Open Review of “A somato-cognitive action network alternates with effector regions in motor cortex” (Gordon et al., 2023)
Dollyane Muret1,2, Tamar Makin3, Jörn Diedrichsen4
1. Inserm, NeuroDiderot, Université Paris Cité, Paris, France
2. CEA, NeuroSpin UNIACT, Université Paris-Saclay, Paris, France
3. MRC Cognition and Brain Sciences Unit, University of Cambridge, Cambridge, UK
4. Western Institute of Neuroscience, Western University, London, Ontario, CA.
In the article “A somato-cognitive action network alternates with effector regions in motor cortex”, published in Nature this week, the authors argue for the existence of “inter-effector regions” in human primary motor cortex (M1) that are intermittently localised with the leg, hand and face territories. Primarily building on functional magnetic resonance imaging (fMRI), the authors construct a new framework for the functional organisation of M1. While this study offers interesting insights and perspectives, in terms of new methodology and explorative approaches, we think that the wealth of data presented here requires more careful analysis and more thoughtful interpretation. The following is not a comprehensive review of the paper, we simply would like to point at some of the key issues that dampen our enthusiasm regarding some claims of the paper.
Context and Novelty
The main issue in our view is to what degree the novel organization proposed in the paper, namely the existence of three “inter-effector regions” that play a functionally different role than the rest of motor cortex is supported by the data presented. Before evaluating this main argument, we believe that it is important to situate the main claims of the paper in the historical context. The authors contrast their new framework with the supposedly classical view, namely a continuous, linearly arranged motor homunculus, where each body part has a specific and distinct territory. While this is certainly the impression that is conveyed in many Neuroscience textbooks through the use of the iconic motor homunculus (Penfield & Rasmussen, 1950), there has in fact never been much empirical evidence that would support a strict linear organization. In fact, even Penfield himself did not make any such claims in his original papers. "The cortical motor sequence of man shows little preservation of the segmental representation of muscles found in spinal cord and brain stem. There was no evidence for separation of the movement of primitive flexors and extensors. Movements produced by cortical stimulation are gross, awkward. They involve multiple joints and numerous muscles." (Penfield, 1947) Furthermore, many researchers have pointed out that motor representations in M1 can show a center-surround property (Meier et al., 2008), be fractured (Schieber & Hibbard, 1993) and display multiple peaks for single body parts (Ejaz et al., 2015) (Huber et al., 2020). This and many other observations have led to the idea that M1 encodes coordinated actions, rather single body parts (Graziano & Aflalo, 2007). Several previous studies have used multimodal MR imaging to characterise M1 organisation, including the highly influential paper by Glasser et al. (2016), who parcelled M1 into 5 distinct sections, based on resting-state functional connectivity, movements-based activity, and structural markers. In addition, Kuehn et al. (2017) used 7T and 3T multimodal imaging (i.e., myelin-sensitive sequences, task-based and resting-state fMRI, including a similar seeding approach as used here, in both area 4 and 3b), and reported the existence of two cortical fields corresponding to the representations of the hand and face separated by a low myelinated septa, where different resting-state connectivity profiles were observed. As such, structural and functional discontinuities in the organization of M1 have been well established.
The stubborn persistence of the idea of a continuous linear motor homunculus therefore is not so much a result of any empirical data, but rather a case study on how a compelling simplifying illustration can distort the thinking of generations of scientists. As such, we definitely welcome a high-profile collection of evidence that demonstrates the inadequacy of the idea. At the same time, we are wary of the attempt to replace the simplifying and partially misleading Penfield-Rasmussen map (Fig 4a), with a potentially similarly misleading alternative. We therefore believe that a more cautious interpretation of the individual findings presented in the paper is required.
The obvious main strength of the paper is that it presents a wealth of datasets, including repeated resting-state scans (termed precision mapping) of 7 typical adult participants, an infant, a child, an adolescent with perinatal stroke, and an adult monkey. In addition, openly available large scale resting-state datasets were utilised, involving adults, adolescents and neonates. This comprehensive gathering of diverse datasets is extensive, and in our view provides an unprecedented opportunity to identify and robustly replicate new candidate organising principles for further research. By re-examining existing datasets through new prisms, this study is a tour de force example of the power of open data.
Another strength of the current study lies in the multimodal and broader approach combining on one hand resting-state, task-related and structural data. While previous studies made important contributions using similar multimodal approaches (Glasser et al. 2016; Kuehn et al. 2017), the current study stands out in its mission to explore potential differences between M1 regions in terms of their structure and functional connectivity with higher order brain regions. While exploratory, this approach could stimulate new hypotheses-driven research.
The main weaknesses of the study are a consequence of the exploratory, and sometimes circular, nature of the analysis used throughout the paper, the over-interpretation of the findings and the flexible inference that dominates the paper. While the sheer volume of evidence presented is overwhelming, many of the individual pieces do not meet the rigorous standards one would typically expect from an empirical paper. We would like to point at some of the key issues we identified:
The key conceptual contribution of the study is the introduction of these body-part (or parcel) boundaries as so-called inter-effector regions that, as the authors argue, form a somato-cognitive action network for mind-body integration. But as clearly demonstrated in the activity maps of the two participants, these regions are housed by representations of the abdomen (dorsal spot), upper-face (middle spot), and throat/inner-mouth (ventral spot). It is therefore unclear to us why the hegemony of these body parts over these cortical territories is dismissed. Instead, we believe it is important to consider the alternative hypothesis that the increased correlation between these sub-regions is caused by their functions. For example, these body parts may be co-engaged during breathing (i.e., involving both abdominal and inner-mouth/throat motion for air inflow/outflow) and blinking, which are some of the main muscular activities taking place during resting-state data collection. Such synchronized activity during rest could contribute to the increased inter-regional correlation that the authors interpret as increased functional connectivity (as demonstrated in Duff et al. (2018)). Conceptually, instead of classifying these areas as inter-effector regions, they could, in our view, also be classified as representing the abdomen, upper-face and throat.
