Abstract
Movement and posture depend on sensory feedback that is regulated by specialized GABAergic neurons (GAD2+) that form axo-axonic contacts onto myelinated proprioceptive sensory axons and are thought to be inhibitory. However, we report here that activating GAD2+ neurons, directly with optogenetics or indirectly by cutaneous stimulation, facilitates sensory feedback to motoneurons in awake rodents and humans. GABAA receptors and GAD2+ contacts adjacent to nodes of Ranvier at branch points of sensory axons cause this facilitation, preventing spike propagation failure that is otherwise common without GABA. GABAA receptors are generally lacking from axon terminals (unlike GABAB) and do not inhibit transmitter release onto motoneurons, disproving the long-standing assumption that GABAA receptors cause presynaptic inhibition. GABAergic innervation of nodes near branch points allows individual branches to function autonomously, with GAD2+ neurons regulating which branches conduct, adding a computational layer to the neuronal networks generating movement and likely generalizing to other CNS axons.
Main
The ease with which animals move defies the complexity of the underlying neuronal circuits, which include corticospinal tracts (CSTs) that coordinate skilled movement, spinal interneurons that form central patterns generators (CGPs) for walking, and motoneurons that ultimately drive the muscles1. Sensory feedback ensures the final precision of such motor acts, with proprioceptive feedback to motoneurons producing a major part of the muscle activity in routine movement and posture2–4, without which severe ataxia occurs5. Proprioceptive sensory feedback is regulated by specialized GABAergic neurons (GAD2+; abbreviated GABAaxo neurons) that form axo-axonic connections onto the sensory axon terminals6–8. These neurons are thought to produce presynaptic inhibition of sensory feedback to motoneurons9–11 and possibly limit inappropriate sensory feedback3, 4, 7. However, during movement the CST, CPG and even sensory neurons all augment GABAaxo neuron activity11–16 right at a time when sensory feedback is known to be increased to ensure precision and postural stability2–4, raising the question of whether GABAaxo neurons have a yet undescribed excitatory action.
The long-standing view that GABAergic neurons produce presynaptic inhibition of proprioceptive sensory axon terminals in adult mammals actually lacks direct evidence, largely because of the difficulty in recording from these small terminals and the technical limitations of previously employed reduced spinal cord preparations (immature)7, 17 or anesthetized animals, since anesthetics themselves modulate GABA receptors9, 18. Thus, in this paper we used optogenetic approaches to directly target GABAaxo neurons in awake animals and in isolated whole adult spinal cord preparations. Surprisingly, we found that optogenetically activating these GABAergic neurons markedly facilitates sensory axon transmission to motoneurons, throwing into doubt presynaptic inhibition.
GABAA receptors on sensory axon terminals have long been assumed to cause presynaptic inhibition, based on indirect evidence11, 18, though their mode of action is rather counterintuitive. Sensory axons, like many other axons, have high intracellular chloride concentrations, leading to an outward chloride ion flow through activated GABAA receptors11, 19, 20. Thus, GABAA receptors cause a depolarization of sensory axons (primary afferent depolarization, PAD)11, 16, 21–23. PAD and associated GABAA receptors have variously been theorized to cause presynaptic inhibition by depolarization-dependent inactivation or shunting of sodium currents at the sensory axon terminals22, 24. However, we do not even know if terminals of large myelinated proprioceptive sensory axons express GABAA receptors at all, despite their demonstrated innervation by GABAaxo neurons6. These terminals appear to lack the α5 subunit of extrasynaptic GABAA receptors16, but this leaves open the possibility that they express synaptic GABAA or GABAB receptors. We thus examined this question here and unexpectedly found that synaptic GABAA receptors are generally not at these axon terminals, but are instead near sodium channels (NaV) at the nodes of Ranvier throughout the myelinated regions of the axon, together with innervation by GABAaxo neurons. What then is the function of such nodal GABAA receptors?
An unexplored possibility is that the depolarizing action of nodal GABAA receptors (and GABAaxo neurons) aids sodium spike propagation between axon nodes. This has not previously been considered, as spikes are thought to securely propagate from node to node, at least in the orthodromic direction16. Myelinated proprioceptive axons branch extensively in the spinal cord16(Fig. 1a) and each branch point poses a theoretical risk for spike propagation failure25, 26. However, branch points are always located near nodes (NaV)16, likely to minimize this failure. Nevertheless, indirect evidence has suggested that propagation failure can occur27–30. Thus, in the present study we sought direct evidence of nodal spike failure at branch points and examined whether nodal GABA and PAD facilitates afferent conduction by preventing this failure. We already know that PAD lowers the threshold for initiating axon spikes by extracellular stimulation31, and even initiates spikes16, but do not know whether it aids normal spike propagation. We found that spike propagation depends so heavily on nodal GABA that blocking GABA action makes the majority of proprioceptive sensory axons fail to propagate spikes to motoneurons, and thus GABA provides a powerful mechanism to turn on specific nodes and branches to regulate sensory feedback.
Results
Nodal GABAA and terminal GABAB receptors
To visualize how GABA regulates axons we examined the distribution of GABAA receptor subunit immunolabelling on nodes and terminals in large myelinated proprioceptive sensory axons, including α5 subunits that are extrasynaptic, α1 and α2 subunits that are mostly synaptic, and ubiquitous γ2 subunits32. We labelled sensory axons with neurobiotin injections in the rat spinal cord (Fig. 1) or fluorescent reporters in VGLUT1Cre/+ mice (Extended Data Fig. 1), and reconstructed them in 3D. GABAA receptors containing α5, α1, α2 and γ2 subunits were expressed on these axons, especially near sodium channels (< 6 µm away; Fig. 1d-h, Extended Data Fig. 1). Specifically, GABAA receptors were on large myelinated 1st and 2nd order branches in the dorsal and ventral horn (Figs. 1d-e,h; Extended Data Fig. 1) near their nodes (identified by large sodium channel clusters, paranodal Caspr, and an axonal taper; < 10 µm away; Figs. 1c,h,j), and on short unmyelinated terminal branches in the dorsal horn (3rd order; Figs. 1b,g), the latter near nodes on 2nd order branches (< 100 µm away). In contrast, GABAA receptors were mostly absent from the long unmyelinated terminal branches contacting motoneurons in the ventral horn (3rd order; Figs. 1b,f,g; Extended Data Fig. 1a-e), which also generally lacked sodium channels16. This left GABAA receptors on average far from the terminal boutons contacting motoneurons (∼500 µm; Fig. 1h) relative to the axon space constant (λS ∼90 µm), the majority in dorsal and intermediate laminae. Nodes were widely spaced, as were branch points (∼50 µm separation, Fig. 1i), but branch points were always near nodes (100%, within 7 µm; NaV; Fig. 1c-e,i) and associated GABAA receptors. While nodes sometimes occurred without branching (49%), the majority of nodes expressing GABAA receptors were at branch points (Fig. 1j), implying an importance for branch point GABA. In contrast, GABAB receptors were found mostly on terminal branches in the ventral horn where boutons had dense receptor expression (Figs. 1f-h), and not usually on larger myelinated ventral or dorsal branches (not at nodes; Figs. 1e,g,j; Extended Data Fig. 1f).
Propagation failure in dorsal horn axon branches
Considering that GABAA receptors are expressed in large myelinated branches of proprioceptive axons, we next directly recorded from of these branches in the dorsal horn of rat and mouse spinal cords (Figs. 2 and 3) to examine whether spike propagation depends on these receptors. When we stimulated the dorsal root (DR) containing the axon branch, an all-or-nothing spike was recorded in many branches (Figs. 2b, 3d) at the latency of the fastest afferent volley that arrived at the spinal cord (group I afferents; EC in Fig. 2b). However, in other axon branches this spike did not occur (∼20%), but at the same latency there was a small all-or-nothing residual spike (failure potential, FP). This FP was indicative of a spike activating a distant node, but failing to propagate further to the recording site, leaving only its passively attenuated potential, with smaller FPs reflecting more distal failure points in the spinal cord (Figs. 2c-g, 3e-f). Failure never occurred in the DR itself (Fig. 2f). The failing branches with FPs were otherwise indistinguishable from non-failing axon branches, exhibiting full spikes (> 60 mV) with current injection pulses (Fig. 2cii, g), and low conductances and resting potentials (∼ −65 mV, Fig. 2h), ruling out penetration injury. With high repetitive DR stimulation rates all branches (100%) exhibited propagation failure and an associated FP (Fig. 2e-g), again with the FP implying that the spike is reliably initiated in the DR, but incompletely propagates within the spinal cord.
Axon spike failure was voltage dependent: in branches with failing spikes (FPs) depolarizations that brought the axon closer to threshold enabled full DR-evoked spikes (via current injection, Fig. 2ci; or spontaneous depolarization Fig. 2d). Also, in branches without spike failure at rest (secure spikes) a steady hyperpolarizing current induced spike failure (FP), with more branches failing with increasing hyperpolarization (Extended Data Fig. 2). With increasing hyperpolarization, nodes failed progressively more distal to the electrode, causing abrupt drops in the overall spike amplitude with each failure and a characteristic delay in the nodal spike prior to failure (Extended Data Fig. 2a-d). Simulating spike propagation by applying a brief current pulse to mimic the current arriving from an upstream node (and FP) yielded similar results, with full spikes evoked at rest, but hyperpolarization leading to a spike delay and then failure (Extended Data Fig. 3). Large depolarizations inactivated spikes, though outside of the physiological range (> - 50 mV, Extended Data Fig. 3b-c).
Nodal spike facilitation by GABA
Since sensory axons are tonically depolarized by spontaneous GABA activity16, we wondered whether this GABA aids propagation. Blocking extrasynaptic α5 GABAA receptors (with L655708) or all GABAA receptors (with gabazine) increased the incidence of spike failure (to ∼45% and 65%, respectively; Fig. 2f) and sensitivity to hyperpolarization (Extended Data Fig. 2e-h), without altering overall spike properties (Fig. 2g), implying that spike propagation is highly dependent on nodal GABA receptors. Application of 5-HT to mimic natural brainstem-derived 5-HT also increased failure (Fig. 2f), likely via its indirect inhibition of GABAA receptor activity33.
Nodal spike facilitation by GABAaxo neuron activation
To examine whether GABAaxo neurons facilitate spike propagation, we expressed light-sensitive channelrhodopsin-2 (ChR2) in GAD2+ neurons in adult GAD2CreER/+;R26LSL-ChR2-EYFP mice (termed GAD2//ChR2-EYFP mice, Fig. 3). A brief light pulse (5 - 10 ms) produced a long-lasting depolarization and spiking in these GABAaxo neurons (Fig. 3a), followed by a longer lasting GABAA-mediated depolarization (PAD) of proprioceptive axons at a monosynaptic latency that was blocked by gabazine (Fig. 3a-b). In these mice, spikes in proprioceptive axons failed with a similar incidence as observed in rats (Figs. 3c-h), but the light-evoked PAD prevented this failure (Fig. 3e-g), similar to direct depolarization. In branches with secure non-failing spikes, light had minor effects (Fig. 3d), but blocking GABAA receptors again increased the incidence of spike failure (Fig. 3h).
In GAD2//ChR2-EYFP or GAD2//ChR2-EYFP//tdTom mice the EYFP and tdTom reporters labelled GABAergic neurons (Fig. 3k; VGAT+, GAD2+ and VGLUT1-) residing near the central canal and throughout much of the dorsal horn (Fig. 3i-o). These neurons densely innervated the dorsal horn with terminal boutons (Fig. 3j,l,n), and less densely innervated both the ventral horn and dorsal columns (Fig. 3l,m,o), allowing GABAergic innervation of sensory axons along their entire length. They made both synaptic (VGAT+) and perisynaptic contacts all along proprioceptive sensory axons, both at nodes and sensory axon terminals on motoneurons (Figs. 3k, and 1e), confirming their identity as GABAaxo neurons.
Computer simulations of branch point failure and rescue by GABA
To establish that spike failure arises at the branch points where GABA can influence them, we generated a computer simulation of a proprioceptive sensory axon arbour in the spinal cord34. With simulated DR stimulation, spike failure occurred distal to complex branch points (at nodes N2 and N3 in Extended Data Fig. 4a-b) that had associated increases in net conductance, which shunted the nodal currents. Simulated nodal GABAA receptor activation rescued these failed spikes, with increasing GABAA activation (gGABA) preventing more branch point failures (Extended Data Fig. 4c). In contrast, when we moved all these GABAA receptors to the terminals, then their activation did not rescue failed spikes (Extended Data 4d). This is because GABAA-induced depolarizations (PAD) were attenuated sharply with distance (λS ∼90 µm); so only nodal, and not terminal, induced PAD was visible at the dorsal columns (Extended Data Fig. 4a,g-h), in agreement with previous terminal recordings16.
