跳到正文
原文
Nature 健康护理主题·· 10 小时前AI 评分35

感觉重组与近端肌肉重组织构成单侧下肢截肢后姿势控制的基础

Sensory reweighting and proximal muscle reorganization underlie postural control after unilateral lower limb amputation

AI 导读

本研究首次量化了单侧下肢截肢患者在直立站立时的感觉重组与近端肌肉协同模式变化,揭示本体感觉减退时代偿性依赖视觉与前庭输入、并通过近端躯干和大腿肌肉协同完成姿势控制的神经运动机制。基于旋转平台的力学扰动结合倒立摆建模提取的本体感觉加权(Wp)和时间延迟(TD)指标,可为截肢者个性化康复训练和假肢设计提供定量生物标志物。

正文

Introduction

The number of amputation cases, including traumatic events, diabetes, vascular diseases, and cancer are on the rise1,2. Regardless of etiology, a lower limb amputation substantially impairs balance control. Individuals with a lower limb amputation are more prone to falls and less confident while walking, which limits their social activities3,4,5,6,7. Many individuals with an amputation also develop secondary musculoskeletal conditions, including back pain, osteoarthritis, and osteoporosis in the residual limb8.

A disrupted postural control leads to further functional limitations, such as reduced ankle joint control and increased weight-bearing on the intact leg9,10,11. These issues are reflected in altered center of pressure (CoP) patterns12 and increased compensatory lower leg and trunk muscle co-activation while standing13, with visual information modulating these adaptations.

With reduced somatosensory feedback, amputees overuse their remaining sensory channels (i.e., vision and vestibular inputs) and rely heavily on the sensations coming from the intact limb and the residual limb to control balance14,15. The loss of somatosensory neurons thus leads to sensory reweighting. Similarly, the loss of muscles and skeletal structures leads to motor reorganization, evidenced by altered muscle synergies during walking16,17 and during quiet standing tasks13.

Somatosensory, vestibular, and visual inputs are continuously integrated during the execution of two possible balance strategies. The ankle strategy is used for small and slower perturbations, in which the body rotates around the ankle and the corrective torque is mainly generated distally. The hip strategy is used for larger or faster disturbances or when distal torque is insufficient18,19, in which counter-rotation at the hip recruits proximal trunk and thigh musculature for a rapid corrective response. These strategies are not used separately but typically combined depending on the standing conditions. In individuals with two intact lower limbs, somatosensory input dominates balance control during normal conditions (~ 70% of sensory weighting)20. Yet, little is known about how much this sensory weighting shifts following a lower limb amputation.

Mechanical perturbations during upright standing can be used to study underlying postural control mechanisms (Fig. 1A). By measuring the center of mass (CoM) responses to small tilting perturbations of the support surface, the frequency-dependent gain and phase relationships can be quantified. Further, fitting the experimental transfer functions with an inverted pendulum model enables the extraction of key neural control parameters (Fig. 1B). Proprioceptive sensory weighting (\({W}_{p}\)) and time delays (\(TD\)) are model-derived measures that quantify how the nervous system processes and integrates balance-related sensory information. \({W}_{p}\) quantifies the relative reliance on peripheral joint afference versus visual and vestibular inputs, making it a sensitive index of sensory channel reorganization following loss of peripheral sensory information. \(TD\) is the time between stimulus and response onset. Excessive time delays lead the phase margin toward zero, threatening closed-loop stability20,21. While this approach has successfully characterized sensory reweighting in neuropathological populations22,23,24, the quantitative characterization of sensory reweighting parameters and their relationship to motor reorganization patterns remains limited in individuals with a lower limb amputation.

Fig. 1

Experimental design and analytical framework. (A) Schematic overview of the experimental setup showing participant stance on the rotating platform with 10 EMG electrode placement on lower limb and trunk muscles and the motion capture system. (B) Sensorimotor control system diagram. The model comprises active neural control (solid lines: PID controller with proprioceptive weighting Wp and time delay TD) and passive biomechanical components (dashed box: intrinsic stiffness Kpas and damping Bpas). During model optimization, proportional (Kp), derivative (Kd), proprioceptive weight (Wp), and time delay (TD) parameters were optimized, while Ki, Kpas, and Bpas were fixed. Body dynamics are modeled as a single-segment inverted pendulum. (C) Muscle Synergy extraction pipeline using non-negative matrix factorization (NMF) to decompose EMG signals into temporal activation patterns (H) and spatial muscle weightings (W).

While a single-segment inverted pendulum model describes whole-body dynamics, muscle synergy analysis further identifies the muscle coordination strategies underlying these dynamics. The brain generates coordinated motor output through muscle synergies, which are co-activation patterns of functionally related muscles acting as building blocks for movement execution25,26. Nonnegative Matrix Factorization (NMF) is a widely used method to characterize muscle synergies during standing balance tasks from electromyographic (EMG) data (Fig. 1C). It has been implemented in clinical populations, such as individuals with chronic ankle instability during standing tasks27, and has revealed altered muscle coordination during walking in amputee participants16,17. However, muscle synergy patterns during continuous standing tasks under perturbation in lower limb amputees remain uncharacterized.

Thus, quantifying sensory and motor adaptation mechanisms may inform the development of targeted interventions that include personalized motor and sensory training. These quantitative biomarkers could also guide new prosthetic design and provide objective measures for tracking rehabilitation progress in lower limb amputees and to deliver feedback to minimize instabilities.

In this work, we quantified for the first time sensory reweighting in lower limb amputees during upright standing. Unilateral transtibial amputees (TTA) and transfemoral amputees (TFA), as well as able-bodied individuals, performed standing tasks under surface perturbation while their CoM kinematics and lower body muscle EMG were recorded. We analyzed postural control biomechanically using transfer function analysis and inverted pendulum modeling of CoM responses, and on the neuromuscular level through muscle synergy extraction from EMG recordings. We hypothesized that lower limb amputees would present reduced proprioceptive weight (\({W}_{p})\) values associated with the level of their amputation (i.e., TFA < TTA < controls), and we expected \({W}_{p}\) to be vision dependent. Time delays (\(TD\)) were expected to be comparable across groups, as amputees are neurologically intact and thus the neural transmission should remain unchanged. However, because \(TD\) is an effective, lumped delay rather than a purely neural one, sensory integration demands and prosthetic mechanics would be expected to shift its estimated value. We further expected the neural control parameters (\({K}_{p}\) and \({K}_{d}\), Fig. 1B) to differ across groups due to altered ankle and knee mechanics, with amputees showing increased effective stiffness resulting from the mechanical properties of the prosthetic foot and compensatory balance strategies. We hypothesized that neuromuscular adaptations would manifest as altered trunk and ankle synergies, reflecting compensatory balance strategies, with a shift of the main torque generation from the ankle to the hip joint. We further anticipated that perturbation amplitude and visual condition would each modulate the measured responses.

Results

Compared with able-bodied individuals and transtibial amputees, transfemoral amputees exhibited impaired sensory processing during balance control, reflected by frequency-dependent phase shifts and reduced coherence, when vision was unavailable and lower proprioceptive weighting with eyes open. Despite these deficits, gain responses and relative synergy activation were preserved across groups. Muscle synergy analysis revealed altered coordination patterns characterized by condition-specific reductions in soleus contributions, increased proximal leg involvement and a redistribution of trunk muscles between synergies, suggesting compensatory neuromuscular adaptations that maintain overall postural performance.

Transfemoral amputees show frequency-specific phase differences, most pronounced without visual feedback

Group differences in phase were frequency-dependent and modulated by visual condition (Fig. 2, visual × group × frequency: \({\chi}^{2}=31.45,\:p=0.0495)\). With eyes closed, transfemoral amputees (TFA) showed significantly increased phase lag compared to controls at 0.4 and 0.55 Hz (\(p\:<\:0.05\)). Transtibial amputees did not differ from controls at any frequency but showed differences from TFA with eyes closed (TFA: increased phase lag at 0.55 Hz and decreased phase lag at 1.1 and 1.75 Hz; \(p\:<\:0.05\)). With eyes open only the highest frequency (2.2 Hz) showed TFA-control difference, with reduced phase lag in TFA (\(p\:=0.015\); Supplementary Tables 1–3).

Fig. 2

Phase lag response of the transfer function analysis. Symbols represent group means and error bars are the standard error of the mean. CTRL: Control group (circles), TTA: transtibial amputee group (triangles), TFA: transfemoral amputee group (squares).

