequence.

Resting state fMRI images were acquired approximately 15 min after the [¹⁸F]‐FDG injection using a single‐shot echo planar imaging (EPI) sequence (TR/TE = 3000/30 ms, number of volumes = 164, number of slices = 40, voxel size = 3 mm isotropic, FOV = 240 mm × 240 mm). A high‐resolution T1‐weighted structural image was acquired using a 3D MPRAGE sequence (TR/TE = 2000/2.98 ms, voxel size = 1 mm isotropic, FOV = 256 mm × 256 mm).

2.2.2. IRCCS SYNLAB SDN

All nine patients underwent FDG‐PET and MRI scans. PET and rs‐fMRI data acquisitions began simultaneously 30 min after an intravenous bolus injection of [¹⁸F]‐FDG (~250 MBq). PET data from 30 to 45 min were reconstructed using an OSEM algorithm (4 iterations, 21 subsets) to one image volume (256 × 256 × 127 matrix, voxel size: 1.21 × 1.21 × 2.031 mm). Images were corrected for scattering, dead time, and decay, and the vendor‐provided UTE sequence was used to correct for attenuation.

Functional MRI data was collected using a single‐shot EPI sequence (TR/TE = 1000/21.4 ms, number of volumes = 350, number of slices = 40, voxel size = 4 mm isotropic, FOV = 256 mm × 256 mm). A high‐resolution structural image was acquired using a 3D T1‐weighted MPRAGE sequence (TR/TE = 2400/2.25 ms, voxel size = 0.8 mm isotropic, FOV = 192 mm × 214 mm).

2.3. Data Pre‐Processing

2.3.1. MRI Pre‐Processing

Pre‐processing of rs‐fMRI data was conducted using the CONN toolbox (Whitfield‐Gabrieli and Nieto‐Castanon 2012), a FC toolbox based on the Statistical Parametric Mapping (SPM) software package (http://www.fil.ion.ucl.ac.uk/spm/software/spm12). First, time series data were corrected for motion, with volumes exhibiting framewise displacement greater than 0.9 mm determined as outliers (Power et al. 2012). Slice time correction was applied, and images were co‐registered to corresponding T1‐weighted structural images. Structural images were then segmented into probabilistic gray matter, white matter, and cerebrospinal fluid (CSF) maps to generate binary masks for later use. The warping parameters obtained from the segmentation process were used to normalize the fMRI data to the standard Montreal Neurological Institute (MNI) template, resampling the images to an isotropic resolution of 2 mm³.

Following spatial normalization, the anatomical component‐based denoising procedure, aCompCor, was performed (Behzadi et al. 2007). This procedure included regressing out the six motion parameters and five principal components of global white matter and CSF signals, which have been shown to more effectively reduce motion‐related artifacts than mean WM/CSF signal regression (Muschelli et al. 2014). To account for the influence of outlier scans on the BOLD signal, these were included as additional nuisance regressors. Finally, linear detrending was applied to the functional data, and band‐pass filtering (0.01–0.1 Hz) was performed to isolate low‐frequency fluctuations.

2.3.2. PET Pre‐Processing

FDG‐PET data were preprocessed using SPM12 and in‐house MATLAB (R2022b, The MathWorks, Natick, MA, USA) scripts. FDG maps were converted into standardized uptake values (SUV) using the subject's weight and injected dose, which were then co‐registered to T1‐weighted structural images. The normalization parameters obtained from T1 segmentation were used to warp the maps to MNI space, and the images were resampled to a 2 mm³ isotropic resolution. Subsequently, the SUV maps were converted into standardized uptake value ratios (SUVr) by normalizing against the mean occipital lobe activity. The cerebellum was not chosen as a reference region due to growing evidence of cerebellar dysfunction in FTD (Chen et al. 2019; Olivito et al. 2022). Finally, the images underwent spatial smoothing with a 6‐mm FWHM Gaussian kernel, followed by z‐standardization with respect to the gray matter mean and standard deviation of the maps.

