Electrophysiological and Morphological Features of Rebound Depolarization Characterized Interneurons in Rat Superficial Spinal Dorsal Horn

Substantia gelatinosa (SG) neurons, which are located in the spinal dorsal horn (lamina II), have been identified as the “central gate” for the transmission and modulation of nociceptive information. Rebound depolarization (RD), a biophysical property mediated by membrane hyperpolarization that is frequently recorded in the central nervous system, contributes to shaping neuronal intrinsic excitability and, in turn, contributes to neuronal output and network function. However, the electrophysiological and morphological properties of SG neurons exhibiting RD remain unclarified. In this study, whole-cell patch-clamp recordings were performed on SG neurons from parasagittal spinal cord slices. RD was detected in 44.44% (84 out of 189) of the SG neurons recorded. We found that RD-expressing neurons had more depolarized resting membrane potentials, more hyperpolarized action potential (AP) thresholds, higher AP amplitudes, shorter AP durations, and higher spike frequencies in response to depolarizing current injection than neurons without RD. Based on their firing patterns and morphological characteristics, we propose that most of the SG neurons with RD mainly displayed tonic firing (69.05%) and corresponded to islet cell morphology (58.82%). Meanwhile, subthreshold currents, including the hyperpolarization-activated cation current (Ih) and T-type calcium current (IT), were identified in SG neurons with RD. Blockage of Ih delayed the onset of the first spike in RD, while abolishment of IT significantly blunted the amplitude of RD. Regarding synaptic inputs, SG neurons with RD showed lower frequencies in both spontaneous and miniature excitatory synaptic currents. Furthermore, RD-expressing neurons received either Aδ- or C-afferent-mediated monosynaptic and polysynaptic inputs. However, RD-lacking neurons received afferents from monosynaptic and polysynaptic Aδ fibers and predominantly polysynaptic C-fibers. These findings demonstrate that SG neurons with RD have a specific cell-type distribution, and may differentially process somatosensory information compared to those without RD.


INTRODUCTION
Rebound depolarization (RD) is a transient membrane depolarization (sometimes accompanied by a series of spikes) following hyperpolarizing pulses. It has been observed in various brain regions, which include the hippocampus (Surges et al., 2006), auditory midbrain (Sun et al., 2020), medial prefrontal cortex (Kurowski et al., 2018), thalamus (Wang et al., 2016;Zhu et al., 2018), and deep spinal dorsal horn (Rivera-Arconada and Lopez-Garcia, 2015), among others. RD substantially relies on channels with hyperpolarization-dependent activation or de-inactivation features. Hyperpolarization-activated cation current (I h ), a mixed inward current consisting of sodium and potassium ions, mediated by hyperpolarizationactivated cyclic nucleotide-gated (HCN) channels has been confirmed to be able to regulate RD. In addition, T-type calcium channel-induced current (I T ) is another vital ionic conductance supporting RD generation (Sangrey and Jaeger, 2010;Engbers et al., 2011;Duan et al., 2018). Functionally, RD has been proposed as an inhibition-excitation converter transforming inhibitory inputs into excitatory signals (Sanchez-Vives and McCormick, 2000;Tadayonnejad et al., 2009;Pedroarena, 2010).
Lamina II of the spinal dorsal horn, namely substantia gelatinosa (SG), is an indisputably important component of the pain pathway. The SG is referred to as the "central gate" since it converses inputs from the primary afferents, local circuit interneurons, and endogenous descending tracts, and may therefore be essential in transmitting and modulating nociceptive information from the periphery (Todd, 2010;Duan et al., 2018). Alterations in SG neuronal excitability have been implicated as catalysts for the development and maintenance of pathological pain Kuner, 2010;Feng et al., 2019). The SG is composed of excitatory and inhibitory interneurons with substantial electrophysiological and morphological heterogeneity (Grudt and Perl, 2002;Maxwell et al., 2007;Yasaka et al., 2010;Graham and Hughes, 2020). RD is a striking biophysical property of SG neurons (Tadros et al., 2012;Hu et al., 2016). Nevertheless, whether SG neurons with RD display distinct morphological features, intrinsic electrophysiological properties, and input of afferent fibers has not yet been well studied. In addition, the ionic basis for RD in SG neurons also remains to be elucidated.
Here, we used the whole-cell patch-clamp technique to record passive and active membrane properties from SG neurons, which were further categorized to the neurons with RD and without RD. Interestingly, we found that a majority of the SG neurons with RD showed tonic firing, and a post hoc morphological study confirmed that most of the RD-expressing neurons exhibited islet morphology. Besides, RD neurons also presented significant differences in intrinsic neural excitability, as well as spontaneous, miniature, and evoked excitatory synaptic transmission. Additionally, we confirmed that I h and I T are vital ionic contributions to RD responses in SG neurons. Therefore, our results may provide new insights for unraveling the role of RD in pain processing.

