Secondary Logo

Journal Logo

Perspective for Progress

Steps Forward in Understanding Backward Gait

From Basic Circuits to Rehabilitation

Hoogkamer, Wouter1*; Meyns, Pieter2*; Duysens, Jacques1,3

Author Information
Exercise and Sport Sciences Reviews: January 2014 - Volume 42 - Issue 1 - p 23-29
doi: 10.1249/JES.0000000000000000
  • Free


Backward walking (BW) and running (BR) (sometimes referred to as retro running, retro locomotion, or reverse running) have been thought to be used already for several decennia in China and Japan not only to get a good physical workout but also to stay mentally fit. While running or walking backward, one has to rely more on senses other than the visual system (like the auditory and sensory system) because one does not have a complete view of the road and obstacles ahead. In the last decade, interest in BR and BW has risen in other parts of the world as well (e.g., United States, United Kingdom, Italy), as BR races and championships are increasingly popular. Furthermore, BW and BR are used as a preferred (rehabilitation) exercise under some conditions, in particular when the impact of heel strike needs to be avoided (35,40). However, it is still unknown whether the effects of BW can be transferred to forward walking (FW). This would be the case if the same neural structures are involved in both types of locomotion. It was indeed suggested that this is the case but it generally is not accepted. If one reverses a video from someone walking backward, one barely can notice any difference with the same subject walking forward (42). This apparent similarity has led some authors to propose that these two forms of locomotion are generated by the same basic neural mechanisms (10,18,23,45). However, some findings, for example, from adaptation studies (4), do not fit well into this framework. Hence, this question needs reexamining based on kinematic and kinetic studies, as well as on experimental work with electromyography. In addition, some studies on reflex modulation also are relevant for this issue. It is argued that both forms of gait can be controlled largely using the same basic neural generators. Furthermore, it is suggested that, for both forms of locomotion, the basis is formed by a central structure that is generating a flexor synergy (flexor burst generator) that is activated periodically when unloading of the legs occurs.


BW and BR are used (increasingly) in sports and in rehabilitation settings. Although several advantages for BW and BR have been advocated compared with FW and forward running (FR) (37), scientific evidence is rather scarce.

BW and BR have been used as part of specific training modalities in different athletic sports for different reasons. For instance, in several sports (e.g., tennis and American football) at some point in a match, players will need to run backward, therefore, the implementation of BW and BR in their individual workout allows for task-specific training. Conforming to this line of reasoning, Agbonlahor et al. (1) have studied BW at different slopes for tennis players and suggested that this could be used as alternate training exercise to complement their sports training modality.

On the other hand, BW and BR are used as a training modality to improve general aspects of fitness (for instance, endurance) that are related to other skills as well. Several studies have shown that BW and BR have a substantially increased cardiopulmonary load compared with FW and FR at similar speeds and elevation angles (for instance, see Flynn et al. (13)). This has led some coaches and trainers to believe that BW and BR are more appropriate to increase aerobic capacity than FW and FR (37). In this case, however, the question arises as to why one would start to train BW or BR at a specific speed and not just increase the speed for FW or FR to increase the cardiopulmonary requirements. The most frequently cited reason to start training BW or BR is that FW and FR are not possible or disadvantageous for an athlete. In that case, BW and BR have been suggested to have the benefit of minimizing knee overload because ground reaction, impact, and patella-femoral joint compression forces are reduced compared with FW and FR (35,40). This benefit of knee load reduction may help an athlete suffering from knee injury (e.g., anterior cruciate ligament injury, patella-femoral pain) to maintain, or even further improve, cardiorespiratory fitness without increased knee joint stress when compared with FW or FR training techniques.

In addition to reducing knee joint stress, BR training also may improve anterior cruciate ligament stability (and anterior lateral rotator stability) in people with knee injury (30). This study, however, did not include an FR control group. The authors suggested that an overstretching of the anterior cruciate ligament was prevented by excessive quadriceps muscle action during the BR training. This is an observation that is supported by some researchers who found that BR has increased power and torque demands on the knee extensors (for instance, see Threlkeld et al. (40)). Flynn and Soutas-Little (14) studied the muscle firing patterns in FR and BR in detail. Their results indicated that muscle action of the knee extensors largely was eccentric and concentric during FR, while they were isometric and concentric during BR, which could be beneficial for the rehabilitation of patellar tendinitis (i.e., decreased eccentric loading of the extensors, which is implicated in these patients). A combination of BW and moving up an inclination has been suggested to enhance further the positive effects of BW on rehabilitation of knee injury because uphill BW requires increased activation of the lower limb muscles in combination with an increased knee flexion and ankle dorsiflexion (5).

