Abstract
During infancy and adolescence, language develops from a predominantly interhemispheric control—through the corpus callosum (CC)—to a predominantly intrahemispheric control, mainly subserved by the left arcuate fasciculus (AF). Using multimodal neuroimaging, we demonstrate that human left-handers (both male and female) with an atypical language lateralization show a rightward participation of language areas from the auditory cortex to the inferior frontal cortex when contrasting speech to tone perception and an enhanced interhemispheric anatomical and functional connectivity. Crucially, musicianship determines two different structural pathways to this outcome. Nonmusicians present a relation between atypical lateralization and intrahemispheric underdevelopment across the anterior AF, hinting at a dysregulation of the ontogenetic shift from an interhemispheric to an intrahemispheric brain. Musicians reveal an alternative pathway related to interhemispheric overdevelopment across the posterior CC and the auditory cortex. We discuss the heterogeneity in reaching atypical language lateralization and the relevance of early musical training in altering the normal development of language cognitive functions.
- arcuate fasciculus
- corpus callosum
- hemispheric specialization
- language lateralization
- left-handedness
- musicianship
Significance Statement
Since the discovery in the nineteenth century that left-handedness predisposes to an atypical lateralization of language, progress in understanding how this condition appears in healthy individuals has been scarce. Here, we introduce a new relevant factor: musical training. We demonstrate how this early and intensive audiomotor learning can potentially modify the hemispheric specialization of language by prompting a differential development of the callosal fibers. Importantly, this perspective also reveals an alternative route to atypical lateralization—unrelated to musicianship—through an underdevelopment of the arcuate fasciculi. In both scenarios, interhemispheric connectivity through the callosum remains prominent, directly or indirectly. Therefore, the historical lack of definitive answers to this phenomenon might be attributed to the existence of multiple potential pathways.
Introduction
Speech and music are closely related. Both processes share an acoustic medium, a communicative goal, and their execution depends on precise control over motor and perceptual systems, existing a parallelism in their cerebral bases (Johansson, 2008). Language depends on left-dominant interactions between auditory posterior regions and anterior premotor regions through the arcuate fasciculus (AF; Price, 2010; Friederici, 2011). Music processing—and specially playing a musical instrument—involves a similar auditory–motor network but on the right hemisphere (Zatorre et al., 2002, 2007). This complementarity also holds in cerebral development. Whether due to neuroplastic changes after training or innate predisposition, musical expertise is associated with larger auditory cortices—particularly in the right hemisphere (Palomar-García et al., 2017)—along with enhanced intrahemispheric structural connectivity across the right AF (Halwani et al., 2011). In parallel, typical language development has been linked to a leftward structural asymmetry in auditory regions (Tzourio-Mazoyer et al., 2017) and a more prominent role of the left AF (Sreedharan et al., 2015). Moreover, auditory asymmetries have been suggested to sustain the processing of speech and music through bottom–up specializations—respectively, left/right auditory cortices more sensitive to temporal/spectral decoding, which is critical for speech/music—and top–down influences (Sihvonen et al., 2019; Albouy et al., 2020; Zatorre, 2022). Thus, cerebral bases supporting speech and music follow similar auditory–motor streamlines but with a leftward or rightward asymmetry.
Going more in-depth into their development, language is biased toward the left hemisphere from infancy (Peña et al., 2003; Dehaene-Lambertz et al., 2010), but its primary control undergoes a gradual ontogenetic shift (starting at ∼7 years old) from a more interhemispheric reliance—through the corpus callosum (CC)—to a more intrahemispheric pattern—habilitated by the left AF (Szaflarski et al., 2006; Friederici et al., 2011; Perani et al., 2011). On the other hand, musical expertise and training have been associated with increasing volume and connectivity in the posterior parts of the CC, especially during infancy (Hyde et al., 2009; Leipold et al., 2021), as well as a potentiated structural and functional connectivity across the right AF (Halwani et al., 2011; Palomar-García et al., 2017). Therefore, both language development and musical learning rely on changes in the CC and the AF. This opens the possibility that intensive musical training may influence language development.
Although this is the most common pattern of language development, some individuals (∼4–6% of right-handers and 22–27% of left-handers) present an atypical lateralization of speech, characterized by a greater—or even exclusive—involvement of the right hemisphere (Mazoyer et al., 2014), which has the potential to impact cognitive performance and certain psychological traits (Mellet et al., 2014; Gerrits et al., 2020; Villar-Rodríguez et al., 2024). One of the proposed mechanisms to explain this rare phenomenon involves interhemispheric hyperconnectivity via the CC, potentially facilitating a hemispherically flexible development for the various cognitive components of language (Tzourio-Mazoyer, 2016). Recent data have indeed revealed that atypical lateralization phenotypes are characterized by a strong interaction between the language processing nodes of both hemispheres, presumably attributed to an enlarged CC (Labache et al., 2020). Considering that musicianship increases the probability of atypical language dominance among left-handers (Villar-Rodríguez et al., 2020) and the parallelism between music and language, we investigated if the inclusion of musicianship might reveal evidence favoring the hyperconnectivity proposal or even represent an alternative pathway to developing atypical lateralization.
We recruited left-handed musicians and nonmusicians and used functional magnetic resonance imaging (fMRI) plus diffusion-weighted imaging to study their speech production, speech perception, functional connectivity at rest, and gray/white matter. We hypothesized that (1) musicianship increases the likelihood of atypical lateralization, using different cerebral mechanisms than nonmusicians (Villar-Rodríguez et al., 2020); (2) atypical lateralization is associated with a shift in auditory cortices functional asymmetries (Albouy et al., 2020); and (3) atypical lateralization is supported by differences in intrahemispheric and interhemispheric connectivity, which are potentiated by musicianship (Halwani et al., 2011; Sreedharan et al., 2015; Palomar-García et al., 2017; Labache et al., 2020).
