Human gait control

Gait depends on both automatic and volitional processes. Supraspinal structures, such as the mesencephalic locomotor region (MLR) and the pontomedullary reticular formation, activate and regulate spinal pattern generators controlling the basic step cycle.6 In addition, functional and efficient gait (navigation through space, obstacle avoidance or adaptation to the goal) depends on higher level control centres, including the basal ganglia (BG), cerebellum, and parietal and frontal cortical areas.7

Gait disturbances are a hallmark of parkinsonian syndromes and can take different forms. In addition to the typical parkinsonian gait, some patients may experience festination and FoG. FoG is a sudden and unexpected inability to walk.1 ,2 Festination is an involuntary shortening and fastening of gait.1 Both symptoms are influenced by factors such as the environment, task demand, medication, attention and emotions.1

The pathophysiology of these disorders remains poorly understood. In PD, the late onset levodopa resistance of the symptom suggests the involvement of cerebral networks different from the ones responsible for akinesia. Specifically, the hypothesis of a progressive extension of the degenerative process to non-dopaminergic structures controlling locomotion has been put forward.8 Cortico-frontal and brainstem lesions have been proposed, as well as disruption in the circuits connecting the BG and MLR, including the PPN.9 In atypical parkinsonism, the association of nigrostriatal dopaminergic loss with either cortical lesions, in the case of CBD and Lewy body disease, and/or brainstem lesions, in the case of PSP, could explain both the severity of the symptoms and the lack of significant improvement following dopaminergic therapy.

Regarding FoG, its occurrence in disorders other than PD also suggests that deficiencies in neurotransmitters other than dopamine could be involved. This idea is supported by the variable response of FoG to dopaminergic drugs. Defects in glutamate, norepinephrine10 and serotonin have been evoked. Thus early onset and severe (up to 70%) norepinephrine depletion occurs in the locus coeruleus and has been suggested to significantly contribute to motor dysfunction in PD.11 ,12 FoG could also result from abnormal norepinephrine release in the frontal lobe.13 Degeneration of the serotoninergic neurons in the raphe nuclei is less severe.13 Finally, the role of cholinergic structures, particularly the PPN, in gait disorders and FoG has attracted a lot of attention in recent years.

The PPN has reciprocal connections with cortical and subcortical structures, such as the cerebral cortex, limbic system, BG and cerebellum, as well as descending projections to the lower brainstem and spinal cord.12 Animal studies have shown that low frequency PPN stimulation (20–60 Hz) can elicit locomotion in decerebrated cats while its pharmacological activation improves locomotor activity in MPTP treated monkeys.14 In contrast, lesion, pharmacological inhibition14 or high frequency stimulation (over 60 Hz) of the PPN causes severe akinesia and postural disorders in monkeys.15 Finally, specific lesions of the cholinergic PPN cells induce postural abnormalities.15 In humans, gait disorders following vascular midbrain lesions or bilateral stroke16 in the PPN area (PPNa) also suggest involvement of this region in locomotion. In addition, post mortem studies have revealed that degeneration of the PPN correlates with the severity of gait difficulties before death.15 ,17 Loss of PPN neurons is even more important in PSP.17 Regarding anatomy, a link has been established between the diameter of the midbrain and the severity of gait disorders in patients with PD18 or chronic hydrocephaly.19

These findings lead to testing the efficacy of low frequency stimulation of the PPNa to alleviate gait disorders in PD. Although early reports described dramatic improvements in gait disorders and FoG,20 ,21 better controlled studies revealed a much more mitigated outcome. Improvement of FoG and decrease in falls related to FoG appear highly variable, ranging from a significant reduction in some patients to a lack of effect in others.22 ,23 Nevertheless, this is the first time stimulation frequencies as low as 20 Hz have been shown to have an effect in movement disorders. Because the relevance of stimulating a degenerating and atrophic mesencephalic structure can be questioned, a better understanding of the mechanisms underlying the influence of PPNa stimulation on gait difficulties is directly needed.

Cerebral imaging appears to be a suitable tool to study the pathophysiological mechanisms underlying gait disorders in advanced PD and parkinsonian syndromes, as well as their modulation by levodopa or PPNa stimulation. Below we review the main neuroimaging studies aimed at identifying the cerebral networks involved in gait control, in both healthy subjects and parkinsonian patients. We also discuss studies exploring modulation of these networks by medical treatment or brain surgery. We describe the methods used in these studies and their limits, and then examine the main results obtained using different locomotion or locomotion-like tasks.

Which tools to study the cerebral networks involved in human gait?

The limits of our understanding of the mechanisms underlying human gait control are mainly due to the methodological and technical constraints of the movement itself. The first difficulty stems from the lack of relevant animal models for studying bipedal gait and gait disorders. Nevertheless, data in rodents, cats and primates have revealed the involvement of cortical, subcortical and brainstem structures in gait regulation. In addition, clinical data relating lesions of the MLR and gait disorders16 strongly suggest its implication in human gait control, although they do not reveal the underlying mechanisms. In this context, neuroimaging studies fill an important gap.

