An ALE meta-analytic review of musical expertise

Through long-term training, music experts acquire complex and specialized sensorimotor skills, which are paralleled by continuous neuro-anatomical and -functional adaptations. The underlying neuroplasticity mechanisms have been extensively explored in decades of research in music, cognitive, and translational neuroscience. However, the absence of a comprehensive review and quantitative meta-analysis prevents the plethora of variegated findings to ultimately converge into a unified picture of the neuroanatomy of musical expertise. Here, we performed a comprehensive neuroimaging meta-analysis of publications investigating neuro-anatomical and -functional differences between musicians (M) and non-musicians (NM). Eighty-four studies were included in the qualitative synthesis. From these, 58 publications were included in coordinate-based meta-analyses using the anatomic/activation likelihood estimation (ALE) method. This comprehensive approach delivers a coherent cortico-subcortical network encompassing sensorimotor and limbic regions bilaterally. Particularly, M exhibited higher volume/activity in auditory, sensorimotor, interoceptive, and limbic brain areas and lower volume/activity in parietal areas as opposed to NM. Notably, we reveal topographical (dis-)similarities between the identified functional and anatomical networks and characterize their link to various cognitive functions by means of meta-analytic connectivity modelling. Overall, we effectively synthesized decades of research in the field and provide a consistent and controversies-free picture of the neuroanatomy of musical expertise.


Results
A total of 1169 records was identified through database searching, and 679 records were initially screened by title and abstract after removing duplicates. Next, 145 articles were assessed for eligibility in the full-text screening stage. From these, 84 studies fulfilled criteria for eligibility and were thus included in the qualitative synthesis. Finally, from the 84 studies, only 58 reported results in stereotactic coordinates (foci), either Talairach or Montreal Neurological Institute (MNI) three-dimensional-coordinate system which were therefore included in the quantitative synthesis (ALE meta-analyses) ( Supplementary Fig. 1). This study used the BrainMap platform 36 , a large-scale database storing results obtained from published brain activation (functional) and brain structure (voxel-based morphometry) studies. ALE meta-analyses were conducted using GingerALE 37 , a software that tests the convergence of coordinates across independently conducted MRI studies investigating the same construct. To complement our findings, meta-analytic connectivity modelling (MACM) 38 was performed using Sleuth 39 , to explore the co-activation patterns of regions-of-interest (ROIs) resulted from the ALE meta-analysis, and to functionally segregate each region's contribution to behavioural domains and paradigm classes according to the BrainMap platform 40-42 . Characteristics of studies. Details of the studies included in our work are provided in Table 1. Eighty-four publications met inclusion criteria and were included in the qualitative synthesis which was comprised of 3005 participants, with 1581 musicians (M) and 1424 non-musicians (NM). Eighteen studies (21%) included amateur musicians, and only 7 studies (8.3%) reported absolute pitch possessors (n = 97). Musical instruments were reported in most of the studies (81%): piano or keyboard (62%), string instruments (41%), wind instruments (26%), percussion instruments (17%), voice (8%), and 19% studies failed to report musicians' instrument. Years  ----------Pro  Y   5  Angulo-P  2014 fMRI  Timbre  28 25  -29  24  53  a,b  28  7  29  9  --------- ---------Pro  Y   50 Liu  2018 fMRI  Tempo + emotion  21 27  -23  25  48  ---------------Pro  Y   51 Matsui  2013 fMRI  Musical structure  12 15  -0  27  27  a  29.1  -23.4  ---------- MRI quality. MRI quality of the included studies in the meta-analysis was assessed following a set of guidelines for the standardized reporting of MRI studies 43,44 . All studies included in the qualitative synthesis (n = 84) reported their MRI design, software package and image acquisition, pre-processing, and analyses. Overall, all the studies followed good MRI practices. Neuroimaging data was acquired in either 1.5 T (39%), or 3 T (56%) scanners, while 5% of studies did not report the magnetic field strength. MRI scanners included Siemens (40%), General Electric (25%), Philips (21%), Bruker (5%), while 7% did not report it. Analysis methods included fMRI (52%), VBM (29%), DTI (18%), and CT (10%). Finally, 84% of studies included GM analyses, while 23% included WM analyses (Supplementary Tables 1, 2, and 3). Structural studies. The structural ALE meta-analysis included 33 experiments and 1515 participants. The contrast M > NM in GM resulted in significant peak clusters in the bilateral superior temporal gyrus (primary auditory cortex), including the bilateral Heschl's gyrus and planum temporale, and the postcentral gyrus (somatosensory cortex, SI), including area 4a of the primary motor cortex (M1 4a). Conversely, the comparison NM > M in GM resulted in a significant peak cluster located in the right precentral gyrus (primary motor cortex, M1). In WM, musicians showed larger tracts of the internal capsule bundle (extending to the thalamus) and corticospinal tract. No significant clusters were identified in the comparison NM > M for WM ( Fig. 1, Table 2).     Table 2).

Meta-analytic connectivity modelling (MACM).
