Functional gradients in prefrontal regions and somatomotor networks reflect the effect of music training experience on cognitive aging

Abstract

Studies showed that the top-down control of the prefrontal cortex (PFC) on sensory/motor cortices changes during cognitive aging. Although music training has demonstrated efficacy on cognitive aging, its brain mechanism is still far from clear. Current music intervention studies have paid insufficient attention to the relationship between PFC and sensory regions. Functional gradient provides a new perspective that allows researchers to understand network spatial relationships, which helps study the mechanism of music training that affects cognitive aging. In this work, we estimated the functional gradients in four groups, young musicians, young control, older musicians, and older control. We found that cognitive aging leads to gradient compression. Compared with young subjects, older subjects presented lower and higher principal gradient scores in the right dorsal and medial prefrontal and the bilateral somatomotor regions, respectively. Meanwhile, by comparing older control and musicians, we found a mitigating effect of music training on gradient compression. Furthermore, we revealed that the connectivity transitions between prefrontal and somatomotor regions at short functional distances are a potential mechanism for music to intervene in cognitive aging. This work contributes to understanding the neuroplasticity of music training on cognitive aging.

music training, cognitive aging, gradient compression, functional connectivity

Introduction

Cognitive aging results in sensory/motor and cognition changes (Albers et al. 2015), which have seriously affected the lives of the elderly. Research showed that it is difficult for the elderly to ignore distracting information in sensory perception (Van Gerven and Guerreiro 2016), which is reflected in hearing (Alain et al. 2012), vision (Solbakk et al. 2008), tactile (Ruitenberg et al. 2019), and motor (Schäfer et al. 2006). The inhibitory efficiency and inhibition ability of the elderly was reduced (Connelly and Hasher 1993) compared with young people, which made it difficult for them to clear information that was no longer relevant to working memory (Hasher et al. 1997). Age-related deterioration in working memory performance has been suggested to be associated with poor inhibitory control over the information of working memory (Hasher and Zacks 1988). Inhibitory control is defined as an attentional selection mechanism including: restraint, access, and deletion functions. Among them, the inhibition function is thought to suppress prominent but irrelevant information, so as to activate weaker but potentially more important information (Zeintl and Kliegel 2007). Thus, age-related decline of inhibitory control results in a deficit of ignoring task-irrelevant information, which makes it difficult to access and retrieve task-relevant information (Stoltzfus et al. 1996; Zacks et al. 1997). It is precisely because of the important role of inhibition control in working memory that many studies have explored through working memory experiments with different difficulties and found age-related susceptibility to active interference information (Bowles and Salthouse 2003; Bunting 2006). The defect of inhibiting irrelevant information may lead to information confusion in working memory and reduce the overall processing efficiency, which is considered to be an important feature of cognitive aging (Zeintl and Kliegel 2007).

The prefrontal cortex (PFC) is one of the most severely affected regions in aging (Resnick et al. 2003; Raz 2005; Takao et al. 2010) and has been demonstrated to be a critical node for ignoring distracting information (Hasher et al. 1997). This region is considered to play an essential role in top-down impulse control, inhibitory control (Jackson et al. 2011; Aron et al. 2014), and regulating neural activity in sensory regions (Picton et al. 2007; Bidet et al. 2015; Amer et al. 2016). The PFC may facilitate and/or suppress sensory processing through thalamocortical connections to the bottleneck of afferent input to auditory, visual, and somatosensory cortices (Scannell et al. 1999; Nakajima et al. 2019). Structural studies found that aging leads to more decline of gray matter density in the PFC than sensory region (Raz et al. 1997), and white and gray matter changes are related to increased auditory- and visual-evoked response latency (Price et al. 2017). Functional study found that a larger amplitude of auditory-evoked potentials (Chao and Knight 1997) is associated with a greater percent of Wisconsin Card Sorting Test preservative errors, which can assess executive functions mediated by prefrontal regions (Gruber et al. 2010; Kim et al. 2011). Prefrontal damages have been associated with changes in sensory-evoked response amplitude (Knight et al. 1989; Yamaguchi and Knight 1990). Moreover, among the functional changes associated with age, a prominent observation is that different neural structures mediate a reduction in the specificity of specific processing roles (Li and Lindenberger 1999). Functional connectivity studies found that the reduction in specificity is particularly common in the fronto-parietal regions, namely, the segregation between it and other networks is reduced (Geerligs et al. 2017; Bethlehem et al. 2020). Meanwhile, research also suggested that decreasing segregation of sensory-motor systems is age-related (Chan et al. 2014). Our recent work further found that, compared with young adults, the elderly exhibited stronger source activity in sensory cortices and stronger neural synchrony between prefrontal and sensory cortices. The positive correlation between neural synchrony and source activity suggested that age differences in sensory-evoked responses amplitude may be linked to ineffective top-down regulation from supramodal prefrontal regions (Alain et al. 2022). It is inline with previous study that the prefrontal regions can exert top-down control over sensory cortices (Knight et al. 1999). Sum up, the above studies suggest that the relationship between the prefrontal lobe and the sensory perception area plays an essential role in cognitive aging.