Materials and Methods
Participants
This study included 112 nonright-handed participants: 56 musicians [30 females; 23.4 ± 5.2 years; 42.8 ± 5.3 Edinburgh Handedness Inventory (EHI) score] and 56 nonmusicians (27 females; 22.6 ± 5.1 years; 42.1 ± 4.6 EHI score). Musicians and nonmusicians did not show statistically significant between-group differences in gender (Pearson's χ2 = 0.32; p = 0.57) or age (Mann–Whitney's U = 1,315; p = 0.14). They were all left-handed or mixed-handed, according to the EHI (Oldfield, 1971) and the scoring method by Bryden (1977; EHI score range 24–37 was considered mixed-handed, and range 38–50 was considered left-handed). Proportion of left-handers/mixed-handers was almost equal among musicians (47/9) and nonmusicians (46/10), with no significant differences detected (Pearson's χ2 = 0.06; p = 0.8). EHI score did not significantly differ between musicians (42.8 ± 5.3; range, 30–50) and nonmusicians (42.1 ± 4.6; range, 26–50; Mann–Whitney's U = 1,468.5; p = 0.56; see Fig. 1 for distributions). A participant was considered a musician if they had received musical training at any music school for at least 6 years. Musicians and nonmusicians did significantly differ in their pitch discrimination capabilities (t(107) = 4.21; p < 0.001; 77.3% ± 10.2% accuracy for musicians; 69.7% ± 8.5% accuracy for nonmusicians), as measured by the Jake Mandell test (Palomar-García et al., 2020). None of the participants had suffered any neurological or psychiatric disorders, and they were tested for hearing loss through pure tone audiometry, confirming normal hearing. Written informed consent was obtained from all participants. All methods were carried out in accordance with the policies and principles contained in the Federal Policy for the Protection of Human Subjects (US Office of Science and Technology Policy) and in the Declaration of Helsinki and were approved by the Human Research Ethics Committee of the Universitat Jaume I (reference, CD/03/2021).
EHI. Distribution of EHI score in the 56 nonmusicians (gray) and 56 musicians (red). Ranging from 24 to 37 (mixed-handedness) and 38 to 50 (left-handedness).
Image acquisition
MRIs were acquired on a 3 T GE Signa Architect scanner. A 3D structural MRI was acquired for each subject using a T1-weighted magnetization-prepared rapid gradient–echo sequence (TR/TE, 8.5/3.3 ms; matrix, 512 × 512 × 384; voxel size, 0.47 × 0.47 × 0.5 mm). Gradient-echo T2*-weighted echo–planar imaging sequences were recorded for the verb-generation task (volumes, 144; TR/TE, 2,500/30 ms; matrix, 64 × 64 × 30; voxel size, 3.75 × 3.75 × 4 mm), the word-listening task (volumes, 378; TR/TE, 2,000/30 ms; matrix, 64 × 64 × 26; voxel size, 3.75 × 3.75 × 4.5 mm), and the eyes-open resting state (volumes, 210; TR/TE, 2,000/30 ms; matrix, 64 × 64 × 27; voxel size, 3.75 × 3.75 × 4.5 mm). Slices were acquired in interleaved sequence and strict axial orientation, covering the whole cortex. A diffusion-tensor-imaging (DTI) sequence was also acquired (diffusion gradient directions, 25; b = 0 + 1,000 mm2/s; TR/TE, 13,000/80; matrix, 256 × 256 × 60; voxel size, 1.01 × 1.01 × 2 mm).
Verb-generation task
We used a computerized Spanish verb-generation task suited for MRI scanners that is described elsewhere (Sanjuán et al., 2010). In summary, it consists of an activation condition in which the participant generates verbs from visually presented concrete nouns and a control condition in which the participant reads aloud visually presented pairs of letters. Stimuli were presented using MRI-compatible goggles (VisuaStimDigital, Resonance Technology), and responses were recorded via a noise-cancelling microphone to ensure that each participant was actively and accurately engaged in the task (FOMRI III+, Optoacoustics). Before entering the scanner, participants practiced with a different version of the task for 1 min.
Word-listening task
Speech perception was investigated using a monaural passive listening fMRI paradigm. Monaural stimulation using words and tones has been found to elicit stronger contralateral cortical responses in the superior temporal gyrus (STG) than binaural stimulation (Jäncke et al., 2002), and contralateral cortical response to monaural stimulation has also been described to differ according to certain characteristics of the stimuli, such as amplitude modulation (Gutschalk and Steinmann, 2015). Because of this, we opted for a monaural design to benefit from auditory contralateral dominance in an effort to maximize the sensitivity of our data to differences in hemispheric lateralization. Participants were instructed to lay in the scanner with eyes closed, trying not to sleep, and to simply pay attention to the words and sounds that they were about to hear in both ears. Stimuli were presented monaurally in a block design consisting of words left-ear (20 s × 6 blocks), words right-ear (20 s × 6 blocks), tones left-ear (20 s × 6 blocks), and tones right-ear (20 s × 6 blocks). The task alternates word and tone blocks, beginning with a word block. Each block consisted of eight stimuli, with a mean interstimulus interval of 2,242 ms. The distribution of left-ear and right-ear blocks was pseudorandomized, and two inverse versions (one beginning with left-ear and the other beginning with right-ear) were presented in a contra balanced way to participants. Blocks were separated by 12 s segments of no stimuli presentation (12 s × 23 segments), which will be referred to as the baseline. In summary, this was the block task design (total duration, 756 s), where A and B denotes different ears depending on the task version (Awords – Btones – Bwords – Atones – Bwords – Atones – Awords – Btones – Bwords – Atones – Bwords – Atones – Awords – Btones – Awords – Btones – Bwords – Btones – Awords – Atones – Awords – Atones – Bwords – Btones). Stimuli were presented using a 30 dB noise-attenuating MRI–compatible headset (VisuaStimDigital, Resonance Technology). The volume level was set at 85 dB—as it was considered both comfortable and loud enough by different subjects during the test runs prior to this study—and it was not altered between participants. Before entering the scanner, participants practiced with a different version of the task for 1 min.
Word stimuli consisted of 50 familiar and 46 unfamiliar trisyllabic words of similar imageability extracted from the EsPal database (Duchon et al., 2013) and pseudorandomly assigned to different blocks. Stimuli were synthetized using the speech package Microsoft Helena Desktop (International Spanish) included in the software Balabolka (balabolka.en.softonic.com/). Stimuli duration ranged from 410 to 770 ms (mean ± SD = 538 ± 67 ms).