Animals
Sprague-Dawley (SD) rats (4-6 weeks old) of both sexes obtained from the Animal Center of Nanchang University were used. Rats were housed in same-sex groupings (4-5 animals per cage) in an air-conditioned room maintained at 24 ± 1 • C and 50-60% humidity, under a 12:12 light/dark cycle (lights on at 7 a.m.). They had ad libitum access to food and water. All animal procedures were performed according to methods approved by the Institutional Animal Care and Use Committee of Nanchang University. Maximal efforts were made to minimize animal pain or discomfort and the number of animals used.
Only neurons with a resting membrane potential (RMP) more negative than −45 mV and showing overshooting action potentials (APs) (i.e., exceeding 0 mV) were included for this study. The series resistance was typically between 10 and 30 M , and cells in which the series resistance changed by more than 20% were discarded from further analysis. No correction for liquid junction potential (approximate 15 mV) was made in this study. All data analyses were performed using Clampfit (Molecular Devices, CA, United States) and Mini Analysis software (Synaptosoft Inc., GA, United States).
Membrane capacitance (Cm) was estimated from the area under the transient capacitive current evoked with a 5-mV depolarizing pulse by Patchmaster software in real-time automatically. Neuronal RMP was recorded within 20 s after the break-in. A hyperpolarizing voltage step was used to estimate the input resistance (Rin) of the tested SG neuron, while negative current pulses (−120 and −140 pA, 1 s) were applied to generate RD responses. The firing pattern of each neuron was determined with a series of depolarizing current pulses (20-140 pA in 20 pA increments) of 1-s duration in the current-clamp mode. The spike adaptation index was obtained by dividing the average of the first two interspike intervals (ISIs) by that of the last two ISIs (Ha et al., 2016). A subthreshold current was evoked by a series of hyperpolarizing voltage steps from −50 to −130 mV in 10 mV decrements (duration 1 s) at a −50 mV holding potential in the voltage-clamp mode.
To investigate the primary afferent inputs, neurons were voltage-clamped at −70 mV. DR-evoked excitatory postsynaptic currents (eEPSCs) were initiated by a constant current source of a pulse (duration 0.1 ms) at a frequency of 0.05 Hz delivered through a suction electrode. The stimulation intensities were set at 50 and 100 µA for the activation of Aδ fibers, 500 µA and 1 mA for the activation of C fibers using a stimulator (Master 8, AMP Instruments Ltd., Israel) and a stimulus isolator (ISO-Flex, AMP Instruments Ltd., Israel). Neurons without an evident monosynaptic response at 1 mA were subsequently stimulated at 3 and/or 5 mA. The Aδ or C-afferent-mediated responses evoked by DR stimulation were distinguished based on the stimulus intensity and the conduction velocity of afferent fibers (<0.8 m/s for C fibers and >1.0 m/s for Aδ fibers). Evoked-EPSCs were judged to be monosynaptic if there were no failures during subsequent repeated stimulation (20 times at 2 Hz for Aδ and 1 Hz for C fibers) and if their latency remained constant in repetitive trials as reported previously (Cui et al., 2016;Iwagaki et al., 2016).

