Agonist-antagonist muscle strain in the residual limb preserves motor control and perception after amputation.

Background: Elucidating underlying mechanisms in subject-specific motor control and perception after amputation could guide development of advanced surgical and neuroprosthetic technologies. In this study, relationships between preserved agonist-antagonist muscle strain within the residual limb and preserved motor control and perception capacity are investigated.
Methods: Fourteen persons with unilateral transtibial amputations spanning a range of ages, etiologies, and surgical procedures underwent evaluations involving free-space mirrored motions of their lower limbs. Research has shown that varied motor control in biologically intact limbs is executed by the activation of muscle synergies. Here, we assess the naturalness of phantom joint motor control postamputation based on extracted muscle synergies and their activation profiles. Muscle synergy extraction, degree of agonist-antagonist muscle strain, and perception capacity are estimated from electromyography, ultrasonography, and goniometry, respectively.
Results: Here, we show significant positive correlations (P < 0.005-0.05) between sensorimotor responses and residual limb agonist-antagonist muscle strain. Identified trends indicate that preserving even 20-26% of agonist-antagonist muscle strain within the residuum compared to a biologically intact limb is effective in preserving natural motor control postamputation, though preserving limb perception capacity requires more (61%) agonist-antagonist muscle strain preservation. Conclusions: The results suggest that agonist-antagonist muscle strain is a characteristic, readily ascertainable residual limb structural feature that can help explain variability in amputation outcome, and agonist-antagonist muscle strain preserving surgical amputation strategies are one way to enable more effective and biomimetic sensorimotor control postamputation.

Introduction

Proprioception is possible due to the presence of sensory organs within peripheral tissues including muscles, tendons, joint capsules, and skin[1,2]. Among these sensory organs, proprioception is primarily mediated by mechanoreceptors called muscle spindles and Golgi tendon organs which sense muscle length, speed, and tension[3]. Proprioceptive neural signaling relies on both microscale mechanotransduction processes[4], and macroscale biomechanically-functional tissue architectures[1]. The realization of such an architecture in a person with biologically intact limbs is implemented by mechanically-coupled antagonistic muscles spanning an articular joint that enables afferent signaling from the mechanoreceptors corresponding to limb movements through agonist-antagonist muscle strain (AMS).

The conventional standard-of-care amputation paradigm permanently disrupts the anatomical and neuromechanical principles of AMS, resulting in perturbed proprioception in people living with limb loss. Many ongoing efforts to restore locomotion for persons with leg amputation involve motor intent classification strategies based on electromyography (EMG) and intrinsic prosthetic signals[5,6]. However, in the absence of visual feedback, postural responses and balance during walking remain challenging for persons with leg amputation[7,8], which may indicate that motor control and proprioceptive percepts are significantly altered in persons that have undergone a conventional amputation procedure[9-11]. Toward better, more biomimetic control of an external prosthesis, invasive nerve interfacing using artificial electrical stimulation has shown great potential in restoring cutaneous[12-14] and proprioceptive sensation[11,12]. However, due to the complexity of afferent signaling through artificial nerve stimulation, and the relatively limited resolution of state-of-the-art implantable devices, it can be challenging to engineer stable neuroprosthetic interfaces that offer natural cutaneous and proprioceptive percepts.

As an alternative approach to neuroprosthetic interface design, surgical methodologies to reconfigure residual limb soft tissues may lower the burden of engineering and offer a more efficacious, biomimetic motor control strategy while also providing feedback from the amputated limb via biological sensory organs[15]. Conventional myocutaneous flap amputation procedures prioritize creating enough muscle padding for prosthetic socket fitting[16]. To enhance neuroprosthetic control, amputation paradigms seeking further reconfiguration of residual limb soft tissues have been developed, including Targeted Muscle Reinnervation (TMR)[17-20], Regenerative Peripheral Nerve Interfaces (rPNIs)[21-23], and the Agonist-antagonist Myoneural Interface (AMI)[24-27]. Each of these techniques have been demonstrated in combination with neuroprostheses[17,19,22,24] for the enhancement of prosthetic control. Nevertheless, neuroprosthetic performance is a system-level evaluation that depends on multiple factors such as subject-specific inherent capacities of residual limb motor control and phantom limb perception, engineered functional feedback, presence of visual feedback, and choice of the neuroprosthetic control paradigm. Consequently, there is a clear and present need to uncover the fundamental nature of how surgical residual limb reconstruction alone impacts clinical outcomes after amputation.