Tone stimuli consisted of 96 different sequences of 3 pure tones, trying to match the trisyllabic structure of word stimuli. Pitches used in the elaboration of tone sequences comprised the ascending scale from C4 to C5 (261.63; 293.67; 329.63; 349.23; 391.99; 440; 493.88; and 523.25 Hz). Stimuli were synthetized and combined using the sound software package Audacity (https://www.audacityteam.org/). Tone sequences were matched in duration to word stimuli, hence creating tone blocks that matched word blocks.
fMRI processing
Task-based functional images were processed using the Statistical Parametric Mapping software package (SPM12; Wellcome Trust Centre for Neuroimaging). Preprocessing followed the default pipeline and included (1) alignment of each participant's fMRI data to the AC-PC plane by using the anatomical image; (2) head motion correction, where the functional images were realigned and resliced to fit the mean functional image; (3) coregistration of the anatomical image to the mean functional image; (4) resegmentation of the transformed anatomical image using a tissue probability map (a symmetric map was used for the word-listening task, as the following processing steps required a voxel-wise correspondence between hemispheres); (5) spatial normalization of the functional images to the MNI (Montreal Neurological Institute) space with 3 mm3 resolution; and (6) spatial smoothing (FWHM, 4 mm). The general linear model for the verb-generation task was defined for each participant by contrasting activation > control blocks. The general linear model for the word-listening task was defined for each participant by contrasting (irrespective of ear): (1) words > tones; (2) words > baseline; and (3) tones > baseline. For both tasks, the BOLD (blood oxygenation level-dependent) signal was estimated by convolving the task's onsets with the canonical hemodynamic response function. Six motion realignment parameters extracted from head motion preprocessing were included as covariates of no interest, and a high-pass filter (128 s) was applied to the contrast images to eliminate low-frequency components. Additionally, we computed whole-brain voxel–wise functional asymmetry maps from the word-listening contrast images. To do so, images were flipped at midline, inverting the right and left hemispheres, and subsequently subtracted from the original unflipped contrast images (Kurth et al., 2015). As a result, functional asymmetry maps present voxel-wise asymmetry indexes, which indicate the proportion of BOLD signal between homotopic voxels on opposite hemispheres. Finally, unsmoothed images were separately processed using the Functionnectome approach (Nozais et al., 2021). Briefly, Functionnectome allows to voxel-wise study the functional role of white matter circuitry during a particular fMRI task. Specifically, BOLD time-series projection was carried out using whole-brain probabilistic priors.
Resting-state functional images were preprocessed using the Data Processing Assistant for the Resting-State toolbox or DPARSFA (Yan et al., 2016). Processing steps included the following: (1) slice-timing correction; (2) head motion correction; (3) coregistration; (4) segmentation; (5) removal through linear regression of nuisance variance derived from head motion, white matter signal, cerebrospinal fluid signal, and global mean signal; (6) spatial normalization to the MNI space; (7) spatial smoothing using a Gaussian kernel at FWHM, 4 mm; (8) removal of the linear trend in the time-series; and (9) band-pass temporal filtering (0.01–0.1). Normalized z mean time courses were extracted using a region of interest (ROI)-to-ROI approach between Broca–Wernicke in both hemispheres and between the interhemispheric Broca–Broca and Wernicke–Wernicke. Broca and Wernicke ROIs were defined using the f2_2 and SMG7 parcellations found in the SENSAAS language atlas (Labache et al., 2018).
Laterality index
Individual functional lateralization was assessed by calculating the laterality index (LI) for the verb-generation contrast images and the word-listening contrast “words > tones.” LI is a proportion of the brain activation between the two hemispheres, thus giving us information about the direction and degree of hemispheric specialization during a particular function in a single individual. LI ranges from +100 (totally leftward function) to −100 (totally rightward function). For its calculation, we used the bootstrap method implemented in the LI-toolbox (Wilke and Lidzba, 2007), based on SPM. For the verb-generation task, we computed the LI encompassing the frontal voxels most relevant in the expressive language function evaluated by this task. We defined this using standard anatomical criteria (Rutten et al., 2002), thus including Brodmann Areas 9, 44, 45, and 46. For the word-listening task, we were especially interested in the functional specialization of the auditory cortices, so we included Brodmann Areas 22, 41, and 42, which approximately correspond to the STG. To create both masks, we employed the Talairach Daemon atlas (Lancaster et al., 2000) included in the WFU PickAtlas toolbox (Maldjian et al., 2003), inserting the Brodmann areas with a 3D dilatation value of 2. Additionally, medial areas were subtracted by a boxcar with dimensions of 20, 100, and 100 and an epicenter at 0, 0, 0. So, we ended up with a language production LI (verb-generation task), a language perception LI (word-listening task), and a language mean LI (average of both LIs).
Voxel-based morphometry
3D structural MRI images were preprocessed using a voxel-based morphometry approach, via the CAT12 toolbox (http://www.neuro-jena.github.io/cat/) for SPM12. Preprocessing followed the default pipeline and included the following: (1) segmentation into gray matter, white matter, and cerebrospinal fluid; (2) registration to the ICBM standard template; and (3) DARTEL normalization of gray matter and white matter segments to the MNI template. Finally, we extracted ROI native gray matter volume values corresponding to the primary auditory cortex or Heschl's gyrus (Brodmann Areas 41 and 42) and native white matter volume values corresponding to the genu, body, and splenium of the CC (Mori et al., 2008).
Diffusion-based tractography
DTI images were corrected for eddy current distortions using FSL (Smith et al., 2004). Afterward, we followed the default preprocessing pipeline of DiffusionToolkit (Wang et al., 2007) to reconstruct the images, track the fibers, and smooth the tracks. Fractional anisotropy was used for the masking procedure during fiber tracking (threshold, 0.2), setting the angle threshold at 35°. Virtual dissection of the AF followed the guidelines provided by previous tractography studies (López-Barroso et al., 2013), utilizing the TrackVis software (trackvis.org). In summary, dissection was carried out for all participants in a single-subject basis by displaying their track reconstruction in native space and performing manual ROI delineations to encompass the AF segments in both hemispheres (see Fig. 2 for an example). The anterior AF segment was delineated by drawing a coronal prefrontal ROI and a sagittal inferior parietal ROI. The posterior AF segment was delineated by drawing a sagittal inferior parietal ROI and an axial superior temporal ROI. The direct AF segment was delineated by drawing a coronal prefrontal ROI and an axial superior temporal ROI. Fiber count was extracted for every AF segment from each participant. The virtual dissection procedure was performed by two different researchers that independently followed the provided guidelines. Inter-rater reliability was evaluated by computing the intraclass correlation coefficient (ICC) between the extracted fiber counts for every segment, indicating a very good reliability (mean ICC, 0.875; p < 0.001). All tractography-derived measures used in the following analyses were averaged across both observations.
