ERR2 and ERR3 promote the development of gamma motor neuron functional properties required for proprioceptive movement control

The ability of terrestrial vertebrates to effectively move on land is integrally linked to the diversification of motor neurons into types that generate muscle force (alpha motor neurons) and types that modulate muscle proprioception, a task that in mammals is chiefly mediated by gamma motor neurons. The diversification of motor neurons into alpha and gamma types and their respective contributions to movement control have been firmly established in the past 7 decades, while recent studies identified gene expression signatures linked to both motor neuron types. However, the mechanisms that promote the specification of gamma motor neurons and/or their unique properties remained unaddressed. Here, we found that upon selective loss of the orphan nuclear receptors ERR2 and ERR3 (also known as ERRβ, ERRγ or NR3B2, NR3B3, respectively) in motor neurons in mice, morphologically distinguishable gamma motor neurons are generated but do not acquire characteristic functional properties necessary for regulating muscle proprioception, thus disrupting gait and precision movements. Complementary gain-of-function experiments in chick suggest that ERR2 and ERR3 could operate via transcriptional activation of neural activity modulators to promote a gamma motor neuron biophysical signature of low firing thresholds and high firing rates. Our work identifies a mechanism specifying gamma motor neuron functional properties essential for the regulation of proprioceptive movement control.


Introduction
In tetrapod vertebrates, stretch-sensitive mechanosensory organs called muscle spindles embedded in skeletal muscle inform the nervous system about muscle length, movement and position, in addition to participating in reflex control over muscle actions [1][2][3]. In amphibians and nonavian reptiles, spindle sensitivity during movements is maintained by axon collaterals of a type of motor neurons, called beta motor neurons [3][4][5][6], while alpha motor neurons exclusively connect to the extrafusal muscle fibers and are responsible for eliciting muscle contractions [7]. In mammals and birds, a third subtype of motor neurons, called gamma motor neurons, do not connect to the force-generating extrafusal skeletal muscle fibers but exclusively innervate the intrafusal fibers of muscle spindles [8][9][10][11][12]. The uncoupling of gamma motor neurons from muscle force generation and from receiving monosynaptic afferent input from muscle spindles proper [13][14][15] is thought to allow the nervous system to control muscle spindle-dependent proprioception independent of and in anticipation of movements [10,16]. Gamma motor neurons may thereby prime muscle spindles to signal expected changes in muscle length, allowing the nervous system to compare spindle signals generated by intended and actual movements, and to correct movements when both signals differ from each other [16][17][18]. Gamma motor neurons are further responsible for generating muscle tone by recruiting the force-generating alpha motor neurons through a monosynaptic spindle afferent feedback loop, thus assisting movement initiation and postural control [1,10].
Apart from their exclusive innervation of muscle spindles, the ability of gamma motor neurons to control muscle proprioception entails their acquisition of intrinsic biophysical properties distinguishing them from the muscle force-generating alpha motor neurons [19]. For instance, their low firing thresholds and ability to rapidly gear up high firing rates appear to be exquisitely suited for achieving near-instant intrafusal fiber tension for maintaining or modulating muscle spindle dynamic range [19]. Another feature allowing gamma motor neurons to effectively control muscle proprioception is their lack of monosynaptic (Ia) afferent feedback, which uncouples them from potential "short circuits" by their own actions on muscle spindle activity [13][14][15]. During embryonic development, combinatorial gene expression programs promote the generation of spinal motor neurons, the organization of their somas into motor columns and pools and axonal projections to peripheral targets [20], but much less is known about how motor neurons subsequently specialize to play different roles in movement generation or movement accuracy [20]. Such specialization of neurons can involve transcriptional programs that implement specific synaptic connectivity patterns, morphologies as well as fundamental biophysical properties [21,22], and which can operate in an overlapping or modular manner to establish neuron subtype identities [23]. Apart from their overall distinction from the gamma motor neurons, alpha motor neurons themselves exhibit a range of functional properties important for the adjustment of muscle force [24]. We and others have previously reported mechanisms underlying the diversification of alpha motor neurons proper, which involved both cell-autonomous actions by the noncanonical Notch ligand Dlk1 [25], as well as non-cell-autonomous signals by region-specific astrocytes [26].
Much has been learned in the past approximately 70 years about the diversification of motor neurons into alpha and gamma types and their respective contributions to movement control [8,16,19]. More recent studies have further begun to identify molecular markers facilitating histological and transcriptomic identification of gamma motor neurons, which include transcription factors, ion channel subunits, and neurotransmitter receptors [27][28][29][30][31][32][33][34][35] that could play roles in their specification and/or their properties. However, which of these or other yet to be identified markers would actually contribute (or how they would contribute) to the specification of gamma motor neurons or their properties remained unaddressed. Here, we studied the specification of gamma versus alpha motor neurons in mice and identify the orphan nuclear receptors ERR2 and ERR3 as determinants of gamma motor neuron-specific functional properties required for muscle spindle-dependent proprioceptive movement control.

Electrophysiological characterization of murine gamma motor neurons
Since the last recordings from mature gamma motor neurons dated from 1978 in the cat [19], we first sought to establish a baseline for gamma motor neuron functional properties in the mouse. We were aided in this by a fortuitous finding that allowed us to selectively record gamma motor neurons in the mouse spinal cord based on the relative efficacy of retrograde Fluoro-Gold (FG) uptake (Figs 1A, 1B, and S1A-S1F), which involves systemic FG administration, endocytosis by motor presynapses at the neuromuscular junction not enveloped by the blood-brain barrier, and retrograde transport to the motor neuron somas [35]. We observed that gamma motor neurons can be distinguished from alpha motor neurons based on their FG uptake, as small-soma motor neurons with high levels of FG retention (FG high ) lacked direct sensory neuron innervation and expressed low or negligible levels of NeuN, when compared to FG low motor neurons (Figs 1B and S1A-S1F). Upon performing whole-cell patch-clamp recordings at 20 to 22 days postnatally, we observed that the FG high motor neurons exhibited a  16) and FG low (n = 22, N = 12) motor neurons exhibit divergent gamma and alpha subtype-defining electrophysiological signatures, respectively, including gamma subtype-specific combination with low rheobase and high gain by FG high motor neurons (see S1 Table for details). (E) FG high motor neurons have significantly lower rheobase (pA) (221.87 ± 31.34), higher firing frequency (Hz) (50.65 ± 3.23), higher gain (Hz/nA) (161.01 ± 9.77), higher input resistance (136.62 ± 14.63), and lower capacitance (76.07 ± 6.01) when compared to FG low motor neuron rheobase (909.09 ± 82.11), firing frequency (20.41 ± 1.70), gain (32.64 ± 4.02), input resistance (36.55 ± 4.16), and capacitance (239.1 ± 17.17), respectively (see S1 Table for details). Data are presented as mean ± SEM. n = # of neurons, N = # of mice. Statistically significant differences between FG high and FG low neurons are indicated as ��� p < 0.001, Student t test). Data for Fig 1D and 1E can be found in S1 Data. biophysical signature matching that previously reported for cat gamma motor neurons [19] (Figs 1C-1E and S1G-S1I). The FG high motor neurons showed a distinctive combination of lower rheobases with higher firing frequencies and gains, as well as higher instantaneous and steady-state firing rates when compared to the larger FG low motor neurons (Figs 1D, 1E, S1G, and S1H and S1 Table). Moreover, FG high motor neurons showed significantly lower membrane input resistance, higher membrane capacitance, and lower AHP-decay times when compared to FG low motor neurons (Figs 1E and S1I and S1 Table). In contrast, FG low motor neurons were characterized by overall higher rheobases and lower firing rates and gains, as well as lower instantaneous and steady-state firing rates, similar to the alpha motor neurons of the cat [24] (Fig 1D and 1E and S1 Table). In addition to the overall quantitative differences in biophysical properties between FG high and FG low motor neurons, FG low motor neurons showed a range of combinations of biophysical properties, from relatively low rheobases plus low firing rates to high rheobases plus higher firing rates characteristic for slow, intermediate, and fast alpha motor neuron subtypes [24,25]. Gamma motor neurons in mouse therefore possess a distinctive biophysical signature matching that previously recorded for cat gamma motor neurons [19].