Gain responses are comparable across groups

Gain responses showed a clear resonance peak but no significant group effects (Fig. 3, frequency main effects: \({\chi}^{2}=422.46,\:p<0.001)\). Unlike phase responses, gain did not differ between groups (main effects and all group interaction effects non-significant), indicating that transfemoral and transtibial amputees preserved normal gain magnitudes across conditions. Gain was modulated by visual condition and stimulus amplitude (eyes-open < eyes-closed, particularly at mid-range frequencies (\({\chi}^{2}=30.50,\:p<0.001)\) and stimulus amplitude (0.5° > 1.0°, especially near the resonance peak; amplitude × frequency: \({\chi}^{2}=41.64,\:p<0.001\); Supplementary Table 4).

Fig. 3

Gain responses of the transfer function. Symbols represent group means and error bars are the standard error of the mean. CTRL: Control group (circles), TTA: transtibial amputee group (triangles), TFA: transfemoral amputee group (squares).

Transfemoral amputees exhibit reduced coherence without visual feedback

Coherence decreased systematically with increased frequency \(({\chi}^{2}=682.97,\:p<0.001\), Fig. 4). A significant group main effect \(({\chi}^{2}=6.91,\:p=0.032\)) was modulated by visual condition and stimulus amplitude (visual × amplitude × group \({\chi}^{2}=6.28,\:p=0.043)\). Specifically, transfemoral amputees showed reduced coherence compared to controls and transtibial amputees (\(p\:\le\:\:0.004\)) with eyes closed at 1.0° showing less input-output coupling under challenging conditions. Unlike phase response, coherence reduction did not show frequency-specific group differences (group × frequency: \({\chi}^{2}=21.78,\:p=0.352)\), indicating that reduced coherence in transfemoral amputees is relatively consistent across the tested frequency ranges (Supplementary Tables 5–8).

Fig. 4

Coherence of the transfer function. Symbols represent group means and error bars are the standard error of the mean. CTRL: Control group (circles), TTA: transtibial amputee group (triangles), TFA: transfemoral amputee group (squares).

Model parameters

Transfemoral amputees show reduced proprioceptive weighting, most pronounced with eyes open

Proprioceptive weighting (\({W}_{p}\)) showed significant group effects (\({\chi}^{2}=38.21,\:p<0.001)\) with transfemoral amputees showing lower \({W}_{p}\) values compared to both controls and transtibial amputees, while transtibial amputees did not differ from controls (Fig. 5). This proprioceptive deficit in transfemoral amputees was modulated by visual condition (visual × group: \({\chi}^{2}=11.19,\:p=0.004)\). With eyes open, transfemoral amputees showed a substantial reduction in \({W}_{p}\). With eyes closed, the deficit persisted but was less pronounced, indicating that the removal of visual input led to greater reliance on remaining proprioceptive signals. Similarly, rotation amplitude modulated the group effect (amplitude × group: \({\chi}^{2}=8.63,\:p=0.013)\). With larger deficits at smaller perturbations (0.5°: ctrl-TFA, TTA-TFA) compared to larger ones (only 1.0° ctrl-TFA but not TTA-TFA; Supplementary Tables 9–14).

Amputation did not affect Kp, Kd and TD after correction.

For proportional gain (\({K}_{p}\)), which quantifies the corrective torque generated responding to CoM sway angle, showed a non-significant group effect (\({\chi}^{2}=5.32,\:p=0.070)\). \({K}_{p}\) seems to be stable across conditions (Fig. 5). Derivative gain (\({K}_{d}\)), which quantifies the corrective torque generated in response to body sway velocity, showed a significant group effect (\({\chi}^{2}=8.00,\:p=0.018)\). However, post-hoc comparisons revealed no pairwise differences reaching significance. Time delay (\(TD\)), representing the time between stimulus and response, showed a group effect (\({\chi}^{2}=6.32,\:p=0.043)\). However, pairwise comparisons did not survive correction for multiple comparisons (Supplementary Tables 15–19).

Fig. 5

Postural control model parameters estimated from CoM transfer functions across visual conditions and rotational amplitudes. Proprioceptive weight (Wp) quantifies reliance on ankle proprioception for postural control (0 = no reliance, 1 = complete reliance); time delay (TD) represents time between stimulus and response. Proportional (Kp) and derivative (Kd) gains are normalized by gravitational torque of the individual (mgh). CTRL: Control group, TTA: transtibial amputee group, TFA: transfemoral amputee group. Significant group differences in Wp are reported in Supplementary Tables 9–14.

Muscle synergy analysis

Five muscle synergies were identified as the optimal rank across groups and conditions. Each synergy was dominated by one or two muscles that defined its functional role, with contributing muscles showing group-dependent patterns. Trunk stabilizer muscles (ES and EO) participated across multiple synergies. Group differences emerged primarily in trunk stabilizers and select dominant muscles, while other dominant muscles remained preserved, with differences observed during quiet standing as well as during perturbed conditions.

Amputees increase bilateral trunk stabilization during quiet standing

Synergy 1 was characterized by BF muscle and elevated trunk muscles (ES, Figs. 6 and 7; Supplementary Tables 20–23). Transtibial amputees recruited less BF than controls \((p=0.007)\) and transfemoral amputees \((p=0.017)\) during eyes-open quiet standing (group main effect: \( \chi ^{2} = 10.07,\:p = 0.006 \)). Transtibial amputees recruited more ES muscle of the dominant side than controls during quiet standing with eyes open and closed (\(p<0.001;\)group\(:\:{\chi}^{2}=16.15,\:p<0.001;\)group × perturbation\(:\:{\chi}^{2}=15.88,\:p=0.003)\) as did transfemoral amputees with eyes closed (\(p=0.010\)). In TTA, this pattern extended to ES muscle on the non-dominant side (group: \(\chi^{2}=28.41,\:p<0.001\); CTRL-TTA for both visual conditions: \(p<0.001\)), reflecting a compensatory bilateral trunk stabilization. In contrast, controls recruited more EO on the non-dominant side than TTA (\(p=0.034\)) and TFA (\(p<0.001\)) during eyes-open quiet standing (group: \({\chi}^{2}=19.62,\:p<0.001\)).

Transtibial amputees show altered distal coordination with compensatory trunk stabilization

Synergy 2 was primarily characterized by distal leg muscles, including the GL and the SOL muscles (Figs. 6 and 7; Supplementary Tables 24–27). At maximum perturbation with eyes open, transtibial amputees recruited less SOL than controls (\(p\:=\:0.002\); three-way: \(\chi^{2}=\:47.39,\:p\:<\:0.001\)) and more ES than TFA on both sides (dominant: \(p=0.012\); non-dominant: \(p=0.034\)) and than controls on the non-dominant side (\(\text{p}=0.006\); three-way: dominant \(\chi^{2}=14.63,\:\text{p}=0.006\); non-dominant \(\chi^{2}=29.12,\:p<0.001)\). With eyes closed, this pattern reversed for SOL (TTA > controls, \(p=\:0.003\)) and ES (controls > TTA, dominant: \(p=0.007\); non-dominant: \(p<\:0.001\)). In the same eyes open condition, TTA recruited less GL than TFA (\(p<0.001\)) and controls (\(p=0.008\); three-way: \({\chi}^{2}=18.68,p<0.001)\). During quiet standing with eyes closed, both amputee groups recruited less SOL than controls (TFA: \(p=0.014\); TTA: \(p=0.049\)), and with eyes open TFA recruited less SOL than TTA (\(p=0.005\)).

Gastrocnemius recruitment is preserved, while transtibial amputees reduce soleus recruitment under challenging conditions

Synergy 3 was primarily characterized by ankle extensors, including the GM and SOL muscles (Figs. 6, 7; Supplementary Tables 28–31). GM recruitment did not differ between groups (group: \({\chi}^{2}=\:4.10,\:p=0.129\); no significant group interactions). SOL showed a three-way interaction (\(\chi^{2}\:=\:37.83,\:p<0.001\)) without group differences during quiet standing. At 1.0° with eyes closed, TTA recruited less SOL than controls and TFA (\(p\:<\:0.001\)).

