Tuning of feedforward control enables stable muscle force-length dynamics after loss of autogenic 1 proprioceptive feedback 2 3 Authors: JC Gordon1, NC Holt2, AA Biewener3 and MA Daley1,4* 4 Affiliations: 5 1 Comparative Biomedical Sciences, Royal Veterinary College, University of London 6 2 Evolution, Ecology & Organismal Biology, University of California, Riverside7 3 Organismic and Evolutionary Biology, Harvard University8 4Ecology and Evolutionary Biology, University of California, Irvine 9 10 Running head: muscle dynamics in running after self-reinnervation 11 12 Address for correspondence: 13 *Author for correspondence: madaley@uci.edu14 University of California, Irvine, 1408 Biological Sciences, Irvine, CA 92697-2525 15 16 Keywords: stability, intrinsic mechanics, passive-dynamics, reflexes, feedback, feedforward, electromyography, 17 running, locomotion, in vivo muscle work loop, guinea fowl 18 19 Accepted for publication in eLife 11 June 2020 Abstract 20 Animals must integrate feedforward, feedback and intrinsic mechanical control mechanisms to maintain stable 21 locomotion. Recent studies of guinea fowl (Numida meleagris) revealed that the distal leg muscles rapidly 22 modulate force and work output to minimize perturbations in uneven terrain. Here we probe the role of reflexes 23 in the rapid perturbation responses of muscle by studying the effects of proprioceptive loss. We induced bilateral 24 loss of autogenic proprioception in the lateral gastrocnemius muscle (LG) using self-reinnervation. We 25 compared in vivo muscle dynamics and ankle kinematics in birds with reinnervated and intact LG. Reinnervated 26 and intact LG exhibit similar steady state mechanical function and similar work modulation in response to 27 obstacle encounters. Reinnervated LG exhibits 23ms earlier steady-state activation, consistent with feedforward 28 tuning of activation phase to compensate for lost proprioception. Modulation of activity duration is impaired in 29 rLG, confirming the role of reflex feedback in regulating force duration in intact muscle.30 Main text 31 Introduction 32 Sensory feedback is widely accepted as an integral component of vertebrate locomotor control (Cohen, 1992; 33 Donelan & Pearson, 2004; Grillner, 2011; Prochazka & Ellaway, 2012; Rossignol et al., 2006). Proprioception 34 from muscle mechanoreceptors contributes to 1) short-latency reflexes via spinal mono- and polysynaptic 35 pathways to regulate the ongoing activity and muscle mechanical output (force, stiffness, impedance and work) 36 and 2) longer-latency responses to coordinate and maintain task-level goals for balance and movement (Grillner, 37 2011; Lam & Pearson, 2002; Frigon & Rossignol, 2006; Prochazka & Ellaway, 2012; Proske & Gandevia, 2012; 38 Rossignol et al., 2006; Safavynia and Ting 2013; Sherrington & Laslett, 1903; Sherrington 1910). Proprioceptive 39 reflexes can occur through autogenic (self-generated) pathways arising from the same muscle, and through 40 heterogenic pathways arising from other muscles via spinal interneurons (Abelew et al., 2000; Frigon & 41 Rossignol, 2006; Lam & Pearson, 2002; Nichols 1989; Ross & Nichols 2009). Thus, the relationship between a 42 specific sensory signal and its resulting effects is complex and dynamic. 43 Despite recognized functions of proprioception, the relative contribution of feedback control in high-speed 44 locomotion remains unclear. Sensorimotor delay constrains how quickly an animal can sense and respond to a 45 stimulus using feedback control (More & Donelan, 2018; More et al., 2010). The fastest possible feedback loop 46 occurs through mono-synaptic reflexes, which involve a delay that increases in proportion to nerve transmission 47 distance. This reflex delay becomes a larger fraction of the stride cycle with increasing speed, limiting time 48 available for reflex-mediated correction. 49 The challenges of long delays relative to stride cycle times likely necessitates greater reliance on feedforward 50 control strategies at higher speeds. Here we use feedforward to refer to the contributions to motor output arising 51 from the motor cortex, descending pathways and rhythmic spinal networks (Frigon & Rossignol, 2006; Pearson, 52 2000; Yakovenko et al. 2004). Rhythmic spinal networks can generate the basic flexion and extension motor 53 pattern for gait, even when proprioceptive feedback is removed (e.g., Pearson et al., 2003; Sharp & Bekoff 54 2015). Normally, however, descending networks act in concert with spinal networks and feedback, using 55 multimodal and distributed sensory inputs to update state estimates, regulate rhythm and control foot placement 56 (Cohen 1992; Drew and Marigold 2015; Marigold and Drew 2017; Pearson et al. 1999; Potocanac et al. 2014; 57 Todorov 2004; Wagner & Smith 2008; Wolpert et al. 2011). Consequently, there is no ‘pure’ feedfoward control 58 within vertebrate systems. We use the term here as a pragmatic distinction, where feedforward refers to 59 anticipatory ‘look-ahead’ control over one or more stride cycles, and feedback refers to reflex-mediated reactive 60 responses to perturbations. 61 Although feedforward networks normally act in concert with feedback, feedforward motor activation coupled to 62 intrinsic muscle properties can be sufficient to produce stable gait (Yakovenko et al., 2004). Consistent with this, 63 the lateral gastrocnemius (LG) of guinea fowl (Numida meleagris) rapidly absorbs energy in response to 64 unexpected drop perturbations (Daley et al., 2009), stabilizing high speed running without a reflex response. The 65 rapid perturbation response arises from the intrinsic mechanical properties of the muscle-tendon tissues and 66 musculoskeletal system (Brown & Loeb 1999; Loeb et al. 1999; Jindrich and Full 2002). Intrinsic mechanical 67 responses can be actively tuned by the specific feedforward pattern of muscle activation. For example, humans 68 hopping on surfaces with randomized, sudden increases in ground stiffness show a feedforward increase in 69 muscle co-activation and knee flexion, increasing mechanical stability (Moritz & Farley 2004). However, many 70 perturbation responses involve multiple control mechanisms that overlap in time. Guinea fowl running over 71 obstacles use a combination of feedforward, intrinsic mechanical and reflex-mediated mechanisms, with a delay 72 of ~40 ms for reflex-mediated increases in muscle force (6ms reflex latency + 34ms force development delay: 73 Daley et al. 2009; Daley & Biewener, 2011). Considering that the feedforward and intrinsic mechanical 74 contributions alter ongoing muscle dynamics before the reflex-mediated response, it is difficult to disentangle 75 the specific contributions of proprioceptive reflexes to the observed perturbation responses (Daley & Biewener, 76 2011; Gordon et al. 2015). 77 Investigating the role of proprioception through self-reinnervation 78 Here we probe the integration of feedforward, feedback and intrinsic mechanical control by eliciting a 79 proprioceptive deficit in the lateral gastrocnemius muscle (LG) of the guinea fowl (Numida meleagris) using 80 bilateral self-reinnervation (Figure 1). Self-reinnervation involves peripheral nerve branch transection and 81 immediate repair, resulting in recovery of motor output with long-term, local loss of autogenic muscle 82 proprioception (Cope et al., 1994; Bullinger et al. 2011). Self-reinnervation occurs through axonal regrowth and 83 reconnection with denervated tissues over a recovery period of 4-8 weeks (Carr, et al., 2010; Cope, et al., 1994; 84 Gordon & Stein, 1982; Vannucci et al., 2019). Reinnervated muscles retain a deficit in the monosynaptic stretch 85 reflex due to synaptic retraction of primary muscle spindle afferents and disconnection from parent motoneuron 86 populations (Alvarez et al., 2011; Brandt et al. 2015). However, intermuscular force and length feedback 87 networks may remain partially intact (Lyle et al. 2016). Cats and rats with reinnervated muscles maintain whole-88 limb function by adjusting inter-joint coordination and muscle co-activation to compensate for loss of reflex-89 mediated ankle stiffness (Abelew et al., 2000; Maas et al., 2007; Chang et al. 2009; Boeltz et al. 2013). These 90 findings highlight the ability of animals to flexibly exploit musculoskeletal plasticity to maintain function and 91 suggest self-reinnervation as a promising tool to investigate sensorimotor control mechanisms. 92 Studying neuromuscular control in the guinea fowl, a bipedal animal model, provides insight into similarities 93 and differences among vertebrates that may relate to locomotor modality, evolutionary history, or both. Birds 94 share features of sensorimotor structure and function with mammals, including muscle tissue properties (Nelson, 95 et al., 2004; Poore et al., 1997) and muscle proprioception through muscle spindle and Golgi tendon organs 96 (Dorward, 1970; Haiden & Awad 1981; Maier, 1992). Ground birds use bipedal walking and running gaits with 97 mechanics and energetics similar to human locomotion (Heglund, et al., 1982; Taylor, et al., 1982; Gatesy & 98 Biewener, 1991; Roberts, et al., 1997; Daley & Birn-Jeffery, 2018). Bipedal gaits involve substantial periods of 99 single-limb contact, which limits the redundancy of balance mechanisms and poses a challenge for stability 100 (Daley, et al., 2009; Clark & Higham, 2011; Daley & Biewener, 2011). Whereas a quadrupedal cat or rat might 101 be able to compensate for deficits by shifting weight bearing among legs, a biped with a bilateral proprioceptive 102 deficit cannot. Accordingly, one goal of the current study is to explore whether or not guinea fowl exhibit a 103 similar response to proprioceptive deficit as observed in quadrupedal vertebrates (Abelew et al., 2000; Maas et 104 al., 2007; Chang et al. 2009; Boeltz et al. 2013) 105 We hypothesize that autogenic proprioceptive deficit will lead to increased reliance on feedforward control 106 mechanisms and intrinsic muscle mechanics to maintain stable locomotion. To test for shifts in stability and 107 control mechanisms, we measured ankle kinematics and in vivo LG muscle dynamics (length, force and 108 activation) during treadmill running on level and obstacle terrain. There are several potential mechanisms to 109 compensate for autogenic proprioceptive deficit: 1) Birds might compensate for proprioceptive deficit by 110 increasing feedforward muscle activation before obstacle contact, as observed in birds negotiating high-contrast 111 obstacles (Gordon et al. 2015). 2) Alternatively, if feedback regulation of LG is essential for stability in fast 112 locomotion, loss of autogenic proprioception may necessitate increased reliance on heterogenic reflex pathways 113 from synergists, with a slight delay compared to intact animals, as suggested by work in cats and rats (Boeltz et 114 al. 2013; Lyle et al. 2016). 