MACM was performed to functionally segregate the behavioural contribution and the patterns of co-activation of each music-related region-of-interest (ROI) resulted from the structural (n = 5) and functional (n = 5) ALE meta-analyses. Five-millimetre ROIs were created using Mango 41 and imported into the BrainMap 42 database separately using Sleuth 39 . Foci from each identified study were extracted and a secondary GingerALE meta-analysis was performed aiming to identify the functional network of each ROI, namely its functional connectivity (FC) 38 (Supplementary Table 4). Finally, the functional characterization of each ROI was described using Sleuth and focused on behavioural domains (e.g., action, perception, emotion, cognition and interoception) and paradigm classes (e.g., pitch discrimination, finger tapping, music comprehension, go/no-go) included in the BrainMap platform (Supplementary Table 5).
Structural ROIs. The right superior temporal gyrus ROI (Fig. 2a) showed co-activation with left superior temporal gyrus, right precentral gyrus, left medial frontal gyrus, left cerebellum, and left thalamus. Relevant behavioural domains within its boundaries include execution, speech, memory, music, emotion, reward, and auditory perception; and experimental paradigms including emotion induction, finger tapping, music comprehension and production, passive listening, reasoning/problem solving, and phonological, pitch, semantic, syntactic, and tone discrimination. The left superior temporal gyrus ROI (Fig. 2b) showed co-activation with right superior temporal gyrus, right insula, right inferior frontal gyrus, bilateral precentral gyrus, and left medial frontal gyrus. Relevant behavioural domains within its boundaries include execution, speech, motor learning, attention, language, speech, memory, music, emotion, and auditory perception; and experimental paradigms including emotion induction, finger tapping, music comprehension and production, passive listening, reasoning/problem solving, visuospatial attention, and oddball, orthographic, phonological, pitch, semantic, and tone discrimination.
The right postcentral gyrus ROI (Fig. 2c) showed co-activation with left medial frontal gyrus, left parietal lobule, right middle frontal gyrus, right thalamus, right superior temporal gyrus, and right cerebellum. Relevant behavioural domains within its boundaries include execution, motor learning, attention, respiration regulation, Figure 1. Anatomic likelihood estimation meta-analytic results for studies comparing brain structure and function between M and NM at cluster level inference p < 0.05 (FWE). The primary outcome included ALE meta-analysis of the contrast M vs NM for structural and functional modalities, independently. M > NM = higher volume/activity in musicians; NM > M = lower volume/activity in musicians; GM grey matter, WM white matter, L left, R right, A anterior, P posterior, Z peak Z-value, IC internal capsule, INS insula, IPL inferior parietal lobule, PostCG postcentral gyrus (primary somatosensory cortex, or S1), PreCG precentral gyrus (primary motor cortex, or M1), STG superior temporal gyrus (primary auditory cortex). www.nature.com/scientificreports/ and auditory perception; and experimental paradigms including finger tapping, passive listening, visuospatial attention, and oddball, tactile, and tone discrimination. The right precentral gyrus ROI ( Fig. 2d) showed co-activation with claustrum and insula. Relevant behavioural domains within its boundaries include execution, attention, speech, temporal processing, and emotional processing; and experimental paradigms including finger tapping and visuospatial attention.
The right internal capsule ROI (Fig. 2e), including the right thalamus as the nearest grey matter, showed coactivation with left thalamus, right medial frontal gyrus, right insula, and left cerebellum. Relevant behavioural domains within its boundaries include execution, speech, attention, memory, reasoning, emotion, reward, and auditory perception; and experimental paradigms including emotion induction, finger tapping, passive listening, reward, and tone discrimination.  Relevant behavioural domains within its boundaries include execution, speech, attention, language, memory, music, reasoning, social cognition, emotion, and auditory perception; and experimental paradigms including encoding, finger tapping, music comprehension and production, passive listening, reasoning/problem solving, reward, and phonological, semantic, tactile, and tone discrimination. The right superior temporal gyrus ROI (Fig. 3b) showed co-activation with bilateral inferior frontal gyrus, bilateral inferior parietal lobule, left medial frontal gyrus, left fusiform gyrus, and caudate. Relevant behavioural domains within its boundaries include execution, speech, attention, language, memory, music, reasoning, social cognition, emotions, and auditory perception; and experimental paradigms including finger tapping, music comprehension and production, passive listening, reasoning/problem solving, reward, theory of mind, and phonological, semantic, tactile, and tone discrimination.
The left superior temporal gyrus ROI (Fig. 3c) showed co-activation with right superior temporal gyrus, right middle temporal gyrus, right claustrum, left insula, and left medial frontal gyrus. Relevant behavioural domains within its boundaries include execution, speech, attention, language, memory, reasoning, social cognition, emotion, and auditory perception; and experimental paradigms including divided auditory attention, emotion induction, emotional body language perception, encoding, finger tapping, music comprehension and production, reasoning/problem solving, theory of mind, visuospatial attention, and oddball, phonological, pitch, semantic, and tone discrimination.