Post hoc Morphological Identification
For morphological identification, an internal pipette solution containing 0.05% neurobiotin 488 was used. After maintaining the stable whole-cell patch-clamp configuration for at least 20 min, the electrode was gently withdrawn from the targeted neuron, and the slice was then fixed at RT for 1 h and then at 4 • C overnight in 4% paraformaldehyde in 0.1 M PB (pH 7.4). Following fixation, slices were rinsed in PBS three times and treated with 50% ethanol for 30 min. After rinsing in PBS, the slices were mounted onto slides with a non-fluorescing mounting medium. Neurobiotin-filled neurons were identified and reconstructed under 20× magnification, 1.0-1.5 zoom, and 1.5 µm stack using a confocal microscope (LSM 700, Zeiss, Germany). Neurons were morphologically grouped on the basis of the following parameters regarding their dendritic arborizations (RC: rostro-caudal extent of dentrites, DV: dorsalventral expansion of dendrites, SR, SC, SD, SV: dendrites spread from center of the soma to rostral, caudal, dorsal and ventral limit, respectively) (Grudt and Perl, 2002;Yasaka et al., 2007). As long as the value of RC/DV exceeded 3.5, cells could be classified as an islet or a central cell. Within this group, islet cells exhibited abundant dendrites elongated in the RC dimension (>400 µm), whilst central neurons possessed strikingly shorter RC extents (<400 µm) of their dendric arbors. Neurons with RC/DV less than 3.5 were identified as vertical or radial cells. Vertical neurons generally had a predominantly ventrally oriented dendritic geometry with SV/SD > 3.5. If the ratio of SV/SD was less than 3.5, the neurons were defined as a radial neuron. Neurons differed from the four morphological categories were identified as unclassified cell.

Statistical Analysis
Statistical data analysis was conducted with GraphPad Prism 7 (GraphPad Software, La Jolla, CA, United States). For data displaying normal distribution and homogeneity of variance indicated by the Shapiro-Wilk normality test and Levene test respectively, ANOVA or t-test was used as appropriate. Numerical data are presented as mean ± SEM. Otherwise, the Mann Whitney Wilcoxon test or Wilcoxon matchedpairs signed-rank test was used for non-parametric statistical comparisons. Data of AP number were analyzed by twoway ANOVA followed by the Bonferroni post hoc test. The Chi-squared test or Fisher's exact test was applied to analyze the difference between groups regarding the proportions of the firing patterns, subthreshold currents, and morphological categories. Differences were considered significant when p < 0.05.

Intrinsic Passive and Active Membrane Properties
As shown previously, the discharge patterns of SG neurons were highly heterogeneous, including tonic firing, phasic firing, initial firing, single firing, delayed firing, and gap firing (Ruscheweyh and Sandkuhler, 2002;Zhu et al., 2019). We found that the distribution of the discharge patterns differed between the two groups (p < 0.001). All six firing categories were encountered in SG neurons lacking RD: tonic (43.81%), phasic (10.47%), initial (12.38%), single (10.47%), delayed (19.05%), and gap (3.81%) (Figures 1B,C). However, delayed firing and gap firing were not identified in SG neurons with RD, of which 69.05% displayed a tonic firing. The proportions of phasic, initial, and single firing in SG neurons with RD were 7.14, 20.24, and 3.57%, respectively (Figures 1B,C).
The results of the passive and active membrane properties of SG neurons with or without RD are summarized in Figure 1D and Table 1. There was no significant difference in Cm. However, the RMP, Rin, and properties of AP of the two populations were significantly different. Neurons with RD had a more depolarized RMP versus those without RD (−54.05 ± 0.58 mV vs −59.18 ± 0.77 mV, p < 0.001). In addition, RD-expressing neurons showed a smaller Rin (0.59 ± 0.05 G vs 0.71 ± 0.05 G , p = 0.013), a more hyperpolarized AP threshold (−31.53 ± 0.53 mV vs −28.59 ± 0.54 mV, p < 0.001), a shorter AP half-width (0.87 ± 0.05 ms vs 1.14 ± 0.05 ms, p < 0.001), a smaller potential difference between AP threshold and RMP (22.52 ± 0.71 mV vs 30.59 ± 0.84 mV, p < 0.001), and a larger AP amplitude (78.13 ± 1.09 mV vs 70.52 ± 1.17 mV, p < 0.001) compared to SG neurons without RD. These results suggest that RD-expressing SG neurons exhibit higher neuronal membrane excitability.
As tonic-firing neuron has repeated APs at relatively regular rates and is the highest proportion in both types of neurons, we next compared the responses of tonic-firing neurons to depolarized current injections (Figure 2A). Tonic-firing neurons from the group with RD showed significantly higher spike frequencies compared to those lacking RD (Figure 2B), suggesting that the former was more excitable. Furthermore, we found that the spike adaptation index was smaller in neurons with RD than those without RD (0.45 ± 0.03 vs 0.58 ± 0.05, p = 0.015) ( Figure 2C).

