The muscle's in vivo force-length and force-velocity relationships dictate the magnitude of activation required to achieve a given force and velocity of shortening (Praagman et al., 2006) at a given joint configuration (muscle length). The force-velocity relationship dictates that force production for a given level of activation is maximal when that force can be developed isometrically (Fenn and Marsh, 1935; Roberts et al., 1997; Biewener, 1998a) and decreases as shortening velocity increases. Stainsby and Lambert (1979) suggest that the major determinant of metabolic cost of muscle contraction in voluntary movement should be motor unit recruitment. This notion is consistent with the observation that there is a unique bicycling cadence associated with the best efficiency (Coast and Welch, 1985) and this cadence corresponds with the cadence with the lowest magnitude of electromyographic (EMG) activity (MacIntosh et al., 2000). Load, shortening, and velocity of shortening have less impact on the magnitude of energy requirement (Stainsby and Lambert, 1979). For submaximal contractions like those exerted during distance running, the level of activation needed to generate a given force can be minimized when the fascicles are allowed to develop force isometrically.

Keeping the underlying factors dictating energy use in muscle in mind (force, length, velocity, shortening, and activation), we now turn our attention to the energy cost of running. The factors that are not affected by training will be considered first. This will be followed by an examination of those factors which can be affected by training.

The energy required to overcome air-resistance is a function of the runner's frontal surface area and drag coefficient, the air density (altitude, humidity, and pressure) and the relative speed of movement of air past the runner. Pugh (1971) found that work required to overcome air resistance was a linear function of the sum of running speed plus wind velocity, squared. As such, when running at speeds approaching the 2-h marathon barrier (5.8 m·s⁻¹) under still wind conditions, the extra energy required to overcome air resistance is ~8% higher compared to running with no air resistance (Pugh, 1970). Presumably the extra energy is required due to the need to generate greater horizontal propelling force. This would relate to the need for increased motor unit recruitment in muscles contributing to the forward propulsion. This extra energy can be nearly abolished by drafting behind other runners, which saves 80% of the extra energy required to overcome wind resistance (Pugh, 1971).

High environmental temperatures lower a runner's ability to dissipate heat. Increasing heat storage impacts cross-bridge turnover, potentially increasing the energy cost of muscle contraction, and also results in sweating and potentially dehydration. Heat exchange between the body and the environment is achieved by conduction, convection, radiation, and heat loss is also achieved by evaporation (Cheuvront and Haymes, 2001). Any imbalance between heat generated by metabolism and net heat lost by these heat dissipating mechanisms will result in change in heat storage (Bergeron, 2014). Wind serves a thermoregulatory function in that cooler air crosses the skin during running, allowing greater heat loss by convection. This accelerated heat loss will result in lower heat storage for a given speed of running, but heat loss by convection would be considerably less while drafting, so less heat generated through a lower E_run will likely be the key feature of a record-attempt at the sub 2-h marathon. A high ratio of surface area to mass will also favor heat loss by conduction, convection, radiation, and evaporation. A lower E_run for a given environmental temperature and humidity will result in less heat storage, and a longer exercise duration is permitted. Similarly, a runner with a comparatively low E_run can run at a faster speed, for the same rate of heat storage. Where less heat is generated, less energy is consumed by the heart for peripheral circulation of warmed blood from the core to the skin for cooling purposes (Rowell et al., 1969; MacDougall et al., 1974).

Running straight ahead at a constant speed on a dry, smooth, flat surface requires friction between shoe (or foot) and surface (Frederick, 1986). When on a slippery or wet surface, where the coefficient of friction is reduced, subjects tend to modify their kinematics (and therefore use a less-than-optimal movement pattern) to compensate for soft, energy-dissipating, and uneven surfaces (Frederick, 1983). For example, a reduced coefficient of friction probably involves greater muscle activation prior to footstrike in order to stabilize posture in uncertain circumstances. Presumably, this also elevates E_run, although further research is required to determine the magnitude of this increase in muscle activation and corresponding increase in energy cost as a result of the less than optimal kinematics caused by lower surface friction.