Spike facilitation by sensory evoked GABAaxo activity
We next examined whether natural activation of GABAaxo neurons affects proprioceptive axon conduction. GABAaxo neurons are indirectly activated by sensory activity via two variants of a trisynaptic circuit, where sensory axons drive excitatory neurons that activate GABAaxo neurons and cause PAD: one driven by cutaneous afferents and the other by proprioceptive afferents (Extended Data Figs. 5a and 6a)16. As expected, following DR stimulation these circuits caused fast synaptic and slower extrasynaptic GABAA receptor mediated depolarizations of proprioceptive axons (termed sensory-evoked phasic PAD and tonic PAD, respectively16) that were blocked by GABAA antagonists, and mimicked by optogenetic activation of GABAaxo neurons (Fig. 4a-d).
Like with direct GABAaxo activation, spike propagation failure was prevented by sensory-evoked phasic PAD, regardless of whether the failure was spontaneous (Figs. 4e-f,h), 5-HT-induced (Fig. 4h), or repetition-induced (Extended Data Fig. 6b-f). The latter is particularly important because sensory axons naturally fire at high rates, where they are vulnerable to spike failure (Fig. 2e-f). This action of phasic PAD was abolished by gabazine but not L655935, supporting a synaptic origin (Fig. 4h). Slow extrasynaptic GABAergic depolarization (tonic PAD; L655935-sensitive16) further facilitated spike propagation (Fig. 4g), especially as it built up with repeated DR stimulation (at 1 Hz; Extended Data Fig. 5b). Cutaneous (Extended Data Fig. 5), proprioceptive (Extended Data Fig. 6) or mixed afferent (Fig. 4e-h) -evoked PAD all helped prevent spike failure.
In secure non-failing axon branches sensory-evoked PAD (or optogenetic GABAaxo activation) sped up the spikes and lowered their threshold (rheobase current; Fig. 3d and Extended Data Fig. 7a-d), as predicted from computer simulations (Extended Data Fig. 4e). Importantly, spike height was only slight reduced during PAD (∼1% or 1 mV) indicating that nodal GABAA receptor conductances have minimal shunting action (Fig. 3 and Extend Data Fig. 7a-d).
Failure of axon conduction to motoneurons and rescue by PAD
To quantify the overall failure of spikes to conduct from the DR to the sensory axon terminals we measured whether axon branches not conducting during failure were not refractory to subsequent stimulation with a microelectrode in the ventral horn (Extended Data Fig. 8). This method indicated that about 50 – 80% of sensory axons failed to conduct to their ventral terminals under resting conditions, especially in long axons, whereas sensory-evoked PAD decreased failure to < 30%. Similar conclusions were reached by directly recording the extracellular afferent volley in the ventral horn produced by the spikes propagating from a DR stimulation to the motoneurons, which was consistently increased by PAD (Extended Data Fig. 9).
Facilitation of sensory feedback to motoneurons by nodal GABAA receptors
To examine the functional role of nodal GABA in regulating sensory feedback to motoneurons, we recorded monosynaptic excitatory postsynaptic potentials (EPSPs) from motoneurons in response to proprioceptive sensory axon stimulation (Fig. 5a). This EPSP was inhibited by optogenetically silencing GABAaxo neurons with light in mice expressing archaerhodopsin-3 (Arch3, induced in GAD2CreER/+;R26LSL-Arch3-GFP mice; abbreviated GAD2//Arch3, Fig. 5a-b,d), consistent with a tonic GABAA receptor tone facilitating spike transmission in axons. Likewise, the EPSP was reduced when sensory axon conduction was reduced by blocking endogenous GABAA receptor tone with antagonists, despite increasing motoneuron and polysynaptic reflex excitability (minimized with APV, Fig. 5c,d). GABAB antagonists slightly increased the EPSP, suggesting a tonic GABAB-mediated presynaptic inhibition (Fig. 5d), though much smaller than the tonic GABAA-mediated nodal facilitation that dominates when all GABA was reduced (in GAD2//Arch3 mice).
Consistent with nodal GABAA receptors (and PAD) facilitating axon conduction, the monosynaptic EPSP was facilitated during, but not after, depolarizing proprioceptive axons (evoking PAD) with an optogenetic activation of GABAaxo neurons in GAD2//ChR2 mice (10 ms light conditioning stimulation; Fig. 5e-f). The EPSP was also facilitated by naturally activating GABAaxo neurons by a sensory conditioning stimulation, including with a conditioning stimulation of cutaneous and/or proprioceptive afferents (Extended Data Fig. 10a,b,e). The latter indicates that proprioceptive activity primes subsequent proprioceptive reflex transmission (self-facilitation). GABAA receptor antagonists (gabazine), but not GABAB antagonists (CGP55845), blocked the EPSP facilitation with sensory (Extended Data Fig. 10e) or light (Fig. 5f) conditioning.
The facilitation of the EPSP by conditioning-evoked PAD arose from axonal GABAA receptors, rather than from postsynaptic actions on the motoneurons, since it occurred with weak conditioning stimuli that produced only a transient background postsynaptic depolarization that terminated before the EPSP testing (at 60 ms; Figs. 5e, Extended Data Fig. 10b,g), followed by a slight hyperpolarization that if anything reduced the EPSP (shunting the synaptic current, Extended Data Fig. 10h). Increasing the DR conditioning intensity produced large background depolarizing conductances in the motoneurons during the EPSP testing, which led to postsynaptic inhibition of the EPSP (shunting inhibition; Extended Data Fig. 10d,g), masking the effect of nodal facilitation. Importantly, sometimes PAD itself induced afferent spikes (Extended Data Fig. 7e; termed DRR spikes), and following these spikes, the EPSP was always smaller than when these spikes were not present (n = 8/8 mice, not shown). This is because these DRR spikes themselves triggered EPSPs, leading to a post activation depression, as noted by Eccles9, and thus we minimized DRR activity by keeping the conditioning-evoked PAD small.
Sensory conditioning was particularly effective when it was repeated to mimic natural firing, which increased tonic PAD for minutes (Fig. 5g). This facilitated the EPSP for ∼3 min after a brief fast DR repetition (200 Hz, 0.5 s conditioning, Fig. 5i, Extended Data Fig. 10e, Tonic), and ∼1 min after slower repetition (0.1 Hz, 2 min conditioning, Extended Data Fig. 10e, After effect), both long outlasting postsynaptic effects from each conditioning pulse (< 1 s). This was blocked by L655708 or gabazine (Extended Data Fig. 10e).
Increases in the probability of unitary EPSPs by nodal facilitation
We often noticed large all-or-nothing EPSPs (unitary EPSPs) spontaneously fluctuating on and off during repeated EPSP testing, leading to discrete large changes in the total EPSP size and time course (Fig. 5j-k). We thought this might be due to spontaneous branch point failures, rather than quantal changes in transmitter release that produce much smaller fluctuations35, as previously suggested28. Indeed, when we increased the axon conduction by activating the GABAaxo neurons and PAD (via a cutaneous conditioning train) the probability of unitary EPSPs occurring increased (Fig. 5k-l), and this sometimes recruited further large unitary EPSPs (Fig. 5k). In contrast, the size of the underlying unitary EPSP was not increased by this conditioning (Fig. 5j-l), ruling out decreases in terminal presynaptic inhibition or postsynaptic inhibition contributing to the increased overall EPSP (Fig. 5i,l).
Facilitation of sensory feedback to motoneurons by GABAaxo neurons in awake mice
To determine whether GABAaxo neurons increase sensory feedback to motoneurons in awake mice we activated these neurons with light applied through a window chronically implanted over the spinal cord of GAD2//ChR2 mice (Fig. 6a), and assessed the monosynaptic reflex (MSR) recorded in tail muscles in response to nerve stimulation (counterpart of EPSPs). As expected, the MSR was facilitated by a conditioning light pulse, but only during, and not after, the expected time of phasic PAD induced on sensory axons (Fig. 6b-d,j). This light-induced facilitation occurred both at rest and when there was a background voluntary contraction, with the latter matched with and without light, again ruling out postsynaptic depolarization related differences in MSR (Fig. 6d). Light alone caused a brief pause in ongoing EMG (∼30 ms post-light; Fig. 6b), indicative of postsynaptic inhibition, which masked nodal facilitation at short intervals.
Facilitation of sensory feedback to motoneurons during PAD in humans and rats
Finally, to estimate the role of GABAaxo neurons in humans (and awake rats) we employed the sensory-evoked depolarization of proprioceptive axons by GABAaxo neurons (sensory-evoked PAD; Fig. 4), which is known to occur in humans36. For this we recorded the MSR in humans in the soleus muscle in response to tibial nerve stimulation (Fig. 7a). Increasing GABAaxo neuron activity with a brief cutaneous stimulation increased the MSR (Fig 7bi, d and Fig. 6e-g) during a period consistent with nodal facilitation by PAD (30 – 200 ms post stimulation; in humans and rats, Fig. 7bii and Fig. 6j). We again kept the conditioning stimulation small enough to not change the background EMG or single motor unit (MU) firing (Fig. 7biii, Fig 6f-g) to rule out postsynaptic actions. When we instead increased PAD by a proprioceptive conditioning in humans (via muscle TA vibration) the soleus MSR was inhibited (for up to 200 ms; Fig. 7ci-ii), as previously reported37. However, the vibration alone inhibited the ongoing MU discharge (Fig. 7ciii), implying that this MSR inhibition was caused in part by postsynaptic inhibition, rather than PAD-mediated presynaptic inhibition37.
Blocking GABAA receptor tone (in rats) decreased the MSR, at matched levels of background EMG (Fig. 6h), suggesting a spontaneous tonic PAD facilitating the MSR. In rats and humans repeated cutaneous conditioning stimulation (trains) to induce a buildup in this tonic PAD caused an associated buildup of the MSR that outlasted the conditioning and its postsynaptic actions by many seconds (after effect; Fig. 7d,e; Fig. 6i).
Importantly, the probability of a single MU contributing to the human MSR was increased by cutaneous conditioning (Fig. 7fi-ii). This occurred without an increase in the estimated EPSP amplitude or rise time (PSF; see Methods; Fig. 7Fiii) or change in the MU firing prior to the MSR testing (Fig. 7Fiv; motoneuron not depolarizing closer to threshold), consistent with an increased probability of unitary ESPSs, as in rats (Fig. 5).
Discussion
Following the pioneering studies of Eccles on inhibition of the monosynaptic connection from sensory axons to motoneurons9, the concept of presynaptic inhibition of axon terminals has stood as a cornerstone of our understanding of mammalian brain and spinal cord function10. While presynaptic inhibition has never been directly confirmed in these sensory axons, recordings from invertebrate sensory axons have firmly established the idea that terminal GABAA receptor-mediated depolarizations (PAD) can cause conductance increases (shunts) or sodium channel inactivation that inhibit transmitter release22, 24. Thus, our finding that GABAA receptors are located too far from the axon terminals to influence terminal depolarizations (relative to short λS) or cause presynaptic inhibition of transmitter release onto motoneurons (EPSPs not inhibited) had not been anticipated. Direct recordings from terminal boutons of other axon types in the mammalian brain (e.g. Calyx of Held) have shown that terminal GABAA or glycine receptors sometimes cause presynaptic inhibition, and at other times facilitate transmitter release, depending on their action on terminal calcium and potassium channels21, 22, 38. However, these terminal actions of GABA are very different from the actions of GABA on the nodes that we have uncovered, where GABAA receptors near sodium channels (nodes) and associated branch points help prevent conduction failure by bringing nodes closer to spike threshold. We found that without nodal GABA provided by GABAaxo neurons, spike transmission fails in many central branches of large myelinated sensory axons, leaving large silent branches, depending on the branching structure and prior history of activity (frequency). Thus, neuronal circuits that control GABAaxo neurons serve to aid branch conduction via nodal facilitation. This concept of nodal facilitation may generalize to other large central axons (e.g. pyramidal cells) that are innervated by GABAergic neurons, branch extensively (and so may fail), and have depolarizing actions of GABA20, 22, 39, 40, allowing selective recruitment of specific axon branches and functional pathways, especially for high frequency firing.
Sensory driven GABAaxo circuits (and PAD) are experimentally convenient, since they allow us to estimate how sensory transmission to motoneurons is modulated with GABAaxo neuron activity in humans. Specifically, the expected PAD evoked by cutaneous conditioning is associated with a potent reflex facilitation in humans and awake rodents. This suggests that a substantial ongoing spike failure (prior to facilitation) that can be alleviated by GABAaxo activity (PAD). Indeed, we found that during PAD the probability of EPSPs occurring (and MU firing) is increased without changing the EPSP amplitude (estimated by PSFs in humans). The latter rules out changes in presynaptic inhibition with PAD that grades the EPSP size, including ruling out previous arguments that MSR facilitation by cutaneous conditioning is due to a removal of presynaptic inhibition41, 42.
