Spinal Hb9::Cre-derived excitatory interneurons contribute to rhythm generation in the mouse

Rhythm generating neurons are thought to be ipsilaterally-projecting excitatory neurons in the thoracolumbar mammalian spinal cord. Recently, a subset of Shox2 interneurons (Shox2 non-V2a INs) was found to fulfill these criteria and make up a fraction of the rhythm-generating population. Here we use Hb9::Cre mice to genetically manipulate Hb9::Cre-derived excitatory interneurons (INs) in order to determine the role of these INs in rhythm generation. We demonstrate that this line captures a consistent population of spinal INs which is mixed with respect to neurotransmitter phenotype and progenitor domain, but does not overlap with the Shox2 non-V2a population. We also show that Hb9::Cre-derived INs include the comparatively small medial population of INs which continues to express Hb9 postnatally. When excitatory neurotransmission is selectively blocked by deleting Vglut2 from Hb9::Cre-derived INs, there is no difference in left-right and/or flexor-extensor phasing between these cords and controls, suggesting that excitatory Hb9::Cre-derived INs do not affect pattern generation. In contrast, the frequencies of locomotor activity are significantly lower in cords from Hb9::Cre-Vglut2Δ/Δ mice than in cords from controls. Collectively, our findings indicate that excitatory Hb9::Cre-derived INs constitute a distinct population of neurons that participates in the rhythm generating kernel for spinal locomotion.

Locomotor behaviors, such as flying, swimming or walking, are complex motor actions that give animals and humans the ability to move and interact with the surroundings. The generation of locomotion in vertebrates is largely determined by neural networks in the spinal cord [1][2][3][4] . These spinal networks are responsible for key features that characterize limbed locomotion in mammals, rhythm generation and the precise coordination of muscle activation. In the ventral spinal cord where the locomotor circuit is located 5 , developmentally expressed transcription factors 6,7 have allowed for the identification and targeted manipulation of selected interneuron populations 3,[8][9][10] . These studies have provided insights into the organization of the neuronal circuits underlying left-right [11][12][13][14][15][16] and flexor-extensor [17][18][19] coordination in mammals. Optogenetic, pharmacological, and lesion studies have provided strong evidence that the rhythmic drive in the mammalian spinal locomotor circuit comes from activity in ipsilaterally-projecting excitatory glutamatergic neurons [1][2][3]9,[20][21][22][23][24] . However, despite the ablation of entire progenitor domain-derived classes of ipsilaterally-projecting excitatory neurons, rhythm-generating interneurons still remain elusive 12,14,16,[25][26][27] . Therefore, there has also been focus on identifying and studying cell populations that span several progenitor domains. Recently, one such study directly linked rhythm generation to a group of excitatory interneurons identified by the embryonic expression of the transcription factor short stature homeobox protein 2 (Shox2) 28 . Shox2 neurons span several of the dorsally and ventrally progenitor domain-derived interneuron classes, and partially overlap with V2a neurons which express the transcription factor Chx10 during development [28][29][30] . Shox2 non-V2a neurons were shown to be involved in rhythm generation 28 . However, eliminating Shox2 non-V2a neurons from the network did not result in complete abolition of locomotor rhythm. Collectively, these data suggest that glutamatergic interneuron populations other than those captured in the current transcription factor schemas are involved in rhythm generation. To target such populations we are examining currently available genetic tools that may define unique neuronal groups to which we can ascribe such a function.