3) Finally, if intrinsic mechanics are mainly responsible for the modulation of muscle 115 force and work, we might expect minimal change in muscle activity patterns (EMG), as observed in birds 116 subjected to unexpected drop perturbations (Daley, et al., 2009). We expect birds to compensate for 117 proprioceptive deficit by tuning gait and feedforward muscle activity to maintain a stable response to obstacle 118 perturbations, as observed in reinnervated rats and cats walking on slopes (Abelew et al., 2000; Maas et al., 119 2007; Chang et al. 2009; Boeltz et al. 2013). If stability is impaired following reinnervation, this should be 120 evident from increased variance and longer time to recover from obstacles. By investigating the shifts in guinea 121 fowl LG muscle force, length and activation dynamics following reinnervation, we hope to gain insight into the 122 mechanisms of sensorimotor integration and plasticity that enable robustly stable and agile bipedal locomotion. 123 124 Results 125 Mechanical function of intact versus reinnervated LG 126 We find that many features of the steady-state in vivo mechanical function of the guinea fowl reinnervated lateral 127 gastrocnemius (rLG) are similar to that previously measured in the intact LG (iLG). In Figure 2, average 128 trajectories (mean±95% confidence interval) are shown for muscle strain, force and electromyographic activity 129 (iLG at top, rLG below), with the average for steady level running in grey. During the swing phase of the stride 130 cycle, both iLG and rLG exhibit a period of passive stretch, followed by rapid shortening. Activation and force 131 development begin in late swing around the time of the transition from stretch to shortening, initiating rapid 132 active shortening until foot contact (Figure 2, triangles). At the time of foot contact, force increases rapidly to a 133 peak before midstance, then declines more slowly. Typically, in level running both iLG and rLG show a near-134 isometric phase in early stance, followed by shortening in late stance, which produces net positive work, as 135 indicated by counter-clockwise force-length work loops (Figure 3). The average magnitude of work output 136 (Wnet) during steady level running is similar between the iLG and rLG, with similar spread of the distribution 137 around the mean (Figure 4). However, rLG shows faster shortening velocity at peak force (VpkF) compared to the 138 iLG across both level and obstacle terrains (Figure 4), indicating a difference in steady state contraction 139 dynamics. 140 Force-length dynamics and work output during obstacle negotiation 141 In obstacle encounters (Figure 2, S 0), foot contact with the obstacle occurs earlier in the stride cycle compared 142 to level terrain, altering force-length dynamics during the obstacle stance period. During obstacle contact, both 143 iLG and rLG remain at longer lengths, force increases rapidly to reach a higher peak force, and the muscle 144 shortens throughout force development, producing positive work (Figure 2, S 0). Both iLG and rLG exhibit 145 increased force and work output in obstacle strides compared to level strides (Figure 3, S 0). The magnitude of 146 the shift in work output (Wnet) in obstacle strides (S 0) is similar between intact and reinnervated cohorts, 147 increasing by 3.60±0.57 Jkg-1 in iLG and 3.88±0.60 Jkg-1 in rLG (mean95% ci, Figure 4B, Table 2). 148 Although the magnitude of the shifts in work output are similar between iLG and rLG, the mechanisms 149 underlying the shift in work output differ between them (Figure 4). In iLG, increased work upon obstacle contact 150 occurs through modest increases in both peak force (Fpk) and shortening velocity (VpkF), compared to level 151 strides. In contrast, rLG exhibits a substantially larger increase in Fpk on obstacle strides and maintains similar 152 VpkF between level and obstacle terrain strides (S 0, Figure 4B, Table 2). Reinnervated LG also exhibits small 153 but significant increases in Fpk in the strides preceding and following obstacle contact (S-1 and S+1, 154 respectively), compared to level terrain. 155 Shifts in activation patterns between intact and reinnervated LG 156 Despite deficits in LG monosynaptic reflex following reinnervation, rLG and iLG show similar increases in total 157 muscle activation intensity (Etot, integral of EMG) in obstacle strides compared to steady level strides (S 0, 158 Figure 5). Intact LG exhibits a 4% increase in duration in obstacle strides (S 0) compared to level; however, 159 there is no significant increase in EMG duration for rLG in S 0 (Figure 5B, Table 2). This suggest that the 160 observed increase in Etot in rLG obstacle strides occurs through increased activation amplitude, not increased 161 duration. 162 Several results suggest a shift in central drive and feedforward activation pattern in rLG compared to iLG. 163 Reinnervated LG exhibits longer steady-state duration of activity (Edur) compared to iLG across all level and 164 obstacle terrain strides, averaging 37% of the stride cycle in rLG compared to 29% in iLG (Figure 5B, Table 2). 165 Additionally, rLG exhibits higher average frequency of EMG activity across all strides compared to the intact 166 cohort (Figure 5, Table 2, Figure 5-figure supplement 1). Finally, the steady-state timing of rLG activation is 167 phase-shifted to 6% (23ms) earlier in the stride cycle relative to the length trajectory, quantified by the variable 168 ‘Ephase’ (Figure 6, Table 1-2). Earlier activation onset may help explain the higher rate of shortening in rLG 169 compared to iLG, reported above. 170 Timing of obstacle-induced changes in EMG activity in iLG and rLG 171 To explore the timing of obstacle-induced shifts in EMG activity relative to perturbations in force and length, we 172 calculated a difference trajectory between the steady-state level and obstacle perturbed stride cycles (S 0 – Lev) 173 for each individual, and then calculated the mean and 95% confidence interval across individuals (Figure 7). 174 Increased EMG activity begins ~30-40 ms before obstacle-induced increases in length and force, for both iLG 175 and rLG, suggesting an anticipatory (feedforward) contribution to increases in Etot (Figure 7, arrows indicating 176 ‘anticipatory increase’). In iLG, the anticipatory increase in EMG starts ~25% of stride period (asterisk in EMG 177 trace and vertical dashed line in Figure 7A). Starting around 58% of stride period, there is another distinct burst 178 of increased activity, suggesting reflex-mediated contribution to increased EMG in obstacle strides (Figure 7, 179 arrow indicating ‘reflex’). In rLG, the ‘anticipatory increase’ in EMG starts around 21% of stride (asterisk and 180 vertical dashed line in 7B); however activity in the latter half of stance is highly variable and idiosyncratic 181 among individuals (Figure 7-figure supplement 1), as indicated by the wide 95% confidence intervals spanning 182 the region of time where the iLG shows a distinct reflex response (compare Figure 7B versus 7A lower panels). 183 Cross-correlation between the obstacle perturbation trajectories for iLG reveals a correlation of 0.82 between 184 length and EMG deviations, and a correlation of 0.85 between force and EMG deviations. For rLG, the cross-185 correlations are reduced to 0.53 between length and EMG deviations, and 0.57 between force and EMG 186 deviations, respectively. The reduced correlations suggest a disrupted reflex-mediated response to muscle load 187 and strain in the latter half of stance in rLG, but which is present in iLG (Figure 7). Considering that the changes 188 in muscle length and force are strongly correlated with each other during obstacle encounters in both intact and 189 reinnervated conditions (Figure 7), it is difficult to distinguish the specific sensory signal eliciting reflex 190 responses. 191 Stability and kinematic changes during obstacle negotiation in intact vs reinnervated birds 192 Compared to the intact cohort, birds with rLG exhibit more pronounced shifts in gait dynamics in obstacle 193 terrain relative to level terrain. Obstacle-induced increases in peak force (Fpk) are larger for rLG compared to 194 iLG (S 0 Figure 4, Table 2), reflecting larger deviations from steady state in response to the same obstacle. 195 Additionally, rLG shows small but significant increases in peak force (Fpk) in the strides preceding and following 196 obstacle contact (S -1, S +1) compared to steady level strides (Figure 4B, Table 2). Multiple significant 197 differences from level strides occur for rLG in S +1, including a 39±22% increase in Etot, a 6±2% decrease in 198 force duration and a 4±3% decrease in stride duration (mean±95% ci, Table 2). In contrast, most variables for 199 iLG have recovered to steady state in S +1 (Table 2). Both iLG and rLG rapidly increase work output during 200 obstacle encounters and face increased activation costs for locomotion in obstacle terrain. However, rLG shows 201 larger deviations from steady state, and a slower recovery to steady state mechanics and activation level 202 compared to iLG, indicating reduced stability. 203 Reinnervated birds also show differences in running kinematics in obstacle terrain compared to the intact cohort, 204 undergoing a more pronounced increase in ankle flexion in obstacle encounters (Figure 8). Reinnervated birds 205 also use a shorter stride duration immediately preceding the obstacle encounter (S -1), suggesting anticipatory 206 preparation that is not observed in intact birds (Figure 8). These observations suggest reduced ankle stiffness and 207 increased anticipatory preparation for obstacle encounters in the reinnervated birds. 208 209 Discussion 210 What is the role of proprioception in the control of high-speed locomotion? 211 We investigated the role of reflexes in the sensorimotor control of running by examining the effects of proprioceptive 212 deficit on the mechanical function of the lateral gastrocnemius muscle (LG) of guinea fowl. Long sensorimotor delays 213 relative to limb cycling times necessitate that animals use a combination of feedforward, feedback and intrinsic 214 mechanical control mechanisms to achieve stable locomotion at high speeds (Brown & Loeb, 2000; Jindrich and Full 215 2002; Daley & Biewener, 2011; Daley et al., 2009; Frigon & Rossignol, 2006; Grillner, 2011; Lam & Pearson, 2002; 216 More & Donelan, 2018; Pearson & Gramlich, 2010; Prochazka & Ellaway, 2012). We hypothesized that an autogenic 217 proprioceptive deficit will lead to increased reliance on feedforward tuning of muscle activity to achieve stable muscle 218 dynamics in obstacle terrain. In birds with intact LG proprioception, the timing of muscle activity in obstacle-perturbed 219 strides is consistent with combined feedforward and feedback control (Daley & Biewener 2011, Gordon et al. 2015). 