Metabolic costs of walking and arm reaching in persons with mild multiple sclerosis

Abstract

Movement slowness is a common and disruptive symptom of multiple sclerosis (MS). A potential cause is that individuals with MS slow down to conserve energy as a behavioral adjustment to heightened metabolic costs of movement. To investigate this prospect, we measured the metabolic costs of both walking and seated arm reaching at five speeds in persons with mild MS (pwMS; n = 13; 46.0 ± 7.7 yr) and sex- and age-matched controls (HCs; n = 13; 45.8 ± 7.8 yr). Notably, the cohort of pwMS was highly mobile and no individuals required a cane or aid when walking. We found that the net metabolic power of walking was approximately 20% higher for pwMS across all speeds (P = 0.0185). In contrast, we found no differences in the gross power of reaching between pwMS and HCs (P = 0.492). Collectively, our results suggest that abnormal slowness of movement in MS—particularly reaching—is not the consequence of heightened effort costs and that other sensorimotor mechanisms are playing a considerable role in slowing.

NEW & NOTEWORTHY Individuals with multiple sclerosis (MS) often move more slowly than those without the disease. A possible cause is that movements in MS are more energetically expensive and slowing is an adaptation to conserve metabolic resources. Here, we find that while walking is more costly for persons with MS, arm-reaching movements are not. These results bring into question the driving force of movement slowness in MS and implicate other motor-related networks contributing to slowing.

INTRODUCTION

An inflammatory disease of the central nervous system (CNS), multiple sclerosis (MS) elicits a myriad of sensory, cognitive, and motor symptoms. The symptoms are consequences of axonal demyelination, or potential total neurodegeneration, which disrupts efferent and afferent signal transmission between brain and body (1–3). One of the primary barriers to our understanding of MS rests in its heterogeneous presentation of neurological disability dependent upon the location of these demyelinated lesions within the CNS (4).

Despite this high person-to-person variability, a prevalent symptom experienced by persons with MS (pwMS) is movement slowness in walking (5–9), upper limb movements (10–14), and even saccades (15–17). There are, however, several potential causes of movement slowness in pwMS that likely depend on the location of inflammation in the CNS. Slowness in walking or reaching could be due to pyramidal tract lesions leading to spasticity (18) or due to motor unit loss contributing to muscle force unsteadiness (19, 20); whereas saccade dysfunction could be linked to cerebellar damage causing tremor or dysmetria (17). Focusing specifically on locomotion, spasticity, for example, is linked to alterations in gait such as increased double support time (9, 21) or heightened spatiotemporal variability (22, 23) that contribute to reduced walking speeds and increased metabolic costs for pwMS (24).

In healthy adults and animals, preferred movement speeds are often selected to reduce the associated metabolic effort, and increasing the metabolic effort of moving consequently leads to slower walking and reaching (25–31). Hence, walking and reaching speeds can be partially understood in terms of their cost function that includes minimizing effort expenditure as an objective (32, 33). Abnormal movement slowness is a prominent feature of many neurological disorders, including MS, which suggests that movements may become more costly, or at least perceived as such, during the disease course.

The energetic costs of walking (5, 7, 9, 24, 34–36) and other mobility tasks (37, 38) trend higher for pwMS, thus walking slowness in MS could be credited to higher effort. Gait-related impairments caused by disruptions such as spasticity (39), loss of dynamic balance (40), or reduced cardiorespiratory fitness (9, 41–45) may contribute to these elevated costs. Rehabilitating mobility for pwMS frequently involves improving physical fitness, which has been shown to both reduce energetic costs and improve walking speeds (44). Overall, regardless of the distinct sensorimotor mechanisms increasing the metabolic costs of walking for pwMS, it appears that walking slowness may be a suitable response to conserve metabolic resources.

In addition, large percentages of pwMS report upper extremity impairment (46), which, like walking, is associated with cognitive function and can reduce the quality of life (14). Part of this MS-related upper limb impairment includes slowness of reaching movements (10–14). Traditionally, the speed of reaching movements is thought to be primarily related to accuracy, such that more accurate reaches beg slower speeds (47). PwMS present more variable and less symmetric reaching velocity profiles as well as a worsened speed-accuracy tradeoff (48–50), which could be driving this reaching slowness. Yet, the energetic costs associated with these upper extremity movements remain unstudied in MS. Could slowness of arm reaching movements for pwMS also be explained by increased metabolic costs? Here, we investigated the costs of seated arm reaching movements, which involve small net metabolic costs (32), and may not elicit a full-fledged sympathetic response to exercise due to the low workloads, unlike locomotion (51, 52). Performing these movements seated also removes many of the covariates linked to increased walking costs for pwMS—namely, lower limb spasticity and balance deficits (39, 40). Seated reaching offers an alternative modality to investigate altered energy demands of movement in MS.

Accordingly, the purpose of our study was to characterize the energetics of both walking and seated arm reaching for persons with mild MS as compared with healthy age- and sex-matched control participants (HCs). We used indirect calorimetry and systematically measured the metabolic costs of walking and reaching across five different speeds. Our goal was to determine whether, similar to the metabolic cost of walking, the metabolic cost of reaching would be elevated in persons with mild MS. We first hypothesized that the metabolic effort of walking would be higher for pwMS. However, due to its separation from many of the potential sensorimotor covariates of heightened walking effort, we hypothesized that the metabolic effort of reaching would not differ for pwMS and HCs. If the costs of reaching are not abnormally high, other mechanisms beyond metabolic effort may be contributing to slowness. If reaching costs are elevated, this may suggest that movement slowing in pwMS in both walking and reaching is a rational response to increased metabolic costs.

METHODS

Participants

Persons with MS (pwMS) (n = 13; 46.0 ± 7.7 yr; 11 females) and healthy, age- and sex-matched (±3 yr) controls (HCs) (n = 13; 45.8 ± 7.8 yr; 11 females) were recruited upon passing a strict prescreening process and participated after written and informed consent was obtained. All procedures were approved by the University of Colorado Boulder Institutional Review Board (Protocol No. 20-0072).