Correlated expression of orphan nuclear receptors ERR2 and ERR3 by gamma motor neurons
It had previously been established that the survival of gamma motor neurons relies on signals released by muscle spindles [27][28][29], which, in addition to the discovery of markers allowing in situ detection of gamma motor neurons [27][28][29][30][31][32][33][34], provided us with entry points for studying mechanisms underlying the diversification of motor neurons into alpha and gamma subtypes. We focused on the estrogen-related receptor (ERR) subfamily of orphan nuclear receptors, which primarily function as ligand-independent transcription factors [36,37], because of their contribution to cell type-specific properties in other contexts [38] and because of the previously reported expression of ERR3 by gamma motor neurons [27]. Upon immunodetection using antibodies specifically recognizing either ERR2 or ERR3 (S2A-S2Q Fig), we further found that the closely related ERR3 paralogue ERR2, with which it shares virtually the same DNA binding sequences [38], was coexpressed with ERR3 by gamma motor neurons (Figs 2A-2J and S3A-S3T), a coexpression that had been independently observed by others using single-cell RNAseq [32]. Through deeper analysis by quantitative immunodetection, we indeed found high levels of correlated expression (Pearson's correlation coefficient r = 0.86) of both ERR2 and ERR3 by motor neurons with relatively small somas characteristic for gamma motor neurons [15,27,28] and low or negligible levels of the alpha motor neuron marker NeuN [27,28] (NeuN low ) (Figs 2I, 2I', S3J, and 3J'). The small-soma ERR2/3 high NeuN low motor neurons lacked VGLUT1 + varicosities on somatic or dendritic membranes (Fig 2L and  2M), indicating absence of monosynaptic spindle afferent input, a defining characteristic of gamma motor neurons [14,15,27]. Similar to ERR3 [27], ERR2 was initially broadly expressed by most motor neurons during embryonic development (S4A- S4C Fig), but high ERR2 levels became increasingly confined to gamma motor neurons during the first 2 postnatal weeks (S4D- S4R Fig). Consistent with the dependency of gamma motor neuron maintenance on spindle-derived signals [27,28], we observed an absence of ERR2/3 high NeuN low motor neurons in adult Egr3-deficient mice with impaired muscle spindle development (Figs 2K, S4S-S4X, and S4S'-S4X'). At the same time, we noted that ERR2/3 low , NeuN high motor neurons were maintained in Egr3-deficient mice (S4S'-S4X' Fig reconstruction studies in control mice showed that, indeed, FG high , ERR2/3 high , NeuN low gamma motor neurons were contacted by few VGLUT1 + terminals, while FG low , ERR2/3 low , NeuN high alpha motor neuron somas were covered with VGLUT1 + varicosities (Fig 2L-2N). Mouse gamma motor neurons thus exhibit high levels of correlated ERR2 and ERR3 expression. We surmise that previous work using qualitative Esrrb mRNA detection by in situ hybridization, which concluded that Esrrb/ERR2 was expressed by slow alpha motor neurons [39], was impeded by the inherent inability of these methods to distinguish between the pronounced quantitative differences of Esrrb/ERR2 expression levels between gamma and alpha motor neurons and to distinguish between the expression of Esrrb/ERR2 by motor neurons and by the interspersed ventral interneurons retained in Egr3-deficient mice.

ERR2/3 are required for the acquisition of a gamma motor neuron biophysical signature
Due to their correlated expression as well as their molecular similarity, we next asked whether ERR2/3 would contribute to gamma versus alpha motor neuron functional diversification by selectively inactivating both Esrrb and Esrrg genes in motor neurons via Cre-mediated recombination in cholinergic neurons in Esrrb flox/flox ;Esrrg flox/flox ;Chat Cre (ERR2/3 cko ) mice (S2O-S2Q Fig). Because Chat Cre -mediated recombinase activity, which is initiated in early postmitotic motor neurons (S5M-S5T Fig), overlapped with endogenous ERR2/3 expression in motor neurons but not in other cholinergic neuron types throughout the nervous system (S5A-S5L and S6A-S6Z Figs), we concluded that ERR2/3 cko mice permitted addressing ERR2/3 function in motor neurons. Since Esrrb flox/flox ;Chat Cre. and Esrrg flox/flox ;Chat Cre single-homozygous mice showed only mild impairment of movements, compared to compound ERR2/3-deficient mice (S1-S6 Movies), we reserved further analysis to the ERR2/3 cko mice. Whole-cell patch-clamp recordings performed in 20-to 22-days-old ERR2/3 cko mice showed that most FG high motor neurons failed to acquire a gamma motor neuron biophysical signature and instead shifted their properties towards a signature resembling those of FG low alpha motor neurons (compare Fig 3B and 3E), with significantly elevated rheobases, as well as lowered gains and firing rates (Figs 3A-3F, S7A, and S6J and S1 Table). Since FG high motor neurons in ERR2/3 cko mice showed average firing rates about the same of those for alpha motor neurons (compare Figs 3C, 3F, S7A, and S7F) but still retained somewhat higher gains and lower rheobases (Figs 3C, 3F, and S7B), their overall biophysical signature seemed to have shifted to somewhere between those of slow or fast alpha motor neuron subtypes, thus to some extent resembling yet immature early postnatal gamma motor neurons [30]. At the same time, we did not observe significant changes in the properties of FG low alpha motor neurons in ERR2/3 cko mice, when compared to control mice (Figs 3D, 3F, and S7F-S7J and S1 Table), suggesting that the low ERR2/3 levels in alpha motor neurons were not sufficient to significantly influence the acquisition of their characteristic properties. ERR2/3 thus appear to be prerequisite for the acquisition of a gamma motor neuron biophysical signature.

