Abstract
Injury that severs peripheral nerves often results in long-lasting motor behavioral deficits and in reorganization of related spinal motor circuitry, neither of which reverse even after nerve regeneration. Stretch areflexia and gait ataxia, for example, emerge from a combination of factors including degeneration of Ia–motoneuron synapses between peripherally damaged Ia muscle spindle afferents and motoneurons. Based on evidence that nerve injury acts via immune responses to induce synapse degeneration, we hypothesized that suppressing inflammatory responses would preserve Ia–motoneuron connectivity and aid in restoring normal function. We tested our hypothesis by administering the anti-inflammatory agent minocycline in male and female rats following axotomy of a peripheral nerve. The connectivity of Ia–motoneuron synapses was then assessed both structurally and functionally at different time points. We found that minocycline treatment overcame the physical loss of Ia contacts on motoneurons which are otherwise lost after axotomy. While necessary for functional recovery, synaptic preservation was not sufficient to overcome functional decline expressed as smaller than normal stretch-evoked synaptic potentials evoked monosynaptically at Ia–motoneuron connections and an absence of the stretch reflex. These findings demonstrate a limited capacity of minocycline to rescue normal sensorimotor behavior, illustrating that structural preservation of synaptic connectivity does not ensure normal synaptic function.
SIGNIFICANCE STATEMENT Here we demonstrate that acute treatment with the semisynthetic tetracycline anti-inflammatory agent minocycline permanently prevents the comprehensive loss of synaptic contacts made between sensory neurons and spinal motoneurons following peripheral nerve injury and eventual regeneration. Treatment failed, however, to rescue normal function of those synapses or the reflex circuit they mediate. These findings demonstrate that preventing synaptic disconnection alone is not sufficient to restore neural circuit operation and associated sensorimotor behaviors.
Introduction
Spinal sensorimotor circuits undergo permanent reorganization following peripheral nerve injury (PNI) (Alvarez et al., 2010, 2020). One prime example comes from the central degradation of the monosynaptic connection between propriosensory Ia afferents and motoneurons (Rotterman et al., 2014). Interruption in this circuit connection deprives the CNS of its most rapid neural response to mechanical perturbations of body movements and posture (Prochazka, 1996; Gandevia et al., 2002). This is reflected by the loss of the stretch reflex and is suspected of contributing to gait and postural disorders and instability that persist even after peripheral nerves successfully reinnervate functionally appropriate muscles (Cope and Clark, 1993; Cope et al., 1994; Haftel et al., 2005; Maas et al., 2007; Sabatier et al., 2011; Lyle et al., 2016). These observations promote interest in preserving synaptic connections between muscle propriosensors and motoneurons, which should improve the capacity of peripheral nerve regeneration to restore sensorimotor behaviors.
Permanent removal of the Ia afferent synapses differs from the classically defined “synaptic stripping” phenomenon in which synapses, predominately excitatory, temporarily retract from axotomized motoneurons and then reattach following peripheral regeneration (Blinzinger and Kreutzberg, 1968; Sumner, 1975; D. H. Chen, 1978; Linda et al., 2000; Berg et al., 2013). However, nerve regeneration largely fails to reattach synapses formed on motoneurons by Ia propriosensory neurons after PNI severs both the presynaptic and postsynaptic axons. Instead, the Ia–motoneuron synapses detach and are permanently degraded through a central neuroinflammatory process mediated through a spinal microglia reaction (Rotterman et al., 2019). Previous investigation of PNI and other neurodegenerative models has suggested that the semisynthetic tetracycline antibiotic known as minocycline can retard synaptic loss (Shermadou, 2013; Bordone et al., 2017; Di Liberto et al., 2018; Squair et al., 2018; Ghosh et al., 2019; Celorrio et al., 2022). How this retention impacts function at the synaptic level remains to be discovered.
Here, we propose two interdependent hypotheses: (1) that minocycline preserves Ia–motoneuron synapses normally lost following injury and repair of the homonymous muscle nerve; and (2) that minocycline restores synaptic and reflex function of preserved Ia–motoneuron synapses. We tested these hypotheses by performing survival surgeries on adult rats in which we immediately rejoined surgically cut ends of the medial gastrocnemius (MG) nerve and began administering the anti-inflammatory compound minocycline for 2 weeks, as this time course coincides with the initial central inflammatory response in rodents after nerve cut (Rotterman et al., 2019). Through a series of time points in our current study, we mapped Ia boutons on retrogradely labeled motoneurons, assessed the functional connectivity and efficacy of Ia synapses on motoneurons, and measured the reflex force evoked by the reinnervated muscle in response to stretch. We established that minocycline treatment following nerve injury and repair prevented a majority of Ia contacts from detaching from the motoneuron, but it failed to rescue the functional monosynaptic connectivity and efficacy of those synapses or the stretch reflex they normally mediate. Our findings emphasize that preserving synaptic connections, while absolutely necessary for functional recovery, is not alone sufficient to improve behavioral outcomes.
