Motor Learning And Savings Of Adaptive Mediolateral Control During Split-Belt Walking

11 Active control of frontal plane mechanics regulates balance in destabilizing environments, 12 such as during asymmetric split -belt walking. Compared to sagittal plane mechanics , 13 mediolateral (ML) kinematic and kinetic adaptations to split-belt perturbations are not as 14 extensively reported. Moreover, the associated metabolic cost of these adaptations as 15 well as the retention of previously learned ML adaptations upon re-exposure to the same 16 perturbation have not been concurrently examined. We investigated adaptations in step 17 width and peak ML ground reaction forces (GRF) during an initial and subsequent 18 perturbation in order to characterize motor learning and motor savings , respectively. 19 Additionally, we examined the extent to which a neuroplasticity inducing stimulus, acute 20 intermittent hypoxia (AIH), affected the magnitude of each adaptation . Although w e 21 observed bilateral increases in step width during the initial adaptation, only the slow leg 22 significantly reduced step width during the subsequent perturbation. Distinct interlimb 23 differences emerged as only the slow leg modulated ML GRF during the braking phase 24 whereas the fast leg increased ML GRF during the propulsive phase. T he AIH group 25 uniquely demonstrated greater motor savings of reduced step width and peak ML GRF 26 strategies during the propulsive phase , suggesting greater retention of prior strategies. 27 Furthermore, we find significant associations between ML kinetic adaptations and 28 reductions in metabolic cost. Together, our findings suggest that unlike the sagittal plane, 29 asymmetrical frontal plane adaptations contribute to ML stability as well as reductions in 30 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION metabolic cost during split -belt walking. These insights could inform clinical training 31 approaches to improve balance and prevent falls in clinical populations. 32 33 NEW & NOTEWORTHY 34 We investigated adaptations in step width and mediolateral ground reaction forces during 35 the braking and propulsive phases of split-belt walking across an initial and subsequent 36 perturbation. We observe that the initial learning and savings of unique interlimb frontal 37 plane coordination strategies contribute to stability and are associated with a reduction in 38 metabolic cost. 39 40

41 Split-belt walking; mediolateral stability; motor learning; motor savings; acute intermittent 42 hypoxia 43 44

45 Maintaining balance during walking is critical for preventing falls, especially in older adults 46 (Rogers et al., 2001; McIlroy & Maki, 1996) and individuals with neurological impairments, 47 such as spinal cord injury (Jørgensen et al., 2016; Arora et al., 2019) . In contrast to 48 predominantly passive control of sagittal plane mechanics (McGeer, 1990; Kuo & 49 Donelan, 2010), it is well documented that active control of frontal plane mechanics is 50 required to maintain mediolateral (ML) stability (Bauby & Kuo, 2000; Donelan et al., 2004; 51 Kuo & Donelan, 2010). To enhance lateral margin of stability, measured as the minimum 52 distance between the base of support and extrapolated center of mass (Hof et al., 2005), 53 able-bodied individuals often regulate foot placement (Bruijn & Van Dieën, 2018; Buurke 54 et al., 2018; Rawal & Singer, 2021) to widen their stance (McAndrew Young & Dingwell, 55 2012; Hak et al., 2013) . Indeed, individuals with neurological impairments often adopt 56 wider strides compared to healthy controls (Curtze et al., 2024). While spatiotemporal and 57 kinetic features of sagittal plane mechanics have been extensively studied to characterize 58 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION balance during gait (Park & Finley, 2017; Debelle et al., 2020), adaptations in frontal plane 59 kinetics are less comprehensively reported. Particularly, the relationship between 60 adaptive step kinematics and their associated ML ground reaction forces (GRF) during 61 sustained destabilizing perturbations has not been thoroughly examined. Given that both 62 ML foot placement and GRF are coupled responses to external destabilization (Rawal & 63 Singer, 2021) , characterizing their simultaneous adaptations is necessary to further 64 elucidate control priorities in frontal plane mechanics. 65 One approach to assess ML balance control is through split -belt walking, which 66 uses single- belt speed perturbations to induce transient spatiotemporal walking 67 asymmetries (Sombric & Torres-Oviedo, 2020; Sánchez et al., 2021). These perturbations 68 elicit sensorimotor error -based motor learning (Reisman et al., 2005; Roemmich & 69 Bastian, 2015; Leech et al., 2022), as evidenced by progressive reductions in step length 70 asymmetry and double support time asymmetry within the sagittal plane (Donelan et al., 71 2002; Sánchez et al., 2021) . Importantly, participants also demonstrate ret ention of the 72 learned sagittal plane motor strategies during subsequent exposure to the same 73 perturbation known as motor savings (Roemmich & Bastian, 2015; Leech et al., 2018; 74 Bogard et al., 2023) . While frontal plane adaptations have been documented during a 75 single exposure to a split -belt speed perturbation (Buurke et al., 2018, 2019, 2021; 76 Cornwell et al., 2024), motor savings have not been directly examined. Motor savings is 77 thought to reflect the retention of previously learned motor patterns (Leech et al., 2018; 78 Huang et al., 2012), as well as shifts in reactive vs. anticipatory control strategies (Rawal 79 & Singer, 2021; Ahuja & Franz, 2022) . Thus, examining the temporal evolution of ML 80 coordination strategies during both the initial and subsequent perturbation exposure could 81 reveal further insights into the adaptive control of balance when stability is challenged. 82 During split-belt adaptation, indices of metabolic cost, like net metabolic power, 83 accompany learned biomechanical adaptations during both motor learning and motor 84 savings (Finley et al., 2013; Sánchez et al., 2017; Price et al., 2023) . Notably, we and 85 others have demonstrated that net metabolic power initially increases (Sánchez et al., 86 2019; Butterfield & Collins, 2022; Bogard et al., 2023) but gradually decreases toward 87 baseline values as participants adopt more energetically efficient coordination strategies. 88 In the sagittal plane, reductions in net metabolic power in able-bodied individuals parallel 89 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION improvements in spatiotemporal symmetry during split-belt adaptation (Finley et al., 2013; 90 Bogard et al., 2023). These findings suggest that energy optimization may play a role in 91 split-belt adaptation (Emken et al., 2007; Sánchez et al., 2017; Stenum & Choi, 2020) as 92 different control strategies exact distinct metabolic demands (Ahuja & Franz, 2022). 93 In contrast with the sagittal plane, concurrent reductions in net metabolic power 94 and adaptations in frontal plane mechanics have been inconsistently observed across 95 split-belt walking trials. For example, while external ML stabilization has been estimated 96 to reduce metabolic cost during normal walking (Donelan et al., 2004; Dean et al., 2007), 97 changes in ML margin of stability or step width during split-belt walking show no consistent 98 relationships with metabolic cost (Buurke et al., 2018). Deviations in step width (Donelan 99 et al., 2004), whole-body angular momentum (Cornwell et al., 2024) and ML foot roll-off 100 (Buurke et al., 2018) suggest that other ML gait parameters influence metabolic cost 101 during split-belt adaptation. To our knowledge, no study has concomitantly characterized 102 frontal plane mechanics and metabolic adaptations during both motor learning and motor 103 savings. Thus, our primary aim was to examine kinematic and kinetic adaptations 104 alongside associated changes in net metabolic power during split-belt walking. 105 We recently demonstrated that brief exposure to low oxygen, known as acute 106 intermittent hypoxia (AIH) , enhances sagittal plane adaptation during split-belt walking, 107 including spatiotemporal and anterior -posterior force asymmetry (Bogard et al., 2023) . 108 Although both the AIH and control groups successfully adapted their walking mechanics, 109 the AIH group achieved a greater reduction in net metabolic power (Bogard et al., 2023). 110 Combined with prior studies that show AIH facilitates descending excitability and reduces 111 performance fatiguability (Bogard et al., 2024; Bogard, Pollet, et al., 2024) , these 112 observations suggest that adaptive changes within the nervous system may facilitate 113 greater motor learning. Indeed, AIH-induced synthesis of brain-derived neurotrophic 114 factor (BDNF) parallel strengthened synaptic plasticity and improvements in motor 115 performance in rodents (Fritsch et al., 2010; Lovett-Barr et al., 2012). Increases in BDNF 116 in humans have also been associated with enhanced motor learning (Rasmussen et al., 117 2009; Leech & Hornby, 2017) . Thus, t o further elucidate the neural control process 118 underlying motor adaptation, a secondary aim of this study was to investigate whether 119 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION repetitive AIH enhances motor learning of frontal plane adaptation and whether such 120 changes drive greater reductions in net metabolic power. 121 Accordingly, we examined time-dependent adaptations in step width and peak ML 122 GRF during braking and propulsive phases of gait in response to split -belt speed 123 perturbations. We further characterized the extent to which these adaptations are retained 124 in response to a second exposure to the same perturbation, as well as the corresponding 125 effect on net metabolic power . We hypothesized that participants would initially adopt a 126 wider step and greater ML GRF and that these coordination strategies would be reduced 127 upon subsequent exposure. Given that AIH has been shown to improve sensorimotor 128 adaptation in sagittal plane mechanics (Bogard et al., 2023) , we tested the secondary 129 hypothesis that adaptations in frontal plane mechanics and reductions in net metabolic 130 power would be more prominent in the AIH group relative to the control group. 131 132

