Divergent Modulation of Nociception by Glutamatergic and GABAergic Neuronal Subpopulations in the Periaqueductal Gray

Abstract The ventrolateral periaqueductal gray (vlPAG) constitutes a major descending pain modulatory system and is a crucial site for opioid-induced analgesia. A number of previous studies have demonstrated that glutamate and GABA play critical opposing roles in nociceptive processing in the vlPAG. It has been suggested that glutamatergic neurotransmission exerts antinociceptive effects, whereas GABAergic neurotransmission exert pronociceptive effects on pain transmission, through descending pathways. The inability to exclusively manipulate subpopulations of neurons in the PAG has prevented direct testing of this hypothesis. Here, we demonstrate the different contributions of genetically defined glutamatergic and GABAergic vlPAG neurons in nociceptive processing by employing cell type-specific chemogenetic approaches in mice. Global chemogenetic manipulation of vlPAG neuronal activity suggests that vlPAG neural circuits exert tonic suppression of nociception, consistent with previous pharmacological and electrophysiological studies. However, selective modulation of GABAergic or glutamatergic neurons demonstrates an inverse regulation of nociceptive behaviors by these cell populations. Selective chemogenetic activation of glutamatergic neurons, or inhibition of GABAergic neurons, in vlPAG suppresses nociception. In contrast, inhibition of glutamatergic neurons, or activation of GABAergic neurons, in vlPAG facilitates nociception. Our findings provide direct experimental support for a model in which excitatory and inhibitory neurons in the PAG bidirectionally modulate nociception.

Previous studies have suggested that the vlPAG has a bidirectional role in the modulation of nociception. Nonspecific activation of vlPAG neurons produces analgesia, whereas inhibition of vlPAG produces hyperalgesia to noxious stimulation (Reynolds, 1969;Liebeskind et al., 1973;Moreau and Fields, 1986;Depaulis et al., 1987;Siegfried and de Souza, 1989). It is tempting to speculate that output from the vlPAG has a purely analgesic action. However, the vlPAG comprises diverse subpopulations of neurons with distinct neurochemical properties that regulate excitatory and inhibitory neurotransmission (Behbehani and Fields, 1979;Moss and Basbaum, 1983;Moss et al., 1983;Moreau and Fields, 1986;Behbehani et al., 1990;Behbehani, 1995;Vaughan et al., 1997;Hahm et al., 2011;Ho et al., 2013). Microinjection of glutamate receptor agonists or GABA antagonists into the vlPAG leads to global activation of neurons and produces antinociceptive effects to noxious stimuli (Moreau and Fields, 1986;Ness and Gebhart, 1987;Carstens et al., 1988;Jacquet, 1988;Jones and Gebhart, 1988;Jensen and Yaksh, 1989;Sandkühler et al., 1989;Carstens et al., 1990;Budai and Fields, 1998;Morgan et al., 2003). In contrast, microinjecting glutamatergic antagonists or GABA agonists, presumably leading to global suppression of neural activity in the vlPAG, produces hyperalgesia (Moreau and Fields, 1986;Depaulis et al., 1987;Siegfried and de Souza, 1989;Behbehani et al., 1990). These studies suggest that in the context of a noxious stimulus, GABAergic neurotransmis-sion in the vlPAG is pronociceptive, although the source of the GABAergic inputs to the vlPAG cannot be determined (Reichling andBasbaum, 1990a, 1990b). Collectively, a large number of prior studies suggest that glutamatergic and GABAergic neurons within the vlPAG play critical and complex roles in processing nociception (Behbehani and Fields, 1979;Moreau and Fields, 1986;Millan et al., 1987;Sandkühler et al., 1989;Siegfried and de Souza, 1989;Vaughan et al., 1997). Based on these studies, the vlPAG GABA disinhibition hypothesis has been proposed (Basbaum and Fields, 1978;Fields, 2004;Lau and Vaughan, 2014). In this hypothesis, GABAergic interneurons exert tonic inhibition over vlPAG glutamatergic neurons, which are thought to be output neurons that project to the RVM to facilitate the descending inhibition of nociception (Vaughan et al., 1997;Budai and Fields, 1998;Wang and Wessendorf, 2002;Maione et al., 2006;Starowicz et al., 2007;Heinricher et al., 2009;Park et al., 2010;Tovote et al., 2016). Despite the wealth of evidence supporting this model, the distinct roles of GABAergic and glutamatergic neuronal populations in descending vlPAG pain modulation have not been directly investigated. In this study, we use cell type-specific chemogenetic manipulations of neuronal activity in the vlPAG to test the hypothesis that GABAergic neurons are pronociceptive and glutamatergic neurons are antinociceptive.

