Abstract
The human brain coordinates a wide variety of motor activities. On a large scale, the cortical motor system is topographically organized such that neighboring body parts are represented by neighboring brain areas. This homunculus-like somatotopic organization along the central sulcus has been observed using neuroimaging for large body parts such as the face, hands and feet. However, on a finer scale, invasive electrical stimulation studies show deviations from this somatotopic organization that suggest an organizing principle based on motor actions rather than body part moved. It has not been clear how the action-map organization principle of the motor cortex in the mesoscopic (sub-millimeter) regime integrates into a body map organization principle on a macroscopic scale (cm). Here we developed and applied advanced mesoscopic (sub-millimeter) fMRI and analysis methodology to non-invasively investigate the functional organization topography across columnar and laminar structures in humans. We find that individual fingers have multiple mirrored representations in the primary motor cortex depending on the movements they are involved in. We find that individual digits have cortical representations up to 3 mm apart from each other arranged in a column-like fashion. These representations are differentially engaged depending on whether the digits’ muscles are used for different motor actions such as flexion movements like grasping a ball or retraction movements like releasing a ball. This research provides a starting point for noninvasive investigation of mesoscale topography across layers and columns of the human cortex and bridges the gap between invasive electrophysiological investigations and large coverage non-invasive neuroimaging.
- high-resolution fMRI
- motor cortex
- cortical layers
- cortical columns
- VASO
Introduction
The human repertoire of motor activity spans an immense variety of movements, and most of our interactions with the environment involve some degree of movement. Voluntary movement is controlled by the central nervous system, particularly the brain’s motor cortex. The primary motor cortex is organized according to the principle of somatotopy, whereby areas controlling different body parts are arranged in a predictable order. Penfield and colleagues (1937) were the first to report the medial-to-lateral leg-to-face somatotopic “homunculus” in the human primary motor cortex; a representation that has been confirmed with neuroimaging for large body parts (Schieber 2002; Chainay 2004) and digits (Siero 2014; Olman 2012; Schellekens 2018) in the spatial regime of centimeters. However, fundamental deviations from this simple linear arrangement of body parts have been reported at a more microscopic level (Barinaga 1995; Strother 2012; Schieber 1993; Penfield 1937; Strick and Preston 1982; Sanes 1995; Meier and Aflalo 2008; Idovina and Sanes 2001; Hlustik 2001; Woolsay 1979; Lemon 1988). First, individual neurons in the primary motor cortex are not tuned to individual body parts but show a gradual overlap of multiple body parts (Schieber 1993, 2002; Indovina 2001). Second, the progression of body parts from face representation towards leg representation does not follow a simple steady linear order. In monkeys, multiple unconnected body part representations were reported (Kwan 1978; Park 2001). As such, it was suggested that the hand is emphasized in a core region, and the wrist, arm, and shoulder are emphasized in a half ring surrounding the core (Strohter 2012; Strick and Preston 1982; Porter and Lemon 1993; Meier and Alflalo, 2008).
In light of these deviations from a linear somatotopic organization, alternative organizational principles for the motor cortex have been investigated. Previous findings show that when a small part in the monkey motor cortex is electrically stimulated, the resulting movement combines joints and muscles in a manner resembling a coordinated action (Graziano et al. 2002), suggesting that the motor cortex organization might follow an action map representation rather than a body part map (Graziano 2016; Sanes 1995). More recently, Dietrichsen et al. also found fMRI evidence, which confirms that large portions of the motor cortex represent complex muscle synergies in humans too (Ejaz 2015). These findings are in agreement with the functioning of so called corticomotoneuronal cells that have connections to multiple muscle groups to facilitate represent complex movement actions (Omrani 2017). These corticomotoneuronal cells are, however, confined to the evolutionary younger subdivision of M1, the Brodmann area BA4p which is the inferior portion of M1 on the pre-central bank of the central sulcus (Rathelot and Strick 2006, 2009; Lemon 2008). For the evolutionary older part of M1, BA4a, it is not clear, how the action-map representation on the microscale is integrated into the body map representation on the macroscopic scale. Due to recent methodological advances in fMRI (Polimeni 2010; De Martino 2015; Huber 2017) the mesoscopic regime of sub-millimeter resolutions becomes accessible for non-invasive functional neuroimaging across columnar and laminar directions in humans. Thus, mesoscopic fMRI could provide insights on how (microscopic) action maps are integrated into (macroscopic) somatotopical body part maps. However these current mesoscopic fMR-methods can only capture small patches of cortex with thick slices and without the ability to resolve columnar structures and laminar structures simultaneously. Thus that they do not allow the investigation of microscale topographic movement representations in humans.
