Noncholinergic excitatory actions of motoneurons in the neonatal mammalian spinal cord

Abstract

Mammalian spinal motoneurons are considered to be output elements of the spinal cord that generate exclusively cholinergic actions on Renshaw cells, their intraspinal synaptic targets. Here, we show that antidromic stimulation of motor axons evokes depolarizing monosynaptic potentials in Renshaw cells that are depressed, but not abolished, by cholinergic antagonists. This residual potential was abolished by 2-amino-5-phosphonovaleric acid and 6-cyano-7-nitroquinoxaline-2,3-dione. In the presence of cholinergic antagonists, motor axon stimulation triggered locomotor-like activity that was blocked by 2-amino-5-phosphonovaleric acid. Some cholinergic motoneuronal terminals on both Renshaw cells and motoneurons were enriched in glutamate, but none expressed vesicular glutamate transporters. Our results raise the possibility that motoneurons release an excitatory amino acid in addition to acetylcholine and that they may be more directly involved in the genesis of mammalian locomotion than previously believed.

Mammalian spinal motoneurons innervate the extrafusal and intrafusal fibers of skeletal muscles (1) and send monosynaptic projections to spinal inhibitory interneurons known as Renshaw cells (2). They are thought to exert their peripheral and central actions exclusively through cholinergic mechanisms and function solely as output elements of the spinal cord. However, several recent reports challenge this idea. For example, in developing chick and mouse embryos, spontaneous episodes of rhythmic discharge can be initiated by motoneuron discharge (3, 4), and Xenopus tadpole motoneurons are interconnected by chemical and electrical synapses and may project onto spinal interneurons involved in swimming (5, 6). In addition, some studies suggest that excitatory amino acid neurotransmitters (EAAs) might be coreleased with acetylcholine from motoneuron terminals (7-10), although the evidence for such corelease is controversial. Glutamate or glutamate transporter immunoreactivities have been identified at the mammalian neuromuscular junction in some studies (7, 11, 12) but not in others (9, 13). A similar disparity exists for studies examining vesicular glutamate transporter immunoreactivity (VGLUT-IR) in the spinal terminals of motoneurons onto Renshaw cells (9, 14) or VGLUT mRNA expression in motoneurons (10, 14-17). Finally, several papers have reported the presence and physiological actions of postsynaptic ionotroptic (18, 19) and metabotropic (20) glutamate receptors at the vertebrate neuromuscular junction, although the origin of the released EAA was not analyzed in detail.

Currently, no direct physiological evidence supports the release of fast neurotransmitters other than acetylcholine from the central terminals of motoneurons onto Renshaw cells. However, it is known that the monosynaptic depolarizing potentials evoked in Renshaw cells by stimulation of motor axons are incompletely blocked by cholinergic antagonists (21-23), raising the possibility that motoneurons might exert noncholinergic excitatory actions on these cells. Furthermore, in the developing spinal cord of chick embryos, ventral root potentials evoked by stimulation of motor axons are depressed by both glutamatergic and cholinergic antagonists (3, 24), consistent with the central release of an EAA from avian motoneurons. Recently, glutamate and acetylcholine corelease was demonstrated at the synaptic terminals of spinal interneurons in the developing Xenopus embryo (25), although motoneurons in this preparation appear to be solely cholinergic (5).

In this paper, we investigate whether motoneurons can exert noncholinergic excitatory effects on Renshaw cells of the neonatal mouse spinal cord and examine their potential involvement in locomotor-like activity. Preliminary results have been presented.§

Methods

Further experimental details are given in Supporting Text and Figs. 4-9, which are published as supporting information on the PNAS web site.

Intracellular Recordings from Renshaw Interneurons and Extracellular Recordings from Ventral Roots to Monitor Locomotor-Like Activity. We used whole-cell current clamp to record in vitro from neonatal (P2-P4) mouse Renshaw cells using the isolated spinal cord. Suction electrodes were used to either stimulate or record from ventral roots, dorsal roots, the sciatic nerve, or the ventro-lateral funiculus.

Acetylcholine-induced currents were recorded from Renshaw cells by using whole-cell voltage clamp in neonatal rat spinal cord slices (300 μm). Renshaw cells were identified after the recordings by the presence of vesicular acetylcholine transporter immunoreactive (VAChT-IR) terminals on their soma and/or dendrites. Acetylcholine was injected with the aid of a Picospitzer.

