Long-Known Music Exposure Effects on Brain Imaging and Cognition in Early-Stage Cognitive Decline: A Pilot Study

Abstract

Background:

Repeated exposure to long-known music has been shown to have a beneficial effect on cognitive performance in patients with AD. However, the brain mechanisms underlying improvement in cognitive performance are not yet clear.

Objective:

In this pilot study we propose to examine the effect of repeated long-known music exposure on imaging indices and corresponding changes in cognitive function in patients with early-stage cognitive decline.

Methods:

Participants with early-stage cognitive decline were assigned to three weeks of daily long-known music listening, lasting one hour in duration. A cognitive battery was administered, and brain activity was measured before and after intervention. Paired-measures tests evaluated the longitudinal changes in brain structure, function, and cognition associated with the intervention.

Results:

Fourteen participants completed the music-based intervention, including 6 musicians and 8 non-musicians. Post-baseline there was a reduction in brain activity in key nodes of a music-related network, including the bilateral basal ganglia and right inferior frontal gyrus, and declines in fronto-temporal functional connectivity and radial diffusivity of dorsal white matter. Musician status also significantly modified longitudinal changes in functional and structural brain measures. There was also a significant improvement in the memory subdomain of the Montreal Cognitive Assessment.

Conclusion:

These preliminary results suggest that neuroplastic mechanisms may mediate improvements in cognitive functioning associated with exposure to long-known music listening and that these mechanisms may be different in musicians compared to non-musicians.

Keywords

Alzheimer’s disease cognitive reserve functional MRI imaging MRI music

INTRODUCTION

Alzheimer’s disease (AD) is a devastating neurological condition associated with declines in cognitive performance and functional abilities. Unfortunately, existing treatments have provided limited benefits to date [1]. As a result, there has been a shift to non-pharmacological interventions. One such approach involves exposure to autobiographically salient music. Music exposure has been shown to have numerous beneficial effects, mediated by elevated brain dopamine levels, neurogenesis, and synaptic plasticity [2, 3], including improved mood and anxiety symptoms [4] and increased arousal [5]. Effects on cognitive performance, however, are unclear, with some studies showing increased performance and other studies showing more limited effects [6]. According to a review by Fang et al. [6], exposure to music-based interventions has been shown to improve cognitive functioning in a number of domains, including autobiographical and episodic memory, psychomotor processing speed, executive dysfunction, and global memory, although properly designed randomized controlled trials are lacking. A systematic review conducted by our group further suggests that music-based interventions may be helpful for patients with cognitive impairment in terms of enhancing cognition and addressing behavioral symptoms, specifically individualized approaches that use playlists [7].

Exposure to long-known or autobiographically salient music may provide benefits above and beyond those associated with music listening in general. Long-known music often evokes an emotional response, thereby reinforcing autobiographically salient memories which have been shown to deteriorate with the progression of AD [8]. Prior clinical studies have demonstrated improvement in autobiographical memory associated with exposure to autobiographically salient music [4, 9] as well as improvement in cognitive test performance [10]. There is also some evidence that musicianship modulates the response to autobiographically salient memories, although this is an area that has received scant attention to date. In a recent study comparing non-musicians to professional musicians, sung words were used as a mnemonic device, with musicians showing better learning of sung information relative to non-musicians [11]. Moreover, some studies have suggested that musicianship may be associated with preservation of implicit musical memory, such as the ability to play a musical instrument, while explicit musical memory relating to recall of tunes may be more susceptible to deterioration [12]. Prior studies have also suggested that professional musicians may have enhanced baseline cognitive abilities relative to non-musicians on a composite score of cognitive control [13].

While studies have examined the effects of repeated exposure to long-known music on cognitive functioning in patients with AD, few studies to date have identified the underlying neural changes associated with repeated exposure to long-known music. Previous work by our group looking at baseline activation in this same cohort of patients demonstrated that exposure to long-known music relative to newly-heard music (newly-composed and heard for the first time 60 min before scanning) resulted in more widespread activation of frontal, temporal, parietal, and subcortical brain regions [14]. In a similar study, 17 subjects with AD were assigned a personalized music playlist. Post-training subjects demonstrated increased activation of the supplementary motor area as well as increased corticocortical and corticocerebellar network functional connectivity [15].

The purpose of our study is to examine whether a program involving listening to long-known music for 1 h a day for 3 weeks results in changes in brain structure and function and corresponding improvements in global memory performance for patients with mild cognitive impairment (MCI) or early AD. We will identify the brain networks associated with long-known and recently-heard music listening, and examine whether the music listening protocol leads to consistent longitudinal changes in the functional organization of the network, including task-related brain activity and resting-state functional connectivity. Furthermore, we will examine whether there are also longitudinal changes in white matter microstructure, indicating corresponding changes in the neuroanatomical substrate. Secondary analyses conducted at each stage will examine the influence of musicianship on the observed changes. This information will be critical to understanding the neural mechanisms through which repeated exposure to long-known music modulates cognitive performance.

METHODS

Data availability statement

Please note that raw data and pipelines required for processing of imaging data will be available upon request.

Participants

Participants were recruited from the St. Michael’s Hospital Memory Disorders Clinic in Toronto, Canada. In order to qualify for the study, participants were required to meet clinical criteria for MCI or AD in accordance with the National Institute on Aging and Alzheimer’s Association (NIA-AA) diagnostic criteria [16]. We chose to focus on patients with MCI/AD given it is difficult to distinguish between these groups clinically, to optimize recruitment and to ensure they have sufficient cognitive reserve to benefit from a music-based intervention in alignment with prior studies [7]. They were also required to be fluent in English, with a Mini-Mental Status exam (MMSE) [17] score above 19 and an available caregiver. Participants with a history of significant medical, neurologic, or psychiatric illness were excluded. Participant assessments included a comprehensive history, objective cognitive testing, and neuroimaging. For the purpose of our study, “musician” was defined as someone who plays music professionally and/or has received formal musical training (minimum bachelor’s education or Royal Conservatory of Music training). All experimental procedures were approved by the St. Michael’s Hospital Research Ethics Board and all participants provided written informed consent.