This study develops advanced high-resolution imaging methods and novel analysis methodology to investigate topographical organization patterns across the ‘columnar’ and ‘laminar1’ cortical dimensions with CBV-senitive fMRI across large patches of cortex. The method advancements could be achieved with conventional 7T MRI hardware and a non-invasive VASO (vascular space occupancy) sequence (Lu 2003; Hua 2013; Huber 2015). We investigated the neuroscientific applicability of the new methodology by mapping the meso-scale (sub-millimeter) topographical representations of the primary motor (M1) and primary sensory cortex (S1). We use this approach to investigate the organizational principle of individual finger representations for a set of simple motor actions including tapping, grasping (flexion) and retraction movements (extension). We find mesoscopic topographic finger representations in M1 (BA4a) on the order of 1.2±0.4 mm. We also find evidence of a simple topographical organizational principle: Fingers are represented in multiple somatotopically organized patches that are preferentially activated for either grasping or retraction movement actions.
Results
To investigate the columnar organization of the primary motor cortex with fMRI, five different experiments (Fig. S1) were conducted. These were:
Individual tapping of two fingers (index and little).
Individual tapping of four fingers (index, middle, ring, and little; no thumb) interspaced with rest periods.
Individual tapping of all five fingers (index, middle, ring, little, and thumb) interspaced with rest periods.
Alternating grasping or retraction of a rubber ball interspaced with rest periods. These tasks engage muscle extension and muscle flexion of every finger.
Resting-state.
Motor tasks were performed with the left hand, while imaging the primary sensorimotor cortex on both sides of the central sulcus in the right hemisphere (Fig. 1A). The right hand was not engaged during any of experiments of this study. For more background on these tasks and explanations, why this task setup was used, see the supplementary information (Fig. S1). Five participants underwent 42 fMRI total sessions of 2 hours each (84h of scanning, Tab. S1). We simultaneously measured changes in the cerebral blood volume (CBV) and blood oxygenation level dependent (BOLD) response using the SS-SI-VASO method (Huber 2014; Lu 2003) with a 3D-EPI readout (Poser 2010) at 7T. The nominal resolution was 0.79 mm in the columnar dimension and 0.99 mm thick slices perpendicular to the precentral bank of the central sulcus – also known as the hand knob (Fig. 1A). Estimates of cortical depths (layers) and cortical distances (columns) across the entire central sulcus were calculated by means of simultaneously acquired anatomical and functional image contrasts in EPI space (Fig. 1B). Details on the data acquisition and analysis approach are provided in the Supplemental Experimental Procedures (Fig. S1-S10). Brain activity changes are estimated as the percent signal change of the fMRI signal during the task, compared to rest. Relative digit dominances are estimated as the activity change for a given finger compared to the activity change for all other fingers.
Functional MRI signal changes in S1 show a linear arrangement of individual digits (Fig. 1) as previously described in high-resolution fMRI (Olman 2012; Panchuelo 2016; Kolasinsky 2016; Schluppeck 2017; Siero 2014; Ejaz 2015). The distance between the representations of the thumb and the little finger in S1 is 16 ± 4 mm. In M1 we find clear deviations from a continuous linear alignment of the digit representations; we find multiple representations of every digit in a mirrored pattern (Fig. 2). These digit representations are significantly smaller than the representations in the sensory cortex. The distance between the representations of the thumb and the little finger in M1 is 6 ± 2 mm. The mirrored pattern of multiple distant digit representations was highly consistent across participants (Fig. 2), across fMRI contrasts (Fig. S5), and across days (Fig. S6).