220 Birds with reinnervated LG (rLG) exhibit a consistent phase shift in EMG onset relative to muscle length, with activation 221 starting 6% earlier (23ms) in the steady state contraction cycle, in both level and obstacle terrain (Figure 6, Table 2). 222 This is consistent with a feedforward tuning of rLG activation timing to enable rapid force development and high muscle 223 stiffness during stance, in the absence of monosynaptic reflexes. Regulation of EMG duration in obstacle strides (S 0) is 224 absent in rLG (Figure 5), suggesting that proprioceptive feedback in late stance normally regulates force duration, which 225 is disrupted following reinnervation. 226 A stable intrinsic mechanical response with neither feedforward- nor feedback-mediated changes in neural drive 227 can occur when a perturbation is encountered at high running speeds (Daley et al. 2009). Rapid changes in 228 muscle length and velocity in response to perturbations can decouple activation and force development (Daley et 229 al. 2009, Daley & Biewener, 2011). In our previous work on intact in vivo muscle dynamics, variation in LG 230 muscle strain during initial foot contact and limb loading explained 60% of the variation in force developed in 231 obstacle encounters, while variation in LG muscle activation explained only 9%. This clearly demonstrates the 232 decoupling between activation and force development that can occur in vivo (Daley & Biewener, 2011). These 233 intrinsic mechanical effects minimize the disturbances in body dynamics that arise from terrain height 234 perturbations, enabling rapid recovery to steady gait. Similar intrinsic mechanical stabilizing responses have 235 been demonstrated in the distal hindlimb joints of hopping and running humans subjected to unexpected changes 236 in terrain height and stiffness (Dick, et al., 2019; Ferris, et al., 1999; Moritz & Farley, 2004). In concert with the 237 stabilizing contributions of the intrinsic muscle-tendon dynamics, guinea fowl with intact proprioception also use 238 feedforward and feedback regulation of muscle activity to maintain stability in obstacle terrain, with greater 239 feedforward contributions when obstacles are visible and high contrast (Daley & Biewener, 2011, Gordon et al. 240 2015). 241 We find that guinea fowl with LG proprioceptive deficit achieve similar increases in total EMG activity during 242 obstacle strides; however, the increases in activity occur early in the stride, before obstacle-induced changes in 243 muscle force and length (Figure 7). This is consistent with anticipatory, feedforward increases in neural drive to 244 the muscle, as observed in birds running over high-contrast visible obstacles (Gordon et al. 2015), and humans 245 hopping on randomized but expected increases in surface stiffness (Moritz & Farley, 2004). These findings are 246 consistent with a hybrid feedforward/feedback control model as conceptualized by Kuo (2002) in which 247 feedforward and feedback gains are balanced to enable accurate state estimation and robust cyclical dynamics in 248 the presence of both disturbances and sensory error. Although the reinnervated LG contributes to an effective 249 obstacle negotiation response, it requires a longer recovery time and increased muscle activity following obstacle 250 contact (Figure 5B). This suggests that the integrated response of the intact neuromuscular system enables robust 251 stability with lower muscle activation costs. 252 Several features of the kinematics and muscle dynamics suggests coordinated plasticity and tuning of feedfoward 253 control to compensate for reflex deficit following recovery from nerve injury (Figure 9). We observe similar 254 increases in work output in response to obstacle encounters in rLG and iLG (Figure 4B). This finding is 255 consistent with the idea that muscle work modulation is an important feature of task-level control for stability in 256 uneven terrain (Daley et al. 2009; Daley & Biewener 2011). However, work modulation is achieved through 257 different underlying mechanisms in rLG and iLG. Earlier steady state activation in rLG (lacking autogenic 258 proprioceptive feedback) enables higher muscle force development in early stance to resist the external load 259 applied at foot contact, which likely contributes to the higher rate of shortening throughout stance. Additionally, 260 rLG exhibits larger increases in peak force in obstacle encounters compared to iLG (Figure 4B). This increase in 261 peak force likely involves both active and passive components: an active contribution from increased 262 feedforward drive and EMG amplitude (Figure 5B), and a passive contribution from increased stretch of 263 connective tissues associated with a more flexed ankle posture at foot contact (Figure 8). Finally, Birds with rLG 264 also show an anticipatory shift in stride duration before obstacle encounters, which is not observed in the intact 265 cohort (Figure 8B) and may help control landing conditions for obstacle encounters (Gordon et al. 2015). These 266 findings suggest that reinnervated birds achieve effective muscle work modulation and stable obstacle 267 negotiation through feedforward tuning of muscle activation and gait to compensate for loss of autogenic 268 proprioception. 269 A recent study by Sawicki et al. (2015) found that earlier onset of activation was associated with a shift to energy 270 absorption in cyclical muscle contractions with a sinusoidal MTU length trajectory. We find here that earlier onset is 271 associated with greater shortening and work production. The specific response of a muscle to a shift in activation phase 272 is likely to be highly sensitive to the specific steady-state length trajectory of the muscle. Muscle force capacity and the 273 activation and deactivation kinetics are substantially influenced by velocity and recent strain history, as demonstrated in 274 controlled studies of in vitro muscle force-length work loops (Askew & Marsh, 1998; Josephson, 1999). Further work is 275 needed to understand how in vivo muscle fascicle length dynamics interact with neural activation patterns and MTU 276 compliance to enable tuning of muscle contraction dynamics to the mechanical demands of cyclical locomotor tasks. 277 278 We do observe shifts in muscle activity in late stance in some reinnervated individuals in response to obstacle 279 encounters, which suggests heterogenic reflex responses in reinnervated LG (Figure 2). However, these 280 responses are variable and idiosyncratic, with some birds showing increased EMG in late stance in obstacle 281 strides, and others showing a decrease (Figure S3). The variable and idiosyncratic reflex responses result in 282 wide confidence intervals for the obstacle-induced EMG response in the latter half of stance, despite consistent 283 force-length trajectories over the same time-period (Figure 7). Idiosyncratic use of heterogenic reflex modulation 284 across individuals following nerve injury recovery is consistent with findings in cats (Lyle et al. 2016; Lyle & 285 Nichols, 2018). Guinea fowl have several agonist muscles to the LG that could contribute to heterogenic 286 feedback modulation, including the medial gastrocnemius and digital flexors (Daley & Biewener, 2011, Gordon 287 et al. 2015). However, no muscle is an exact synergist of the LG, because each has a unique combination of 288 moment arms, fiber length, pennation angle and connective tissue compliance (Daley & Biewener 2003; Cox et 289 al. 2019). Consequently, it is unlikely that proprioception from agonists can completely restore accurate sensing 290 to regulate LG force and work output. Additionally, in the presence of increased sensory error and noise, birds 291 may learn over time to compensate through sensory integration in higher CNS pathways, leading to updated 292 central coordination and feedforward drive to rhythm generating networks. 293 It was previously unknown whether guinea fowl would respond to reinnervation and proprioceptive deficit in a 294 manner similar to quadrupedal mammals. We find that our results are consistent with the findings on rats and 295 cats. Reinnervated cats and rats exhibit shifts in feedforward muscle activity and inter-joint coordination during 296 slope walking, to compensate for loss of reflex-mediated ankle stiffness (Abelew et al., 2000; Maas et al., 2007; 297 Chang et al. 2009; Boeltz et al. 2013). Cats and rats also preserve task level features of gait, such as leg length 298 and body motions, despite variance in muscle and joint dynamics (Chang et al. 2009; Boeltz et al. 2013). These 299 findings suggest performance of task-level goals as a target of sensorimotor optimization. Also similar to cats, 300 guinea fowl exhibit variation among individuals in heterogenic compensation for loss of the autogenic stretch 301 reflex, as suggested by the variable tendon tap responses and variation in late-stance EMG activity in obstacle 302 perturbed steps (Figure 7; Figure S3). Work on cats suggests complex intermuscular feedback connectivity, 303 which can recovery to varying degrees following reinnervation (Cope et al. 1994; Pearson et al. 1999; Lyle et al. 304 2016, Lyle and Nichols 2018). This complexity and variability among individuals following recovery from nerve 305 injury reflects the complexity and plasticity of proprioceptive feedback networks. Nonetheless, the recovery of 306 consistent task-level mechanical function supports the idea that sensorimotor control is optimized to maintain 307 task-level performance goals such as stable body dynamics (Chang et al. 2009; Safavynia and Ting 2013). 308 Limitations and future directions: 309 One of the major limitations in the current study is the potential for multiple differences between the intact and 310 reinnervated cohorts that were not controlled, because the experiments on the two cohorts were conducted over 311 different periods of time. The self-reinnervation procedure requires a long-term recovery period and results in a 312 chronic sensory deficit, which is likely to lead to a complex array of changes in the musculoskeletal tissues and 313 the sensorimotor networks. The current study did not include a sham-surgery experimental control, and we did 314 not strictly monitor the ages of the original intact cohort at the time of the in vivo muscle-tendon surgeries. 315 Nonetheless, it is reassuring that our findings are consistent with similar studies in cats and rats, suggesting 316 feedforward tuning of muscle activation and ankle kinematics to maintain stability following loss of 317 proprioception. In future experiments, it will be important to control the timing and amount of exercise training 318 in intact and reinnervated experimental groups, considering the potential for exercise to influence the recovery 319 process (Boeltz et al. 2013; Brandt et al. 2015). 