An excitatory lateral hypothalamic circuit orchestrating pain behaviors in mice

Understanding how neuronal circuits control nociceptive processing will advance the search for novel analgesics. We use functional imaging to demonstrate that lateral hypothalamic parvalbumin-positive (LHPV) glutamatergic neurons respond to acute thermal stimuli and a persistent inflammatory irritant. Moreover, their chemogenetic modulation alters both pain-related behavioral adaptations and the unpleasantness of a noxious stimulus. In two models of persistent pain, optogenetic activation of LHPV neurons or their ventrolateral periaqueductal gray area (vlPAG) axonal projections attenuates nociception, and neuroanatomical tracing reveals that LHPV neurons preferentially target glutamatergic over GABAergic neurons in the vlPAG. By contrast, LHPV projections to the lateral habenula regulate aversion but not nociception. Finally, we find that LHPV activation evokes additive to synergistic antinociceptive interactions with morphine and restores morphine antinociception following the development of morphine tolerance. Our findings identify LHPV neurons as a lateral hypothalamic cell type involved in nociception and demonstrate their potential as a target for analgesia.


Introduction
Responding appropriately to environmental stimuli is vital to an organism's survival. Nociception facilitates survival via the detection of dangerous environmental stimuli, which organisms use to escape and avoid these threats (Bolles and Fanselow, 1980;Tovote et al., 2015). However, maladaptive processes following injury or infection can cause the transition to chronic pain, a clinical condition with great economic burden that is not well-addressed by current therapeutics (Price et al., 2018;Grace et al., 2014). The widespread failure of preclinical pain therapies to translate to the clinic may be due to the historical focus on studying acute, pain-stimulated nocifensive behaviors in naive animals such as paw withdrawal to heat, which are not maladaptive and necessitate the examination of off-target effects like sedation in separate assays (Negus et al., 2015). Rather than physical sensitization to painful stimuli, the more problematic components of chronic pain in humans are likely the loss of ability to perform standard daily life activities and development of comorbid depression (Asmundson and Katz, 2009;Cleeland and Ryan, 1994;Dworkin et al., 2005;Elman et al., 2013;Negus et al., 2006). As such, rodent studies searching for new analgesics have begun to investigate ethological behaviors like nesting that are suppressed by noxious stimulation (e.g., forgoing standard life activities) as well as the affective/emotional component of nociception with assays of noxious stimulus-induced aversion (e.g., comorbid depression) (Negus et al., 2015;Johansen et al., 2001;Corder et al., 2019). Identifying specific brain pathways capable of managing these multiple components of chronic pain behavior and developing strategies for targeting them for translational use will advance the search for novel analgesics.
While the LH circuits controlling food intake and reward have received intense focus over the past several years (Jennings et al., 2015;Jennings et al., 2013;Qualls-Creekmore et al., 2017;Navarro et al., 2016;Barbano et al., 2016;Nieh et al., 2016), those governing nociception have been understudied by comparison. Thus, with its diverse array of neuronal populations (Mickelsen et al., 2019), uncovering genetically defined LH circuits that regulate pain behavior may bring forth novel therapeutic targets. We previously described a small population of fast-spiking glutamatergic LH neurons expressing parvalbumin (LH PV neurons) that forms functional excitatory synapses in the ventrolateral periaqueductal gray area (vlPAG) and regulates acute thermal and chemical nociception Kisner et al., 2018). However, the broader therapeutic potential of LH PV neurons and their specific targets within the vlPAG have not yet been fully assessed.
Using in vivo calcium imaging, we demonstrate that LH PV neurons exhibit time-locked responses to acute hot or cold stimuli as well as increased activity following the administration of a persistent inflammatory irritant. Additionally, we show that chemogenetic modulation of LH PV neurons alters not only reflexive nociceptive behaviors over a timescale of hours but also restores noxious stimulussuppressed behavior and ameliorates noxious stimulus-associated negative affect. In models of persistent inflammatory or neuropathic pain, optogenetic activation of LH PV neurons or their axonal projections in the vlPAG attenuates nociception. Furthermore, neuroanatomical tracing using modified rabies virus revealed that LH PV neurons preferentially target nociception-suppressing glutamatergic neurons over nociception-facilitating GABAergic neurons in the vlPAG. Interestingly, we observed that activation of an LH PV neuron pathway to the lateral habenula (LHb) can mediate aversion-like behavior but not nociception, suggesting pathway-specific behavioral effects of these neurons. Finally, we report that LH PV neuronal activation evokes additive to synergistic antinociceptive interactions with morphine and restores morphine antinociception following the development of morphine tolerance. Our findings identify LH PV neurons as a lateral hypothalamic cell type intricately involved in nociception and demonstrate their potential as a novel target for analgesic treatment or for use in combination therapies with current analgesics.

Results
In vivo functional imaging of LH PV neurons LH PV neurons bidirectionally modulate responses to acute noxious stimuli , but their activity in response to noxious stimuli in vivo has not yet been studied. To investigate this, we used the combination of in vivo endomicroscopy with a genetically encoded calcium indicator (GCaMP) to measure intensity fluctuations of calcium-sensitive fluorophores as an indicator of neuronal activity in LH PV cells during behavior. First, we expressed a green fluorescent calcium indicator in LH PV neurons by injecting a Cre recombinase-dependent viral vector driving the expression of GCaMP6s (Chen et al., 2013) in the LH of Pvalb Cre transgenic mice (Hippenmeyer et al., 2005). For detection of GCaMP6s fluorescence, we implanted a GRIN lens above the LH PV nucleus and interfaced the lens with a detachable miniscope (Figure 1a, b). In conjunction with established and opensource computational algorithms for data processing (Friedrich et al., 2017;Zhou et al., 2018;Pnevmatikakis and Giovannucci, 2017), we were able to visualize ( Figure 1c) and extrapolate calcium (Ca 2+ ) traces from individual LH PV neurons over periods of behavioral testing (Figure 1d). We first monitored Ca 2+ dynamics in LH PV neurons (n = 87 neurons, three mice) in response to an acute thermal hot plate stimulus and clustered the neurons according to their response properties (Figure 1dÀf). In a subset of the recorded LH PV neurons ('Cluster 1,' n = 35/87 neurons), we observed time-locked increases in fluorescence in response to the 51˚C hot plate relative to a roomtemperature innocuous stimulus of similar visual and tactile properties, suggesting that this subpopulation of LH PV neurons becomes active in response to a thermal stimulus (Figure 1g). Another subset of neurons ('Cluster 2,' n = 16/87 neurons) exhibited an average decrease in activity in response to the hot plate relative to control stimulus ( Figure 1h). We observed a similar profile of time-locked responses to a 4˚C cold stimulus relative to a control innocuous stimulus (Figure 1i, j). One subset of the recorded neurons ('Cluster 1,' n = 15/53 neurons) was significantly activated in response to the cold plate relative to control stimulus (Figure 1k), while another subset ('Cluster 2,' n = 11/53 neurons) displayed significantly lower activity following the cold plate stimulus as compared to the control stimulus (Figure 1l). Within each cluster, we trained a support vector machine (SVM) classifier using averaged 10 s traces following contact with the noxious (hot/cold) or neutral surfaces and tested whether it could predict the stimulus type when given unlabeled traces. Remarkably, neuronal activity from each cluster except cluster 2 from the cold plate test could decode the correct stimulus type above chance levels ( Figure 1-figure supplement 1). Cluster 1 and 2 neurons were observed in each of the mice tested. Thus, we registered cells across the hot plate and cold plate sessions to examine whether LH PV neurons exhibited consistent response profiles across tests (Sheintuch et al., 2017). Of the 33 total neurons that were detected in both sessions, only 6 remained in the same Figure 1 continued filtered traces from individual LH PV neurons. Dotted lines represent contacts with hot plate. (e) Z-scored Ca 2+ traces of LH PV neurons (87 neurons, three mice) averaged across exposures to a 51˚C hot plate or a room temperature control surface. Dotted line represents contact with plate or control surface. (f) Clustering of 87 units by mean max amplitude and mean area under the curve (AUC) following hot plate surface contact. Dotted lines indicate the thresholds for inclusion into cluster 1 (mean max amplitude ! 1 and mean AUC ! 0) or cluster 2 (mean max amplitude 1 and mean AUC 0). (g) Neurons in cluster 1 (n = 35/87) displayed time-locked increases in activity in response to the hot plate as compared to the control room temperature surface. Two-way repeated-measures ANOVA on average Z-score per second revealed a significant time Â stimulus interaction (F(29, 986) = 10.47, p<0.0001). Bonferroni multiple comparisons post-test significant between-stimulus differences are represented by the bolded red line. Red and gray shaded areas represent s.e.m. Tan shaded region represents average contact time with hot plate stimulus. (h) Neurons in cluster 2 (n = 16/87) displayed average decreases in activity in response to the hot plate as compared to the control room temperature surface. Two-way repeated-measures ANOVA on average Z-score per second revealed a significant time Â stimulus interaction (F(29, 435) = 7.61, p<0.0001). Bonferroni multiple comparisons post-test significant between-stimulus differences are represented by the bolded red line. Red and gray shaded areas represent s.e.m. Tan shaded region represents average contact time with hot plate stimulus. (i) Z-scored Ca 2+ traces of LH PV neurons (53 neurons, three mice) averaged across exposures to a 4˚C cold plate or a room temperature control surface. Dotted line represents contact with plate or control surface. (j) Clustering of 53 units by mean max amplitude and mean AUC following cold plate surface contact. Dotted lines indicate the thresholds for inclusion into cluster 1 (mean max amplitude ! 1 and mean AUC ! 0) or cluster 2 (mean max amplitude 1 and mean AUC 0). (k) Neurons in cluster 1 (n = 15/53) displayed time-locked increases in activity in response to the cold plate as compared to the control room temperature surface. Two-way repeated-measures ANOVA on average Z-score per second revealed a significant time Â stimulus interaction (F(29, 406) = 5.94, p<0.0001). Bonferroni multiple comparisons post-test significant between-stimulus differences are represented by the bolded blue line. Blue and gray shaded areas represent s.e.m. Tan shaded region represents average contact time with cold plate stimulus. (l) Neurons in cluster 2 (n = 11/53, top) displayed average decreases in activity in response to the hot plate as compared to the control room temperature surface. Two-way repeated-measures ANOVA on average Z-score per second revealed a significant time Â stimulus interaction (F(29, 290) = 2.05, p=0.0016). Bonferroni multiple comparisons post-test significant between-stimulus differences are represented by the bolded blue line. Blue and gray shaded areas represent s.e.m. Tan shaded region represents average contact time with cold plate stimulus. (m) Illustration of fluorescent trace deconvolution to estimated periods of neuronal firing. (n) Average deconvolved events per 5 min period following no injection (n = 46 neurons) or formalin injection in the hindpaw ipsilateral (n = 67 neurons) or contralateral (n = 51 neurons) to the brain hemisphere implanted with a GRIN lens. (o-q) Formalin induced fluctuations in LH PV neuronal activity in each phase of the formalin test. Mann-Whitney U-tests with Holm-Sidak correction for multiple comparisons revealed significantly higher Ca 2+ event frequency following contralateral formalin injection in the (o) acute (p=0.048), (p) interphase (p=0.0078), and (q) inflammatory phases (p=0.048) relative to no injection, whereas no significant differences were found between ipsilateral formalin and no injection (acute p=0.80, interphase p=0.18, inflammatory p=0.86). Lines and error bars indicate mean ±95% CI. See also    cluster, and the area under the curves of the fluorescent traces of all 33 neurons did not significantly correlate between sessions, suggesting that the responses of LH PV neurons were generally variable across testing (Figure 1-figure supplement 2). Together, these results demonstrate that LH PV neuronal activity is modulated in response to acute thermal stimuli.
We next tracked LH PV neuronal activity over a longer timescale in response to a hindpaw injection of the chemical irritant formalin. Formalin induces discrete acute (0-5 min) and inflammatory (15-45 min) phases of pain behavior, separated by a brief interphase period (Alhadeff et al., 2018;Dubuisson and Dennis, 1977), allowing us to monitor changes in neuronal activity during each phase. Relative to recording sessions without formalin injection, we observed that the frequency of deconvolved Ca 2+ transients ( Figure 1m) appeared to be generally higher following formalin injections in the hindpaw contralateral to the brain hemisphere in which the GRIN lens was implanted (Figure 1n), and statistical analyses of Ca 2+ event frequency within each period supported this observation ( Figure 1oÀq). Cell registration revealed that only three neurons were detected in more than three of these imaging sessions, thus we could not examine whether a neuron being in cluster 1 or 2 in the hot plate and cold plate tests impacted its response properties in the formalin tests (Figure 1-figure supplement 2). Together, these findings indicate that LH PV neurons display changes in spontaneous activity in response to several stimulus modalities, including both acute thermal stimuli and ongoing chemical inflammation.