Amputees weight synergy 4 toward the rectus femoris and away from the erector spinae

Synergy 4 was primarily characterized by the RF and EO muscles on the dominant side. Marked group differences were observed for the RF muscle (three-way interaction: \(\chi^{2}=57.33,\:p<0.001\)). Both amputee groups recruited more RF than controls at 0.5° with eyes open and in both visual conditions during quiet standing (\(p<0.001\)). At 0.5° with eyes closed, TFA recruited less RF than controls (\(p<0.001\)) and TTA (\(p=0.007\); Figs. 6 and 7; Supplementary Tables 32–35). EO muscles on the dominant side did not differ between groups, whereas amputees recruited more EO on the non-dominant side than controls during quiet standing (TFA eyes open: \(p=0.006\), eyes closed: \(p=0.008\); TTA eyes closed: \(p=0.014\); group: \({\chi}^{2}=9.43,\:p=0.009)\). Controls recruited more ES than amputees on both sides during quiet standing and at 0.5° with eyes open (non-dominant:\(p\le\:0.031;\) group: \({\chi}^{2}=24.62,\:p<0.001\); dominant: \(p\le\:0.050\) except CTRL-TTA in eyes-open quiet standing; three way\(:\:{\chi}^{2}=12.47,\:p=0.014).\:\)At 1.0°, there were no group differences in synergy 4.

Ankle dorsiflexor patterns are preserved while transfemoral amputees reduce knee extensor recruitment

Synergy 5 was primarily characterized by ankle dorsiflexors (TA) and knee extensors (RF). There were no group differences for the TA muscle (group main effect: \(\chi^{2}=2.15,\:p=0.342\)), indicating preserved ankle dorsiflexor coordination patterns across groups (Supplementary Tables 36–39). In contrast, there was a significant group main effect for the RF muscle (\(\chi^{2}=12.11,\:p=0.002\)) and a three-way interaction (\({\chi}^{2}=13.33,\:p=0.010)\) with transfemoral amputees recruiting this muscle less than controls during eyes-open quiet standing (\(p=0.001\); Figs. 6 and 7) and at 0.5° (\(p=0.001\)) and less than transtibial amputees during eyes-open quiet standing (\(p=0.046\)). Among the secondary contributors, controls recruited more ES on the dominant side than TFA at 1.0° with eyes open (\(p=0.039;\:\)three-way: \(\chi^{2}=14.80,\:p=0.005\)).

H matrix did not show major group differences, indicating a preserved use of each synergy

Mean relative synergy activation (H-matrix coefficients) showed no group main effects or group interactions for any synergy (Supplementary Fig. 1, Supplementary Tables 40–44), indicating preserved recruitment of synergies across groups.

Fig. 6

Nonnegative matrix factorization muscle synergy weights for the eyes open condition. Shown are group means with standard errors; the grey dashed line is the threshold activity. CTRL: Control group, TTA: transtibial amputee group, TFA: transfemoral amputee group, SOL: Soleus, TA: Tibialis anterior, GM: Gastrocnemius medialis, GL: Gastrocnemius lateralis, RF: Rectus femoris, BF: Biceps femoris, EO: External obliques (dominant and non-dominant), ES: Erector spinae (dominant and non-dominant).

Fig. 7

Muscle Synergies weights for the eyes-closed condition. Group means with standard errors are shown; the grey dashed line indicates the activity threshold. CTRL: Control group, TTA: transtibial amputee group, TFA: transfemoral amputee group, SOL: Soleus, TA: Tibialis anterior, GM: Gastrocnemius medialis, GL: Gastrocnemius lateralis, RF: Rectus femoris, BF: Biceps femoris, EO: External obliques (dominant and non-dominant), ES: Erector spinae (dominant and non-dominant).

Discussion

In this work, we used perturbed standing experiments to model postural control in lower limb amputees. The model quantified sensory weightings for proprioception, along with neural controller parameters, and a transfer function was used to examine gain and phase relationships between perturbations and CoM responses. In parallel, lower body muscle synergies were extracted to study possible neuromuscular mechanisms underlying the compensatory strategies found in the postural model.

As we hypothesized, TFA exhibited a markedly reduced proprioceptive contribution to postural control. Contrary to the graded relationship we predicted across amputation levels, transtibial amputees did not differ from controls as sensory reweighting in the model emerged at the transfemoral, not the transtibial level. Neural controller gains did not behave as predicted as the motor gain did not differ and the time delays were comparable across groups. The predicted muscle-synergy reorganization was confirmed. The deficit in TFA was therefore confined to the sensory and temporal domains while the magnitude of motor output was preserved. These findings indicate that reduced proprioceptive weighting is accompanied by compensatory proximal muscle reorganization preserving response amplitude but constraining temporal coordination.

TFA patients maintain response magnitude but show frequency- and vision-dependent phase shifts

TFA participants presented frequency-specific phase differences, while gain magnitudes remained preserved. With eyes closed, TFA patients had increased phase lag compared to controls at low-to-mid frequencies (0.4–0.55 Hz), a difference not present when vision was available, suggesting that visual feedback compensates for the underlying temporal control disruptions28. At higher frequencies the direction is reversed with TFA showing reduced phase lag compared to TTA with eyes closed (1.1 and 1.75 Hz) and compared to controls with eyes open (2.2 Hz). The reduced phase lag at high frequencies likely reflects a shift toward a stiffer, proximally anchored strategy: With limited rapid torque from the passive prosthetic ankle and TFA relying more on trunk and hip co-contraction, which is consistent with the synergy findings, than on distal ankle modulation, reducing the frequency-dependent phase lag29,30. Gain magnitudes were preserved after amputation, consistent with a dissociation between amplitude control and temporal execution. This “what versus when” distinction contrasts the magnitude of the corrective torque (preserved gain), with the timing relative to the perturbation (altered phase). It suggests that compensatory strategies successfully generate sufficient corrective torque but reorganize its temporal structure rather than slowing the response31. This altered timing without change in response amplitude, suggest that compensatory strategies avoid increased controller effort, which would otherwise risk oscillatory overshoot21,32.

TFA patients exhibit less consistent coupling between perturbation responses

TFA participants showed reduced coherence, indicating that the coupling between the perturbation input and the CoM adjustments was less consistent, as higher coherence values reflect more reliable and reproducible responses. This effect was highlighted with increased challenging conditions (eyes closed and 1.0° rotation) in TFA compared to controls and transtibial amputees. The reduced coherence in TFA was consistent across frequencies, with no group-by-frequency interaction. This indicates that TFA show less consistent sensorimotor responses to repeated perturbation regardless of the perturbation frequency, possibly caused by increased sensorimotor noise or prosthesis related biomechanical nonlinearities9,33. A reduction of coherence values was observed only with eyes closed at 1.0°, reinforcing that visual feedback serves as a compensatory role in stabilizing postural control when proprioceptive reliability is compromised28. From a clinical perspective, reduced coherence values may represent increased fall risk even when average performance (i.e., response gain) appears normal, as inconsistent responses mean that environmental disturbances in daily life can produce unpredictable postural outcomes34.

Evidence of reduced proprioceptive reliance under visual feedback conditions in TFA patients

The proprioceptive weight parameter (\({W}_{p}\)) quantifies the nervous system’s relative reliance on proprioceptive versus visual and vestibular feedback for postural control. Transfemoral amputees showed markedly reduced \({W}_{p}\) values compared to controls during the eyes open condition. Transtibial amputees, by contrast, were indistinguishable from controls, so the graded reduction across amputation levels did not hold and indicate a threshold-like effect specific to a certain amputation level. This reduction was smaller with eyes closed (Fig. 5). These results suggest active sensory reweighting20, where reliance on the proprioceptive feedback is downweighed when visual information is available35, probably due to major loss of ankle and knee joint receptors and inferior residual limb feedback31,36. With eyes closed, TFA still showed lower \({W}_{p}\) than controls even under forced reliance on proprioception37. The smaller deficit at higher perturbation amplitudes suggests that stronger mechanical signals partially overcome sensory noise, consistent with signal detection principles, in which increased stimulus magnitude improves detection thresholds38. This channel-switching represents successful sensory-level adaptation, distinct from the motor-level compensations we observe in biomechanical strategies and muscle coordination patterns. The surface-tilt stimulus perturbs the proprioceptive channel directly and \({W}_{p}\) is identified from the shape of frequency response. It describes how strongly the body follows the surface and is therefore less sensitive to the ankle-versus-hip torque ambiguity that confounds the absolute gains \({K}_{p}\) and \({K}_{d}\).