2.4. Local Connectivity Metrics

ReHo was calculated (Zang et al. 2004) using Kendall's coefficient of concordance (KCC) (Kendall and Gibbons 1990), and fALFF was calculated as the ratio of power in the low‐frequency band (0.01–0.1 Hz) to the total power across the entire frequency spectrum (Zou et al. 2008). Both metrics have been used to demonstrate disrupted intrinsic connectivity in clinical populations (Farb et al. 2013; Marchitelli et al. 2018; Zhang et al. 2012). ReHo and fALFF maps were calculated within each subject's gray matter mask (generated at a 50% probability threshold) using the REST toolbox (Song et al. 2011). ReHo was calculated considering a cluster of 27 nearest neighbor voxels. Spatial smoothing was then applied to both sets of connectivity maps using a 6‐mm FWHM Gaussian kernel. Afterward, each metric's gray matter mean and standard deviation were used to standardize the maps into z‐scores.

2.5. Regions of Interest (ROI)

Image analysis focused on the SN and the DMN as the SN is preferentially targeted by bvFTD (Ferreira et al. 2022) and there is evidence of DMN involvement as well (Ferreira et al. 2022). Hub regions in the SN and DMN networks were derived from the CONN toolbox (Whitfield‐Gabrieli and Nieto‐Castanon 2012), which are derived from CONN's independent component analysis (ICA) of the Human Connectome Project (HCP) dataset comprising 497 subjects. Given its role as a critical hub of the SN, the anterior insular region was subdivided into dorsal and ventral regions to reflect their established functional differences: the dorsal portion is broadly involved in cognitive control, salience detection, and decision‐making, while the ventral portion contributes more to emotional and autonomic regulation (Droutman et al. 2015; Menon 2025). The masks for the dorsal and ventral anterior insula were derived from the Human Brainnetome Atlas (Fan et al. 2016); further details regarding the parcellation are provided in the Supporting Information material (Figure S1). The final set of ROIs included the anterior cingulate cortex (ACC), bilateral dorsal and ventral anterior insula (dorsal anterior insula [dAI] and ventral anterior insula [vAI]), and bilateral rostral prefrontal cortex (RPFC) in the SN, and the medial prefrontal cortex (MPFC), posterior cingulate cortex (PCC), and bilateral parietal cortex (LP) in the DMN. The MNI coordinates of ROIs (n = 11) are provided in Supporting Information material (Table S1).

2.6. Region to Region FC

Resting‐state FC analysis was performed using the CONN toolbox (Whitfield‐Gabrieli and Nieto‐Castanon 2012) to examine the communication within and between networks of interest (DMN and SN). FC was calculated as the bivariate Pearson correlation between the mean time series of pairs of ROIs. This was applied across all sets of ROIs of both networks. The resulting correlation coefficients (r) were converted to z values using Fisher's r‐to‐z transformation, and two‐sample t‐tests were conducted for group comparisons (p < 0.05). Correction for multiple comparisons was applied using the false discovery rate (FDR) approach.

2.7. Voxel‐Wise Comparisons

Maps of FDG‐SUVr and rs fMRI metrics (ReHo and fALFF) between patients and controls were compared voxel‐wise. SUVr maps were compared for a subset of 21 patients and 18 healthy controls, while rsfMRI metrics were analyzed for the entire dataset. Voxel‐wise differences were investigated through independent two‐sample t‐tests implemented using the program 3dttest++ available in the Analysis of Functional NeuroImages (AFNI) software package (Cox 1996). As bvFTD is characterized by significant gray matter atrophy (Marino et al. 2019; Pan et al. 2012), which could potentially bias voxel‐wise comparisons (Xie et al. 2015), probabilistic maps of gray matter were added as covariates. In addition to gender and age, the scan site was also added as a scalar covariate. Statistical t‐maps were transformed into z‐scores. AFNI's 3dClustSim correction was used to correct for multiple comparisons. After identifying clusters surviving the initial voxel‐level threshold p, this program uses Monte Carlo simulations (10,000 iterations in this case) to simulate noise‐only random datasets to determine the probability (α) of false‐positive clusters. Statistically significant clusters were identified by adopting p < 0.05 and α < 0.05. This correction strategy was previously applied in an Alzheimer's study that also investigated voxel‐wise differences in FDG uptake and rs fMRI measures (ReHo and fALFF) (Marchitelli et al. 2018). As recommended by Poldrack et al. (2017), the resulting unthresholded z‐score maps will be made publicly available via the NeuroVault repository for visualization and future meta‐analytic use.

