Salience network resting-state activity predicts progression in frontotemporal dementia

Abstract

Importance:

Non-invasive measures of activity within intrinsic brain networks may be clinically relevant, providing a marker of neurodegenerative disease and predicting clinical behaviors.

Objective:

To correlate baseline resting-state measures within the salience network and changes in behavior in patients with frontotemporal dementia.

Design:

Baseline resting-state functional magnetic resonance imaging data and longitudinal clinical measures were obtained from prospectively-accrued patients over an 8-week interval.

Setting:

Tertiary academic care center specializing in assessment and management of patients with neurodegenerative disease.

Participants:

Fifteen patients with clinically-diagnosed frontotemporal dementia (5 behavioral variant, 10 semantic dementia).

Main Outcome Measure:

Baseline resting-state fMRI data measured within regions of interest were regressed on serial behavioral measures from prospectively-accrued patients with frontotemporal dementia to determine the ability of baseline resting-state activity to account for changes in behavior.

Results:

Low-frequency fluctuations in the left insula significantly predicted changes in frontal behavioral inventory scores (std. beta=0.51, p=0.049), accounting for 28% of the change variance. The trend was driven by changes in measures of apathy, independent of dementia severity.

Conclusions and Relevance:

Baseline measures of salience network connectivity involving the left insula may predict behavioral changes in patients with frontotemporal dementia.

Introduction

Advances in functional neuroimaging have provided a window into the brain’s intrinsic connectivity, leading to the discovery of the salience network (SLN), comprised of the anterior cingulate, insula, striatum and amygdala. The SLN is activated in healthy patients during tasks requiring attentional selection, task-switching and self-regulation of behavior,¹ and is an important neural substrate in frontotemporal dementia (FTD),² with dysfunction confirmed on histopathology³ and resting-state functional magnetic resonance imaging (fMRI).^1,2 Within the SLN, the insula has emerged as a nodal point of particular importance for frontolimbic function² and dysfunction. Supporting this assertion, insular atrophy is recognized as one of the earliest structural biomarkers in behavioral variant FTD (bvFTD) and semantic dementia (SD),^3,4 with insular loss correlated with worsening behavioral inventory scores,⁵ and progressive accumulation of FTD-associated pathologic inclusions within insular von Economo neurons and fork-cells.^6,7

Abnormal activity within intrinsic brain networks may be clinically relevant, indicative of neurodegenerative disease.^1,2 Resting-state fMRI may provide a non-invasive biomarker for the diagnosis and longitudinal monitoring of patients with FTD. It remains to be determined, however, whether this emerging technique can be used to identify patterns of network disruption in patients prior to development of changes on clinical examination or structural neuroimaging. We explored the ability of baseline resting-state connectivity measures to predict behavioral changes in participants with bvFTD and SD over an 8-week interval.

Methods

Baseline resting-state fMRI data were collected from 15 participants with FTD (5 bvFTD, 10 SD) prior to initiation of study medication as part of the protocol for a prospective open-label clinical trial. Details concerning study recruitment have been published previously.⁸ In the present study, the subtype diagnosis for two patients has been amended from bvFTD to SD, as symptoms of the semantic variant of primary progressive aphasia manifested after clinical trial completion.⁹ Control participants were not explicitly recruited for the open-label trial; however, resting-state data was available from 16 age-matched healthy volunteers recruited for a parallel study and was included in comparison analyses. Control participants did not differ in age, gender or education.¹ The term “FTD” refers here to both the bvFTD and SD subtypes of FTD.

Behavioral measures were obtained at baseline and at two-months from FTD participants using the Frontal Behavioral Inventory (FBI) total score, with apathy and disinhibition sub-scores,¹⁰ and the more global Clinical Dementia Rating score.¹¹ Participants with FTD received the clinical intervention, memantine hydrochloride 10 mg, twice daily. All procedures were approved by institutional ethics boards. Written informed consent was obtained from patients or their substitute decision makers.

The fMRI protocol directed participants to lie with eyes closed during image acquisition. Data was preprocessed using the Data Processing Assistant for Resting-State fMRI (Song Xiaowei, http://www.restfmri.net). To identify SLN hubs for our analysis, we compared resting-state activity in participants with FTD to healthy controls and selected ROIs based upon regions of maximal group distinction. Three regions of interest (ROIs) were identified: the right and left insulae, and medial anterior cingulate cortex extending into both hemispheres (eTable 1).¹

Resting-state activity within the right insula, left insula, and anterior cingulate ROIs were separately assessed using two distinct measures of voxel-wise signal power and homogeneity. Signal power was measured using fractional amplitude of low-frequency fluctuation (fALFF)¹², a voxelwise ratio between low-frequency power (i.e., 0.01–0.1 Hz) and the broader frequency spectrum of resting-state activity (i.e., from 0–0.25 Hz). The fALFF reported for an ROI is the first eigenvariate of the fALFF scores from all voxels within that region. fALFF has emerged as a functional measure of local signal strength of connections within neural networks, providing a measure of integrity (i.e., health) of individual nodal points within a network. Regional homogeneity (REHO), on the other hand, provides a measure of local coherence in the brain, calculated as the cross-correlation between each voxel and its neighbors, reflecting coherence within an ROI.¹³ When applied to analysis of the spontaneous low-frequency fluctuations observed during the resting-state, REHO is argued to represent local brain network integrity.¹⁴ Changes in fALFF and REHO within SLN structures can reliably discriminate patients with neurodegenerative disease from normal controls.^1,2

