RORβ spinal interneurons gate sensory transmission during locomotion to secure a fluid walking gait

SUMMARY

Animals depend on sensory feedback from mechanosensory afferents for the dynamic control of movement. This sensory feedback needs to be selectively modulated in a task- and context- dependent manner. Here, we show that inhibitory interneurons (INs) expressing the RORβ orphan nuclear receptor gate sensory feedback to the spinal motor system during walking and are required for the production of fluid locomotor rhythm. Genetic manipulations that abrogate inhibitory RORβ IN function result in an ataxic gait characterized by exaggerated flexion movements and marked alterations to the step cycle. Inactivation of RORβ in inhibitory neurons leads to reduced presynaptic inhibition and changes to sensory-evoked reflexes, arguing that the RORβ inhibitory INs function to suppress the sensory transmission pathways that activate flexor motor reflexes and interfere with the ongoing locomotor program.

eTOC blurb

Koch et al. identify an inhibitory spinal circuit that is required for fluid rhythmic stepping movements. Inhibitory RORβ⁺ neurons in the spinal cord selectively gate proprioceptive transmission during locomotion by a presynaptic mechanism

INTRODUCTION

The locomotor central pattern generator (CPG) functions as a malleable neural network that is capable of generating simple reflexes or complex motor behaviors such as locomotion (Grillner, 2006). The manner by which this shared neural network generates these different behavioral outputs is not well understood. While reflexive behaviors are largely sensory driven, goal directed movements are controlled by centrally-driven motor programs and descending supraspinal pathways (Armstrong, 1986; Grillner, 2006; Rossignol et al., 2006). Nonetheless, centrally-driven motor behaviors such as locomotion are strongly modulated by sensory feedback (Conway et al., 1987; Dietz, 1992; Hasan and Stuart, 1988; Kiehn et al., 1992; McCrea, 2001; Pearson, 2008; Rossignol et al., 2006; Stein and Capaday, 1988; Windhorst, 2007). One such example is the stumbling corrective reflex, where activating innocuous cutaneous mechanoreceptors on the dorsal surface of the foot during the swing phase of locomotion induces flexion of the lower limb to facilitate clearance of the obstacle (Eng et al., 1994; Forssberg, 1979; Forssberg et al., 1975, 1977; Quevedo et al., 2005). By contrast, stimulation of the plantar surface of the foot during stance phase promotes limb extension and stability (Duysens and Pearson, 1976).

In addition to exteroceptive touch-related sensory signals, the locomotor CPG receives multiple streams of interoceptive information from proprioceptors and joint receptors that are activated when the body and limbs are moving. During ongoing locomotion, this incoming somatosensory information needs to be filtered and gated to prevent abnormal reflexive responses that can disrupt rhythmic stepping movements. Gating occurs at the level of the spinal cord (Rossignol et al., 2006; Zehr and Stein, 1999, and it involves a reduction in proprioceptive reflexes that can be attributed in large part to presynaptic inhibition (Gossard et al., 1991; Rossignol et al., 2006). Presynaptic inhibition also contributes to step cycle-dependent changes in cutaneous transmission (Gossard et al., 1990; Ménard et al., 1999; Rossignol et al., 2006; Rudomin and Schmidt, 1999), thereby limiting the recruitment of cutaneous reflexes during stepping. However, in spite of the overwhelming evidence indicating sensory transmission is dynamically modulated during locomotion, the identity of the spinal neurons that gate this sensory transmission during ongoing locomotion has remained elusive. Moreover, it is unclear to what extent subpopulations of inhibitory INs in the spinal cord have specific roles in gating sensory transmission in a task-dependent manner.

Inhibitory interneurons (INs) in the dorsal and intermediate spinal cord are primarily derived from embryonic dI4/dIL_A progenitors that express the Ptf1α and Pax2 transcription factors (Glasgow et al., 2005; Gross et al., 2002; Müller et al., 2002; Pillai et al., 2011). Many of these inhibitory INs express GAD2 (Betley et al., 2009), and are thought to play pivotal roles in gating sensory transmission to the spinal cord. Recent studies have revealed specific roles for multiple populations of dorsal inhibitory INs in gating itch, noxious mechanical stimuli, and light touch (Bourane et al., 2015a; Duan et al., 2014; Foster et al., 2015; Hilde et al., 2016; Peirs et al., 2015; Petitjean et al., 2015; Ross et al., 2010). This raises the question as to whether there are specialized neurons in the spinal cord that have a dedicated role in gating sensory transmission during locomotion. Although the ablation of GAD2-expressing neurons in the cervical spinal cord markedly perturbs forelimb reaching movements (Fink et al., 2014), their loss did not result in marked changes to locomotion. Consequently, the identity of the neurons in the spinal cord that gate sensory afferent transmission during locomotion remains unknown.

Prior studies have shown that mice lacking the orphan nuclear hormone receptor RORβ develop a duck gait phenotype (Andre et al., 1998; Wiltschko et al., 2015), however, the nature and cellular basis of this deficit is not known. Our previous finding that RORβ is expressed in inhibitory neurons in the dorsal spinal cord (Del Barrio et al., 2013) led us to hypothesize that these spinal RORβ INs modulate the motor output during walking by gating sensory afferent transmission. In this study, we show that inhibitory RORβ IN function is required to restrict flexor motor activity during stepping. The RORβ INs form inhibitory synapses on flexor proprioceptive afferents and inhibit them presynaptically. In RORβ mutant mice, this presynaptic inhibition of sensory transmission is degraded. Taken together, our results reveal that RORβ IN-derived inhibition restricts flexor muscle activity during the swing phase of the step cycle in order to produce a fluid and rhythmic walking gait.

RESULTS

RORβ defines two populations of interneurons in the dorsal and intermediate spinal cord

Our observation that RORβ is expressed in the dorsal spinal cord (Del Barrio et al., 2013) raised the possibility that the loss of RORβ function in these spinal INs may underlie the RORβ mutant locomotor gait deficit (Andre et al., 1998; Wiltschko et al., 2015). To begin testing this, a RORβ^Cre knock-in allele (Harris et al., 2014) was used to gain genetic access and characterize the spinal RORβ INs. First, we verified that the RORβ^Cre allele recapitulates the endogenous expression pattern of RORβ (Figures 1A-C) at various postnatal times. We then counterstained spinal cord sections from P42 RORβ^Cre; Thy1∷LSL-YFP mice with a RORβ in situ hybridization probe to assess co-expression (Figures 1D-G). 84.6 ± 3.8% of the RORβ^Cre-derived YFP⁺ neurons located in laminae IIi-IV and in medial laminae V-VI showed strong co-localization with RORβ mRNA transcripts (Figure 1G), demonstrating RORβ^Cre mediated recombination is highly specific and recapitulates the endogenous pattern of RORβ expression in the spinal cord.

Figure 1. Characterization of RORβ IN subpopulations in the spinal cord.