Experimental design

Pre-experiment phone interview

Participants were contacted by phone prior to the first experimental in-person session to identify musical preferences. Long-known music was defined for the purpose of the study as instrumental or vocal music recognized by the participant as his/her preferred music that was known to them for at least 20 years and that held special meaning for them (e.g., the music they danced to at their wedding). This approach has been validated in previous studies [18, 19]. A musical playlist was compiled based on participant responses and, during a subsequent follow-up, the participant was asked to listen to songs from the playlist in order to verify that the music held special meaning. The music was purchased and downloaded from the iTunes music store. Recorded versions of music were selected over live versions. Factors that influenced the selection of music included meaningfulness, quality of recording and length of the piece.

Music listening interview

Before scanning, each participant met with a research assistant, to verify that the music on the playlist was known by the participant and to expose them to a novel musical composition. At this time, music clips were played in random order. The participant was prompted to identify the music as long-known or unknown and responses were recorded. Unknown music is hereafter referred to as “recently-heard” music.

Music listening intervention

Playlists were loaded onto an MP3-player that was given to the participant for the duration of the study. Participants were asked to listen to their playlist for approximately 1 h per day, at approximately the same time of day for a period of 3 weeks. We selected a period of 3 weeks as we felt this time frame was supported by the literature [7], was sufficiently long for a pilot study to incur benefit yet sufficiently short to maximize feasibility. During music listening, they were instructed to focus on the music and not perform any other tasks concurrently. Caregivers were encouraged to participate in the listening sessions and to ask questions such as “what does this music remind you of” during the music listening session. Caregivers were asked to keep a log of participation and a research assistant from the Faculty of Music at the University of Toronto called several times a week to remind the subject and to troubleshoot any issues with the MP3 player.

Clinical outcome measures

Participants completed a demographics questionnaire in addition to a variety of cognitive measures. The Montreal Cognitive Assessment (MoCA) [20] was administered prior to and after completion of the listening period to all participants as the primary clinical outcome measure. Alternate versions were used to minimize practice effects. The MoCA is a screening tool that identifies mild cognitive complaints [20]. It includes eight cognitive domains, including visuospatial and executive function, naming, attention, language, abstraction, memory delayed recall, and orientation. The MoCA is previously established to have high validity, specificity and sensitivity [21] in differentiating individuals with MCI and AD from those with normal cognition.

Functional MRI (fMRI) task design

Participants underwent fMRI before initiating and after completing the music intervention. Prior to the MRI all participants underwent a brief training session. Each MRI scan consisted of a structural scan and task-based fMRI. The task-based MRI consisted of participants listening to music clips that were 20 s in duration and consisted of 12 long-known clips and 12 recently-heard clips transmitted into the MRI machine that were matched in style and orchestration to the long-known music and first heard 60 min prior to scanning. The matching was completed by a research assistant based at the Faculty of Music, University of Toronto trained in music-related research. Recently-heard and long-known music clips were alternated and play to the participant with 16 s of Gaussian white noise in between. The fMRI protocol was created on E-Prime software (version 2.0). Participants were instructed to reflect on the music while listening to music in the MRI and try to recall if they had heard it previously. Participants were instructed to try not to move their heads, or other parts of their body (such as finger tapping, toe tapping, humming, or mouthing the words), which can result in additional motor activation, and keep their eyes closed, as is standard with MRI procedures. Following the 3-week music intervention, all MRI sequences were repeated to ascertain changes in brain function and structure.

Magnetic resonance imaging

All MRI scanning was conducted at 3.0 Tesla field strength on research-dedicated scanner with 32-channel head coil (Siemens Magnetom Syngo Skyra). As part of the imaging protocol, a T1-weighted magnetization prepared rapid gradient-echo (MPRAGE) scan was acquired to measure neuroanatomy at 1 mm isotropic voxel resolution (scan parameters: echo time (TE) = 2.54 ms, repetition time (TR) = 2000 ms, 176 sagittal slices, 1.0 mm slice with 0 mm gap, field of view (FOV) = 256 mm). Blood-oxygenation-level dependent functional MRI (BOLD fMRI) was subsequently acquired during a music listening task and during a 6 min resting-state protocol. In addition, a diffusion tensor imaging (DTI) protocol was used to assess white matter microstructure. Acquisition and preprocessing details are provided below.

Functional MRI

The task and resting-state fMRI data were acquired via multi-slice T2^* echo planar imaging sequence (scan parameters: TE/TR = 30/2000 ms, θ=70°, 32 oblique-axial slices with 200×200 mm FOV, 64×64 matrix, 4.0 mm slice with 0.5 mm gap, 3.125×3.125 mm in-plane, 2298 Hz/px BW). Data processing was performed as described in prior publications [22 –24], with key details summarized below. Processing software included Analysis of Functional Neuroimages (AFNI; afni.nimh.nih.gov), the FMRIB Software Library (FSL; https://fsl.fmrib.ox.ac.uk) and customized software developed in-house. For each fMRI sequence, the first four images were discarded and AFNI 3dvolreg used to correct for rigid-body head movement. After, high-variance outlier volumes were detected and replaced with interpolated values using SPIKECOR (nitrc.org/projects/spikecor) and AFNI 3dTshift used to correct for slice-timing differences. AFNI 3dmerge was then used to spatially smooth images by convolving with a 6 mm Full Width at Half Maximum (FWHM) 3D Gaussian kernel (isotropic). Regression of nuisance covariates then included the removal of linear+quadratic trends and variance associated with the 6 rigid-body motion parameter timeseries obtained from 3dvolreg. To control for physiological noise, PHYCAA+ (nitrc.org/projects/phycaa_plus) was used to down-weight voxels with high non-neural signal content, followed by removal of signal from white matter (WM) and cerebrospinal fluid (CSF); the latter steps were performed after spatial normalization, as described below.

