Spinal V2b neurons reveal a role for ipsilateral inhibition in speed control

The spinal cord contains a diverse array of interneurons that govern motor output. Traditionally, models of spinal circuits have emphasized the role of inhibition in enforcing reciprocal alternation between left and right sides or flexors and extensors. However, recent work has shown that inhibition also increases coincident with excitation during contraction. Here, using larval zebrafish, we investigate the V2b (Gata3+) class of neurons, which contribute to flexor-extensor alternation but are otherwise poorly understood. Using newly generated transgenic lines we define two stable subclasses with distinct neurotransmitter and morphological properties. These V2b subclasses synapse directly onto motor neurons with differential targeting to speed-specific circuits. In vivo, optogenetic manipulation of V2b activity modulates locomotor frequency: suppressing V2b neurons elicits faster locomotion, whereas activating V2b neurons slows locomotion. We conclude that V2b neurons serve as a brake on axial motor circuits. Together, these results indicate a role for ipsilateral inhibition in speed control.


Introduction
Rhythmic, coordinated body movements require selective recruitment of motor neurons by spinal and supraspinal premotor circuits. Most vertebrates locomote via alternating left-right contractions that travel from rostral to caudal; tetrapods additionally alternate between flexors and extensors to regulate limb movements. Due in part to the technical challenges in identifying and manipulating specific classes of neurons in the spinal cord, the underlying circuitry of locomotion remains only poorly worked out.
Traditionally, patterned locomotion has been modeled as an alternation between excitation and inhibition, which dominate motor neurons during contraction and extension portions of the cycle, respectively (Danner et al., 2017). Recently, however, evidence from fish, frogs, and turtles has challenged the notion that inhibition is minimal during the contraction of the cycle, i.e., in-phase with excitation. Instead, inhibitory conductances appear to be significant both in-and out-of-phase Kishore et al., 2014;Berg et al., 2007;Petersen et al., 2014;Li and Moult, 2012), suggesting that simultaneous recruitment of excitation and inhibition during the contraction is important for regulating motor neuron firing (Petersen and Berg, 2016).
In-phase inhibition is thought to derive from two spinal interneuron classes, the V1 and V2b populations. The V1 population includes Renshaw cells (Alvarez et al., 2005;Sapir et al., 2004), which provide recurrent inhibition onto motor neurons with potentially significant shunting effects (Bhumbra et al., 2014). To date, most analysis of drive from V2b neurons has focused on the shared contributions of V1s and V2bs to reciprocal inhibition governing flexor/extensor alternation in limbed animals (Britz et al., 2015;Zhang et al., 2014;McCrea and Rybak, 2008). However, this does not shed light on potential functions of ipsilateral inhibition in gain control for regulation of motor neuron firing during contraction, as opposed to suppression of motor neuron firing during extension.
In-phase inhibition increases in amplitude for faster locomotor movements (Kishore et al., 2014) suggesting a potential role in speed control. Here we investigated whether V2b neurons could indeed provide direct inhibition to motor neurons for speed control, taking advantage of the speeddependent organization of zebrafish motor circuits (McLean et al., 2007;Menelaou and McLean, 2012;McLean et al., 2008). V2b neurons are good candidates for in-phase gain control because they are exclusively inhibitory in mouse and zebrafish (Batista et al., 2008) with ipsilateral, descending axons within the spinal cord (Britz et al., 2015;Lundfald et al., 2007). They arise from a final progenitor division that produces pairs of V2a and V2b neurons (Kimura et al., 2008). Given the role of V2a neurons in triggering motor output Eklö f-Ljunggren et al., 2012) particularly through speed-specific circuits for titrating levels of motor excitation (Zhong et al., 2011;Crone et al., 2009;Ampatzis et al., 2014;Menelaou et al., 2014), it seems plausible that their sister V2b neurons exert an opposing, inhibitory role in speed control. However, the V2b class has not been well characterized at anatomical or neurochemical levels outside of very early development.
Here, we define two subclasses of V2b neurons in larval zebrafish based on differential transmitter expression and anatomy and further show that these neurons directly inhibit axial motor neurons in speed-specific circuits. Optogenetic suppression of V2b activity elicits faster locomotion whereas optogenetic activation of V2b activity reduces tail frequency, consistent with a role for ipsilateral inhibition in speed control.
In a typical spinal segment, V2b soma position spanned the dorsoventral and mediolateral axes of the spinal cord ( Figure 1B). Gata3 is expressed in not only V2b neurons but also the mechanosensory cerebrospinal fluid contacting neurons (CSF-cN) as well as in intraspinal serotonergic neurons  Figure 1-figure supplement 2) (Petracca et al., 2016;Montgomery et al., 2018). Both classes were readily excluded from further anatomical analysis based on their distinct anatomical features: CSF-cNs exhibit large soma size, ventral position, and stereotyped extension into the central canal (triangles, Figure 1B), while ISNs have rectangular somata located along the ventral edge of the spinal cord (stars, Figure 1B). On average, each hemisegment contained 17.2 + /-2.5 (mean + /-SD) V2b neuron somata with relatively little variation from rostral to caudal segments ( Figure 1C).

V2b axons extend throughout the spinal cord
To visualize V2b axonal trajectories within the spinal cord, we labeled individual neurons via either single cell dye-electroporation or Kaede photoconversion in a Tg(gata3:Gal4, UAS:Kaede) line ( Figure 1D) (Ando et al., 2002). No difference in axon length or trajectory was observed between the two methods. In all 59 neurons, the axon descended caudally and ipsilaterally, with an extent ranging from 2 to 15 segments. V2b axons originated on the ventral aspect of the soma and projected laterally into the white matter. Putative en passant boutons were seen as swellings distributed along the axon. Most V2b axons projected short collaterals into the soma-dense medial spinal cord along the axon extent. V2b dendrites extended from the main axon branch near the soma, ( Figure 1D), similar to identified mixed processes in V2a neurons (Menelaou et al., 2014). However, in contrast to V2a neurons (Menelaou et al., 2014;Azim et al., 2014), no V2b neurons extended rostral axons beyond the segment of origin. Single-cell Kaede photoconversions made at different positions along the rostrocaudal extent of the spinal cord revealed that axonal projections were longest for V2b somata located in the midbody range relative to V2b located in the rostralmost and caudalmost segments ( Figure 1E and F, p=3.17Â10 À3 and p=4.44Â10 À4 , ANOVA and Tukey's test). Overall, these data reveal that zebrafish V2b neurons exclusively innervate areas ipsilateral and caudal to the soma, with the greatest territory of axonal coverage originating from mid-body neurons with long axons.