(A-C) Sections through the lumbar spinal cord of RORβ^Cre; Thy1∷LSL-YFP mice showing the location of the RORβ INs at P0 (A) and P21 (B-C). At P21, the RORβ INs (green) have formed are localized to dorsal laminae V-VI and lamina III (B), with the dorsal subpopulation of RORβ INs located ventral to PKCY⁺ INs in lamina II_i (red) and CGRP⁺ afferents in lamina I (blue) (C). (D) Section through P42 RORβ^Cre; Thy1∷LSL-YFP spinal cord counter-stained with a RORβ in situ hybridization probe (red). (E-F) High magnification images of highlighted sections in (E, lamina III), and (F, lamina V/VI) showing the YFP reporter (green) is largely co-expressed with RORβ (arrowheads). (G) Quantification of the overlap between YFP-positive neurons and RORβ mRNA positive neurons (n=4 cords). (H) Schematic of RORβ IN subpopulations. (I-K) Transverse section through a P42 RORβ^Cre; Thy1∷LSL-YFP spinal cord showing YFP (green) is co-expressed with GAD1 and GlyT2 (red). (J-K) High magnification images from (I) showing coexpression of YFP and GAD1/GlyT2 mRNA (arrowheads) in lamina III (J) and laminae V-VI (K). (L) Quantification of excitatory and inhibitory marker expression (see also Figure S1). RORβ^Cre; R26^LSL-HTB; GAD67∷GFP and RORβ^Cre; R26^LSL-HTB; GlyT2∷GFP mice were used for the GAD1⁺/RORβ⁺ and GlyT2⁺/RORβ⁺cell counts, respectively. (M) Transverse section through a P42 RORβ^Cre; Thy1∷LSL-YFP spinal cord showing GAD2 mRNA (red) expression in lamina V/VI RORp INs (green). Inset shows high magnification image of the overlap between RORβ-YFP and GAD2 mRNA in lamina V (arrowheads). (N) Quantification of YFP⁺/GAD2⁺ INs in lamina III and laminae V/VI. Scale Bar: 100 pm (A-D, I), 50 μm (E-F, J-K, M (insert)). See also Figure S1.

In view of our previous finding that the RORβ IN population comprises a mixture of excitatory and inhibitory cells types (Del Barrio et al., 2013), we examined the neurotransmitter phenotype of the RORβ INs in laminae IIi-IV and laminae V-VI in more detail. In sections from P42 RORβ^Cre; Thy1∷LSL-YFP mice, 58.5 ± 4.2% of the RORβ INs expressed the inhibitory marker Pax2, whereas 43 ± 6.6% of all RORβ cells expressed the excitatory marker vGluT2 (Figure 1L, Figures S1A and S1B). These excitatory RORβ INs were primarily located in laminae IIi-IV, and they have been shown to co-express RORα (Del Barrio et al., 2013) By contrast, inhibitory Pax2+ RORβ INs were found in laminae IIi-III and in lamionae V-VI (Figure S1G).

In situ co-localization with probes to GAD1 (GAD67) and GlyT2 revealed that 59.8 ± 7.8% of the spinal RORβ INs expressed either GAD1 and/or GlyT2 (Figures 1I-L), demonstrating the inhibitory RORβ INs comprise a mixture of glycinergic and GABAergic neurotransmitter phenotypes. Most importantly, we found that these inhibitory RORβ INs are largely distinct from other previously characterized inhibitory populations in the dorsal spinal cord, which are marked by the expression of Satb2, galanin, parvalbumin, dynorphin and NPY (Figures S1C-F, S1H).

We were able to further subdivide the inhibitory RORβ INs according to their location and expression of the inhibitory GABA transporter GAD2 (GAD65). Whereas the majority (82.9 ± 2.9%) of the RORβ INs in laminae V/VI expressed GAD2 transcripts, only 20.2 ± 2.1% of the RORβ INs in laminae IIi-IV co-localized with GAD2 (Figure 1M-N). Sparse labeling of the RORβ INs with EnvA pseudotyped rabies virus (Bourane et al., 2015b) revealed that the RORβ INs in laminae IIi-IV possessed both excitatory (vertical, central) and inhibitory (islet) cell morphologies (Figures S1I and S1K; Grudt and Perl, 2004). By contrast, those in laminae V–VI were exclusively islet-like (Figures S1J and S1K), which is consistent with their inhibitory phenotype. RORβ therefore defines two phenotypically diverse populations of spinal INs: the RORβ INs in laminae III–IV comprising a mixture of excitatory and inhibitory phenotypes and cells in laminae V-VI that predominantly display a GAD2⁺ inhibitory phenotype.

Inactivation of RORβ in Pax2 inhibitory neurons recapitulates the global RORβ knockout motor phenotype

Mice lacking RORβ display a duck gait phenotype and abnormal clasping reflexes (Andre et al., 1998; Wiltschko et al., 2015). While this behavioral phenotype is consistent with a deficit in spinal sensorimotor gating, RORβ is expressed outside of the spinal cord, most prominently in excitatory pyramidal neurons in neocortex (Andre et al., 1998; Nakagawa and O’Leary, 2003; Schaeren-Wierners et al., 1997). To test whether RORβ activity is required in spinal inhibitory neurons for normal locomotor function, multiple Cre drivers were used to recombine and inactivate a conditional floxed allele of RORβ (RORβ^fl/fl, Figures S2A and 2B) in different neuronal populations. Specifically, RORβ was inactivated in: (a) all neurons (Nestin∷Cre; (Graus-Porta et al., 2001; Tronche et al., 1999), (b) in excitatory neurons throughout the cerebral cortex (Emx1^Cre; (Gorski et al., 2002)), (c) in excitatory spinal cord INs and Purkinje cells that express RORα (RORα^Cre; (Bourane et al., 2015b; Chou et al., 2013)), and (d) in inhibitory spinal INs (Pax2∷Cre; (Ohyama and Groves, 2004)).

Pan-neuronal inactivation of RORβ in Nestin∷Cre; RORβ^fl/− mice resulted in a duck gait phenotype that closely matched that seen in the germline RORβ knockout mice, indicating the locomotor phenotype is neuron-specific (c.f. Figures 2B and 2C). Deleting RORβ from excitatory cortical neurons using Emx1^Cre failed to recapitulate the RORβ mutant motor phenotype (Figure 2D). There was also no marked change in locomotion when RORα^Cre was used in combination with the RORβ^fl/fl allele to selectively delete RORβ from RORα excitatory neurons (Figure 2E). By contrast, inactivating RORβ in Pax2⁺ inhibitory INs (Pax2-RORβ) produced a strong locomotor duck gait phenotype (Figure 2F), similar to that seen in germline RORβ mutant mice (Figure 2B). These results indicate the duck gait phenotype is due to the loss of RORβ function in spinal inhibitory neurons, and it is consistent with our hypothesis that the gating of sensory afferent transmission is reduced when the RORβ mutant mice are locomoting.

Figure 2. Inactivating RORβ in spinal inhibitory interneurons causes hyperflexion of the hindlimbs during stepping.

(A-F) Still images from high-speed kinematic videos showing the position of the left hindlimb at mid-swing phase with corresponding schematics of genetic expression patterns. The mice shown have the following genotypes: (A) control, (B) RORβ^−/−; (C) Nestin∷Cre; RORβ^fl/−, (D) Emx1^Cre; RORβ^fl/fl, (E) RORα^Cre; RORβ^fl/fl, (F) Pax2∷Cre; RORβ^fl/fl. The schematics associated with each panel show Cre expression, except for (B), which depicts the expression pattern of RORβ. Control mice show a normal walking gait (A, see Movie S1). Note the duck-gait phenotype when RORβ is inactivated in inhibitory Pax2⁺ INs (F, see Movie S2), but not in RORα^Cre-derived INs or in cortical Emx1^Cre-derived neurons (D and E). See also Figure S2.

Previous studies have implicated RORβ in axon guidance, neuronal migration and synaptic plasticity (Jetten and Joo, 2006). To assess whether the duck gait phenotype might be due to a non-cell autonomous effect on sensory axon innervation we examined the trajectory and spinal termination patterns of sensory afferents in the Pax2-RORβ mutant spinal cord the trajectory. No detectable changes in sensory innervation were observed in the Pax2-RORβ mutant spinal cord (Figures S2C-S2E’). Moreover, there was no difference in the number and distribution of dorsal IN cell types following the deletion of RORβ from spinal inhibitory INs as assessed by a battery of cell type specific markers (Figure S2F).