In this study, we investigate motor control and phantom limb perception capacities of 14 clinical research subjects having unilateral transtibial amputation spanning a range of ages and etiologies. Of the 14 participants, 7 subjects had received the AMI amputation, and 7 had undergone a non-AMI amputation. The transtibial AMI amputation comprises the surgical creation of dynamic agonist-antagonist muscle pairs for the enhancement of AMS. One muscle pair is constructed for the ankle joint comprising the lateral gastrocnemius linked to the tibialis anterior, and a second muscle pair for the subtalar joint comprising the peroneus longus linked to the tibialis posterior[22]. The study focuses on the impact of AMS preservation within the residual limb on motor control and phantom limb perception. We hypothesize that enhanced levels of residual limb AMS will improve motor control naturalness and proprioceptive perception postamputation in persons with transtibial amputation. To evaluate this hypothesis, the study clinically evaluates the naturalness of motor control and limb perception capacity during ankle and subtalar joint movements without visual or any other functional feedback. We collected muscle electromyography patterns and two degrees-of-freedom (2-DoF) kinetic data during bilateral, mirrored movements between the intact and phantom ankle-foot limbs of each subject. For these mirrored movements, we assessed the degree of residual limb AMS using ultrasonography. Because motor control in biologically-intact limbs is executed by the activation of combinations of muscle synergies, we evaluate motor control naturalness and limb perception of amputees using muscle synergy analysis[28-30]. The study findings support the hypothesis that enhancing AMS in the residual limb improves motor control naturalness and perception after amputation, underscoring the importance of surgical techniques such as the AMI that create a residuum tissue structure that preserves agonist-antagonist muscle dynamics.

Methods

Study design and clinical evaluation

The present work begins to investigate one outcome measure—economy of motion—of our ongoing clinical trial, NCT03913273, although we do not report any pre-specified endpoints of that trial in the present work. In overall scope the NCT03913273 trial investigates if AMIs can (i) improve voluntary free-space prosthetic control, (ii) improve voluntary and involuntary (reflexive) prosthetic terrain adaptations, and (iii) serve as a bidirectional human-device interface after transtibial amputation (https://clinicaltrials.gov/). The relationship of the present work to that trial is to obtain preliminary data and an algorithmic framework—a muscle synergy model—and thus inform our assessment of the pre-specified outcome measures for NCT03913273. Computational tools such as the model investigated in the present work are a critical part of studying and quantifying voluntary motor control postamputation in free space, ambulatory ascent and descent of stepping stairs while wearing a prototype multi-degree-of-freedom prosthesis, and potential for closed-loop prosthetic control by functional electrical stimulation.

All data in the present study, were collected at the Massachusetts Institute of Technology (MIT) under IRB approval from our Committee on the Use of Humans as Experimental Subjects (protocol 1812634918). All participants signed informed consent forms prior to data collection. The work followed the same prospective, non-randomized study design described in NCT03913273. The time period of recruitment and data collection was June 12, 2019 through September 19, 2021. Eligibility criteria included transtibial amputee subjects within an age range of 18 to 65 years, a fully healed amputation site, proficiency in the use of a standard lower-extremity prosthesis, and capability for ambulation with variable cadence (K level 3 and 4[31]). Exclusion criteria included one more of the following underlying health conditions: cardiopulmonary instability manifest as coronary artery disease, chronic obstructive pulmonary disease, extensive microvascular compromise, as well as persons who are pregnant and/or active smokers.

Table 1 of the present work summarizes the 14 study participants, listing a total of 7 AMI subjects and 7 non-AMI control (CTL) subjects. The age of the subjects ranged from 25 to 62 years. The male to female ratio was 5:2. The subjects represented different amputation types: AMI[25] (7/14), conventional[16] (6/14), and Ertl osteomyoplasty[32,33] (1/14). AMI amputation surgeries had been done according to Partner’s Institutional Review Board protocol p2014001379 as in previous reports[24,25].

Table 1

Study population.