E_run is also elevated on soft and uneven surfaces such as sand compared to grass or concrete; (Zamparo et al., 1992; Lejeune et al., 1998; Pinnington and Dawson, 2001). The elevated E_run on sand has been attributed to a reduction in the re-utilization of elastic energy and/or the energy lost due to backwards translation of the foot during push-off. It has also been hypothesized that an elevated muscle-tendon work while running on sand contributes to the elevated E_run (Lejeune et al., 1998). In terms of the muscle energetics presented earlier, these mechanisms (foot slip, increased work and decreased tendon strain energy release) translate to an increased muscle shortening and probably increased motor unit recruitment. It would certainly require increased muscle coactivation for stability when running on an unstable surface. Both of these factors would increase the energy cost of muscle contraction, contributing to increased E_run.

The additional mass of footwear predictably increases E_run by ~1% per 100 grams of added mass per shoe (Frederick et al., 1984). This 1% increase in E_run is fairly consistent across a range of running speeds (Franz et al., 2012) and also degrades running performance (e.g., 3,000 m time-trial time) to a similar extent (Hoogkamer et al., 2016). It has been suggested that a potential mechanism by which footwear might reduce E_run is because footwear serves to reduce some of the impact shock. A reduction in E_run of 3% with well-cushioned shoes compared to poorly-cushioned ones supports this notion (Frederick, 1986). These authors developed a “cost of cushioning” hypothesis whereby a portion of the measured E_run in well-cushioned shoes is reduced because less muscle activation is required to brace for the force of impact with the ground. To support this hypothesis, E_run was compared between well-cushioned shod and unshod conditions. The former condition would incur an estimated increase in E_run as a result of the mass of the shoes. Despite the added mass of the shoes, E_run was not different between shod and unshod conditions.

Stearne et al. (2016) proposed that a portion of the metabolic energy required during running could be saved by the arch spring. These authors demonstrated that restricting foot arch compression increased E_run by 6%. This elevated metabolic cost was not seen during walking or incline running, due apparently to the smaller role of elastic energy savings in these gaits. Their results further support the notion that orthotic insoles and arch-support footwear, which are often prescribed to runners, may reduce the foot arch's elastic energy storage and result in an elevated E_run while wearing these types of orthotics or shoes (Berg and Sady, 1985).

It has also been speculated that the design of footwear midsole construction may enhance energy return, and therefore E_run. Worobets et al. (2013) have shown a small (0.9–1.1%) but significant difference in V˙O2 between an energy-return midsole (Adidas Boost™) and a conventional ethyl vinyl acetate midsole in runners running below the anaerobic threshold. To eliminate any confounding factors such as shoe construction, Tung et al. (2014) isolated the effect of cushioning on E_run by attaching the same cushioning foam to the belt of a treadmill. In so doing, E_run was reduced by 1.6% when runners ran unshod on the cushioned belt in comparison to running unshod without the cushioning. Interestingly, E_run was not different between shod and unshod conditions on a normal treadmill belt, likely because the beneficial effects of cushioning were balanced by the detrimental effects of added shoe mass. These results suggest (1) shoe mass can have a meaningful influence on the measured E_run and (2) there exists a trade-off between running in very light running shoes at the expense of extra cushioning in order to minimize E_run. Runners are also able to assess shoe comfort reliably (Hennig et al., 1996) and it has been hypothesized that comfort could relate to performance (Nigg, 2001). In fact, oxygen cost was 0.7% lower in shoes deemed “most comfortable” compared to those deemed “least comfortable” (Luo et al., 2009). Further, studies into the specific mechanism for a lower E_run with particular reference to muscle energetics associated with some footwear design (kinematics, kinetics, muscle activity etc.) are needed.

E_run is determined primarily by the energy needed for muscle contraction of sufficient average force to support body weight for the full stride duration (Kram and Taylor, 1990). Therefore, average muscle force and thus muscle energy cost is related to the average vertical force (F_z) during stance, as dictated by body mass and the F_z moment arm and the moment arm of the Achilles tendon (Ker et al., 1987; Carrier et al., 1994). The ratio of F_z moment arm to that of the Achilles tendon is referred to as the gear ratio. Often, the F_z moment arm length is interpolated from known forefoot length. In this case, the ratio of forefoot length to AT moment arm is referred to as the foot lever ratio (Kunimasa et al., 2014). The F_z moment arm can be altered by changing ankle joint kinematics during the stance phase. This has important implications to E_run since changes in joint angle configuration at touch-down result in changes in the F_z moment arm. The relative change in the gear ratio for a given F_z will determine the magnitude of the required muscle force. Reductions in the gear ratio result in a reduction in muscle forces and this should be reflected as lower energy cost.