A pressing question that remains is how can nearly a century of research on sensory transmission and presynaptic inhibition be reconciled with GABA-mediated nodal facilitation and reflex facilitation (Table S1)? Sensory axon conduction failure has repeatedly been noted from indirect observations28–30, but GABAA receptors and PAD were previously thought to cause, rather than prevent, conduction failure30, even though computer simulations showed physiological GABA levels unlikely to block spike propagation34, as we confirmed. Furthermore, the fundamental assumption that GABAA receptors cause presynaptic inhibition that reduces transmitter release from sensory axons was from the outset circumspect, based mainly on the observation that a conditioning stimulation (on a flexor nerve) caused an inhibition of the MSR (evoked in extensor muscles) that was somewhat correlated to the time-course of PAD caused by this conditioning (in extensor afferents)9. However, in retrospect this PAD is too brief to account for the much longer (up to 1 s) inhibition caused by this conditioning9, 11, 43, and GABAB receptor antagonists block much of this MSR inhibition17, 43. This fits with GABAB receptors being at the terminals (Fig. 1) and primarily responsible for presynaptic inhibition in proprioceptive axons, as in other neurons21, 22, though further study of GABAB receptor function is now needed. This predominant GABAB action in proprioceptive axon terminals does not rule out GABAA-mediated presynaptic inhibition in other sensory axons that have dense terminal GABAA receptor expression, such as in cutaneous afferents16.
Anatomical studies have shown that GABAaxo neuron activation is inevitably accompanied by some postsynaptic inhibition, since most GABAaxo contacts on afferent terminals also contact motoneurons, in a triad8, 44. Indeed, we find that GABAaxo neuron activation produces an inhibition of motoneurons (Fig. 6b) and associated MU firing (Fig. 7c) that masks, and at times overwhelms, the facilitation of the MSR by nodal GABAA receptors (as with muscle vibration). This postsynaptic inhibition was likely previously mistaken for presynaptic inhibition, though this needs to be re-examined37. The argument that presynaptic inhibition with conditioning should be evident from reductions in the EPSP without changing its time course45 now seems untenable, especially as unitary EPSPs differ markedly in shape and conditioning increases the number of unitary EPSPs contributing to the EPSP, as different axon branches are recruited (Fig. 5k)28.
Early on Barron and Matthews23 and later others7, 16 established that sensory-evoked PAD (or light-evoked) excites axons by directly inducing spiking, including spikes in the sensory axons mediating the MSR itself, raising a further contradiction with presynaptic inhibition. While these PAD-triggered spikes only sometimes fully propagate antidromically out the DR46, they are more likely to conduct orthodromically16 where they activate the motoneurons7, 9, 47, making these axons and their motoneuron synapse refractory to subsequent testing9. This contributes to a long-lasting post-activation depression of the MSR pathway that is GABAA-mediated (sensitive to GABAA antagonists, like PAD) and is thus readily mistaken for GABAA-mediated presynaptic inhibition7, 17, 18, 37.
Functionally, nodal facilitation and regulation of branch point failure by GABAaxo-driven GABAA receptors acts like a global switching system that recruits entire silent sensory or motor circuits. This works in concert to terminal presynaptic inhibition (including GABAB receptor action) that locally fine tunes reflex gains, to optimize the stability and compliance of movement3, 4, 7. The direct activation of GABAaxo neurons (and PAD) by cortical (CST) and spinal (CPG) circuits12, 13, 48, and inhibition by the brainstem (5-HT)33, 49, suggests that nodal facilitation is under explicit central control during reaching and locomotion (Fig. 2). The widespread action of PAD (occurring simultaneously over many spinal segments)11, 16 implies that nodal facilitation acts over large regions of the spinal cord to ready sensory axons for action during cortical, spinal or sensory evoked activity, reminiscent of the Jendrassik maneuver50, ensuring that adequate sensory feedback aids postural stability and walking. More generally, our results imply that each axonal branch point has the capacity to function separately (like separate neurons), depending on their GABAergic innervation, increasing the complexity of sensory processing in the spinal cord.
Methods
Adult mice, rats and humans used
Recordings were made from large proprioceptive group Ia sensory afferents, GABAergic neurons, motoneurons and muscles in adult mice (2.5 – 6 months old, both female and male equally; strains detailed below) and rats (3 - 8 months old, female only, Sprague-Dawley). All experimental procedures were approved by the University of Alberta Animal Care and Use Committee, Health Sciences division. Recordings were also made from the soleus muscle of neurologically intact adult humans (female and male equally), aged 21 to 58, with written informed consent prior to participation. Experiments were approved by the Health Research Ethics Board of the University of Alberta (Protocols 00023530 and 00076790) and conformed to the Declaration of Helsinki. No effects of sex were noted and data from both sexes were combined for analysis.
Mice used for optogenetics and imaging
We evaluated GABAergic neurons in a strain of mice with Cre expressed under the endogenous Gad2 promotor region. Gad2 encodes the Glutamate decarboxylase 2 enzyme GAD2 (also called GAD65), which is unique to axoaxonic contacting GABAergic neurons that project to the ventral horn, whereas all GABAergic neurons express GAD16. These GAD2+ neurons were activated or inhibit optogenetically using channelrhodopsin-2 (ChR2)51, 52 or archaerhodopsin-3 (Ach3)53, 54, respectively. The following mouse strains were employed (Supplementary Table 2):
Gad2tm1(cre/ERT2)Zjh mice (abbreviated Gad2CreER mice; The Jackson Laboratory, Stock # 010702; CreERT2 fusion protein expressed under control of the endogenous Gad2 promotor)55,
B6;129S-Gt(ROSA)26Sortm32(CAG–COP4*H134R/EYFP)Hze mice (abbreviated R26LSL-ChR2-EYFP mice; The Jackson Laboratory, Stock # 012569; ChR2-EYFP fusion protein expressed under the R26::CAG promotor in cells that co-express Cre because a loxP-flanked STOP cassette, LSL, prevents transcription of the downstream ChR2-EYFP gene)56,
B6.Cg-Gt(ROSA)26Sortm14(CAG–tdTomato)Hze and B6.Cg-Gt(ROSA)26Sortm9(CAG–tdTomato)Hze mice (abbreviated R26LSL-tdTom mice; The Jackson Laboratory, Stock # 007914 and #007909; tdTomato fluorescent protein expressed under the R26::CAG promotor in cells that co-express Cre)57,
B6;129S-Gt(ROSA)26Sortm35.1(CAG–aop3/GFP)Hze mice (abbreviated R26LSL-Arch3-GFP mice; The Jackson Laboratory Stock # 012735; Arch3-GFP fusion protein expressed under the R26::CAG promotor in cells that co-express Cre)56, and
B6;129S-Slc17a7tm1.1(cre)Hze mice (abbreviated VGLUT1Cre mice; The Jackson Laboratory, Stock # 023527; Cre protein expressed under control of the endogenous Vglut1 promotor; kindly donated by Dr. Francisco J. Alvarez)58.
Heterozygous GAD2CreER mice (i.e., GAD2CreER/+ mice) were crossed with homozygous reporter strains to generate GAD2CreER/+; R26LSL-ChR2-EYFP, GAD2CreER/+; R26LSL-tdTom and GAD2CreER/+; R26LSL-Arch3-GFP mice that we abbreviate: GAD2//ChR2, GAD2//tdTom and GAD2//Arch3 mice. Offspring without the GAD2CreER mutation, but with the effectors ChR2, Arch3 or tdTom were used as controls. We also used mice bred by crossing homozygous VGLUT1Cre mice with R26lsl-tdTom reporter mice to obtain mice with VGLUT1 labelled sensory axons 59.
CreER is an inducible form of Cre that requires tamoxifen to activate 60, which we applied in adult mice to prevent developmental issues of earlier induction of Cre. Specifically, mice were injected at 4 - 6 weeks old with two doses of tamoxifen separated by two days, and studied > 1 month later, long after washout of tamoxifen. Each injection was 0.2 mg/g wt (i.p.) of tamoxifen dissolved in a corn oil delivery vehicle (Sigma C8267). These tamoxifen-treated mice were denoted GAD2//ChR2+ and GAD2//Arch3+, and non treated mice were used as controls and denoted GAD2//ChR2- and GAD2//Arch2-. For all mice, genotyping was performed according to the Jackson Laboratories protocols by PCR of ear biopsies using primers specific for the appropriate mutant and wild type alleles for each of the mouse lines (see Key Resources Table for primer details).
Ex vivo recording from axons and motoneurons in whole adult spinal cords
Mice or rats were anaesthetized with urethane (for mice 0.11 g/100 g, with a maximum dose of 0.065 g; and for rats 0.18 g/100 g, with a maximum dose of 0.45 g), a laminectomy was performed, and then the entire sacrocaudal spinal cord was rapidly removed and immersed in oxygenated modified artificial cerebrospinal fluid (mACSF), as detailed previously 61–63. This preparation is particularly useful as the small sacrocaudal spinal cord is the only portion of the adult spinal cord that survives whole ex vivo, allowing axon conduction to be assessed along large distances. Further, this segment of cord innervates the axial muscles of the tail that are readily assessable for reflex recording in awake animals, and has proven to be a useful model of motor function in normal and injured spinal cords 62, 64. Spinal roots were removed, except the sacral S3, S4 and caudal Ca1 ventral and dorsal roots on both sides of the cord. After 1.5 hours in the dissection chamber (at 20° C), the cord was transferred to a recording chamber containing normal ACSF (nACSF) maintained at 23 - 32°C, with a flow rate > 3 ml/min. A one-hour period in nACSF was given to wash out the residual anaesthetic prior to recording, at which time the nACSF was recycled in a closed system. The cord was secured onto tissue paper at the bottom of a rubber (Silguard) chamber by insect pins in connective tissue and cut root fragments. The dorsal surface of the cord was usually oriented upwards when making intracellular recording from afferents in the dorsal horn, whereas the cord was oriented with its left side upwards when making recordings from motoneurons or afferent terminals in the ventral horn. The laser beam used for optogenetics was focused vertically downward on the GAD2 neurons, as detailed below.
Optogenetic regulation of GABAaxo neurons
The GAD2//ChR2 or GAD2//Arch3 mice were used to optogenetically excite or inhibit GAD2+ neurons (with 447 nm D442001FX and 532 nM LRS-0532-GFM-00200-01 lasers from Laserglow Technologies, Toronto), respectively, using methods we previously described 65. Light was derived from the laser passed through a fibre optic cable (MFP_200/220/900-0.22_2m_FC-ZF1.25 and MFP_200/240/3000-0.22_2m_FC-FC, Doric Lenses, Quebec City) and then a half cylindrical prism the length of about two spinal segments (8 mm; 3.9 mm focal length, Thor Labs, Newton, USA,), which collimated the light into a narrow long beam (200 µm wide and 8 mm long). This narrow beam was focused longitudinally on the left side of the spinal cord roughly at the level of the dorsal horn, to target the epicentre of GABAaxo neurons, which are dorsally located (Fig. 3). ChR2 rapidly depolarizes neurons 52, and thus we used 5 – 10 ms light pulses to activate GABAaxo neurons, as confirmed by direct recordings from these neuron (see below). Light was always kept at a minimal intensity, 1.1x T, where T is the threshold to evoke a light response in sensory axons, which made local heating from light unlikely. Arch3 is a proton pump that is activated by green light, leading to a hyperpolarization and slowly increased pH (over seconds), both of which inhibit the neurons 52, 66. Thus, we used longer light pulses (∼200 ms) to inhibit GABAaxo neurons.
To directly confirm the presence of functional GAD2 expression in neurons (GABAaxo neurons) we recorded from them with similar methods and intracellular electrodes used to record from motoneurons (see below). Electrodes were advanced into these cells through the dorsal horn (with the dorsal surface oriented upwards), and their identity established by a direct response to light activation of the ChR2 construct (5 – 10 ms light pulse, 447 nm, mW; in GAD2//ChR2 mice), without a synaptic delay (<1 ms) and continued light response after blocking synaptic transmission.
Dorsal and ventral root stimulation
Dorsal and ventral roots (DR and VR) were mounted on silver-silver chloride wires above the nASCF of the recording chamber and covered with grease (a 3:1 mixture of petroleum jelly and mineral oil) for monopolar stimulation 16, 64, 67. This grease was surrounded by a more viscous synthetic high vacuum grease to prevent oil leaking into the bath flow. Bipolar stimulation was also used at times to reduce the stimulus artifact during recording from ventral roots (detailed below). Roots were stimulated with a constant current stimulator (Isoflex, Israel) with short pulses (0.1 ms). Note that proprioceptive afferents are selectively activated by low intensity DR stimulation (1.1 – 1.5 x threshold, T) and cutaneous afferents are additionally activated by higher intensity DR stimulation (2 – 3xT). DRs were dissected to be as long as possible, and the distal end of this root was stimulated, so it was ∼20 mm way from the spinal cord. In this way the DR stimulation site itself (at wire, and threshold for stimulation) could not be affected by axonal depolarizations in the spinal cord, since dorsal root potentials from spinal events (PAD) are only observed very close to the cord (within a few mm, see below), and drop exponentially in size with distance 16.