The authors use connectivity ‘network sub-divisions’ in their resting-state datasets to identify the so-called “inter-effector” regions. However, rather than providing a systematic approach based on independent measures (anatomical and/or task-related activity markers), it seems that these regions (black outlines in Fig 1b) were defined based on their resting-state functional connectivity differences. Taking the fact that these regions (after defining them thus) have different functional connectivity profiles as a basis for the argument of a new organisation scheme in M1 seems circular to us.
Perhaps the most speculative claim of the study is that these so-called inter-effector regions function as somato-cognitive integrative regions. The authors used exploratory functional connectivity analysis with higher order areas to support their interpretation. However, the function of an area should not be inferred post-hoc and solely based on a thresholded resting-state connectivity map, especially considering that these higher order areas have been associated with a large range of functions. Labelling the primary function of the brain areas as a somato-cognitive interface for the whole-body action plans certainly doesn’t seem to provide a sufficiently precise hypothesis to critically test. A more detailed functional investigation of the inter-effector regions, for example using representational similarity analysis to identify task-relevant information content, and TMS to infer on consequences of stimulation of these regions, is required to test the hypothesis put forward here.
Where the authors do evaluate the function of the hypothesized inter-effector regions, the statistical evaluation is often insufficient. For example, the authors claim that the inter-effector regions have a special role in the planning of whole-body movements. This claim relies on task-related fMRI data (Fig 3e) from n=2 participants, making an assessment of the inter-subject variability of the results impossible. The statistical argument appears to rely on the fact that the planning-related activity was higher than the execution related activity, but that in other regions this effect was not significant. Here the authors seem to rely on a difference in significance rather than showing the significance of a difference, a common statistical fallacy (Nieuwenhuis et al., 2011) Inspection of Fig 3e however clearly also reveals planning related activation in the mouth region, which, similarly to the “inter-effector” regions was less involved in the execution of the movement. Thus, it is likely that the reported regional differences were due to the high execution-related activity in hand and foot regions (and the results would not replicate if the authors were to study coordination of muscles controlling the trunk and upper-face). For example, in our recent work, we did not find elevated activity or movement encoding in the supposed inter-effector regions during movement planning (Ariani et al., 2022).
During the task-related functional mapping, behaviour was not sufficiently rigorously monitored or in fact considered when interpreting the data. For example, would it be possible that some of the double-peaks observed using the Gaussian analysis could correspond to the representation of two body parts? For example, when moving the shoulder, the hand and wrist will receive sensory input through the induced movement, which would in turn impact M1 activity. We also think that the interpretation that differences in connectivity profiles between the neonate and the other samples (where breathing and blinking are co-occurring), interpreted as ontogenetic development, could have been influenced by task execution. Specifically, there are multiple differences during the ‘resting’ state of the neonates, who were asleep. Beyond blinking that is obviously absent when sleeping, sleep patterns are known to change rapidly during the first months of life (Weerd & Bossche, 2003). In addition, even if restrained, neonates, while sleeping, are more likely to move in the scanner, something that was not monitored here, and which could have influenced the resulting ‘connectivity’ maps. This criticism (recording during sleep and motion more likely) also holds for the macaque dataset.
Overall, we welcome the paper as an attempt to overhaul an oversimplified diagram that has had a disproportionate influence on the neuroscientific understanding of M1. The idea of an interspersed network that works qualitatively differently (and functionally on a hierarchically higher level) than the rest of M1 is an intriguing hypothesis. Given the lack of hard evidence for this claim, however, we believe it is premature to suggest that the new diagram of M1 (Fig 4b) provides a more adequate characterization.
Penfield & Rasmussen (1950). The Cerebral Cortex of Man: A Clinical Study of Localization of Function.
Penfield (1947). Some observations on the cerebral cortex of man. Proceedings of the Royal Society of London. Series B, Biological sciences.
Meier et al. (2008). Complex organization of human primary motor cortex: a high-resolution fMRI study. Journal of neurophysiology.
Schieber & Hibbard (1993). How somatotopic is the motor cortex hand area?. Science.
Ejaz et al. (2015). Hand use predicts the structure of representations in sensorimotor cortex. Nat Neurosci.
Huber et al. (2020). Sub-millimeter fMRI reveals multiple topographical digit representations that form action maps in human motor cortex. NeuroImage.
Graziano & Aflalo (2007). Mapping behavioral repertoire onto the cortex. Neuron.
Glasser et al. (2016). A multi-modal parcellation of human cerebral cortex. Nature.
Kuehn et al. (2017). Body Topography Parcellates Human Sensory and Motor Cortex. Cerebral cortex (New York, N.Y. : 1991).
Weerd & Bossche (2003). The development of sleep during the first months of life. Sleep Medicine Reviews.
Nieuwenhuis et al. (2011). Erroneous analyses of interactions in neuroscience: a problem of significance. Nature neuroscience.
Ariani et al. (2022). Motor planning brings human primary somatosensory cortex into action-specific preparatory states. eLife.
Duff et al. (2018). Disambiguating brain functional connectivity. NeuroImage.