One of the reasons that BW and BR are advocated for athletes (e.g., long distance runners) is that it would be one of the few “natural” ways of strengthening the quadriceps (37). This claim actually is supported by previously mentioned findings that BR requires increased knee extensor activations. It also is suggested that, this way, the imbalance between the knee extensors and flexors (i.e., quadriceps/hamstring strength ratio) can be realigned toward the ideal 60/40 (37). In that case, maintaining this muscle balance would prevent possible knee injuries. However, it is unclear why one would start a training paradigm with BW or BR because we have found no evidence that BW would result in a faster strengthening of the quadriceps muscle than other (conventional or analytic) strength exercises.

Related to the prevention of injury are the results described by Whitley and Dufek (44). They reported that hamstring flexibility increased (measured with the sit-and-reach test) after a 4-wk BW intervention, which could explain further the possible preventive effect of BW or BR on knee injuries (44). However, no control group was included that received an FW intervention to be able to determine whether the positive effect on hamstring flexibility is caused by BW specifically. The authors measured the effect of BW training on both hamstring flexibility and low back range of motion because it is suggested that BW or BR training could have a positive effect on low back pain (7). A 10- to 15-min BW training program for 3 days per week for 3 wk significantly decreased low back pain and increased low back range of motion (ROM) in five athletes (who experienced low back pain) (7). Low back ROM also was increased in five control (nonathletic, pain-free) subjects. The study by Whitley and Dufek (44), however, could not duplicate the low back ROM results in 10 normal control subjects after a similar training paradigm. None of these studies included a control group that received FW or FR training. Hence, it still is not possible to conclude unequivocally that BW or BR is responsible for these effects rather than walking or running as such.

The combination of possible preventive benefits, the increased cardiopulmonary requirements, and its relatively novel character have made BW and BR a source of inspiration for research in normal (nonathlete) adults (38), children (21), and elderly (16,29). In adult women, BW and BR training had a positive effect on cardiorespiratory fitness and body composition compared with a control group that had no training. Although, again, no FW and FR training control group was included in this study, this study has indicated that a BW and BR training program does have a positive effect on oxygen consumption during an FR test as well (38). The growing interest in the use of BW and BR for rehabilitation has led some researchers to focus on its advantages on a population that might benefit even more than young adults, that is, the elderly. Several studies have suggested that BW and BR could be used in an elderly population; however, caution is warranted because some of the elderly have been found to experience difficulties during BW (29). These difficulties could be related to instability during BW because it was found that BW measures were even more sensitive than FW measures to differentiate elderly who are prone to fall from those not prone to fall (16). The proposition that BW is related to instability has resulted in a study that examined whether balance would improve after BW training in school-aged children (21). After 8 wk of BW training (twice per week), balance was improved significantly compared with that of a control group that received a normal physical education class. A suggested possible explanation for this effect of BW on balance was that children could not use visual input during BW and, therefore, learned to rely more on other sources of sensory input (cutaneous, proprioceptive, and vestibular senses).

Because of this reduced ability to rely on vision during BW or BR, caution is warranted when implementing BW or BR into training. This is of specific importance in rehabilitation settings where patients have less balance control and are more vulnerable to fall. Nonetheless, training with BW or BR has found its place in neurorehabilitation as well. In several populations with neurological impairments, BW has been implemented in the gait rehabilitation program. For several decades, physiotherapists have been treating patients with neurological disorders (e.g., stroke or cerebral palsy) using BW to improve specific components of FW required to develop an independent gait pattern (3). In these rehabilitation concepts, BW is used to increase the motor control of the patient and to reduce the use of aberrant movement patterns. For instance, some children with cerebral palsy present with a crouched gait pattern (i.e., the ankle in a dorsiflexion and the hip and knee are in a flexed position). Gait rehabilitation with BW puts emphasis on positioning the foot behind the body and, thus, facilitates hip extension while performing a knee flexion (rather than using the crouched pattern). To the best of our knowledge, however, no studies have investigated this mechanism and whether there is an actual transfer from practicing BW to improving FW.

Two studies have examined the effects of BW training combined with conventional training in poststroke patients (43,47). Yang et al. (47) compared a control group, which received conventional rehabilitation training, with an experimental group, which additionally received a BW training program. The patients in the experimental group showed superior improvement in several gait characteristics (e.g., walking speed, stride-length, and symmetry) compared with the control group. The improvement in the experimental group, however, simply could be caused by the increased amount of training it received. Weng et al. (43) performed a similar training study. However, their control group received the same amount of conventional training as the experimental group received BW training. Their results also indicated that walking speed was significantly more improved in the experimental group. Furthermore, they found that balance function and the motor function of the lower limbs were increased significantly in the BW group compared with those of the control group. Only one study was found that investigated BW training in children with cerebral palsy (25). This study was aimed at examining whether a BW training program would improve the gait characteristics in children with cerebral palsy. After 8 wk of training, significant improvements were found for several gait parameters (such as walking speed, weight-bearing symmetry, stride, and step length). Because this study did not include a control group, we question whether any of the observed improvements also would be present in children who did not receive BW training.