During running, the ankle angle at touchdown is nearly 90°, and the excursion during stance in good runners is nearly 10° more than in elite runners (Cavanagh et al., 1977; Williams and Cavanagh, 1987). A small excursion in the same amount of time translates to lower angular velocity and a corresponding slower velocity of muscle contraction of the ankle plantarflexors. A slower velocity of contraction results in lower level of activation needed to generate a given force and consequently lower energy cost of muscle contraction for these muscles. Using measurements of hemoglobin desaturation during blood flow occluded plantarflexion exercise (which was assumed to be proportional to energy use), Fletcher et al. (2013a) showed that a faster muscle fascicle shortening and/or velocity of fascicle shortening elevates energy use. The relationship between muscle energy cost, shortening, and shortening velocity is shown in Figure 3.

A shorter AT moment arm, measured at rest, has been associated with a lower oxygen cost of running (Scholz et al., 2008; Mooses et al., 2014). The advantage of a short AT moment arm in reducing E_run has been attributed to increases in the elastic energy storage/release from the AT during running since larger AT forces for a given joint moment are required with a short AT moment arm; more elastic strain energy is stored and released in a tendon when AT forces are higher. To claim the benefit of a short moment arm is dependent on the estimated extra elastic energy storage ignores the fact that additional muscle energy cost would be required to generate the additional force production of the muscles. This extra muscle energy cost has been estimated to be considerably higher than the extra energy stored in and subsequently released from the tendon (Fletcher and MacIntosh, 2015).

Shorter moment arms also permit slower muscle shortening velocity to achieve a given joint angular velocity (Nagano and Komura, 2003) and require less shortening for a given joint excursion. This slower velocity would permit a higher force without much increase in activation. As previously suggested however, the elevated AT force associated with a shorter moment arm may also incur a substantial muscle energy cost (Fletcher and MacIntosh, 2015) and as such, a longer AT moment arm may help reduce E_run by reducing the required muscle force and level of muscle activation to sustain a given joint moment. To support this hypothesis, elite Kenyan long-distance runners, a population known for their exceptionally-low E_run (Larsen, 2003; Wilber and Pitsiladis, 2012), have longer AT moment arm lengths and shorter forefoot lengths compared to similarly-trained Japanese distance runners (Kunimasa et al., 2014). van Werkhoven and Piazza (2017) recently found a significant relationship between peak Achilles tendon force and heel length, suggesting subjects with shorter heels experience larger Achilles tendon forces. These larger forces, however, were not associated with a reduced oxygen uptake measured at a common speed for all subjects. These authors, and others contend that while larger Achilles tendon forces should increase tendon stretch and strain energy storage, larger forces also require higher muscle activation and elevated metabolic cost (Perl et al., 2012; van Werkhoven and Piazza, 2017). It remains unclear why these authors did not find the same negative correlation between oxygen uptake and heel length, as has been reported previously (Scholz et al., 2008; Raichlen et al., 2011; Mooses et al., 2014).

The ratio of forefoot to AT moment arm length is known as the foot-lever or gear ratio; a low gear ratio is associated with better E_run (Mooses et al., 2014). The gear ratio determines the force and length change needed in the gastrocnemius muscle and these factors will affect the energy cost of muscle contraction. Reducing the gear ratio from 2 to 1.5, by reducing forefoot length and keeping the AT moment arm constant reduces the estimated triceps surae muscle energy cost by nearly 40% (Fletcher and MacIntosh, 2015), assuming the same amount of shortening. Less force is needed if the F_z moment arm is shorter. Similarly, reducing the gear ratio by increasing the AT moment arm will also reduce the required muscle force but also necessitates a greater amount of shortening for a given angular displacement so the energy savings would not be this large.

The length of the F_z moment arm is dictated by the footstrike pattern and the length of the forefoot. Forefoot length is another anatomical feature (along with presumably short or long moment arm lengths) for which humans have evolved, presumably to favor economical walking and running compared to other primate species. In relation to body mass, humans possess extremely short forefoot lengths (Rolian et al., 2009). This evolutionary adaptation has long been assumed to benefit bipedal locomotion since short toes require smaller plantarflexor forces to balance the large dorsiflexion moments as a result of F_z (Weidenreich, 1923; Mann and Hagy, 1979). Using kinematics, force and plantar pressure measurements, this hypothesis was tested in a sample of human subjects (Rolian et al., 2009). It was demonstrated that subjects with relatively long forefoot lengths had to generate more than four times the peak flexor force over a single stance phase compared to short-forefoot individuals. The authors suspected that such an increase in force output would lead to at least a small increase in the metabolic cost of running. This seems very likely given that the elevated muscle force would result in a greater active muscle volume and a concomitant increase in energy cost. Thus, it seems logical to suggest that it is the ratio of Fz moment arm to AT moment arm length, rather than the absolute AT moment arm length itself which dictates the muscle energy cost.

