Metachronal coupling between spinal neuronal networks during locomotor activity in newborn rat
- Corresponding author J.-R. Cazalets: CNRS Unité Mixte de Recherche 5227, Université Bordeaux 2, Zone nord Bat 2, 146, rue Léo Saignat, 33076 Bordeaux Cedex, France. Email: jean-rene.cazalets{at}u-bordeaux2.fr
Abstract
In the present study, we investigate spinal cord neuronal network interactions in the neonatal rat during locomotion. The behavioural and physiological relevance of metachronally propagated locomotor activity were inferred from kinematic, anatomical and in vitro electrophysiological data. Kinematic analysis of freely behaving animals indicated that there is a rhythmic sequential change in trunk curvature during the step cycle. The motoneurons innervating back and tail muscles were identified along the spinal cord using retrograde labelling. Systematic multiple recordings from ventral roots were made to determine the precise intrinsic pattern of coordination in the isolated spinal cord. During locomotor-like activity, rhythmic ventral root motor bursts propagate caudo-rostrally in the sacral and the thoracic spinal cord regions. Plotting the latency as a function of the cycle period revealed that the system adapts the intersegmental latency to the ongoing motor period in order to maintain a constant phase relationship along the spinal axis. The thoracic, lumbar and sacral regions were capable of generating right and left alternating motor bursts when isolated. Longitudinal sections of the spinal cord revealed that both the bilateral antiphase pattern observed for the sacral region with respect to the lumbar segment 2 as well as the intersegmental phase lag were due to cross-cord connections. Together, these results provide physiological evidence that the dynamic changes observed in trunk bending during locomotion are determined by the intrinsic organization of spinal cord networks and their longitudinal and transverse interactions. Similarities between this organization, and that of locomotor pattern generation in more primitive vertebrates, suggest that the circuits responsible for metachronal propagation of motor patterns during locomotion are highly conserved.
Locomotion in humans as well as in quadrupeds involves complex synergistic postural regulation requiring the integrated functioning of all the body musculature, including hind- and forelimb, trunk and neck muscles. The trunk, which is the main body mass, is a complex, articulated and biomechanically constrained structure. Although numerous kinematic, biomechanical and electrophysiological studies have focused on understanding how the central nervous system controls hindlimb movements (Rossignol, 1996; Duysens & Van de Crommert, 1998; Orlovski et al. 1999), very few studies have been devoted to understanding the functioning of neuronal networks that activate trunk muscles in coordination with limb movements (Thorstensson et al. 1982; Koehler et al. 1984; Zomlefer et al. 1984; Geisler & Gramsbergen, 1998). These studies have shown that back muscles are rhythmically activated during locomotion, suggesting that the preservation of the dynamic equilibrium needed to propel the body can only be achieved through appropriate coordination of trunk and hindlimb premotoneuronal networks. This issue is of particular importance in the context of functional rehabilitation, because one of the most challenging problems in the control of movement in patients with spinal cord injuries is the preservation of postural body balance.
The neuronal networks that generate the basic motor patterns underlying limb movements (central pattern generator; CPGs), have been located within the lumbar spinal cord for the hindlimbs (for review see Rossignol, 1996; Cazalets, 2000) and the cervical spinal cord for the forelimbs (Ballion et al. 2001), and details on intra- and interhindlimb coordination (Rossignol, 1996; Cazalets, 2000; Butt et al. 2002; Butt & Kiehn, 2003), as well as on fore- and hindlimb coordination (Juvin et al. 2005), have been reported. An unresolved question is whether these identified neuronal networks contribute to the coordinated behaviour of the entire body during actual locomotion. We have recently found that the spinal cord of newborn rat contains propriospinal pathways involved in a metachronal propagation of motor activity along the spinal axis (Cazalets, 2005). However, in this earlier sequential study the motor activation was observed during extreme pharmacological manipulation including a generalized blockade of fast inhibitory synaptic transmission. In the present study, we have analysed the interactions between the various parts of the cord during centrally generated spinal cord activity under more physiological conditions. To assess the global functioning of spinal circuitry and to understand how the thoracic, lumbar and sacral segments interact, we have recorded from up to 16 ventral roots simultaneously along the thoraco-lumbo-sacral spinal cord axis during sequences of fictive locomotor activity. Furthermore, we have combined in vitro observations with analysis in intact animals to study the simultaneous functioning of the trunk and hindlimbs during actual locomotion. We found that motor bursts propagate rostrally and caudally from the lumbar region to the most distant cord segments.
Methods
Experiments were carried out on newborn Wistar rats aged from P1 (P0 defined as the first 24 h after birth) to P4. The protocols were approved by the local ethics Committee at the University of Bordeaux 2 and were also in accordance with NIH guidelines.
Two-dimensional kinematic measurements
Four-day-old rat pups were used for kinematic analysis (n = 5). Each animal was tagged with nine dots (1 mm diameter) using a felt tip marker (Fig. 1B) at the following locations: tail (tip, middle and base), spine (sacrum, pelvis, midway between scapuli and pelvis and scapuli), head and right hindfoot. Rats were induced to walk on a rubber strip over 10 trials (approximately eight steps each) and animals were filmed with a CCD video camera (Sony, 25 frames s−1). Sustained locomotion was reliably initiated using the protocol of olfactory stimulation as developed by Jamon & Clarac (1998). Briefly, a tube containing nest material was presented to the pup. As the animal moved towards the scent of the material, the experimenter slowly withdrew the tube so that the animal followed it. Only forward walking was elicited with this method. In a given sequence, the first and last steps were excluded from the subsequent analysis. Video sequences were analysed with ImageJ software (Rasband, W.S., ImageJ, US National Institutes of Health, Bethesda, MD, USA; http://rsb.info.nih.gov/ij/, 1997–2005) and a manual tracking plug-in (implemented by F.P. Cordelières, Institut Curie, Orsay, France). Skin slippage was not compensated for because it appeared minimal in the back region. The x and y coordinates of individual dots were determined manually on each frame under visual inspection. Angles were computed using Microsoft Excel and stick diagrams and angles were plotted using Igor-Pro software (Wavemetrics, OR, USA).