Primary inclusion criteria involved: 1) between the ages of 18 and 65 yr old; 2) within ±3 yr of age to a pwMS (HC-group only); 3) clinical diagnosis of MS (pwMS-group only) and free of other neurological disorders; 4) Patient Determined Disease Steps (PDDS) score ≤ 3 or ambulatory without an assistive device (pwMS-group only); and 5) on stable doses of symptomatic-treating medications (pwMS-group only). Primary exclusion criteria involved: 1) a relapse or administration of systemic steroids within the last 30 days (pwMS-group only); or 2) the presence of other risk factors, including other major diseases, cognitive impairment, mental disorders, drug use, spasticity, or recent orthopedic injury that would otherwise limit exercise involvement.

Protocol

Participants visited the laboratory on two separate occasions with 7 to 14 days between visits. On the first day of testing, participants fasted 2 to 4 h in advance. After obtaining written, informed consent, participants first completed a battery of questionnaires [Modified Fatigue Impact Scale (MFIS), the Patient-Reported Outcomes Measurement Information System (PROMIS) Sleep Disturbance, PROMIS Sleep Impairment, and MS Walking Scale (MSWS-12)] and physical assessments [weight, Grooved Pegboard test (GPT), maximal voluntary contraction (MVC) grip, and Timed 25-foot Walk Test (T25FW)].

A standing baseline metabolic trial was performed which involved 5 min of quiet standing on the treadmill while breathing through the metabolic mouthpiece. Participants then walked on the treadmill at five speeds (0.6, 0.85, 1.1, 1.35, and 1.6 m/s) for 5 min each while breathing into the mouthpiece (Fig. 1A). The order of speeds was randomized, and participants were required to take at least 5 min of seated rest between walking conditions. While walking, participants were instructed to “breathe and walk as naturally as possible” and asked to refrain from holding the handrails of the treadmill, if possible. Participants performed a second, standing baseline metabolic trial at the conclusion of the first visit.

Figure 1.

Experimental methods. Metabolics were collected via indirect calorimetry while participants: walked on the treadmill at five speeds, ranging from 0.6 to 1.6 m/s (A); and performed horizontal planar arm reaching movements at five speeds while grasping the handle of the robot arm (B). C: simulated minimum jerk velocity trajectories for a 20 cm reach completed within the five enforced time durations. The enforced times to complete a reach in each speed condition were 1,850, 1,150, 800, 500, and 250 ms ± 50 ms corresponding to the “very slow” (VS), “slow” (S), “medium” (M), “fast” (F), and “very fast” (VF) conditions, respectively. D: audiovisual display presented on the monitor during reaching. Participants controlled the yellow cursor with the robot arm and moved from the home circle to the target circle. Upon arrival at the target circle, participants received differing audiovisual feedback dependent on whether they arrived at the target “Too fast,” “Right on time,” or “Too slow.” After a 300-ms intertrial interval in which participants kept the cursor still in the target circle, a new target appeared in the previous home location and the next trial began.

The second day of testing involved reaching metabolics. Participants were required to fast for a longer period (6–8 h) beforehand, typically overnight. Participants then made seated reaching movements using their dominant arm while breathing into the mouthpiece (Fig. 1B). When reaching, participants grasped the handle of the shoulder-elbow robot and controlled the position of a virtual cursor displayed on a vertically positioned monitor at eye level. Reaches were 20 cm in distance and performed in an out-and-back fashion along an anteroposterior axis originating near the participant’s sternum. Like with walking, participants reached at five different speeds ranging from “very slow” (∼10 cm/s) to “very fast” (∼80 cm/s) for at least 5 min each (Fig. 1C). The order of speeds was again randomized, and participants were required to take at least 5 min of seated rest between conditions. Participants completed a seated baseline metabolic trial at the beginning and at the conclusion of the visit.

Unlike walking on the treadmill, participants could not be forced to reach at a given speed. To enforce reaching speeds, we used audiovisual feedback to encourage participants to reach from one target to the next in a specified amount of time. The position of the handle controlled a cursor (r = 0.3 cm) on a vertical computer monitor in front of the subject positioned at eye level. To begin a trial, participants moved the cursor within the home circle (r = 0.8 cm) and remained still for 300 ms. After this 300 ms delay, the home circle disappeared, and a red circular target (r = 0.8 cm) appeared 20 cm away. Participants were then instructed to reach “out” (i.e., primarily elbow extension) to the new target with this cue and stop within it. After another 300 ms delay with the cursor inside the target, the next target appeared in the position of the previous home circle, and the reach “back” (i.e., primarily elbow flexion) could begin. Audiovisual feedback was provided when the cursor reached the target and was based on the time it took for the participant to move the 20 cm from the home circle to the target circle. There were five duration requirements that consequently enforced average reaching speed: 1,850, 1,150, 800, 500, and 250 ms corresponding to the “very slow,” “slow,” “medium,” “fast,” and “very fast” conditions, respectively (Fig. 1C). If the cursor reached the target within ±50 ms of the specified duration, the target would flash yellow and play a pleasant tone. If the cursor reached the target outside of this ±50 ms time window, the target would flash gray or green, indicating the movement was “too slow” or “too fast,” respectively (Fig. 1D). Participants learned to associate the audiovisual stimuli to the speed of their reaching movements during a 50-trial practice block before beginning the main reaching protocol. Each reaching condition also began with 20 familiarization trials followed by a 30-s break to allow participants to experience the upcoming speed.

Data Acquisition

Metabolic cost.