To conduct group-level analyses of task and resting data, all fMRI datasets were co-registered to the standard Montreal Neurological Institute 152 (MNI152) atlas. For each fMRI sequence, FSL flirt computed the 6-parameter alignment of the subject’s mean fMRI image to their T1 image, and the 12-parameter alignment of their T1 image to the template. Then, convert_xfm computed the net transform from fMRI to template space, with was then applied, combined with resampling at 3 mm isotropic voxel resolution. After, WM and CSF signals were regressed out. This was achieved by running the FSL fslvbm protocols on subject T1 images, producing cohort-specific maps of grey matter, WM, and CSF in the MNI template space, which were subsequently downsamples to 3 mm isotropic voxel resolution and smoothed with a 6 mm FWHM Gaussian kernel (isotropic). For both WM and CSF maps, a mask was made of the upper 95th percentile of voxels, and BOLD signal was averaged separately within cortical and brain stem voxels (WM) and within left and right ventricle voxels (CSF), producing 4 seed timeseries in total. These timeseries were then regressed from the fMRI dataset.

Diffusion tensor imaging

A DTI protocol was performed consisting of 30 diffusion-weighting directions and nine b0 scans (scan parameters: TE/TR = 83/7800 ms, b = 700 s/mm², 66 axial slices with FOV = 240×240 mm, 120×120 matrix, 2.0 mm thickness, 2.0×2.0 in-plane resolution, BW = 1736 Hz/Px). Data processing software included FSL utilities and custom software. The FSL eddy protocol corrected for the effects of eddy currents and rigid-body head motion and FSL bet masked out non-brain voxel. Afterwards, FSL dtifit calculated voxel-wise measures summarizing local water diffusion properties. This included indices of fractional anisotropy (FA), axial diffusivity (AD), and radial diffusivity (RD). Afterwards, to conduct group-level analyses, the DTI maps were co-registered to a common template using the Diffusion Tensor Imaging ToolKit (DTI-TK; http://dti-tk.sourceforge.net/). To initialize alignment, IXI Aging DTI Template 3.0 was used as a reference, followed by construction of a cohort-specific template. The dti_template_bootstrap script was then used to generate a crude initial “bootstrap” estimate, the dti_affine_population script updated this template via affine alignment, and afterwards the dti_diffeomorphic_population script further refined the template via diffeomorphic alignment. To facilitate interpretation of results, the cohort template was then transformed into MNI space using the the IIT Human Brain Atlas’ mean tensor template as a reference. Sequential applications of dti_rigid_reg, dti_affine_reg and dti_diffeomorphic_reg then obtained increasingly refined alignments to this template. The DTI parameter maps were then warped into MNI space by computing the net diffeomorphic transform with dfRightComposeAffine and then applying the transform with deformationScalarVolume. Aligned images were also resampled into 3 mm isotropic voxel resolution, with a 6 mm FWHM 3D Gaussian smoothing kernel (isotrpic) applied afterwards. Analysis was restricted to a mask of white matter voxelswhere FA > 0.30 in the group template, with manual segmentation and exclusion of brain stem areas.

Statistical analyses

We have used effect sizes and confidence intervals to evaluate goodness of fit. We also ran standard visual checks of the residual plots and used non-parametric bootstrap analyses which makes minimal modelling assumptions.

Task-based analysis

For task fMRI data, subject-level analyses were conducted using a general linear model (GLM) in which the conditions of unfamiliar music listening (UFAM), familiar music listening (FAM), and fixation (fix) were modelled using a binary design matrix convolved with a standard double-gamma “SPMG1” basis function (afni.nimh.nih.gov/pub/dist/doc/program_help/3dDeconvolve.html). We obtained beta maps for all possible contrasts UFAM-fix, FAM-fix, and FAM-UFAM, analyzed in the NPAIRS split-half framework [25] to produce reproducible z-scored statistical parametric maps (rSPMZs). Preliminary group-level analyses were then conducted on the full set of pre-intervention rSPMZs. Voxel-wise t-tests were performed to obtain group activation maps for each contrast, with voxel-wise thresholding at p < 0.005, followed by cluster-size thresholding at an adjusted p = 0.05, using AFNI 3dFWHMx to estimate spatial smoothness of maps and 3dClustSim to obtain the corresponding cluster size threshold.

Along with the thresholded maps, we defined nodes within a “musical network” as follows: the full list of clusters was retained from the three group-level activation maps (UFAM-fix, FAM-fix, FAM-UFAM) and any overlapping clusters between activation maps were merged by taking their intersection. For each cluster in the new set, a spherical region of interest (ROI) was created by setting a value of 1 at the cluster center of mass and smoothing with a 3D isotropic Gaussian kernel. Kernel full width at half maximum (FWHM) was chosen to match average cluster size (in accordance with the matched filter theorem), by calculating average cluster volume and the diameter for a sphere of equivalent volume, resulting in an FWHM of 17.97 mm. Clusters were localized using the BNA atlas (http://atlas.brainnetome.org) to identify cortical regions and the SUIT atlas (http://www.diedrichsenlab.org/imaging/propatlas.htm) to identify cerebellar domains. These atlases were also used to verify that individual clusters did not span multiple lobes; for the FAM-UFAM contrast only, we manually divided a single subcortical cluster into left and right hemispheres.

The pre-intervention activation t-statistics were reported for each ROI of the musical network. For all subjects with both pre- and post-intervention imaging, we then calculated the paired-measures t-statistics on longitudinal change for the three task contrasts, along with the corresponding p-values. Significant longitudinal effects were identified at a False Discovery Rate (FDR) threshold of 0.05, applied to each task contrast. Secondary analyses examined the impact of musicianship on longitudinal change, by performing GLM analysis of the effect of musician status (binary variable, where 0 = non-musician, 1 = musician) and testing for significant longitudinal effects at an FDR of 0.05, applied successively to each task contrast. For ROIs showing significant longitudinal effect, the mean activation value was also plotted from pre- to post-intervention, for the whole group, and plotted separately for non-musicians and musicians. Bootstrap resampling (1000 iterations) was used to generate empirical distributions on the means, and on the longitudinal and cross-sectional differences in ROI activation z-scores.

Resting-state analysis

For each participant and imaging session, the mean resting-state BOLD time course was obtained for all musical network ROIs. The mean pre-intervention functional connectivity pattern shown for the whole group. For each pairwise functional connection, we then measured the longitudinal change in connectivity using a paired-measures bootstrap resampling approach, with reporting of the bootstrap ratio (BSR; a z-distributed measure of standardized effect based on the ratio of mean / standard error) and empirical p-value. Significant connections were identified at an FDR of 0.05. Secondary analyses examined the impact of musicianship on longitudinal change, by performing bootstrapped GLM analysis of the effect of musician status and testing for significant longitudinal effects at an FDR of 0.05. For all significant ROI pairs, the mean connectivity value was also plotted pre- and post-intervention, for the whole group, and also plotted separately for non-musicians and musicians. Bootstrap resampling (1000 iterations) was used to generate empirical distributions on the means, and on the longitudinal and cross-sectional differences in connectivity values.