One group of interest is the group of neurons labeled by the transcription factor Hb9. Hb9 is usually described as a postmitotic motor neuron marker 31 , however, it is also transiently expressed in a larger group of progenitor cells and postmitotic neurons embryonically 32 . As it is commonly done when targeting spinal cord interneuron populations, we use conditional genetics and rely on Cre-dependent recombination to reliably target excitatory interneurons identified by the developmental expression of Hb9. This approach includes the small medial subset of interneurons retaining HB9 protein expression postnatally, which have been extensively studied and found to be excitatory neurons displaying cellular properties consistent with being potential rhythm generating neurons 33-39 . In the current study, we selectively silence glutamatergic synaptic transmission in Hb9::Cre-derived excitatory neurons while retaining motor neuron output. We demonstrate that although reporter expression observed in Hb9::Cre mice represents a much larger population of neurons than those expressing the HB9 protein postnatally, the population identified is consistent and can be reliably used to target and manipulate a novel excitatory neuronal population in the spinal cord. We also show that excitatory Hb9::Cre-derived interneurons do not overlap with the Shox2 non-V2a population. Synaptically silencing the excitatory subset of Hb9::Cre-derived interneurons by a targeted deletion of the vesicular glutamate transporter 2 (Vglut2) leads to a significant reduction in locomotor frequency without any significant effect in pattern formation, suggesting a role in rhythm generation. Taken together, our findings indicate that excitatory Hb9::Cre-derived interneurons constitute a second population of neurons, distinct from the Shox2 non-V2a, which appear to be involved in the rhythm-generating kernel for mammalian locomotion.

Results
Hb9::Cre-derived INs are located in the ventral and dorsal spinal cord. Although Hb9::Cre 40 and Hb9-GFP 41 mice were generated with an analogous strategy to the Hb9-LacZ mice 31 , reporter expression in Hb9::Cre-reporter-labeled (Hb9::Cre;Rosa26-FP) and Hb9-GFP mice is not restricted to motor neurons (MNs). In the Hb9-GFP mouse, a ventral population of interneurons is marked 41 , whereas in the Hb9::Cre;Rosa26-FP mouse line a dorsal population of cells is also captured. The increase in the number and laminar distribution of fluorescent reporter cells observed in Hb9::Cre;Rosa26-FP mice may be attributed to the transient embryonic expression of Hb9 in these cells 32,42 .
YFP expression in Hb9::Cre;Rosa26-YFP mice was detected through lamina I to VI in addition to lamina VII, VIII and ventral X (Fig. 1A) throughout the lumbar spinal cord. This is in contrast to the GFP expression in Hb9-GFP mice 33 , which is restricted to lamina VII, VIII and ventral X (Fig. 1B). We will collectively refer to these dorsal (lamina I-VI) and ventral neuronal populations in Hb9::Cre;Rosa26-FP mice as Hb9::Cre-derived INs.
Despite the larger number of reporter-expressing cells in Hb9::Cre;Rosa26-FP mice, reporter expression identifies a consistent group of interneurons throughout the lumbar spinal cord. Thus, we use the Hb9::Cre mouse as a genetic tool to study the possible roles of excitatory Hb9::Cre-derived INs in locomotion. In Hb9::Cre;Rosa26-YFP mice, an overlap of YFP and HB9 protein was evident in canonical Hb9 INs, motor neurons (MNs) and sympathetic preganglionic neurons (Fig. 1A). We indeed found that the majority of canonical Hb9 INs co-express HB9 protein and the reporter protein, YFP, in Hb9::Cre;Rosa26-YFP mice (86% ± 7%, N = 3, 18 sections). Conversely, canonical Hb9 INs make up less than 1% (0.86% ± 0.37%) of the Hb9::Cre-derived IN population (N = 3, 18 sections) (Fig. 1C). The number of canonical Hb9 neurons (193 ± 61 cells between Th12 and L3, N = 3) is similar to what was previously reported 33 . Taken together these data indicate that most of the canonical Hb9 INs are part of Hb9::Cre-derived IN population, but they account for a very small percentage of this population. Therefore, hereinafter whenever we refer to the Hb9::Cre-derived INs, the canonical Hb9 INs in Th12-L3 will be implicitly included.
with the exclusively excitatory and ipsilateral IN markers, Isl1, Chx10 and Shox2 ( Fig. 3A and B respectively). All experiments were performed at embryonic stage E11.5 to capture the peak of expression of these early neuronal markers.
The LIM homeodomain transcription factor 1 (Isl1) is expressed in MNs and is involved in MN differentiation 31 . Its expression defines a class of interneurons, the dI3 INs 43,44 , which have been shown to be involved in the cutaneous regulation of paw grasp 26 but not in locomotor rhythm or patterning. Therefore, we sought to investigate whether or not dI3 INs and Hb9::Cre-derived INs were overlapping. We found that Hb9::Cre-derived INs seldomly co-expressed Isl1 (6% ± 0.2%, N = 2, 22 sections) (Fig. 3C), indicating that the dI3 IN population shows no appreciable overlap with the Hb9::Cre-derived INs.