Preserved motor parameters suggest sensory input and not motor output deficits

Contrary to our hypothesis that prosthetic mechanics and compensatory stiffening would elevate \({K}_{p}\) and \({K}_{d}\), neither gain nor time delay showed group differences surviving correction. This preservation is central to the interpretation as it localizes the deficit to sensory and temporal control rather than to the capacity to generate corrective torque against the gravitational torque. The single-segment inverted pendulum model cannot differentiate the torque generated by the ankle from the one generated by the hip joint. The maintained system-level stiffness and damping are likely due to hip-generated compensatory torque rather than preserved ankle control15,39. This interpretation is supported by our muscle synergy findings showing elevated proximal activation alongside condition-specific reductions in soleus recruitment. It also exemplifies motor redundancy, as multiple biomechanical solutions achieve the same control parameters, resulting in functional equivalence through non-equivalent movement patterns. Preserved \(TD\) values after an amputation also suggest that amputee participants have normal response times to the stimulus. The observed phase differences therefore may stem from altered sensory weighting and integration times (i.e., visual and vestibular pathways have different temporal characteristics20,40 than proprioceptive feedback) and biomechanical constraints of multi-segment coordination, and not from slowed neural processing (this is evidenced by upper and lower body transfer function phase responses, Supplementary Fig. 2). These findings suggest that the primary challenge within this experimental context for TFA is degraded sensory input rather than the magnitude of corrective motor output. Because muscle morphology and strength are changed after transfemoral amputation41, the motor output may be limited in more demanding tasks.

Altered muscle recruitment within preserved coordination structure reveals proximal compensation strategy

Muscle synergy analysis revealed how amputees achieve preserved motor control gains and synergy recruitment through reorganized muscle contributions, in line with previous work on walking16. Mean relative synergy activation showed no statistically significant group differences (Supplementary Fig. 1). Averaged over the trial, amputees thus recruited synergies with similar relative intensity despite reorganized muscle contributions. This may reflect preserved synergy recruitment, but trial-averaged activation can also mask group differences in the timing of recruitment. Because the perturbation followed a pseudorandom sequence signal, future work could relate activation time series of each synergy directly to platform movement, for example with the transfer function approach applied here to body sway, to test whether amputees recruit synergies with altered timing, as the phase differences in TFA would suggest. However, the weights within the modules suggested neuromuscular adaptations in amputees, with reduced distal muscle and increased proximal leg recruitment and a redistribution of the trunk muscles between synergies. This was the case for the soleus muscle, which both amputee groups recruited less during eyes closed quiet standing (synergy 2) and TTA recruited less under the largest perturbation (synergies 2 with eyes open and synergy 3 with eyes closed), whereas gastrocnemius recruitment in synergy 3 was preserved. Similarly, the distal muscles in synergy 2 displayed different co-activation patterns (i.e., at the largest perturbation with eyes open, TTA recruited less GL and SOL but more ES). Proximal compensation was reflected in an increased activation of the muscles of the proximal leg (RF in synergy 4), in TTA and TFA, although TFA recruited less RF in some conditions (synergies 4 and 5). The trunk muscles were redistributed rather than generally increased: amputees weighted the ES more in synergy 1, together with the BF, and TTA in synergy 2 at the largest perturbation with eyes open, whereas controls weighted the ES more in synergy 4 during quiet standing and small perturbations. The kinematics support the same reading: the upper and lower body did not respond in a single rigid segment with differing gains and phases across frequency ranges (Supplementary Fig. 2). Because the body is multi-segmental, hip and trunk-generated torque is folded by the single-segment model into apparent ankle gain. Together with proximal muscle reorganization, this is why the preserved neural controller parameters in amputees most plausibly reflect hip compensation rather than maintained ankle control. This proximal shift explains the preserved \({K}_{p}\) and \({K}_{d}\) values in the model, as torque generated by the hip joint maintains whole-body postural stiffness, which the single-segment model can only attribute to the ankle control mechanism15,39. Importantly, these changes emerged during quiet standing (synergies 1, 2, 4 and 5), suggesting a fundamental reorganization that is not limited to perturbation-specific reactions. Vision and perturbation amplitude modulated recruitment magnitudes and vision reversed some of them (SOL and ES in synergy 2, RF in synergy 4), indicating vision-dependent rather than fixed compensation. Not all primary muscles showed group differences (e.g., GM in synergy 3 and TA in synergy 5, where primary are the muscles that dominate the synergy), indicating a selective rather than a general neuromuscular adaptation. This compensatory mechanism at the neuromuscular level explains the dissociation between sensory loss and preserved neural controller parameters (\({K}_{p}\), \({K}_{d}\)). Preserved gain was achieved through altered neuromuscular patterns, but temporal deficits arose from multi-segment coordination constraints inherent to proximal control strategies.

The muscle synergy findings directly complement the single-segment model’s limitations and reveal the mechanisms underlying preserved control parameters. Because proximal reorganization accompanies reduced SOL activity, the single-segment model attributes hip-generated torque to the ankle, so the two cannot be separated. This illustrates motor redundancy42, in which multiple biomechanical implementations achieve equivalent motor control despite altered neuromuscular coordination. At the sensory level, the reduced proprioceptive weight reflects sensory redundancy20, in which the nervous system maintains stability through redistribution of available feedback channels. Together, these adaptations allow amputees to preserve magnitude control while temporal precision remains compromised by biomechanical constraints. Future work should employ multi-segment models to explicitly represent ankle and hip contributions, using synergy patterns to constrain parameter estimation.

Limitations

The findings of this study must be considered within its limitations. First, the single-segment inverted pendulum model cannot separate the contribution of hip and trunk to postural control, which is most limiting for the hip-dominant control that our challenging standing conditions tend to produce. Another limitation of this study is the relatively low sample size of the patient group (7 TFA and 7 TTA) and that perturbation trials were only repeated once. Therefore, weaker effects may be masked. Moreover, the findings may not generalize to individuals early after amputation or with less mobility than the sample recorded in this study as they showed good mobility (Table 1). None of the amputee participants had a vascular or diabetic cause of amputation or a neurological condition affecting balance. The reported deficits are therefore not explained by neuropathy of the intact limb and likely underestimate those of less mobile or dysvascular amputees. Body mass and height were accounted for in the model (moment of inertia, normalization of \({K}_{d}\:\) and \({K}_{p}\) by \(mgh\)), whereas age, time since amputation and prosthetic components varied within groups. The muscle synergy analysis is not without limitations. While NMF is a widely adopted method to quantify muscle synergies, its solutions are not unique, which is partially mitigated by the consistent synergy structure recovered across participants, conditions, and groups. Moreover, the synergy alignment procedure is critical. As NMF first finds the optimal number of ranks and then fixes all ranks across all trials, this may result in either overfitting or underfitting and may also miss important adaptations.

Materials and methods

Experimental design

This study aimed to quantify sensory reweighting and muscle synergies underlying postural control in individuals with unilateral limb amputation. Three groups were compared: transtibial amputees (TTA), transfemoral amputees (TFA), and able-bodied controls (CTRL) approximately matched for age, height, and weight. Primary outcomes were proprioceptive weighting (Wp), and neural control parameters derived from transfer function modelling. Muscle weights give insight on muscle synergy adaptations reflecting neuromuscular coordination. Analytical approaches were specified prior to data collection and inspection of results. Each participant completed perturbed and quiet standing trials under two visual conditions (eyes open and eyes closed) at two perturbation amplitudes (0.5° and 1°). Trial order was partially randomized with first trial fixed for familiarization. Two complementary analytical approaches were applied: transfer function analysis to estimate model parameters, and non-negative matrix factorization (NMF) of surface EMG to extract muscle synergies.