White matter microstructure analysis

For each of the white matter parameters, including FA, AD, and RD, we measured voxel-wise longitudinal changes in DTI parameters using a paired-measures bootstrap resampling approach, with reporting of the BSR and empirical p-value. Significant clusters were identified as in the task-based fMRI analyses above. Secondary analyses examined the impact of musicianship on longitudinal change, by performing bootstrapped GLM analysis of the effect of musician status and similarly performing cluster-size thresholding. For ROIs showing significant longitudinal effect, the mean DTI parameter value was also plotted from pre- to post-intervention, for the whole group, and plotted separately for non-musicians and musicians. Bootstrap resampling (1000 iterations) was used to generate empirical distributions on the means, and on the longitudinal and cross-sectional differences in DTI parameter values.

RESULTS

Behavioral data

Fourteen of 17 subjects completed follow up assessments and the home-based music intervention (mean age = 73.1; mean years of education 15.2; Female, n = 11; mean MMSE 27.3, mean MoCA 23.2) as reported in Table 1. Of the sample that completed the experimental procedures, 6 were musicians and 8 were non-musicians. Mean and standard deviations (SD) were computed for all continuous variables. Categorical variables were assessed for significance using the chi-square (χ²) test of independence (see Table 1). According to review of caregiver logs, all participants completed the home-based music intervention and there was no difference in completion rates between musicians and non-musicians. The range of musical styles featured in the personalized playlists included pop, jazz, classical (both instrumental and vocal), folk, soul, country, songs from musicals, and international music (Indian, Italian, French). In general, musicians selected more classical and jazz music compared to non-musicians. Across both groups, music that contained vocals were selected more than purely instrumental music. The novel music was newly composed and can be described as modern classical music with abstract elements that was built from traditional Western scales and followed common rhythmic principles. There were two sets of novel music, instrumental and vocal, and we matched these accordingly during the task-based scan. See the Supplementary Material for a full list of songs selected.

Table 1

Demographic characteristics, baseline cognitive presentation participants participating in intervention (n = 14)

Age, y	73.1±9.3 (56–88)
Sex, n (%)	11 females (79%)
	3 males (21%)
Education, y	15.2±2.5 (12–20)
Bilingual, n (%)	6 (43%)
Depression, n (%)	0 uncontrolled (0%)
	2 controlled (14%)
	12 none (86%)
Average time listening to music (min)	72.9±14.4 (60–105)
MMSE total score	27.3±1.3 (25–29)
MoCA total score (pre-MBI)	23.2±2.1 (19–25)

A comparison of MoCA subscores pre- and post-intervention is reported in Table 2. Performance on visuospatial/executive function, naming, attention, language, abstraction, orientation, and total score did not differ significantly post-intervention compared to baseline. A Wilcoxon Signed-Rank test showed a significant change in the memory subscore of the MoCA post-intervention (p = 0.034, uncorrected). When participants were separated into non-musician and musician subgroups, the only significant effect was seen for the memory sub-scores of non-musicians (p = 0.025, uncorrected).

Table 2

Comparison of Pre- and Post-intervention MoCA scores in all participants

	Baseline MoCA score (n = 14)	Post MBI MoCA score (n = 14)	p
Visuospatial/Executive function (/5)	3.3±0.91 (2–5)	3.7±0.61 (2–4)	0.130
Naming (/3)	2.6±0.50 (2–3)	2.8±0.43 (2–3)	0.317
Attention (/6)	5.6±0.50 (5–6)	5.8±0.43 (5–6)	0.414
Language (/2)	2.2±0.70 (1–3)	1.9±0.95 (0–3)	0.132
Abstraction (/2)	1.6±0.65 (0–2)	1.6±0.50 (1–2)	0.739
Memory (/5)	2.0±1.9 (0–5)	2.4±1.20 (0–5)	0.034 ^*
Orientation (/6)	5.9±0.36 (5–6)	5.9±0.36 (5–6)	1.000
Total Score (/30)	23.4±2.06 (20–25)	24.2±2.64 (20–27)	0.100

Task-based analysis

Pre-intervention group-level analysis results are shown in Fig. 1. For the UFAM-fix contrast, increased activity was seen most prominently in the bilateral temporal lobes, with a smaller cluster spanning right precentral and postcentral gyri. Decreases in activation were also seen prefrontally in bilateral middle frontal and anterior cingulate areas. For the FAM-fix contrast, increased activity was again seen temporally, along with clusters bilaterally within the cerebellum, globus pallidus (GP) of the basal ganglia, inferior frontal gyri (IFG) and superior frontal gyri. Decreases in activity were also seen in lateral occipital lobes. For the FAM-UFAM contrast, increases in activity were limited to the GP and superior frontal gyri (along with activity decreases in the lateral occipital lobes), while more extensive cerebellar activations were uniquely detected for this contrast. The 12 musical network ROIs associated with task-positive activation are shown in Fig. 1D and summarized in Table 3.

Fig. 1

Pre-intervention group-level activation maps, obtained via 1-sample t-statistics, with cluster size adjustment for multiple comparisons. Maps are shown for task contrasts, comparing conditions of unfamiliar music (UFAM), familiar music (FAM) and fixation (fix). The bottom panel also shows the 12 distinct musical network ROIs that were identified from at least one of the three task contrasts. Axial slices are labelled with MNI space coordinates below.