Inactivating RORβ IN function phenocopies the RORβ mutant phenotype

Two complementary genetic approaches were then employed to directly test the contribution RORβ IN dysfunction makes to the duck gait phenotype. First we used RORβ^Cre; R26^LSL-TeNT mice to selectively express the tetanus light chain subunit (TeNT; (Zhang et al., 2008 and block synaptic transmission in all neurons that express RORβ. These RORβ^Cre; R26^LSL-TeNT mice exhibited a duck gait phenotype that was largely indistinguishable from that seen in the RORβ null mutants (Figures 3A and 3C-E, Figure S3). The second approach used RORβ^Cre; hCdx2∷FlpO; Tau^ds-DTR mice to selectively ablate the RORβ INs in the caudal spinal cord, (Bourane et al., 2015b; Britz et al., 2015). Following DTX-treatment, which resulted in an 86.0 ± 1.7% reduction in the number of RORβ INs at lumbar levels, we observed a strong duck gait phenotype (Figures 3B-E, Figure S3). Taken together, these findings demonstrate that abrogating RORβ inhibitory IN function in the spinal cord is the likely cause of the motor deficits seen in RORβ null and Pax2-RORβ conditional knockout mice.

Figure 3. Inactivating spinal RORβ interneurons recapitulates the RORβ mutant motor phenotype.

(A) Quantification of maximum and minimum hip joint angles during stepping for Pax2-RORβ mutant, RORβ IN-ablated (RORβ^Cre; hCdx2∷FlpO; Tau^ds-DTR) and RORβ IN-silenced (RORβ^Cre; R26^LSL-TeNT) mice and their respective littermate controls (Pax2-RORβ, control: n=8; mutant: n=6; RORβ^Cre; hCdx2∷FlpO; Tau^ds-DTR, control: n=5; mutant: n=5; RORβ^Cre; R26^LSL-TeNT: control: n=5 ; mutant: n=5). In all instances, the minimum hip angle of was significantly decreased compared to their littermate controls (*p < 0.05, **p < 0.01, ***p < 0.001) (B) Quantification of maximum and minimum ankle joint angles during stepping for Pax2-RORβ, RORβ^Cre; hCdx2∷FlpO; Tau^ds-DTR and RORβ^Cre; R26^LSL-TeNT mice and their respective littermate controls. The minimum ankle angle of the Pax2-RORβ mutants was significantly decreased compared to controls (***p < 0.001). (C) Representative stick figure diagrams showing two complete hindlimb step cycles (swing and stance) for (top to bottom): control (grey), Pax2-RORβ (red), RORβ^Cre; hCdx2∷FlpO; Tau^ds-DTR (green), and RORβ^Cre; R26^LST-TeNT (blue) mice. Asterisks indicate the flexion phase. (D) Still image from a high-speed kinematic video showing the position of the left hindlimb at midswing phase after synaptic silencing of RORβ INs in a RORβ^Cre; R26^LST-TeNT mouse. (E) Still image from a high-speed kinematic video showing the position of the left hindlimb at midswing phase two weeks after ablation of caudal RORβ INs in a RORβ^Cre; hCdx2∷FlpO; Tau^ds-DTR mouse. See also Figure S3.

Pax2-RORβ mice show an increase in flexor muscle activity during walking

To assess the nature of the duck gait phenotype in more detail, high speed kinematic analyses (Britz et al., 2015; Pearson et al., 2005; Zhang et al., 2008) were used to track the hip and ankle joints throughout the step cycle and measure the angular changes and duration of each step cycle phase in the Pax2-RORβ mutant mice as compared to their littermate controls (see Movies S1 and S2). During the swing (flexion) phase, the minimum angle of the hip and ankle joints were significantly more acute in the Pax2-RORβ mutant mice compared to their control littermates (Figures 3A–3C; hip: control: 46.1 ± 3.1° Pax2-RORβ mutant: 34.2 ± 2.9°, p < 0.05; ankle: control: 30.5 ± 2.9°, Pax2-RORβ mutant: 9.5 ± 1.5°, p < 0.05; genotype: p < 0.05; control: n=8, Pax2-RORβ mutant: n=6). This is consistent with the pronounced hyperflexion of the hindlimb that occurs when the Pax2-RORβ mutant mice are walking (Figure 2). Most importantly, we observed no hyperflexion of the hindlimbs at rest, indicating the RORβ INs only gate flexor motor activity during ongoing locomotion. Similar changes in the minimum hip and ankle joint angles were observed in both the RORβ-TeNT mice and RORβ^Cre IN-ablated mice (Figures 3A–3E), providing further evidence that the cell autonomous loss of spinal RORβ IN function is the primary cause of the duck gait phenotype.

The hyperflexion phenotype was further confirmed by EMG recordings showing a marked increase in the duration of flexor TA activity in Pax2-RORβ mutant mice during walking (Figure S3A). In keeping with this selective increase in TA flexor muscle activity, the duration of the swing phase was increased with little or no effect on the duration of stance phase (Figure S3B; swing: control, 0.1 ± 0.01 sec, n=8; Pax2-RORβ mutant, 0.3 ± 0.07 sec, n=6; swing phase: p < 0.05, control vs Pax2-RORβ mutant). Interestingly, while RORβ^Cre; R26^LSL-TeNT mice displayed a prolongation of the swing phase similar to that seen in the Pax2-RORβ mutant mice (Figures S3B and S3C), this was less pronounced in the RORβ^Cre IN-ablated mice (Figure S3D), possibly due to the incomplete killing of RORβ INs in the lumbar cord. Although we did observe a trend toward an increase in swing phase duration, this did not reach statistical significance due to the marked step-to-step variability in the trajectory of the hindlimb when these mice were walking.

Interestingly, inactivating RORβ also caused the pelvis and lower body to become elevated during walking, indicating the opposing hindlimb is overextended during walking (Figure 2F and 3C, Figure S3). While the Pax2-RORβ mutant mice did show a trend toward an increase in the maximum joint angle for the hip and ankle (hip: control: 100.6 ± 5.7° mutant: 112.6 ± 9.6° ankle: control: 112.6 ± 4.3° Pax2-RORβ mutant: 123.6 ± 5.0°), these changes they did not reach statistical significance.

Pax2-RORβ mice display a decreased threshold for sensory afferent evoked reflexes

The increased flexor activity seen during locomotion in mice when RORβ IN function is abolished, together with experiments showing that the RORβ INs are innervated by proprioceptors and low-threshold mechanoreceptors (Figure S4), prompted us to ask if there is a change in sensory input-motor output function in the Pax2-RORβ mutant spinal cord. To test this, recordings of dorsal root-evoked ventral root potentials (VRPs) were made from P8 control and Pax2-RORβ mutant spinal cords to measure the threshold and amplitude of the evoked ventral root response (Figure 4A-D). Pax2-RORβ mutant spinal cords displayed normal monosynaptic and polysynaptic VRP responses to dorsal root stimulation, both with respect to amplitude and latency (Figures 4B and C), indicating the loss of RORβ does not alter postsynaptic excitatory transmission from sensory afferents to motor neurons. However, in the Pax2-RORβ mutant cord these VRPs were evoked with lower amplitude stimulation (Figure 4D). This suggests that motor neurons are recruited by lower intensity afferent stimulation in the Pax2-RORβ mutant cord as compared to control cords (VRP threshold: control: 7.8 ± 0.37 μA, n=5; Pax2-RORβ mutant: 4.7 ± 0.9 μA, n=6; p < 0.05). This change in threshold is consistent with reduced presynaptic inhibition of sensory afferent transmission in the Pax2-RORβ mutant, raising the possibility that the increased flexor activity that the Pax2-RORβ mutant mice display is due to a loss of presynaptic inhibition.

Figure 4. Inactivation of RORβ in inhibitory interneurons disrupts motor neuron excitability and presynaptic inhibition.