Neurotransmitter expression defines subpopulations of V2b neurons
Previous work has established that V2b neurons in embryonic zebrafish, as identified by Gata3 RNA expression, are exclusively inhibitory and predominantly GABAergic (Batista et al., 2008). To describe the neurotransmitter profile of V2b neurons in larvae, Gata3+ neurons were evaluated for coexpression with transgenic markers for glyt2, a glycine vesicular transport protein, and gad1b, a GABA synthesis enzyme. Nearly all larval V2b neurons expressed Glyt2 in 5 dpf larvae (Figure 2A), in contrast to embryonic stages. In contrast, Gad1b is expressed in approximately half of the V2b population ( Figure 2B).
Inhibitory neurotransmitter switching is posited to occur at early developmental stages in zebrafish (Higashijima et al., 2004a). Therefore, we examined whether the variation of neurotransmitter expression in V2b neurons at 5 dpf represented a transient developmental stage or a stable pattern of expression. We assessed coexpression of the neurotransmitter markers in midbody V2b neurons at 5, 10, and 15 dpf ( Figure 2C), after which V2b neurons are not reliably labeled by transgenic lines (data not shown). Gad1b and Glyt2 expression in V2b neurons remains unchanged across these ages, with~52% of neurons expressing Gad1b and~91% expressing Glyt2 ( Figure 2C). These ratios indicate at least two subclasses of V2b neurons: those expressing Glyt2 but not Gad1b, and those expressing both Glyt2 and Gad1b. A possible third subclass of V2b neurons which express only Gad1b and not Glyt2 may exist; however, incomplete transgenic line label in Tg(glyt2:GFP) (see Table 1) could also account for Glyt2 expression below 100%.
Are GABAergic and non-GABAergic neurons distributed similarly throughout the neuraxis? By plotting dorsal-ventral (DV) position relative to spinal boundaries, we found that on average, GABAergic V2b somata are located slightly ventral to non-GABAergic V2b somata, but that both populations span the same DV range ( Figure 2D). Therefore, soma position is not predictive of neurotransmitter expression. In the rostrocaudal axis, the percentage of GABAergic V2b cells is highest (~80%) in rostral segments, then decreases to~50% by midbody and throughout the rest of the spinal cord. In contrast, Glyt2 robustly colabels with V2b cells throughout the entire spinal cord ( Figure 2E). These data indicate that the Gad1b+ and Gad1b-populations comprise distinct and persistent subclasses. V2b neurons expressing both Glyt2 and Gad1b will be referred to as V2b-mixed, in reference to their mixed neurotransmitter expression, whereas V2b neurons that solely express Glyt2 will be referred to as V2b-gly.

Axonal morphology varies by subpopulation identity
The classic axonal morphology of zebrafish ventral longitudinal descending (VeLD) neurons is ventral, with little change in DV position from the onset (Batista et al., 2008). However, some V2b neuron fills exhibited axons with much more dorsal trajectories (e.g. Figure 1D). To resolve whether these represent different subclasses, we investigated axonal morphology of identified V2b-mixed and V2b-gly neurons using single-cell dye electroporation in the double transgenic line Tg(gata3:loxP- The following source data is available for figure 2: Source data 1. Data underlying quantification of Gad1b/GlyT2 co-expression. DOI: https://doi.org/10.7554/eLife.47837.008 DsRed-loxP:GFP; gad1b:GFP), in which expression of GFP (Gad1b) differentiates between the mixed and glycinergic subclasses. Although both V2b-gly and V2b-mixed neurons extend axons caudally and ipsilaterally, consistent with data in Figure 1, the DV position of their axons was different. GABAergic V2b-mixed neurons projected axons ventrally along the spinal cord, with an average axon location found between 0.24-0.33 DV (example, Figure 3A). In contrast, axons from V2b-gly neurons typically make an initial ventral dip but then turn more dorsally, ranging from 0.31 to 0.65 in the DV axis (example, Figure 3B). Traces from all filled neurons are shown in Figure 3C, and averaged trajectories in Figure 3D. Somata were filled in segments ranging from 14 to 18; the traces are shown aligned at the soma for ease of visualization.
Other features of anatomy also varied between the two subtypes. The axon DV position of V2bgly neurons is positively correlated to the soma DV position, that is a more dorsal soma projects a more dorsally positioned axon ( Figure 3G). However, this trend is not realized for V2b-mixed cells, which project axons ventrally to a narrow spinal cord region regardless of soma position. Putative en passant boutons were found in both cell populations. Most filled axons (22/24) extended vertical collaterals from the main axon. The number of collaterals per axon did not significantly vary between populations (V2b-mixed, median = 3, range 0-5; V2b-gly, median = 5, range 0-23, Mann-Whitney Wilcoxon test p=0.056). However, collaterals of V2b-mixed and V2b-gly axons cover largely distinct DV regions of the spinal cord ( Figure 3H).
What is the significance of differential DV axon trajectories between V2b-gly and V2b-mixed subclasses? Previous work has shown that motor neurons active during fast movements are located more dorsally within the spinal cord, whereas those for slower movements are located more ventrally (McLean et al., 2007). Therefore, we compared population averages of the V2b-gly and V2b-mixed axons ( Figure 3D) to a plot of motor neuron dendritic territory ( Figure 3E; see Materials and methods). Notably, V2b axon position of the two classes overlaps with two peaks in the motor neuron density profile. Consequently, we next investigated the direct influence of V2b neurons on motor neurons.