Virtual dissection of the AF in an example subject. Unnormalized white matter tracks are displayed on top of a normalized three-dimensionally reconstructed brain.
Statistical analyses
Homogeneous distribution of typical and atypical lateralization of speech production between musicians and nonmusicians was tested using three χ2 tests. For these tests, participants were considered as typically lateralized or atypically according to their LI during the verb-generation task. We used three different cutoffs (hence the three different χ2 tests): LI > 0 = typical; LI > 20 = typical, and LI > 40 = typical. Additionally, we calculated a Mann–Whitney's U to test if musicians and nonmusicians quantitatively differed as groups in their LI of speech production. We also performed Spearman's correlations and a stepwise linear regression to investigate the possible association between this LI and several musicianship variables among musicians: age of onset of musical training, duration of musical training (in years), number of trained instruments, and Barcelona Music Reward Questionnaire or BMRQ score (Mas-Herrero et al., 2013).
We ran different group analyses to investigate the functional asymmetry, functional connectivity, Functionnectome, and structural characteristics of both white matter and gray matter. However, all of them followed a similar ANCOVA or MANCOVA design, which aimed at answering if lateralization of language (continuous independent variable) was related to a particular measure of interest (continuous dependent variable) while considering if musicianship (categorical independent variable) played any role in that relation.
Functional asymmetry maps (Kurth et al., 2015) of the word-listening contrast “words > tones” were voxel-wise analyzed, examining a potential relation between verb-generation task LI and functional asymmetry, including musicianship as a factor [FWE-corrected at p < 0.05 via threshold-free cluster enhancement (TFCE)]. Because functional asymmetry maps are mirrored along the Y axis (changing only the sign of the values between both hemispheres), analyses were restricted to the left hemisphere. Additional ROI-wise analyses explored this design in the mean functional asymmetry of the medial geniculate body (MGB) and the inferior colliculus, subcortical nuclei that participate in auditory processing just before the primary auditory cortex. We defined these ROIs by drawing 6 mm spheres at the peak coordinates described in Gutschalk and Steinmann (2015). Importantly, it should be noted that the contrast between words and tones informs us about relative differences in hemispheric specialization, but it is not enough to extract conclusions about absolute differences. To address this, we explored the functional asymmetry of relevant clusters (those found in the ANCOVA design, the MGB, and the inferior colliculus) in the two contrasts against baseline: “words > baseline” and “tones > baseline.”
Functionnectome maps (Nozais et al., 2021) of the verb-generation task and the word-listening task were voxel-wise analyzed, testing a possible relation between mean language LI (averaged LI of the verb-generation task and the word-listening task) and the BOLD projection in white matter, including musicianship as a factor (voxel-wise FDR-corrected at p < 0.05). Voxels with a probability of being gray matter >50% were discarded from these tests.
Structural measures were analyzed using different repeated-measure MANCOVA designs. Note that all volumetric and diffusion designs included total intracranial volume, EHI score, and gender as covariates. We tested a possible relation between mean language LI and Heschl's gyrus volume, including hemisphere (left/right) as a within-subject variable and musicianship as a factor. We tested a possible relation between mean language LI and AF streamlines, including hemisphere (left/right) and segment (anterior/posterior/direct) as within-subject variables and musicianship as a factor. We tested a possible relation between mean language LI and CC volume, including segment (genu/body/splenium) as within-subject variables and musicianship as a factor.
Post hoc, we also calculated the INTER/INTRA index for every participant. This index was computed by subtracting the standardized CC volume from the standardized AF volume. Thus, it tells us whether the INTER/INTRA white matter ratio of these tracts in a particular brain leans toward more intrahemispheric (higher values) or more interhemispheric (lower values) connectivity. We tested a possible relation between mean language LI and INTER/INTRA index via a regular MANCOVA, including musicianship as a factor.
ROI-to-ROI functional connectivity at rest was analyzed using two different repeated-measure MANCOVA designs. On the one hand, we tested a possible relation between mean language LI and intrahemispheric functional connectivity (Broca–Wernicke), including hemisphere (left/right) as a within-subject variable and musicianship as a factor. On the other hand, we tested a possible relation between mean language LI and interhemispheric functional connectivity (left Broca–right Broca and left Wernicke–right Wernicke), including region (Broca/Wernicke) as a within-subject variable and musicianship as a factor.
Post hoc, we also conducted Spearman's correlations between the two interhemispheric resting-state functional couplings (left Broca–right Broca and left Wernicke–right Wernicke) and their respective white matter neuroanatomical measures (callosal genu, body, and splenium volume), as well as between the two intrahemispheric resting-state functional couplings (left Broca–left Wernicke and right Broca–right Wernicke) and their respective white matter neuroanatomical measures (number of streamlines across the left anterior AF and the right anterior AF).
Data availability
All data presented in this study are available in the following public dataset: https://doi.org/10.6084/m9.figshare.24449827.v1. Request for raw data should be submitted to the corresponding author.