LH PV neurons regulate sensory and affective aspects of pain over long timescales
We next examined whether manipulating LH PV neuronal activity can alter noxious stimulation-suppressed behavior and negative affect, which may be better indicators of clinical utility than stimulusevoked behaviors (e.g., reflexive withdrawal to acute thermal stimuli). To investigate this, we targeted these neurons for chemogenetic manipulations by bilaterally injecting Cre recombinasedependent viral vectors driving the expression of either the excitatory designer receptor hM3D, the inhibitory designer receptor hM4D, or the fluorophore mCherry as control into the LH of Pvalb Cre transgenic mice (Figure 2a). Activation of the designer receptors via administration of the ligand clozapine-N-oxide (CNO, 1 mg/kg, i.p.) evoked significant increases and decreases in PWL HP in LH PV : hM3D and LH PV :hM4D mice, respectively, as compared to mCherry controls, with effects detectable between 1 and 18 hr post-injection ( Figure 2b). Thus, chemogenetic manipulations of these neurons alter nociception over a long timescale.
Pain in basic research is traditionally assessed by measuring 'pain-stimulated behavior' or the elicited reactions to noxious stimuli (e.g., paw withdrawal). However, clinical pain disorders often impact quality of life more profoundly by deterring actions normally performed when healthy (Cleeland and Ryan, 1994;Dworkin et al., 2005). Therefore, we next examined the effects of LH PV neuronal activity in a model of 'pain-suppressed behavior,' which measures a decrease in behavioral output following a noxious stimulus (Negus et al., 2006). Healthy mice normally collect nestlet pieces distributed throughout the home cage within 30 min and begin nest building, a natural behavior (Negus et al., 2015;Diester et al., 2021a;Diester et al., 2021b;Figure 2c); this was not affected by CNO administration across groups (Figure 2d, 'CNO + saline'). However, administration of acetic acid (0.6%; i.p.) significantly decreases nesting behavior; this was apparent in all three groups without LH PV manipulations ( Figure 2d, 'Vehicle + acid'). Interestingly, activation of LH PV neurons in LH PV :hM3D mice prior to acetic acid injection prevented the reductions in nesting behavior ( Figure 2d, 'CNO + acid') as compared to tests without CNO and to LH PV :Ctrl mice. In contrast, no changes were observed in LH PV :hM4D mice. Thus, LH PV activation not only decreases noxious stimulus-evoked behavior but also restores behaviors normally suppressed by noxious stimulation.
Pain results not only in overt behavioral changes but also negative affect, as made evident by the high comorbidity between pain and mood disorders (Asmundson and Katz, 2009;Elman et al., 2013). We sought to determine the role of LH PV neuronal activity on the affective, or emotional, component of a painful experience. For this, we used a place conditioning paradigm in which mice avoid a context paired with an aversive event (Johansen et al., 2001;Alhadeff et al., 2018;Figure 2e). After assessment of initial side preference of a two-chamber apparatus, we passively conditioned the mice by administering CNO (i.p.) with intra-plantar formalin to induce inflammation in the initially preferred side and CNO with intra-plantar saline in the initially less-preferred side. Mice were conditioned twice in each context on alternating days and then were given Chemogenetic activation and inhibition of LH PV neurons evoked long-lasting significant increases and decreases in thermal pain thresholds, respectively (n = 11 mice per group; twoway mixed-model ANOVA group Â time interaction, F(16, 240)=14.15, p<0.0001). Significant differences from LH PV :mCherry mice were determined by Bonferroni multiple comparisons tests and are represented graphically, *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. (c) Schematic for pain-suppressed nesting assay. (d) Chemogenetic activation of LH PV neurons prevented the reductions in nesting behavior induced by i.p. injection of 0.6% acetic acid (10 ml/kg). Two-way mixed-model ANOVA revealed a significant group Â test interaction (n = 11 mice per group; F(4, 60) = 4.17, p=0.0048). Bonferroni multiple comparisons post-tests revealed no differences in normal nesting behavior from clozapine-N-oxide (CNO) injections (p=0.92), and that acetic acid injection decreased nesting behavior across groups when administered without CNO (p<0.0001). Administration of CNO before acetic acid increased nesting behavior in LH PV :hM3D mice relative to LH PV :mCherry control mice (p=0.008; Cohen's d = 1.09) and tests without CNO (p=0.0002). (e) Schematic of the formalin place conditioning experiment. (f) Chemogenetic modulation of LH PV neurons altered the effects of formalin on place conditioning. Two-way mixed-model ANOVA revealed a significant group Â test interaction (n = 11 mice per group; F(2,28) = 3.89, p=0.032). Bonferroni multiple comparisons post-tests showed significant shift in chamber preference in LH PV :mCherry (p=0.0001) and LH PV :hM4D mice (p<0.0001) but not Figure 2 continued on next page free access to both chambers during a post-test to assess changes in place preference. As expected, LH PV :Ctrl mice lost preference to the formalin-paired context as compared to pre-formalin preference levels ( Figure 2f). However, activation of LH PV neurons during conditioning attenuated this loss of place preference, whereas inhibition of LH PV neurons during conditioning permitted the loss of place preference ( Figure 2f). Furthermore, the time spent paw licking in these sessions was bidirectionally affected by chemogenetic LH PV neuronal activation or inhibition (Figure 2g, h). LH PV neuronal activation decreased paw licking during the acute but not inflammatory phase, whereas inhibition increased paw licking in the inflammatory but not acute phase. Together, these results support a role for LH PV neurons both in pain behaviors and associated negative affect.