(A) Recorded VRP traces of dorsal root-evoked VRPs for P8 control and Pax2-RORβ mutant spinal cords. (Inset) Schematic of recording set up. (B-D) Quantification of the area under the curve of the VRP (B). Panels C and C show latency to the polysynaptic VRP (C), and threshold for VRP recruitment (D) (control: n=6 cords; Pax2-RORβ mutant: n=5 cords). Note the lower threshold for VRP recruitment in the Pax2-RORβ mutant cord. (E) Recorded traces showing primary afferent fiber depolarization (PAD) in P8 control and Pax2-RORβ mutant cords. (Inset) Schematic of recording set up. (F-G) Quantification of the amplitude of the PAD (F) and threshold to PAD recruitment (G) (control: n=10; Pax2-RORβ mutant: n=9). The Pax2-RORβ mutant cords displayed decreased PAD compared to control cords. (H) Quantification of the number of parvalbumin (PV) and vGluT 1 double positive terminals that are contacted by GAD2⁺ boutons in the lumbar cord (laminae V-VI) of P14 control and Pax2-RORβ mutant mice (control: n=4; Pax2-RORβ mutant: n=4). *p < 0.05, **p < 0.01. Scale bar as marked. See also Figure S4.

To investigate whether changes in presynaptic inhibition underlie the RORβ mutant motor phenotype, we examined the contribution RORβ INs make to presynaptic inhibition as measured by evoked dorsal root reflexes, which we term primary afferent depolarization or PAD. Upon dorsal root stimulation, the cords of Pax2-RORβ mutant mice exhibited reduced levels of PAD compared to control littermate mice with no change in threshold or excitability (mean mutant PAD: 52.6 ± 5.9% of control; control mice: n=7, mutant mice: n=9; p < 0.01; Figures 4E-G). In keeping with the reduction in evoked PAD in the Pax2-RORβ mutant cord, the number of parvalbumin⁺ (PV⁺) afferent terminals that are contacted by GAD2⁺ boutons in laminae V-VI was greatly reduced in the Pax2-RORβ cord (Figure 4H; control: 64.6 ± 8.5, n=4; Pax2-RORβ mutant: 33.3 ± 3.1, n=4; p < 0.05).

RORβ INs provide presynaptic inhibition to myelinated sensory afferents

To further investigate whether the RORβ INs are a source of spinal presynaptic inhibition, we measured light evoked PAD in RORβ^Cre; R26^LsL-Ai32 spinal cords that selectively express channelrhodopsin (ChR2) in RORβ INs. Kynurenic acid was used in combination with mephenesine to block glutamatergic neurotransmission and ensure all recorded potentials were monosynaptic in nature (Ziskind-Conhaim, 1990). This removed the confound of PAD being activated indirectly by descending corticospinal pathways or by the excitatory RORβ INs that coexpress RORα (Del Barrio et al., 2013). Optogenetic stimulation (470 nm) of P10-14 RORβ^Cre; R26^LSL-Ai32 spinal cords produced back-propagating PAD currents in the L5 dorsal root, which are characteristic of presynaptic inhibition (Figure 5A and B). These light-evoked RORβ PAD currents (28.9 ± 5.9μV) were blocked by the GABAergic antagonist bicuculline (Figure 5C, n=7), which is consistent with a GABAergic presynaptic mechanism.

Figure 5. Inhibitory RORβ interneurons presynaptically inhibit myelinated flexor afferents.

(A) Recorded traces of primary afferent fiber depolarization (PAD) from isolated spinal cords of RORβ^Cre; R26^LSL-Ai32 mice. (Inset) Schematic of recording set up. Optogenetic activation of RORβ INs (light blue trace) evokes PAD, highlighting RORβ IN-induced presynaptic inhibitory control of primary afferents. (B) Quantification of PAD amplitude following optogenetic activation of RORβ INs at 23°C (n=6 cords). (C) Quantification of RORβ optogenetically-induced PAD at baseline, after application of the GABAergic antagonist bicuculline, and after drug washout, showing RORβ IN presynaptic inhibitory PAD is GABA_AR mediated (n=7 cords). (D) Quantification of PAD amplitude following optogenetic activation of RORβ INs at 23°C (left), at 33°C (middle) and after normalization to baseline 23°C recording temperature (right) (n=6 cords). The decrease in amplitude following temperature increase suggests inhibition is predominantly onto large myelinated afferents (see text). (E-F) Transverse spinal section of P14 RORβ^Cre; Thy1∷LSL-YFP immunostained with antibodies raised against GAD2 (red) and PV (blue) highlighting RORβ axo-axonic inhibitory synapses onto myelinated afferents (arrows). (G) Quantification of Thy1⁺/VGAT⁺ double positive contacts onto CTb⁺ labeled afferent terminals from the hip flexor (iliopsoas) or hip extensor (biceps femoris) respectively. Counts were performed in the intermediate lumbar cord of P14 RORβ^Cre; Thy1∷LSL-YFP mice (flexor: n=3; extensor: n=3; p < 0.001). RORβ INs preferentially target hip flexor afferents over hip extensors. 2T, 2x threshold for PAD recruitment; DR, dorsal root. Scale Bar: 5 pm (E-F), and as marked in A. See also Figure S5.

Previous studies have shown that recording evoked PAD currents at 33°C and at 23°C can be used to distinguish between presynaptic inhibition of cutaneous versus proprioceptive afferents (Barron and Matthews, 1938; Eccles et al., 1961; Fink et al., 2014). To assess the relative contribution the RORβ INs make to cutaneous versus proprioceptive presynaptic inhibition, a series of recordings were performed at 33°C to isolate the residual cutaneous afferent component of PAD. At 33°C light-evoked RORβ-induced PAD was reduced to near zero as compared to the same preparation recorded at 23°C (Figure 5D, n=6 cords). This loss of PAD was fully reversed when the spinal cord preparation was returned to 23°C (Figure 5D), indicating the decrease in amplitude at higher temperatures is not due to the deterioration of the signal or to associated changes in the spinal circuitry. It therefore appears that RORβ IN-evoked presynaptic inhibition is largely directed toward large myelinated sensory axons, which are primarily proprioceptors.

In the spinal cord, inhibitory presynaptic terminals apposed to PV⁺ proprioceptive afferents express high levels of the GAD2 protein (Betley et al., 2009). Our observation that the RORβ INs located in laminae V-VI express high levels of GAD2 (Figure 1N), together with the evidence that the RORβ INs presynaptically inhibit myelinated afferents (Figure 5D), prompted us to search for RORβ IN-derived GAD2⁺ inhibitory presynaptic terminals on PV⁺ afferents in medial laminae V-VI, where many PV⁺ proprioceptive afferents are located. Multiple PV⁺ processes were found to be decorated with RORβ IN-derived GAD2⁺ terminals (Figures 5E and 5F), providing further evidence that the RORβ INs contribute to the presynaptic inhibition of proprioceptive sensory afferents in the lumbar spinal cord. Interestingly, we observed a strong bias with respect to RORβ-derived synaptic contacts onto hip flexor and extensor muscle afferents, with approximately twice as many contacts on iliopsoas (hip flexor) afferents as compared to biceps femoris (hip extensor) afferents (Figure 5G). We also observed a reduction in the number of GAD2⁺ contacts onto CTb-labeled sensory afferents in the Pax2-RORβ mutant cord (Figures S5A-B). Interestingly, injection of AAV2/1-hSyn-DIO-sypHTomato virus into the lumbar cord of RORβ^Cre mice revealed relatively few RORβ IN-derived inhibitory contacts on V0c INs, motor neurons and Chx10⁺ V2aINs (Figures S5C-E), indicating the loss of postsynaptic inhibition may not make a strong contribution to the duck gait phenotype.

