All study procedures were approved by the ethics committees of Sophia University and Juntendo University and the study conformed to the tenets of the Declaration of Helsinki. All subjects provided written informed consent prior to study participation.

We recruited 41 graduate and undergraduate music school students (mean age, 23.4 years; age range, 19–30 years). All participants were healthy, right-handed, Japanese women, without a history of neurological or neuropsychiatric disease. Students majoring in music had begun musical training at the age of 3–5 years (i.e., all participants had more than 15 years of musical training) and had actively participated in concert performances. These students specialized in classical music played on various instruments: 15 played the piano, 8 the violin, 4 the clarinet, and 14 were opera vocalists.

All participants completed two fMRI sessions: the first was the imagined music performance session followed by the resting-state session. Each session lasted 6 min and 40 s. During the imagined music performance session, the participants were asked to imagine the act of music performance in a concert hall as vividly as possible without performing actual movements with their eyes closed. The music “performed” was chosen freely from their repertoires. For example, pianists chose a piece of piano music (e.g., Ballade No. 1 by Frederic Chopin), violinists chose a piece of violin music (e.g., Violin Sonata No. 1 by Robert Schumann), and vocalists chose an opera aria (e.g., “Regnava nel silenzio” from Lucia di Lammermoor by Donizetti). The performance was truncated at the end of each session. In the resting-state session, the participants were instructed not to think anything in particular with their eyes closed.

Imaging data were preprocessed using the CONN toolbox (Whitfield-Gabrieli and Nieto-Castanon, 2012) run on MATLAB (R2016b, MathWorks, Inc.). The individual fMRI data were co-registered to the T1 images. The first four volumes were discarded and the remaining 196 volumes were subjected to preprocessing. The fMRI data were slice-timed, realigned, and subsequently normalized to the standard Montreal Neurological Institute (MNI) template as implemented in the Statistical Parametric Mapping (SPM) software platform. Image artifacts originating from head movement were processed using the ART scrubbing procedure. Signal contributions from white matter, cerebrospinal fluid, and micro head movements (six parameters) were regressed out from the data. Finally, the fMRI data were band-pass filtered (0.008–0.09 Hz) and functional images were spatially smoothed using a Gaussian filter kernel (full width at half-maximum = 8 mm) for the subsequent seed-to-voxel analysis.

A seed-to-voxel functional connectivity analysis was performed using the CONN toolbox. The seed was either the left or right SMA. Pearson's correlation coefficients were calculated between the time course of the left or right SMA and the time courses of all other voxels in the gray matter, which provided a seed-to-voxel connectivity matrix. Positive and negative correlation coefficients defined positive and negative functional connectivity, respectively (Whitfield-Gabrieli and Nieto-Castanon, 2012). The correlation coefficients were then converted to normally distributed scores using Fisher's transformation and subsequently used in the population-level analysis. A height threshold of p < 0.001, uncorrected, was applied to individual voxels to define clusters. The extracted clusters were then set to p < 0.05 with the family-wise error (FWE) correction for results reporting.

In the present study, the SMA exhibited increased connectivity with the SMG during imagined music performance. The increased connectivity with the left SMG was more marked than that with the right SMG for both the left and right SMA. The left SMG is a node of the dorsal language pathway (Catani et al., 2005; Gierhan, 2013) and involved in speech production and perception as well as verbal working memory (Hartwigsen et al., 2010; Turkeltaub and Coslett, 2010; Zevin et al., 2010; Deschamps et al., 2014; Simmonds et al., 2014; Simonyan and Fuertinger, 2015). One fMRI study demonstrated that the SMG was activated during immersive second language learning in monolingual adults, and that activity was positively correlated with second-language acquisition capability (Barbeau et al., 2016). In a study of bilingual individuals, functional connectivity between the SMG and IFG was enhanced in those who became simultaneously bilingual, compared with those who became sequentially bilingual, and the degree of connectivity was correlated with the age of second language acquisition in sequential bilingual individuals (Berken et al., 2016). Music performance also requires cooperating auditory and motor processing. This is evidenced by fMRI studies indicating the involvement of the SMG in music processing using auditory and motor tasks in professional pianists (Bangert et al., 2006; Bengtsson and Ullén, 2006). The SMG is also associated with pitch memory and regulation (Zarate and Zatorre, 2008; Schaal et al., 2015a,b). Therefore, our finding of increased functional connectivity between the SMA and SMG during imagined performance is consistent with previous findings, supporting a relationship between music performance and speech production as suggested in previous studies (Koelsch et al., 2002; Brown et al., 2006). The SMG participates with the IFG in a network that mediates the abstract coding of audiovisual speech (Hasson et al., 2007). Studies suggest that the IFG, especially in the right hemisphere, mediates music syntax processing (Koelsch, 2005), and that the volume of the gray matter in the right IFG is larger in musicians than in non-musicians (Sato et al., 2015). The IFG is structurally and functionally connected with the SMG and STG (Kelly et al., 2010), forming a part of the perisylvian network for language processing (Catani et al., 2005; Xiang et al., 2010; Friederici, 2011; Gierhan, 2013). It should be noted that the connectivity of the SMA with the IFG, SMG, and STG was increased during imagined music performance. To our knowledge, our functional connectivity analysis is the first to show the recruitment of this network by the SMA during imagined music performance. Thus, it appears that the SMA contributes to music syntax processing during imagined music performance by interacting with the perisylvian network. In contrast to language syntax processing (den Ouden et al., 2012), the increased connectivity between the SMA and perisylvian network was not restricted to the left hemisphere. It seems that the network for music syntax processing is bilateral or even rightward lateralized, as indicated by rightward lateralized activation during music syntax processing (Koelsch, 2005; Musso et al., 2015). It should be noted, however, that there remains a debate concerning the laterality of syntax processing in music (Patel, 2003; Asano and Boeckx, 2015). A recent study suggests a common syntax processing role of the left IFG or the Broca's area in music and language (Kunert et al., 2015). Further investigations are needed for elucidating the network for music syntax processing.