Retrograde labelling
Pups (n = 13) from P2 to P4 were anaesthetized by hypothermia until reflexes were lost and placed on a cold pad in order that they remained anaesthetised during tracer injection. They showed no reflex reactions such as withdrawal in response to the muscle injections. The retrograde tracer, cholera toxin B subunit Alexa Fluor 488 conjugate (CTB at 1% in distilled water, Molecular Probes, OR, USA) was injected using a Hamilton syringe into the right back muscles (longissimus or multifidus) and into the left gastrocnemius. The rats were then returned to their mother for 2 days, after which they were killed and their spinal cords were dissected out and fixed in 4% paraformaldehyde in 0.2 m phosphate buffer (3 h at 4°C). After dehydration, the spinal cord was cleared in methyl-salicylate for 1 h, and then examined using a fluorescence microscope.
In vitro isolated spinal cord
Locomotor-like activity was elicited by bath-application of a mixture of either serotonin (5-HT, 2 × 10−5m) or dopamine (DA, 2 × 10−4m) plus N-methyl-d,l-aspartate (NMA, 1.5–2 × 10−5m) (Sqalli-Houssaini et al. 1993; Barriere et al. 2004).
Pups (n = 33) from P1 to P3 were anaesthetized by hypothermia until reflexes were lost, then decapitated and eviscerated. As described by Squalli et al. (1991), the spinal cord was isolated and pinned down ventral side up in a recording chamber. All dissection and recording procedures were performed under continuous perfusion with Krebs saline containing (mm): NaCl 130, KCl 3, CaCl2 2.5, MgSO4 1.3, NaH2PO4 0.58, NaHCO3 25 and glucose 10; bubbled with 95% O2–5% CO2, adjusted to pH 7.4 with HCl and maintained at room temperature (24–26°C). Spinal cords were sectioned at the T1 level at the beginning of the experiment.
Up to 16 ventral roots were recorded simultaneously (Fig. 4), using stainless steel pin electrodes insulated from the bath with Vaseline (the location of the extracellular electrodes is indicated with dots in the figures). During an episode of fictive locomotion, a steady state was reached within 10 min (Sqalli-Houssaini et al. 1993), after which measurements were made. Signals, amplified 5000 × using custom-made amplifiers, were acquired at 2 kHz on 16 channels using a Digidata 1322A interface driven by Axograph software (Axon Instruments, CA, USA). They were then processed and analysed using Axograph analysis plug-ins. Cord sections were made in the recording chamber with sharp MC-52 scissors (Moria, Paris).
The raw data were rectified and integrated. For this, each burst of activity was detected using a threshold value that was set at 30% of the maximum amplitude of the activity and a minimum period value (according to the mean period in the measured sequence) was used to preclude a given burst from being counted twice. We used L2 as the reference trace because it invariably exhibited the best signal-to-noise ratio (Fig. 4B). Each time a burst was detected in the reference trace, it was sampled with all other associated recorded traces. Individual cycles were superimposed and then averaged. This process allowed switching from a continuous mode of recording to an episodic one. The aim was to increase the signal-to-noise ratio in order to obtain a more accurate detection of burst peaks. Subsequent measurements were automatically performed on the averaged trace using a method based on peak detection (at 30% of maximum peak amplitude). Burst phase was computed by dividing the latency between the reference peak and the analysed trace by the averaged period. Lag graphs were plotted using Igor-Pro software.
Statistical analyses were performed on raw data by descriptive statistics of circular distribution (Zar, 1984). In each experiment, the phase (Φ) was calculated and the mean phase 
was determined with the formula:
with n the number of experiment.
Phase values were plotted in a circular representation (0–360 deg in the trigonometric direction), with the mean phase being
indicated by the direction of the vector, and its length (range from 0 to 1) indicating the strength of the mean. The latter
was calculated by the formula:
The Rayleigh test was used to determine the coupling strength. Circular statistical analyses (circular linear correlations) were performed using R software (Team, 2005) or Oriana (KCS, UK). The significance threshold was taken to be P < 0.05 unless otherwise specified. Other analyses (linear regression, correlation and one-way ANOVA) were performed using Prism software (Graphpad software, CA, USA). All data values in the text are means ± s.e.m.