Results
Higher incidence of atypically lateralized speech among musicians
We used the verb-generation task to assess the hemispheric lateralization of speech production. Whole-brain one–sample analysis of this task elicited activations in the left inferior frontal gyrus (IFG), left superior temporal sulcus, left middle temporal gyrus (MTG), bilateral insula, and bilateral SMA (Fig. 3). As in previous studies, we used the IFG subjectwise activation for the calculation of LIs. As expected, musicians as a group presented a significantly more atypical LI than nonmusicians (Mann–Whitney's U = 1,225.5; p = 0.046; musicians mean ± SD LI = 23.2 ± 64.4; nonmusicians mean ± SD LI = 47.2 ± 54.8). For replication purposes (Villar-Rodríguez et al., 2020), we confirmed that this mostly held true using categorical approaches. That is, atypical lateralization of speech production was slightly more frequent among musicians when compared with nonmusicians, using a cutoff criterion of +20 (Pearson's χ2 = 4.22; p = 0.04) and +40 (Pearson's χ2 = 3.89; p = 0.049; Fig. 4), with the cutoff of 0 also close to statistical significance (Pearson's χ2 = 3.77; p = 0.052). Respectively, our total sample included 72/78/83 typically lateralized participants (LI > 40/20/0) and 40/34/29 atypically lateralized participants (LI < 40/20/0; incidence, 44.6%/39.3%/33.9% in musicians, 26.8%/21.4%/17.9% in nonmusicians).
Voxel-wise one–sample t test for “activation > control,” during the verb-generation task. FWE-corrected at p < 0.05 via TFCE. The color bar represents TFCE value. L, left hemisphere; R, right hemisphere.
Higher incidence of atypically lateralized speech among musicians. Distribution of language production LI (10-intervals) in musicians (red) and nonmusicians (gray). Ranging from −100 (completely right-lateralized) to +100 (completely left-lateralized) and using +40 as cutoff.
We also investigated the potential correlation between verb-generation LI and various variables associated with musicianship (age of onset of training, years of training, number of trained instruments, and BMRQ score). We observed a negative relationship between LI and the years of training (ρ(52) = −0.33; p = 0.014), indicating that a longer period of musical training was linked to a more atypical lateralization of the IFG. The number of trained instruments was positively correlated with LI (ρ52 = 0.27; p = 0.048). Upon categorizing musicians as typical or atypical, we observed that the proportion of musicians specializing in a single instrument doubled among atypical individuals, increasing from 30 to 62%. LI was also negatively correlated with BMRQ score (ρ(48) = −0.29; p = 0.041), specifically with the musical seeking subscale (ρ(48) = −0.35; p = 0.013). Notably, both musical training variables remained statistically significant when utilized as inputs in a stepwise linear regression model to predict LI (p = 0.011 and 0.024, respectively; model's F = 5.92; p = 0.005), but not the BMRQ score. Thus, the two significant training variables contribute to explaining distinct parts of LI variance. The age of onset of musical training appeared to be unrelated to LI.
Atypically lateralized speech is related to a rightward specialization for the hearing of words > tones and a leftward specialization for the hearing of tones > words
We used the word-listening task to assess the relation between the hemispheric lateralization of speech perception and speech production. Voxel-wise functional asymmetry ANCOVA analyses contrasting the perception of words and tones revealed an asymmetrical network linked to speech production lateralization (FWE-corrected at p < 0.05 via TFCE; Fig. 5). That is, the voxels depicted in Figure 5a present a significant relationship between their functional asymmetry when contrasting the listening of words and tones and the LI calculated during the verb-generation task. Specifically, as we can see in Figure 5b, as the verb-generation LI shifted rightward (more atypical), the auditory network also displayed increased rightward asymmetry during word listening compared with tones. Notably, this also meant that the more rightward verb-generation LI was, the more leftward asymmetrical this auditory network was during the listening of tones compared with words and vice versa. Therefore, atypical lateralization of verb generation is associated with a relative hemispheric reversal of auditory processing between words (leftward) and tones (rightward). Components of this hemispherically mirrored network were defined via the AAL atlas (Rolls et al., 2020) into the following: IFG, STG (including Heschl's gyrus, the location of the primary auditory cortex), MTG, ITG, fusiform gyrus, caudate, putamen, and pallidum. Additional ROI analyses focused on auditory subcortical nuclei also confirmed a mirrored relative asymmetry in the MGB (ρ(110) = −0.24; p = 0.01) and a diminished relative asymmetry in the inferior colliculi (ρ(110) = −0.21; p = 0.03). No significant main effect or interaction of musicianship was found in any of these analyses. Hence, regardless of musicianship, atypical dominance for speech production is group-wise associated with the processing of heard words > tones predominantly in the right auditory cortex and heard tones > words predominantly in the left auditory cortex; this reversed pattern extends to some degree to the auditory subcortical nuclei.
Atypically lateralized speech is related to a rightward specialization for the hearing of words > tones and a leftward specialization for the hearing of tones > words. Functional hemispheric asymmetry during the word-listening task. a, Voxels presenting a significant correlation between their functional asymmetry during the word-listening task (comparison between words and tones processing) and the verb-generation task LI. Voxel-wise significance maps are displayed in three-dimensional reconstructions plus coronal slices. Coordinates are reported in MNI space. FWE-corrected at p < 0.05 via TFCE. The color bar represents TFCE value. b, Graph depicting the mean functional asymmetry values of the significant clusters from the previous panel, in the words > tones contrast. For illustrative purposes, two means are presented: one for typical participants (verb-generation LI > 40; blue bars) and one for atypical participants (verb-generation LI < 40; orange bars). Negative values indicate leftward asymmetry, whereas positive values indicate rightward asymmetry. Significant clusters from the voxel-wise regression were divided into cerebral regions using the AAL atlas. Note the inclusion of the values resulting from the ROI analysis of the MGB and IC, marked in red. Also note that every cluster goes through a rightward shift as LI becomes atypical (with the exception of IC, which experiments a decrease in its leftward asymmetry but not a rightward shift). c, Graph depicting the mean functional asymmetry values of the significant clusters from the (a) panel, in the words > baseline contrast (left) and the tones > baseline contrast (right). IFG, inferior frontal gyrus; Thal, thalamus; Caud, caudate; Put, putamen; Pallid, pallidum; FusG, fusiform gyrus; STG, superior temporal gyrus; MTG, middle temporal gyrus; MGB, medial geniculate body; and IC, inferior colliculus.