2.8. FDG/fMRI Correlation

To assess functional‐metabolic linkage and the effect of bvFTD on this association, voxel‐wise Spearman correlations between the z‐scored maps of FDG‐SUVr and the two FC metrics were computed in MATLAB within the ROIs of the SN and DMN. Correlations over whole‐brain gray matter were also computed. Spearman was chosen because the variable distributions deviated from normality, as determined by the Shapiro–Wilk test. Voxel‐wise GM probability values were added as covariates when computing correlations to control for atrophy‐related biases. As recommended by prior voxel‐wise correlation studies, each region contained a sufficient number of voxels (minimum ~500) to maintain statistical robustness (Riedl et al. 2016; Scherr et al. 2021).

Multiple linear regression was conducted to investigate group‐level differences in FDG/fMRI correlations while controlling for age, sex, and scan site. FDR correction was applied to correct for multiple comparisons.

2.9. Clinical Correlation Analysis

To examine potential associations between FDG/fMRI coupling and clinical measures, Spearman correlations were computed between regional coupling values and Mini‐Mental State Examination (MMSE) and Frontal Behavioral Inventory (FBI) scores across bvFTD patients (n = 17). As the variable distributions were non‐normal, as indicated by the Shapiro–Wilk test, a rank‐based test was used. Analyses were performed separately for FDG/ReHo and FDG/fALFF coupling. Statistical significance was defined as p < 0.05.

3. Results

3.1. Clinical Characteristics of Participants

Demographic characteristics comparing bvFTD patients and controls are presented in Table 1. Patient and control groups did not differ significantly in terms of age or sex. Compared to controls, the patient group had significantly lower mean MMSE scores. Table 2 summarizes site‐wise demographic and clinical characteristics of the bvFTD cohort. No significant differences were observed between the two sites in age, sex, or MMSE scores. However, FBI scores were significantly higher in patients from Lawson compared to those from IRCCS.

3.2. FC Analysis

Z‐scores of ROI‐to‐ROI connectivity were compared between 31 bvFTD patients and 41 controls within and between the ROIs of the SN and DMN. Figure 2 presents all statistically significant changes. Within the SN, bvFTD was associated with significant reductions in connectivity between the ACC and all subdivisions of the AI (dAI and vAI, left and right). Connectivity between the right AI (dAI and vAI) and left RPFC was also reduced. Between networks, reduced connectivity was observed between the left vAI and MPFC, and between the right vAI and PCC.

FIGURE 2.

Impact of bvFTD on ROI‐to‐ROI functional connectivity (FC) within the SN and DMN. All statistically significant FC changes (p < 0.05, FDR corrected) between patients and controls are shown. Additional statistical details are provided in Supporting Information (Table S2).

In contrast, bvFTD patients showed increased connectivity within the DMN and between the SN and DMN. Specifically, connectivity increased between the PCC and left and right lateral parietal cortex (LP) within the DMN, and between the left RPFC (SN) and right LP (DMN).

3.3. Voxel‐Wise Comparisons

Results from the voxel‐wise t‐tests for FDG‐SUVr, ReHo, and fALFF are shown in Figure 3 and Supporting Information material (Table S3). Patients displayed significant cortical hypometabolism predominantly in the frontal and temporal lobes, with metabolic disruptions extending to posterior regions. The reductions were observed in both hemispheres, showing an almost symmetrical pattern.

FIGURE 3.

Voxel‐wise differences in metabolism and connectivity metrics between patients and controls. Top row: FDG, middle row: ReHo, and bottom row: fALFF. T values of unpaired t tests converted to z‐scores are shown.

Both fMRI metrics (ReHo and fALFF) showed significant disruptions in the bvFTD group. ReHo alterations were primarily observed in the cingulate cortex, involving the paracingulate gyrus and the anterior and posterior divisions of the cingulate gyrus. Changes in the posterior division extended to the precuneus. Frontal regions were spared. Additional decreases were found in the left hemisphere.

Fractional ALFF displayed a more widespread pattern of disruption. While it showed significant involvement of the cingulate cortex like ReHo, the alterations in BOLD signal amplitude involved multiple frontal regions. Reduced fALFF was found in the insular cortex, covering all insular sub‐regions in both hemispheres. Additionally, the caudate nucleus (L and R) and the hippocampus (L and R) were significantly affected.

3.4. Association Between Metabolism and Connectivity Metrics

Significant correlations (p < 10⁻⁵) were found between glucose metabolism and connectivity metrics in all regions of interest. Notably, the strength of the coupling varied among regions, with DMN regions showing stronger correlations. Scatter plots of voxel‐wise correlations between group‐averaged SUVr and fMRI maps for major nodes of the DMN and SN are shown in Figure 4.

FIGURE 4.