A forward linear regression analysis used baseline resting-state scores (REHO and fALFF) within ROIs to predict percentage change on behavioral scales at the end of the two-month interval. Prior to the forward regression, baseline FBI total scores were entered as a first step in the model to control for variation in initial symptom severity. A predictor variable was selected for inclusion in the model if it improved the model fit at a significance level of p<0.05. Post-hoc linear regression was performed to quantify the extent to which baseline resting-state activity predicted behavioral change in patients with FTD. Analyses were performed using IBM SPSS Statistics 20 (IBM Corporation, NY).

Results

Table 1 lists demographic and clinical characteristics of the participants; all had mild-to-moderate severity dementia. The groups with bvFTD and SD did not differ in clinical or demographic measures. Changes in participants’ FBI scores were heterogeneous across the 8-week study interval, without discernible improvement or degradation (Table 2). Resting-state measures within the left insula differentiated controls from participants with SD and bvFTD (Figure 1).

Table 1:

Demographic and clinical characteristics of study participants

Variable	bvFTD	SD	P-value
Age (mean ± std. dev)	62.6 (±4.2)	59.1 (±8.9)	0.42a
Male, n (%)	2 (40%)	6 (60%)	0.44b
Education	16.6 (±2.2)	16.6 (±2.8)	1.0a
Duration of Illness	5.4 (±4.2)	3.9 (±2.2)	0.37a
Clinical Dementia Rating Sum of Boxes Score (mean ± std. dev)	1.8 (±1.3)	1.3 (±0.7)	0.29a
Frontal Behavioral Inventory (total score):
baseline (mean ± std. dev)	29.6 (±16.9)	25.0 (±9.9)	0.51a
8 weeks (mean ± std. dev)	29.8 (±15.8)	23.6 (±9.1)	0.35a
% change (mean ± std. dev)	3.8 (±22.9)	−3.6 (±6.5)	0.34a

^aTwo independent sample t-test, assuming equal variance

^bChi-Square test, df=1

Table 2:

Baseline left insula fALFF, behavioral measures and change at 8 weeks for individual patients.

Patient	FBI-Total		FBI-Apathy		FBI-Disinhibition
	Baseline	Change (%)	Baseline	Change (%)	Baseline	Change (%)
bvFTD
003	32	−2 (−6.3)	25	0 (0)	7	−2 (−28.6)
004	49	1 (2.0)	31	2 (6.4)	18	−1 (−5.5)
006	42	−4 (−9.5)	23	−4 (−17.4)	19	0 (0)
018	9	−1 (−11.1)	9	−1 (−11.1)	0	0 (0)
026	16	7 (4.4)	11	7 (63.6)	5	0 (0)
SD
002	33	−4 (−12.1)	27	−2 (−7.4)	6	−2 (−33.3)
005	14	−1 (−7.1)	8	0 (0)	6	−1 (−16.7)
007	33	−3 (−9.1)	18	0 (0)	15	−3 (−20.0)
011	17	0 (0)	15	0 (0)	2	0 (0)
015	26	1 (4.3)	15	3 (20.0)	11	−5 (−45.5)
019	26	1 (3.8)	14	0 (0)	12	1 (8.3)
020	41	−1 (−2.4)	20	−1 (−5.0)	21	0 (0)
021	23	0 (0)	11	0 (0)	12	0 (0)
022	8	0 (0)	5	0 (0)	3	0 (0)
028	29	−4 (−13.8)	16	−6 (−37.5)	13	2 (15.4)

Figure 1:

Left Insula fALFF Scores and Symptom Change. Top left: bar graph demonstrating the identification of the left insula fALFF through the contrast of participants with FTD < Healthy Controls. Top right: scatter plot showing participant FBI changes over 8 weeks and fALFF measures. The line represents regression for the entire group (n=15). FBI change scores are residualized controlling for baseline FBI scores. Bottom left: scatter plot showing FBI-Apathy changes over 8 weeks and fALFF measures. Bottom right: scatter plot showing FBI-Disinhibition changes and fALFF measures.

Forward linear regression for the entire sample revealed a predictive relationship between low-frequency fluctuations in the left insula (i.e, fALFF) and changes in behaviors captured with the FBI total scores (std. beta=0.51, p=0.049; eTable 2). Higher left insular fALFF activity predicted an interval worsening (increase) in FBI scores, accounting for 28% of the change variance (Figure 1). The trend appeared to be driven by changes in the FBI apathy sub-scores. Left insula fALFF predicted increases in apathy sub-scores (std. beta=0.66, p=0.006). Although right insula resting-state measures did not independently predict changes in overall FBI scores, after controlling for left fALFF, right fALFF measures did improve predictions of changes in apathy sub-scores: higher right insula fALFF identified those least likely to experience increases in apathy sub-scores (std. beta=−0.49, p=0.034). Neither insular REHO nor resting-state measures within the anterior cingulate cortex accounted for changes in behavior. Additionally, no correlation was observed between baseline resting-state measures and duration of illness, global Clinical Dementia Rating scores, or the magnitude of FBI scores (eTable 3).

