Loss of finger control complexity and intrusion of flexor biases are dissociable in finger individuation impairment after stroke

Reviewer #1 (Public Review):

Summary:
Impairment in hand function is a challenge for stroke rehabilitation, and its neural underpinnings are of paramount importance for the field of biomedical science and neuroscience. The present study uses a novel finger force measurement device to measure individual fingers' force production in three dimensions when one finger is needed to produce an independent isometric force. Enslavement, i.e., the unwanted coactivation of non-intended fingers, is exaggerated in stroke survivors. The study started out by noting that the contribution of underlying factors (the loss of corticospinal drive, intrusion of flexor synergy due to a loss of regulation on subcortical pathways, and/or biomechanical changes) is not well understood. Detailed analysis for the inter-dependence between finger forces shows that the covariation between finger forces showed stroke-specific changes in shape and magnitudes, and these changes are not caused by biomechanical constraints. The important message that the study tries to convey is that the magnitude change in finger coactivation of the paretic hand is caused by the two dissociable factors, i.e., a loss of complexity in finger control and an intrusion of flexor bias.

Strengths:
The targeted topic of individual control of fingers for stroke survivors is of both theoretical and applied importance. The methodology of using isometric finger force to fulfill simple yet relevant motor tasks for stroke patients is also novel and sound. The paper is concisely written with excellent figures.

Weaknesses:
I have three major concerns about the study: 1) the link between the analysis results and two of the study's main conclusions is weak, specifically for the conclusion that a loss of complexity in finger control and the intrusion of flexor bias is dissociable. 2) using hand posture measures to quantify the influence of biomechanical factors in stroke patients is not well justified. 3) only a limited number of stroke patients were recruited (n=13).

First, the conclusion that the two factors contributing to the magnitude of finger covariation pattern are dissociable is not well reasoned. For example, the reasoning is clearly stated (Line 434) as: "Given the above converging evidence that Angular Distance is a measure of complexity of the geometric shape of finger coactivation, whereas Euclidean Distance is more sensitive to the magnitude change of these patterns across task goals if the two factors are dissociable, the intrusion of flexor bias would predict the magnitude (Euclidean Distances), but not the shape (Angular Distances) of the enslavement patterns. "

The logic behind this statement is unclear. Suppose the "two factors" are the complexity loss (shown by Angular Distance) and intrusion of flexor bias (shown by Bias). In that case, we cannot just use the predictability and the lack of predictability of the measure of intrusion of flexor bias (Bias) to reach the above conclusion, i.e., the Bias (for the intrusion of flexor bias) and changes in Angular Distance (for the loss of complex loss) is dissociable. Why not just test the association between Bias and Angular Distance directly?

Another conclusion is that the changes of Euclidian Distance and Angular Distance from the pattern similarity analysis of finger coactivation patterns inform us that the coactivation shape is preserved but its magnitude is increased in the paretic hand. However, the shape measure (Angular Distance) shows a decrease in paretic hands, indicating the coactivations for different task requirements become similar in the paretic hand. It becomes similar across task conditions, but this does not mean the coactivation shape for each task requirement is preserved in patients. In fact, one possible sign of reservation might be an unchanged function of distance measure (varied by intended fingers or directions) between groups (ideally shown in the format as Figure 5B). As we can see from the figure, the shape is preserved in the mild group but not so in the severe group if we compare the data between groups. Statistically, it is better to do ANOVA and use the group*fingers and group*directions interaction to show the reservation of "shape." The same logic applies to the Euclidean Distance measure (Figure 7B and 7D). Again, the connection between data analysis results and conclusions should be clarified.

Second, the use of hand posture measures to quantify biomechanical factors for hand impairments is not validated.

Based on two hand posture measures, the study rules out the contribution of biomechanical factors for enslavement in patients entirely (Line 390). However, the alternative explanation for the negative effect of posture variables is that these two specific variables (Mount Distance and Angle) might not reflect the postural changes (and biomechanical factors in hand function) in patients. Note these two measures are not about the resting hand posture of the patient, which is often affected. It is the posture when the hand is inserted into the apparatus, and the total force readings are minimal. The force readings would be quite small if people are good at relaxing their muscles and inhibiting unwanted reflexes in a specific posture. Healthy hands can remain a small force for rather different postures. Thus, healthy hands can produce a range of possible minimum-force postures, making the reliability of these "minimum" posture measures questionable. For patients, on the other hand, since a minimum-force posture is related to the ability to relax the muscles, it probably reflects both biomechanical changes (muscles and tendons, etc.) and subcortical influence. Thus, using these two measures to rule out the possibility of biomechanical factors needs further justification.

