Low-intensity, Kilohertz Frequency Spinal Cord Stimulation Differently Affects Excitatory and Inhibitory Neurons in the Rodent Superficial Dorsal Horn

Since 1967, spinal cord stimulation (SCS) has been used to manage chronic intractable pain of the trunk and limbs. Compared to traditional high-intensity, low-frequency (<100 Hz) SCS that is thought to produce paresthesia and pain relief by stimulating large myelinated fibers in the dorsal column (DC), low-intensity, high-frequency (10 kHz) SCS has demonstrated long-term pain relief without generation of paresthesia. To understand this paresthesia-free analgesic mechanism of 10 kHz SCS, we examined whether 10 kHz SCS at intensities below sensory thresholds would modulate spinal dorsal horn (DH) neuronal function in a neuron type-dependent manner. By using in vivo and ex vivo electrophysiological approaches, we found that low-intensity (sub-sensory threshold) 10 kHz SCS, but not 1 kHz or 5 kHz SCS, selectively activates inhibitory interneurons in the spinal DH. This study suggests that low-intensity 10 kHz SCS may inhibit pain sensory processing in the spinal DH by activating inhibitory interneurons without activating DC fibers, resulting in paresthesia-free pain relief.


INTRODUCTION
Since 1967, spinal cord stimulation (SCS), either at low (<100 Hz) or at high (>1.5 kHz) stimulation frequency, has been clinically used to manage chronic intractable pain of the trunk and limbs, but its pain-relieving mechanism(s) remain unclear (Shealy et al., 1967;Kumar et al., 2008;Foreman and Linderoth, 2012;Guan, 2012;Kapural et al., 2016;Crosby et al., 2017;Li et al., 2018). The analgesic mechanisms of lowfrequency SCS have been investigated since the early 1970's and are generally based upon the Gate Control Theory of pain (Melzack and Wall, 1965). Although its definitive primary mechanism has not been determined yet, it is thought that large diameter Ab fibers must be activated for low-frequency SCS to produce pain relief, and the degree of inhibition of dorsal horn (DH) wide-dynamic range (WDR) neurons is positively correlated with stimulation intensity (Yang et al., 2014). This proposed mechanism has some clinical validation; it is known that paresthesia (defined as any abnormal sensation, caused by Ab fiber activation, including what is often perceived by patients as tingling, buzzing, pins and needles, pressure, etc.) must be experienced by patients, and the body areas of paresthesia must overlap the patients' painful areas to result in pain relief (North et al., 1991;Oakley and Prager, 2002;Krames et al., 2008). This overlap is an ostensible indication of convergence of Ab terminal projections to hyperexcitable regions of the DH (Smits et al., 2012).
However, this ''Ab fiber activation" seems unlikely to play a mechanistically significant role in the pain relief by high-frequency (e.g., 10 kHz) SCS. In contrast to low-frequency, paresthesia-based SCS, patients do not experience stimulation-induced abnormal sensations with low-intensity 10 kHz SCS, and the overlap between Ab fiber-mediated paresthesia and painful areas targeted by 10 kHz SCS appears uncorrelated to pain relief (De Carolis et al., 2017). Recent in vivo work using rats demonstrated that low-intensity 10 kHz SCS significantly reduced the wind-up response of superficial spinothalamic tract neurons close to the SCS electrodes (McMahon et al., 2017). Since the stimulation intensities used in the study were too low to activate nearby large https://doi.org/10.1016/j.neuroscience.2019.12.031 0306-4522/Ó 2020 The Authors. Published by Elsevier Ltd on behalf of IBRO. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). myelinated Ab fibers, it remains unknown how the SCS produced such inhibitory modulation of DH neuronal activity.
In the present work, therefore, we explored potential mechanisms for how 10 kHz SCS inhibits pain sensory processing in the spinal DH, by examining its effects on excitatory and inhibitory DH neurons. Performing in vivo electrophysiological experiments, we found that 10 kHz SCS at intensities below sensory thresholds preferentially activates putative inhibitory interneurons with no effect on putative excitatory interneurons, which was corroborated by ex vivo electrophysiological recordings of identified GABAergic inhibitory neurons in the DH.

EXPERIMENTAL PROCEDURES
All experimental procedures were carried out in accordance with the National Institute of Health Guide for the Care and Use of Laboratory Animals and approved by the Animal Care Committee at Vivocore Resources Inc and the Institutional Animal Care and Use Committee at the University of Texas Medical Branch. All efforts were made to minimize the number of animals used and their suffering in this study.