133 Participants 134 Thirty able- bodied i ndividuals with no prior history of neurological impairments were 135 recruited for the study. All participants provided informed consent approved by the 136 Colorado Multiple Institutional Review Board (COMIRB no. 20- 0689). The procedures 137 complied with the standards of the Declaration of Helsinki and the study was registered 138 on clinicaltrials.org (NCT05341466). Participants were randomized into an AIH group (n 139 = 15, 9 females, 6 males, age 23.5 + 2.3 years, body mass 65.9 + 11.5 kg) or a control 140 group (n = 15, 7 females, 8 males, age 25.3 + 5.3 years, body mass 73 .9 + 15.1 kg). 141 Criteria for exclusion included prior exposure to split -belt walking, altitude sensitivity, 142 cardiovascular or pulmonary diseases, syncope, or being pregnant at the time of the 143 study. 144 145 Protocols 146 Acute Intermittent Hypoxia 147 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION For the AIH group, low-oxygen air was delivered through an altitude generator (HYP 123; 148 Hypoxico, Inc., USA) while trained researchers continuously monitored heart rate (HR) 149 and oxygen saturation levels (SpO 2) along with measuring blood pressure (BP) every 5 150 normoxic cycles (Masimo; Irvine, CA, USA). Similar to Tan et al., 2021, AIH consisted of 151 90 s bouts of hypoxic air (9% O 2) followed by 60 s bouts of normoxic air (21% O2) for a 152 total of 15 cycles. Treatment was paused if individuals de-saturated below 70% SpO2 and 153 resumed when they re -saturated above 80% (Tan et al., 2020) . The experiment was 154 discontinued if any of the following conditions occurred: systolic BP exceeded 140 mmHg, 155 diastolic BP exceeded 90 mmHg , or HR values surpassed 160 bpm; however, th ese 156 thresholds were not exceeded (Tan et al., 2020) . Termination criteria also included 157 reported or visible signs of adverse events such as dizziness, numbness, tinnitus, blurred 158 vision, or diaphoresis. This protocol was repeated for five consecutive days at 159 approximately the same time each day. All participants tolerated the hypoxic dose and 160 completed the full procedure. 161 162 Gait Mechanics 163 Kinematic data was recorded with a 10-camera system at a rate of 100 Hz (Vicon Nexus 164 v2.8.1; Vicon Motion Systems, Denver, CO, USA) . Thirty-four reflective markers were 165 placed as shank and thigh clusters and on the following anatomical landmarks: anterior 166 and posterior superior iliac spines, iliac crests, greater trochanters, medial, and lateral 167 femoral epicondyles, medial and lateral malleoli, calcanei and first and fifth metatarsals 168 (Montgomery & Grabowski, 2018) . P articipants were instructed to avoid us ing the 169 handrails unless ne cessary for safety (Buurke, Lamoth, Van Der Woude, & Den Otter, 170 2019) and were secured with a single passive harness that neither affected movement 171 nor provided body weight support. A mirror was positioned in front of the treadmill to help 172 individuals avoid crossing their feet onto the opposite belt. Time-synchronized kinetic 173 variables were measured on an instrumented split-belt treadmill at a rate of 1000 Hz, and 174 the belt speeds were controlled independently (M-Gait, D-flow v3.34.3; Motek Medical , 175 Houten, NL). 176 177 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION Metabolic Data 178 Metabolic rate was measured using an open circuit spirometry system (TrueOne 2400; 179 ParvoMedics Inc., Salt Lake City, UT, USA). The rate of oxygen consumption ( 𝑉𝑉̇O2) and 180 carbon dioxide production (𝑉𝑉̇CO2) was utilized to calculate respiratory exchange ratios 181 (RER = 𝑉𝑉̇ CO2 /𝑉𝑉̇O2). All participants had RER values below 1, indicating that aerobic 182 pathways were primarily being used (Huang et al., 2012) . Resting metabolic rate (RMR) 183 was estimated from each individual’s average 𝑉𝑉̇O2 and 𝑉𝑉̇CO2 during the last 2 minutes of a 184 5-minute standing trial for the AIH (1.51 + 0.26 W/kg) and control (1.50 + 0.20 W/kg) 185 groups. We estimated energetic cost during walking trials by calculating metabolic power 186 using the regression formula provided in Equation 1 (Péronnet & Massicotte, 1991). 187 Metabolic Power (W) = 16.98 𝑉𝑉̇O2 L s� + 6.98 𝑉𝑉̇CO2 L s� (1) 188 Net metabolic power (W/kg) was obtained from the difference between metabolic 189 power and average RMR and normalized to body mass (Finley et al., 2013) . Data from 190 the initial 60 s of each trial were excluded, as this reflects the duration needed for net 191 metabolic power to stabilize to its average value (Finley et al., 2013). 192 193 Split-Belt Walking Protocol 194 We followed the split-belt walking protocol described by Bogard et al., 2023 which 195 included four trials of two tied-belt and two split-belt walking conditions. A single belt was 196 randomly selected to speed up without forewarning (Reisman et al., 2005) throughout the 197 experiment. The first trial was ‘baseline,’ which included walking at a tied- belt speed of 198 1.0 m/s for 300 strides . Following baseline, participants performed an ‘adapt 1’ trial that 199 involved tied-belt walking at 1.0 m/s for 15-30 strides before a sudden increas e of one 200 belt to 2.0 m/s for 300 strides of split -belt walking at a 2:1 belt speed ratio. Next, the 201 participants performed a ‘washout’ trial where they walked with tied-belt speed of 1.0 m/s 202 for 350 strides. Finally, participants performed an ‘adapt 2’ trial by walking at 1.0 m/s tied-203 belt speed for 15-30 strides followed by a second exposure to the split-belt perturbation 204 at a 2:1 belt speed ratio for 300 strides. The trials were performed consecutively, stopping 205 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION after each trial with an optional 1- minute break during which individuals were instructed 206 to stand still (Leech et al., 2018; Sombric & Torres -Oviedo, 2020) . Participants in the 207 intervention group performed this protocol 15 minutes after receiving their final low-208 oxygen treatment. 209 210 Data Analysis 211 Motion-captured m arkers were labeled in Vicon to create lower body models for gait 212 analysis using custom pipelines created i n Visual3D ( v2021.11.3; HAS-motion Inc., 213 Germantown, MD, USA ). Ground-reaction forces (GRF) were filtered using Lowpass 214 Butterworth with a cutoff frequency of 20 Hz. Heel-strike and push -off gait events were 215 identified at vertical GRF thresholds of 30 N (Karakasis & Artemiadis, 2021). Step widths 216 were calculated in Visual3D as the perpendicular distance between a stride vector formed 217 by consecutive heel-strikes of the contralateral limb and the calcaneus position at heel-218 strike of the ipsilateral limb. Midstance events were defined as the instance when anterior-219 posterior GRF crossed zero when plotted, marking the transition from braking into 220 propulsion phase of gait (Masani et al., 2002). Therefore, the time intervals between heel-221 strike and midstance defined the braking phases, and the propulsive phases were defined 222 between midstance and push-off. We identified peak ML GRF during both the braking and 223 propulsive phases within each step. 224 Step width and peak ML GRF values were averaged for each trial. The last 20 225 steps of the baseline were analyzed for comparison with the perturbation trials . The 226 perturbation trials were divided into separate learning phases. Namely, ‘early learning’ 227 was defined as the first 5 steps immediately following the change in belt speed while ‘late 228 learning’ was the last 20 steps of each trial. The early and late learning phases within a 229 single perturbation trial are referred to as motor learning. To characterize the retention of 230 motor strategies, we compared the learning phases across perturbation trials (Bogard et 231 al., 2023). ‘Early savings’ compared motor adaptions between early adapt 1 and early 232 adapt 2, while ‘late savings’ compared motor adaptations between late adapt 1 and late 233 adapt 2 . Comparisons of learning phases across perturbation trials are described as 234 motor savings. 235 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION 236 Statistical Analysis 237 Kinematic, kinetic, and metabolic d ata were processed in MATLAB (R202 3b; The 238 MathWorks, Inc., Natick, MA, USA). Statistical analysis was performed in R Studio 239 (v2024.04.2) with the significance set to p < 0.05. Normality and homogeneity of the data 240 were tested using Shapiro- Wilks and Levene’s tests, respectively. An a pr iori power 241 analysis using preliminary data determined that a sample size of 14 participants per group 242 would be adequate to observe differences between and within groups (power = 0.85, 243 Cohen’s f = 0.76, α = 0.05, F(1,12) = 4.74) (Cohen, 1988). Two-way repeated measures 244 analysis of variance ( ANOVA) tests were utilized to examine the effect of leg speed on 245 step width. Mean step width was compared between the fast and slow legs at baseline, 246 early adapt 1, late adapt 1, early adapt 2, early adapt 2, and late adapt 2. Similarly, two-247 way ANOVAs were conducted to examine the effect of leg speed on peak ML GRF. Peak 248 ML GRF s were compared between the fast and slow legs during the braking and 249 propulsive phases at baseline, early adapt 1, late adapt 1, early adapt 2, and late adapt 250 2. Furthermore, two-way ANOVAs were conducted separately for each limb to investigate 251 the effects of AIH intervention on early and late savings of step width and peak ML GRF. 252 Additional two-way ANOVAs analyzed the effects of motor learning and motor savings on 253 net metabolic power. Average net metabolic power was compared between early and late 254 learning of adapt 1 and 2 as well as early and late savings . For analyses of repeated 255 measures that violated the assumption of sphericity ( P < 0.05) as determined by 256 Mauchly’s Test for Sphericity, adjustments were made using the Greenhouse-Geisser 257 (GG) correction for epsilon (GGe) values below 0.75 or the Huynh- Feldt (HF) correction 258 for GGe values above 0.75. Tukey's Honestly Significant Difference (HS D) tests were 259 performed to identify pairwise interactions for main effects and interactions that reached 260 significance. Linear regression analyses were generated to investigate relationships 261 between changes in net metabolic rate and changes in step width and peak ML GRF. 262 263 264 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION

265 Kinematic Adaptations 266 Reductions in Step Width During Motor Learning 267 During adapt 1, learning phase had a main effect on step width (F(1.38, 39.9) = 18.2, P < 268 0.001), in which both legs increased step width during early learning then decreased in 269 late learning toward baseline levels (Fig. 1A). Post-hoc comparisons revealed significant 270 increase from baseline to early learning (fast leg: t(60.4) = - 5.35, P < 0.001; slow leg: 271 t(60.4) = -5.14, P < 0.001) and decrease from early learning to late learning (fast: t(60.4) 272 = 5.18, P < 0.001; slow: t (60.4) = 5. 03, P < 0.001). N o main effect s of leg speed ( P = 273 0.491) nor an interaction of leg speed and learning phase (P = 0.749) on step width were 274 observed. 275 Similarly, step width significantly narrowed throughout learning during adapt 2 (F(2, 276 58) = 6.19, P = 0.004), but neither leg speed (P = 0.348) nor an interaction of leg speed 277 and learning phase (P = 0.482) had significant effects (Fig. 1B). Pairwise t-tests showed 278 significantly larger widths during early learning compared to baseline for both legs (fast 279 leg: t(62.1) = -3 .12, P = 0.008; slow leg: t(62.1) = - 3.49, P = 0.002) . Interestingly, 280 differences between limbs began to emerge during late learning of the second 281 perturbation, with significant narrowing observed in the slow leg ( t(62.1) = 2.74, P = 282 0.021). 283 284 Reductions in Step Width During Motor Savings 285 To examine how step width adapts independently for each limb, we analyzed the fast and 286 slow limbs separately, highlighting motor savings strategies of AIH and control groups . 287 For the fast leg, we observed reduced step widths as a main effect of perturbation 288 between early learning phases (F(1, 28) = 13.4, P = 0.001), but no effect of group ( P = 289 0.717) nor an interaction of perturbation and group ( P = 0.149) (Fig. 1 C). P airwise 290 comparisons identified early savings strategies in the AIH group (t(28) = 3.64, P = 0.001) 291 but not th e controls (P = 0.013). We observed no late savings on the fast leg between 292 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION perturbation (P = 0.481), within groups (P = 0.053), nor the interaction of these factors (P 293 = 0.123), indicating that participants in both groups had adapted by this timeframe. 294 Similarly, a repeated measures ANOVA for the slow leg revealed reductions in step 295 width as a main effect of perturbation during early learning phases (F(1, 28) = 9.27, P = 296 0.005) (Fig. 1D). Post-hoc revealed early savings of smaller step widths in the AIH group 297 (t(28) = 2.66, P = 0.013) but not the control group (P = 0.111). There were no late savings 298 across perturbations (P = 0.403), within groups (P = 0.606), nor an interacti on of either 299 (P = 0.120). 300 301 Kinetic Adaptations 302 Braking Phase During Motor Learning 303 Peak ML GRF was significantly larger than baseline during early learning for both legs 304 then decreased for just the slow leg during late learning (Fig. 2A ). We observed main 305 effects of leg speed (F(1,29) = 47.2, P < 0.001), learning phase (F(1.40,40.8) = 61.4, P < 306 0.001), and interaction between leg speed and learning phase ( F(1.66,48.2) = 34.0, P < 307 0.001) for peak ML GRF during the braking phase of adapt 1. Post-hoc analyses showed 308 a significant increase for the fast leg between baseline and early learning (t(93.1) = -9.96, 309 P < 0.001) and between baseline and late learning (t(93.1) = -9.90, P < 0.001). The slow 310 leg similarly increased peak ML GRF between baseline and early learning ( t(93.1) = -311 8.95, P < 0.001) and decreased from early to late learning (t(93.1) = 6.67, P < 0.001). 312 Furthermore, we observed significant reductions in force in the slow leg compared to the 313 fast leg during late learning (t(86.9) = 10.7, P < 0.001). 314 During adapt 2 there were main effects of leg speed ( F(1,29) = 192, P < 0.001), 315 learning phase (F(2,58) = 85.9, P < 0.001), and an interaction between leg speed and 316 learning phase (F(2,58) = 83. 8, P < 0.001) (Fig. 2B). Post-hoc t-tests found significant 317 increase between learning phases for the fast leg (b aseline vs. early learning: (t(97.1) = 318 -14.5, P < 0.001); baseline vs. late learning: (t(97.1) = -16.12, P < 0.001)) and the slow 319 leg (baseline vs. early learning: ( t(97.1) = -5 .11, P < 0.001); baseline vs. late learning: 320 (t(97.1) = -2.81, P = 0.016). Additionally, the fast leg maintained significantly higher peak 321 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION force magnitudes compared to the slow leg during early learning ( t(82.2) = 11.5 , P < 322 0.001) and late learning (t(82.2) = 16.3, P < 0.001). 323 324 Braking Phase During Motor Savings 325 To examine motor savings strategies between AIH and control groups, peak ML GRF of 326 the fast and slow legs were independently analyzed . For the fast leg, w e observed no 327 main effects of group ( early savings: P = 0.332; late savings: P = 0.965), perturbation 328 (early savings: P = 0.165; late savings: P = 0. 755), nor interaction of group and 329 perturbation (early savings: P = 0.428; late savings: P = 0.658) (Fig. 2C). This suggests 330 that the fast leg’s strategy was to maintain larger peak ML GRF throughout adaptation. 331 Conversely, the slow leg showed main effects of perturbation (F(1,28) = 42.1, P < 0.001), 332 but no group effect (P = 0.164) nor interaction of group and perturbation (P = 0.113) for 333 early savings (Fig. 2D). Pairwise comparisons showed that both groups demonstrated 334 early savings strategies (AIH: t(28) = 5.75, P < 0.001); controls: (t(28) = 3.43, P = 0.002), 335 reflecting a reduction in response magnitudes upon re-exposure to asymmetric walking. 336 Neither group exhibited late savings for the slow leg (P = 0.663) between perturbations 337 (P = 0.279), nor an interaction of group and perturbation (P = 0.305). 338 339 Propulsive Phase During Motor Learning 340 We observed main effects of leg speed ( F(1,29) = 54.5, P < 0.001), learning phase 341 F(1.38,40.1) = 32. 2, P < 0.001) , and an interaction of leg speed and learning phase 342 (F(1.44,41.9) = 65.8, P < 0.001) for peak ML GRF during the propulsive phase of adapt 343 1. The fast leg significantly increased force magnitudes during early learning (t(93.5) = -344 11.11, P < 0.001) which then decreased during late learning (t(93.5) = 10.78, P < 0.001) 345 (Fig. 3A). Pairwise analysis revealed a significant difference in magnitude between the 346 legs during early learning ( t(87) = 13.5, P < 0.001), with the fast leg exhibiting larger 347 magnitudes while the slow leg maintaining forces approaching baseline values. 348 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION During adapt 2, we identified main effects of leg speed (F(1,29) = 31.7, P < 0.001), 349 learning phase ( F(2,58) = 5.56, P = 0.006), and interaction of leg speed and learning 350 phase ( F(1.62,47.1) = 68. 5, P < 0.001) (Fig. 3 B). Pairwise comparisons showed 351 significant increase for the fast leg from baseline to early learning (t(95.4) = - 7.62, P < 352 0.001) and decrease from early to late learning: (t(95.4) = 7.19, P < 0.001). The slow leg 353 significantly reduced peak ML GRF below baseline during early learning (t(95.4) = 4.01, 354 P < 0.001) and late learning (t(95.4) = 2.45, P = 0.042). Additionally, there were evident 355 interlimb differences during early learning ( t(85.7) = 1 2.4, P < 0.001), with the fast leg 356 increasing and the slow leg decreasing relative to baseline. 357 358 Propulsive Phase During Motor Savings 359 A two-way ANOVA investigated early savings of peak ML GRF on the fast leg during the 360 propulsive phase, which showed a main effect of perturbation (F(1,28) = 43.0, P < 0.001). 361 There were no effects of group (P = 0.559) nor interaction of group and perturbation (P = 362 0.303). Both the AIH (t(28) = 5.38, P < 0.001) and control (t(28) = 3.90, P < 0.001) groups 363 demonstrate early savings of reduced peak ML force magnitude (Fig. 3C). In late savings, 364 there were no significant main effects of group (P = 0.417), perturbation (P = 0.833), nor 365 interaction of group and perturbation (P = 0.208). 366 Additionally, an analysis for the slow leg showed a main effect of perturbation 367 (F(1,28) = 7.08, P = 0.013), but no effect of group (P = 0.433) nor interaction of group and 368 perturbation (P = 0.225). Post-hoc analysis revealed early savings of peak ML GRF during 369 the propulsive phase, specifically lower peak ML GRF in the AIH group (t(28) = 2.76, P = 370 0.010) but not in the controls (P = 0.324) (Fig. 3D). There were no significant late savings 371 between perturbations (P = 0.946), within groups (P = 0.852), nor interaction of group and 372 learning phase (P = 0.971). 373 374 Metabolic Adaptations 375 Net Metabolic Power During Motor Learning 376 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION A two-way ANOVA showed a main effect of learning phase for net metabolic power 377 (F(1,28) = 42.48, P < 0.001) but no group effect (P = 0.903) nor interaction between group 378 and learning phases (P = 0. 