Materials and Methods
All animal experiments and procedures were approved by the Institutional Animal Care and Use Committee at Georgia Institute of Technology. Survival surgeries were conducted in adult Wistar rats (250-300 g, Charles River Laboratory), which included both males and females to account for potential sex differences. Data presented here from electrophysiological studies included a total of 42 rats (157 motoneurons), motoneuron reconstructions included a total of 41 rats (408 reconstructions), and microglia analyses included 17 rats (167 sections).
Experimental design
Animal groups
Rats were randomly placed in one of four groups for investigation: control (Ctl), control treated with minocycline (Ctl + Mino), MG nerve cut-repair treated with vehicle (Cut + Veh), or MG nerve cut-repair treated with minocycline (Cut + Mino). For all survival surgeries, rats were first deeply anesthetized by isoflurane inhalation (induction: 5%; maintenance: 1%-3% in 100% O2) and were treated with an initial subcutaneous injection of slow-release buprenorphine (0.1 mg/kg) immediately after surgery to treat possible pain prophylactically, and thereafter as needed. Animals were observed daily for 1 week after survival surgeries to monitor for any signs of pain and/or distress. All surgical procedures were conducted using sterile technique.
Time points
Terminal experiments took place at three different time points: 14 d when central inflammation is exacerbated and majority of the synaptic boutons, specifically glutamatergic synapses, are actively undergoing synaptic stripping; at 3 months when motor axons have regenerated in the periphery; and our final comparison, at 6+ months (6-12 months) when both motor and proprioceptor Ia sensory axons have reinnervated the muscle allowing functional recovery to be evaluated.
Nerve repairs
Once a surgical plane of anesthesia was achieved, the left hindlimb was secured and a vertical incision was made in the skin along the posterior side of the leg, passing through the biceps femoris, to expose the MG nerve. Surrounding connective tissue was cleared away and the nerve was completely transected. A 3 mm2 piece of subdermal silicon sheet (Dow Corning, catalog #501-1) was placed under the two nerve stumps. Then a fibrin glue mixture was prepared containing thrombin, fibronectin, and fibrinogen was used to rejoin the nerve stumps (Guest et al., 1997; Akhter et al., 2019). When the glue dried, the biceps femoris muscle was sutured back together with absorbable suture and the incision was closed using wound clips.
Retrograde tracing
Similarly, as described above, the MG muscle was exposed. Small boluses of 2-3 µl of 5% wheat germ agglutinin conjugated to an AlexaFluor-555 (Invitrogen, catalog #W32464) or 1.5% Fast Blue (Polysciences, catalog #17 740) were injected into the MG muscle belly with a total volume of tracer equaling ∼35 µl per rat. For 14 d survival surgeries, Fast Blue was injected before the nerve injury. For long-term survival experiments (≥ 3 months), WGA-555 was injected 3-5 d before the terminal surgery.
Drug treatment
Rats were treated 1× daily with 35 mg of minocycline hydrochloride (TCI, catalog #M2288) for 14 d starting immediately after the nerve injury procedure. Minocycline was mixed in a Nutra-Gel Diet (Bio-Serv, catalog #F4798-KIT) as per the manufacturer's instructions, and a single food cube was placed in the home cage of the animal in addition to the standard chow. Total cube consumption was monitored daily and any animal consuming on average <90% of the total volume was excluded from the study. Identical cubes containing no drug and were used for vehicle experiments.
Terminal physiological experiments
Anesthesia and vital signs
Rats included in physiological studies were prepared and monitored for up to 12 h during data collection. First, rodents were deeply anesthetized with isoflurane inhalation (induction: 5% in 100% O2) to and ensure a complete absence of withdrawal reflex to noxious stimuli. Once a surgical plane of anesthesia was achieved, a tracheal canula was inserted to maintain isoflurane inhalation through the duration of the experiment (1%-3% in 100% O2). In preparation for stretch reflex experiments, both carotid arteries were tied off with 6–0 suture to interrupt cortical blood flow before decerebration. The rodent's conditions were continuously monitored and assessed by respiration rate (40-60 breaths per min), end-tidal CO2 (2.5%-5%), oxygen saturation (>95%), heart rate (300-500 beats per min), and core body temperature (36°C-38°C). These values were modulated by adjusting isoflurane concentrations. Subcutaneous delivery of Lactated Ringer's solution was also provided throughout the experiments. In some cases, a paralytic drug (pancuronium bromide 0.2 mg/kg) was delivered because of membrane potential oscillations introduced by respiratory artifact, and rats were then placed on a ventilator. This approach promoted stable intracellular recordings.
Surgical preparation
Previously described procedures by our laboratory were used to isolate the MG muscle in the left hindlimb and the corresponding MG nerve from surrounding tissue (Bullinger et al., 2011; Nardelli et al., 2017; Rotterman et al., 2021). All other nerve branches supplying the lower limb were crushed. Rats were then secured in a stereotaxic recording frame, and a laminectomy was performed to expose the spinal cord (S1-L3). The MG muscle was secured to a servomotor lever by the muscle tendon to control length and record force (Aurora Scientific, model 305). A monopolar silver hook electrode was placed under the MG nerve for antidromic stimulation. A pair of silver wire electrodes were placed under the corresponding dorsal root for extracellular sensory nerve recording. All neural and muscle tissue were bathed in warm mineral oil through the duration of the experiment.