Several researchers have indicated a beneficial effect of BW (or BR) training on FW, however, the underlying mechanisms of the transfer of these improvements are understood poorly. It is important to point out that transfer characteristics may be either related to common gains in cardiovascular fitness (or in musculoskeletal properties) or caused by common neural structures. Training that results in cardiovascular (e.g., aerobic fitness) or musculoskeletal (e.g., strength, flexibility) gains may be expected to transfer to other tasks that place similar demands on these systems. Thus, it might be possible that the BR training places similar demands on the cardiovascular and musculoskeletal systems as FR, and therefore transfer of gains in these systems may be expected (e.g., aerobic fitness, strength, flexibility). In contrast, for less physically demanding tasks, it is thought that common neural structures could explain transfer. In particular, walking is less demanding than running and, therefore, the transfer of FW and BW could rely more on the use of common neural structures. For podokinetic adaptation (walking on a circular treadmill) some transfer was observed from FW to BW (originating from changes in trunk rotation) (12). In contrast, using a different paradigm (moving platform), Reynolds and Bronstein (34) showed a lack of transfer from forward to BW. The transfer data of Choi and Bastian (4) go in the same direction. In intact subjects, they failed to find transfer of FW split-belt adaptation on BW. This was taken as evidence that FW and BW are generated by “networks for each leg which are largely nonoverlapping.” At first sight, this seems to contradict the suggestion that FW and BW are generated by the same circuitry (10,18,23,27,28,45). Hence, it is worthwhile to review briefly the evidence for and against this proposal.


The idea that the same neural control structures are involved in both FW and BW stems from the observation that there are many similarities in kinematics and kinetics. If one reverses a video from someone walking backward, one barely can notice any difference with the same subject walking forward (42). Not surprisingly, this apparent similarity has led some authors to believe that these two forms of locomotion are generated by the same basic neural mechanisms (10,18,23,24,27,28,39,45).

Kinetics, Kinematics, and Muscle Activity

To further explore this hypothesis, it is essential to first examine the similarities and differences between BW and FW for the various aspects of gait (kinetics, kinematics, and muscle activity). Since the classic work of Thorstensson (39) on this topic, many studies have compared kinematics and kinetics during FW and BW (18,24,32,45). For example, Winter et al. (45) compared joint angles, moments, and powers for the ankle, knee, and hip. They observed similar knee and hip angle patterns but different ankle angle patterns. Hip and ankle moment patterns were similar, but a difference in knee moment during the stance phase was noted. For all joints, the power curves of FW and BW were almost exact mirror images. These findings in the sagittal plane have been confirmed for the anatomical joint angles (18,24). Meyns et al. (32) compared elevation angles of several lower and upper limb segments during FW and BW. Their results indicated that also the upper limb kinematics of FW correlated highly to reversed BW kinematics. The findings seemed to be consistent with the proposal that control of FW and BW may be similar for the upper and lower limbs (both in adults and children).

For muscle activity, the similarity between FW and BW also has been investigated in multiple studies (10,18,22–24,39,45). After analyzing tibialis anterior (TA), rectus femoris (RF), hamstrings, lateral gastrocnemius (LG), vastus lateralis (VL), and gluteus maximus (GM) muscle activity, Thorstensson (39) concluded that “drastic changes” occur between FW and BW. Winter et al. (45) compared activity in the same first three muscles, together with the medial gastrocnemius, vastus medialis, and soleus muscles. They attributed differences in the amplitude of the muscle activation between FW and BW to changes between concentric and eccentric tasks, and they concluded that somewhat similar, but time-reversed, muscle activation patterns could be used in both modes. Grasso et al. (18) used cross correlation and principal component analysis to address the similarities between muscle activation patterns in FW and BW. They analyzed TA, RF, LG, VL, GM, and biceps femoris (BF) muscle activity. The activity of all these muscles in BW strikingly was different from that during FW with r2 coefficients below 0.2 for all muscles. In addition, for all muscles but the BF, it was impossible to predict BW waveforms starting from FW waveforms, despite using all 10 principal components. The differences in muscle activity patterns between FW and BW again were confirmed by Ivanenko et al. (23) in a study on the spatiotemporal organization of α-motoneuron activity in the human spinal cord during gait. The different muscle activity patterns resulted in different spatiotemporal spinal activity maps, with a more intense rostrocaudal banding in BW. However, the temporal structure of the motor output was observed to be similar between FW and BW, meaning that both gaits can be constructed from the same collection of patterns. The shape of those patterns does not change between FW and BW, but the timing and weight of the patterns may change considerably (27). Note that the method of pattern extraction was different from that of Grasso et al. (18), who failed to predict BW patterns starting from FW patterns.