Altogether, these data indicate that although excitatory Hb9::Cre-derived INs are a heterogeneous group of neurons that span several excitatory progenitor domains, they make up a distinct group of excitatory neurons that only marginally overlaps with the Shox2 non-V2a population.
In order to selectively target excitatory Hb9::Cre-derived INs and investigate their role in locomotor network activity, Hb9::Cre mice were crossed with mice carrying a conditional floxed Vglut2 allele to produce offspring with a Cre-dependent selective loss of Vglut2 (see Experimental Procedures), thereby disrupting synaptic transmission exclusively in excitatory Hb9::Cre-derived INs. As shown in previous work from our lab and others, removal of Vglut2 efficiently blocks action potential mediated synaptic transmission from the affected neurons 17,26,28 , and hence results in the functional removal of these neurons from the network.
To assess the loss of Vglut2 transcript, we performed in situ hybridization with a Vglut2 probe on spinal cord tissue from both mice which have complete removal of Vglut2 from Hb9::Cre-derived neurons (Hb9::Cre;Vglut2 Δ/Flox ; Rosa26YFP, referred to as Hb9::Cre-Vglut2 Δ/Δ ) and littermates which have one copy of Vglut2 removed but retain normal synaptic function (Hb9::Cre;Vglut2 Flox/+ ;Rosa26-YFP, referred to as controls). Vglut2 signal was seen extensively throughout the gray matter in cords from both control (    (Fig. 5B). There was no significant difference in pattern generation as assessed by coordination of activity between the left and right sides of the cord (out of phase activity between right L2 and left L2) and in ipsilateral flexor-extensor coordination (out of phase activity between right L2 (flexor) and right L5 (extensor)) between control and Hb9::Cre-Vglut2 Δ/Δ cords (Watson-William's Test, p > 0.09). Mean left-right phase values (Fig. 5C, left)  In addition to drug-evoked locomotion, we also tested the effects of the loss of excitatory Hb9::Cre-derived INs on locomotion evoked by neural stimulation. It was possible to evoke locomotor-like activity by stimulating fibers descending from the brainstem in all control cords tested and in ~80% of the Hb9::Cre-Vglut2 Δ/Δ cords (N = 13/16). The bursting frequency was reduced in Hb9::Cre-Vglut2 Δ/Δ cords by 15-21% compared to controls at stimulation strengths of 250 μ A − 1 mA (p < 0.02) (Fig. 5D). Additionally, during electrical stimulation, the bursting was more difficult to discriminate in spinal cords from Hb9::Cre-Vglut2 Δ/Δ mice.
Together, the lower drug-evoked and stimulus-evoked locomotor frequencies seen in spinal cords from Hb9::Cre-Vglut2 Δ/Δ mice compared to controls suggest that excitatory spinal Hb9::Cre-derived INs play a significant role in rhythm generation.
Loss of glutamatergic synaptic transmission in motor neurons does not affect the frequency of drug-induced or neural-evoked locomotor activity. Hb9 is expressed by motor neurons 31 and mammalian motor neurons co-release glutamate and acetylcholine from central collaterals 17,45,46 . Since locomotion can be initiated by ventral root stimulation 46,47 , we sought to determine if loss of Vglut2-mediated glutamatergic synaptic transmission in motor neurons would contribute to the locomotor phenotype observed in Hb9::Cre-Vglut2 Δ/Δ spinal cords.