Intracellular data collection
Borosilicate glass microelectrodes (6-12 mΩ filled with 2 m K-acetate) coupled to an electrometer (Axoclamp) were advanced through the L4-L5 spinal cord with a micromanipulator device (Transvertex Microdrive) to penetrate spinal motoneurons in the L4-L5 region of the spinal cord. MG motoneurons were identified via electrical antidromic nerve stimulation, and motor pool membership was confirmed by the presence of a muscle twitch. Once a steady membrane potential was achieved and antidromic action potentials exceeded 60 mV, a series of biophysical properties were recorded. Properties included rheobase (minimal 50 ms depolarizing current pulse to elicit a single action potential), input resistance (average membrane potential change in response to 50 ms −1 and −3 nA hyperpolarizing current), and afterhyperpolarization (AHP, suprathreshold 0.5 ms current pulse) (Bichler et al., 2007; Rotterman et al., 2021). In a subset of experiments, dorsal rootlets overlaid on silver-wire bipolar electrodes were penetrated intra-axonally with sharp micropipettes (∼20 mΩ) to record single muscle afferents. Specific afferent types were classified as previously described (Vincent et al., 2017). All physiological data were digitized at 20 kHz with a 1401 DAC (Cambridge Electronic Design) and analyzed in Spike2 software.
Stretch-evoked synaptic potentials (SSPs)
Muscle was stretched to a constant resting length (Lo) corresponding to 10 grams of tension maintained by the servomotor. Vibration stimulation (100 Hz at 80micron amplitude for 1 s) was used to elicit SSPs. SSPs were averaged over 10-20 trials and compared with resting membrane potential before the stretch.
Stretch reflex experiments
The rat was decerbrated by removing all brain tissue rostral to intercollicular transecrtion of the brainstem. By rendering rats insensate, decerebration enabled cessation of anesthesia as needed to disinhibit reflex responsiveness. Data collection began ∼30 min after anesthesia delivery had ceased, although 100% O2 was continuously delivered during the duration of the experiment. Muscle force was recorded together with EMG, obtained from bipolar silver wires inserted in the muscle, in response to tendon vibration. In a subset of animals, tetanic stimulation of the sural nerve was evoked to ensure the animal was indeed reflexive even when a stretch reflex was absent (see Fig. 5B) The determination of reflexive or nonreflexive was gaited on the presence of a corresponding EMG signal in response to the muscle stretch. If no EMG was detected, the individual trial was classified as no reflex; therefore, force was reported as zero.
Tissue collection and immunofluorescence
Tissue harvesting
Following electrophysiological recordings, rats received a lethal dose of isoflurane anesthesia (5%) until respiration rates dropped below 10 beats/min. Then, the animals were transcardially perfused with chilled vascular rinse containing heparin followed by ice-cold 4% PFA in 0.1 m PB, pH 7.4. Spinal cords were extracted and postfixed overnight in 4% PFA at 4°C. Tissue was transferred and stored in 30% sucrose. Transverse sections of the L4-L5 spinal cord region were sectioned at a thickness of 50 or 75 µm on a freezing sliding microtome (Leica Biosystems, model SM2010R) and collected free-floating.
Histologic methods and immunofluorescence
Spinal cord sections containing retrogradely labeled motoneurons (WGA-555 or Fast Blue) were immunohistochemically processed to label Ia afferent synapses with an antibody against the vesicular glutamate transporter isoform 1 (VGluT1) (Table 1). Spinal cord sections from 14 d postinjury animals were processed with antibodies against the ionized calcium binding adaptor molecule 1 (Iba1) and colony stimulating factor 1 (CSF1) (Table 1).
Free-floating sections were washed 3× (10 min per wash) in 0.01 m PBS with 0.3% Triton X-100 (PBST) and then blocked for 1 h in 10% normal donkey serum diluted in PBST. Serum was then aspirated, and tissue sections were incubated overnight in a primary antibody mixture diluted in PBST at room temperature (Table 1). The following day, tissue sections were washed again 3× in PBST and immunoreactive sites were revealed with specie specific secondary antibodies raised in donkey (Table 1). Following a 2 h incubation in the secondary mixture at room temperature, sections were washed in PBS (no Triton), mounted, and coverslipped with Vectashield (Vector Labs, catalog #H1000).
Confocal imaging
Spinal cord and muscle sections were imaged using laser scanning confocal microscopy (Zeiss Biosystems, 900). Z stacks of spinal cord sections containing motoneurons were acquired with a 63× objective (N.A. = 1.4) at a step size of 0.5 µm and a magnification of 0.5×.