Third, the number of stroke patients is limited (n=13), especially when one important test is to compare the mild group and the moderate-severe subgroups. The group comparison thus has small statistical power with a medium split.

As the study aims to tease out the contributions of biomechanical, subcortical, and cortical input to the observed impairment of enslavement, we need to be careful about whether the selected behavioral variables are justified to reflect these factors and whether the data analysis results coherently support the conclusions. As it currently stands, the paper still has room to improve to achieve its aims.

Reviewer #2 (Public Review):

Summary:
This study addresses the factors affecting the loss of independent control of finger forces after stroke. As central and peripheral factors contribute to this impairment, the authors used a novel apparatus and task to rigorously quantify the specific features of loss of finger individuation across all digits. The analyses ruled out the role of biomechanical constraints and revealed that the loss of independent control of finger forces is primarily driven by the interaction of two factors: loss of complexity in finger control (shape of enslavement patterns) and involuntary coactivations of task-irrelevant fingers (flexion bias).

Strengths:
1. The device and 3D finger individuation task are major strengths of the study, setting this work apart from previous work and enabling novel insights.
2. The analyses are thorough and well-designed. Of particular value is the analysis of finger force control in 3D Cartesian space and the use of Representational Similarity Analysis of finger enslavement pattern magnitude and shape.
3. A major contribution of this work is the teasing out of the effects of top-down factors versus biomechanical constraints affecting impairment of finger force control.
4. I found the discussion about complexity of finger control (lines 541-553) very interesting. The topic of adaptability of finger coactivation patterns in the context of dexterous manipulation is a key topic in robotics and neuroscience. In robotics, finger forces are decomposed into a grasp and manipulation component. In human motor control studies, this approach has identified their temporal coordination (work by Latash and Zatsiorsky, e.g., Gao et al., 2005) and potentially distinct sensorimotor control mechanisms (Wu and Santello, 2023). The authors might wish to discuss how coactivation patterns might contribute to the coordination of grasp and manipulation forces.

Weaknesses:
None (only minor clarifications, e.g., the term biomechanical constraints should be defined earlier in the paper).

Reviewer #3 (Public Review):

This paper seeks to characterize finger enslavement impairment after stroke-"the unwanted coactivation of non-intended fingers in individuated finger movements." In the past, three possible neuromuscular mechanisms contributing to finger enslavement were suggested: passive musculotendon properties, an intrusion of flexor bias, and a loss of complexity in finger control repertoire. To tease apart these factors, the authors simultaneously recorded all five fingertip forces using a sensitive isometric force measurement device, which allowed characterizing patterns of enslavement for all fingers in a variety of instructed tasks. This novel experimental design opened new opportunities to study finger enslavement in more detail. To analyze this multi-dimensional dataset, new metrics were introduced, and many detailed analyses were conducted. Here is a brief account of the important results as best as I can summarize them.

1. Gross finger individuation ability is lower in the paretic hand of stroke patients than in non-paretic or healthy hands. Enslavement worsens with the severity of overall stroke impairment.
2. The enslavement patterns - unintended finger forces as functions of an instructed force in a different finger - show smaller "complexity" in paretic than nonparetic hands. I.e., the directions of unintended finger forces in the paretic hand remain similar across various instructed tasks. This reduced complexity also correlates with the severity of stroke.
3. The enslavement patterns show larger magnitude differences in the paretic than non-paretic hands; i.e., the unintended fingers' forces show a larger shift when comparing two instructed force directions in a paretic finger.
4. Finger force biases exist in paretic and non-paretic hands and correlate with the severity of stroke. Biases are more pronounced in flexion than ab/adduction direction.
5. The resting hand posture does not correlate with finger force bias or enslavement patterns.
6. Finger force biases correlate with enslavement patterns in the paretic hand, but not in the non-paretic hand.
7. Flexor bias (force biases in flexor direction) does not correlate with gross individuation ability in the ab/adduction direction in the non-paretic hand, but it correlates with the ab/adduction individuation ability in the paretic hand.
8. Finger force biases do not correlate with directional differences in enslavement patterns on either hand. However, biases correlate with the magnitude of force shift in the enslavement pattern.
9. The intrusion of flexor bias (difference of finger force biases in paretic and non-paretic hands) does not correlate with directional differences in enslavement patterns in either hand, but it correlates with force shifts in enslavement patterns in both hands.
10. More principal components (in principal component analysis, PCA) are required to explain similar levels of variance of enslavement patterns in paretic than non-paretic hands.