Participant ID	Amputation type	Age (years)	Time since amputation (years)	Amputation etiology	Biological sex	Height (m)	Weight (kg)
AMl-1/BIO-A1	AMI	43	1.6	Thermal Injury	Female	1.68	81
AMl-2/BIO-A2	AMI	55	2.7	Trauma	Male	1.73	77
AMl-3/BIO-A3	AMI	50	1.0	Trauma	Female	1.68	81
AMl-4/BIO-A4	AMI	58	1.2	Trauma	Male	1.90	93
AMl-5/BIO-A5	AMI	32	0.5	Trauma	Male	1.75	75
AMl-6/BIO-A6	AMI	29	0.6	Trauma	Male	1.68	84
AMl-7/BIO-A7	AMI	48	0.5	Trauma	Male	1.70	75
CTL-1/BIO-C1	Standard	25	1.4	Oncological	Female	1.64	54
CTL-2/BIO-C2	Standard	62	2.7	Trauma	Female	1.65	81
CTL-3/BIO-C3	Standard	25	2.0	Talipes Equinovarus	Male	1.78	108
CTL-4/BIO-C4	Standard	39	2.7	Trauma	Male	1.60	63
CTL-5/BIO-C5	Ertl osteo-myoplasty	62	2.6	Trauma	Male	1.80	97
CTL-6/BIO-C6	Standard	61	5.3	Trauma	Male	1.73	91
CTL-7/BIO-C7	Standard	46	8.7	Trauma	Male	1.78	75
Mean ± s.d.		45 ± 14	2.4 ± 2.2			1.72 ± 0.08	81 ± 14

AMI Agonist-antagonist Myoneural Interface, residual limbs of participants who underwent an AMI amputation (AMI-1-7), residual limbs of participants who underwent a Non-AMI control amputation (CTL-1-7), unaffected biologically-intact limbs (BIO-A1-7 and BIO-C1-7).

As noted in the introduction section, the AMI transtibial amputation procedure creates mechanical linkages between two pairs of natively vascularized and innervated muscles within the residual limb; one pair for the missing ankle joint and another pair for the missing subtalar joint[22,23]. For the ankle joint AMI construct, the tibialis anterior (TA) was linked to the lateral gastrocnemius (GA), and for the subtalar joint AMI construct, the tibialis posterior (TP) was linked to the peroneus longus (PL). The AMI amputation aims to emulate physiological antagonistic actuation between the residual limb muscles to restore AMS (Fig. 1a). In distinction, some non-AMI amputations may perturb AMS by severing or restricting residual agonist-antagonist muscle movements. Agonist-antagonist muscle couplings were not specifically reconstructed during conventional amputation or Ertl osteomyoplasty amputation procedures[16,32,33]. Thus, the study population represented differing degrees of AMS within the residual musculature.

Fig. 1

Clinical evaluation of sensorimotor responses and the degree of agonist-antagonist muscle strain (AMS) for participant’s residual limb muscles.

Shown in a, the Agonist-antagonist Myoneural Interface (AMI) amputation seeks to emulate physiological actuation of antagonistic muscle contraction and stretch. Ankle and subtalar AMI constructs are devised to create direct agonist-antagonist coupling for ankle dorsi and plantarflexion and for subtalar eversion and inversion. For the ankle joint AMI construct, the tibialis anterior (TA) is linked to the lateral gastrocnemius (GA), and for the subtalar joint AMI construct, the tibialis posterior (TP) is linked to the peroneus longus (PL). In b and c, the experimental setup is shown. Motor control and phantom limb perception capacity are assessed in free space without visual or any other functional feedback. Perturbed motor control and perception are anticipated if a critical degree of AMS is not preserved in the limb. Representations of the dorsi and plantarflexion synergic motor outputs are shown in green and red, respectively; efferent and afferent neural signals are shown in brown and yellow, respectively. Eversion and inversion were also tested but are not shown here. In d, AMS is computed from muscle fascicle changes during cyclic phantom ankle and subtalar joint movements. Here, the size and positioning of elements are representative and not to scale.