Anatomical evaluations
Motoneuron reconstructions and VGluT1 quantification
Ten confocal image stacks were chosen per animal in each group based on strict criteria: (1) the entire soma must have been included in the tissue section (2) the motoneuron must include at least 5 parent dendritic branches (3) the motoneuron must be within the known MG pool located in the L4-L5 region of the spinal cord (Nicolopoulos-Stournaras and Iles, 1983; X. Y. Chen and Wolpaw, 1994); and (4) a dendritic arbor that expanded a minimum of 300 µm in total length. Image stacks meeting this criterion were uploaded to Neurolucida (MicroBrightField, version 2019.1.1), and the soma and proximal dendritic arbor were manually reconstructed in 3D through each optical plane (morphologic parameters are presented in Table 2). VGluT1-immunoreactive boutons that were in direct contact with the surface of the motoneuron were identified and a “marker” was placed at the exact XYZ coordinate to calculate both the somatic density (reported per 1000 µm2) and dendritic linear density (reported per 100 µm) (see Fig. 1B1,B2,C). Quantification of VGluT1 synapses was conducted blindly to the experimenter.
Sholl analysis
To investigate differences in synaptic density along specific regions of the dendritic arbor, we performed a Sholl analysis to account for potential discrepancies in overall total dendrite sampled from each cell. This allowed us to determine whether the VGluT1 synapses were better preserved in specific regions along the dendritic compartment after injury with minocycline treatment (see Fig. 2A). Each Sholl bin accounts for the total dendritic length and total number of VGluT1 synapses in increasing increments of 50 µm moving distally from the middle of the soma. VGluT1 dendritic densities, per 100 µm of dendrite, are reported for three different bins: 50, 100, and 150 µm.
Microglia quantification
Ten confocal image stacks were chosen per animal in each group based on specific criteria. Each section quantified included at least one retrogradely labeled MG motoneuron and was from the L4-L5 region of the spinal cord with the motoneuron centered in the middle of the image. Image stacks were uploaded to Neurolucida, and the total number of Iba1+ cell bodies were counted per 50-µm-thick tissue section. All quantification was performed blindly by the experimenter.
Experimental design and statistical analysis
Linear mixed effect models
We used linear mixed effect statistical models in our comparisons to account for interanimal variance and reduce Type I error rates (Yu et al., 2022). These were based on previously described models using packages lme4, nlme, icc, and emmeans in R (Yu et al., 2022). Pairwise post hoc comparisons were performed using the Kenward–Roger method to determine degrees of freedom, and p value adjustments were derived from a Tukey Method approach. Statistical values reported for control conditions from pairwise comparisons are from 6+ months analyses unless stated otherwise. Alpha values set to determine significance are <0.05. Box plots representing these data were made in R using package ggplot2. All data points are displayed as individual points and color-coded based on the animal in which they were collected from. This is used for visualization purposes; however, for statistical analyses, data points were not treated as independent samples. In comparing the SSP amplitude in response to vibration, we performed a nonparametric analysis using a Kruskal–Wallis test followed by a Wilcoxon Rank Sum test as the number of responders after injury resulted in a substantially smaller sample size that contained a large amount of variance.
Statistical modeling
Bayesian data analytic approaches were used to empirically derive the full joint posterior probability distribution. Our models describe uncertainty in the response variable, for example, probability of response to stimuli, VGluT1 density, y, conditional on unknown parameters θ (e.g., regression coefficients) and predictors (e.g., experimental treatment group, sex), x, as well as the a priori uncertainty about these parameters and predictors (Rotterman et al., 2021). Bayes theorem describes the proportional relationship (∝) between our prior knowledge about the parameters (before observing the data) and our posterior beliefs about the parameters (after observing the data) as follows:
Somatic and dendritic VGluT1 densities and microglia counts were modeled with R's brms package, specifically the brm function, to construct a hierarchical Gaussian Bayesian model. We modeled the probabilities of stretch reflex and SSP responses using Bayesian logistic regression. Marginally informative priors were applied to model parameters and variance components such that inferences were driven predominantly by the experimental data to explicitly answer our central questions. Prior specifications were based on results from previous preliminary data and published studies by our group (Rotterman et al., 2014; Rotterman and Alvarez, 2020; Housley et al., 2021).
All models were fit using Hamiltonian Markov Chain Monte Carlo sampling to compute credible parameter values. Each model was run with four independent chains for 500 warmup and 10,000 sampling steps. Steps to perform model evaluation and validation have been extensively described in our previous work (Horstman et al., 2019; Housley et al., 2020a, b, 2021; Rotterman et al., 2021). Briefly, for all parameters, the number of effective samples was >2000, convergence was assessed and assumed to have reached the stationary distribution by ensuring that the Gelman–Rubin shrinkage statistic for all reported parameters was <1.05. Trace plots were examined and indicated clear stationarity and good mixing, and numerical checks of sampling quality indicated convergence.
We then generated forward statistical models by drawing 10,000 samples from the posterior distribution (posterior_samples function in the brms package). This forward model allowed generalized comparison of the effects of nerve cut, minocycline, and their combination across time, for example, what is the probability that any given motoneuron from an animal that underwent Cut + Veh or Cut + Mino responds to vibration (see Fig. 4C). We display these forward model predictions as density estimates where: Ctl, Cut + Veh, Cut + Mino groups are green, red, and purple traces, respectively. The Ctl + Mino group was also included in these analyses; however, for visualization purposes, these distributions were excluded from some graphical displays.