320 The recovery process almost certainly involves coupled changes across multiple systems, including connective 321 tissue compliance, muscle activation kinetics, fiber type distribution, motor unit size and distribution, spinal 322 intraneuronal connectivity, and sensory integration in higher CNS centers for state estimation and movement 323 planning. Due to the complex nature of these adaptations, it is challenging to fully tease apart individual 324 contributions and mechanisms from in vivo experimental measures alone. In future studies, the coordinated 325 mechanisms of sensorimotor adaptation and plasticity could be systematically explored through a combination of 326 integrative experimental and computational approaches. These approaches could include 1) closed loop 327 neuromechanical simulations to enable predictive hypothesis testing (Ijspeert, 2014; Roth, et al., 2014), 2) 328 combined use of in vivo measures of muscle dynamics with in vitro testing of muscle contractile dynamics, to 329 replicate biologically realistic force-length contraction dynamics, 3) histological studies to examine changes in 330 muscle fiber type distribution and connective tissue characteristics following reinnervation, and 4) perturbation 331 approaches that probe both short and long term adaptation processes. 332 It remains unclear how the specific length trajectory and velocity features of in vivo muscle dynamics contribute 333 to the intrinsic stability and control of movement. Dynamic measurement techniques are needed to address this 334 challenge and to develop realistic models for in vivo muscle-tendon function. In addition to widely recognized 335 force-length and force-velocity ‘Hill-type’ properties, muscle exhibits short and long-term history-dependent 336 changes in force capacity in response to stretch and shortening (Edman, 1975; Edman et al. 1978; Edman, 1980; 337 Josephson, 1999; Herzog, 2004; Edman 2012; Herzog 2014; Rode et al. 2009; Nishikawa et al. 2012; Yeo et al. 338 2013; Nishikawa, et al., 2018). Recent developments in biorobotic platforms that enable controlled muscle 339 experiments with realistic loading and length trajectories are promising tools for advancing our understanding of 340 the role of intrinsic muscle dynamics in the control of movement (Clemente & Richards, 2012; Richards, 2011; 341 Robertson & Sawicki, 2015). Integrative neuromechanical studies using multiple techniques will be essential for 342 unravelling mechanisms of muscle function, sensorimotor integration and plasticity. Findings from these studies 343 have important implications for many human health conditions, including acute nerve injury, diabetic 344 neuropathy, neurodegenerative disorders, cerebral palsy, and muscular dystrophies. 345 346 Materials and Methods 347 Animals and treadmill training 348 We obtained and reared six hatchling guinea fowl keets (Numida meleagris) from a breeder (Hidden Hollow 349 Acres, Whitehouse Station, NJ), to allow re-innervation surgeries in juveniles with at least 12 weeks for recovery 350 before in vivo muscle procedures (see below). At the time of the in vivo muscle measurements, the guinea fowl 351 had reached adult size, averaging 1.810.28 kg body mass (meanS.D.). Birds had primary feathers clipped and 352 were trained to run on a level motorized treadmill (Woodway, Waukesha, WI, USA). Training sessions were 15-353 20 minutes in duration, with breaks for 2 minutes as needed. All experiments were undertaken at the Concord 354 Field Station of Harvard University, in Boston (MA, USA), and all procedures were licensed and approved by 355 the Harvard Institutional Animal Care and Use Committee (AEP #20-09) in accordance with the guidelines of 356 the National Institutes of Health and the regulations of the United States Department of Agriculture. 357 We also include data previously reported in Daley and Biewener (2011) from intact individuals (n=6, 358 1.770.63 kg body mass) to serve as a control group for statistical comparison to the new dataset (n=6 359 reinnervated individuals). We re-analyzed the intact cohort dataset alongside the reinnervated cohort, to ensure 360 consistency in data processing and statistics. The intact data includes a larger sample than reported in Daley and 361 Biewener (2011), because the analysis here includes all strides in the level and obstacle terrain collected for 362 running speeds between 1.3-2.0 ms-1. Trials were recorded only for speeds that each bird could comfortably 363 maintain on the treadmill belt for at least 30 seconds, allowing for 10-minute rest periods between trials with 364 access to food and water. We focus on running speeds (>1.3ms-1), to avoid the confounding effect of different 365 sensorimotor control strategies in walking versus running (Gordon et al. 2015). Due to variation among 366 individuals in the successful trials recorded, the intact cohort dataset includes a wider speed range (1.3-2.0 ms-1) 367 than the reinnervated cohort dataset (1.7-2.0 ms-1). However, the analysis is focused on obstacle perturbations 368 compared to steady gait at the same speed, and the datasets include comparable samples of obstacle encounters 369 between the two cohorts: 128 for intact and 133 for reinnervated birds, respectively. In total, the dataset includes 370 1027 strides for reinnervated and 1512 strides for intact individuals and excludes 81 strides as outliers that were 371 non-obstacle encounter strides with Z-score > 4. No obstacle encounter strides were excluded as outliers. The 372 complete datasets for reinnervated and intact cohorts are available through DataDryad.org, including metadata 373 and Matlab processing scripts (Daley et al. 2020, https://doi.org/10.7280/D11H49). 374 Anesthesia and post-operative care 375 Birds were induced and maintained on a mid-plane of anesthesia using isoflurane (2 - 3%, mask/intubation 376 delivery). We administered perioperative enrofloxacin and flunixin intramuscularly for analgesia after induction 377 and continued for three days after each surgery. Birds recovered to bilateral weight bearing within 20 minutes 378 following completion of surgical procedures. 379 Reinnervation surgery 380 The timing of surgeries was planned based on a pilot study, which found full recovery of LG motor activity by 6 381 weeks following reinnervation surgeries, and continued absence of calcaneal tendon reflex one year later, 382 indicating continued absence of autogenic stretch reflexes (Carr et al., 2010). We bilaterally transected and 383 immediately repaired the peripheral nerve branch supplying the LG muscle in maturing guinea fowl between 7-384 12 weeks of age. We allowed time for full reinnervation recovery of motor output and growth to adult size 385 before a subsequent surgery to implant muscle transducers (Figure 1). In the reinnervation surgery, a lateral 386 incision was made posterior-distal to the knee to expose the underlying muscle. Blunt dissection enabled 387 exposure and identification of relevant nerve branches, and the identity of the correct nerve branch was 388 confirmed using an isolated nerve stimulator (SD48, Grass Instruments, Warwick, RI) to visualize contraction in 389 the LG. After pre-placement of single longitudinal throw of 6-0 braided non-absorbable silk (Silk, Ethicon, 390 Somerville, NJ, USA) through a 3 mm nerve section, we transected the nerve branch and sutured to appose the 391 cut nerve endings. Fibrin glue (bovine thrombin in CaCl2, fibrinogen, fibronectin from bovine plasma) was 392 applied over the apposed nerve endings as an additional repair scaffold (Carr et al., 2010; Spotnitz, 2010). We 393 closed the fascia and skin with 3-0 braided absorbable polyglactin (Vicryl, Ethicon, Somerville, NJ, USA). 394 In the immediate post-operative period, bilateral limb posture was visibly more crouched compared to ‘intact’ 395 birds and Achilles tendon tap revealed no stretch reflex response. LG atrophy was qualitatively observed during 396 the first 2 weeks of the recovery period. Within 1 week of surgery, bird activity levels appeared comparable to 397 intact conspecifics, with limb posture partially recovered. From 2-3 weeks onwards, differentiating reinnervated 398 from ‘intact’ birds was not possible from grossly observable limb morphology, posture and gait. We conducted 399 regular treadmill training from 7 weeks after reinnervation surgery. Exercise was started at 7 weeks to ensure 400 synaptic withdrawal of primary afferents before recovery, eliciting a proprioceptive deficit in the self-401 reinnervated LG. In rats, synaptic withdrawal and resulting proprioceptive deficit is minimized if training is 402 initiated on the 3rd day after nerve injury (Boeltz et al. 2013; Brandt et al. 2015). During training and 403 experiments, birds did not stumble or fall with noticeably greater frequency than observed in intact birds and 404 were able to maintain treadmill position over a similar speed range. At the time of muscle recordings, Achilles 405 tendon tap revealed variable latencies of 4943ms (meanS.D., range 10.3-91.5ms). This may reflect variable 406 recovery of intermuscular reflex connectivity, consistent with observations in cats and rats (Boeltz et al. 2013; 407 Brandt et al. 2015; Lyle et al. 2016). In comparison, the Achilles tendon tap reflex latency in intact birds was 408 6.11.2ms (meanS.D), consistent with the mono-synaptic stretch reflex (Nishikawa et al. 2007; Daley and 409 Biewener 2011). 410 Transducer implantation surgery 411 When the birds were 23-28 weeks old (13-16 weeks following bilateral reinnervation surgeries), we performed a 412 2nd surgery for transducer placement, following similar procedures as Daley and Biewener (2003). The surgical 413 field was plucked of feathers and gently cleaned with antiseptic solution (Prepodyne, West Argo, Kansas City, 414 MO, USA). We tunneled transducer leads subcutaneously from a 1–2 cm incision over the synsacrum to a 415 second 4–5 cm incision over the lateral left shank. Sonomicrometry crystals (2.0 mm; Sonometrics Inc., London, 416 Canada) were implanted into the lateral head of the gastrocnemius (LG) along the fascicle axis in the middle 417 1/3rd of the muscle belly. Crystals were placed in small openings using fine forceps, approximately 3–4 mm 418 deep and 15 mm apart. We verified signal quality using an oscilloscope and secured the crystals by closing the 419 overlying muscle fascia and lead wires with separate 4-0 silk sutures (Silk, Ethicon, Somerville, NJ, USA). Next 420 to the crystal pair, we implanted bipolar EMG electrodes constructed from two strands of 38-gauge Teflon-421 coated stainless steel (AS 632, Cooner Wire Co., California, USA) with staggered 1 mm exposed regions spaced 422 1.5 mm apart. Electrodes were placed using sew-through methods and surface silicon anchors (3 x 3 x 2 mm) 423 positioned with a single square knot at the muscle surface-electrode interface (Deban & Carrier, 2002). An “E”-424 type stainless-steel tendon buckle force transducer insulated with a polyurethane coating (Micro-Measurements, 425 Raleigh NC) was implanted on the common gastrocnemius tendon, equipped with a metal foil strain gauge (type 426 FLA-1, Tokyo Sokki Kenkyujo). We connected transducers to a micro-connector plug (15-way Micro-D, Farnell 427 Ltd, Leeds, UK) sutured to the bird’s dorsal synsacrum. 