Selective inactivation of TrkB in RORβ neurons recapitulates the RORβ mutant duck gait phenotype

In view of prior studies showing: 1) inhibitory RORβ INs in laminae V/VI express high levels of GAD2 (Figure 1) and form presynaptic contacts onto proprioceptive afferents (Figures 5E-F, see also Figure 6F), 2) GAD2 is enriched in presynaptic GABAergic terminals onto proprioceptive afferents (Betley et al., 2009) and 3) presynaptic proprioceptive synapse differentiation is dependent on BDNF/TrkB signaling (Betley et al., 2009; Fink et al., 2014), we asked whether disrupting inhibitory presynaptic terminal differentiation the RORβ INs might also precipitate the duck gait phenotype. RORβ and TrkB mRNA show overlapping expression throughout the dorsal spinal cord, including extensive co-expression in medial laminae V-VI (Figure 6A). In RORβ^Cre; TrkB^fl/fl mice where TrkB is selectively from RORβ-expressing neurons, we observed a large decrease in the number of GAD2⁺ boutons on vGluT1⁺ afferents in laminae V-VI (Figure 6G-I; control: 84.4 ± 5.3 vGluT1⁺ terminals with GAD2⁺ boutons, n=3; RORβ^Cre; TrkB^fl/fl mutant: 54.7 ± 0.8 vGluT1⁺ terminals with GAD2⁺ boutons, n=3 cords for each genotype, p < 0.001). This loss of GAD2⁺ presynaptic boutons on proprioceptive afferents was accompanied by a pronounced duck gait phenotype (Figure 6B and C). Decreases in the minimum hip and ankle joint angles were seen during the swing phase of locomotion (Figure 6D-E; minimum hip angle: control: 50.5 ± 2.4° RORβ^Cre; TrkB^fl/fl mutant: 25.2 ± 4.8° p < 0.001; minimum ankle angle: control: 25.7 ± 5.1°, n=5; RORβ^Cre; TrkB^fl/fl mutant: 5.4 ± 1.7° n=5; p < 0.01). Changes in the maximum hip angle (control: 103 ± 2.2° RORβ^Cre; TrkB^fl/fl mutant: 97.13 ± 12.7° p= 0.7) and maximum ankle angle (control: 103.2 ± 6.5° RORβ^cre; TrkB^fl/fl mutant: 132.4 ± 11.1° p= 0.06) were less pronounced and in line with the changes observed in the Pax2-RORβ and RORβ^Cre; R26^LSL-TeNT mutant mice. Notably, there were no marked changes in cutaneous sensory reflexes in the RORβ^Cre; TrkB^fl/fl mice (Figures S5A-C Taken together, these findings suggest that the loss of RORβ-derived inhibitory presynaptic contacts onto proprioceptors is sufficient to recapitulate the RORβ mutant duck gait phenotype.

Figure 6. Inactivation of TrkB in RORβ neurons recapitulates the RORβ mutant duck gait phenotype.

(A) Transverse section through a P42 RORβ^Cre; Thy1∷LSL-YFP spinal cord showing TrkB mRNA expression (red). Inset shows high magnification image of the overlap between RORβ- YFP and TrkB mRNA in lamina V (arrowheads). (B-C) Still images from high-speed kinematic videos showing the position of the left hindlimb at midswing phase in (B) control and (C) after selective deletion of the BDNF receptor TrkB from RORβ INs using RORβ^Cre; TrkB^fl/fl mice. (D-E) Quantification of maximum and minimum joint angles in the hip (D), and ankle (E) during locomotion in control and RORβ^Cre; TrkB^fl/fl mice, showing hyperflexion of the hindlimb as seen with RORβ mutant mice (n=5 for each genoptype). (F) Schematic of TrkB-expressing RORβ INs, which presynaptically target BDNF-expressing proprioceptors. (G-H’) Transverse spinal sections of control (G), and RORβ^Cre; TrkB^m (H) mice immunostained with antibodies to GAD2 (red) and vGluT1 (blue). Note the GAD2⁺ inhibitory contacts onto large vGluT1 terminals in laminae V-VI (arrowheads). (I) Number of vGluT1⁺ terminals containing GAD2⁺ boutons in laminae V-VI of P42 control and RORβ^Cre; TrkB^fl/fl spinal cords (control: n=3; mutant: n=3). YFP immunofluorescence (green) was visualized without amplification. *p < 0.05, **p < 0.01, ***p < 0.001. Scale Bar: 5 pm (G-H’).

Attenuating low threshold afferent input partially rescues the RORβ motor phenotype

The decrease in presynaptic inhibition in the Pax2-RORβ mutant mice (Figure 4), together with the phase specificity of the Pax2-RORβ phenotype (Figure 3, Figure S3), suggested to us the RORβ motor phenotype might be due to a loss of sensory afferent gating resulting in increased peripheral drive to the locomotor CPG. To assess this possibility, we blocked proprioceptive and cutaneous afferent input from the hindlimb by injecting a local anesthetic (2% lidocaine, 0.2% QX-314) bilaterally in the perisciatic space (Figure 7A-B). 10 minutes after administering the perisciatic block, the minimum joint angle for the hip in the Pax2-RORβ mutant increased significantly (hip: control baseline: 46.1 ± 3.1°, control post anesthetic: 48.1 ± 3.9° n=5; mutant baseline: 34.2 ± 2.9°, Pax2-RORβ mutant post anesthetic: 45.1 ± 3.3°, n=5; Pax2-RORβ mutant baseline vs post anesthetic: p < 0.05; Figure 7C), suggesting pharmacological blockade of sensory transmission can attenuate the Pax2-RORβ mutant limb hyperflexion phenotype. Although efferent transmission may have been affected by the perisciatic block, our analyses were only performed when the mice were fully mobile. This suggests that the pharmacological rescue we observe is primarily due to the attenuation of proprioceptive afferent transmission.

Figure 7. Blocking peripheral nerve transmission attenuates the RORβ mutant duck gait phenotype.

(A-B) Still images from high-speed kinematic videos showing the position of the left hindlimb at midswing phase in a Pax2-RORβ mutant mouse at (A) baseline and (B) 10 minutes after perisciatic anesthetic injection. (C) Quantification of the maximum and minimum hip joint angles in control and Pax2-RORβ mutants at baseline and 10 minutes after applying either a cutaneous or perisciatic blockade. Pax2-RORβ mutants show a normalization of hip joint movements after perisciatic anesthetic injection (control: n=5; Pax2-RORβ mutant: n=5). (D) Quantification of step cycle duration after sensory blockade in control and Pax2-RORβ mutants at baseline and 10 minutes following perisciatic blockade. Peripheral sensory blockade significantly decreases the duration of the swing phase in Pax2-RORβ mutant mice compared to their baseline measurements. *p < 0.05, **p < 0.01

The elongation of the swing phase duration seen in the Pax2-RORβ mutant mice was also reduced to control levels (swing phase duration: control baseline, 0.1 ± 0.01 sec; control post anesthetic: 0.09 ± 0.01 sec, n=5; Pax2-RORβ mutant baseline, 0.3 ± 0.07 sec, Pax2-RORβ mutant post anesthetic: 0.1 ± 0.03 sec, n=5; mutant baseline vs post anesthetic: p < 0.05; Figure 7D). By contrast, cutaneous blockade of the paw by means of a topical anesthetic, while sufficient to block sensory cutaneous reflexes, did not significantly affect the locomotor pattern or swing phase deficits that occur in the Pax2-RORβ mutant mice (Figure 7C). This conclusion is consistent with our observation that the RORβ INs form multiple presynaptic inhibitory contacts onto PV⁺ afferents (Figure 5E-F), the majority of which are proprioceptive, and that GAD2⁺ presynaptic contacts onto PV⁺ afferents are markedly reduced when RORβ is inactivated in Pax2⁺ inhibitory INs (Figure 4H).

DISCUSSION

This study identifies an inhibitory IN population in the spinal cord that selectively gates sensory transmission during locomotion. Although previous studies have shown that synaptic transmission is strongly modulated by inhibition during locomotion (Gosgnach et al., 2000; Gossard, 1996; Gossard et al., 1991; Hultborn et al., 1987; Perreault et al., 1999; Rossignol et al., 2006), the nature and source of this inhibition remained unknown. We now show the inhibitory Pax2-RORβ mutant have a dedicated and specific role in modulating sensory input to the spinal motor system during locomotion, and that this inhibition is necessary for fluid stepping movements. The abrogation of RORβ function, either by ablating RORβ INs in the spinal cord or by blocking neurotransmission with TeNT, argues that a cell autonomous defect in inhibitory spinal RORβ INs underlies the duck gait phenotype. Our demonstration that PAD is reduced in the Pax2-RORβ mutant spinal cord (Figure 4), and the RORβ INs presynaptically inhibit myelinated sensory afferents (Figure 5 and Figure 6), argue that the RORβ INs are an essential source of the presynaptic inhibition that gate sensory afferent transmission during locomotion.