It should be noted, however, that contrasting words and tones are relevant to describe relative differences in functional asymmetry between the two stimuli. But it does not inform about the absolute hemispheric specialization present during the processing of words or the processing of tones. To know this, we explored the functional asymmetry of the prior cerebral regions in the baseline contrasts of the word-listening task: “words > baseline” and “tones > baseline” (Fig. 5c). For the processing of words, reversal was found only in IFG, thalamus, caudate, and—to a lesser extent—the pallidum. Temporal regions were characterized by a decreased leftward asymmetry (almost disappearing in the MTG) as LI became more atypical, but they did not present a reversal. MGB and IC presented near to zero asymmetry. For the processing of tones, reversal was found only in the MGB. Surprisingly, temporal regions were also leftward asymmetrical when processing tones, but they experimented a decrease in their leftward asymmetry the more typical LI became. IFG, thalamus, caudate, putamen, and pallidum presented near to no asymmetry. IC was rightward asymmetrical for both ends of LI.
Next, we calculated word-listening LI in the STG for the hearing of words compared with tones for each participant. When examining these LIs in relation to the verb-generation LIs on a per-subject basis, it was observed that 30% of participants exhibited a discrepancy. That is, when categorized as typical (LI > 40) or atypical (LI < 40), these discrepant individuals did not adhere to a typical–typical or atypical–atypical LI pairing (Fig. 6). In light of this, we computed a mean language LI for each subject by averaging both LIs. These mean language LIs were utilized in the subsequent analyses.
Hemispheric specialization of speech production and word listening. Correspondence between LIs from the verb-generation task (measured in the IFGs) and the word-listening task (measured in the STG). Note the important presence of discrepant individuals in the top-left and bottom-right quadrants. Gray circles, nonmusicians; red circles, musicians.
Lastly, we explored whether this mean LI correlated with musical training variables. Similar to the speech production LI, the mean LI demonstrated a negative association with the duration of training (ρ(52) = −0.31; p = 0.02), although it did not reach a statistical significance when examined with the number of trained instruments (ρ(52) = 0.23; p = 0.09) nor BMRQ score (ρ(48) = −0.23; p = 0.11). The regression analysis closely mirrored that of the speech production LI, with both variables significantly predicting the mean LI (p = 0.016 and 0.049, respectively; model's F = 4.9; p = 0.01). The age of onset of musical training showed no apparent association with the mean LI.
Atypical lateralization of language is functionally supported by a rightward participation of the AF but modulated by musicianship
We employed the Functionnectome approach to investigate the functional implication of white matter pathways during both language production and perception. This innovative methodology addresses the question by probabilistically projecting the BOLD cortical signal onto externally defined white matter priors (Nozais et al., 2021). Functionnectome ANCOVA analyses revealed significant effects during the verb-generation task across the AF, the cerebral peduncles, and the CC (Fig. 7). Voxel-wise main effects of mean language LI showed that atypical lateralization was related to a decrease in activation along the left AF and the left cerebral peduncle, as well as to an increase in activation across the right AF, the right cerebral peduncle, and the anterior CC (voxel-wise FDR-corrected at p < 0.05).
Atypical lateralization of language is functionally supported by a rightward participation of the AF, but modulated by musicianship. Functionnectome and language lateralization during the verb-generation task. Colored regions correspond to white matter voxels that present a significant correlation between Functionnectome-derived activation and mean LI, depending on musicianship. Red voxels decrease their activation with atypical lateralization among musicians. Green voxels increase their activation with atypical lateralization among nonmusicians. Voxel-wise FDR-corrected at p < 0.05. The scatterplots depict these correlations (top plot, mean red region; bottom plot, mean green region) for both musicianship groups.
However, significant voxel-wise interaction effects between mean language LI and musicianship revealed that these relations were not uniformly applicable to both groups (voxel-wise FDR-corrected at p < 0.05). In nonmusicians, atypical lateralization was linked to the rightward increase (R2 = 0.385; p < 0.001), but not to the leftward decrease (R2 = 0.006; p = 0.95). Conversely, in musicians, atypical lateralization was related to the leftward decrease (R2 = 0.297; p < 0.001), but not to the rightward increase (R2 < 0.001; p = 0.79). Thus, musicianship modulates the functional implication of white matter pathways during the verb-generation task in atypically lateralized individuals: either by a leftward decrease (musicians) or by a more extensive rightward increase which includes the anterior body of the CC (nonmusicians). No significant main effects or interactions were found when applying the Functionnectome approach during the word-listening task. No statistically significant correlation of any kind was found between Functionnectome results and musicianship variables.
Different anatomical correlates of atypical lateralization according to musicianship: disruption of the interhemispheric and intrahemispheric equilibrium
We examined neuroanatomical data by focusing on the gray matter volume of the auditory cortices and the white matter volume of the CC (using voxel-based morphometry), along with the connectivity of the AF (via diffusion-based deterministic tractography). Repeated-measure multivariate analyses unveiled two different correlates of atypical language lateralization, contingent on musicianship. In nonmusicians, as the mean LI became more atypical, the AF of both hemispheres exhibited reduced connectivity (number of streamlines and extremely similar results using tract volume) across their anterior segment (interaction of mean LI × musicianship × segment, F = 6.99; p = 0.001; Fig. 8a). Conversely, in musicians, as the mean LI became more atypical, the CC increased in volume (interaction of mean LI × musicianship, F = 7.75; p = 0.006), particularly in the callosal splenium (interaction of mean LI × musicianship × segment, F = 5.71; p = 0.004; Fig. 8b). It should be noted that gender and EHI did not present similar interactions with LI. So, together with the lack of between-group differences in gender and EHI, we can safely assume that those variables are not affecting the results. Additionally, irrespective of language lateralization, musicianship was related with larger auditory cortices (F = 4.37; p = 0.04) and more streamlines in the right AF (F = 4.51; p = 0.013). Consequently, atypical language lateralization was structurally marked by either (1) a reduction of the intrahemispheric white matter tracts (as seen in nonmusicians) or (2) an enlargement of the interhemispheric white matter tracts (as revealed in musicians). No statistically significant correlation of any kind was found between structural results and musicianship variables.
Different anatomical correlates of atypical lateralization according to musicianship: disruption of the interhemispheric and intrahemispheric equilibrium. a, Correlation of mean language LI with anterior AF streamlines, in nonmusicians. b, Correlation of mean language LI with callosal splenium volume, in musicians. c, Correlation of mean language LI with INTRA/INTER index (lower values correspond to larger interhemispheric white matter). Streamlines and volume values were adjusted for total intracranial volume.
