Elimination of glutamatergic transmission from Hb9 interneurons does not impact treadmill locomotion

The spinal cord contains neural circuits that can produce the rhythm and pattern of locomotor activity. It has previously been postulated that a population of glutamatergic neurons, termed Hb9 interneurons, contributes to locomotor rhythmogenesis. These neurons were identified by their expression of the homeobox gene, Hb9, which is also expressed in motor neurons. We developed a mouse line in which Cre recombinase activity is inducible in neurons expressing Hb9. We then used this line to eliminate vesicular glutamate transporter 2 from Hb9 interneurons, and found that there were no deficits in treadmill locomotion. We conclude that glutamatergic neurotransmission by Hb9 interneurons is not required for locomotor behaviour. The role of these neurons in neural circuits remains elusive.

Scientific Reports | (2021) 11:16008 | https://doi.org/10.1038/s41598-021-95143-y www.nature.com/scientificreports/ But there are several problems and potential pitfalls with such an approach. Firstly, it was noted that in Hb9::eGFP transgenic mice, there is GFP expression beyond Hb9 INs and MNs; that is, eGFP expression did not represent "true" post-natal Hb9 expression. Furthermore, Cre expression in Hb9 cre mice is not limited to Hb9 INs and MNs, but is expressed in widespread populations including those that expressed Hb9 transiently early in development 11 . Ergo, excision of vGluT2 using Hb9 cre mice would not be limited to Hb9-expressing neuronal populations, making interpretation of results using that strategy problematic.
To address this concern, we generated a novel inducible Cre mouse line, Hb9::CreER T2 , using a BAC transgene. Our goal was to determine whether Hb9 INs play a discernable role in overground locomotion. We first characterised the pattern of tamoxifen inducible recombination and demonstrated the specificity and sensitivity of recombination in Hb9-expressing MNs and Hb9 INs. Next, we crossed this line with a vGluT2 fl/fl line 10,12 to eliminate vGluT2 expression in Hb9-expressing neurons (and called the line Hb9-vGluT2 OFF ). We then demonstrated that glutamatergic transmission by Hb9 INs does not contribute to treadmill locomotion of varying speeds. We conclude that the role, if any, of Hb9 INs thus remains obscure.

Results
Recombination in Hb9::CreER mice is restricted to Hb9-expressing cells. Like others 11 , we initially used an Hb9 cre mouse line, but in early experience, found widespread expression of Cre-reporter throughout the embryo (Fig. 1A), all levels of the spinal cord, and in neurons throughout all laminae in the spinal cord ( Fig. 1B; noted over many mice, but n = 2 formally analyzed for this study). It was thus clear that we should abandon this strategy.
We therefore developed an inducible Hb9::CreER T2 mouse line ( Fig. 1C; hereafter called Hb9::CreER), and proceeded to study the selectivity of recombination following administration of tamoxifen (TAM) initially between E9 and E14 in Hb9::CreER;Rosa26-lox-stop-lox-tdTomato (together Hb9::CreER;tdTom). Expression of tdTom was evident as early as 24 h following TAM. At early embryonic stages (E9; Fig. 1D), TAM led to widespread expression of tdTom in the spinal cord, presumably due to early expression of Hb9 outside of the MN lineage in the caudal neural plate at this time point (Fig. 1D). This result is similar to the pattern of expression seen in Hb9 cre ;tdTom mice (cf. Fig. 1A).
On the other hand, administration of TAM even 12 h later (E9.5) led to more restricted expression patterns, with clear specificity of recombination (Fig. 1E). In particular, ventral horn neurons in the region of motor pools expressed the reporter along with various other neurons (Fig. 1F). But it was evident that there were two clusters of tdTom ON MNs, one which maintained Cre-ER expression at E12.5, and one which had very low Cre-ER protein levels (Fig. 1F, arrow). These clusters corresponded to two distinct motor columns, the lateral and medial portions of the lateral motor columns (LMC-L and LMC-M), respectively. The populations of somatic MNs that showed recombination dependent expression of tdTom depended on the timing of TAM. While the lateral aspect of the lateral motor column (LMC-L) expressed tdTom with administration any time after E9.5, the medial portion (LMC-M) only expressed tdTom if TAM was given prior to E11.5. This downregulation of Hb9::CreER is consistent with previous observations that endogenous Hb9 expression is down-regulated in LMC-M motor neurons by E12.5 13,14 . Note also that motor neurons that have not yet settled in lamina IX express CreER at this embryonic stage (Fig. 1F, arrowhead). Thus, embryonic administration of TAM in Hb9::CreER mice leads to specific activation of Cre recombinase, accurately marking Hb9-expressing motor neurons (Hb9 ON MN) in time and space.