Development of morphologically distinguishable gamma motor neurons in the absence of ERR2/3
Since we observed that loss of ERR2/3 in gamma motor neurons in ERR2/3 cko mice led to the loss of a gamma motor neuron biophysical signature, we next asked whether ERR2/3 would also be required for the specification of other aspects of gamma motor neurons, including morphology, marker expression, and synaptic connectivity. In addition to their unique electrophysiological features, gamma motor neurons characteristically possess small soma sizes, lack 1a sensory presynaptic boutons, and innervate muscle spindle intrafusal fibers [13,14,26,27]. Notably, the lack of ERR2/3 in ERR2/3 cko mice did not affect the generation of small-soma FG high and NeuN low motor neurons (Fig 4A-4I), consistent with the unaltered input resistance measured in these neurons (S7E Fig). Similar to control mice (Fig 4E, 4F and 4I), the small-soma FG high , NeuN low motor neurons in ERR2/3 cko mice lacked VGLUT1 + varicosities on somatic or dendritic membranes (Fig 4G, 4H and 4I). Moreover, FG low and NeuNhigh alpha motor neurons retained VGLUT1 + terminals in both ERR2/3 cko (Fig 4C, 4D and 4I) and control mice (Fig 4A, 4B and 4I). We next investigated whether lack of ERR2/3 in motor neurons would affect the innervation and/or development of muscle spindles. All muscle spindles analyzed in ERR2/3 cko mice exhibited overall organization indistinguishable from control mice, including normally formed annulospiral sensory endings in the central segments (Figs 4J-4O and S8A-S8F; n = 10 muscle spindles in control, n = 8 in ERR2/3 cko mice), while 100% of the postsynaptic boutons in the peripheral segments of intrafusal muscle fibers were  Fig 1D). (F) No significant differences seen in ERR2/3 cko FG low alpha motor neuron subtype rheobase (855.55 ± 124.85), firing frequency (25.03 ± 3.76), and gain (43.04 ± 9.81) when compared to control FG low alpha motor neuron subtype rheobase (909.09 ± 82.11), firing frequency (20.41 ± 1.70), gain (32.64 ± 4.02), respectively (see S1 Table for details). Data are presented as mean ± SEM. n = # of neurons and N = # of mice. Statistically significant differences between FG high and FG low neurons are indicated as � p < 0.05, �� p < 0.01, ��� p < 0.001, n.s. = not significant, Student t test). Data for Fig 3B, 3C, 3E, and 3F can be found in S1 Data.
https://doi.org/10.1371/journal.pbio.3001923.g003 supplied by motor axon termini in both control and ERR2/3 cko mice (arrowheads in Figs 4J-4O and S8A-S8F; n = 71 and n = 115 intrafusal neuromuscular synapses, respectively). We further found that the expression of the gamma motor neuron molecular marker GFR1A was maintained in the small-soma FG high , NeuN low motor neurons in ERR2/3 cko mice (S9A- S9H  Fig). Moreover, consistent with the absence of detectable impacts of ERR2/3 removal on alpha motor neuron biophysical properties, 100% of the neuromuscular junctions with extrafusal muscle fibers analyzed were supplied by motor axon termini and showed characteristic pretzel-like morphologies in ERR2/3 cko mice, indistinguishable from control mice (S8G-  could not be strictly ruled out. The function of ERR2/3 in gamma motor neurons thus appears to be to a large degree restricted to implementing their characteristic biophysical signature.

ERR2/3-dependent gamma motor neuron biophysical properties are required for movement accuracy
The biophysical properties of gamma motor neurons are thought to be exquisitely suited to effect instant intrafusal fiber peak tension for maintaining and modulating spindle dynamic range and thus muscle proprioception [19]. We therefore asked how a shift towards an alpha motor neuron-like biophysical signature in gamma motor neurons would impact movements relying on proprioceptive feedback from muscles. Because of the exclusive association of high ERR2/3 levels with gamma motor neurons, and the lack of a significant impact of ERR2/3 loss on other motor neuron subtypes, we predicted that beta motor neuron function would be preserved in ERR2/3 cko mice, thus allowing us to study the contribution of gamma motor neuron function to movement control in the otherwise intact animal. Similar to Egr3-deficient mice lacking muscle spindles [40,41], ERR2/3 cko mice exhibited marked postural and gait alterations (S1-S6 Movies), including changes in metrics related to foot placement, weight bearing, stride, stance, braking and propulsion (Figs 5A, 5B, and S10A-S10E), consistent with the predicted contributions of gamma motor neuron-assisted spindle function to posture, gait phase transitions, and force generation during locomotion [10,15,17,42]. ERR2/3 cko mice were nevertheless able to sustain the same range of speeds as control mice in a treadmill locomotion task, suggesting that apart from the perturbance of gamma motor neuron-dependent muscle proprioception, the overall integrity of the skeletomuscular system was preserved (S10A-S10C Fig). However, ERR2/3 loss from motor neurons triggered a failure to handle precision tasks, such as navigating a narrow horizontal beam (Fig 5C and 5D and S7 and S8 Moviess) or a horizontal ladder (Fig 5E and 5F and S9 and S10 Movies), consistent with the particular sensitivity of precision tasks towards perturbations of muscle proprioception [40,41]. These data suggested that biophysical signature implemented by ERR2/3 in gamma motor neurons is prerequisite for effective modulation of muscle spindle-dependent proprioception and, thereby, the execution of precision movements. To further test this idea, we recorded spindle afferent responses via suction electrodes in nerve-muscle preparations [43,44] derived from ERR2/3 cko or control mice. In these preparations, muscle stretch applied by a force transducer elicited similar Ia afferent responses in ERR2/3 cko and control mice (Figs 5G and S10F-S10I), consistent with the morphologically normal spindle assembly in these animals. In contrast to control spindles (Figs 5G, 5G', S10F, and S10G), however, ERR2/3 cko spindle afferents frequently exhibited reduced firing rates at muscle resting length (Figs 5G, 5G', S10H, and S10I), possibly due to a decrease in basal intrafusal fiber contractility caused by chronic disruption of gamma motor neuron input. The ERR2/3-dependent implementation of gamma motor neuron biophysical properties therefore appears to be prerequisite for regulating spindle-mediated muscle proprioception and movement control.