Optogenetic activation of LH PV neurons attenuates persistent inflammatory pain-associated behaviors
Since LH PV neurons ameliorated moderately long-lasting behavioral effects of pain, we next sought to determine whether they could also alter nociceptive thresholds in traditional models of persistent pain behavior. We targeted LH PV neurons for optogenetic manipulations with bilateral injections of a Cre recombinase-dependent viral vector driving the expression of either channelrhodopsin (ChR2: tdTomato; light-sensitive neuronal activator) or GFP (control fluorophore) in the LH of Pvalb Cre transgenic mice and implanted optical fibers bilaterally above these neurons ( Figure 3a). Activation of LH PV neurons in naive mice significantly increased paw withdrawal latency in response to a 51˚C hot plate (PWL HP , Figure 3b). However, activating these neurons did not change paw withdrawal threshold in the von Frey filament test (PWT VF , Figure 3c), suggesting that these neurons regulate acute thermal but not mechanical nociception in healthy mice. Since LH PV neurons are glutamatergic Kisner et al., 2018) and activation of LH neurons expressing the vesicular glutamate transporter 2 (SLC17A6; LH VGLUT2 ) is aversive (Jennings et al., 2013), we also assessed the effects of LH PV neuronal activation in a real-time place preference (RTPP) assay in which photostimulation was paired with one-half of the behavioral arena. Activation of LH PV neurons was mildly aversive as mice spent significantly less time on the photostimulation-paired side (Figure 3d), suggesting that these neurons may play a role in reward-and aversion-like behaviors. Next, we injected complete Freund's adjuvant (CFA), a well-known inflammatory reagent (Alhadeff et al., 2018;Fehrenbacher et al., 2012;Nagakura et al., 2003), into the right hindpaw to cause inflammation and induce persistent hypersensitivity. We observed a significant decrease in nociceptive thresholds for both thermal and mechanical stimuli following these CFA injections ( Figure 3-figure supplement 1a, b). Interestingly, activation of LH PV neurons after CFA evoked significant increases in both PWL HP and PWT VF (Figure 3e, f). Additionally, activation of LH PV neurons no longer triggered place avoidance ( Figure 3g). Furthermore, we observed that the magnitude of the PWL HP response depends on the photostimulus frequency ( Figure 3-figure supplement 1c) and that LH PV neuronmediated antinociception was not strictly photostimulus-bound ( Figure 3-figure supplement 1d) as the antinociceptive effects persisted for several minutes after photostimulation ceased. Together, these results indicate that LH PV neuronal activation attenuates hypersensitivity to both thermal and mechanical stimuli following the onset of inflammation.
LH PV neurons target excitatory circuits within the vlPAG to regulate pain behaviors LH PV neurons send dense projections to the vlPAG (Kisner et al., 2018;Celio et al., 2013), where they form functional excitatory synapses. We next examined whether this LH PV !vlPAG pathway also regulates nociception in models of persistent pain behavior. To specifically target and manipulate the LH PV !vlPAG pathway, we bilaterally injected a Cre recombinase-dependent viral vector driving LH PV :hM3D mice (p=0.094). (g) Time spent paw licking was altered by LH PV neuronal modulation (n = 10 LH PV :mCherry, 7 LH PV :hM3D, and 10 LH PV : hM4D mice; two-way mixed-model ANOVA group Â time interaction, F(22, 264) = 1.99, p=0.0064). (h) Acute and inflammatory phase paw licking were differentially altered by LH PV neuronal activation and inhibition (n listed above; two-way mixed-model ANOVA group Â phase interaction, F(2, 24) = 4.33, p=0.025). Activation of LH PV neurons in LH PV :hM3D mice decreased acute (p=0.0004, Cohen's d = 2.07) but not inflammatory phase paw licking (p=0.50), whereas LH PV neuronal inhibition in LH PV :hM4D mice increased inflammatory (p=0.0049, Cohen's d = 1.29) but not acute phase paw licking (p>0.99). . However, in contrast to somatic manipulations, no effects were observed for either LH PV !vlPAG activation or inhibition in the RTPP test ( Figure 4f), suggesting that there were no changes in the overall affective state of the mice that may have contributed to these bidirectional effects on nociception. In healthy mice, we also observed that the magnitude of the PWL HP response during activation of the LH PV !vlPAG pathway depends on photostimulus frequency. Moreover, these responses were not affected by systemic administration of the cannabinoid receptor 1 (CNR1 or CB1) antagonist/inverse agonist rimonabant (3 mg/kg, i.p.; Figure 4g) despite the PAG being an important site for cannabinoid-mediated antinociception (Esmaeili et al., 2017;Finn et al., 2003;Maione et al., 2006). These results suggest that blocking CB1 receptors does not affect antinociception driven by LH PV !vlPAG circuitry. We next investigated the effects of activating the LH PV !vlPAG pathway in the spared nerve injury (SNI) model of neuropathy. Five days post-SNI, we observed significant decreases in thermal and mechanical thresholds (  Figure 4j). Due to the modest effects of LH PV !vlPAG activation on mechanical thresholds, we predicted that this pathway may be more effective during inflammatory than neuropathic conditions. Therefore, in a new cohort of mice, we activated the LH PV !vlPAG pathway before ( Together, these results show that the LH PV !vlPAG pathway regulates nociception in at least two models of persistent pain behavior and that its activation is more effective in attenuating inflammatory than neuropathic hypersensitivity.

Pvalb
Within the vlPAG, GABAergic and glutamatergic neurons play opposing roles in regulating nociception and defensive behavior (Samineni et al., 2017;Tovote et al., 2016). Although we previously showed that LH PV neurons form functional excitatory synapses with vlPAG neurons , the identity of these post-synaptic targets remains unknown. Thus, we used a monosynaptic retrograde viral tracing strategy with a modified rabies virus (Wickersham et al., 2007a;Wickersham et al., 2007b) to identify the targets of LH PV neurons in the vlPAG. In Slc17a6-Cre and Slc32a1 Cre mice (Vong et al., 2011), we injected starter cells in the vlPAG with Cre recombinase-dependent helper virus containing rabies glycoprotein G and the EnvA receptor for avian sarcoma leukosis virus (TVA) to express the proteins required for uptake and monosynaptic showed that LH PV :ChR2 mice had significantly higher PWL HP during the photostimulation epoch than LH PV :Ctrl mice (p=0.0001; Cohen's d = 2.08). (f) Optogenetic activation of LH PV neurons in mice 6 days following CFA injection triggers increases in PWT VF (n = 9 ChR2 mice and 10 Ctrl mice). Two-way mixed-model ANOVA revealed a significant group Â epoch interaction (F(2, 34) = 11.28, p=0.0002), and Bonferroni multiple comparisons post-test showed that LH PV :ChR2 mice had significantly higher PWL HP during the photostimulation epoch than LH PV :Ctrl mice (p=0.003; Cohen's d = 1.11). (g) LH PV :ChR2 mice did not display significant real-time place avoidance to photostimulation relative to controls 7 days post-CFA, p=0.75 (n = 9 ChR2 mice and 10 Ctrl mice). See also propagation of modified rabies virus ( Figure 5a). Three weeks later, we injected the EnvA-pseudotyped G-deleted rabies virus RVdG-mCherry(EnvA) into the vlPAG. After an additional 3 weeks, mice were perfused, and brains were processed for histological assessment. LH-containing sections were immunostained with an anti-parvalbumin antibody and imaged using confocal microscopy ( Figure 5b). Quantitative analyses revealed that more vlPAG VGLUT2 neurons (14.89%; n = 63 of 423 neurons, three mice) than vlPAG VGAT neurons (6.96%; n = 22 of 316 neurons, six mice) are synaptically targeted by LH PV neurons (Chi-square = 11.18, ***p=0.0008, Figure 5c). Since activation of glutamatergic neurons in the PAG was shown previously to decrease pain (Samineni et al., 2017), our findings suggest a potential role for LH PV neurons as an excitatory input to glutamatergic vlPAG circuitry.
To determine whether other lateral hypothalamic circuits encode for nociception, we examined the effects of manipulating LH leptin receptor expressing (LH LEPR ) neurons, which also project to the vlPAG, albeit a slightly more posterior region (Schiffino et al., 2019;Leinninger et al., 2009). In contrast to LH PV neurons, LH LEPR neurons are predominantly GABAergic and their vlPAG axonal projections are more broadly distributed than those of LH PV neurons. Cre-dependent viruses driving ChR2 or GFP expression were injected into the LH of Lepr Cre mice (Leshan et al., 2006; Figure 4figure supplement 3a), and we found that activation of the LH LEPR !vlPAG pathway potentiated both thermal and mechanical nociception in healthy mice (Figure 4-figure supplement 3bÀe). Moreover, activation of the LH LEPR !vlPAG pathway was rewarding as LH LEPR :ChR2!vlPAG mice spent more time on the photostimulation-paired side of the chamber than LH LEPR :Ctrl!vlPAG control mice (Figure 4-figure supplement 3f). These results demonstrate that activation of lateral hypothalamic glutamatergic (LH PV ) and GABAergic (LH LEPR ) populations that project to the vlPAG attenuates and potentiates nociception, respectively.  ). (f) Optogenetic activation or inhibition of the LH PV !vlPAG pathway did not affect real-time place preference behavior in naive mice (n = 7 mice per group; one-way ANOVA, F(2, 18) = 0.28, p=0.76). (g) LH PV !vlPAG activation-induced antinociception is dependent on photostimulus frequency but is not attenuated by the CB1 receptor antagonist rimonabant (3 mg/kg, i.p.; '3 RIM'). Two-way mixed-model ANOVA revealed a significant group Â epoch interaction (n = 6 ChR2 mice and 7 Ctrl mice; F(7, 77) = 14.27, p<0.0001). Bonferroni multiple comparisons post-tests revealed between-group differences during the '50 Hz' and '50 Hz + 3 RIM' epochs (p<0.0001), but no within-group differences between these epochs (p>0.99). (h) Optogenetic activation of the LH PV !vlPAG pathway evokes increases in PWL HP on day 5 post-spared nerve injury (SNI) (n = 7 mice per group; two-way mixed-model ANOVA group