428 Transducer recordings 429 A lightweight shielded cable was used to connect the microconnector to data acquisition systems. 430 Sonomicrometry data were collected via a Sonometrics TRX analog data-acquisition device and PC interface 431 (TRX Series 8, Sonometrics, Ontario, Canada). Crystals were tested before surgery in a saline bath to confirm 432 distances measured by digital caliper matched those measured by the software. Occasional drop-outs and level-433 shift artifacts in the sonomicrometry length signal (arising from variation in signal-to-noise characteristics) were 434 corrected within the Sonometrics software where possible and smoothed using cubic smoothing spline with a 435 tolerance of 0.1 in MATLAB (‘spaps’ function, Mathworks, Inc.; Natick, MA, USA). Tendon buckle signals 436 were fed through a bridge amplifier (Vishay 2120, Micro-Measurements, Raleigh, NC), and EMG signals were 437 amplified and bandpass filtered (10Hz and 3kHz) using GRASS pre-amplifiers (P511, Grass Instruments, 438 Warwick, RI). Signals were recorded at 10kHz using a 16-channel, 16-bit Biopac A/D acquisition device 439 (MP150, BIOPAC systems, Gotleta, CA, USA). Following experiments, birds were euthanized using an 440 intravenous injection of sodium pentobarbital (100 mg kg−1) while under deep isoflurane anesthesia (4%, mask 441 delivery). 442 443 Muscle morphology 444 Postmortem, we recorded the morphology of the muscle and the location of transducers to confirm muscle 445 fascicle and tendon alignment. In the reinnervated cohort (this study) LG mass was 10.72.7 g and total 446 gastrocnemius mass was 26.55.1 g. In the intact cohort (Daley and Biewener 2011), LG mass was 10.24.3 g, 447 and total gastrocnemius mass was 23.57.4 g. These muscle masses are a comparable to those measured from 448 intact guinea fowl in previous studies, representing approximately 0.5% body mass for LG and 1.3% body mass 449 for total gastrocnemius mass (Daley & Biewener 2003; Higham & Biewener 2008). This suggests full recovery 450 from denervation-induced muscle atrophy in the reinnervated cohort. Fascicle lengths for were 17 1 mm and 451 182 mm and pennation angles were 255º and 245º for reinnervated and intact LG, respectively. Crystal 452 alignment relative to the fascicle axis (α) was within 2°, indicating that errors due to misalignment were <0.1%. 453 We calibrated the tendon force buckle in situ post mortem by applying a series of known cyclical loads using a 454 force transducer (model 9203, Kistler, Amherst, MA), which yielded linear least-squares calibration slopes with 455 R2 > 0.97. 456 Level and obstacle terrain conditions 457 We recorded trials on i) uniform level terrain and ii) terrain with repeating 5 cm obstacles, at the same treadmill 458 speeds, as in Daley and Biewener (2011). The treadmill belt (Woodway, Waukesha WI) was slatted black 459 rubber-coated steel with running surface 55.8 cm x 172.7 cm with clearance for obstacles beneath. Obstacles 460 were constructed from styrofoam reinforced with cardboard covered with black neoprene to form a light, stiff 461 surface. Waterproof glue (Shoe Goo, Eugene, OR, USA) secured heavy-duty fabric hook and loop fastener 462 (Velcro, Cheshire, UK) to the obstacle and treadmill surface. Four sequential slats of obstacles produced a 463 20 cm2 continuous obstacle surface. Obstacles were encountered approximately every 4-5 strides, with some 464 variation due to varied stride length and treadmill station keeping. We recorded high-speed video at 250 Hz 465 (Photron, San Diego, CA, USA) for analysis of ankle kinematics, detection of stride timing and statistical coding 466 of strides in relation to obstacle encounters. 467 468 Data Processing 469 We assigned strides categories in relation to the obstacle encounters, using the approach in Daley and Biewener 470 (2011), based on the stride sequence of the instrumented leg: the stride prior to an obstacle contact (S -1), 471 obstacle contact strides (S 0), the stride following obstacle contact (S +1), and strides in flat terrain between 472 obstacles (S +2). All level terrain strides were assigned the same stride category (L). Note, this is simpler stride 473 coding than presented in Gordon and colleagues (2015), which also considered the timing of obstacle encounters 474 by the contralateral leg. Strides in which the contralateral limb had made contact with the obstacle in the 475 previous step are grouped here into the (S +2) category, for simplicity. This does not substantially alter the 476 findings, because the coding was similar between intact and reinnervated birds, and the current analysis is 477 focused on the shifts in LG force-length dynamics related to a direct obstacle encounter by the instrumented leg. 478 Features of LG activation, force-length dynamics and work output were measured, similar to Daley and 479 Biewener (2011). Raw EMG signals were used to calculate myoelectric intensity in time and frequency domains 480 using wavelet decomposition (Daley et al. 2009; Gordon et al. 2015). This was used to calculate total 481 myoelectric intensity per stride (Etot) and mean frequency of muscle activation (Efreq). We calculated fractional 482 fascicle length (L) from sonomicrometry data using mean length in level terrain as a reference length (L o ). Note, 483 however, that L o is not directly related to sarcomere length or optimal length for the isometric force-length curve, 484 which were not measured. Fractional fascicle length was differentiated to obtain fascicle velocity (V, in lengths 485 per second, Ls-1). Shortening strains are negative. We multiplied fascicle velocity (in ms-1) by tendon force (in 486 Newtons, N) to calculate muscle power (Watts), which was integrated through time to calculate total work per 487 stride (Joules, J; with shortening work being positive), and then normalized by muscle mass to obtain mass-488 specific muscle work (Jkg-1). We also recorded muscle (fascicle) length, velocity and force at specified times to 489 evaluate how strain and activation factors influence muscle force and work output. All data processing was 490 completed using MATLAB (Mathworks, Inc.; Natick, MA, USA). 491 Statistics 492 The statistical analysis approach is similar to that used in Gordon and colleagues (2015) to investigate obstacle 493 perturbation responses relative to steady state level terrain strides. We used a linear mixed-effects model 494 ANOVA to test for significant effects of treatment (intact/reinnervated cohorts) and stride category (stride ID: 495 Level, S -1, S 0, S +1, S +2) as fixed categorical factors, with individual (ind) included as a random effect. 496 Statistical analysis was completed in Matlab using ‘fitlme’ and associated functions in the Statistics and Machine 497 Learning Toolbox (Mathworks, Inc.; Natick, MA, USA). Several linear mixed effects models were evaluated: 498 1) 'Y ~ 1 + (1|ind)' 499 2) 'Y ~ 1 + treatment + (1|ind)' 500 3) 'Y ~ 1 + stride_ID + (1|ind)' 501 4) 'Y ~ 1 + stride_ID + treatment + (1|ind)' 502 5) 'Y ~ 1 + stride_ID *treatment + (1|ind)' 503 504 Model 5 (with the interaction term between fixed effects) was used as the final model, because it had the lowest 505 AIC for 13 of 14 of the variables analyzed (AIC, Akaike, 1976). Model 4 had the lowest AIC for mean EMG 506 frequency (Efreq), but the difference between Models 4 and 5 was not significant according to a likelihood ratio 507 test. Therefore, for consistency, Model 5 was used for all variables. Posthoc pairwise comparisons were 508 calculated for the mean difference  95% confidence interval between intact and reinnervated treatment cohorts 509 and for the mean differences between level and obstacle strides categories within each treatment cohort. Pairwise 510 comparisons were calculated after removing the random effect of individual, because the intact and reinnervated 511 datasets came from different cohorts of individuals. We used False Discovery Rate to calculate an adjusted p-512 value threshold to maintain a 5% false positive rate across all statistical tests, including fixed effects tests and 513 post-hoc pairwise comparisons (Benjamini, and Hochberg 1995). 514 515 Acknowledgements This research was supported by NIH grant NIAMS 5R01AR055648 to AAB, grant 516 BB/H005838/1 to MAD from the Biotechnology and Biological Sciences Research Council (BBSRC), and a 517 doctoral training studentship from the BBSRC to JCG supervised by MAD. Thanks to Jennifer A Carr for 518 assistance in the experiments. Thanks to Reviewers N Cowan and L Ting and Reviewing Editor K 519 VijayRaghavan for thoughtful and constructive feedback. 520 521 References 522 Abelew, T. A., Miller, M. D., Cope, T. C. and Nichols, T. R. (2000). Local loss of proprioception results in disruption of 523 interjoint coordination during locomotion in the cat. Journal of Neurophysiology 84, 2709-2714. 524 Alvarez, F. J., Titus-Mitchell, H. E., Bullinger, K. L., Kraszpulski, M., Nardelli, P. and Cope, T. C. (2011). Permanent 525 central synaptic disconnection of proprioceptors after nerve injury and regeneration. I. Loss of VGLUT1/IA 526 synapses on motoneurons. Journal of Neurophysiology 106, 2450-2470. 527 Askew, G. N. and Marsh, R. L. (1998). Optimal shortening velocity (V/Vmax) of skeletal muscle during cyclical 528 contractions: length-force effects and velocity-dependent activation and deactivation. Journal of Experimental 529 Biology 201, 1527-1540. 530 Askew, G. N. and Marsh, R. L. (2001). The mechanical power output of the pectoralis muscle of blue-breasted quail 531 (Coturnix chinensis): the in vivo length cycle and its implications for muscle performance. Journal of 532 Experimental Biology 204, 3587-3600. 533 Azizi, E., Brainerd, E. L. and Roberts, T. J. (2008). Variable gearing in pennate muscles. Proceedings of the National 534 Academy of Sciences of the United States of America 105, 1745-1750. 535 Benjamini, Y. and Hochberg, Y. (1995) Controlling the false discovery rate: a practical and powerful approach to multiple 536 testing. Journal of the Royal Statistical Society: series B (Methodological), 57, 289-300. 