672). Pairwise t-tests showed that net metabolic power 379 decreased in late learning of adapt 1 compared to early learning for both the AIH (t(28) = 380 4.91, P < 0.001) and control groups ( t(28) = 4.30, P < 0.001) (Fig. 4A). During adapt 2, 381 there were main effects of learning phase ( F(1,28) = 4.50, P = 0.043) and interaction of 382 group and learning phase (F(1,28) = 5.29, P = 0.029) for net metabolic power. Post-hoc 383 comparisons revealed that the AIH group increased net metabolic power during late 384 learning (t(28) = -3.13, P = 0.004) (Fig. 4B). 385 386 Net Metabolic Power During Motor Savings 387 We compared early learning phases across both perturbations and observed main effects 388 of perturbation (F(1,28) = 98. 6, P < 0.001 ) and interaction of group and perturbation 389 (F(1,28) = 4.61, P = 0.041) on early savings of net metabolic power, but no group 390 differences ( P = 0.477) (Fig. 4 C). Tukey’s comparison showed significant reductions 391 between early adapt 1 and early adapt 2 for both AIH (t(28) = 8.54, P < 0.001) and control 392 groups (t(28) = 5.50, P < 0.001). We also observed late savings of net metabolic power 393 between perturbations (F(1,28) = 10.8, P = 0.003), but no effects of group (P = 0.799) nor 394 interaction of group and perturbation (P = 0.978). Post-hoc analysis showed magnitude 395 reductions in net metabolic power during late learning in both the AIH (t(28) = 2.34, P = 396 0.027) and control groups (t(28) = 2.30, P = 0.029) (Fig. 4D). 397 398 Correlations of Kinematic, Kinetic, and Metabolic changes 399 We investigated correlations between changes in net metabolic power and changes in 400 kinematic and kinetic variables. There were non-significant positive relationships between 401 changes in net metabolic power and step width i n the fast (R2 = 0.203, P = 0.092) and 402 slow leg (R2 = 0.254, P = 0.055). We observed significant correlations between reductions 403 in net metabolic power and decreases in peak ML GRF during the first perturbation, 404 notably during the propulsive phase for the fast leg (R2 = 0.728, P < 0.001) and during 405 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION the braking phase for the slow leg (R 2 = 0.328, P = 0.026). These results suggest that 406 individuals with the largest decreases in step width and peak ML GRF drove reductions 407 in net metabolic cost. 408 409

410 This study examined adaptations in step width, peak ML GRF, and their association with 411 changes in net metabolic power during split-belt treadmill walking. We observe distinct 412 coordination strategies between the fast and slow limbs during the initial perturbation 413 exposure. We further demonstrate motor savings of these distinct interlimb adaptations 414 during a subsequent perturbation as well as novel associations between changes in peak 415 ML GRF and reductions in net metabolic power. R epetitive AIH treatments further 416 enhanced both motor learning and motor savings of adaptive frontal plane mechanics. 417 418 Step Width Reductions During Motor Learning and Motor Savings 419 During adapt 1, we observed that step width increased bilaterally during early learning, 420 followed by a reduction toward baseline as learning progressed. Congruent with our 421 hypothesis, participants utilized an initial compensatory widening of gait to increase 422 stability under the newly altered belt speeds, which was gradually reduced as subjects 423 adapted to the perturbation. Other studies similarly observed initial increases in step width 424 (Fettrow et al., 2021) in addition to increases in ML margin of stability during early learning 425 (Buurke et al., 2018; Brinkerhoff et al., 2024) followed by reductions in step width 426 throughout adapt 1. One novel finding was that the leg on the slow belt followed a similar 427 pattern in foot placement adaptation during a subsequent perturbation, while the fast leg 428 maintained wider steps into late learning. The maintenance of wider steps can likely be 429 interpreted as the fast leg’s strategy to preserve stability (McAndrew Young et al., 2012). 430 These observations indicate that bilateral foot placement adaptations are primarily 431 adopted in the early learning phase of asymmetric walking, whereas re-exposure reveals 432 a distinct shift in fast and slow foot placement strategies. These results contrast with 433 previously reported sagittal plane mechanics during split -belt walking, which shows a 434 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION return to step length symmetry as adaptation progresses (Bogard et al., 2023; Leech et 435 al., 2018) . This suggests that ML spatial adaptations require distinct mechanical 436 strategies between limbs as interlimb symmetry in the frontal plane is not a control priority. 437 The AIH group uniquely demonstrated early savings of step width in both limbs, 438 indicating that receiving low-oxygen treatments enhances retention of previously learned 439 strategies. These findings support our previous observations that repetitive AIH exposure 440 facilitates improved motor savings of spatiotemporal asymmetry (Bogard et al., 2023). In 441 contrast, we did not observe late savings, indicating that both groups consolidated their 442 adaptation strategies at an earlier timescale regardless of intervention. This interpretation 443 is consistent with sagittal plane spatial adaptations observed in healthy participants 444 (Bogard et al., 2023) and individuals with Parkinson’s disease (Thompson & Reisman, 445 2022). 446 447 Distinct Interlimb Adaptation Patterns in Peak ML GRF During Motor Learning and 448 Savings 449 The fast and slow legs differentially adapted peak ML GRF during the braking phase of 450 both perturbation trials. The braking phase of gait, occurring immediately after heel-strike 451 in early stance, is critical for re- stabilization and may explain why higher forces were 452 maintained on the fast leg (Rawal & Singer, 2021). Similarly, our group previously noted 453 larger braking forces on the fast leg in the sagittal plane during split-belt walking (Bogard 454 et al., 2023) . In contrast, the leg on the slow belt adapted by decreasing its peak force 455 magnitude, emphasizing each leg’s unique role in maintaining ML stability during split-456 belt adaptation. Contrary to our findings, Roper et al., 2017 observed no interlimb 457 differences in ML GRF impulses during the braking phase. This discrepancy may arise 458 from the limitations of time integration of force, which might not fully capture key details 459 such as peak magnitudes (Deffeyes & Peters, 2021), particularly in dynamically adaptive 460 environments. One limitation is that we analyzed peak force magnitudes when ML GRF 461 was directed medially, not capturing adaptations of laterally directed forces (John et al., 462 2012) which could provide further insights into kinetic adaptation strategies. 463 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION During the propulsive phase, peak ML GRF of the fast leg reactively increased 464 during early learning before returning to baseline in both perturbation trials. Conversely, 465 peak ML GRF of the slow leg did not alter during adapt 1 but was reduced below baseline 466 during adapt 2. The unique adaptive responses to each belt speed align with previously 467 observed changes in ML GRF at varying walking speeds, as well as magnitude shifts 468 throughout the gait cycle that reflect our observed peak force values during the braking 469 and propulsive phases (John et al., 2012). Although interlimb adaptation patterns during 470 the propulsive phase are distinct, they appear less pronounced compared to the braking 471 phase. Interestingly, the slow leg demonstrated savings o f peak ML GRF during the 472 propulsive phase, suggesting that changes during push-off primarily contributed to the 473 observed strong correlation with net metabolic power . Applying external lateral 474 stabilization reduced energetic costs by minimizing excessive ML movement (Dean et al., 475 2007), while external horizontal aiding forces similarly lowered metabolic rate by offsetting 476 the high energy cost of generating propulsive forces during walking (Gottschall & Kram, 477 2003). Interestingly, increasing propulsive demand through inclined split -belt walking 478 appears to improve the magnitude of motor learning in the sagittal plane in participants 479 with stroke as evidence d by reduc tions in step length asymmetry (Sombric & Torres -480 Oviedo, 2020). This suggests that the kinetic demands of propulsion influence not only 481 impact th e energetic cost but also shape the magnitude of motor adaptation. Our 482 observations emphasize the contribution of the ML forces generated during the propulsive 483 phase in metabolic adaptation , underscoring the importance of optimizing p ush-off 484 mechanics to reduce energy expenditure. Another consideration is that Roper et al. 485 observed that the ML force impulse of the slow limb during late adaptation to be more 486 medially directed than the fast limb. This highlights an important limitation that our 487 analyses of the peak force magnitudes do not adequately capture adaptations in ML GRF 488 directionality and their corresponding effect on net metabolic power (Roper et al., 2017). 