Research that examines how the body adapts to new movements is shedding new light on how the nervous system learns, and could help to inform a wide range of applications, from customized rehabilitation and athletic training to wearable systems for healthcare. The research is published this week in the journal Current Biology.
“How does our brain figure out how to best move our body? It turns out that this can be a challenging problem for the nervous system, considering we have hundreds of muscles that can be coordinated hundreds of times per second – with more possible coordination patterns to choose from than moves on a chessboard,” says study senior author and Simon Fraser University (SFU) professor Max Donelan, director of SFU’s Locomotion Lab.
“We often experience changes to our body and our environment. Perhaps you enjoy a long run on a Saturday morning – your muscles may fatigue as the length of the run increases. Perhaps you choose to run on the beach on vacation – the sand may be uneven and loose in comparison to the pavement on the sidewalk. While we might register that these changes have occurred, we might not appreciate how our body adapts to these changes.”
Donelan’s team of neuroscientists who study motor learning collaborated with a Stanford University team of mechanical engineers who design human-robot systems. Together, they tracked the walking characteristics of study participants wearing exoskeletons.
Researchers found that the nervous system solves the problem of learning a new movement coordination pattern by first exploring and evaluating many different coordination patterns. This exploration was measured as a general increase in variability spanning the levels of the whole movement, joint, and muscle.
With experience, the nervous system adapts specific aspects of movement and simultaneously decreases variability along these aspects. The researchers also found that these adaptive changes improved movement overall, reducing the energy cost of walking by about 25 percent.
“We created new contexts using exoskeletons that act to assist walking, and then studied how people explore new movements and learn more optimal ones,” says Sabrina Abram, the study lead author and former graduate student in the Locomotion Lab. Participants experienced walking in this context over six days, resulting in about 30 hours of lab time for each and an extraordinary amount of data collected by co-author Katherine Poggensee.
While the nervous system appears to benefit from first searching among many different coordination patterns, it also benefits from reducing this search space over time, Abram adds. “This is because continuing to search among coordination patterns that already reduce energy can in turn increase energy, as well as add to the already challenging problem of figuring out the best way to move.”
Understanding how the brain searches for and figures out how to best move the body is important for a runner navigating new terrain, as well as a patient recovering from spinal injury or stroke.
For example, knowing when the body has adapted to a new training regimen can help coaches identify at which point an athlete should transition to learning new skills. This can also be useful for designing wearable systems – such as exoskeletons and prosthetics – by facilitating learning, and then evaluating people’s optimal responses to a range of designs.
“We would all like to move in the best way possible. For healthy people, it seems that with the right circumstances, the brain can take care of this. For those recovering from an injury, we might learn about how to best rehabilitate this injury from a better understanding of how the nervous system learns to adapt,” says Donelan.
When we move an individual limb or body part like one of our fingers, many different cortical areas in frontal and parietal lobes show elevated levels of activity [1–4]. However, it is far from clear how the many different brain regions contribute to motor output. Even in primary motor cortex (M1), which shows the highest correlation with localized muscle activity [5,6], it is not fully understood how the neuronal activity contributes to the actual movement [7–9].
Exemplary of this lack in understanding is that M1 has been reported to exhibit both a somatotopic organization (i.e. the orderly topography of cortical body part representations, [4,10–14]), as well as efferent connections exceeding the range of individual body parts or localized muscle groups [15–17]. In our previous study, we proposed that a Gaussian population Receptive Field (pRF) model may help to reconcile these multiple M1 interpretations .
Our pRF model showed that M1 neuronal populations (i.e. small ensembles of neurons within MR-voxels) can both contain a preferred finger representation (pRF center) constituting the somatotopy, as well as connections to adjacent fingers reflected by the pRF size. How fingers or other body parts relate to each other within small neuronal populations can illustrate how motor cortices are wired and what functions they perform with respect to individual body part movements. Since many body parts can move in conjunction, the mutual relation between different body parts is not trivial.
Our previous study investigated the movement of fingers only and, additionally, assumed a rigid order of fingers (from thumb to little finger), predefining the internal pRF structure. The limited number of body parts in combination with an a priori assumption on their reciprocal relations prevents quantification of body part relationships. Thus, while our previous study indicates that pRF modeling is able to model cortical motor activity, it is unknown how body parts relate to each other and how body parts are ordered within the response profile of neuronal populations.
In the current study, we investigate the relationship between body part representations in human motor cortices following an 18-body-part motor task, using pRF modeling and high-field 7 Tesla Blood-Oxygenation-Level-Dependent (BOLD) fMRI. At this point we note that in light of cortical motor activity the term ‘population Response Field’ is better suited than ‘population Receptive Field’, since cortical motor activity cannot be solely receptive in nature.