Scatter plots showing voxel‐wise Spearman correlations (r) between group‐averaged z‐scores from FDG‐PET and each of the fMRI metrics (ReHo and fALFF). Analysis was conducted in five brain regions: Posterior cingulate cortex (PCC), dorsal anterior insula (dAI, left and right), and ventral anterior insula (vAI, left and right). Each dot represents the mean voxel value averaged across all subjects for either controls (blue) or patients (red).

Group‐level comparisons showed that the strength of the FDG‐fMRI coupling in major network hubs was diminished in bvFTD (p < 0.05, FDR corrected) (Figure 5). The strongest reductions were observed in the PCC, which showed significantly decreased coupling in both FDG/ReHo and FDG/fALFF analyses. Within the SN, FDG/ReHo correlation was significantly reduced in the left and right dAI and the left RPFC, while no significant changes were observed in the vAI. In the FDG/fALFF analysis, no additional regions survived FDR correction. In all cases, age, sex, and study location had no significant contribution to the observed group differences. On a whole‐brain level, while FDG/ReHo correlations were significantly reduced in patients (0.51 ± 0.07 vs. 0.44 ± 0.06, p = 0.022), the FDG/fALFF association did not reach statistical significance (0.41 ± 0.11 vs. 0.36 ± 0.08, p = 0.39).

FIGURE 5.

Voxel‐wise correlation between FDG and fMRI metrics within the nodes of the default mode network (DMN) and the salience network (SN). (A) Metabolism versus ReHo. (B) Metabolism versus fALFF. Significant differences are indicated as follows: *p < 0.05, **p < 5 × 10⁻³, ***p < 5 × 10⁻⁵ (all FDR adjusted). Full statistical details for all ROIs are provided in the Supporting Information (Table S4).

3.5. Regional Coupling and Clinical Measures

Significant associations were observed between regional FDG/fMRI coupling and clinical scores in bvFTD patients. Specifically, FDG/ReHo coupling in the PCC was positively correlated with MMSE score (r = 0.64, p = 0.006) and negatively correlated with total FBI scores (r = −0.69, p = 0.0026) (Figure 6).

FIGURE 6.

Association between clinical measures and metabolic‐functional coupling in the posterior cingulate cortex (PCC) in bvFTD patients (n = 17). Spearman correlations (r) and p values are shown.

4. Discussion

In this study, simultaneously acquired FDG‐PET and rs‐fMRI data were used to investigate how FC, glucose metabolism, and their interplay were affected by bvFTD. Significant alterations in FC were identified within the SN and DMN in bvFTD patients. Glucose hypometabolism, as indicated by SUVr, was most pronounced in the frontal and temporal lobes in the patient group. Although both intrinsic connectivity metrics (ReHo and fALFF) exhibited disease‐related changes, their involvement in frontotemporal regions was less pronounced compared to FDG‐PET. Strong correlations between FDG‐PET and each of the connectivity metrics were observed across the SN and DMN, but this functional/metabolic coupling was significantly reduced in bvFTD, particularly within the dorsal AI and PCC. Our results support the hypothesis that the dissociation of metabolism and FC observed previously in patients with AD is also found in patients with FTD, suggesting it is a shared trait across neurodegenerative dementias, with these changes being most evident in regions associated with the specific disease.

Neurodegenerative disorders are known to target macroscale functional brain networks defined by synchronous rs activity (Seeley et al. 2009; Zhou et al. 2012) and are thought to spread along these networks through the propagation of pathological proteins (Drzezga 2018; Frost and Diamond 2010; Raj et al. 2012). The SN, initially identified by Seeley and colleagues (Seeley et al. 2007), is a primary target of functional dysfunction in bvFTD (Kamalian et al. 2022; Seeley 2010; Seeley et al. 2009; Zhou et al. 2010). We found aberrant reductions in FC within the SN, specifically between the ACC and both dorsal and ventral subdivisions of the bilateral AI, as well as between the right AI and left RPFC. The finding of reduced connectivity within the SN is consistent with previous studies of bvFTD (Borroni et al. 2012; Ferreira et al. 2022; Filippi et al. 2013; Ng et al. 2021). This network, anchored by the ACC and bilateral AI, plays a pivotal role in integrating relevant internal and external stimuli to guide appropriate behavioral and emotional responses (Menon and Uddin 2010; Seeley 2019; Seeley et al. 2007)—an attribute that is critically impaired in bvFTD.