Our analysis showed increased connectivity of the SMA with the precuneus, PCC, lateral occipital cortex (LOC), lingual gyrus, intracalcarine cortex, and OP during imagined music performance. The precuneus and PCC have been previously associated with the mental representation of visual scenes (Cavanna and Trimble, 2006; Johnson and Johnson, 2014; Tanaka and Kirino, 2016, 2017). Therefore, connecting with visual areas may enable the SMA to process mental imagery of body movements and visual scenes. The LOC includes the middle temporal (MT)/V5 and extrastriate body area (EBA). The MT/V5 mediates visual motion perception (Born and Bradley, 2005), generally showing higher responses to moving body parts than to moving objects (Spiridon et al., 2006). The EBA mediates the visual perception of body parts (Astafiev et al., 2004) and is thought to contribute to understanding goal-directed actions by representing dynamic aspects of human motion (Takahashi et al., 2008). Activity of the EBA is modulated by limb (arm and foot) movements and by motor imagery without actual movements (Astafiev et al., 2004). Based on this premise, increased connectivity between the right SMA and visual cortex is intriguing, as participants kept their eyes closed during the imaging session. Increased connectivity between the SMA and these visual areas suggests that imagined music performance includes the processing of visual scene and imagined body movements.

The SMA showed increased functional connectivity with posterior temporal regions (pSTS, pMTG, toMTG) during imagined music performance. Activation of the pSTS/MTG has been associated with audiovisual semantic integration (Li et al., 2016). The pMTG is thought to be relatively specialized for semantic control (Noonan et al., 2013). Indeed, posterior regions of temporal lobe (pMTG and toMTG) contribute to semantic cognition and control (Whitney et al., 2011a,b; Jefferies, 2013; Davey et al., 2015, 2016). Semantic control and action understanding are processed by a network that includes the pMTG (Davey et al., 2015, 2016). The pMTG represents object properties (Martin, 2007), and semantic judgments about manipulable objects are represented in a network that includes the pMTG (Davey et al., 2015). The pMTG is functionally connected to regions of the frontoparietal network and shows increased activity during semantically demanding tasks (Tune and Asaridou, 2016). Accordingly, the pMTG has been associated with tool and action knowledge (Martin, 2007; van Elk et al., 2014). An fMRI study demonstrated that the pSTS was activated during the multimodal perception of nonverbal vocal and facial emotional signals and associated trait emotional intelligence with activity in the right pSTS (Kreifelts et al., 2010). In an experiment using naturalistic audiovisual movie clips, the pSTS was suggested to be a hub for social perception (Lahnakoski et al., 2012). Another study reported right hemisphere dominance for the emotional processing of scenes with social contents (Calvo et al., 2015). Consistent with this finding, the increased connectivity between the SMA and the posterior temporal regions in the present study has revealed to be rightward lateralized. Taken together, enhanced connectivity between the SMA and posterior temporal regions suggests that semantic and socio-emotional processing is included in the SMA network for imagined music performance.

A major question addressed in this study was the exact role of the SMA network during imagined music performance. Our result that the SMA increased its functional connectivity with many brain regions during imagined performance is consistent with the idea that the SMA integrates multimodal information required for the performance. The left SMA showed increased functional connectivity with lateral prefrontal regions (SFG/MFG/IFG), especially in the posterior regions, during our study task. These regions have been associated with central executive functions and working memory manipulation as well as maintenance (Nee and D'Esposito, 2016). Moreover, the right lateral prefrontal regions contribute to episodic memory retrieval: The activation of the right prefrontal cortex (PFC) is a consistent finding in neuroimaging studies of retrieval from episodic memory (Henson et al., 1999; Allan et al., 2000; Nyberg et al., 2000). Therefore, the increased connectivity between the SMA and the lateral prefrontal regions, especially in the right hemisphere, suggests that, under top-down control by the central executive system, the SMA construct “the internal representation of music performance” for the upcoming performance by integrating multimodal information represented in several brain regions. Just as working memory is not localized to a single brain region (D'Esposito, 2007), the executive imagery for music performance may be an emergent property of the interaction between the PFC and the SMA. This view is consistent with the rostro-caudal gradient organization theory of PFC functions, in which the posterior PFC represents behavioral and motor information (Badre, 2008; Depue et al., 2016). Complex cognitive control and time estimation, which are also required in music performance, engage the overlapping brain regions including the SMA, PMC, SPL, and precuneus (Radua et al., 2014). Our result is consistent with their meta-analysis. The enhanced network of the SMA during imagined music performance also include the temporal and occipital regions. The inclusion of these regions into the network indicates that the network mediates multimodal integration of the information needed for performance, specifically, the integration of visual imagery, syntactic, semantic, and socio-emotional information into motor imagery. Because various motor imagery tasks activate brain regions including the SMA, SPL, IFG, and preCG (Hanakawa et al., 2008; Hétu et al., 2013), our results suggest that the SMA network processes motor imagery. Therefore, we believe that information processing by the enhanced SMA network mediates multimodal integration into motor imagery for imagined music performance. Regarding functional networks, however, no comparison has ever been made between imagined and actual music performance. To date, an fMRI study reported that imagined and actual playing of short piano pieces in pianists activated brain regions in common: the sensorimotor, premotor, posterior parietal, and visual cortical areas, precuneus, and cerebellum (Meister et al., 2004). Among these regions, the sensorimotor and posterior parietal cortical areas were activated more during actual playing, which suggests that motor execution required the higher level of visuomotor integration.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.