Hence, the abbreviation pRF will refer to ‘population Response Field’ from here on. We postulate that reciprocal relationship between body parts can be elucidated by estimation of the internal structure of whole-body pRFs. Conventional pRF modeling tries to fit a Gaussian pRF across a rigid functional space, e.g. visual field locations  or auditory frequencies , which has also been applied to finger space in combination with somatosensory [21,22] and motor tasks . However, to adequately assess the internal structure of neuronal populations with respect to motor activity, we cannot simply assume that each pRF consists of a rigid ordering or body parts, e.g. similar to the conventional cortical homunculus ordering of body parts .
Therefore, we developed a novel non-rigid pRF method that does not assume a rigid ordering of body parts. Rather than fitting a variable Gaussian function along an unchanging dimension of body parts, variably positioned body parts are fitted within a static Gaussian shaped pRF. The non-rigid pRF method can be regarded as a Gaussian shaped theoretical response field, which is populated with a set of functions (body part movements in the current study).
Properties that are common to conventional pRF methods, such as pRF center and size, can likewise be extracted from the non-rigid pRF method on the basis of position, number and spread of functions within the theoretical response field. Additionally, the non-rigid pRF approach allows for the investigation of pRF composition: one can address which body parts constitute the total pRF, including the proximity between body parts. Thus, the novel non-rigid pRF center allows for estimation of conventional pRF properties such as pRF center and size, and allows for the investigation of occurrence and proximity of body part representations within a pRF without making assumptions on the intrinsic structure of the pRF.
In order to estimate a whole-body pRF, eighteen body parts were selected for movement that encompass the lower limb, midsection, upper limb and face. The distribution of selected body parts is not uniform in terms of physical size, but was instead determined by the ability to be moved on cue. Therefore, the upper limb and face consist of more body parts that are cued for movement, compared to the lower limb and midsection.
In order of appearance on the cortical homunculus, those body parts are: toes, ankle, knee, abdomen, shoulder, elbow, wrist, little finger, ring finger, middle finger, index finger, thumb, forehead, eyelid, nostril, lip, jaw, and tongue (Fig 1A). Each neuronal population will represent these body part movements within its pRF in its own unique way. Through averaging the pRFs from neuronal populations with the same body part preference (i.e. pRF center), the mean body part pRF is obtained, which represents the average response profile for any given body part movement.
The relationship between body parts can then be assessed with graph theory on the basis of the mean body part pRF [24–26]. Whole-body graphs are constructed by correlation of the mean body part pRFs, representing the linkage and connection strength between body parts. For each body part representation we calculate graph theory metrics that reflect relevant aspects of body part relations: the connectivity (degree), clustering coefficient and betweenness centrality coefficient.
The connectivity metric estimates the connectedness of body parts based on the similarity of their respective mean body part pRFs: the larger the connectivity, the more similar a body part’s response field is compared to other body parts. The clustering metric is a measure of ‘clique-formation’, representing the interconnectedness of a body part and its neighboring body parts [27,28]. Betweenness centrality is measure of body part influence: here it represents the (indirect) involvement of a particular body part when other body parts move [29,30]. Lastly, we define modules of body part representations, based on shared characteristics of the mean body part pRFs (Fig 1B and 1C), using Louvain modularity [31,32
In the current study, we investigate the relationships among 18 different body parts in the following cortical areas related to motor control: primary motor cortex (M1), primary somatosensory cortex (S1), supplementary motor area (SMA), dorsal and ventral premotor cortex (PMd and PMv, respectively), insular cortex (Insula), and superior and inferior parietal cortex (sPC and iPC, respectively).
Body part relationships are scrutinized in several distinct ways. Using our novel non-rigid pRF model, we first estimate pRF center and size, approximating the neuronal population’s body part preference and the size of the population’s response field. We hypothesize that the non-rigid pRF centers reveal somatotopic structures in cortical motor areas that have previously been reported to exhibit a somatotopy: M1, S1, SMA and the insula [4,11–14].
Additionally, we hypothesize that the non-rigid pRF sizes will be smallest for primary sensorimotor cortices (M1 and S1), since activity profiles from primary sensorimotor cortices are thought to correlate to individual body parts to a greater extent than activity from secondary motor cortices [33,34]. Second, we quantify relationships between body parts as observed within the non-rigid pRF. We hypothesize that body parts that are adjacent on the cortical homunculus share a high proximity within response fields. Finally, the graph theory metrics describe the relations of body part representations in different cortical areas from the brain’s perspective. The uniqueness of body parts is given by the connectivity measure, the cliqueness is given by the clustering coefficient and body part influence is given by the betweenness centrality coefficient. The modules reflect which body part response profiles share similar characteristics.
reference link : https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1009955
More information: Sabrina J. Abram et al, General variability leads to specific adaptation toward optimal movement policies, Current Biology (2022). DOI: 10.1016/j.cub.2022.04.015