Modulation of sensory transmission by RORβ interneurons

Our findings suggest RORβ function is required for the integrity of a RORβ IN-proprioceptive afferent presynaptic inhibitory circuit. This conclusion is supported by: 1) the reduction in the number of presynaptic inhibitory terminals on PV⁺ myelinated afferents in the Pax2-RORβ mutant cord (Figure 4H), 2) the accompanying decrease in the amplitude of evoked PAD (Figure 4F) and 3) our demonstration that abolishing TrkB signaling in RORβ inhibitory INs produces a hyperflexion phenotype and reduces the number of GAD2⁺ contacts on vGluT1⁺ afferents (Figures 6B-I). Taken together, these findings reveal a defect in the formation and functioning of RORβ presynaptic inhibitory synapses in the RORβ mutant spinal cord, and they argue that the loss of RORβ⁺/GAD2⁺ IN-derived presynaptic inhibition is the likely cause of the the duck gait phenotype. Although it appears that the primary mechanism by which the RORβ INs regulate hindlimb movement during locomotion is by presynaptically gating proprioceptive afferent transmission, we cannot rule an additional contribution from RORβ IN-derived postsynaptic inhibition. Many of the RORβ INs are glycinergic (Figure 1), and the RORβ INs form sparse inhibitory postsynaptic contacts onto other ventral neurons including V0c and V2a glutamatergic premotor INs and motor neurons (Figure S5).

The observation that blocking cutaneous transmission fails to reverse the RORβ knockout duck gait/hyperflexion phenotype (Figure 7) is consistent with RORβ IN mediated inhibition being primarily directed at proprioceptive transmission pathways. However, since it was not possible to selectively block proprioceptive transmission in Pax2-RORβ mutant mice while leaving cutaneous sensory pathways intact, the contribution that RORβ IN inhibition of cutaneous afferents makes to modulating limb movements remains to be determined. Our finding that the RORβ⁺/GAD2^high INs contribute to the gating of proprioceptive input during locomotion is in general agreement with the finding of Fink et al. (2014) that GAD2-expressing neurons are critical for controlling proprioceptive feedback to facilitate smooth reaching movements. However, while they described defects in a skilled forelimb motor task, we observed a hindlimb- specific locomotion phenotype. Aside from the different behavioral assays used, differences in the motor circuitry are likely to underlie the distinct behavioral deficits. One such example is the C4 propriospinal-lateral reticular nucleus circuit that is necessary for skilled forelimb reaching movements (Alstermark and Isa, 2012; Azim et al., 2014). It is also known that the rodent cervical cord receives comparatively more innervation from the cortex than the lumbar spinal cord (Gribnau et al., 1986), with broader areas of the cortex innervating cervical versus lumbar segments (Kamiyama et al., 2015).

Presynaptic inhibitory circuits in the spinal cord

Presynaptic inhibition provides a powerful mechanism for gating of sensory feedback from different sources, and is thus well suited to regulating the central trafficking of sensory information in a state- and task-dependent manner (Gaudry and Kristan, 2009; Rudomin and Schmidt, 1999). This has led to the suggestion that discrete subpopulations of GABA⁺ neurons may be recruited to inhibit particular sensory transmission pathways in a task-dependent manner. In ablating the GAD2⁺ GABA neurons in the cervical spinal cord, Fink et al. (2014) saw both altered reaching movements and increased scratching, indicating deleting the GAD2 population affects both propriospinal and cutaneous afferent transmission. Our results reveal a more selective behavioral deficit when RORβ-derived inhibition is attenuated. Whereas flexion movements during walking are disrupted in the Pax2-RORβ knockout mice (Figure 2), scratching and reflexive responses to von Frey, pinprick and temperature remain unchanged (Figure S6; SCK, data not shown).

The selective nature of the RORβ mutant phenotype in which the RORβ INs gate flexion movements during walking, but not at rest, underlines the precise role that the RORβ INs play in gating sensory feedback during ongoing locomotion. It also reveals a high degree of functional specialization in the inhibitory IN circuits that modulate sensory feedback to the spinal cord. Support for select inhibitory spinal IN populations having dedicated roles in gating specific sensory modalities comes from recent studies showing dorsal horn inhibitory neurons that express dynorphin gate noxious mechanical stimuli (Duan et al., 2014), while those that express bHLHb5 gate chemical itch (Kardon et al., 2014; Ross et al., 2010). Moreover, Bourane et al. (2015a) have shown inhibitory neurons that developmentally express neuropeptide Y gate low- threshold mechanical stimuli and suppress mechanical itch, but they do not inhibit chemical itch or noxious mechanical pathways, nor do they regulate locomotion.

Regulation of RORβ IN activity during locomotion

An important question going forward is how the RORβ IN inhibitory circuit is recruited during locomotion to gate proprioceptive inputs to the spinal cord. Given the demonstrated role of RORβ IN inhibition in attenuating flexor motor activity during stepping, it seems likely that the RORβ INs are biased in the actions on flexor- versus extensor-derived afferents. In support of this, we find greater numbers of RORβ-derived inhibitory contacts on iliopsoas sensory afferents as compared to biceps femoris afferents (Figure 5G). Sensory afferents are differentially active during the step cycle (Gossard et al., 1990, Hayes et al., 2012; Pilyavskii et al., 1988) with phasic antidromic discharges in flexor and bifunctional flexor/extensor afferents being maximal during the flexion phase (Gossard et al., 1991). It is therefore worth noting that the RORβ INs are innervated by low-threshold cutaneous mechanoreceptors and by proprioceptors (Figure S5), and that dorsal root potentials in the rat have been strongly correlated to phasic changes in primary afferent input during locomotion (Yakhnitsa et al., 1988). Finally, the RORβ INs are innervated by local excitatory vGluT2⁺ INs (data not shown) indicating RORβ IN inhibition may be modulated, either directly or indirectly, by the CPG during the step cycle. This would be consistent with previous studies showing central pathways, including the locomotor CPG, regulate presynaptic inhibition (Eccles et al., 1962a, b; Rossignol et al., 2006; Rudomin and Schmidt, 1999).

Our demonstration that the RORβ INs receive inputs from multiple low-threshold afferents and in turn presynaptically inhibit proprioceptors is consistent with their involvement in the flexion reflex (Eccles et al., 1962b; Gossard et al., 1991; Jankowska and Riddell, 1995; Lundberg et al., 1987; Perreault et al., 1999; Riddell et al., 1995; Schomburg et al., 1998; Figure S4). Interestingly, recordings of dorsal root-evoked ventral root potentials from Pax2-RORβ mutants revealed lowered thresholds for polysynaptic potentials, but not monosynaptic evoked potentials (Figures 4A-D). This suggests that the inhibitory actions of the RORβ INs are primarily on group Ib, II and III polysynaptic afferents, as opposed to group Ia monosynaptic afferents.

In summary, this study shows that spinal RORβ INs form an integral part of a low-threshold afferent inhibitory feedback circuit that is recruited during locomotion to limit flexor motor activity. We propose that these RORβ INs act as sensory filters to presynaptically gate proprioceptive afferent transmission and prevent abnormal flexor reflexes that disrupt the ongoing locomotor program, thereby securing the smooth rhythmic limb movements that are required for a fluid walking gait.

STAR METHODS

Contact for Reagent and Resource Sharing

Further information for resources and reagents should be directed to the Lead Contact, Martyn Goulding at the Salk Institute for Biological Studies (email: goulding@salk.edu).

