Distribution of Central Pattern Generators for Rhythmic Motor Outputs in the Spinal Cord of Limbed Vertebratesa OLE KIEHNb AND OLE KJAERULFF Section of Neurophysiology, Department of Medical Physiology, University of Copenhagen, Blegdamsvej 3, 2200 Copenhagen N, Denmark

ABSTRACT: Neuronal networks in the spinal cord are capable of producing rhythmic movements, such as walking and swimming, when the spinal cord itself is isolated from the brain and sensory inputs. These spinal networks, also called central pattern generators or CPGs, serve as relatively simple model systems for our understanding of brain functions. In this paper we concentrate on spinal CPGs in limbed vertebrates and in particular address the question: Where in the spinal cord, in the longitudinal and transverse planes, are they located? We will review the use of lesions to isolate the rhythm and pattern-generating parts of the CPG network, indirect methods like activity-dependent labeling with [14C]-2-deoxyglucose, c-fos, sulforhodamine 101, and WGA-HRP, which label presumed rhythmically active neurons en bloc, and direct methods such as calcium-imaging, extracellular and intracellular recordings, which identify rhythmically active cells directly. With this review we hope to highlight the scientific disagreements and the consensus, which have emerged from these studies with regard to the distribution of the CPG networks in the spinal cord.

I

n modern neuroscience it is a general view that specific behaviors arise from localized regions of the brain. This localization has been found for complex mental functions such as language, which primarily is coded in cortical centers located in the frontal and temporal lobes, and face recognition, which seems to take place primarily in the inferior temporal cortex. Similarly, rhythmic motor behaviors are generated by localized neuronal networks in the central nervous system. In vertebrates, for example, neuronal networks in the spinal cord are capable of producing rhythmic locomotor movements, such as walking and swimming, when the spinal cord is isolated from the brain and sensory inputs.1–3 These spinal networks, also called central pattern generators or CPGs, serve as relatively simple model systems to understanding how neural assemblies produce complex functions, and have been the subject of intense research in a range of different limbed and nonlimbed animals. A comprehensive understanding of the spinal CPGs for swimming has been obtained in two relatively simple vertebrates, the tadpole and the lamprey. The mechanisms for rhythm and pattern-generation and modulation of motor output in these species are discussed elsewhere4–6 (see Grillner et al. and Roberts et al., this volume). In this review we concentrate on spinal CPGs in limbed vertebrates and, in particular, we address the question of where in the spinal cord in the longitudinal and transverse planes these CPGs are located. Attempts to answer this question in a variety of preparations with a variety of methods have generated new ideas about the function a This work is supported by the Danish Medical Research Council, the European Union, and the Novo Foundation. Ole Kiehn is a Hallas Moeller Associate Professor supported by the Novo Foundation. b Corresponding author; e-mail: [email protected] 110

KIEHN & KJAERULFF: CPGS FOR RHYTHMIC MOTOR OUTPUTS

111

of the spinal CPGs, and these ideas have provided direction for more detailed investigations of the cellular elements of these networks. Localization studies have also sometimes given rise to scientific disagreements both between different preparations and within the same preparation. It is our hope with this review to highlight these scientific disagreements, as well as the consensus which emerges from the studies. LONGITUDINAL DISTRIBUTION OF MOTOR CPG NETWORKS IN THE SPINAL CORD The first demonstration that networks located in the spinal cord can generate the details of a rhythmic motor output came from Graham Brown’s7 work in the beginning of this century. He reported that rhythmic locomotor-like alternating activity in antagonist muscles in each hindlimb could be generated for a short time following a transection of the thoracic spinal cord in an animal that had previously had its dorsal roots sectioned. With these experiments Brown unequivocally demonstrated that neuronal networks in the spinal cord deprived from sensory inputs and supraspinal influences can generate a coordinated rhythmic motor output. It is now clear that such “autonomous” motor networks—or central pattern generators (CPGs) as they were later called—are found in all animals. In limbed vertebrates there are different CPGs for controlling the forelimbs and for controlling the hindlimbs. In this review we focus on the CPGs controlling hindlimb movements. Although it was clear from Brown’s initial discovery that the lumbar spinal cord itself could generate a rhythmic alternating motor output in the hindlimbs, these experiments did not determine the rostrocaudal extent of the CPG network. It later became possible to approach this issue when Jankowska, Lundberg, Grillner and their colleagues8,9 found that L-dopa injected intravenously in combination with a monoamine-oxidase inhibitor reliably evoked locomotor movements in cats with an acute transection in the lower thoracic cord (T12-13). Grillner and Zangger10 then made a series of additional transections at different levels caudal to the first transection and examined the effects of these lesions on the production of locomotion. They found that stable alternation of the muscles controlling the ankle joint (LG and TA muscles) persisted as long as part of the L5 segment was connected to the caudal lumbar cord. Since the cord was also transected between S1 and S2, they concluded that “three intact, but deafferented segments (L6, L7, S1 and maybe part of L5) seem to be sufficient for the generation of alternating activity in LG and TA muscles nerves.”10 These transections disconnect the ankle motoneurons, which are located in the lower lumbar cord (L6-S1), from motoneurons located in the upper lumbar cord (L3-L5), such as those acting at the hip. Therefore, these experiments convincingly demonstrate that the isolated caudal part of the lumbar cord can generate a rhythmic output. This led Grillner and Zangger to propose that (1) the CPG controlling L-dopa-induced hindlimb locomotion is distributed throughout the lumbar enlargement (L3-S1) and (2) instead of being composed of one flexor and one extensor half-center, the CPG is composed of multiple unit burst generators, controlling individual joints or muscle synergies.1 These unit burst generators might then be distributed along the cord. The idea of a distributed network was further supported by studies in the cat of another rhythmic movement: scratching. Scratching in the cat can be evoked by sensory stimulation of the ear (“pinna” stimulation), and the CPG network for this movement is located in the lumbosacral enlargement.11 In preliminary experiments Berkinblit et al.12 demonstrated that the rostral part of the cord containing L4 and more rostral segments is capable of generating rhythmic scratch-like movements in rostral muscles. In a subsequent study Deliagina et al.13 confirmed this finding (FIG. 1A-A1). By performing specific destruction or cooling of different parts of the cord, these authors also showed that the caudal part of the cord from L5 and caudally is capable of generating a normal rhythmic output in