We find that different patches of the hand knob in M1 are preferentially activated during grasping and retraction movements, respectively. These patches are 6 ± 2 mm in size and columnar distance. We find that the outlines of grasping and retraction preference patches contain complete sets of all digit preferences (copper and turquoise outlines in Fig. 2B).
We next quantified relative preferences for body parts (i.e., different fingers) and movements (i.e., grasping versus retraction) along the medial-lateral axis in both the primary motor and primary sensory cortex. Columnar profiles along the central sulcus are shown in Fig. 3, for one representative participant. In the primary motor cortex, we find a sharp switch in preference between grasping and retraction at a certain point along this axis (Fig. 3A left, echoing results shown in the last row of Fig. 2). In the primary sensory cortex, we find a strong preference to grasping tasks only. There are no clear areas that have a stronger S1 response to retraction versus grasping (Fig. 3A, right). This is likely because common grasping actions are more often associated with stimulations of the sensory receptors (exteroception on the fingertips and inside of the hand) compared to retraction actions. Representations of sensory proprioception in the primary sensory cortex might be similarly engaged for grasping and retraction movements and, thus, do not introduce a preference to either of the two motor actions in the sensory cortex. The columnar profiles in Fig. 3B also clearly demonstrate the multiple mirrored finger representations in M1, versus the single linear representation in S1. Finally, we quantified relative finger preferences across both columnar and layer dimensions (Fig. 3C). We find that there is a more gradual transition between preferred fingers (i.e., more overlap across finger representations) in superficial layers (solid lines) than in deeper layers (dashed lines), where the transition is sharper. This result is also confirmed in resting-state data (Fig. S8).
Besides task-based topographical descriptions, we also characterized the topography of finger representations in restingstate functional connectivity with the sensory cortex. The aim of these investigations was to confirm the multiple digit representations in the motor cortex on data that does not rely on a specific task design. As expected from the task-based data, we find the same multi-stripe patterns in the primary motor cortex for A) task-induced motor activity, B) resting-state seed-based correlation, and C) selected independent components of FSL-melodic ICA (Fig. 4). In all three cases, the two investigated fingers (index and little) are represented as two clear patches in the sensory cortex (red and blue arrows in bottom left of Fig. 4A-C). In the motor cortex, however, multiple red and blue patches can be identified. The resulting stripe-pattern of blue-red-blue is very similar for task and resting-state results (red and blue arrows in top right of Fig. 4A-C). We find that the digit representations in the sensory cortex have a stronger connectivity with the grasping patches in the motor cortex compared to its retraction patches (Fig. 4E-D). This is consistent with the task results showing that the sensory cortex has stronger activity for grasping motor actions compared to retraction motor actions.
Discussion
The data presented here suggest a mesoscopic topographical organization principle of individual fingers in the primary motor cortex (BA4a) at an unprecedented spatial resolution. The primary motor cortex has been extensively studied with cytoarchitecture, functional representations using invasive electrophysiology, human non-invasive electrophysiology and neuroimaging. No comprehensive view, however, has emerged regarding how the representations of actions, as found with electrical microstimulation (Strick and Preston 1928, Graziano 2002, 2016) translate into topographical somatotopic organization principles, as detected for larger body parts. Here, using high-resolution CBV-fMRI neuroimaging in humans and topographical analysis tools, we reveal how these different organizational principles are integrated across cortical columns and layers.
We find evidence for action specific representations of fingers in M1, as well as for finger specific representations. Our results reveal how these multiple organizational principles are elegantly combined in M1: We find multiple mirrored representations of individual fingers that are differently engaged for specific movement actions (Fig. 1, 2, 3). Thus, M1 is organized as an action map as well as a topographical finger map in small (0.9-1.2 mm) representations along the columnar dimension within 4-9 mm action patches.
