Anatomical bases of fast parietal grasp control in humans: A diffusion-MRI tractography study

The dorso-posterior parietal cortex (DPPC) is a major node of the grasp/manipulation control network. It is assumed to act as an optimal forward estimator that continuously integrates efferent outflows and afferent inflows to modulate the ongoing motor command. In agreement with this view, a recent per-operative study, in humans, identified functional sites within DPPC that: (i) instantly disrupt hand movements when electrically stimulated; (ii) receive short-latency somatosensory afferences from intrinsic hand muscles. Based on these results, it was speculated that DPPC is part of a rapid grasp control loop that receives direct inputs from the hand-territory of the primary somatosensory cortex (S1) and sends direct projections to the hand-territory of the primary motor cortex (M1). However, evidence supporting this hypothesis is weak and partial. To date, projections from DPPC to M1 grasp zone have been identified in monkeys and have been postulated to exist in humans based on clinical and transcranial magnetic studies. This work uses diffusion-MRI tractography in two samples of right- (n = 50) and left-handed (n = 25) subjects randomly selected from the Human Connectome Project. It aims to determine whether direct connections exist between DPPC and the hand control sectors of the primary sensorimotor regions. The parietal region of interest, related to hand control (hereafter designated DPPChand), was defined permissively as the 95% confidence area of the parietal sites that were found to disrupt hand movements in the previously evoked per-operative study. In both hemispheres, irrespective of handedness, we found dense ipsilateral connections between a restricted part of DPPChand and focal sectors within the pre and postcentral gyrus. These sectors, corresponding to the hand territories of M1 and S1, targeted the same parietal zone (spatial overlap > 92%). As a sensitivity control, we searched for potential connections between the angular gyrus (AG) and the pre and postcentral regions. No robust pathways were found. Streamline densities identified using AG as the starting seed represented less than 5 % of the streamline densities identified from DPPChand. Together, these results support the existence of a direct sensory-parietal-motor loop suited for fast manual control and more generally, for any task requiring rapid integration of distal sensorimotor signals.


Introduction
The primate hand is an extraordinarily sophisticated and flexible sensorimotor system. It is able to grasp and manipulate various objects and tools with astonishing precision. For a large part, this skillfulness relies on the contribution of powerful control loops that continuously adjust the ongoing efferent command to compensate for biological ( Guigon et al., 2008 ;Harris and Wolpert, 1998 ), dynamic To date, there is still no consensus on the neural circuits involved in the control of these skilled hand actions. In monkey, the grasp/manipulation network is thought to be mediated by a dorsolateral circuit cascading from the hand proprioceptive sector of the parietal operculum (SII region), to the anterior intraparietal area (aIP) to the ventral premotor region (F5) and, finally the primary motor hand region (F1) Grafton, 2010 ;Jeannerod et al., 1995 ). Converging observations suggest that a similar network exists in humans ( Davare et al., 2011 ;Filimon, 2010 ;Turella and Lingnau, 2014 ). During the last decade Transcranial Magnetic Stimulation (TMS) studies have shown, for instance, that transient virtual lesions of the rostral part of the intraparietal sulcus impairs grasping behaviors ( Dafotakis et al., 2008 ;Davare et al., 2007 ), prevents on-line adjustments of hand shaping ( Rice et al., 2007 ;Rice et al., 2006 ;Tunik et al., 2005 ) and bias neural activity in the ventral premotor region ( Davare et al., 2010 ).