Immunohistochemistry and Image Analysis Using Confocal Microscopy. Spinal cords were either removed from the animal after transcardial perfusion with 4% paraformaldehyde, or if fluorescent dyes were used for tracing purposes (Texas red dextran, 10,000 MW, Cascade blue dextran 10,000 MW, Molecular Probes), the spinal cords were immersion fixed in 4% paraformaldehyde for 3-4 h. The cords were subsequently processed for immunohistochemistry by using cryostat (20-40 μm) or Vibratome (50-70 μm) sections. The primary antibodies used in this study were: VAChT (guinea pig, 1:1,000; goat, 1:2,000), choline acetyltransferase (goat, 1:100), VGLUT1 (guinea pig, 1:1,000), VGLUT2 (guinea pig, 1:2,000), and VGLUT3 (guinea pig, 1:1,000) obtained from Chemicon (Temecula, CA); glutamate (rabbit, 1:500-1:1,000) obtained from Alpha Diagnostics (San Antonio, TX); calbindin D28K (rabbit, 1:2,000 or mouse, 1:500) obtained from Swant (Bellinzona, Switzerland); and VGLUT3 (rabbit, 1:2,000) kindly donated by R. H. Edwards (University of California School of Medicine, San Francisco). Immunoreactive sites were revealed with several fluorochrome-conjugated (AMCA, FITC, Rhod-RX, or Cy5) secondary antibodies (dilution: 1:50 for FITC, RRX, Cy3, and Cy5 and 1:10 for AMCA, Jackson Laboratories). All images were obtained confocally by using either a four-channel two-photon laser (Chameleon, Coherent) 510META (Zeiss, Germany) confocal microscope equipped with three single photon lasers (488, 543 and 650 nm lasers), or a dual-channel Olympus FX Fluoview system or triple-channel Leica TCS system.

Results

Synaptic Potentials Evoked in Renshaw Cells by Stimulation of Motor Axons. Renshaw cells were identified by their ventral location in the spinal cord (Fig. 4 a and b), their high frequency of firing (range, 83-122 Hz; mean ± SD, 102 ± 14 Hz, n = 8) in response to a single ventral root (or sciatic nerve) stimulus (Fig. 4c), and the presence of a monosynaptic potential after ventral root stimulation (Fig. 4 d and e). The latency of the potential evoked in Renshaw cells was 3.6 ± 0.7ms(n = 12 cells) after ventral root stimulation and 5.6 ± 0.6ms(n = 5 cells) after sciatic nerve stimulation (dorsal roots were cut) at a distance of ≈7-8 mm from the spinal cord.

Bath application of a mixture of nicotinic (50 μM mecamylamine and 50 μM dihydro-β-erythroidine) and muscarinic (5 μM atropine) cholinergic antagonists depressed, but did not abolish, the monosynaptic potential evoked in Renshaw cells by ventral root or sciatic nerve stimulation (Fig. 1 a-c). In three cells, the amplitude of the residual potential was 1.6 ± 1.0 mV or 31.2% of the control response measured 40 min after application of the cholinergic antagonists. The remaining synaptic potential was almost completely blocked by bath application of the NMDA and α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptor antagonists 2-amino-5-phosphonovaleric acid (APV, 100 μM) and 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX, 10 μM). The mean (n = 3 cells) latency of the cholinergic-resistant potential (3.1 ± 0.6 ms) was not significantly different from that of the control potential (3.0 ± 0.6 ms) suggesting that it was also mediated monosynaptically.

Fig. 1.

