Functional data from the mapping/fixation task runs were preprocessed via Statistical Parametric Mapping software (SPM12; Wellcome Department of Cognitive Neurology, UCL, UK; http://www.fil.ion.ucl.ac.uk/spm) and Matlab 16a (http://www.mathworks.com). The first 5 volumes of each run were removed to allow for T₁ equilibrium effects. We applied a slice-timing correction using the time center of the volume as reference. Then, head motion correction was applied using rigid-body realignment. These realignment parameters were used as covariates during the statistical analysis to model out potential nonlinear head motion artifacts. Functional and morphological images were spatially normalized into standard MNI space using SPM's default templates. Functional data were finally smoothed using a 6-mm full-width half maximum Gaussian kernel (Friston et al., 1995a,b,c). A 128-s temporal “high-pass filter” regressor set was included in the design matrix to exclude low-frequency confounds.

Each task trial was modeled with impulse regressors at the time of the presentation of the fixation point that initiated the performance of the hand, tongue, and saccadic eye movements, as well as ocular fixation. These regressors were then convolved with the canonical hemodynamic response function and entered into a general linear model (GLM) of each subject's fMRI data. The 6 scan-to-scan motion parameters produced during realignment were included as additional regressors in the GLM to account for residual effects of subject movement. Statistical significance in the resulting single-subject contrast images was assessed with a p_uncorrected < 0.001 voxel-wise threshold.

The preprocessing of resting-state scans was also performed with SPM 12. The first 5 volumes of each run were removed to allow for T₁ equilibrium effects. We then applied a slice-timing correction using the time center of the volume as reference. The head motion correction was applied using rigid-body realignment and, then, using the AFNI software (Cox, 1996), a temporal filtering was applied to extract the spontaneous slowly fluctuating brain activity (0.01–0.1 Hz). Finally, a linear regression was used to remove nuisance variables (the six parameter estimates for head motion, the cerebrospinal fluid and white matter signals from the SPM segmentation, the number of heart pulses per TR, and the mean pupil size per TR). Note that a customized Matlab program was used to detect and count the number of peaks in the PPG signal in each TR window. Automatic peak detection was performed using the “findpeaks” function from the Signal Processing Toolbox (https://www.mathworks.com). Raw pupil diameter data were extracted from the Eyelink datafile and subsequently averaged in each TR window via Matlab. Finally, a spatial smoothing with a 6-mm FWHM Gaussian kernel was applied to the output of the regression.

The main goal of the resting-state fMRI analyses was to investigate, on a subject-by-subject basis, the differential functional connectivity associated with the various CMAs as well as their individual motor representations (i.e., tongue, eye, and hand). Specifically, our analyses were focused on the left intra-hemispheric connectivity of the CMAs (and their motor representations) with the lateral prefrontal and motor cortical areas, as well as with connectivity between the CMAs. For this analysis, a total of 15 locations in the left hemisphere of each subject were identified: in the CMAs (8 seeds), prefrontal cortex (5 ROIs), and motor cortex (3 ROIs) (see below and Figure 3).

For each subject, 8 CMA seed locations in the left cingulate cortex were identified based on activation peaks from the fMRI motor-mapping task. These included the hand, eye, and tongue motor representations in RCZa and RCZp, and the two hand motor representations in CCZ (see Figures 2, 3).

For each subject, 3 ROIs within the left motor cortex were identified based on activation peaks from the fMRI motor-mapping task. These included the hand motor region (the precentral knob) within the central sulcus –M-H– (Boling et al., 1999; Amiez et al., 2006), the tongue motor region within the ventral part of the posterior part of the precentral gyrus –M-T– (Weiss et al., 2013), the frontal eye field –FEF– within the ventral branch of the superior precentral sulcus (Amiez et al., 2006). Monkey studies have shown that the CMAs are anatomically linked with the motor cortex (e.g., Dum and Strick, 1991; Procyk et al., 2016). The motor ROIs were selected to verify the connectivity profiles of the individual motor representations in the various CMAs with the motor system. Considering that the CMAs contain tongue, eye, and hand motor representations (Amiez and Petrides, 2014), we assessed the connectivity of these regions with the primary tongue area, the FEF, and the primary hand area. Note that the CMAs also contain foot motor representations but these had not been mapped in our mapping task. As such, in the present study, we did not assess CMAs' connectivity with the primary foot motor cortex (Amiez and Petrides, 2014). Figure 3 displays the location of these ROIs.