Results
Kinematic analysis in intact animal
Body movements were studied in newborn rats in order to provide behavioural data that could be correlated to the activity recorded under in vitro conditions. Figure 1A presents images obtained at three different time intervals during one step of olfactory stimulus-induced locomotion. Visual examination of marker dots on the pup clearly shows that the back flexes alternately during stepping. The locomotor parameters (period, stance and swing duration) that we observed with our kinematic analysis were comparable to those previously reported by Jamon & Clarac (1998). The mean cycle period for all steps collected in five different animals was 0.8 ± 0.03 s. The schematic diagram in Fig. 1Ba shows the model that was used for analysing the region-specific movements illustrated in Fig. 1Bb where the various colour lines show the trajectories for various points on the back and tail following manual tracking. Figure 1C shows stick diagrams established for a single step that was broken down into swing (Fig. 1Ca) and stance (Fig. 1Cb) phases with respect to the reference right hindlimb. During both phases, there was a clear bending in the trunk axis as the tail orientation concomitantly changed. Figure 2 presents the angular variations at various trunk levels measured from a pup as shown schematically in Fig. 2A. For each plot in Fig. 2B and C, the coordinate axis corresponds to the average changes in angular position during a duplicated cycle (duty cycle normalized from 0 to 1 in the x axis). During each step, the angle at all levels changed cyclically so that the trunk curvature alternated symmetrically between the right (above 180 deg) and left sides (below 180 deg).
Moreover the angle changes observed in Fig. 2Ba and C also indicate that bending propagates rhythmically through all trunk regions until the base of the tail. To visualize the global changes in trunk bending, we plotted the maximum peak amplitude on one side for each segment analysed during one cycle (Fig. 2D). Each point in Fig. 2D indicates the cycle phase at which a given angle (y axis numbered from 1 to 5) was maximal on the right side (above 180 deg in each plot of Fig. 2B). This revealed that the amplitude was maximal in the lower back at the transition between the swing and stance phase, and then propagated rostrally along the trunk axis in a metachronal sequence.
Anatomical identification of motoneurons
Trunk and tail movements are mainly due to the involvement of the epaxial muscular system, including the multifidus, longissimus and sacroccygeus muscles which run laterally along the body axis. Before further investigating the properties and generation of trunk motor activity, we sought to identify the origin and central distribution of motoneurons supplying axial musculature in the newborn rat because these motoneurons have only been partially localized in adult rats (Brink et al. 1979). This was achieved by injecting the retrograde marker cholera toxin B subunit (see Methods) at various levels into the right axial muscles as indicated in Fig. 3A, as well as into the contralateral (left) gastrocnemius. Figure 3B shows three examples of labelled motoneuron pools, with three different injection sites (tail, Fig. 3Ba; lower back muscles, Fig. 3Bb; mid-back muscles, Fig. 3Bc). Tail muscle injections always resulted in bilateral labelling due to small muscle size. Figure 3Bb shows an example of the labelling on the left side resulting from an injection into the left gastrocnemius (G) and the labelling on the right side corresponding to a right back muscle injection (B) at level 4 indicated in Fig. 3A. Under our experimental conditions, a back muscle injection was found to label motor columns spanning two segments. Moreover, at the lumbar level, there was a substantial overlap between motoneuron pools innervating the back muscles (B) and the motoneuron pools supplying the hindlimb (G, Fig. 3Bb). The summary diagram of Fig. 3C illustrates the motoneuron location (dots), identified following injections at the various sites indicated in Fig. 3A. We have observed a relationship between the rostro-caudal location of the injected muscle site and the rostro-caudal location of the corresponding motor columns, with motoneurons innervating the tail being located in the most distal sacral and coccygeal cord regions. These data therefore showed that motoneurons innervating the axial musculature are distributed along the entire spinal cord, from thoracic to coccygeal segments and including the lumbar segments where they colocalize with motoneurons that innervate the hindlimbs.
Coordinated motor patterns in vitro
Next we recorded motor activity in the isolated spinal cord preparation during chemically evoked sequences of locomotor-like activity. Figure 4 illustrates simultaneous recordings from 16 ipsilateral ventral roots (indicated by the dots in Fig. 4A) during bath-application of a mixture of 5-HT and NMA (see Methods). One characteristic of the multiple-site motor pattern recorded under these conditions was that the signal-to-noise ratio (compare background activity in Fig. 4Ba to pharmacologically induced motor activity in Fig. 4Bb) was always greater in the low thoracic and lumbar ventral roots, suggesting that more units are active at these levels. In addition, there was typically a one-to-one relationship between bursts recorded in all the recorded ventral roots. Occasionally (4 out of 29 experiments) we observed two bursts of activity in either low lumbar (L4–L6) or sacral ventral roots (Cazalets et al. 1992; Cazalets & Bertrand, 2000a). When viewed on an expanded time scale, it is apparent that the motor bursts were not synchronous, as shown in Fig. 5, where the onset of bursts at various cord levels is compared to burst onset in L2 (Fig. 5A) and S4 (Fig. 5B). This shows that there was a progressive delay in onset of burst generation which was observed both in sacral and lumbo-thoracic segments. In both cases, the propagation of motor activity was in a caudal-to-rostral direction. In these series of experiments, we always recorded all lumbar and sacral ventral roots and the positions of the electrodes on the thoracic ventral roots were varied from one experiment to another in order to monitor the entire length of the thoracic cord region to establish the phase diagram as seen in Fig. 7.
To accurately determine motor burst onsets, the raw data were processed in order to obtain averaged measurements of sequential bursts (Fig. 6). Figure 6A presents rectified, integrated recordings from 16 ipsilateral ventral roots during a sequence of locomotion. With L2 ventral root activity as the reference (see above), activity recorded at all 16 electrode positions was partitioned into individual cycles, superimposed (Fig. 6B) and averaged (Fig. 6C). To highlight phase relations between recording locations, averaged cycles were then replicated in a continuous sequence (Fig. 6D). For all traces, the temporal relationships between the burst peaks were then calculated with respect to the reference L2 trace (Fig. 6E), with the burst phase (Φ) being expressed in degrees (0–360 deg).