V2b subpopulations provide differential inputs to fast and slow circuits
Anatomical evidence indicates that V2b neurons make contact onto limb motor neurons where they are partially responsible for enforcing flexor/extensor alternation (Britz et al., 2015;Zhang et al., 2014). However, to date there are no physiological recordings of synaptic connections from V2bs onto motor neurons or other targets. We first validated that optogenetic stimulation in the Tg (gata3:Gal4; UAS:CatCh) line was sufficient to elicit action potentials in V2b neurons ( Figure 4A). We then targeted spinal motor neurons for in vivo recording in Tg(gata3:Gal4, UAS:CatCh) larvae at 4-6 dpf ( Figure 4B). Optogenetic activation of V2b neurons with a 20-50 ms pulse of light delivered 3-7 segments rostral to the recording site elicited robust IPSCs in motor neurons ( Figure 4D and E). Synaptic conductance amplitudes exhibited a median of 139 pS (25-75% range, 97-174 pS). Although the Tg(gata3:Gal4) line labels CSF-cNs in addition to V2bs ( Figure 1B), CSF-cNs exhibit short ascending axons that do not contact motor neurons other than the CaP (Hubbard et al., 2016). To validate that V2b neurons are providing these inhibitory inputs, we used a digital micromirror device to deliver targeted squares of light stimuli (~20 mm x 20 mm) to dorsal spinal cord areas containing V2b but not CSF-cN or ISN somata. These localized stimuli still elicited reliable IPSCs in both primary and secondary motor neurons ( . Together these results indicate that the optogenetically elicited inhibitory inputs arise from monosynaptic V2b to motor neuron connections. The striking difference in dorsal-ventral targeting of V2b-gly and V2b-mixed axonal trajectories ( Figure 3D) suggests a potential relationship with the well-described dorsal-ventral distribution of motor neurons based on size and speed at recruitment. Large motor neurons with low input resistance are located dorsally within the motor pool and are recruited for the fastest speeds of swimming, whereas more ventrally located motor neuron somata exhibit higher input resistance and are recruited during slower movements (McLean et al., 2007;Menelaou and McLean, 2012).
Accordingly, we tested whether the glycinergic and GABAergic components of the IPSC differed between fast and slow motor neurons. Bath application of 10 mM strychnine to block glycine receptors abolished a median of 91% of the V2b-evoked IPSC in fast motor neurons, but only 71% of the V2b-evoked IPSC in slow motor neurons ( Figure 4D and E,C; p=0.003, Wilcoxon Rank test). Similar results were obtained with measurement of the charge integral (over 100 ms: 91% blocked by strychnine in fast motor neurons; 66% block in slow motor neurons). The GABA A receptor antagonist  Axon traces for V2b neurons, aligned at the segment of origin, relative to the spinal cord dorsoventral boundaries (V2b-mixed, green, N = 12; V2b-gly, orange, N = 12 from eight clutches). All axons were exclusively descending. (D) Mean and SEM of V2b-gly and V2b-mixed axon trajectories. (E) Motor neuron dendrite fluorescence intensity, measured in Tg(mnx:GFP), relative to the same dorsoventral landmarks. (F) Mean axon position for each traced axon. ***p<0.0001 (p=8.23Â10 À5 ), Student's t-test. (G) Average axon position of V2b-mixed (green, left) and V2bgly (orange, right) relative to soma position for each neuron. A correlation between soma position and axon position is observed for V2b-gly but not V2b-mixed neurons. V2b-gly: r 2 = 0.33, p<0.05, V2b-mixed: r 2 = 0.0059, p=n .s. (H) Axon collaterals of V2b-gly neurons also innervate more dorsal spinal cord territory than V2b-mixed axons. DOI: https://doi.org/10.7554/eLife.47837.009 The following source data is available for figure 3:  and slow (bottom) motor neurons, as identified by soma location and input resistance. Response during baseline (black), after application of strychnine (dark gray), and after additional application of gabazine (light gray). In all cases, the IPSC was entirely abolished by the combination of strychnine and gabazine. N = 7 slow motor neurons from two clutches, N = 9 fast motor neurons from four clutches. (E) Percentage peak current reduction by strychnine in fast and slow motor neurons. *p<0.01 (p=2.0Â10 À3 ). (F) Elicited spike counts in motor neurons (fast motor neurons, top; slow, bottom) during 30 ms depolarizing steps, ranging from sub-to supra-threshold, with (blue) and without (black) concurrent optogenetic activation of V2bs. Input resistance varied across slow motor neurons, and accordingly the number of current steps relative to threshold is shown rather than the actual injected current (stepsize range: 5-25 pA). DOI: https://doi.org/10.7554/eLife.47837.011 The following source data and figure supplement are available for figure 4: Source data 1. Characteristics of V2b-evoked inhibition in motor neurons, source data. DOI: https://doi.org/10.7554/eLife.47837.013 Figure 4 continued on next page gabazine (SR-95531, 10 mM) eliminated the remaining IPSC in all cases. Therefore V2b-mediated inhibition onto fast motor neurons is carried out predominantly by glycinergic synapses, whereas V2b inhibition onto slow motor neurons is carried by mixed glycinergic/GABAergic transmission. These results are consistent with the idea that V2b-gly preferentially inhibit more dorsally located fast motor neurons, whereas V2b-mixed inhibit the more ventrally located slow motor neurons. Does V2b-mediated inhibition suppress spiking in motor neurons? We tested this by delivering 30 ms pulses of depolarizing current injection to motor neurons, with or without conjunctive optogenetic activation of V2b neurons in the Tg(gata3:Gal4; UAS:CatCh) line. Interestingly, V2b activation did not change thresholds or gain relationships on average, in both fast and slow motor neurons ( Figure 4F). Within fast motor neurons, 3/14 neurons did exhibit a clear increased threshold to spike during V2b activation, and in two of those neurons, strychnine completely abrogated the effects (the third was not tested). However, the overall results indicate that V2b suppression of motor neuron spiking is not robust, despite the clear evidence of connectivity. These findings may be attributable to the putative location of V2b synapses on motor neuron dendritic arbors and consequent effects on dendritic integration rather than somatically-elicited spiking.

V2b cell physiology does not distinguish between subtypes
Intrinsic physiological characteristics, including input resistance and spiking properties, can be used to subdivide some spinal interneuron populations into distinct subpopulations (Borowska et al., 2013;Song et al., 2018;Bikoff et al., 2016). We examined whether the V2b-gly and V2b-mixed subgroups exhibited differences in intrinsic physiology by targeting whole-cell recordings to these neurons. V2b neurons were silent at rest, in contrast to CSF-cNs which exhibited spontaneous spiking (data not shown). Spikes were elicited by depolarizing current steps, ( Figure 5D and E) which usually led to one or a few spikes, with only 3/10 V2b-gly and 3/9 V2b-mixed neurons able to sustain spiking across the step. Spikes were typically small in amplitude, similar to what is seen in zebrafish V1 neurons (Higashijima et al., 2004b), likely reflecting action potential initiation at some electrotonic distance from the soma. There was no difference in input resistance ( Figure 5F) or spike shape ( Figure 5G) between the V2b-gly and V2b-mixed neurons. Therefore, the two V2b subpopulations are indistinguishable at the level of intrinsic physiology despite their anatomical differences.