Central Pattern Generators Coordinating FW and BW

Overall, these data listed above gave rise to the proposition that BW could be mostly the reverse of FW as far as neural control is concerned. What is the basis of this control?

In cats, it is well established that the core control units for walking consists of spinal circuitry labeled “central pattern generators” (CPG), but in humans, there is only indirect evidence for CPG (11). Nevertheless, as gait is usually quite automated, it is reasonable to expect that the same “spinal” automatisms are used for FW and BW (as also proposed by Earhart et al. (12) and Ivanenko et al. (23)). The best evidence for a role of spinal CPG comes from infant stepping, where BW can be generated, despite the absence of mature corticospinal projections (28). Infants, when held under their arms with their feet touching a moving treadmill belt, display stepping movements. This behavior has been observed in the majority of 52 infants (2 to 11 months old) in forward, sideways, and backward direction. Opposed to adults, the FW kinematic patterns were not similar to the time-reversed BW patterns. However, differences in muscle activity between FW and BW were similar to those in adults. Furthermore, it was observed that children can make transitions between forward and sideways stepping in a continuous way. The authors argued that those findings are consistent with the concept of a common locomotor network controlling walking in different directions (28). More supportive data came from work by Selionov et al. (36), who studied induced air-stepping movements in humans. When such movements were performed unilaterally, it was found that there were spontaneous transitions between forward and backward stepping movements. The authors explained their data by assuming that spinal CPG could generate both types of movements.

Although parts of the CPG possibly may be used in common for FW and BW, it is too simplistic to expect that there would be a simple reversal of the pattern (as we and others originally proposed, see Duysens et al. (10) and Zehr et al. (49)). FW and BW have very different constraints; hence, a simple reversal is out of the question. The point is, however, that there is no need for a completely different circuitry either. Basic features of the CPG can be used in FW and BW. This is perhaps most elegantly shown in the previously mentioned work in which the muscle activation patterns are analyzed with respect to basic underlying synergies (23,27). Although the muscle activity patterns of FW and BW differ, they still can be reconstructed using just a few of these synergies.

Reflex Studies

Further insights on the control of BW come from reflex studies. It was proposed that reflexes are modulated during the step cycle, and that there may be similarities in this modulation during FW and BW if there are common elements in the control of these two modes of locomotion (48,50). A series of studies by the group of Charles Capaday (for overview, see Ung et al. (41)) explored the soleus H-reflex during BW. In FW, the soleus H-reflex followed the classic pattern of reciprocal inhibition between antagonistic muscles, but in BW, the modulation pattern was different. In BW, the soleus H-reflex was increased markedly during the swing phase. At first sight, this could be taken as evidence that BW was controlled differently, but it soon appeared that the differences were to be explained differently. After training of the BW task, or when participants held on to handrails, the marked reflex increase in swing was no longer present. This suggested to the authors that the reflex increase in BW was related to elements of instability (the unfamiliarity and reduced balance in BW). This explanation was supported by the observation that, after training, the marked reflex increase reappears when subjects close their eyes during BW. Because changes in H-reflex were not caused by changes in motor activity or kinematics, they likely are part of the motor program controlling BW. It was hypothesized initially that the motor cortex would be involved in the control of the H-reflex during BW through the corticospinal tract (41). However, so far, this hypothesis could not be confirmed because transcranial magnetic stimulation (TMS) did not result in high motor-evoked potentials in the soleus in the same period as the high H-reflex was observed (41).

For cutaneous reflexes, the phase-dependent modulation is more pronounced and, therefore, an obvious target to compare FW and BW (50). Duysens et al. (10) compared the reflex activity after cutaneous nerve (sural) stimulation during BW and FW. Based on their observations of similar phase-dependent modulation of reflex activity in the muscles of the stimulated leg during FW and BW, the authors suggested that the reflex modulation could rely largely on a common locomotor network for FW and BW. This was supported by findings from cutaneous reflex studies on backwards (arm) cycling (48,49). For arm cycling, it was observed that reflex patterns at equivalent positions were similar between forward and backward motion (48). Along the same line, for leg cycling, reflex patterns were similar at equivalent functional phases (49). More insights came from observations on the phase-dependent reflex modulation during BW in the contralateral leg muscles (22). Selective reflex activity was observed in the TA of the contralateral leg during stance phase. This reflex activity was mostly independent of background muscle activity. The selective TA reflex activity during stance was suggested to be related to the reduced stability in BW. However, reflex activity and gait stability measures were uncorrelated. More likely, the increased reflex activity was related to the TA’s decelerating function during the stance phase in BW (24). The low correlation between the reflex activity and the background muscle activity suggested again some premotoneuronal modulation, but the source remains unclear. The modulation patterns of BW in reverse showed some similarities with FW modulations, but with major deviations. Hence, overall, these data are only partly compatible with a possible common source for FW and BW. These differences in modulation patterns between FW and BW could reflect cortical input in gait control (for overview, see Hoogkamer et al. (22)).