For individual averaged ventral root traces in each experiment, we established the phase relationship between each recording location and L2, as seen in Fig. 7A, where each point in the polar plots represents the mean phase value obtained in one experiment. The vector direction in each polar plot represents the mean value and its length the coupling strength. Inspection of the polar plots reveals that the phase values were more variable between experiments in the sacral area than in the lumbar or thoracic segments. From the L2 segment there was a progressive phase lag increase that was maximal (180 deg, i.e. complete antiphase) at the S4 segmental level. The lag in the upper thoracic cord (T3 level) reached about 25% (75 deg) of an L2 burst cycle. For the sake of clarity, a global presentation of motor burst propagation was used in which a phase graph was plotted with the segmental level given on the y axis and the associated mean phase values on the x axis (Fig. 7B). Inspection of the curve revealed several distinct propagation zones (delimited by three ovals). In the first region (T), spanning from L2 to the upper thoracic segments, a linear caudo-rostral propagation of motor bursts occurred. A circulo-linear correlation analysis performed by allocating a number from 1 to 22 to each segment (see Methods) revealed a systematic and significant change in the phase value with distance (r = 0.94, P < 0.001) in this region. In the second region (L), spanning L5 and L6, motor activity was out of phase with the activity recorded from L2, and corresponded to the previously observed alternation between ipsilateral flexor and extensor burst units (Cazalets et al. 1992; Cazalets, 2000). In the third region (S), which spanned the sacral segments, the most caudal S4 motor burst was in antiphase with the L2 activity and from this point there was a progressive phase shift up to S1. Correlation analysis showed that this increase in phase value from S4 to S1 was significant (r = 0.99, P < 0.05). The dashed line, fitted from the data points in the sacral area, represents the temporal decay that would have occurred if motor bursts had propagated uniformly throughout the lumbar region. It is likely that due to the concomittent localization of both axial and hindlimb motoneurons at this level, this linear propagation is masked by the massive activation of hindlimb flexor and extensor motor units (see Discussion).
In addition to the relative values provided by the polar plots and phase diagram of Fig. 7, we plotted the mean total propagation time (Fig. 8A, black bars, left-hand y axis) through the lumbo-thoracic spinal cord (L2–T2), the sacral spinal cord (S1–S4) and the associated propagation time between individual segments (Fig. 8A, grey bars, right-hand y axis). The intersegmental delay was much shorter in the thoracic segments (67 ms segment−1 or 0.015 mm ms−1 because each segment measured 1 mm) than in the sacral segments (185 ms segment−1 or 0.005 mm ms−1). Moreover plots of latency versus motor burst period (filled circles in Fig. 8B) for representative segments from the three different zones, revealed that there was a systematic and significant increase in latency with the burst period (correlation analysis between the motor period and the latency, P < 0.01). Slope comparisons indicated that the variation in latency in the thoracic versus sacral segments was significantly different (P < 0.01), with slope being less in the thoracic region (T10 in the present study) than in the lumbar (L5) or sacral (S2) areas. By contrast, the plots of phase value versus period (open squares, Fig. 8B) show that there was no correlation between burst phase and the motor period. On this basis, therefore, it appears that spinal motor networks adjust the intersegmental latency to the ongoing motor period in order to maintain a constant phase relationship of activity along the spinal axis.
Generation of rhythmic motor patterns in the isolated cord
The above data show that coordinated motor patterns with a specific temporal organization are expressed in the isolated spinal cord. This raises two questions: from where does the propagated motor activity originate, and what determines the phase shifts? To address these questions, we performed cord section experiments in order to assess whether the different zones (T, L and S) identified in Fig. 8B have the capacity for rhythmogenesis. In the control recordings from an intact isolated spinal cord as in Fig. 9A, motor bursts were coordinated 1: 1 in all the ventral roots recorded with a burst period of 2.8 ± 0.05 s (Fig. 9C). Right and left alternation occurred in thoracic segments in all experiments where thoracic segments were recorded bilaterally (n = 5). We further performed cord transection experiments (n = 8 preparations), at levels indicated by the vertical dashed lines in Fig. 9B. Rhythmic motor patterns were still expressed in all three separated pieces of spinal cord, albeit with substantial differences in activity (Fig. 9C). As previously reported (Cazalets & Bertrand, 2000a), a sustained rhythmic activity was observed in the isolated lumbar area with a burst period (3.2 ± 0.1 s, n = 30 cycles) similar to that recorded in the intact cord, whereas a slightly slower rhythmic bursting pattern was still observed in the isolated sacral region (3.9 ± 0.2 s, n = 30 cycles). An alternation between the L2 (flexor units) and the L5 (extensor units) persisted. Similarly a left–right alternating motor root pattern also occurred in the isolated thoracic region although at an even slower cycle period (4.85 ± 0.1 s, n = 30 cycles). Such slow and sometimes irregular rhythmic motor activity was observed in the isolated thoracic T2–T12 spinal cord in six out of eight experiments, with bath-application of NMA plus either 5-HT or DA. Although slower and even more irregular, rhythmic activity was recorded in shorter pieces (T5–T12 segments) of the thoracic spinal cord. As bursting properties can be expressed in the various spinal areas including the thoracic cord, it is likely that all segments actively participate in motor wave propagation.