The left inferior parietal lobule ROI (Fig. 3d) showed co-activation with left medial frontal gyrus, right inferior frontal gyrus, and left precentral gyrus. Relevant behavioural domains within its boundaries include execution, speech, motor learning, attention, memory, music, reasoning, social cognition, emotion, and auditory perception; and experimental paradigms including emotion induction, finger tapping, motor learning, reasoning/problem solving, reward, visuospatial attention, and phonological, semantic, tactile and tone discrimination.
The left precentral gyrus ROI (Fig. 3e) showed co-activation with left precuneus, left superior frontal gyrus, right inferior frontal gyrus, right inferior parietal lobule, right claustrum, left fusiform gyrus, left thalamus, and right middle frontal gyrus. Relevant behavioural domains within its boundaries include execution, speech, motor learning, attention, language, memory, music, reasoning, social cognition, temporal processing, emotion, sleep, and auditory perception; and experimental paradigms including divided auditory attention, emotion induction, encoding, finger tapping, music comprehension, reasoning/problem solving, reward, theory of mind, visuospatial attention, and oddball, phonological, pitch, semantic, tactile, and tone discrimination.

Discussion
The link between musical expertise and humans' cognitive functions has been explored with great interest since the times of Pythagoras. Recent years reveal a renewed and more than vivid attention to the topic, as reflected in the rising number of empirical research in the past half-century 1,2 . Decades of investigations in psychology, cognitive and translational neuroscience have attempted to foster our understanding of the neurocognitive processes underlying musical expertise. Thus, long-term musical training has been associated with neuro-anatomical and -functional specializations in brain regions engaged in multimodal (audio-visual) sensory and sensorimotor perception, integration and predictions as well as fine movement control 17,18 . Furthermore, the duration and intensity of training has been associated with improvements in general cognition, ranging from working memory, intelligence, executive functions and inhibitory control 6,[9][10][11] . However, as mentioned before, this rapidly growing field of research is also characterized by some methodological inconsistencies (e.g., sample differences and neglected background variables), and sometimes shows discrepant results and controversial interpretations of the findings. Such limitations, alongside with the absence of a meta-analysis, has prevented the plethora of variegated findings to ultimately converge into a unified picture of the neuroanatomy of musical expertise.
To address this lack in the literature, we performed a comprehensive and quantitative meta-analysis of neuroanatomical and -functional studies investigating brain changes associated with long-term musical training. Our coordinate-based anatomic/activation likelihood estimation (ALE) meta-analysis effectively summarizes decades of research in the field and finally provides a consistent and controversies-free picture of the core brain regions engaged in and influenced by long-term music processing and production. To better characterize the emergent neural network of musical expertise, we performed meta-analytic connectivity modelling analyses (MACM) and functionally linked each node of the music network to specific cognitive functions. By discussing the main results of the meta-analysis alongside with the observations derived from MACM, we ultimately provide a comprehensive view of the anatomical, functional, and cognitive substrates of musical expertise. This Co-activated areas: CAU caudate; claustrum, CRBL cerebellum, FusG fusiform gyrus, IFG inferior frontal gyrus, IPL inferior parietal lobule, INS insula, MedFG medial frontal gyrus (pre-motor), MidFG middle frontal gyrus (pre-frontal), PostCG postcentral gyrus (primary somatosensory cortex or S1), PreCG precentral gyrus (primary motor cortex or M1), PCN precuneus, PUT putamen, SFG superior frontal gyrus, STG superior temporal gyrus (primary auditory cortex), THA thalamus. To conduct MACM, music-related ROIs were created in Mango (http:// rii. uthsc sa. edu/ mango// userg uide. html) with a 5 mm-radius sphere. For visualization purposes, the music-related ROI radius was increased to 10 mm, while co-activated areas were created with a 5 mm-radius sphere. Last search in Sleuth, 10.10.2021 (http:// www. brain map. org/ sleuth/). www.nature.com/scientificreports/ discussion is organized in three main paragraphs: 'the ear' , 'the body' and 'the heart' , elaborating on the emergent fronto-temporal, sensorimotor and interoceptive networks respectively. Notably, MACM further allows us to strengthen the notion that musical training represents a stimulating multisensory experience engaging not only sensory and motor functions strictly related to acoustic and motor processes, but a wide variety of high-order cognitive functions from working memory, attention, executive functions and emotional regulation 11,[16][17][18]45 .
To conclude, we argue that the observed music-related neuroanatomical and -functional changes represent an interface between nature and nurture effects. Namely, gene-environment interactions and other background variables likely interacted with brain maturation processes ultimately influencing the neuroplasticity mechanisms responsible for the observed training-specific neuroanatomical and -functional changes.