Experimental Model and Subject Details

Mouse lines

All protocols for animal experiments were approved by the IACUC of the Salk Institute for Biological Studies according to NIH guidelines for animal experimentation. Male and female mice were used in all studies. Animals were randomized to experimental groups and no sex differences were noted.

The RORβ^Cre transgenic mouse line used in this study was generated by the Allen Brain Institute (Harris et al 2014). The specificity of Cre recombination and reporter expression in RORβ INs was determined by crossing RORβ^Cre mice with a Thy1∷LSL-YFP reporter line (Madisen et al., 2012). Conditional RORβ^fl/fl mutant mice were generated by ES cell gene targeting. Sequences encoding loxP sites were either side of exon 2 that encodes the RORβ first zinc finger of the binding domain. The F1 generation was crossed with Flp-deleter mice to remove the PGKneo cassette. Genotyping was performed using CCAGGAAGGCCATTCAAATA (forward) and GAGCTGCAGATCAGTTAGACAAAA (reverse) primers to distinguish between wild type and conditional RORβ alleles (Figure S2).

Other mouse lines used in this study were: RORα^Cre (Chou et al., 2013), Pax2∷Cre (Ohyama and Groves, 2004), Nestin∷Cre (Tronche et al., 1999), and Emx1^Cre (Gorski et al., 2002), TrkB^fl/fl (Xu et al., 2000), R26^LSL-Ai32 mice (Madisen et al., 2012), R26^ds-HTB (Stam et al., 2012), Tau^ds-DTR; hCdx2∷FlpO (Britz et al., 2015), R26^LSL-TVA mice (Seidler et al., 2008), and R26^LSL-TeNT (Zhang et al., 2008), Chx10∷CFP (Zhong et al. 2010), GAD1∷GFP (Tamamaki et al., 2003) and GlyT2∷GFP (Zeilhofer et al., 2005). All mice were maintained on a mixed background and littermates were used as controls for all experiments. Littermates lacking the Pax2∷Cre allele were used as controls for experiments using the RORβ^fl/fl allele. Littermates lacking the RORβ^Cre allele were used as controls for experiments using the R26^LSL-TeNT and TrkB^fl/fl alleles. Littermates lacking hCdx2∷FlpO allele were used as controls for Tau^ds-DTR experiments. R26^LSL-TVA and R26^ds-HTB mice (Bourane et al., 2015b; Stam et al., 2012) were used for the morphological and connectivity studies, in conjunction with spinal injection of EnvA- pseudotyped, G-deleted-mCherry rabies virus. P0, P10, P14 or P42 day old mice were used for the immunohistochemical analyses. P4 mice were used for the rabies transsynaptic tracing experiments. P8-P14 day old mice were used for the electrophysiological experiments. P42 day- old mice were used for all behavioral experiments, the in situ hybridization studies and all morphological analyses.

Method Details

Immunohistochemistry

Mice were perfused with 4% paraformaldehyde in PBS prior to dissecting the spinal cord (lumbar levels L4 and L5) and DRG (lumbar levels L4 and L5). Spinal cords were post-fixed for 2 hours at 4C in 4% paraformaldehyde/PBS. Tissues were then washed 3 times (10 minutes each) with cold PBS, cryoprotected in 30% sucrose/PBS and embedded in OCT (Tissue-Tek). Cryostat sections (14 pm, 20 pm or 50 pm) were collected and stored at −20°C for further analysis. Sections were incubated overnight with primary antibodies at 4°C. Primary antibody staining was detected with fluorophore-conjugated secondary antibodies (1:500; Jackson Laboratories). Images were captured using a Zeiss LSM 700 confocal microscope. Immunofluorescence was evaluated using ImageJ software with thresholds set according to signal intensity (Jensen, 2013).

In Situ Hybridization

Spinal cord sections (lumbar levels L4 to L5; 14μm) were hybridized overnight at 65°C with an antisense RNA probe. The slices were washed twice in 1× SSC, 50% formamide, and 0.1% Tween-20 at 65°C for 30 minutes and blocked with a solution of 2% blocking reagent and 20% inactivated sheep serum for 2 hours. The slides were then incubated with anti-DIG-alkaline- phosphatase (AP)-conjugated antibody (Roche Diagnostics) overnight. Sections were washed and developed with NBT/BCIP staining solution. For double staining analyses of YFP fluorescence coupled with DIG in situ hybridization, we directly acquired the YFP fluorescent signal without amplification using a Zeiss fluorescent microscope before performing the in situ hybridization steps. In situ hybridization signals were pseudo-colored and superposed on the YFP signal.

Rabies Virus Tracing and Morphological Analyses

Injections of EnvA-pseudotyped, G-deleted-mCherry rabies virus (~1×10⁸ units per ml) were made into the lumbar spinal cord of P42 RORβ^Cre; R26^LSL-TVA mice to obtain sparse labeling of the RORβ INs and reconstruct their morphology (see Bourane et al., 2015b). For the transsynaptic tracing studies, injections were made into the lumbar cord of RORβ^Cre; hCdx2∷FlpO; R26^ds-HTB animals at P4. Mice were anesthetized by administering 2.5% isoflurane via a nose cone. The skin overlaying the lumbar region of the spinal cord was incised and a laminectomy performed at the T13-L1 level. After removing the dura mater with a fine needle and exposing the spinal cord, a fine glass capillary was inserted on the left side of the dorsal spinal cord. 250 nl of EnvA-pseudotyped, G-deleted-mCherry rabies virus (~1×10⁹ units per ml) was injected into the dorsal cord, and the capillary was left in the cord for 1 min after injection to prevent viral spread. The skin was then closed using tissue adhesive (3M Vetbond) and a Reflex skin closure system. The animals used for both morphological analysis and transsynaptic studies were perfused 5 days post-injection and processed for immunohistochemistry.

AAV Virus Tracing of Synaptic Connections

Injections of AAV2/1-hSyn-DIO-SypHTomato (1.6×10¹² units per ml) were made into the lumbar spinal cord of P28 RORβ^Cre and RORβ^Cre; Chx10∷CFP mice. Injections were performed as outlined above for the rabies virus morphological analysis. Mice were were perfused two weeks after injection (P42) before fixing and processing the tissue.

Injection of Cholera Toxin Subunit B (CTb)

Two cohorts of mice, P8 RORβ^Cre; Thy1∷LSL-YFP mice and Pax2-RORβ mice together with control littermate mice were used for these studies. Prior to injection, all mice were anesthetized with 2.5% isoflurane in O₂ administered through a nose cone 0.25%. Alexa 647-conjugated CTb (Invitrogen) in 0.9% saline solution was then injected into a single hindlimb iliopsoas (IL, hip flexor) or biceps femoris (BF, hip extensor) muscle. Five days after CTb injection, the mice were perfused and processed for immunohistochemistry.

Behavioral Testing and Analysis

All the behavioral tests were performed blind to the genotype of the animals. Both sexes were used. Animals were acclimatized to the behavioral testing apparatus for 30 min, each day, 3 to 5 days prior to experimentation and data collection. After habituation, baseline measures were recorded on two consecutive days for each behavioral test prior to surgery or anesthetic injection. For behavioral experiments on RORβ^Cre IN-ablated mice, 6-7 week old RORβ^Cre; Tau^ds-DTR (control) and RORβ^Cre; hCdx2∷FlpO; Tau^ds-DTR mice were injected 3 times at 48 hour intervals intraperitoneally with diphtheria toxin (DTX; List Biological laboratories; 10 ng/g). Behavioral experiments were performed 2 weeks after the final injection of DTX.

Dynamic Touch Test

To measure light touch sensitivity, mice were placed on an elevated wire grid and habituated for 15 min on the day of the experiment. The plantar hindpaw was stimulated by light stroking with a fine paintbrush, in a heel to toe direction. The test was repeated five times at 10 sec intervals. For each test, no evoked movement was scored as 0. Walking movements and brief paw lifting (~1 sec or less) were scored as 1. For each mouse, the cumulative score from three tests (15 trials, expressed as a percentage) was used as a measure of the touch response.

