Exoskeletons assist and augment movement, but their effects on proprioceptive feedback remain poorly understood due to challenges in making direct measures of sensory signals in humans. Here, we leveraged a benchtop animal model to explore how mechanical context akin to an elastic exoskeleton operating on a human lower limb joint influences primary muscle spindle firing. In an anesthetized rat preparation, we applied controlled stretches to the medial gastrocnemius with engineered springs (0–0.5 N/mm) in parallel to the muscle-tendon unit (MTU) while modulating muscle activation to maintain overall system stiffness. Fascicle length was measured with sonomicrometry, force and MTU length with a servo motor, and spindle instantaneous firing rate (IFR) using dorsal root recordings. Trading off increases in exoskeleton stiffness with reductions in muscle activation decreased biological muscle force (3.1±0.6 N to 1.6±0.6 N, p<0.001) and stiffness (4.4±1.5 N/mm to 2.3±1.3 N/mm, p<0.01), and increased fascicle length (7.9±1.3 mm to 8.3±1.5 mm, p<0.005). We found significant correlations between spindle firing and each independent fascicle dynamics factor we investigated (p<0.005). Thus, parallel stiffness modified muscle fascicle dynamics but did not alter spindle firing, possibly due to internal trade-offs in the salient fascicle dynamics that drive spindle behavior. Leveraging in-situ experiments that enable monitoring of afferent feedback in complex mechanical contexts, such as added parallel stiffness, can provide a window into the effects of wearable devices on underlying sensory systems.

IFR was measured via a glass microelectrode inserted intracellularly into the L5 dorsal root while it was suspended on bipolar hook electrodes to record the activity of single cells. Afferent firing was recorded while imparting a predetermined ramp length change with a 3 mm amplitude, 0.125 s rise and fall time (24 mm/s rising speed), and 1 s hold using a servo motor (Aurora Scientific (Ontario, Canada)) attached to the Achilles tendon. Primary afferents were identified by their responsiveness to a 100 Hz, low-amplitude vibration imparted through the servo motor. The femur was clamped in place, and the Medial Gastrocnemius (MG) was isolated from the other plantar flexors, and activation of the isolated muscle was modulated with electrical stimulation imparted through the ventral root efferents. 100 Hz electrical stimulation (pulse width was 0.5ms, duration was 130ms) was imparted at amplitudes dependent on the total MTU and exoskeleton force required during the up ramp of the ramp and hold stretches. Muscle fascicle length was measured using sonomicrometry crystals embedded in the MG muscle belly (Sonometrics (Ontario, Canada)), taking care to avoid the aponeurosis. Various springs were attached in order of highest to lowest spring stiffness (ranging from 0.1 to 0.5 N/mm) in parallel to the medial gastrocnemius MTU (at the to the servo motor distally and the femur clamp proximally) to simulate a passive exoskeleton (Fig. 1A). The resulting length-clamped, force-matched contractions across various exoskeleton stiffness conditions, as depicted in Fig. 1A, resembled standing balance paradigms with ramp platform perturbations with and without an exoskeleton. All data were collected with a Cambridge Electronic Design Power 1401 and using Spike2 software and analyzed in MATLAB.

The springs used to represent a passive exoskeleton were attached via a custom tensioning mechanism by which the preset tension in the springs could be modified to have a consistent initial exoskeleton force acting across all spring conditions. The extension springs (McMaster-Carr) had hook ends and were all the same initial length, so that once the tensioning mechanism was adjusted for the specific rat model’s MTU length, all springs could be easily swapped throughout the experiment to test across all experimental conditions.

Measurements of motor position, motor force, and fascicle displacement were low-pass filtered with a fourth-order Butterworth filter with a 50 Hz cutoff. They were then differentiated with respect to time to get the yank (dF/dt), the velocity (dL/dt), and the acceleration ((d^2L)/(dt2 )). Muscle fascicle power, estimated as the muscle-tendon force multiplied by the fascicle velocity, was also analyzed as it potentially relates more directly to the dynamics experienced by the intrafusal fibers. The total MTU and exoskeleton system displacement and force were measured by the servomotor’s displacement and measured force. The exoskeleton force was calculated as the passive spring stiffness multiplied by the displacement of the spring, which was equivalent to the displacement of the total MTU and exoskeleton system. The tensioning mechanism was adjusted for each animal to ensure that the springs always start at resting length (no load) and the initial exoskeleton force is therefore zero. The biological MTU force was then calculated as the total force minus the exoskeleton force. This was further validated experimentally by comparing passive stretches of the total MTU and exoskeleton system across all spring conditions (no spring, 0.1 N-mm, 0.3 N-mm, 0.5 N-mm). This was done by subtracting the passive, no-spring stretch condition from the other spring conditions and then fitting a linear function to the resulting force-length curve (i.e., the stiffness of the spring in the exoskeleton’s force contribution).

To validate the effects of the parallel exoskeleton, the length change of the fascicle relative to the MTU, the stiffness of the MTU, and the stiffness of the fascicle were computed for each exoskeleton condition type. Stiffness (N/mm) was computed by fitting a linear function to force/length curves during the up-ramp of the ramp-and-hold stretches. Stiffness (in N/mm) was calculated by computing the slope of the tangent line at these force thresholds.

Spike count during the up-ramp, dynamic responses, and static responses were extracted from firing rates of the ramp-and-hold stretches (Fig. 2A-B). The dynamic response was taken as the instantaneous firing rate at the peak MTU force production, when it first reaches its max stretch. The static response was taken as the firing rate approximately 0.5 s into the hold phase, and the dynamic index is the difference between the dynamic and static response.

Data were excluded from analysis if the sensory afferent action potentials could not be discriminated from the cell recording. Data were also excluded if the muscle fascicle dynamics during the trials did not reflect those expected from standing balance trials with an exoskeleton (i.e., decreased muscle force production and increased muscle stretch with increased parallel exoskeleton stiffness. If we could not retain stable measurements from the cell long enough to record all exoskeleton conditions, it was also excluded from analysis.

Data were collected from 24 unique primary afferent cells across 8 animals and reduced to 13 cells across 4 animals after filtering based on the exclusion criteria mentioned above. In total, 117 ramp-and-hold stretches across the 4 exoskeleton conditions were analyzed.