However, recent evidence also suggests the existence of a more direct control loop linking the primary sensory cortex (S1) to the dorsoposterior-parietal cortex (DPPC) to the primary motor cortex (M1). In monkeys, anatomical studies have identified direct projections from DPPC to M1 areas controlling hand/arm movements Kaas and Stepniewska, 2016 ;Rizzolatti et al., 1998 ). A recent study, in particular, reported connections between the grasp regions of DPPC (medial bank of the IPS and area 5) and M1 ( Gharbawie et al., 2011 ). In humans, it was reported that a conditioning-TMS pulse delivered over DPPC could significantly depress motor evoked potentials triggered, in intrinsic hand muscles, by a test pulse delivered over the ipsilateral M1 ( Koch et al., 2007 ;Mackenzie et al., 2016 ). Likewise, a recent per-operative study recorded short-latency inhibitory effects of superior parietal stimulation on corticospinal excitability of distal upper limb muscles ( Cattaneo et al., 2020 ). Moreover, it has been shown that focal lesions within the superior parietal lobule (SPL) can cause unwanted hand movements through a release of inhibition within M1 ( Assal et al., 2007 ). Regarding sensory inputs, several studies have reported that upper-limb somatosensory evoked potentials recorded in DPPC are abolished following S1 excision ( Allison et al., 1991a ;Allison et al., 1991b ). Consistent with all these observations, we recently identified neural sites, in DPPC, that display the expected properties of a neural on-line sensorimotor controller ( Desmurget et al., 2018 ). First, these sites instantly disrupt ongoing open/close hand movements when electrically stimulated. Second, they receive short-latency somatosensory inputs from intrinsic hand muscles.
To sum up, the results above point to the possible existence of a rapid feedback loop involved in grasp/manipulation control, mediated by neural activity in DPPC, receiving sensory signals from the hand area of S1 and projecting to the hand area of M1 ( Fig. 1 ). This hypothesis was evaluated using diffusion Magnetic Resonance Imaging (dMRI) tractography in two successive samples of right-and left-handed subjects randomly selected from the Human Connectome Project (HCP) database. The dorso-posterior parietal hand control region (DPPC hand ) was used as the seed region of interest (ROI) for the tractography. This region was defined as the 95% confidence area of the parietal sites that disrupted hand movements in our previous per-operative study ( Desmurget et al., 2018 ) ( Fig. 1 ). To avoid ambiguity, it may be worth mentioning that this area is likely to be oversized, i.e. to be substantially larger than the parietal region truly involved in hand control. Indeed, DPPC hand was computed from a limited number of per-operative points ( n = 12) collected in brain tumor patients ( Desmurget et al., 2018 ) known to exhibit a high level of cortical reorganizations ( Desmurget et al., 2007 ). As a consequence, we expect the posterior parietal area connected with M1 and S1 hand regions, if any, to be substantially smaller than our seed parietal region.

Material and methods
The HCP is a unique dataset aiming to be used by the scientific community to shed light on the anatomical and functional connectivity of Putative anatomy of the parietal inhibitory network. S1: primary somatosensory cortex. M1: primary motor cortex. DPPC: dorso-posterior parietal region. Confidence ellipsoid of DPPC hand was computed from the parietal sites where electrical stimulation was found to trigger a selective disruption of hand movements in a previous per-operative study ( Desmurget et al., 2018 ). The yellow border displays the 95% confidence border of this ellipsoid. the human brain ( Glasser et al., 2013 ;Van Essen et al., 2013 ). Tractography analyses were carried out on dMRI multi-shell data of 50 righthanded subjects (22 to 35 years old; 31 females) randomly selected from the HCP database. To evaluate the generality of the results obtained from this main sample and to evaluate the existence of potential hemispheric asymmetries related to handedness, we also considered an additional population of 25 left-handed subjects (randomly selected from the HCP database; 25 to 33 years old; 16 females). The 75 multi-shell data-sets (18 images with b = 0 s.mm-2 and 270 diffusion-weighted images with b = 1000, 2000 and 3000 s.mm-2 applied in 90 non-collinear diffusion directions) were acquired with an isotropic spatial resolution of 1.25 mm and corrected for distortion artifacts. . Tissue segmentations as well as cortical pial and white matter surfaces computed for each subject using FreeSurfer ( Fischl, 2012 ) were taken from the structural HCP database.
Within this dataset, precentral (PreC) and postcentral (PostC) ROIs were defined from Destrieux parcellation ( Destrieux et al., 2010 ). The DPPC hand ROI was defined on each individual pial surface, in MNI coordinates, as the 95% confidence area of all the parietal sites that disrupted hand movements when electrically stimulated in our per-operative study ( Desmurget et al., 2018 ). Tractography analyses were performed from the parietal portion of the ellipse (i.e. after the small anterior fraction of the ellipse involving the postcentral gyrus was truncated) ( Figs. 1 and S1).