The multiple representations of individual fingers can be seen in both task-based fMRI and resting-state functional connectivity with S1 (Fig. 3). We find that the finger representations in the motor cortex are very similar across cortical depths. There are only small differences between superficial cortico-cortical input layers II/III and corticospinal output layers Vb/VI. Namely, the size of the finger representations is slightly smaller in deeper layers compared to superficial layers. One potential origin of the sharper body part representation in deeper layers might be associated with previously described phenomena of surround inhibition in M1. This kind of surround inhibition in M1 has been proposed to serve the function of improved discriminiability between motor representations. The phenomenon has been described as the reduction of corticospinal excitability in muscles that are non-active but adjacent to the active muscles (Beck and Hallett, 2011). During voluntary finger flexion, the person’s muscle shows increased corticospinal excitability, whereas the surrounding neighboring muscle shows decreased corticospinal excitability. However, due to the lack of invasive layer-dependent electrophysiology measurements in the literature about M1, it is unclear whether this is expected to be different across cortical layers, as it is in sensory brain areas (Hubel and Wiesel, 1968, 1972). In resting-state fMRI, we find that the functional connectivity strength varies, as it also does with task-based fMRI, with more tightly constrained connectivity in deeper layers compared to superficial layers (see Fig. S8).
As opposed to previous columnar fMRI studies, we do not only try to depict known structures with known shape and size as proof-of-principle for a method as previous studies. Instead here, we are finding previously unknown organization principles of sub-millimeter representations in M1. This is a fundamentally new approach and a paradigm shift for the field of columnar and laminar fMRI.
It is important to note that the motor cortex representations of grasping vs. retraction are not completely orthogonal to each other: that is, the entire hand knob shows increased activity to both grasping as well as extension movements, compared to rest. The respective grasping and retraction patches show differential sensitivity to either motor action (Fig. S6C) only. Other potential movements could theoretically engage the two patches even more deferentially: for example, spreading fingers vs. clenching fist; maintaining force without movement (i.e., isometric muscle contraction) vs. movement with minimal force, and others. These potential movement actions are only partly segregated for the particular flexion vs. extension task used here. For the finger tapping tasks used here, the individual finger movements underwent both extension and flexion movements. Thus, the double-hand map can relate to grasping and retraction task that engages flexion and extension differently.
In the primary somatosensory cortex, we find no deviations from the homunculus model as shown previously in humans (Schluppeck 2017; Olman 2012; Kolasinski 2016; Shellekens 2018).
Previous digit mapping studies using GE-BOLD fMRI could not clearly identify the mirrored finger representation in the primary motor cortex. This might be due to the fact that GE-BOLD fMRI suffers from poor localization specificity due to the presence of large draining veins (Turner 2002; Menon 2002; Kim and Ogawa, 2012; Kennerley 2015). Most human fMRI studies that investigated somatotopic organization likely did not have the sub-millimeter specificity to investigate deviations from the linear representation without methodological challenges (Olman 2012; Siero 2014; Dechent and Frahm 2003, Hlustik 2001; Shellekens 2018). We believe the columnar specificity of GE-BOLD could be limited by two potential sources of venous contaminations. First, large principal veins are described to have tangential branch lengths of 2-6 mm (Fig. S8E,; Duvernoy 1981). Thus, these tangential veins can result in more tangential signal mixing of GE-BOLD signal compared to their respective neural representations. Second, in some participants the sensory and motor banks of the central sulcus can be drained by the same pial veins. BOLD signal changes in the precentral gyrus can thus be contaminated by signal leakage from the postcentral gyrus. Other methodological challenges to identify the double-hand signature may arise from sensitivity limitations, task design, analysis design and effective resolution. Our results show that CBV-based fMRI has a higher localization specificity than conventional GE-BOLD fMRI (Fig. S2, S5), thus overcoming the specificity limitations of previous studies. Even though CBV-based fMRI has a lower sensitivity and requires longer scan durations to exceed the detection threshold, it provides clearer results. The higher localization specificity of CBV-fMRI has previously been shown in humans with VASO fMRI (Huber 2017) across cortical layers. In this study we extend this finding and also show the higher localization specificity of VASO across cortical columns. Note that the CBV weighting in VASO has been extensively validated by comparisons with gold-standard methods in rats and monkeys across layer and columns (Huber et al., 2015a-c; Kennerley et al., 2013).