We collected EMG simultaneously from the TA, GA, TP, and PL muscles of both the residual limbs (AMI, CTL) and unaffected biologically intact limbs (BIO-A, BIO-C). A 2-DoF goniometer was also placed on the posterior aspect of the unaffected BIO limb spanning the ankle-foot complex (Supplementary Fig. 1a–c) to record mirrored movements between the intact and perceived phantom limb. We provided multiple motor control task instructions via on-screen and audio recordings. During motor control trials, no visual or other functional feedback was provided in an effort to focus on investigating the impact of proprioceptive feedback on motor control and limb perception capacity (Fig. 1b, c). To compute AMS, we utilized ultrasonography to record from the residual limbs while each subject repeated cyclic plantarflexion-dorsiflexion (PF-DF) and inversion-eversion (IN-EV) mirrored phantom limb movements (Fig. 1d, Supplementary Fig. 1d, e). The maximum muscle fascicle strains were then estimated from ultrasound video recordings which were further normalized by the nominal muscle fascicle strain ranges from a computational musculoskeletal limb model[34]. The average PF-DF and IN-EV AMS values were utilized to represent the degree of AMS within the residuum. Given this definition, the degree of AMS ranges from 0 to 1, where zero indicates that the subject preserved none of the AMS present in the biologically intact limb, and 1 indicates fully preserved biological AMS.

Surface electrodes placements and EMG processing

Bipolar surface electrodes were acutely placed over each of the target muscles for EMG recording. The target muscles include GA, TA, TP, and PL of both the residuum and unaffected limb (Supplementary Fig. 1a). Ultrasound imaging was used to guide electrode placement when it was needed. All the EMG signals were off-line high-pass filtered (fourth-order zero-lag Butterworth filter, 20 Hz cut-off frequency). The filtered EMG signals were full-wave rectified and low-pass filtered (fourth-order zero-lag Butterworth filter, 5 Hz cut-off frequency) to compute muscle activation patterns. All EMG signals were normalized to calibrated maxima for each muscle.

Muscle synergy extraction and synergy activation profile

The motor control of residual limbs was evaluated by muscle synergy extraction from muscle activation patterns during discrete ankle and subtalar joint movement trials. Subjects were asked to sequentially make PF, IN, DF, and EV movements of both the residual (phantom) and biologically intact limbs. The order of discrete movement trials was not randomized in an effort to identify existing motor capabilities for the 4 principal movements as accurately as possible. Discrete movements were repeated 40 times. We used a generally accepted mathematical model[29] for the representation of motor outputs as muscle synergy combinations aswhere is the muscle activation patterns at time t; is the ith muscle synergy vector; is the time-varying coefficient, or synergy activation, for i-th muscle synergy vector; N is the total number of muscle synergy vectors composing the muscle activation patterns; and is the residual. Briefly, this model represents the muscle activation patterns as linear combinations of a set of time-invariant muscle synergy vectors that are activated by time-varying coefficients. As is generally accepted in the field, we consider as a muscle synergy profile that has a structural basis in the nervous system and consider as an index of motor commands, or synergy activations. Muscle synergies were extracted by the non-negative matrix factorization (NMF) algorithm[35]. The NMF was started with the initialization of time-varying coefficients and muscle synergy vectors to random positive values in the [0 1] interval. The goodness of fit metric of decomposed matrices was evaluated by variance accounted for (VAF)[36]. The NMF was continued until the change in VAF in 50 consecutive iterations was less than a tolerance of . To reduce the probability of finding a local minimum solution for the NMF optimization, the same procedures were repeated 30 times with different sets of initial conditions, and the solution with the most VAF was selected. The number of muscle synergies was selected as the least number of synergies that could adequately reconstruct the muscle activation patterns, as determined by VAF > 0.95[36]. To enable intrasubject comparisons of motor commands, the average vectors of synergy activation profiles were used, orwhere is a unit vector in synergy space indicating activation of synergy vector ; is a time period of a given task. To identify motor commands for PF, DF, IN, and EV, the muscle activation patterns for each discrete movement were gathered and synergy activation vectors were computed independently. Then, the synergy activation vectors were normalized for further analysis.

Naturalness of muscle synergy and synergy activation

We quantified the naturalness of muscle synergy and synergy activation profiles of AMI and CTL groups by computing their similarities to the average normalized values of the BIO group. Specifically, a muscle synergy of one subject was considered to correspond to a muscle synergy of another subject when the maximum of the scalar products was found among others. After sorting the muscle synergy vectors, the representative muscle synergy vectors of the BIO group were determined as the normalized average muscle synergy vectors of the BIO group. Finally, the naturalness of one’s muscle synergy was calculated by plotting the mean scalar products with the representative of the BIO group. Similarly, the naturalness of one’s synergy activation vectors was calculated by plotting the mean scalar products with the normalized average synergy activation vectors of the BIO group. For the BIO group, a leave-one-out procedure was used for computing their dot products. The universal number of synergy vectors for similarities analysis was unified to 3 muscle synergy vectors, which was the number of synergy vectors of all subjects in the BIO group. When fewer synergy vectors were identified previously by synergy extraction procedures, vectors were added to muscle synergy and synergy activation vectors to match the dimensionality of vectors.