PLOS BIOLOGY
ERR2/3 implement gamma motor neuron properties required for precision movements mouse gamma motor neurons. Indeed, forced expression of ERR2 or ERR3 partially shifted motor neuron properties towards a biophysical signature resembling that of mouse or cat gamma motor neurons [19], including a combination of high firing rates and lower rheobases ( Fig 6E-6J and S2 Table), although the absolute values of the parameters differed between chick and mouse, likely due to differences in maturation-stage and soma sizes between the pre-hatching chick and postnatal mouse motor neurons analyzed. The effects on motor neuron properties were enhanced by fusing ERR2 or ERR3 to the heterologous transcriptional activation domain VP16 (Fig 6H-6J and S2 Table), but not by fusion to the engrailed transcriptional repressor domain (EnR) (Fig 6H-6J and S2 Table), suggesting that in this context, ERR2 and ERR3 primarily function as transcriptional activators. Finally, forced coexpression of ERR2 and ERR3 in chick motor neurons recapitulated the effects on motor neuron biophysical properties by forced expression of either factor alone (Fig 6H-6J and S2 Table), consistent with their apparent mutual redundancy in mouse. Although it remained unclear whether the properties of avian gamma motor neurons would be mediated by the same mechanisms as those in mammals, these results nevertheless suggested the possibility that ERR2/3 could not only be necessary, but also sufficient to promote gamma motor neuron-like functional properties by operating as transcriptional activators.
ERR2/3 may operate through neural activity modulators, including the shaker K + channel subunit Kcna10/Kv1.8 To identify through which intermediate factors ERR2/3 would operate to promote gamma motor neuron-like electrophysiological properties in chick, we performed comparative transcriptome profiling by RNA sequencing of chick motor neurons forcedly expressing ERR2 (Figs 7A and S11A). In these experiments, elevated ERR2 levels significantly activated a set of genes largely distinct from the gene signature activated by the previously identified (fast) alpha motor neuron determinant Dlk1 [25] (Figs 7A, S11A, and S11B and S3 Table). The set of genes whose expression was altered by ERR2 expression were enriched in potential neural activity modulators (S11C and S11D Fig), including Kcna10, which encodes Kv1.8, a member of the shaker family of voltage-gated K + channels implicated in neuronal excitability [45] (Figs 7A and S11B). We found that the promoter region of the Kcna10 genomic locus contained an evolutionary conserved region (ECR) with 3 clustered ERR2/3 DNA binding motifs (Fig 7B) that were conserved between mouse and chick (S11E Fig). In chick motor neurons, moreover, ERR2 boosted reporter gene activity driven by the Kcna10 ECR (Figs 7C and S11F-S11K), but not upon introducing mutations into the ECR's ERR2/3 binding motifs (Figs 7B, 7C, S11N, and S11K). Ultimately, forced expression of Kcna10 shifted motor neuron electrophysiological properties in a manner partially recapitulating those elicited by ERR2/3, including lower rheobases and higher firing rates (Fig 7D-7H and S2 Table). However, not all parameters associated with murine gamma motor neurons changed significantly (Fig 7G and S2 Table), suggesting that in chick, ERR2 would operate through additional neural activity modulators (S3 Table). While small motor neurons in birds express NKAα3 [46], a marker for gamma motor neurons in mouse [31], and high levels of Esrrb and Esrrg are expressed by a subset of motor neurons in late-gestation chick in a pattern resembling that previously observed for Esrrb in mouse motor neurons [39] (S11L-S11O Fig), the extent to which the mechanisms underlying gamma motor neuron specification or function would be conserved between avian and mammals, if at all, remained unclear. Intriguingly, we found Kv1.8 to be selectively expressed by small soma-size NeuN low gamma motor neurons in mouse (Fig 8A-8D and 8I). Moreover, the expression of Kv1.8 in small soma-size NeuN low motor neurons was abolished in ERR2/3 cko mice (Fig 8E-8H and 8J), thus neatly complementing the observed up-regulation of Kcna10 upon ERR2 overexpression in chick. Taken together, the similarity of the biophysical signatures promoted by ERR2/3 in chick motor neurons with the ERR2/3-dependent gamma motor neuron biophysical signature in mouse, the presence of conserved clustered binding sites for ERR2/3 in the promotor region of both mouse and chick Kcna10 genomic loci, the promotion of Kcna10 expression by ERR2/3 in chick depending on these binding sites, and the ERR2/3-dependent expression of Kv1.8 in gamma motor neurons in mouse suggest that ERR2/3 operate as conserved transcriptional activators, by implementing a gene expression program encoding neural activity modulators, to promote a gamma motor neuron biophysical signature. However, despite compelling similarities, the extent to which chick and mouse gamma motor neurons and the mechanisms underlying their specification or properties would be comparable remains to be determined.

Discussion
The contribution of gamma motor neurons to proprioceptive movement control in mammals has long been established [8][9][10], but despite recent progress in assigning markers and gene expression signatures to these neurons [27][28][29][30][31][32][33][34], how they are specified and/or acquire their unique properties remained unaddressed. In the present study, we identified ERR2/3 as determinants of the divergence of gamma and alpha motor neuron biophysical properties, which may involve the transcriptional activation of genes encoding neural activity modulators ( Fig  7K). The ERR2/3-dependent establishment of a gamma motor neuron biophysical signature, in turn, appears to be necessary for regulating muscle spindle-dependent muscle proprioception underlying movement accuracy.

Gamma motor neuron generation and maturation
Previous results suggest that between P0 and P6, the biophysical properties gamma motor neurons are yet immature and lie in between those of slow and fast alpha motor neurons [30], thus somewhat resembling the properties of ERR2/3-deficient gamma motor neurons. Therefore, while gamma motor neurons proper are generated before birth [29], and slow and fast alpha motor neuron properties are already distinguishable by P7 [25], a distinctive gamma motor neuron biophysical signature emerges sometime between P7 [30] and P20. This period may therefore approximate the time window of gamma motor neuron maturation instigated by ERR2/3, raising the question whether their actions could be superimposed on an initial alpha motor neuron-like groundstate [25,26]. Two observations suggest that the latter is not the case. First, the gamma motor neuron biophysical signature is quite distinct from those of the alpha motor neuron subtypes, for instance, combining high gains and firing rates with low rheobases, in contrast to the combination of either low gains/firing rates with intermediate rheobases (slow alpha motor neurons) or intermediate gains/firing rates with high rheobases (fast alpha motor neurons) [24,25], which would be inconsistent with a purely additive effect of ERR2/3 on immature (alpha motor neuron-like) gamma motor neuron properties. Second, we observe that not only activation, but also (likely indirect) repression of potential neural activity modulator genes upon forced ERR2/3 expression in chick (S3 Table), suggesting that ERR2/3 may actually reconfigure the combination of such genes, including voltage-gated K + and Ca 2+ channel subunits, expressed by gamma motor neurons. About the significance of the expression of ERR2/3 by most motor neurons before birth, we can at present only speculate. It is conceivable, however, that ERR2/3 actions are delayed by the postnatal activation of yet to be identified cofactors (possibly by muscle spindle-derived signals), critical for the ability of ERR2/3 to activate genes that together promote mature gamma motor neuron biophysical properties.