112

ANNALS NEW YORK ACADEMY OF SCIENCES

FIGURE 1. The CPG network generating rhythmic scratching in the hindlimb is distributed along the lumboscaral spinal cord. (A-A2) Rostral scratching when the rostral and caudal lumbar cords are disconnected by cooling the L5 segment (indicated by the hatched area in A). (A1) Pinna-evoked scratching before cooling recorded as activity in the sartorius (Sart) nerve (motoneurons located in L4-L5) and in the gastrocnemius (G) nerve (motoneurons located in L7-S1). The upper trace shows recording from an interneuron located in L4. (A2) After cooling L5 the rhythmic scratching continued in the Sart-nerve and in the L4-interneurons. In this case the activity also vanished in the lower lumbar cord. (B-C2) Caudal scratching after lesioning the gray matter in L3-L4 (B) or L4-L5 (C). (B1-B2) The scratching recorded in the Sart-nerve and the tibialis anterior (TA) nerve (motoneurons in L6-L7) was unchanged after destruction of the L3-L4 segments. (C1-C2) When the L4-L5 segments were destroyed no activity could be recorded in the Sart-nerve (because of destruction of the motoneurons) and the period of rhythmic activity in TA was longer than before the destruction. (Adapted from Deliagina et al.13)

response to pinna stimulation (FIG. 1B-B2). These findings clearly demonstrate that the scratch CPG is distributed throughout the entire hindlimb enlargement in the cat. However, when the gray matter in L4 and L5 was destroyed, pinna stimulation produced only slow short-lasting rhythmic motor output in the L6-S1 segments (FIG. 1C-C2). After increasing the excitability of neurons in the L6-S1 segments by slight cooling (see ref. 14 and references therein) sensory stimulation now generated a prolonged rhythmic motor output, which appeared to be independent of the rhythmic influences from the rostral cord. These

KIEHN & KJAERULFF: CPGS FOR RHYTHMIC MOTOR OUTPUTS

113

latter findings suggest that although the CPG network for scratching is distributed along the cord and is possibly composed of unit burst generators, the excitability of the individual burst generators is unequally distributed. In particular, rostral components of the network appear to have a higher excitability than more caudal ones. In the combined network the unit burst generators in the rostral lumbar cord might therefore be leading in a chain of unit burst generators distributed along the cord.13 A rostrocaudal excitability gradient was also found in studies of three different forms of hindlimb scratch reflexes in the seawater turtle.15 Here, key elements of the three forms of the scratch reflex could be generated by neuronal elements in D8-D10 either in isolation or when these segments were connected to segments rostral to the hindlimb enlargement (the lumbar enlargement is constituted by D8-D10 and S1-S2). In contrast, the rhythmogenic and pattern-generating potential was much lower in the isolated caudal hindlimb enlargement (S1-S2), where only a slow and feeble rhythm could be evoked. Although the studies of scratch reflexes in the cat and turtle are not exactly comparable because the reflexes were initiated by different sensory pathways, they both converge on the same conclusion: namely, that the scratch CPG is distributed along the hindlimb enlargement, but that there is a decreasing rhythmogenic potential in the rostrocaudal direction. Later studies in the in vitro chick embryo expanded this notion of a distributed but rostrally biased organization of the CPGs underlying scratch reflexes to also include locomotor activity.16 By using transverse sectioning of the cord, it was shown that the capacity to generate rhythmic activity in the chick embryo was distributed along the hindlimb enlargement. Further, isolated rostral segments generated more cycles in response to electrical stimulation than did isolated caudal segments, suggesting that, as with scratch reflexes in the cat and turtle, there is a rostrocaudal gradient in the rhythm-generating capability. The conjecture about distributed CPG networks in the hindlimb enlargements in vertebrates, which have emerged from studies in cat, turtle, and chick, was recently challenged by Cazalets et al.17 By exploiting the in vitro conditions of the neonatal rat, lumbar spinal cords were partitioned by building Vaseline walls at various rostrocaudal levels. When the upper lumbar enlargement (L1-L2) was exposed to a mixture of 5-HT (100 µM) and Nmethyl-D,L-aspartate (NMA, 20 µM), rhythmic locomotor-like activity could be recorded in all lumbar segments (FIG. 2A). In contrast, when the same combination and concentration of drugs were bath applied to the lower lumbar cord (L3-L6), only tonic activity was induced in lower lumbar segments (FIG. 2B). These and other observations lead the authors to conclude that “in the rat, the spinal network producing rhythmic activity is located between segments T13 and L2, while the lower segments, which contain most of the motoneurons innervating the hindlimbs, do not participate at all in the rhythm genesis.” According to this view the hindlimb locomotor CPG in neonatal rats is not segmentally distributed but confined to a restricted area, as has also been proposed for the forelimb locomotor CPG in the adult mudpuppy.18 This view was obviously at odds with the previous experiments in other vertebrates described above. It was also in conflict with previous preliminary observations in the neonatal rat showing that the isolated L4-L5 segments could produce an alternating activity in ankle flexors and extensors when the cord was exposed to 20–30 µM NMA alone.19 A series of subsequent experiments using transverse sectioning supported these latter findings by producing results that were incompatible with the hindlimb locomotor CPG being restricted to L1-L2 in the neonatal rat. Thus, using transverse sectioning and bath application of a combination of 4.5–9 mM NMDA and 4.5–30 µM 5-HT, Kjaerulff and Kiehn20 showed that the ability to generate rhythmic alternating activity was distributed to all lumbar segments (FIG. 3A). However, the modulation amplitude of the motor output was lower and the period cycle length was longer in the caudal lumbar segments compared with the rostral lumbar segments. Furthermore, rhythm was only observed in 33% of the isolated L4-L5/L6 segments, whereas rhythmic activity was always observed in

114

ANNALS NEW YORK ACADEMY OF SCIENCES

FIGURE 2. Rhythmogenic potential in the longitudinal plane in the lumbar cord of the neonatal rat investigated by a transverse split-bath technique. (A) Adding a combination of 5-HT and NMA to the upper lumbar cord (L1-L2) initiated rhythmic locomotor-like activity in L1-L2 ventral roots (left) or in all lumbar roots including L3-L5 (right). Vaseline walls were placed rostrally between the T13 and L1 segments and caudally between the L2 and L3 segments (indicated by a stippled line). (B) Only tonic activity was elicited when the same combination of drugs were applied to the caudal cord. R-L, right lumbar; L-L, left lumbar; NMA, N-methyl-D,L-aspartate; 5-HT, 5-hydroxytryptamine. (Adapted from Cazalets et al.17)

isolated rostral T12/13-L1/L2 segments. Essentially similar findings were obtained by Kremer and Lev-Tov,21 who showed that transecting the cord at the mid-L3 level blocked activity induced by 5-HT (20 µM) and NMDA (3 µM) in the detached lower cord, while it continued in the upper cord. Increasing the NMDA concentration to 5 µM reinstated the rhythm in the lower cord, although with a lower frequency than the one seen in the upper FIGURE 3. Rhythmogenic potential in the longitudinal plane in the lumbar cord of the neonatal rat investigated by transverse sectioning. (A, left) Rhythmic locomotor-like activity recorded in ventral roots in a rostral (T12-L1) preparation and a caudal (L2-L6) preparation after application of a combination of 5-HT and NMDA. The period was longer in the caudal preparation. (A, right) After deleting one segment from both preparations the rhythm remained almost unchanged in the rostral preparation, whereas the modulation amplitude was reduced in the caudal preparation. (Adapted from Kjaerulff and Kiehn.20) (B) Rhythmic locomotor-like activity recorded in ventral roots before (left) and after a transverse section at the mid-L3 level (right) in a T4 to conus medullaris (end) preparation. The NMA concentration was 3 µM before transection and was increased to 5 µM after the transection in order to obtain rhythmic activity in the caudal cord. Note the long cycle period in the caudal cord. (Adapted from Kremer and Lev-Tov.21) (C) NMA-induced rhythm in caudal segments from the L3/L4 junction to the end of the cord. (Adapted from Cowley and Schmidt.23) T, thoracic; NMDA, N-methyl-D-aspartate.