The phase diagram of Fig. 7 indicates that the low lumbar and sacral areas are active in antiphase with bursting recorded in L2. Could this antiphase relationship be due to cross-connections between the two sides of the cord? To answer this question we analysed the effects on phase relationships of splitting spinal cords (n = 6) along their midlines. The traces in Fig. 10 show the rhythmic activity generated under various conditions of longitudinal cord section. In control conditions (Fig. 10A), both L5 and S2 motor bursts were themselves phase coupled, but out-of-phase with L2 motor bursts. In the isolated hemicord (Fig. 10B), L2 ventral root activity remained out of phase with L5 activity (flexor/extensor activity, see above and Cazalets et al. 1992), but S2 now fired concomitantly with L2. In the polar plots of Fig. 10D, the open circles show the phase values and means (thin line) obtained in control experiments at five different lumbar/sacral segmental levels (Fig. 10A). The filled circles and associated dotted lines, respectively, indicate the phase values and mean from all six hemicord experiments (as in Fig. 10B). It is significant that the progressive caudo-rostral shift in phase value (from 170 to 250 deg) observed along the sacral region in control conditions, was abolished in hemi-spinal cords so that the mean phase value was now constant at around 15 deg (Fig. 10Da–e). By contrast, L5 motor bursts remained phase-locked at around 180 deg (Fig. 10Da).
Finally, as the L2/sacral phase opposition appeared to depend on cross-cord coupling, we used partial hemi-sections to determine the level at which this transverse connectivity occurred. For this, the spinal cord was split from sacral until the low lumbar segments (Fig. 10C). In this condition (n = 5 experiments), the L2/sacral activity remained out-of-phase (Fig. 10C), as did L2 activity with the L5 (thick line in Fig. 10Da), but the intersegmental delay in the sacral area was no longer evident (asterisks and associated thick lines in Fig. 10D). The phase graphs in Fig. 10E summarize the data in the three experimental conditions and indicate that different pathways determine the phase relationships in the lumbar and sacral cord regions with the L2/sacral antiphase pattern and intersegmental delay being organized essentially by cross-cord coupling that occurs above the L5 segment.
Discussion
In the present study we have found that locomotion in the newborn rat is accompanied by a strict sequence of undulating trunk and tail movements. Using isolated spinal cord preparations, we present evidence for a central origin of the metachronal motor pattern that drives back muscles, and provide some insights into the organization of the neuronal networks that underlie it. Figure 11 summarizes the data reported here and previously (Cazalets, 2005). In this model, each segmental level is represented by bilateral circles that are highlighted to indicate that some bursting activity can be expressed at the segmental level. The double-headed arrows between circles indicate local intercircuit interactions that contribute to the propagation of motor activity. The horizontal blue lines represent cross connections that occur along the spinal cord, while the red lines denote long propriospinal pathways involved in coordination. The various aspects of this model will be further discussed below.
Organization of spinal cord networks
The exploration of the ability of spinal motor networks to generate rhythmic motor output through section experiments (Figs 9 and 10), revealed that three zones (T, L and S; Figs 7B and 9B) possess the intrinsic capacity for rhythmogenesis. While this was already described for the sacral (Cazalets & Bertrand, 2000a; Lev-Tov & Delvolve, 2000) and lumbar cord areas (Cazalets et al. 1995; Cazalets & Bertrand, 2000a), we demonstrate here that the isolated thoracic spinal cord can also generate coordinated activity with rhythmic left–right alternation. Previous studies (Cowley & Schmidt, 1997; Kremer & Lev-Tov, 1997; Ballion et al. 2001) presented contradictory data. Kremer & Lev-Tov (1997) noted that rhythmic activity (both alternating or non-alternating) or random bursting activity could be recorded in some thoracic segments, but this activity disappeared following a cord section at the low thoracic level. A similar finding was reported by Ballion et al. (2001). On the other hand, Cowley & Schmidt (1997) noted that thoracic segments when isolated could generate rhythmic activity, but their study did not give any indication of right–left phase relationships. The discrepancy between these different studies may be attributable to different experimental conditions and to the fact that a complete systematic study on the rhythm-generating properties of the thoracic spinal cord was not performed.
The rhythmic activity generated by the three different cord areas is qualitatively different, with the thoracic (Fig. 9) and sacral (Fig. 10; Cazalets & Bertrand, 2000a) regions generating slower rhythms than the lumbar area. The model of Fig. 11 correponds closely to that recently proposed by Magnuson et al. 2005), who studied the functional consequences of a spinal cord contusion at the lumbar level in adult rats. According to their model, rhythmic hindlimb locomotor activity elicited physiologically by stimulation of descending commands (see green symbols in Fig. 11) or by drug application is dominated by circuitry in the L1 and L2 segments of the intact spinal cord. In addition, the finding that further reduced preparations (e.g. isolated L3–L6) can exhibit drug-induced bursting (Kjaerulff & Kiehn, 1996; Cowley & Schmidt, 1997; Kremer & Lev-Tov, 1997; Magnuson & Trinder, 1997; Bonnot et al. 1998) demonstrates that neuronal circuitry capable of rhythmogenesis (‘rhythmic elements’) is distributed throughout the lumbar enlargement. This in turn led to the proposition (Magnuson et al. 2005) that the CPG located in the rostral part of the lumbar enlargement provides rhythmic locomotor output to neighbouring rhythmic elements that in turn transfers and modulates the output to segmental motor neurons in the adult animal. The model that we propose in Fig. 11 expands on this view to incorporate the entire spinal cord. In an intact spinal cord, the lumbar area probably imposes its own timing on the thoracic spinal cord generators, as it does for the sacral segmental rhythm (Cazalets & Bertrand, 2000a). The pre-eminent role of lumbar locomotor networks was also postulated in a recent study on the coordination between the motor circuitry controlling the forelimbs and hindlimbs (Juvin et al. 2005). In various studies (Kjaerulff & Kiehn, 1996; Cowley & Schmidt, 1997; Kremer & Lev-Tov, 1997; Bonnot et al. 1998), bursting has been found to occur in isolated pieces of the spinal cord under pharmacological activation. It cannot be excluded that some of these observations may in fact be attributable to bursting occurring in axial motoneurons and that the rhythmogenic ability of the various spinal compartments controlling axial musculature may have contributed to different interpretations concerning the organization of hindlimb locomotor networks (Cazalets, 2000; Cazalets & Bertrand, 2000b).