Gas exchange (V̇o₂ and V̇co₂) was measured breath-by-breath by the metabolic cart (TrueOne2400, ParvoMedics, Sandy, UT). Participants wore a nose clip and breathed through a 2-way, non-rebreathable value (Hans Rudolph, Kansas City, MO) while walking on a split-belt treadmill (Bertec, Columbus, OH) and reaching with the arm while grasping the handle of a shoulder-elbow robot (Interactive Motion Technologies, Watertown, MA) (Fig. 1, A and B). Participants were required to fast before metabolic testing (2–4 h before walking; 6–8 h or overnight before reaching) and performed any given walking or reaching movements for a minimum of 5 min to allow metabolic steady state to be reached. The longer fasting period before reaching was implemented because our previous work involving reaching metabolics (17, 18) found that detecting the small net costs of reaching were sensitive to the thermic effects of food (19). The metabolic cart V̇o₂ and V̇co₂ sensors were calibrated before each visit using certified gas mixtures, and the flowmeter was calibrated using flow rates from a 3-L syringe. Baseline metabolic rates were measured for a minimum of 5 min while quietly standing and quietly sitting both at the start and end of each of the walking and reaching visits, respectively.

The V̇o₂ and V̇co₂ exchange values over the final 2 min of each walking or reaching condition were averaged. Gross metabolic rates (W) were then estimated using the averaged steady-state gas exchange values in the Brockway equation (53). Net metabolic rates (W) were estimated by subtracting the lower of the two baseline metabolic rates from the gross metabolic rates. Normalized metabolic rates were calculated by dividing the gross or net rates by the participant’s body mass (W/kg) for walking, and by dividing the rates by estimated arm mass for reaching. Estimated masses of the upper arm, lower arm, and hand were summed to arrive at an estimate of mass for the entire upper limb (54).

Reaching kinematics.

Robot handle position and velocity data were acquired at 200 Hz. Position data were filtered using a fourth-order Butterworth filter with 10 Hz cutoff. Velocity, acceleration, and jerk along each axis were obtained by differentiating the filtered position signals. Movement onset was determined using a custom algorithm that used a threshold on the standard deviation of tangential velocity. Movement offset was determined as the first time point that the cursor traveled 20 cm along the anteroposterior axis.

Fatigue.

The self-reported level of fatigue was measured with the 21-item Modified Fatigue Impact Scale (MFIS) (55–57). Questions are answered based on how fatigue had impacted an aspect of day-to-day life ranging from 0 (Never) to 4 (Almost always) within the past 4 weeks. We used the total MFIS score, which incorporates the physical, cognitive, and psychosocial subscales. Higher scores indicate increased feelings of fatigue, and a cutoff score of 38 is often used to discriminate “fatigued” from “non-fatigued” individuals (56).

Sleep quality.

Aspects of sleep quality and its impact were measured with two questionnaires: the Patient-Reported Outcomes Measurement Information System (PROMIS) Sleep-Related Impairment and Sleep Disturbance scales (58, 59). The PROMIS Sleep-Related Impairment is a 16-item self-assessment measuring the impact of sleep on quality of life in the past seven days and is rated on a five-point Likert scale ranging from “Not at all” to “Very much.” Similarly, the PROMIS Sleep Disturbance is a 27-item self-assessment measuring the quality of sleep within the past seven days and is rated on the same aforementioned five-point Likert scale. Higher scores indicated worse sleep and heightened detriments owing to poor sleep. Sleep quality was measured due to its high correlation with fatigue (60).

Walking ability.

Self-assessed walking ability was measured with two questionnaires: the Patient Determined Disease Steps (PDDS) (61) and MS Walking Scale (MSWS-12) (62). The PDDS is a 1-item assessment asking how well pwMS can walk ranging from 0 (normal) to 8 (bedridden). We used a PDDS score of ≤3 (walking only mildly impaired, do not typically require a cane/walking aid) as a primary inclusion criterion. The MSWS-12 is a 12-item assessment asking pwMS about their limitations with walking in the past two weeks and is rated on a five-point Likert scale ranging from 1 (Not at all) to 5 (Extremely).

Maximum walking speed was measured with the timed 25-foot walk test (T25FW) (63). The time it took participants to walk a prescribed 25-foot distance “as quickly, but as safely, as possible” was timed with a stopwatch. The fastest of two trials was recorded.

Upper limb ability.

Upper-limb ability was measured with two physical assessments: the grooved pegboard test (GPT) (64, 65) and maximal voluntary contraction (MVC) grip (66). The GPT provided a measure of manual dexterity. Participants placed 25 pegs into the holes as quickly as possible, twice with each hand in alternating fashion. The fastest time for each hand was recorded. The MVC grip provided a measure of hand-grip strength. Participants held a dynamometer and squeezed as hard as possible, three times with each hand in alternating fashion. The largest MVC (kg) for each hand was recorded.

Statistical Analyses

Data were analyzed using R version 4.0.5 (67). All HCs were able to complete all reaching and walking speeds. All pwMS were able to complete the five reaching speeds and slowest three walking speeds; 12 of 13 pwMS were able to complete the second fastest (1.35 m/s) and 11 of 13 the fastest (1.6 m/s) walking speeds. One pwMS did not complete the second day of reach testing. These uncompleted speeds were treated as missing data points in subsequent statistical analyses.

Participant demographics, questionnaire scores, and baseline physical assessment performance are described in Table 1. Numeric outcome variables were compared with the nonparametric two-sample Kolmogorov–Smirnov test. Nominal data were compared with the χ²-test. Effect sizes were calculated with Cohen’s d. Comparisons of reach kinematics (i.e., speed, accuracy, and jerk) were made using two-sample t tests, and P values were reported with Bonferroni corrections for multiple comparisons.

Table 1.