Gamma motor neuron subtypes and beta motor neurons
Two types of gamma motor neuron output, static or dynamic, are thought to normally tune muscle spindle sensitivity during different aspects of movements [10,16,47]. Three observations led us to conclude that the actions of ERR2/3 do not distinguish between both types of gamma motor neurons. First, we found ERR2/3 to be expressed by all gamma motor neurons. Second, consistently, gamma motor neuron biophysical properties appeared to be disrupted as a whole upon loss of ERR2/3. Third, the range of movements affected by ERR2/3 removal from motor neurons (from basic locomotion to precision movements) further suggest that ERR2/3 function would be required for both static and dynamic modulation of spindle function. The mechanistic bases underlying the two different outputs of gamma motor neurons to spindles thus remain to be addressed. We noted that Kv1.8 was not expressed by all gamma motor neurons, which may provide a starting point for investigating the basis for the distinction between the two predicted gamma motor neuron subtypes. How about the beta motor neurons? Beta motor neurons connect to both extrafusal and intrafusal muscle fibers but are thought to be otherwise morphologically indistinguishable from alpha motor neurons [48][49][50][51]. Since beta motor neurons, like the alpha motor neurons, possess relatively large soma sizes and receive monosynaptic Ia afferent input [50,51], we were able to rule out that ERR2/3 were also operating in beta motor neurons. Since beta motor neurons so far have mostly been studied in the cat and have been identified solely based on their simultaneous innervation of intrafusal and extrafusal fibers [51], both the abundance of beta motor neurons and their significance for movement control in mouse awaits further study. However, the identification by single-cell transcriptome profiling of a motor neuron gene expression signature distinct from those of gamma and alpha motor neurons could aid this quest [33].

Gamma motor neuron gene expression signatures
Apart from Esrrb/g proper, we observed little, but notable overlap between the genes activated by ERR2/3 in chick motor neurons and previously published single-cell RNAseq data linked to murine gamma motor neurons [32][33][34] (S3 Table). One notable example was Cacng3 [34], the protein product of which can be incorporated with that of Cacna1c, also linked to mouse gamma motor neurons [34], into the same L-type Ca 2+ holochannels [34]. Similarly, while Kcna10/Kv1.8 did not show up in any of the single-cell data sets [32][33][34], one of them contained Kcna5 [32], a related shaker channel subunit gene. The limited overlap between the chick and mouse data sets likely had three reasons. First, despite their obvious power for revealing the cell type composition of organs and tissues, these single-cell approaches inherently capture only part of a cell's transcriptome [52]. This is exemplified by the absence of Kcna10/Kv1.8 and other gamma motor neuron markers identified in mouse (for instance, Atp1a3/NKAα3 [46] and Wnt7a [29]) from the independent single-cell transcriptome data sets, as well as the relatively limited overlap between these data sets proper [32][33][34]. Second, ERR2/3 likely targets only a fraction of gamma motor neuron genes, while the list of potential target genes we identified in chick is unlikely to be complete. This may have also been because the chick motor neurons were analyzed at stages at which they were not yet mature, which, in turn, could have resulted in insufficient chromatin accessibility of ERR2/3 target sites and may have also contributed to the paradoxical down-regulation by ERR2 of some genes associated with mouse gamma motor neurons, including Grin2a [34] and Ret [32] (S3 Table). Third, there may be species-specific differences in the genes targeted by ERR2/3 and/or low levels of conservation of the corresponding genes/proteins leading to differences in annotation, which together can obscure actual relationships. An example of this is ENSGALG00000015953 (S3 Table), encoding a protein with similarity to Serotonin receptor genes linked to mouse gamma motor neurons [34]. Because of these limitations, it will be interesting to apply single-cell transcriptomics to mouse motor neurons lacking ERR2/3 to more directly appraise the set of genes through which these factors likely promote gamma motor neuron properties in mouse.

Gamma motor neurons of birds and mammals
Whether the presence of gamma motor neurons in both birds and mammals reflect a common phylogenetic origin of these neurons or rather convergent adaptations remains unclear [53]. On the one hand, the observation that subsets of chick motor neurons share molecular markers with mouse gamma motor neurons [46], including Esrrb/g, together with the regulation of Kcna10 and other genes linked to mouse gamma motor neurons by ERR2/3 in chick , could point to a common phylogenetic origin of avian and mammalian gamma motor neurons and the mechanisms underlying their specification. On the other hand, it is also conceivable that a preexisting ERR2/3-dependent gene regulatory program was adopted independently in the avian and mammalian lineages to promote similar neuron subtype-specific properties, which, in turn, contributed in parallel to improved motor control by facilitating alpha motor neuron-independent regulation of muscle prorioception [53]. A more fine-grained comparison of motor neuron gene expression between avian and mammalian motor neurons, as well as extant vertebrates with more ancestral spinal motor systems, could shed more light on the possible phylogenetic relationships and origins of these neuron types.

Outlook
In addition to gamma motor neurons, ERR2

Microscopy, image analysis, and quantification
Fluorescence microscopy was performed using Zeiss LSM 710 and LSM 800 laser scanning microscopes. ApproximatelyAU : PerPLOSstyle; numeralsarenotallowedatthebeginningofasentence:Ple 6 to 28 optical sections were obtained at a step-size of 0.8 to 1.5 μm. Care was applied to avoid oversaturation and distortion of relative expression levels during image acquisition. Raw images were imported into ImageJ and z projected at maximum intensity. For quantification of expression levels, raw pixel intensities were quantified in individually outlined motor neuron nuclei "region of interests" (ROIs) using Adobe Photoshop CS5.1. using background levels and the neuron with the highest fluorescent intensity for normalization. For quantification of VGLUT1 + varicosities, a step-size of 0.80 μm was used to obtain an average of 20 optical sections. Raw Z-stack Carl Zeiss files (.czi) were imported into Imaris 8.0 (Bitplane AG, Zurich, Switzerland). Neuronal surfaces were rendered to detect VGLUT1 varicosities ("spots") on motor neuron somas and dendrites using the "find spots close to surface" function [61] and guided by specific parameters.