Tamoxifen administered at E12-when MNs are specifically labeled within the spinal cord-also led to reporter expression outside the central nervous system ( Fig. 2A-C). Delaying TAM administration until E14 did not alter this non-neuronal recombination ( Fig. 2D-I). Sites of recombination included forelimb and hindlimb mesenchyme ( Fig. 2A We next turned to postnatal (up to P7) TAM, which also resulted in efficient recombination in Hb9 ON motor neurons: reporter expression in brain stem MNs was specific to Hb9-expressing ventral MNs (vMNs), including abducens ( Fig. 3A) and hypoglossal (Fig. 3B) nuclei. We noted no other supraspinal expression of reporter except in the choroid plexus (note, for example, lack of expression in the facial nucleus (VII), which is of branchial origin, Fig. 3A). In spinal MNs, postnatal tamoxifen induced highly efficient recombination in Hb9 ON MNs (Fig. 3C). During development, LMC-M motor neurons down-regulate Hb9 (see above), and correspondingly these MNs, identified by their location and expression of cholinergic markers VAChT (Fig. 3D) or ChAT (Fig. 3E), were not labelled with tdTom. However, in Hb9-expressing populations such as the medial motor column (MMC) and LMC-L motor neurons, recombination was efficient, with almost all motor neurons undergoing recombination. In addition, sympathetic preganglionic neurons (SPNs) express Hb9 4 , and they too showed recombination following TAM (Fig. 3H) (Note that, consistent with patterns of Hb9 expression, other cholinergic neurons did not).
We next examined Hb9::CreER post-natal recombination in α-and γ-MN subtypes, as non-BAC based Hb9::GFP transgenic mice display selective expression in α-MN 15 . We identified γ-MNs as small, ChAT ON neurons with a visible nucleus that did not express NeuN (Fig. 3F), or those that had a paucity of primary afferent (labelled by vGluT1) or C-bouton (labelled by vAChT) inputs (Fig. 3G). Using postnatal TAM regimens, we identified efficient recombination in both NeuN OFF , VGluT1 low , VAChT low γ-MNs as well as NeuN ON , VGluT1 high , VAChT high α-MNs (Fig. 3F,G). Therefore, within Hb9 ON motor columns the Hb9::CreER BAC line efficiently recombines in all MN subtypes.
We next turned our attention to Hb9 interneurons: a population of glutamatergic neurons in medial lamina VIII above the second lumbar segment (L2) 7 . Following postnatal TAM administration, there was CreER  www.nature.com/scientificreports/ activation in this region as noted by reporter expression (Fig. 3I). These neurons were positive for Hb9 and negative for ChAT, indicating that these are Hb9 INs (Fig. 3I). Virtually all visually-identified Hb9 INs, identified by their location, size, and Hb9 expression, expressed the reporter (see β-gal expression below for quantification). Together, these data reflect the specificity of our approach, in which recombination is seen primarily during the times when, and in the neurons in which, endogenous Hb9 is expressed. That is, recombination in Hb9::CreER mice induced by TAM any time after E10.5 and including the early post-natal period was seen in MNs that endogenously express Hb9: LMC-L MNs, MMC MNs, SPNs, vMN brain stem MNs, and Hb9 INs. But when TAM was administered prior to E11.5, there was also recombination in LMC-M MNs (Fig. 1F). Administration prior to E9.5 led to widespread, non-specific recombination in caudal regions (Fig. 1D). These data indicate that these inducible Hb9::CreER mice can be used for spatially and temporally specific and sensitive recombination strategies aimed at Hb9-expressing MNs and Hb9 INs.