An interesting feature of the phase-lag diagram of Fig. 7B is the temporal nature of motor burst propagation in the sacral area. In the intact spinal cord, sacral activity propagates caudo-rostrally, in a phase-opposite sequence to the ipsilateral L2 segments. Our cord section experiments also revealed important insights into the underlying synaptic organization (see Fig. 11). First, they indicate that rhythm generation does not require cross-cord connections (horizontal blue lines in Fig. 11) as also demonstrated in the Xenopus embryo (Soffe, 1989) and recently in the lamprey (Cangiano & Grillner, 2003). The vertical blue lines in Fig. 11 indicate that the reciprocal organization of ipsilateral flexor (L2) and extensor (L5) activity bursting (Cazalets et al. 1992) also does not rely on cross-cord connections but is completely organized ipsilaterally (Kudo & Yamada, 1987). By contrast, the ipsilateral L2/sacral antiphase pattern relies on contralaterally projecting connections (vertical brown lines in Fig. 11) because in a complete hemicord preparation (Fig. 10B), sacral bursting activity switched to an in-phase motor pattern with L2 bursting. It is surprising that when the spinal cord was split only from the sacral to L5 segments, the antiphase pattern persisted, although the intersegmental latency was dramatically reduced to produce almost single phase sacral motor activity (Fig. 10C). This suggests that different cross-cord connections are involved both in setting the timing of intersegmental phase shifts as well as in the reciprocal bilateral organization. Commissural interneurons that are active during locomotion have been shown to play a major role in setting the contralateral relationships in various species (Buchanan, 1996; Soffe et al. 2001; Butt & Kiehn, 2003; Jankowska et al. 2005; Zhong et al. 2006). In neonatal rat, however, the involvement of different classes of these commissural interneurons (both segmental and intersegmental) with their firing patterns occurring at all phases of the locomotor cycle has led to the suggestion that they may also be involved in other aspects of locomotor rhythm generation (Butt & Kiehn, 2003; Zhong et al. 2006). The results of our study suggest that one of these roles is in setting the phase delay between segments along the spinal cord as has also been suggested in Xenopus embryo cord (Green & Soffe, 1998). This should be further investigated by determining whether the anatomical distribution of commissural interneurons along the spinal cord is commensurate with such a specific role.
Temporal organization of motor activity in the spinal cord
Our phase analysis revealed several discrete zones along the spinal axis in which motor activity propagates sequentially. This demonstration was possible because we recorded from neighbouring ventral roots, which allowed the establishment of the phase relationships between adjacent segments. In lower vertebrates, such as lamprey and frog tadpole, axial body movements involved in locomotion result from a rostro-caudal metachronal wave of motoneuron discharge that gives rise to muscle contractions with appropriate phase delay in adjacent body segments (Cohen, 1987; Matsushima & Grillner, 1992; Roberts et al. 1998; McClellan & Hagevik, 1999; Grillner & Wallen, 2002). Similarly, our in vitro data strongly suggest that the metachronal changes observed in trunk bending (Fig. 2) rely on intrinsic spinal cord properties. In a previous study performed under particular pharmacological conditions (strychnine and bicuculline), it was concluded that the spinal cord contains axial propriospinal pathways that may be involved in intersegmental coordination (Cazalets, 2005). Although this system may account for the phase lag observed here, the reported speed of longitudinal propagation in the presence of strychnine and bicuculline was much higher (Cazalets, 2005). In the present study without blockade of inhibitory synaptic transmission, the mean propagation of locomotor bursts along the thoracic cord was 800 ms (Fig. 8), indicating that propagation between each segment may also involve local interactions that effectively slow propagation. If long spinal fibres that distribute to each segmental level were solely involved (red lines in Fig. 11), presumably propagation would be faster because it would only take the time for spike conduction along the fibres, as well as synaptic delay (i.e. about 50 ms). As suggested in previous work (Cazalets, 2005), and as also shown in the lamprey (McClellan & Hagevik, 1999; Miller & Sigvardt, 2000), it is therefore likely that both long intersegmental neuronal tracts and local circuit interactions between adjacent segmental oscillators are involved in coupling. Furthermore, the fact that both thoracic and sacral areas can express intrinsic bursting properties (Fig. 9), albeit weaker than the lumbar region, may also contribute to slowing motor propagation. To date the exact nature of the propriospinal systems involved in the coordinating process is unknown. In the lamprey, the excitatory interneurons that traverse from two to six segments have been suggested to account for intersegmental coordination (Dale, 1986). Juvin et al. (2005) have suggested that interactions between lumbar and cervical cord regions may be mediated by asymmetric propriospinal pathways arising in the lumbar area and relaying through the thoracic level caudo-rostrally. In the newborn rat, a class of ipsilateral excitatory interneuron (Kiehn & Butt, 2003) could also be partly responsible for the segment–segment interactions. In addition, we have previously shown that long propriospinal pathways probably exist that participate in motor burst propagation (Cazalets, 2005).