KIEHN & KJAERULFF: CPGS FOR RHYTHMIC MOTOR OUTPUTS

115

116

ANNALS NEW YORK ACADEMY OF SCIENCES

cord (FIG. 3B). Experiments with strychnine and bicuculline-induced bursting have also shown that the caudal segments can generate a rhythm when disconnected from the rostral cord.22 Taken together, these transverse sectioning experiments indicate that the rhythmogenic network controlling hindlimb movements in the neonatal rats is distributed along the lumbar enlargement and shows a rostrocaudal activity gradient, as described for the cat, turtle, and chick. How do we explain the discrepancies between these transverse sectioning experiments and the partitioning experiments performed by Cazalets et al.? One possibility is that sectioning releases a caudal rhythm-generating capability, which is normally not observable when the cord is intact, because of suppression from rostral segments. Another possibility is that the physical extent of the CPG network is dependent on the class and concentration of the transmitter(s) used to induced rhythmic activity. Cowley and Schmidt23 found that the neonatal rat lumbar cord needed to be connected to the lower thoracic cord for 5-HT alone to induce an alternating rhythmic activity in the hindlimb. In contrast, NMA or acetylcholine in combination with an acetylcholine esterase inhibitor was capable of inducing rhythmic activity in isolated parts of both the rostral and caudal (FIG. 3C) lumbar cord. Subsequent addition of concentrations of 5-HT to the NMA-containing bath, comparable to those used by Cazalets et al.17 but in general much higher than those used by Kjaerulff and Kiehn20 and by Kremer and Lev-Tov,21 resulted in loss of the rhythmic activity in the caudal spinal cord, while the rhythm persisted rostrally. A unified theory encompassing these results is that 5-HT decreases reciprocal inhibition in the neonatal cord. Because the reciprocal inhibition involved in left right alternation is weaker in caudal segments than in rostral segments,20 high concentrations of 5-HT will result in tonic activity in the caudal segments, whereas the rostral segments will maintain their rhythmicity. A 5HT-mediated reduction of glycinergic reciprocal inhibition has been described for the tadpole spinal cord,24 and it is possible that a similar effect is present in the neonatal rat. In other words, when 5-HT is present in high concentrations, the network is locked in a nonbursting mode caudally because of relative lack of inhibition. An asymmetrical distribution of other 5-HT influenced synaptic or cellular mechanisms for rhythm-generation may, of course, also contribute. Whether this is also a sufficient explanation for the inability of 5-HT to induce locomotor activity in the isolated lumbar spinal cord when applied alone is still an open question. Cowley and Schmidt23 have proposed that 5-HT activates a rhythmogenic network primarily located in the supralumbar region and different from the lumbar rhythmogenic network activated, for example, by NMA and acetylcholine. Given our knowledge about the different locomotor patterns induced by the activation of different receptors in the neonatal rat,25,26 this is certainly a possibility. It does perhaps seem counterintuitive, however, that the 5-HT-controlled CPG network should be “dislocalized” from the lumbar cord where all the hindlimb motoneurons are situated. We do favor the idea, however, that the locomotor CPG in the neonatal rat is composed of unit burst generators and that these unit burst generators are reorganized into different rhythm and pattern-generating networks by different transmitters (see ref. 26 for a discussion). This idea of reconfiguring CPG networks has been elaborated extensively in small motor CPGs.27 Clearly, these ideas remain theoretical for limbed vertebrates as long as we have not identified neuronal elements of the CPG networks. The discussion illustrates, however, that we should be cautious when considering the distribution of the lumbar CPG networks (or any network controlling a group of muscles), because the extent of the networks is likely not to be fixed, but determined by the identity of the rhythmogenic transmitter. This possibility of transmitter-dependent configuration of locomotor CPGs must of course influence our results when we physically try to isolate the networks.

KIEHN & KJAERULFF: CPGS FOR RHYTHMIC MOTOR OUTPUTS

117

LOCALIZATION OF SPINAL CPG NETWORKS IN THE TRANSVERSE PLANE The localization of the hindlimb motor CPGs in the transverse plane has attracted as much attention as their extent in the longitudinal plane. In addition to the traditional repertoire of electrophysiological extracellular recording techniques, researchers have applied more recently developed techniques such as intracellular tight-seal whole cell recordings, activity-dependent labeling, calcium-imaging, and microlesioning to map the transverse extent of hindlimb CPG networks. All of these techniques have their own problems of interpretation, which is perhaps one reason why a less clear picture of the localization of hindlimb CPG networks in the transverse plane has emerged than for their localization in the longitudinal plane, where the conclusion is mainly based on transverse sectioning studies. The existence of phylogenetic differences might be another reason. We discuss below the results obtained with the different methods and pinpoint their shortcomings. Microlesion Studies The most direct attempt to localize the hindlimb CPGs in the transverse plane comes from lesion studies using horizontal and sagittal sectioning of the cord. Such studies have been performed by Ho and O’Donovan16 in chick embryos and in the neonatal rat by Kjaerulff and Kiehn.20 In the chick, rhythmic activity with alternation between hindlimb flexors and extensors persisted in lumbar spinal cords in which the dorsal and medial half had been removed (FIG. 4A and C). Following more severe lesions leaving lateral and ventral strips of the cord (the average amount of remaining gray matter was 42 and 22%, respectively), the alternating rhythmic activity was transformed into synchrony, with the same cycle period as before lesioning (FIG. 4B and C). This suggests that the neuronal elements for rhythm-generation in the chick are located ventrally and laterally in the cord. Since both ventral and lateral strips can generate a rhythm, the rhythmogenic network must be distributed to both of these or overlapping regions (FIG. 4D). Furthermore, critical elements responsible for alteration are located dorsomedial to the lateral motor nucleus, separated from the rhythmogenic network.16 That result suggests a separation of rhythm- and pattern-generating areas in the chick (FIG. 4D). In the neonatal rat, the isolated ventral part of the spinal cord could generate normal alternating motor output (FIG. 4E-E1), whereas lateral fragments of the cord showed little or no rhythmogenic potential compared to intact cords (FIG. 4F-F1). These results suggest that the locomotor CPG in the neonatal rat is primarily located ventrally in the cord,20 and is therefore less distributed in the transverse plane than in the chick. Alternatively, rhythm-generating networks might also reside dorsolaterally in the rat but were not revealed because they project to medially located relayneurons before reaching motoneurons, or to medially extending motoneuron dendrites. Activity-dependent Labeling: Indirect Methods to Localize Active Cells In recent years, several attempts have been made to use activity-dependent labeling to identify neurons in the hindlimb enlargement involved in the production of rhythmic motor behaviors. The idea behind these experiments is that the kernel of the hindlimb CPGs and the follower cells driven by the CPGs (e.g., motoneurons) should be specifically labeled because they are highly active during the rhythmic motor output, in contrast to cells that are not related to the rhythmic motor output. A number of different activity-dependent