Taken together, the authors use these results to claim that: 1) enslavement impairment is unrelated to passive biomechanical properties, and 2) loss of complexity and flexor bias both contribute to enslavement, but possibly via different mechanisms.

The first argument is supported by the result that resting hand posture does not explain gross individuation ability or enslavement patterns. Although these results are valid, biomechanical contributions are not ruled out altogether in my opinion. The experiment starts from the optimal posture in which minimal finger forces are recorded in a relaxed state, essentially an "equilibrium" posture where all forces from muscles, ligaments, and other soft tissues are balanced. However, this equilibrium posture alone does not represent potential asymmetry in passive biomechanical properties (e.g., at equilibrium, flexion may face less stiffness than extension), nor does it take into account complex interactions between muscles of the hand. A simple finger force requires the co-activation of several intrinsic and extrinsic hand muscles as well as those of the wrist, some of which may be weak, shortened, stiff, painful, etc. Even if neural activity is present, compensation from other muscles may be needed, which may lead to unintended forces in other fingers. Although my "hunch" agrees with the author's claim that neural contributions outweigh biomechanical factors in enslavement, I believe resting posture on its own cannot account for "all" biomechanical factors. Additionally, the results comparing biases in paretic and non-paretic hands (line 389) are unrelated to biomechanics. It is reasonable to believe that the passive biomechanical properties of the paretic hand are different from those of the non-paretic hand if long enough time has passed since the stroke. So biomechanical of one hand is not representative of the other hand. Even if biases in the non-paretic hand could explain those in the paretic hand, I find it hard to extend the conclusion that biomechanics is a factor.

The authors further presented detailed analyses to tease apart contributions of flexor bias and loss of complexity to enslavement. The flexor bias is straightforward to define, and its correlation with enslavement (or the absence of correlation in the non-paretic hand) is supported by the results. However, the arguments about complexity are less straightforward. Two separate definitions of complexity are used: one is the directional differences between enslavement patterns, and the other is based on the number of principal components. This is one source of confusion as to which definition is used when referring to "loss of complexity". Nonetheless, both complexities are shown to decrease with the severity of stroke. The first type of complexity is also shown to be uncorrelated to flexor bias. However, I did not find evidence among the results that directly linked complexity to enslavement. Could complexity, similar to biomechanical properties, be ruled out? This paper provides no evidence for or against the contribution of complexity to enslavement.

My last point is about the neural correlates of these characteristics. The authors frequently use the terms "low-level", "subcortical", "top-down cortical", etc. throughout the paper, while the results are exclusively at a behavioral level. This issue is also present in the abstract where the authors state that: "we aim to tease apart the contributions of lower biomechanical, subcortical constraints, and top-down cortical control to these patterns in both healthy and stroke hands"; however, the methods and the results are unrelated to neural aspects of control, and the authors only refer to other studied to link these behavioral effects to "potential" neural causes. Further, the intrusion of flexor bias is usually associated with "subcortical" neural pathways in Results. The authors have properly discussed these possible neural correlates in the Discussion, but mentioning these terms in the Results is unjustified and unsupported by the results or the methods. This paper does not provide any standalone evidence to directly link complexity or bias to their neural correlates.

Comments on the representational distance matrices (RDM):
Two types of RDM were defined: "by-Finger" RDM (Fig 4A), and "by-Target Direction" RDM (Fig 4D). While I understand the by-Finger RDM and it physically makes sense to me, I cannot fully wrap my head around the by-Target RDM. I leave the interpretation of these results to the reader.