Functional implications of metachronal propagation
Body displacement in elongated animals such as tadpoles (Roberts et al. 1998; Soffe et al. 2001; Tunstall et al. 2002), lamprey (Cohen, 1987; McClellan & Hagevik, 1999; Miller & Sigvardt, 2000; Grillner & Wallen, 2002) and snakes (Gasc et al. 1989) are driven by trunk muscle contractions that occur sequentially along the body length. Rhythmic activation of back muscles during locomotion has also been reported to occur in various quadrupeds such as the cat (Carlson et al. 1979; Zomlefer et al. 1984), adult rat (Geisler & Gramsbergen, 1998), newt (Delvolve et al. 1997) and human (Thorstensson et al. 1982). By using retrograde staining, we determined here that the motoneuron pools that innervate the trunk and tail muscles are distributed along the spinal cord, in a manner that matches the distribution of muscles along the body. These results add to those of a previous study in adult rats (Brink et al. 1979), in which the central locality of motoneurons innervating muscles that specifically intervene in lordosis were determined. Moreover our in vitro electrophysiological data appear to support the physiological involvement of trunk muscle activity in locomotion. Although the thoracic ventral roots innervate other body regions, particularly the respiratory (Monteau & Hilaire, 1991) and abdominal musculature (Iscoe, 1998), because the spinal cord was transected at the T1 level in our experiment, respiratory motoneurons were not involved in the recorded rhythmic patterns (Monteau & Hilaire, 1991). Furthermore, an interesting finding is that back muscle motoneurons are colocalized with the hindlimb muscle motoneurons at the lumbar spinal cord level (Fig. 3Bb; see also Nicolopoulos). Thus, the possibility arises that segmental lumbar output may concomitantly reflect different sources of rhythmogenesis: the lumbar hindlimb generators themselves and the networks responsible for back muscle activation. Alternatively, the pre-eminent activation of hindlimb motoneurons by the lumbar generators may mask the activation of axial motoneurons in this cord region.
In the present study, we have found evidence suggesting that trunk curvature observed during locomotion is due to a sequential propagation of motor output-related activity along the spinal cord in newborn rat. Our kinematic data provide evidence for a caudo-rostral progression in the maximum angular deviation that occurs with a specific phase relationship (Fig. 2D). Our in vitro data match observations from the intact animal because a comparable caudo-rostral propagation of motor bursts was observed (Fig. 7B). Another interesting observation is that the intersegmental latency in the sacral area is longer than in the thoracic region. This may reflect the fact that the central nervous system differentially controls spinal curvature, which is greater in the lower back (Fig. 1A; compare angle variations in Fig. 2Ba and 2Bb and c, Fig. 2D) and the tail. Such an organization would therefore seem to support the dynamic control of posture by the performance of fluent movements during locomotion. Moreover our data suggest that the networks responsible for metachronal propagation of motor patterns during locomotion may correspond to those observed in invertebrates or lower vertebrates, and thus are highly conserved.
Acknowledgements
The authors thank S. Bertrand J. Simmers and N. Mellen for their valuable comments on the manuscript and for correcting the English. We also thank R. Nargeot for help in statistical analysis. This work was funded in part by the Ministère de la Recherche (ACI Neuroscience; Action Concertée Incitative (ACI) Plateformes Technologiques 032645), the Région Aquitaine, the Direction Générale de l'Armement (DGA) (Ministère de la Défense no. 0334045) and by the Institut pour la Recherche sur la Moelle Epinière (I.R.M.E.).
Footnotes
-
(Received 26 June 2006; accepted after revision 18 December 2006; first published online 21 December 2006)
- 2007 The Authors. Journal compilation © 2007 The Physiological Society
References
Figure 1. Two-dimensional kinematic analysis of trunk movements in newborn rat A, three frames showing trunk bending during a single step. The cursors on the bars below each video frame indicate the time at which each frame occurred during the step cycle (reference limb is right hindlimb). The horizontal bar indicates the swing and stance duration for the considered step. Ba, drawing indicating the position of the points analysed for the trajectory as C. Bb, trajectories of the corresponding dots in Ba following manual tracking during 4 s of walking. C, stick diagram reconstruction of trunk (black lines) and tail (blue lines) movements. For the sake of clarity, traces were staggered vertically in order to avoid overlaps. Time is indicated besides the traces. Coloured dots on the first swing and stance plots correspond to those in Ba. The arrows indicate the direction of movement. H, head; T, tail.
Figure 2. Analysis of trunk curvature during stepping A, drawing of the rat with dots indicating the analysed segments (S, sacral; H, head; T, tail; L, left; R, right). B, plots of angular change from lower (Ba) through to the upper back (Be) during two normalized cycles. Each curve represents the mean angular changes calculated from five steps collected from five different animals. For clarity, two successive normalized cycles are presented. Dashed lines indicate the confidence interval (± 2 s.e.m.). The filled dots above each plot present the analysed angle. The horizontal line at 180 deg indicates the time at which there is no bending. C, mean angular change reflecting complete trunk bending. D, plot of maximum peak amplitude on both sides. The y axis indicates the angle number considered in Ba–e. In B and C the horizontal bar below each plot represents two successive normalized cycles, while in D it represents one normalized cycle.