Similarly to our approach with Hb9::Cre mice, ChAT::Cre mice were used to generate ChAT-Vglut2 Δ/Δ mice in which glutamatergic synaptic transmission was selectively lost in cholinergic neurons, including MNs. ChAT-Vglut2 Δ/Δ mice are viable and show no clear phenotype; however, all experiments were performed at P0 for consistency. Locomotor-like activity was induced by bath application of NMDA (5 μ M-10 μ M) and 5-HT (8 μ M), and it was evoked in all control (N = 5/5) (Fig. 6A, upper traces) and ChAT-Vglut2 Δ/Δ spinal cords (5 μ M, N = 7/9; 7 μ M and 10 μ M NMDA, N = 9/9) (Fig. 6A, lower traces). The locomotor frequency obtained in ChAT-Vglut2 Δ/Δ cords was not significantly different from that of controls at any NMDA concentration tested (p > 0. To further verify if there were any deficits to the rhythm generating network due to the loss of glutamatergic transmission in motor neurons, we also used descending stimulation to evoke locomotion. Locomotor-like activity was evoked in all control (N = 6/6) and ChAT-Vglut2 Δ/Δ (N = 9/9) spinal cords. In keeping with the results obtained from drug-induced locomotor studies, upon neural-evoked stimulation, no significant difference in locomotor frequency was observed between ChAT-Vglut2 Δ/Δ and control cords at any of the stimulation intensities tested, 100 μ A − 1 mA (p > 0.3) (Fig. 6D).
Collectively, these data indicate that the reduction in locomotor frequency we observe in Hb9::Cre-Vglut2 Δ/Δ cords is due to the loss of glutamatergic synaptic transmission in Hb9::Cre-derived INs and not in motor neurons.

Discussion
In this study, we have used the Hb9::Cre mouse line to identify Hb9::Cre-derived INs, target the excitatory subset of these neurons for removal from the spinal network, and investigate the consequences of their removal on motor output. Our findings provide insight into the involvement of these neurons in mammalian locomotor rhythm generation.
Hb9::Cre mice 40 were generated with an analogous strategy to the Hb9-LacZ mice 31 . However, a larger population of ventral as well as dorsal neurons is apparent in the Hb9::Cre;Rosa26-FP mouse. This is possibly due to the transient embryonic expression of Hb9 in progenitor cells and postmitotic neurons 32 which is captured by the Cre-line. Here, we have collectively referred to the dorsal and ventral populations visualized in Hb9::Cre;Rosa26-FP mice as Hb9::Cre-derived INs.
Since in limbed animals the spinal locomotor networks are distributed rostro-caudally along the ventral lumbar spinal cord 5,20 , the fact that the Hb9::Cre mouse also includes cells in the dorsal spinal cord does not hinder its use as a tool to study the role of excitatory Hb9::Cre-derived INs in locomotion. Moreover, despite the large number and laminar distribution of Hb9::Cre-derived INs, they comprise a reliably identifiable population of interneurons. The same neurons are visualized by reporter expression in all Rosa26 lines throughout the lumbar spinal cord. Additionally, as demonstrated by Vglut2 mRNA in situ, this same visualized population is manipulated in the Hb9::Cre-Vglut2 Δ/Δ mice.
Therefore, by using Cre-Lox recombination, we were able to reliably manipulate a group of excitatory INs which does not correspond to any of the known major excitatory ipsilaterally-projecting classes of neurons. In particular, excitatory Hb9::Cre-derived INs delineate a group of neurons that is distinct from the only other excitatory neuronal population that has been shown to play a role in mammalian locomotor rhythm generation, the Shox2 non-V2a neurons 28 .
The cardinal feature that should characterize rhythm-generating neurons is that their selective manipulation should have a direct impact on locomotor frequency. Their activation should be able to initiate locomotor rhythm and/or change the frequency of the ongoing rhythm whereas a selective reduction in their number should reduce the frequency of the ongoing locomotor rhythm 1,4,9,23,28,48,49 . The most conclusive test to determine the effects of a given group of neurons on locomotor activity is achieved by an acute activation or inactivation of these neurons. However, the combination of channelrhodopsin (or other cell activator or inactivator) mouse lines with Hb9::Cre mice would lead to expression in both inhibitory and excitatory Hb9::Cre-derived INs as well as in MNs, thus confounding results. The inclusion of MNs in the population makes optogenetic or chemogenetic perturbation experiments impossible to perform. Therefore, we instead pursued the functional removal of glutamatergic Hb9::Cre-derived INs from the network by selectively deleting Vglut2 from the excitatory Hb9::Cre-derived population. By silencing the output of glutamatergic Hb9::Cre-derived INs and studying motor output in an isolated spinal cord preparation, we have shown that excitatory Hb9::Cre-derived INs are likely part of the locomotor rhythm generator, as demonstrated by the reduced locomotor frequency. In contrast, silencing the output of Vglut2-expressing Hb9::Cre-derived INs had no effect on patterning, including left-right and flexor-extensor alternations, unlike what has been described for the manipulation of excitatory V2a, V0 or V3 neurons [12][13][14]16,50 . We acknowledge, however, that distinguishing between a role in tonic drive to the rhythm generator and a role in rhythm generation per se is difficult and nearly impossible experimentally, and therefore, we cannot exclude that this decrease in frequency might be partly due to a reduction in tonic drive to the rhythm generating core.
