The descending motor tracts are different in dancers and musicians

Abstract

Long-term motor training, such as dance or gymnastics, has been associated with increased diffusivity and reduced fiber coherence in regions including the corticospinal tract. Comparisons between different types of motor experts suggest that experience might result in specific structural changes related to the trained effectors (e.g., hands or feet). However, previous studies have not segregated the descending motor pathways from different body-part representations in motor cortex (M1). Further, most previous diffusion tensor imaging studies used whole-brain analyses based on a single tensor, which provide poor information about regions where multiple white matter (WM) tracts cross. Here, we used multi-tensor probabilistic tractography to investigate the specific components of the descending motor pathways in well-matched groups of dancers, musicians and controls. To this aim, we developed a procedure to identify the WM regions below the motor representations of the head, hand, trunk and leg that served as seeds for tractography. Dancers showed increased radial diffusivity (RD) in comparison with musicians, in descending motor pathways from all the regions, particularly in the right hemisphere, whereas musicians had increased fractional anisotropy (FA) in the hand and the trunk/arm motor tracts. Further, dancers showed larger volumes compared to both other groups. Finally, we found negative correlations between RD and FA with the age of start of dance or music training, respectively, and between RD and performance on a melody task, and positive correlations between RD and volume with performance on a whole-body dance task. These findings suggest that different types of training might have different effects on brain structure, likely because dancers must coordinate movements of the entire body, whereas musicians focus on fewer effectors.

Appendices

Supplementary Material

Sample-specific template

A sample-specific template with its parcellation of GM and WM regions was created from the raw images in order to define the center of gravity (COG) of the hand motor regions and standardize some operations across the sample. For instance, a sample-specific parcellation of GM and WM, implemented with the Freesurfer’s Desikan-Killiany Atlas, was automatically calculated on the sample-specific template to localize the precentral motor cortices to guide the regions of interests (ROIs) localization. To make the template, all individual raw brain images were first projected to Freesurfer’s operational space, named conformed space, applying the Freesurfer’s tool recon_all. With this command, the individual labels of the GM and WM parcellated regions, necessary at different steps of the seed mask creation, were also produced. With a combination of Freesurfer’s and FSL’s tools, the structural brain images were transformed into the MNI152 standard space and averaged to constitute the sample-specific template in the MNI space.

Fig. 5

Template creation. a Individual structural brain image. b Individual brain image in MNI space. c Averaged brain image in MNI space. d Averaged brain image in conformed space, i.e., template

More specifically, the brain structural images were transformed from the structural space (5a) to the MNI space (5b) with linear and nonlinear transformations, using FSL’s FLIRT and FNIRT tools. They were then averaged, with FSL’s Fslmaths tool, in the MNI space (5c) and transformed into the conformed space (5d) with recon-all, which also created the GM and WM parcellations of the template.

Seed mask creation

In order to separately track the primary motor pathways connecting the head, hand, trunk or leg representations, we defined, in each hemisphere, the four WM regions subjacent the motor cortex that topographically correspond to the head, hand, trunk/arm or leg/foot representations. To do this, we developed a procedure that permits to define the seed masks for tractography, using an approach that is at the same time sample-specific and reproducible across subjects. See the diagram in Fig. 6 for the procedure followed.

Fig. 6

Diagram of the steps followed to create the seed masks. The yellow windows indicate the user-defined steps; the light orange windows indicate the steps that were standardized across subjects

Creation of sample-specific hand centers of gravity (COGs)

The hand is the only body part whose topographical location within the motor cortex that can be identified using well-established gross anatomical landmarks (Yousry et al. 1997; Caulo et al. 2007). Therefore, the first step was to identify the hand motor region individually, created a sphere based on its center of gravity (COG) and then to locate the other body parts along M1 relative to its position. Hand motor regions for each individual were manually labeled in three dimensions for both hemispheres in Freesurfer’s conformed space (7a). To do this, we overlaid the GM label of the precentral cortex provided by the Desikan-Killiany Atlas on each individual structural brain. To create the hand masks, we identified the hand regions defined by the hand landmarks, previously described as an omega or epsilon shaped portion of the cortex, with their variants (Yousry et al. 1997; Caulo et al. 2007). The user-drawn hand regions were then nonlinearly projected onto the MNI152 standard space with the FSL’s tool applywarp (7b). The hand masks were then averaged (7c), with the FSL’s tool Fslmaths, and projected again onto the sample-specific template space (7d), using the Freesurfer’s tool mri_vol2vol. Finally, the COGs of the average hand masks were calculated, in the template conformed space, using the FSL’s tool Fslstats.