Participants

Seven individuals with a unilateral transtibial amputation (age: 57.00 ± 8.96 years; height: 179.20 ± 5.07 cm; weight: 76.94 ± 9.42 kg; 6 males, 1 female) and seven individuals with a unilateral transfemoral amputation (age: 51.43 ± 17.74 years; height: 186.43 ± 7.83 cm; weight: 76.29 ± 11.08 kg; 7 males) participated in this study. A control group of 21 able-bodied individuals also participated, matching the amputee groups in age, height, and weight (Table 1). Because kinematic recordings of two controls (controls 21–22) and EMG recordings of one other control (control 1) were missing or unusable, the transfer function and model analysis included 19 controls (age: 53.79 ± 8.39 years; height: 176.09 ± 10.05 cm; weight: 72.89 ± 13.30 kg; 10 males, 9 females) and the synergy analysis included 20 controls (age: 54.85 ± 9.03 years; height: 177.18 ± 10.04 cm; weight: 74.95 ± 14.81 kg; 12 males, 8 females). Eighteen controls were included in both analyses and all amputee participants were included in both analyses. All participants had normal or corrected-to-normal vision. One additional control participant was excluded because of impaired vision in one eye. Participants with any other additional neurological, orthopedic, or cardiovascular conditions affecting balance were excluded. The experiment was approved by the University of Freiburg’s ethics committee (ethical approval numbers 21-1245 and 24-1444_1-S1), and all participants provided informed consent before participating. All methods were performed in accordance with relevant guidelines and regulations, and in accordance with the Declaration of Helsinki. Amputee participants were recruited via a prosthetist (Sanitätshaus Pfänder, Freiburg) or a collaborating hospital physician among clients/patients that are able to stand unassisted and express unassisted mobility (see Table 1, Mobility Grades).

Table 1 Demographics of the participant groups. TFA: Transfemoral Amputees, TTA: Transtibial Amputees, Ctrl: Controls, height and weight were measured with prosthesis and shoes. Control 15 was excluded from analysis, Control 1 was only used in the postural control model while controls 21–22 were only used in the EMG analysis due to missing kinematic data. “–” is used when information was not available/applicable. Mobility grade: 1: Indoor walker, 2: Restricted outdoor walker, 3: Unrestricted outdoor walker, 4: Unrestricted outdoor walker with particularly high demands (e.g., performing sports or heavy labor43).

Full size table

Data collection

Participants stood with their feet shoulder-width apart and their arms crossed at chest level on a Kistler force platform (type 9286, Winterthur, Switzerland) integrated into a motorized platform that rotated in the anterior-posterior (AP) direction (Fig. 1A). Surface rotation was used to directly perturb the ankle proprioceptive channel that can be parameterized by the inverted-pendulum model, making it the established stimulus for quantifying proprioceptive weighting. The movement stimulus followed a pseudorandom ternary sequence (PRTS20) with amplitudes of 0.5° and 1°, spanning a wide frequency range of 0.05–2.50 Hz and keeping responses in the linear small-angle range. The PRTS stimulus design ensured unpredictable perturbations that while eliciting responses at relevant frequencies and avoid pushing participants into hip/stepping strategies violating model assumptions. Participants completed 8 trials in total as follows. Four PRTS trials (2 amplitudes × 2 visual conditions) and four quiet-standing trials (2 eyes open and 2 eyes closed). The first trial (eyes open, no perturbation) was fixed for familiarization, and the remaining trials were randomized. Trial duration was 65 s, with perturbation onset at 2.5 s. Participants rested for 30 s between trials.

Motion capture data were recorded using a markerless motion capture system sampled at 100 Hz (The Captury GmbH, Saarbrücken, Germany). Muscle activity was recorded from 10 muscles using wireless Trigno Avanti sensors (Delsys Inc., Natick, MA, USA) at a sampling rate of 1259 Hz and a bandwidth of 20–450 Hz. Skin preparation included wiping sensor sites with alcohol wipes to remove excess oil and dead skin cells. EMG sensor placement followed the SENIAM standard44. The external oblique (EO) and erector spinae (ES) muscles were measured on both sides (dominant/intact and non-dominant/prosthetic) as they were accessible on both sides for all participants and unilateral amputation loads the trunk asymmetrically. The rectus femoris (RF), biceps femoris (BF), tibialis anterior (TA), gastrocnemius medialis (GM), gastrocnemius lateralis (GL), and soleus (SOL) muscles were measured only on the dominant/intact side. Dominant leg was determined by questionnaire: Which leg is used for stepping on a stair, which leg is preferred to use to kick a ball. Platform movements were controlled by a single computer synchronized via timestamps. The Captury and the Delsys EMG device were hardware-triggered.

Data processing of kinematic data and transfer function analysis

Anthropometric segment heights were measured for each participant. Three-dimensional positions of the head, shoulder (upper body), and hip (lower body) segments were tracked and recorded during each trial using the Captury system. The angles between body segments were then calculated in the anterior-posterior (AP) plane using the small-angle approximation. For the lower body, the hip position was normalized by hip height. For the upper body, the relative position between the shoulder and hip was normalized by the trunk segment height (shoulder-hip distance). The CoM position was estimated as the hip position plus one-quarter of the trunk segment displacement, yielding a CoM location of approximately 57% of body height in upright standing45.

A transfer function \({H}_{exp}\left(f\right)\) was then created to quantify the frequency-dependent relationship between the platform rotation \(({\theta\:}_{Platform}\left(f\right)\), input in Eq. 1) and the body segment angular responses \(({\theta\:}_{CoM}\left(f\right)\), output in Eq. 1)20.

$$ H_{{exp}} \left( f \right) = ~\frac{{\theta _{{COM}} \left( f \right)}}{{\theta _{{Platform}} \left( f \right)}} $$

(1)

During each PRTS trial, the stimulus was repeated three times, and transfer functions were computed separately for each cycle by calculating the discrete Fourier transform of both the stimulus and response signals (sampled at 100 Hz). The fundamental frequency (determined by the cycle length) was 0.0501 Hz for rotation trials (cycle length = 1997 samples at 100 Hz), yielding 50 frequency components as integer multiples of this fundamental frequency. The cross-spectral density between the stimulus and the response was computed and averaged over the last two cycles (the first cycle was discarded due to transients). Spectral estimates were averaged within overlapping frequency bands to produce 11 final frequency points (0.05–2.2 Hz), reducing variance while maintaining frequency resolution at critical frequencies35.

Several transfer function metrics were evaluated, including the gain (magnitude of \(H\left(f\right)\)), calculated as the cross-spectral density divided by the input power spectral density; the phase, defined as the angle of the complex transfer function \(H\left(f\right)\); and coherence, quantifying input-output coupling Eq. 2.

$$ \gamma ^{2} = \frac{{\left| {S_{{xy}} \left( f \right)} \right|^{2} }}{{S_{{xx}} \left( f \right)S_{{yy}} \left( f \right)}} $$

(2)

In Eq. 2, \({S}_{xy}\) is the cross-spectral density, \({S}_{xx}\) is the input power spectral density, and \({S}_{yy}\) is the output power spectral density. Coherence values range from 0 (no coupling) to 1 (perfect linear coupling). The transfer function was only computed for trials with a rotation stimulus.

Inverted pendulum model and parameter estimation

A single-segment inverted pendulum model was used to estimate neural control parameters from the experimental transfer functions20,23,24,35. The model represents body rotation about the ankle joint, relating platform rotation (input) to CoM angle (output) through biomechanical dynamics and neural control mechanisms (Fig. 1B). Biomechanical components were computed from participant-specific anthropometrics. The moment of inertia was computed using Eq. 3.

$$\begin{array}{c}J=m\cdot\:{\left(k\cdot\:H\right)}^{2}\end{array}$$

(3)

In Eq. 3, \(m\) is body mass (in kilograms), \(H\) is body height (in meters), and \(k\) = 0.55 is the coefficient of the radius of gyration derived by combining body segments, approximating whole-body rotation about the ankle45 and yielding individual moments of inertia appropriate for each participant’s body mass and height. Body dynamics were represented using Eq. 4.

$$\begin{array}{c}{B}_{i}\left(\omega\:\right)=J\cdot\:{s}^{2}-mgh\end{array}$$

(4)

In Eq. 4, \(s=j\omega\:\) is the Laplace variable (\(\omega\:=2\pi\:f)\). The term \({B}_{i}\) represents the body dynamics capturing inertial resistance, while \(mgh\) is the gravitational (destabilizing) torque, with \(h\) as the height of the CoM above the platform, estimated as \(h={h\:}_{hip}+\frac{1}{4}({h\:}_{shoulder}-{h\:}_{hip})\) and \(g\:\)as the gravitational acceleration.