Cortical Control

There are many indications that, in addition to basic locomotor patterns generated by spinal CPG, some features of gait are controlled strongly by the cortex. This particularly is true for the control of muscles such as TA and VL, which are important in the control of touchdown (which depends heavily on vision in the case of FW). This cortical control of gait recently has been studied intensively by TMS and by electroencephalography (for overview, see Duysens et al. (9)). This cortical input is not distributed evenly but preferentially focuses on some muscles. Dietz (6) suggested that the corticospinal tract is most closely linked with the segmental circuits controlling the flexors such as TA. Studies using TMS have confirmed this. TA receives strong facilitatory input from fast corticospinal connections, whereas the soleus and gastrocnemius muscles receive a less developed corticomotorneuron drive, at least when the size of the contributing cortical surface is measured (2). In the context of the present review, the question arises whether FW and BW have different cortical representations. The few data available indicate that this is not the case. The same areas seem to be involved but to a different degree. For example, using imagery tasks, Godde and Voelcker-Rehage (17) showed that BW as compared with FW required larger activations in the primary motor cortex, supplementary motor area, parietal cortex, thalamus, putamen, and caudatum but less activation in the cerebellum and brainstem. Similarly, using fNIRS and real walking, Kurz et al. (26) showed that oxyHb was greater in the supplementary motor area, precentral gyrus, and superior parietal lobule when participants walked backward rather than forward. This was suggested to show that BW presents more of a stability challenge than FW. Indeed, in human bipedal gait, the motor cortex is especially important because of the added need for control of stability (46).

This added need is likely to be higher in BW than in FW because BW often is reported to be more variable than FW (20,22,26). Specifically, stride time, stride length, knee and hip range of motion, and relative stance phase have been reported to be more variable during BW (20,26). In addition to being more variable, BW is observed to have lower local dynamic stability than FW (22). This lower local dynamic stability makes BW a more demanding task from the perspective of control. However, it is important to note that, in most studies, the familiarization time for the BW task was limited. Whether differences in variability and stability are still apparent when participants are more accustomed to BW after repetitive training of the task is still an important open question. A second notable point is that it is important to have matched speeds to compare FW and BW variability and stability. Self-selected walking velocity commonly is observed to be reduced in BW (20). When BW and FW at equal speed are compared, cadence is higher in BW (18,39).

Kurz et al. (26) have addressed the variability and cortical activation in FW and BW simultaneously and have indeed found an increase in sensorimotor cortical activation measured (using fNIRS) and a greater stride-time variability during BW. However, they were not able to find a relationship between the amount of stride-time variability during BW and the amount of cortical activation of the sensorimotor cortex. The lack of correlation suggests that another gait parameter would be related to the increased cortical processing during BW. However, it should be noted that all participants in this study were required to hold on to handrails during the walking trials. Holding on to the handrails most probably has improved the balance of BW, which likely could have artificially decoupled the relationship between stride-time variability and changes in the cortical activations.

Further insights on the role of the cortex in BW come from a study by Meyns et al. (31). They have compared the gait and coordination patterns during FW and BW in children with cerebral palsy and age-matched healthy control subjects to examine to what extent cortical deficits influenced the kinematic reversal from FW to BW. Both groups presented with the same basic type of interlimb coordination pattern during FW and BW. However, compared with the control subjects, the children with cerebral palsy showed a significant decrease in coordination between the legs and significantly worse coordinative stability between the legs. These results indicate that, in persons with a cortical deficit, the reversal to BW elicits greater coordinative problems in the lower limbs.

Conceptual Framework

How can the same circuitry be used in FW and BW? In the cat, the experiments of Musienko et al. (33) convincingly show that afferent input is likely to be a main contributor. However, in the intact animal or human, there probably also is an important contribution from supraspinal sources. In adults with spinal cord injury, BW can be induced, but there is no transfer of learning (19). Learning to walk FW had no effect on the performance of (untrained) BW indicating again that BW may require a “higher level of supraspinal control” for this type of learning (see also Ung et al. (41)). Contributions of supraspinal structures in gait control also have been observed in brain imaging studies using motor imagery of gait (for overview, see Godde and Voelcker-Rehage (17)). Specifically, imagery of BW required larger activation of cortical areas (e.g., for visual-spatial processing and sensorimotor control) and less activation in the brainstem and cerebellum, as compared with imagery FW (17). Hence, BW seems to be less automated than FW. The cerebellum and brainstem may be needed for fine tuning and learning. At that level, there may be separate circuits supporting either type of gait (Figure). This notion also is supported by the cat experiments of Musienko et al. (33). They compared locomotion evoked from the brainstem (by stimulation of the mesencephalic locomotor region (MLR)) with locomotion evoked by epidural stimulation of the spinal cord. They found that MLR stimulation evoked well-coordinated stepping movements only if the treadmill was moving in the front to rear direction. They concluded that “supraspinal commands … select one of the numerous forms of operation of the spinal limb controllers, namely, the forward walking” (33). Presumably, there are other supraspinal networks where BW is fine tuned.