489 We further demonstrated that during re- exposure to the split -belt walking 490 perturbation, individuals in the AIH group uniquely retained motor strategies that reduced 491 peak ML GR F in addition to narrowing step width. Even though both groups similarly 492 demonstrated early savings d uring the propulsive phase o n the fast leg, only the AIH 493 group displayed reduced peak ML GRF on the slow leg during adapt 2 . Although we 494 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION previously observed that anterior-posterior force a symmetry was ultimately reduced 495 during split-belt walking (Bogard et al., 2023), differences between limbs were maintained 496 in frontal plane kinetics during braking and propulsive phases. These distinct adaptations 497 of the fast and slow leg may be partly attributed to independent neural control of each 498 limb, which is modulated based on unique sensory feedback from afferent inputs to each 499 limb (Choi & Bastian, 2007) . Furthermore, it has been shown that intralimb adaptations 500 occur at a quicker timescale than interlimb gait parameters (Sato & Choi, 2022) , as 501 evidenced by each leg independently adapting its peak ML GRF during adapt 1. Thus, 502 differences in plane-specific adaptations may indicate distinct control priorities where the 503 sagittal plane adaptations prioritize a more symmetrical AP force generation to propel the 504 body forward, while frontal plane adaptations allow asymmetrical stepping mechanics to 505 maintain dynamic stability. 506 507 Energetic Optimization 508 We observed that both groups concurrently reduced net metabolic power as they adapted 509 their frontal plane mechanics , supporting the notion that adaptive control of dynamic 510 balance is related to the reduction in energy expenditure (Donelan et al., 2001; Finley et 511 al., 2013; Selinger et al., 2015; Buurke et al., 2018) . Although the AIH group increased 512 net metabolic power from early to late learning during Adapt 2, they exhibited about 0.3 513 W/kg less than the controls during early learning and continued to decrease throughout 514 late learning. These observations suggest that the retention of motor adaptations is more 515 pronounced after repetitive AIH, as indicated by the trend in reduced net metabolic power. 516 The savings of frontal plane adaptations align with numerous studies that examined how 517 biomechanical adaptations reduce energetic expenditure upon re- exposure to the 518 perturbation, including decreases in step length asymmetry (Finley et al., 2013; 519 Roemmich & Bastian, 2015; Buurke et al., 2022), optimization of step frequency (Selinger 520 et al., 2015), adjustments in peak force asymmetry (Bogard et al., 2023), and reductions 521 in positive mechanical work (Sánchez et al., 2019) . Given previous work that shows 522 savings of biomechanical adaptations are improved with exposure to larger perturbations 523 that utilize great split -belt speed ratios (Leech et al., 2018) , we speculate that energetic 524 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION savings may reach a plateau as split-belt speed ratios increase. The findings of this study, 525 which used a 2:1 speed ratio, contrast with our previous observations using a 1.5:1 speed 526 ratio in which we noted continual reductions in net metabolic power during the subsequent 527 perturbation (Bogard et al., 2023). Nevertheless, both groups demonstrated early savings 528 of lower net metabolic power, with AIH participants demonstrating greater adaptations. 529 Late savings showed similar changes between groups, indicating that gains in metabolic 530 power had stabilized. 531 532 Correlations Between Kinematics, Kinetics, and Net Metabolic Power 533 Narrowed step widths were positively correlated with reductions in net metabolic power. 534 Other studies have reported reductions in metabolic cost when frontal plane strategies 535 are optimized to walk at preferred step width (Donelan et al., 2001) and exploit frontal 536 plane passive dynamics (Fettrow et al., 2021). In the sagittal plane, reductions in net 537 metabolic power paralleled increases in step time asymmetry (Bogard & Tan, 2024) and 538 decreases in step length asymmetry during motor learning and savings (Sánchez et al., 539 2019; Bogard et al., 2023). These findings show that the metabolic determinants of split-540 belt adaptation are driven by distinct yet complementary kinematic adaptations across 541 both planes (Buurke et al., 2020; Buurke & den Otter, 2021) as well as limb-specific 542 kinetics that are likely interdependent . For example, t he correlations between ML force 543 production and net metabolic power showed stronger relationships than with kinematic 544 changes. Notably, kinematic adjustments respond more quickly to perturbations and are 545 driven by sensory feedback whereas kinetic adaptations are more gradual (Mawase et 546 al., 2013) , supporting the view that ML kinetic demands largely regulate changes in 547 energy cost. There was a significant correlation between ML GRF of the fast leg during 548 the propulsive phase and net metabolic power , suggesting that as the limb transitions 549 from push- off to the swing phase, control of ML GRF may be critical in regulating 550 metabolic cost during gait adaptations. Indeed, modulating propulsive force in the sagittal 551 plane influences walking energetics (Grabowski et al., 2005; Pieper et al., 2021) and our 552 findings suggest that ML mechanics contribute to the overall metabolic cost. Regulation 553 of frontal plane stability is critical in driving initial adaptation, as recent studies show that 554 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION ML adjustments adapt at a faster timescale than in the sagittal plane (Brinkerhoff et al., 555 2024). Furthermore, it has been proposed that improvements in sagittal plane symmetry 556 may come at the cost of frontal plane parameters, (Cornwell et al., 2024), a pattern also 557 observed in individuals post-stroke (Buurke et al., 2020). Together with our findings, these 558 observations emphasize the importance of examining adaptations across both planes to 559 gain a comprehensive understanding of how their interactions influence energetic 560 optimization. 561 562 Improved Adaptations Following AIH 563 We demonstrated enhanced kinematic and kinetic adaptations following repetitive AIH 564 treatments. Notably, the AIH group exhibited early savings of reduced step width for both 565 legs and peak ML GRF during the propulsive phase for the slow leg. The neural 566 mechanisms underlying AIH -induced improvements in motor performance and motor 567 learning remain unclear. Previous studies in spinally injured rats show that BDNF -568 dependent mechanisms were enhanced following daily AIH, leading to improved 569 horizontal ladder walking (Lovett-Barr et al., 2012). In humans, emerging evidence from 570 our lab and others indicate that AIH strengthens descending neural excitability 571 (Christiansen et al., 2018; Bogard et al., 2023; Bogard, Hembree, et al., 2024) as well as 572 decreases performance fatiguability (Bogard, Pollet, et al., 2024). Together, these findings 573 suggest that similar BDNF -dependent mechanisms underly improvements in motor 574 learning. Indeed, preliminary evidence indicates that BDNF affects other forms of motor 575 learning including visuomotor adaptation and use-dependent plasticity (Fritsch et al., 576 2010; Joundi et al., 2012; Mang et al., 2014; Helm et al., 2016) . We speculate that AIH-577 induced increases in BDNF enhance synaptic plasticity (Fritsch et al., 2010), leading to 578 improved sensorimotor adaptation (Bogard et al., 2023) and motor savings observed in 579 the present study. Previous research suggests that the nervous system facilitates postural 580 adjustments by sending commands to anticipate and react to postural threats (Cesari et 581 al., 2022). The savings of reductions in step width and ML GRF during adapt 2 support 582 the interpretation of a shift from generalized anticipatory control strategies to task-specific 583 reactive control to preserve stability following a perturbation (Ahuja & Franz, 2022) . 584 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION Although we did not measure step- to-step adjustments , quantifying any persistent 585 increases step variability throughout adapt 2 may further elucidate shifts towards task -586 specific reactive control that accommodates flexible foot placement strategies (Ahuja & 587 Franz, 2022). While the current data set cannot parse the underlying neural mechanisms, 588 it remains plausible that AIH -induced synaptic plasticity may enhance these adaptive 589 control mechanisms , leading to more effective dynamic balance during destabilizing 590 perturbations. 591 AIH treatments have significant potential for i ntegration into the design of 592 rehabilitation training paradigms aimed at improving walking stability and reducing fall 593 risks. Indeed, both l ong-term and progressive training paradigms have been previously 594 demonstrated to induce improvements in motor learning (Christiansen et al., 2020), lateral 595 balance control (Sawers et al., 2013), and enhanced walking performance after repetitive 596 AIH exposure (Hayes et al., 2014; Tan et al., 2021) . Moreover, prolonged exposure to 597 optical flow perturbations has the potential to be used as a training method to improve 598 corrective motor adjustments while walking in older adults (Richards et al., 2019) . 599 Improving frontal plane stability is especially critical for older populations and individuals 600 with neurological disorders , considering their diminished motor control and adap tive 601 capabilities (Sato & Choi, 2022; Fettrow et al., 2021; Arora et al., 2020). Therefore, future 602 studies may more critically examine how implementing targeted practice paradigms 603 combined with AIH treatments improves dynamic balance in persons with neurological 604 deficits such as SCI (Navarrete-Opazo et al., 2017) as well as healthy older adults. Our 605 findings further underscore the coupling between adaptive ML control an d walking 606 energetics, which may further inform the detection of balance decline as well as the 607 tailoring of patient-specific rehabilitation strategies to improve postural control. 608 609 DATA AVAILABILITY 610 The authors confirm that the data supporting the findings of this study are fully available 611 and presented in the supporting information of the manuscript. Correspondence and 612 requests for materials should be addressed to A.Q.T. 613 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION 614 GRANTS 615 This work was funded by the National Institute of Health ’s National Center of 616 Neuromodulation for Rehabilitation (NM4R) [NIH P2CHD086844] , ABNEXUS Award 617 (AQT), and the Boettcher Foundation Webb Waring Biomedical Research Award (AQT). 618 NMN was supported by The Shurl and Kay Curci Foundation. 619 620