Optogenetic V2b suppression increases tail beat frequency
What are the functional consequences of V2b inhibition onto ipsilateral motor circuits? To better understand this role, we carried out high-speed behavioral recordings during optogenetic inactivation of V2b neurons with a light-gated Clchannel, ZipACR (Bergs et al., 2018). To eliminate contributions from Gata3 expressing neurons in the brain, we used a spinally transected preparation. Tail movements were induced pharmacologically with application of N-methyl-d-aspartate (NMDA, 200 mM) (McDearmid and Drapeau, 2006). NMDA induces tail movements with episodic, left-right alternations that mimic the natural beat-and-glide swims of 5 dpf larvae (McDearmid and Drapeau, 2006;Wiggin et al., 2014;Wiggin et al., 2012).
Spinal V2b neurons, CSF-cNs and ISNs are labeled in BAC-generated Gata3 transgenic lines ( Figure 1B). However, a CRISPR-generated Tg(gata3:ZipACR-YFP) knock-in shows robust expression of the fluorescent ZipACR protein in V2b neurons and ISN but only sparse, dim expression in CSF- We first validated the efficacy of the ZipACR construct in suppressing V2b firing under high-intensity light (Figure 6-figure supplement 2A and B, n = 4). Under lower-intensity light conditions, identical to those of the behavioral recordings, action potentials were completely suppressed in 4 out of 6 V2b cells and partially suppressed in one additional cell ( Figure 6B and C), in Tg(gata3: ZipACR-YFP; gata3:loxP-DsRed-loxP:GFP) animals. In contrast, identical stimulation partially suppressed spiking in only 1 of 5 CSF-cNs ( Figure 6C and Figure 6-figure supplement 2C). Therefore, we used these light stimulation parameters, under which V2b neurons are mostly if not entirely suppressed whereas CSF-cNs are not substantially affected, to carry out behavioral experiments assessing the effects of suppressing V2b neurons on locomotion.
Animals were head-embedded with a free tail and high-speed (200 Hz) recordings were acquired to capture fictive locomotion. Swim dynamics were recorded and evaluated both with and without optogenetic stimulation ( Figure 6D). Kinematic analysis of the high-speed video was performed with code adapted from Severi et al. (2018). The total tail displacement and quantity of tail movements did not significantly differ during optogenetic stimulation ( Figure 6D and F). Strikingly, tail beat frequency (TBF), a proxy for locomotor speed (Müller and van Leeuwen, 2004), increased in Tg(gata3: ZipACR-YFP)+ animals during light stimulation but not in their ZipACR negative clutchmates ( Figure 6E and G, paired Student's t-test p=0.0025). The average TBF change was 1.4 Hz (control TBF mean = 13.4 Hz, stimulation TBF mean = 14.7 Hz) and was robustly observed in animals from  Brustein et al., 2003). Thus stimulation of ISNs is unlikely to contribute to the observed TBF changes. We conclude that suppression of V2b neurons increases locomotor frequency.

V2b activation reduces TBF during evoked swims
We next investigated how the converse manipulation, V2b activation, influences locomotor rhythms. Tg(gata3:Gal4; UAS:CatCh) animals exhibited variable non-specific expression of CatCh in some muscle fibers, precluding experiments on freely swimming or head-embedded preparations. Instead, we recorded fictive swim from the motor ventral root in paralyzed, intact preparations. Brief electrical stimuli were used to elicit fictive locomotion, and in alternate trials were paired with localized optogenetic stimulation at 3-4 segments rostral to the recording ( Figure 7A). Optogenetic stimulation rostral to the ventral root recordings, targeting V2b neurons that provide inhibition to the recorded segment, led to a decrease in the first inter-burst interval (IBI) relative to trials without light stimulation ( Figure 7D and E, paired Student's t-test p=1.72Â10 À4 ). IBIs across the entire swim bout, which progresses from faster to slower, were also reduced in frequency ( Figure 7F, p=1.51Â10 À3 , paired Student's t-test) though to a lesser degree than the first IBI (Mann-Whitney Wilcoxon test p=1.23Â10 À3 ).
Several control experiments were used to ensure that these effects were attributable to V2b activation, rather than CSF-cN activation. First, to test whether connections from CSF-cNs could slow locomotion, optogenetic activation was delivered caudally to the recorded segment, to target the predominantly ascending axons of CSF-cNs. Caudal stimulation did not affect IBI frequency ( Figure 7G and H, paired Student's t-test, p=0.495). Additionally, we repeated these experiments in a set of Tg(gata3:Gal4; UAS:CatCh) animals with low CatCh expression in V2b neurons to account for CSF-cN contributions (<3 V2b neurons per hemisegment, akin to that used in Figure 4-figure supplement 1) ( Figure 7C). Optogenetic stimulation in low V2b CatCh expressing animals did not lead to a significant change in tail frequency ( Figure 7E, paired Student's t-test p=0.174, CatCh-sibling control group p=0.427). Finally, to determine whether visually-induced effects contributed to the locomotor slowing, we delivered a red-shifted excitation (~550 nm) incapable of activating CatCh to rostral segments in the normally-expressing Tg(gata3:Gal4; UAS:CatCh) line. Tail beat frequencies of the red-shifted stimulation did not vary from controls ( Figure 7G and H, paired Student's t-test, p=0.467). Activation of V2b neurons in the original experiment did reduce swim bout duration ( Figure 7I, Student's t-test p=4.65Â10 À3 ). However, given the expression of CatCh in ISNs and their known role in influencing swim bout incidence  it is not possible to ascertain whether this result arises from V2b inhibition, ISN serotonergic modulation, or a superposition of both. Bout length was not significantly altered in animals with low V2b but strong CSF-cN CatCh expression ( Figure 7I, Student's t-test p=0.113), though we note that CSF-cN activation after swim onset (as opposed to throughout) has been shown to reduce bout duration (Fidelin et al., 2015). All together these experiments demonstrate that stimulating V2b neurons reduces the speed of rhythmic tail movements during swim. In combination with our finding that V2b suppression increased swim frequency, these results provide evidence that V2b neurons play a role in speed control in motor circuits by acting as a brake on locomotor frequency.