Table 3

Significant clusters, with center of mass brain region based on either the BNA atlas (cortical and subcortical areas) or the SUIT atlas (cerebellar areas). Center of mass is also given in MNI coordinates. The ROI t-statistics are also given for each of the three task contrasts, comparing unfamiliar music (UFAM), familiar music (FAM) and fixation (fix). A ‘^*’ denotes clusters that were significant for each contrast after adjusting for multiple comparisons at an FDR of 0.05. GP, globus pallidus; A22c, caudal area 22; A22r, rostral area 22; A44op, opercular area 44; A45c, caudal area 45; A6cvl, caudal ventrolateral area 6; A8m, medial area 8

	Brain region	center of mass	UFAM-fix	FAM-fix	FAM-UFAM
1	Cerebellum R (VIIb)	21, –69, –51	2.54	3.38^*	1.86
2	Cerebellum L (VI)	–27, –60, –30	1.90	3.69^*	2.92^*
3	Cerebellum R (Crus-I)	33, –63, –30	0.66	2.77	4.39^*
4	Cerebellum (I–IV)	0, –48, –15	0.07	2.90	3.60^*
5	Basal ganglia L (GP)	–21, –6, 3	1.20	4.93^*	5.32^*
6	Basal ganglia R (GP)	21, 0, 3	2.68	6.35^*	4.73^*
7	Superior temporal gyrus L (A22c)	–57, –21, –3	6.22^*	5.70^*	1.92
8	Superior temporal gyrus R (A22r)	57, –21, 0	6.36^*	5.46^*	1.89
9	Inferior frontal gyrus L (A44op)	–39, 27, 3	2.02	4.82^*	3.06
10	Inferior frontal gyrus R (A45c)	51, 30, 12	4.60	6.78^*	2.19
11	Precentral gyrus R (A6cvl)	48, 3, 45	3.45	6.06^*	3.55
12	Superior frontal gyrus (A8m)	0, 6, 63	1.38	4.66^*	5.10^*

Analyses of the longitudinal change in network ROI activity are shown in Fig. 2, for each of the three task contrasts. Significant longitudinal changes were seen in a subset of ROIs (Fig. 2A), including bilateral GP and the right IFG. In these regions, negative t-statistics indicate reduced activation post-intervention. Secondary analyses of musician status identified significant effects for the GP alone (Fig. 2B), with musicians having a positive t-statistic, indicating a diminished effect of time on brain activity for this subgroup. This is clarified in Fig. 2C-E, where the mean change plots are given for GP and right IFG. For left and right GP, non-musician activation z-scores decline significantly over time (mean, 95%CI and p-value of the left: –0.84, [–1.06, –0.60], p < 0.001; right: –1.07 [–1.31, –0.83], p < 0.001) whereas musician values do not change significantly (left: 0.08, [–0.14, 0.32], p = 0.498; right: –0.15, [–0.51, 0.22], p = 0.430). Pre-intervention, musicians have significantly lower activity levels than non-musicians (left: –0.60, [–0.10, –1.08], p = 0.012; right: –0.68, [–0.10, –1.13], p = 0.014), but post-intervention, their activity levels do not differ (left: 0.32 [–0.18, 0.89], p = 0.228; right: 0.24 [–0.32, 0.96], p = 0.491). For the right IFG, by contrast, the overall longitudinal change is (–0.97, [–1.65, –0.36], p = 0.001) and musicians are comparable to non-musicians pre-intervention (–0.22, [–1.44, 1.21], p = 0.697) and post-intervention (1.02 [–0.11, 2.21], p = 0.070). These results indicate that there are spatially sparse but significant decreases in musical network activity over time, with effects that are predominantly driven by non-musicians.

Fig. 2

Longitudinal change in activity of musical network ROIs, corresponding to clusters listed in Table 3. A) t-statistics of longitudinal change, and B) t-statistics of effect of musician status on longitudinal change. In both plots, ‘^*’ denotes significant effects at an FDR of 0.05. C-E) Plots of mean activation pre-intervention (PRE) to post-intervention (POST) for significant ROIs, with error bars denoting standard error of the mean.

Resting-state analysis

The results of functional connectivity analyses are depicted in Fig. 3. As shown in Fig. 3A, the pre-intervention network had relatively high connectivity bilaterally between homologous areas (0.6 < ρ < 0.8), but otherwise relatively lower inter-regional connectivity values (0.1 < ρ < 0.4). As depicted in Fig. 3B, significant longitudinal change was limited to the inter-hemispheric connection between right superior temporal gyrus and left IFG. By contrast, as shown in Fig. 3C, musician status affected longitudinal connectivity changes for intra-hemispheric connections of left GP to IFG, and right GP to IFG. The effects are clarified in Fig. 3D-F, where mean change plots are shown. For the inter-hemispheric connection, overall longitudinal change in connectivity values is significant (–0.115, [–0.200, –0.032], p = 0.004) and musicians are comparable to non-musicians pre-intervention (0.015, [–0.156, 0.179], p = 0.867) and post-intervention (–0.053, [–0.258, 0.134], p = 0.602). For the intra-hemispheric connections, however, musician connectivity values decline significantly over time (left: –0.136, [–0.197, –0.079], p < 0.001; right: –0.145, [–0.287, –0.008], p = 0.32), whereas non-musician values do not change significantly (left: 0.021, [–0.063, 0.117], p = 0.796; right: 0.091, [–0.009, 0.172], p = 0.069). The musicians have similar connectivity values as non-musicians pre-intervention (left: –0.005, [–0.148, 0.146], p = 0.968; right: 0.031, [–0.107, 0.162], 0.708) but lower values than non-musicians post-intervention (left: –0.162 [–0.009, –0.322], p = 0.031; right: –0.204 [–0.048, –0.362], p = 0.006).

Fig. 3

Longitudinal change in functional connectivity of musical network ROIs, corresponding to clusters listed in Table 3. A) Mean pre-intervention functional connectivity pattern of the musical network. B) Bootstrap ratios (BSRs) for connections with significant longitudinal change, and C) BSRs for connections with significant effect of musician status on longitudinal change. In both plots, significant connections are identified at an FDR of 0.05. C-E) plots of mean connectivity pre-intervention (PRE) to post-intervention (POST) for significant ROI connections, with error bars denoting standard error of the mean.