Body composition and distribution of mass may be another relevant feature in dictating muscle energy cost, and therefore E_run. Active skeletal muscle is primarily responsible for the energy use, so a body mass consisting of a high proportion of skeletal muscle mass and low fat mass should be advantageous in reducing the absolute energy cost of running over a fixed distance (J·m⁻¹). This lower absolute energy cost has advantages in less heat generation, and lower relative muscle activation needed for running at a given speed. Although stored fat contributes to the provision of metabolic energy during running, it is considered that in most cases much more than enough energy is available. Transporting metabolically-inactive tissue like fat would come at a metabolic cost. In fact, Kenyan boys have smaller calf circumference than boys of similar age from other continents (Larsen, 2003). This suggests that even lower muscle mass may be advantageous, as long as sufficient muscle mass is available to provide the required forces and support the metabolic rate.

It is estimated that the oxygen cost of running (measured as V˙O2, ml·kg⁻¹·min⁻¹) was elevated by 4.5% for every additional kg of load carried distally on the legs (500 g distributed across both legs) whereas the energy cost was only elevated by 1% when that same mass was carried on the trunk (Jones et al., 1986). Therefore, minimizing the mass of the swinging limbs, by minimizing fat and unnecessary muscle mass in these areas should reduce E_run, as long as the muscle mass necessary to generate the forces and movements is maintained.

Since running involves rotation of the limbs, a substantial portion of the limb's mass should be located at a close proximity to the joint center of rotation. This serves to minimize the limb moment of inertia. Moving the limbs comprises a substantial portion of the total metabolic cost of running; the joint moment needed to impart an angular acceleration is proportional to the moment of inertia (Fenn, 1930; Cavagna et al., 1964). Swinging the limbs during running should come at a substantial energy cost. Modica and Kram (2005) used a device that assisted the forward swing of the leg, reducing the need of the muscles to swing the leg directly. Their results revealed a reduction in metabolic cost by 20%. This estimate was later refined to ~7% of the metabolic cost of running (Warddrip, 2007), the difference likely a result of the device used by Modica and Kram also aiding in forward propulsion (Arellano and Kram, 2014a).

The metabolic cost of arm swing has also been addressed (Arellano and Kram, 2014b). Swinging the arms incurred a small metabolic cost; however, the arm swing also serves to reduce the amplitude of shoulder and torso rotation. Without arm swing, shoulder and torso rotation must increase to counterbalance the rotational angular momentum created by swinging of the legs. Thus, running with a normal arm swing (i.e., arm carriage with a bent elbow) incurs the lowest E_run since moment of inertia of the upper limbs is lowest. The benefit of normal arm swing during running relates to a reduced E_run of ~3%. Hypothetically though, E_run could be reduced by an even greater extent without the extra mass of the arms, since the reduction in metabolic cost is nearly proportional to the reduced mass (Arellano and Kram, 2014b). Future investigations into the energy cost of elite Paralympic arm amputee distance runners may help resolve this hypothesis.

Above, we have attempted to outline those factors not affected by training which likely alter the energy cost of muscle contraction, and therefore, serve as determinants of whole-body E_run. There exist specific anthropometric (e.g., limb length) and morphological (e.g., ankle and foot anatomy) characteristics that influence the measured E_run.