Animal preparation for in vivo experiments
In vivo experiments were performed on adult male Sprague Dawley rats obtained from Charles River, Montreal. A total of 12 rats was used for this study. Under urethane anesthesia (20% in saline; 1.2 g/kg, intraperitoneally), a laminectomy was performed to expose L4-S1 segments of the spinal cord. The rat was then placed in a stereotaxic frame, and its vertebrae were clamped above and below the recording site to immobilize the spinal cord. The left hind paw was immobilized in plasticine with the plantar surface facing upwards to receive mechanical stimuli. Rectal temperature was kept at 37°C using a feedbackcontrolled heating pad (TR-200; Fine Science Tools, Taunton, MA) throughout the experiment.

Single unit extracellular recordings
A 4-shank probe with a total of 16 recording microelectrodes (A4x4-3mm-50-125-177-A16, NeuroNexus) was implanted at the L5 spinal level in the left DH. The array was oriented parallel to the rostrocaudal axis so that each electrode was at roughly the same mediolateral position. Depth of electrode insertion was monitored to estimate electrode position within presumed Rexed lamina(e). Electrode shank tips were lowered 354 ± 87 lm (mean ± standard deviation) below the dorsal surface. Since recording sites are spaced at 50 lm intervals up each shank of the electrode, we estimated the depth of recorded neurons to be positioned within lamina I or II. Neurons that responded to limb displacement, indicating that they received proprioceptive input, were excluded. Measured extracellular signals of neuronal activity were amplified, band-pass filtered between 500 Hz and 10 kHz, digitized at 40 kHz with OmniPlex Data Acquisition System (Plexon, Dallas, TX, USA), and stored with stimulus markers on disk. Single units were isolated using Offline Sorter v3 software (Plexon) and were analyzed with NeuroExplorer 4 (Plexon).

Mechanical stimulation
Cutaneous receptive fields were identified on the basis of observed spiking evoked by mechanical stimuli applied using a brush or von Frey (VF) filaments to the glabrous skin of the left hind paw. By using weak search stimuli (to avoid peripheral sensitization), we specifically targeted neurons receiving low-threshold afferent input. Each stimulus comprised 10 one-second-long applications of the brush or VF filament (2, 4, 6, 8, and 10 g) repeated at 2-s intervals. Each stimulus was applied twice onto different locations within the receptive field for each test condition.

SCS
SCS was applied via a miniature in-line quadripolar electrode array (420 lm diameter, 1 mm-long contact leads interspaced with 1.5 mm-long nonconductive polymers), positioned epidurally over the L5-L6 dorsal spinal segments (innervating the left hind paw). The SCS electrode was medially positioned within 1-2 mm of the recording microelectrode probe. Two contacts on the miniature SCS lead were used to provide bipolar stimulation, delivered from a modified version of the same trial stimulator used clinically for 10 kHz SCS therapy (Trial Stimulator, Nevro, Redwood City, USA). Stimulation was delivered as a continuous train of symmetric, biphasic, 30 ls-per-phase pulses, at a variety of kHz frequencies and pulse amplitudes (see 'Protocol'). Initially, to determine the motor threshold (MT), stimulation at each tested frequency was gradually increased until motor twitch (most often in the hindpaw, but also observed in the paraspinal musculature/trunk). To allow for clear visualization of muscle activity, the stimulation was duty cycled at 300ms ON, and 2-s OFF.

General protocol
After insertion of the recording microelectrode probe, mechanical stimuli were applied to the hindpaw to determine neurons responsive to innocuous brush stimuli and strong VF (10 g). Next, SCS was applied at 1 kHz, 5 kHz, and 10 kHz, at intensities of 10% and 30% of MT, which are reportedly below the sensory threshold (Song et al., 2014), as well as at 60% and 90% MT, while neuronal response was recorded ( Fig. 1). SCS was then administered for periods of 40 s at each stimulation setting while DH neuronal response was recorded. SCS was left OFF for at least 5 min between measurement periods.