Neural control was implemented as a proportional-integral-derivative (PID) controller with a time delay (Eq. 5).

$$\begin{array}{c}{PID}_{TD}\left(\omega\:\right)=\left({K}_{d}\cdot\:{s}^{2}+{K}_{p}\cdot\:s+{K}_{i}\right)\cdot\:TD\left(\omega\:\right)\end{array}$$

(5)

In Eq. 5, \({K}_{p}\), \({K}_{d}\), and \({K}_{i}\) are the proportional, derivative, and integral gains representing neural responses to position, velocity, and integrated position errors. The formulation follows the transfer function convention of Peterka (2002), in which the controller and plant dynamics are expressed as combined polynomial in s (dividing Eq. 5 by s gives \(K_{p} + ~K_{d} \cdot s~ + ~K_{i} /s \)). The time delay was modeled as in Eq. 6.

$$\begin{array}{c}TD\left(\omega\:\right)=\text{cos}\left(\omega\:\cdot\:{T}_{d}\right)-j\cdot\:\text{sin}\left(\omega\:\cdot\:{T}_{d}\right)\end{array}$$

(6)

In Eq. 6, \(TD\) represents effective time delays, the time from stimulus to a response (in seconds). Sensory weighting (\({W}_{p},\) range 0–1) quantified the relative contribution of proprioceptive feedback to postural control. Passive mechanical properties (Eq. 7) were also included as passive stiffness (\({K}_{pas})\) and damping (\({B}_{pas})\).

$$\begin{array}{c}Passive\left(\omega\:\right)={K}_{pas}\cdot\:s+{B}_{pas}\cdot\:{s}^{2}\end{array}$$

(7)

The complete model transfer function from platform rotation to CoM angle is shown in Eq. 8.

$$\begin{array}{c}H\left(\omega\:\right)=\frac{{W}_{p}\cdot\:{PID}_{TD}+{K}_{pas}\cdot\:s+{B}_{pas}\cdot\:{s}^{2}}{s\cdot\:{B}_{i}+{PID}_{TD}+{K}_{pas}\cdot\:s+{B}_{pas}\cdot\:{s}^{2}}\end{array}$$

(8)

In Eq. 8, the numerator represents sensory pathways and passive mechanics, and the denominator represents body dynamics and active control mechanisms.

Parameter optimization

Four parameters were optimized to minimize the error between experimental and model transfer functions, including the proportional gain, the derivative gain, the proprioceptive sensory weighting, and the effective time delay. Proportional and derivative gains were normalized by \(mgh\) (gravitational destabilizing torque each participant must counteract) to remove anthropometric confounds. Three parameters were fixed to reduce model complexity, including integral gain (\( K_{i} = 100\:{\text{N}}\;{\text{m}}\;{\text{rad}}^{{ - 1}} \;{\text{s}}^{{ - 1}} \) ), passive stiffness (\( K_{{pas}} = ~90\;{\text{N}}\;{\text{m}}\;{\text{rad}}^{{ - 1}} \)), and damping (\( B_{{pas}} = 60\:{\text{N}}\;{\text{m}}\;{\text{s}}\;{\text{rad}}^{{ - 1}} \)). The cost function was the mean squared error (MSE). Optimization was performed using the sequential quadratic programming (SQP) algorithm (MATLAB’s fmincon function, Release 2022b, MathWorks, Inc., Natick, Massachusetts, USA) with parameter constraints to avoid unrealistic values20,23,24. Starting values: \({K}_{p}=950,\:{K}_{d}=250,\:{W}_{p}=0.8,\:TD=0.15\), lower bounds: \({K}_{p}=500,\:{K}_{d}=100,\:{W}_{p}=0,\:TD=0\), upper bounds: \({K}_{p}=1500,{K}_{d}=500,\:{W}_{p}=1,\:TD=0.40\)). Time delays below 0.1 s were evaluated as 0.1 s in the cost function, thus 0.1 s acts as effective lower bound and estimates are reported accordingly. To avoid local minima, optimization was performed from five starting points around these values with parameter combinations spanning the allowable range. The solution with the lowest MSE across all starting points was selected as the final fit. Convergence criteria were set to MaxFunctionEvaluations = 100,000 and MaxIterations = 100,000 in the fmincon function. Model fitting was performed separately for each combination of visual condition (eyes open or eyes closed) and rotation amplitude (0.5° or 1.0°).

Muscle synergy analysis

Muscle synergies were computed to quantify patterns of neuromuscular coordination by identifying low-dimensional modules of co-activated muscles46. First, raw EMG data were processed to extract muscle activation envelopes from both perturbation and quiet standing trials. An analysis window spanning 2.5–62.5 s was extracted from each trial, and each muscle channel underwent the following sequential processing steps. First, EMG data were high-pass filtered at 35 Hz (8th-order Butterworth, zero-phase) to remove movement artifacts and low-frequency noise. Second, full-wave rectification was applied to obtain absolute activation amplitude. Finally, EMG data were low-pass filtered at 14 Hz (5th-order Butterworth, zero-phase) to extract the activation envelope27. Following envelope extraction, signals were downsampled by a factor of 8 (downsampled from 1259 to 157.4 Hz) to reduce computational demands while preserving temporal dynamics well above the Nyquist frequency for the 14 Hz envelope. To ensure all muscles contributed proportionally to the extracted synergies regardless of baseline activation differences or recording artifacts, each muscle’s envelope was normalized by its root mean square (RMS) amplitude computed across the entire 60-s trial window (Fig. 1C), serving the same purpose as other normalizations of previous studies47,48 but including tonic activity of standing EMG.

The optimal rank (number of synergies) was determined for each trial as the smallest rank for which an additional synergy improved the variance accounted for (VAF) (Eq. 9) by less than 5%, or at which VAF reached 90%47,48,49 (while the 90% criterion was not often met).

$$\begin{array}{c}VAF=\:\left(1-\frac{{\sum\:}_{i,j}{\left({X}_{ij}-{\left(WH\right)}_{ij}\right)}^{2}}{{\sum\:}_{i,j}{\left({X}_{ij}-\stackrel{-}{X}\right)}^{2}}\right) \cdot 100\end{array}$$

(9)

VAF was computed in its centered form49, where X is the original EMG matrix, \(W\:H\) the NMF reconstruction, and \(\stackrel{-}{X}\) is the grand mean of the entire matrix. The uncentered form counts the mean activity as explained variance and in our data was close to 100% already with one or two synergies. Because the rank distribution showed no significant group effects (ordinal generalized linear mixed model), a fixed rank of 5 was chosen across all experimental conditions (the mode of the rank distribution). The five-synergy solution accounted for adequate variance across all conditions and groups (median ≈ 80% VAF or higher; Supplementary Fig. 3). NMF was implemented using scikit-learn’s NMF module (version accessed via the R package reticulate 1.44.1, within a conda environment with Python 3.10.18 and scikit-learn version 1.7.1). Each muscle activation matrix M (muscles × time) was decomposed via NMF into synergy weight vectors W (muscles × synergies) and temporal activation patterns H (synergies × time) such that M ≈ WH. The multiplicative update algorithm was used with the Frobenius norm as the reconstruction error metric, with 2000 maximum iterations and a tolerance of 0 (no early stopping). To mitigate sensitivity to random initializations, 20 trials were run with distinct random seeds, and the solution with the lowest reconstruction error was retained. To address label-switching across decompositions, synergies were aligned using hierarchical clustering with the Hungarian algorithm50, enforcing a one-to-one mapping between synergies across conditions. Alignment proceeded hierarchically as follows: (1) within experimental conditions, (2) between quiet standing conditions, (3) across vision conditions within perturbation level, (4) within perturbation levels across groups, and (5) across all perturbation levels. Cosine distance quantifies synergy pattern similarity.

Following the NMF, we computed the mean weight of each muscle within a synergy across the cohort. A muscle was defined as primary contributor in the synergy if its mean weight exceeded the median weight of all muscles within that specific synergy. This simple threshold ensured that statistical analysis focuses on the primary contributors within a synergy. Because the reconstruction distributes residual variance across all synergies, muscles that do not define a given synergy still receive small, comparatively uniform weights that constitute a noise-dominated baseline. Restricting the analysis to primary contributors (above the within-synergy median) therefore focuses comparisons on functionally meaningful weights and avoids testing this baseline activation. The H matrix (activation coefficients) quantifies when and to what extent each synergy is activated. Because the scale factor of W and H is only defined in each decomposition, each synergy weight vector was normalized to sum to 1 across muscles and accordingly the row of H was multiplied with the same factor, leaving WH unchanged. Because of the RMS normalization, W describes the contribution of each muscle relative to the other muscles, and the mean H describes how the normalized activity of a trial is distributed across synergies. After aligning H matrices according to the W matrices mean relative activation was calculated by averaging across the time series. This approach is appropriate for continuous perturbation tasks where trial-averaged recruitment magnitude is more relevant than event-locked temporal structure, which would require analyses beyond the scope of this study.