118

ANNALS NEW YORK ACADEMY OF SCIENCES

labels have been used in an attempt to identify the neurons involved in the production of rhythmic behaviors. [14C]-2-deoxyglucose. In a preliminary study Viala et al.28 used [14C]-2-deoxyglucose (2-DG) as a metabolic marker of neuronal activity (FIG. 5, 1) in spinalized and curarized rabbits in which locomotion was evoked by activating noradrenergic receptors. Compared to control animals, the locomoting animals showed a specific intense labeling in the intermediate gray just medial to the dorsal motoneuron nucleus (FIG. 6A), and the authors suggested that this area is involved in rhythm-generation in the rabbit. Because other areas

KIEHN & KJAERULFF: CPGS FOR RHYTHMIC MOTOR OUTPUTS

119

(including the motor nuclei) were more weakly labeled, this conclusion is based on the assumption that rhythm-generating neurons have a higher metabolic rate than follower neurons. At present, however, such a correlation has not been established. In fact, the exact relationship between glucose-dependent energy consumption and neuronal activity, such as spike generation, transmitter-release, channel-opening, etc., is simply not known. Furthermore, the 2-DG technique has the disadvantage that it does not allow identification of individual neurons, which makes it less usable for a cellular analysis. c-fos and sulforhodamine. c-fos expression and sulforhodamine-labeling are two other activity-dependent methods. c-fos belongs to a family of immediate early genes or transcription factors (FIG. 5, 2). The fos protein can be identified with immunohistochemistry. The rapid and transient expression of c-fos has made it widely used as marker of neuronal activity following specific stimulation or behavioral paradigms (ref. 29 and references therein). Sulforhodamine 101 (SR) is a fluorescent dye which is thought to be taken up in neurons in an activity-dependent manner by transmitter-related endocytose30,31 (FIG. 5, 3). c-fos expression and SR-labeling have the cellular resolution that 2-DG lacks. This increased resolution makes these methods potentially very informative because they can be combined with intracellular staining techniques. This allows researchers to demonstrate whether fos or SR-labeled neurons are active, and if so, what kind of neuronal activity causes labeling. Furthermore, the techniques can be combined with immunohistochemistry targeting neuronal transmitter substances. FIGURE 4. Rhythmogenic potential in the transverse plane in the lumbar cord of the chick (A–D) and the neonatal rat (E–F1) investigated by horizontal and sagittal sectioning. (A) Electrically evoked rhythmic motor activity in the hip flexor sartorius (Sart) and the knee extensor femerotibialis (Fem) in the chick embryo (E9). Data are shown from the intact cord (thoracic segment T7 to lumbar segment LS3; upper traces) and after a combined moderate sagittal and horizontal sectioning leaving an isolated piece of the ventrolateral cord (lower traces). In general, the alternation between the flexor and the extensor persisted after moderate sagittal and horizontal lesions (C). The arrow indicates a single pulse of electrical stimulation. (B) Evoked rhythmic motor activity in Sart and Fem-nerves in a chick embryo (E9) in the intact cord (T7-LS4; upper traces) and after a severe sagittal section leaving an isolated strip of the lateral cord (lower traces). In general, severe sagittal and horizontal sectioning (C) disrupted the alternation between flexors and extensors but allowed rhythmic activity to be generated. (C) Schematics illustrating the extent of moderate and severe lesions in the sagittal and horizontal planes. The dotted outline shows the part of the cord which was removed; the continuous outline the part that remained after the lesion. The straight lines defines the level of the section. (D) Proposed distribution of regions involved in rhythmogenesis in the chick lumbar cord. Rhythmic activity persists in cords with severe sagittal (upper hatched area) or horizontal (lower hatched area) lesions, but alternation is depressed or abolished. When the stippled region is left intact, flexor-extensor alternation persists. (A–D adapted from Ho and O’Donovan.16) (E) Rhythmic locomotor-like activity in the neonatal rat before (left) and after (right) a horizontal section removing the dorsal half of a T12-L6 preparation. The level of the section is indicated by the numeral 1 in E1. Note the strict alternation between homolateral roots (R-L2 and R-L5) and the left/right alternation in bilateral corresponding roots (R-L2 and L-L2). The rhythm was induced by a combination of 5-HT (7.5 µM) and NMDA (7.5 µM). (E1) Schematic of all the horizontal lesion studies. The numeral 2 indicates the most ventral section which allowed alternating rhythmic activity. After lesions indicated with 3, all rhythmic activity was lost. (F) Before sagittal sectioning (left), 20 µM 5-HT and 7.5 µM NMDA induced a regular alternating pattern in a T12-S1 preparation. After sectioning (right), the same drug concentration failed to induce rhythm in the isolated small lateral segment (R-L2/R-L5; indicated by 2 in F1), whereas the rhythm remained in the large fragment of the cord (L-L2/L-L5), although with a lower cycle period. (F1) Summary of the sagittal lesion studies. The vertical solid lines indicate the medial border of the small lateral fragment (upper schematic) and the lateral border of the large fragment (lower schematic). Rhythmic activity was obtained in all large fragments, whereas rhythmic activity was only observed in one lateral fragment resulting from the most medial lesion (indicated by 1). The vertical stippled lines indicate the midline (M) of the cord and the lateral border of the gray matter (L). (E-F1 adapted from Kjaerulff and Kiehn.20)

120

ANNALS NEW YORK ACADEMY OF SCIENCES

FIGURE 5. Schematic of the mechanisms for activity-dependent labeling. (1) [14C]-2-deoxyglucose (2-DG) to measure local glucose uptake in neurons. 2-DG is taken up in neurons in response to increased neuronal activity. Since 2-DG is not metabolized in the cell, it can, after being trapped in the neuron, be visualized by autoradiography. (2) Expression of fos proteins as a measure of neuronal activity. Calcium influx through calcium-channels and/or the NMDA-receptor (NMDAR) complex triggers cytoplasmic synthesis and subsequent nuclear translocation of the fos protein, which can be visualized with an antibody. (3) Sulforhodamine (SR) uptake in response to transmitter-release. SR is taken up by activity-dependent endocytosis and presumably transported back to the soma (“?”) where it can be fixed and visualized. (4) Transneuronal transport of wheat-germ agglutinated horseradish peroxidase (WGA-HRP). WGA-HRP is retrogradely transported to the soma, where it can cross to presynaptic terminals. The crossing appears to be facilitated by activity in the presynaptic cell.