537 Birn-Jeffery, A.V., Hubicki, C.M., Blum, Y., Renjewski, D., Hurst, J., and Daley, M.A. (2014). Don't break a leg: Running 538 birds from quail to ostrich prioritize leg safety and economy in uneven terrain. Journal of Experimental Biology 539 217, 3786-3796. 540 Brandt, J., Evans, J.T., Mildenhall, T., Mulligan, A., Konieczny, A., Rose, S.J. and English, A.W. (2015) Delaying the onset 541 of treadmill exercise following peripheral nerve injury has different effects on axon regeneration and motoneuron 542 synaptic plasticity. Journal of neurophysiology, 113, 2390-2399. 543 Boeltz, T., Ireland, M., Mathis, K., Nicolini, J., Poplavski, K., Rose, S.J., Wilson, E. and English, A.W. (2013) Effects of 544 treadmill training on functional recovery following peripheral nerve injury in rats. Journal of neurophysiology, 109, 545 2645-2657. 546 Brown, I. E., & Loeb, G. E. (2000). A reductionist approach to creating and using neuromusculoskeletal models. In 547 Biomechanics and Neural Control of Posture and Movement (pp. 148-163). Springer New York. 548 Bullinger K.L., Nardelli P., Pinter M.J., Alvarez F.J., Cope T.C. (2011). Permanent central synaptic disconnection of 549 proprioceptors after nerve injury and regeneration. II. Loss of functional connectivity with motoneurons. Journal of 550 Neurophysiology 106: 2471–2485. 551 Carr, J. A., Chao, L. and Biewener, A.A. (2010). The effects of denervation and self-reinnervation in the guinea fowl lateral 552 gastrocnemius Online abstract: American Society of Biomechanics, 26(24), 22. 553 Chang, Y. H., Auyang, A. G., Scholz, J. P. and Nichols, T. R. (2009). Whole limb kinematics are preferentially conserved 554 over individual joint kinematics after peripheral nerve injury. Journal of Experimental Biology 212, 3511-3521. 555 Clark, A. J. and Higham, T. E. (2011). Slipping, sliding and stability: locomotor strategies for overcoming low-friction 556 surfaces. Journal of Experimental Biology, 214, 1369-1378. 557 Clemente, C. J., & Richards, C. (2012). Determining the influence of muscle operating length on muscle performance 558 during frog swimming using a bio-robotic model. Bioinspiration & Biomimetics, 7(3), 036018. 559 Cohen, A. H. (1992). The role of heterarchical control in the evolution of central pattern generators. Brain, Behavior and 560 Evolution, 40(2–3), 112–124. 561 Cope, T. C., Bonasera, S. J. and Nichols, T. R. (1994). Reinnervated muscles fail to produce stretch reflexes. Journal of 562 Neurophysiology, 71, 817-820. 563 Cox S.M., Easton K.L., Lear M.C., Marsh R.L., Delp S.L., Rubenson J. (2019). The interaction of compliance and 564 activation on the force-length operating range and force generating capacity of skeletal muscle: a computational 565 study using a guinea fowl musculoskeletal model. Integrative Organismal Biology, 2019 Sep 3;1(1):obz022. 566 Daley, M. A. and Biewener, A. A. (2003). Muscle force-length dynamics during level versus incline locomotion: a 567 comparison of in vivo performance of two guinea fowl ankle extensors. Journal of Experimental Biology. 206, 568 2941-2958. 569 Daley, M. A. and Biewener, A. A. (2011). Leg muscles that mediate stability: mechanics and control of two distal extensor 570 muscles during obstacle negotiation in the guinea fowl. Philosophical transactions of the Royal Society of London. 571 Series B, Biological sciences, 366, 1580-1591. 572 Daley, M. A., & Birn-Jeffery, A. (2018). Scaling of avian bipedal locomotion reveals independent effects of body mass and 573 leg posture on gait. Journal of Experimental Biology, 221(10), jeb152538. 574 Daley, Monica A; Gordon, Joanne C; Biewener, Andrew A (2020), Dataset for 'Tuning of feedforward control enables 575 stable muscle force-length dynamics after loss of autogenic proprioceptive feedback', v6, UC Irvine, Dataset, 576 https://doi.org/10.7280/D11H49 577 Daley, M. A., Voloshina, A., & Biewener, A. A. (2009). The role of intrinsic muscle mechanics in the neuromuscular 578 control of stable running in the guinea fowl. The Journal of Physiology, 587(11), 2693-2707. 579 https://doi.org/10.7280/D11H49 Deban, S. M. and Carrier, D. R. (2002). Hypaxial muscle activity during running and breathing in dogs. Journal of 580 Experimental Biology, 205, 1953-1967. 581 Dick, T. J., Punith, L. K., & Sawicki, G. S. (2019). Humans falling in holes: Adaptations in lower-limb joint mechanics in 582 response to a rapid change in substrate height during human hopping. Journal of the Royal Society Interface, 583 16(159), 20190292. 584 Donelan, J. M. and Pearson, K. G. (2004). Contribution of force feedback to ankle extensor activity in decerebrate walking 585 cats. Journal of Neurophysiology, 92, 2093-2104. 586 Dorward, P. K. (1970). Response characteristics of muscle afferents in the domestic duck. Journal of Physiology, 211(1), 1–587 17. 588 Drew, T. and Marigold, D.S. (2015) Taking the next step: cortical contributions to the control of locomotion. Current 589 opinion in neurobiology, 33:.25-33 590 Edman, K. A. P. (1975). Mechanical deactivation induced by active shortening in isolated muscle fibres of the frog. Journal 591 of Physiology, 246, 255–275 592 Edman, K. A., Elzinga, G., & Noble, M. I. (1978). Enhancement of mechanical performance by stretch during tetanic 593 contractions of vertebrate skeletal muscle fibres. Journal of Physiology, 281(1), 139-155. 594 Edman, K. (1980). Depression of mechanical performance by active shortening during twitch and tetanus of vertebrate 595 muscle fibres. Acta Physiologica Scandinavica, 109(1), 15–26. 596 Edman, K. A. P. (2012). Residual force enhancement after stretch in striated muscle. A consequence of increased 597 myofilament overlap? Journal of Physiology, 590(6), 1339-1345. 598 Ferris, D. P., Liang, K., & Farley, C. T. (1999). Runners adjust leg stiffness for their first step on a new running surface. 599 Journal of Biomechanics, 32(8), 787–794. 600 Frigon, A. and Rossignol, S. (2006). Experiments and models of sensorimotor interactions during locomotion. Biological 601 Cybernetics 95, 607-627. 602 Gatesy, S. M., & Biewener, A. A. (1991). Bipedal Locomotion—Effects of Speed, Size and Limb Posture in Birds and 603 Humans. Journal of Zoology, 224, 127–147. 604 Gordon, J. C., Rankin, J. W., & Daley, M. A. (2015). How do treadmill speed and terrain visibility influence neuromuscular 605 control of guinea fowl locomotion? Journal of Experimental Biology, 218(19), 3010-3022. doi:10.1242/jeb.104646 606 Gordon, T. and Stein, R. B. (1982). Time course and extent of recovery in reinnervated motor units of cat triceps surae 607 muscles. Journal of Physiology, 323, 307-323. 608 Grillner, S. (2011). Control of Locomotion in Bipeds, Tetrapods, and Fish. Comprehensive Physiology 2011, Supplement 2: 609 Handbook of Physiology, The Nervous System, Motor Control: (1179-1236). 610 Haiden, G.J. and Awad, E.A., (1981). The ultrastructure of the avian Golgi tendon organ. The Anatomical Record, 200(2), 611 153-161. 612 Heglund, N. C., Cavagna, G. A., & Taylor, C. R. (1982). Energetics and mechanics of terrestrial locomotion. III. Energy 613 changes of the centre of mass as a function of speed and body size in birds and mammals. Journal of Experimental 614 Biology, 97(1), 41–56. 615 Herzog, W. (2004). History dependence of skeletal muscle force production: Implications for movement control. Human 616 Movement Science, 23(5), 591–604. 617 Herzog, W. (2014). Mechanisms of enhanced force production in lengthening (eccentric) muscle contractions. Journal of 618 Applied Physiology, 116(11), 1407-1417. 619 Ijspeert, A. J. (2014). Biorobotics: Using robots to emulate and investigate agile locomotion. Science, 346(6206), 196–203. 620 Josephson, R. K. (1999). Dissecting muscle power output. Journal of Experimental Biology 202, 3369-3375. 621 Kuo, A. D. (2002). The relative roles of feedforward and feedback in the control of rhythmic movements. Motor control 6, 622 129-145. 623 Lam, T., and Pearson K.G. (2002) The role of proprioceptive feedback in the regulation and adaptation of locomotor 624 activity. In Sensorimotor Control of Movement and Posture (pp. 343–355). Springer, Boston, MA. 625 Loeb, G. E., Brown, I. E., & Cheng, E. J. (1999). A hierarchical foundation for models of sensorimotor control. 626 Experimental Brain Research, 126(1), 1-18. 627 Lyle M.A., Prilutsky B.I., Gregor R.J., Abelew T.A., Nichols TR. (2016) Self-reinnervated muscles lose autogenic length 628 feedback, but intermuscular feedback can recover functional connectivity. Journal of Neurophysiology. 629 116(3):1055-67. 630 Lyle, M. A., & Nichols, T. R. (2018). Patterns of intermuscular inhibitory force feedback across cat hindlimbs suggest a 631 flexible system for regulating whole limb mechanics. Journal of Neurophysiology, 119(2), 668–678. 632 Maas, H., Prilutsky, B. I., Nichols, T. R. and Gregor, R. J. (2007). The effects of self-reinnervation of cat medial and lateral 633 gastrocnemius muscles on hindlimb kinematics in slope walking. Experimental Brain Research 181, 377-393. 634 Maier, A. (1992). The avian muscle spindle. Anatomy and Embryology, 186(1), 1–25. 635 Marigold, D.S. and Drew, T. (2017) Posterior parietal cortex estimates the relationship between object and body location 636 during locomotion. eLife, 6, p.e28143. 637 More, H. L., & Donelan, J. M. (2018). Scaling of sensorimotor delays in terrestrial mammals. Proceedings of the Royal 638 Society B: Biological Sciences, 285(1885), 20180613. 639 More, H. L., Hutchinson, J. R., Collins, D. F., Weber, D. J., Aung, S. K., & Donelan, J. M. (2010). Scaling of sensorimotor 640 control in terrestrial mammals. Proceedings of the Royal Society B: Biological Sciences, 277(1700), 3563–3568. 641 Moritz, C. T., & Farley, C. T. (2004). Passive dynamics change leg mechanics for an unexpected surface during human 642 hopping. Journal of Applied Physiology, 97(4), 1313–1322. https://doi.org/10.1152/japplphysiol.00393.2004 643 Nelson, F. E., Gabaldón, A. M., & Roberts, T. J. (2004). Force–velocity properties of two avian hindlimb muscles. 644 Comparative Biochemistry and Physiology Part A: Molecular & Integrative Physiology, 137(4), 711–721. 645 Nichols, T.R. (1989) The organization of heterogenic reflexes among muscles crossing the ankle joint in the decerebrate cat. 646 The Journal of Physiology, 410(1), 463-477. 647 Nishikawa, K., Biewener, A. A., Aerts, P., Ahn, A. N., Chiel, H. J., Daley, M. A., Daniel, T. L., Full, R. J., Hale, M. E., 648 Hedrick, T. L. et al. (2007). Neuromechanics: an integrative approach for understanding motor control. Integrative 649 and Comparative Biology 47, 16-54. 650 Nishikawa, K. C., Monroy, J. A., Uyeno, T. E., Yeo, S. H., Pai, D. K., & Lindstedt, S. L. (2012). Is titin a ‘winding 651 filament’? A new twist on muscle contraction. Proceedings of the Royal Society of London B: Biological Sciences, 652 279, 981-990. 653 Nishikawa, K. C., Monroy, J. A., & Tahir, U. (2018). Muscle function from organisms to molecules. Integrative and 654 Comparative Biology, 58(2), 194–206. 