Participant characteristics

Variable	HC (n = 13)	MS (n = 13)	P Value	Effect Size (95% CI)
Physiological assessments
Sex
F	11 (85%)	11 (85%)
M	2 (15%)	2 (15%)
Age, yr	45.9 (7.8)†	46 (7.7)	1‡	−0.02 (−0.83; 0.79)
Height, cm	166.6 (6.2)	166.1 (8.7)	0.88	0.06 (−0.75; 0.87)
Mass, kg	65.9 (7.6)	76.9 (17.9)	0.13	−0.80 (−1.64; 0.04)
MS diagnosis, yr		9.37 (8.52)
MVC grip, kg	34.2 (7.5)	32.1 (7.8)	0.57	0.27 (−0.54; 1.09)
Pegboard time, s	54.7 (7.6)	61.9 (10.0)	0.13	−0.80 (−1.64; 0.04)
T25FW, s	3.42 (0.45)	4.33 (1.13)	0.015	−1.05 (−1.91; −0.19)
Questionnaires
MFIS	9.6 (14.8)	36.7 (20.6)	0.0039	−1.51 (−2.43; −0.59)
PDDS		1.23 (1.17)
MSWS-12		28 (11.40)
Sleep disturbance	49.2 (18.2)	64.7 (24.0)	0.13	−0.73 (−1.56; 0.11)
Sleep impairment	26 (10.7)	40.2 (18.7)	0.13	−0.93 (−1.78; −0.08)

HC, Healthy controls; MFIS, Modified Fatigue Impact Scale; MS, multiple sclerosis; MSWS-12, MS Walking Scale; MVC grip, maximum voluntary contraction grip force; PDDS, Patient Determined Disease Steps; T25FW, timed 25-foot walk.

^†Values reported as means (SD); ‡P values and effect sizes from nonparametric two-sample Kolmogorov–Smirnov test and Cohen’s d, respectively. Bold P values indicate a significant difference between groups at the 0.05 level.

Walking and reaching metabolic data were analyzed using linear mixed-effects regression models (LMER) predicting metabolic power. Due to the established nonlinear relationship between movement speed and metabolic power in both walking and reaching (29, 68), power was log-transformed to best satisfy the standard regression assumptions of linearity, homoscedasticity, and normality. Fixed effects were the movement speed, a binary MS indicator (pwMS = 1), and their interaction. Random intercepts on participants were included to capture within-subject variability. Interactions were kept in the models regardless of significance, due to the models’ purpose being testing of a priori hypotheses:

$$\begin{array}{c}\log \left({\text{MP}}_{ij}\right)={\text{β}}_0+{\text{β}}_1\text{MS}+{\text{β}}_{2}V+{\text{β}}_3\left(\text{MS}\times V\right)+u_{0j}+e_{ij}\end{array}.$$ (1)

LMERs with both log-transformed power and log-transformed walking speed, as well as gamma-distributed generalized linear mixed effects (GLMER) models with log links were also tested and compared with our LMERs; however, the LMERs were deemed to provide better overall fits via residual distributions, log-likelihoods, and AIC/BIC scores, though results did not significantly change across choice of model.

Preferred walking speed is often related to the movement speed that minimizes metabolic cost per distance, or the cost of transport. To calculate the walking cost of transport as a function of speed, we began with the equation for metabolic rate (32):

$$\begin{array}{c}{\dot{e}}_{\text{w}}=a+b{v}^j\end{array}.$$ (2)

Net metabolic rates normalized to body mass (${\dot{e}}_{\text{w}}$) and walking velocities (v) were used to fit parameters a, b, and j. This metabolic rate was then divided by walking velocity to arrive at an equation for cost per distance normalized to body mass, or cost of transport (COT) of walking:

$$\begin{array}{c}{e}_{\text{w}}=\frac{{\dot{e}}_{\text{w}}}{v}\end{array}.$$ (3)

We did the same for reaching metabolic rate using a slightly different form of the equation that better captures the metabolics of discrete movements (32) that includes distinct terms for distance and time, rather than just velocity:

$$\begin{array}{c}{\dot{e}}_{\text{r}}=a+b\frac{d^i}{T^j}\end{array}.$$ (4)

Distance (d) was fixed to 0.2 m and its exponent (i) to 1. Parameters a, b, and j were fit from the empirical movement durations (T) and gross rates ${\dot{e}}_{\text{r}}$. Integrating with respect to time, we get a measure for cost of transport of reaching, or cost per movement (CPM):

$$\begin{array}{c}{e}_{\text{r}}={\dot{e}}_{\text{r}}T\end{array}.$$ (5)

Both the COT (e_w) and CPM (e_r) are U-shaped with distinct minima. Parameter estimates for these rate equations (Eqs. 2 and 4) were obtained via nonlinear least squares (nls function in R) and confidence intervals were obtained via 1,000 bootstrap replicates (nlstools in R). Parameter estimates and 95% confidence intervals for walking were: a_pwMS = 1.31 [0.61, 1.69], b_pwMS = 0.84 [0.42, 1.65], j_pwMS = 2.46 [1.41, 3.82]; and a_HC = 1.16 [0.82, 1.41], b_HC = 0.74 [0.49, 1.14], j_HC = 2.68 [1.95, 3.48]. For reaching: a_pwMS = 89.74 [76.39, 101.50], b_pwMS = 53.44 [22.91, 111.17], j_pwMS = 2.53 [1.86, 3.31]; and a_HC = 81.85 [38.14, 101.51], b_HC = 104.78 [27.77, 356.52], j_HC = 1.69 [0.77, 2.89].

Exploratory statistical analyses were performed to investigate the potential modulatory effect of other factors (i.e., demographic data, questionnaire scores, physical ability, etc.) on the relationship between metabolic power, speed, and MS. We used the caret and glmnet packages in R (67) to build elastic net regression models predicting log transformed walking or reaching metabolic power (69). Data were split 75%/25% into training and testing subsets, respectively, and hyperparameters were tuned using five repeats of 10-fold cross-validation. Variables were standardized during the fitting procedure. Root mean square error (RMSE), mean absolute error (MAE), and R² were calculated to describe the models’ performance in capturing the training data and predicting unseen testing data.