For each subject, 5 ROI locations within the left prefrontal cortex were identified based on local anatomy. These included area 10 –a10– (located at the intersection between the vertical segment of the intermediate frontal sulcus, the lateral and the medial frontomarginal sulcus, see Petrides, 2014), dorsolateral prefrontal areas 46 –a46– [within the dorsal paraintermediate frontal sulcus (pimfs-d)] and 9/46 –a9/46– [within the anterior segment of the posterior middle frontal sulcus (pmfs-a), see Amiez and Petrides, 2007], area 44 –Broca– [Broca's area, in the center of the pars opercularis, i.e., between the anterior ramus of the lateral fissure (aalf) and the inferior precentral sulcus (iprs), see (Petrides, 2014)], and the frontal operculum –Fo– (intersection between the frontal operculum and the circular sulcus, see Amiez et al., 2016). These ROIs were selected because of their known frequent co-activation with the MCC in a large range of cognitive tasks (Amiez and Petrides, 2007; Amiez et al., 2012a,b, 2013). ROI locations on the lateral surface of the prefrontal cortex are shown in Figure 3. Note that the center of each of the ROIs located in the sulci was positioned about 5 mm below the surface given the chosen size of the radius sphere (i.e., 4 mm, see below).

Based on the above 15 locations, seed/ROI spheres with a 4 mm radius were generated for each subject using the AFNI software and the mean signal form these regions was extracted. Note that this radius size was selected in order to allow the separate assessment of functional connectivity of seeds located in the CGS from those located in the PCGS when present. Indeed, the PCGS and CGS are most often separated of about 8–12 mm, so a larger radius would have prevented us to assess this putative dissociation. For each subject, correlation coefficients between the different CMA seeds with the various ROIs in the prefrontal cortex, motor cortex and other CMA seeds were computed and normalized using the Fisher's r-to-z transform formula. Significant threshold at the individual subject level was Z = ±0.2 (p < 0.001) and Z = ±0.15 (p < 0.01). These normalized correlational coefficients, which corresponded to the functional connectivity strength between each CMA seed and each ROI/other CMA seed in individual subjects, were subsequently processed with R statistical software (https://www.r-project.org/) for all the following analyses.

To compare the connectivity between the different CMAs and lateral frontal cortex, we first averaged the normalized correlation coefficients for each seed-ROI pairing across subjects. Next, Euclidean distance vectors were separately computed for the 8 CMA seeds (based on their mean connectivity values with the ROIs) and the 8 prefrontal/motor ROIs (based on their mean connectivity values with the CMA seeds). Finally, unsupervised hierarchical clustering was performed for the CMA seeds and ROIs separately. This clustering was generated using the hclust function in R which uses the complete linkage method for hierarchical clustering. This particular clustering method defines the cluster distance between two clusters to be the maximum distance between their individual components. At every stage of the clustering process, the two nearest clusters are merged into a new cluster. The process is repeated until the whole data set is agglomerated into one single cluster. This clustering method assumes a representation of data in an Euclidean space in which it is possible to summarize a collection of points by their centroid (i.e., the average of the points) (for more details, see http://www.r-tutor.com/gpu-computing/clustering/hierarchical-cluster-analysis). The outcome was used to construct dendrograms and heatmaps. To better display clusters across ROIs, values in the heatmaps were normalized (z-scored) by column. Hence values (and sign) in the heatmap do not represent actual connectivity measures.

To compare the connectivity profile of each CMA with the various lateral frontal ROIs, we constructed boxplots corresponding to the correlation strength of each CMA location (RCZa, RCZp, CCZa, and CCZp) with each of the 8 ROIs (Figure 4B). Note that we divided the CCZ into an anterior and posterior part in view of the fact that there were two distinct CCZ hand representations. This allowed us to compare the connectivity profiles of the two CCZ hand representations. The correlation strength of each CMA location with a particular ROI was obtained by averaging the correlation values from its constituent motor representations with the same ROI. The CCZa and CCZp locations contained the anterior and posterior CCZ hand representations respectively. Based on these boxplots, it can be discerned that more frontal CMAs had stronger connectivity with prefrontal regions and weaker connectivity with premotor and motor areas, whereas the opposite was true for more posterior CMAs –reflecting a rostro-caudal coupling of the CMAs with the lateral frontal cortex. The statistical significance of these effects with connectivity z values was tested via a general linear model with CMA location (RCZa, RCZp, CCZa, and CCZp), ROI zone (prefrontal and motor cortex) and their interaction as factors. Note that in the second factor, the prefrontal and motor zones were, respectively, obtained by pooling the correlation values of the 5 prefrontal ROIs (a10, a46, a9/46, Broca, Fo) and 3 motor/premotor ROIs (FEF, M-H, and M-T).