Characteristics of the included studies. The publications included in this systematic review and metaanalysis reported a clear research question, inclusion and exclusion criteria for participants, description of methods and explicit results. Most of the studies used state-of-the-art techniques and computational MRI-tools, important for the support of standardization and reproducibility of neuroimaging studies. However, some of the studies lacked important demographic data such as the years of education, age of musical training onset, and current time of musical practice, which may influence behavioural tasks and neuroimaging data. Thus, our research encourages to adopt in future studies standardized tools specifically designed and validated for assessing musical expertise 46 .
Structural and functional neuroplasticity in musical expertise. Our results highlight that expert musicians exhibited higher GM volume in the bilateral superior temporal gyri and right postcentral gyrus and greater WM volume in the right internal capsule bundle and corticospinal tract, as compared to non-musicians. Additionally, musicians exhibited higher activity of the bilateral superior temporal gyri, left inferior frontal gyrus, left precentral gyrus, and left insula. On the other hand, musicians had lower GM volume in areas of the sensorimotor cortex and no WM structure was found to have larger volume in non-musicians as compared to musicians. Finally, musicians exhibited lower neurofunctional activation of the inferior parietal lobule and motor cortex during a variety of cognitive tasks.
The ear: enhanced frontotemporal auditory network in musicians. One of our main findings shows enlargement of GM volume in musicians located in medial and posterior superior temporal regions, with clusters extending into primary and secondary auditory cortices. These regions include neuronal assemblies dedicated to encoding of spectro-temporal features of sounds relevant to music 47 , such as the discrete pitches forming the Western chromatic scale and fine changes in pitch intervals 48 . More specifically, it seems that the posterior supratemporal regions are more involved in encoding the height of pitch, whereas the anterior regions are representing the chroma, that is the pitch category irrespectively of the octave 49 . Moreover, these areas participate in auditory imagery of melodies 50 and in the processing of the contour and Gestalt patterns of melodies, allowing for recognition and discrimination of mistakes 51 . Beyond music-related functions, functional characterization analyses of our ROIs (Supplementary Table 5) show that superior temporal regions are usually recruited for phonological processing and multimodal integration of sensory information. Accumulating evidence has shown that the superior temporal sulcus and posterior superior temporal gyrus, together with early auditory regions (HG), are involved for the processing of speech sounds, abstract representation of speech sounds, as well as more general language, phonology and sematic processing and audio-visual integration. Therefore, temporal regions seem to represent fundamental structures for both language and music processing 52 . MACM further revealed that auditory cortices tend to co-activate with insula, (pre)motor regions, inferior and medial frontal gyri, thalamus and cerebellum, confirming the relevance of extended cortico-subcortical audio-motor coupling for rhythm processing in language and music [53][54][55][56] . Further supporting this view, MACM showed that the inferior frontal gyrus co-activates with motor areas in the cortex and the basal ganglia (see next paragraph, 'The body'), and with parietal areas related to the dorsal auditory pathway.
The inferior frontal gyrus has been described as an important hub of both the dorsal and ventral auditory streams. The dorsal auditory stream connects the auditory cortex with the parietal lobe, which projects in turn to the inferior frontal gyrus pars opercularis (Brodmann area 44). The inferior frontal gyrus has been related to the articulatory network, dedicated to specific functions of speech comprehension and production, and highly connected to premotor and insular cortices 57 . The ventral auditory stream connects the auditory cortex with the middle temporal gyrus and temporal pole, which in turn connects to the inferior frontal gyrus pars triangularis (Brodmann area 45). This area has been associated with semantic processing 58 . These two regions within the inferior frontal gyrus constitute Broca's area. The supramarginal gyrus is also a relay of the dorsal auditory stream involved in processing of complex sounds, including language and music 59 . As such, it is considered an integration hub of somatosensory input 60 .
The parietal lobe has been also described as an integration area of sensory inputs. The superior parietal lobule includes Brodmann areas 5 and 7, which are involved in somatosensory processing and visuomotor coordination, respectively. The inferior parietal lobule includes Brodmann areas 39 and 40, the angular gyrus and supramarginal gyrus, respectively. The angular gyrus has been related to projection of visual information to Wernicke's area, memory retrieval and theory of mind 61 . MACM revealed that the parietal lobe co-activates with sensorimotor cortices and the inferior frontal gyrus.