Discussion
In this study we demonstrate that V2b neurons exert direct control over axial musculature in the larval zebrafish. The V2b population comprises two stable subclasses, defined by neurotransmitter identity: one subclass is exclusively glycinergic and the other mixed glycinergic/GABAergic. These distinct V2b-gly and V2b-mixed subclasses preferentially inhibit fast and slow motor neurons, respectively, analogous to the speed-dependent connectivity found in a diverse range of zebrafish spinal interneurons (McLean and Fetcho, 2009;McLean et al., 2008;Ampatzis et al., 2014;Svara et al., 2018). Moreover, we found that V2b activity serves as a brake on locomotor speeds: driving V2b neurons slows locomotion whereas suppressing V2b neurons speeds it up. Together, these results indicate that inhibition from V2b neurons is not restricted to enforcing agonist-antagonist muscle coordination but also influences locomotor speed through in-phase modulation of axial motor circuits.

V2b conservation across species
We demonstrated that V2b neurons are inhibitory and extend axons ipsilaterally and caudally in zebrafish (Figures 1 and 3) similar to V2b neurons in mice (Lundfald et al., 2007;Britz et al., 2015). Gata3+ V2b neurons are widely present in vertebrates but have also been identified in the nerve cord of a marine annelid, indicating an ancestral persistence in motor circuitry (Karunaratne et al., 2002;Vergara et al., 2017). V2b neurotransmitter profiles appear to vary throughout development and across species. Gata3-expressing cells in the embryonic 24 hr post fertilization zebrafish are predominantly GABAergic with a smaller subset expressing or co-expressing glycine (Batista et al., 2008), an inversion of our finding that V2b cells in 5-15 dpf zebrafish are all glycinergic with approximately half co-expressing GABA. Early in development, murine V2b neurons broadly co-express GABA and glycine (Zhang et al., 2014). By P0 in mouse, however, nearly all V2bs are glycinergic and~25% are GABAergic (Lundfald et al., 2007), which is broadly similar to our results. In zebrafish, these two subclasses persist out to 15 dpf, implying that they are stable identities. Consistent with the idea that V2b-gly and V2b-mixed are distinct identities, the two subclasses exhibit different axon trajectories in the DV axis, perhaps indicating responsiveness to different axon guidance cues. In mouse, V2b subpopulations have not been directly shown. However, the differential expression of the transcription factors Gata2/Gata3/BhlhB5 in non-overlapping neural subsets may imply their presence (Francius et al., 2013). More broadly, our finding of subclasses in the V2b population is parallel to previously identified genetically and anatomically distinct subclasses within the V0, V1, and V2a populations in the mouse and zebrafish (Bikoff et al., 2016;Hayashi et al., 2018;Menelaou et al., 2014;Song et al., 2018;Bjö rnfors and El Manira, 2016). Do V2b-gly and V2b-mixed populations match existing zebrafish neural classes?
Historically, zebrafish spinal neurons have been classified by anatomy. V2b (Gata3+) neurons are thought to correlate to ventral longitudinal (VeLD) neurons, an anatomically defined cell class with a characteristic longitudinal ventral-positioned axon (Batista et al., 2008;Hale et al., 2001;Bernhardt et al., 1992). How do the subpopulations we have described here relate to the VeLD population? Based on the ventral axon morphology and GABA co-expression, the V2b-mixed subtype represents a matured version of the embryonic VeLD neurons. In contrast, V2b-gly neurons are distinct in morphology and neurotransmitter profile from VeLDs, indicating either that they have not been previously identified in embryonic stages or that they develop at a later time.
In mice, the V2 progenitor domain gives rise to a third class of neurons called V2c, which express Sox1 and only transiently Gata3 in very early development (prior to E12.5) before later downregulation (Panayi et al., 2010). It is unclear whether zebrafish have a homologous V2c population, although Sox1a/b is present in the 24 hpf spinal cord and notably also colabels with Gata3 (Andrzejczuk et al., 2018). A possible V2c homolog, referred to as V2s, has been identified as a Sox1a/b+ glycinergic cell type deriving from the V2 domain with long, ipsilateral, caudally projecting axons (Gerber et al., 2019) similar to the V2b-gly neurons described here. However, Gata3 expression was not investigated in V2s neurons, leaving it unclear whether V2s neurons are in fact V2b-gly (Gerber et al., 2019). Given the persistent, distinguishing expression of Gata3 in both V2b-gly and V2b-mixed subtypes, our data are consistent with the designation of two subclasses within V2b, not a V2c homolog or additional V2s class. Further detailed investigation of Sox1a/b gene expression in these neurons will be required to clearly separate these classes.

Speed specific inputs to motor circuits
Locomotion at faster versus slower speeds engages different sets of spinal interneurons, both within a genetically defined population (McLean and Fetcho, 2009;McLean et al., 2008;Ampatzis et al., 2014;Zhong et al., 2011) andacross populations (McLean et al., 2007;Talpalar et al., 2013). Given the observation that slow motor neurons likely receive more input from V2b-mixed neurons whereas fast motor neurons receive largely V2b-gly input, it would be of interest to explore whether the V2b subpopulations are recruited at different speeds of locomotion. An alternative prediction is that, because V2b-gly and V2b-mixed somata extend over a shared DV range ( Figure 2D), they are both likely to be recruited at over the same range of locomotor speeds, in line with the principle that DV position dictates speed-dependent recruitment of both motor neurons and interneurons (Kimura and Higashijima, 2019;McLean et al., 2007;McLean et al., 2008). Overall the specific postsynaptic consequences of glycinergic and mixed glycinergic/GABAergic inhibition remains unresolved. We posit that the fast decay kinetics of glycine-mediated transmission generates a restricted window of inhibition onto fast motor neurons, which undergo a faster recruitment cycle than slow motor neurons. In contrast, a combination of fast glycine and slower GABA-mediated currents might be expected to offer a longer window of postsynaptic current integration in slow motor neurons. In much of the nervous system, glycine receptor-mediated currents traditionally show faster decay rates than those from GABA receptors (Awatramani et al., 2005), including at interneuron inputs onto motor neurons (Jonas et al., 1998;O'Brien and Berger, 1999;Russier et al., 2002); with some exceptions (Dumoulin et al., 2001;Eggers and Lukasiewicz, 2006). Moreover, GABA can act as a coagonist on glycine receptors and in fact shorten the postsynpatic inhibitory response (Lu et al., 2008). Simultaneous whole cell recordings from V2bs and motor neurons will be necessary to elucidate the postsynaptic effects of V2b-gly versus V2b-mixed inputs.
Intra-V2b connectivity suggests a possible 'gear shift' within the V2bs, with the V2b-gly and V2bmixed populations potentially inhibiting each other to enforce a given speed of swim. One caveat in interpretation of these results is that despite their different somatic positions, fast and slow motor neurons have overlapping dendritic fields (Svara et al., 2018). Therefore, it is possible that V2bmixed neurons make synapses onto all motor neurons, and differential receptor expression is responsible for their IPSC pharmacology ( Figure 4); meanwhile, V2b-gly might be responsible for other functions, such as suppression of dorsal horn sensory interneurons (Li et al., 2004). Paired recordings or higher resolution anatomical experiments will be required to distinguish these possibilities.