Robustness of synergy activation

We quantified the robustness of ankle and subtalar volitional control based on the degree of decoupling between synergy activations for different target movements. The angle between two average vectors of synergy activations for two different target movements indicates the tolerance to variance in motor commands of two corresponding targeted movements. When the tolerance of motor commands is larger than the expected variance, it indicates that a subject can reliably produce distinguishable synergy activation for two discrete movements of interest. Therefore, the margin of synergy activation is given aswhere ϕ is the margin in synergy activations to have distinguishable patterns between targeted i and j discrete movements; T is the interval time of i discrete movement. Note that the first term in the right-hand side of the equation is the angle between two synergy activation average vectors (tolerance); the second and third terms are variability in motor commands corresponding to the two discrete movements. The thresholds of 0 were selected for synergy activation margin to determine robustly decoupled discrete movements.

Synergy space (U-space) and motor intent decoding (α-space)

We decoded the motor intents from arbitrary muscle activation patterns based on the extracted muscle synergy and synergy activation average vector of the 4 principal movements (PF, DF, IN, EV). The time-varying coefficients of the arbitrary muscle activation patterns were decomposed by revised NMF, fixing synergy vectors as the extracted muscle synergy from discrete movement trials during NMF iterations[37]. The same initialization and iteration protocols were utilized as those of muscle synergy extraction procedures for the rest of decomposition procedures. Given the revised NMF, the arbitrary muscle activation patterns can be reflected into the muscle synergy space of the discrete ankle and subtalar joint movements. This reflected time-varying coefficients at time t, , is further decoded into motor intents of ankle and subtalar movements by using the synergy activation vector of 4 discrete movements, , , , and , aswhere and indicate directions of desired movements in ankle and subtalar DoF, respectively. Given these definitions, and ranges from −1 to 1 and the -space consists of and served as a phase domain of motor control. The universal dimensionality of decomposed was unified as 3, which was the dimensionality of decomposed of all subjects in the BIO group. When fewer synergy vectors were identified previously by synergy extraction procedures, vectors were added in to match dimensionality of vectors.

2-DoF motor controllability

We quantified simultaneous multi-DoF motor controllability by investigating the transitions in directionality of motor intent during 10 cycles of the drawing-a-circle tasks. First, the was reflected into -space. To draw an ideal circle in joint space, the directionality of motor intent in both ankle and subtalar DoF needs to be changed simultaneously. This is equivalent to simultaneous changes in both and , resulting in diagonal trajectory in -space. Meanwhile, if the subject is only able to perform a single DoF motor control at a time, only changes in or is found at a time. This is equivalent to a horizontal or vertical trajectory in -space. Therefore, 2-DoF motor controllability was calculated by integrating diagonal components of trajectories within -space to evaluate the simultaneous multi-DoF motor controllability, orwhere is the number of trajectories in j-th quadrant of -space; and indicate the start and end time of ith trajectory in jth quadrant, respectively; and are a velocity vector in -space and unit diagonal vector, given as [], respectively. The trajectories in -space were evaluated by each quadrant independently to investigate multi-DoF motor controllability of different combinations of discrete movements. The average diagonal components of trajectories on each quadrant were computed and normalized by . Finally, the mean values of diagonal components of all quadrants were calculated. If no full trajectories were present in a quadrant, it was considered as zero diagonal component for that quadrant. Note that the second term on the right-hand side of the equation converges to zero and 2-DoF motor controllability becomes 1 when all the trajectories in all quadrants are composed by only diagonal components. Thus, given this definition, 2-DoF motor controllability ranges from 0 to 1.