The fact that some of the spinal cords from Hb9::Cre-Vglut2 Δ/Δ mice did not respond to descending fiber stimulation may suggest that glutamatergic Hb9::Cre-derived INs play a role in mediating the initiation of locomotion. Thus, it is possible that the descending command system which initiates locomotion may target glutamatergic Hb9::Cre-derived INs and, due to the functional inactivation of these glutamatergic interneurons in Hb9::Cre-Vglut2 Δ/Δ mice, locomotion is less readily evoked by stimulation of the descending fibers.
With the genetic tools at hand, it is not possible to determine whether the canonical Hb9 INs are partly or solely accountable for the locomotor phenotype we observe in Hb9::Cre-Vglut2 Δ/Δ spinal cords. However, canonical Hb9 INs make up less than 1% of the Hb9::Cre-derived INs, and 33% of these Hb9::Cre-derived INs are excitatory. Therefore, the canonical Hb9 INs would be a negligible portion of the excitatory Hb9::Cre-derived INs (< 3%). Accordingly, in Hb9::Cre-Vglut2 Δ/Δ cords, the small number of canonical Hb9 neurons as compared to the total number of excitatory Hb9::Cre-derived INs, makes it less likely that the canonical Hb9 interneurons alone are responsible for the observed phenotype.
It should be emphasized that, similarly to what was concluded in studies on Shox2 neurons 28 , we believe that it is unlikely that excitatory Hb9::Cre-derived INs are the sole rhythm-generating neurons in the mammalian locomotor network. In the absence of excitatory Hb9::Cre-derived INs, locomotor frequency is reduced but locomotor rhythm is not abolished. This is seen more clearly when compared to experiments where all excitatory neurons are silenced, which leads to a complete cessation of the rhythm 23 . These observations suggest that several molecularly distinct groups of neurons may contribute to rhythm generation. A distribution of rhythm generation between molecularly distinct groups of excitatory neurons is also seen in young zebrafish where excitatory commissural neurons [51][52][53] and V2a domains are involved in rhythm generation [54][55][56] . However, the corresponding Shox2 INs and excitatory Hb9::Cre-derived INs have not yet been identified in the zebrafish.
Our classification of neurons as Hb9::Cre-derived INs does not sprout from the initial progenitor domain-defined classes of interneurons 7 . Instead, Hb9::Cre-derived INs, here, are those identified in the Hb9::Cre mouse line 40 . We have shown that excitatory Hb9::Cre-derived INs are a heterogeneous group of neurons that span several of the transcription factor-defined interneuron classes. Similarly, studies on Shox2 INs have shown that even within Shox2 non-V2a neurons, rhythm generation may be distributed amongst neurons derived from several progenitor domains 28 . Collectively, these data suggest that glutamatergic neurons spanning several classes of the known molecularly defined groups of neurons are responsible for rhythm generation. Therefore, it is plausible that, in the same way as new markers for acetylcholinergic spinal neurons 57,58 and inhibitory neurons 59,60 were identified, a comprehensive transcriptome screening to define a finer-grained molecular code of spinal glutamatergic neurons may define new makers for excitatory neurons that may help to clarify the identity of the functional subgroups involved in mammalian locomotor rhythm generation. For neurotransmitter phenotyping Vglut2-GFP, Gad67-GFP or Glyt2-GFP mice were crossed with the reporter line Rosa26-tdTomato to obtain the double transgenic lines Vglut2-GFP;Rosa26-tdTomato, Gad67-GFP;Rosa26-tdTomato or Glyt2-GFP;Rosa26-tdTomato. These double transgenic lines were then crossed with Hb9::Cre mice to generate the triple transgenic lines Hb9::Cre;Rosa26-tdTomato;Vglut2-GFP, Hb9::Cre;Rosa 26-tdTomato;Gad67-GFP; or Hb9::Cre;Rosa26-tdTomato;Glyt2-GFP.