RESULTS

Walking Energetics Were Greater in PwMS

The log-transformed net power normalized to body mass (W/kg) increased with walking speed (β_Speed = 1.039; P < 2e-16) (Fig. 2B). On average, a 0.1 m/s increase in walking speed resulted in a 10.95% (95% CI: [10.26, 11.65]) increase in net normalized power. A significant main effect of group suggested that pwMS used an average of 20.54% (95% CI: [3.33, 40.60]) more metabolic power for walking at a given speed than HCs (β_MS = 0.187; P = 0.0185) (Fig. 2B). The group-by-speed interaction (β_MS:Speed = −0.084; P = 0.0709) was nonsignificant, suggesting the rates of change of metabolic power with speed were not different between groups.

Figure 2.

Walking metabolics. A: gross metabolic power of walking increased nonlinearly with speed. Persons with mild multiple sclerosis (pwMS) (green) required more gross metabolic power to walk at a given speed, on average, as compared with healthy controls (HCs) (purple). B: same as in A but displaying net metabolic power normalized to body mass. C: net cost of transport was higher for pwMS. The speed that minimized the cost of transport was no different between groups. (*Effect of group is significant at the 0.05 level).

Net metabolic power normalized to body mass is a standard metric for walking. Nonetheless, we also analyzed the gross costs. Log-transformed gross power (W) also increased linearly with speed (β_Speed = 0.656; P < 2e-16). The effect of group on gross metabolic power was nonsignificant (β_MS = 0.131; P = 0.106) (Fig. 2A).

Walking net COT was higher overall for pwMS (P = 0.014), but the metabolically optimal speeds between HCs (0.97 m/s; 95% CI [0.848, 1.086 m/s]) and pwMS (1.02 m/s; 95% CI [0.853, 1.193 m/s]) did not differ (Fig. 2C).

Reaching Energetics Are Similar in HC and PwMS

The log-transformed gross power (W) increased with reaching speed (β_Speed = 1.798; P < 2e-16). On average, a 0.1 m/s increase in hand speed resulted in a 19.70% (95% CI [17.60, 21.83]) increase in gross power (Fig. 3A). There was no main effect of group (β_MS = −0.053; P = 0.492), which suggests that pwMS and HCs had similar average costs of reaching at a given speed (Fig. 3A). Like walking, the group-by-speed interaction was nonsignificant (β_MS:Speed = 0.207; P = 0.147).

Figure 3.

Reaching metabolics. A: gross metabolic power of reaching increased nonlinearly with speed. Persons with mild multiple sclerosis (pwMS) (green) and healthy controls (HCs) (purple) required similar average gross power to reach at a given speed. Crosses indicate the average power (±SE) at the average reaching speed (±SE) within a condition. B: same as in A but displaying net metabolic power normalized to estimated arm mass. C: gross reaching cost per 20 cm movement was similar between groups. The speed that minimized the cost per movement was likewise no different.

The metabolic costs of reaching are reported as gross metabolic power, because, in contrast to walking, the center of mass does not need to be transported during reaching tasks and net costs are modest. Rather, we opted to analyze the net metabolic power of reaching normalized to mass of the arm, which is transported during a reach, using reported estimates based on body mass (54). Log-transformed net normalized metabolic power (W/kg) of reaching increased linearly with reaching speed (β_Speed = 4.95; P < 2e-16). The effect of group remained unchanged (β_MS = −0.029; P = 0.911), suggesting that pwMS and HCs had similar net costs of reaching at a given speed (Fig. 3B).

Reaching CPM was not different between groups (P = 0.477), and the metabolically optimal speed was nearly identical between HCs (0.411 m/s; 95% CI [0.386, 0.436 m/s]) and pwMS (0.395 m/s; 95% CI [0.366, 0.424 m/s]) (Fig. 3C).

We compared reaching kinematics between pwMS and HCs across the five speed conditions (“very slow” to “very fast”) (Fig. 4, A and B). There were no significant differences in average reaching duration (Fig. 4C) or speed (Fig. 4D) between pwMS and HCs in any of the five conditions (P > 0.05). We also compared reach performance and kinematics in each speed condition. There were no significant differences in average sum of squared jerk between pwMS and HCs for any reach speed (P > 0.05) (Fig. 4F). Reach accuracy was calculated as the Euclidean distance from the cursor to the center of the target at the first moment velocity went below 2.5 cm/s. Reach accuracy did not significantly differ between pwMS and HCs in any condition (Fig. 4E). Furthermore, there were no differences between groups on average overshoot or undershoot error (P > 0.05), nor in the proportion of trials that were overshot or undershot (P > 0.05), per condition. Together, the similar average speeds, jerks, and accuracies suggested that any findings in reaching metabolics were likely not credited to kinematic differences.

Figure 4.

Reaching kinematics. Average reaching speed profiles in the “very slow” (VS), “slow” (S), “medium” (M), “fast” (F), and “very fast” (VF) conditions in healthy controls (HCs) (A) and persons with mild MS (pwMS) aligned to peak speed (B). Average movement durations (C) and reaching speeds did not significantly differ between HCs (purple) and pwMS (green) (D). E: average endpoint error in each condition did not differ between HCs (purple) and pwMS (green). The endpoint of a reach was defined as the last moment reach speed exceed 2.5 cm/s. F: average sum of squared jerk did not significantly differ between HCs (purple) and pwMS (green). Shading (A and B), vertical (C–F), and horizontal (E and F) error bars represent means ± SE.

Baseline Metabolic Rates Are Similar between HC and PwMS

Previous work has suggested that the baseline V̇o₂ or metabolic rates do not differ between pwMS and HCs (34, 36, 41). We found that the standing (MS: 79.72 ± 13.36 W; HC: 84.39 ± 10.67 W) and sitting (MS: 68.73 ± 9.98 W; HC: 69.93 ± 10.32 W) baseline rates were not different between groups (P > 0.05). However, the baseline rate while standing was significantly higher than that of sitting for both HCs and pwMS (P < 0.05). Note that the average baselines in the sitting and standing positions were quantitatively less for pwMS than HCs—this slight discrepancy possibly helps explain why the net normalized metabolic power of walking was significantly higher for pwMS (β_MS = 0.187; P = 0.0185), but gross metabolic power was not (β_MS = 0.131; P = 0.106). This further implicates reduced exercise tolerance or fitness contributing to elevated net walking energetics in pwMS.