Subthreshold Currents
Based on previous studies, three subthreshold currents, I h , I T , and A-type currents (I A ) were identified in SG neurons (Figure 3Aa; Tadros et al., 2012). In this study, we found that the expression and distribution of the subthreshold currents in RD-lacking SG neurons varied markedly from those in neurons with RD (p < 0.001) (Figure 3Ab). Overall, 76.00 and 71.11% of SG neurons with RD expressed I h and I T , respectively. I A was rarely encountered (2.22%) in this population. However, in neurons without RD, the prevalence rates of I h and I T dropped to a relatively low level (30.61 and 20.41%, respectively), while I A became the dominant subthreshold current accounting for 35.71%. Measurement of the amplitude and time constant of subthreshold currents showed that no significant differences were observed regarding the activation and inactivation time constants of I T between RD-positive and RD-negative neurons. However, RD-expressing neurons possessed faster time constant of I h , larger current amplitude and density of both I h and I T (Supplementary Figure 3 and Tables 1, 2).
Because I h along with I T has been proposed to be the major ionic basis responsible for RD in neurons of the deep cerebellar nuclei, periventricular preoptic area, etc. (Engbers et al., 2011;Zhang et al., 2013), we next studied the effects of I h and I T on RD in SG neurons. Bath application of ZD7288 (10 µM), a specific HCN channel blocker, increased RD latency from 42.64 ± 7.93 ms to 117.17 ± 20.11 ms (p = 0.011) but did not affect the number of spikes generated (3.63 ± 1.25 vs 3.13 ± 0.77, p = 0.750) ( Figure 3B). In addition, the blockage of I T by TTA-P2 (10 µM) markedly decreased RD spiking number from 4.90 ± 1.06 to 1.50 ± 0.27 (p = 0.002), and increased RD latency from 58.59 ± 10.07 ms to 97.79 ± 23.47 ms (p = 0.034) ( Figure 3C).
In attempt to better addressing the influence of I h or I T on RD properties, TTX (0.5 µM) was applied to the bath solution to eliminate rebound discharge. In the presence of TTX, blockage of I h led to a substantial delay in latency (66.71 ± 13.42 ms to 122.55 ± 14.38 ms, p = 0.008) without alterations of the RD amplitude ( Figure 3D). In addition, the blocking effect of TTA-P2 almost eliminated RD. As illustrated in Figure 3E, the RD amplitude dropped by approximately 64.97% ± 8.03% (p = 0.016), and the RD latency raised from 64.67 ± 8.55 ms  to 90.05 ± 13.32 ms (p = 0.020) with bath application of TTA-P2( Figure 3E). Taken together, these results suggest that RD in SG neurons is mainly mediated by I T , while I h facilitate its onset.