vGlut2-dependent transmission by Hb9
INs does not affect treadmill locomotion. As our goal was to study mature mice, we next ensured that expression induced by postnatal TAM administration led to persistent and specific expression in adult mice (Fig. 4A). When crossed into Hb9 lacZ knock-in mice 16 , www.nature.com/scientificreports/ reporter expression in the adult was 95% concordant with β-gal expression (205/216 tdTom neurons in 3 mice also expressed β-gal), and was evident in MNs, SPNs, and Hb9 INs (Fig. 4A,B). We did not find β-gal-expressing neurons that did not express tdTom. That is, recombination was efficient at this age, and there was no evidence of substantial off-target recombination. Given that Hb9 interneurons are glutamatergic and express vGluT2 7 (and not vGluT1 or vGluT3 17 ), whereas the primary neurotransmitter of MNs is acetylcholine, we reasoned that we could functionally remove Hb9 INs from circuits by crossing Hb9::CreER mice with vGluT2 flox/flox mice 12 , to yield Hb9-vGluT2 OFF mice (Fig. 4C), in which glutamatergic transmission from neurons expressing Hb9 would be eliminated 10 . This floxed vGluT2 mouse strain has been used with numerous Cre drivers where it has been shown to be effective [10][11][12] . Furthermore, in situ hybridisation for vGluT2 mRNA in Hb9-vGluT2 OFF mice confirmed that this strategy was effective (n = 2; Fig. 4D; cf. Figure 3A in 7 ).
We next assessed these mice in the sixth post-natal week at various treadmill locomotor speeds, quantifying a number of locomotor parameters as well as their variability (Figs. 5,6). Parameters related to the pattern of locomotion-homolateral (Fig. 5A) and homologous (Fig. 5B) coupling, and rear track width (Fig. 5C)-did not differ between the control (n = 12, of which n = 8 could maintain the highest speed) and Hb9-vGLuT2 OFF (n = 11, of which n = 9 could maintain the highest speed) mice. That is, while there were some differences related to speed (homolateral coupling and rear track width), there were no differences between genotypes (Table 1). Furthermore, while the variability in these parameters differed depending on speed (homolateral and homologous coupling), there was no effect of genotype on these variabilities (lower panels of Fig. 5A-C; Table 1). Finally, we analysed the data taking into consideration the sex using a one way ANOVA, and found no differences in the 4 (sex × genotype) groups, except for the variability of the homolateral coupling at the slowest speed (Table 2). Multiple comparisons demonstrated that this difference arose from a difference between females and males within the control group (P = 0.04; Table 3). That is, there were no differences between the control and Hb9-vGLuT2 OFF mice. These data indicate that glutamatergic transmission by Hb9 INs does not contribute to the generation of the locomotor pattern required for treadmill locomotion.
As we and others 6,7,9,18 had previously suggested that Hb9 INs may provide a "clock" function for locomotor circuits (see "Discussion"), we suspected there may be differences, in particular increased variability, in timing parameters [stride length (Fig. 6A), step cycle duration (Fig. 6B), stance duration (Fig. 6C), and swing duration ( Fig. 6D)] in Hb9-vGluT2 OFF mice. While most of these parameters differed depending on speed (Table 4), there were no significant differences based on genotype. Furthermore, the coefficients of variability for each of these parameters (Fig. 6, lower panels), while in some cases speed-dependent, did not differ between the different genotypes. Analysis of the data taking into consideration the sex using a one way ANOVA failed to find differences between the 4 (sex × genotype) groups (Table 2). Finally, the relationships between step cycle phases and cycle duration (Fig. 7) were the same in Hb9-vGluT2 OFF and control mice. That is, the absence of glutamatergic transmission by Hb9 INs did not influence the parameters related to locomotor rhythm of mice walking on a treadmill.
Taken together, we were unable to find evidence to support the hypothesis that glutamatergic transmission by Hb9 INs plays a significant role in treadmill locomotion.

Discussion
Hb9 is a homeodomain protein expressed in sub-populations of somatic motor neurons and a population of ventral spinal interneurons termed Hb9 interneurons whose function is unknown. In order to gain genetic access to these neurons, we made a new BAC transgenic mouse line using a tamoxifen-inducible Cre recombinase strategy, Hb9::CreER. After demonstrating the specificity of this line, we proceeded to ask whether Hb9 INs significantly contribute to locomotor activity in adult mice. To do this, we removed glutamatergic neurotransmission from Hb9 INs using an intersectional approach to delete vGluT2 from Hb9-expressing neurons. We demonstrated that doing so does not significantly impact treadmill locomotion.