c-fos expression and SR-labeling have recently been used by different groups in the search for the locomotor CPGs in cats and rats. The two methods have given remarkably similar results. Thus, when locomotion was induced by stimulating in the mesenchephalic locomotor region (MLR) in decerebrated and curarized cats, fos labeling was found along the length of the lumbar cord (from L3-S1) with labeled cells concentrated in the medial part of the intermediate gray, lamina VIII and around the central canal.32,33 An almost identical labeling pattern was found with SR in the thoracolumbar (T12-L6) spinal cord of the isolated in vitro neonatal rat (FIG. 6B), which had been exposed to SR for four hours during which it was made to locomote by a combination of 5-HT and NMDA.34 fos labeling was also found in the intermediate gray and around the central canal in the lumbar spinal cords from adult deafferented rats performing a Rota-Rod walking task.35 Following unilateral scratching in the cat, fos-labeled cells were also found in the intermediate gray, but in a more lateral position than during locomotion, and in lamina VIII36 (FIG. 6C). Despite the apparent consistency in the labeling patterns, there are reasons to interpret the results with c-fos and SR with caution. First, the nature of the type of neuronal activity that causes labeling is not known. Under in vitro conditions c-fos expression is believed mainly to rely on calcium-influx through calcium channels or the NMDA receptor gated channel. However, such a clear relationship has not been established in vivo (see ref. 29 for a discussion). SR is believed to be internalized in neurons by endocytosis, presumably in the axon terminal and then transported back to the soma. But the exact relationship to neuronal activity has not been determined. Because of these uncertainties, both false negative as well as false positive labeling are expected to occur. In fact, false negative labeling has already been observed with both c-fos and SR: with both methods the motoneurons in the ventral

KIEHN & KJAERULFF: CPGS FOR RHYTHMIC MOTOR OUTPUTS

121

horn were only sparsely and weakly stained.32–36 Furthermore, Carr et al.37 combined intracellular staining of interneurons with the fos reaction in the cat spinal cord and found that several interneurons which were strongly modulated during locomotion did not express the fos protein. With respect to false positive labeling, Hochman and colleagues38 have recently reexamined SR-labeling in the neonatal rat and found that when 5-HT was used alone to elicit locomotor movements the cellular staining in the lumbar cord was more sparse than when using a combination of NMDA and 5-HT. Moreover, although the bulk of cells were found in the intermediate gray, the staining pattern was more diffuse than that reported by Kjaerulff et al.34 Hochman and colleagues, therefore, concluded that NMDA recruits cells that are not or only weakly correlated with locomotor movements. Although there are alternative explanations to these differences (e.g., that the 5-HT and 5-HT/NMDA-activated networks really are different), the experiments clearly demonstrate that false positive labeling presents a problem in the interpretation of these experiments. Recent pilot studies have also shown that SR-labeling is diffuse and unspecific in the tadpole spinal cord (K. T. Sillar, J. R. McDearmid, O. Kjaerulff and O. Kiehn unpublished data). Finally, SR has been reported not to work reliably in several mammals.39 Transneuronal labeling. Wheat-germ agglutinin conjugated with horseradish peroxidase (WGA-HRP) is transported retrogradely and transneuronally40 (FIG. 5, 4). The transneural transport appears to be activity dependent. WGA-HRP can, therefore, after initial injection into specific muscles be used to label last-order interneurons, which terminate on motoneurons and are active during particular behaviors. By use of this method, labeled WGA-HRP interneurons were found in the lumbar enlargements of cat in the intermediate gray, around the central canal, in the dorsal horn ipsilateral to the labeled motoneurons, and in lamina VIII contralaterally in cats that had been trained to jump and were otherwise walking freely41 (FIG. 6D). A similar labeling pattern was found after stimulating the pyramidal tracts in anesthetized animals. Although these experiments reveal overlapping patterns with the fos and SR-labeling patterns, there is clearly also a more diffuse labeling with WGA-HRP. Because long survival times (up to several days) are usually required for the transneuronal transport to take place, this might be expected since it is difficult to restrict the motor acts to a stereotyped behavior. Moreover, if the neurons belonging to the locomotor CPG are not last-order interneurons they will not be labeled by this method. Direct Methods to Localize Areas Containing Rhythmically Active Neurons Extracellular and intracellular recording techniques and calcium imaging are direct methods used to localize rhythmically active neurons. Unlike activity-dependent labeling, these methods will not create false positive results (although false negatives may occur). Furthermore, these methods give information about the timing of interneuronal activity with respect to flexor and extensor phases. However, recording rhythmic activity does not tell whether the cells are actively engaged in generating the rhythm and pattern or are only passive followers driven by the network. A priori it seems reasonable, however, that rhythmically active cells that are not directly driven by the afferent inputs during locomotion or conveying information to the brain are somehow part of the spinal CPG networks. In a broader sense, this of course also includes identified groups of interneurons, such as Renshaw cells and Ia interneurons, which are active during locomotion but do not seem to be essential for rhythm-generation (see ref. 2 for references). Below we will briefly describe activity mapping experiments. In these studies no attempts have been made to identify the cells in terms of afferent input or projections to motoneuron pools. For a description of neurons that have been identified by their afferent input and are