Music training has been widely used in cognitive aging intervention studies. Behavior study found that music practice can promote cognitive reserve, which can be helpful for cognitive decline due to aging (Fauvel et al. 2013). Neuroimaging study showed that music training contributes to the maintenance of music-related memory (Jacobsen et al. 2015). Meanwhile, positive music performance can promote working memory (Bugos 2019). Our longitudinal music training research found that music training could change theta oscillations in the fronto-central region to affect inhibitory control of the elderly with the underlying mechanism that functional connectivity changes between the PFC and sensory cortices (Lu et al. 2022). Only a few studies on music intervenes in the aging have focused on the functional connections between prefrontal and sensorimotor regions, and they have used mainly regions of interest (ROIs) analysis. It is necessary to study it through a more integrated vision. Functional gradient can capture continuous spatial patterns of connectivity in the brain (Atasoy et al. 2016; Margulies et al. 2016; Haak et al. 2018) and get more characteristics of how the sensory and high-level cognitive networks assemble together (Huntenburg et al. 2018) in music intervenes aging. Meanwhile, the principal gradient reflects the distances among connectivity patterns of brain regions across the primary sensory network toward the transmodal regions of the default-mode network (DMN; Margulies et al. 2016; Bethlehem et al. 2020; Xia et al. 2022). It provides a new perspective to study the segregation and integration between the prefrontal and sensory regions in the process of music intervention in cognitive aging. Here, we calculate the principal gradient of young musicians, young control, older musicians, and older control and provide evidence for the changes in brain hierarchy under music intervention. Then, stepwise functional connectivity (SFC) is used to further demonstrate that music training enhances the functional separation between the regions across prefrontal and somatomotor networks, which may represent the mechanism that music affects inhibitory function in working memory during aging.

Materials and methods

Subjects

A total of 82 subjects were recruited, of which 25 young nonmusicians from the University of Electronic Science and Technology of China were recruited as young control group, 22 young musicians and 16 older musicians from Sichuan Conservatory of Music were recruited as young and older musicians group, and 19 age matched older control from community were recruited as older control group. All subjects were right-handed according to the Edinburgh Handedness Inventory (Oldfield 1971) and had normal hearing ability. They were paid for participating in the study. All musicians had formal musical education and could play at least one instrument. All nonmusicians who could not play instrument had no formal musical education. The level of music ability of all subjects was confirmed by the Montreal Music History Questionnaire (MMHQ; Coffey et al. 2011), the Goldsmiths Musical Sophistication Index (Gold-MSI) (Gold et al. 2019), and the Barcelona Music Reward Questionnaire (BMRQ; Mas-Herrero et al. 2013). We also collected Mini-Mental State Examination (MMSE), Self-Rating Depression Scale (SDS), Self-Rating Anxiety Scale (SAS), and the Big Five Inventory (Caprara et al. 1993) to further explore the subjects’ individual backgrounds and differences. MMSE was used as a screening test. The information is provided in Table 1. We collected resting-state functional Magnetic Resonance Imaging (fMRI) data and neuroelectric brain activity (EEG) under the paradigm of N-back. The behavior data of N-back were used as an indicator to characterize the degree of cognitive aging in the further analysis. Four young musicians and one older musician who did not complete the N-back EEG experiment were excluded from the correlation analysis.

Table 1

Participant demographic data.