Activation of LH PV axonal projections to the LHb triggers aversion
LH PV neurons also target other brain regions including the LHb (Kisner et al., 2018;Celio et al., 2013). Therefore, we examined whether LH PV neurons also modulate nociceptive processing via projections to the LHb. For this, we bilaterally injected a Cre recombinase-dependent viral vector driving the expression of either channelrhodopsin (ChR2:tdTomato) or the fluorophores GFP or tdTomato (control) into the LH of Pvalb Cre mice and implanted optical fibers bilaterally above the LHb to specifically activate the LH PV !LHb pathway ( Figure 6a). Interestingly, activation of this LH PV !LHb circuitry did not evoke changes in nociceptive responses to an acute noxious thermal or mechanical stimulus in healthy mice (Figure 6bÀd) or in mice with SNI-induced neuropathy when tested at 5 and 25 days post-surgery ( Figure 6-figure supplement 1). Because the LHb is a brain region associated with reward-and aversion-related behaviors (Stamatakis et al., 2016;Faget et al., 2018), we also sought to determine whether activation of the LH PV !LHb pathway triggers such behaviors. We found that activation of this pathway in healthy mice was aversive as mice spent significantly less time on the photostimulation-paired side in the RTPP assays (Figure 6eÀg). These results are consistent with previous findings demonstrating that broad activation of lateral hypothalamic glutamatergic axonal projections in the LHb is aversive (Stamatakis et al., 2016). Together, these findings demonstrate that LH PV neurons encode for distinct behavioral outputs depending on their targeted downstream regions: nociceptive processing via projections to the vlPAG and aversion-related behaviors through connections to the LHb.

Antinociceptive interactions between LH PV neuronal activation and morphine
Since activation of LH PV neurons appears to reduce nociception as monotherapy, we last sought to examine the interaction between the antinociception induced by these neurons and the m-opioid pain reliever morphine. For this, we performed a dose-addition analysis of CNO and morphine in LH PV :hM3D and LH PV :Ctrl mice. First, we determined the individual dose-response curves of CNO and morphine using a cumulative dosing procedure. As expected, CNO evoked dose-dependent PWL HP increases in LH PV :hM3D (Figure 7a) but not LH PV :Ctrl mice (Figure 7b), whereas morphine produced dose-dependent increases in both groups (Figure 7a, b). Next, the two drugs were combined in fixed proportions (1:1, 1:3, and 3:1) according to their relative potencies (ED 50 ) in the LH PV : hM3D group. For example, the 1:1 ratio consisted of one unit of the morphine ED 50 (10.31 mg/kg) for every one unit of the CNO ED 50 (0.78 mg/kg). Fractions of these mixtures (e.g., the combined 0.125 Â, 0.25 Â, 0.5 Â, and 1 Â ED 50 values of morphine and CNO) were administered consecutively by a cumulative dosing procedure to complete one dose-response curve test (Figure 7a, b). The shared dose-response curves were used to calculate the ED 50 of each drug within each mixture; these equi-effective points were plotted on an isobologram to visualize the nature of each interaction (Figure 7c, d). For LH PV :hM3D mice, 1:3 and 1:1 morphine:CNO combinations fell within the range of additivity. Remarkably, the 3:1 morphine:CNO combination fell below the range of additivity, suggesting synergistic interactions between morphine and LH PV neuronal activation, indicating that activation of LH PV neurons enhanced the antinociceptive potency of morphine. Formal statistical comparison of expected and experimental ED 50 values confirmed this observation (Student's paired t-test: t(7) = 2.92, p=0.022). For LH PV :Ctrl mice, no combinations significantly differed from the range of additivity, suggesting that CNO did not affect the antinociceptive potency of morphine in control subjects. Finally, using the same mice, we investigated the effects of LH PV neuronal activation following the development of morphine tolerance. We administered morphine (32 mg/kg, i.p.) twice per day for 3 days, which caused a significant decrease in morphine-induced antinociception (Figure 7e). On day 5, CNO (1 mg/kg) evoked a significant increase in PWL HP and restored morphine-induced antinociception as compared to control mice (Figure 7f). We then treated these mice once per day over the following three days with a combination of 1 mg/kg CNO and 32 mg/kg morphine to assess the potential development of tolerance to this combination. However, no differences were observed on the day 9 test in LH PV :hM3D mice as compared to day 5 ( Figure 7f). Thus, activating LH PV neurons not only increases morphine potency acutely but also rescues morphine tolerance and may prevent subsequent tolerance development.