The body: enhanced sensorimotor functions in musicians. The precentral and postcentral gyri represent the primary motor and somatosensory cortex, respectively. These two areas are divided by the central sulcus, whose www.nature.com/scientificreports/ extension represent the sensation and motion of segregated body parts. Our findings show both convergent and divergent effect of musical training in these areas, suggesting a more complex picture than previously thought. For example, neuroadaptations in the sensorimotor system may vary depending on the musical instrument of use 62 . MACM revealed that the primary motor cortex co-activates with an extensive network that includes the frontal pole, limbic areas such as the anterior cingulate cortex and insula, and parietal areas such as the precuneus. It also revealed that the primary somatosensory cortex co-activates with motor and pre-motor areas, basal ganglia, thalamus, and the cerebellum. A dedicated temporal processing network has been described by Kotz and Schwartze 54 including such areas, which are important for implementing sequential actions, as well as to form predictions about the timing of external events. Healthy motor performance relies on a functional loop established by the basal ganglia and supplementary motor area that maintains adequate preparation for sequential movements. The supplementary motor area prepares for predictable forthcoming movements, keeping the system "ready". Once the movement starts, the supplementary motor area's readiness activity stops. This cycle engages with BG discharges after each sub-movement within an automatized sequence 63 . The loop requires an internal cue to coordinate the cycle. The basal ganglia are nuclei of neurons important for the initiation and suppression of movements. In the motor loop of the basal ganglia (BG), inputs from motor cortices project to the dorsal striatum, composed by the putamen and caudate. In the presence of adequate dopaminergic signalling, the 'direct pathway' (cortexstriatum-internal pallidum-thalamus-cortex) works to facilitate movement, while the 'indirect pathway' suppresses it (cortex-striatum-external pallidum-subthalamic nucleus-thalamus-cortex). Zooming into inhibitory processes, the striatum transiently inhibits the pallidum, and in turn, the motor area of the thalamus is disinhibited and is free to project back to the motor cortex, initiating a motor program that flows down the corticospinal tract. Similarly, the subthalamic nucleus in the indirect pathway is transiently inhibited when suppressing movement, increasing the inhibition of the pallidum over the thalamus, therefore blocking the motor cortex activity 64 .
Our findings show neuroadaptive processes in the putamen and caudate of musicians (striatum), presumably reflecting effective inhibitory mechanisms as seen by fine movement control. Furthermore, our findings strengthen the notion that basal ganglia circuits are involved in motor sequence learning, and in particular in the learning and control of fine-movement sequences acquired through music practice 65,66 .
The cerebellum has been shown to play a crucial role in multiple cognitive processes such as sensory discrimination, rhythmic perception and production, working memory, language, and cognition 67 . Previous fMRI studies in humans suggest that the cerebellum shows segregated activations for motor and cognitive tasks. Motor tasks seem to activate lobules IV-VI in the superior parts of the anterior cerebellum. In contrast, attentional or working memory tasks activate posterior cerebellar hemispheres, namely lobule VIIA, which is divided to crus I and crus II, as well as lobule VIIB 68 . Musicians and non-musicians show GM volume differences in the cerebellum, specifically in area Crus I. In our study, this area did not survive correction for multiple comparisons, however MACM revealed that the cerebellum is functionally connected to auditory cortices, somatosensory cortices, and the thalamus. It has been demonstrated that the activity in crus I/II has a specific relationship with cognitive performance and is linked with lateral prefrontal areas activated by cognitive load increase 69 . In other words, the crus I/II seems to optimize the response time when the cognitive load increases. Additionally, it has been suggested that crus I/II is associated with beat discrimination thresholds. Thus, there is a positive correlation between GM volume in crus I and beat discrimination performance, evidenced by enhanced ability in musicians 70 .
The heart: enhanced interoceptive areas in musicians. Among the other results, our meta-analysis reported higher functional activation of left insula in musicians as compared to non-musicians. MACM analyses reported the left insula in a functional network that connects inferior frontal gyrus with precentral gyrus, middle frontal gyrus and parietal lobule bilaterally (Supplementary Table 4).
It has been proposed that the insula and the anterior cingulate cortex (ACC) are part of the salience network, and coordinate interactions between the default-mode network and the central executive network 71 . The ACC has been related to cognitive and emotional processing. The cognitive component projects to prefrontal, motor, and parietal areas to process top-down and bottom-up stimuli. The emotional component features connections from the insula to amygdala, nucleus accumbens, hypothalamus and hippocampus, with the scope to assess the salience of emotional and motivational information 72 . Moreover, the insula integrates information from the internal physiological state, and projects to the ACC, ventral striatum and prefrontal cortex to initiate adaptive responses 73 . Thus, enhanced function of these areas after musical training may be associated with a more efficient coordination between interoceptive, emotional, salience and central executive networks.
White matter. M exhibited larger clusters of WM as compared to NM in the internal capsule and cortico-spinal tract. While previously thought to be rather passive tissues, WM tracts are now consistently associated with an active modulatory role in information flow between brain regions 64 . Indeed, myelin regulates the speed of action potential transfer within and between GM structures and further provide metabolic support to local neural cells. WM changes are commonly observed during learning and associated with fast, accurate and coordinated motor sequences 27 .
The internal capsule is a WM structure which connects basal ganglia regions and carries information from and to surrounding cerebral cortex. Connecting fibres in basal ganglia might be thickened by musical expertise because of their involvement in motor control, rhythmic processing, sequence learning, reinforcement learning and memory processes 65 . In general, basal ganglia structures are recruited during working memory processing for musical motifs 74 and the most ventral regions are a core structure of the reward circuit. Interestingly, they are found to be more active in musicians as compared to non-musicians while listening to expressive music 75  www.nature.com/scientificreports/ The corticospinal tract allows the motor plans originated in the cortex to be transferred to motor nuclei in the spinal cord and to finally regulate the activity of muscle effectors. The myelination and integrity of the corticospinal tract has been observed to be increased in expert musicians 27,28 , and is further influenced by the time of onset of musical practice 31 with early onset musician showing the greatest axial diffusivity.