V2b manipulation indicates role in speed control
One role of ipsilateral inhibition is to mediate flexor-extensor alternations via Ia reciprocal inhibition from V2b and V1 populations (Britz et al., 2015;Zhang et al., 2014). Ipsilaterally descending propriospinal neurons may also stabilize left-right alternation, although specific ablation of inhibitory ipsilaterally descending neurons has not been tested (Ruder et al., 2016;Danner et al., 2017). Our work establishes that selective activation of V2b neurons decreases tail beat frequency (Figure 7) while suppression leads to its increase ( Figure 6). In contrast, genetic ablation of V1 neurons in mouse and larval zebrafish both led to a reduction in fictive locomotor speeds (Gosgnach et al., 2006;Kimura and Higashijima, 2019). In-phase V1 inhibition was found to limit slow motor neuron recruitment during fast movements through both direct and indirect (V1 onto V2a) synaptic pathways (Kimura and Higashijima, 2019). From this we surmise that ipsilateral inhibition from V1 and V2b shape distinct features of locomotion. V1 neurons may act to terminate motor neuron burst cycles whereas V2b neurons may limit overall speed, much like a brake.
The exact mechanism of V2b-associated speed changes remains unclear but is unlikely to be due to direct motor neuron spike shunting (Figure 4). One possible source of modulation is through local gating of excitatory drive on the dendrites, with V2b-mediated inhibition shunting coincident excitatory inputs during locomotion. Another possibility is that V2b neurons modulate motor circuitry via synapses onto other premotor targets, for example through V2b-V2a connectivity (M. Sengupta, data not shown), mirroring the V1 influence over motor circuitry via direct synapses onto both motor neurons and V2a intermediaries (Kimura and Higashijima, 2019). These results point to a need for greater exploration of connectivity among spinal interneurons.

Overall role of V2b in motor circuits
What is the functional role of V2b-mediated ipsilateral inhibition onto motor circuits? Three broad categories of V2b function occur. First, as discussed above, enforcing flexor-extensor and potentially forelimb-hindlimb alternation in limbed animals. Second, as supported in this work, V2b neurons may serve to titrate excitatory drive onto motor neurons differentially across varying speeds of movement. Measuring inhibitory conductances in motor neurons in vivo has revealed, surprisingly, that inhibition in-phase with excitation actually increases for increasingly strong movements (Kishore et al., 2014;Berg et al., 2007;Berg et al., 2008;Li and Moult, 2012), rather than diminishing to allow more powerful contractions. In this context, shunting ipsilateral inhibition might serve to enforce tight temporal control over spike timing via shortening membrane time constants. Surprisingly, ipsilateral inhibition does not appear to act as a source of gain control (Figure 4).
Thirdly, ipsilateral inhibition may act to isolate movements in certain behaviors that engage dedicated premotor circuitry, for example through the selective inhibition of interneurons during scratching versus swimming in turtle (Berkowitz, 2002;Berkowitz, 2007). Some V2b neurons, by virtue of their direct connections with ipsilateral motor circuits, could form part of the locomotor 'switch' from one behavior to another. A thorough investigation of V2b-gly and V2b-mixed recruitment during natural behaviors, such as speed transitions, turning, or balance, will allow us to better understand the similar or distinct ways that V2b subclasses influence locomotion. Animal care

Materials and methods
Adult zebrafish (Danio rerio) were maintained at 28.5˚C with a 14:10 light:dark cycle in the Washington University Zebrafish Facility following standard care procedures. Larval zebrafish, 4-7 days post fertilization (dpf), were kept in petri dishes in system water or housed with system water flow. Animals older than 7 dpf were fed rotifers daily. All procedures described in this work adhere to NIH guidelines and received approval by the Washington University Institutional Animal Care and Use Committee.

Single-cell photoconversion
Fluorescent protein photoconversion was performed on anesthetized and embedded 5 dpf Tg (gata3:Gal4; UAS:Kaede) animals using an Olympus FV1200 microscope. Single-plane confocal images were continuously acquired to monitor conversion progress while 500 ms bursts of 405 nm light (100% intensity) were applied to an ROI~1/10 th the size of the targeted soma to elicit photoconversion. Animals were removed from agarose and allowed to recover in system water for 1-3 hr. After recovery, fish were anesthetized, embedded, and imaged as above. Tiled image stacks were acquired over an area ranging from the most rostral processes to the most caudal with a minimum of 10% area overlap between adjacent fields of view to aid the image stitching process.

Single-cell dye electroporation
Tg(gata3:LoxP-dsRed-LoxP:GFP; gad1b:GFP) animals (5-6 dpf) were anesthetized in 0.02% MS-222 and three electroetched tungsten pins were placed through the notochord securing the animal to a Sylgard-lined 10 mm well dish. Forceps and an electroetched dissecting tool were used to remove skin and one segment of muscle fiber to expose the spinal cord. A pipette electrode filled with 10% Alexa Fluor 647 anionic dextran 10,000 MW (Invitrogen) in internal recording solution, was positioned to contact the soma of the target neuron. Dye was electroporated into the cell via one or more 500 ms, 100 Hz pulse trains (1 ms pulse width) at 2-5 V (A-M systems Isolated Pulse Stimulator Model 2100). Confocal imaging was performed as described above, after >20 min for dye filling.