Between‐network reductions were also observed between the left ventral AI and the MPFC, and between the right ventral AI and PCC. We also found elevated connectivity within the DMN between the lateral parietal cortices and PCC. Although less frequently observed, increased activity within the DMN in bvFTD has been reported previously (Ferreira et al. 2022) and is associated with reduced connectivity in the anti‐correlated SN (Zhou et al. 2010). It has been speculated that elevated DMN activity reflects a compensatory mechanism in response to neuronal dysfunction (Borroni et al. 2012; Meijboom et al. 2017; Zhou et al. 2010). Increased connectivity was also observed between the SN and DMN, specifically between the left RPFC and right lateral parietal cortex. Since these two networks are anti‐correlated, increased FC could be interpreted as increased dissociation between these critical networks, as speculated in a recent meta‐analysis on bvFTD (Kamalian et al. 2022). It is worth noting that findings related to the DMN in bvFTD have been mixed, with both increases and decreases in connectivity reported across studies (Kamalian et al. 2022). This variability may reflect small sample sizes and the inherent pathological heterogeneity of bvFTD. Nonetheless, it is widely accepted that the DMN is affected in bvFTD, while reduced FC in the SN remains the most robust and reproducible network‐level finding (Kamalian et al. 2022), which our results also confirm.

Voxel‐wise comparisons of PET SUVr maps revealed significantly reduced cortical metabolism in bvFTD, primarily involving the frontal and temporal gyri, orbitofrontal cortex, and paracingulate gyri, with posterior extensions to the supramarginal and angular gyri. These reductions were largely symmetrical across hemispheres, with no significant metabolic decline observed in subcortical structures. The observed bilateral frontotemporal hypometabolism aligns with the known pattern of metabolic deficits in bvFTD (Foster et al. 2007; Minoshima et al. 2022). Compared to FDG‐PET, changes captured by fMRI metrics were less extensive, with fALFF reductions observed in some frontotemporal regions, insular cortex, hippocampus, and caudate nucleus. However, similar changes were not observed in ReHo maps. Notably, all three metrics converged on reductions in the anterior cingulate and paracingulate, consistent with their established importance in bvFTD (Kamalian et al. 2022). The more widespread spatial extent of FDG‐PET findings may reflect modality‐specific signal characteristics with a high signal‐to‐noise ratio (SNR) for FDG‐PET compared to the rs‐fMRI metrics, which are susceptible to physiological noise and motion‐related artifacts. In addition, the relative sparing of frontal regions by fMRI metrics may reflect reduced signal quality in atrophic regions, where voxel‐wise correction for GM may not fully compensate for atrophy‐related reductions in signal quality. Similar findings have been reported in AD, where FDG‐PET revealed broader spatial involvement than fMRI‐based measures (Marchitelli et al. 2018). As an additional test of robustness, we applied a more conservative thresholding approach using FDR correction. Under FDR, the spatial patterns in the FDG‐PET and fALFF maps remained largely consistent, though with reduced spatial extent. The ReHo effects did not survive, except for a small (> 100 voxels) cluster near the posterior cingulate. These results are presented in Supporting Information material (Figure S2).

The key finding of the current study was reduced coupling between glucose metabolism and intrinsic FC associated with bvFTD. This aligns with prior observations of diminished voxel‐wise functional‐metabolic correlations in AD on a whole‐brain level (Marchitelli et al. 2018). While this whole‐brain decoupling (FDG/ReHo) was significant, it was less pronounced than reported previously for AD (i.e., an effect size of 1.07 compared to 2.95 in AD; Marchitelli et al. 2018) and not significant for the FDG/fALFF relationship. These differences may reflect the distinct sensitivities of functional‐metabolic coupling to the pathological processes underlying each disorder. In AD, Scherr et al. proposed that the decoupling of metabolism and FC is driven by amyloid‐beta (Aβ) accumulation (Scherr et al. 2018), the primary pathological substrate of the disease (Reiss et al. 2018). Different proteinopathies, such as those of bvFTD and AD, could exert distinct influences on metabolic processes, including glycolysis, glucose transport, insulin signaling, and mitochondrial function (Garrett and Niccoli 2022; Rhein and Eckert 2007). These variations in metabolic modulation could contribute to the difference in whole‐brain functional‐metabolic coupling between AD and bvFTD. It has been suggested that reduced coupling between metabolism and connectivity in AD could partly reflect cerebrovascular dysfunction (Marchitelli et al. 2018). Though less well characterized, recent evidence suggests FTD has a vascular component (Chakraborty et al. 2024; Chu et al. 2024; Gerrits et al. 2022), which could contribute to the observed whole‐brain decoupling. Future studies may further examine the contribution of vascular factors to functional‐metabolic decoupling in neurodegenerative diseases.