Participant Characteristics

Comparisons between the 13 pwMS and 13 HCs are enumerated in Table 1. Briefly, neither age, height, nor weight differed between groups (P > 0.05). There were no differences in PROMIS Sleep-Related Impairment or Sleep Disturbance scores between pwMS and HCs. Neither grooved pegboard times nor MVC grip of the dominant hand differed between groups (P > 0.05). Self-reported fatigue measured via the MFIS was higher for pwMS (MS: 36.69 ± 20.62; HC: 9.62 ± 14.79; P = 0.0039). Seven of thirteen pwMS and one of thirteen HCs scored higher than the standard cutoff score of 38, indicating a clinically relevant level of fatigue (56, 70).

Time since initial disease diagnosis for the pwMS-group was 9.37 ± 8.52 yr. PwMS scored 1.23 ± 1.17 on the PDDS and 28 ± 11.40 on the MSWS-12, indicating overall low levels of mobility impairment. Nonetheless, T25FW time was still slower for pwMS (MS: 4.33 ± 1.13 s; HC: 3.42 ± 0.45 s; P = 0.0022).

Exploratory Analyses

To account for the potential modulatory influence of participant characteristics and covariates on metabolic costs, we used elastic net regression to construct a model using a subset of predictors selected from a full set of potential predictors.

For the walking model, all thirteen predictor variables remained in the model; however, many predictors were near zero, so we report the top four variables here (Fig. 5A). Log-transformed net power of walking normalized to body mass was predicted to increase with walking velocity (β_Speed = 0.9700), increase with T25FW time (β_T25FW = 0.0.0497), and increase with PDDS score (β_PDDS = 0.0129). Critically there was a significant main effect of group (β_MS = 0.0696), again indicating that pwMS required more power, on average, than HCs to walk. A fortuitous outcome was that the other top predictors (i.e., walking speed, T25FW, and PDDS) are all metrics related directly to walking.

Figure 5.

Elastic net model predicting the net metabolic power of walking normalized to body mass. A: scaled variable importance for the set of potential predictor variables. All predictors remained in the model after training. B: observed vs. predicted metabolic power. Open circles are the training data (R² = 0.876 ± 0.066); filled triangles are the unseen test data (R² = 0.927). The dashed line is the line of unity. MFIS, Modified Fatigue Impact Scale; MS, multiple sclerosis; MSWS-12, multiple sclerosis Walking Scale; MVC, maximal voluntary contraction; PDDS, Patient Determined Disease Steps; T25FW, Timed 25-foot Walk Test.

The model explained the variability in the training data very well (R² = 0.876 ± 0.0656; RMSE = 0.148 ± 0.0372; MAE = 0.121 ± 0.0313) and performed accurately predicting the unseen test data (R² = 0.927; RMSE = 0.102; MAE = 0.081) (Fig. 5).

Only six predictor variables were selected out of sixteen for the reaching analysis (Fig. 6A). Log gross metabolic power (W) was predicted to increase with reaching velocity (β_Speed = 1.721), arm mass (β_Mass = 0.127), and T25FW time (β_T25FW = 0.021). The remaining three predictors were near zero, but remained in the model: log gross metabolic power was also predicted to decrease with lower quality sleep (β_Sleep = −0.002), increase with higher resting rate (β_Resting = 0.0004), and decrease with more fatigue (β_MFIS = −1.07e-5). Noticeably, any categorical main effect of MS was discarded, reinforcing the primary finding that pwMS do not exhibit heightened costs of reaching as compared with HCs.

Figure 6.

Elastic net model predicting the gross metabolic power of reaching. A: scaled variable importance for the set of potential predictor variables. Six predictors remained in the model after training (orange) and ten predictors were excluded (gray). B: observed vs. predicted metabolic power. Open circles are the training data (R² = 0.800 ± 0.111); filled triangles are the unseen test data (R² = 0.735). The dashed line is the line of unity. MFIS, Modified Fatigue Impact Scale; MS, multiple sclerosis; MSWS-12, multiple sclerosis Walking Scale; MVC, maximal voluntary contraction; PDDS, Patient Determined Disease Steps; T25FW, Timed 25-foot Walk Test.

This model likewise explained the variability in the training data well (R² = 0.800 ± 0.1110; RMSE = 0.174 ± 0.0414; MAE = 0.142 ± 0.0367) and performed accurately predicting the unseen test data (R² = 0.735; RMSE = 0.155; MAE = 0.126) (Fig. 6).

DISCUSSION

Does increased metabolic cost contribute to movement slowing in multiple sclerosis? To answer this question, we measured the metabolic rates of both walking and seated arm reaching movements in persons with mild multiple sclerosis (pwMS) compared with healthy age- and sex-matched controls (HCs). We found that while the metabolic power of walking was higher at any speed for pwMS, the metabolic demands of seated arm reaching movements were no different than HCs. These results suggest that complementary mechanisms beyond metabolic cost, such as accuracy, perceived effort, or reward valuation, are slowing reaching movements for pwMS.

Walking Costs Were Elevated for Persons with MS

The metabolic cost of walking is contingent on such variables as speed, incline, mass, and age (25, 29, 71, 72). We found that the metabolic rate of walking increases with speed similarly in both HCs and pwMS, but that there is an MS-dependent inflation in power of around 20%.