Discussion
The LH is an important site for numerous survival-critical processes such as sleep, feeding, and reward (Carter et al., 2009;Bonnavion et al., 2016;Stuber and Wise, 2016). New technologies have enabled the identification of specific lateral hypothalamic populations associated with certain behaviors and the understanding of how the activity of such neurons drives behavior and relates to external factors. However, the cell types mediating many other LH-associated behaviors have received less attention. Nociception has historically been a less LH-prototypical process than one such as feeding, but LH circuits were nevertheless previously shown to respond to noxious stimuli, to control nociception, and to affect downstream circuits in the PAG, a critical brain region for pain regulation (Cox and Valenstein, 1965;Lopez et al., 1991;Dafny et al., 1996;Fuchs and Melzack, 1995;Behbehani et al., 1988). Cell-type-specific optogenetic manipulations showed that a small cluster of fast-spiking glutamatergic LH PV neurons projects to the vlPAG and modulates acute nociception in a m-opioid-independent manner . However, much remained to be learned as to how LH PV neurons respond to noxious events and whether they could be targeted for therapies in scenarios outside of acute sensory stimulation.
One of the great challenges in understanding how dynamics in neuronal circuits control behavioral output is to determine when specific cell types are active, as well as the nature of the . Each point represents the ED 50 ± 95% CI of each drug alone or in a mixture; ordinates represent the ED 50 value of morphine and abscissae represent the ED 50 value of CNO. In LH PV :hM3D mice, the 3:1 morphine:CNO mixture was significantly more potent than predicted by the hypothesis of additivity (paired Student's t-test, t(7) = 2.92, p=0.022). (e) Both groups of mice developed significant antinociceptive tolerance to 32 mg/kg morphine when administered twice per day for 3 days. Three-way mixed-model ANOVA revealed a significant morphine Â test interaction (n = 8 mice per group; F(1, 14) = 134.7, p<0.0001), and Bonferroni multiple comparisons post-tests showed the antinociceptive effects of 32 mg/kg morphine were significantly lower on day 4 than day 1 (both p<0.0001). (f) Activation of LH PV neurons restored morphine potency, and further tolerance did not develop to combination treatment. Three-way mixed-model ANOVA revealed a significant treatment Â group interaction (n = 8 mice per group; F(2, 28) = 42.10, p<0.0001). Bonferroni multiple comparisons post-tests revealed that there were between-group differences in PWL HP evoked on day 5 by CNO (p=0.0006) and morphine (p<0.0001) and on day 9 by CNO (p=0.016) and morphine (p<0.0001). However, no within-group differences were observed between day 5 and 9 in LH PV :hM3D mice during CNO (p>0.99) or morphine treatment (p>0.99).
relationship between this activity and behavior. Although direct manipulations of neuronal activity followed by behavioral examination are important for understanding this relationship, measuring changes in the activity patterns of neurons in awake behaving mice provides information as to how this circuit functions in the absence of experimenter-driven input. Using functional imaging to measure calcium dynamics, we gained insight as to how the activity of LH PV neurons correlates with nociception and show for the first time that LH PV neurons exhibit an array of time-locked responses to acute noxious thermal events. The involvement of LH PV neurons in holding information related to noxious events is supported by the finding that the neuronal activity could be used to decode noxious from innocuous stimuli. LH PV neuronal activity was also altered during formalin-induced inflammation. Formalin injection, which causes discrete phases of acute and inflammatory forms of pain behavior, into the hindpaw contralateral to the imaged LH hemisphere evoked increases in calcium transient frequencies that were more pronounced than those observed during injection of the paw ipsilateral to the imaged LH hemisphere, likely reflecting decussation of the nociceptive signal at the spinal level (Dafny et al., 1996;Yoshida et al., 2019;Yamada et al., 2012). It is worthwhile to note the advantages of using a single-photon miniscope, which enables single-cell resolution of neuronal activity. Other methods such as fiber photometry would likely not have revealed the changes we observed in LH PV neuronal activity to acute thermal stimuli, for which there were heterogeneous responses across neurons, as well as to formalin, which was reflected as an elevated rate of calcium transients that were asynchronous across neurons. Together, our functional imaging data suggest that LH PV neurons may become active during noxious events to signal or suppress nociception in mice.
In chemogenetic experiments designed to assess the broader therapeutic potential of LH PV neuron manipulation in pain disorders, we found that these neurons can bidirectionally modulate thermal nociception over long timescales, and thus may represent potential targets for extended duration analgesia. Moreover, activation of LH PV neurons significantly attenuated acetic acidreduced nesting behavior, demonstrating that this manipulation not only decreases sensory pain but also permits the resumption of species-specific natural behaviors that are suppressed by noxious events. This model may be analogous to clinical interventions that allow patients undergoing chronic pain to resume daily activities such as exercising or performing occupational duties as opposed to removing pain at the expense of a reduced motivational capacity. We also observed that inhibition of LH PV neuronal activity in hM4D-expressing mice did not decrease nesting behavior in control tests, suggesting that inhibition of these neurons does not cause pain directly, but likely rather enhances sensitivity to noxious stimuli. In support of this, LH PV neuronal activation reduced formalinassociated negative affective pain, whereas this was nearly enhanced by inhibiting LH PV neurons. Behavioral scoring showed that sensory pain behavior was also bidirectionally modulated in this experiment, suggesting that LH PV neurons modulate both sensory and affective experiences.
Optogenetic activation of LH PV neurons decreased both thermal and mechanical nociception following the induction of a commonly used inflammatory pain model. These effects were likely attributable to LH PV projections to the vlPAG as LH PV !vlPAG activation decreased thermal and mechanical thresholds in neuropathic and inflammatory models. In contrast, projections of LH PV neurons to the LHb regulated aversion as previously shown for the broader LH glutamatergic population (Stamatakis et al., 2016), but not nociception, suggesting that LH PV neurons regulate different behavioral outputs via different downstream projection areas. While systemic CB1 or m-opioid antagonism does not affect LH PV !vlPAG activation-induced antinociception , the finding that sustained antinociception following extended LH PV somatic or LH PV !vlPAG activation suggests that LH PV neurons may co-release neuropeptides that interact with downstream receptors to attenuate nociception or that other efferent circuits for antinociception are recruited during such activation that may function to decrease nociception. Importantly, in conjunction with the functional imaging data, the increased sensitivity observed upon inhibition of LH PV somas or the LH PV !vlPAG pathway suggests that LH PV neurons may become active in response to a noxious stimulus to decrease its severity. It is important to briefly note that while a study found that sustained activation of archaerhodopsin evokes spontaneous synaptic release in ex vivo preparations (Mahn et al., 2016), this phenomenon has not been observed during in vivo electrophysiological recordings or precluded the observation of behavioral effects in the direction associated with the loss of presynaptic input when using photoinhibition times equal to or longer than the ones we employed here (Jennings et al., 2013;Rozeske et al., 2018).
Although LH PV neurons are functionally connected to neuronal circuits within the vlPAG, the heterogeneous behavioral effects driven by the intermingled vlPAG neuronal populations made it challenging to draw a clear circuit map from LH PV neurons to behavior through the vlPAG pathway. For instance, activation of vlPAG glutamatergic and GABAergic neurons decreased and increased nociception, respectively (Samineni et al., 2017). Therefore, we used a retrograde monosynaptic rabies tracing strategy to identify the preferred post-synaptic vlPAG targets of LH PV neurons. We found a higher proportion of LH PV neurons labeled following uptake of RVdG-mCherry(EnvA) in vlPAG VGLUT2 compared to vlPAG VGAT neurons, suggesting that LH PV neurons may preferentially, yet not exclusively, target glutamatergic vlPAG neurons. In the context of previous work, excitatory input from LH PV neurons to vlPAG VGLUT2 neurons would thus form a discrete antinociceptive pathway. However, experiments using techniques such as ChR2-assisted circuit mapping (CRACM) from LH PV :ChR2+axonal projections onto postsynaptic vlPAG neurons followed by single-cell RT-qPCR analysis will be needed to elucidate how LH PV neurons regulate vlPAG microcircuitry and how activation of this LH PV !vlPAG pathway modulates nociceptive responses to noxious stimuli. The opposing behavioral outcomes during photostimulation of a GABAergic LH population, LH LEPR neurons, in the vlPAG further demonstrate the complex, heterogeneous nature of LH!PAG pathways and highlight the need for future circuit characterizations. Moreover, it is still unknown whether LH PV axonal projections to their target regions follow a one-to-one or one-to-many architecture. This is certainly an important question that has yet to be determined. However, our behavioral data suggest that these might be independent LH PV populations since we did not observe aversive-like effects during LH PV !vlPAG stimulation or antinociception during LH PV !LHb stimulation. Furthermore, the LH PV !LHb data also demonstrate that the antinociception evoked by activating the LH PV !vlPAG pathway was not due to antidromic stimulation effects. Future experiments will be needed to determine that these are indeed independent populations of LH PV neurons.
In a final series of experiments, we investigated the antinociceptive interactions between LH PV neuronal activation and the m-opioid receptor agonist morphine. For a novel analgesic therapy to be useful, it must meet one of these three criteria: (1) possess analgesic properties alone, (2) facilitate analgesic action of existing treatments, or (3) decrease unwanted effects of existing treatments to make them more suitable for extended use (Li and Zhang, 2011). Our observation that LH PV neuronal activation attenuates nociception suggests that this manipulation meets the first criterion. Therefore, our last experiments were designed to assess the remaining criteria. To address the second criterion, we performed a dose-addition analysis between morphine and LH PV DREADD receptor activation by CNO as a standard pharmacological agent. We found that, depending on the proportion of drugs in the mixture, LH PV neuronal activation and morphine produced additive to synergistic interactions on thermal antinociception. The combination exhibiting the highest level of synergism required only a small stimulation of LH PV neuronal activity to greatly enhance morphine's potency. Importantly, we included a group of control mice without hM3D receptors, in which CNO did not alter morphine's potency. To address the third criterion above, we last investigated the effects of activating LH PV neurons following the development of tolerance to morphine-induced antinociception effect in the hot plate test. Here, chemogenetic activation of LH PV neurons evoked significant antinociception in morphine-tolerant mice, and more importantly, significantly restored morphineinduced antinociception. Remarkably, similar antinociceptive effects were maintained through another period of concurrent LH PV neuronal activation and morphine administration. Together, these findings show that LH PV neuronal activation can synergistically enhance acute morphine antinociception and restore its antinociceptive effects following the development of tolerance. Thus, activation of these LH PV neurons could be used to reduce the effective antinociceptive dose of morphine, helping to attenuate unwanted side effects such as respiratory depression and slow the rate of morphine tolerance.
An important point warranting further discussion is the contrast between our observations of divergent clusters of response patterns in LH PV neuronal activity during the hot and cold plate tests and the dominant behavioral phenotype of pain suppression when bulk activating these neurons. First, there is precedent to this contrast between diverse spontaneous activity patterns and more uniform behaviors driven by causal manipulations. For instance, LH GABA neurons responded to food locations, appetitive, or consummatory behaviors in a heterogeneous manner by either increasing or decreasing in activity at each event, yet when these neurons were bulk activated, mice ate voraciously, and when these neurons were ablated, mice ate less (Jennings et al., 2015). In the current study, we observed that most neurons identified as cluster 1 or 2 in the hot plate assay did not maintain the same designation in the cold plate assay, and vice versa. Therefore, it seems likely that, in general, the responses of LH PV neurons are either not consistent over time or dependent on the type of stimulus applied. For instance, some LH PV neurons may specifically suppress heat pain, others may suppress cold pain, others may suppress chemical pain, and so on. As such, we think that the bulk activation of LH PV neurons with optogenetics or chemogenetics during stimulation with one specific noxious stimulus (e.g., heat) likely activates a stimulus-specific cluster of LH PV neurons (heat) as well as the clusters specific for other stimuli (cold, chemical, mechanical, etc.) to evoke antinociception, as opposed to having populations of LH PV neurons that are exclusively pronociceptive or antinociceptive, the effects of which could be potentially diluted during bulk activation of these neurons. However, further work will be needed to elucidate how noxious stimuli are responded to and encoded by LH PV neuronal activity.
Here, we provide a detailed characterization of LH PV neurons, clearly demonstrating that these neurons modulate nociception through a distinct downstream circuit. Moreover, we measured and correlated LH PV neuronal activity patterns during noxious events. Finally, we found that chemogenetic modulation of these neurons could potentially be used as a standalone analgesic therapy or in combination with current analgesics such as morphine. These results support the continued investigation of LH PV neurons as a target for novel analgesics and warrant new efforts to identify neuronal populations in humans for targeting in clinical settings. Further information and requests for resources and reagents should be directed to and will be fulfilled by Yeka Aponte (yeka.aponte@nih.gov).