Intracellular recording from sensory axon branches in the dorsal horn
Electrode preparation and amplifier
Recording from fine afferent collaterals in the spinal cord without damaging them or disturbing their intracellular milieu required specialized ultra-sharp intracellular electrodes modified from those we developed for motoneuron recording 61. That is, glass capillary tubes (1.5 mm and 0.86 mm outer and inner diameters, respectively; with filament; 603000 A-M Systems; Sequim, USA) were pulled with a Sutter P-87 puller (Flaming-Brown; Sutter Instrument, Novato, USA) set to make bee-stinger shaped electrodes with a short relatively wide final shaft (∼1 mm) that tapered slowly from 30 to 3 µm over its length, and then abruptly tapered to a final tip over the final 20 µm length. The tip was subsequently bevelled to a 100 nm hypodermic-shaped point, as verified with electron microscope images (Harvey et al. 2006). This very small tip and wide shaft gave a combination of ease of penetrating axons in dense adult connective tissue, and good current-passing capabilities to both control the potential and fill the axons with neurobiotin. Prior to beveling, electrodes were filled through their tips with 2 M K-acetate mixed with varying proportions of 2 M KCl (to make intracellular Cl-concentrations ranging of 0, 100, 500, and 1000 mM) or 500 mM KCl in 0.1 Trizma buffer with 5 - 10% neurobiotin (Vector Labs, Birmingame, USA). They were then beveled from an initial resistance of 40 - 150 MΩ to 30 - 40 MΩ using a rotary beveller (Sutter BV-10). GABAergic chloride-mediated potentials (PAD) were the same with different concentrations of KCl, without passing large amounts of negative current, as we have previously detailed 16, indicating that the ultra-sharp tips impeded passive fluid exchange between the electrode and intracellular milieu, with in particular electrode Cl- not affecting the axon; thus, recordings were mostly made with electrodes with 1 M K-acetate and 1 M KCl, when not filling cells with neurobiotin.
Intracellular recording and current injection were performed with an Axoclamp2B amplifier (Axon Inst. and Molecular Devices, San Jose, USA). Recordings were low pass filtered at 10 kHz and sampled at 30 kHz (Clampex and Clampfit; Molecular Devices, San Jose, USA). Sometimes recordings were made in discontinuous-single-electrode voltage-clamp (gain 0.8 –2.5nA/mV; for Ca PICs) or discontinuous-current-clamp modes (switching rate 7 kHz), as indicated (the latter only when injecting current, for example during recording of input resistance or the voltage dependence of spikes).
Axon penetration
Electrodes were advanced into myelinated afferents with a stepper motor (Model 2662, Kopf, USA, 10 µm steps at maximal speed, 4 mm/s), usually at the boundary between the dorsal columns and dorsal horn gray matter. Extracellular tissue (especially myelin in the white matter) often impeded and blocked the electrode tip following a forward step, as determined by an increase in resistance to small current pulses passed from the tip of the electrode (20 ms, −0.3 nA, 1 Hz), and this was cleared with a brief high frequency current (from capacitance overcompensation buzz) and moving backwards slowly, the latter which helped prevent tissue dimpling. Prior to penetrating afferents, we recorded the extracellular (EC) afferent volley following dorsal root (DR) stimulation (0.1 ms pulses, 3xT, threshold, where T = ∼3 uA, repeated at 1 Hz), to determine the minimum latency and threshold of afferents entering the spinal cord. The group Ia afferent volley occurs first with a latency of 0.5 - 1.0 ms, depending on the root length (which were kept as long as possible, 10 - 20 mm), corresponding to a conduction velocity of about 16 - 24 m/s, as previously described for in vitro conduction at 23 C 16, 68. When a forward step penetrated an axon, small slow movements were made to stabilize the recordings. Penetrations were usually in the myelinated portion of the axon between nodes, rather than at nodes, because the chance of penetrating a node is low since they only make up a small fraction of the total axon length (Fig. 1). The spikes from the two nodes adjacent to the electrode were readily detected separately when testing for the spike threshold with current injection pulses (20 ms; rheobase test), because just at threshold the current sometimes evoked a spike from just one node and not the other, which usually halved the total spike height, consistent with the penetration being about halfway between the two nodes.
Proprioceptive afferent identification
Upon penetration, afferents were identified with direct orthodromic spikes evoked from DR stimulation. We focused on the lowest threshold (T) proprioceptive group Ia afferents, identified by their direct response to DR stimulation, very low threshold (< 1.5 x T), short latency (group Ia latency, coincident with onset of afferent volley), and antidromic response to ventral horn afferent terminal microstimulation (∼ 10 µA stimulation via tungsten microelectrode to activate Ia afferent terminals; tested in some afferents, detailed below)16. Clean axon penetrations without injury occurred abruptly with a sharp pop detected on speakers attached to the recorded signal, the membrane potential settling rapidly to near – 70 mV, and > 70 mV spikes readily evoked by DR stimulation or brief current injection pulses (1 – 3 nA, 20 ms, 1 Hz). Sensory axons also had a characteristic >100 ms long depolarization following stimulation of a dorsal root (primary afferent depolarization, PAD, at 4 - 5 ms latency, detailed below) and short spike afterhyperpolarization (AHP ∼ 10 ms), which further distinguished them from other axons or neurons. Injured axons had higher resting potentials (> - 60 mV), poor spikes (< 60 mV) and low resistance (to current pulse; Rm < 10 MΩ) and were discarded.
Quantification of spike conduction failure in the dorsal horn: failure potentials (FPs)
Sometimes healthy intracellular penetrations were made into a sensory axon branch (e.g. < −60 mV rest, large PAD), but dorsal root stimulation did not evoke a full spike, even though a full > 60 mV spike could be readily evoked by intracellular current injection. Instead, DR stimulation evoked a partial spike at the latency and threshold of group Ia afferents, indicating that this was a branch of a Ia afferent that failed to fully conduct spikes to the electrode, with only the passively attenuated spike from the last node to spike prior to conduction failure recorded at the electrode (failure potential, FP; also referred to as electronic residue by Luscher69). The size of the FP reflected how far away the spike failure occurred, with spatial attenuation corresponding to a space constant of about 90 µm (see Results), and so FPs became exponentially smaller with distance from failure and undetectable when many mm away (nodes separated by about 50 µm). Occasionally axons were penetrated with undetectable DR evoked spikes or FPs, but otherwise they had characteristics of a Ia afferent (PAD, Rm similar). These were likely afferents with FPs too distal to detect, but were usually excluded from the main analysis to avoid ambiguity, though this underestimates the incidence of failure. However, some of these axons exhibited short latency, low threshold DR spikes when depolarized by a prior DR stimulation (PAD) of an adjacent DR, in which case they were unequivocally Ia afferents and included in the analysis (Fig. 4f).
Both during extracellular and intracellular recording the group Ia afferent volley (small negative field) was observed as the first event after DR stimulation (the latter subthreshold to a spike), though this was usually small in relation to intracellular events and ignored. However, this was sometimes removed from the intracellular record by subtracting the extracellular potential recorded just outside the same axon to determine the actual transmembrane potential 16. This was necessary to see the very smallest FPs following DR stimulation in some afferents, as the negative volley from other nearby afferents obscured the FPs.
After quantifying the axons spikes and conduction failures (FPs) under resting conditions, we then examined the changes in spike conduction with changes in membrane potential induced by either directly injecting current into axons or inducing GABA-mediated changes in membrane potential by pharmacological methods, optogenetic methods (activating ChR2 on GABAaxo neurons to induce PAD) or more naturally evoking PAD with a DR stimulation.
Neurobiotin filling of axons
Some of the proprioceptive afferents that we recorded intracellularly were subsequently filled with neurobiotin by passing a very large positive 2 - 4 nA current with 90% duty cycle (900 ms on, 100 ms off) for 10 - 20 min. The identity of group Ia proprioceptive afferents were then confirmed anatomically by their unique extensive innervation of motoneurons 16. Prior to penetrating and filling axons with neurobiotin filled electrodes, a small negative holding current was maintained on the electrodes to avoid spilling neurobiotin outside axons.
Quantification of spike conduction failure in the ventral horn
Wall’s method
To measure whether spikes fail during propagation to their fine terminals in the ventral horn we examined whether failed axon segments were relatively less refractory to activation after spike conduction failure, using a double pulse method adapted from Wall 30, 70. The essence of the method is that after DR activation all nodes that generate spikes become relatively refractory for a few ms, whereas nodes that fail to spike are not refractory to activation. Thus, a microelectrode placed near these failing nodes more readily activates them if they fail rather than generate spikes with DR stimulation and orthodromic conduction. For this we placed a tungston microelectrode (12 MΩ, #575400, A-M Systems, Sequim, USA) in the ventral horn near the axons terminals on motoneurons, to activate the branches/nodes of the axon projecting to the motoneuron that may have failed (VH stimulation).
Spikes from VH or DR stimulation were recorded intracellularly in a proprioceptive axon penetrated in the dorsal columns directly above the VH stimulation site or in an adjacent segment, with two combinations of double axon stimulations. First, we applied two rapidly repeated VH stimuli (VH doublet; two 0.1 ms pulses) at a ∼4 ms interval to make the axon relatively refractory to stimulation and determine both the threshold current to activate the first spike (TVH1, with VH1 stimulation) and the higher threshold current to overcome this the inactivation and generate a second spike (TVH2, with VH2 stimulation). Second, we repeated this double spike activation, but with the first activation from a supra-threshold DR stimulation (at 1.5x DR threshold) and the second from a VH stimulation at the TVH2 intensity from B (DR-VH pair). In this case the VH stimulation readily activates the axon spike if the orthodromic DR evoked spike does not propagate to the ventral horn, leaving the silent portion of the axon non refractory. Accordingly, we also determined the threshold current to activate the VH after the DH in this arrangement (termed TDR,VH), which was lower than TVH2. For comparison to the spike inactivation with VH doublets, we adjusted the DR-VH pair timing slightly so that the pairs of spikes (or expected spikes, at vertical lines) are separated by the same interval (∼ 4 ms) when they reach the recording site, to compensate for DR conduction delays. The putative spike failure with DR stimulation happens at a node somewhere between the recording site and the VH, because we only studied axons that securely conducted single DR pulses to the recording site, and thus failure was not directly visible.
We quantified the spike failure based on the following considerations: If the DR-evoked spike entirely fails to propagate to the VH, then the threshold for subsequently activating the ventral horn (TDR,VH) should be the same as the threshold without any prior activation (TVH1 = TDR,VH), whereas if it does not fail, then the threshold for activating the ventral horn should be the same as with a VH doublet (TVH2 = TDR,VH). In between these two extreme scenarios, the DR evoked spike may only partially fail to propagate spikes to the ventral horn (by only some of its branches failing or conducting only partially to the VH); in this case TDR,VH should be between TVH1 and TVH2, with the difference TVH2 -TVH1 representing the range of possible thresholds between full failure and full conduction. Thus, overall the failure was quantified as: Conduction failure = (TVH2 -TDR,VH) / (TVH2 -TVH1) x 100%, which is 100% at full failure and 0% with no failure. This estimate is predicated on the assumption that the failed spikes are only relatively refractory to conduction and increased stimulation can overcome this failure, which is reasonable for the interspike intervals we used, and means that the computed % failure reflects the number of nodes that failed to spike, with more dorsal branch point failures giving more failed nodes. On the other hand, we used interspike intervals that were short enough for the DR stimulation not to evoke PAD that affected the subsequent spike threshold (∼ 4 ms), in contrast to the longer intervals where PAD can help DR doublet firing (DR-DR in Extended Data Fig. 6, ∼ 5 - 10 ms).
Extracellular recording from sensory axon terminals
To directly record spike conduction in proprioceptive afferent terminal branches in the VH we used our intracellular glass pipette electrode (∼30 MΩ) positioned just outside these axons (extracellular, EC), to avoid penetration injury in these fine axon branches. The DR was stimulated near threshold (1.1xT) to evoke the EC response in a few single axons near the electrode, and many trials were averaged to remove noise from these small signals (20 – 50 trials at 3 s intervals). The EC field was multiphasic as previously described for other axons 71–73, with a small initial positive field resulting from passively conducted axial current from sodium spikes at distant nodes (closer to the DR; outward current at electrode), some of which fail to propagate spikes to the VH recording site, making this field a measure of conduction failure 71, 73. Following this, a larger negative field arises, resulting from spikes arising at nodes near the electrode (inward current), making this negative field a measure of secure conduction. A relatively large stimulus artifact is present prior to these fields, due to the small size of the EC fields themselves, and we truncated this.