Animals
All experiments were conducted in accordance with the National Institutes of Health guidelines and with approval from the Animal Care and Use Committee of Washington University School of Medicine. Male, 8-to 12-week-old, heterozygous Slc32a1 tm2L°wl (Vgat-ires-Cre, selectively targets Vgat ϩ GABAergic inhibitory neurons), Slc17a6 tm2L°wl (Vglut2ires-Cre, selectively targets Vglut2 ϩ glutamatergic excitatory neurons, and C57BL\6J mice were used (Vong et al., 2011). Mice were purchased from Jackson Laboratories (C57BL\6J, Vgat Cre, stock number 016962 and Vglut2 Cre stock number 016963) and colonies were established in our facilities. Experimenters were blind to treatment and genotype.

Chemogenetic manipulation
Three weeks after viral injections, mice were injected intraperitoneally with clozapine N-oxide (CNO, BML-NS105 from Enzo life sciences) 60 min before beginning behavioral assessment, and data were collected between the second and third hour after injection. All baselines for thermal and mechanical sensitivity were recorded two weeks after the viral injections and one week before the CNO administration. The doses of CNO were chosen based on preliminary pilot experiments designed to determine the minimal dose needed to activate designer receptor exclusively activated by designer drugs (DREADD) receptors in the vlPAG of C57BL\6J mice, Vgat-ires-Cre and Vglut2-ires-Cre mice. We assessed the response of these different mouse lines to various doses of CNO to identify the minimal doses required to activate DREADDs (data not shown). In non-Cre-dependent studies, we administered 1 mg/kg CNO for both hM3Dq activation and hM4Di inhibition. In Vgat Cre mice, we administered 3 mg/kg CNO for both hM3Dq activation and hM4Di inhibition. In Vglut2 Cre mice, we administered 2 mg/kg CNO for both hM3Dq activation and hM4Di inhibition.

Pain behavior assessment
To evaluate nociception, mechanical withdrawal thresholds and thermal withdrawal latencies were assayed. Mice were tested for baseline responses to mechanical and thermal stimuli, as previously described (O'Brien et al., 2013). For the assessment of mechanical withdrawal threshold, von Frey filaments (North Coast Medical) were applied bilaterally to the hind paws of the mice using the up-down method. Two to three trials were performed on each hind paw for each mouse. The average 50% withdrawal threshold was calculated for each paw individually and then averaged to obtain a threshold value for each mouse. The Hargreaves test was performed to evaluate heat sensitivity thresholds, measuring latency of withdrawal to a radiant heat source (IITC Life Science, Model 390). We applied the radiant heat source bilaterally to the hind paw and measured the latency to evoke a withdrawal. Three to five replicates were acquired per hind paw per mouse, and values for both paws were averaged.

Electrophysiology
To determine the functional effects of chemogenetic manipulations in vlPAG neurons, we performed targeted whole-cell patch-clamp recordings in acute coronal slices from both Vgat-and Vglut2-Cre mice expressing either hM3Dq or hM4Di receptors. Mice used for electrophysiology and behavioral studies were between 8 and 12 weeks of age. Three weeks after viral infection of vlPAG neurons, coronal slices containing the vlPAG were prepared as previously described (Siuda et al., 2015). GABAergic and glutamatergic neurons in the vlPAG were visualized through a 40ϫ objective using IR-differential interference contrast (DIC) microscopy on an Olympus BX51 microscope, and mCherryϩ cells were identified using epifluorescent illumination with a green LED (530 nm; Thorlabs), coupled to the back fluorescent port of the microscope. Whole-cell recordings of vlPAG GABAergic and glutamatergic neurons expressing hM3Dq-mCherry and hM4Di-mCherry were performed using a Heka EPC 10 amplifier (Heka) with Patchmaster software (Heka). Following stable 5-min whole-cell recordings (baseline), the direct effects of either hM3Dq or hM4Di receptor expression on cellular excitability was isolated by blocking AMPA/KARs (10 M NBQX, Abcam), NMDARs (50 M D-APV, Abcam), GABA A Rs (100 M picrotoxin, Abcam), and GABA B Rs (50 M baclofen, Abcam). aCSF solution containing 10 M CNO added to the antagonist cocktail above was bath applied to the brain slice.