Electroencephalography (EEG) and magnetoencephalography (MEG) measure cerebral activity with good temporal resolution. EEG precisely records rhythmic electrical signals and provides data regarding possible functional deteriorations of their dynamics. However, although it can be used to measure the cerebral electric activity preceding gait initiation in the upright position,24 recording the signal during walking is restricted by movement artefacts. In addition, EEG monitors the activity of neuronal ensembles located at a distance from the electrodes. While high density EEG partly resolves these limits,25 the spatial resolution of the signal is not as good as that of positron emission tomography (PET) or functional MRI (fMRI). Therefore, EEG is often used together with fMRI or PET. MEG consists of measuring the magnetic fields produced by neuronal electric activity. However, similar to EEG, MEG measures the signal generated by a number of synchronously activated neurons, which somewhat limits its spatial resolution. In addition, the technique requires the subject to sit still, precluding the performance of whole body movements. To our knowledge, no MEG study to date has been carried out on gait disorders.

Single photon emission CT (SPECT), near infrared spectroscopy (NIRS), fMRI and PET enable the study of neurovascular coupling—that is, the link between increased neuronal activity and increased metabolism occurring during performance of a specific task. Specifically, SPECT and PET monitor the disintegration kinetics of a radiotracer—such as ¹⁸FDG or H₂0¹⁵ for PET, ^99m[Tc]-HM-PAO or ¹²³[I]-IMP for SPECT—while fMRI measures the BOLD (blood oxygenation level dependent) signal variations associated with changes in the blood oxygen concentration downstream of the activated neurons. NIRS detects small changes in haemoglobin oxygenation, and its physical principles are similar to those of fMRI. NIRS can be used in ambulatory mode and thus can be used to study actual gait performance. However, the infrared light can only penetrate a few centimetres below the skull surface, making it impossible to acquire data from deep brain structures. In addition, both spatial and temporal resolutions are weak.

The half-life duration of the radiotracers used in SPECT is long. Once injected, they are quickly distributed and are retained in the brain for several hours. This allows for execution of a specific task, such as freely moving gait, during which the radiotracer is fixed in the active brain regions, followed by image acquisition. The whole brain, including the BG and cerebellum, can be scanned using SPECT. Thus SPECT could prove useful to differentiate the neural substrates involved in the different types of gait disorders in parkinsonism (shuffling, akinetic or trembling FoG, instability) in well characterised groups of patients. However, if SPECT enables the description of the cerebral structures involved in gait maintenance, it does not shed light on gait initiation, as subjects must walk before and after tracer injection. SPECT also limits the study to a few experimental conditions because of the long half-life of the radiotracers. Finally, the spatial resolution of SPECT is also somewhat limited.

PET and fMRI have the best spatial resolution and enable image acquisition of deep brain structures as well as cortical areas. Their spatial resolution is better than that of SPECT, enabling the comparison of several conditions or tasks within a single experimental session. However, fMRI use in patients carrying intracerebral electrodes raises safety issues. Thus PET or SPECT are usually used in patients with DBS. Finally, using specific ligands and well characterised groups of patients, PET and SPECT could also be useful to study the neurotransmitter alterations associated with particular gait symptoms. However, both techniques involve radiation exposure.

Although fMRI and PET are the two favoured techniques for studying the cerebral network involved in voluntary movements, obtaining good quality images requires that the head of the participant does not move. Therefore, these techniques mainly allow the study of small amplitude upper arm movements or repetitive leg or foot movements26 ,27 while lying or seating, excluding the actual execution of gait. This constraint is likely responsible for the relative scarcity of cerebral imaging studies of locomotion. As no portable system enabling the real time recording of cerebral activity during locomotion currently exists, alternative protocols have been developed.

From metabolic studies to functional neuroimaging of normal and parkinsonian gait

Motor imagery: a new way to study the cerebral networks of gait?

In the past few years, a number of studies have used mental motor imagery (MI) to better understand the neuronal basis of normal or pathological locomotion. MI has long been used for training motor skills in sports48 and or relearning motor skills in rehabilitation settings.48 In addition, MI allows investigation of the internal dynamics of movement planning and preparation, while avoiding sensory and motor compounds related to motor execution.49–51 Thus, associated with brain imaging, MI enables the exploration of the cerebral networks involved in upper limb movements52 as well as in gait control. Furthermore, it has been shown that the cerebral plasticity occurring during skill acquisition could be seen both during actual movement and MI.