We conducted three control experiments to confirm the relation of these EC fields to spike conduction. First, in the dorsal horn where we can readily intracellularly record from large proprioceptive axon branches, we compared intracellular (IC) recordings from axons to EC recordings just outside the same axon, to confirm that the DR evoked spike (IC) arrives at about the time of the negative EC field. Second, we locally applied TTX to the DR near the recording site (10 µl bolus of 100 µM TTX over DR) which eliminated the negative field and left only the initial positive field, confirming that the positive field is from distal nodes upstream of the TTX block, and generated by passive axial current conduction. This is important, since some investigators have argued on theoretical grounds that the positive field can instead result from the closed end electrical properties of axons at their terminals 74, rather than spike failure, though others have refuted this73. Finally, we improved nodal spike conduction by reducing the divalent cations Mg++ and Ca++ in the bath medium, since divalent cations normally cause a gating or guarding action on the sodium channel, the latter by one charge binding to the membrane and the other raising the local extracellular positive charge, and overall raising the local voltage drop across the channel and its spike threshold75. This decreased the failure-related initial positive field and increased the main EC negative field, indicating improved conduction, and again confirming the use of these fields as measures of conduction, similar to previous conclusions for the motor endplate 71 and mathematical consideration of axon cable properties 76.
To quantify the EC fields we estimated the overall conduction to the recording site as:
Conduction Index = nf / (nf + pf) x 100%, where pf and nf are the positive and negative EC field amplitudes. This conduction index approaches 100% for full conduction (pf ∼=0) and 0% for no conduction (nf = 0). The absolute EC field potential amplitudes are highly variable between different recordings sites, and thus are difficult to quantify across animals and sites, whereas this ratio of field amplitudes (nf / (nf + pf)) eliminates the variability, and can effectively be viewed as a normalization of the negative field (nf) by the total field peak-to-peak size (nf + pf).
Intracellular recording from motoneurons
The same intracellular glass electrode, stepper motor and amplifier used for recording sensory axons were used for intracellular recording from motoneurons, except that the electrodes were bevelled to a lower resistance (30 MΩ). The electrode was advanced into motoneurons with fast 2 µm steps and brief high frequency currents (capacitance overcompensation) guided by audio feedback from a speaker. After penetration, motoneuron identification was made with antidromic ventral root stimulation, and noting ventral horn location, input resistance and time constant (> 6 ms for motoneurons) 62. The monosynaptic excitatory postsynaptic potentials (EPSPs) and associated currents (EPSCs) were measured in motoneurons following stimulation of dorsal roots (at 1.1-1.5 xT, 0.1 ms, 3 – 10 s trial intervals). These were identified as monosynaptic by their rapid onset (first component), lack of variability in latency (< 1 ms jitter), persistence at high rates (10 Hz) and appearance in isolation at the threshold for DR stimulation (< 1.1xT), unlike polysynaptic EPSPs which varying in latency, disappear at high rates, and mostly need stronger DR stimulation to activate.
Dorsal and ventral root grease gap recording
In addition to recording directly from single proprioceptive axons and motoneurons, we employed a grease gap method to record the composite intracellular response of many sensory axons or motoneurons by recording from dorsal and ventral roots, respectively, as previously detailed for similar sucrose and grease gap methods, In this case, a high impedance seal on the axon reduces extracellular currents, allowing the recording to reflect intracellular potentials16, 76–78. We mounted the freshly cut roots onto silver-silver chloride wires just above the bath, and covered them in grease over about a 2 mm length, as detailed above for monopolar recordings. Return and ground wires were in the bath and likewise made of silver-silver chloride. Specifically for sensory axons, we recorded from the central ends of dorsal roots cut within about 2 - 4 mm of their entry into the spinal cord, to give the compound potential from all afferents in the root (dorsal roots potential, DRP), which has previously been shown to correspond to PAD, though it is attenuated compared to the intracellular recordings of PAD16. The signal attenuation has two reasons. First the voltage PAD is attenuated along the length of nerve in the bath, as detailed in the next paragraph. Second, the grease does not completely remove the extracellular fluid around the nerve, even though we deliberately allowed the nerve to dry for a few seconds before greasing, and this causes a conductance that shunts or short circuits the recorded signal, reducing it by about half71, 78. For optogenetic experiments we additionally added silicon carbide powder (9 % wt, Tech-Met, Markham) to the grease to make it opaque to light and minimize light induced artifactual current in the silver-silver chloride recording wire during optogenetic activation of ChR2 (detailed below). Likewise, we covered our bath ground and recording return wires with a plastic shield to prevent stray light artifacts. The dorsal root recordings were amplified (2,000 times), high-pass filtered at 0.1 Hz to remove drift, low-pass filtered at 10 kHz, and sampled at 30 kHz (Axoscope 8; Axon Instruments/Molecular Devices, Burlingame, CA).
These grease gap recordings of PAD on sensory afferents reflect only the response of largest diameter axons in the dorsal root, mainly group I proprioceptive afferents, because of the following considerations. First, the largest axons in peripheral nerves have a nodal spacing of about 1 mm79, 80, and length constants λS are estimated to be similar, at about 1 – 2 times the nodal spacing81, Further, in our recordings we were only able to get the grease to within about 2 mm of the spinal cord. Thus, the centrally generated signal (PAD) is attenuated exponentially with distance x along the axon length in the bath (x = 2 mm). This is proportional to exp(– x / λS) (see 76), which is 1 / e2 = 0.11 for x = 2 λS, as is approximately the case here. This makes a central PAD of about 4 mV appear as a ∼0.4 mV potential on the root recording (DRP, 10 times smaller), as we previously reported16. Furthermore, the nodal spacing and λS decrease linearly with smaller axon diameters76, 79, making the voltages recorded on the smaller afferents contribute to much less of the compound root potential (halving the diameter attenuates PAD instead by 1/e4 or 0.012, which is 99% attenuation). Finally, unmyelinated sensory axons attenuate voltages over a much shorter distance than myelinated axons, since that membrane resistance (Rm) drops markedly without myelin and λS is proportional to (where Ri is axial resistance; Stein 1980). Thus, any centrally generated change in potential in these small axons is unlikely to contribute to the recorded signal 2 mm away.
The composite EPSPs in many motoneurons were likewise recorded from the central cut end of ventral roots mounted in grease (grease gap), which has also previously been shown to yield reliable estimates of the EPSPs (though again attenuated by the distance from the motoneurons)82. The monosynaptic EPSPs were again identified as monosynaptic by their rapid onset (first component, ∼1 ms after afferent volley arrives in the ventral horn; see below), lack of variability in latency (< 1 ms jitter), persistence at high rates (10 Hz) and appearance in isolation at the threshold for DR stimulation (< 1.1xT), unlike polysynaptic reflexes which varying in latency, disappear at high rates, and mostly need stronger DR stimulation to activate.
Analysis of synaptic responses in sensory axons (PAD) and motoneurons (EPSPs)
When we recorded from sensory axons of an associated dorsal root (directly or via the dorsal roots) stimulation of an adjacent dorsal root (not containing the recorded axon; 0.1 ms, 1 – 3xT) evoked a characteristic large and long depolarization of the afferents, previously demonstrated to be mediated by GABAergic input onto the sensory axons 16. This depolarization is termed primary afferent depolarization (PAD). PAD occurs at a minimal latency of 4 – 5 ms following the afferent volley, consistent with its trisynaptic origin 14, 16, making it readily distinguishable from earlier events on the axon. PAD has a fast synaptic component evoked by a single DR stimulation (rising within 30 ms and decaying exponentially over < 100 ms; termed phasic PAD) and a slower longer lasting extrasynaptic component (starting at about 30 ms and lasting many seconds) that is enhanced by repeated DR stimulation (tonic PAD, especially with cutaneous stimulation) 16. We used this sensory activation of PAD or direct optogenetic activation of PAD to examine the action of GABA on sensory axon spike transmission to motoneurons, usually evoking phasic PAD about 10 – 60 ms prior to spikes or associated EPSPs on motoneurons (during phasic PAD), though we also examined longer lasting effects of tonic PAD evoked by repeated DR stimulation. Sometimes PAD is so large that it directly evokes spikes on the afferents, and these travel out the dorsal root, and thus they have been termed dorsal root reflexes (DRRs) 16, 23. We usually minimized these DRRs by keeping the DR stimulus that evokes PAD low (1.1 - 3.0 xT), though there were inevitably some DRRs, as they even occur in vivo in cats and humans 9, 36, 46.
When we recorded from motoneurons (directly or via ventral roots) stimulation of proprioceptive afferents in a dorsal root (0.1 ms, 1.1-1.5xT) evoked a monosynaptic EPSP, and associated monosynaptic reflex (MSR, spikes from EPSP). This EPSP is depressed by fast repetition (rate depended depression, RDD) 83, and thus to study the EPSP we evoked it at long intervals (10 s, 0.1 Hz rate) where RDD was less. However, even with this slow repetition rate (0.1 Hz), at the start of testing the first EPSP was often not similar to the steady state EPSP after repeated testing. Thus, to avoid RDD we usually ran the 0.1 Hz EPSP testing continuously throughout the experiment, at least until a steady state response was reached (after 10 minutes). We then examined the action of activating (or inhibiting) GABAaxo neurons on this steady state EPSP, by introducing light or sensory conditioning that activated these neurons at varying intervals (inter-stimulus intervals, ISIs) prior to each EPSP stimulation (control, GAD2//ChR2 mice and GAD2//Arch3 mice). We averaged the EPSP from ∼10 trials (over 100 s) just before conditioning and then 10 trials during conditioning, and then computed the change in the peak size of the monosynaptic EPSP with conditioning from these averages. After conditioning was completed EPSP testing continued and any residual changes in the EPSP were computed from the 10 trials following conditioning (after-effect). Finally, EPSP testing continued over many minutes after which the original steady state EPSP was established. The background motoneuron potential, membrane resistance (Rm) and time constant just prior to the EPSP was also assessed before and after conditioning to examine whether there were any postsynaptic changes that might contribute to changes in the EPSP with conditioning. Along with the VR recordings, we simultaneously recorded PAD from DRs by similar averaging methods (10 trials of conditioning), to establish the relation of changes in EPSPs with associated sensory axon depolarization PAD.
Drugs and solutions
Two kinds of artificial cerebrospinal fluid (ACSF) were used in these experiments: a modified ACSF (mACSF) in the dissection chamber prior to recording and a normal ACSF (nACSF) in the recording chamber. The mACSF was composed of (in mM) 118 NaCl, 24 NaHCO3, 1.5 CaCl2, 3 KCl, 5 MgCl2, 1.4 NaH2PO4, 1.3 MgSO4, 25 D-glucose, and 1 kynurenic acid. Normal ACSF was composed of (in mM) 122 NaCl, 24 NaHCO3, 2.5 CaCl2, 3 KCl, 1 MgCl2, and 12 D-glucose. Both types of ACSF were saturated with 95% O2-5% CO2 and maintained at pH 7.4. The drugs sometimes added to the ACSF were APV (NMDA receptor antagonist), CNQX (AMPA antagonist), gabazine (GABAA antagonist), bicuculline (GABAA, antagonist), L655708 (α5 GABAA, antagonist), CGP55845 (GABAB antagonist; all from Tocris, USA), 5-HT, kynurenic acid (all from Sigma-Aldrich, USA), and TTX (TTX-citrate; Toronto Research Chemicals, Toronto). Drugs were first dissolved as a 10 - 50 mM stock in water or DMSO before final dilution in ACSF. DMSO was necessary for dissolving gabazine, L655708, bicuculline and CGP55845, but was kept at a minimum (final DMSO concentration in ACSF < 0.04%), which by itself had no effect on reflexes or sensory axons in vehicle controls (not shown). L655708 was particularly difficult to dissolve and precipitated easily, especially after it had been exposed a few times to air; so immediately after purchase we dissolved the entire bottle and froze it at −40°C in single use 5 - 20 µl aliquots, and upon use it was first diluted in 100 µl distilled water before dispersing it into ACSF.
Recording monosynaptic reflexes in awake mice and rats, and PAD activation
Window implant over spinal cord
In GAD2//ChR2+ mice and control GAD2//ChR-mice a glass window was implanted over the exposed spinal cord to gain optical access to the sacrocaudal spinal cord, as described previously 65. Briefly, mice were given Meloxicam (1 mg/kg, s.c.) and then anesthetized using ketamine hydrochloride (100 mg/kg, i.p.) and xylazine (10 mg/kg, i.p.). Using aseptic technique, a dorsal midline incision was made over the L2 to L5 vertebrae. Approximately 0.1 ml of Xylocaine (1%) was applied to the surgical area and then rinsed. The animals were suspended from a spinal-fork stereotaxic apparatus (Harvard Apparatus) and the muscles between the spinous and transverse processes were resected to expose the L2 to L5 vertebrae. The tips of modified staples were inserted along the lateral edge of the pedicles and below the lateral processes of L2 and L5, and glued in place using cyanoacrylate. A layer of cyanoacrylate was applied to all of the exposed tissue surrounding the exposed vertebrae followed by a layer of dental cement to cover the cyanoacrylate and to form a rigid ring around the exposed vertebrae. A modified paperclip was implanted in the layer of dental cement to serve as a holding point for surgery. A laminectomy was performed at L3 and L4 to expose the spinal cord caudal to the transection site. Approximately 0.1 ml of Xylocaine (1%) was applied directly to the spinal cord for 2 – 3 s, and then rinsed. A line of Kwik-Sil (World-Precision Instruments) was applied to the dura mater surface along the midline of the spinal cord and a glass window was immediately placed over the exposed spinal cord. The window was glued in place along the outer edges using cyanoacrylate followed by a ring of dental cement. Small nuts were mounted onto this ring to later bolt on a backpack to apply the laser light (on the day of experimentation). Saline (1 ml, s.c.) and buprenorphine (0.03 mg/kg, s.c.) was administered post-operatively, and analgesia was maintained with buprenorphine (0.03 mg/kg, s.c.) every 12 hours for two days. Experimentation started 1 week after the window implant when the mouse was fully recovered.