Hybrid imaging studies have highlighted significant regional variability in the relationship between metabolic activity and neuronal function in the healthy brain (Aiello et al. 2015; Shokri‐Kojori et al. 2019; Volpi et al. 2024), driven by factors such as the regional distribution of metabolic pathways (Vaishnavi et al. 2010) and differing activity levels of rs networks and their hubs (Palombit et al. 2022). To assess possible regional changes, the current study focused on metabolic/FC coupling in the major hubs of the SN and DMN. Marked dissociations were found within the nodes of the SN, especially in bilateral dorsal AI, where FDG/ReHo coupling was significantly reduced. A similar pattern was observed for FDG/fALFF in the right dorsal AI (uncorrected p = 0.021), though this did not survive FDR correction (FDR‐adjusted p = 0.057). In contrast, the ventral AI showed no coupling changes in either hemisphere, despite being involved in the broader connectivity breakdown of the SN.

The differential impact of bvFTD on functional–metabolic coupling in the dorsal and ventral AI may reflect their distinct functional specializations. The ventral region is more engaged in affective and emotional processing (Chang et al. 2013; Kurth et al. 2010), while the dorsal region primarily supports cognitive control, including attention, decision‐making, and goal‐directed behavior (Droutman et al. 2015; Eckert et al. 2009; Menon 2025; Menon and Uddin 2010). Most notably, the dAI plays a critical role in regulating dynamic transitions between the DMN and the central executive network (CEN). This role has been consistently observed across both resting and task‐engaged conditions, where the dAI appears to act as a switch that enables the brain to shift from internally focused thoughts to goal‐directed actions (Menon and Uddin 2010; Sridharan et al. 2008). Furthermore, dynamic FC studies highlight the unique role of the dAI within the insular cortex, showing its ability to alternate between brain networks over time (Nomi et al. 2016). Considering the role of the dAI in dynamically coordinating multiple networks, it likely depends on tight coupling between functional activity and metabolism. The observed decoupling may therefore reflect a breakdown in this balance, potentially impairing its capacity to support flexible network transitions in bvFTD. Notably, this dissociation emerged despite the AI exhibiting only a few hypometabolic voxels in the FDG SUVr maps, highlighting the sensitivity of PET/rs‐fMRI coupling metrics to detect subtle disruptions in the relationship between metabolism and neuronal function.

In the DMN, the PCC exhibited strong reductions in the coupling between FDG and the two FC metrics. Similar to the dAI, the functional‐metabolic decoupling in the PCC was observed despite no significant hypometabolism. As a major functional hub of the brain (Buckner et al. 2009; Tomasi and Volkow 2010), the PCC is highly metabolically active and integral to the DMN, maintaining robust structural and functional connections with other brain regions (Hagmann et al. 2008). The PCC has long been recognized as a consistent target in AD (Bonte et al. 2004; Boxer et al. 2003). Recent studies have demonstrated that the PCC is not exclusive to AD pathology and is also affected in FTD disorders (Bergeron et al. 2020; Cayir et al. 2024; Scheltens et al. 2018). Together, these findings underscore the vulnerability of metabolically active network hubs, such as the PCC, across neurodegenerative diseases.

It is worth noting that the correlations between FDG and ReHo were more sensitive to disease‐related changes than those between FDG and fALFF. This difference likely reflects inherent distinctions between FDG–ReHo and FDG‐fALFF coupling. Prior work modeling the extent to which regional FDG variance is explained by rsfMRI metrics has shown that ReHo exhibits the strongest correspondence with FDG among commonly used fMRI measures (Volpi et al. 2024). In our data, FDG‐ReHo Spearman correlations were higher than FDG‐fALFF in all regions except the PCC (Figure 4), consistent with reports that synchrony‐based metrics show stronger associations with FDG than amplitude‐based measures such as fALFF (Bernier et al. 2017; Jiao et al. 2019; Volpi et al. 2024). Given the stronger relationship between FDG and ReHo, this coupling is likely more sensitive to disease‐related alterations and therefore showed more significant reductions in FTD compared to FDG‐fALFF. Interestingly, the PCC, which was the only region with a significant reduction in FDG‐fALFF coupling (Figure 5), was also the region with a strong FDG‐fALFF association in healthy controls (Figure 4). We believe that the weaker FDG‐fALFF correlation in other regions contributed to fewer detectable disease‐related changes relative to FDG‐ReHo. Accordingly, the study was likely underpowered to detect subtle FDG‐fALFF decreases outside the PCC.