Statistical analysis

Transfer function metrics (gain, phase and coherence) and model parameters (\({K}_{p},\:{K}_{d},{W}_{p}\), and \(TD\)) were analyzed using generalized linear mixed models (GLMM). Distributions were fitted according to each data set. Fixed effects were set to group (three levels: controls, TTA and TFA), visual condition (two levels: eyes open and eyes closed), and rotation amplitude (two levels: 0.5° and 1.0°). The models of the transfer function metrics additionally included frequency (11 levels). For the phase and gain metric, a frequency-dependent variance structure was observed; thus, these models included a dispersion formula with frequency as a predictor to account for heterogeneous variance across frequencies. The participant ID was in all models used as a random effect.

Each active muscle in a synergy was analyzed separately using GLMM with the normalized weights as the outcome variable, without additional transformation (range < 0.001 to 0.92 and no values of exactly 0 or 1). Fixed effects were set to group (three levels: control, TTA, and TFA), visual condition (two levels: eyes open and eyes closed), and perturbation (three levels: quiet standing, 0.5°, and 1°), including all main effects and interactions. Participant ID was included as a random factor. To control the false discovery rate across the models fitted within each synergy, the Benjamini-Hochberg correction was applied to the omnibus tests of all primary-contributor muscles in that synergy (5 muscles × 7 terms = 35 tests).

All model diagnostics were assessed using the DHARMa package to verify distributional assumptions. All hypothesis testing for fixed effects was conducted using Type III Wald chi-square tests and significant effects were further investigated using estimated marginal means with Tukey- and Sidak-corrected pairwise comparisons. Post hoc comparisons among groups and among perturbation levels were Tukey-adjusted within each combination of the remaining factors. Contrasts of vision and amplitude were pooled across all combinations of the remaining factors and Sidak-adjusted for their total number. Statistical significance was set at \(\alpha\:=0.05\). All statistical analyses were performed in R version 4.5.2 (2025-10-31).

Data availability

The code and output of this study are available on https://doi.org/10.5281/zenodo.23917792.

References

  1. Moxey, P. W. et al. Lower extremity amputations—a review of global variability in incidence. Diabet. Med.28, 1144–1153 (2011).

    Article  CAS  PubMed  Google Scholar 

  2. Yuan, B., Hu, D., Gu, S., Xiao, S. & Song, F. The global burden of traumatic amputation in 204 countries and territories. Front. Public. Health. 11, 1258853 (2023).

    Article  PubMed  PubMed Central  Google Scholar 

  3. Dite, W., Connor, H. J. & Curtis, H. C. Clinical identification of multiple fall risk early after unilateral transtibial amputation. Arch. Phys. Med. Rehabil.88, 109–114 (2007).

    Article  PubMed  Google Scholar 

  4. Hagberg, K. & Brånemark, R. Consequences of non-vascular trans-femoral amputation: A survey of quality of life, prosthetic use and problems. Prosthet. Orthot. Int.25, 186–194 (2001).

    Article  CAS  PubMed  Google Scholar 

  5. Horne, C. E. & Neil, J. A. Quality of life in patients with prosthetic legs: A comparison study. JPO J. Prosthet. Orthot.21, 154 (2009).

    Google Scholar 

  6. Hunter, S. W. et al. Risk factors for falls in people with a lower limb amputation: A systematic review. PMR9, 170–180e1 (2017).

    Article  PubMed  Google Scholar 

  7. Miller, W. C., Deathe, A. B., Speechley, M. & Koval, J. The influence of falling, fear of falling, and balance confidence on prosthetic mobility and social activity among individuals with a lower extremity amputation. Arch. Phys. Med. Rehabil.82, 1238–1244 (2001).

    Article  CAS  PubMed  Google Scholar 

  8. Gailey, R., Allen, K., Castles, J., Kucharik, J. & Roeder, M. Review of secondary physical conditions associated with lower-limb amputation and long-term prosthesis use. J. Rehabil. Res. Dev.45, 15–29 (2008).

    Article  PubMed  Google Scholar 

  9. Curtze, C., Hof, A. L., Postema, K. & Otten, B. The relative contributions of the prosthetic and sound limb to balance control in unilateral transtibial amputees. Gait Posture36, 276–281 (2012).

    Article  PubMed  Google Scholar 

  10. Hlavackova, P., Franco, C., Diot, B. & Vuillerme, N. Contribution of each leg to the control of unperturbed bipedal stance in lower limb amputees: New insights using entropy. PLOS ONE6, e19661 (2011).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  11. Toumi, A. et al. Standing posture and balance modalities in unilateral transfemoral and transtibial amputees. J. Bodyw. Mov. Ther.27, 634–639 (2021).

    Article  PubMed  Google Scholar 

  12. Claret, C. R. et al. Neuromuscular adaptations and sensorimotor integration following a unilateral transfemoral amputation. J. Neuroeng. Rehabil.16, 115 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

  13. Krauskopf, T. et al. Unilateral transfemoral amputees exhibit altered strength and dynamics of muscular co-activation modulated by visual feedback. J. Neural Eng.19, 016024 (2022).

    Article  Google Scholar 

  14. Fuchs, K. et al. Influence of augmented visual feedback on balance control in unilateral transfemoral amputees. Front. Neurosci.15, 727527 (2021).

    Article  PubMed  PubMed Central  Google Scholar 

  15. Nadollek, H., Brauer, S. & Isles, R. Outcomes after trans-tibial amputation: The relationship between quiet stance ability, strength of hip abductor muscles and gait. Physiotherapy Res. Int.7, 203–214 (2002).

    Article  Google Scholar 

  16. De Marchis, C. et al. Modular motor control of the sound limb in gait of people with trans-femoral amputation. J. Neuroeng. Rehabil.16, 132 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

  17. Mehryar, P. et al. The impact of different self-selected walking speeds on muscle synergies in transfemoral amputees during transient-state gait. Biomechanics4, 14–33 (2024).

    Article  Google Scholar 

  18. Nashner, L. M. & McCollum, G. The organization of human postural movements: A formal basis and experimental synthesis. Behav. Brain Sci.8, 135–150 (1985).

    Article  Google Scholar 

  19. Runge, C. F., Shupert, C. L., Horak, F. B. & Zajac, F. E. Ankle and hip postural strategies defined by joint torques. Gait Posture. 10, 161–170 (1999).

    Article  CAS  PubMed  Google Scholar 

  20. Peterka, R. J. Sensorimotor integration in human postural control. J. Neurophysiol.88, 1097–1118 (2002).

    Article  CAS  PubMed  Google Scholar 

  21. Sipahi, R., Niculescu, S., Abdallah, C. T., Michiels, W. & Gu, K. Stability and stabilization of systems with time delay. IEEE Control Syst. Mag.31, 38–65 (2011).

    Article  MathSciNet  Google Scholar 

  22. Feller, K. J., Peterka, R. J. & Horak, F. B. Sensory re-weighting for postural control in Parkinson’s disease. Front Hum. Neurosci. 13, 126 (2019).

  23. Kneis, S. et al. A new approach to characterize postural deficits in chemotherapy-induced peripheral neuropathy and to analyze postural adaptions after an exercise intervention. BMC Neurol. 20, 23 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  24. Kneis, S., Bruetsch, V., Dalin, D., Hubbe, U. & Maurer, C. Altered postural timing and abnormally low use of proprioception in lumbar spinal stenosis pre- and post-surgical decompression. BMC Musculoskelet. Disord.20, 183 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

  25. Chvatal, S. A. & Ting, L. H. Common muscle synergies for balance and walking. Front. Comput. Neurosci.7, 48 (2013).

    Article  PubMed  PubMed Central  Google Scholar 

  26. Tresch, M. C., Cheung, V. C. K. & d’Avella, A. Matrix factorization algorithms for the identification of muscle synergies: Evaluation on simulated and experimental data sets. J. Neurophysiol.95, 2199–2212 (2006).

    Article  PubMed  Google Scholar 

  27. Ghislieri, M. et al. Balance and muscle synergies during a single-limb stance task in individuals with chronic ankle instability. IEEE Trans. Neural Syst. Rehabil. Eng.PP, 1–1 (2023).

    Google Scholar 

  28. Jeka, J. et al. Sensory reweighting with translational visual stimuli in young and elderly adults: the role of state-dependent noise. Exp. Brain Res.174, 517–527 (2006).