Experimental model and subject details Animals
All experimental protocols were conducted in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and with the approval of the National Institute on Drug Abuse and Michigan State University Animal Care and Use Committees. Male and female heterozygous Pvalb Cre mice (RRID:IMSR_JAX:008069; C57BL/6J background, The Jackson Laboratory, Bar Harbor, ME, USA), Slc32a1 Cre mice (RRID:IMSR_JAX:028862, C57BL/6J background, The Jackson Laboratory), Slc17a6 Cre mice (RRID:IMSR_JAX:028863, C57BL/6J background, The Jackson Laboratory), and Lepr Cre mice (RRID:IMSR_JAX:032457; C57BL/6J background, kindly provided by M.G. Myers Jr., University of Michigan Medical School, MI, USA) were used in this study. Mice were maintained at the National Institute on Drug Abuse animal facility under standard housing conditions. Up to five mice of the same sex were group housed under a 12 hr light-dark cycle at 20-24˚C and 40-60% humidity with free access to water and food (PicoLab Rodent Diet 20, 5053 tablet, LabDiet/ Land O'Lakes Inc, St. Louis, MO, USA). For behavior experiments, 6-to 8-week-old male and female mice (~18-25 g) were randomly assigned to experimental groups while maintaining littermate or age-matched and gender-matched controls. Following stereotaxic surgeries, mice were individually housed.
In all experiments, biological replicates were defined as 'parallel measurements of biologically distinct samples that capture random biological variation,' and technical replicates were defined as 'repeated measurements of the same sample that represent independent measures of the random noise associated with protocols or equipment' (Blainey et al., 2014).

Surgical procedures
For in vivo functional imaging experiments, mice were anesthetized with isoflurane and placed onto a stereotaxic apparatus (David Kopf Instruments, Tujunga, CA, USA). After exposing the skull by a minor incision, a small hole (<1 mm diameter) was drilled unilaterally (bregma, À1.78 mm; midline, +1.38 mm) for virus injection and GRIN lens insertion. A sterile, beveled 25-gauge needle was inserted into the center of the craniotomy stopping approximately 50 mm above the dorsal-ventral coordinate for the lens implant and remaining in place for 4-5 min to create a path for the implant. Next, rAAV2/9-CAG-FLEX-GCaMP6s-WPRE-SV40 was injected offset from the center of the craniotomy (100 nl; rate: 25 nl/min; RRID:Addgene_100842; Addgene viral prep 100842-AAV9; titer: 5.0 Â 10 12 GC/ml) into the LH of Pvalb Cre mice (bregma, À1.78 mm; midline, +1.365 mm; skull surface, À5.38 mm) by a pulled glass pipette (20-30 mm tip diameter) with a micromanipulator (Narishige International USA Inc, Amityville, NY, USA) controlling the injection speed. The injection is offset to avoid damaging the tissue in the lens field of view. After injection, a 500-mm-diameter GRIN lens (Snap-in Imaging Cannula Model L-V; Doric Lenses Inc, Qué bec, QC, Canada) was lowered into the center of the craniotomy (bregma, À1.78 mm; midline, +1.38 mm; skull surface, À5.28 mm). Implants were affixed to the skull with C&B Metabond Quick Adhesive Cement System (Parkell, Inc, Edgewood, NY, USA). Subsequently, mice were individually housed for 3À4 weeks for post-surgical recovery and viral transduction.

In vivo functional imaging
A miniature microscope with an integrated LED was used to image GCaMP6s fluorescence in LH PV neurons through an implanted GRIN lens (Basic Fluorescence Snap-In Microscopy System -Deep Brain; Doric Lenses Inc). LH PV :GCaMP6s mice underwent five imaging sessions (hot plate, cold plate, ipsilateral formalin, contralateral formalin, and no formalin). Before each imaging session, GRIN lenses were briefly cleaned with isopropanol and mice were gently restrained while the snap-in microscope was secured to the baseplate for alignment with the implanted GRIN lens. Mice were then given approximately 5 min to acclimate to the microscope and tether. Grayscale TIFF images were collected at 10 frames per second (100 ms exposure) using Doric Neuroscience Studio software version 5.1 (RRID:SCR_018569). The LED power was calibrated between 10% and 50% (0.2-1.2 mW of 458 nm blue light). At the beginning of each session, imaging was synchronized with behavioral video recordings for later alignment. Sample size estimates were derived from a previous study using miniscope recordings in the hypothalamus (Betley et al., 2015).
For the hot plate tests, mice were placed on a 51˚C hot plate (IITC Life Science, Woodland Hills, CA, USA) or a room temperature black cardboard surface with similar visual and tactile properties for 10-12 trials per stimulus. Mice were removed from the hot plate when typical behavioral responses were observed (e.g., paw withdrawal or paw licking). For the cold plate tests, mice were placed on a 4˚C aluminum block or a room temperature white cardboard surface for 8-9 trials per stimulus. Mice were removed from the cold plate when paw checking or withdrawal responses were observed. 1À2 min interstimulus intervals were used for these tests. For the formalin tests, mice received a 20 ml intra-plantar injection of 2% formalin (Cat # 5016-02; Macron Fine Chemicals/Avantor, Radnor, PA, USA) diluted in saline. 47 min videos were captured, and formalin was injected into one of the hindpaws at the 2 min mark. For the 'no injection' test, no formalin was administered. These tests were separated by at least 5 days to minimize photobleaching and inter-test effects.

Image processing
Image analyses were performed using MATLAB scripts available in the Miniscope Analysis pipeline (https://github.com/etterguillaume/MiniscopeAnalysis). First, images were motion-corrected using the Non-Rigid Motion Correction (NoRMCorre) package (Pnevmatikakis and Giovannucci, 2017) and downsampled spatially and temporally by factors of 3. Motion-corrected, downsampled videos were then processed using Constrained Non-negative Matrix Factorization for Endoscopic data (CNMF-E) to extract individual neural segments, denoise their signals, demix signals from nearby neurons, and deconvolve calcium transients for estimation of neuronal firing (Friedrich et al., 2017;Zhou et al., 2018;Pnevmatikakis et al., 2016).

Imaging and behavioral analysis
For the hot plate and cold plate tests, filtered traces were Z-score normalized and smoothed with a rolling average of 3 frames. The 30 s activity traces surrounding stimulus presentations (10 s before to 20 s after) were averaged within each stimulus to form an average peri-stimulus activity trace per neuron. Neurons were assigned into clusters for further analysis if they displayed one of the two following phenotypes: (cluster 1) the average peak Z-score amplitude was !1.0 and the AUC of the trace following the stimulus was positive, or (cluster 2) the average peak Z-score amplitude was À1.0 and the AUC of the trace following the stimulus was negative. The remaining neurons that did not meet either of these criteria were considered non-responsive to the noxious stimuli and were not analyzed further.
For decoding analysis, average traces for each neuron were constructed for hot plate, cold plate, or neutral stimulus trials for the 10 s period following stimulus onset. Principal component analysis was used to reduce the dimensionality of the averaged traces while maintaining 99% of the variance. 80% of the resulting traces from each cluster of neurons (see above) were used to train an SVM classifier in MATLAB (built-in function) to distinguish between activity resulting from a neutral stimulus and a hot or cold plate stimulus. The resulting classifier was used to predict which stimulus generated the remaining traces, and the predictions were compared to the known stimuli labels to determine the accuracy of the classifier. This process was repeated 100 times with random subsets of training data to obtain a distribution of test accuracies. To determine the significance of the test accuracy distribution, the labels of the testing dataset were randomly shuffled 100 times. Each label permutation was compared to the predictions obtained from one of the previously trained classifiers to form a null distribution of chance accuracies. A cumulative Gaussian curve was fit to the cumulative frequency distributions (0.10 bin size) of the test and null accuracies, and the distribution means were compared using the extra sum-of-squares F-test in Prism.
For the formalin experiments, deconvolved signals were used to bin estimated Ca 2+ transients for every 5 min period of the test. For statistical comparison, we averaged the number of events per 5 min period within each phase of the formalin test (0-5 min, acute; 6-15 min, interphase; 16-45 min, inflammatory).

Optical manipulations
Optical fiber implants were coupled to patch cords connected to lasers (Doric Lenses Inc) via rotary joints mounted over behavioral testing areas. Optical fiber implants were custom-made and assessed for output efficiency !80%. Laser output was controlled by Doric Neuroscience Studio software version 5.1 (RRID:SCR_018569). For photostimulation experiments, 450 nm laser diodes were used to deliver 5 ms pulses of 10-15 mW light at a frequency of 5-100 Hz. For photoinhibition experiments, 520 nm laser diodes were used to deliver 10-15 mW of constant light.

Behavioral experiments
Mice were habituated to experimenter handling for 3 days prior to experiments, and all experiments were performed during the light cycle. Mice were acclimated to behavioral rooms for at least 1 hr before experiments began. Across experimental and control groups, mice were gender-matched and age-matched or littermates. By design, sample sizes were 8-12 mice based on (Bolles and Fanselow, 1980) previous literature using similar procedures (Negus et al., 2015;Jennings et al., 2015;Jennings et al., 2013;Siemian et al., 2019;Alhadeff et al., 2018) and (Tovote et al., 2015) estimates of exclusion rates following histology. Mice were excluded from analysis if viral expression and fiber placement were not observed in at least one hemisphere after histological assessment (see Histology).

