The neuronal networks that generate vertebrate movements such as walking and swimming are embedded in the spinal cord1,2,3. These networks, which are referred to as central pattern generators (CPGs), are ideal systems for determining how ensembles of neurons generate simple behavioural outputs. In spite of efforts to address the organization of the locomotor CPG in walking animals2,4,5,6, little is known about the identity and function of the spinal interneuron cell types that contribute to these locomotor networks. Here we use four complementary genetic approaches to directly address the function of mouse V1 neurons, a class of local circuit inhibitory interneurons that selectively express the transcription factor Engrailed1. Our results show that V1 neurons shape motor outputs during locomotion and are required for generating ‘fast’ motor bursting. These findings outline an important role for inhibition in regulating the frequency of the locomotor CPG rhythm, and also suggest that V1 neurons may have an evolutionarily conserved role in controlling the speed of vertebrate locomotor movements.
Several genetically defined classes of neurons have been identified in the developing spinal cord7, including a class of ipsilaterally projecting inhibitory neurons that innervate motor neurons, the Engrailed1 (En1)-expressing V1 neurons8,9,10,11. To assess the function of V1 neurons in the locomotor CPG, we used two mouse models that have selective loss of the En1-expressing V1 neuronal population (Fig. 1). This was achieved through the altered specification of V1 neurons in Pax6-knockout (Pax6-/-) mice and by the selective ablation of these neurons in En1Cre; R26-lacZflox/DTA (En1-DTA; ref. 12) mice. At embryonic day (E)12.5, a marked reduction in En1-positive V1 neuron cell numbers was apparent in spinal cords isolated from Pax6-/- and En1-DTA embryos (Fig. 1a, e). This was confirmed using an En1Cre lacZ or green fluorescent protein (GFP) reporter system9, which also showed that significantly fewer En1-derived V1 neurons are present at E18.5 (Fig. 1b, f). Both Pax6-/- and En1-DTA mice had normal numbers of lumbar spinal motor neurons that were appropriately organized into lateral and medial motor columns. (Fig. 1c, g). Spinal cords from Pax6-/- mice did show an increase in commissural V0 and V3 interneuron cell numbers, coupled with a slight decrease in ipsilaterally projecting Chx10-positive V2 neurons13 (Chx10 is a marker of V2 neurons; Supplementary Fig. S1). However, there were no changes in ventral interneuron specification in the En1-DTA mice (Supplementary Fig. S1).
Limbed walking movements in vertebrates are characterized by the repetitive oscillatory bursting of motor neurons, in which flexor–extensor and left–right motor activities alternate. These locomotor-like oscillations can be induced in the isolated spinal cord by excitatory neurotransmitter agonists, and typically show a step-cycle period of 2–4 s (refs 5, 14, 15 and Fig. 2a). We asked whether V1 neurons regulate the coordination of flexor–extensor motor neurons during locomotion, because many V1 neurons differentiate into Renshaw cells and Ia inhibitory interneurons, two inhibitory cell types that have been proposed to coordinate flexor and extensor muscle activity around ipsilateral limb joints9,16,17,18,19. Spinal cords lacking V1 neurons showed a normal pattern of alternating flexor (L2) activity and extensor (L5) activity (Fig. 2a–c). Left–right alternation (that, is lL2 versus rL2) was also normal. However, ‘locomoting’ Pax6-/- and En1-DTA spinal cords did show a significant lengthening of both the step-cycle period and motor neuron burst duration (Fig. 2b–e).
Whereas 5 µM NMDA (N-methyl-d-aspartate) in combination with 5 µM serotonin (5-hydroxytryptamine or 5-HT) typically induces robust locomotor-like activity in wild-type spinal cords15 (Fig. 2a), we were only able to elicit a stable pattern of locomotor-like activity in Pax6-/- spinal cords by increasing the concentration of 5-HT to 7.5–20 µM (Fig. 2b, e). In experiments using 10 µM 5-HT, the step-cycle period (8.4 ± 2.9 s, mean ± s.d.) and burst duration (4.4 ± 1.0 s) in Pax6-/- spinal cords (n = 7) were increased significantly (P < 0.001, Student's t-test) compared to wild-type and heterozygous (Pax6+/-) spinal cords (step-cycle period 3.8 ± 0.8 s, burst duration 1.7 ± 0.8 s; n = 10). This relative increase in the step-cycle period was seen over a wide range of 5-HT concentrations (Fig. 2e). Similar lengthening of the step cycle (Fig. 2c, d) and burst duration was seen in En1-DTA mice (9.3 ± 2.3 s and 5.4 ± 1.1 s, respectively; n = 5 spinal cords; P < 0.001, t-test). V1 neurons are therefore dispensable for flexor–extensor coordination, but have an essential role in determining the speed of the locomotor rhythm.