Fig. 7

Calculation of the hand center of gravity (COG) for the right hemisphere, from manually drawn hand ROIs. a Individual manually drawn hand ROI in the subject Freesurfer’s conformed space. b Individual manually drawn hand ROI in the MNI structural space. c Average of all subjects manually drawn hand ROIs in MNI 2 mm space. d Average of manually drawn hand ROIs in template conformed space (thresholded at 30%)

Creation of seed masks from COGs across subjects

Once the hand COGs were calculated, the appropriate radii were estimated to draw the hand spheres around them. These parcellations were critical as visual checks for the determination of the radii of the spheres, allowing us to ensure that the spheres included all the relevant portions of the precentral cortex parcellation (Desikan-Killiany Atlas). The trunk, leg and hand spheres were created relative to the hand sphere positions by shifting the hand COGs along the precentral cortex (Table 4). In particular, the trunk sphere was entirely shifted along the x-direction medially, so that the sphere included a similar portion of the cortex with minimal overlapping. For the leg and head, the spheres were first shifted along the x-direction—medially and laterally, respectively—and then along the z-direction (both spheres) and the y-direction (head sphere only), in order to include all the cortex.

Table 4 Coordinates of the centers of gravity (COGs) in the MNI space (mm) for the four motor regions

The radii were also adjusted to include all the relevant portion of the precentral motor cortex. In the end, the hand and trunk spheres had the same radii, whereas the leg and the head regions, being more elongated, had bigger radii. Because there was some overlap between the hand and the trunk spheres, the trunk spheres were subtracted from the hand spheres; this was not necessary for the other spheres. While the hand, head and trunk spheres were masked with the precentral WM parcellation (see below), the leg spheres were masked with the paracentral WM parcellation; therefore there was no problematic overlapping with the trunk regions. It is worth noting that the paracentral WM underlies both the pre- and post-central cortices, and therefore, the leg tracts were not restricted to the descending primary motor fibers, but included also the ascending sensorimotor fibers.

Once obtained the bilateral COGs and spheres, we projected the spheres onto the structural subject space and masked them with the individual precentral WM masks. The choice of masking the spheres with the individual WM masks was done to take advantage of the more precise individual parcellations compared to the template parcellation. In detail, the spheres were projected from the template conformed space (Fig. 7a) into the standard MNI152 space (Fig. 7b) applying the mri_vol2vol tool by Freesurfer. In this passage, the transformation matrix came from the previously calculated projection of the MNI space to the template conformed space, performed with tkregister, and was here inverted with the -inv option. A nonlinear warping procedure was then applied to project the spheres from the MNI space (Fig. 3b) to the individual structural space (Fig. 7c), using applywarp fed with linear and nonlinear warping, previously calculated with FLIRT and FNIRT. Then, by means of Fslmaths (‘–mas’ option), the spheres were masked with the WM individual parcellation in the structural space, previously transformed from the individual conformed space with FLIRT. Finally, the selected WM ROIs were projected onto the diffusion FA space with a linear transformation (FLIRT). The WM ROIs in FA space were now ready to be used as seed masks for tractography (Fig. 8).

Fig. 8

Seed mask creation from spheres to final masks in the right hemisphere. a Spheres and COGs are shown in the template conformed space for the hand (yellow sphere, COG indicated by red arrow in a1), trunk (light blue sphere, COG in a2), leg (green sphere, COG in a2) and head (fuchsia sphere, COG in a3). b Spheres in MNI152 space. c Spheres in one subject individual structural space overlapping with the precentral/paracentral cortex; bottom raw shows the head transformation not visible in the same slice as the other body-part regions. d Individual final seed masks in one subject FA diffusion space. The light blue arrows between images indicate the transformations between images, the commands used, above the arrow, and the software, below the arrows (FSL and Freesurfer, FS). Above the main images transformations, the transformation matrix estimations are shown to indicate where the transformation matrices come from. For instance, to project the spheres from the template space to the MNI space, the inverted matrix created with the command tkregister from the MNI space to the template conformed space was applied (see Fig. 5b in Supplementary Material$\rightarrow$d)