Characteristic	Old control (n = 19)	Old musicians (n = 16)	Young control (n = 25)	Young musician (n = 22)	Ab	P value Bc	Cc
Gender (Male)	7	11	15	8	0.17a
Age	61.58 ± 4.02	65.25 ± 3.85	22.16 ± 2.03	23.95 ± 3.52	8.18e−61	0.04	0.01
Years of education	10.88 ± 2.22	13.88 ± 0.86	16.88 ± 1.51	16.40 ± 6.89	1.44e−22	0.52	7.36e−06
BMRQ Music seeking Emotion evocation Mood regulation Sensori-motor Social Music reward Overall score	46.74 ± 9.95 48.16 ± 7.75 45.58 ± 9.75 41.79 ± 8.56 48.74 ± 8.37 44.42 ± 8.93 72.79 ± 7.22	57.79 ± 6.68 56.21 ± 6.04 46.14 ± 8.28 48.43 ± 8.79 58.50 ± 8.52 54.29 ± 8.07 81.86 ± 8.04	50.36 ± 8.73 49.88 ± 6.85 46.68 ± 8.39 46.60 ± 9.44 49.88 ± 10.06 47.80 ± 9.52 73.52 ± 7.84	59.45 ± 4.96 54.23 ± 7.57 48.27 ± 6.96 51.00 ± 7.72 57.82 ± 7.15 55.32 ± 6.43 80.23 ± 5.55	4.44e−06 0.01 0.77 0.01 6.34e−04 2.62e−04 2.22e−04	0.0001 0.05 0.50 0.10 0.004 0.004 0.002	0.001 0.003 0.87 0.04 0.003 0.003 0.003
The Big Five Inventory Extraversion Agreeableness Conscientiousness Neuroticism Openness	3.26 ± 0.60 4.10 ± 0.54 3.77 ± 0.46 2.39 ± 0.67 3.32 ± 0.56	3.21 ± 0.48 4.19 ± 0.49 4.20 ± 0.58 2.12 ± 0.58 4.11 ± 0.47	3.27 ± 0.49 3.74 ± 0.52 3.20 ± 0.53 2.97 ± 0.53 3.42 ± 0.58	3.19 ± 0.55 3.82 ± 0.46 3.29 ± 0.57 3.01 ± 0.57 3.85 ± 0.48	0.99 0.02 1.18e−06 2.53e−05 1.24e−04	0.64 0.58 0.63 0.82 0.01	0.84 0.63 0.03 0.25 0.0002
GDS-30	5.58 ± 2.96	3.14 ± 3.29	6.40 ± 4.48	8.81 ± 7.49	0.02	0.19	0.04
SAS	30.42 ± 6.04	28.57 ± 4.66	27.38 ± 5.84	35.45 ± 7.72	5.05e−04	0.0002	0.36
SDS	34.21 ± 7.55	30.71 ± 9.05	28.50 ± 6.15	38.18 ± 9.19	0.001	0.0001	0.25
MMSE	28.26 ± 1.52	29.08 ± 7.59	29.88 ± 0.52	29.73 ± 0.69	1.02e−05	0.40	0.13
Training yearsm	N/A	44.44 ± 15.78	N/A	10.91 ± 5.53	N/A	1.03e−10	N/A
Age of startm	N/A	10.63 ± 4.62	N/A	7.82 ± 3.93	N/A	0.06	N/A
Training timem (h/week)	N/A	38.66 ± 16.90	N/A	10.55 ± 7.45	N/A	7.04e−08	N/A
Listening musicg (h/day)	N/A	4.81 ± 2.49	N/A	4.34 ± 6.11	N/A	0.78	N/A
Performanceg (last year)	N/A	8.25 ± 2.97	N/A	5.05 ± 1.85	N/A	0.0003	N/A
Training timeg (h/day)	N/A	4.47 ± 1.01	N/A	5 ± 1.65	N/A	0.27	N/A

Abbreviations: Values are shown as mean ± SD, unless otherwise noted. BMRQ: The Barcelona Music Reward Questionnaire (Mas-Herrero et al. 2013). The Big Five Inventory: The Big Five Inventory (Caprara et al. 1993). GDS-30: The Geriatric Depression Scale. Performance: Time of participating in live music performances. P-values refer to: A: significance across all four groups. B: significance between the 2 young groups. C: significance between the 2 old groups.

^aChi square test was used.

^bOne-way analysis of variance was used.

^cTwo-sample t-test was used.

^gThe Gold-MSI (Gold et al. 2019).

^mMMHQ (Coffey et al. 2011).

The study was done with the approval of the Ethics Committee of the University of Electronic Science and Technology of China (No. 1061420210305026). All procedures were carried out in adherence to approved guidelines. All the subjects were fully informed about the nature and procedures of the study before the experiment, and informed written consent was obtained from participants prior to participation in the study.

N-back task

Ten different single digits served as stimuli in the N-back. Stimuli were spoken by a female voice and lasted 500 ms. Responses were required independently for each modality whenever the current stimulus matched the stimulus one or two positions back in the sequence (depending on the load level); no response was requested to nontargets. For the 0-back condition, serving as a baseline condition with minimal memory demands, the participant had to respond to a prespecified target. The N-back task consisted of 10 test blocks: two 0-back test blocks, four 1-back test blocks, and four 2-back test blocks. In order to ensure that subjects are familiar with the task, three practice blocks (0, 1, 2-back) were presented before the experiment. Instructions were provided before each new N-level, and 10-s breaks were given between blocks (Kirchner 1958).

fMRI scan

Functional MRI scans were acquired using a 3 T magnetic resonance imaging (MRI) scanner (MR750; GE Discovery, Milwaukee, WI, USA) at the MRI Research Center of University of Electronic Science and Technology of China. During scanning, foam padding and earplugs were used to reduce head motion and scanning noise, respectively. The functional images were acquired using an echo-planar imaging sequence (slices = 35, slice scan order: interleave, echo time [TE] = 30 msec, repetition time [TR] = 2000 msec, flip angle [FA] = 90°, field of view [FOV] = 24 × 24 cm², matrix = 64 × 64, slice thickness/gap = 4 mm/0.4 mm, and slices = 255). Structural T₁-weighted images were acquired using a 3D fast spoiled gradient echo sequence (echo time [TE] = 1.984 msec, repetition time [TR] = 6.008 msec, flip angle [FA] = 90°, field of view [FOV] = 25.6 × 25.6 cm², matrix = 256 × 256, slice thickness/gap = 1 mm/0 mm, and 152 slices).

