Humans trade off whole-body energy cost to avoid overburdening muscles while walking

Abstract

Metabolic cost minimization is thought to underscore the neural control of locomotion. Yet, avoiding high muscle activation, a cause of fatigue, often outperforms energy minimization in computational predictions of human gait. Discerning the relative importance of these criteria in human walking has proved elusive, in part, because they have not been empirically decoupled. Here, we explicitly decouple whole-body metabolic cost and 'fatigue-like' muscle activation costs (estimated from electromyography) by pitting them against one another using two distinct gait tasks. When experiencing these competing costs, participants (n = 10) chose the task that avoided overburdening muscles (fatigue avoidance) at the expense of higher metabolic power (p < 0.05). Muscle volume-normalized activation more closely models energy use and was also minimized by the participants' decision (p < 0.05), demonstrating that muscle activation was, at best, an inaccurate signal for metabolic energy. Energy minimization was only observed when there was no adverse effect on muscle activation costs. By decoupling whole-body metabolic and muscle activation costs, we provide among the first empirical evidence of humans embracing non-energetic optimality in favour of a clearly defined neuromuscular objective. This finding indicates that local muscle fatigue and effort may well be key factors dictating human walking behaviour and its evolution.

Keywords: cost of transport; energetics; fatigue; locomotion; muscle effort; optimality.

Figure 1.

Experimental design. Participants were asked to select between crouch walking, which moderately affects metabolic cost and penalizes activation cost, and a series of incline levels that incrementally increase both metabolic (M) and activation (A) costs. A competing-cost pair was established when participants walked on an incline level that incurred a higher metabolic cost but provided an activation cost advantage relative to the crouch condition. A virtual ceiling and the participant's silhouette, displayed on a screen anterior to the treadmill, provided feedback regarding crouch walking task performance. (Online version in colour.)

Figure 2.

Walking data for the (a–d) mean (±s.d.) crouch trial and five levels of incline (from 0 to 24%) and (e–h) mean (±s.d.) crouch trial and pre- and post-transition inclines. (a–e) Metabolic power (C_met,P; W kg⁻¹; n = 10), (b,f) mean of squared activations (${\overline{C}}_{a^2}$; unitless; n = 8), (c,g) maximal muscle activation (${\overline{C}}_{a,\max }$; unitless; n = 8), and (d,h) volume-weighted muscle activations (${\overline{C}}_{a,\mathrm{vol}}$; unitless; n = 8). The number of participants that selected incline (i) or crouch (c) in each condition are indicated in (a) (grey text). *Significantly different (p < 0.017) from crouch (e–h only); ^Significantly different (p < 0.017) from pre-transition (e–h only). (Online version in colour.)

Figure 3.

Average (n = 8) average normalized activation costs during crouch walking (level treadmill) and incline walking on treadmill slopes of 0–24%. Seven lower limb muscles were analysed; gluteus maximus (Gmax), biceps femoris (BF), rectus femoris (RF), vastus medialis (VM), medial gastrocnemius (MG), soleus (SOL) and tibialis anterior (TA). The unitless quantity A_ij was established by integrating the linear envelope and expressing this as a rate (a_ij), before normalizing data for each muscle to the average value obtained from the 0% incline trial. These normalized data were then averaged over five strides. (Online version in colour.)

Figure 4.

Competing-cost pairs were established for seven of the eight participants for which this was feasible. All seven participants selected the incline condition that favoured the mean of squared activations (${\overline{C}}_{a^2}$) and penalized metabolic power (C_met,P). In doing this, they rejected the crouch walking condition that favoured metabolic power and penalized the mean of squared activations. Data are presented as the mean ± s.d. *Significantly different (p < 0.05) from crouch C_met,P; ^Significantly different (p < 0.05) from crouch ${\overline{C}}_{a^2}$. (Online version in colour.)