Fluorescent hybridization chain reaction (HCR)
Animals were fixed at 5 dpf in 4% paraformaldehyde and in situ hybridization was performed according to the HCR v3.0 protocol (Choi et al., 2018) with noted modifications. Preparation, dehydration and rehydration steps 1 through 14 were replaced with steps 2.1.1 through 2.2.8 with a Heat Induced Antigen Retrieval (HIAR) option in place of Proteinase K treatment (King and Newmark, 2018;King and Newmark, 2013). In situ probes were designed and distributed by Molecular Technologies (Beckman Institute, Caltech) to target gata3, gad1b, glyt2 (slc6a5), DsRed, mCherry, and GFP. Samples were kept in 4x saline-sodium citrate solution at 4˚C prior to imaging. Samples were mounted in Vectashield (Vector Laboratories) or low-melting point agarose (Camplex SeaPlaque Agarose, 1.2% in system water) and positioned laterally on a microscope slides with #1.5 coverslip glass.
Confocal imaging 5-7 dpf larvae were anesthetized in 0.02% MS-222 and embedded in low-melting point agarose in a 10 mm FluoroDish (WPI). Images were acquired on an Olympus FV1200 Confocal microscope equipped with high sensitivity GaAsP detectors (filter cubes FV12-MHBY and FV12-MHYR), and a XLUMPLFLN-W 20x/0.95 NA water immersion objective. A transmitted light image was obtained along with laser scanning fluorescent images. Sequential scanning was used for multi-wavelength images. Z-steps in 3D image stacks range from 0.8 to 1.4 microns. Fluorescent in situ hybridization samples were imaged with an UPLSAPO-S 30x/1.05 NA and silicone immersion media. Spectral images were collected for Tg(gata3:zipACR-YFP; gata3:loxP-DsRed-loxP:GFP) animals to distinguish between expression patterns of overlapping fluorophores. Samples were excited with a 515 nm laser. Emission was collected with a PMT detector from 10 nm wide spectral windows across the emission range 525-625 nm for each z-plane. Spectral deconvolution was performed with Olympus Fluoview software.
The Tg(gata3:Gal4; UAS:CatCh) line labels both V2b and Kolmer-Agduhr/cerebrospinal fluid-contacting neurons (CSF-cNs). To ensure that evoked IPSCs derived from presynaptic V2bs rather than CSF-cNs, epifluorescent illumination was targeted 3-7 segments rostral to the recorded segment. A Polygon400 Digital Micromirror Device (Mightex) was used to provide patterned illumination in indicated recordings. Previous studies found that CSF-cNs have short ascending axons and do not contact any motor neurons besides the caudal primary (CaP) (Hubbard et al., 2016). CatCh expression in the Tg(gata3:Gal4; UAS:CatCh) line is variegated, with some animals showing strong CatCh expression throughout both CSF-cNs and V2bs, and others showing good expression in CSF-cNs and minimal expression in V2b cells. Additional control experiments were performed in animals with minimal V2b label to demonstrate the absence of contribution of CSF-cN synapses in these experiments. Strychnine and gabazine (SR-95531) were applied at 10 mM.
In current clamp experiments examining V2b effects on motor neuron firing, 30 ms depolarizing current steps were delivered to motor neurons in increasing amplitude steps (primary motor neurons: 25 pA steps; secondary motor neurons: 5-25 pA steps), alternating between delivery with and without V2b activation targeted 3-5 segments rostral to the recording. V2b neurons were activated with a 10-20 ms pulse of blue light, delivered either via epifluorescence of via the Polygon DMD, as above, that initiated 10 ms prior to the onset of the depolarizing step in order to maximize the peak of V2b inhibition during the current step. At resting membrane potentials, a modest (1-5 mV) nonspecific depolarizing current was often seen in response to this optogenetic stimulation; it was insensitive to strychnine, gabazine, and two blockers of serotonergic receptors (cyproheptadine and mianserin, both at 1 mM; data not shown), and therefore seems likely to arise from gap junctional networks of unclear identity. To prevent this from interfering with recordings, small squares of illumination were delivered via DMD to target just V2b neurons, or epifluorescence apertures were narrowed to the smallest possible, minimizing the depolarizing current. As a positive control for the efficacy of this protocol in driving V2b-mediated inhibition, we found that spiking in V2a neurons was significantly reduced by V2b optogenetic activation (data not shown), indicating that the absence of effect on motor neuron spiking was not due to a failure of V2b activation.
Motor neurons were identified by axon fill that extended into the musculature and/or by retrograde dye labeling from the muscle. For retrograde labeling, 4 dpf larvae were anesthetized (0.02% MS-222) and laid flat on an agarose plate. A Narishige micromanipulator in conjunction with a microinjection pump (WPI, Pneumatic Picopump) was used to deliver small volumes of dye (Alexa Fluor 568 dextran, 3000 MW) via glass pipette into the muscle. Fish recovered in regular system water and were subsequently used for recordings at 5-6 dpf.
Data were imported into Igor Pro using NeuroMatic (Rothman and Silver, 2018). Spike threshold was defined as 10 V/s, and custom code was written to determine spike width and afterhyperpolarization of the initial spike elicited by pulse steps, code available on Github (Bagnall-lab, 2019a; copy archived at https://github.com/elifesciences-publications/V2b_paper_igor_code). Input resistance was calculated by an average of small hyperpolarizing pulses. To isolate IPSCs, 10 mm NBQX was present in the bath and neurons were voltage clamped at the EPSC reversal potential.
Motor neurons at the dorsal extent of the distribution (>50% of distance from bottom of spinal cord to top) exhibited lower input resistances (mean ± SD: 287 ± 75 MW) and were considered 'fast' and the remainder, which exhibited higher input resistances (885 ± 367 MW) considered 'slow' (McLean et al., 2007). These groups mostly correspond to primary and secondary motor neurons, but some dorsally located bifurcating secondaries may be included in the fast group (Menelaou and McLean, 2012).