Basic organization of forward walking (FW) and backward walking (BW). Some basic features of the central pattern generator ((CPG) such as the rhythm generator part) can be used in FW and BW, whereas parts of the spinal pattern generators may be more specific for FW and BW. Hence, the CPG can work in different configurations. The choice of the configurations and the fine tuning are made by separate supraspinal networks (that may be located at specialized separate areas). The F1 to F3 refer to different flexors and E1 to E3 to different extensors. The scheme indicates that some motoneuron pools get input from both the CPG and from supraspinal sources, but this input could go to either flexors or extensors, irrespective of the direction of gait (hence, no direction-specific bias for F or E is implied). For simplicity, the feedback from afferent sources to the various levels is not depicted.

This conceptual framework also could explain why there is a lack of transfer in FW and BW adaptations (see previously). The lack of transfer is not necessarily in conflict with the idea that there may be common generator parts being involved in both FW and BW. Choi and Bastian (4) propose that “independently adaptable locomotor networks exist for each leg in humans. The basic locomotor networks are configured and modulated by feedforward commands from descending brain signals as well as via sensory feedback from afferent input.” Hence, CPG may provide basic rhythms, but the details of the pattern are dictated both by spinal and supraspinal circuits, some of which are specialized for FW, others for BW. Hence, the idea arises of generators that have elements that can be reconfigured so that they can be used for different locomotion modes (15). On the basis of such models, one can predict that some muscles show a simple reversal of the activity patterns but that others would show quite different patterns for FW and BW.

This concept has been confirmed and expanded on by a recent modeling study of our group (24) using forward simulations to study the effects of muscles on the acceleration of the center of mass in FW and BW (24). It was observed that there was a reversal in function in the muscle control of the horizontal movement of the center of mass (e.g., TA and gastrocnemius). However, in the control of the vertical movement, some muscles showed direction-specific contributions (dorsiflexors, including TA). It was concluded that the changes in muscle contributions imply that a simple time reversal would be insufficient to produce BW from FW, and that BW uses extra elements, presumably supraspinal, in addition to a common spinal drive. These additions are needed for propulsion and to meet the specific constraints of BW, in particular, those related to the ankle joint. This requires that supraspinal sources can give rise to a partial reconfiguration of lower-level common networks. Other parts of the spinal generator could be used easily in both types of gait. Control of antigravity muscles is likely through activation of circuits sensitive to load receptors, hence, there is no functional reversal needed in the circuits controlling basic “extension” (as confirmed by Jansen et al. (24)). Similarly, in both types of gait, there is a need to flex the limb at the different joints, and this function could be achieved by a flexor “synergy,” based on flexor reflex circuitry (8).


In recent years, BW and BR are increasingly used in sports and rehabilitation settings. Backward gait is observed to have biomechanical and cardiopulmonary benefits over forward gait under specific conditions. Training that results in cardiovascular (e.g., aerobic fitness) or musculoskeletal (e.g., strength, flexibility) gains may be expected to transfer to other tasks that place similar demands on these systems. From studies evaluating the neural control of gait, it is known that FW and BW can be controlled largely by the same basic neural mechanisms but with the addition of circuits that are more specialized for FW or BW. Transfer of any functional outcomes, motor learning, or adaptation from backward gait to forward gait has not been observed convincingly, indicating that BW also is at least partly controlled by specialized circuits.

We thank K. Jansen, L. Heyrman, G. Verheyden, A.J. Bastian, and J.T. Choi for helpful suggestions and stimulating discussions. We also acknowledge the very helpful suggestions and additions of the reviewers of this article. We also like to apologize for not citing some relevant work in view of the restrictions related to the maximum number of citations.

This work was supported by grants from Research Foundation Flanders (F.W.O.; grants G.0.756.10 and G.0901.11), from Bijzonder Onderzoeksfonds KU Leuven (OT/08/034 and PDMK/12/180), and from the Hercules stichting.