Results
Test the hypothesis that minocycline preserves Ia–motoneuron synapses otherwise lost following injury and repair of the homonymous muscle nerve
We hypothesized that minocycline treatment following PNI would preserve VGluT1+ synapses on both the soma and proximal dendritic arbor of axotomized spinal motoneurons compared with those receiving vehicle alone. To test this hypothesis, we examined VGluT1 synaptic density on MG motoneurons sampled from rats treated with minocycline or vehicle and studied each at different times during or after transection and regeneration of the MG nerve (Fig. 1).
Synaptic preservation
We found that VGluT1 contacts began to retract in a proximal to distal gradient (i.e., from soma to dendrites) in Cut + Veh rats as we detected a significant depletion of VGluT1 synapses on the soma 14 d following injury (p = 0.018), but no significant differences on the proximal dendritic arbor (p = 0.329) (Fig. 1F1,F2). However, in the Cut + Mino condition, we completely prevented the loss of VGluT1 synapses on the soma, even at 14 d when synaptic stripping is otherwise readily observable (soma: p = 0.631, dendritic: p = 0.750) (Fig. 1F1,F2). Minocycline alone had no detectable effect on VGluT1 synaptic density in Ctl rats (Fig. 1A1,A2,D1,D2). We also constructed a generative statistical model to compare simulated VGluT1 synaptic density using a Hamiltonian Markov Chain Monte Carlo sampling approach (see Materials and Methods). Both statistical approaches are hereafter deployed to evaluate and compare observed and simulated data to establish the greatest confidence in conclusions drawn from the described experiments. At 14 d after injury without minocycline, the distribution of synaptic density shifted well to the left and outside the range predicted for control values (Fig. 1G). With minocycline treatment, that shift was not observed; but instead, distributions for Ctl and Cut + Mino groups overlapped extensively, giving strong evidence of their comparability. These findings suggest that minocycline preserved somatic VGluT1 synapses by preventing the initial stripping of synapses from occurring.
We also performed a Sholl analysis (bins: 50, 100, and 150 µm) to determine whether the preservation of VGluT1 was specific to defined regions along the dendritic arbor (Fig. 2A). In vehicle-treated rats, there was a trend toward a loss of synapses in all three bins by 14 d after injury, although none of these reached a level of significance compared with their control counterparts (50 µm: p = 0.128, 100 µm: p = 0.538, 150 µm: p = 0.287). Additionally, animals treated with minocycline did not exhibit detectable synaptic loss in any of the Sholl bins following injury compared with Ctl + Mino rats (Fig. 2D1). Generative simulations support these findings by predicting a lower VGluT1 density in vehicle-treated rats, compared with those receiving minocycline, at 14 d in both the 50 and 100 µm bin (Fig. 2E1). Again, minocycline alone did not result in any detectable differences in VGluT1 density compared with control rats within any of the three bins (Fig. 2B,C). These data support the conclusion that minocycline treatment prevents detachment of Ia–motoneuron contacts otherwise induced by PNI, particularly on the somatic region.
Synaptic retention
To investigate whether preservation of synapses persisted after removing minocycline treatment, we quantified and compared VGluT1 contact densities at 3 and 6+ months following nerve injury in vehicle- and minocycline-treated animals (Fig. 1E3-E6). We found that the density of VGluT1 contacts in minocycline-treated rats remained indistinguishable from Ctl + Mino (3 months; soma: p = 0.956, dendritic: p = 1.000, 6+ months; soma: p = 0.491, dendritic: p = 0.883) (Fig. 1F1,F2; Table 2). In the absence of minocycline treatment, synaptic loss continued to progress; and by 3 months, significant depletion was detected on both the soma and dendrites in Cut + Veh animals (soma: p = 0.003, dendritic: p < 0.0001) (Fig. 1F1,F2). Last, there was no recovery in synaptic density even well after peripheral regeneration was complete (soma: p = 0.015, dendritic: p = 0.0001) (Fig. 1F1,F2). Generative models predict modest decreases in VGluT1 density in Cut + Mino animals at 3 and 6+ months; however, the extent of this loss is minimal compared with vehicle-treated animals alone (Fig. 1G2,G3). In the Cut + Veh group, the predicted density continues to decline from 14 d with no substantial improvement even after peripheral regeneration is complete by 6 months (Fig. 1G2,G3). These data demonstrate that the brief period of minocycline treatment (2 weeks) introduced at the time injury preserved the majority of VGluT1 synapses for several months, and presumably longer, which is a prerequisite for functional recovery to occur once peripheral regeneration is complete.
Finally, the Sholl analysis did not detect any significant synaptic loss in any of the three bins in the Cut + Mino group compared with Ctl + Mino (Fig. 2D2,D3). These findings were further supported by simulation studies (Fig. 2E2,E3). Without minocycline, however, Cut + Veh resulted in a significant decline in VGluT1 density in both the 50 µm (3 months: p = 0.0007, 6+ months: p = 0.0027) and 100 µm (3 months: p = 0.0013, 6+ months: p = 0.0185) bins but not in the 150 µm (3 months: p = 0.271, 6+ months: p = 0.5058) (Fig. 2D2,D3). These data support the conclusion that synaptic preservation was long-lasting and uniformly distributed across proximal dendritic compartments of motoneuron in those animals receiving minocycline after nerve injury.