Processing

Preprocessing was performed using NIT (Neuroscience Information Toolbox, http://www.neuro.uestc.edu.cn/NIT.html) (Dong et al. 2018) and SPM12 software (Statistical Parametric Mapping; http://www.fil.ion.ucl.ac.uk/spm). To avoid MRI machine field effects and eliminate head movements of the participants, a series of preprocessing steps were performed, including discarding the first five volumes, slice time correction, 3D motion detection and correction, and spatial normalization (using parameters from individual T₁ segmentation, and normalizing to MNI space with 3 × 3 × 3 mm³). Then, the white matter and cerebrospinal fluid signals, linear trend signals, whole brain mean signal, and 12 head motion parameters were regressed to reduce nonneural noise and artifacts. Temporal bandpass filtering (pass band 0.01–0.08 Hz) was conducted using a phase-insensitive filter, which was used to reduce the effects of low-frequency drift and high-frequency noise.

Functional gradient

The calculation of functional gradient referred to the research of Dong et al. (2020). Association matrices for each subject were calculated by using the Pearson correlation among 400 ROIs based on Schaefer’s report (Schaefer et al. 2018). After obtaining the matrix of Pearson correlation, Fisher z transformation was applied to normalize the variance in r-values. Then, based on BrainSpace toolbox (Vos de Wael et al. 2020), functional gradient was computing by using a series of processing steps. First, we set a threshold for the matrix with the top 10% of connections per row retained, whereas all others were zeroed (the negative connections were zeroed as well) based on the previous studies (Coifman et al. 2005; Guell et al. 2018). Second, cosine distance was used to generate a similarity matrix that reflected similarity of connectivity profiles between each pair of voxels. Third, we used diffusion map embedding (Coifman et al. 2005), a nonlinear dimensionality reduction technique, to identify a low-dimensional embedding from a high-dimensional connectivity matrix. The parameter α in manifold learning was set to 0.5, which helped retain the global relations between data points in the embedded space (Margulies et al. 2016; Hong et al. 2019; Yang et al. 2020). Finally, the functional gradient of each subject was aligned to the group level gradient component template, which was generated from the average connectivity matrix based on all subjects (Hong et al. 2019). The principal gradient is anchored at one end by the somatosensory/motor, visual, and auditory regions. The other end includes regions such as angular gyrus, middle, superior frontal gyri regions, etc. Meanwhile, the principal gradient has been reported to account for the greatest variance in connectivity in the human brain (Margulies et al. 2016). Thus, the principal gradient was selected for the subsequent statistical analysis. In addition, we also calculated the statistical differences among the four groups in the second and third gradients and reported it in the Supplementary Material.

Stepwise functional connectivity

SFC is a method that can represent how brain systems are bound together (Sepulcre et al. 2012) and allow us to understand the gradient as a sequence of steps in connectivity space. Similarly, it reveals a consistent connectivity evolution from primary sensory to DMN systems that recapitulate functional connectivity gradients (Huntenburg et al. 2018; Hong et al. 2019). Differently, functional connectivity gradient results from an unsupervised dimension reduction, whereas SFC is initiated from defined seeds to construct connectivity evolution (Hong et al. 2019). SFC analysis can depict connectivity transitions at different steps, which efficiently examines how brain systems reconfigure their modes of operation along the axis of brain hierarchy (Sepulcre et al. 2012; Lee et al. 2022).

First, we defined ROIs based on the statistical analysis results of functional gradient and calculated functional connections among these regions. By comparing the FC difference between normal aging and aging with music training experience, we further identified the brain regions that we focused on in the SFC analysis. The calculation of SFC analysis is referenced in previous studies (Sepulcre et al. 2012; Pretus et al. 2019; Costumero et al. 2020; Lee et al. 2022) with the following steps: first, we calculated the matrix of Pearson correlation after Fisher z transformation. Only positive correlations of the association matrix were retained, as positive connectivity has been proved to drive functional connectivity network topology in the human brain (Qian et al. 2018). Then, false discovery rate (FDR) correction (Benjamini and Hochberg 1995) (P < 0.05) was used to remove spurious associations (positive correlations that did not reach the correction threshold). Finally, based on a previous study (Costumero et al. 2020), the degree of stepwise connectivity for a given step distance is computed from the power of adjacency matrices. Matrices were then normalized between 0 and 1. As reported in the previous study, the SFC analysis showed different models in the first to sixth step and became stable in the seventh step (Sepulcre et al. 2012). Thus, we computed the SFC at the first seven step distances. Meanwhile, for each step, we computed the SFC degree between the ROIs and used it in the subsequent comparison and correlation analysis.