Patch-clamp recording of DH neurons in ex vivo spinal cord slices
Transgenic mice expressing green fluorescent protein (GFP) in GABAergic neurons (FVB-Tg(GadGFP) 45704Swn/J, 4-6 weeks old, both sexes, produced inhouse from breeders purchased from The Jackson Laboratory, Bar Harbor, ME) were used for ex vivo experiments. A total of five mice (three males, two females) was used for this study.
For 10 kHz SCS ex vivo, a miniature SCS electrode array was placed dorsal to the DC ( Fig. 4) with its first lead being 0.1-mm distanced from the DC by an attached nonconductive polymer spacer. SCS, 30 lsper-phase biphasic pulses at 10 kHz, was delivered for 300 ms at 2-s intervals and stimulation intensities ranging from 2 mA to 8 mA (2-mA step).

Data analysis
Data are expressed as mean ± standard error of the mean (SEM) with n, the number of cells and N, the number of animals. For the in vivo experiments, to quantify and characterize the spiking evoked by mechanical stimulation of the hindpaw and by SCS, we measured the mean firing rate of individual neurons during each stimulus, and subtracted the spontaneous firing rate measured from the 10-s epoch preceding each stimulus. As we sampled multiple neurons from a single animal (i.e., nested design), we performed multilevel analysis (Aarts et al., 2014) using a linear mixed model. Statistically analyzing the difference in cellular location or spontaneous firing rates between two cell types, we considered cell type a fixed categorical factor and individual rats a higher level random factor. When firing rates during SCS were analyzed, SCS intensity (% MT), SCS frequency, and their interaction were considered fixed categorical factors, while rats and the cells (nested within rats) were regarded as random categorical factors; the SCS intensity was a repeated variable.
In ex vivo experiments, the mean number of APs discharged during the 300-ms SCS was measured after applying an offline 1 kHz low-pass filter. Similar to the statistical analysis used for in vivo results, we performed multilevel analysis: we regarded SCS intensity (mA), neuron types (e.g., GFP(+/À) or AP firing-types), and their interaction as fixed categorical factors, and mice and the neurons (nested within mice) as random categorical factors; the SCS intensity was a repeated variable. In case that no significant interaction between SCS intensity and neuron types was detected (results for Fig. 5), the dataset was re-analyzed using a statistical model without interaction.
When statistically analyzing firing rates or AP numbers, despite our understanding that the F-test is robust in terms of Type I error even if the normality assumption is violated (Blanca et al., 2017), we took a conservative approach to meet both the normality and equal variance assumptions by transforming our data of AP firing (log e transformation; 0.01 was added to raw values to make all >0 before the transformation). Differences between multiple groups with Bonferroni-adjusted p 0.05 (e.g., unadjusted p 0.016 for three comparisons in Figs. 3 and 5) were considered significant.

RESULTS
Spinal DH neurons recorded in vivo fall into two distinct groups: adapting and non-adapting neurons Examining the effect of SCS on DH excitatory and inhibitory neurons separately is crucial, as simultaneous activation of these neurons might cancel each other's influence on pain processing and thus produce no apparent aggregate outcome. Therefore, as in Fig. 2  Previous reports have suggested that SCS administered at intensities of 40-60% of motor threshold (MT) correlates to the clinical paresthesia onset threshold (Song et al., 2014). By testing the entire range of SCS intensity over multiple frequencies, the effects of frequency and amplitude can be independently assessed.
by their AP firing pattern to 10-g VF filament stimulation, depth from the surface, and existence of spontaneous firing. We first identified the neuronal type as 'non-adapting' and 'adapting' cells based on their responses to VF (10 g) stimulation ( Fig. 2A). Next, we investigated the distribution of recording depths for all adapting (n = 57, N = 10) and non-adapting (n = 30, N = 12) neurons (Fig. 2B). Average depth for non-adapting cells (267 ± 14 lm) was greater than that for adapting cells (149 ± 11 lm, F(1,86.9) = 46.1, p 0.001). Also, the proportion of adapting (65.5%) and nonadapting cells (34.5%) was about expected 2:1 ratio for excitatory and inhibitory interneurons in superficial laminae (Polgar et al., 2003;Todd, 2010Todd, , 2017 (Fig. 2C). The spontaneous firing rate of non-adapting cells (6.75 ± 0.99, n = 30, N = 12) was higher than that of adapting cells (0.41 ± 0.08, n = 57, N = 10; F (1,86.9) = 60.8, p 0.001 using log e transformed data). In addition, more non-adapting cells (93%, 28 out of 30, N = 12) spontaneously fired APs than adapting cells (61%, 35 out of 57, p 0.005 by Fisher's exact test). Based on the depth of recordings, we determined that the measured neurons were located within the superficial DH. Collectively, these data suggest that we sampled two distinct populations of superficial DH neurons, presumably excitatory (adapting cells) and inhibitory (nonadapting cells) subtypes as shown previously (Lee and Prescott, 2014;Lee et al., 2019). Furthermore, the relative distribution of these two cell types across the DH laminae (Fig. 2B) is consistent with the increasing proportion of inhibitory neurons from lamina I to III (Polgar et al., 2013).
When we applied 10 kHz SCS at 60% and 90% of MT intensities, which are likely above sensory threshold, adapting cells (n = 18, N = 5) fired 6.3 ± 2.1 and 12.9 ± 3.8 spikes/s at 60% and 90% MT after subtracting baseline firing rates, respectively; and non-adapting cells (n = 14, N = 5) 17.6 ± 5.9 and 36.4 ± 9.1 spikes/ s. Together, these data suggest that 10 kHz SCS, but not 1 kHz or 5 kHz SCS, is able to activate nonadapting, putative inhibitory interneurons at sub-sensory threshold levels of stimulation intensity (i.e., 30% MT) ; not all neurons were tested across all stimulation intensities) fired more APs than GFP(À) neurons that were non-tonic-firing upon depolarizing current injections (n = 7-11, N = 3). ** p 0.01 vs. GFP (À) neurons following multilevel analysis using log e transformed data. ; not all neurons were tested across all stimulation intensities) were more readily activated by the SCS than delayed-firing neurons (n = 6-9, N = 3). No significant difference in AP numbers was detected between tonic-and phasic-firing neurons across the range of 4-8 mA SCS intensities when log e transformed data were used for statistical analysis. *Adjusted p 0.05 vs. delayed-firing neurons by Bonferroni multiple comparison test following multilevel analysis using log e transformed data. without activating adapting, presumed excitatory interneurons. However, when SCS intensity is at suprasensory threshold, such differential activation between the two neuron types is no longer attained.