Current-clamp recordings of motor neurons in wild-type and Pax6-/- spinal cords (Fig. 3a–d) revealed that Pax6-/- motor neurons show prolonged periods of membrane potential depolarization, and that they continue to fire action potentials throughout this depolarized phase (Fig. 3c). This is consistent with flexor- and extensor-related CPG neurons remaining active for longer periods of time. These intracellular recordings also revealed that the alternating excitatory and inhibitory drive that normally underlies rhythmic changes in motor neuron membrane potential20 is still present in the Pax6-/- spinal cord (Fig. 3e). Importantly, Pax6-/- motor neurons still receive the phasic inhibitory inputs that are essential for coordinating alternating flexor and extensor activity (Fig. 3f). The origin of these phasic inhibitory inputs is not known—although they could in principle be derived from contralateral commissural interneurons21,22, this seems not to be the case, as ipsilateral locomotor coordination is normal in the Pax6-/- cord after spinal cord hemisection (data not shown).
We then used En1tlZ/tlZ null ‘knock-in’ mice (see ref. 8) to address the function of V1 neurons in the intact animal. Spinal cords isolated from E18.5 En1tlZ/tlZ mice show a 3–4-fold slowing of the locomotor step cycle and lengthening of the burst duration (Fig. 4a, b; 11.2 ± 2.3 s and 6.1 ± 1.8 s, respectively; n = 10 spinal cords), which closely resemble the behavioural changes seen in Pax6-/- and En1-DTA spinal cords (compare with Fig. 2). Using a Wnt1En1 transgene23 to rescue the En1 mutant midbrain–hindbrain phenotype, we obtained adult En1tlZ/tlZ mice in which the V1 neuron connectivity defects in the spinal cords8,9 are not rescued (H.S., unpublished observation). When rotarod tests were performed on 4–6-month-old En1tlZ/tlZ; Wnt1En1 mice and age-matched controls, En1tlZ/tlZ; Wnt1En1 mice showed a clear deficit in their ability to walk and maintain their balance at higher rotarod speeds (Fig. 4c; see Methods). Although the duration (and the speed) for which the mutant animals walked on the rotarod increased during the trial period, at no time did they attain walking speeds above 18 r.p.m. However, these animals did walk on the rotarod for extended periods of times at idling speed (5 r.p.m., data not shown). In contrast, wild-type animals walked and maintained their balance at much higher speeds (approximately 22 r.p.m. on day 1 and up to 35 r.p.m. on day 5). This suggests that adult En1 mutants are impaired in their ability to perform quick stepping movements.
Next, we used transgenic mice (AlstR192 mice) that conditionally express the Drosophila allatostatin G-protein-coupled receptor (AlstR) to test whether acutely suppressing V1 neuron excitability causes a similar lengthening of the locomotor step cycle (Fig. 5a). AlstR couples to endogenous mammalian GIRK channels (inward-rectifying K+ channels), causing a decrease in cellular input resistance and neuronal excitability24. In control experiments, allatostatin (100 nM–5 µM) had no effect on motor rhythm generation, locomotor coordination or the step-cycle period (Supplementary Fig. S2a). In contrast, spinal cords from nestinCre; AlstR192 mice showed strong allatostatin-dependent depression in rhythmic motor activity (Supplementary Fig. S2b).
En1Cre; AlstR192 mice were then used to selectively express AlstR in V1 neurons (Fig. 5a, b). In spinal cord slices prepared from En1Cre; AlstR192 mice, we observed a pattern of GFP reporter expression that was identical to the normal distribution of V1 neurons (Fig. 5c). Whole-cell recordings from identified GFP-positive V1 neurons (n = 7 cells) revealed a reversible decrease in neuronal excitability in response to current steps (Fig. 5d) and ramps (Fig. 5e) after the addition of allatostatin (10 nM). The decrease in excitability (2–3-fold), although not as pronounced as in isolated adult neurons24, was nevertheless substantial, and it was not seen in control, AlstR-negative neurons (Supplementary Fig. S2c).
