Data from two groups of participants were used. In one study group, both structural and functional MRI data were acquired (n = 20), but in the other study group, only structural MRI data were available (n = 30).
Structural and functional MRI data from the performance of motor tasks were acquired from 20 participants (12 females, 18–40 years [27.4 ± 5.6, mean ± SD], 5 self-reported left-handers). All participants were self-reported native English speakers (two were raised bilingually from infancy and three were fluent in a second acquired language) and had no history or diagnosis of speech disorders. All had normal hearing, normal or corrected-to-normal vision, and no neurological impairments. The participants were part of a study that had been approved by the Central University Research Ethics Committee of the University of Oxford (CUREC, R55787/RE001) in accordance with the regulatory standards of the Code of Ethics of the World Medical Association (Declaration of Helsinki). All participants gave informed consent for their participation and were monetarily compensated.
In addition, we used cortical brain surface reconstructions from 30 participants provided by the Human Connectome Project (HCP), WU-Minn Consortium (Principal Investigators: David Van Essen and Kamil Ugurbil; 1U54MH091657) funded by the 16 NIH Institutes and Centers that support the NIH Blueprint for Neuroscience Research; and by the McDonnell Center for Systems Neuroscience at Washington University (Van Essen et al. 2013). The minimally pre-processed datasets of the first 31 participants (16 female, age range 22–35 years, handedness information not accessed) of the Q2 release were used. One participant was excluded because of a technical problem in the automatic FreeSurfer parcellation.
MRI data acquisition
MRI data acquisition parameters differed for the two groups of participants. Data from the participants that took part in the functional study were obtained at the Oxford Centre for Human Brain Activity (OHBA) using a 3 T Siemens Prisma scanner with a 32-channel head coil. Two structural images of the whole brain had been acquired at 1 mm isotropic resolution: a T1w image (MPRAGE sequence) and a T2w image (SPACE sequence). For task-fMRI, whole head T2*-weighted echo planar images were acquired at 2.4 mm3 isotropic resolution (TE = 30 ms, multiband factor 6, TR = 0.8 s, Casey et al. 2018).
Details of data acquisition and preprocessing methods of the HCP participants are provided in Glasser et al. (2013) and Uǧurbil et al. (2013). T1w images had been acquired using an MPRAGE sequence at 0.7 mm isotropic resolution.
Structural MRI data analysis
Data from the participants who took part in the functional study were pre-processed using the HCP-pipeline (Glasser et al. 2013). The automatic processing pipeline includes cortical surface reconstruction using FreeSurfer based on the contrast from the T1w and the T2w images and automatic assignment of neuroanatomical labels. Cortical surface reconstructions of the HCP participants were derived using FreeSurfer based on the T1w scans directly provided by the database. A linear transformation (12 degrees of freedom) from the high-resolution T1w anatomical scans to standard MNI space (nonlinear 6th generation atlas, Fonov et al. 2011) was derived using FSL’s FLIRT (Jenkinson and Smith 2001; Jenkinson et al. 2002) and further refined using FNIRT nonlinear registration (Andersson et al. 2007).
We identified the following sulci and sulcal segments in the structural data from all 50 participants: five segments of the central sulcus from dorsal to ventral (cs_1 to cs_5), the lateral and opercular segments of the anterior subcentral sulcus (ascs_lat, ascs_op) and the posterior subcentral sulcus (pscs) (Fig. 1a). Sulcal labels for one example participant are shown in Fig. 1b and Fig. 1c and a three-dimensional rendering of this individual’s segments is provided as animation in the supplementary material (Online Resource 1) generated using FSLeyes (McCarthy 2020). Note that all sulcal labels and surface reconstructions are openly accessible to allow interactive inspection in a 3D viewer. Sulcal labels were drawn manually onto the native surface mesh (approximately 136,000 vertices) in Connectome Workbench’s wb_view (www.humanconnectome.org/software/connectome-workbench.html). Surface features of both pial and white matter surfaces were inspected in conjunction with the participant’s T1w scan. The identification of the central sulcus segments was based on changes in direction of the sulcus, and on the location of gyral ‘plis de passage’, which are small gyral bridges connecting the postcentral with the precentral gyrus, or based on more subtle ‘wall pinches’. These landmarks can be most easily identified on the white matter surface and are not always visible on the pial surface alone (Germann et al. 2020). The following description of morphological criteria was sufficient to identify the sulci in all 50 participants.