Properties of Spontaneous and Miniature Excitatory Postsynaptic Currents
Next, we examined whether SG neurons with RD exhibited any difference in the excitatory synaptic communication compared to those lacking RD. As shown in Figure 4, the frequencies of sEPSCs recorded from neurons with RD tended to be approximately 56% of those without RD (2.89 ± 0.43 Hz vs 5.10 ± 0.60 Hz, p = 0.010), while the difference of amplitudes of sEPSCs was not significant (19.02 ± 1.57 pA vs 22.58 ± 1.91 pA, p = 0.182) (Figures 4A-D). SG neurons with and without RD did not differ in mEPSCs amplitudes (19.02 ± 1.91 pA vs 22.50 ± 1.56 pA, p = 0.169), but neurons with RD had considerably lower mEPSCs frequencies compared with neurons lacking RD (2.36 ± 0.37 Hz vs 3.91 ± 0.52 Hz, p = 0.032) (Figures 4E-H).

Morphological Characterization
It has been well established that SG neurons exhibit complex and diverse morphologies. In this study, 46 recordings were attempted to recover morphology among which 33 were finally morphologically identified, and examples were illustrated in Figure 6A. All five morphological categories (islet, radial, central, vertical, and unclassified) were observed in the examined neurons, and the morphology distributions between the two groups were statistically similar ( Figure 6B) (p = 0.332). Islet cell was clearly the type with the highest prevalence encountered in both RD-expressing and -lacking SG neurons (58.82 and 43.75%, respectively). Given the crucial role of dendritic dimensions in morphological classification, we further conducted morphometric analysis by measuring the extents of the dendritic trees of neurons from the two groups. The mean rostro-caudal dendritic lengths for neurons with or without RD were 513.78 ± 56.95 µm and 361.88 ± 31.34 µm, respectively, and these were significantly different (p = 0.029). A similar comparison indicated that dendritic extents in the dorsal-ventral axis did not differ significantly between the two subpopulations (190.81 ± 24.12 µm vs 156.85 ± 21.62 µm, p = 0.305) ( Table 3).

DISCUSSION
The aim of the present descriptive study is to define the electrophysiological and morphological characteristics of SG neurons with RD, and several key conclusions can be drawn from this study. First, compared to RD-lacking SG neurons, RD-expressing neurons showed distinctive electrophysiological signatures including more depolarized RMP, smaller Rin, more hyperpolarized AP thresholds, and increased spike frequencies, representing a higher membrane excitability. Meanwhile, SG neurons with RD predominantly exhibit a tonic firing pattern. Second, we found that I T was a vital ionic mechanism basis responsible for the generation of RD in SG, while the onset of RD was regulated by I h . Third, neurons with RD receive monosynaptic as well as polysynaptic excitatory inputs from both Aδ and C afferents, and they showed lower frequencies regarding sEPSCs together with mEPSCs than neurons without RD. Finally, we found anatomical diversity in RD-expressing neurons, with islet cells constituting half of the population.