Synaptic potentials evoked in Renshaw cells by stimulation of motor axons are mediated by cholinergic, NMDA, and AMPA/kainate receptors. (a) Bath application of cholinergic receptor blockers (50 μM mecamylamine, 50 μM dihydro-β-erythroidine, and 5 μM atropine) resulted in an 82% reduction of the amplitude of the evoked synaptic potential. APV (100 μM) and CNQX (10 μM) abolished the remaining response. (Inset) The superimposed traces at a faster sweep speed to illustrate the unchanged latency (arrow) after cholinergic blockade. (b) During cholinergic blockade, the amplitude of the VR-evoked potential showed no significant changes 20 and 40 min after drug application (ANOVA on ranks), indicating that the antagonist had attained its maximum effect. Addition of CNQX and APV further reduced the amplitude within 5 min of exposure and abolished it by 20 min (P < 0.05, ANOVA on ranks; ^*, P < 0.05 compared to control; ‡, P < 0.05 compared to last cholinergic blockade time point; ANOVA on ranks). (c) VR-evoked subthreshold responses recorded from another Renshaw cell demonstrating the presence of NMDA and AMPA/kainate receptor activation. In this cell, cholinergic receptor blockade (35 min in mecamylamine, dihydro-β-erythroidine, and atropine) reduced the response by ≈70%. Exposure to CNQX (15 min) reduced it further by ≈49%, and APV (5 min) abolished the response. Washout in cholinergic blockers resulted in a recovery that reached 25% of the initial control amplitude (20 min). (d) Recordings from a Renshaw cell (identified by sciatic nerve stimulation), in which the glutamatergic antagonists were first applied. Application of CNQX and APV for 25 min resulted in a 34% reduction in the response amplitude. The response fully recovered after 22-min washout. Subsequent exposure to cholinergic blockers resulted in an 82% depression in the amplitude of the potential. To confirm the effect and specificity of the antagonists, we also monitored responses evoked by stimulation of the dorsal roots (lowest traces in d). Note the abolition of these responses under APV and CNQX and the lack of effect of cholinergic blockers. (Insets) The responses of another Renshaw cell to suprathreshold stimulation under control conditions, during bath application of APV/CNQX and under cholinergic blockade showing that either pharmacologically isolated component of the potential can evoke action potentials when the cell is near its resting potential (-51mV). (Scale bars, 20 mV and 40 ms.) (e) The impedance of Renshaw cells remains largely unchanged throughout the pharmacological protocol (n = 3; not significant, one-way ANOVA). The impedance was calculated from the steady-state voltage response to a small negative current pulse (as shown in d). (f) Percentage depression in the amplitude of VR-evoked potentials (expressed with respect to control amplitude) under NMDA/AMPA/kainate receptor blockade, washout, and then cholinergic receptor blockade averaged from six cells. (Right) A similar cholinergic component in the responses of three Renshaw cells challenged only with the nicotinic and muscarinic receptor antagonists. For further details, see Supporting Text.

To ensure that the cholinergic-resistant potential was glutamatergic and not because of incomplete blockade, we applied the NMDA and AMPA receptor antagonists first. In the presence of APV and CNQX, ventral root-evoked potentials in Renshaw cells were depressed by 1.9 ± 0.8 mV (n = 6 cells) representing 31.7 ± 5.3% of the control response (Fig. 1f), whereas the cholinergic component (tested after washout) was 78.6 ± 12.5% of the control response. The percentage depression of the evoked potential when cholinergic antagonists were applied first was similar (78.8 ± 8.9% of control; n = 3 cells).

The variations of input impedance and membrane potential were not significantly different during the pharmacology, indicating that they cannot be responsible for the changes in the amplitude of the evoked potentials. In addition, we found that the current evoked in Renshaw cells (identified by during voltage-clamp recordings by the characteristics of their spontaneous synaptic currents; ref. 26) by local application of acetylcholine was completely blocked by cholinergic antagonists, indicating that acetylcholine does not activate NMDA or AMPA/kainate receptors (see Supporting Text).

Locomotor-Like Activity Can Be Induced by Ventral Root Stimulation. In 13 experiments, we found that a train of stimuli (4 Hz for 10 s) applied to a ventral root or the sciatic nerve (DR cut) initiated an episode of locomotor-like activity. The rhythm was triggered preferentially by stimulation of the L5 or L6 ventral roots. The evoked activity comprised 5-18 cycles (cycle period 0.6 ± 0.1 s; n = 8) in which the discharge and slow potentials alternated between the left and right rostral (L1 or L2) ventral roots (Fig. 2a1) and between the ipsilateral rostral and caudal lumbar roots. The rhythmic cycles were superimposed on a tonic depolarization that lasted for the duration of the train. In the same experiments, trains of stimuli were also applied to the cut dorsal roots L5 or L6 (Fig. 2b1). As previously reported (27, 28), this type of stimulation also evokes locomotor-like activity, and it served as a control for the subsequent pharmacological experiments.

Fig. 2.