DISCUSSION
Traditional low-frequency SCS has been assumed to rely upon the activation of the large myelinated axon projections of Ab fibers in the dorsal column (DC) (Oakley and Prager, 2002). Clinically, this results in paresthesia, likely from the SCS-induced activity in the DC pathway travelling orthodromically through the DC nuclei, the thalamus, and ultimately driving perception upon reaching the cortex. Still, the majority of preclinical work suggests that the pain-relieving effects of SCS are segmental, due to the APs delivered to the DH of hyperexcitable spinal segments via collateral projections of those same DC fibers (Yowtak et al., 2011). Preclinical studies have shown that the higher the stimulation amplitude, the more apparent the inhibitory effects, while much less effect is observed at stimulation amplitudes believed to be below the DC fiber activation threshold (Shechter et al., 2013;Yang et al., 2014). The translational relationship between preclinical stimulation intensity levels, clinical paresthesia and MT has not, to our knowledge, been determined in a systematic or definitive way. Like other investigators, we used the MT as an upper limit on clinically relevant stimulation intensity (Russo et al., 2018). Previous reports suggested that SCS administered at intensities of 40-60% of MT correlate to the clinical paresthesia onset threshold, likely from the activation of DC fibers (Song et al., 2014). Thus, we assumed that stimulation intensities of no greater than 30% of MT would translate to 'paresthesia-free' clinical settings not activating DC fibers. In this study, we observed that 10 kHz SCS is capable of selectively driving inhibitory interneurons in the superficial DH at 30% of MT, whereas it activates both excitatory and inhibitory DH neurons at 60% and 90% MT. Therefore, low-intensity 10 kHz SCS may be able to modulate inhibitory interneurons (and thus relieve pain) without triggering spinal nociceptive circuit activation and interrupting somatosensory mechanisms mediated by excitatory neurons in the spinal DH. We acknowledge that classifying excitatory and inhibitory cells in vivo based on extracellular recordings of firing, locational, and proportional properties as described in this study is not definitive although such classification system has been previously utilized (Lee and Prescott, 2014;Lee et al., 2019). Therefore, as a corroborative approach, we identified inhibitory and excitatory neurons in ex vivo spinal cord slices based on GFP-expression and tonic-firing pattern and compared their responses to a range of SCS intensities with those of their counterparts. The ex vivo study showed that stimulation-response profiles of identified GFP(+) GABAergic neurons or tonic-firing neurons were leftshifted, compared to those of GFP(À) neurons or phasic/delayed-firing neurons, supporting the in vivo findings that low-intensity 10 kHz SCS preferentially activates spinal inhibitory neurons over their excitatory counterparts. The mechanistic basis of this neuron typedependent effect of low-intensity 10 kHz SCS is currently unclear, and investigation of such mechanisms is beyond the scope of present work. It would be of interest in future study to examine the nature of SCS-responsive apparatus in the DH neurons and functional/expressional difference of the apparatus between inhibitory and excitatory neurons. As it is also possible that activation of excitatory neurons by low-intensity 10 kHz SCS is hampered by simultaneous, robust activation of inhibitory interneurons in the DH neural network, it will be necessary in future study to assess the contribution of local inhibitory neurotransmission to the neuron type-dependent effect of lowintensity 10 kHz SCS.