For dMRI tractography, we used the state-of-the-art MRtrix3 software ( Tournier et al., 2019 ). First, for all tissue types (white matter (WM), grey matter (GM) and cerebrospinal fluid) we computed the response function estimations using an automated unsupervised algorithm ( Dhollander et al., 2016 ). Second, we computed the multishell, multi-tissue constrained spherical deconvolution ( Jeurissen et al., 2014 ). A whole-brain probabilistic tractogram ( Tournier et al., 2010 ) was then computed with the Anatomically-Constrained Tractography ( Smith et al., 2012 ) and GM / WM interface seeding. We used the ensemble tractography approach ( Takemura et al., 2016 ) with variations of the step size [0.3, 0.6 and 1.25 mm], of the angle [30°, 45°and 60°] and of the cutoff [0.05, 0.1, 0.15] to achieve a 54 million of streamlines tractogram. The final whole brain tractogram of each subject was automatically filtered to identify the left and right ipsilateral tracts connecting DPPC hand and PostC; and DPPC hand and PreC. This filtering was performed using purely anatomical criteria. Streamlines passing through both ipsilateral DPPC hand and PostC ROIs form the DPPC hand /PostC bundle, while the DPPC hand /PreC bundle is defined by all streamlines crossing both ipsilateral DPPC hand and the PreC ROIs. Supplementary Table  1 summarizes the average number of streamlines obtained for each of the DPPC hand /PostC and DPPC hand /PreC bundles.
The end-point density surfaces of DPPC hand /PostC and DPPC hand /PreC bundles were computed for each subject as the amount of streamlines endpoints projected onto each vortex of white matter surface. Then, the obtained endpoints density surfaces were averaged for all subjects, normalized over the range of 0 to 1 and displayed on the mean pial surface after a heat kernel smoothing ( Chung et al., 2005 ) with a Full Width at Half Maximum (FWHM) of 4.9. The overlap between tracts from pre and postcentral regions within the DPPC hand pial surface was computed using the Jaccard index (intersection over union of both surfaces of interest).
3D-streamlines density maps of DPPC hand /PostC and DPPC hand /PreC tracts (representing the amount of streamlines traversing each voxel of the brain) were also computed for each subject using track-density imaging ( Calamante et al., 2011 ). Then, the 3D-streamline density maps of DPPC hand /PostC tracks and DPPC hand /PreC tracts were averaged for all the subjects, normalized over the range of 0 to 1 and overlaid on the T1 study template using the buildtemplateparallel procedure of the ANTS toolkit ( Avants et al., 2011 ). These tracks were then compared with the three subdivisions of the superior longitudinal fasciculi (SLF) bundles included in the white matter atlas of the FSL XTRACT toolbox ( https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/XTRACT ) ( Warrington et al., 2020 ), thresholded at 10 % of the maximal probability.
As a methodological sensitivity control of the tractography approach used in this study, we replicated all the analyses above to investigate the Angular Gyrus AG/PostC and AG/PreC connectivity. The AG ROI used for tractography was defined as the inferior part of the angular gyrus from Destrieux's parcellation ( Destrieux et al., 2010 ) situated outside the DPPC hand (supplementary Figure S1). AG was selected in light of converging evidence suggesting that this area has only sparse direct connections with the primary sensorimotor areas ( Koch et al., 2010 ;Schulz et al., 2015 ;Seghier, 2013 ).
All ROIs (DPPC hand , AG, PostC and PreC) are available on request to the corresponding author.

Main analyses: right-handed subjects
As illustrated in Fig. 2 , which displays the mean normalized density maps of streamline endpoints projected to the average pial surface, streamlines from DPPC hand converge within circumscribed areas of the pre-and postcentral gyri (see supplementary Table 2 for MNI coordinates of the centers of gravity of these projections; see also supplementary Fig. S2 for more views of these endpoint surface densities). These areas are in the sector of the sensorimotor strip where hand/finger movements are typically represented ( Desmurget et al., 2014 ;Guzzetta et al., 2007 ;Kuhtz-Buschbeck et al., 2008 ;Penfield and Boldrey, 1937 ;Szameitat et al., 2012 ). More precisely, they are close the middle knee of the central lobe ( Ribas, 2010 ), an anatomical marker that is typically assumed to define the functional hot-spot for distal hand control, both in the primary motor ( Vigano et al., 2019 ;Yousry et al., 1997 ) and the primary sensorimotor ( Sastre-Janer et al., 1998 ;White et al., 1997 ) cortex. These pre and postcentral regions of maximum density are located, respectively, within M1 (Brodmann area 4) and S1 (mainly Brodmann area 2, with some extension into area 1; see Fig. 2 and supplementary Fig. S2). Supplementary Fig. S3 further illustrates these results by showing streamline bundles (DPPC hand /PreC, DPPC hand /PostC) and density of streamline endpoints for three randomly selected subjects.