Body mass may or not be a trainable feature. Long-distance runners are shorter and lighter than middle-distance runners (Cavanagh et al., 1977). Also, elite African runners appear to be of lower body mass (Coetzer et al., 1993) and BMI (Saltin et al., 1995) compared to their Caucasian counterparts. These anthropometric differences appear to have persisted since childhood (Larsen et al., 2004). However, body fat mass can also be reduced through large volumes of aerobic and high-intensity training, through increases in caloric expenditure. Also, transporting metabolically-inactive tissue would elevate the energy cost. While little research has examined why low body mass confers an athletic advantage, several factors specifically related to the energy cost of muscle contraction may explain this. For example, it is well-established that F_z expressed relative to body mass is increased as a function of running speed (Keller et al., 1996). Thus, at a given running speed, the absolute F_z is lower in lighter runners compared to heavier runners. As such, there should be lower energy cost required by the active muscles. Over a wide range of body mass, across different species, Taylor et al. (1980) showed that E_run at a particular speed is proportional to the force exerted by the muscles active during stance. By manipulating the required muscle force by the addition of extra weight it was shown that the increased energy cost was proportional to the weight of the carried load.

At speeds below the anaerobic threshold, where E_run is most appropriately measured, the lowest E_run in humans is generally thought to occur at stride frequencies of 83–91 strides per minute (Hunter and Smith, 2007). The freely-chosen stride frequency closely corresponds to the stride frequency associated with the lowest energy cost (Högberg, 1952; Cavanagh and Williams, 1982; Hunter and Smith, 2007) particularly in trained runners. At a given running speed, a change in stride frequency must result in an opposite change in stride length; however, the naturally chosen stride frequencies and stride lengths, both increase with faster running speeds. There is a proportionally greater increase in stride length compared to the increase in stride frequency, at least at submaximal speeds where the measurement of E_run is valid (Cavanagh and Kram, 1989).

Small animals use more energy (per kg of body mass) to run a given distance than do large animals (Kram and Taylor, 1990) since small animals must take many strides to cover the same distance a large animal can cover in one stride. The mass-specific energy cost is highest in small animals since the muscle fascicles of these animals must develop force and relax more quickly, thus requiring greater rates of cross-bridge cycling and Ca²⁺ pumping (Barany, 1967; Rome, 1992). Considering human runners, those runners with longer legs, and thus longer stride lengths should have a lower energy cost; they will take fewer strides to cover a given distance than a runner with small strides. However, the relationship between stride length (expressed either in absolute terms or relative to height or leg length) and E_run in human runners, unlike the relationship seen across a wide-range of stride frequencies in animals, is moderate at best (Cavanagh and Williams, 1982; Williams and Cavanagh, 1987; Cavanagh and Kram, 1989).

Running is often considered a bouncing gait whereby humans literally bounce along the ground (Cavagna et al., 1964), storing, and recovering kinetic and potential energy as the center of mass rises and falls with each stride, thus closely resembling a simple spring. To address this, simple studies of humans hopping have revealed optimal conditions for energy storage and release. By having subjects hop at various speeds on a treadmill, Farley et al. (1991) were able to deduce that a range of hopping frequencies existed whereby the body behaved like a spring, storing, and recovering elastic energy. However, at faster than optimal frequencies, the time available to apply force to the ground was necessarily shorter, and more contacts per unit time would be required. At slightly slower than optimal frequency, the body did not behave in a spring-like manner and the recovery of elastic energy was reduced. Clearly there is a trade-off between ground contact time, and the requirement to generate force rapidly and the ability to sustain large forces over a relatively long period during the stance phase, which serves to minimize E_run.

The fact that runners tend to choose a stride frequency slightly lower than optimal frequency suggests a greater importance is placed on maintaining ground contact time (and thus allowing a lower recruitment of muscle fibers) over maximizing the storage and release of elastic energy. The self-selected stride frequency should be the one at which the metabolic cost of operating the springs is the lowest (Farley et al., 1993) since muscle metabolic energy is required in order to store and release elastic strain energy from the tendons (Alexander, 1986; Fletcher and MacIntosh, 2015). An interesting situation that avoids this issue is that the horse has a vestigial “muscle,” with tendon but no muscle fibers (Biewener, 1998b) that provides a pure energy storage and return without the metabolic cost.

Modeling running as a simple spring-mass system can characterize the mechanics of the body's center of mass quite well (McMahon and Cheng, 1990; Farley et al., 1993); however, it does not adequately explain the energetics of running. Theoretically a perfectly-elastic spring could supply all of the mechanical work on the center of mass during running, and no net metabolic cost would be required (Arellano and Kram, 2014a). A complement to the spring-mass model hypothesis, the “cost of generating force hypothesis” was proposed (Taylor et al., 1980).