At this juncture, we questioned whether these two distinct structural correlations might signify different pathways to a similar outcome: the partial disruption of the interhemispheric/intrahemispheric equilibrium in favor of a more interhemispheric brain. To investigate this possibility, we conducted a post hoc computation of the INTRA/INTER index (normalized AF volume minus normalized CC volume). Briefly, this index quantifies the volumetric ratio between the AF and the CC (higher values indicating a more intrahemispheric brain and vice versa). Remarkably, we discovered a direct relationship between this index and mean LI (main effect of mean LI, F = 8.04; p < 0.01; Fig. 8c), confirming that atypical lateralization—regardless of musicianship—implied a relatively more interhemispheric white matter structure. Importantly, as previously demonstrated, this outcome could be reached through two distinct structural mechanisms, dependent on changes across the AF or the CC, and modulated by musicianship.
Atypical lateralization of language is related to a reduced intrahemispheric but stronger interhemispheric functional connectivity at rest between speech cortical hubs
We approached the resting-state data by calculating ROI-to-ROI functional connectivity—both intrahemispheric and interhemispheric—between the main speech hubs of the cortex: Broca's area and Wernicke's area. Two repeated-measure MANCOVA analyses (one considering intrahemispheric connectivities as left/right hemisphere and one considering interhemispheric connectivities as anterior/posterior region) showed an association between the functional coupling patterns of these essential areas and language lateralization. Specifically, as the mean LI became more atypical, interhemispheric connectivity between left and right IFG increased (interaction of mean LI × region, F = 6.96; p = 0.01; Fig. 9a), while intrahemispheric connectivity between the frontal and parietal inferior cortices decreased bilaterally (main effect of mean LI, F = 4.05; p = 0.047; Fig. 9b). No significant main effect nor interaction of musicianship was observed in these analyses.
Atypical lateralization of language is related to a reduced intrahemispheric but stronger interhemispheric functional connectivity at rest between speech cortical hubs. a, Correlation of mean language LI with interhemispheric FC between Broca's areas. b, Correlation of mean language LI with mean (left + right) intrahemispheric FC between Broca's and Wernicke's areas.
Furthermore, to better comprehend the functional implications of the neuroanatomical findings presented in the previous section, we conducted post hoc correlations between these resting-state functional couplings and the neuroanatomical measures of the underlying white matter tracts (Tables 1, 2). Notably, we found a direct correlation between connectivity across the left anterior segment of the AF and our ROI-to-ROI measure of left intrahemispheric functional connectivity (ρ(105) = 0.26; p = 0.01). Additionally, the volumes of all three CC segments (genu, body, and splenium) were inversely correlated with posterior interhemispheric functional connectivity (respectively, ρ(105) = −0.2; p = 0.04; ρ(105) = −0.25; p = 0.01; and ρ(105) = −23; p = 0.02).
Interhemispheric structural–functional connectivity
Intrahemispheric structural–functional connectivity
Discussion
In this study, we have approached atypical language dominance from a new perspective by considering how musical ability may influence its development. We demonstrated that (1) atypical lateralization of language in the IFG is observed mostly among left-handed musicians; (2) the atypically lateralized group exhibits a mirrored organization of language functions—in contrast to tones processing—affecting the auditory cortex, subcortical auditory areas, and the AF (Figs. 5, 7); (3) left-handed musicians achieve atypical lateralization of language through an overdevelopment of the CC, whereas nonmusicians do so due to a lack of anatomical development of the AF (Fig. 8a,b); and (4) both pathways converge to enhance the prominence of interhemispheric connectivity within the brain, increasing the probability of shifting language function to the right hemisphere (Fig. 8c).
Previous functional neuroimaging studies in right-handers have demonstrated that a left bias is observed in the STG when hearing speech (Binder et al., 1996; Peña et al., 2003; Dehaene-Lambertz et al., 2010). In line with this, we have showed that the generation of verbs using the right IFG—atypical lateralization of speech—is associated with a decrease in this leftward bias of the STG during the processing of word stimuli. This decrease expands to other temporal areas such as the middle and fusiform gyri and reaches hemispheric reversal—rightward implication—in the IFG, caudate, and thalamus. Thus, it encompasses the prelexical pathway of auditory speech processing and the semantic processing of familiar words (Hickok and Poeppel, 2007). Importantly, when contrasting the hearing of speech and tones [as in Binder et al. (1996) and Dehaene-Lambertz et al. (2010)], a hemispherically reversed cortical network was found in frontal and temporal regions (leftward for words, rightward for tones). Beyond the cerebral cortex, it was also evident in auditory subcortical structures including the MGB of the thalamus and the inferior colliculi. Previous research has implicated these structures in sound-to-speech representation (Mihai et al., 2019), integrating spectral and temporal information (Miller et al., 2002), auditory learning (Gilad et al., 2020), and the transformation from low-level acoustical analysis to words (Pannese et al., 2015). Thus, in comparison with tone processing, the entire network involved in auditory word processing exhibits some degree of lateralization shift when speech production is rightward dominant.
Speech production tasks also elicit activations in the right posterior areas of the temporal and parietal cortices (Szaflarski et al., 2006). However, there is evidence of mismatch in the lateralization of different language components among some healthy controls (Labache et al., 2020) and patients with brain lesions (Lee et al., 2008). A recent proposal has suggested that a two-factor model of lateralization, which separates language production and comprehension, provides a better explanation for this phenomenon than a single-factor approach (Parker et al., 2022). Consistent with this proposed theory of independence, while at the group level our data demonstrate a strong correspondence between the lateralization of verbal generation and word listening, subjectwise analysis reveals discrepancies between these functions in 30% of studied left-handers. This should be considered when extrapolating our results to other studies with different tasks.