Fluorescence in situ hybridization (FISH)
Following rapid decapitation of mice, brains were flash frozen in Ϫ50°C 2-methylbutane and stored at Ϫ80°C for further processing. Coronal sections containing the PAG, corresponding to the injection coordinates used in the behavioral experiments, were cut at 20 m at Ϫ20°C and thaw-mounted onto Super Frost Plus slides (Fisher). Slides were stored at -80°C until further processing. FISH was performed according to the RNAScope 2.0 Fluorescent Multiple Kit User Manual for Fresh Frozen Tissue (Advanced Cell Diagnostics) as described previously (Wang et al., 2012). Slides containing PAG coronal brain sections were fixed in 4% PFA, dehydrated, and pretreated with protease IV solution for 30 min. Sections were then incubated with target probes for mouse Vglut2 (slc17a6, accession number NM_080853.3, probe region 1986-2998), Vgat (slc32a1, accession number NM_009508.2, probe region 894-2037), and Cre (accession number KC845567.1, probe region 1058-2032). All target probes consisted of 20 double Z oligonucleotides and were obtained from Advanced Cell Diagnostics. Following probe hybridization, sections underwent a series of probe signal amplification steps (AMP1-4) followed by incubation of fluorescent probes (Alexa Fluor 488, Atto 550, Atto 647), designed to target the specified channel associated with the probes. Slides were counterstained with 4',6diamidino-2-phenylindole (DAPI; RNAScope), and coverslips were mounted with Vectashield Hard Set mounting medium (Vector Laboratories). Images were obtained on a Leica TCS SPE confocal microscope (Leica), and Application Suite Advanced Fluorescence (LAS AF) software was used for analyses.

Statistics
Throughout the study, researchers were blinded to all experimental conditions. At least two to three replicate measurements were performed and averaged in all behavioral assays. The studies were designed to compare behavioral readouts following CNO to baseline values before CNO administration. This was done using paired t tests to account for interindividual variability among mice across different cohorts. The control eGFP group was included throughout our study to determine whether CNO administration or DREADD expression had off-target effects in the behaviors that we tested. All datasets were evaluated for normality using the D'Agostino and Pearson omnibus normality test. A parametric test was used only when normality was confirmed. If normality could not be confirmed, the nonparametric Wilcoxon matched pairs test was used.

The role of vlPAG GABAergic and glutamatergic neuronal populations in nociceptive processing
The vlPAG is comprised of both inhibitory GABAergic and excitatory glutamatergic neurons, and we hypothesized that these neuronal populations differentially regulate nociceptive processing. Double-label RNA-FISH in the vlPAG of c57-mice revealed that GABAergic neurons (Vgat transcripts) and glutamatergic neurons (Vglut2 transcripts) show no overlap in expression in the PAG and, thus, are distinct populations ( Fig. 2A,B). To selectively test the functional contributions of vlPAG GABAergic and glutamatergic neurons in modulating nociceptive behaviors, we used Vgat-ires-Cre and Vglut2-IRES-Cre mice to target and manipulate the activity of GABAergic and glutamatergic neurons, respectively. To determine the specificity of Cre in targeting Vgat ϩ neurons in Vgat-IRES-Cre mice or in targeting Vglut2 ϩ neurons in the Vglut2-IRES-Cre mice, we performed RNA-FISH using probes for Vgat, Vglut, and Cre in vlPAG slices obtained from Vgat Cre and Vglut2 Cre mice. We observed 79 Ϯ 4.1% of Vglut2 ϩ transcripts in the vlPAG colabel with Vglut2 Cre-expressing neurons, and 97.5 Ϯ 2.5% Vglut2 Cre-expressing neurons in the vlPAG colabel with Vglut2 ϩ transcripts (Fig. 2C,E-J). We also observed 92 Ϯ 4.5% of Vgat ϩ transcripts in the vlPAG colabel with Vgat Cre-expressing neurons, and 95.5 Ϯ 4.2% Vgat Cre-expressing neurons in the vlPAG colabel with Vgat ϩ transcripts (Fig. 2D,K-P). Our doublelabel RNA-FISH studies revealed that Vgat and Vglut2 Cre mice faithfully label vlPAG GABAergic and Vglut2 ϩ glutamatergic neuronal populations, as described previously for other brain regions (Vong et al., 2011).