White matter microstructural analysis

For DTI analyses, group-level analyses found no significant effects for the FA and AD parameters. For RD, however, there were significant longitudinal changes and effects of musicianship, depicted in Fig. 4 with clusters summarized in Table 4. Effects of time were focal and limited to a negative effect within a single cluster in the right superior corona radiata. The whole group showed a longitudinal decline in diffusivity values (–1.01×10^–5, [–1.35, –0.67]×10^–5, p < 0.001), but there were no significant effects of musician status pre-intervention (3.12×10^–5, [–1.74, 7.06], p = 0.185) and post-intervention (2.47×10^–5, [–2.47, 6.60]×10^–5, p = 0.299). Effects of musicianship were much more spatially extensive but were again predominantly right-lateralized and seen in the superior corona radiata. In this group, different trajectories of change were seen, with non-musicians having a modest increase over time (0.98×10^–5, [0.47, 1.44]×10^–5, p < 0.001), while musicians had a larger decrease over time (–1.86×10^–5, [–2.52, –1.30]×10^–5, p < 0.001). However, the groups were not substantially different pre-intervention (0.54×10^–5, [–3.50, 4.86]×10^–5, p = 0.289) or post-intervention (–2.29×10^–5, [–6.49, 1.82]×10^–5, p = 0.853), indicating the inter-subject differences were relatively subtle.

Fig. 4

Effects of intervention on white matter microstructural parameter of radial diffusivity (RD), with clusters summarized in Table 4. A) Areas of significant longitudinal change, with effect sizes reported as bootstrap ratio (BSR) values, and B) plots of mean RD values pre-intervention (PRE) to post-intervention (POST) averaged over significant voxels, with error bars denoting standard error of the mean. C) Areas with a significant effect of musicianship on longitudinal change, and D) plots of mean RD values. Brain maps are shown as maximum intensity projections in the three Cartesian planes.

Table 4

Significant clusters, with center of mass location based on the JHU atlas. Center of mass is given in MNI coordinates. The ROI bootstrap ratio (BSR) values are also provided

		Brain region	center of mass (MNI)	cluster volume (mm³)	Peak value (BSR)
main	1	superior corona radiata R	27, 9, 15	1053	–5.09
music	1	superior corona radiata R	30, 12, 39	3456	–33.93
	2	anterior corona radiata R	15, 42, 6	1296	–23.68
	3	superior corona radiata R	12, 0, 57	972	–10.09
	4	superior corona radiata L	–18, 27, 39	864	–8.28
	5	superior corona radiata L	–15, 0, 60	783	–8.53
	6	internal capsule (ant. limb) R	18, 18, 9	756	–27.83

DISCUSSION

In the present study, we were able to design a home-based music intervention that participants found both useful and pleasant as indicated by their level of adherence to the study protocol. Fourteen of 17 participants initially recruited completed the music-based intervention and as confirmed by review of caregiver logs all participants were fully compliant with all study procedures, with no differences noted between musicians and non-musicians. This is in keeping with the literature, which suggests that music-based interventions are enjoyable and motivating [7]. Music in general has been linked to enhanced memory performance, irrespective of whether it is long-known or recently-heard. Potential mechanisms of memory enhancement include elevated brain dopamine levels [2, 3], improved mood and anxiety symptoms [4], increased arousal [5], neurogenesis, and synaptic plasticity. Long-known music has the added benefit of evoking autobiographical memories, potentially augmenting sense of self which has shown to deteriorate with the progression of AD [10].

As an additional key study outcome, we also characterized brain networks associated with music listening, and further identified consistent functional and structural correlates of the music-based intervention. Pre-intervention analyses of task-based fMRI identified robust musical networks distinguishing recently learned music listening from long-known music listening. In the former condition, activation was mainly limited to temporal regions associated with auditory processing. In the latter condition, activation was more extensive and included distributed cerebellar, subcortical, and frontal regions. These findings are broadly congruent with studies of autobiographical memory, where there is a tendency to identify frontal and temporal activations with right-lateralization of clusters [26]. More specifically, the medial temporal, lateral temporal, and cerebellar activations seen in this study are consistent with the “core” network identified in meta-analyses [27], although we did not observe the medial prefrontal, temporo-parietal and posterior cingulate activations that are also typically seen. Interestingly, the lateral prefrontal, motor and basal ganglia activations seen in this study are far less commonly identified [27]. These differences are likely attributable to unique aspects of the present task modality, including the music-related cue and potentially the unconstrained nature of autobiographical recall during music listening. It is notable that there is generally a high proportion of network nodes implicated in motor planning, coordination, and execution. This suggests that action memory may be a relatively consistent aspect of music-related memory for this cohort [28]. The absence of occipital activation suggests that, conversely, visual imagery is not consistently engaged during task-induced recall.

Longitudinal analyses conducted in this study further indicated that only a subset of musical memory nodes showed consistent changes from pre-intervention to post-intervention. In particular, there was a consistent decline in activation subcortically within the GP, along with the right IFG, during long-known music listening. The decline in GP activation is notable, given its role in movement regulation and motor learning in particular [29]; the decreased activation over time may reflect increased efficiency of action memory representation during repeat exposure. Conversely, however, the GP plays a role in reward pathways [30], and longitudinal changes may also be in response to the pleasurable aspects of the musical memory task. The right IFG, by contrast, has been implicated in domains of emotion and memory retrieval [31 –33], thus the decreased activation over time may reflect greater processing efficiency within these domains. When comparing musicians and non-musicians, it is notable that the former group shows less longitudinal change in task-related activity. This suggests that to some extent, the former group already has a relatively efficient musical memory network, which is minimally affected by the intervention. Musicians may also show more heterogeneous patterns of network change, due to differences in their neural representation of music, thus diminishing any group-level effects. Further research is needed to distinguish between these (non-exclusive) mechanisms.

Longitudinal analysis of resting-state functional connectivity indicates that the musical intervention may also have lasting effects on network functional integration, outside of the music listening context. Significant longitudinal effects, however, are constrained to moderate strength connections between non-homologous regions. The primary effect was a reduction in connectivity between temporal and frontal network nodes. The limited extent of effects may be due to the smaller sample size and unconstrained nature of resting state. Nevertheless, results may reflect disengagement between lateral temporal and frontal regions, further suggesting network efficiency and perhaps a reduced need of top-down control in as the task becomes more familiar. Interestingly, musicianship again played a modulatory role primarily within the GP, in this case altering connectivity with frontal networks. Differing from the task data, in this case it is the musicians who showed a more marked decline in connectivity over time, compared to the minimal effects in non-musicians. This suggests that habituation to the music task is implemented differently in non-musicians and musicians, with the former showing greater changes to in-task activation response, but the latter having greater effects on out-of-task network integration.