Experimental
In Situ Hybridization, Immunohistochemistry and Cell Counts. Spinal cords from embryonic mice at stage E11.5 and lumbar spinal cords from newborn mice at P0 were immersion fixed in 4% paraformaldehyde (2 hours at room temperature for immunohistochemistry or 24 hours at 4 °C for in situ hybridization), cryopreserved in 30% sucrose overnight, and stored at − 80 °C. Serial transverse sections of embryonic spinal cords and newborn lumbar spinal cords were cryostat sectioned (20 μ m) and mounted on Menzel-Glaser SuperFrost slides. All slides were stored at − 80 °C. Sections were incubated for 24 hours with one or several of the following primary antibodies: rabbit anti-Shox2 #860 (1:32,000, generated against the peptide CKTTSKNSSIADLR), sheep anti-Chx10 (1:400, Chemicon), mouse anti-Isl1 (40.2D6 and 39.4D5) (1:250, DSHB), rabbit anti-Hb9 (1:16000, DSHB). The specificity of these antibodies has been evaluated and described in previous publications (see refs 12, 28 and 35). Secondary antibodies were obtained from Jackson Immunoresearch or Invitrogen, used at 1:500 and incubated for 2 hours at room temperature.
Slides were mounted in Vectashield reagent (Vector Laboratories), and images were scanned on a LSM510 Confocal Microscope (Zeiss Microsystems) using 10x and 20x objectives. Multiple channels were scanned sequentially to prevent fluorescence bleed through and false-positive signals. A contrast enhancement and noise reduction filter were applied in ImageJ for publication images.
Counts were done manually using a cell-counter plug-in in ImageJ, in 4-11 non-adjacent sections per mice (N = 2-4) in each condition. Cellular counts per hemi-section were averaged per individual animal, and a grand-mean ± standard error of the mean (SEM) was calculated across animals to produce percentage of cell per hemi-section bar-graphs.
Recordings were obtained using glass suction electrodes attached to lumbar ventral roots (VRs) L2 or L5 on the left and right sides of the cord. VR signals were AC recorded and band-pass filtered from 100 to 1000 Hz (gain 5-10,000), digitized, and acquired with pCLAMP software (version 10; Molecular Devices).
Locomotor analysis. Locomotor frequency (Hz) (defined as the reciprocal of locomotor cycles) was calculated from 3-5 min of activity, taken at least 15 min after the initial burst of drug-induced activity, when locomotor-like activity had stabilized. Locomotor-like activity was analysed using rectified, smoothed (time constant of 0.2 s) and filtered AC signals of VR activity in Spike2 (Cambridge Electronic Design). Statistical significance was determined using two-way ANOVA and Sidak's multiple comparisons test or multiple T test using the Holm-Sidak method, with alpha = 5.000% (p < 0.05). All p-values were adjusted to account for multiple comparisons. Data are expressed as mean ± standard error of the mean (SEM).
For the analysis of left-right and flexor-extensor coordination, 50 random locomotor cycles were selected. Left (l) L2 VR was chosen as the reference trace while right (r) L2 and rL5 VRs were used as test traces. Circular statistics were used to determine the phase relationship between ipsilateral and contralateral VRs during locomotor-like rhythmic activity as previously described 13,17,20 .
The onset of a locomotor burst in the reference trace corresponds to the phase value of 0 in the circular plot and the onset of the next burst corresponds to a phase value of 1. For test traces, a phase value of 0.5 corresponds to alternation while a phase value of 1 (or 0) corresponds to synchrony. The mean timing of the 50 phase values was calculated as a vector in the circular plot where the resulting vector length (r) reflects the concentration of phase values around the mean. Rayleigh's test 67 was used to determine whether r values reached statistical significance (p < 0.05). Points falling inside the inner circle are not significant while those outside the inner circle are significant. Data are expressed as grand mean for each drug concentration. Watson-Williams test 67 was used to determine significant differences amongst conditions (p < 0.05).