Structural connectivity analyses comparing M vs NM are scarce in the literature with high variability of designs, methods, and results 76 . However, it is suggested that neuroadaptive effects of musical expertise relies on effective structural and functional communication between cortical and subcortical sensorimotor areas through thalamic radiations and the internal capsule. Moreover, differences in the corpus callosum connecting both hemispheres have been reported in studies using DTI 77,78 , which may reflect the bimanual coordination and related inter-hemispheric connections required for playing most musical instruments. Notably, such differences appear to be more salient in musicians that started musical training before the age of seven 27,79,80 . Taken together, our results and previous results suggest that the acquisition of musical skills will develop structural and functional connections between auditory, sensorimotor, timing, and reward areas of the brain reflecting the network-like nature of the human brain.
(Dis)similarities between anatomical and functional studies. The meta-analyses on neuroanatomical and -functional changes coherently show greater GM volumes and increased functional engagement of superior temporal gyrus bilaterally, together with pre-and post-central gyri in expert musicians as compared to laypersons. Functional studies further agree on the pivotal involvement of left inferior frontal gyrus (BA9, BA44) along with superior temporal gyrus bilaterally in musicians. However, dissimilarities emerge when looking at pre-and post-central regions: right precentral gyrus (right primary motor cortex (M1)) is reduced in musicians, while the right postcentral gyrus [right primary somatosensory cortex (S1)] is increased. Functional studies show, instead, that there is increased activity in left precentral gyrus in the inferior frontal gyrus-superior temporal gyrus-insula network, and reduced activity of the left precentral gyrus in a cluster which extends into the left parietal lobule.
While results pertaining to the frontotemporal auditory network and the sensorimotor network have been discussed in 'The ear' and 'The body' paragraphs above, we here speculate that the enlargement of S1 in musicians is associated with a more sophisticated representation of the sensorimotor periphery 19 and that the increased left inferior frontal gyrus-precentral gyrus-superior temporal gyrus-insula activation, at the expense of the M1-parietal lobule network may be related to the acquisition of accurate and automatized motor programs in musicians 4 . This view is further corroborated by the connectivity observed between primary somatosensory cortex, motor and pre-motor areas, basal ganglia, thalamus, and the cerebellum in the MACM analyses. In agreement with early studies, we lastly argue that the hemispheric asymmetry may be related to the music instrument played and the dominant hand of the musicians 81 , but interhemispheric transfer effects are possible with motor sequence learning 82 . However, longitudinal studies should further elucidate on the heterogeneity of structural and functional adaptations associated with intensive and long-lasting motor training.
Limitations and future perspectives. This comprehensive review and meta-analysis had the scope to summarize decades of research investigating neuro-anatomical and -functional changes associated with musical expertise. Our qualitative review highlights that previous studies in this field are characterized by heterogeneity of methods, paradigms, and sample backgrounds, as well as relevant missing information. While arguing that the field will benefit from more clarity (e.g., thorough description of methods) and consistency, we also delineate limitations for our meta-analysis. For example, we set a contrast based on the comparison M vs NM with the aim to narrow down the heterogeneity of the sample and methods in use. However, by doing so we relied on two assumptions: (1) the data we pool is based on best research practices; (2) the validity of the GingerALE method. Indeed, to conduct the ALE meta-analysis, we pooled peak coordinates derived from the included studies, rather than using original raw structural MRI images. Thus, the accuracy of our findings relies on the result of a statistical estimation of coordinate-based anatomic foci (input), treated as spatial probability distributions centred at the given coordinates. The heterogeneity of the methods in use in previous studies (ranging from preprocessing software, smoothing, statistical thresholds and participants' characteristics) are not under our control and represent potential confounders for the results. Perhaps a regression-based assessment of the influence of those heterogenous factors on the findings would sharpen the results. However, meta-regression analysis is not compatible with GingerALE. When assessing publication bias using the Fail Safe-N analysis, we found adequate robustness of our results, with only 2 ROIs showing an FSN below of the minimum imposed in each of the ALE within contrasts (BA2, BA4 in the structural ALE and BA22, BA6 in the functional ALE), thus, indicating an overall robust convergence of foci our study (further information is reported in Supplementary Table 6).