1. Agbonlahor EI, Agwubike EO, Ikhioya GO, Osagiede II, Agbonlahor LN. Cardiorespiratory responses of tennis players to backward slope walking on differential treadmill gradients. Ozean. J. Appl. Sci. 2009; 2: 195–203.
2. Bawa P, Chalmers GR, Stewart H, Eisen AA. Responses of ankle extensor and flexor motoneurons to transcranial magnetic stimulation. J. Neurophysiol. 2002; 88: 124–32.
3. Bobath B. Adult Hemiplegia: Evaluation and Rreatment. 1st ed. London (UK): William Heinemann Medical Books Ltd; 1970, p. 59–62.
4. Choi JT, Bastian AJ. Adaptation reveals independent control networks for human walking. Nat. Neurosci. 2007; 10: 1055–62.
5. Cipriani DJ, Armstrong CW, Gaul S. Backward walking at 3 levels of treadmill inclination — an electromyographic and kinematic analysis. J. Orthop. Sports Phys. Ther. 1995; 22: 95–102.
6. Dietz V. Human neuronal control of automatic functional movements: interaction between central programs and afferent input. Physiol. Rev. 1992; 72: 33–69.
7. Dufek JS, House A, Mangus B, Melcher G, Mercer J. Backward walking: a possible active exercise for low back pain reduction and enhanced function in athletes. J. Exerc. Physiol. Online. 2011; 14: 17–26.
8. Duysens J, De Groote F, Jonkers I. The flexion synergy, mother of all synergies and father of new models of gait. Front Comput. Neurosci. 2013; 7: 14.
9. Duysens J, Severens M, Nienhuis B. How can active cycling produce less brain activity than passive cycling? Clin. Neurophysiol. 2013; 124: 217–8.
10. Duysens J, Tax AA, Murrer L, Dietz V. Backward and forward walking use different patterns of phase-dependent modulation of cutaneous reflexes in humans. J. Neurophysiol. 1996; 76: 301–10.
11. Duysens J, Van de Crommert HW. Neural control of locomotion: the central pattern generator from cats to humans. Gait Posture. 1998; 7: 131–41.
12. Earhart GM, Jones GM, Horak FB, Block EW, Weber KD, Fletcher WA. Forward versus backward walking: transfer of podokinetic adaptation. J. Neurophysiol. 2001; 86: 1666–70.
13. Flynn TW, Connery SM, Smutok MA, Zeballos RJ, Weisman IM. Comparison of cardiopulmonary responses to forward and backward walking and running. Med. Sci. Sports Exerc. 1994; 26: 89–94.
14. Flynn TW, Soutaslittle RW. Mechanical power and muscle action during forward and backward running. J. Orthop. Sports Phys. Ther. 1993; 17: 108–12.
15. Frigon A. Central pattern generators of the mammalian spinal cord. Neuroscientist. 2012; 18: 56–69.
16. Fritz NE, Worstell AM, Kloos AD, Siles AB, White SE, Kegelmeyer DA. Backward walking measures are sensitive to age-related changes in mobility and balance. Gait Posture. 2013; 37: 593–7.
17. Godde B, Voelcker-Rehage C. More automation and less cognitive control of imagined walking movements in high- versus low-fit older adults. Front Aging Neurosci. 2010; 2: 139.
18. Grasso R, Bianchi L, Lacquaniti F. Motor patterns for human gait: backward versus forward locomotion. J. Neurophysiol. 1998; 80: 1868–85.
19. Grasso R, Ivanenko YP, Zago M, Molinari M, Scivoletto G, Lacquaniti F. Recovery of forward stepping in spinal cord injured patients does not transfer to untrained backward stepping. Exp. Brain Res. 2004; 157: 377–82.
20. Hackney ME, Earhart GM. Backward walking in Parkinson’s disease. Mov. Disord. 2009; 24: 218–23.
21. Hao WY, Chen Y. Backward walking training improves balance in school-aged boys. Sports Med. Arthrosc. Rehabil. Ther. Technol. 2011; 3: 24.
22. Hoogkamer W, Massaad F, Jansen K, Bruijn SM, Duysens J. Selective bilateral activation of leg muscles after cutaneous nerve stimulation during backward walking. J. Neurophysiol. 2012; 108: 1933–41.
23. Ivanenko YP, Cappellini G, Poppele RE, Lacquaniti F. Spatiotemporal organization of alpha-motoneuron activity in the human spinal cord during different gaits and gait transitions. Eur. J. Neurosci. 2008; 27: 3351–68.
24. Jansen K, De Groote F, Massaad F, Meyns P, Duysens J, Jonkers I. Similar muscles contribute to horizontal and vertical acceleration of center of mass in forward and backward walking: implications for neural control. J. Neurophysiol. 2012; 107: 3385–96.
25. Kim SG, Ryu YU, Je HD, Jeong JH, Kim HD. Backward walking treadmill therapy can improve walking ability in children with spastic cerebral palsy: a pilot study. Int. J. Rehabil. Res. 2013; 36: 246–52.