Percutaneous EMG wire implant and fibre optic cable attachment
On the day of experimentation, the mouse was briefly sedated with isoflurane (1.5 %) and fine stainless steel wires (AS 613, Cooner Wire, Chatsworth, USA) were percutaneously implanted in the tail for recording EMG and stimulating the caudal tail trunk nerve, as we previously detailed (wires de-insulated by 2 mm at their tip and inserted in the core of 27 gauge needle that was removed after insertion)62. A pair of wires separated by 8 mm were inserted at base of the tail for recording EMG the tail muscles, and second pair of wires was inserted nearby for bipolar activation of the caudal trunk nerve to evoke reflexes. A fifth ground wire was implanted between the EMG and stimulation wires. Following this a backpack was bolted into the nuts imbedded in the dental cement ring around the window. This backpack held and aligned a light fibre optic cable that was focused on the centre of the S3 – S4 sacral spinal cord. The Cooner wires were secured to the skin with drops of cyanoacrylate and taped onto the backpack so that the mouse could not chew them. The isoflurane was removed, and the mouse quickly recovered from the anesthesia and was allowed to roam freely around an empty cage during recording, or was sometimes lightly restrained by hand or by a sling. The fibre optic cable was attached to a laser (447 nM, same above) and the Cooner wires attached to the same models of amplifiers and stimulators used for ex vivo monosynaptic testing detailed above.
MSR testing
The monosynaptic reflex (MSR) was recorded in the tail EMG at ∼6 ms latency after stimulating the caudal tail trunk nerve at a low intensity that just activated proprioceptive afferents (0.2 ms current pulses, 1.1 xT), usually near the threshold to activate motor axons and an associated M-wave (that arrived earlier). We studied the tail MSR reflex because our ex vivo recordings were made in the corresponding sacral spinal cord of adult mice and rats, which is the only portion of the spinal cord that survives whole ex vivo, due to its small diameter 64. This reflex was verified to be of monosynaptic latency because it was the first reflex to arrive, had little onset jitter, and had the same latency as the F wave (not shown; the F wave is evoked by a strong stimulation of all motor axons, at 5xT, which causes a direct motoneuron response on the same axons, while the monosynaptic EPSP is blocked by collision at this intensity) 84. The MSR also underwent rate dependent depression (RDD) with fast repeated stimulation and so was synaptic and not a direct muscle response (M-wave), which occurred earlier at sufficient intensity to recruit the motor axons (not shown).
Conditioning of the MSR by optogenetic activation of GABAaxo neurons
As with in vitro EPSP testing, the MSR was tested repeatedly at long 5 – 10 s intervals until a steady state MSR was achieved. Then testing continued but with a conditioning light pulse applied just prior to the MSR stimulation (40 – 120 ms), to examine the effect of PAD evoked during this time frame on sensory transmission to motoneurons. Background EMG just prior to MSR testing was assessed to estimate the postsynaptic activity on the motoneurons. The changes in MSR and background EMG with light were quantified by comparing the average response before and during the light application, computed from the mean rectified EMG at 6 – 11 ms after the nerve stimulation (MSR) and over 20 ms prior to the nerve stimulation (background just prior to the MSR, Bkg). Because awake mice spontaneously varied their EMG, we plotted the relation between the MSR and the background EMG, with as expected a positive linear relation between these two variables 85, computed by fitting a regression line. In trials with conditioning light applied the same plot of EMG vs background EMG was made and a second regression line computed. The change in the MSR with conditioning at a fixed matched background EMG level was then computed for each mouse by measuring the difference between the regression line responses at a fixed background EMG. This ruled out changes in MSRs being due to postsynaptic changes. Two background levels were assessed: rest (0%) and 30% of maximum EMG, expressed as a percentage of the control pre-conditioning MSR. The change in background EMG with light was computed by comparing the EMG just prior to the light application (over 20 ms prior) to the EMG just prior to the MSR (over 20 ms prior, Bkg), and expressed as a percentage of the maximum EMG.
Cutaneous conditioning of the MSR in rats
A similar examination of how PAD affected the MSR was performed in rats with percutaneous tail EMG recording. However, in this case PAD was evoked by a cutaneous conditioning stimulation of the tip of the tail (0.2 ms pulses, 3xT, 40 – 120 ms prior to MSR testing) using an additional pair of fine Cooner wires implanted at the tip of the tail (separated by 8 mm). In rats the MSR latency is later than in mice due to the larger peripheral conduction time, ∼12 ms (as again confirmed by a similar latency to the F wave). This MSR was thus quantified by averaging rectified EMG over a 12 – 20 ms window. Also, to confirm the GABAA receptor involvement in regulating the MSR, the antagonist L655708 was injected systemically (1 mg/kg i.p., dissolved in 50 µl DMSO and diluted in 900 µl saline). Again, the MSR was tested at matched background EMG levels before and after conditioning (or L655708 application) to rule out changes in postsynaptic inhibition.
Conditioning of the MSRs in humans
H-reflex as an estimate of the MSR
Participants were seated in a reclined, supine position on a padded table. The right leg was bent slightly to access the popliteal fossa and padded supports were added to facilitate complete relaxation of all leg muscles. A pair of Ag-AgCl electrodes (Kendall; Chicopee, MA, USA, 3.2 cm by 2.2 cm) was used to record surface EMG from the soleus muscle. The EMG signals were amplified by 1000 and band-pass filtered from 10 to 1000 Hz (Octopus, Bortec Technologies; Calgary, AB, Canada) and then digitized at a rate of 5000 Hz using Axoscope 10 hardware and software (Digidata 1400 Series, Axon Instruments, Union City, CA) 62. The tibial nerve was stimulated with an Ag-AgCl electrode (Kendall; Chicopee, MA, USA, 2.2 cm by 2.2 cm) in the popliteal fossa using a constant current stimulator (1 ms rectangular pulse, Digitimer DS7A, Hertfordshire, UK) to evoke an H-reflex in the soleus muscle, an estimate of the MSR 37. Stimulation intensity was set to evoke a test (unconditioned) MSR below half maximum. MSRs recorded at rest were evoked every 5 seconds to minimize RDD 86 and at least 20 test MSRs were evoked before conditioning to establish a steady baseline because the tibial nerve stimulation itself can presumably also activate spinal GABAergic networks, as in rats. All MSR were recorded at rest, except when the motor unit firing probabilities were measured (see below).
Conditioning of the MSR
To condition the soleus MSR by cutaneous stimulation, the cutaneous medial branch of the deep peroneal (cDP) nerve was stimulated on the dorsal surface of the ankle using a bipolar arrangement (Ag-AgCl electrodes, Kendall; Chicopee, MA, USA, 2.2 cm by 2.2 cm), set at 1.0xT, where T is the threshold for cutaneous sensation. A brief burst (3 pulses, 200 Hz for 10 ms) of cDP stimuli was applied before evoking a MSR at various inter-stimulus intervals (ISIs; interval between tibial and cDP nerve stimuli) within the window expected for phasic PAD evoked by cutaneous stimuli, presented in random order at 0, 30, 60, 80, 100, 150 and 200 ms ISIs. Seven conditioned MSR at each ISI were measured consecutively and the average of these MSR (peak-to-peak) was used as an estimate of the conditioned MSR. This was compared to the average MSR without conditioning, computed from the 7 trials just prior to conditioning.
The cDP nerve was also stimulated with a 500 ms long train at 200 Hz to condition the MSR, and examine the effect of tonic PAD evoked by such long trains, as in rats. Following the application of at least 20 test MSRs (every 5 s), a single cDP train was applied 700 ms before the next MSR and following this the MSR continued to be evoked for another 90 to 120 s (time frame of tonic PAD). We also conditioned the soleus MSR with tibialis anterior (TA; antagonist muscle, flexor) tendon vibration (brief burst of 3 cycles of vibration at 200Hz) to preferentially activate Ia afferents, as has been done previously 37.
Motor unit recording to examine postsynaptic actions of conditioning
Surface electrodes were used to record single motor units in the soleus muscle during low level contractions by placing electrodes on or near the tendon or laterally on the border of the muscle as detailed previously 87. Alternatively, single motor unit activity from the soleus muscle was also recorded using a high density surface EMG electrode (OT Bioelettronica, Torino, Italy, Semi-disposable adhesive matrix, 64 electrodes, 5×13, 8 mm inter-electrode distance) with 3 ground straps wrapped around the ankle, above and below the knee. Signals were amplified (150 times), filtered (10 to 900 Hz) and digitized (16 bit at 5120 Hz) using the Quattrocento Bioelectrical signal amplifier and OTBioLab+ v.1.2.3.0 software (OT Bioelettronica, Torino, Italy). The EMG signal was decomposed into single motor units using custom MatLab software as per 88. Intramuscular EMG was used to record MUs in one participant as detailed previously 89 to verify single motor unit identification from surface EMG.
To determine if there were any postsynaptic effects from the conditioning stimulation on the motoneurons activated during the MSR, we examined whether the cDP nerve stimulation produced any changes in the tonic firing rate of single motor units, which gives a more accurate estimate of membrane potential changes in motoneurons compared to compound EMG. Single motor units were activated in the soleus muscle by the participant holding a small voluntary contraction of around 5% of maximum. Both auditory and visual feedback were used to keep the firing rates of the units steady while the conditioning cutaneous was applied every 3 to 5 seconds. The instantaneous firing frequency profiles from many stimulation trials were superimposed and time-locked to the onset of the conditioning stimulation to produce a peri-stimulus frequencygram (PSF, dots in Fig. 7biii), as previously detailed 89, 90. A mean firing profile resulting from the conditioning stimulation (PSF) was produced by averaging the frequency values in 20 ms bins across time post conditioning (thick lines in Fig. 7biii and 7ciii). To quantify if the conditioning stimulation changed the mean firing rate of the tonically firing motor units, the % change in the mean PSF rate was computed at the time when the H reflex was tested (vertical line in Fig. 7bii-iii).
Unitary EPSP estimates from PSF
To more directly examine if the facilitation in MSR resulted from changes in transmission in Ia afferents after cutaneous afferent conditioning, we measured changes in the firing probability of single motor units (MUs) during the brief MSR time-course (typically 30 to 45 ms post tibial nerve stimulation) with and without cDP nerve conditioning. Soleus MSRs were as usual evoked by stimulating the tibial nerve, but while the participant held a small voluntary plantarflexion to activate tonic firing of a few single motor units. The size of the MSR was set to just above reflex threshold (and when the M-wave was < 5% of maximum) so that single motor units at the time of the MSR could be distinguished from the compound potential from many units that make up the MSR 91. For a given trial run, test MSRs were evoked every 3-5 s for the first 100 s and then MSR testing continued for a further 100s, but with a cDP-conditioning train (50 ms, 200 Hz) applied 500 ms prior to each MSR testing stimulation. These repeated high frequency trains evoke a tonic PAD in rats that facilitates sensory conduction. A 500 ms ISI was used to ensure the firing rate of the motor unit returned to baseline before the MSR was evoked, and this is also outside of the range of phasic PAD. Approximately 40-50 usable test and conditioned firing rate profiles were produced for a single session where the motor units had a steady discharge rate before the cDP nerve stimulation. Sessions were repeated 3-6 times to obtain a sufficient number of frequency points to construct the PSF (∼ 200 trials). To estimate the EPSP profile, motor unit (MU) firing was again used to construct a PSF, as detailed above, but this time locked to the tibial nerve stimulation used to evoke the MSR, so that we could estimate the motoneuron behaviour during the MSR (EPSP). When more than one MU was visible in the recordings firing from these units (usually 2 – 3) were combined into a single PSF. Overall this gave about of 100 – 600 MU MSR test sweeps to generate each PSF. Firing frequency values were averaged in consecutive 20 ms bins to produce a mean PSF profile over time after tibial nerve stimulation, for both unconditioned and conditioned MSR reflex trials. The mean background firing rate within the 100 ms window immediately preceding the tibial stimulation was compared between the test and conditioned MSR trials to determine if the conditioning cDP nerve stimulation produced a change in firing rate, and thus post-synaptic effect, just before the conditioned MSR was evoked. For each PSF generated with or without conditioning, the probability that a motor unit discharged during the MSR window (30 to 45 ms after the TN stimulation) was measured as the number of discharges during the time of the MSR window divided by the total number of tibial nerve test stimuli. As an estimate of EPSP size, the mean firing rate during the MSR window was also measured (this was computed with smaller PSF bins of 0.5 ms during the MSR).