We then characterized the rostro-caudal functional axis based on the correlation profiles of the CMAs with the lateral frontal cortex by estimating linear trends in the correlation strength for each CMA location (RCZa, RCZp, CCZa, and CCZp) with the 8 rostro-caudal lateral frontal ROIs. The 8 ROIs were first ranked along a rostro-caudal axis based on their averaged Y coordinate values across subjects and recoded into a numeric axis variable (ROIline): a10 (most anterior)-1, a46 - 2, a9/46 - 3, Broca - 4, Fo - 5, FEF - 6, M-T - 7, M-H (most posterior) - 8. We then performed linear regressions of the connectivity z values on CMA location and the linear axis variable (ROIline).

Additionally, we were interested in whether the connectivity profiles of the distinct motor representations (hand, eye, tongue) in the CMAs adhered to the same rostral-caudal functional organization. We focused on the two rostral CMAs (RCZa and RCZp) that both had a complete set of hand, tongue and eye motor representations. Given that the locations of the face motor representations (eye and tongue) in the RCZs varied depending on cingulate sulcal morphology (i.e., whether a PCGS was present), we also questioned if the connectivity profiles of the CMA motor representations were affected by the presence of a PCGS. To test the possible effects of the type of motor representation and cingulate morphology on the rostro-caudal connectivity organization in each RCZ, we linearly regressed the connectivity z values on motor representation type (hand, eye, and tongue), cingulate morphology (with or without PCGS) and the linear axis variable (ROIline). Here, the significance of the motor representation-ROIline and cingulate morphology-ROIline interaction terms would indicate, respectively, if motor representation and cingulate morphology significantly impacted the rostro-caudal connectivity organization in the RCZs.

Importantly, all linear model fits were evaluated using graphic diagnostic tools to ensure that all linear model assumptions were met.

The goal of the task-related fMRI study was to map the hand, tongue, and eye motor representations within the various CMAs as seeds for the subsequent resting-state connectivity analyses.

To map the hand representations in the various CMAs, we compared the BOLD signal during right hand movements with the BOLD signal during ocular fixation. Subject-level analysis was performed to assess the impact of the presence of a paracingulate sulcus on the location of this activity. Four hand-motor related activity peaks were observed: (1) at the intersection of the CGS and the vertical paracingulate sulcus (vpcgs) (anterior peak), (2) at the intersection of the CGS and the pre-paracentral sulcus (prepacs) (middle peak), (3) at the intersection of the CGS and the paracentral sulcus (PACS) and (4) posterior to this intersection. Thus, the locations of the two anterior peaks related to the hand fell within the RCZa and RCZp, respectively. The locations of the two posterior peaks related to the hand fell within the CCZ. Indeed, as previously shown (Amiez and Petrides, 2014), the CCZ displays 2 peaks related to hand movement. In the following resting-state data analysis, note that we dissociate the peak located at the intersection of the CGS and the PACS (CCZa) from the peak located posterior to this intersection (CCZp), in an effort to disentangle their relative functional connectivity pattern.

It was also evident that the presence of a PCGS did not have an impact on the location of these hand-related activity peaks in RCZa and RCZp (note that a PCGS is never observed at the level of CCZ). Note that the activity remained in the CGS even when a PCGS was present (with the exception of 2/21 subjects who showed hand motor-related activity in the PCGS in RCZa). These results were in line with our previous findings (Amiez and Petrides, 2014). Finally, we also found that the performance of right handed movements evoked activity in the CMA hand motor representations, bilaterally. Average MNI coordinates of hand movement representations in the cingulate cortex in left and right hemispheres with or without a PCGS are shown in Table 2 and Figure 2.

To map the tongue movement representations in the various CMAs, we contrasted the BOLD signal during tongue movements with the BOLD signal during ocular fixation. Tongue-motor related activity was observed: (1) at the intersection of the CGS and the vpcgs when no PCGS was present, and at the intersection of the PCGS and the vpcgs when a PCGS was present (anterior peak), (2) at the intersection of the CGS and prepacs when no PCGS was present and at the intersection of the PCGS and prepacs when a PCGS was present (middle peak). These observations suggested that the anterior and the middle peaks were located in the RCZa and RCZp, respectively. Note that both peak locations were clearly influenced by the presence of a PCGS: when a PCGS was present, tongue-motor related activity was found in the PCGS (with the exception of 2/21 subjects who had tongue motor activity in the CGS in RCZa). The absence of a third posterior peak corroborated our earlier observation that the CCZ does not contain tongue movements representation. These results are congruent with our previous observations (Amiez and Petrides, 2014). Finally, as with hand movements, tongue movements evoked bilateral activation in the CMA tongue regions. The average MNI coordinates of tongue movement representations in the cingulate cortex in left and right hemispheres with and without a PCGS are in Table 2 and Figure 2.