Activation of locomotor-like activity by stimulation of motor axons. (a1-a4) Locomotor-like activity recorded from the left and right ventral roots (L1 segment) in an isolated spinal cord of a P3 mouse in vitro. The activity was evoked by a train of stimuli (4 Hz, 0.2-ms pulse width, 10-s duration) applied to the sciatic nerve (a1-a4; sciatic nerve stimulation/dorsal roots cut; stimulus intensity, 50 μA) or to the cut dorsal roots (b1-b4; DR-L6 stimulation; stimulus intensity, 25 μA). The evoked activity comprised rhythmic discharges that alternated between left and right lumbar ventral roots (L1 left and right, see Supporting Text) and between rostral and caudal ventral roots (L1 and L5 right, not shown). (a1 and b1) Control recordings. (a2 and b2) Recordings obtained 1 h after blockade of cholinergic receptors (50 μM mecamylamine, 50 μM dihydro-β-erythroidine, and 5 μM atropine) still show an alternating rhythm. (a3 and b3) Bath application of 100 μM APV abolished the rhythmic activity and reduced the slow synaptic depolarization evoked by either dorsal or ventral root trains. (a4 and b4) The locomotor-like activity reappeared in response to either sciatic nerve or dorsal root stimulation after washout of all antagonists (recordings made 1 h after start of washout). Stimulus artefacts have been erased for clarity. (c1 and c2) Two cycles of rhythmic activity, expanded from the dotted boxes shown in a1 and a2, illustrate the unchanged phase relations of the ventral root discharge (L1 right, L1 left, L5 right) before (c1, control) and after 1 h of cholinergic blockade (c2). Arrowheads indicate single stimuli. (d) The upper two traces compare the potentials evoked in the right L5 ventral root by trains applied to the sciatic nerve (black) and the dorsal root (gray). Only the first four stimuli are shown. (e) The same responses as in d expanded further and superimposed. Dorsal root stimulation resulted in a short latency monosynaptic (arrow in gray trace) and longer latency polysynaptic potentials. In contrast, sciatic nerve stimulation resulted in a long latency potential after the second and subsequent stimuli (double arrow in black trace).

We found that motor axon stimulation still evoked locomotor-like activity during blockade of cholinergic receptors (≈1 h of bath-application). Under these conditions, the phase relations between the discharge and slow potentials of the left and right and the ipsilateral rostral and caudal ventral roots were preserved (Fig. 2 a2 and c2). The amplitude of the slow potential was similar to control, but motoneuron discharge was clearly reduced. Application of APV (100 μM) abolished all rhythmic depolarizations from both ventral and dorsal root stimulation within 5-10 min of application (Fig. 2a3). The effect of APV was the same whether it was given alone or in the presence of the cholinergic antagonists. During the ventral root train, the buildup of slow-depolarizing events still occurred, although it began later (third to fifth stimulus) and the amplitude of the evoked tonic depolarization was significantly reduced (0.8 ± 0.1 mV, control vs. 0.5 ± 0.0 mV under APV alone or in the presence of the cholinergic antagonists and APV; P < 0.0005, n = 5). The delay of the onset, time to peak, and amplitude of the slow depolarizing events evoked by motor axon stimulation were not significantly altered in the presence of APV.

The ability of motoneuron stimulation to generate locomotor-like activity recovered fully or partially after washout, depending on whether APV had been applied alone or together with the cholinergic blockers. The locomotor-like activity evoked by dorsal root stimulation was affected in a similar manner to the motoneuron-evoked activity during drug application and washout.

To establish whether activation of the locomotor network by motor axon stimulation depended on functional gap junctions between motoneurons and interneurons, we applied the gap junction blocker carbenoxolone (100 μM) in four experiments. In the presence of carbenoxolone, the locomotor activity induced by either dorsal or sciatic nerve (or ventral root) stimulation gradually deteriorated over 1-2 h. However, alternating locomotor-like activity could still be evoked by either stimulus for up to 45 min in the presence of the drug. In the spinal cord of the neonatal rat, gap junctions between motoneurons have been reported to be blocked after 20 min in carbenoxolone, and prolonged exposure to carben-oxolone causes drug-induced locomotor-like activity to deteriorate (29). We found that, even when locomotor-like activity was no longer expressed (after 1-2 h in carbenoxolone), ventral root stimulation still produced a tonic slow depolarization on which the long-latency potentials were present, indicating that these evoked effects are unlikely to be the result of electrical coupling between motoneurons and spinal interneurons.