SG Neurons With RD Exhibit Distinct Intrinsic Neuronal Characteristics
Recent studies have linked I h /I T to neuronal electrical properties. For instance, intracellular recording from primary sensory neurons found that I h blockage by ZD7288 changed RMP toward the negative direction, prolonged AP duration, and resulted in diminished effects on AP frequency (Hogan and Poroli, 2008). In our prior investigations, we have noted that impeding I h by using drugs interfering with the function of HCN channels robustly decreased firing rate, delayed RD latency, and lower post-hyperpolarization spike frequency in SG neurons, indicating an excitatory influence of I h (Liu et al., 2015;Hu et al., 2016). Likewise, a number of groups have suggested the role of I T in establishing intrinsic excitability. We have shown previously that there were remarkable differences in properties of SG neurons with or without I T . For I T -expressing neurons, they displayed a more hyperpolarized AP threshold, a smaller difference of AP threshold and RMP, as well as elevated firing frequency (Wu et al., 2018). Meanwhile, a recent study also showed that AP amplitude was decreased in Cav3.2-ablated SG neurons (Candelas et al., 2019). Thus, as suggested by identifying I h /I T in RD-expressing SG neurons, the intrinsic excitability of this population might involve a contribution from these two subthreshold currents. Previous work on deep dorsal horn neurons, which are also of paramount importance for integrating somatosensory information, found that AP threshold was more hyperpolarized and the difference of AP threshold and RMP tended to be smaller in neurons with a high-amplitude RD; while RMP, Rin and AP amplitude in RD-expressing deep dorsal horn neurons were comparable to those lacking it (Rivera-Arconada and Lopez-Garcia, 2015). Here, marked differences in active and passive membrane properties were identified in SG neurons with or without RD. In line with the aforementioned study from the mouse deep dorsal horn, we found neurons with RD in superficial spinal dorsal horn exhibited significantly lower AP threshold and smaller difference of AP threshold and RMP. However, we also observed a significant difference in RMP, Rin, AP duration along with AP amplitude between the two groups. The differences in our observations raise the possibility that the intrinsic excitability of different spinal cord neurons may correlate with RD in unique ways. Besides, we found tonic-firing neurons with RD generated considerably increased spike frequencies in response to defined depolarizing currents. Together, these data indicate that SG neurons with RD display higher membrane excitability.
Additional support for the relationship between RD and neuron excitability came from the result of the spike adaption index. Diverse degrees of firing adaptation have been observed in SG neurons (Ruscheweyh and Sandkuhler, 2002;Olschewski et al., 2009;Melnick, 2011). An interesting study analyzed the major factors determining the appearance of spike frequency adaption in SG neurons and found that lower Na + conductance was critical for the generation of adapting firing (Melnick et al., 2004). Given that RD-expressing neurons exhibited stronger adaptation than RD-lacking neurons, the present study implies that the expression or function of Na + channels may differ between these two populations. Moreover, due to neurons with strong adaptation have been suggested to be nociceptive neurons with specific cutaneous afferent input (Lopez-Garcia and King, 1994), it is reasonable to assume

SG Neurons With RD Show Specific Cellular Distributions
The prevalence of rebound spikes in randomly sampled mouse SG neurons was around 30% according to previous reports (Tadros et al., 2012;Candelas et al., 2019). However, the percentage of RD in SG neurons raised to 44.44% in this study. Reasons for this incidence discrepancy may attribute to the different species we used in our studies. Furthermore, recent literature segregating SG neurons in a geneticallydefined approach found two-thirds of cholinergic interneurons in mouse dorsal horn displayed RD following hyperpolarization (Mesnage et al., 2011), whilst the fraction of RD in delayed  Data reported as mean ± SEM. RC, rostro-caudal extent of dentrites; DV, dorsal-ventral expansion of dendrites; SR, SC, SD, SV, dendrites spread from center of the soma to rostral, caudal, dorsal, and ventral limit, respectively. * p < 0.05. * * * p < 0.001.
(short-latency) firing SG interneurons expressing neuropeptide Y (NPY) Y1 receptor from mice was 32% (Sinha et al., 2021). This quantitative difference suggests that RD might distribute in SG neurons in a specific cellular manner. Substantia gelatinosa neurons are a very heterogeneous population of interneurons with varied electrophysiological features, morphologies, and molecular profiles (Lu et al., 2006;Todd, 2017;Peirs et al., 2020). SG neurons can be divided into five broad categories on structural grounds, namely islet, central, vertical, radial, and unclassified cells. A series of pioneering works well established that islet cells are predominantly inhibitory (Todd and Sullivan, 1990;Maxwell et al., 2007;Yasaka et al., 2010), while other morphologies have been proposed to be associated with both excitatory and inhibitory phenotypes in the literature (Hantman et al., 2004;Lu and Perl, 2005;Maxwell et al., 2007;Duan et al., 2014;Punnakkal et al., 2014;Smith et al., 2015;Boyle et al., 2019).
Given that the most common morphological type in our sample of SG neurons with RD was "islet, " and that the majority of these cells discharged tonically when depolarized, a vital characteristic of inhibitory neurons (Lu and Perl, 2003;Zheng et al., 2010), a major finding of this study is that RDexpressing interneurons in SG have probably an inhibitory nature. However, a strict conclusion could finally be drawn based on future immunocytochemical studies. Furthermore, combined with transgenic technologies, recent publications have added greatly to our understanding of the functional heterogeneity of inhibitory interneurons in the superficial spinal dorsal horn. So far, several molecularly distinct inhibitory populations in the superficial region of the dorsal spinal cord have been described according to neurochemical classification scheme established based on the expression of NPY, nitric oxide synthase (nNOS), parvalbumin (PV), galanin, calretinin (CR), and receptor tyrosine kinase (Tiong et al., 2011;Polgár et al., 2013;Smith et al., 2015;Cui et al., 2016;Iwagaki et al., 2016). Future work that identifies the expression of molecular compounds in RDexpressing SG neurons will be required to elucidate their exact neurochemical subtype and hence be useful for addressing their functional role in-depth.