Consistent with the predictions of our initial hypothesis ( Fig. 1 ) and the results of our previous per-operative study ( Desmurget et al., 2018 ), streamlines identified in PreC and PostC originated in the same subregion of our initial parietal seed ROI ( Fig. 3 and supplementary Fig. S4). Within this subregion (hereafter designated sDPPC hand ), spatial overlap between pre and postcentral streamlines was extensive. It reached 92.2% and 92.4% for the left and right hemispheres respectively. Also, in both hemispheres, the center of gravity of sDPPC hand was located in the anterior part of the intraparietal sulcus (aIP; supplementary Table 2). As a control, we compared the sum of streamlines density for PostC/PreC and sDPPC hand . We found no statistical difference, in agreement with the conclusion that the fibre projections found in sDPPC hand reflect the sum of the projections identified in PostC and PreC (sign test; Left hemisphere p > .65, Right hemisphere p > .10).
Overall, these results are also consistent with the claim that our seed parietal area was oversized (see introduction). Indeed, sDPPC hand appears substantially smaller than DPPC hand (Left: 55% of the seed surface; Right: 64%). Its centre of gravity is also slightly shifted in the dorsoanterior direction, along IPS (Left: 17 mm; Right: 16 mm).
Finally, it may be worth noting that no sign of lateralization was visible in our data. The total number of streamlines was statistically comparable in the left and right hemispheres for both bundles (sDPPC hand /PostC and sDPPC hand /PreC, sign test, ps > .10; see supplementary Table 3 for details).
When the analyses above were replicated with AG as the seed ROI instead of DPPC hand , very few streamlines were identified (as expected from previous studies; see methods). Streamline densities measured from AG to PostC and from AG to PreC represented less than 5% of the streamline densities identified from DPPC hand . This difference was unlikely to be related to differences in the size of the two ROIs. Indeed, as shown by additional analyses, AG and DPPC hand have comparable surfaces (47 cm 2 and 49 cm 2 respectively).

Supplementary analyses: left-handed subjects
Analyses involving the second sample of left-handed subjects fully replicated the conclusions above. As shown in Fig. 4 , we found dense ipsilateral connections between a restricted area in the anterior part of DPPC hand and focal sectors within PreC (M1; Brodmann area 4) and PostC (S1; Brodmann area 2 with some extension into area 1) hand territories (see supplementary Table 4 for MNI coordinates of the centers of gravity of these projections; see also supplementary Figure S5). This parietal subregion (sDPPC hand ) was substantially smaller than our initial seed region (Left: 60% of the seed surface; Right: 62%). Within sDPPC hand , we found a strong overlap between streamlines originating from pre and post central structures (Left hemisphere: 92.5%; Right hemisphere: 95.2%; Fig. 5 and supplementary Figure S6). In addition, there was no statistical difference between the sum of streamlines density found in PostC/PreC and sDPPC hand (sign test; Left hemisphere p > .10, Right hemisphere p > .40).
In agreement with the results obtained in right-handed subjects, tractography revealed no significant sign of lateralization in this sample of left-handed individuals. The total number of streamlines was statistically comparable in the left and right hemispheres for both bundles (sign test, ps > .20; see supplementary Table 3 for details). We conducted additional analyses to identify potential asymmetries between right-and left-handed subjects. Again, no significant difference was found. The total number of streamlines of the two bundles (sDPPC hand /PostC and sDPPC hand /PreC) was statistically comparable in each hemisphere between the two samples (Kolmogorov-Smirnov two-sample test, ps > .10; see supplementary Table 5 for details).