122

ANNALS NEW YORK ACADEMY OF SCIENCES

KIEHN & KJAERULFF: CPGS FOR RHYTHMIC MOTOR OUTPUTS

123

rhythmically active during cat locomotion the reader is referred to chapters by Hultborn et al. and McCrea et al. in this volume. Calcium imaging. O’Donovan and colleagues42 have used calcium imaging to visualize rhythmically active cells in the transverse and horizontal planes of the cord in the chick embryo. Lumbar spinal interneurons were loaded with the calcium sensitive marker fura2am by superfusion. Changes in the excitation spectrum of fura-2am indicate calcium influx and therefore that cellular activity is taking place. During spontaneous rhythmic activity, active cells were found along the length of the hindlimb enlargements. Eighty percent of the active cells were located in the lateral half of the cord, with a dominance in the ventral compared to the dorsal part (FIG. 7A). These results complement elegantly the previous lesion studies,16 which also suggested a ventrolateral (in addition to a ventromedial) localization of rhythmogenic areas in the chick spinal cord. The problem of false positive results is not as significant with this method as with the passive labeling methods, because it can determine directly whether the activity of cells is modulated or not. On the other hand, the possibility exists for false negative labeling with this method, because cells might not be sufficiently loaded with fura-2am for a detectable signal to be emitted. This has been recognized as a problem in the chick where a group of medial interneurons which project contralaterally show little or no activity after bath-applied fura-2am. However, when the same group of cells was retrogradely loaded with calcium-sensitive dyes, thereby producing a higher intracellular concentration of these dyes than that produced by bath application, a subset of them was observed to be strongly modulated during rhythmicity.42 Extracellular and intracellular recordings. The published data on the location of unidentified locomotor-related rhythmically active cells in the hindlimb enlargement using extracellularly and intracellularly recording techniques are surprisingly sparse. Using extracellular recording in the deafferentated MLR cat, Orlowskii and Feldman43 found that 68% of all rhythmically active cells were located lateral in the ventral horn and in the lateral intermediate area (FIG. 7B). Ninety percent of the cells recorded just dorsolateral to the ventral motoneurons were modulated during locomotion. This region corresponds to the region where many cells are activated in a reciprocal pattern by high-threshold afferent stimulation after L-dopa injection in spinalized cats,8 and to the area in the rabbit with a particularly strong 2-DG labeling and with many strongly modulated interneurons.44 Although these studies might concur on the importance of the lateral intermediate area in FIGURE 6. Distribution of activity-dependent labeling in the transverse plane. (A) [14C]-2-deoxyglucose labeling in the sacral cord of the rabbit before (control) and after 45 min Nialamide and L-dopainduced locomotion. The panel to the left shows cross sections of the sacral cord with arrows pointing to the ventral and dorsal motor nuclei, while the panel to the right shows the corresponding autoradiograms. Note the strong increase in labeling medial to the dorsal motor nucleus after locomotion. (Adapted from Viala et al.28) (B) Distribution of sulforhodamine-labeled cells in the spinal segments L1L6 in an isolated spinal cord preparation from the neonatal rat. Locomotor activity was maintained by a combination of NMDA and 5-HT for 4 h in the presence of sulforhodamine. Each diagram includes labeled cells in three representative anatomical sections in each double segment, and each dot represent a cell. Large dots indicate labeled motoneurons. Cells are primarily located in the medial part of the intermediate gray and in the area around the central canal. (Adapted from Kjaerulff et al.34) (C) Schematic drawing of the L7 and S1 segments in the cat with the distribution of fos-positive nuclei (dots) after induction of unilateral scratching. The contralateral side serves as a control. Each map was derived from the distribution of cells in 10 sections. Labeled cells are distributed to the area around the central canal and in the lateral intermediate gray. (Adapted from Barajon et al.36) (D) Distribution of transneuronally labeled cells in the L3 segment and the rostral half of the L4 segment (left) and the caudal L4 segment (right) of the cat spinal cord. The filled circles represent labeled cells in an anesthetized cat in which the pyramidal tract was stimulated after WGA-HRP injection. Open circles represent labeled cells from a cat that was walking freely and trained to jump after WGA-HRP injection. The injection side is indicated by an arrow. Note the pronounced labeling in the contralateral lamina VIII. (Adapted from Jankowska and Skoog.41)

124

ANNALS NEW YORK ACADEMY OF SCIENCES

FIGURE 7. (A) Maps of activated cells during an episode of spontaneous rhythmic activity in the chick embryos. Cells were loaded with fura-2am. Data from two different preparations. LMC, lateral motor nucleus; DL, dorsolateral; DM, dorsomedial; VM, ventromedial; VL, ventrolateral. (Adapted from O’ Donovan et al.42) (B) Distribution of rhythmically modulated (left) and nonmodulated (right) neurons during locomotion in decerebrated cats. Locomotor activity was elicited by stimulation of the mesencephalic locomotor region. Filled circles in the left panel denote cells which were rhythmically modulated before efferent motor activity was observed. (Adapted from Orlowskii and Feldman.43)

FIGURE 8. Expression of bursting properties in an interneuron in the isolated neonatal spinal cord. (A) Diagrammatic view of the isolated spinal cord preparation. The cord was split midsagittally from L1-L5, and suction electrodes were placed on ipsilateral ventral L2 and L5 roots. A whole-cell patch electrode placed near the central canal is indicated in the drawing. (B) Approximate positions of 99 intracellularly recorded interneurons. Closed dots indicate cells that displayed rhythmic activity during transmitter-induced root activity; open dots indicate cells which did not show locomotor-related activity. The roman numerals indicate spinal cord laminae. (C) Rhythmic membrane potential oscillations in an interneuron and ventral root activity during superfusion of 5-HT (7.5 µM) and NMDA (7.5 µM) superfusion. IC, intracellular; VR, ventral roots. (D) Current clamp recording form the same cell and ventral root recordings 1.5 min after removing the transmitters from the bath. The cell continues to oscillate in a voltage-dependent manner even after the rhythmic ventral root bursts have disappeared. (E) Current clamp from the same cell and ventral root recordings 25 min into the control wash. The cell stops oscillation and fires tonically with depolarization. Bias Current values are indicated below the intracellular traces in C–E. (Adapted from Kiehn et al. 47)

KIEHN & KJAERULFF: CPGS FOR RHYTHMIC MOTOR OUTPUTS

125

126

ANNALS NEW YORK ACADEMY OF SCIENCES

rhythm-generation, it is difficult to assess how systematic the tracking in the cord has been. Therefore, areas other than these lateral regions of the spinal cord might also be involved. In fact, in recent experiments where short-latency extracellular field potentials were systematically mapped in L4-L7 during MLR-evoked locomotion, it was found that the MLRinduced volley activated interneurons in both lateral and medial parts of the intermediate gray.45 Moreover, extracellular recording relies on the observation of spike activity to “see” a cell. Therefore, this technique cannot determine the ratio between modulated and nonmodulated cells in a given region of the cord. This can be accomplished, however, with intracellular recording techniques, which have the additional advantage of giving information about intrinsic cellular properties. Recording from cells intracellularly with sharp electrodes, which primarily target large cells, MacLean et al.46 found that 75% of recorded cells were rhythmically active in the ventromedial part of the intermediate gray during 5-HT/NMDA-induced locomotor activity in the neonatal rat spinal cord. Only one-third of the rhythmically active cells were recruited above spike threshold. In a larger study using tight-seal whole-cell recording (FIG. 8A) which allows intracellular recordings from neurons of all sizes, Kiehn and colleagues47–49 have shown that about 60% of the cells located around the central canal and in the medial intermediate gray in the neonatal rat were rhythmically modulated during transmitter-induced locomotor activity (FIG. 8B). About 12% of the active cells displayed transmitter-induced bursting and plateau capabilities (FIG. 8C–F). Bursting properties in a small subpopulation of spinal interneurons surrounding the central canal were also found by Hochman et al.50 using whole-cell recordings in a thin slice preparation of the neonatal rat spinal cord. The contribution of these cells to rhythmic motor output could, however, not be determined because the slice did not support network activity. The studies in the neonatal rat were all limited in their exploration of the transverse plane, and systematic explorations of all areas of the cord with the whole-cell recording technique are needed to determine the actual density of rhythmically active cells as well as the distribution of bursting cells. These studies in the neonatal rat clearly demonstrate, however, that within the regions of the cord implicated in rhythmogenesis by lesion studies, a large percentage of neurons are in fact rhythmically activated. Whether the cells with bursting properties are the primary drivers of the rhythm is, of course, still an open question (see ref. 47 for a discussion). CONCLUSION Based on surgical isolation experiments, strong evidence exists that the CPG networks controlling rhythmic hindlimb movements in vertebrates are distributed throughout the hindlimb enlargement and most probably also into the lower thoracic cord. This distributed nature of the network supports the idea that the CPG controlling hindlimb movements is not a unitary entity, but can be subdivided into unit burst generators controlling single muscles or joints. The rhythmogenic potential in the networks is asymmetric with the highest excitability in the rostral regions of the cord. Observed differences in the rostrocaudal extent of the networks are likely to reflect the fact that different transmitters or sensory inputs do not activate exactly the same rhythmogenic elements in the cords. With regard to the localization of hindlimb CPGs in the transverse plane, at the moment it is not possible to make a clear statement. Lesion studies in the chick and rat have pointed to a distributed network located in the lateral and ventral cord. Activity-dependent labeling in the cat and the rat have suggested that the medial part of the intermediate gray and the area around the central canal are important for rhythm-generation during locomotion. Furthermore, whole-cell recordings in the rat have demonstrated that these areas contain cells with intrinsic bursting capabilities. In contrast, extracellular recordings in locomoting cats and activity-dependent labeling in the locomoting rabbit and the scratching cat