To selectively test the functional contributions of vlPAG GABAergic and glutamatergic neurons in modulating nociceptive behaviors, we used Cre-dependent DREADD expression in Vgat-ires-Cre and Vglut2-ires-Cre mice to target GABAergic and glutamatergic neurons, respectively. Virus carrying Cre-dependent stimulatory (hM3Dq) or inhibitory (hM4Di) DREADDs fused with mCherry were injected into the vlPAG of Vgat-Cre mice or Vglut2-Cre mice. Three weeks after DREADD injection, we prepared acute coronal slices of the vlPAG from Vgat-Cre and Vglut2-Cre mice and targeted mCherry ϩ neurons in vlPAG for whole-cell recordings (Fig. 3A). hM3Dq-expressing vlPAG neurons were held at hyperpolarized membrane potentials, and a brief bath application of 10 M CNO caused a transient depolarization and robust action potential firing in both Vgat  and Vglut2 neurons (Fig. 3B). To test the effects of G icoupled inhibition with hM4Di, we monitored neuronal activity while holding cells with a depolarizing current injection, which elicited persistent action potential firing in both Vgat and Vglut2 neurons. Bath perfusion with 10 M CNO resulted in prolonged membrane hyperpolarization and decreased firing of both cell types (Fig. 3C). Quantification of hM3Dq-expressing neurons showed that CNO depolarized neurons by an average of 3.6 mV and caused a small decrease in the input resistance (Fig. 3D,G). In contrast, activation of hM4Di hyperpolarized neurons by an average of 5.6 mV, and substantially reduced input resistance, consistent with CNO-induced G i -coupling to inwardly rectifying K ϩ channels (Sternson and Roth, 2014; Urban and Roth, 2015; Fig. 3D,G). We also investigated how CNO modulates membrane excitability in response to depolarizing step current injections. In hM3Dq-expressing neurons, we observed a large increase in the number of action potentials elicited during a 1ϫ rheobase current following CNO stimulation in slices from both Vgat-and Vglut2-Cre mice (Fig. 3E,H). Suprathreshold current injections of 2ϫ rheobase elicited sustained high-frequency action potential firing in both neuronal subtypes (Fig. 3F,I, blue traces), and CNO application dramatically reduced membrane excitability to an identical suprathreshold stimulus in mice injected with New Research hM4Di constructs (Fig. 3F,I, green traces). This confirmed that we are able to bi-directionally modulate GABAergic and glutamatergic neuron excitability in the vlPAG using hM3Dq and hM4Di DREADDs, and led us to explore the contributions of these neuronal populations in nociceptive processing.

Discussion
Here, we report that global chemogenetic activation of vlPAG neurons attenuates thermal nociception while inhibiting vlPAG neurons cause thermal and mechanical hypersensitivity. Using cell type-specific chemogenetic manipulations, we found that activating GABAergic, or inhibiting glutamatergic, neurons in vlPAG, causes thermal and mechanical hypersensitivity. In contrast, inhibiting GABAergic or activating glutamatergic neurons attenuates thermal sensitivity but has no effect on mechanical sensitivity. The differential effects on thermal and mechanical sensitivity suggest that distinct circuit elements within the vlPAG regulate these two sensory modalities. These results provide new insights into the functional role of GABAergic and glutamatergic neurons in the vlPAG in the modulation of nociception.