Our findings diverged slightly from previous studies investigating walking costs in MS. For one, we only found an approximate 20% increase in walking power for pwMS compared with HCs, which is not as dramatic as previously reported two-to-threefold increases in walking costs (7, 24, 38). This disparity is likely due in-part to the novelty of our data set: it consists of a relatively homogeneous cohort of pwMS who did not require a cane or walking aid, and a majority of whom did not rely on any contact with the handrails of the treadmill. Larger differences in cost were found in studies with 30% (34) to 50% (7) to 100% (38) of their MS groups requiring walking aids.

Furthermore, most studies investigating walking in MS use the oxygen cost of walking as their proxy for energy expenditure (5, 7, 9, 24, 34, 36, 38). The oxygen cost of walking is the amount of oxygen required to move one kilogram of mass one meter. Instead, we used the Brockway equation (53) to convert oxygen and carbon dioxide exchange rates into metabolic power and then calculated costs per distance, which may have contributed to our smaller differences.

Work by Chung et al. (34) found that the oxygen cost of walking for pwMS was ∼78.5% higher at a slow walking speed of 0.6 m/s, but not at a preferred (around 1.0 m/s) or fast walking speed of 1.4 m/s. Although we did not find a convergence of metabolic power at faster speeds, we did see similar predicted walking and reaching speeds that minimize net COT between our groups (34, 73, 74). Walking slowness exhibited by pwMS suggests other factors besides minimizing cost of transport are prioritized when determining preferred walking speed, such as maintaining stability or counteracting spasticity (22, 39, 40). PwMS experience proprioceptive-based deficits and reduced dynamic stability while walking (40, 75), which correlate negatively with walking speed (9) and encourage pwMS to walk more quickly to exploit the passive dynamics of walking (76).

Comparable increases in walking effort are observed in healthy older adults. Energetic costs of walking for older adults are ∼23% higher than young adults (72). Young and older adults walking at 1.3 m/s were found to utilize 2.57 ± 0.09 W/kg and 2.96 ± 0.11 W/kg, respectively (77). Our cohort of pwMS, though much younger in age, required ∼20% more metabolic power on average, and 3.08 ± 0.18 W/kg when walking at a similar 1.35 m/s. MS, although not an age-related disease, heightens locomotor energy demand in a manner mirroring healthy aging.

Reaching Costs Were Not Elevated for Persons with MS

Similar to walking, reaching metabolic costs vary with factors such as speed, mass, distance, and age (32, 68). We appear to be the first to measure the metabolic costs of arm reaching in pwMS and demonstrate that across a range of feasible reach speeds, the costs were not related to MS disease state.

The closest analog to reaching metabolics performed previously is arm crank ergometry (43), although this was done in a maximal exercise performance context. When performing arm crank ergometry, pwMS had reduced peak aerobic capacity, weakened respiratory muscles, and impaired cardiopulmonary responses. In arm reaching, interestingly, the gross power of reaching at the fastest speed was no different than walking at the slowest in both groups, implicating the physical demand of reaching very quickly is akin to walking slowly. The body’s sympathetic cardiopulmonary and metabolic response to exercise may have come into play at these fastest reach speeds which, with peak velocities exceeding 1.0 m/s, were approaching the limit of achievable hand speeds (51, 52). While reaching faster than what we required in the fastest condition would be challenging, pwMS may have greater costs if moving even more quickly; however, for standard day-to-day reaching requirements, the energetic demands during arm reaching do not differ between pwMS and HCs.

Nonetheless, pwMS still tend to reach more slowly. A recent study had pwMS and controls perform horizontal planar reaches and found that pwMS reached with slower, jerkier, and more inaccurate movements than controls, while also adopting altered muscle activation patterns and synergies (12). Pellegrino et al. (12) recruited a cohort of pwMS with a higher disability status and required participants to reach to one of eight radial targets as accurately as possible. We may have not found differences in reach kinematics due to the lower disability of our cohort and the relative simplicity of our task: reaching out-and-back between two targets with end point accuracy, though enforced, not as critical to the protocol as was speed-matching accuracy. It remains possible then that pwMS slow upper limb movements to maintain task accuracy in the face of compromised sensorimotor upper limb control (47, 49, 50, 78).

Pathophysiological Changes in Metabolism Caused by MS

Why are walking but not reaching costs elevated in MS? One explanation is that the costs of walking for pwMS are heightened due to worsened physiological fitness. Deconditioning occurs as a secondary consequence of the symptoms of MS, such as fatigue, which disincentives exertion and leads to a more sedentary lifestyle (9, 43, 45). Although walking may induce a sympathetic exercise response to reveal differences in fitness between pwMS and HCs, reaching movements may not. Changes in muscle sympathetic nerve activity may only become significantly higher from rest at workloads exceeding some percent V̇o₂ threshold (51, 52). Here, gross reaching power only approached ∼250 W at the maximum, whereas walking costs were larger. Our suggestion here is that this sympathetic response to exercise (i.e., increased heart rate, increased ventilation, etc.) was not exceeded for most of the reaching movements performed, so this fitness-related inflation of metabolic power was not prevalent.

Exercise-based rehabilitation has been shown to both reduce the energetic cost of walking and increase the preferred speeds in pwMS (44), and more active pwMS have lower costs than those less active (35). However, whether the benefits of exercise would improve reaching ability or restore locomotory performance to levels comparable with control individuals is uncertain. Regardless of the exact mechanism, the slower gait speeds in pwMS could be explained in-part by the increase in metabolic effort as speed is reduced to minimize expenditure (24, 73).

A second explanation is that there are distinct metabolic shifts caused directly by MS pathology. There can be irreversible neurological damage that occurs in MS that causes permanent mobility disability (79), changes to resting muscle oxygen consumption (80), or compensatory axonal mechanisms that alter central nervous system (CNS) energetic demand (81). Thus, despite adequate rehabilitation, pwMS may neither regain full mobility nor restore typical energetic costs of movement due to axonal injury, loss of motor units, decayed muscle strength, and other such adaptations (19, 20).