We found frequency to be a critical factor for selective activation of DH neurons. At low intensities, SCS using 10 kHz, not 1 kHz or 5 kHz, could solely activate nonadapting inhibitory neurons. It may be that certain channels that predominate on DH neuronal membranes are sensitive to the repetition rate of pulse delivery. Temporal summation is a well-known phenomenon of synaptic transmission, where more rapid release of neurotransmitter increases the likelihood of post-synaptic cell firing (Kandel et al., 2000). Non-synaptic neural membranes also appear to demonstrate a form of temporal summation, where the effect is mediated by nonsynaptic mechanisms. Bromm (1975) demonstrated that frequencies greater than approximately 4 kHz could repetitively activate sodium channels to the point of AP generation. Reilly et al. (1985) used computational modeling of mammalian axons to demonstrate that reduced stimulation pulse amplitudes could still generate APs if >5 kHz sinusoidal frequencies were used. Similarly, Schoen and Fromherz (2008) used weak capacitive currents administered at >1.5 kHz to repetitively activate Nav1.4 channels in rat neurons, and ultimately generate APs. In our experiments, if SCS-induced AP firings below sensory threshold intensity were triggered by temporal summation of cation channel-mediated small depolarization, the summated depolarization must be able to reach AP threshold only in inhibitory neurons at 10 kHz.
In performing our in vivo experiments, we placed recording electrodes only in the local DH. We did not measure more rostral activity (e.g., cord dorsum potentials, gracile nuclei activity, etc.), so we do not know the actual stimulation amplitude range that activates DC fibers. Thus, we do not know precisely which observed effects may be influenced by synaptic drive from activated DC collaterals. The same limitation also applies to our data obtained from ex vivo spinal cord slices. Still, our assumption that stimulation amplitudes 30% of MT would not activate DC axons is in keeping with most prior preclinical literature. In addition, it is noteworthy that in ex vivo spinal cord slices prepared from the same transgenic mouse strain in a similar fashion, Ab fiber input-evoked activation of GABAergic neurons was seldom observed presumably due to damages in the incoming Ab fibers along their DC passage (Schoffnegger et al., 2006;Daniele and MacDermott, 2009). Therefore, it could be that in our ex vivo experiments, contribution of DC collaterals to SCS-induced GABAergic neuronal activation is physically limited.
Without measurement of neuronal response to antidromic stimulation at supraspinal sites in our experiments, we cannot determine whether the DH neurons we measured were local interneurons or projection neurons. We presume that, given that projection neurons are absent in lamina II and represent only up to 5% of neurons in the lumbar segments of lamina I in the rat (Spike et al., 2003), it is likely that the majority of neurons we studied were local interneurons, about 1/3 of which are inhibitory and the remainder are excitatory (Todd, 2017).
Our experiments were in anesthetized animals, and stimulation was only applied for seconds-minutes. Thus, what we measured are acute, rapid onset electrophysiological effects. It has been observed clinically that pain relief onset from 10 kHz SCS can typically be appreciated hours to days after starting SCS. Thus, we speculate that our measurements relate to the start of a cascade of changes that may ultimately result in perceivable pain relief. Study of these effects in longer-term models, including behavioral assays would be valuable to truly assess mechanisms for pain relief by low-intensity 10 kHz SCS.
In conclusion, by using in vivo and ex vivo electrophysiological approaches, we found that lowintensity 10 kHz SCS, but not 1 kHz or 5 kHz SCS, selectively activates inhibitory interneurons in the spinal DH. This study suggests that low-intensity 10 kHz SCS may inhibit pain sensory processing in the spinal DH by activating inhibitory interneurons without activating DC fibers, ultimately resulting in paresthesia-free pain relief.