Pain-suppressed nesting assay
Single-housed mice were tested in their home cages, which were initially supplemented with nestlet. Mice were acclimated to the procedure room for at least 1 hr before testing and had access to food and water in their cages throughout test sessions. At the start of each test, mice were pretreated with saline or CNO (1 mg/kg, i.p.; PUBCHEM:135445691; Cat # 4936; Tocris Bioscience, Minneapolis, MN, USA). After 1 hr, the existing nest was removed from each home cage, and a new nestlet cut into six small, equal-sized pieces was placed into the home cage, distributed across zones divided by a 3 Â 2 grid in the cage. The mouse was then given an i.p. injection of 0.6% acetic acid (PUBCHEM:176; Cat # 320099; Sigma-Aldrich) in saline (10 ml/kg) or saline alone (10 ml/kg) and returned to the home cage. Measurements of the number of nestlet pieces collected were taken at 10, 30, 60, and 100 min post-acetic acid injection (Negus et al., 2015) by an experimenter blinded to the treatment group. The data from the 30 min time point were presented. At least 5 days separated tests to minimize inter-test effects (Negus et al., 2015).

Formalin place conditioning
Place conditioning experiments were performed in a two-chamber apparatus separated by a wall with a small door that could be closed with a divider. The chambers were defined by tactile, visual, and olfactory cues. One chamber had a metal grid floor, walls decorated with tan and black alternating vertical stripes, and almond scent. The other chamber had a smooth white floor, walls decorated with white circles on a tan background, and orange scent. The front wall of each chamber remained clear, and sessions were recorded using video cameras aimed through this wall using ANY-maze software. Pilot experiments showed that mice consistently preferred the metal grid side at a rate of 60-70% per 15 min test. In comparison to using an unbiased design, this biased design permits preassigning groups at surgery with less potential for mismatched side preference at pretest. All mice used in this study preferred the metal grid side in the 15 min pretest on day 1, and this side was assigned for pairing with formalin treatment.
Over the next 4 days, mice received one training session per day with the center door closed and only one chamber accessible; formalin sessions were video recorded for later behavioral scoring of paw licking behavior by a blinded scorer; some videos were difficult to view the mouse to score licking behavior and were removed from this analysis (two mCherry, three hM3D, and one hM4D). All sessions were preceded by an injection of CNO (1 mg/kg, i.p.; Tocris Biosciences) to control for potential subjective effects of LH PV manipulation in the absence of inflammatory pain. On even days (sessions 2 and 4), mice received an intra-plantar injection of saline (20 ml) in the hindpaw and were immediately placed in the initially non-preferred side for 60 min. On odd days (sessions 3 and 5), mice received a 20 ml hindpaw intra-plantar injection of 2% formalin (Cat # 5016-02; Macron Fine Chemicals/Avantor, Radnor, PA, USA) diluted in saline and were placed in the initially preferred chamber for 60 min. The formalin-treated paw was different on each of the two condition sessions. On day 6, untreated mice were placed back in the testing arena with free access to both chambers and the sessions were analyzed with ANY-maze video tracking system v5 (RRID:SCR_014289; Stoelting Co., Wood Dale, IL, USA).

Thermal nociception (hot plate test)
A cylindrical plexiglass enclosure was placed on a 51˚C hot plate (IITC Life Science). For optogenetic experiments, patch cords were connected, and mice were placed in a holding chamber for an initial 3 min period. Mice were gently transferred to the hot plate and the latency to paw withdrawal (PWL HP ) was measured. A latency of 20 s was defined as complete analgesia and used as a cutoff time to avoid tissue injury. Following this measurement, mice were removed from the hot plate and photomanipulations commenced for 3 min in the holding chamber after which mice were placed back on the hot plate for a second PWL HP measurement. Photomanipulations ceased for another 3 min period in the holding chamber before a final PWL HP measurement. For frequency-response experiments, this procedure was repeated for each frequency, except only one 3 min 'laser-OFF' period separated photostimulation epochs. For experiments examining the effects of longer photostimulation, 50 Hz photostimulation was delivered every other second over 20 min, and PWL HP was measured at the end of the photostimulation period and at 5, 10, and 20 min post-photostimulation. For experiments examining the effects of rimonabant on photostimulation-induced antinociception, rimonabant (3 mg/kg, i.p., dissolved in a vehicle of 8% Tween-80 in saline; PUBCHEM:5360515; Cat # 9000484; Cayman Chemical, Ann Arbor, MI, USA) was administered in a volume of 10 ml/kg 30 min prior to photostimulation. For chemogenetic experiments, CNO (1 mg/kg, i.p.; PUB-CHEM:135445691; Cat # 4936; Tocris Bioscience) was administered after the second PWL HP measurement and measurements were taken periodically after (0.5-72 hr).

Mechanical nociception (von Frey test)
Mice were habituated for 20 min in cylindrical plexiglass enclosures on a fine mesh grid floor. For optogenetic experiments, patch cords were connected, and mice were placed in a holding chamber for an initial 3 min period. Von Frey filaments ranging from 0.008 g to 4 g were used to determine paw withdrawal threshold (PWT VF ), which was defined as the lowest strength filament eliciting a behavioral response in at least two out of three applications. Briefly, measurements started with the lowest strength filament, and the filament strength was increased until paw withdrawal responses reliably occurred in at least two out of three applications. This procedure was repeated for each hindpaw in three epochs as described above for PWL HP measurements: pre-photostimulation, photostimulation, and post-photostimulation.
Real-time place preference RTPP sessions were performed in a standard rat cage with opaque black siding filled with a thin layer of clean rodent bedding, except for a subset of LH PV !LHb mice that were also tested in a threechamber apparatus consisting of two identical black-walled chambers separated by a narrow hall section, and the entire apparatus was filled with a thin layer of clean rodent bedding. Patch cords were connected, and mice were placed into the chamber. Photostimulation (50 Hz) or photoinhibition was paired with one side of the chamber, which remained constant across all tests. For LH LEPR !vlPAG experiments, 20 Hz photostimulation was used (Schiffino et al., 2019). Tests lasted for 10 min (LH PV somatic manipulations) or 20 min (axonal projection manipulations). At the end of the sessions, the percentage of time spent on the laser-paired side was calculated by ANY-maze video tracking system v5 (RRID:SCR_014289; Stoelting Co.).

Persistent inflammatory pain
Following initial behavioral tests after stereotaxic surgery and viral transduction, CFA (Cat # F5881; Sigma-Aldrich, St. Louis, MO, USA) was diluted 1:1 in saline and injected (20 ml) into the plantar surface of one hindpaw under brief isoflurane anesthesia (Alhadeff et al., 2018). Behavioral tests resumed 5 days post-CFA.

Persistent neuropathic pain
Following initial behavioral tests after stereotaxic surgery and viral transduction, the SNI model was used for induction of neuropathic pain. Briefly, under isoflurane anesthesia, the tibial and common peroneal nerves were axotomized while the sural nerve was spared (Decosterd and Woolf, 2000;Suter et al., 2003). Behavioral tests resumed 5 days post-SNI.

Dose-addition analysis
For the experiment examining interactions between CNO and morphine, tests were conducted according to a cumulative dosing procedure, in which PWL HP measurements are taken immediately prior to i.p. drug administration, then 60 min after drug administration immediately before the next drug administration. When administered alone, CNO was tested across a dose range of 0.1-3.2 mg/ kg, and morphine was tested across a dose range of 3.2-32 mg/kg. For combination tests, these dose-measurement cycles continued until near 100% maximal effect was achieved corresponding to the predetermined cutoff time of 20 s. Raw PWL HP values for CNO and morphine were transformed into percent maximum possible effect (%MPE) values according to the formula %MPE = [(post-drug PWL HP -pre-drug PWL HP ) / (cutoff time -pre-drug PWL HP ) Â 100]. %MPEs were averaged within each group (± s.e.m.) and plotted as a function of dose. Log(ED 50 ) values were determined from the %MPE dose-response curve via linear regression and averaged within the group to calculate the ED 50 (±95% confidence interval [CIs]) for each drug, except for CNO in the mCherry control group, which did not produce 50% effect levels. Morphine was obtained from the National Institute on Drug Abuse Drug Supply Program (PUBCHEM:5288826).
To examine the antinociceptive interactions between CNO and morphine, a fixed-proportion dose-addition analysis method was used Tallarida, 2010;Negus et al., 2009). For this analysis, CNO and morphine were combined in fixed proportions (1:1, 1:3, and 3:1) and administered using the cumulative dosing procedure as described. The actual doses of the drugs in the combination were determined by the relative potencies of each drug (based on the ED 50 values) in the LH PV :hM3D group. For example, the 1:1 ratio consisted of one unit of the morphine ED 50 (10.31 mg/kg) for every one unit of the CNO ED 50 (0.78 mg/kg). By this method, the 1:3 ratio contained 0.5 Â ED 50 of morphine and 1.5 Â ED 50 of CNO and the 3:1 ratio contained 1.5 Â ED 50 of morphine and 0.5 Â ED 50 of CNO. Fractions of these mixtures (the combined 0.125 Â, 0.25 Â, 0.5 Â, 1 Â, and 2 Â ED 50 values of morphine and CNO) were administered consecutively by the cumulative dosing procedure to complete one dose-effect curve test. At least 1 week separated each test to avoid the development of tolerance and inter-test effects. Furthermore, a morphinealone dose-response curve was taken 1 week after the last combination test, which showed that the morphine ED 50 had not significantly changed (mCherry mice first morphine ED 50 10.51 mg/kg, second morphine ED 50 10.26 mg/kg; hM3D mice first morphine ED 50 10.31 mg/kg, second morphine ED 50 10.45 mg/kg). The shared dose-response curves were used to calculate the ED 50 of each drug within each mixture. Isobolograms plotting the ED 50 values of each drug were constructed to visually represent the nature of the drug interactions as additive, infra-additive, or supra-additive (synergistic).
Dose-addition analysis was performed as described previously Tallarida, 2000). When both drugs were active in an assay, expected additive ED 50 values (±95% CL) (Z add ) were calculated from the equation Z add = fA + (1 À f)B, where A is the ED 50 of morphine alone, B is the ED 50 of CNO alone, and f is the fractional multiplier of A in the computation of the additive total dose (e.g., f = 0.5 when fixed ratio was 1:1). When only one drug was active (i.e., morphine in the mCherry control group), the hypothesis of additivity predicts that the inactive drug (i.e., CNO) should not contribute to the effects of the mixture, and the equation reduces to Z add = A/A, where A is the proportion of morphine in the total drug dose. Experimental ED 50 values (Z mix ) were determined from the 1:3, 1:1, and 3:1 combinations and were defined as the sum of the ED 50 values of both drugs in the combination. Given the within-subject experimental design, Z add and Z mix values were analyzed with paired two-tailed Student's t-tests to determine differences between expected and experimental ED 50 values.