The vlPAG is known to be an essential component of neural pathways that mediate stimulation and stressinduced analgesia (Reynolds, 1969;Yeung et al., 1977;Jones and Gebhart, 1988;Morgan et al., 1989;Hohmann et al., 2005;Samineni et al., 2011). Consistent with these previous studies, we found that global chemogenetic activation of vlPAG neurons produced antinociceptive effects. We also show that chemogenetic inhibition of vlPAG neurons leads to nociceptive hypersensitivity, consistent with a bidirectional role of the vlPAG in the modulation of nociception (Moreau and Fields, 1986;Depaulis et al., 1987;Heinricher et al., 1987;Ness and Gebhart, 1987;Carstens et al., 1988;Jacquet, 1988;Jones and Gebhart, 1988;Jensen and Yaksh, 1989;Siegfried and de Souza, 1989;Behbehani et al., 1990;Carstens et al., 1990). The magnitude of antinociceptive effects observed after global chemogenetic activation of vlPAG neurons is modest relative to the robust effects produced by electrical stimulation of the PAG or by microinjection of morphine, GABA A receptor antagonists, or glutamate agonists into the vlPAG (Reynolds, 1969;Liebeskind et al., 1973;Moreau and Fields, 1986;Depaulis et al., 1987;Ness and Gebhart, 1987;Carstens et al., 1988;Sandkühler et al., 1989;Siegfried and de Souza, 1989;Carstens et al., 1990). Such a difference could be expected if the number of neurons that are transduced with stimulatory DREADDs is small relative to the large number of neurons impacted by electrical or pharmacologic approaches.
Although the vlPAG has been extensively studied for its role in endogenous descending pain modulation (Reynolds, 1969;Liebeskind et al., 1973;Basbaum and Fields, 1978;Sandkühler et al., 1989;Behbehani et al., 1990;Vaughan et al., 1997;McGaraughty et al., 2003;Starowicz et al., 2007;Morgan et al., 2008;Waters and Lumb, 2008;Heinricher et al., 2009;Samineni et al., 2011;Wang et al., 2012;Ho et al., 2013;Lau and Vaughan, 2014;Tovote et al., 2016), previous studies have not yet determined how distinct subpopulations of vlPAG neurons modulate pain transmission. Inhibitory neurotransmission in the vl-PAG is known to modulate nocifensive behaviors, since microinjecting a GABA agonist produces pronociceptive effects, while decreasing inhibitory neurotransmission by microinjection of GABA antagonists into the vlPAG produces antinociceptive effects to noxious stimuli (Moreau and Fields, 1986;Depaulis et al., 1987;Sandkühler et al., 1989;Budai and Fields, 1998;Morgan et al., 2003). Such studies led some to propose the vlPAG GABA disinhibition analgesia hypothesis (Basbaum and Fields, 1978;Fields, 2004). Corroborating this hypothesis, many studies have shown that mu opioid receptor agonists have direct inhibitory effects on GABAergic neurons of vlPAG. This causes analgesia when directly administered into the vlPAG, suggesting that inhibition of GABAergic vlPAG neuronal activity may be a major mechanism for opioidinduced analgesia (Chieng and Christie, 1996;Vaughan et al., 1997). The cellular mechanisms underlying the analgesic and hyperalgesic effects of manipulating inhibitory and excitatory neurotransmission in the vlPAG have not been directly evaluated (Moreau and Fields, 1986;Carstens et al., 1988;Jacquet, 1988;Sandkühler et al., 1989;Vaughan et al., 1997;Budai and Fields, 1998;Fields, 2004;Maione et al., 2006;Starowicz et al., 2007). It is not known how distinct neuron subpopulations in the vlPAG engage complex downstream circuits of the descending pain modulation pathway. For the first time, we show that chemogenetic activation of vlPAG GABAergic neurons causes hypersensitivity to nociceptive stimuli while their inhibition causes antinociception, consistent with the proposed role of GABAergic vlPAG neurons in pain modulation.
It has been hypothesized that GABAergic interneurons exert tonic inhibition of vlPAG glutamatergic neurons, which are thought to be output neurons that project to the RVM (Jacquet, 1988;Roychowdhury and Fields, 1996;Vaughan et al., 1997;Wang and Wessendorf, 2002;Morgan et al., 2008;Park et al., 2010;Hahm et al., 2011;Ho et al., 2013). These glutamatergic neurons have been hypothesized to play a role in an analgesic modulatory pathway (McGaraughty et al., 2003;Maione et al., 2006;Starowicz et al., 2007), but this has not been selectively demonstrated. In agreement with the GABA disinhibition hypothesis, our RNA-FISH studies show that GABAergic and glutamatergic neurons in the PAG are distinct populations. We also found that chemogenetic activation of vlPAG glutamatergic neurons is antinociceptive, whereas chemogenetic inhibition of these neurons is pronocicep-tive. Conversely, chemogenetic activation of GABAergic neurons in the vlPAG produces hypersensitivity while inhibition of these neurons produces analgesia.