SG Neurons With RD Have Differential Excitatory Synaptic Inputs and Primary Afferent Innervations
Prior anatomical work on SG neurons from both rats and mice showed islet, central tonic, PV-positive, and CR-positive cells, which were identified with an inhibitory phenotype by morphology, electrophysiology, or the expression of particular molecular markers, received relatively weak excitatory inputs, as assessed by low-frequency sEPSCs (Grudt and Perl, 2002;Hughes et al., 2012;Smith et al., 2015;Graham, 2020). In keeping with these findings, our observation of excitatory synaptic inputs revealed that although the amplitudes of excitatory events (sEPSCs and mEPSCs) were comparable between the two clusters of SG neurons, neurons with RD presented lower frequencies of both sEPSCs and mEPSCs. Thus, our results mirror these recent reports indicating that inhibitory interneurons receive a lower excitatory drive. Because functional dendritic spines are recognized as sites of presynaptic inputs (Alvarez and Sabatini, 2007), it is tempting to speculate that a possible cause of the phenomenon reported here is RD-expressing SG neurons in our sample may exhibit less dendritic spine density in spite of larger dendritic areas. Another potential source of inconsistency regarding the EPSC frequency is that the presynaptic release probabilities for neurons with or without RD from local excitatory interneurons may differ. Nevertheless, these possible mechanisms need to be further confirmed by dendritic spine analysis along with a paired patch-clamp recording.
Numerous studies have investigated synaptic input onto inhibitory populations in dorsal horn superficial layers, and interpretations from these datasets suggest that discrete inhibitory SG interneurons have differing synaptic afferents. One of the previous studies from rats showed that islet cell normally received monosynaptic excitatory input from unmyelinated afferents (C afferents), this result was supported by an in vitro electrophysiological experiment using mice where NPY-expressing neurons displayed responses to monosynaptic inputs originated from C fibers (Grudt and Perl, 2002;Iwagaki et al., 2016). In contrast, immunolabelling work from Hughes et al. found that PV-immunoreactive cells, a subpopulation of islet neurons, were directly associated with VGLUT1expressing terminals derived from myelinated afferent fibers, including both Aδ and Aβ fibers specifically (Hughes et al., 2012). Other inhibitory neurons belonging to the galanin or nNOS populations were likely to form contacts from both myelinated and unmyelinated classes of primary afferent either directly or indirectly, with many cells receiving convergent inputs (Hantman et al., 2004;Ganley et al., 2015). Although evidence from different groups showed disagreement about the primary innervations onto inhibitory SG neurons, the above evidence demonstrated that SG neurons with an inhibitory identity were likely recruited in gating innocuous as well as noxious information. In agreement with results reported by Ganley et al. and Hantman et al., we found that SG neurons with RD received both monosynaptic and polysynaptic inputs from Aδ and C fibers, the presumed nociceptive primary afferents. As we proposed that RD-expressing neurons were inhibitory interneurons based on their morphological and electrophysiological features, they may counterbalance excitatory drive through feedforward, feedback, or lateral inhibition in dorsal horn circuits, and hence contribute to nociceptive sensation and modulation.