Retention of VGluT1 synapses is not dependent on peripheral regeneration
We next investigated whether minocycline interfered with muscle reinnervation by motor and sensory neurons. Motor recovery was demonstrated when motoneuron action potentials induced by intracellular current injection elicited muscle contraction (Fig. 3A–C). Sensory recovery was assessed from afferent firing responding to tendon vibration stimulation (Fig. 3D1). We have previously reported that vibration stimuli predominantly, if not exclusively, recruits Ia afferents at high frequencies (Vincent et al., 2017). Intriguingly, we found that vibration (Fig. 3D2) could not elicit a dorsal root response 3 months following a nerve Cut + Mino (n = 4 rats), demonstrating that sensory encoding had not recovered at this time point, although motor axons have reconnected with the periphery.
Not until 6 months after nerve injury did we observe muscle afferents that were classifiable as Ia based on their firing entrainment to muscle vibration and initial burst firing at the onset of ramp stretch (n = 10 afferents, 2 rats, Fig. 3E1, E2, G1, G2). This delay establishes that regeneration restores peripheral function more slowly for sensory than for motor neurons. The delay also establishes that minocycline's preservation of central contacts with motoneurons did not rely upon the afferents' recovery of peripheral function.
Test the hypothesis that minocycline restores synaptic and reflex function of preserved Ia–motoneuron synapses
We recorded rat MG motoneurons intracellularly in vivo at multiple time points ranging from 6 to 12 months after injury and found no differences in their biophysical or electrical properties within this time frame (Tables 2, 3). Therefore, these animals were grouped together into one cohort referred to as “6+ months” for both vehicle- and minocycline-treated rats. Intracellular response to vibration stimulation (100 Hz) was first recorded, and motoneurons were classified as “responders” or “nonresponders” based on the strict criteria that a discernable synaptic potential was detected in the motoneuron and associated with each individual vibration cycle (Fig. 4A). This criterion excludes nominal EPSPs in the microvolt range and were only detected by averaging membrane potential in nonresponder motoneurons over thousands of individual vibration cycles.
Since each individual Ia afferent projects to, essentially, every homonymous motoneuron in the spinal cord, vibration stimulation results in 100% of MG motoneurons producing an SSP in Ctl and Ctl + Mino animals (Ctl: n = 36 motoneurons, 17 rats), Ctl + Mino: n = 37 motoneurons, 5 rats) (Mendell and Henneman, 1971) (Fig. 4A,B). However, after Cut + Veh, a large proportion of Ia synapses are lost and the number of responder motoneurons drops to 37.8% (n = 37 motoneurons, 9 rats) (Fig. 4B). Furthermore, we found no improvement in response rates in Cut + Mino rats, although the majority of VGluT1 synapses were retained in this group, with only 22.2% of motoneurons now producing an SSP (n = 36 motoneurons, 9 rats) (Fig. 4B). In line with the observed data, investigation of the posterior predicted densities suggests an almost certainty of producing a synaptic potential in response to vibration in Ctl (green curve). However, in both vehicle and minocycline rats, the probability of producing a synaptic potential drops drastically with no overlap compared with Ctls (Fig. 4C). These findings demonstrate that minocycline's preservation of physical connectivity did not translate into maintenance of functional connectivity in the monosynaptic Ia-motoneuron pathway.
We next investigated the amplitude of the SSPs to determine whether synaptic efficacy was preserved in motoneurons that did respond to vibrations (Fig. 4D1-D4). All SSP values of zero, also known as “nonresponders,” were excluded from this analysis. Here, we measured the amplitude of the first and third SSP response peak (Fig. 4E1,E2, inset). We found that, in both vehicle- and minocycline-treated rats, the amplitude of the SSP was significantly smaller compared with their control counterparts for both peaks (Cut + Veh: peak 1 p = 0.0023, peak 3 p = 0.0002; Cut + Mino: peak 1 p = 0.0008, peak 3 p = 0.0036) (Fig. 4E1,E2), suggesting no improvement in synaptic efficacy with minocycline treatment. Furthermore, there were no detectable differences in amplitude between vehicle and minocycline. Simulations predicting amplitude distributions for both cut with minocycline or vehicle alone are far shifted to the left and overlap substantially for both peak1 and peak 3, suggesting the model cannot distinguish these two cohorts from one another (Fig. 4F1,F2).