Use of Hb9::CreER mice to study motor neurons. Several methods have been used to gain genetic access to somatic motor neurons, but all-including ours-have caveats. The mouse line that we describe here can be used to study and manipulate LMC-L MNs, or with precise timing of TAM administration at E9, possibly even all MNs. While the limitation to LMC-L would be disadvantageous for many studies, it could provide the ability to compare affected (LMC-L) vs unaffected (LMC-M) motor neurons in the same animal with any given genetic manipulation. This could be useful, for example, in ALS studies 19 .
In contrast to our line, Hb9-ires-Cre knock-in mice (Mnx1 tm4(cre)Tmj , termed Hb9 cre here) have been used in several studies to genetically manipulate cervical motor neurons 20,21 . To ensure specificity, these investigators used an intersectional approach. Specifically, Hb9-cre;Isl2-lox-stop-lox-DTA (or DTX) crosses were used to eliminate most cervical motor neurons 20,21 . Of note, the focus in those studies was the neuromuscular junction. This intersectional approach would not affect neuronal populations other than motor neurons as isl2-expressing INs do not express Hb9 21 . Hb9-driven Cre expression was not extensively investigated in these studies, but appeared to be confined to lamina IX (motor neurons) in a section of cervical spinal cord; this may have led others to the erroneous assumption that Hb9-ires-Cre was motor neuron specific throughout the spinal cord (see http:// www. infor matics. jax. org/ refer ence/ allele/ MGI: 24477 93? typeF ilter= Liter ature). Several years later, Cre-mediated recombination was studied in these Hb9 cre mice, and a gradient of non-MN recombination was evident: there was reporter expression in relatively few cells outside motor pools in the cervical cord, but extensive expression throughout the lumbar cord [22][23][24] . That is, there is expression of Cre beyond MN pools in these mice, and this "ectopic" expression increases from rostral to caudal. This pattern likely results from early Hb9 expression in the caudal neural plate and tail bud area, where all cell progeny derived from these neural plate progenitor cells  27 , we elected to test this hypothesis in treadmill locomotion, because if no effects could be observed, circuit analysis in reduced preparations would be unnecessary. Hb9 INs were identified almost 2 decades ago, and their rhythmic, conditional bursting properties along with their location in the ventromedial upper lumbar spinal cord 3 led to the suggestion that they perform a "clock" role to maintain the rhythm of locomotion 7,9 . Since that time, these neurons have been studied in a number of additional labs as well 18,28,29 . Some investigators have cast doubt on a role for Hb9 INs in locomotion 11 , and others concluded that, given that Hb9 INs are active following the onset of motoneuron activity, they are unlikely to be solely responsible for rhythm generation 29 . Their function in behavior has therefore remained opaque, and a putative role in rhythm generation continues to be discussed-for recent reviews, see 30,31 .
Using the Hb9 cre knock-in line, it was suggested that eliminating glutamatergic transmission from "Hb9::Crederived" neurons (vide supra) impacted locomotor rhythm in isolated neonatal spinal cord preparations 11 . But as these authors discussed, it is difficult to draw conclusions from this approach about the role of Hb9 INs, as early expression of Hb9 leads to Cre-mediated recombination in many neurons through the spinal cord, i.e. Cre recombination is too widespread and Hb9 INs too few in number relative to Cre expression to draw solid conclusions 11 . Thus, to allow more specific genetic access, it was necessary first to make a new mouse that could obviate the problems resulting from early caudal Hb9 expression. Using an inducible CreER system, we indeed found specific and sensitive recombination in Hb9 expressing neurons when activated with appropriate timing of TAM administration.
With this new mouse, we generated Hb9-vGluT2 OFF mice by multi-generational breeding with vGluT2 flox mice, which we and others have used successfully before 3,10,12 , and used these mice to quantify treadmill locomotion in the absence of glutamatergic transmission from Hb9 INs. The pattern and timing of treadmill locomotion were the same in mutant and control mice. While a statistical difference was seen in the variability of homolateral coupling at the slowest speed, multiple comparison analysis revealed that this difference was between females and males of the control group ( Fig. 5A; Table 3; P = 0.04), and was not due to genotype.