The distinction between the Euclidean and Angular distance is also vague to me. Angular distance is a valid similarity measure for the directions of two vectors and it is unrelated to the norms of vectors. However, Euclidean distance is not fully independent of the Angular distance as the authors claim; it changes with both the norms of the two vectors and the angle between them. If the angular distance is small, then Euclidian distance mostly represents norm differences, but the statement "Euclidean distances are sensitive to the length difference between two force vectors but insensitive to direction differences" is not generally correct. This issue is particularly important because the averaging of distances (see my next point) masks details of individual distance values, which hinders the interpretation of the results.

The enslavement patterns and RDMs are potentially valuable metrics, however, the way they are condensed in the final statistical analyses reduces their value. The way I understand it, all elements of the RDM matrix are averaged into a single value. This averaging masks the details of individual pairs of comparisons, which not only reduces the information resolution but also seriously hinders rigorous analysis and interpretation of the results.

Author Response

We thank the reviewers and editorial team for their positive and thoughtful comments and recommendations for our paper. We will provide a detailed point-to-point response accompanying a revised version of our paper to carefully incorporate all the recommendations and clarify several confusing points. Here we provide a brief provisional response to summarize the key points.

1. Are the two factors in the enslavement patterns after stroke, changes in shape (loss of complexity) and magnitude (intrusion of flexor bias), dissociable? Our results show both a loss of shape (Fig. 5) and an increase of magnitude (Fig. 7) in enslavement patterns in the paretic hand. We agree with the reviewers that the key measures for these two factors, Angular (Cosine) and Euclidean Distances, are not mathematically orthogonal because, while Angular Distance is indeed only influenced by shape, Euclidean Distance is influenced by both magnitude and shape changes of the enslavement patterns. However, our LME results show that increased flexor bias in the paretic hand strongly predicts Euclidean Distance but not Angular Distance (Fig. 9), thereby suggesting that pattern shape change cannot be fully accounted for by flexor intrusion. This analysis was also recommended by Reviewer 1. In the revised version, we will further clarify the dissociation of the two components.

2. Can biomechanical factors be ruled out from the enslavement patterns in the paretic hand? We agree with the reviewers that resting hand posture measures alone cannot fully assess biomechanical factors, given that biomechanical constraints during action and abnormal postures due to neural loss after stroke were not captured in these measures. In the paper, however, we used three analyses to justify this point. In the first analysis, we showed that resting hand posture (Mount Distance and Mount Angle) could not account for the Biases in all groups (healthy, paretic, non-paretic). In the second analysis, we showed that resting hand posture could not account for Enslavement in all groups. In the third analysis, we showed that Biases in the non-paretic hand could not predict Biases or Enslavement in the paretic hand within the same patients. The third analysis was done based on the existing literature that secondary biomechanical change after stroke was likely not the major contributor in the hand impairment, where passive muscle stimulation could successfully evoke a similar level of fingertip forces in both stroke and control hands (Hoffmann et al. 2016) and median nerve stimulation could significantly reduce intrusion of finger flexion (Kamper et al. 2003). The resting hand posture and non-paretic hand biases would include both biomechanical and neural factors, but since none of these measures could predict enslaving patterns, we maintain that biomechanical factors would not be a contribution to the enslavement in the paretic hand.

3. Neural correlates of behavioral changes were not tested, therefore claims such as "low-level," "subcortical," and "top-down cortical" contributions are not fully justified. We agree with the reviewers, and we will clear references to these neural correlates from the text of the Results section in the revised version of the paper. These neural correlates will only be discussed in the Discussion section.

4. RDM construction for "by-Target Direction" was not clearly explained. We agree with the reviewer that the diagram in Fig. 4D was a little confusing. To construct these matrices, we analyzed differences in coactivation patterns of the non-instructed fingers when two fingers move in the same target direction. A cleaner pattern comparison should exclude both the two instructed fingers to be compared from the enslavement matrices. This will be clarified in the revised version.

References

Hoffmann, Gilles, Megan O. Conrad, Dan Qiu, and Derek G. Kamper. 2016. “Contributions of Voluntary Activation Deficits to Hand Weakness after Stroke.” Topics in Stroke Rehabilitation 23 (6): 384–92. https://doi.org/10.1179/1945511915Y.0000000023.

Kamper, D G, R L Harvey, S Suresh, and W Z Rymer. 2003. “Relative Contributions of Neural Mechanisms versus Muscle Mechanics in Promoting Finger Extension Deficits Following Stroke.” Muscle & Nerve 28 (3): 309–18. https://doi.org/10.1002/mus.10443.