655 Pearson, K. (2000). Motor systems. Current Opinion in Neurobiology, 10(5), 649–654. 656 Pearson, K., & Gramlich, R. (2010). Updating neural representations of objects during walking. Annals of the New York 657 Academy of Sciences, 1198(1), 1-9. 658 Pearson, K. G., Misiaszek, J. E. and Hulliger, M. (2003). Chemical ablation of sensory afferents in the walking system of 659 the cat abolishes the capacity for functional recovery after peripheral nerve lesions. Experimental Brain Research 660 150, 50-60. 661 Poore, S. O., Ashcroft, A., Sanchez-Haiman, A., & Goslow, G. (1997). The contractile properties of the M. 662 supracoracoideus in the pigeon and starling: A case for long-axis rotation of the humerus. Journal of Experimental 663 Biology, 200(23), 2987–3002. 664 Potocanac, Z., de Bruin, J., van der Veen, S., Verschueren, S., van Dieën, J., Duysens, J. and Pijnappels, M. (2014) Fast 665 online corrections of tripping responses. Experimental brain research, 232(11): 3579-3590. 666 Prochazka, A. and Ellaway, P. (2012). Sensory systems in the control of movement. Comprehensive Physiology 2, 2615-667 2627. 668 Proske, U., & Gandevia, S. C. (2012). The proprioceptive senses: their roles in signaling body shape, body position and 669 movement, and muscle force. Physiological Reviews, 92(4), 1651-1697. 670 Richards, C. T. (2011). Building a robotic link between muscle dynamics and hydrodynamics. Journal of Experimental 671 Biology, 214(14), 2381–2389. 672 Roberts, T. J., Marsh, R. L., Weyand, P. G. and Taylor, C. R. (1997). Muscular force in running turkeys: the economy of 673 minimizing work. Science 275, 1113-1115. 674 Robertson, B. D., & Sawicki, G. S. (2015). Unconstrained muscle-tendon workloops indicate resonance tuning as a 675 mechanism for elastic limb behavior during terrestrial locomotion. Proceedings of the National Academy of 676 Sciences, 112(43), E5891–E5898. 677 Rode, C., Siebert, T., & Blickhan, R. (2009). Titin-induced force enhancement and force depression: A ‘sticky-spring’ 678 mechanism in muscle contractions? Journal of Theoretical Biology, 259(2), 350-360. 679 Ross, K. T. and Nichols, T. R. (2009). Heterogenic feedback between hindlimb extensors in the spontaneously locomoting 680 premammillary cat. Journal of Neurophysiology 101, 184-197. 681 Rossignol, S., Dubuc, R. and Gossard, J. P. (2006). Dynamic sensorimotor interactions in locomotion. Physiological 682 Reviews 86, 89-154. 683 Roth, E., Sponberg, S., & Cowan, N. (2014). A comparative approach to closed-loop computation. Current Opinion in 684 Neurobiology, 25, 54–62. 685 Safavynia, S.A. and Ting, L.H. (2013) Long-latency muscle activity reflects continuous, delayed sensorimotor feedback of 686 task-level and not joint-level error. Journal of neurophysiology, 110(6),1278-1290. 687 Sawicki, G.S., Robertson, B.D., Azizi, E. and Roberts, T.J., (2015). Timing matters: tuning the mechanics of a muscle–688 tendon unit by adjusting stimulation phase during cyclic contractions. Journal of Experimental Biology, 218(19), 689 3150-3159. 690 Sharp, A. A., & Bekoff, A. (2015). Pyridoxine treatment alters embryonic motility in chicks: Implications for the role of 691 proprioception. Developmental Psychobiology, 57(2), 271-277. 692 Sherrington, C. S., & Laslett, E. E. (1903). Observations on some spinal reflexes and the interconnection of spinal segments. 693 Journal of Physiology, 29(1), 58–96. 694 Sherrington, C. S. (1910). Remarks on the reflex mechanism of the step. Brain 33.1 1-25. 695 Spotnitz, W. D. (2010). Fibrin sealant: past, present, and future: a brief review. World Journal of Surgery 34, 632-634. 696 Taylor, C. R., Heglund, N. C., & Maloiy, G. (1982). Energetics and mechanics of terrestrial locomotion. I. Metabolic energy 697 consumption as a function of speed and body size in birds and mammals. Journal of Experimental Biology, 97(1), 698 1–21. 699 Todorov, E. (2004). Optimality principles in sensorimotor control. Nature Neuroscience, 7(9), 907-915. 700 Vannucci, B., Santosa, K. B., Keane, A. M., Jablonka‐Shariff, A., Lu, C., Yan, Y., … Snyder‐Warwick, A. K. (2019). What 701 is Normal? Neuromuscular junction reinnervation after nerve injury. Muscle & Nerve.60:604–612 702 Wagner, M. J., & Smith, M. A. (2008). Shared internal models for feedforward and feedback control. Journal of 703 Neuroscience, 28(42), 10663-10673. 704 Wolpert, D. M., Diedrichsen, J., & Flanagan, J. R. (2011). Principles of sensorimotor learning. Nature Reviews 705 Neuroscience, 12(12), 739-751. 706 Yakovenko, S., Gritsenko, V. and Prochazka, A. (2004). Contribution of stretch reflexes to locomotor control: a modeling 707 study. Biological Cybernetics 90, 146-155. 708 Yeo, S. H., Monroy, J. A., Lappin, A. K., Nishikawa, K. C., & Pai, D. K. (2013). Phenomenological models of the dynamics 709 of muscle during isotonic shortening. Journal of Biomechanics, 46(14), 2419-2425. 710 711 Figure legends and Tables 712 Figure 1. Protocol for bilateral self-reinnervation of the lateral gastrocnemius (LG), followed by transducer implantation 713 for in vivo recordings of muscle force (tendon buckle), fascicle length (sonomicrometry crystals) and electromyographic 714 activity (EMG). 715 Figure 2. Muscle trajectories (mean 95% ci) during obstacle negotiation for intact and reinnervated lateral 716 gastrocnemius (iLG: blue, top, rLG: orange, bottom), averaged over the stride cycle from mid-swing to mid-swing. 717 Averages are shown for a 4-stride sequence in obstacle terrain, with steady level terrain means as a reference, in grey. The 718 shaded box is an obstacle encounter (S 0). Obstacle terrain strides are coded as in Daley and Biewener 2011, for strides 719 preceding (S -1), on (S 0) and following obstacle contact (S +1), with S +2 including all other strides between obstacles. 720 Trajectories are fractional muscle fascicle length (top), muscle-tendon force (middle) and rectified myoelectric activity 721 (EMG). Triangles indicate the timing of foot-ground contact (grey: level terrain, black: obstacle terrain). Example data is 722 shown from one individual in each treatment cohort. See Figure 2- figure supplement 1 for details on stride-cycle cutting 723 and categorization in an example stride sequence in obstacle terrain. 724 Figure 2-figure supplement 1. Example 6-stride sequence of in vivo muscle recordings of the reinnervated lateral 725 gastrocnemius (rLG) in the right leg, running at 1.7ms-1 on the obstacle treadmill. Muscle length (top, orange), force 726 (bottom, black) and activation (rectified EMG, bottom, orange) are shown, with triangles indicating the time of foot-727 ground contact, a shaded box indicating an obstacle encounter stride (S 0), and vertical lines indicating the mid-swing cut 728 points between stride cycles. Stride categories were identified from video. Grey silhouettes at the top illustrate the leg 729 posture at the time of foot contact. Strides cycles were cut based on a minimum in muscle-tendon force after it was low-730 pass filtered with a 6th order Butterworth filter with a cutoff frequency of 3.4Hz. This resulted in a sinusoidal trajectory 731 with a mid-swing minimum, which was confirmed against video to correspond to when the swing leg crossed vertical. 732 Note that between the last two strides, the contralateral leg stepped on the obstacle, leading to a downward step of the 733 instrumented leg in the final stride. For simplicity, these strides are group with the ‘mid-flat’ strides S+2 (as in Daley and 734 Biewener 2011) because the focus of the current analysis is the direct response to the obstacle encounter (S 0). 735 Figure 3. Force-length work loops for iLG (top) and rLG (bottom), for a single individual from each treatment cohort 736 (intact/reinnervated, mean ± 95% ci). Level mean in grey and obstacle strides in colored lines (iLG: blue, rLG: orange). 737 Stride categories as in Figure 2, where the shaded box is an obstacle encounter (S 0). Triangles indicate the timing of foot-738 ground contact and arrows indicate the direction of the work loop, with a counter-clockwise loop corresponding to net 739 positive muscle work. 740 Figure 4. A) Distributions of muscle total work output (Wnet), peak force (Fpk), velocity and length at peak force (VpkF, 741 LpkF) across stride categories for iLG (blue) and rLG (orange). Circles indicate group means. Lines connect means 742 between stride categories, to highlight the shifts in relation to obstacle encounters (S 0). B) Pairwise mean differences 743 (mean ± 95% ci) for fixed effect categories, between intact and reinnervated treatment cohorts (grey bar), and between 744 obstacle stride categories compared to level means, within treatment cohorts (colored bars). See Tables 1& 2 for full 745 statistics results and summary data. 746 Figure 5. A) Distributions of total intensity EMG activity (Etot), duration of activity (Edur) and mean frequency of activity 747 (Efreq) across stride categories for iLG (blue) and rLG (orange). Circles indicate group means. Lines connect means 748 between stride categories, to highlight the shifts in relation to obstacle encounters (S 0). B) Pairwise mean differences 749 (mean ± 95% ci) for fixed effects, as presented as in Figure 4. See Tables 1& 2 for full statistics results and summary data. 750 Figure 5-figure supplement 1. Distribution of EMG activation frequency (mean ± 95% ci) for intact LG (top, blue) and 751 reinnervated LG (bottom, orange), during level running (dark grey lines) and obstacle encounters (S 0, colored lines). 752 Note the shift in peak frequency of EMG activity in the reinnervated LG for both level terrain and obstacle strides, 753 suggesting recruitment of faster motor units. 754 Figure. 6. Phase relationship (Ephase) between length and EMG activation A) Average steady-state length and activation 755 trajectories for iLG and rLG in level terrain, aligned in time based on peak length during the swing phase, before foot-756 substrate contact. Black dot and vertical dashed line indicate the time of peak fascicle length. Triangles indicate timing of 757 foot contact. B) Pairwise mean differences in Ephase (mean ± 95% ci) between intact and reinnervated treatment cohorts 758 (grey), and obstacle stride categories compared to level terrain within each cohort (colored bars). Ephase is reported in the 759 ANOVA tables as a fraction of the stride cycle but is reported in milliseconds here. 760 Figure. 