Image analysis
Image analysis was performed with ImageJ (FIJI) (Schindelin et al., 2012). Igor Pro 6 was utilized for data analysis and statistics unless otherwise noted. V2b cell counts and neurotransmitter coexpression was quantified manually by two researchers (R.C. and M.J.); no significant differences in quantification were detected. Gata3+ V2b cells were identified and marked (ImageJ Cell Counter) relative to spinal cord and segment boundaries, giving total V2b/segment quantities. Subsequently, each cell was evaluated for expression of fluorescent proteins marking Gad1b or Glyt2.
Transgenic line validation was performed with in situ hybridization and quantified by two researchers (M.B. and R.C.) with no significant discrepancy in results. A~3-5 mm z-stack projection was made in a cell-dense area of spinal cord spanning two to three segments for each animal. ROIs of neurons were drawn in one channel before checking whether there was colocalization in the other channel. Samples were quantified twice: once for completeness (percentage of endogenous RNA positive neurons also expressing the transgene) and once for accuracy (percentage of transgene labeled neurons positive for endogenous RNA). 3-7 animals were evaluated in each line.
For axon tracing, stitched projection images were made with the Pairwise stitching (Preibisch et al., 2009) ImageJ plugin. The overlap of the fused image was smoothed with linear blending and was registered based on the fill channel or the average of all channels. Photoconversion cell fill images underwent an extra processing step in which the bleached green channel was subtracted from the photoconverted red channel. The Simple Neurite Tracer plugin (Longair et al., 2011) was used to trace the axon projection and branching relative to marked spinal cord boundaries. Axon lengths are reported as the number of segments transversed.
Motor neuron dendrites were quantified from confocal z-stack images of Tg(mnx:GFP) 5 dpf animals. Images were cropped to a single hemisegment. The Weka Trainable Segmentation plugin (Arganda-Carreras et al., 2017) was used to segment the motor neuron image into three classifiers; somata, axons exiting the spinal cord, and dendrites. Classification was based on Hessian training features. Training was performed iteratively for each image. The binary segmented images were applied to mask all non-dendrite fluorescence (n = 4 hemisegments/animal; n = 3 animals.) Fluorescence was maximum intensity projected in the z-dimension, collapsed along the horizontal plane and normalized to give an estimate of motor neuron dendrite density in the dorsoventral plane of the spinal cord.
NMDA-induced behavior and ZipACR optogenetic supression 5 dpf Tg(gata3:zipACR-YFP) animals and clutchmates were anesthetized in 0.02% MS-222 and placed on an agar plate under a dissecting microscope. A complete spinal cord transection was made with Vannas spring microscissors, plus a sharpened pin if necessary, between spinal cord segments 2 and 5. Tail blood flow was monitored post-transection and throughout the preparation; animals with significantly reduced blood flow were euthanized and not used for recordings. After transection the animal briefly recovered in extracellular solution and then was embedded in a dorsal up position in 1.2% low melting point agarose. Solidified agarose surrounding the tail caudal to the transection was removed with a dissection scalpel. 200 mM NMDA (Sigma Aldrich) in extracellular solution was added to the dish. Recordings were initiated after tail movement began, typically 2-10 min later.
Behavior experiments were performed with a Scientifica SliceScope upright microscope equipped with a Fastec HiSpec1 camera and an Olympus Plan N 4x/0.10 objective. Image collection was made with Fastec acquisition software. Images were acquired at 200 Hz for 5 s. Optical stimulation was made with 100% intensity full field epi-illumination from a CoolLED pe300ultra source routed through a GFP filter cube (Chroma 49002). Recordings with optical stimulation were alternated with recordings without stimulation; n = 6-17 recordings for each animal.
Analysis was run in MATLAB R2017a with custom code adapted from Severi et al. (2018), available for download from Github (Bagnall-lab, 2019b; copy archived at https://github.com/elifesciences-publications/V2b_behavior). The caudal edge of the transection and the tail periphery were manually selected as tail boundaries and 10 points for tracking were evenly distributed along the body. The caudal-most tail point was used to calculate tail speed (mm/s) at each frame of the recordings. A tail speed threshold of 0.5 mm/s was used to distinguish true movement from tail drift. Tail movement amplitude was calculated as the maximum tail displacement in the initial second of each recording. Tail beat frequency was computed from left-to-right tail oscillations during manually identified movement bouts; 6-30 consecutive peaks were averaged for each recording.
Evoked swim with optogenetic activation 5-6 dpf Tg(gata3:Gal4; UAS:CatCh) and sibling animals were paralyzed and pinned to Sylgard dishes with three sharpened Tungsten pins. 3-4 segments of skin were removed and the recording electrode was placed on the body near the intermyotomal cleft. Swim was induced with a bipolar Tungsten electrode placed on an intact portion of skin~5 segments caudal to the recording site with an empirically determined stimulation level set to reliably evoke swims (5-50 V). Optogenetic stimulation was made with epifluorescent light through standard filter cubes and a 40x objective at 5% light intensity with 20 ms on/off light cycles under the following conditions: no light, GFP excitation, and RFP excitation. Electical stimulation occured 100 ms post light stimulation. Trials in which the animal initiated swim prior to the electrical stimulation were excluded from analysis. Optical stimulation was made 3-5 segments rostral and 3-5 segments caudal to the peripheral motor recording to assess the contributions of descending (V2b) and ascending (CSF-cN) inhibition to the motor activity. Each recording paradigm was repeated for 16-32 trials. Peripheral motor nerve signal was rectified and inter-beat interval frequencies were manually tabulated in Matlab by experimenter blind to treatment conditions. Trial averages are reported for each recording condition in each animal.

Resource sharing
Transgenic fish lines generated in this work are available from the authors upon request. Example data and statistics for soma quantification, V2b cell traces, motor neuron dendrite mapping, whole cell electrophysiology and optogenetic connectivity, behavior experiments, and example images are available for download from Dryad (DOI: https://doi.org/10.5061/dryad.1d78mt2). Custom written code for spike analysis and tail tracking can be downloaded from Github (https://github.com/bagnall-lab/). tail tracking code, Marquise Jones for cell counting work, and Dr. Paul Stein for his thoughtful insights. We gratefully acknowledge the Washington University Zebrafish Facility. Imaging experiments were performed in part through the use of Washington University Center for Cellular Imaging (WUCCI) supported by Washington University School of Medicine, The Children's Discovery Institute of Washington University and St. Louis Children's Hospital and the Foundation for Barnes-Jewish Hospital. This work was supported by funding through the National Institute of Health (NIH) R00 DC012536 (MWB), R01 DC016413 (MWB), a Sloan Research Fellowship (MWB), The Children's Discovery Institute of Washington University (MWB), NINDS F32 NS103247 (RAC), and the National Bio-Resource Project in Japan (SH). MWB is a Pew Biomedical Scholar and a McKnight Foundation Scholar. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.