The objective of the next experiment was to determine whether the synaptic current generated from preserved VGluT1 contacts could elicit a stretch reflex response in decerebrate rats in which neural excitability is enhanced (Fig. 5A1–A3). In Ctl rats, an average of 96.4% (range: 87.9%-100%, n = 5 rats) of all trials resulted in a detectable reflex with 10 g of background force applied and 100% with 20 g of force (n = 5 rats) in response to vibration (Fig. 5A2,D1,D2). The total force integral was then quantified with an average force of 2.12 ± 0.83 g (SD) at 10 g and 3.18 ± 0.83 g at 20 g in Ctl animals (Fig. 5C,E1,E2). Nerve injury resulted in substantial impairments in eliciting a reflex with zero trials producing any reflex at either 10 or 20 g in those animals treated with vehicle (n = 6 rats) (Fig. 5D1,D2). Similarly, in the Cut + Mino group, the majority of trials resulted in absolutely no reflex. There was, however, one animal that did produce some detectable reflex when the background force was increased to 20 g (1 of 4 rats in total), although the amount of force generated was considerably less compared with Ctl rats (0.5 vs 3.18 g) (Fig. 5D2,E2). The other 3 Cut + Mino rats did not produce any reflex at background forces of 10 or 20 g. These data further support that, although structural connectivity did improve with minocycline treatment, there was essentially no improvement in functional recovery through monosynaptic pathways.
Discussion
The objective of this study was to prevent the loss of Ia–motoneuron synaptic connections and, possibly, also their dysfunction following peripheral nerve injury. Our results demonstrated that temporary treatment with minocycline starting immediately after injury preserved and retained VGluT1+ contacts, presumably made by synapses between Ia afferents and motoneurons. Retaining synaptic contacts did not rescue function; however, neither functional connectivity (i.e., the number of motoneurons responding to Ia monosynaptic input) nor the synaptic efficacy with those motoneurons that did respond. Predictably, muscle reflexes normally produced through this pathway were not restored, even under conditions of generally heightened neural responsiveness. These findings led us to conclude that retaining central connections, although necessary, is not sufficient to restore function.
Maintenance of VGluT1 contacts following minocycline treatment
The mechanism of VGluT1 preservation remains unknown, but present findings rule out microglia proliferation and migration to the axotomized motoneuron. Historically, the microglia response after axotomy was thought to be responsible, either directly or indirectly, for the detachment, or “lifting,” of synaptic boutons from the surface of the motoneuron after injury (Blinzinger and Kreutzberg, 1968; Sumner and Sutherland, 1973), and this proposed mechanism continues to be cited (Kreutzberg, 1996; Cullheim and Thams, 2007; Kettenmann et al., 2013; Z. Chen and Trapp, 2016; for review, see Alvarez et al., 2020). Though contested (Yamada and Jinno, 2011; Campos et al., 2021), in our study we found that minocycline prevented neither the proliferation of microglia (Fendrick et al., 2005) nor the motoneuron's production of CSF1, the molecule responsible for inducing a microglia reaction (Fig. 6) (Rotterman et al., 2019). This is consistent with our recent findings, and others, demonstrating that microglia are not responsible for the initial detachment of synapses (Berg et al., 2013) but are instead involved in the specific degeneration of injured Ia afferent synapses after detachment (Rotterman et al., 2019).
The possibility remains, however, that minocycline achieved synaptic preservation via some effect on microglial function. Among its large repertoire of effects on cellular mechanisms, minocycline has been shown to increase microglial production of neurotrophins, including BDNF and others (X. Chen et al., 2012; Lu et al., 2018; Miao et al., 2018). These factors have been shown to modify synaptic connectivity and efficacy at the Ia synapse (Seebach et al., 1999; H. H. Chen et al., 2002; Mentis et al., 2010; Boyce and Mendell, 2014; Calvo et al., 2018, 2020). Furthermore, BDNF release in response to exercise has been shown to retain VGluT1 synapses after nerve injury (Krakowiak et al., 2015), and neurotrophic support from the periphery could also be responsible for VGluT1 retention after reflex conditioning (Y. Chen et al., 2010). Contrary to these similarities, minocycline failed to preserve synaptic function achieved by BDNF or NT3, demonstrating some fundamental differences in underlying mechanisms of action in the context of nerve injury and neurodegeneration.
Minocycline may also influence cell signaling pathways responsible for synapse preservation or degeneration. Recent reports posit a likely role of retrograde injury signaling cascades in initiating intracellular programs that promote synaptic stripping. Pathways of interest involve several kinase proteins, such as the p38 mitogen-activated protein kinase (p38 MAPK) and extracellular signal-regulated kinases (Rishal and Fainzilber, 2014; DeFrancesco-Lisowitz et al., 2015). Of interest to our study, minocycline has been shown to directly suppress components of this intracellular cascade in cell culture, suggesting it could act postsynaptically on the motoneuron to retain synapses (Wilkins et al., 2004; Nikodemova et al., 2006, 2007; Jordan et al., 2007). However, minocycline has many sites of action opening additional potential mechanisms requiring investigation relative to synapse preservation described here (Jordan et al., 2007; Garrido-Mesa et al., 2013a, b; Moller et al., 2016).
Failure to restore monosynaptic function
Muscle tendon vibration enables assessment of monosynaptic function between Ia's and motoneurons (Westbury, 1972). Specifically, high-frequency vibration (≥100 Hz) selectively activates only Ia afferents, thereby enabling isolation of monosynaptic potentials (Vincent et al., 2017). In untreated Ctl rats, Ia afferents supplying the MG muscle exhibit extensive functional connectivity with the homonymous motor pool, meaning that individual Ia afferents produce an EPSP with >90% of homonymous motoneurons (Mendell and Henneman, 1971; Bullinger et al., 2011). The size of aggregate EPSPs evoked in single motoneurons by concomitant activation of all homonymous Ia afferents defines synaptic efficacy. After Cut + Veh, both functional connectivity and efficacy drop below 40% of normal and minocycline did not prevent these losses. This lack of functional recovery is inconsistent with significant retention of contacts.