Results

Functional gradient

We first divided the subjects into two groups according to age and compared the functional gradient between young and older subjects regardless of the music training experience to obtain a more stable aging effect. Then, we focused on these regions with the stable influence of aging and further explored the effect of music training, which can help us to exclude the bias due to the small sample. We found that the older subjects exhibited higher principal gradient scores in some regions across the bilateral somatomotor networks and lower principal gradient scores in the right dorsal and medial prefrontal regions compared with young subjects. The significantly different regions were reported in Table 2 and Fig. 1 (FDR-corrected). A two-way ANOVA between age and music training experience (young/older; musicians/nonmusicians) was then used to determine the influence of aging and music intervention. We focused on the regions across the prefrontal and bilateral somatomotor networks and found the following results: first, for the RH_Default_PFCdPFCm_3, we found a main effect of age (F = 12.65, P = 0.0006) with the principal gradient score being lower in older subjects group than young subjects group. Pairwise comparisons revealed a significant difference between young musicians and older control (P = 0.0448), and older musicians (P = 0.0155). Second, we averaged the principal functional gradient of the regions across the somatomotor network, which had exhibited a significant difference in previous statistical analysis (LH_SomMot_1, LH_SomMot_4, LH_SomMot_5, LH_SomMot_8, LH_SomMot_10, RH_SomMot_1, RH_SomMot_3, RH_SomMot_9, RH_SomMot_11, and RH_SomMot_13). We found a main effect of age (F = 28.27, P = 9.783e−7) and music (F = 8.19, P = 0.0054). More importantly, the age × music training experience interaction trended toward significance (F = 8.1, P = 0.0057). Pairwise comparisons revealed that the principal gradient scores of young musicians (P = 6.59e−7), young control (P = 3.43e−7), and older musicians (P = 0.0018) are higher than older control. Finally, we computed the difference of principal functional gradient between the RH_Default_PFCdPFCm_3 and the Bilateral_SomMot. We found that the difference showed a main effect of age (F = 28.67, P = 8.388e−07) with the principal gradient score being lower in the older subjects group than the young subjects group. The results of pairwise comparisons were young musicians > older musicians (P = 0.0052), young musicians > older control (P = 4.78e−5), young control > older musicians (P = 0.0306), and young control > older control (P = 4.38e−4) (Fig. 2).

Table 2

The significant difference in principal gradient scores between young and older subjects.

ROI name	Network name	Full component name	T value
LH_Vis_6	visual	visual	−3.3378
LH_SomMot_1	somatomotor	somatomotor	−3.8703
LH_SomMot_4	somatomotor	somatomotor	−3.4088
LH_SomMot_5	somatomotor	somatomotor	−3.2268
LH_SomMot_8	somatomotor	somatomotor	−3.4655
LH_SomMot_10	somatomotor	somatomotor	−3.2521
LH_Limbic_OFC_3	limbic	orbital frontal cortex	4.2012
LH_Default_PFC_4	default	PFC	3.3139
RH_SomMot_1	somatomotor	somatomotor	−3.3111
RH_SomMot_3	somatomotor	somatomotor	−3.344
RH_SomMot_9	somatomotor	somatomotor	−3.4709
RH_SomMot_11	somatomotor	somatomotor	−3.6333
RH_SomMot_13	somatomotor	somatomotor	−3.4303
RH_SalVentAttn_FrOperIns_1	salience/ventral attention	frontal operculum insula	−3.2283
RH_SalVentAttn_FrOperIns_3	salience/ventral attention	frontal operculum insula	−3.2785
RH_SalVentAttn_FrOperIns_4	salience/ventral attention	frontal operculum insula	−5.3526
RH_Limbic_OFC_3	limbic	orbital frontal cortex	3.3381
RH_Default_PFCdPFCm_3	default	dorsal PFC medial PFC	3.5216

Fig. 1

A) Principal functional gradient of young and older subjects. B) Compressed gradient pattern in older subjects shown in density histogram (gray: young subjects, light red: older subjects). C) Significant difference map between the two groups (t map, FDR-corrected P < 0.05).

Fig. 2

A) Principal functional gradient of young musicians, young control, older musicians, and older control. B) Histogram of statistical analysis results among four groups (light blue: young musicians, blue: young control, light purple: older musicians, purple: older control; ^**: P < 0.0001, ^*: P < 0.05).

Stepwise functional connectivity

Based on the results of the principal functional gradient (Table 2), we calculated the functional connections among these ROIs. The different matrices of functional connectivity were shown in Fig. 3. We found stronger connections between the right dorsal and medial prefrontal region and other regions in older subjects than in young subjects. Further comparisons across the four groups revealed a similar trend under normal aging, while no such trend was found for aging under music training experience.

Fig. 3

Difference of functional connectivity based on the ROIs in results of functional gradient. A) Comparison of the functional connections between young subjects and older subjects (regardless music training experience). B) Changes of functional connections induced by normal aging. C) No changes in functional connections caused by aging under music training experience. (FDR-corrected P < 0.05).