Using fluorescent labeling methods, we were unable to identify any ventral root axons originating from the dorsal root ganglion suggesting, in contrast to some earlier reports (30, 31), that few (if any) sensory fibers were present in the L5 and L6 ventral roots (Fig. 6). Thus, it is unlikely that stimulation of glutamatergic ventral root afferents can explain our findings.

Immunocytochemistry of Intraspinal Motoneuron Terminals for Glutamate and VGLUTs. In two P7 and one P10 animal, we found that a subset of putative motoneuron synaptic terminals (expressing both VAChT and retrograde labeling) in contact with either Renshaw cells or motoneurons (32) were enriched in glutamate (Fig. 3). To quantify the glutamate-immunofluorescence in motoneuron synaptic terminals, we estimated the average pixel intensity of glutamate immunofluorescence within VAChT-IR and retrogradely labeled motor axon boutons and compared it to a control region immediately surrounding the terminal. The difference in immunofluorescence intensity between the boutons and surrounding neuropil was normalized to the standard deviation of the control region (see Supporting Text). A similar analysis was performed on glutamatergic primary afferent terminals identified by anterograde labeling from the dorsal roots and expression of VGLUT1 (33). We found that the average glutamate-IR inside motoneuron terminals was 4.4 ± 14.1 SD above the intensity of the surrounding neuropil. By contrast, this difference was 40.2 ± 50.7 SD for primary afferent synaptic boutons (Fig. 7). Because the value for motoneurons was obtained for all putative motoneuron terminals, it included many without elevated glutamate-IR. Nevertheless, the distribution of glutamate-immunofluorescence levels inside motor axons terminals was shifted toward higher values compared to a population of similarly sized regions randomly distributed in the neuropil. These findings suggest that motoneuron synaptic boutons are enriched with glutamate-IR, although to a lesser extent than the VGLUT1-IR terminals from dorsal root afferents.

Fig. 3.

Glutamate is enriched in some motoneuron synaptic boutons contacting Renshaw cells and motoneurons. (a1-a5) Confocal images of motoneuron varicosities apposed to the dendrite of a calbindin-IR (a1, white) neuron in ventral lamina VII, presumed to be a Renshaw cell. The motoneurons and motor axons were retrogradely labeled with Texas red dextran (a3, red) applied to the L5 ventral root (vr), and the sections were also immunolabeled for VAChT (blue; AMCA) and glutamate (green; FITC). (a1) Overlay of all four labels. (a2) VAChT- and calbindin-IR. (a3) Axon collateral terminals from motoneurons labeled retrogradely with Texas Red and calbindin-IR. (a4) Glutamate and calbindin-IR (superimposed on the calbindin-IR dendrite enlarged from the box in a1). Five of the varicosities exhibited both VAChT-IR and the retrograde label, indicating they were motoneuron synaptic terminals. Three of these contained elevated glutamate-IR as shown in line scan profile of the fluorescence intensity along the calbindin-IR dendrite (a5, terminals 1, 3, and 5; see Inset for the line profile). The peak intensity of the glutamate fluorescence coincides with that of VAChT and Texas red dextran for terminals 1, 3, and 5, whereas glutamate-IR of terminals 2 and 4 is absent. (b1-b4) Putative synaptic contacts between motoneurons were analyzed by labeling two adjacent ventral roots (L4 and L5) with two different fluorochromes: Cascade blue dextran (blue) applied to vr-L5 and Texas red dextran (red) applied to the adjacent vr-L4. VAChT-IR varicosities (b1, white) were found apposed to motoneuron somata and dendrites (blue cell). Some of these were derived from other motoneurons because they were also labeled for the retrograde tracer applied to the adjacent ventral root (b2, red), and some were enriched in glutamate (b3, green). (b1-b3 Insets) The varicosity indicated with an arrow at a higher magnification. A line scan profile for all four fluorochromes (b4 and Inset) shows colocalization of VAChT, Texas red dextran (motoneuron collateral from L4) and glutamate. Note that glutamate is also elevated in the motoneuron soma. (c1-c4) Rendered 3D reconstructions of the varicosity on the motoneuron shown in b4 Inset. Note the colocalization of the three fluorochromes in the presynaptic varicosity (arrow in c1-c3) and in the overlay (c4).