Pinprick Test

For the pinprick test, mice were placed in a plastic chamber on an elevated wire grid and the plantar surface of the hindpaw was stimulated with an Austerlitz insect pin (Tip diameter: 0.02 mm; Fine Science Tools). The pin was gently applied to the plantar surface of the hindpaw without moving the paw or penetrating the skin. The pin stimulation was repeated 10 times on different paw areas with a 1-2 min interval between trails and the percentage of positive paw withdrawal trials was calculated.

von Frey Assay

For the von Frey assay, mice were placed on an elevated wire grid and the lateral plantar surface of the hindpaw was stimulated with calibrated von Frey monofilaments (0.008-1.4 g). The paw withdrawal threshold for the von Frey assay was determined by Dixon’s up-down method (Chaplan et al., 1994).

Kinematic Analysis

Six week-old mice of both sexes were tested. Limb position and movements were tracked with light reflective markers or black marks positioned on the iliac crest, hip, knee, ankle, paw, and tip of fourth digit of the left hindlimb as described previously (Pearson et al., 2005). Mice were allowed to run on a clear Plexiglas runway and their movements were recorded at 250 frames/sec using an InLine high-speed digital camera (Fastec Imaging Corporation, San Diego, CA). Video files were processed using the MaxTRAQ software package (Innovision Systems, Columbiaville, MI). The markers were identified and digitized manually or using the auto tracking feature. Data files were further analyzed in MaxMATE, a plug-in for Microsoft Excel.

Sensory blockade

Perisciatic blockage was performed after baseline kinematic recordings had been collected. 20μl of a 0.2% QX-314 (Sigma-Aldrich) 2% lidocaine solution (Lidoject, Henry Schein) was injected bilaterally into each hindlimb. Recordings were repeated every five minutes for 20 minutes after administration of the perisciatic block, by which time the nociceptive reflex had fully recovered. Maximum blockade was seen at 10 minutes, and this time point was used for all subsequent analysis.

Cutaneous blockade was achieved by applying a thin film of a topical anesthetic cream (DermaPlanet; 6% lidocaine, 20% benzocaine, 4% tetracaine) to both hindpaws. The cream was allowed to dry completely before placing the mice back on the walkway for recording. Recordings were repeated every five minutes after administering the anesthetic. Maximum blockade was seen at 5 minutes, and this time point was used for all subsequent analysis.

Electrophysiology

Dorsal Root Potential Recordings

Dorsal root potentials were recorded in P10-14 isolated spinal cords of RORβ^Cre; R26^LSL-Ai32 and Pax2∷Cre; RORβ^fl/fl mice. Dorsal roots were stimulated at twice the minimum threshold for primary afferent depolarization or ventral root potential recruitment (2T). Recordings were made at room temperature to recruit low threshold afferents (23°C, see Fink et al., 2014). P10 to P14 RORβ^Cre; R26^LSL-Al32 heterozygous mice, and Pax2∷Cre; RORβ^fl/fl and littermate controls were anesthetized using an intraperitoneal cocktail of ketamine and xylazine. The spinal cord was promptly dissected, isolated, and immersed in oxygenated ice-cold ACSF solution (rACSF - NaCl, 125 mM; KCl, 2.5 mM; NaHCO₃, 26 mM; NaH₂PO₄H₂O, 1.25 mM; MgCl₂, 1 mM; CaCl₂, 2 mM; glucose, 20 mM). Meningeal membranes were carefully removed to avoid lumbar roots damage and the spinal cords transferred to the recording chamber and superfused with oxygenated ACSF. All drugs used for these analyses (1 mM kynurenic acid, 1 mM mephenesine, 20 μM bicuculline) were superfused into the recording chamber. The recording chamber temperature was controlled with a perfusion temperature controller (Warner Instruments) and experiments were conducted at room temperature (23°C) unless otherwise stated. A 4 min ramp was required to increase the recording chamber temperature from 23°C to 33°C.

Suction pipettes were used to stimulate the L4 dorsal root (2T, 0.1 msec; A360 WPI stimulus isolator) and record from the adjacent L5 dorsal root.

For RORβ^Cre; R26^LSL-Ai32 optogenetic stimulation experiments, light stimulation (pulse duration 100 msec) was performed using a single LED optic fiber source (2.5 mW output at 470 nm; Mightex Systems) placed 10 mm from the spinal cord. A light beam collimator was used to restrict the beam to the area of interest (5 mm in diameter). A differential EXT 10-2F extracellular amplifier (NPI electronic) was used to acquire all signals, which were then amplified and filtered at DC and 1 kHz. Each trial was repeated 5 times and data collected at 2.5 kHz using pClamp 10.4. at 0.1 Hz before being averaged offline for further analysis.

Ventral Root Recordings

P8 Pax2∷Cre; RORβ^fl/fl and littermate controls were anesthetized and spinal cords prepared as for dorsal root potential recordings above and as previously published (Thompson et al., 1992). Experiments were conducted at room temperature. Suction pipettes were for stimulating the L4 dorsal root (2T, 0.2 msec pulse; A360 WPI stimulus isolator) and recording the ipsilateral L4 ventral root. Signals were acquired using a differential EXT 10-2F extracellular (NPI electronic) before being amplified and filtered at DC and 2 kHz. Each trial was repeated 5 times at 0.1 Hz and averaged offline for analysis. Statistical analyses were performed on control and mutant datasets using the two-sided unpaired parametric t test.

Electromyographic (EMG) Recordings

Electromyographic recordings were performed on 6 week old Pax2∷Cre; RORβ^fl/fl and littermate controls as described in Pearson et al. (2005). Briefly, animals were anesthetized by isoflurane anesthesia (2.5% in O₂) and incisions were made in skin dorsally at cervical levels and above each of the hindlimb muscles receiving electrode implants. Fine recording EMG electrodes were led under the skin from neck incision to the muscles. A 30G needle attached to the end of each electrode was used to guide the recording wire through the targeted muscle. Each electrode was then knotted at the distal end to prevent movement. A connector was then cemented to the neck area and the electrodes attached. All incisions were sutured and allowed to recover for two days prior to recording.

Quantification and Statistical Analysis

All data are presented as the mean ± standard error of the mean (SEM) with n indicating the number of mice analyzed unless otherwise stated in the figure legend. Prism 5 software was used for all statistical analyses, with p values below 0.05 considered to be statistically significant.

Neuronal and synaptic counts were determined by analyzing 3-6 spinal cords (5-10 sections per cord) per genotype. Synaptic counts were performed by imaging the intermediate spinal dorsal horn at 63x magnification. 3 μm z stacks at 0.3μm separation were taken and synaptic contacts were counted blind to the genotype within single planes. Statistical analyses were performed using a two-tailed, unpaired Student’s t-test or two-way ANOVA with Bonferroni post hoc test Asterisks above histograms in the figures indicate post hoc significance between groups as indicated in figure legends.

For kinematic analyses, statistical were performed using two-way ANOVA with Bonferroni post hoc test. Asterisks above histograms within figures indicate post hoc significance between groups as indicated in figure legends.

For physiological studies, statistical analyses were performed using unpaired Student’s t-test (Figure 4) and two-way ANOVA with Bonferroni post hoc test (Figure S3). Asterisks above histograms within figures indicate post hoc significance between groups as indicated in figure legends.

Supplementary Material

1

Supplemental Movie S1. Control mouse on walkway (related to Figure 2).

Supplemental Movie S2. Pax2∷Cre; RORβ^fl/fl mouse on walkway (related to Figure 2).

Highlights.

• Mice lacking RORβ inhibition display a hyperflexion locomotor phenotype

• Lamina V-VI RORβ⁺ GAD2⁺ interneurons presynaptically inhibit myelinated afferents

• Presynaptic inhibition is reduced in Pax2-RORβ mutant mice

• RORβ⁺ interneurons gate sensory transmission to facilitate a fluid locomotor gait