Significant associations were observed between functional/metabolic coupling and clinical measures in the PCC. These findings suggest that regional functional/metabolic coupling may be sensitive to clinical severity in bvFTD, consistent with prior works in AD highlighting FDG/fMRI coupling as a marker of disease‐related functional disruption (Scherr et al. 2018, 2021). Lower FDG/ReHo coupling in the PCC of patients was associated with diminished cognitive performance (MMSE) and behavioral deficits (FBI). Beyond its role in memory and attention, the PCC has been implicated in behavioral flexibility (Pearson et al. 2011). Seminal electrophysiological studies in non‐human primates suggest that the PCC monitors the outcomes of actions and signals when adjustments in strategy are needed, supporting adaptive responses to changing environmental demands (Hayden et al. 2008, 2009; Pearson et al. 2009). Disruption in this functional/metabolic linkage may therefore contribute to the behavioral inflexibility and social difficulties observed in bvFTD. While behavioral symptoms in FTD have largely been examined through the lens of SN disruption, the DMN—in particular the PCC—has also been increasingly implicated in the disorder. Although the contribution of the PCC to behavior in FTD is not well understood, its established role in monitoring environmental changes and guiding behavioral adjustments suggests it may play a broader role than previously appreciated.

The current study has some limitations, including pathological heterogeneity inherent to bvFTD. Behavioral‐variant FTD can be categorized based on underlying proteinopathy, with tau, TDP‐43, and fused‐in‐sarcoma being the three primary proteins (Mann and Snowden 2017). While each subtype may modulate glucose metabolism in distinct ways (Garrett and Niccoli 2022), differences in regional metabolic alterations associated with these subtypes are unknown. Considering the rarity of bvFTD syndromes, obtaining a sufficient sample size that accounts for these subtypes poses a substantial challenge. Our findings of network‐related dysfunction, despite possible pathological heterogeneity, underscore the vulnerability of specific networks to the disease (Seeley et al. 2009).

Another limitation was the amalgamation of data from two centers, which introduces the possibility of site‐related variability in group comparisons. To mitigate this concern, all PET and fMRI maps were generated using z‐standardization to minimize inter‐subject variability. Additionally, regression analyses revealed no significant effect of scan site on the results. However, patients recruited from the Lawson site showed more severe behavioral symptoms. Since coupling was calculated within each subject, the analysis is less affected by overall differences between patient groups from the two sites.

An important consideration is the spatial definition of ROIs used for the coupling analysis. While we relied on anatomically defined regions from standard atlases, the metabolism–function relationship may vary within these regions. Data‐driven approaches, such as voxel‐wise clustering (e.g., K‐means), could help identify subregions with distinct coupling profiles, as shown by Shokri‐Kojori et al. (2019). Incorporating such methods in future work could enhance sensitivity to regionally specific effects of neurodegenerative disease.

5. Conclusion

Simultaneously acquired FDG/PET and rs‐fMRI data were used to investigate network‐level dysfunction in terms of the coupling between energy metabolism and functional communication in bvFTD. Weakened functional/metabolic coupling was found, extending previous studies of AD. Both the SN and DMN were targeted in bvFTD, with their main hubs being primary targets of disruptions in the critical relationship between energy metabolism and neuronal activity. Our findings further highlight the vulnerability of functional hubs to neurodegenerative diseases, which may contribute to the disintegration of intrinsic connections due to impaired metabolic processes, which could eventually lead to network‐level functional breakdowns. Future studies are needed to determine how metabolism/activity decoupling is related to the molecular pathological signatures of bvFTD and how this decoupling relates to neurocognitive deficits.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1: hbm70388‐sup‐0001‐Supinfo.docx.

Joshy, M. , Liu L., Dassanayake P., et al. 2025. “Disrupted Coupling Between Cerebral Glucose Metabolism and Intrinsic Functional Connectivity: A Hybrid PET/fMRI Study on Frontotemporal Dementia.” Human Brain Mapping 46, no. 15: e70388. 10.1002/hbm.70388.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.