Lastly, on a more theoretical perspective, our results contribute but do not solve the long-standing "nature vs nurture" debate. Indeed, based on evidence that musical training stimulates higher-cognitive functions, auditory-motor integration, attention, memory and engages reward networks, some have suggested that it may be particularly effective in driving neuroplastic mechanisms 78 . However, we are indeed blind to whether the highlighted differences emerging when comparing M vs NM are training-dependent or due to innate predispositions. Altogether, the most reasonable conclusion is that the observed neuro-anatomical and -functional changes may be attributed to the interaction between brain maturation processes and gene-environment interactions 13,50,85 . Notably, multiple studies demonstrated a strong correlational link between the length of musical training and neuroanatomical and -functional changes 83,84 . For instance, the study conducted by Gaser and Schlaug 85 reported that amateur musicians showed an intermediate increase in gray matter volume when compared to NM and M, supporting the idea of use-dependent structural changes. The same pattern was found when comparing cognitive abilities, with amateurs showing higher cognitive abilities than NM, but lower than M 11 . To be noted, however, www.nature.com/scientificreports/ this research field suffers of the paucity of longitudinal (f)MRI studies conducted with children, which thus far amount only to seven [4][5][6][7][8]86,87 , next to one 15-week long study in adults 88 . Longitudinal studies are the only ones promising to better elucidate on the causal link between musical training and neural adaptations. Our work, on the other hand, pools a large quantity of anatomical and functional MRI studies conducted over > 20 years of world-wide research. By doing so, it bears the potential to achieve an unprecedented signal-to-noise ratio, so to filter out the mediating influence of background, psychological and other environmental factors, and to effectively isolate music-related neuroplastic changes. Thus, we here provide, within the delineated limits, a consistent view of the neuroanatomy of neural expertise. Furthermore, we explore the connections and functions of the brain areas that appear to be key in the acquisition of musical skills. Such regions include auditory, limbic, and sensorimotor regions that reflect the network-like nature of the human brain. We hope our work would better inform future basic and comparative research in the field of auditory and cognitive neuroscience and that we encouraged translational approaches bridging to the clinical field 89,90 .

Conclusions
The neuroanatomical and functional changes observed in the musician's brain have been repeatedly regarded as the ideal scenario to investigate neuroplastic mechanisms. Yet, decades of research in cognitive neuroscience have provided a scattered and partially controversial series of findings. The present coordinate-based metaanalysis represents a comprehensive and quantitative attempt to summarize existing literature and provide a unified picture of the neuroanatomy of musical expertise. We show that music experts exhibit bilateral corticosubcortical neuroanatomical and -functional differences as compared to laypersons. This systematic review and meta-analysis strengthens the view that musical training represents a beneficial and stimulating multisensory experience which engages a wide variety of neurocognitive functions.

Methods
Literature search, screening, and extraction. This systematic review and meta-analysis followed procedures from the Cochrane Handbook for Systematic Reviews 91 and from the Centre for Reviews and Dissemination (Centre for Reviews and Dissemination, 2014). The review protocol was registered with PROSPERO No.
[CRD42017060365]. This review was carried in accordance with the PRISMA statement 92 . Systematic search was performed using PubMed, PsycInfo and Scopus, of publications that reported brain structural or functional differences between M and NM. The search (March 2021) included MeSH terms ("music", "education", "brain", "motor skills", "magnetic resonance imaging") and key words ("musical training", "musician"). No years or places of publication were imposed.
For qualitative synthesis, studies were included if they met the following criteria: (1) studies comparing brain structure and function between musicians and non-musicians, (2) in adult population, (3) by means of magnetic resonance imaging, in either structural modality (e.g., voxel-based morphometry [VBM]) or functional modality (e.g., functional magnetic resonance imaging [fMRI]). For the final quantitative synthesis (meta-analysis), studies were included only if the results were reported in stereotactic coordinates either Talairach or Montreal Neurological Institute (MNI) three-dimensional-coordinate system.
Two reviewers (AC and VP) independently screened by title and abstract and selected articles for full-text review and performed full-text reviews. Screening and data extraction were performed using the Covidence tool 93 . Any disagreements that arose between the reviewers were resolved through discussion or by a third and/ or fourth reviewer (LB, EB).
From each study, the following variables were extracted: first author, year of publication, population of interest, number of participants, age, sex, absolute pitch, musical feature, years of education, years of musical training, age of musical training onset, weekly training, musical instrument, MRI-system, MRI-model, head-coil, image acquisition parameters of T1, T2* and DWI sequences, repetition time (TR), echo time (TE), voxel size, analysis method and software. The main outcome to extract was any difference in structure or function, in stereotactic coordinates, comparing a musician group and a non-musician group. If any of these points were not reported in the original article, authors were contacted to retrieve this information. Six authors were contacted, with 2 positive answers.
Quality assessment of MRI studies. Criteria for MRI quality reporting was selected from a set of guidelines for the standardized reporting of MRI studies 43,44 . Such guidelines dictate a more consistent and coherent policy for the reporting of MRI methods to ensure that methods can be understood and replicated.

Activation likelihood estimation (ALE).