KIEHN & KJAERULFF: CPGS FOR RHYTHMIC MOTOR OUTPUTS

127

have shown intense activity in the dorsolateral intermediate gray. Other activity-dependent labeling studies during rhythmic behaviors in the cat and the rat have shown more diffuse labeling in both lateral and medial parts of the intermediate gray. Clearly, more research— perhaps most importantly, a systematic mapping of rhythmically active cells in different animals—is needed to clarify the exact transverse location of the CPG networks controlling hindlimb movements in vertebrates. It is possible that this search will define areas which encode specific aspects of locomotion (e.g., rhythm- or pattern-generating), and that the extent of the network, as in small invertebrate motor systems, is dynamically modulated by transmitters. ACKNOWLEDGMENT We are grateful to Matthew Tresch for scrutinizing the English in an earlier version of this manuscript. REFERENCES 1. GRILLNER, S. & P. WALLEN. 1985. Central pattern generators for locomotion, with special reference to vertebrates. Annu. Rev. Neurosci. 8: 233–261. 2. ROSSIGNOL, S. 1996. Neural control of stereotypic limb movements. In Handbook of Physiology, Section 12. Exercise: Regulation and Integration of Multiple Systems. L. B. Rowell & J. T. Sheperd, Eds.: 173–216. American Physiological Society, Bethesda, MD. 3. KIEHN, O., J. HOUNSGAARD & K.T. SILLAR. 1997. Basic building blocks of vertebrate spinal central pattern generators. In Neurons, Networks and Motor Behavior. P. S. G. Stein, S. Grillner, A. I. Selverston & D. G. Stuart, Eds.: 47–59. MIT Press, Cambridge, MA. 4. GRILLNER, S. & T. MATSUSHIMA. 1991. The neural network underlying locomotion in lamprey— synaptic and cellular mechanisms. Neuron 7: 1–15. 5. ROBERTS, A., S. R. SOFFE & R. PERRINS. 1997. Spinal networks controlling swimming in hatchling Xenopus tadpoles. In Neurons, Networks and Motor Behavior. P. S. G. Stein, S. Grillner, A. I. Selverston & D. G. Stuart, Eds.: 83–90. MIT Press, Cambridge, MA. 6. SILLAR, K. T., O. KIEHN & N. KUDO. 1997. Chemical modulation of vertebrate motor circuits. In Neurons, Networks and Motor Behavior. P. S. G. Stein, S. Grillner, A. I. Selverston & D. G. Stuart, Eds.: 183–194. MIT Press, Cambridge, MA. 7. BROWN, T. G. 1911. The intrinsic factor in the progression of the mammalian. Proc. R. Soc. Lond. B 44: 308–319. 8. JANKOWSKA, E., M. G. JUKES, S. LUND & A. LUNDBERG. 1967. The effect of DOPA on the spinal cord. 6. Half-centre organization of interneurones transmitting effects from the flexor reflex afferents. Acta Physiol. Scand. 70: 389–402. 9. GRILLNER, S. 1969. Supraspinal and segmental control of static and dynamic gamma-motoneurones in the cat. Acta Physiol. Scand. Suppl. 327: 1–34. 10. GRILLNER, S. & P. ZANGGER. 1979. On the central generation of locomotion in the low spinal cat. Exp. Brain Res. 34: 241–261. 11. DELIAGINA, T. G, A. G. FELDMAN, I. M. GELFAND & G. N. ORLOVSKY. 1975. On the role of central program and afferent inflow in the control of scratching movements in the cat. Brain Res. 100: 297–313. 12. BERKINBLIT, M. B., T. G. DELIAGINA, A. G. FELDMAN, I. M. GELFAND & G. N. ORLOVSKY. 1978. Generation of scratching. I. Activity of spinal interneurons during scratching. J. Neurophysiol. 41: 1040–1057. 13. DELIAGINA, T. G., G. N. ORLOVSKII & G. A. PAVLOVA. 1983. The capacity for generation of rhythmic oscillations is distributed in the lumbosacral spinal cord of the cat. Exp. Brain Res. 53: 81–90. 14. ANDERSEN, P. & E. I. MOSER. 1995. Brain temperature and hippocampal function. Hippocampus 5: 491–498. 15. MORTIN, L. I. & P. S. STEIN. 1989. Spinal cord segments containing key elements of the central pattern generators for three forms of scratch reflex in the turtle. J. Neurosci. 9: 2285–2296.