To map the saccadic eye movement representations in the various CMAs, we compared the BOLD signal during saccadic eye movements with that during ocular fixation. Saccadic eye movement-related activity was observed: (1) at the intersection of the CGS and the vPCGS when no PCGS was present and at the intersection of the PCGS and the vPCGS when it was present (anterior peak), (2) at the intersection of the CGS and prepacs when no PCGS was present and at the intersection of the PCGS and prepacs when it was present (middle peak). These observations suggest that the anterior and the middle peaks belong, respectively, to RCZa and RCZp. The location of the two peaks were impacted by the presence of PCGS: both saccadic movement-related activity peaks were located in the PCGS when it was present (with the exception of 1/21 subjects which showed saccadic eye movement activity in the CGS in RCZp). As with tongue movements, a third posterior saccadic movement related peak was absent confirming that there is no saccadic eye movement representation in the CCZ. The above results were in line with our previous observations (Amiez and Petrides, 2014). Finally, we also observed that saccadic eye movements induced bilateral activation in the CMAs. The average MNI coordinates of saccadic eye movement representations in the cingulate cortex in left and right hemispheres with and without a PCGS are shown in Table 2 and Figure 2.

Across all subjects, the 8 motor-related peaks could be observed in the cingulate cortex (3 peaks in RCZa, 3 in RCZp, and 2 in CCZ) in both hemispheres. Note that not all subjects showed all of the 8 peaks. On average, subjects display 6.1 peaks ± 1.2 stdev in the left hemisphere and 5.1 peaks ± 1.7 stdev in the right hemisphere, consistently with the results of Amiez and Petrides (2014).

We used the same contrasts described above to identify the primary hand motor region (hand vs. fixation), primary tongue motor region (tongue vs. fixation), and FEF (saccades vs. fixation). Subject by subject analysis confirmed, for each subject, the location of the hand motor region in the precentral knob in the central sulcus, the location of the tongue motor region in the posterior part of the ventral precentral gyrus, and the location of the FEF in the ventral branch of the superior precentral sulcus. The average x, y, z coordinates ± s.e.m. of precentral motor ROIs in each of the 21 subjects are presented in Table 3.

To demonstrate the pattern of intra-hemispheric connectivity of RCZa, RCZp, CCZa, and CCZp seeds with the lateral prefrontal and motor ROIs, we first performed a hierarchical clustering of the seeds and ROIs based on their inter-correlations across all subjects (see Method). Resulting dendrograms from the seed and ROI clustering are displayed in Figure 4A along with a heatmap reflecting the correlation strength between each pair of seed-ROI clusters. This analysis demonstrated the existence of three CMA clusters based on their functional connectivity with the prefrontal and the motor cortex: RCZa, RCZp, and CCZ.

The Boxplots in Figure 4B further depict the average Z values of correlations between seeds and ROIs across subjects and, therefore, how the activity of each CMA is differentially correlated with the activity of the prefrontal/motor ROIs. We tested these differences in connectivity z values with a generalized linear model with CMA location (RCZa, RCZp, CCZa, and CCZp) and ROI zones (prefrontal zones: a10, a46, a9/46, Broca, and Fo, and motor zones: FEF, Face –M-T– and Hand –M-H–) as fixed effect. The results indicated that the connectivity profiles with the prefrontal and motor zones differ significantly between CMA (ANOVA, ROIs × CMAs, df = 7, F = 21.296, p < 2 × 10⁻¹⁶): (1) the activity of RCZa is more correlated with that of the prefrontal zone but less correlated with that of the motor areas than the RCZp, CCZa, and CCZp, and (2) the activity of RCZp, CCZa, and CCZp are more correlated with that of the motor areas, but less correlated with that of prefrontal areas in comparison with RCZa. This analysis thus further suggests a topological organization of the correlation profiles of the CMAs with lateral frontal areas. We also assessed whether this correlation profiles can be a function of physical distance between seeds and ROIs. We calculated the Euclidian distances between the different seeds and the ROIs tested (Figure 4C). Results strongly suggest that the z-scores (displayed in Figure 4A) do not strictly varies as a function of distance as the shorter distances are not systemically associated with the higher z-scores and vice versa.