We then examined the expression of the three currently known vesicular transporters for glutamate (VGLUT1-3) in motoneuron synaptic terminals that were retrogradely labeled and/or expressed VAChT. Analyses were preformed in P3, P10, and P15 motoneurons in both mice and rats. Although VAChT-IR was readily demonstrable in most retrogradely labeled motor axon varicosities (82% at P3; 90% at P10), they lacked significant immunoreactivity for any of the three known VGLUT isoforms. Accordingly, we could not detect colocalization between VAChT-IR and any VGLUT isoform in ventral laminae VII and IX (Fig. 8 a-k).

Nevertheless, calbindin-IR putative Renshaw cells were densely innervated by VGLUT1- and VGLUT2-IR varicosities, although at 3- to 4-fold lower density than VAChT-IR contacts. No overlap was found between VGLUT1- or VGLUT2-IR and VAChT-IR boutons contacting Renshaw cells (Fig. 8 l and m). VGLUT1-IR varicosities were always intensely labeled, whereas VGLUT2 and VGLUT3 boutons expressed a range of labeling intensities, despite the use of tyramide amplification (for VGLUT2 and -3) to increase sensitivity and allow detection of even the most weakly labeled varicosities. VGLUT3-IR fibers occur at much lower density than VGLUT1 or VGLUT2 varicosities in the spinal cord, but they are preferentially distributed at the border between ventral lamina IX/LVII and the ventral funiculus, a region enriched with Renshaw cells. Despite this overlap in distribution, VGLUT3-IR boutons were seldom found in contact with Renshaw cells. Approximately 24% (6 of 26) of the sampled calbindin-IR putative Renshaw cells received a single VGLUT3-IR contact on their proximal dendrites, and on two occasions two contacts (Fig. 8n). It should be noted that without VAChT-IR it is difficult to be certain that a retrogradely labeled process is an axon terminal. Furthermore, some motoneurons and their dendrites express glutamate-IR, which could lead to the erroneous identification of glutamate-IR within motoneuron axons. Nonetheless, we did not detect any Glut-IR, in retrograde varicosities in contact with Renshaw cell somas or dendrites, suggesting that such fibers, if they exist, are rare.

Similar conclusions were obtained for five Renshaw cells (P2-P4) intracellularly labeled during physiological experiments and then immunolabeled with different combinations of VGLUT antibodies (Fig. 9, which is published as supporting information on the PNAS web site). Collectively, these findings indicate that, if glutamate or another excitatory amino acid neurotransmitter is released from motoneuron synapses, it must use mechanisms independent of currently known vesicular glutamate transporters.

Discussion

Motoneuron Evoked NMDA and AMPA/Kainate Excitatory Postsynaptic Potentials (EPSPs) in Renshaw Cells. We have demonstrated that EPSPs evoked in Renshaw cells by stimulation of motoneuron axons are mediated by the activation of NMDA, AMPA/kainate, and acetylcholine receptors. The latency of NMDA and AMPA/kainate components was the same as the monosynaptic cholinergic EPSP, indicating that they are unlikely to be mediated by polysynaptic activation of glutamatergic interneurons presynaptic to Renshaw cells. This finding may explain the puzzling observation that cholinergic antagonists fail to completely suppress either the synaptic activation of Renshaw cells by motoneurons in the adult cat (21) or the recurrent IPSP evoked in neonatal rat motoneurons by ventral root stimulation (22).

The most parsimonious mechanism to account for this finding is that glutamate or a related amino acid is released from motoneuron terminals in addition to acetylcholine. Such corelease may be more common than previously thought given the recently discovered plasticity of neurotransmitter phenotype (34) and the finding that glutamate and acetylcholine are both released from the synaptic terminals of a class of spinal interneuron in the developing Xenopus embryo (25). Although glutamate is the predominant neurotransmitter released by motor axons in insects (35) and crayfish (36), evidence for the release of excitatory amino acids from motor axons in mammals is controversial. Several reports have provided evidence for the presence of VGLUTs on mammalian motoneuron terminals, but we could not confirm these findings. However, our results are consistent with studies indicating the absence of VGLUT mRNA expression in motoneurons (14, 15). Our failure to identify any of the three vesicular glutamate transporters in motoneurons or their terminals suggests that release occurs from a source other than the motoneuron terminal or, alternatively, that there is another, as yet unknown vesicular glutamate transporter that could store EAAs inside synaptic vesicles. This latter idea is plausible because other known glutamate-releasing synapses in the spinal cord (i.e., some C-fiber afferents) contain only weak IR for presently known vesicular glutamate transporters (33, 37).

