Circuit-selective pharmacological targeting of prefrontal cortex-projecting locus coeruleus neurons drives antinociception

SUMMARY

Selective manipulation of neural circuits using optogenetics and chemogenetics holds great translational potential but requires genetic access to neurons. Here, we demonstrate a general framework for identifying genetic-tool-independent, pharmacological strategies for neural-circuit-selective modulation. We developed an economically accessible calcium-imaging-based approach for large-scale pharmacological scans of endogenous receptor-mediated neural activity. As a test bed for this approach, we used the mouse locus coeruleus due to the combination of its widespread, modular efferent neural circuitry and its wide variety of endogenously expressed G-protein-coupled receptors (GPCRs). Using machine-learning-based action potential deconvolution and retrograde tracing, we identified an agonist cocktail that selectively inhibits medial prefrontal cortex-projecting locus coeruleus neurons. In vivo, this cocktail produces synergistic antinociception, consistent with selective pharmacological blunting of this neural circuit. This framework has broad utility for selective targeting of other neural circuits under different physiological and pathological states, facilitating non-genetic translational applications arising from cell-type-selective discoveries.

Graphical abstract

In brief

Kuo and McCall introduce a strategy for neural-circuit-selective pharmacology. Using ex vivo calcium imaging combined with pharmacological profiling of the locus coeruleus, the authors identify a combination of projection-selective GPCR agonists. In vivo, this combination synergistically enhances antinociception by dampening the noradrenergic projection to the medial prefrontal cortex.

INTRODUCTION

The development of cell-type-selective neuroscience tools such as optogenetics and chemogenetics has enabled unprecedented dissection of neural circuitry.^1–4 These advanced tools dramatically improve our understanding of neural circuit function and provide translational insight for strategies to alleviate neurological and neuropsychiatric disorders. These approaches, however, rely exclusively on genetic targeting to particular cell populations,^5–8 largely hindering immediate clinical translation.^9–12 To overcome this limitation, we developed a genetic-tool-independent approach for neural-circuit-selective pharmacology.

To do so, we leveraged the endogenous properties of the mouse locus coeruleus-noradrenergic (LC-NE) system. LC-NE neurons in the dorsal pons are the main source of norepinephrine for the mammalian central nervous system, have nearly ubiquitous efferent circuitry throughout the forebrain and spinal cord, and participate in a variety of brain functions, including cognitive control, salience detection, memory, arousal, stress processing, and pain regulation.^13–18 Furthermore, LC-NE neurons express a diverse population of G-protein-coupled receptors (GPCRs) and undergo complex physiological regulation by endogenous ligands.^19–22 For example, optogenetic excitation of corticotropin-releasing factor (CRF)-secreting axons in the LC drives anxiety-related avoidance behaviors, while pharmacological activation of orexin receptors (OXARs) and mu-opioid receptor (MOR) in the LC facilitates synaptic plasticity in the hippocampus and interferes with performance in cognitive function, respectively.^23–25 These various LC GPCRs are ideal for pharmacological targeting.

Despite historically being considered a homogeneous structure, recent work has emerged demonstrating anatomically defined modular organization of LC efferent projections during distinct behaviors.^26–30 This heterogeneity is common among central neuromodulatory systems^31–33 and makes the LC an excellent test bed for neural-circuit-selective control by additive, synergistic, or competitive pharmacology. However, challenges arise from the lack of understanding of how LC-NE GPCRs reallocate neural activity across LC modules in response to environmental stimuli. The consistent response of LC-NE neurons to noxious nociceptive stimuli has been used as an inclusion criterion for single-unit recordings,^34,35 and acute inflammatory pain and chronic pain augment LC expression of neural excitation markers such as cFos and pERK.^36–38 Furthermore, pharmacological LC inhibition increases the paw withdrawal threshold to noxious stimuli.^39–41 In a separate and contemporaneous study, we showed that genetic knockout of MOR in the LC increased baseline mechanical and thermal sensitivity.⁴² LC modularity is clear and apparent in pain regulation.^{28–30,42–52} Notably, the LC-NE system bidirectionally modulates nociception. Efferent spinal cord projections mediate descending analgesia via spinal adrenergic receptors, while activation of LC efferents targeting the medial prefrontal cortex (mPFC) is pronociceptive and hyperalgesic after injury.^28,42,43 This LC-mPFC efferent projection is also involved in many cognitive functions, such as attention, decision-making, fear recall, and stress regulation.^{15,23,27,43,53,54} Given that LC-mPFC projections are crucial for regulating cognition and nociception, it is reasonable that differential modulation from distinct GPCRs across efferent-defined LC modules may help achieve rapid, flexible shifts in LC-NE neural activity upon detection of nociceptive stimuli.