To optimally separate individual digit representations, we included long rest periods into the experiment and used a pseudo random ordering of individual fingers. This is different to previously employed “phase-encoding” paradigms and it comes at the cost of less efficient task design requiring longer scan durations. However, it can be more sensitive to detect deviations of continuously linearly aligned representations.
Most of the knowledge on the functional representation of movements in the primary motor cortex has been obtained from countless experiments in monkeys over the last century. The current state of consensus in the field is nicely summarized by Paul Cheney in (Omrani 2017; see also references therein); Overall, corticomotoneuronal cells in the primary motor encode muscle-related parameters of movement such as muscle activity and muscle force. Although some corticomotoneuronal cells in the primary motor cortex (particularly those involved with finger movements) have their terminations confined to motoneurons of single muscles, a large amount of corticomotoneuronal cells are not rigidly coupled to the activity of its target muscles but show specialization for particular movements or categories of muscle activity. Namely, almost half of the corticomotoneuronal cells facilitate muscles involving at least one distal and one proximal joint and are specialized for specific muscle synergies, E.g. for reach-to-grasp movements. With respect to action representations shown in Fig. 2B, it is important to note that Cheney and Fetz (1985) had previously identified the muscle fields of neighboring corticomotoneuronal cells. They showed that neighboring corticomotoneuronal cells had muscle fields that were very similar. Hence, the notion of cortical patches that are preferentially activated for grasping and retraction actions, as shown in Fig. 2B, has its basis in previous monkey data and could refer to these previously described muscle fields.
A previous study by Ejaz et al. (2015) already identified deviations from linear organizations for finger representations in the human motor cortex with GE-BOLD at 2.5 mm and 1.4 mm resolutions, respectively. These data already showed some indications of multiple finger representations (e.g. Fig. 1 in (Ejaz et al. 2015)). However, these data were not discussed with respect to an alternative geometric somatotopic organization principle such as a mirrored representation.
Previous studies with PET had shown multiple hand movement representations across both sub-areas of the primary motor cortex (Geyer 1996), BA4a and BA4p, respectively. The mirrored double-hand representation shown here (Fig. 2-3) is located in the anterior side (BA4a) of the primary motor cortex, only. Thus, the hand representations in BA4p, as discussed in (Geyer 1996) refers to yet another representation of the fingers. This is the evolutionary younger part of M1 that is located deep in the central sulcus. In this part of M1, individual body parts are largely overlapping (probably to facilitate complex hand movement) and thus, in this part of the motor cortex finger dominance maps might be misleading ways of depicting the complex representation principle (Ejaz et al., 2015). We find that this finger representation in BA4p is actually not completely separated from the one in BA4a. It is partly connected to the representation in BA4a and does not show a mirrored pattern (Fig. S9).
In contrast to human fMRI literature, electrophysiological recordings have shown deviations from a linear somatotopic organization. However, much of our understanding from the electrophysiological organization in M1 comes from experiments in which stimulations or recordings are performed from few cortical points only, whereas neuroimaging samples continuously and uniformly across space. Thus, continuously sampled maps of the motor cortex representations with electrophysiology at the sub-millimeter resolution have not been collected across columns and layers yet. And consequently, most of these studies only conclude that the representation pattern is complicated and nonlinearly organized (Schieber 1993; Hatsopoulos 2010) without proposing an alternative organization principle.