Discussion
To summarize, our results identify an anatomical network linking the cortical territories devoted to hand control in the primary motor,  the primary somatosensory and the dorso-posterior parietal areas. This network provides the required architecture for the existence of a functional, short-latency, sensory-parietal-motor loop dedicated to the rapid control of hand movements.
Of interest with respect to this conclusion is the degree to which grasping/manipulation behaviors differ from reaching activities. In humans, aIP is clearly linked to fine distal control. Convergent evidence have shown that permanent (anatomical) and transient (TMS-induced) lesions within this region disrupt the grasp component of prehension movements, without affecting the transport phase ( Binkofski et al., 1998 ;Rice et al., 2007 ;Rice et al., 2006 ;Tunik et al., 2005 ). The ongoing control of this transport phase has been shown to rely on the medial intraparietal sulcus (mIP) ( Chib et al., 2009 ;Desmurget et al., 1999 ) and/or the SPL ( Desmurget et al., 2001 ;Diedrichsen et al., 2005 ;Wolpert et al., 1998 ). Several modeling experiments have also established the computational viability of dissociating the neural representations of distal (grasp-manipulation) and proximal (limb transport) move-ment components for object-oriented actions ( Hoff and Arbib, 1993 ;Ulloa and Bullock, 2003 ).
However, in contrast with this dissociative view, many neuroimaging studies have reported reach related activity in the human aIP, irrespective of any grasping behavior (for a review, Filimon, 2010 ). In addition, it was found that the so-called hand-knob was not as selective as initially described by Penfield ( Penfield and Boldrey, 1937 ;Penfield and Rasmussen, 1950 ). Within this area, distal (wrist/finger) and proximal (arm/forearm) representations overlap generously ( Branco et al., 2003 ;Desmurget et al., 2014 ;Desmurget and Sirigu, 2015 ). Moreover, in the present study, although parietal streamlines were concentrated around the hand-knob in M1 and S1, diffuse projections were identified medially and laterally with respect to this knob (yellow gradients in Figures 2 ,  4 , S2 and S5). Based on these observations, it cannot be determined whether the functional network identified in the present study is selectively associated with fine distal motor control or whether it is involved  in the regulation of more general proximo-distal reach-to-grasp activities.
In connection with this issue, it may be worth mentioning that the parietal region here defined as sDPPC hand is a complex structure ( Andersen et al., 2014 ;Freedman and Ibos, 2018 ;Janssen and Scherberger, 2015 ). Certainly, as emphasized in the introduction, it plays a central role in motor planning and control (for a recent review, Sirigu and Desmurget, 2020 ). However, it is also a key node of many other functional networks involved in action selection ( Lindner, 2018 ), spatial attention ( Corbetta and Shulman, 2002 ), coordinate transformations ( Andersen et al., 1997 ), multisensory integration ( Huang et al., 2012 ), crossmodal evaluations ( Grefkes et al., 2002 ), etc. As a consequence, the role of the direct anatomical connections here identified between sDPPC hand and S1 and sDPPC hand and M1 is unlikely to be restricted to on-line motor control. For instance, this pathway could be involved in haptic object processing. Indeed, convergent data suggest that sDPPC hand contribute to this function through combining somatosensory and motor information ( Burton et al., 2008 ;James et al., 2007 ;Sathian, 2016 ).