Cs_1 is the most dorsal segment of the central sulcus, which runs more or less in a vertical straight direction. Its ventral boundary was drawn at the location where a gyral bridge provides a prominent landmark on the posterior bank of the central sulcus. Cs_2 has a characteristic curvature in the shape of the Greek omega letter (see Fig. 1b, c), which is known as the ‘hand knob’ (Yousry et al. 1997). The hand knob can be split into two smaller knob-like curves, the so-called epsilon configuration (Yousry et al. 1997). If this configuration were present, both knobs were labeled together as cs_2 and the pinch on the posterior wall of the central sulcus in the middle was not treated as a boundary between these labels. The boundary between cs_2 and cs_3 was drawn at the location where the central sulcus changes direction and where a gyral passage can be observed on the posterior bank. In some brains, an additional convexity of the central sulcus can be observed on the posterior bank in the middle of cs_3. The ventral boundary of cs_3 was drawn ventral to this convexity, if present, at the location where a small gyral bridge forms a landmark on the anterior bank of the central sulcus. The last two segments of the central sulcus, cs_4 and cs_5 are smaller in extent, shallower and more variable in their morphology (see “Results” section for a more detailed anatomical characterization). The boundary between cs_4 and cs_5 was defined based on a gyral bridge on the posterior bank of the central sulcus. Cs_5, which is the most ventral part of the central sulcus, can form an additional curve or run straight in a variable direction.
The labels for ascs and pscs were assigned based on an atlas of human brain morphology (Petrides 2019). For ascs, we labeled two distinct segments: a lateral and an opercular segment (ascs_lat, ascs_op). The course of ascs_lat, ascs_op and pscs was highly variable and a detailed description of the sulcal anatomy in the subcentral region is reported in the results section.
The morphological patterns of the ventral subcentral region were categorized into five types depending on the configuration of the ascs_lat. The classification was based on the location of the ascs_lat on the cortex and its spatial relation to other sulci. Sulcal segments were considered as ‘merged’, when there was a clear continuation on the pial surface, although, in some cases, a discontinuity between the merged sulci was still observed on the white matter surface.
Spatial probability maps
To characterize the inter-individual morphological variability of the labeled sulci, we generated probability maps in surface and volume space. To obtain surface probability maps, all surface labels were resampled from native to a standard mesh (32k_fs_LR) following FreeSurfer’s registration (Glasser et al. 2013). At each vertex, the 50 surface maps were binarized, summed and then normalized to create a surface label with intensities ranging from 0% to 100% at the maximal possible overlap of all 50 participants. For visualization, the surface probability maps were displayed onto an inflated template surface.
For the generation of volumetric probability maps, individual surface labels were mapped to volume space. For surface-to-volume mapping, we used the participant’s native pial surface registered to 32k_fs_LR mesh and non-linearly transformed to MNI-space (i.e. the surfaces in the subfolder /MNINonLinear/fsaverage_LR32k/). We performed the mapping to 0.5 mm resolution volume space using wb_command and ‘-nearest-vertex’ option. Given that the volumetric labels spatially varied in 3 rather than in 2 dimensions, the labels were smoothed (Gaussian kernel with FWHM of 2 mm) to enhance overlap. The smoothed labels were then thresholded at 0.1, binarized and then summed at each voxel. For visualization, volume probability maps were overlaid onto the MNI average brain.
Anatomical characterization of sulcal segments
We characterized the sulcal segments using several anatomical descriptors. Inter-individual variability was determined as the maximal value within the surface probability map. For the measure of sulcal depth, we computed the minimal value in each individual’s sulcal-segment labels using FreeSurfer’s sulcal depth map. The depth value is based on an individual’s pial surface and expressed as a normalized value relative to a baseline running along the sulcal banks, rather than in real-world units. The depth values range from positive values at the crown of a gyrus to negative values at the fundus of a sulcus. Furthermore, we computed the mean cortical thickness in each label, as provided by FreeSurfer. As a measure for the size of the sulcal segment, we computed the number of vertices that the surface label spanned on the native surface.
Functional MRI experimental design
The 20 participants who provided structural and functional data took part in an fMRI study on speech production and laryngeal motor control. The experimental design, processing, and fMRI results of this study have been reported elsewhere in detail (Eichert et al. 2020a) and are here only briefly described. In a functional localizer task, participants were asked to perform repeated lip protrusion or tongue retraction at a rate of approximately 1–2 reps/s. The participant’s breathing pattern was explicitly controlled using the fixation symbol on the screen, instructing them to inhale for 1.5 s and exhale for 4 s. A ‘breathing only’ condition, during which the participants followed the same breathing pattern, was acquired as baseline condition. Each task condition was performed in blocks lasting 22 s followed by a rest period of 8 s with normal breathing. The conditions were presented in a fixed pseudo-random order following a balanced Latin-square design wherein each condition was repeated four times.
In a separate task, participants were instructed to produce a syllable sequence (/la leɪ li la leɪ li/) under four different conditions: overt speech, silent mouthing, only vowel production, and covert speech. Breathing instructions, task timing and randomization of the four blocks were the same as described for the localizer task, except that each condition was repeated five times. In a third task, participants performed a task that required both phonological and semantic judgements. Participants had to indicate a yes/no response by pressing a button with the right index or the middle finger every 3 s. This task was analyzed as a localizer for the hand region in the left hemisphere.
Functional MRI data analysis and statistical analysis
Functional MRI data processing was carried out using FEAT (FMRI Expert Analysis Tool) Version 6.00, part of FSL (FMRIB's Software Library, www.fmrib.ox.ac.uk/fsl) including motion correction of the images and unwarping using a fieldmap (Jenkinson 2003). Time-series statistical analysis was based on a general linear model (GLM) implemented in FILM with local autocorrelation correction (Woolrich et al. 2001). Standard motion correction parameters and individual volumes that were motion outliers, determined using fsl_motion_outliers, were included as separate regressors at the first level for each participant. Registration to the high-resolution structural scan and standard 2-mm MNI template was carried out using FLIRT. Registration from high-resolution structural to MNI space was then further refined using FNIRT nonlinear registration (Andersson et al. 2007).