Our main contribution is the demonstration of a direct correlation across the entire sample—regardless of musicianship—between atypical lateralization of language and an increased interhemispheric brain connectivity. This applies both to brain structure (measured by indexing the normalized volumes of the AF and CC) and function (assessing ROI-to-ROI inter- and intrahemispheric functional connectivity at rest). This stronger interhemispheric control would increase the likelihood of shifting language control from the left to the right hemisphere (Labache et al., 2020). Our results align with the hypothesized role of the CC in hemispheric specialization: (1) in primates and humans, CC volume and connectivity decrease as cognitive functions increase in lateralization (Hopkins and Rilling, 2000; Karolis et al., 2019); (2) CC agenesis is related to a failure in developing left lateralization of language (Hinkley et al., 2016); and (3) larger CC volume has been reported in a substantial sample of atypically lateralized individuals (Labache et al., 2020).
Crucially, our results also align with the developmental trajectory of language and its involved white matter tracts. The normal ontogeny of language implies a shift from a predominantly interhemispheric control through the CC to a left intrahemispheric control facilitated by the AF (Szaflarski et al., 2006; Friederici et al., 2011; Perani et al., 2011). Consistently with this, recent evidence has linked the functional lateralization of cognitive functions with callosal shrinkage (Karolis et al., 2019), which has been attributed to processes of fiber myelination, redirection, and pruning (Luders et al., 2010), influencing cognitive function for at least the first 30 years of life (Danielsen et al., 2020). Similarly, the anterior segment of the AF—connecting the inferior parietal cortex with the premotor cortex—constitutes the dorsal stream of language responsible for sound-to-articulation mapping (Saur et al., 2008). The maturation of this fasciculus is not complete until adolescence (Uda et al., 2015; Tak et al., 2016), determining not only the development of language capabilities (López-Barroso et al., 2013) but also the intrahemispheric potential of IFG-STG functional connectivity (Reynolds et al., 2019). Therefore, these two white matter tracts are critically involved in language development, and, due to their slow maturation, they play that role over an extended period of an individual's life.
The first pathway is associated with musicians exhibiting a positive relationship between atypical hemispheric specialization of language and posterior CC volume. This association is consistent with investigations showing that (1) musical training transversally and longitudinally enhances posterior CC connectivity (Hyde et al., 2009; Elmer et al., 2016; Leipold et al., 2021); (2) the age of onset and duration of musical training during childhood and adolescence increase connectivity in the posterior CC (Bengtsson et al., 2005; Steele et al., 2013); and (3) acquired amusia is associated with lesions in the posterior CC (Sihvonen et al., 2017). Thus, increased posterior CC volume—as a consequence of training (Bengtsson et al., 2005; Steele et al., 2013)—may constitute the relevant factor that alter the normal development of language, shifting the control of these functions to the right hemisphere. Interestingly, it has been reported in right-handers that stimulation to the right (but not the left) auditory cortex results in widespread inhibitory influences on auditory and motor-related networks at rest, with its interhemispheric propagation directly related to the size of posterior CC fibers (Andoh et al., 2015). So, it is plausible that the reduced implication of the left AF during language production—observed in the Functionnectome analysis of the atypical musicians—reflects the effects of right-to-left inhibition of the auditory cortex through the posterior CC. Furthermore, we have observed that atypical lateralization was mostly found in the musicians that (1) had more years of training; (2) actively pursued musical experiences; and (3) devoted their training to a single instrument. We must, however, consider that fMRI-based lateralization does not provide the causality demonstrated by other techniques such as the Wada test or transcranial magnetic stimulation (TMS). Consequently, we should not rule out the possibility that this atypical pathway—linked to musicianship and a callosal enlargement—could represent an adaptation or cognitive strategy for enhanced efficiency in language processing. That is, the capability of their left hemisphere to produce language remains uncertain. Future studies might investigate this aspect by means of TMS.
The second pathway to atypical language lateralization, observed in nonmusicians, involves reduced structural connectivity across the anterior segments of the bilateral AF. The observed pattern of diminished intrahemispheric connectivity—both structurally and functionally—and enhanced interhemispheric functional connectivity resembles the pattern displayed during early infancy (Friederici et al., 2011; Perani et al., 2011). Thus, it may reflect a failure in the transition from a more interhemispheric to a more intrahemispheric language control, which typically occurs between the ages of 7 and 18 in conjunction with left AF maturation (Szaflarski et al., 2006; Friederici et al., 2011; Tzourio-Mazoyer, 2016; Reynolds et al., 2019). The exact causes of this failure remain unclear, and we can only speculate about the reported relationships between the maturation of the language system and the characteristics of the early exposure to vocalizations and speech (Estrada et al., 2023).
Our results have two broader implications for future research. First, the consideration of two different developmental pathways to atypical lateralization may facilitate the comprehension of this rare condition (Mazoyer et al., 2014), its consequences (Mellet et al., 2014; Gerrits et al., 2020; Villar-Rodríguez et al., 2024), and its relation to certain neurological disorders (Sommer et al., 2001; Li et al., 2007; Illingworth and Bishop, 2009; Jouravlev et al., 2020). It may even explain the difficulties encountered when investigating anatomical and functional correlates of this phenomenon in the past, as previous studies were attempting to describe a unitary mechanism when at least two different mechanisms may be at play. Second, this study supports the notion of musical training as an excellent model for investigating brain reorganization (Herholz and Zatorre, 2012). We have described how this training may, in some cases, modify the normal development of other cognitive functions of the brain, such as language.
Footnotes
We thank all the participants for their collaboration in this study, as well as the radiographers at the clinic ASCIRES-Castellón for their valuable assistance during data acquisition. We also thank Victor Nozais for his support with the Functionnectome toolbox. E.V.-R. was supported by an FPU predoctoral grant funded by the Spanish Ministry of Education, Culture and Sports (FPU18/00687) as well as a research stay grant funded by the Spanish Ministry of Education and Professional Formation (EST23/00268). L.M.-M. was supported by a MSCA Postdoctoral Fellowship funded by Horizon Europe Guarantee and UKRI (EP/Y014367/1). C.A. received funding from the Spanish State Research Agency (PID-2019-108198GB-l00), Generalitat Valenciana’s Ministry for Innovation, Universities, Science and Digital Society (IDIFEDER/2021/2021/026), and the Universitat Jaume I (UJI-B2021-11).
The authors declare no competing financial interests.
- Correspondence should be addressed to César Ávila at avila{at}psb.uji.es.