Then, we computed the SFC at a given distance of steps 1–7 and selected the right dorsal and medial prefrontal regions as the ROI. Subsequently, statistical analysis of SFC degree from it to somatomotor regions was performed for each step. Statistical differences between normal aging and aging with music intervention were only found at a given distance of step 3. Statistical analysis results of SFC degree showed a higher SFC degree in older subjects than young subjects (P = 0.0091, t = 2.6715). The further comparison revealed the trend that older control > older musicians > young control > young musicians in step 3. We found older control > young control (P = 0.0428, t = 2.0888) and older control > young musicians (P = 0.0299, t = 2.2533). Nevertheless, there is no significant difference between older musicians and young musicians (Fig. 4).

Fig. 4

A) Difference of SFC degree between young and older subjects. B) SFC degree of four groups for a given step distance (step3) (light blue: young musicians, blue: young control, light purple: older musicians, purple: older control).

N-back

The behavior data of N-back are shown in Table 3. We calculated the main effect of age and music training and their interaction effect for 0, 1, and 2-back. We found that only the accuracy of 2-back exhibited three significant effects simultaneously. Specifically, we found a main effect of age (F = 27.57, P = 1.47e−6) and music (F = 8.43, P = 0.0049). More importantly, the age × music training experience interaction trended toward significance (F = 8.44, P = 0.0049). Pairwise comparisons revealed that the accuracy of the young control is higher than the older control (P = 3.85e−7, t = 6.01); the accuracy of young musicians is higher than older control (P = 1.04e−6, t = 5.90); the accuracy of the older musicians is higher than older control (P = 0.0015, t = 3.47). Additionally, the accuracy of the 0-back experiment of older musicians is higher than the older control (P = 0.0615, t = 1.94). The accuracy of the 1-back experiment of older musicians is higher than the older control (P = 0.0417, t = 2.13). The reaction time of the 1-back experiment of older musicians is lower than the older control (P = 0.0313, t = −2.26). The reaction time of the 2-back experiment of older musicians is lower than the older control (P = 0.0058, t = −2.96). Additional statistical results are provided in the supplementary material.

Table 3

Behavioral results of N-back.

N-back	Old control (n = 19)	Old musicians (n = 15)	Young control (n = 25)	Young musician (n = 18)	Difference between older musicians and older control (p)
0-back:
Accuracy	0.91 ± 0.09	0.97 ± 0.06	0.98 ± 0.05	0.97 ± 0.03	0.0615
Reaction Time(s)	0.79 ± 0.14	0.73 ± 0.11	0.63 ± 0.10	0.67 ± 0.15	0.1529
1-back:
Accuracy	0.91 ± 0.07	0.96 ± 0.04	0.96 ± 0.05	0.97 ± 0.03	0.0417
Reaction Time(s)	0.76 ± 0.10	0.68 ± 0.11	0.61 ± 0.12	0.65 ± 0.14	0.0313
2-back:
Accuracy	0.83 ± 0.06	0.89 ± 0.06	0.93 ± 0.05	0.93 ± 0.04	0.0015
Reaction Time(s)	0.78 ± 0.08	0.70 ± 0.10	0.62 ± 0.11	0.65 ± 0.13	0.0058

Correlation analysis

Correlation analysis was used to find out the relationship between music training, cognitive behavior data, and principal functional gradient as well as SFC degree. We first found that there is a negative correlation between the difference of principal functional gradient (the RH_Default_PFCdPFCm_3—the Bilateral_SomMot) and SFC degree (P = 6.88e−11, r = −0.6437, Fig. 5a). Then, we found negative correlations between age and gradient distribution distance in musicians and control (P = 0.0037, r = −0.46, P = 8.21e⁻⁶, r = −0.6169, Fig. 5b). There is a positive correlation between age and Principal gradient score of bilateral SomMot (P = 3.81e⁻⁷, r = 0.6801, Fig. 5b) and positive correlations between age and SFC degree (P = 0.0744, r = 0.2928, P = 0.0225, r = 0.3433, Fig. 5c). Second, as shown in Fig. 6A, we found a positive correlation between accuracy and principal gradient score (P = 0.0024, r = 0.7216), a negative correlation between accuracy and age of music training start (P = 0.0767, r = −0.4706), and a positive correlation between accuracy and gradient distribution distance (P = 0.0062, r = 0.6704). Finally, there is a negative correlation between the age of music training start and principal gradient score (P = 0.063, r = −0.4749) and a positive correlation between the age of music training start and SFC degree (P = 0.0859, r = 0.4428) (Fig. 6B).

Fig. 5

Correlation results. A) Negative correlation between SFC degree and gradient distribution distance (prefrontal-somatomotor). B) Correlation between age and principal gradient score of the regions across somatomotor network and gradient distribution distance (prefrontal-somatomotor). C) Correlation between age and SFC degree. Red: All subjects. Purple: Musicians. Blue: Control.