It is unlikely that the observed effects are mediated by ventral root glutamatergic primary afferents fibers. First, in agreement with recent studies (38, 39), we failed to identify (using retrograde and anterograde labeling) the sensory afferent axons or DRG neurons that were earlier proposed to project into the spinal cord through ventral roots (30). Second, we found that terminals labeled retrogradely from the ventral roots did not express VGLUT1-IR, a characteristic of many myelinated glutamatergic dorsal root primary afferents (33, 37). Finally, if indeed there are a few small unmyelinated glutamatergic C-fibers (VGLUT1 negative) entering the spinal cord via the ventral roots, these are likely to have conduction velocities slower than those from motor axons and their putative effects should exhibit delayed latencies compared to the cholinergic motor axon-evoked EPSPs.

Another possibility is that synchronous activation of motoneurons by the antidromic stimulus might activate glutamatergic interneurons (or their axons) ephaptically or through gap junctions. This is unlikely because such coupling would introduce an additional delay between the stimulus and the onset of the glutamatergic synaptic potential that we never observed. The measured latency to the onset of the intracellularly recorded antidromic action potential in motoneurons after a ventral root stimulus was 1.5 ms. If we assume that enough current flows across the putative gap junctions to fire interneurons by the time the motoneuronal action potential has risen to 50% of its peak value (3.3 ms measured in 11 motoneurons), then the total time required to activate the interposed interneuron would be 1.5 + 3.3 ms = 4.8 ms. This time does not include the synaptic delay, which we estimate to be ≈2 ms (see Supporting Text). Because the measured latency of the glutamatergic component of the Renshaw potential was 3.1 ms, it seems very unlikely that this mechanism can account for the findings. Consistent with these observations, electrical or dye coupling has never been described between mammalian motoneurons and interneurons, although it is known that synergistic motoneurons are coupled during development (40, 41).

A further possibility is that an excitatory amino acid other than, or in addition to, glutamate might be released from motoneuron terminals. It is known that l-aspartate and l-homocysteic acid can activate NMDA and non-NMDA receptors on Renshaw cells (42). Moreover, aspartate and glutamate can be colocalized and coreleased from several central synapses in other brain regions (43, 44), and none of the three known VGLUT isoforms transport aspartate into synaptic vesicles.

Locomotor Activity Triggered by Motor Axon Stimulation. We have shown that activation of motor axons can trigger locomotor-like activity in the neonatal mouse spinal cord. The effect persisted in cholinergic blockers, but was abolished by APV. Although it is possible that APV interfered with the subsequent genesis of locomotor-like activity at the network level, we note that bath-applied APV does not abolish locomotor-like activity induced by dopamine and serotonin in the neonatal mouse cord (27). A more plausible explanation is that lumbar motoneurons release glutamate and/or aspartate thereby triggering an NMDA-mediated mechanism in the network. Postsynaptic targets that could mediate network activation include Renshaw cells that, at this age, may produce depolarizing GABA/glycine potentials in other spinal interneurons. Alternatively, NMDA activation of postsynaptic motoneurons might directly trigger the expression of rhythmic activity. In the rat spinal cord, bath-application of NMDA and serotonin can trigger rhythmic activation of motoneurons under conditions of synaptic blockade (29). However, even if motoneurons become rhythmically active in response to glutamate release, the phasing of activity in the different ventral roots presumably requires activation of an interneuronal locomotor network. It is unlikely that activation of the locomotor network depends on electrical coupling between motoneurons and spinal interneurons because it persisted for up to 45 min in the presence of the gap junction blocker carbenoxolone.

In conclusion, we show that mammalian motoneurons are able to activate NMDA and AMPA/kainite receptors in Renshaw cells, perhaps by releasing EAAs through a hitherto unknown mechanism. Our results also suggest that excitatory amino acid receptor activation by motoneurons can play an important role in mammalian locomotion, thereby shedding light on the involvement of mammalian motoneurons in spinal network function.

Note Added in Proof. While our paper was in review, Nishimaru et al. (45) published similar findings.