To test the convergence of findings from the neuroimaging studies, we used the anatomic/activation likelihood estimation (ALE) method implemented in the GingerALE software v3.0.2 35 , a widely used technique for coordinate-based meta-analysis of neuroimaging data. Statistically significant foci from between-group contrasts were extracted and recorded for each study. If necessary, coordinates were converted from Talairach coordinates to MNI space using the Lancaster transform (icbm2tal) incorporated in GingerALE 37,94 . The ALE method uses activation foci (input) not as single points, but as spatial probability distributions centred at the given coordinates. Therefore, the algorithm tests to what extent the spatial locations of the foci correlate across independently conducted MRI studies investigating the same construct and assesses them against a null distribution of random spatial association between experiments 46  www.nature.com/scientificreports/ significance of the ALE scores was determined by a permutation test using cluster-level inference at p < 0.05 (FWE), with a cluster-forming threshold set at p < 0.001. First, we used the ALE meta-analytic technique to identify brain differences measured by MRI between musicians (M) and non-musicians (NM) with the aim of comprehensively examine the neural signatures of musical expertise. Two independent ALE meta-analyses were conducted for structural studies and functional studies. To test the directionality of the M vs NM contrast, foci were pooled reporting higher volume/activity in musicians (M > NM) and lower volume/activity in musicians (NM > M) for both structural and functional studies.
Meta-analytic connectivity modelling (MACM). Meta-analytic connectivity modelling (MACM) was performed to analyse co-activation patterns of music-related regions-of-interest (ROI) resulted from the structural (n = 5) and functional (n = 5) ALE meta-analyses, independently, and to functionally segregate each region's putative contribution to behavioural domains and paradigm classes according to the BrainMap platform [40][41][42] .
Large-scale databases such as BrainMap store results obtained from published brain activation (functional) and brain structure (voxel-based morphometry) studies 36,95 . Such databases can be taken into advantage with a meta-analytic approach focusing on the co-activation of brain regions with a specific ROI across all kinds of different mental processes, rather than to a specific mental process. Thus, MACM identifies the functional network of the ROI, namely, its functional connectivity (FC). Traditionally, in fMRI studies, two brain regions are functionally connected when there is a statistical relationship between the measures of neuronal activity, by means of the blood-oxygen-level-dependent signal (BOLD), both during resting-state (task-free FC) or performing a specific task (task-dependent FC). In contrast, MACM relies on patterns of co-activation across many different tasks and allows to examine task-based FC in a general manner 40,96,97 . Thus, MACM provides a data-driven and unbiased approach to determine the connectivity "signature" of a given ROI.
Co-activation analyses were performed using Sleuth 42 and GingerALE 35 from the BrainMap platform. To identify regions of significant convergence, an ALE meta-analysis was performed over all foci retrieved after searching Sleuth by each music-related ROI independently and included the experiment level search criteria of "context: normal mapping" and "activations: activation only". Music-related ROIs were created in Mango 98 with a 5 mm-radius sphere. The results of each ROI search were exported to GingerALE, and a permutation test was conducted using cluster-level inference at p < 0.05 (FWE), with a cluster-forming threshold set at p < 0.001.
Finally, MACM allows to conduct functional profiling of ROIs to study brain-behaviour relationships at a meta-analytic level. In other words, through the BrainMap platform, it is possible to objectively characterize a given ROI in terms of its cognitive/behavioural function which are based on the meta-data that is stored in the database 38 . Thus, tasks in the database are coded in a way that is possible to conduct a behavioural profile of ROIs that resulted from an ALE meta-analysis. The tasks are coded in two dimensions: behavioural domains (BD) and paradigm classes (PC). As the present study has two independent meta-analyses, one for structural studies and one for functional studies, MACM was divided into ROIs that resulted from the structural ALE meta-analysis and ROIs that resulted from the functional ALE meta-analysis. The functional characterization of music-related ROIs was based on the BD meta-data categories available for each neuroimaging study in the database which include action, perception, emotion, cognition and interoception. PC refer to paradigms that have been used repeatedly by different researchers with only minor changes. Such paradigms have become widely known and accepted by the neuroimaging field (e.g., pitch discrimination, finger tapping, music comprehension, go/no-go). A BD refers to the categories and sub-categories of mental operations likely to be isolated by the experimental contrast; a PC is the experimental task isolated by the contrast of interest. Notably, multiple BDs and PCs may apply for a given experiment 99 .
All meta-analytic results (ALE maps) were visualized using Mango 41 on the MNI152 1 mm standard brain, and resulting coordinates were cross-referenced to the Harvard-Oxford Cortical and Subcortical Atlas and the Juelich Histological Atlas via NeuroVault 100 and FSLeyes 101 , respectively. N analysis (FSN). As all meta-analyses, coordinate-based meta-analyses such as ALE can be subject to different forms of publication bias which may impact results and invalidate findings (e.g., the "file drawer problem"). Thus, the Fail-Safe N analysis (FSN) 102 was performed as a measure of robustness against potential publication bias. It refers to the amount of contra-evidence that can be added to a meta-analysis before the results change and can be obtained for each cluster that survives thresholding in an ALE meta-analysis. For normal human brain mapping, it is estimated that a 95% confidence interval for the number of studies that report no local maxima varies from 5 to 30 per 100 published studies. Therefore, the minimum FSN was defined as 30% of total studies for each CBMA. A higher FSN indicates more stable results and hence a higher robustness.