Nevertheless, the above analyses do not provide any estimation of the linearity of trends of correlation with lateral frontal areas along the rostro-caudal axis. To test and quantify these linear trends from anterior prefrontal to motor areas, we recoded the various lateral frontal ROIs into a numeric axis variable (ROIline) that corresponded to their relative posterior-to-anterior positions (see Method). Based on this coding, the lowest value (1) corresponds to a10 (the most anterior ROI) and the largest value (8) corresponds to M-H (the most posterior ROI). We then performed multiple linear regressions on the correlation values with CMA location and ROIline as predictors. The analysis revealed a significant CMA x ROIline interaction (Figure 4D, df = 3, F = 66.3, p < 2 × 10⁻¹⁶, ANOVA with CMAs and ROIline as factors), which indicates that the linear trends in connectivity within the rostral-to-caudal axis of lateral frontal cortex differed between the CMAs: a negative linear trend was observed for RCZa (more strongly correlated with rostral prefrontal areas), positive slopes were observed for the other CMAs (more strongly correlated with posterior motor areas). Thus, the connectivity profiles of CMAs with the lateral frontal cortex regions follow a rostro-caudal organization: the anterior CMA has stronger functional coupling with the prefrontal cortex and weaker with the motor cortex; the posterior CMAs have stronger coupling with the motor cortex and weaker with the prefrontal cortex.

We further investigated whether the same rostro-caudal cognitive-to-motor pattern of connectivity existed for each motor representation in the CMAs. Because only RCZa and RCZp contain both hand and face (tongue and saccadic eye) movement representations, this analysis was performed only with these two CMAs, independently. Seeds for resting-state data analyses were derived from the CMA motor activation peaks described above, and they are named “motor representations” in the description below. Overall, the rostro-caudal anatomo-functional organization described above is observed for all motor representations in both the RCZa and RCZp (Figure 5A) (Hand motor representation: df = 7, F = 3.6, p < 0.001; Tongue motor representation: df = 7, F = 17.436, p < 2.2 × 10⁻¹⁶; Eye motor representation: df = 7, F = 6.1, p < 1.6 × 10⁻⁶, ANOVA with CMAs type (i.e., RCZa and RCZp) and ROIs as factors). The slopes of seed-ROI connectivity strength against ROIline slightly varied between seed motor representations in RCZa (after stepwise model selection: ROIline × Seed interactions, df = 2, F = 10.2, p < 1 × 10⁻⁴, ANOVA). This is obviously related to the specific connections between each motor representations with the corresponding primary motor fields (see Figure 5A). Concerning RCZp, these slopes varied marginally (after stepwise model selection: ROIline x Seed interactions, df = 2, F = 3.02, p < 0.05, ANOVA) but they were all negative for RCZa (stronger connectivity with more rostral lateral regions) and all positive for RCZp (stronger connectivity with more caudal regions) (Figure 5B).

We then assessed whether cingulate morphology (i.e., the presence or absence of a PCGS, see Table 1) affected the connectivity profiles of RCZa and RCZp. Note that a PCGS is never observed at the level of CCZ. The results show that the rostro-caudal anatomo-functional organization is similar for all motor representations in both the RCZa and RCZp independently of the presence of a PCGS (Figure 6A) (In hemispheres with PCGS: interaction between CMAs identity and ROIs: df = 7, F = 13.77, p < 2 × 10⁻¹⁶ but no interaction between CMAs identity, ROIs, and motor representations: df = 14, F = 1.63, p < 0.07; In hemispheres with no PCGS: interaction between CMAs identity and ROIs: df = 7, F = 7.96, p < 4.75 × 10⁻⁹ but no interaction between CMAs identity, ROIs, and motor representations: df = 14, F = 0.62, p < 0.84, ANOVA). Figure 6B shows that when morphology (presence/absence of a PCGS) is taken into account, connectivity with lateral frontal areas always follows negative and positive slopes for RCZa and RCZp seeds, respectively (Figures 6C,D). In RCZa, these slopes were not influenced by the presence of a PCGS (interaction dropped by model selection) although the main effect of morphology was significant (df = 1, F = 15.6, p < 1 × 10⁻⁴, ANOVA) showing that overall correlation values (Z) between RCZa and lateral frontal regions were lower in the presence of a PCGS. In RCZp, the morphology did interact with the rostro-caudal organization (interaction after stepwise selection, RCZp, df = 2, F = 11.87, p < 1 × 10⁻³, ANOVA) as slopes of ROIline for Hand and Tongue representations were steeper than the gradient for Hand representation in presence of a PCGS. Yet, the rostro-caudal slopes of ROIline was positive in all cases in RCZp.

Taken together, these findings suggest that the type of motor representation and cingulate morphology had no qualitative impact on the rostral-caudal functional connectivity organization between the CMAs and the lateral frontal cortex.