621 We thank our study participants and collaborators. 622 623 DISCLOSURES 624 The authors have no conflicts of interest. 625 626 AUTHOR CONTRIBUTIONS 627 AQT and ATB designed the study protocol; ATB, AKP, and LP conducted the experiments; 628 ATB, NMN, and LMP analyzed the data; NMN drafted the original manuscript. All authors 629 contributed to the interpretation of the results and revision of the manuscript; All authors 630 approved the final version of the manuscript. 631 632

633 Ahuja, S., & Franz, J. R. (2022). The metabolic cost of walking balance control and 634 adaptation in young adults. Gait and Posture, 96, 190–194. 635 https://doi.org/10.1016/j.gaitpost.2022.05.031 636 Arora, T., Musselman, K. E., Lanovaz, J. L., Linassi, G., Arnold, C., Milosavljevic, 637 S., & Oates, A. (2019). Walking stability during normal walking and its 638 association with slip intensity among individuals with incomplete spinal cord 639 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION injury. Journal of Injury, Function and Rehabilitation, 11(3), 270–277. 640 https://doi.org/10.1016/j.pmrj.2018.07.012 641 Arora, T., Oates, A., Lynd, K., & Musselman, K. E. (2020). Current state of balance 642 assessment during transferring, sitting, standing and walking activities for the 643 spinal cord injured population: A systematic review. The Journal of Spinal Cord 644 Medicine, 43(1), 10–23. https://doi.org/10.1080/10790268.2018.1481692 645 Bauby, C. E., & Kuo, A. D. (2000). Active control of lateral balance in human 646 walking. Journal of Biomechanics, 33(11), 1433–1440. 647 https://doi.org/10.1016/S0021-9290(00)00101-9 648 Bogard, A. T., Hembree, T. G., Pollet, A. K., Smith, A. C., Ryder, S. C., Marzloff, G. 649 E., & Tan, A. Q. (2024). Intermittent hypoxia-induced enhancements in 650 corticospinal excitability predict gains in motor learning and metabolic 651 efficiency. Research Square, PREPRINT. https://doi.org/10.21203/rs.3.rs-652 4259378/v1 653 Bogard, A. T., Hemmerle, M. R., Smith, A. C., & Tan, A. Q. (2023). Enhanced motor 654 learning and motor savings after acute intermittent hypoxia are associated with 655 a reduction in metabolic cost. The Journal of Physiology, 1–21. 656 https://doi.org/10.1113/JP285425 657 Bogard, A. T., Pollet, A. K., & Tan, A. Q. (2024). Intermittent hypoxia enhances 658 voluntary activation and reduces performance fatigability during repeated lower 659 limb contractions. Journal of Neurophysiology, 132(6), 1717–1728. 660 https://doi.org/10.1152/jn.00385.2024 661 Bogard, A. T., & Tan, A. Q. (2024). Author response to Panza et al. The Journal of 662 Physiology. https://doi.org/10.1113/JP287844 663 Brinkerhoff, S. A., Sánchez, N., Culver, M. N., Murrah, W. M., Robinson, A. T., 664 Mccullough, J. D., Miller, M. W., & Roper, J. A. (2024). The dual timescales of 665 gait adaptation: Initial stability adjustments followed by subsequent energetic 666 cost adjustments. Journal of Experimental Biology, 227(23). 667 https://doi.org/https://doi.org/10.1242/jeb.249217 668 Bruijn, S. M., & Van Dieën, J. H. (2018). Control of human gait stability through foot 669 placement. Journal of The Royal Society Interface, 15(143). 670 https://doi.org/10.1098/rsif.2017.0816 671 Butterfield, J. K., & Collins, S. H. (2022). The energy cost of split-belt walking for a 672 variety of belt speed combinations. Journal of Biomechanics, 132. 673 https://doi.org/10.1016/j.jbiomech.2021.110905 674 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION Buurke, T. J. W., & den Otter, R. (2021). The relationship between the 675 anteroposterior and mediolateral margins of stability in able-bodied human 676 walking. Gait & Posture, 90, 80–85. 677 https://doi.org/10.1016/j.gaitpost.2021.08.008 678 Buurke, T. J. W., Lamoth, C. J. C., Van Der Woude, L. H. V, & Den Otter, R. (2019). 679 Handrail holding during treadmill walking reduces locomotor learning in able-680 bodied persons. IEEE Transactions on Neural Systems and Rehabilitation 681 Engineering, 27(9), 1753–1759. https://doi.org/10.1109/tnsre.2019.2935242 682 Buurke, T. J. W., Lamoth, C. J. C., Vervoort, D., van der Woude, L. H. V., & den 683 Otter, R. (2018). Adaptive control of dynamic balance in human gait on a split-684 belt treadmill. Journal of Experimental Biology, 221(13). 685 https://doi.org/10.1242/jeb.174896 686 Buurke, T. J. W., Lamoth, J. C., Van Der Woude, L. H. V, Hof, A. L., & Otter, R. den. 687 (2019). Bilateral temporal control determines mediolateral margins of stability 688 in symmetric and asymmetric human walking. Scientific Reports, 9(12494). 689 https://doi.org/10.1038/s41598-019-49033-z 690 Buurke, T. J. W., Liu, C., Park, S., den Otter, R., & Finley, J. M. (2020). Maintaining 691 sagittal plane balance compromises frontal plane balance during reactive 692 stepping in people post-stroke. Clinical Biomechanics, 80, 105135. 693 https://doi.org/10.1016/j.clinbiomech.2020.105135 694 Buurke, T. J. W., Sharma, N., Swart, S. B., van der Woude, L. H. V., den Otter, R., & 695 Lamoth, C. J. C. (2022). Split-belt walking: An experience that is hard to forget. 696 Gait & Posture, 97, 184–187. https://doi.org/10.1016/j.gaitpost.2022.08.003 697 Cesari, P., Piscitelli, F., Pascucci, F., & Bertucco, M. (2022). Postural Threat 698 Influences the Coupling Between Anticipatory and Compensatory Postural 699 Adjustments in Response to an External Perturbation. Neuroscience, 490, 25–700 35. https://doi.org/10.1016/j.neuroscience.2022.03.005 701 Choi, J. T., & Bastian, A. J. (2007). Adaptation reveals independent control 702 networks for human walking. Nature Neuroscience, 10(8), 1055–1062. 703 https://doi.org/10.1038/nn1930 704 Christiansen, L., Larsen, M. N., Madsen, M. J., Grey, M. J., Nielsen, J. B., & 705 Lundbye-Jensen, J. (2020). Long-term motor skill training with individually 706 adjusted progressive difficulty enhances learning and promotes corticospinal 707 plasticity. Scientific Reports, 10(1). https://doi.org/10.1038/s41598-020-72139-708 8 709 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION Christiansen, L., Urbin, M. A., Mitchell, G. S., & Perez, M. A. (2018). Acute 710 intermittent hypoxia enhances corticospinal synaptic plasticity in humans. 711 ELife, 7. https://doi.org/10.7554/eLife.34304 712 Cohen, J. (1988). Statistical Power Analysis for the Behavioral Sciences (2nd ed.). 713 Routledge. https://doi.org/10.4324/9780203771587 714 Cornwell, T., Novotny, R., & Finley, J. M. (2024). Associations between asymmetry 715 and reactive balance control during split-belt walking. Journal of Biomechanics, 716 172, 112221. https://doi.org/10.1016/j.jbiomech.2024.112221 717 Curtze, C., Shah, V. V., Stefanko, A. M., Dale, M. L., Nutt, J. G., Mancini, M., & 718 Horak, F. B. (2024). Stride width and postural stability in frontal gait disorders 719 and Parkinson’s disease. Journal of Neurology, 1–10. 720 https://doi.org/10.1007/S00415-024-12401-5 721 Dean, J. C., Alexander, N. B., & Kuo, A. D. (2007). The effect of lateral stabilization 722 on walking in young and old adults. IEEE Transactions on Biomedical 723 Engineering, 54(11), 1919–1926. https://doi.org/10.1109/tbme.2007.901031 724 Debelle, H., Harkness-Armstrong, C., Hadwin, K., Maganaris, C. N., & O’Brien, T. 725 D. (2020). Recovery From a Forward Falling Slip: Measurement of Dynamic 726 Stability and Strength Requirements Using a Split-Belt Instrumented Treadmill. 727 Frontiers in Sports and Active Living, 2, 522764. 728 https://doi.org/10.3389/fspor.2020.00082 729 Deffeyes, J. E., & Peters, D. M. (2021). Time-integrated propulsive and braking 730 impulses do not depend on walking speed. Gait & Posture, 88, 258–263. 731 https://doi.org/10.1016/j.gaitpost.2021.06.012 732 Donelan, J. M., Kram, R., & Kuo, A. D. (2001). Mechanical and metabolic 733 determinants of the preferred step width in human walking. Proceedings of the 734 Royal Society Biological Sciences, 268(1480), 1985–1992. 735 https://doi.org/10.1098/rspb.2001.1761 736 Donelan, J. M., Kram, R., & Kuo, A. D. (2002). Mechanical work for step-to-step 737 transitions is a major determinant of the metabolic cost of human walking. 738 Journal of Experimental Biology, 205(23), 3717–3727. 739 https://doi.org/10.1242/jeb.205.23.3717 740 Donelan, J. M., Shipman, D. W., Kram, R., & Kuo, A. D. (2004). Mechanical and 741 metabolic requirements for active lateral stabilization in human walking. 742 Journal of Biomechanics, 37(6), 827–835. 743 https://doi.org/10.1016/j.jbiomech.2003.06.002 744 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION Emken, J. L., Benitez, R., Sideris, A., Bobrow, J. E., & Reinkensmeyer, D. J. (2007). 745 Motor adaptation as a greedy optimization of error and effort. Journal of 746 Neurophysiology, 97(6), 3997–4006. https://doi.org/10.1152/jn.01095.2006 747 Fettrow, T., Hupfeld, K., Reimann, H., Choi, J., Hass, C., & Seidler, R. (2021). Age 748 differences in adaptation of medial-lateral gait parameters during split-belt 749 treadmill walking. Scientific Reports 2021 11:1, 11(1), 1–17. 750 https://doi.org/10.1038/s41598-021-00515-z 751 Finley, J. M., Bastian, A. J., & Gottschall, J. S. (2013). Learning to be economical: 752 The energy cost of walking tracks motor adaptation. Journal of Physiology, 753 591(4), 1081–1095. https://doi.org/10.1113/jphysiol.2012.245506 754 Fritsch, B., Reis, J., Martinowich, K., Schambra, H. M., Ji, Y., Cohen, L. G., & Lu, B. 755 (2010). Direct current stimulation promotes BDNF-dependent synaptic 756 plasticity: Potential implications for motor learning. Neuron, 66(2), 198–204. 757 https://doi.org/10.1016/j.neuron.2010.03.035 758 Gottschall, J. S., & Kram, R. (2003). Energy cost and muscular activity required for 759 propulsion during walking. Journal of Applied Physiology, 94, 1766–1772. 