Electrophysiology of gamma versus alpha/beta motor neurons in mice
Mice (P20-22) were intraperitoneally injected with 0.5% to 2% (w/v) FG (Flurochrome, Denver, CO) dissolved in PBS (pH = 7.2) at a volume of 0.10 ml/10 g body weight. The animals (1-day post-FG injection) were intraperitoneal injected with 100 mg/kg body weight of Ketamine, 20 mg/kg body weight Xylazine in PBS (pH = 7.2) at a volume of 0.10 ml/10 g of body weight. After losing their righting reflex, they were placed on a bed of ice until loss of toe pinch response. Immediately after, the animals were decapitated and quickly eviscerated. The torso was placed in chilled Dissecting aCSF (DaCSF) solution (in mM): 191 sucrose, 0.75 K-gluconate, 1.25 KH 2 PO 4 , 26 choline bicarbonate (80% solution), 4 MgSO 4 , 1 CaCl 2 , 20 dextrose, 2 kynurenic acid, 1 (+)-sodium L-ascorbate, 5 ethyl pyruvate, 3 myo-Inositol. The solution was maintained at pH 7.3 using carbogen (95% O 2 −5% CO 2 ), and osmolarity was adjusted to approximately 305 to 315 mOsm with sucrose. Vertebrectomy was performed to extract the spinal cord. Ventral roots were cut and the meninges were removed from the spinal cord. The thoracolumbar region (T10-L5) of the spinal cord was isolated and embedded in agar block (4% agar in Recording aCSF (RaCSF)) using 20% gelatin in RaCSF). Slices (370 μm) were obtained using Leica VT1200 S (Leica Biosystems, GmbH, Nussloch Germany). The slices were incubated in 35˚C RaCSF for 30 minutes and 30 minutes at room temperature before the recordings. Motor neurons (MNs) were recorded in the RaCSF solution (mM): 121 NaCl, 3 KCl, 1.25 NaH 2 PO 4 , 25 NaHCO 3 , 1.1 MgCl 2 , 2.2 CaCl 2, 15 dextrose, 1 (+)-sodium L-ascorbate, 5 ethyl pyruvate, 3 myo-Inositol. The solution was maintained at pH 7.4 using carbogen (95% O 2 −5% CO 2 ), and osmolarity was adjusted to approximately 305 to 315 mOsm with sucrose. Whole-cell patch-clamp recordings were performed from FG-labeled motor neurons in the ventral horn from control and ERR2/3 cko animals. FG high and FG low MNs were visually identified using Olympus BX51W1 microscope (Olympus Europa SE & Co. KG, Hamburg, Germany) equipped with an FG longpass filter set (350 nm bandpass and 425 nm longpass filter) (AHF analysentechnik AG, Tübingen Germany). The patch pipette (resistances of 3 to 6 MΩ) was filled with intracellular solution (mM): 131 K-methanesulfonate (or MeSO 3 H), 6 NaCl, 0.1 CaCl 2 , 1.1 EGTA-KOH, 10 HEPES, 0.3 MgCl 2 , 3 ATP-Mg 2+ salt, 0.5 GTP-Na + salt, 2.5 L-glutathione reduced, 5 phosphocreatine di(tris) salt. The solution pH was adjusted to 7.25 with KOH, and the osmolarity was adjusted to 300 mOsm using sucrose. Data analysis was performed offline using Axograph X Version 1.6. Previously established protocols were applied to obtain membrane properties of rheobase, input resistance, capacitance, F-I curve, AHP amplitude, half-width and half-decay times [25,[62][63][64][65][66]. For obtaining the F-I curve for discharge properties, spikes were elicited by applying 20 pA, 1,000 ms square current pulses to cells. Currents up to 1 nA were injected for all neurons. For the mouse recordings, currents of up to 1 nA were injected into FG high and 3 nA into FG low MNs from control and ERR2/3 cko spinal cord slices. The firing frequency (Hz) was defined as the inverse of the duration between first two spikes (instantaneous firing frequency), or 0.25 to 0.75 seconds of the current pulse (steady-state firing frequency), or 1 second current pulse (mean firing frequency). The gain (Hz/nA) was defined as the slope of the regression line of mean firing frequency upon current injection [65].

Gait analysis
Control (n = 9) and ERR2/3 cko (n = 8) mouse locomotion on a treadmill was recorded through automated high-speed motion capture (DigiGait, Mouse Specifics, Framingham, USA) as described previously [25,67]. This method generates over 50 different gait variables, which exceed the number of observations (8 to 9 animals per genotype) and are partially redundant. We therefore used the Partial Least Squares (PLS) regression, which is optimized for predictive modeling of multivariate data and to deal with multicollinearity among variables. Orthogonal Signal Correction PLS (OSC-PLS) was used as an extension of PLS to separate continuous variable data into predictive and uncorrelated information for improved diagnostics as well as more easily interpreted visualization. The method seeks to maximize the explained variance betweengroups in a single dimension and separates the within-group variance (orthogonal to classification goal) into orthogonal dimensions or components. We modified an existing R script [68,69], originally designed for chemometrics analysis [70], to adapt it to our behavior data. The OSC-PLS method was applied to the complete centered-scaled data set in order to define a model. A model with an optimal number of 2 components was used for subsequent analysis in both fore and hind limbs. Our model coefficient of determination Q2, i.e., the model's fit to the training data, and its Root Mean Square Error of Prediction (RMSEP), i.e., the model's predictive ability on the testing data were calculated using the Leave-One-Out method [68]. The model was finally validated to ensure that it was performing better than a random model, while not being overfitted. We conducted an internal cross-validation by performing permutations in the original data, from which 2/3 was used to fit a model. This model was then used to predict group memberships on the remaining 1/3 testing set. The process was repeated 100 times, and Q2 and RMSEP values were averaged over the repeats. We finally compared our model's Q2 and RMSEP values to the mean Q2 and mean RMSEP values of the permuted models. The results of the two-sample Student t tests used for the comparisons indicated a probability much inferior to 0.1% of achieving a performance similar to our model by random chance.