Evaluation of spatiotemporal motor control under time constraints

Spatiotemporal motor control was evaluated from two metrics, motor control performance and economy of motion under increasing time constraints from 2.0 s to 1.5 s, 1 s, 0.8 s, and 0.5 s. For the visualization, an index of difficulty (ID) of speed-accuracy task for each time constraint (2.0–0.5 s) was calculated as the logarithm of the inverse of the time constraint and scaled to range from 0 to 1 (-), inspired by Fitts’ law[38-40]. Speed-accuracy tasks comprised 10 repetitions in each of discrete PF, DF, IN, EV movements in a random order for each of 5 time-interval settings. The motor control performance was analyzed based on the tracking errors between the decoded motor intent , and ideal targets of j discrete movement in -space, as

Ideal target movements in -space , , , and were defined as (0, −1), (0, 1), (−1, 0), and (1, 0), respectively. Given this definition, the motor control performance shows how one can maintain motor intent corresponding to given motor tasks. Economy of motion was computed by the ratio of effective synergy activation for targeted j discrete movements to total synergy activation, orwhere is effective synergy activation for targeted j discrete movements, determined as for PF and DF and for IN and EV. Given this definition, the economy of motion indicates the trajectory straightness of movements that were produced to achieve the target discrete movements.

Assessment of phantom limb perception capacity

A psychometric task was used to assess limb perception capacity of the phantom limbs across the full perceived ranges of motion (ROM) for DF-PF and IN-EV. The mirrored perceived phantom limb positions were measured by goniometry from the subjects’ BIO limb and were normalized by the ROM of the BIO limb () to allow comparison when plotted against the intended phantom limb positions () assessed from the EMG data. The intended limb position was assessed by the average value of and for each movement which show the both direction and amplitude of desired movements. To vary the phantom limb position while considering the range of motion of each joint, the subjects were guided to perform 25, 50, 75, and 100% PF and DF range of motion during separate, 40 randomized PF and DF trials, and 50 and 100% range of motion during 30 randomized IN and EV trials. The limb perception capacity was estimated based on the relationship between the intended phantom limb positions and the mirrored perceived phantom limb positions, determined as the mean value of the ranges between 5 and 95% of the psychometric functions. When a subject reported either zero phantom limb sensation or inconsistent phantom limb sensation to the intended limb positions, the limb perception capacity was determined as zero. When 5 or 95% of the psychometric function was not reached, the minimum or maximum value of the psychometric function was selected.

Assessment of phantom limb sensations

The reported phantom limb sensation scores focused on the vividness of their phantom limb sensations compared to actual sensations of their biologically intact limb during ankle, subtalar, and ankle/subtalar joint rotations. Subjects self-reported vividness of these phantom sensations on a scale of 0-to-10, where values of 0 and 10 respectively indicated no sensation or equivalent sensations for their phantom joint to their BIO limb. The full findings are given in Supplementary Table 1.

Statistics

No statistical methods were used to predetermine sample size a priori, but effect sizes were determined for the main outcomes using Cohen’s d values between 1.26-to-1.35 (Supplementary Table 2). Data collection and analysis were not performed blind to the conditions of the experiments. The unaffected limbs of all subjects served as the biologically intact limb population when appropriate, thus no separate non-amputated subjects were recruited. In all experiments, except when specifically noted, the order of the motor control tasks was randomized as described in the relevant Methods sections and in the Nature Research Reporting Summary. Six sensory-motor response variables were correlated with the degree of AMS for the pooled dataset of all 14 subjects’ residual limbs (CTL-1-7 and AMI-1-7). A non-parametric correlation, Kendall’s tau (), and P value were computed to address a positive association between each response variable and the degree of AMS. A first order exponential response curve was fitted to address a critical degree of AMS () that preserves 95% of each sensory-motor response variable. Jackknife mean ± standard deviation (s.d.) of and R2 values of fitted response curve for each response variable were reported. The normality of the motor control data was tested by a Shapiro-Wilk test at a significance level of α = 0.05. To consider within-subject limb differences (AMI:BIO-A and CTL:BIO-C), paired one-tailed t-tests were used at a significance level of α = 0.05, as all motor control data did not violate the data normality. To consider between-subject residual limb differences (AMI:CTL), unpaired two-tailed t-tests were used at a significance level of α = 0.05. Interactive effects between limb subgroups (AMI:CTL x affected:unaffected limb) were analyzed by 2-way ANOVA at a significance level of α = 0.05. The full statistics are reported in the Supplementary Table 2.