26. Kurz MJ, Wilson TW, Arpin DJ. Stride-time variability and sensorimotor cortical activation during walking. Neuroimage. 2012; 59: 1602–7.
27. Lacquaniti F, Ivanenko YP, Zago M. Patterned control of human locomotion. J. Physiol. 2012; 590: 2189–99.
28. Lamb T, Yang JF. Could different directions of infant stepping be controlled by the same locomotor central pattern generator? J. Neurophysiol. 2000; 83: 2814–24.
29. Laufer Y. Effect of age on characteristics of forward and backward gait at preferred and accelerated walking speed. J. Gerontol. A Biol. Sci. Med. Sci. 2005; 60: 627–32.
30. Mackie JW, Dean TE. Running backward training effects on upper leg musculature and ligamentous instability of injured knees. Med. Sci. Sports Exerc. 1984; 16: 151.
31. Meyns P, Molenaers G, Desloovere K, Duysens J. Interlimb coordination during forward walking is largely preserved in backward walking in children with Cerebral Palsy. Clin. Neurophysiol. 2013 (in press). doi: 10.1016/j.clinph.2013.08.022.
32. Meyns P, Desloovere K, Molenaers G, Swinnen SP, Duysens J. Interlimb coordination during forward and backward walking in primary school-aged children. PLoS One. 2013; 8: e62747.
33. Musienko PE, Zelenin PV, Lyalka VF, Gerasimenko YP, Orlovsky GN, Deliagina TG. Spinal and supraspinal control of the direction of stepping during locomotion. J. Neurosci. 2012; 32: 17442–53.
34. Reynolds RF, Bronstein AM. The moving platform aftereffect: limited generalization of a locomotor adaptation. J. Neurophysiol. 2004; 91: 92–100.
35. Roos PE, Barton N, van Deursen RW. Patellofemoral joint compression forces in backward and forward running. J. Biomech. 2012; 45: 1656–60.
36. Selionov VA, Ivanenko YP, Solopova IA, Gurfinkel VS. Tonic central and sensory stimuli facilitate involuntary air-stepping in humans. J. Neurophysiol. 2009; 101: 2847–58.
37. Stevenson RK, Feola A. Backwards Running. Robert Stevenson Consultant, 1981.
38. Terblanche E, Page C, Kroff J, Venter RE. The effect of backward locomotion training on the body composition and cardiorespiratory fitness of young women. Int. J. Sports Med. 2005; 26: 214–19.
39. Thorstensson A. How is the normal locomotor program modified to produce backward walking. Exp. Brain Res. 1986; 61: 664–8.
40. Threlkeld AJ, Horn TS, Wojtowicz G, Rooney JG, Shapiro R. Kinematics, ground reaction force, and muscle balance produced by backward running. J. Orthop. Sports Phys. Ther. 1989; 11: 56–63.
41. Ung RV, Imbeault MA, Ethier C, Brizzi L, Capaday C. On the potential role of the corticospinal tract in the control and progressive adaptation of the soleus h-reflex during backward walking. J. Neurophysiol. 2005; 94: 1133–42.
42. Viviani P, Figliozzi F, Campione GC, Lacquaniti F. Detecting temporal reversals in human locomotion. Exp. Brain Res. 2011; 214: 93–103.
43. Weng CS, Wang J, Pan XY, et al. [Effectiveness of backward walking treadmill training in lower extremity function after stroke]. Zhonghua Yi Xue Za Zhi. 2006; 86: 2635–38.
44. Whitley CR, Dufek JS. Effects of backward walking on hamstring flexibility and low back range of motion. Int. J. Exerc. Sci. 2011; 4: 192–8.
45. Winter DA, Pluck N, Yang JF. Backward walking — a simple reversal of forward walking. J. Mot. Behav. 1989; 21: 291–305.
46. Yang JF, Gorassini M. Spinal and brain control of human walking: implications for retraining of walking. Neuroscientist. 2006; 12: 379–89.
47. Yang YR, Yen JG, Wang RY, Yen LL, Lieu FK. Gait outcomes after additional backward walking training in patients with stroke: a randomized controlled trial. Clin. Rehabil. 2005; 19: 264–73.
48. Zehr EP, Hundza SR. Forward and backward arm cycling are regulated by equivalent neural mechanisms. J. Neurophysiol. 2005; 93: 633–40.
49. Zehr EP, Hundza SR, Balter JE, Loadman PM. Context-dependent modulation of cutaneous reflex amplitudes during forward and backward leg cycling. Motor Control. 2009; 13: 368–86.
50. Zehr EP, Stein RB. What functions do reflexes serve during human locomotion? Prog. Neurobiol. 1999; 58: 185–205.

central pattern generator; cortical; locomotion; running; sports; stability; walking

© 2014 American College of Sports Medicine