760 https://doi.org/10.1152/japplphysiol.00670.2002 761 Grabowski, A., Farley, C. T., & Kram, R. (2005). Independent metabolic costs of 762 supporting body weight and accelerating body mass during walking. Journal of 763 Applied Physiology, 98(2), 579–583. 764 https://doi.org/10.1152/japplphysiol.00734.2004 765 Hak, L., Houdijk, H., Steenbrink, F., Mert, A., van der Wurff, P., Beek, P. J., & Van 766 Dieën, J. H. (2013). Stepping strategies for regulating gait adaptability and 767 stability. Journal of Biomechanics, 46(5), 905–911. 768 https://doi.org/10.1016/j.jbiomech.2012.12.017 769 Hayes, H. B., Jayaraman, A., Herrmann, M., Mitchell, G. S., Rymer, W. Z., & 770 Trumbower, R. D. (2014). Daily intermittent hypoxia enhances walking after 771 chronic spinal cord injury. Neurology, 82(2), 104–113. 772 https://doi.org/10.1212/01.wnl.0000437416.34298.43 773 Helm, E. E., Tyrell, C. M., Pohlig, R. T., Brady, L. D., & Reisman, D. S. (2016). The 774 presence of a single-nucleotide polymorphism in the BDNF gene affects the 775 rate of locomotor adaptation after stroke. Experimental Brain Research, 776 234(2), 341–351. https://doi.org/10.1007/s00221-015-4465-8 777 Hof, A. L., Gazendam, M. G. J., & Sinke, W. E. (2005). The condition for dynamic 778 stability. Journal of Biomechanics, 38, 1–8. 779 https://doi.org/10.1016/j.jbiomech.2004.03.025 780 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION Huang, H. J., Kram, R., & Ahmed, A. A. (2012). Reduction of Metabolic Cost during 781 Motor Learning of Arm Reaching Dynamics. Journal of Neuroscience, 32(6), 782 2182–2190. https://doi.org/10.1523/jneurosci.4003-11.2012 783 John, C. T., Seth, A., Schwartz, M. H., & Delp, S. L. (2012). Contributions of 784 muscles to mediolateral ground reaction force over a range of walking speeds. 785 Journal of Biomechanics, 45(14), 2438–2443. 786 https://doi.org/10.1016/j.jbiomech.2012.06.037 787 Jørgensen, V., Butler Forslund, E., Franzén, E., Opheim, A., Seiger, Å., Ståhle, A., 788 Hultling, C., Stanghelle, J. K., Wahman, K., & Skavberg Roaldsen, K. (2016). 789 Factors Associated With Recurrent Falls in Individuals With Traumatic Spinal 790 Cord Injury: A Multicenter Study. Archives of Physical Medicine and 791 Rehabilitation, 97(11), 1908–1916. https://doi.org/10.1016/j.apmr.2016.04.024 792 Joundi, R. A., Lopez-Alonso, V., Lago, A., Brittain, J. S., Fernandez-Del-Olmo, M., 793 Gomez-Garre, P ., Mir, P., Jenkinson, N., Cheeran, B., & Brown, P. (2012). The 794 effect of BDNF val66met polymorphism on visuomotor adaptation. 795 Experimental Brain Research, 223(1), 43–50. https://doi.org/10.1007/s00221-796 012-3239-9 797 Karakasis, C., & Artemiadis, P. (2021). Real-time kinematic-based detection of foot-798 strike during walking. Journal of Biomechanics, 129, 110849. 799 https://doi.org/10.1016/j.jbiomech.2021.110849 800 Kuo, A. D., & Donelan, J. M. (2010). Dynamic Principles of Gait and Their Clinical 801 Implications. Physical Therapy, 90(2), 157–174. 802 https://doi.org/10.2522/ptj.20090125 803 Leech, K. A., & Hornby, T. G. (2017). High-Intensity Locomotor Exercise Increases 804 Brain-Derived Neurotrophic Factor in Individuals with Incomplete Spinal Cord 805 Injury. Https://Home.Liebertpub.Com/Neu, 34(6), 1240–1248. 806 https://doi.org/10.1089/neu.2016.4532 807 Leech, K. A., Roemmich, R. T., & Bastian, A. J. (2018). Creating flexible motor 808 memories in human walking. Scientific Reports 2017 8:1, 8(1), 1–10. 809 https://doi.org/10.1038/s41598-017-18538-w 810 Leech, K. A., Roemmich, R. T., Gordon, J., Reisman, D. S., & Cherry-Allen, K. M. 811 (2022). Updates in Motor Learning: Implications for Physical Therapist Practice 812 and Education. Physical Therapy, 102(1), 1–9. 813 https://doi.org/10.1093/ptj/pzab250 814 Lovett-Barr, M. R., Satriotomo, I., Muir, G. D., Wilkerson, J. E. R., Hoffman, M. S., 815 Vinit, S., & Mitchell, G. S. (2012). Repetitive Intermittent Hypoxia Induces 816 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION Respiratory and Somatic Motor Recovery after Chronic Cervical Spinal Injury. 817 Journal of Neuroscience, 32(11), 3591–3600. 818 https://doi.org/10.1523/jneurosci.2908-11.2012 819 Mang, C. S., Snow, N. J., Campbell, K. L., Ross, C. J. D., & Boyd, L. A. (2014). A 820 single bout of high-intensity aerobic exercise facilitates response to paired 821 associative stimulation and promotes sequence-specific implicit motor learning. 822 Journal of Applied Physiology, 117(11), 1325–1336. 823 https://doi.org/10.1152/japplphysiol.00498.2014 824 Masani, K., Kouzaki, M., & Fukunaga, T. (2002). Variability of ground reaction 825 forces during treadmill walking. Journal of Applied Physiology, 92(5), 1885–826 1890. https://doi.org/10.1152/japplphysiol.00969.2000 827 Mawase, F., Haizler, T., Bar-Haim, S., & Karniel, A. (2013). Kinetic adaptation 828 during locomotion on a split-belt treadmill. Journal of Neurophysiology, 109, 829 2216–2227. https://doi.org/10.1152/jn.00938.2012 830 McAndrew Young, P. M., & Dingwell, J. B. (2012). Voluntary changes in step width 831 and step length during human walking affect dynamic margins of stability. Gait 832 & Posture, 36(2), 219–224. https://doi.org/10.1016/j.gaitpost.2012.02.020 833 McAndrew Young, P. M., Wilken, J. M., & Dingwell, J. B. (2012). Dynamic margins 834 of stability during human walking in destabilizing environments. Journal of 835 Biomechanics, 45(6), 1053–1059. 836 https://doi.org/10.1016/j.jbiomech.2011.12.027 837 McGeer, T. (1990). Passive Dynamic Walking. The International Journal of Robotics 838 Research, 9(2), 62–82. https://doi.org/10.1177/027836499000900206 839 McIlroy, W. E., & Maki, B. E. (1996). Age-related Changes in Compensatory 840 Stepping in Response to Unpredictable Perturbations. The Journals of 841 Gerontology: Series A, 51A(6), M289–M296. 842 https://doi.org/10.1093/gerona/51A.6.M289 843 Montgomery, J. R., & Grabowski, A. M. (2018). The contributions of ankle, knee and 844 hip joint work to individual leg work change during uphill and downhill walking 845 over a range of speeds. Royal Society Open Science, 5(8). 846 https://doi.org/10.1098/rsos.180550 847 Navarrete-Opazo, A., Alcayaga, J., Sepúlveda, O., Rojas, E., & Astudillo, C. (2017). 848 Repetitive Intermittent Hypoxia and Locomotor Training Enhances Walking 849 Function in Incomplete Spinal Cord Injury Subjects: A Randomized, Triple-850 Blind, Placebo-Controlled Clinical Trial. Journal of Neurotrauma, 34(9), 1803–851 1812. https://doi.org/10.1089/neu.2016.4478 852 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION Park, S., & Finley, J. M. (2017). Characterizing Dynamic Balance during Adaptive 853 Locomotor Learning. IEEE Engineering in Medicine and Biology Society, 50–854 53. https://doi.org/10.1109/embc.2017.8036760 855 Péronnet, F., & Massicotte, D. (1991). Table of nonprotein respiratory quotient: an 856 update. Canadian Journal of Sport Science, 16(1), 23–29. 857 Pieper, N. L., Baudendistel, S. T., Hass, C. J., Diaz, G. B., Krupenevich, R. L., & 858 Franz, J. R. (2021). The metabolic and mechanical consequences of altered 859 propulsive force generation in walking. Journal of Biomechanics, 122, 110447. 860 https://doi.org/10.1016/j.jbiomech.2021.110447 861 Price, M., Huber, M. E., & Hoogkamer, W. (2023). Minimum effort simulations of 862 split-belt treadmill walking exploit asymmetry to reduce metabolic energy 863 expenditure. Journal of Neurophysiology, 129(4), 900–913. 864 https://doi.org/10.1152/jn.00343.2022 865 Rasmussen, P., Brassard, P., Adser, H., Pedersen, M. V., Leick, L., Hart, E., Secher, 866 N. H., Pedersen, B. K., & Pilegaard, H. (2009). Evidence for a release of brain-867 derived neurotrophic factor from the brain during exercise. Experimental 868 Physiology, 94(10), 1062–1069. 869 https://doi.org/10.1113/expphysiol.2009.048512 870 Rawal, Y. R., & Singer, J. C. (2021). The influence of net ground reaction force 871 orientation on mediolateral stability during walking. Gait & Posture, 90, 73–79. 872 https://doi.org/10.1016/j.gaitpost.2021.08.009 873 Reisman, D. S., Block, H. J., & Bastian, A. J. (2005). Interlimb coordination during 874 locomotion: What can be adapted and stored? Journal of Neurophysiology, 875 94(4), 2403–2415. https://doi.org/10.1152/jn.00089.2005 876 Richards, J. T., Selgrade, B. P., Qiao, M., Plummer, P., Wikstrom, E. A., & Franz, J. 877 R. (2019). Time-dependent tuning of balance control and aftereffects following 878 optical flow perturbation training in older adults. Journal of NeuroEngineering 879 and Rehabilitation, 16(1). https://doi.org/10.1186/s12984-019-0555-3 880 Roemmich, R. T., & Bastian, A. J. (2015). Two ways to save a newly learned motor 881 pattern. Journal of Neurophysiology, 113(10), 3519–3530. 882 https://doi.org/10.1152/jn.00965.2014 883 Rogers, M. W., Hedman, L. D., Johnson, M. E., Cain, T. D., & Hanke, T. A. (2001). 884 Lateral Stability During Forward-Induced Stepping for Dynamic Balance 885 Recovery in Young and Older Adults. The Journals of Gerontology: Series A, 886 56(9), M589–M594. https://doi.org/10.1093/gerona/56.9.M589 887 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION Roper, J. A., Roemmich, R. T., Tillman, M. D., Terza, M. J., & Hass, C. J. (2017). 888 Split-belt treadmill walking alters lower extremity frontal plane mechanics. 889 Journal of Applied Biomechanics, 33(4), 256–260. 890 https://doi.org/10.1123/jab.2016-0059 891 Sánchez, N., Park, S., & Finley, J. M. (2017). Evidence of Energetic Optimization 892 during Adaptation Differs for Metabolic, Mechanical, and Perceptual Estimates 893 of Energetic Cost. Scientific Reports, 7(1), 1–14. 894 https://doi.org/10.1038/s41598-017-08147 895 Sánchez, N., Simha, S. N., Donelan, J. M., & Finley, J. M. (2019). Taking advantage 896 of external mechanical work to reduce metabolic cost: the mechanics and 897 energetics of split-belt treadmill walking. The Journal of Physiology, 597(15), 898 4053–4068. https://doi.org/10.1113/jp277725 899 Sánchez, N., Simha, S. N., Donelan, J. M., & Finley, J. M. (2021). Using asymmetry 900 to your advantage: Learning to acquire and accept external assistance during 901 prolonged split-belt walking. Journal of Neurophysiology, 125(2), 344–357. 