Precision movement tasks
Precision movements were successively tested using a custom-built setup with a 100-cm horizontal beam of 20, 25, and 30 mm width, respectively [71]. Age-matched control (n = 4) and ERR2/3 cko (n = 4) 10-month-old female mice were trained to move across the beam into a home cage at its end. The animals were trained for 3 days (4 trials/animal, bidirectional on the beam) and tested on the fourth day (4 to 5 trials/animal, bidirectional on the beam). Furthermore, a custom-built setup to record skilled locomotion on a 100-cm horizontal ladder, with 3 mm rungs spaced at 14 mm (similar to a setup previously described) [43,44] was used to test age-matched 8-weeks-old female control (n = 5) and ERR2/3 cko (n = 5) mice. Mice were trained to move across the ladder into a home cage placed at its end and were trained for 2 days (3 trials/animal, single direction on the ladder) for 2 weeks and tested on the third day of the second week (4 to 5 trials/animal, single direction on the ladder). The animals were rested in their home cage for 1 minute between trials for both tasks. Animal locomotion was recorded using GoPro HD Hero2 (GoPro, San Mateo, USA) fitted to a custom-built slider track. The videos were acquired at 120 fps at an image size of 848 × 480 and stored as MP4 files. The videos were processed using GoPro Studio Version (2.5.4) and proDAD Defishr Version 1.0 (pro-DAD GmbH, Immendingen, Germany). The figure videos were slowed to 25% to 40% of original speed and reduced to 60 fps using GoPro Studio Version (2.5.4). The fish-eye view was removed from videos using proDAD Defishr Version 1.0 (proDAD GmbH, Immendingen, Germany) with Mobius A Wide preset and Zoom adjusted to 110.0 or 180.0. A "miss" was scored when the mouse paw failed to locate the rung or the beam leading to the animal to slip or to halt until the paw regained its footing. An average of approximately 40 steps per trial were analyzed for each mouse.

Muscle spindle afferent recordings
Ex vivo extensor digitorum longus (EDL) muscle-nerve preparations were used to study the response of muscle sensory neurons to stretch and was essentially performed as described [43]. Briefly, the dissection of muscle (with the nerve attached) was performed in low calcium and high magnesium solution [43,72]. Then, the muscle was placed in recording solution 22 to 24˚C and equilibrated with carbogen and was then hooked to a dual force and length controller-transducer (300C-LR, Aurora Scientific, Aurora, Canada) with the help of 5-0 sutures tied to its tendons. Following the determination of resting length (Lo) as described in previous studies, a suction electrode (tip diameter 50 to 80 μm) connected to an extracellular amplifier (EXT-02F, npi Electronics GmbH, Tamm, Germany) was used to sample muscle spindle afferent activity. Data acquisition was performed with LabChart 8 connected to PowerLab 8/35 (ADInstruments, Oxford, UK). Afferent activity, when obtained was checked for the presence of a characteristic pause following a series of 30 twitch contractions at 1 Hz and if the pause was present, the unit was identified as a muscle spindle afferent. Then a series of 9 ramp and hold stretches were delivered to the muscle (2.5% Lo, 5% Lo, and 7.5% Lo at 40% Lo/second, protocols were kindly provided by K.A. Wilkinson lab). The data were recorded and analyzed offline with a custom written MATLAB code. Spikes were detected using KMEANS [2]. For each afferent, resting discharge (RD), dynamic peak discharge (DP), dynamic index (DI), and static response (SR) were calculated. A total of 10 control and 10 ERR2/3 cko animals were used for recordings, from which 8 control and 9 ERR2/3 cko spindle afferents were analyzed.

Molecular cloning
Mouse Esrrb (NM_011934.4) (ERR2) and Mouse Esrrg (NM_011935.3) (ERR3) open reading frames were isolated using PrimeScript 1st cDNA synthesis Kit Takara Bio Europe SAS, Saint-Germain-en-Laye, France) following manufacturer's directions from E18.5 mouse spinal cord total RNA and cloned after PCR amplification with the following oligonucleotide primers: Esrrb The chick Kcna10 (NP_989793) open reading frame was isolated from E5.5 chick embryo total RNA as above using the following primers: Kcna10 Thus, synthesized cDNAs were subcloned into an expression vector between CMV promoter in frame with an Aphthovirus 2A peptide-GFP fusion sequence for cotranslational cleavage [73]. This entire cassette was flanked by Tol2 sites facilitating genomic integration upon cotransfection with Tol2 transposase as described [25]. ERR2VP16 was generated by fusing the open reading frame for the herpes simplex virus-1 (HSV-1) VP16 (amplified from pActPL-VP16AD plasmid, Addgene plasmid #15305, Watertown, USA) activation domain to ERR2. ERR2EnR was generated by fusing the open reading frame of the transcriptional repressor domain of Drosophila Engrailed (amplified from CAG-EnR plasmid, Addgene plasmid #19715, Addgene, Watertown, USA) to ERR2.

Chick motor neuron electrophysiology
Chick embryos electroporated at E2.7 to E3.0 (HH stages 14 to 18) with appropriate DNA constructs were harvested at E12 to E15 (HH stages 38 to 41) and processed as described previously [25]. Briefly, chick embryos were placed on ice for 5 minutes, extracted from the egg, decapitated, and dissected in a petri dish containing cold chick aCSF (CaCSF) solution (mM): 139 NaCl, 3 KCl, 1 MgCl 2 , 17 NaHCO 3 , 12.2 dextrose, 3 CaCl 2. The solution pH was adjusted to 7.25 with KOH and the osmolarity was adjusted to approximately 315 mOsm using sucrose. The thoracolumbar region of the spinal cord (with the vertebral column intact) was isolated and embedded in an agarose block (4% agarose in CaCSF) using 20% gelatin in CaCSF. Slices (370 μm) were obtained using a Leica VT1200 S vibrating blade microtome (Leica Biosystems GmbH, Nussloch, Germany) and incubated in CaCSF for 30 minutes at room temperature (22˚C). Motor neurons were visualized by GFP expression using 4× air objective (Olympus UPlan FL N) and 40× water-immersion objective (Olympus UPlan FI N) equipped on an Olympus BX51W1 microscope. The patch pipette (resistances of 3 to 6 MΩ) was filled with intracellular solution (mM): 130 MeSO 3 H, 10 KCl, 2 MgCl 2 , 0.4 EGTA, 10 HEPES, 2 ATP-Mg 2+ salt, 0.4 GTP-Na + salt, 0.1 CaCl 2. The solution pH was adjusted to 7.3 with KOH, and the osmolarity was adjusted to approximately 315 mOsm using sucrose. The intracellular solution contained 25 μM Alexa Fluor 568 dye (Thermo Fisher Scientific, Waltham, USA) to label recorded motor neurons. Current clamp recording signals were amplified and filtered using MultiClamp 700B patch-clamp amplifier (Molecular Devices, San Jose, USA). The signal acquisition was performed at 20 kHz using Digidata 1322A digitizer (Molecular Devices, San Jose, USA) and pCLAMP 10.4 software (Molecular Devices, San Jose, USA).