We measured the metabolic costs of walking and arm reaching with the justification that seated arm reaching movements are far less costly and may remove a large proportion of the fitness-based changes in energetics. The lack of difference in reaching costs suggests that these MS-related pathophysiological shifts in CNS metabolism were, at the very least, not detectable, if occurring at all, and that worsened attributes of fitness may be a principal basis of expensive walking.

Implication of Reward-Related CNS Regions Contributing to Movement Slowing

If the goal of movement is to maximize a reward-effort tradeoff over time (i.e., a net reward rate) (33), then upper limb slowness (10–14) in MS is not accounted for through metabolic effort. One possibility is that instead, individuals are appraising effort on a relative scale. Similar to the fitness hypothesis for walking differences, pwMS have a lower aerobic capacity (41, 43, 45), thus equivalent gross metabolic costs of reaching in pwMS and HCs would equate to a larger proportion of maximal capability for pwMS. Perhaps, then, pwMS are reaching and walking more slowly as a behavioral response to diminished physiological potential.

Alternatively, there is reward, which increases movement speed so that we may maximize the rate of reward accrual (82–85). Mechanistically, dopamine tends to encode reward or its expectation (86–88) in regions of the basal ganglia. Dopamine is released moments before movement onset increases the excitability of primary motor cortex (89, 90), facilitating more vigorous movements.

Recent evidence suggests that gray matter demyelination and impaired dopamine mediation within the basal ganglia may emerge as the disease progresses (2, 91–95). PwMS present smaller cortical gray matter volume, and deep gray matter atrophy—consisting of thalamus, putamen, globus pallidus, caudate, and amygdala—tracks with disease progression (93, 94). Accordingly, pwMS improperly value reward and make nonoptimal reward-based decisions (96–100). Diminished reward responsiveness in pwMS may hinder the assessment of expected reward rate and slow movement because of an undervalued reward-effort tradeoff.

Lesions to the basal ganglia caused by MS may additionally impair movement through heightened effort sensitivity. Dopamine is similarly linked to invigorating the expenditure of effort to obtain said rewards (101, 102). For example, rats with dorsal striatum lesions move toward rewarding stimuli more slowly while maintaining similar reward-seeking behavior (102). As a result, gray matter lesions in MS may result in slow movements due to a paucity of reward-mediated motivation or increased sensitivity to effort; however, this link between reward, effort, dopamine, and movement invigoration in MS remains an open question.

Limitations

Our approach is not without its limitations. One shortcoming is that we did not collect self-reported measures of perceived effort or state fatigue (55) between blocks of walking or reaching. Using a Visual Analog Scale (VAS) or Borg RPE would have provided us with a measure of how subjective task difficulty changed over time and how it compared between pwMS and HCs. Higher perceived effort could be inflating the total effort of movement for pwMS, thus further contributing to movement slowness. In healthy individuals, for example, measures of perceived fatigability tend to correlate with the perceived effort of reaching movements (103), whereas pwMS tend to subjectively rate exercise tasks as more effortful (38, 104). Despite no differences in physical reaching effort between pwMS and HCs, the possibility exists that reaching is perceived as more effortful in pwMS and contributes to movement slowness.

Although not the primary focus of the study, a second limitation is that we did not measure preferred walking or reaching speeds. Both preferred walking and reaching speeds are often selected to minimize the cost of transport (J/kg/m) of the movement (29, 32, 68). Recent evidence suggests that pwMS minimize the walking cost of transport when selecting preferred speeds (44, 73), but that the costs tend to be more expensive than controls. Measuring preferred walking and reaching speeds in this study would have also revealed whether our cohort of pwMS was selecting slower speeds than our HCs. That being said, obtaining preferred speeds, especially in reaching, is not a trivial task (105–107).

Finally, our cohort of pwMS was a convenience sample of predominantly women (12 of 13) with RRMS, which is not necessarily representative of the overall sex distribution of the disease, which sees females 2–3 times more likely to develop MS (108). The inclusion criteria and advertising description of this fairly strenuous study likely introduced selection bias in attracting more physically active pwMS (35). Recruiting a more physiologically diverse group of participants may have lent itself to even greater differences in metabolic costs of movement. The strict inclusion criteria for this study were largely established to ensure safe walking ability on the treadmill; future work could flexibly expand inclusion criteria especially if the seated reaching movements are of primary interest. Furthermore, while an a priori power calculation suggested that a total sample size of ∼20 total was sufficient to achieve 80% power, the relatively small sample size (n = 26) may have still left us underpowered, especially for detecting interactions which had small, estimated effect sizes. Again, expanding inclusion criteria and thereby increasing sample size will lend itself to more sensitive analyses regarding reaching energetics in MS.

Conclusions

Individuals with mild MS expended ∼20% more metabolic energy when walking compared with persons without MS, but there was no difference between the two groups in metabolic energy consumption during seated arm reaching. These increased costs of walking were found even in a highly mobile cohort of individuals with MS who did not require walking aids. Moreover, the lack of differences found in reaching costs was not attributed to disease-related altered kinematics: speeds, accuracies, and jerk were comparable between the MS and control groups. Our results suggest that movement slowness occurring with MS is not altogether a consequence of energy conservation. Instead, other MS-related mechanisms, such as fitness, magnified perceived effort, or diminished reward sensitivity, may additionally compel slower movements and should be considered when treating mobility symptoms.

DATA AVAILABILITY

Data will be made available upon reasonable request.

GRANTS

This work was supported by National Institutes of Health under Grant 1R01NS096083 and the National Science Foundation CAREER Award 1352632 (to A. Ahmed) and a University of Colorado Boulder Summer Graduate School Fellowship (to R. Courter).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

R.C., E.A., R.E., and A.A. conceived and designed research; R.C. performed experiments; R.C. analyzed data; R.C., R.E., and A.A. interpreted results of experiments; R.C. and A.A. prepared figures; R.C. drafted manuscript; R.C., R.E., and A.A. edited and revised manuscript; R.C., R.E., and A.A. approved final version of manuscript.