Surprisingly, we find that activation of vlPAG glutamatergic neurons or inhibition of GABAergic neurons attenuated thermal but not mechanical sensitivity. While there are several possible explanations for this finding, we propose the hypothesis that distinct populations of inhibitory neurons regulate mechanical and thermal nociceptive modulatory pathways emanating from the vlPAG. Here, we posit the presence of tonically active GABAergic neurons regulating descending pathways for thermal nociception, while the population regulating mechanical nociceptive modulation might by quiescent. This would explain the difference in effects of activating GABAergic neurons or inhibiting glutamatergic neurons on mechanical nociception. It is possible that the differences in the effects of DREADD-dependent regulation are simply due to the basal state of the neurons in question. That is, if the neurons are quiescent at baseline, then activation of a G i -coupled DREADD might not affect descending modulation. Similarly, a neuron that is firing at a relatively high frequency at baseline might not be further stimulated by a G q -coupled DREADD. Future studies are necessary to determine the differences in neuronal populations and circuits that code for mechanical versus thermal sensitivity. It is also possible that these results are simply due to a ceiling effect in our von Frey measurements. That is, that von Frey testing is not able to detect analgesic effects at baseline. We do not favor this hypothesis, as treatment with analgesic drugs can indeed increase PWTs in mice (Anseloni and Gold, 2008).
We believe that our results should not be interpreted as absolute, and we recognize that the behavioral changes that we report should not be attributed to the entirety of either the Vgat or Vglut2 vlPAG neuronal populations. The PAG is comprised of molecularly diverse neuronal subpopulations that express fast neurotransmitters and/or neuropeptides (Mantyh, 1982;Moss and Basbaum, 1983;Moss et al., 1983;Smith et al., 1994). In the hypothalamus, recent genetic analysis of anatomically defined neurons has identified subpopulations that coexpress a variety of neuroactive substances that were previously thought to be exclusive to certain clusters (Romanov et al., 2017). Therefore, it is likely that vlPAG GABAergic and glutamatergic neurons can be further subdivided into subpopulations based on their genetic identity and physiology. Although we have only attempted to dissect the roles of glutamatergic versus GABAergic vlPAG neurons in pain modulation, future studies should examine the interplay between other neuronal populations within the vlPAG that can be defined with the expression of other makers, such as neuropeptides, to assess their roles in regulating nociceptive processing.
While the stimulation of vlPAG is predominantly associated with antinociceptive effects (Reynolds, 1969;Liebeskind et al., 1973;Carstens et al., 1988;Morgan et al., 1989;Sandkühler et al., 1989;Hohmann et al., 2005;Maione et al., 2006;Starowicz et al., 2007;Samineni et al., 2011), recent studies also identified facilitatory effects of the vlPAG in the maintenance of neuropathic pain (Pertovaara et al., 1996;Pertovaara et al., 1997;Heinricher et al., 2004;Guo et al., 2006;Lü et al., 2010). We show here that chemogenetic activation of GABAergic neurons or inhibition of glutamatergic neurons can lead to hypersensitivity to mechanical and thermal stimuli, suggesting that any disruption of the balance between activity of the vlPAG excitatory and inhibitory neurons might contribute to the maintenance of chronic pain (Hahm et al., 2011;Ho et al., 2013;Lau and Vaughan, 2014). Future studies should explore the plastic changes in GABAergic and glutamatergic neurons that might contribute to the maintenance of chronic pain.
The vlPAG has been shown to be instrumental in the descending modulation of pain processing. The vlPAG is known to form strong connections with the RVM, and the locus coeruleus (Beitz et al., 1983;Behbehani, 1995;Antal and Odeh, 1998;Odeh and Antal, 2001;Bowman et al., 2013). Based on the data we present here, it will be of great interest to determine which of these, or other, projection targets mediate the differential modulation of nociception by glutamatergic and GABAergic projections from the vlPAG.