It is important to note, however, that these data do not completely exclude Hb9 INs from participating in locomotion. For example, our protocol involved post-natal TAM administration, and Hb9 INs could be important during embryonic development. More interestingly, however, Hb9 INs are electrotonically coupled to other types of neurons 16 ; electrotonic transmission can play an important role in synchronicity and rhythmogenesis in other systems (e.g. [32][33][34] ). The Hb9::CreER mouse described here could be used to try to sort this out, for example by post-natally eliminating Cx36 from Hb9-expressing neurons, when motor neuron expression of Cx36 is already decreasing 35 . Selectively eliminating Hb9 INs using, for example, genetic expression of diphtheria toxin fragment A 36 driven by the promoter for vGluT2 could also be used.
It is possible that glutamatergic transmission by Hb9 INs is involved in intrinsic rhythm generation, such as that which can be studied in isolated spinal cord preparations. We elected to test their role in treadmill locomotion, during which rhythmic sensory inputs from the limbs can entrain locomotion 37 and affect swing and stance vs cycle time relationships 38 . That is, sensory inputs may minimize any deficits from eliminating a central rhythmogenic network. Furthermore, "adjustments" via supraspinal inputs could also be involved.
In recent years, other neurons, in particular Shox2 INs, have been identified as strong candidates for locomotor rhythmogenesis 34 . Whether there are identifiable pacemaker neurons that are necessary for locomotor rhythm remains to be seen: this question has been difficult to answer in the respiratory system, where there have been discussions over decades about whether the rhythm is generated by pacemaker neurons and/or circuit  www.nature.com/scientificreports/ mechanisms, with no single neuronal substrate found [39][40][41] . For locomotion, it is interesting to consider that two populations of neurons that reciprocally inhibit each other are a prime substrate for rhythm generation 42 , and that locomotor circuits have many places for such pairs of direct or indirect reciprocal inhibition (governing movement across joints, between joints, and between limbs, for example 43 ). Thus there are many microcircuits in which rhythm could be generated with or without neuronal pacemaker neurons. In retrospect, perhaps, it would have been a surprise to see profound changes in the rhythm of treadmill locomotion when removing chemical transmission from this one neuron type-Hb9 INs-from spinal cord circuits.

Methods
Animals. All experimental procedures at Dalhousie University were approved by the University Committee on Laboratory Animals and were in accordance with the Canadian Council on Animal Care guidelines and those procedures at Columbia University were approved in accordance with IACUC guidelines. The study was carried out in compliance with ARRIVE guidelines (Table S1). Hb9 lacZ/+4 , Hb9 cre/+4 , and vGluT2 flox/flox12 have been previously described. Rosa26-lox-stop-lox-reporters (tdTomato (Ai14, Jax#007908); EYFP (Jax#006148); or ZsGreen (Ai6; Jax#007906)) were obtained from the Jackson Laboratory. (CHORI BACPAC resources) containing the mouse Mnx1 (Hb9) gene was modified using recombineering bacterial strains (NCI, http:// web. ncifc rf. gov/ resea rch/ brb/ recom binee ringI nform ation. aspx). The BAC was modified to remove the loxP and loxP511 sites in the pTARBAC vector backbone using homologous targeting cassettes containing ampicillin and spectinomycin resistance, respectively. A shuttle vector containing short homology arms to desired recombination sites was used to introduce CreER T2 -BGH polyA and a Frt-Zeo-Frt selection cassette into Hb9 exon 1, replacing the Hb9 coding region. In the modified BAC, the CreER T2 cassette 44 was flanked by 69.5 kb of 5′ DNA and 87 kb of 3′ DNA. The Frt-Zeo-Frt cassette is retained in the transgene but lacks a eukaryotic promoter. Both Dalhousie University and UCL have material transfer agreements with both Columbia University, where the mice were engineered, and Novartis Forschungsstiftung, Zweigniederlassung, Friedrich Miescher Institute for Biomedical Research ("FMI"), for the CreER T2 .
Transgenic mouse generation. Modified BAC DNA was digested by Not1 to remove pTARBAC vector, run out on low melting point agarose, the insert gel purified, and recovered by beta-agarase digestion. DNA was dialyzed into injection buffer and oocyte injections were done in the transgenic core at the Irving Cancer Center, Columbia University Medical Center. Three Hb9::CreER founder lines (4, 8, and 10) on B6CBA hybrid backgrounds were confirmed for tamoxifen induced recombination in motor neurons between E11-E14 using a ROSA26-lox-stop-lox-EYFP reporter. Line 10 gave low level inducible activity and was terminated. Line 4 and Line 8 displayed similar inducible recombination activity in the desired cell populations and were maintained on C57BL/6J backgrounds. Both lines are viable if bred to homozygosity, but were maintained as heterozygotes. Because the lines were indistinguishable, we have used line 8 for the majority of studies described herein.