Fig. 6

Correlation results of older musicians. A) Correlation between the accuracy of 2-back and principal gradient score of prefrontal regions, age of music training start, and gradient distribution distance (prefrontal-somatomotor). B) Correlation between age of music training start and principal gradient score of prefrontal regions and SFC degree. Purple: Older musicians. Blue: Older control.

Discussion

We provided evidence for the compression of the principal prefrontal regions-to-somatomotor gradient in cognitive aging. The compression of the somatomotor of the gradient was mitigated by music training. Correlation results revealed that the gradient distribution of the two regions is correlated with the accuracy of 2-back test. Functional connectivity between prefrontal regions and the somatomotor network could be a potential mechanism underlying it. We reveal the mechanism of music training affects inhibitory function in working memory during aging and offer a new perspective for understanding brain neuroplasticity under the influence of aging and music intervention.

First, our results found that older adults showed a compressed gradient pattern compared with young adults, which supported the opinion that aging is commonly associated with changes to segregation and integration of functional brain networks. As reported in previous studies, the principal gradient recapitulated the sensory-to-transmodal organization (Margulies et al. 2016; Paquola et al. 2019). Thus, the compressed gradient pattern in our results may represent the changes in somatomotor-to-prefrontal organization related to aging. Studies suggested that aging resulted in reduced segregation between networks, particularly in the default mode and fronto-parietal networks (Chan et al. 2014; Geerligs et al. 2017; Bethlehem et al. 2020). On average, the strength of functional connections within the resting state network decreases, while the connectivity between networks increases (Betzel et al. 2014; Zonneveld et al. 2019). For example, compared with young adults, older adults exhibited greater connectivity between the frontal–parietal control system and other systems, which may suggest that increased age could lead to the decreased independence of brain systems (Chan et al. 2014). Furthermore, age-related differences in dispersion are thought to be a significant mediator of the negative relationship between fluid intelligence and age (Bethlehem et al. 2020). Thus, we thought that the right dorsal and medial prefrontal regions play a crucial role in cognitive aging. Combined with correlations with accuracy in the working memory paradigm, we suggested that the gradient compression pattern of the two regions is the mechanism that causes the ability to ignore task-irrelevant information decline in working memory tasks in the elderly.

Further exploration revealed that the compression was affected by music training because the principal gradient score of the bilateral somatomotor of older musicians is significantly lower than older control. Previous studies revealed the structure and functional plastic changes of musicians (Elbert et al. 1995; Schulz et al. 2003; Hosoda and Furuya 2016) and the increased functional connectivity in somatomotor networks after music training (Li et al. 2018), which might relate to the auditory–motor interaction, while the significant interaction effect of age × music training indicated that music training mitigates the functional compression of the somatomotor network due to aging. Actually, research on speech in noise has reported the functional connectivity changes between auditory and speech motor regions as well as frontal speech motor cortices, which involved in sensorimotor integration (Du et al. 2014; Du and Zatorre 2017). Therefore, we consider that the closer gradient distribution of the right dorsal and medial prefrontal regions and the bilateral somatomotor regions on the principle gradient axes in older adults reflects the decreased functional segregation (Bethlehem et al. 2020). Correlation results with age displayed changing trends in principal gradient scores in the bilateral somatomotor network and in the gradient distribution distance of the two regions on principle gradient axes. Accordingly, music training intervenes in the aging process results in further gradient distribution of the two regions by influencing the principal gradient score of the bilateral somatomotor network. That is why the gradient distribution pattern of the two regions in the older musicians is closer to that of young adults than that of the older control.

Furthermore, we explored functional connectivity based on the significantly different regions of principal functional gradient between young and older subjects. The results showed the different functional connection patterns of normal aging as well as aging under music intervention. We found that the functional connectivity from the right dorsal and medial prefrontal regions in older control was higher than that in young control. As we mentioned above, the closer gradient distribution indicated decreased functional segregation and resulted in stronger functional connectivity. Higher connectivity of older control in the results provided evidence for this view. Meanwhile, higher functional connectivity in older adults was also in line with the result of our previous study that aging lead to stronger early sensory-evoked responses and stronger neural synchrony between prefrontal and sensory cortices (Chao and Knight 1997; Alain et al. 2022). We thought that the higher connectivity revealed a higher demand for the top-down control of prefrontal to somatomotor regions, which may be due to the worse perceptual function (Hutchinson et al. 2012; Füllgrabe et al. 2014) in the elderly. The stronger connectivity between the prefrontal and somatomotor regions also could explain the phenomenon that perceptual and cognitive losses are increasingly common together as aging (Kluger et al. 1997; Gates et al. 2011; Lin et al. 2013; Albers et al. 2015; Kondo et al. 2020; Danka Mohammed 2021). The perceptual decline is worse in people with cognitive impairments such as Alzheimer's disease (AD) (Albers et al. 2015). Take hearing as an example, age-related deafness is more prevalent among people with AD than in healthy control, and in some extreme cases, older subjects with impaired hearing are at greater risk of developing cognitive impairment than those with normal hearing (Pichora-Fuller et al. 2017). Therefore, the disappearance of functional connectivity differences between young and older musicians suggested the effect of music training on aging that enhances the functional segregation between the right dorsal and medial prefrontal regions and the bilateral somatomotor regions.