902 https://doi.org/10.1152/jn.00416.2020 903 Sato, S., & Choi, J. T. (2022). Neural Control of Human Locomotor Adaptation: 904 Lessons about Changes with Aging. Neuroscientist, 28(5), 469–484. 905 https://doi.org/10.1177/10738584211013723 906 Sawers, A., Kelly, V. E., Kartin, D., & Hahn, M. E. (2013). Gradual training reduces 907 the challenge to lateral balance control during practice and subsequent 908 performance of a novel locomotor task. Gait & Posture, 38(4), 907–911. 909 https://doi.org/10.1016/j.gaitpost.2013.04.019 910 Selinger, J. C., O’Connor, S. M., Wong, J. D., & Donelan, J. M. (2015). Humans 911 Can Continuously Optimize Energetic Cost during Walking. Current Biology, 912 25(18), 2452–2456. https://doi.org/10.1016/j.cub.2015.08.016 913 Sombric, C. J., & Torres-Oviedo, G. (2020). Augmenting propulsion demands during 914 split-belt walking increases locomotor adaptation of asymmetric step lengths. 915 Journal of NeuroEngineering and Rehabilitation, 17(1), 1–15. 916 https://doi.org/10.1186/S12984-020-00698 917 Stenum, J., & Choi, J. T. (2020). Step time asymmetry but not step length 918 asymmetry is adapted to optimize energy cost of split-belt treadmill walking. 919 The Journal of Physiology, 598(18), 4063–4078. 920 https://doi.org/10.1113/jp279195 921 Tan, A. Q., Papadopoulos, J. M., Corsten, A. N., & Trumbower, R. D. (2020). An 922 automated pressure-swing absorption system to administer low oxygen 923 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION therapy for persons with spinal cord injury. Experimental Neurology, 333, 924 113408. https://doi.org/10.1016/j.expneurol.2020.113408 925 Tan, A. Q., Sohn, W. J., Naidu, A., & Trumbower, R. D. (2021). Daily acute 926 intermittent hypoxia combined with walking practice enhances walking 927 performance but not intralimb motor coordination in persons with chronic 928 incomplete spinal cord injury. Experimental Neurology, 340, 113669. 929 https://doi.org/10.1016/j.expneurol.2021.113669 930 Thompson, E. D., & Reisman, D. S. (2022). Split-Belt Adaptation and Savings in 931 People with Parkinson Disease. Journal of Neurologic Physical Therapy, 46(4), 932 293–301. https://doi.org/10.1097/npt.0000000000000411 933 934 935 936 937 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION 938 Figure 1. Step width adaptation. A) Both legs increased their step width from baseline 939 during early learning then significantly narrowed during late learning of adapt 1. B) During 940 adapt 2, both legs widened their steps during early learning but only the slow leg 941 significantly decreased its width during late learning. The AIH group demonstrated early 942 savings of step width in both the fast leg ( C) and slow leg (D) compared to controls. Bar 943 graphs represent mean step width and standard error (SE) . ∗ P < 0.05, ∗∗ P < 0.01, ∗∗∗ 944 P < 0.001. 945 A C Adapt 2 ** ** * Baseline Early Learning Late Learning 0 0.1 0.2 0.3 Ste p Wid th (m) Early Savings for F ast Leg ** Adapt 1 Adapt 2 0 0.1 0.2 0.3 Ste p Wid th (m) Adapt 1 *** *** *** *** Baseline Early Learning Late Learning 0 0.1 0.2 0.3 Ste p Wid th (m) Fast Leg Slow Leg Both Legs B Early Savings for Slow Leg * Adapt 1 Adapt 2 0 0.1 0.2 0.3 Ste p Wid th (m) AIH Control D .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION 946 Figure 2. Peak mediolateral ground reaction forces (ML GRF) during braking phase. 947 A) peak ML GRF increased from baseline for both legs in early learning and decreased 948 for the slow leg in late learning of adapt 1. B) between-limb differences show higher peak 949 ML GRF maintained by the fast leg during early and late learning of adapt 2 compared to 950 the slow leg. There were no early savings strategies observed for the fast leg ( C), whilst 951 the slow leg significantly decreased magnitude responses during the second adaptation 952 for both groups (D). Bar graphs show averaged peak ML GRF magnitudes and standard 953 error (SE). ∗ P < 0.05, ∗∗ P < 0.01, ∗∗∗ P < 0.001. The red line represents significan t 954 differences between limbs. 955 A B C Braking Phase during Adapt 1 *** *** *** *** *** Baseline Early Learning Late Learning 0 0.1 0.2 0.3 Peak Mediolateral Force (BW) ********* *** *** * Baseline Early Learning Late Learning 0 0.1 0.2 0.3 Peak Mediolateral Force (BW) Braking Phase during Adapt 2 Early Savings during Braking Phase for Fast Leg Adapt 1 Adapt 2 0 0.1 0.2 0.3 Peak Mediolateral Force (BW) Fast Leg Slow Leg Both Legs Early Savings during Braking Phase for Slow Leg *** ** Adapt 1 Adapt 2 0 0.1 0.2 0.3 Peak Mediolateral Force (BW) AIH Control D .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION 956 Figure 3. Peak mediolateral ground reaction forces (ML GRF) during propulsive 957 phase. A) Only the fast leg increased peak ML GRF during early learning which 958 decreased in late learning of adapt 1 while the slow leg stayed at baseline. B ) During 959 adapt 2, the fast leg shows a similar adaptation strategy as in adapt 1 and the slow leg 960 decreases peak ML GRF from baseline in early and late learning. C) Both the AIH and 961 control groups showed early savings on the fast leg. D) Only the AIH group exhibited early 962 savings on the slow leg. Bar graphs show mean peak ML GRF magnitudes and standard 963 error (SE). ∗ P < 0.05, ∗∗ P < 0.01, ∗∗∗ P < 0.001. The red line represents significan t 964 differences between limbs. 965 A B Propulsive Phase during Adapt 1 *** ****** Baseline Early Learning Late Learning 0 0.1 0.2 0.3 Peak Mediolateral Force (BW) Propulsive Phase during Adapt 2 *** ****** *** * Baseline Early Learning Late Learning 0 0.1 0.2 0.3 Peak Mediolateral Force (BW) Early Savings during Propulsive Phase for Fast Leg *** *** Adapt 1 Adapt 2 0 0.1 0.2 0.3 Peak Mediolateral Force (BW) C D Fast Leg Slow Leg Both Legs Early Savings during Propulsive Phase for Slow Leg * Adapt 1 Adapt 2 0 0.1 0.2 0.3 Peak Mediolateral Force (BW) AIH Control .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION 966 Figure 4. Net metabolic power. A) Both groups decreased net metabolic power during 967 late learning of adapt 1. B) During adapt 2, the AIH group increased net metabolic power 968 during late learning. Both groups demonstrated early savings (C) and late savings (D) of 969 net metabolic power. Bar graphs show average net metabolic power (W/Kg) and standard 970 error (SE). ∗ P < 0.05, ∗∗ P < 0.01, ∗∗∗ P < 0.001 971 972 973 Adapt 2 AIH Control ** Early Learning Late Learning 5.5 7.5 9.5 Net Metabolic Power (W/kg) Adapt 1 *** *** Early Learning Late Learning Net Metabolic Power (W/kg) 5.5 7.5 9.5 A B DC Early Savings *** *** Adapt 1 Adapt 2 Net Metabolic Power (W/kg) 5.5 7.5 9.5 Late Savings * * Adapt 1 Adapt 2 Net Metabolic Power (W/kg) 5.5 7.5 9.5 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION 974 Figure 5. Metabolic regressions for fast and slow legs during adapt 1 in the control 975 group. A) Nonsignificant positive relationship between decreases in net metabolic power 976 and step width on the fast leg (R 2 = 0.203, P = 0.092). B ) N onsignificant positive 977 relationship between decreases in net metabolic power and step width on the slow leg 978 (R2 = 0.254, P = 0.055). C) S ignificant positive correlation between decreases in net 979 metabolic power and peak ML GRF during propulsive phase on the fast leg (R 2 = 0.728, 980 P < 0.001). D) Significant positive correlation between decreases in net metabolic power 981 and decreases in peak ML GRF during the braking phase on the slow leg (R2 = 0.328, P 982 = 0.026). Area within blue lines represents 95% confidence and grey points are individual 983 participants’ data. ∗ P < 0.05, ∗∗ P < 0.01, ∗∗∗ P < 0.001. 984 985 BA 0 0.05 0.1 -0.5 0 0.5 1 1.5 R 2= 0.728 Peak MLForce during Propulsive Phase for Fast Leg (BW) P< 0.001 Greater reduction in peak M Lf orce *** -0.05 0 0.05 0.1 -0.5 0 0.5 1 1.5 P= 0.092 Step Width for Fast Leg (m) Greater reduction in step width R 2= 0.203 0.04 -0.02 0 0.02 0.04 0.06 -0.5 0 0.5 1 1.5 P = 0.026 Peak MLForce during Braking Phase for Slow Leg (BW) Greater reduction in peak M Lf orce R 2= 0.328 * -0.05 0 0.05 0.1 -0.5 0 0.5 1 1.5 P = 0.055 Step Width for Slow Leg (m) Greater reduction in step width R 2= 0.254 C D .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION Step Width (mm) Adapt 1 Adapt 2 AIH Control AIH Control Early Learning 0.158 (0.058) 0.176 (0.030) 0.130 (0.055) 0.142 (0.032) Late Learning 0.145 (0.032) 0.144 (0.065) 0.137 (0.038) 0.147 (0.045) Table 1. Mean (SD) values of step width (mm) during early and late learning during adapt 986 1 and adapt 2 for the AIH and Control groups. 987 988 Peak ML GRF (BW) Adapt 1 Adapt 2 AIH Control AIH Control Braking Phase Early Learning 0.141 (0.046) 0.123 (0.023) 0.117 (0.030) 0.111 (0.024) Late Learning 0.114 (0.032) 0.115 (0.029) 0.114 (0.028) 0.111 (0.029) Propulsive Phase Early Learning 0.106 (0.042) 0.096 (0.034) 0.078 (0.018) 0.078 (0.020) Late Learning 0.069 (0.015) 0.073 (0.019) 0.071 (0.011) 0.072 (0.014) Table 2. Mean (SD) values of peak ML GRF (BW) d uring the propulsive and braking 989 phases of early and late learning during adapt 1 and adapt 2 for the AIH and Control 990 groups. 991 992 993 994 995 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint MEDIOLATERAL ADAPTATIONS TO DESTABILIZATION Net Metabolic Power (W/kg) Adapt 1 Adapt 2 AIH Control AIH Control Early Learning 6.959 (0.516) 6.957 (0.825) 6.026 (0.430) 6.355 (0.774) Late Learning 6.468 (0.475) 6.527 (0.751) 6.283 (0.427) 6.345 (0.872) Table 3. Mean (SD) values of net metabolic power (W/kg) during early and late learning 996 of adapt 1 and adapt 2 for the AIH and Control groups. 997 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 29, 2024. ; https://doi.org/10.1101/2024.12.29.630661doi: bioRxiv preprint