To confirm the generalization of these finding to FTD subtypes, we repeated the regression analysis separately for patients with bvFTD and SD. The results were more robust for the bvFTD group: Baseline fluctuations in low-frequency resting-state activity in the left insula strongly predicted increases in the FBI (std. beta=1.017, p=0.028, R²=0.80), especially increases in apathy sub-scores (std. beta=1.072, p=0.041, R²=0.85; eTable 4). A similar correlation with left fALFF activity was observed in the SD group (std. beta=0.61, p=0.040, R²=0.37; FBI-apathy measures: std. beta=0.724, p=0.019. R²=0.52; eTable 5). Correlations were confirmed with parametric and non-parametric statistical measures (eTable 6), suggesting that the described relationship was not driven by outliers.

Discussion

Patterns of connectivity within and between SLN structures reliably distinguish patients with FTD from healthy controls¹ and from patients with Alzheimer’s disease.² The results of this study corroborate the importance of the insula within the SLN, suggesting that baseline measures of SLN connectivity involving the left insula may predict changes in behavior in patients with FTD, as measured with the FBI. Measures of low-frequency signal within the left insula did not serve as an indicator of disease severity, as no association was observed between baseline measures of resting-state activity and clinical features/measures of disease severity at the time of entry into the trial.

Rapid behavioral change early in the course of FTD is a well-described, yet poorly understood clinical phenomenon.¹⁵ This effect is not seen in patients with Alzheimer’s disease,¹⁵ raising the possibility that accelerated functional decline in FTD may be explained by a model of neurodegenerative disease that emphasizes breakdown of network connectivity, preceding structural changes detected on standard neuroimaging. The pattern of breakdown is presumed to be distinct from that seen in Alzheimer’s disease, accounting for differences in clinical progression, and facilitating differentiation between FTD and Alzheimer’s disease with resting-state measures.^2,3 In line with this, the increased tonic signalling measured within the left insula of study participants may reflect a compensatory response resulting from loss of regional connections. Similar increases are recorded in the mean firing rate of neurons within the subthalamic nuclei of patients with Parkinson’s disease undergoing deep brain electrode-implantation,¹⁶ suggesting that hyperactive neuronal discharges in the subthalamic nucleus associate with motor dysfunction. The magnitude of resting-state activity measured within the SLN of our participants with FTD was less than that measured in healthy controls, emphasizing the importance of interpreting resting-state measures relative to other patients with FTD. In our study, a relative increase in low-frequency fluctuations was predictive of behavioral worsening over as short a period as the next 8 weeks in participants with FTD. Within this population, measures of low-frequency fluctuations within the left insula may provide a marker of a dysregulated network at greatest risk of collapse. An alternate explanation for the observed correlations may be that higher left insula resting-state activity selected for participants with a less advanced clinical stage of dementia, identifying those with relatively preserved behavioral functions, and thus, the greatest potential for change (i.e., the most to lose). No correlation was found, however, between resting-state activity and clinical measures approximating disease severity, favoring the assertion that resting-state measures may predict behavioral change independent of assessable clinical measures. Resting-state measures may provide a functional neuroimaging correlate for the precipitous behavioral decline detailed in FTD,¹⁵ permitting evaluation of the role of SLN dysfunction in this process.

Anterior insular atrophy is reported early in the course of patients with bvFTD, with the extent of right hemisphere involvement exceeding that on the left on structural neuroimaging.¹⁷ Correspondingly, low-frequency fluctuations are reduced in the right insula of patients with bvFTD and SD, compared to controls.¹ Extending these findings, higher right insula fALFF appeared protective against worsening of apathy in FTD (after controlling for left fALFF activity). Relative preservation of right insula low-frequency fluctuations may be protective against behavioral decline. The contributions of dysfunction within right and left SLNs to the FTD phenotype is deserving of further study.

The relatively small sample size reported in this study likely limited the ability to draw correlations between resting-state measures and behavior. No significant changes in behavioural measures were reported across the two month study period for the total sample, the bvFTD group or the SD group, indicating that a longer study with a larger sample might be more informative. Additionally, all participants received memantine hydrochloride. It is unlikely, however, that open-label use of memantine significantly altered behavior, given the published randomized control trial that demonstrated no effect in participants with FTD,¹⁸ as well as our own finding that FBI scores did not improve over the open label trial study period.⁸ Future studies could control for medication use, and include more resting-state measurements over the illness course. Two months may be too short a time to detect clinically significant changes in network connectivity.

Limitations notwithstanding, the results of this analysis expand upon prior studies. Resting-state measures of neural connectivity may provide a non-invasive means of assessing network functioning in neurodegenerative disease.