White matter microstructural analysis provided some evidence that the music intervention caused measurable structural changes in conjunction with the observed functional changes. While measures of FA and AD were not significantly affected, RD showed significant longitudinal changes that varied between non-musicians and musicians. In terms of longitudinal effects, there was significantly reduced RD in the right superior corona radiata, denoting microstructural change within this region. While the exact interpretation of DTI parameter change is often ambiguous, studies of pathophysiology indicate that a decline in RD in the absence of AD change may correspond to an increase in myelination post-intervention [34, 35]. There were also significant effects of musicianship on RD change of the superior corona radiata, as musicians show further decline in the identified regions, while non-musicians show longitudinal increase. If the myelination interpretation is correct, this would suggest that musicians see further beneficial structural change, compared to non-musicians. However, there are other possible interpretations of the study findings, as RD change may be driven (in part or in whole) by altered extracellular water content, e.g., due to changes in inflammatory state [36, 37]. It must also be emphasized that the overall differences between musicians and non-musicians remain limited, as groups were not significantly different before or after intervention. In all, the DTI findings suggest subtle structural changes accompanying the musical intervention, but further investigation is needed to validate these findings.

Though our sample size was small, and these results require validation in larger samples, we demonstrated significant improvements on MoCA scores for memory, after participants engaged in a 3-week individualized music-based intervention, which consisted of listening to long-known music for one hour per day, corresponding to the observed changes in brain structure and function. Our results are consistent with and extend previous studies that have shown a beneficial effect of long-known music exposure on memory performance [7 , 10]. The present study did not include a measure of autobiographical memory but instead looked at sub-domains of the MoCA and found an improvement in the memory sub-score. Relative to other studies ours was of shorter duration, making decrements in cognitive test performance less likely. Additionally, our study did not use a control group. Prior studies that used a control group consisting of usual care generally found lower rates of cognitive improvement in particular in frontal-executive domains or greater rates of cognitive decline relative to the music intervention group [7]. Moreover, we did not use any measure of self-consciousness. Despite the methodological differences between this study and previous literature, these findings collectively point towards a generally beneficial impact of long-known music listening on memory performance with further research with larger samples needed to consolidate the different protocols and indices of patient outcome.

While our paper revealed a number of conclusions worthy of further investigation, there are some limitations that should be acknowledged. First, our sample size was relatively small, limiting the power of our conclusions. Second, we combined patients with MCI and AD, which constitute different disorders with varying levels of baseline pathology. Although the present study examined differences between musicians and non-musicians, the subgroup sample sizes were modest and must be replicated in future work. Moreover, the subgrouping was based on formal music training and more detailed background (e.g., formal training, years experience, instrument(s)) may be relevant to findings. Finally, we did not have an appropriate control group that did not receive the intervention or a non-music autobiographical cue.

In this study, we demonstrated modest improvements in cognitive functioning that correlated with task-based de-activation in specific regions involved in processing music including the GP and the right IFG. Moreover, we were able to demonstrate longitudinal effects of the intervention on music-related brain activity, which were detected during long-known but not recently-heard music. As well, we demonstrated significant intervention effects on functional connectivity, outside of the music listening context. Additionally, we found evidence of corresponding structural changes in white matter, including reduced RD in the superior corona radiata over time using DTI. Musicianship also appeared to affect the longitudinal changes in activation, functional connectivity, and white matter microstructure. These findings suggest that repeated exposure to long-known music may induce cognitive effects through consistent changes in brain activation and functional connectivity of nodes in the identified musical network, along with corresponding white matter changes, and that musicianship may play an important modulatory role in these processes. This data, though preliminary, provides support for the use of long-known music listening as a potential therapy for AD. Future studies with a larger number of subjects, of longer duration, and incorporating a control cohort are required to validate these findings, establish whether musicianship moderates response and identify the specific cognitive domains most sensitive to music-based interventions.

Footnotes

ACKNOWLEDGMENTS

We wish to acknowledge the patients who participated in this study without whom this research would not have been possible. This work was funded by a grant from Canadian the Consortium on Neurodegeneration and Aging as well as VPR/Provost Start Up Fund to Dr. Michael Thaut.

Authors’ disclosures available online ().

The supplementary material is available in the electronic version of this article: .

References

Lanctôt

, Herrmann

, Yau

, Khan

, Liu

, LouLou

, Einarson

(2003) Efficacy and safety of cholinesterase inhibitors in Alzheimer’s disease: A meta-analysis. CMAJ 169, 557–564.

Rickard

, Toukhsati

, Field

(2005) The effect of music on cognitive performance: Insight from neurobiological and animal studies. Behav Cogn Neurosci Rev 4, 235–261.

Angelucci

, Fiore

, Ricci

, Padua

, Sabino

, Tonali

(2007) Investigating the neurobiology of music: Brain-derived neurotrophic factor modulation in the hippocampus of young adult mice. Behav Pharmacol 18, 491–496.

Irish

, Cunningham

, Walsh

, Coakley

, Lawlor

, Robertson

, Coen

(2006) Investigating the enhancing effect of music on autobiographical memory in mild Alzheimer’s disease. Dement Geriatr Cogn Disord 22, 108–120.

Thompson

, Schellenberg

, Husain

(2001) Arousal, mood, and the Mozart effect. Psychol Sci 12, 248–251.

Fang

, Ye

, Huangfu

, Calimag

(2017) Music therapy is a potential intervention for cognition of Alzheimer’s disease: A mini-review. Transl Neurodegener 6, 2.

Leggieri

, Thaut

, Fornazzari

, Schweizer

, Barfett

, Munoz

, Fischer

(2019) Music intervention approaches for Alzheimer’s disease: A review of the literature. Front Neurosci 13, 132.

El Haj

, Fasotti

, Allain

(2012) The involuntary nature of music-evoked autobiographical memories in Alzheimer’s disease. Conscious Cogn 21, 238–246.