Tamoxifen administration. Tamoxifen (Sigma T-5648) was dissolved to 20 mg/mL in 90% sesame oil/ 10% Ethanol, briefly heated to 50 °C to facilitate dissolution, and stored in frozen aliquots. For administration, stock aliquots were warmed to 37 °C for 10 min prior to injection. Pregnancies were timed by vaginal plugs, and pregnant dams (~ 30 g bodyweight) were intraperitoneally (IP) injected a single time with 0.1 mL stock (2 mg) at . The coefficients of variation were also the same at all speeds (lower panel). Number of mice used at the 4 speeds for the control group (total 12) were 11 (5 females), 12 (5 females + 1 not specified), 11 (5 females + 1 not specified), and 8 (4 females), and for the Hb9-vGluT2 OFF group (total 11) were 10 (6 females), 11 (6 females + 1 not specified), 9 (5 females + 1 not specified), and 9 (5 females www.nature.com/scientificreports/ indicated ages. In our preliminary characterization of the mice using the ROSA26::lox-stop-lox-EYFP reporter, we found that a single intraperitoneal injection administered between E10 and E16 of tamoxifen into a pregnant female Hb9::CreER mouse was highly efficient at inducing recombination in embryonic spinal motor neurons. In postnatal experiments, pups of both sexes (ages P2-P7) were injected subcutaneously above the dorsal neck fat pad with 0.1 mL (2 mg) stock solutions and maintained on a 37 °C heating pad for 1 h before returning to their nursing cage. Due to leakage from the subcutaneous injection site, the exact dose in pups was likely variable. We did not observe any toxicity when tamoxifen was administered by these approaches.
Tissue preparation and labeling. Under deep ketamine/xylazine anesthesia, mice were perfused with 4% paraformaldehyde (PFA) in 0.2 M PB. Following perfusion, the spinal cords were extracted and post fixed in PFA at 4 °C overnight, then transferred to a 30% sucrose solution for 24-48 h before sectioning.
Spinal cord blocks were sectioned (50-70 μm) using a vibrating microtome and processed for immunohistochemistry on the same day, or stored in glycerol at − 20 °C for later use.
Floating sections were washed in PBS for 10 min before incubation in 50% ethanol for 30 min to enhance antibody (Ab) penetration. Sections were washed with double salt PBS (dsPBS) 3 × 10 min and transferred to blocking solution containing 10% donkey serum in 0.3 M Triton X and PBS solution (PBST) for 30 min at room temperature.
Tissues were transferred to primary Ab (Anti-DsRed-Rabbit, Rockland antibodies and assays PA, USA 1:2000; Anti-β-gal Goat or Mouse, Santa Cruz Biotechnology, Inc, Dallas, Tx, USA 1:500; Anti-GFP-Sheep or Mouse, GenWay Biotech, Inc. San Diego, CA, USA and Novus Biological inc. Littleton, CO, USA 1:500) diluted in a In situ hybridization combined with immunohistochemistry. Tissues were processed for combined fluorescent in situ hybridization and immunohistochemistry as previously described 7,10 , and based on 45  Image acquisition. Confocal images were acquired with either a Zeiss LSM 710 Laser Scanning Confocal Microscope (Dalhousie) or Leica TCS SP5 (Columbia). For imaging synaptic puncta (VGlut1, VAChT), vibratome sections were imaged in Z stacks with a 20× or 63× objective (2 µm optical Z resolution). Images were captured either as 2D snapshots or 3D z-stacks with intervals of 0.8-1.88 μm and total Z-thickness up to approximately 30 μm. Pinhole size was set to 1 airy unit for all channels.