In this study we show for the first time that individual fingers have multiple representations in a mirrored pattern along the lateral-medial axis of primary motor cortex, with each whole-hand instance corresponding to a distinct movements. This mirrored representation across grasping and retraction patches gives rise to neighboring representations for movement synergies (d’Avella and Bizzi 2005). As already suggested by Penfield (1937), our data show that the Penfield homunculus model is an oversimplification. With high resolution fMRI, we can confirm deviations of the homunculus model in M1 at very high resolutions corresponding to small body parts. Our data agree with the hypothesis that in the submillimeter regime, the motor cortex is organized as an action map (Graziano 2016).
The present data corroborate several findings from invasive electrophysiologic recordings and microstimulation experiments in rodents and monkeys, that have elucidated the organizational principle of M1, and extend those to the level of the human M1. The presented methodology allows noninvasive recording of mesoscale functional representations in human M1 that were previously inaccessible. This might help to further understand the blueprint of cortical control over movement dynamics and kinematics. Beyond that, functional imaging of M1 at the layer and columnar level may help to elucidate how aberrations in the organizational principle could lead to movement disorders with largely unknown pathophysiology, such as dystonia.
Summary
Here we used advanced non-invasive neuroimaging in humans to provide novel insights in the organisational principles of the primary motor and sensory cortices at the mesoscale. We demonstrate for the first time that individual fingers are represented multiple times in the primary motor cortex in a columnar fashion following a mirrored pattern, and these representations are differentially engaged during specific motor actions (i.e., grasping versus retraction movements). By using new imaging and analysis technology that bridges the gap between invasive electrophysiological recordings and non-invasive large coverage mesoscopic fMRI in humans, we resolve previous controversies of M1 representation principles as ‘body map’ at the macro-scale vs. ‘action maps’ at the micro-scale.
Author contributions
L.H., conducted the experiments. L.H., P.B. designed the tasks. L.H., E.F. designed the analysis. L.H., S.K., D.G. wrote the analysis code. B.P., L.H., D.I. designed the MR-sequence, D.H., M.B., N.P., S.M., J.G. provided advice on experimental design, sequence, analysis, and research direction. L.H. wrote the draft of the manuscript. All authors contributed to the study story line and edited the manuscript.
Data and Software availability
All raw, anonymized MRI data of this study can be anonymously downloaded (https://goo.gl/1rdfLe). Fully processed data of two participants can be downloaded (https://activecho.cit.nih.gov/t/9ewfvz13). All custom written software (source code) and evaluation scripts are available on Github (https://github.com/layerfMRI/Tolopoly_strips). The authors are happy to share the 3D-VASO MR sequence upon request via a SIEMENS C2P agreement. A complete list of scan parameters used in this study is available on Github (https://github.com/layerfMRI/Sequence_Github/tree/master/Topology).
Acknowledgements
The research was supported by the NIMH Intramural Research Program (ZIA-MH002783). We thank Kenny Chung and Harry Hall for radiographic assistance. The study was approved under NIH Combined Neuroscience Institutional Review Board protocol 93-M-0170 (ClinicalTrials.gov identifier: NCT00001360). Laurentius Huber was funded form the NWO VENI project 016.Veni.198.032 for part of the study. Portions of this study used the high-performance computational capabilities of the Biowulf Linux cluster at the National Institutes of Health, Bethesda, MD (biowulf.nih.gov). We thank Hartwig Siebner and Mark Hallett for comments on the manuscript. We thank James Kolasinsky for discussions about intermediate results of this study and task design. This preprint is formatted based on a LATEXclass by Ricardo Henriques.
Footnotes
↵1 Note, that in the context of fMRI, ‘columnar’ and ‘laminar’ do not refer to cytoarchitectonically and myelo-architectonically defined layers and receptive field defined columns (see discussion in the supplementary information 6). Instead they refer to mesoscopic structures that are aligned along (tangential) and perpendicular (radial) to the cortical depths. Also known as “cortical depths” and “hypercolumns”.
↵2 Such numbers of participants are considered reasonable even from the more conservative estimates (http://www.russpoldrack.org/2018/04/how-can-one-do-reproducible-science.html), as long as every participant is investigated for more than 4 hours (e.g., in visual neuroscience).