The occurrence and amplitude of synaptic potentials elicited by muscle stretch (i.e., SSPs) after reinnervation depend on activation the number of Ia's that regain stretch sensitivity in the periphery (Haftel et al., 2005). It is probable that reduced functional connectivity reported here arises in part from failure by some regenerated Ia's to regain stretch sensitivity (e.g., as a result of reinnervating foreign receptors) (Collins et al., 1986; Koerber et al., 1989, 1995). However, limited recovery in the periphery is insufficient to account for the emergence of nonresponder motoneurons, which actually increase in number in Cut + Mino group compared with Veh-treated animals. Together, these observations demonstrate that minocycline failed to restore function of retained central contacts deeming them “silent” (Isaac, 2003; Voronin and Cherubini, 2004).
Other examples of silent sensory synapses expressed during maturation and after nerve injury have been characterized in the literature and support our findings presented here. For example, Ia afferents form functional connections with Renshaw cells in the neonatal spinal cord early in development, but these synapses fail to proliferate through maturation as the Renshaw cell expands in size and these contacts become essentially nonfunctional (Mentis et al., 2006). Remaining synapses also become smaller in overall size, which is known to correlate with a decrease in synaptic vesicles and the density of active zones, leading to a decrease in effective synaptic current (Pierce and Mendell, 1993), a similar finding for decreased terminal size has been reported for remaining VGluT1 synapses after nerve injury (Rotterman et al., 2014). Furthermore, with genetic manipulation resulting in abnormal muscle spindle development there has been a reported 30% decrease in the number of VGluT1 synapses on motoneurons but an 80% drop in the evoked monosynaptic EPSP (Shneider et al., 2009; for review, see Mentis et al., 2010).
Candidate mechanisms underlying silent VGluT1 contacts are manifold and not determined here. One possibility is that the VGluT1 contacts do not represent structurally intact synapses. Arguing against this possibility, our earlier studies of regenerated Ia afferents demonstrate that VGluT1's punctate aggregations distribute along regenerated Ia afferents as they do on uninjured Ia afferents (Alvarez et al., 2011). Furthermore, 95.4% of all quantified VGluT1 contacts made by Ia afferents colocalized on motoneurons with the presynaptic active zone marker, bassoon, raising confidence that these are indeed synapses (Rotterman et al., 2014). However, presynaptic transmission relies on a wide array and interdependency of protein structures and signaling pathways that we did not examine for Ia–motoneuron synapses following PNI (Crawford and Mennerick, 2012; Sudhof, 2012). Neither did we investigate postsynaptic deficits that may also contribute to silencing synapses (i.e., downregulation and/or overbalanced internalization of AMPARs).
Another mechanism that might explain the failure of rescued synapses to transmit is presynaptic axonal malfunction (Hari et al., 2022). Although we found that the peripheral limb of regenerated Ia axons was competent to conduct action potentials, intraspinal axon branches were not examined. In the absence of treatment, we know that Ia axon projections retract from laminae IX (Alvarez et al., 2011; Rotterman et al., 2014). While minocycline prevents axon retraction judging from preservation of synaptic contacts, the functional status of those axons is unknown. It is possible that action potential conduction through those grossly preserved axons is compromised along their length or at branch points where conduction is susceptible to blockage. Prior work by Luscher et al. (1979) demonstrated that greater numbers of synaptic boutons (more boutons may reflect more branch points) increase the probability of propagation failure leading to a smaller EPSP within the motoneuron. These observations promote the possibility that some reconfiguration or functional deterioration of branch points prevents synaptic transmission by blocking action potential invasion. Although we cannot rule out this possibility, some additional factor(s) would be needed to explain why synaptic transmission and requisite action potential conduction are competent in 20% of motoneurons in Cut + Mino rats.
In conclusion, our results demonstrated a means of preserving synaptic connections made within the CNS by neurons axotomized in peripheral nerve injury. Although a necessary step, synapse retention by minocycline did not restore either synaptic or reflex function. These observations challenge assertions that minocycline restores synaptic function based solely on the presence of individual synaptic protein, synaptophysin, and restoration of general motor behavior (Di Liberto et al., 2018). The disparate effects of minocycline on structure and function uncovered here may provide insight into independent processes of synapse recovery and provide a critical step toward restoring normal animal behavior after nerve injury that is otherwise permanently impaired.
Footnotes
We thank the imaging core facility at the Parker H. Petit Institute for Bioengineering and Bioscience at Georgia Institute of Technology for the use of their equipment, services, and expertise; and Dr. Francisco Alvarez for providing comments on this manuscript.
The authors declare no competing financial interests.
- Correspondence should be addressed to Travis M. Rotterman at trotterman3{at}gatech.edu