By measuring the metabolic cost of carrying various loads, these authors observed the metabolic cost increased in direct proportion to the added load. Therefore, it was proposed that the metabolic cost of running arose in association with the cost of generating force over time, rather than generating mechanical work. The metabolic cost is proportional to the average vertical force applied to the ground and inversely proportional to the ground contact time over which the force can be applied (Kram and Taylor, 1990). The required peak vertical force must be proportionally higher as speed increases, elevating the metabolic cost since muscles must generate more force while shortening at a faster velocity. To generate the higher force, while compensating for the decreased force per motor unit due to the force-velocity relationship, more motor units must be recruited (Roberts et al., 1998b). In further support the cost of generating force hypothesis, several authors have shown an inverse relationship between E_run and ground contact time (Williams and Cavanagh, 1987; Chapman et al., 2012; Di Michelle and Merni, 2014). These results suggest the speed-associated increase in E_run is a result of the elevated muscle energy cost associated with generating greater peak forces over a shorter period of time.

It appears that a rearfoot strike pattern is more economical than either a midfoot or forefoot strike pattern (Gruber et al., 2013; Ogueta-Alday et al., 2014). Heelstrike reduces the plantarflexor moment at the ankle because the center of pressure resides under the heel of the foot during the first half of stance and this reduces the length of the corresponding (F_z) moment arm (Cavanagh and Lafortune, 1980; Williams and Cavanagh, 1987). Conversely, the center of pressure during stance, a surrogate of the Fz moment arm length, is centered under the ball of the support foot in the forefoot landing pattern. Thus, heel strike pattern is associated with a substantially reduced EMG of the lateral gastrocnemius and soleus muscles compared to forefoot strike (Cunningham et al., 2010). Williams and Cavanagh (1987) found the most economical runners were those with a heelstrike pattern. Forefoot striking may result in a longer F_z moment arm, resulting in higher necessary TS force.

The main issue with examining differences in E_run between forefoot and heel strike patterns is that many studies artificially impose an unnatural gait on the subject. Thus, a lower E_run measured under one condition may be the result of runners being unfamiliar with the novel gait pattern. Gruber et al. (2013) measured E_run in habitual forefoot and heelstrike runners and found no difference in V˙O2 between groups when running with their habitual footstrike pattern. Interestingly, at all running speeds (3–4 m·s⁻¹ was the range evaluated), runners habituated to the heel strike pattern showed a higher V˙O2 when asked to forefoot strike, which was not seen when the forefoot group ran with a heel strike pattern. Only at high speeds was the heel strike pattern less economical in the habitual forefoot runners. When the muscle-tendon unit of the triceps surae was modeled to assess the muscle mechanics and energetic differences between foot strike patterns, it was shown that the forefoot strike pattern resulted in a near-isometric contraction during stance. This allows a lower muscle energy cost for a given force compared to the heel strike pattern, where high contraction velocities during stance were demonstrated. A significant difference in the metabolic energy cost, however, could not be shown.

Despite the general belief among runners and coaches that greater flexibility may result in improved E_run (Craib et al., 1996), there is very little evidence to support this notion. A lower flexibility (measured during a sit and reach test) is associated with a lower E_run (Gleim et al., 1990; Craib et al., 1996; Trehearn and Buresh, 2009). Various suggestions have been made by which a lower flexibility may decrease E_run: (1) reducing the trunk muscle energy cost to maintain stability (Craib et al., 1996) and/or (2) increasing the storage and return of elastic energy (Jones, 2002). The latter mechanism appears unlikely given that such a small amount of mechanical energy is stored and released as elastic energy. This mechanical energy represents only a small savings in total metabolic energy (Ker et al., 1987; Fletcher and MacIntosh, 2015) and mechanically, a stiff AT stores less strain energy for a given force compared to a more compliant tendon. It should be pointed out that if muscle had been required to produce the work done by the tendon, the energy cost would have been 4–5 times greater than the work accomplished by the tendon. As we have previously suggested, an optimal tendon stiffness exists and therefore, a delicate balance between the amount of stretching training (with the intention that stretching training will reduce AT stiffness Kubo et al., 2002; Morse et al., 2008) and strength training [to increase tendon stiffness (Kubo et al., 2001a,b) may result in less than optimal tendon mechanical properties where optimal properties minimize muscle energy cost (Fletcher et al., 2013a)]. Another possible mechanism for low flexibility contributing to low E_run is that low flexibility may relate to short muscle fascicles and shorter muscles should use less energy when velocity of shortening is not important.