When locomotor-like activity was induced in postnatal day P0–P2 En1Cre; AlstR192 spinal cords (5 µM NMDA + 10 µM 5-HT; Fig. 5f), we observed a step-cycle period of 3–4 s (3.3 ± 0.8 s, n = 15 spinal cords), comparable to that seen in wild-type animals of a similar age. Application of allatostatin (1–5 µM) to these spinal cords resulted in marked and significant lengthening of the step-cycle period (7.9 ± 1.8 s, n = 15 spinal cords; P = 0.002, paired t-test), which was reversed upon washing out the neuropeptide (Fig. 5f). As such, lengthening of the locomotor step cycle that results from the defects in V1 neuron development can be reproduced when V1 neuronal activity is suppressed. Two recent studies have provided evidence that modifying neuronal activity during development can alter the neurotransmitter phenotype of spinal neurons25 and produce small changes in CPG motor outputs26, indicating that activity may have a role in organizing these locomotor networks However, our results show that the changes in locomotor CPG outputs following the acute suppression of V1 neurons compared with the early loss of V1 neurons are highly concordant. This suggests that the absence of V1 neurons during development does not markedly reconfigure the spinal locomotor CPG network.
Our findings outline an essential role for V1 neurons in mammals in regulating the duration of the locomotor step cycle, and hence the speed of the locomotion. This function is specific for V1 neurons, as we have not seen marked changes in the frequency or duration of the step cycle when either V0 or V3 neurons are deleted from the spinal cord5 (M.G., unpublished observations). Our conclusion that V1 neurons are necessary for ‘fast’ locomotor outputs in mice is supported by studies in Xenopus tadpoles showing that aIN neurons, the homologues of mouse V1 neurons, are the primary source of early cycle inhibition to the swimming locomotor CPG10. Although the role of aIN neurons in regulating the speed of Xenopus swimming movements was not examined, a strong correlation was seen between aIN-derived inhibitory inputs to CPG neurons and the frequency of swimming movements10. This raises the possibility that the Xenopus homologues of V1 neurons may facilitate fast swimming movements, in much the same way that their mammalian counterparts are required for fast ‘walking’ movements.
The exact contribution individual V1 interneuron subtypes make to regulating the duration of motor neuron bursting during fictive locomotion remains to be determined. Inhibition of Renshaw cells with cholinergic blockers causes a small increase in the step-cycle period26 (S.G. and O.K., unpublished observations), suggesting that a group of non-Renshaw V1 neurons or the V1 population as a whole regulates the step-cycle period during fictive locomotion. Although Renshaw cells and Ia inhibitory interneurons have described roles in coordinating flexor–extensor motor neurons during spinal reflexes27, our study demonstrates that the V1 population as a whole does not have a primary role in regulating locomotor CPG flexor–extensor activity. It therefore seems that an additional group of non-V1 interneurons have been incorporated into the walking CPG of terrestrial vertebrates to secure flexor–extensor alternation.
Our study also demonstrates that the acute silencing of a select population of neurons using genetic approaches can be used to elucidate their function with respect to a defined behaviour such as locomotion. Our results highlight the feasibility of using ligand-activated GIRK channels to selectively manipulate neuronal excitability in the vertebrate nervous system, and the conditional AlstR192 mice described here can be used to selectively silence neurons throughout the nervous system. This and similar genetic approaches now make it feasible to selectively probe the function of small populations of neurons, which should facilitate the mapping of neural circuits at higher resolutions than was previously possible.
The generation and genotyping of the Pax6lacZ, Sey, En1Cre, En1tlZ, Wnt1En1, nestinCre, R26-lacZflox/DTA, R26lacZ and R26GFP alleles in mice has been previously described8,9,12,23,28. Embryos and tissues were obtained from timed matings.
Generation of conditional AlstR mice
The Drosophila AlstR coding sequence, followed by an IRES-EGFP24, was inserted downstream of a loxP-flanked βgeo/polyA stop sequence in the conditional Z/EG construct29 to produce the Z/AlstR expression vector (see Fig. 4a). The linearized Z/AlstR construct was electroporated into the embryonic stem cell line W9.5. G418-resistant clones were screened for uniform and robust lacZ expression, and for single transgene integrants. Two clones gave germline transmission (AlstR172 and AlstR192). AlstR192 mice that showed lacZ expression throughout the nervous system were used for all further experiments.