Treadmill locomotion. These experiments were conducted at postnatal day 35-38 (15.9 ± 1.7 g) in control (cre negative with one or both floxed vGluT2 alleles, or cre positive with only one floxed vGluT2 allele) and experimental (Hb9-vGluT2 OFF ) mice. All mice were given tamoxifen as described above. Locomotor behaviour was studied on a treadmill (Cleversys, Inc.) equipped with a transparent belt and a high speed camera (Basler, USA). The treadmill belt and chamber were cleaned with Peroxigard prior to every experimental session. Mice were placed in a 161 cm 2 chamber by an experimenter (LMK) not blinded to genotype, and gait was recorded with the supplied software (BCam Capture Version 2.00, Cleversys, Inc.) at a frame rate of 100 frames/s. Locomotor behaviour was examined in the afternoons of the light cycles of the animal at belt speeds of 15, 20, 27, and 30 cm/s. Faster speeds were attempted, however mice could not maintain locomotion at those speeds (e.g. 40 cm/s and 50 cm/s) for the full 20 s. The speeds were chosen to represent low, medium, and high speeds 46 .
Locomotor activity was not facilitated through means such as electrical shocks or air puffs. Video recording started as soon as the treadmill speed was stable and lasted for 20 s per trial. Mice were allowed to rest for 1-2 min between each trial. In some cases, they were studied on a different day if they did not continuously locomote for 20 s at the test speeds. A total of 11 mice (6 female) were in the experimental group, and 12 mice (5 females) in the control group, which were littermates of the experimental mice. The sex of one mouse per group was not recorded. Power analysis revealed that a 50% increase of the step cycle CV (our primary measure) of 0.15 ± 0.05 in control mice with a = 0.05 and 80% power could be detected with 7 mice in the control and experimental groups.
Gait analysis. Gait parameters were analyzed and recorded automatically using TreadScan Version 3.00 (Cleversys, Inc.). Prior to each experiment, a ruler was placed along the treadmill and imaged for scale calibration. An artificial foot model was created by drawing a polygon over the foot of interest at different frames, and  Fig. 1A). To exclude the instances when the mouse did not keep up with speed (e.g. when rearing or grooming), we included only sequences during which the mouse kept up with the treadmill for the full 20 s, and thus was near the front of the treadmill chamber. For each step, analysis parameters were obtained including: stance duration, which is the time during which the foot is in contact with the treadmill belt; swing duration, during which the foot is not in contact with the treadmill belt; cycle duration, which is the sum of stance and swing durations ( Supplementary Fig. 1B); and stride length, which is defined as the distance travelled by the limb in a complete step cycle, and was calculated as (running www.nature.com/scientificreports/ speed * stride time) + displacement, where displacement could be positive or negative depending on the change of position of the foot in the camera frame. In addition, homologous coupling, defined as the coupling between the two hind limbs, homolateral coupling, defined as the coupling between ipsilateral limbs (e.g. right rear and right front), were calculated, with 50% meaning the limbs were out of phase. Rear track width, measured as the distance between the two hindlimbs measured perpendicular to the long axis of the body and is a measure of the hindlimb base of support was also calculated ( Supplementary Fig. 1A,C). The reliability of each step was confirmed by examining all frames to ensure that each foot was adequately captured before exporting the raw data to an Excel file for analysis. The right hindlimb was used to quantify single limb parameters, both hindlimbs for homologous, and right fore-and hind-limbs for homolateral coupling (Supplementary Fig. 1).
Statistics. For each step cycle parameter, normality was tested using the D' Agostino-Pearson test for the control vs experimental groups, and the Shapiro-Wilk test for the smaller sample sizes of females and males within these groups. The majority of distributions presented here passed these normality tests, so we proceeded with parametric tests to measure statistical significance. As the same animals were tested at different speeds, we used a repeated measures two way ANOVA (speed × genotype; sphericity assumed), but if there were missing values (i.e. if the animal failed to locomote at all speeds), a mixed effects model was used instead. For intra-and inter-group differences between females and males, we used unpaired one way ANOVA and for any significant differences, proceeded with Bonferroni multiple comparison tests with a single pooled variance to determine where the difference arose. For the comparison of swing or stance vs cycle duration (Fig. 7), linear regression analysis was done using ANCOVA-type calculations to compare 47 . All tests were done using Prism (GraphPad, San Diego, CA 9. 1.1(225)). The experimental unit was taken as the mouse. P-values < 0.05 were considered to be significant. (Of note, because not all data sets were normally distributed, we also proceeded with a nonparametric Mann-Whitney U-test to compare control and experimental data, and a Kruskal-Wallis test for the 4 groups (genotype × sex), followed by a Bonferroni-Dunn test for any significant findings. Results were the same as the parametric analyses).