Temperature, latency and PAD considerations
Large proprioceptive group Ia sensory afferents conduct in the peripheral tail nerve with a velocity of about 33 m/s (33 mm/ms) in mice 92. Motor axons are similar, though slightly slower (30 m/s) 93. Thus, in the awake mouse stimulation of Ia afferents in the mouse tail evokes spikes that take ∼ 2 ms to conduct to the motoneurons in the spinal cord ∼70 mm away. Following ∼1 ms synaptic and spike initiation delay in motoneurons, spikes in the motor axons take a further ∼2 ms to reach the muscles, after which the EMG is generated with a further 1 ms synaptic and spike initiation delay at the motor endplate to produce EMG. All told this gives a monosynaptic reflex latency of ∼6 ms. The motor unit potentials within the EMG signal have a duration of about 3 – 5 ms, and thus we averaged rectified EMG over 6 – 11 ms to quantify the MSR. We have shown that similar considerations hold for the rat where tail nerve conduction velocities are similar, except the distance from the tail stimulation to the spinal cord is larger (150 mm), yielding a peripheral nerve conduction delay of ∼10 ms and total MSR delay of ∼12 ms 94. In humans the MSR latency is dominated by the nerve conduction latency (50 – 60 m/s) over a large distance (∼800 mm), yielding MSR latencies of ∼30 ms.
In our ex vivo whole adult spinal cord preparation the bath temperature was varied between 23 and 32°C. All data displayed is from 23 – 24°C, though we confirmed the main results (facilitation of sensory axon transmission to motoneuron by PAD) at 32°C. The Q10 for peripheral nerve conduction (ratio of conduction velocities with a 10 °C temperature rise) is about 1.3 95, yielding a conduction in dorsal roots of about 20 m/s at 23 – 24 °C, as we directly confirmed (not shown). Thus, when the DR is stimulated 20 mm from the cord the latency of spike arrival at the cord should be about 1 ms, which is consistent with the time of arrival of afferent volleys that were seen in the intracellular and extracellular recordings from sensory axons (e.g. Figs. 2b and 4e).
When we found that PAD evoked in sensory axons can prevent failure of spikes to propagate in the cord after DR stimulation, we worried that PAD somehow influenced the initiation of the spike by the dorsal root stimulation at the silver wire. However, we ruled this out by stimulating dorsal roots as far away from the spinal cord as possible (20 mm), where PAD has no effect, due to the exponential attenuation of its dorsal root potential with distance (see above), and found that PAD still facilitated sensory axon spike transmission to motoneurons. The added advantage of these long roots is that there is a clean 1 ms separation between the stimulus artifact and the afferent volley arriving at the spinal cord, allowing us to quantify small FPs and afferent volleys that are otherwise obscured by the artifact.
We did not consistently use high temperature ex vivo baths (32°C) because the VR and DR responses to activation of DRs or PAD neurons are irreversibly reduced by prolonged periods at these temperatures, suggesting that the increased metabolic load and insufficient oxygen penetration deep in the tissue damages the cord at these temperatures. Importantly, others have reported that in sensory axons PAD-evoked spikes (DRRs) are eliminated in a warm bath and argued that this means they are not present in vivo, and not able to evoked a motoneuron response 7, despite evidence to the contrary 9, 46. However, we find that PAD itself is reduced in a warm bath by the above irreversible damage, and it is thus not big enough to evoke spikes in sensory axons; thus, this does not tell us whether these spikes should be present or not in vivo. Actually, in vivo we sometimes observed that with optogenetic activation of GABAaxo neurons and associated PAD there was a direct excitation of the motoneurons (seen in the EMG) at the latency expected for PAD evoked spikes (not shown). However, this was also at the latency of the postsynaptic inhibition produced by this same optogenetic stimulation, which often masked the excitation (Fig. 6). In retrospect, examining the GABAaxo evoked motoneuron responses during optogenetic-evoked PAD (Fink et al.)7, 17, or sensory-evoked PAD17, 18, there is a small excitation riding on the postsynaptic IPSPs from the activation of there GABAaxo neurons. This is consistent with the PAD evoked spike activating the monosynaptic pathway, which inhibits subsequently tested monosynaptic responses by post activation depression (see Discussion).
The latency of a single synapse in our ex vivo preparation at 23 – 24°C was estimated from the difference between the time arrival of the sensory afferent volley at the motoneurons (terminal potential seen in intracellular and extracellular recordings) and the onset of the monosynaptic EPSP in motoneurons. This was consistently 1 – 1.2 ms (Fig. 5b and e). This is consistent with a Q10 of about 1.8 – 2.4 for synaptic transmission latency 96, 97, and 0.4 ms monsynaptic latency at body temperature 98, 99. Based on these considerations we confirm that the PAD evoked in sensory axons is monosynaptically produced by optogenetic activation of GABAaxo neurons with light, since it follows ∼1 ms after the first spike evoked in GABAaxo neurons by light (Fig. 3a). This first spike in GABAaxo neurons itself takes 1 – 2 ms to arise and so the overall latency from light activation to PAD production can be 2 - 3 ms (Fig. 3f), as seen for IPSCs at this temperature in other preparations 100. With DRs stimulation PAD arises with a minimally 4 – 5 ms latency, which is consistent with a trisynaptic activation of the sensory axon, after taking into account time for spikes to arise in the interneurons involved (Fig. 4a,e).
Immunohistochemisty
Tissue fixation and sectioning
After sensory axons in mouse and rat spinal cords were injected with neurobiotin the spinal cord was left in the recording chamber in oxygenated nACSF for an additional 4 – 6 hr to allow time for diffusion of the label throughout the axon. Then the spinal cord was immersed in 4% paraformaldehyde (PFA; in phosphate buffer) for 20-22 hours at 4°C, cryoprotected in 30% sucrose in phosphate buffer for 24-48 hours. Alternatively, afferents were labelled genetically in VGLUT1Cre/+; R26lsl-tdTom mice, which were euthanized with Euthanyl (BimedaMTC; 700 mg/kg) and perfused intracardially with 10 ml of saline for 3 – 4 min, followed by 40 ml of 4% paraformaldehyde (PFA; in 0.1 M phosphate buffer at room temperature), over 15 min. Then spinal cords of these mice were post-fixed in PFA for 1 hr at 4°C, and then cryoprotected in 30% sucrose in phosphate buffer (∼48 hrs). Following cryoprotection all cords were embedded in OCT (Sakura Finetek, Torrance, CA, USA), frozen at −60C with 2-methylbutane, cut on a cryostat NX70 (Fisher Scientific) in sagittal or transverse 25 µm sections, and mounted on slides. Slides were frozen until further use.
Immunolabelling
The tissue sections on slides were first rinsed with phosphate buffered saline (PBS, 100 mM, 10 min) and then again with PBS containing 0.3% Triton X-100 (PBS-TX, 10 min rinses used for all PBS-TX rinses). For the sodium channel antibody, we additionally incubated slides three times for 10 min each with a solution of 0.2% sodium borohydride (NaBH4, Fisher, S678-10) in PB, followed by a PBS rinse (4x 5 min). Next, for all tissue, nonspecific binding was blocked with a 1 h incubation in PBS-TX with 10% normal goat serum (NGS; S-1000, Vector Laboratories, Burlingame, USA) or normal donkey serum (NDS; ab7475, Abcam, Cambridge, UK). Sections were then incubated for at least 20 hours at room temperature with a combination of the following primary antibodies in PBS-TX with 2% NGS or NDS: rabbit anti-α5 GABAA receptor subunit (1:200; TA338505, OriGene Tech., Rockville, USA), rabbit anti-α1 GABAA receptor subunit (1:300; 06-868, Sigma-Aldrich, St. Louis, USA), guinea pig anti-α2 GABAA receptor subunit (1:500; 224 104, Synaptic Systems, Goettingen, Germany), chicken anti-γ2 GABAA receptor subunit (1:500; 224 006, Synaptic Systems, Goettingen, Germany), rabbit anti-GABAB1 receptor subunit (1:500; 322 102, Synaptic Systems, Goettingen, Germany), mouse anti-Neurofilament 200 (NF200) (1:2000; N0142, Sigma-Aldrich, St. Louis, USA), guinea pig anti-VGLUT1 (1:1000; AB5905, Sigma-Aldrich, St. Louis, USA), rabbit anti-Caspr (1:500; ab34151, Abcam, Cambridge, UK), mouse anti-Caspr (1:500; K65/35, NeuroMab, Davis, USA), chicken anti-Myelin Basic Protein (MBP) (1:200; ab106583, Abcam, Cambridge, UK), chicken anti-VGAT (1:500; 131 006, Synaptic Systems, Goettingen, Germany), rabbit anti-VGAT (1:500; AB5062P, Sigma-Aldrich, St. Louis, USA), rabbit anti-EYFP (1:500; orb256069, Biorbyt, Riverside, UK), goat anti-RFP (1:500; orb334992, Biorbyt, Riverside, UK), rabbit anti-RFP (1:500; PM005, MBL International, Woburn, USA), rabbit anti-GFP (1:500, A11122, ThermoFisher Scientific, Waltham, USA), and mouse anti-Pan Sodium Channel (1:500; S8809, Sigma-Aldrich, St. Louis, USA). The latter is a pan-sodium antibody, labelling an intracellular peptide sequence common to all known vertebrate sodium channels. Genetically expressed EYFP, tdTom (RFP) and GFP were amplified with the above antibodies, rather than rely on the endogenous fluorescence. When anti-mouse antibodies were applied in mice tissue, the M.O.M (Mouse on Mouse) immunodetection kit was used (M.O.M; BMK-2201, Vector Laboratories, Burlingame, USA) prior to applying antibodies. This process included 1h incubation with a mouse Ig blocking reagent. Primary and secondary antibody solutions were diluted in a specific M.O.M diluent.
The following day, tissue was rinsed with PBS-TX (3x 10 min) and incubated with fluorescent secondary antibodies. The secondary antibodies used included: goat anti-rabbit Alexa Fluor 555 (1:200; A32732, ThermoFisher Scientific, Waltham, USA), goat anti-rabbit Alexa Fluor 647 (1:500, ab150079, Abcam, Cambridge, UK), goat ant-rabbit Pacific orange (1:500; P31584, ThermoFisher Scientific, Waltham, USA), goat anti-mouse Alexa Fluor 647 (1:500; A21235, ThermoFisher Scientific, Waltham, USA), goat anti-mouse Alexa Fluor 488 (1:500; A11001, ThermoFisher Scientific, Waltham, USA), goat anti-mouse Alexa Fluor 555 (1:500; A28180, ThermoFisher Scientific, Waltham, USA), goat anti-guinea pig Alexa Fluor 647 (1:500; A21450, ThermoFisher Scientific, Waltham, USA), goat anti-chicken Alexa Fluor 405 (1:200; ab175674, Abcam, Cambridge, UK), goat anti-chicken Alexa Fluor 647 (1:500; A21449, ThermoFisher Scientific, Waltham, USA), donkey anti-goat Alexa Fluor 555 (1:500; ab150130, Abcam, Cambridge, UK), donkey anti-rabbit Alexa Fluor 488 (1:500; A21206, ThermoFisher Scientific, Waltham, USA), Streptavidin-conjugated Alexa Fluor 488 (1:200; 016-540-084, Jackson immunoResearch, West Grove, USA) or Streptavidin-conjugated Cyanine Cy5 (1:200; 016-170-084, Jackson immunoResearch, West Grove, USA) in PBS-TX with 2% NGS or NDS, applied on slides for 2 h at room temperature. The latter streptavidin antibodies were used to label neurobiotin filled afferents. After rinsing with PBS-TX (2 times x 10 min/each) and PBS (2 times x 10 min/each), the slides were covered with Fluoromount-G (00-4958-02, ThermoFisher Scientific, Waltham, USA) and coverslips (#1.5, 0.175 mm, 12-544-E; Fisher Scientific, Pittsburg, USA).
Standard negative controls in which the primary antibody was either 1) omitted or 2) blocked with its antigen (quenching) were used to confirm the selectivity of the antibody staining, and no specific staining was observed in these controls. Most antibodies had been previously tested with quenching for selectivity, as detailed in the manufacture’s literature and other publications 16, but we verified this for the GABA receptors with quenching. For antibody quenching, the peptides used to generate the antibodies, including anti-α5 GABAA receptor subunit (AAP34984, Aviva Systems Biology, San Diego, USA), anti-α1 GABAA receptor subunit (224-2P, Synaptic Systems, Goettingen, Germany) and anti-γ2 GABAA receptor subunit (224-1P, Synaptic Systems, Goettingen, Germany), were mixed with the antibodies at a 10:1 ratio and incubated for 20 h and 4°C. This mixture was then used instead of the antibody in the above staining procedure.