Another key point to be discussed concerns the novelty of our results. The sensory-parietal and motor-parietal tracts identified in this study are part of the superior longitudinal system (SLS) recently described by Mandonnet and colleagues ( Mandonnet et al., 2018 ). This system comprises the three branches of the SLF (SLF-I to III) ( Catani and Thiebaut de Schotten, 2012 ;Parlatini et al., 2017 ;Schmahmann and Pandya, 2006 ). Based on this observation, we performed specific analyses to determine more precisely whether the pathways isolated in the present study are part of a specific branch of the SLF. To this end, for the right-hand sample, we superimposed our sDPPC hand /PostC and sDPPC hand /PreC bundles with the three subdivisions of the SLF tract, using the white matter atlas available through the XTRACT toolbox ( Warrington et al., 2020 ;see methods). As illustrated in Fig. 6 , the longitudinal portion of the  ( Warrington et al., 2020 ) overlaid on a sagittal view of the mean T1 template computed from all the subjects. Bottom row: Mean normalized streamline density maps of DPPC hand (DPPC h ; dark blue) to PreC (M1 area, green) and PostC (S1 area, green) tracts superimposed on SLF II (left panel) and SLF III (right panel) branches. Most of the overlap occurs with SLF-II (75%). Figure computed from the right-handed sample. See text for details. sDPPC hand /PostC and sDPPC hand /PreC bundles closely follow the SLF tract. 75% of the overlap occurs with SLF-II. This is much more than the percentages found for SLF-III (22 %) and SLF-I (3%). Anatomically, SLF-II originates from the posterior-inferior parietal region and predominantly targets the dorsal premotor cortex and dorsolateral prefrontal cortex ( Catani and Thiebaut de Schotten, 2012 ;Parlatini et al., 2017 ;Schmahmann and Pandya, 2006 ). This tract has been associated with various functions, including motor control, spatial attention and some aspects of language production ( Hecht et al., 2015 ;Parlatini et al., 2017 ;Thiebaut de Schotten et al., 2014 ;Wang et al., 2016 ). Our data extend these observations by suggesting that the connections here identified between specific cortical territories devoted to hand control in M1, S1 and DPPC, run through the SLF-II bundle. Of course, further studies are now required to confirm this hypothesis.
In the present study, based on tractograms, we failed to identify asymmetries: (i) between the right and left hemispheres for a given hand preference (right or left); and (ii) between right-and left-handed participants for a given hemisphere (right or left). This result is consistent with TMS experiments showing a strict contralateral control for grasping movements ( Rice et al., 2007 ). It is also supported, at group level, by previous studies showing no relation between handedness and hemispheric asymmetry, for SLF-II ( Budisavljevic et al., 2017 ;Hecht et al., 2015 ;Howells et al., 2018 ;Makris et al., 2005 ). However, contradictory observations have been reported ( Wang et al., 2016 ) and one cannot exclude that our sample size was not large enough to allow identification of discrete interhemispheric asymmetries. As a consequence, this negative result should be considered with caution until further studies are conducted.
Finally, an important methodological issue needs to be raised. It is known that tractography algorithms tend to produce a substantial amount of false-positive bundles ( Maier-Hein et al., 2017 ). However, the hypothesis-driven model used in this research dramatically minimizes this risk of erroneous identification. As shown in several studies, while tractography might produce disputable conclusions when used in isolation ( David et al., 2019 ), it represents a powerful cross-validation tool when used predictively, in conjunction with other techniques such as functional MRI ( Guye et al., 2003 ), histology ( Dell'Acqua et al., 2013 ), polarized light imaging ( Mollink et al., 2017 ), tract-tracing , deep brain stimulation ( Calabrese, 2016 ) or direct electrical stimulation ( Kamada et al., 2009 ). In relation to this point, one may mention our inability to identify robust pathways using AG as the seed ROI. Together with previous studies showing sparse direct connections between AG and primary sensorimotor areas ( Koch et al., 2010 ;Schulz et al., 2015 ;Seghier, 2013 ), this result pleads against the conclusion that the sensory-parietal-motor tracks identified in this study reflect false positive reconstructions of the tractography procedure. Whether the few streamlines identified in this study between AG and primary sensorimotor regions reflect false positive or real sparse connections ( Koch et al., 2010 ;Schulz et al., 2015 ;Seghier, 2013 ) cannot be determined from this study.
To summarize, this study identifies a direct anatomical loop connecting the cortical territories harboring hand representations in the primary motor, primary somatosensory and dorso-posterior parietal cortices. This loop is suited for fast manual control and more generally, any task requiring rapid integration of distal sensorimotor signals. Whether the circuit we identified should be considered a new branch of the SLF-II needs to be confirmed.

Declaration of Competing Interest
The authors declare no conflict of interest.

Data availability
All MRI data used in this study are available in the HCP open access dataset. All the software used are open source code. The data that support the findings of this study are available from the corresponding author, BH, upon reasonable request.

Supplementary materials
Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.neuroimage.2020.117520 .