In the functional localizer task for lip and tongue movements, activity during each condition was assessed relative to the ‘breathing only’ condition. For the syllable production task, the conditions were analyzed in a factorial model that allowed separation of the (supra-laryngeal) articulation and the (laryngeal) vocalization components of the task. Brain activity associated with the control of supra-laryngeal articulation was defined as (‘overt speech’ minus ‘vowel production’) plus (‘silent mouthing’ minus ‘covert speech’) and the main contrast for vocalization was derived from the contrast (‘overt speech’ minus ‘silent mouthing’) plus (‘vowel production’ minus ‘covert speech’). The task activations from the ‘articulation’ contrast are not further discussed in this manuscript.
In both tasks described above, the rest blocks with normal breathing served as baseline, which means that they were not modeled in the GLM. For the hand localizer task, we derived a contrast of all conditions involving button presses relative to a resting baseline. Note that this task only provided data to analyze the hand representation in the left hemisphere.
Individual surface activation maxima
To assess inter-individual variability of the fMRI results, we derived the location of individual activation maxima for hand, lip, and tongue movements, and larynx activity during vocalization. Activation maxima were derived using the steps described in Eichert et al. (2020a), and are reported here only briefly. ROI definitions are described in more detail in the supplementary material.
Different volumetric ROI masks were used for the different motor representations based on individual anatomy in both hemispheres. We used an ROI of the whole central sulcus for hand, lip and tongue, based on the Destrieux Atlas. For the dorsal larynx representation, we used a more limited portion of the central sulcus ROI (MNI z-coordinates: 50 – 30). For the ventral larynx representation, we manually defined an ROI based on individual surface landmarks.
Individual volumetric ROIs were linearly transformed from FreeSurfer’s anatomical to functional space of the respective task fMRI scan. Within the ROI, the voxel of maximal intensity was determined from the uncorrected z-statistic image. Activation maxima were manually inspected in the participant’s native volume space to confirm that the systematic approach described below captured task-related activations. It should be noted that, for some participants, this local maximum did not achieve the corrected voxel-wise significance threshold (left hemisphere: hand n = 3, dorsal larynx n = 6, ventral larynx n = 5; right hemisphere: dorsal larynx n = 5, ventral larynx n = 4). Using a lower uncorrected threshold is justified given our goal to visualize and assess spatial variability of the activation maxima. The activation maxima were mapped to the individual’s native mid-thickness surface, resampled to the 32 k standard (fs_LR) surface mesh using the FreeSurfer registration, smoothed (FWHM = 1 mm), and binarized to form a small circular patch.
Next, we examined the spatial relationship between the sulcal segments, and the task activation peaks at the individual participant level. This analysis was performed in the 20 participants who contributed both task activation and structural data. Individual task activation peaks were mapped onto the individual’s cortical surface as described above for sulcal segments. To characterize the structure-to-function relationship at the group level, we aligned all individual surfaces based on the anatomical surface labels and then we applied the same registration to individual task activation peaks. This approach allowed us to visualize individual variability in the spatial distribution of task activation peaks in a common group-level space.
The registration of sulcal labels was driven by the binary labels for cs_1, cs_2, cs_3, cs_4, cs_5, ascs_lat, ascs_op and pscs and performed using the multimodal surface matching tool (MSM, Robinson et al. 2014), which is part of FSL. As target, or reference, for the MSM-based registration, we used the normalized and thresholded (> 0.4) average labels after projecting all of them to the same regular sphere (approximately 32,000 vertices) without using any other anatomical priors. Each participant’s sulcal maps and the reference sulcal maps were merged into a combined file with six data arrays, i.e. they were provided as metric ‘func.gii’ file with six maps. Then, we derived a registration using MSM for each participant. We used default settings for MSM and the following configuration parameters were empirically determined: levels = 3; sigma_in = 10, 5, 2; sigma_ref = 10, 5, 2; lambda = 0.1, 0.01, 0.01; it = 10, 10, 3; opt = DISCRETE, DISCRETE, DISCRETE; CPgrid = 2, 3, 4; SGgrid = 4, 5, 6; datagrid = 4, 5, 6.
We explored how manual labeling of sulcal segments can help to align data across participants. Thus, we compared our registration based on sulcal segments with the FreeSurfer registration and a registration based on sulcal depth maps. The FreeSurfer registration and sulcal depth maps were provided by the HCP-processing pipeline. To derive a registration based on sulcal depth, we used MSM default settings. To characterize the effect of the registrations, we quantified the spatial spread of the fMRI activation peaks: For each effector, we computed the median geodesic distances across all 20 activation peaks. The median distance for each effector was, therefore, based on 190 datapoints. A smaller spatial spread of activation peaks indicates a better registration performance.