Confocal and epifluorescence microscopy
Image acquisition was performed by confocal (Leica TCS SP8 Confocal System) and epifluorescence (Leica DM 6000 B) microscopy for high magnification 3D reconstruction and low magnification imaging, respectively. All the confocal images were taken with a 63x (1.4 NA) oil immersion objective lens and 0.1 µm optical sections that were collected into a z-stack over 10–20 µm. Excitation and recording wavelengths were set to optimize the selectivity of imaging the fluorescent secondary antibodies. The same parameters of laser intensity, gain and pinhole size was used to take pictures for each animal, including the negative controls. Complete sagittal sections were imaged with an epifluorescence 10x objective lens using the Tilescan option in Leica Application Suite X software (Leica Microsystems CMS GmbH, Germany). Sequential low power images were used to reconstruct the afferent extent over the whole spinal cord, using CorelDraw (Ottawa, Canada), and to identify locations where confocal images were taken.
3D reconstruction of afferents and localization of GABA receptors
The fluorescently labelled afferents (neurobiotin, tdTom), GABA receptors, VGLUT1, VGAT, NF200, Caspr, MBP and sodium channels were analyzed by 3D confocal reconstruction software in the Leica Application Suite X (Leica Microsystems CMS GmbH) 16. To be very conservative in avoiding non-specific antibody staining, a threshold was set for each fluorescence signal at a minimal level where no background staining was observed in control tissue with the primary antibody omitted, less 10%. Signals above this threshold were rendered in 3D for each antibody. Any GABA receptor, Caspr or NaV expression within the volume of the neurobiotin filled axon (binary mask set by threshold) was labelled yellow, pink and white respectively (Fig. 1), and the density within the afferents quantified using the same Leica software. Receptor densities were measured for all orders of branch sizes (1st, 2nd, 3rd etc.; see below), for both branches dorsal to the central canal (dorsal) and ventral to the central canal (ventral). Nodes were identified with dense bands of Caspr or Na channel labelling (and lack of MBP). Branch points were also identified. We also examined raw image stacks of the neurobiotin afferents and receptors, to confirm that the automatically 3D reconstructed and identified receptors labelled within the afferent (yellow) corresponded to manually identified receptors colocalized with neurobiotin (Fig. 1). This was repeated for a minimum of 10 examples for each condition, and in all cases the 3D identified and manually identified receptors and channels were identical. Many receptors and channels lay outside the afferent, and near the afferent these were difficult to manually identify without the 3D reconstruction software, making the 3D reconstruction the only practical method to fully quantify the receptors over the entire afferent. We also optimized the reconstruction of the neurobiotin filled afferents following the methods of Fenrich101, including brightening and widening the image edges slightly (1 voxel, 0.001 µm3) when necessary to join broken segments of the afferent in the final afferent reconstruction, to account for the a priori knowledge that afferents are continuous and neurobiotin signals tend to be weaker at the membrane (image edges) and in fine processes.
GABA receptors usually occurred in the axons in distinct clusters. The distances between these receptor clusters and nodes or branch points was measured and average distances computed, from high power confocal images evenly sampled across the axon arbour (employing ∼100 receptor clusters from ∼10 images per labelled Ia afferent). The average distance between the receptor clusters and the nearest axon terminals on the motoneurons was also computed, but this was complicated by the very large distances often involved, forcing us to compute the distances from low power images and relate these to the high power images of receptors sampled relatively evenly along the axon arbour. For this distance calculation, to avoid sampling bias in the high power images, we only admitted images from branch segments (1st, 2nd and 3rd order, detailed below) that had a receptor density within one standard deviation of the mean density in branch types with the highest density (1st order ventral branches for GABAA receptors and 3rd order ventral terminal branches for GABAB receptors; i.e. images from branches with density above the dashed confidence interval lines in Fig. 1g were included); this eliminated very large distances being included from branch segments with relatively insignificant receptor densities. We also confirmed these calculations by computing the weighted sum of all the receptor distances weighted by the sum of the receptor density for each branch type (and divided by the sum of all receptor densities), which further eliminated sampling bias. This gave similar average distance results to the above simpler analysis (not shown).
Sensory axon branch order terminology
The branches of proprioceptive axons were denoted as follows: dorsal column branches, 1st order branches that arose of the dorsal column and project toward the motoneurons, 2nd order branches that arose from the 1st order branches, and 3rd order branches that arose from the 2nd order branches. Higher order branches occasionally arose from the 3rd order branches, but these were collectively denoted 3rd order branches. First and second order branches were myelinated with large dense clusters of sodium channels at the nodes in the myelin gaps, which were characteristically widely spaced. As the second order branches thinned near the transition to 3rd order branches, they became unmyelinated, and at this point sodium channel clusters were smaller and more closely spaced (∼6 µm apart, not shown). These thinned branches gave off 3rd order (and higher) unmyelinated terminal branches with chains of characteristic terminal boutons. The 1st order branches gave off 2nd order branches along most of their length as they traversed the cord from the dorsal columns to the motoneurons, but we separately quantified 1st, 2nd and 3rd order branches in the dorsal and ventral horn.
Node identification
Nodes in myelinated axon segments nodes were identified either directly via direct Na channel clusters and paranodal Caspr, or indirectly by their characteristic paranodal taper. That is, in the paranodal region the neurobiotin filled portion of the axon tapered to a smaller diameter, likely because the Caspr and presumably other proteins displaced the cytoplasmic neurobiotin, which also made the intracellular neurobiotin label less dense (Fig. 1c, black regions in taper). Regardless of the details, this taper made nodes readily identifiable. This taper forces the axial current densities to increase at the nodes, presumably assisting spike initiation, and consistent with previous reconstructions of myelinated proprioceptive afferents 102.
Computer simulations
All computer models and simulations were implemented in NEURON ver7.5 103. The geometry and myelination pattern of the model were extracted from a previous study that used serial-section electron microscopy to generate about 15,000 photomicrographs to reconstruct a large myelinated proprioceptive Ia afferent collateral in the cat (Nicol and Walmsley, 1991). This structure was used in a prior modeling study 34. Four classes of segment were defined in the model: myelinated internodes, nodes, unmyelinated bridges, and terminal boutons. Data from 18 of the 83 segments were missing from the original study. The missing data were estimated using mean values of the same segment class. The cable properties of the model were determined from diameter-dependent equations previously used for models of myelinated axons104 and included explicit representation of myelinated segments using the double cable approach104–106. Hodgkin-Huxley style models of voltage gated sodium (transient and persistent) and potassium channels were adopted from a previous study, at 37°C104. All three voltage-gated conductances were colocalized to unmyelinated nodes and segments throughout the modelled axon collateral. The density of sodium and potassium conductances was adjusted to match the size and shape of experimentally recorded action potentials. To be conservative, sodium channels were placed at each node and bouton (gNa = 1 S/cm2), even though bouton immunolabelling for these channels was not common in our terminal bouton imaging (Fig. 1), since disperse weak sodium channel labelling may have been missed. Removing these bouton sodium channels did not qualitatively change our computer simulation results (not shown). Current clamp stimulation was applied to the middle of the first myelinated segment (pulse width 0.1 ms, amplitude 2 nA; near dorsal root) to initiate propagating action potentials in the model. Voltage at multiple sites of interest along the collateral was measured to assess propagation of action and graded potentials through branch points. Transient chloride conductance (i.e. GABAA receptors) was modeled using a double-exponential point process (Eq. 1); parameters were manually fit to experimental data. GABAA receptors were localized to nodes at branch points to match experimental data. The amplitude and time course of the modeled PAD (also termed PAD) was measured from the first myelinated internode segment, similar to the location of our intra-axonal recordings. The parameters at all synapses were the same: time constant of rise (τrise) = 6ms, time constant of decay (τdecay) = 50ms, default maximum conductance (gmax) = 1.5nS (varied depending on simulation, see figure legends), and chloride reversal potential (ECl-) = −25mV (i.e. 55 mV positive to the resting potential to match our experimental data) 16. Space constants (λS) were computed for each segment of the afferent, from subthreshold current injections (100 ms) on the distal end of each branch segment and fitting an exponential decay (with space constant λS) to the passive depolarization along its length, and then repeating this with current injected in the proximal end to get a second λS, and finally averaging these two space constants.
QUANTIFICATION AND STATISTICAL ANALYSIS
Data were analyzed in Clampfit 8.0 (Axon Instruments, USA) and Sigmaplot (Systat Software, USA). A Student’s t-test or ANOVA (as appropriate) was used to test for statistical differences between variables, with a significance level of P < 0.05. Power of tests was computed with α = 0.05 to design experiments. A Kolmogorov-Smirnov test for normality was applied to the data set, with a P < 0.05 level set for significance. Most data sets were found to be normally distributed, as is required for a t-test. For those that were not normal a Wilcoxon Signed Rank Test was instead used with P < 0.05. Categorical data was instead analyzed using Chi-squared tests, with Yate’s continuity correction used for 2 × 2 contingency tables and again significant difference set at P < 0.05. Data are plotted as bar graphs of mean ± standard deviation (SD, error bar representing variability) or as box plots representing the interquartile range and median and error bars extend to the most extreme data point within 1.5 times the interquartile range (mean also shown as thick lines in boxes).
Data availability
All data are available in the manuscript or the supplementary materials. Raw data are available upon request to the corresponding authors. This study did not generate data sets or new unique reagents.
Code availability
The computer code used to perform the axon simulations (Extended Date Fig. 4) are publicly available on the github repository: https://github.com/kelvinejones/noah-axon.git
Author information
Contributions. K.H, and A.M.L-O. designed the study, carried out the animal experiments and analyzed data. K.M. and M.A.G. designed and performed the human experiments. N.P. and K.E.J. designed and performed the computer simulations. S.L., S.B., A.M., M.J.S. and R.S. assisted with animal electrophysiology. K.F. and A.M.L-O provided confocal microscopy. K.K.F, S.L., D.A.R. and K.H. developed the transgenic mice and performed the optogenetic experiments. Y.L. and D.J.B. conceived and designed the study, carried out experiments and analyzed data. D.J.B, K.H., and Y.L. wrote the paper, with editing from other authors. These authors contributed equally: Krishnapriya (Veni) Hari and Ana M. Lucas-Osma. These authors jointly supervised this work as senior authors: Yaqing Li, Keith K. Fenrich and David J. Bennett.
Ethical declarations
All authors declare no competing interests.
Supplementary information
Acknowledgements
We thank Leo Sanelli, Jennifer Duchcherer, Babak Afsharipour and Christopher K. Thompson for technical assistance, and Shawn Hochman, CJ Heckman, FJ Alvarez and Tia Bennett for discussions and editing the manuscript. VGLUT1Cre mice were kindly donated by Dr. Francisco J. Alvarez. This research was supported by the Canadian Institutes of Health Research (MOP 14697 and PJT 165823 D.J.B.) and the US National Institutes of Health (NIH, R01NS47567; D.J.B. and K.F.).
Footnotes
References
- 1.↵
- 2.↵
- 3.↵
- 4.↵
- 5.↵
- 6.↵
- 7.↵
- 8.↵
- 9.↵
- 10.↵
- 11.↵
- 12.↵
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.↵
- 20.↵
- 21.↵
- 22.↵
- 23.↵
- 24.↵
- 25.↵
- 26.↵
- 27.↵
- 28.↵
- 29.↵
- 30.↵
- 31.↵
- 32.↵
- 33.↵
- 34.↵
- 35.↵
- 36.↵
- 37.↵
- 38.↵
- 39.↵
- 40.↵
- 41.↵
- 42.↵
- 43.↵
- 44.↵
- 45.↵
- 46.↵
- 47.↵
- 48.↵
- 49.↵
- 50.↵
- 51.↵
- 52.↵
- 53.↵
- 54.↵
- 55.↵
- 56.↵
- 57.↵
- 58.↵
- 59.↵
- 60.↵
- 61.↵
- 62.↵
- 63.↵
- 64.↵
- 65.↵
- 66.↵
- 67.↵
- 68.↵
- 69.↵
- 70.↵
- 71.↵
- 72.
- 73.↵
- 74.↵
- 75.↵
- 76.↵
- 77.
- 78.↵
- 79.↵
- 80.↵
- 81.↵
- 82.↵
- 83.↵
- 84.↵
- 85.↵
- 86.↵
- 87.↵
- 88.↵
- 89.↵
- 90.↵
- 91.↵
- 92.↵
- 93.↵
- 94.↵
- 95.↵
- 96.↵
- 97.↵
- 98.↵
- 99.↵
- 100.↵
- 101.↵
- 102.↵
- 103.↵
- 104.↵
- 105.
- 106.↵
- 107.↵
- 108.↵
- 109.
- 110.↵
- 111.↵
- 112.↵
- 113.↵
- 114.↵
- 115.↵
- 116.↵
- 117.
- 118.
- 119.
- 120.
- 121.
- 122.
- 123.
- 124.
- 125.
- 126.
- 127.
- 128.
- 129.
- 130.
- 131.
- 132.
- 133.
- 134.
- 135.
- 136.
- 137.
- 138.