7. Deviations from steady state in the stride cycle trajectories of muscle length, force and activation, between 761 obstacle strides (S 0) and level strides (grand mean ±95% ci across individuals). The horizontal zero line indicates no 762 difference from steady state in S 0. The stride cycle is from mid-swing to mid-swing, as in Figure 2. A black asterisk (*) 763 indicates the first timepoint in each trajectory that differs significantly from the level mean. The dashed vertical line and 764 arrow indicating ‘anticipatory increase’ highlights a significant increase in EMG that starts before deviations length and 765 force in S 0. In A) (iLG), solid vertical lines and yellow fill indicates a 2nd period of significantly increased EMG in late 766 stance that correlates with increased fascicle length and force, suggesting a reflex response. In B) (rLG), the anticipatory 767 increase in EMG is present; however, wide confidence intervals for EMG in late stance indicates inconsistent patterns of 768 activity across individuals, despite similar increases in length and force as iLG. This suggests disrupted autogenic 769 feedback and idiosyncratic heterogenic feedback patterns across individuals (Figure. 7-figure supplement 1). 770 Figure. 7-figure supplement 1. Average stride cycle trajectories (mean±95% ci) for fascicle length, muscle-tendon force 771 and myoelectric activity (EMG) during obstacle encounter strides (S 0, orange) compared to the level terrain averaged 772 (grey), for all birds with reinnervated LG. Although the deviations in length and force during obstacle encounters is 773 qualitatively similar across individuals, the shifts in EMG activation in the latter half of stance varies substantially across 774 individuals, with some individuals showing reflex inhibition of activity (Ind 13, Ind 25) and others showing reflex 775 excitation. 776 Figure. 8. Ankle kinematics in guinea fowl with intact and reinnervated lateral gastrocnemius (LG). A) Example ankle 777 joint angle trajectories for a bird with intact LG (blue, top) and a bird with reinnervated LG (orange, below), running in 778 obstacle terrain (solid lines) with level terrain (grey dashed lines). B) Pairwise mean differences (mean ± 95% ci) between 779 intact and reinnervated treatment cohorts (grey), and obstacle stride categories compared to level terrain within each 780 treatment cohort (intact: blue, reinnervated: orange). In obstacle strides (S 0, shaded box), the ankle is more flexed at foot 781 contact in reinnervated compared to intact birds. Reinnervated birds show a shorter stride period in S -1, preceding the 782 obstacle encounter, suggesting increased anticipatory preparation. (See Figure. 8- source data 1 for statistical results on 783 ankle angle at the time of foot contact). 784 Figure. 8- source data 1. ANOVA results for ankle angle at time of foot contact. F-statistics, p-values and posthoc 785 pairwise comparisons for linear mixed effect model ANOVA with fixed effects of treatment cohort (treatment: intact, 786 reinnervated) and stride category (stride ID) and the interaction treatment x stride ID. Posthoc pairwise mean differences 787 (mean ± 95% ci) are shown between intact and reinnervated treatment cohorts (left), and between obstacle stride 788 categories compared to level stride means, within treatment cohorts (intact/reinnervated). 789 Figure 9. Schematic illustration of neuromechanical control system, indicating hypothesized mechanisms for the 790 observed changes in self-reinnervated lateral gastrocnemius of guinea fowl. Green text indicates the in vivo 791 experimental measures used to infer sensorimotor control mechanisms. Differences in muscle dynamics between intact 792 and reinnervated cohorts suggest that guinea fowl use a combination of feedforward and intrinsic mechanical mechanisms 793 to compensate for disrupted proprioceptive reflexes, suggesting interconnected plasticity of neural and musculoskeletal 794 mechanisms in the recovery from nerve injury. 795 796 797 Table 1. F-statistics for linear mixed effect model ANOVA with fixed effects of treatment cohort (treatment: intact, 798 reinnervated) and stride category (stride ID) and the interaction treatment x stride ID on measures of muscle contraction 799 mechanics and activation. Bolding indicates statistical significance using FDR corrected threshold (p <= 0.0263, see 800 Methods). Degrees of freedom for fixed effects were treatment = 1, stride ID = 4, interaction = 4, and error = 2529. See 801 Table 1 - source data 1 for p-values. 802 F-statistic Variable treatment stride ID: interaction Wnet 2.44 172.04 7.44 Fpk 0.52 30.12 61.73 LpkF 1.52 238.03 39.52 VpkF 11.43 28.09 8.73 Tforce 0.27 99.09 18.32 Tstride 0.10 12.17 17.82 Etot 0.01 49.05 3.51 Efreq 3.93 8.71 1.85 Ephase 5.72 2.34 7.64 Edur 10.02 8.86 10.69 803 Table 1 - source data 1. P-values linear mixed effect model ANOVA with fixed effects of treatment cohort (treatment: 804 intact, reinnervated) and stride category (stride ID) and the interaction treatment x stride ID. 805 Table 2. Pairwise mean differences (mean ± 95% ci) between intact and reinnervated treatment cohorts (left), 806 and between obstacle stride categories compared to level stride means, within treatment cohorts 807 (intact/reinnervated). Bolding indicates statistical significance using FDR corrected threshold (p <= 0.0263). See 808 Table 2- source data 1for p-values. 809 810 Variable Treatment cohort Intact Reinnervated S -1 S 0 Str +1 S +2 S -1 S 0 Str +1 S +2 Wnet -0.47±0.59 -0.91±0.60 3.60±0.57 0.00±0.59 -0.86±0.34 0.34±0.55 3.88±0.60 0.49±0.70 -0.19±0.46 Fpk -0.02±0.06 -0.04±0.07 0.17±0.06 0.04±0.07 -0.04±0.04 0.09±0.06 0.62±0.07 0.10±0.08 0.05±0.05 LpkF -0.02±0.04 -0.02±0.01 0.10±0.01 -0.01±0.01 -0.02±0.01 -0.05±0.01 0.14±0.01 -0.06±0.02 -0.04±0.01 VpkF -1.89±1.10 0.07±0.45 -1.26±0.43 0.00±0.45 0.11±0.26 0.60±0.42 -0.39±0.46 0.63±0.53 0.81±0.35 Tforce 0.02±0.07 -0.01±0.02 0.11±0.02 -0.01±0.02 -0.01±0.01 -0.07±0.02 0.07±0.02 -0.06±0.02 -0.06±0.02 Tstride 0.00±0.02 -0.01±0.02 0.03±0.02 -0.02±0.02 0.01±0.01 -0.07±0.02 0.04±0.02 -0.04±0.03 -0.03±0.02 Etot 0.01±0.22 0.12±0.18 0.76±0.18 0.11±0.18 0.12±0.11 0.16±0.17 0.74±0.19 0.39±0.22 0.09±0.14 Efreq 55.70±55.12 3.05±9.44 -15.05±9.04 -1.46±9.34 -0.46±5.43 -2.86±8.75 -22.06±9.52 -6.81±11.04 -7.68±7.28 Ephase -0.06±0.05 -0.01±0.02 0.00±0.02 0.00±0.02 -0.01±0.01 0.01±0.02 0.04±0.02 -0.01±0.02 0.00±0.01 Edur 0.08±0.05 0.00±0.03 0.04±0.02 -0.01±0.02 0.00±0.01 -0.04±0.02 -0.01±0.03 0.00±0.03 -0.03±0.02 811 Table 2- source data 1. P-values for posthoc pairwise mean differences between intact and reinnervated 812 treatment cohorts (left column) and between obstacle stride categories compared to the level terrain means, 813 within treatment cohorts (intact/reinnervated). 814 Lateral gastrocnemius incision site nerve cut, repaired tendon buckle sono crystals EMGlateral gastrocnemius (LG) bilateral dennervation, immediate repair 13-16 weeks (re-innervation) muscle transducers implanted motor recovered, sensory deficit Figure 1 Intact Lateral Gastrocnemius (iLG) Time (s) 0 1 2 0 1 0.8 1.0 1.2 foot contact S -1 S 0 S +1 S +2 M -T F or ce (F /F m ax ,L ev el ) Le ng th (L /L m ea n) EM G (E /E m ax ,L ev el ) 0 1 Time (s) 0 1 2 0.8 1.0 1.2 M -T F or ce (F /F m ax ,L ev el ) Le ng th (L /L m ea n) EM G (E /E m ax ,L ev el ) 200ms foot contact obstacle contact S -1 S 0 S +1 S +2 Reinnervated Lateral Gastrocnemius (rLG) level terrain obstacle contact level terrain 200ms Figure 2 M -T F or ce (F /F m ax ) Le ng th (L /L m ea n) Time500 ms foot contact S -1 S 0 S +1 S +2(+) S +2(+) obstacle contact downward step S +2(+) Figure 2-figure supplement 1 0.5 1.5 0 1 2 0.5 1 1.5 0.5 1 1.5 0.5 1 1.5 0 1 2 1 Intact lateral gastrocnemius (iLG) foot contact level Reinnervated lateral gastrocnemius (iLG) S -1 S +1 S +2S 0 Fo rc e (F /F pe ak ,le ve l) Fo rc e (F /F pe ak ,le ve l) Fractional length(L/Lmean,level) obstacle foot contact level obstacle Figure 3 0 10 20 -2 0 2 4 0 1 2 3 0 0.2 0.4 0.6 -5 0 5 -4 -2 0 2 0.8 1 1.2 S +2S +1S 0 S -1 ste ad y lev el S -1 S 0 S +1 S +2 -0.1 0 0.1 Wnet (Jkg-1) Fpk (F/Fc) VpkF (Lcs-1) LpkF (L/Lc) iLG rLG ANOVA posthoc pairwise mean differences Int ac t- Rein ne rva ted obstacle terrain mean difference from level obstacle terrain Distributions by stride category obstacle steady level Net work Peak force Velocity at peak force Length at peak force A B Figure 4 0 2 4 6 0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 -0.1 -0.05 0 0.05 0.1 200 400 -20 0 20 40 S +2S +1S 0 S -1 ste ad y lev el S -1 S 0 S +1 S +2 iLG rLG ANOVA posthoc pairwise mean differences Int ac t- Rein ne rva ted obstacle terrain mean difference from level obstacle terrain Distributions by stride category obstacle steady level Etot (E/Elevel) Efreq (Hz) Edur (T/Tstride, level) Integrated EMG EMG duration EMG mean frequency A B Figure 5 62 331 804 0 0.1 0.2 Intact LG Wavelet Center Frequency (Hz) 0 0.1 0.2 Reinnervated LG 62 331 804 In te ns ity (m V2 ) Figure 5-figure supplement 1 0.0 0.5 1.0 0.8 1.0 1.2 50ms Time Le ng th (L /L m ea n, L) EM G (E /E m ax ,L ) S 0S -1 S +1 Int ac t- Rein ne rva ted 15 0 -15 -30 ANOVA posthoc pairwise mean differences ph as e sh ift (m s) A B obstacle terrain mean difference from level iLG rLG Length-EMG phase in level terrain foot contact * * S +2 phase advance Figure 6 Obstacle stride deviations from steady-state level strides EM G (E S0 - E c) Fraction of Stride Cycle 0 0.5 1 Fo rc e (F S0 - F c) 0 0.1 0.2 Le ng th (L S0 - L c) * * * * 0 0.2 0.4 0.6 0.8 1 -0.4 0 0.4 * 0 0.2 0.4 0.6 0.8 1 * A B reflex Intact LG Reinnervated LG reflex absent anticipatory increase anticipatory increase Figure 7 0 0.5 1 1.5 2 0 0.2 0 0.5 1 1.5 2 0.8 1 1.2 1.4 Ind 23 0 0.5 1 1.5 0 0.2 0 0.5 1 1.5 2 0.8 1 1.2 1.4 Ind 13 0 0.5 1 1.5 2 0 0.2 0 0.5 1 1.5 2 0.8 1 1.2 1.4 Ind 21 0 0.5 1 1.5 2 0 0.2 0 0.5 1 1.5 0.8 1 1.2 1.4 Ind 24 0 0.5 1 1.5 0 0.2 0 0.5 1 1.5 2 0.5 1 1.5 Ind 25 0 0.2 0 0.5 1 1.5 2 0 1 2 0.8 1 1.2 1.4 Ind 3 M -T F or ce (F /F m ax ,L ) Le ng th (L /L m ea n, L) EM G (E /E m ax ,L ) Time (s) Figure 7-figure supplement 1 Stride durationAnkle angle at contact An kl e jo in t a ng le (° ) 80 120 160 Time (s) 80 120 160 A B ANOVA posthoc pairwise mean differences obstacle level ** S -1 S 0 S +1 Int ac t- Rein ne rva ted obstacle terrain difference from level S -1 S 0 S +1 Int ac t- Rein ne rva ted obstacle terrain difference from level -15 0 15 Ti m e (m s) -30-40 -20 0 An gl e (° ) obstacle level Ankle kinematics 200ms Reinnervated Intact Figure 8 tim e sc al e ~1-3 steps 100ms 30ms 5ms brain muscle-dynamics body-dynamics intrinsic mechanics neural networks spinal networks vision, balance, hearing proprioception leg & joint mechanics muscle force, length, velocity EMG activity descending drive CPG flexext flexext environment ↑ EMG frequency ↑ shortening velocity ↑ force development ↑ ankle flexion +/- heterogenic reflexes (idiosyncratic by individual) ↑ feedforward drive ↑ early EMG activation - lost regulation of EMG duration Figure 9 Cover Page Article File Figure 1 Figure 2 Figure 2 figure supplement 1 Figure 3 Figure 4 Figure 5 Figure 5 figure supplement 1 Figure 6 Figure 7 Figure 7 figure supplement 1 Figure 8 Figure 9