In situ hybridization and immunohistochemistry
Immunostainings were performed as previously described5. Images were captured using a Zeiss LSM510 confocal microscope and assembled using Adobe Photoshop. Cell numbers are indicated as mean ± s.d. per section per spinal cord side, and were determined on counts of at least three embryos (five sections each). In situ hybridization and staining for β-galactosidase activity were performed as previously described5,8.
Electrophysiology: electroneurogram recordings
Electrophysiological experiments were performed on E18.5–P2 mice in accordance with the ethical rules stipulated by NIH and the Swedish government. Animals were anaesthetized and decapitated, and spinal cords were dissected out in ice-cold Ringers' solution4. Recordings were made in Ringers' solution at 20 °C by placing bipolar suction electrodes on three of the second and fifth lumbar ventral roots (that is rL2, lL2, rL5, lL5). Electroneurogram signals were amplified, band-pass filtered (100 Hz–1 kHz), digitized and collected using Axoscope software (Axon Instruments). Rhythmic locomotor activity was induced by adding NMDA (5 µM) and 5-HT (5–30 µM) to the Ringers' solution. Effects of allatostatin (10 nM–5 µM) were examined by adding the peptide to the perfusion solution.
Electrophysiology: analysis of locomotor activity
Step-cycle period and burst duration were determined by analysing lL2 or rL2 activity4. Step-cycle period and burst duration averages were determined from all recorded bursts after the onset of stable locomotor-like activity. The effects of allatostatin were measured 10 min after application to allow for drug wash-in. All measurements are given ± s.d. P values less than 0.05 were considered significant. Circular statistics19 were used to determine the coupling strength between left and right ventral roots, as well as between flexor-related (L2) and extensor-related (L5) ventral roots on the same side of the spinal cord. L2 bursts were used as a reference. Vector points representing the phase values between 0 and 1 were plotted for each experiment, and show the mean phase as well as the concentration of phase values around the mean. Appropriate left–right or flexor–extensor alternation is represented by phase values around 0.5.
Electrophysiology: intracellular recordings
Whole-cell patch-clamp recordings from motor neurons were made using recording pipettes with a resistance of 4–5 MΩ. The microelectrode was driven into the spinal cord through a pial patch in the ventrolateral surface. Cells were patched in the motor neuron area and recorded using an AxoClamp 2B amplifier (Axon Instruments).
Electrophysiology: acute spinal cord slice recordings
For slice recordings, spinal cord slices 250–300-µm thick were cut using a Leica VY1000E vibrating microtome. After a 1-h recovery period, slices were transferred to a recording chamber, mounted on an Olympus BX51W1 microscope and perfused with oxygenated Ringers' solution at room temperature. GFP-positive cells were visualized using a DAGE-MTI IR-1000 CCD camera and patched visually using a Sutter MPC-325 micromanipulator. Recordings were made in current-clamp mode using a MultiClamp 700B amplifier (Axon Instruments).
Rotarod test were performed during the light phase of a 12 h:12 h light:dark cycle as previously described30. Mice were placed on a Rotamex 4/8 (Columbus Instruments) idling at 5 r.p.m. Rotarod speed was set to increase gradually from 5 to 65 r.p.m. over the course of 3 min. Retention time on the rod was recorded for three trials per mouse per day for five consecutive days, using three mutant and three wild-type mice.
We thank J. Rivier for the allatostatin peptide, P. Gruss and A. Mansouri for Pax6 mutant mice, P. Soriano for Rosa26lacZ mice, A. McMahon for Wnt1En1 mice, S. Narayan for generating ZnG reporter mice, and M. Gross, A. Joyner, S. Pfaff and P. Slesinger for materials. We thank G. Lemke, S. Pfaff, P. Gray and K. Quinlan for their comments on the manuscript. This research was supported by grants from the National Institutes of Health (M.G., O.K.) and the Human Frontiers Science Program (O.K. and M.G.). G.M.L. was supported by an HFSP postdoctoral fellowship. Author Contributions are provided in the Supplementary Information.