As we mentioned in the method, SFC analysis is a complementary method which allows researcher to understand gradient as a sequence of steps in connectivity space. The two methods both exhibit the connectivity evolution between primary sensory and DMN systems (Margulies et al. 2016; Huntenburg et al. 2018; Hong et al. 2019; Bethlehem et al. 2020; Xia et al. 2022). After calculating the SFC degree from the prefrontal to somatomotor regions, we found a negative correlation between the SFC degree and functional gradient distribution distance. It provided evidence that the SFC degree could characterize the hierarchy relationship between the two regions, such as functional gradient. Meanwhile, as reported in the previous study, enhanced connectivity of the frontal–parietal control system suggests the decreased specialization in brain function (Chan et al. 2014), which is associated with age. Thus, stronger functional connectivity represents a decrease in the specialization of the two regions, which is associated with poorer functional separation due to gradient compression. However, changes in distinct networks may cascade multiform changes to other networks in the human brain (Costumero et al. 2020). Therefore, the issue of concern is that although we demonstrated the top-down control of the right dorsal and medial prefrontal regions exerted over the somatomotor network, but we have less understanding of the complex connectivity transitions that take place from the high-order cognitive system to the primary system. SFC analysis, a method that could untangle the functional connectome and transitions from PFC to somatomotor cortices, was used to analysis the effect of music intervention on connectivity transitions (Sepulcre et al. 2012). We found that SFC degree from prefrontal to somatomotor regions in step 3 showed a significant difference between young and older subjects when the subjects were divided into two groups regardless of music training experience. When subjects were divided into four groups, the difference in step 3 between young and older musicians disappeared. The correlation with age showed that music training could slow age-related increase in SFC degree of step 3. The deviation suggests that music training influences connectivity patterns at short functional distances. SFC degree between sensory seeds and higher order integration nodes at short functional distances is related to attention deficit (Pretus et al. 2019). It has also been demonstrated that alterations of the fronto-parietal network and the DMN affect attention (Silk et al. 2008; Castellanos and Proal 2012; Lin et al. 2015). Inhibitory control is considered as the central neurocognitive process in attention deficit (Alderson et al. 2007; Wright et al. 2014; Fosco et al. 2019). Meanwhile, prefrontal regions have been proposed to play an essential role in inhibitory control (Kilavik et al. 2013; Aron et al. 2014) and ignoring distracting information (Hasher et al. 1997). Therefore, together with our correlation results between the accuracy of 2-back and principal gradient scores, we suggested that the connectivity transitions between prefrontal and somatomotor regions at short functional distances are a potential mechanism for music to intervene in cognitive aging.

In addition, although the results are marginally significant, it is interesting to note that the later the age of music training starts, the greater SFC degree of musicians, the lower the functional gradient score of prefrontal regions, and the lower accuracy of 2-back. Previous studies have mentioned that the age at which training begins affects neural development (Steele et al. 2013; Bailey et al. 2014; Baer et al. 2015). Thus, we suggested that the age of music training starts should be considered as a potential influencing factor in music intervention on inhibitory function in the elderly. Previous studies have suggested that the right PFC is specific for inhibition (Aron et al. 2003; Volle et al. 2012; Hornberger and Bertoux 2015). Modulation of the right dorsal and medial prefrontal regions on somatomotor networks in our results further demonstrates the right lateralization of prefrontal regions in inhibitory function in working memory.

Conclusion

The present study demonstrates that the ability of inhibitory control over the information of working memory in healthy aging is associated with functional gradient distribution distance between the prefrontal and somatomotor regions. Through music training, the trend of indicator change is altered in a direction closer to that of young adults. The mechanism behind it may be the connectivity transitions between prefrontal and somatomotor regions at a short functional distance. Correlation results further suggest the potential effect of the age of music training start. In a nutshell, we demonstrate the top-down control of prefrontal regions to the somatomotor network, which is associated with inhibitory function and represents a potential marker of cognitive aging, and reveal that music training may work by affecting the connectivity between the two regions. Although this work has investigated the neuroplasticity of music on cognitive aging by recruiting subjects of different age spans, the present study did not include the study of longitudinal changes of the same group. Further studies should include longitudinal follow-up of the same groups over time to more accurately evaluate the effect of music intervention on the process of cognitive aging.

Acknowledgments

We would like to thank Ying Liu, Junchen Zhou, and Yan Liu, who helped with subject recruitment and the experiment preparation. We also thank all the participants in our study.

Data availability

The data underlying this article will be shared on reasonable request to the corresponding author.

Funding

STI 2030 (2022ZD0208500); Sichuan Science and Technology Program (2021YFS0135).

Conflict of interest statement: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as apotential conflict of interest.

Author notes

Jing Lu and Dezhong Yao contributed equally to the article.