    Article  PubMed  Google Scholar 

  29. Kuo, A. D. & Zajac, F. E. A biomechanical analysis of muscle strength as a limiting factor in standing posture. J. Biomech.26, 137–150 (1993).

    Article  PubMed  Google Scholar 

  30. Winter, D. Human balance and posture control during standing and walking. Gait Posture3, 193–214 (1995).

    Article  Google Scholar 

  31. Vrieling, A. H. et al. Balance control on a moving platform in unilateral lower limb amputees. Gait Posture28, 222–228 (2008).

    Article  CAS  PubMed  Google Scholar 

  32. Åström, K. J. & Murray, R. Feedback Systems: An Introduction for Scientists and Engineers. In Feedback Systems: An Introduction for Scientists and Engineers (2008).

  33. Clites, T. et al. Proprioception from a neurally controlled lower-extremity prosthesis. Sci. Transl. Med.10, eaap8373 (2018).

    Article  PubMed  Google Scholar 

  34. Maki, B. E., Holliday, P. J. & Topper, A. K. A prospective study of postural balance and risk of falling in an ambulatory and independent elderly population. J. Gerontol.49, M72–84 (1994).

    Article  CAS  PubMed  Google Scholar 

  35. Maurer, C., Mergner, T. & Peterka, R. J. Multisensory control of human upright stance. Exp. Brain Res.171, 231–250 (2006).

    Article  CAS  PubMed  Google Scholar 

  36. Chen, L., Feng, Y., Chen, B., Wang, Q. & Wei, K. Improving postural stability among people with lower-limb amputations by tactile sensory substitution. J. Neuroeng. Rehabil. 18, 159 (2021).

    Article  PubMed  PubMed Central  Google Scholar 

  37. Horak, F. B. Postural orientation and equilibrium: what do we need to know about neural control of balance to prevent falls? Age Ageing. 35, ii7–ii11 (2006).

    Article  PubMed  Google Scholar 

  38. Fitzpatrick, R. & McCloskey, D. I. Proprioceptive, visual and vestibular thresholds for the perception of sway during standing in humans. J. Physiol.478 (Pt 1), 173–186 (1994).

    Article  PubMed  PubMed Central  Google Scholar 

  39. Hendershot, B. D. & Wolf, E. J. Three-dimensional joint reaction forces and moments at the low back during over-ground walking in persons with unilateral lower-extremity amputation. Clin. Biomech. (Bristol). 29, 235–242 (2014).

    Article  PubMed  Google Scholar 

  40. Day, B. L., Steiger, M. J., Thompson, P. D. & Marsden, C. D. Effect of vision and stance width on human body motion when standing: Implications for afferent control of lateral sway. J. Physiol.469, 479–499 (1993).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  41. Kroes, C. M. et al. Muscle morphology and function following a transfemoral amputation: A systematic review. Can. Prosthet. Orthot. J.9, 47289 (2026).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  42. Latash, M. L. The bliss (not the problem) of motor abundance (not redundancy). Exp. Brain Res.217, 1–5 (2012).

    Article  PubMed  PubMed Central  Google Scholar 

  43. Ottobock MOBIS: The Mobility Classification System (2020).

  44. Hermens, H. J., Freriks, B., Disselhorst-Klug, C. & Rau, G. Development of recommendations for SEMG sensors and sensor placement procedures. J. Electromyogr. Kinesiol.10, 361–374 (2000).

    Article  CAS  PubMed  Google Scholar 

  45. Winter, D. A. Biomechanics and Motor Control of Human Movement (Wiley, 2009).

  46. Lee, D. D. & Seung, H. S. Learning the parts of objects by non-negative matrix factorization. Nature401, 788–791 (1999).

    Article  CAS  PubMed  Google Scholar 

  47. Roh, J., Rymer, W. Z. & Beer, R. F. Evidence for altered upper extremity muscle synergies in chronic stroke survivors with mild and moderate impairment. Front. Hum. Neurosci.9, 6 (2015).

    Article  PubMed  PubMed Central  Google Scholar 

  48. Torres-Oviedo, G. & Ting, L. H. Subject-specific muscle synergies in human balance control are consistent across different biomechanical contexts. J. Neurophysiol.103, 3084–3098 (2010).

    Article  PubMed  PubMed Central  Google Scholar 

  49. Israely, S., Leisman, G., Machluf, C. C. & Carmeli, E. Muscle synergies control during hand-reaching tasks in multiple directions post-stroke. Front Comput. Neurosci. 12, 10 (2018).

  50. Kuhn, H. W. The Hungarian method for the assignment problem. Naval Res. Logistics Q.2, 83–97 (1955).

    Article  MathSciNet  Google Scholar 

Download references

Acknowledgements

We want to thank all participants for their valuable contribution to this study. We also want to thank Dr. Nina Stobbe for preparing the ethical approval for this study. This work was supported by the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG, grant no. 505063807). Some parts of this work are also part of BrainLinks-BrainTools which was funded by the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG, grant no. EXC 1086) and is currently funded by the Federal Ministry of Economics, Science and Arts of Baden-Württemberg within the sustain-ability program for projects of the excellence initiative II.

Funding

Open Access funding enabled and organized by Projekt DEAL. This work was supported by the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG, grant no. 505063807). Some parts of this work are also part of BrainLinks-BrainTools, which was funded by the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG, grant no. EXC 1086) and is currently funded by the Federal Ministry of Economics, Science and Arts of Baden-Württemberg within the sustainability program for projects of the excellence initiative II.

Author information

Author notes

  1. Thomas Stieglitz and Cristian Pasluosta: Equal senior contribution.

Authors and Affiliations

  1. Laboratory for Biomedical Microtechnology, Department of Microsystems Engineering, University of Freiburg, Freiburg, Germany

    Thomas Krauskopf, Thomas Stieglitz & Cristian Pasluosta

  2. BrainLinks-BrainTools Center, University of Freiburg, Freiburg, Germany

    Thomas Krauskopf, Christoph Maurer, Thomas Stieglitz & Cristian Pasluosta

  3. Department of Neurology and Neuroscience, Faculty of Medicine, Medical Center, University of Freiburg, Freiburg, Germany

    Clémence Ott & Christoph Maurer

  4. Department of Orthopaedics and Trauma Surgery, Faculty of Medicine, Medical Center, University of Freiburg, Freiburg, Germany

    Georg W. Herget & Eva Johanna Kubosch

  5. Sanitätshaus Pfänder, Freiburg, Germany

    Marco Häberlin

  6. Bernstein Center Freiburg, University of Freiburg, Freiburg, Germany

    Thomas Stieglitz

Authors

  1. Thomas Krauskopf
  2. Clémence Ott
  3. Georg W. Herget
  4. Eva Johanna Kubosch
  5. Marco Häberlin
  6. Christoph Maurer
  7. Thomas Stieglitz
  8. Cristian Pasluosta

Contributions

T.K., C.M., C.P. and T.S. conceived the study. T.K., C.M. and C.P. designed the methodology. T.K. and C.O. conducted the investigation and curated the data. T.K. produced the visualizations. T.K., C.M. and C.P. validated the results. T.K. and C.M. developed the software. T.S., C.M. and C.P. acquired funding. T.S., C.M., G.W.H., E.J.K., M.H. and C.P. provided resources. T.S., C.M. and C.P. supervised the work. T.K. and C.P. administered the project. T.K. wrote the original draft. C.O., M.H., G.W.H., E.J.K., C.M., T.S. and C.P. reviewed and edited the manuscript. All authors reviewed the manuscript.

Corresponding author

Correspondence to Cristian Pasluosta.

Ethics declarations

Competing interests

The authors declare no competing interests.

Ethics approval

The experiment was approved by the University of Freiburg’s ethics committee (approval numbers 21-1245 and 24-1444_1-S1). All participants provided informed consent before participating.

Additional information

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Krauskopf, T., Ott, C., Herget, G.W. et al. Sensory reweighting and proximal muscle reorganization underlie postural control after unilateral lower limb amputation. Sci Rep 16, 38904 (2026). https://doi.org/10.1038/s41598-026-81402-3

Download citation

  • Received: 23 June 2026

  • Accepted: 16 November 2026

  • Published: 20 November 2026

  • Version of record: 20 November 2026

  • DOI: https://doi.org/10.1038/s41598-026-81402-3

Keywords

来源:Nature 健康护理主题 · nature.com