Motor imagery of locomotion in cerebral imaging: from healthy subjects to PD patients

Having overcome the problems raised by the actual performance of complex movements, several teams have used MI coupled with fMRI or PET to decipher the cerebral substrates involved in the control of locomotion in healthy subjects.15 ,26 ,60–77 Activations in the fronto-parietal (including SMA) and visuo-occipital (including fusiform and parahippocampic gyri) cortices, BG and cerebellum have been reported during both MI of gait (table 1), and gait related tasks (table 2). These studies have also shown that a number of areas contribute to adjusting the locomotor behaviour to the environmental constraints63 or to the task complexity.69 Using fMRI, Jahn et al64 ,65 reported activation in the dorsal pons during MI of standing and activation in the mesencephalic and cerebellar locomotor regions during MI of walking and running. In healthy subjects, Jahn et al66 also described lesser activation of the vestibular and somatosensory cortical areas during automated locomotion than during obstacle avoidance. Using MI of gait on a curved path, Wagner et al71 observed greater activation in the striatum contralateral to the turn, as well as parahippocampal and fusiform gyri activation. These two structures are involved in visually guided navigation. Comparing actual and imagined locomotion with PET and fMRI techniques, La Fougère et al73 reported activations in the primary motor and somatosensory cortices only during actual locomotion whereas the SMA and BG were activated during imagined locomotion. Moreover, the authors described greater activation within the pontomesencephalic tegmentum (including both the PPN and cuneiform nucleus) during imagined gait. Finally, Karachi et al15 reported activation of the PPN and cuneiform nucleus during imagery of fast gait when comparing imagining normal versus fast gait.

These results were confirmed by fMRI studies in which the actual engagement of the participants in the MI task was controlled.59–61 ,75 Figure 1 summarises the main results of these imaging studies.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Schematic representation of the executive and planning networks of locomotion, illustrating hierarchical control from the cortex to the spinal cord. Solid arrows represent the ascending pathways; broken arrows represent the descending pathways. CB, cerebellum; CLR, cerebellar locomotor region; DLPFC, dorsolateral prefrontal cortex; MLR, mesencephalic locomotor region; M1, primary motor cortex; Pa, pallidum; PMRF, ponto-mesencephalic reticular formation; sCPG, spinal central pattern generator; SMA, supplementary motor area; St, striatum; STN, subthalamic nucleus; TL, thalamus.

To date, two studies have explored gait control in PD.61 ,76 Hypoactivation within the SMA and parietal lobule, as well as hyperactivation in the MLR, were seen during MI of gait in patients with FOG, compared with patients without FoG and matched controls.61

Limitations of motor imagery

These studies call for some comments. First, the MI perspective spontaneously used by the subjects is seldom controlled in spite of the differences in cerebral networks recruited during kinaesthetic and visual MI. In addition, the two perspectives do not engage motor areas in the same way,57 ,72 even if some common neural mechanisms exist.70 Second, all individuals do not have similar imagery skills and it thus appears necessary to assess both their MI ability and performance. The Movement Imagery Questionnaire78 or the Kinaesthetic and Visual Imagery Questionnaire79—recently validated for PD patients80—enable one to measure the vividness of sensations perceived during visual and kinaesthetic MI and therefore the reliability of, say, their kinaesthetic imagery ability. Regarding gait, because it is an over learnt automated movement, it might be particularly difficult to voluntarily imagine. Yet, an adapted training focusing the subject's attention on the sensations triggered by gait execution can decrease visual processes and increase kinaesthetic ones, as demonstrated by Sacco et al.67

Some studies provided MI preliminary training15 ,60–69 71–77 but did not control the correct performance of the MI tasks during image acquisition. Yet, to increase the reliability of MI of gait data, it appears necessary to make sure not only that participants use a kinaesthetic perspective and train them but also to monitor MI performance when the subjects are in the scanner, in order to control for proper task execution. Different methods, based mainly on the temporal similarities of actual and imagined movements, can be used to ascertain that participants actually imagine the target movement during brain image acquisition. The Time Dependant Motor Imagery screening test consists in recording the number of movements imagined over a defined period of time, following the assumption that the number of imagined movements will increase with time. The Temporal Congruence Test involves recording the duration of actual and represented movements to evaluate their chronometric concordance. Bakker et al60 used Fitts' law, asking healthy subjects to imagine walking along paths of various lengths and widths, and then to actually walk along the same paths (figure 2). The results confirmed that actual and imagined walking times were similarly affected by the changes in difficulty of the task. Bakker et al also used a visual imagery task, during which subjects were to imagine seeing a small disk moving along the path. The authors reported that, unlike imagined walking times, the visual imagery times increased with path length, but were less sensitive to its width.60

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Pictures illustrating a motor imagery protocol using Fitts' law to control correct performance of the task. The protocol involves paths of various lengths and widths on which the subjects have to imagine walking and actually walk. Both imagined and actual walking times increase with path length and decrease with path width. A visual imagery task serves as an additional control. Subjects are asked to imagine a blue disk moving along the path. Visual imagery time is influenced by path length but not by path width (adapted from Bakker et al59). (A) Short distance, wide path; (B) long distance, narrow path; (C) long distance, wide path; (D) short distance, narrow path.

Finally, one might question the use of motor imagery (MI) protocols which require high level cognitive processing in patients with gait disorders. Indeed, the link between frontal lobe dysfunction and gait disorders is clearly established.81 ,82 However, although patients with executive function impairments often display gait impairments and falls, a number of PD patients with gait disorders and FOG have well preserved executive functioning.80 In addition, providing specific kinaesthetic MI training to patients before data acquisition both increases the quality of these processes in the selected individuals and contributes to the detection and exclusion of patients whose level of executive functioning can no longer support reliable MI.