RNA sequencing and transcriptome analysis
E12.5 (HH St. 38 to 39) chick lumbar spinal motor columns transfected with either CMV:: eGFP, or CMV:: ERR2VP16.2A.eGFP or Dlk1.IRES.eGFP plasmids were identified by GFP fluorescence, dissected and collected. RNA isolation and RNA sequencing were carried out as described previously [73]. Briefly, RNA was isolated using Tri-Reagent (Sigma-Aldrich Chemie GmbH, Taufkirchen, Germany) and Phenol-Chloroform extraction according to the manufacturer's protocol. RNA quality was assessed using Nanodrop 2000 (Thermo Fisher Scientific, GmbH) and RNA integrity number (RIN) was evaluated by using the Agilent 2100 Bioanalyzer (Agilent Technologies, USA). RNA was reverse transcribed to cDNA using Transcriptor High Fidelity cDNA synthesis kit (Roche Diagnostics Deutschland GmbH, Mannheim, Germany) and RNA-Seq libraries were obtained using TruSeq RNA Sample Preparation v2 kit (Illumina, San Diego, USA). To analyze the library quality, Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, USA) was used and the concentration was measured by a Qubit R dsDNA HS Assay kit (Thermo Fisher Scientific, Waltham, USA). The concentration was adjusted to 2 nM prior to sequencing (50 bp) on a HiSeq 2000 sequencer (Illumina, San Diego, USA) using TruSeq SR Cluster kit v3-cBot-HS (Illumina, San Diego, USA) and TruSeq SBS kit v3-HS (Illumina, San Diego, USA) based on manufacturer's instructions. RNAsequencing quality was evaluated by utilizing raw reads using the FastQC quality control tool version 0.10.1 [74,75]. Bowtie2 v2.0.2 using RSEM version 1.2.29 with default parameters was utilized to align sequence reads (single-end 50 bp) to chicken reference genome (Galgal5) [76,77]. Prior to indexing, GFP, ERR2, Dlk-1, VP16, and IRES sequences and annotations were added to the reference genome (FASTA file) and annotations (GTF file). Ensembl annotations (version 86.5) with rsem-prepare-reference from RSEM software was used to index chicken reference genome [78]. Furthermore, sequence alignment of sequence reads and gene quantity was obtained through the use of rsem-calculate-expression. Rsem-calculate-expression resulted in sequence read count and TPM value (transcripts per million) for individual genes. DESeq2 package was used to carry out differential expression analysis [79]. Finally, genes with less than 5 reads (baseMean) were filtered, while genes with an adjusted pvalue < 0.05 were classified as differentially expressed. Gene ontologies and categorization was performed using the DAVID Gene Functional Classification Tool [80].

Enhancer identification and promoter assays
Evolutionary conserved noncoding genomic regions (ECRs) around the Kcna10 genomic locus were identified using the ECR Browser [81] and screened for potential ERR2/ERR3 transcription factor binding sites using the JASPAR CORE database [82]. A 240-bp ECR 3.5 kb upstream of the Kcna10 transcription start site with 3 putative ERR2/ERR3 binding sites was amplified from mouse genomic DNA using the following primers: Forward 5 0 -TCTCA-CAGCCCTGCTCATC-3 0 and Reverse 5 0 -CTTGCCTGAGAACCTGATCTCC-3 0 and subcloned into a reporter vector containing a minimal promoter followed by tdTomato coding sequence, which together were flanked by Tol2 sites to facilitate stable genomic integration. To test promoter activity and potential regulation by ERR2/3, Lohmann LSL fertilized chick eggs were incubated until E2.7 to E3.0 (HH stages 14 to 18) and chick embryo neural tubes were electroporated in ovo using the ECM 830 electroporation system (BTX/Harvard Apparatus, Holliston, USA) as described [25].  [27] in the FG high , NeuN low motor neurons (arrowheads). (C) FG high motor neurons are NeuN low (arrowheads) (D) and ERR3 high (arrowheads). (E, F) Scatter plots of motor neuron fluorescence levels over soma sizes at P21 (n = 159) (note: data from same experiment depicted in S4Q and S4R Fig): identification of gamma motor neurons based on a combination of soma size and FG levels. (E, F) Gamma motor neurons with small somas and relatively high FG levels that express high ERR3 levels (E), but low or negligible levels of NeuN (F). Larger motor neurons with lower but substantial FG levels express low or negligible levels of ERR3 (E), but mostly higher levels of NeuN (F). (G-I) Control FG high (gamma) motor neurons (black bars) exhibit significantly higher instantaneous firing frequency (G), instantaneous gain (H), and lower AHP-half decay time (I) when compared to FG low (alpha) motor neurons (gray bars). Statistically significant differences are indicated as: � p < 0.05, �� p < 0.01, ��� p < 0.001, n.s. = not significant, Student t test). Data for S1G-S1I Fig can be found in S1 Data and for S1E and S1F  S1 Table for details). (F-J) No significant differences between ERR2/3 cko FG low (alpha) motor neurons (red bars) and control FG low (alpha) motor neurons (red bars) when comparing instantaneous firing frequency (F), instantaneous gain (G), AHP-half decay time (H), capacitance (I), and input resistance (J), respectively (see S1 Table for  Optimized model prediction was used to assign data sets for fore and hind limbs to either genotype (control versus ERR2/3 cko ) and the two components of the models were plotted against each other. Each one of the tested animals is represented by a single dot, while polygons group the animals of the same genotype. The amount of betweengroups variance explained by each component in the model is expressed in percent of the total between-groups variance. The two components of our optimized models captured more than 25% of the variance in the predictors in both fore and hind limb at all treadmill speeds. These scores indicate that the method was able to capture the maximum variance between genotypes in the first dimension, which is also shown by the absence of overlap between the two groups on the x-axis, together indicating that ERR2/3 cko exhibit significant gait alterations compared to control mice. Yet, all ERR2/3 cko mice analyzed were able to successfully complete the treadmill locomotion tasks at all speeds tested (provided as the number of animals "n" running until "completion" for each speed). (D, E) Ranking of the variables' predictive capacities in the forelimb (D) and hind limb (E) models. The most predictive parameters display the highest loadings (arbitrary units) independent of their sign. The mean value of all mice from the same group was calculated for each parameter and depicted in the bar charts. The sign of the loadings only indicates the direction in which gait parameters are affected (increased, red, or reduced, blue, in ERR2/3 cko compared to control mice). (F-I) Examples of Ia afferent recordings from extensor digitorum longus (EDL) nerve-muscle preparations. Red traces: Ia afferent firing, blue traces: relative muscle length, green traces: relative muscle tension upon application of muscle stretch with a force transducer. (F, G) In the control preparation, Ia afferents exhibit relatively low tonic firing rates at resting length, which rapidly increases upon application of muscle stretch with two characteristic bursts during stretch onset and offset. (H, I) In Err2/3 cko mice, Ia afferents frequently fall silent during resting length, while exhibiting a responsiveness towards stretch similar to control Ia afferents. Data for S10A-S10E