Figure 3. Retrograde motoneuron labelling using cholera toxin subunit B Alexa Fluor 488 conjugate A, drawing of the newborn rat indicating the sites of injection. B, photographs of labelled motoneurons from tail muscles (Ba), lower back muscles (Bb) and mid-back musces (Bc). The bar under Bb is the scale. G, gastrocnemius muscle-labelled motoneurons; B, back muscle-labelled motoneurons. C, summary drawing indicating the respective location of labelled motoneurons following injections at the various sites. Each numbered dot in A corresponds to a numbered motor column in C. Each site was injected in at least two animals.
Figure 4. Simultaneous recordings of 16 ventral roots during locomotor-like activity A, drawing of the preparation with dots indicating electrode locations. B, sample recordings of thoracic (T), lumbar (L), sacral (S) and coccygeal (Co) ventral roots in the absence of activity (Ba) or during locomotor-like activity induced by bath-application of 5-HT and N-methyl-d,l-aspartate (Bb).
Figure 5. Caudo-rostral propagation of motor bursts in the isolated spinal cord Left panel in A and B shows rectified, integrated ventral root recordings. The vertical dashed lines indicate the onsets of motor bursts in L2 (A) and S4 (B), respectively. The oblique continuous line was drawn through the onsets of motor bursts in all recorded ventral roots. The right panels show the corresponding recording sites with the arrows indicating the direction of motor propagation.
Figure 6. Data processing and analysis of locomotor-related activity A, rectified integrated simultaneous recordings of 16 ventral roots. B, taking L2 ventral root activity as a reference, the traces associated with all cycles were individualized, normalized to the same data point number and superimposed. C, average of the superimposed cycles presented in B. D, the average cycle was replicated in a continuous sequence to reveal the rhythmic nature of the activity. E, definition of the measured temporal parameters.
Figure 7. Analysis of the phase-lag between ventral root activity in the isolated spinal cord A, circular plots presenting the phase value for six ventral roots (indicated in boxes) relative to the L2 ventral root. Each point is the phase value collected from one experiment. The vector direction (thick line in each plot) represents the mean phase value and its length depends on the strength coupling. B, phase plot of mean values cumulated over the various experiments. Each spinal segment (indicated on y axis) is associated with its corresponding phase lag value (x axis). Error bar is s.e.m. The plot reveals three main phase-lag areas that are denoted by ellipses (T, thoracic; L, lumbar; S, sacral).
Figure 8. Temporal analysis of motor burst propagation A, the bar plot presents the total propagation time (black bars, left y axis) and speed (grey bars, right y axis) from L2 to T2 and from S1 to S4 segment. Propagation speed was calculated by dividing the time for total propagation by the length of the considered cord assuming that each segment measured 1 mm. B, correlation analysis of motor period versus latency (•). Slopes obtained with linear regression for S2 (0.8 ± 0.1), L5 (0.7 ± 0.07) and T10 (0.3 ± 0.1). Slopes were significantly different (P < 0.01). Correlation analysis of phase versus period (□).
Figure 9. Generation of rhythmic activity in various spinal cord areas A, control locomotor-like activity was induced by bath-application of N-methyl-d,l-aspartate (NMA) plus 5-HT. Dots on the drawing indicate the recording sites. B, transections were performed at the T12–T13 and L6–S1 junctions. After transection, a new bath-application of NMA/5-HT re-induced rhythmic activity in all three isolated regions of the spinal cord. A and B are from the same experiment. C, plot of the mean cycle period before and after sectioning in each isolated part of the spinal cord (s.e.m., 30 cycles).
Figure 10. Effects of longitudinal cord sectioning on motor burst phase relationship A, top, schematic diagram of the experimental design. Control locomotor-like activity was induced by bath-application of N-methyl-d,l-aspartate plus 5-HT. Vertical dashed lines in average traces indicate one cycle of activity. The L5 and the S2 ventral roots fired out-of-phase with the L2 ventral root. B, rhythmic bursting recorded in a hemicord (see top drawing). The L5 and L2 ventral roots now fired out-of-phase while the S2 and L2 ventral roots were now in phase. C, rhythmic bursting recorded in a partially split spinal cord (see top drawing). As in the intact cord, the L5 and S2 ventral roots fired out-of-phase with the L2 ventral root. In A–C traces are replicated averaged cycles as presented in Fig. 6D. D, circular plots presenting the mean phase value in intact (thin vectors), hemi-spinal cord (dotted vectors) and partially split spinal cord (thick vectors), showing the mean values for the intact spinal cord (○), the hemi-spinal cord (•) and the partially split spinal cord (*). E, phase plot of mean value (in deg) cumulated over the various experiments in control (Ea), hemi-spinal cord (Eb) and partially split (Ec) conditions. Each spinal segment (indicated on y axis) was associated with its corresponding phase lag value (x axis). Error bar is s.e.m.
Figure 11. Diagram summarizing intersegmental spinal cord interactions Three cord regions (thoracic, lumbar and sacral) are distinguished based on bursting abilities and temporal relationships. Circles, rhythmic segmental elements; yellow square, rostral lumbar CPG; vertical red lines, long propriospinal pathways; horizontal blue lines, cross connections; vertical blue lines, flexor/extensor connections; brown lines, cross connections involved in L2/sacral phase relationships; green lines, descending commands; double-headed arrows, local coupling.