RD in SG Neurons Is Regulated by I h and I T
Although several channels, for instance, inwardly-rectifying potassium channels (Wang et al., 2016), low-threshold TTXresistant sodium channels (Kurowski et al., 2018), persistent sodium channels (Sangrey and Jaeger, 2010), and high-threshold calcium channels (Zheng and Raman, 2009) have been confirmed to be involved in regulating RD by hyperpolarizing RMP, decreasing neuronal input resistances, or enhancing synaptic inhibition that contributes to post-inhibitory depolarizations, currents mediated by HCN channels and T-type calcium channels are recognized as pivotal ionic factors underlying RD (Boehme et al., 2011;Engbers et al., 2011). With respect to the role of I T in RD, pharmacological blockade of T-type calcium channels by using blockers such as NiCl 2 , mibefradil, or NNC 55-0396 could produce a significant block on the total number of spikes generated in RD (Alvina et al., 2009). In addition, in keeping with a recent study reporting that RD was barely detected in SG neurons without I T (Wu et al., 2018), data from Candelas et al. also found that ablation of Cav3.2 but not Cav3.1 downregulated the proportion of RD in SG (Candelas et al., 2019). On the other hand, I h has been proposed to be a key factor in determining the initiation of the first spike in RD, as both pharmacological blockade and genetic knockout of HCN2 could considerably delay RD latency .
It has been demonstrated that various isoforms of T-type calcium channels along with HCN channels are present in SG (Peng et al., 2017;Candelas et al., 2019;Cheng et al., 2019). Alterations in the expression and function of these channels are responsible for the development and maintenance of chronic pain (Papp et al., 2010;Feng et al., 2019). Therefore, it is possible to assume a potential effect of RD on nociceptive processing. In the present work, we also found that abolishment of I T significantly reduced the RD amplitude, while I h blockade modulated the onset of the rebound firing. Our observations are in line with preceding reports showing the role of I h and I T in controlling rebound spike responses.

Functional Implications of RD in SG Neurons
It has been postulated that as an intrinsic biophysical property in the sensory system, RD may play a possible role in encoding stimulus intensity (Large and Crawford, 2002;Oswald et al., 2004;Rivera-Arconada and Lopez-Garcia, 2015). As indicated by pilot observations, different firing patterns were linked with distinct neuronal functions. For instance, neurons showing tonic firing are proposed to be involved in encoding both the strength and duration of input signals. In addition to tonic-firing, initial-firing neurons also contribute to encoding the intensity of afferent excitation (Ruscheweyh and Sandkuhler, 2002). Thus, as the tonic and initial firing were the top two patterns recorded in RD-expressing SG neurons, our results indicate that this group of cells plays a substantial role in coding the intensity of ascending impulses. Additionally, RD has also been speculated to be an inhibition-excitation converter that transforms an inhibitory input into excitatory output (Sanchez-Vives and McCormick, 2000;Surges et al., 2006). However, our results only provide information about excitatory inputs onto RD-expressing neurons; future studies addressing their presynaptic inhibitory circuits involving primary afferents may further constitute circumstantial evidence in favor of this possibility.

CONCLUSION
In summary, our study revealed that SG neurons with RD are not only electrophysiologically distinct from those lacking RD, but also differ in synaptic transmission and thus may play a unique physiological role in the nociception network.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding authors.

ETHICS STATEMENT
The animal study was reviewed and approved by Institutional Animal Care and Use Committee of Nanchang University.

AUTHOR CONTRIBUTIONS
TL and DZ interpreted the data and designed and supervised the study. MZ and TL drafted and edited the manuscript. MZ and YY performed the electrophysiological and morphological experiments. XC, FZ, GX, WS, FL, and LL assisted in analyzing and interpreting the data. ZW, YZ, and XZ provided discussion in study design and interpretation of data. All authors read and approved the final manuscript.

FUNDING
The study was supported by the National Natural Science Foundation of China (Nos. 81860216, 81560198 to DZ and 32060186 to TL).