128

ANNALS NEW YORK ACADEMY OF SCIENCES

16. HO, S. & M. J. O’DONOVAN. 1993. Regionalization and intersegmental coordination of rhythmgenerating networks in the spinal cord of the chick embryo. J. Neurosci. 13: 1354–1371. 17. CAZALETS, J. R., M. BORDE & F. CLARAC. 1995. Localization and organization of the central pattern generator for hindlimb locomotion in newborn rat. J. Neurosci. 15: 4943–4951. 18. WHEATLEY, M., K. JOVANOVIC, R. B. STEIN & V. LAWSON. 1994. The activity of interneurons during locomotion in the in vitro necturus spinal cord. J. Neurophysiol. 71: 2025–2032. 19. KUDO, N. & T. YAMADA. 1987. N-methyl-D,L-aspartate-induced locomotor activity in a spinal cord-hindlimb muscles preparation of the newborn rat studied in vitro. Neurosci. Lett. 75: 43–48. 20. KJAERULFF, O. & O. KIEHN. 1996. Distribution of networks generating and coordinating locomotor activity in the neonatal rat spinal cord in vitro: A lesion study. J. Neurosci. 16: 5777–5794. 21. KREMER, E. & A. LEV-TOV. 1997. Localization of the spinal network associated with generation of hindlimb locomotion in the neonatal rat and organization of its transverse coupling system. J. Neurophysiol. 77: 1155–1170. 22. BRACCI, E., L. BALLERINI & A. NISTRI. 1996. Localization of rhythmogenic networks responsible for spontaneous bursts induced by strychnine and bicuculline in the rat isolated spinal cord. J. Neurosci. 16: 7063–7076. 23. COWLEY, K. C. & B. J. SCHMIDT. 1997. Regional distribution of the locomotor pattern-generating network in the neonatal rat spinal cord. J. Neurophysiol. 77: 247–259. 24. MCDEARMID, J. R., J. F. S. WEDDERBURN & K. T. SILLAR. 1997. Presynaptic modulation of glycine release, in a locomotor network by two biogenic amines. J. Physiol. 503: 111–117. 25. COWLEY, K. C. & B. J. SCHMIDT. 1994. A comparison of motor patterns induced by N-methyl-Daspartate, acethylcholine and serotonin in the in vitro rat spinal cord. Neurosci. Lett. 171: 147–150. 26. KIEHN, O. & O. KJÆRULFF. 1996. Spatiotemporal characteristics of 5-HT and dopamine-induced hindlimb locomotor activity in the in vitro neonatal rat. J. Neurophysiol. 75: 1472–1482. 27. MARDER, E. & R. L. CALABRESE. 1996. Principles of rhythmic motor pattern generation. Physiol. Rev. 76: 687–717. 28. VIALA, D., C. BUISSERET-DELMAS & J. J. PORTAL. 1988. An attempt to localize the lumbar locomotor generator in the rabbit using 2-deoxy-[14C]glucose autoradiography. Neurosci. Lett. 86: 139–143. 29. MUNGLANI, R. & S. P. HUNT. 1995. Proto-oncogenes: basic concepts and stimulation induced changes in the spinal cord. Prog. Brain Res. 104: 283–298. 30. LICHTMAN, J. W., R. S. WILKINSON & M. M. RICH. 1985. Multiple innervation of tonic endplates revealed by activity-dependent uptake of fluorescent probes. Nature 314: 357–359. 31. KEIFER, J., D. VYAS & J. C. HOUK. 1992. Sulforhodamine labeling of neural circuits engaged in motor pattern generation in the in vitro turtle brainstem-cerebellum. J. Neurosci. 12: 3187–3199. 32. DAI, X., J. R. DOUGLAS, J. I. NAGY, B. R. NOGA & L. M. JORDAN. 1990. Localization of spinal neurons activated during treadmill locomotion using the c-fos immunohistochemical method. Soc. Neurosci. Abstr. 16: 889. 33. DAI, X., J. R. DOUGLAS, B. R. NOGA & L. M. JORDAN. 1998. Localization of spinal neurons activated during locomotion using the c-fos immunohistochemical method. J. Neurosci. Submitted. 34. KJAERULFF, O., I. BARAJON & O. KIEHN. 1994. Sulphorhodamine-labelled cells in the neonatal rat spinal cord following chemically induced locomotor activity in vitro. J. Physiol. 478: 265–273. 35. JASMIN, L., K. R. GOGAS, S. C. AHLGREN, J. D. LEVINET & A. I. BASBAUM. 1994. Walking evokes a distinctive pattern of fos-like immunoreactivity in the caudal brainstem and spinal cord of the rat. Neuroscience 58: 275–286. 36. BARAJON, I., J. P. GOSSARD & H. HULTBORN. 1992. Induction of fos expression by activity in the spinal rhythm generator for scratching. Brain Res. 588: 168–172. 37. CARR, P. A., A. HUANG, B. R. NOGA & L. M. JORDAN. 1995. Cytochemical characteristics of cat spinal neurons activated during fictive locomotion. Brain Res. Bull. 37: 213–218.

KIEHN & KJAERULFF: CPGS FOR RHYTHMIC MOTOR OUTPUTS

129

38. CIMA, C., D. FYDA, L. SONG, M. SAWCHUK, L. M. JORDAN & S. HOCHMAN. 1996. Locomotorlabelled neurons are diffusely distributed in the neonatal rat lumbar cord. Soc. Neurosci. Abstr. 22: 1378. 39. ANGLESON, J. K. & W. J. BETZ. 1997. Monitoring secretion in real time: Capacitance, amperometry and fluorescence compared. TINS 20: 281–287. 40. HARRISON, P. J., H. HULTBORN, E. JANKOWSKA, R. KATZ, B. STORAI & D. ZYTNICKI. 1985. Labelling of interneurones by retrograde transsynaptic transport of horseradish peroxidase from motoneurons in rats and cats. Neurosci. Lett. 45: 15–19. 41. JANKOWSKA, E. & B. SKOOG. 1986. Labelling of midlumbar neurones projecting to cat hindlimb motoneurones by transneuronal transport of a horseradish peroxidase conjugate. Neurosci. Lett. 71: 163–168. 42. O’DONOVAN, M. J., S. HO & W. YEE. 1995. Calcium imaging of rhythmic network activity in the developing spinal cord of the chick embryo. J. Neurosci. 14: 6354–6369. 43. ORLOWSKII, G. N. & A. G. FELDMAN. 1972. Classification of lumbosacral neurons by their discharge pattern during evoked locomotion. Neurophysiology (Kiev) 4: 410–417. 44. VIALA, D., G. VIALA & M. JORDAN. 1991. Interneurones of the lumbar cord related to spontaneous locomotor activity in the rabbit. I. Rhythmically active interneurones. Exp. Brain Res. 84: 177–186. 45. NOGA, B. R., P. A. FORTIER, D. J. KRIELLAARS, X. DAI, G. R. DETILLIEUX & L. M. JORDAN. 1995. Field potential mapping of neurons in the lumbar spinal cord activated following stimulation of the mesencephalic locomotor region. J. Neurosci. 15: 2203–2217. 46. MACLEAN, J. N., S. HOCHMAN & D. S. MAGNUSON. 1995. Lamina VII neurons are rhythmically active during locomotor-like activity in the neonatal rat spinal cord. Neurosci. Lett. 197: 9–12. 47. KIEHN, O., B. R. JOHNSON & M. RAASTAD. 1996. Plateau properties in mammalian spinal interneurons during transmitter-induced locomotor activity. Neuroscience 75: 263–273. 48. RAASTAD, M., B. R. JOHNSON & O. KIEHN. 1996. The number of single postsynaptic currents necessary to produce locomotor-related cyclic information in neurons in the neonatal rat spinal cord. Neuron 17: 729–738. 49. RAASTAD, M., B. R. JOHNSON & O. KIEHN. 1997. An analysis of excitatory and inhibitory postsynaptic currents carrying information about rhythmic activity in the isolated spinal cord from neonatal rats. J. Neurophysiol. 78: 1851–1859. 50. HOCHMAN, S., L. M. JORDAN & J. F. MACDONALD. 1994. N-methyl-D-aspartate receptor-mediated voltage oscillations in neurons surrounding the central canal in slices of rat spinal cord. J. Neurophysiol. 72: 565–577.