Morphine tolerance study
One week after the last morphine-alone dose-response curve, we induced morphine tolerance in LH PV :Ctrl and LH PV :hM3D mice by administering 32 mg/kg morphine (i.p.) twice per day, separated by approximately 8 hr. We measured PWL HP before and 1 hr after the first injection on day 1 and the first injection on day 4 (the seventh injection overall) to verify tolerance development. On day 5, three PWL HP measurements were taken: pre-injection, 1 hr post-CNO injection, and 1 hr post-morphine injection. On days 6-8, a combined injection of 32 mg/kg morphine and 1 mg/kg CNO was administered once per day. On day 9, the day 5 test was repeated to measure potential development of tolerance to the morphine/CNO mixture.

Histology
Mice were deeply anesthetized with isoflurane and transcardially perfused with 1Â phosphate buffered saline (PBS) followed by 4% paraformaldehyde (PFA) in 1Â PBS. Whole brains were removed and post-fixed in 4% PFA overnight at 4˚C and subsequently transferred to 1Â PBS for storage at 4C until further processing. Coronal brain sections (50 mm thick) were collected in 1Â PBS using a Leica VT1200 vibratome (Leica Biosystems GmBH, Wetzlar, Germany). In some instances, DsRed immunostaining was required to visualize viral transduction. Sections were blocked for 1 hr at room temperature in 1Â PBS with 0.3% Triton X-100% and 3% normal goat serum. After blocking, sections were incubated with rabbit anti-DsRed antibody (1:1000 Cat # 632496/RRID:AB_10013483; Takara Bio, Inc, Mountain View, CA, USA) in block solution for 20 hr at 4˚C. Tissue was then washed 4 Â 10 min in 1Â PBS followed by incubation in goat anti-rabbit Alexa Fluor 488 antibody (1:500 Cat # A11034/RRID:AB_2576217; Thermo Fisher Scientific, Waltham, MA, USA) in block solution for 1.5 hr at room temperature. After secondary antibody incubation, sections were washed 4 Â 10 min in 1Â PBS. All sections were mounted with DAPI-Fluoromount-G aqueous mounting medium (Electron Microscopy Sciences, Hatfield, PA, USA) onto Superfrost Plus glass slides (VWR International, Radnor, PA, USA). Images were taken with an AxioZoom.V16 fluorescence microscope (Carl Zeiss Microscopy LLC, Thornwood, NY, USA).

Recombinant rabies virus tracing
For retrograde monosynaptic tracing experiments, Slc32a1 Cre and Slc17a6 Cre mice were anesthetized with ketamine/xylazine (90/10 mg/kg i.p.) and placed onto a stereotaxic apparatus (David Kopf Instruments). After exposing the skull by a minor incision, a small hole (<1 mm diameter) was drilled unilaterally for helper virus injection. 40 nl of Cre-dependent AAV8/hSyn-FLEX-TVA-Rabies B19G (TVA+) was injected unilaterally (rate: 10 nl/min; titer: 4 Â 10 12 GC/ml) into the vlPAG (bregma, À3.90 mm; midline, ±0.2 mm; skull surface, À3.20 mm) by a 25-gauge Hamilton syringe (500 nl). 3-4 weeks later, mice were injected with 100 nl of the recombinant rabies viral vector (EnvA-DG-Rabies-mCherry; titer: 1 Â 10 10 pfu/ml) at the same vlPAG coordinate. Both viruses were graciously provided by the Michigan Diabetes Research Center Molecular Genetics Core, University of Michigan. 3-4 weeks after the recombinant rabies virus injection, mice were deeply anesthetized with isoflurane and transcardially perfused with 1Â PBS followed by 4% PFA in 1Â PBS. Whole brains were removed and post-fixed in 4% PFA overnight at 4˚C and subsequently cryoprotected by equilibration in 30% sucrose in 1Â PBS at 4˚C, flash frozen in isopentane on dry ice, and stored at À80˚C. Tissue was embedded in Tissue-Tek O.C.T. Compound (Sakura Finetek USA, Inc, Torrance, CA, USA) for cryosectioning. Coronal brain sections (50 mm thick) were collected in 1Â PBS using a Leica CM3050 S cryostat (Leica Biosystems GmBH, Wetzlar, Germany). Sample size estimates were derived from a previous study using the same methodology (Jennings et al., 2013).
For parvalbumin (PVALB) immunostaining, sections containing the hypothalamus were blocked for 1 hr at room temperature in 1Â PBS with 0.3% Triton X-100% and 3% normal donkey serum. After blocking, sections were incubated with guinea pig anti-PVALB antibody (1:300 Cat # GP72; RRID:AB_2665495; Swant, Marly, Switzerland) in block solution for 16 hr at 4˚C. Tissue was then washed 4 Â 10 min in 1Â PBS followed by incubation in donkey anti-guinea pig Alexa Fluor 488 or 647 antibody (1:500 Cat # 706-545-148/RRID:AB_2340472 or Cat # 706-605-148/RRID:AB_2340476; Jackson ImmunoResearch Laboratories, Inc, West Grove, PA, USA) in block solution for 1.5 hr at room temperature. After secondary antibody incubation, sections were stained for 5 min with 4',6diamidino-2-phenylindole, dilactate (DAPI 1:5000; Thermo Fisher Scientific) in 1Â PBS followed by 3 Â 10 min washes in 1Â PBS. Sections were mounted with Fluoromount-G aqueous mounting medium (Electron Microscopy Sciences) onto Superfrost Plus glass slides (VWR International). Z-stacks (30 mm) containing the LH PV region were imaged with an LSM 700 microscope using a 20Â air objective (Carl Zeiss Microscopy LLC). Maximum intensity projections were manually counted using Fiji v1.52p software (RRID:SCR_002285) with the cell counter plugin (Schindelin et al., 2012). Sections were anatomically matched to ensure that the same regions were analyzed across samples. Additionally, sections containing the PAG were mounted as described, and images were taken with an AxioZoom.V16 fluorescence stereomicroscope using a 7Â digital magnification to assess the injection site for mistargeted or lacking virus expression. After PAG assessment, one Slc32a1 Cre and one Slc17a6 Cre sample was excluded from the analysis as viral expression was not observed in the vlPAG.

Statistics
Graphs and statistics for behavioral experiments were prepared with GraphPad Prism 8 software (RRID:SCR_002798; GraphPad, La Jolla, CA, USA). All data are plotted as mean ± s.e.m., except for Ca 2+ event frequency data and isobolograms, which are plotted as mean ± 95% CI, and cell counts, which are plotted in 'part-of-whole' format. Paired or unpaired Student's two-tailed t-tests, one-way, two-way, or three-way mixed model ANOVAs with Bonferroni or Dunnett's post-tests for multiple comparisons corrections were used to analyze all behavioral data, as appropriate. Mann-Whitney Utests with Holm-Sidak correction for multiple comparisons were used to analyze Ca 2+ event frequency data from the formalin tests. A chi-square test was used to compare cell counts in the rabies tracing experiment. For all statistical tests, p<0.05 was considered significant.
for technical assistance with histology, and NIDA IRP Visual Media, in particular A Russell and L Brick, for brain slice drawings. Mouse clip art was adapted from Openclipart.org (Creative Commons CC0). Modified rabies tracing vectors were graciously provided by the Michigan Diabetes Research Core, funded by NIH P30-DK020572. AJ Robison and AL Eagle are supported by NIMH R01-111604, NIDA R01-040621, NICHD R01-072968, and NINDS R01-085171. GM Leinninger is supported by NIDDK RO1-DK103808. Y Aponte is supported by the National Institute on Drug Abuse Intramural Research Program (NIDA IRP), U.S. National Institutes of Health (NIH). National Institute on Drug Abuse and Michigan State University Animal Care and Use Committees. All of the animals were handled according to approved institutional animal care and use committee protocols (NIDA 19-CNRB-116, 19-CNRB-127, and 20-CNRB-132;MSU 201900103). Surgeries were performed under either isoflurane or ketamine/xylazine anesthesia, and every effort was made to minimize suffering.

Data availability
All data generated or analyzed during this study are included in the manuscript and supporting files. Source data files have been provided for Figure 1.