El Haj

, Postal

, Allain

(2012) Music enhances autobiographical memory in mild Alzheimer’s disease. Educ Gerontol 38, 30–41.

10.

Arroyo-Anlló

, Díaz

, Gil

(2013) Familiar music as an enhancer of self-consciousness in patients with Alzheimer’s disease. Biomed Res Int 2013, 752965.

11.

Baird

A SS

, Miller

, Chalmers

(2017) Does music training facilitate the mnemonic effect of song? An exploration of musicians and nonmusicians with and without Alzheimer’s dementia. J Clin Exp Neuropsychol 39, 9–21.

12.

Baird

, Samson

(2009) Memory for music in Alzheimer’s disease: Unforgettable? Neuropsychol Rev 19, 85–101.

13.

Amer

, Kalender

, Hasher

, Trehub

, Wong

(2013) Do older professional musicians have cognitive advantages? PLoS One 8, e71630.

14.

Thaut

, Fischer

, Leggieri

, Vuong

, Churchill

, Fornazzari

, Schweizer

(2020) Neural basis of long-term musical memory in cognitively impaired older persons. Alzheimer Dis Assoc Disord 34, 267–271.

15.

King

, Jones

, Goldberg

, Rollins

, MacNamee

, Moffit

, Naidu

, Ferguson

, Garcia-Leavitt

, Amaro

, Breitenbach

, Watson

, Gurgel

, Anderson

, Foster

(2019) Increased functional connectivity after listening to favored music in adults with Alzheimer dementia. J Prev Alzheimers Dis 6, 56–62.

16.

McKhann

, Knopman

, Chertkow

, Hyman

, Jack

, Kawas

, Klunk

, Koroshetz

, Manly

, Mayeux

(2011) The diagnosis of dementia due to Alzheimer’s disease: Recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement 7, 263–269.

17.

Folstein

, Folstein

, McHugh

(1975) “Mini-mental state”: A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res 12, 189–198.

18.

Davis

, Thaut

(1989) The influence of preferred relaxing music on measures of state anxiety, relaxation, and physiological responses. J Music Ther 26, 168–187.

19.

Thaut

, Davis

(1993) The influence of subject-selected versus experimenter-chosen music on affect, anxiety, and relaxation. J Music Ther 30, 210–223.

20.

Nasreddine

, Phillips

, Bédirian

, Charbonneau

, Whitehead

, Collin

, Cummings

, Chertkow

(2005) The Montreal Cognitive Assessment, MoCA: A brief screening tool for mild cognitive impairment. J Am Geriatr Soc 53, 695–699.

21.

Freitas

, Prieto

, Simoes

, Santana

(2014) Psychometric properties of the Montreal Cognitive Assessment (MoCA): An analysis using the Rasch model. Clin Neuropsychol 28, 65–83.

22.

Churchill

, Hutchison

, Graham

, Schweizer

(2019) Mapping brain recovery after concussion: From acute injury to one year after medical clearance. Neurology 93, e1980–e1992.

23.

Churchill

, Hutchison

, Graham

, Schweizer

(2020) Scale-free functional brain dynamics during recovery from sport-related concussion. Hum Brain Mapp 41, 2567–2582.

24.

Churchill

, Hutchison

, Graham

, Schweizer

(2021) Brain function associated with reaction time after sport-related concussion. Brain Imaging Behav 15, 1508–1517.

25.

Strother

, Anderson

, Hansen

, Kjems

, Kustra

, Sidtis

, Frutiger

, Muley

, LaConte

, Rottenberg

(2002) The quantitative evaluation of functional neuroimaging experiments: The NPAIRS data analysis framework. Neuroimage 15, 747–771.

26.

Conway

, Pleydell-Pearce

, Whitecross

(2001) The neuroanatomy of autobiographical memory: A slow cortical potential study of autobiographical memory retrieval. J Mem Lang 45, 493–524.

27.

Svoboda

, McKinnon

, Levine

(2006) The functional neuroanatomy of autobiographical memory: A meta-analysis. Neuropsychologia 44, 2189–2208.

28.

Senkfor

, Petten

, Kutas

(2002) Episodic action memory for real objects: An ERP investigation with perform, watch, and imagine action encoding tasks versus a non-action encoding task. J Cogn Neurosci 14, 402–419.

29.

Turner

, Desmurget

(2010) Basal ganglia contributions to motor control: A vigorous tutor. Curr Opin Neurobiol 20, 704–716.

30.

Howell

, Prescott

, Lozano

, Hodaie

, Voon

, Hutchison

(2016) Preliminary evidence for human globus pallidus pars interna neurons signaling reward and sensory stimuli. Neuroscience 328, 30–39.

31.

Ritchey

, LaBar

, Cabeza

(2011) Level of processing modulates the neural correlates of emotional memory formation. J Cogn Neurosci 23, 757–771.

32.

Swart

, Liemburg

, Kortekaas

, Wiersma

, Bruggeman

, Aleman

(2013) Normal brain activation in schizophrenia patients during associative emotional learning. Psychiatry Res 214, 269–276.

33.

Drobyshevsky

, Baumann

, Schneider

(2006) A rapid fMRI task battery for mapping of visual, motor, cognitive, and emotional function. Neuroimage 31, 732–744.

34.

Klawiter

, Schmidt

, Trinkaus

, Liang

H-F

, Budde

, Naismith

, Song

S-K

, Cross

, Benzinger

(2011) Radial diffusivity predicts demyelination in ex vivo multiple sclerosis spinal cords. Neuroimage 55, 1454–1460.

35.

Song

S-K

, Yoshino

, Le

, Lin

S-J

, Sun

S-W

, Cross

, Armstrong

(2005) Demyelination increases radial diffusivity in corpus callosum of mouse brain. Neuroimage 26, 132–140.

36.

Lodygensky

, West

, Stump

, Holtzman

, Inder

, Neil

(2010) In vivo MRI analysis of an inflammatory injury in the developing brain. Brain Behav Immun 24, 759–767.

37.

Wang

, Wang

, Haldar

, Yeh

F-C

, Xie

, Sun

, Tu

T-W

, Trinkaus

, Klein

, Cross

(2011) Quantification of increased cellularity during inflammatory demyelination. Brain 134, 3590–3601.