The functional separation of LC-mPFC from other LC-NE neurons could be an ideal target for circuit-selective antinociception. To identify circuit-selective pharmacological strategies for targeting the LC-mPFC projection, we developed an accessible ex vivo calcium-imaging-based approach for large-scale pharmacological scans of endogenous receptor-mediated neural activity. Our method takes advantage of (1) newly developed ultrafast and highly sensitive genetically encoded calcium indicators (i.e., GCaMP8f⁵⁵) and (2) machine-learning-based algorithms for spike deconvolution.⁵⁶ Conventionally, pharmacological effects on neural activity can be approximated by subtracting differences in cellular calcium-associated fluorescence.^57–59 In contrast, our assay enables efficient functional screening of ligands for 18 different GPCRs expressed in the LC.^19,22 By directly comparing changes in deconvoluted neural activity, this approach largely avoids concerns arising from action-potential-independent fluctuations in intracellular calcium.^56,60,61 Importantly, this approach can be combined with anatomical tracing to identify pharmacological action at discrete neural circuits, enabling quick identification of circuit-selective, receptor-mediated responses. We demonstrate this efficiency using retrograde labeling of LC-mPFC neurons. Cell-type-selective genetic approaches have shown LC-mPFC to promote hypersensitivity during nociception.²⁸ Our method shows that both MOR and 5HT1a receptor (5HTR1a) agonists more robustly inhibit LC-mPFC neurons compared to other LC neurons. Likewise, mACh1 receptor agonism preferentially excites non-LC-mPFC neurons. Translating these findings in vivo, we found that local infusion of a cocktail of these slice-imaging-identified agonists provides synergistic thermal antinociception—a finding consistent with shifting neural activity away from the LC-mPFC projection through circuit-selective pharmacology. Moreover, genetic deletion of MOR in LC-mPFC demonstrates that interaction between LC modules is required for pharmacological synergy. Finally, we show that combinatorial systemic agonism of these receptors at non-antinociceptive doses also produces significant antinociception. Altogether, we introduce an easily accessible pipeline of techniques to identify non-genetic, circuit-selective pharmacological approaches for neural modulation.

RESULTS

Deconvolution of spontaneous firing of LC-NE neurons

While most evidence of modular LC function relies on differential cell-type-selective control of efferent circuity with optogenetics or chemogenetics,^{27–30,54,62} endogenous control of these anatomical modules may be mediated by neuromodulators acting at the diverse array of GPCRs on these neurons. To test whether GPCR-mediated regulation of LC-NE neurons can give rise to modular function, we selectively expressed GCaMP8f in LC-NE neurons by injecting AAV1-DIO-GCaMP8f into the LC of dopamine beta hydroxylase-Cre (Dbh-Cre⁶³) mice (Figures 1A and 1B). Consistent with our previous findings, ex vivo cell-attached recordings showed that most LC-NE neurons fire spontaneous action potentials in the 1–3 Hz range^42,64 (Figure 1C). Interestingly, we found robust rhythmic fluctuations in the GCaMP8f calcium signal from LC-NE neurons. We then extracted the spatiotemporal information of individual regions of interest (ROIs) from calcium-imaging recordings taken on our typical epifluorescence slice electrophysiology microscope (Figures 1B and S1A; see STAR Methods). Simultaneous cell-attached recordings demonstrate a strong relationship between spontaneous action potential firing and calcium fluctuations (Figure 1D), suggesting the possibility of extracting LC action potentials from the calcium signal alone.⁶⁵ To do so, we adopted calibrated spike inference of calcium data using deep networks (CASCADE), a machine-learning-based algorithm, to precisely deconvolute individual action potential spikes from the calcium waveforms.^56,66 Using approximately 6 h of simultaneous cell-attached and calcium-imaging recordings, we were able to fit a network capable of predicting the temporal location of individual action potentials. We held back 15% of total learning materials for model evaluation (Figure 1E). We then converted calcium signals to the likelihood of spike occurrence, and peaks representing a higher probability of action potential generation were detected and aligned with the electrophysiological ground truth, showing clearly identified action potentials from simultaneous cell-attached recordings (Figures 1F and S1B). To ensure spike inference quality, we included only cells located within 25–75 μm of the slice surface. This avoids decreased signal-to-noise ratio when cells are located outside the focal plane. Furthermore, ROIs without spontaneous calcium fluctuations and cells with insufficient fluorescence were usually not automatically extracted or subsequently manually discarded due to poor signal-to-noise ratios in these conditions (Figure S1C). Our model displays an excellent predictive accuracy (99.2%) for LC-NE neurons with spontaneous firing rates <4 Hz, and this value only dropped to 82.1% over 5 Hz (Figure 1G). This drop in predictive accuracy could be a sampling rate limitation. Higher frame rates are likely required for detecting the calcium fluctuations with the lower signal-to-noise ratios that arise with higher firing frequencies (Figures S1C and S1D). Although decreasing exposure time to increase frame rate could offer such advantages, it inherently impacts the signal-to-noise ratio, and we found 20 frames per second to be the optimal trade-off between exposure time and predictive accuracy with our optical setup. Together, these results provide sufficient action potential deconvolution from LC-NE neurons under normal slice conditions (Figures 1C and 1F). Further analysis shows that most of the decreased predictive accuracy came from missing existing spikes rather than erroneous predictions. This issue was exaggerated in ROIs with higher firing rates due to limits on temporal resolution from the imaging acquisition and smaller calcium fluctuations that occur during higher firing rates in LC-NE neurons (Figure S1D). This can also be seen in the temporal difference between the predicted spike and the ground truth (Figure S1E). To avoid repeated ROIs in both somatic and dendritic components from the same cell, a cross-correlation matrix was made based on the traces of spike likelihood across extracted ROIs (Figure S1F). Interestingly, the distribution of these ex vivo correlation coefficients exhibits a high similarity to those from in vivo single-unit recordings previously reported.⁶⁷ These synchronous dynamics could arise from electrical coupling through dendritic gap junctions between LC-NE neurons, whereas the negative relationship could be driven by alpha2-adrenergic receptor-mediated inhibition from neighboring cells.^65,68–71 Therefore, we used a correlation coefficient cutoff threshold between ROI pairs due to the transient increase in complete synchronicity in ROI pairs with higher coefficients. Pairs with a correlation coefficient greater than 0.3 were considered together as a single electrical compartment (Figure S1G). Together, spike deconvolution from ex vivo calcium imaging enabled efficient recording of individual neuron activity across the whole LC.

Figure 1. Deconvolution of individual spikes in LC-NE neurons using calcium imaging.

(A) Cartoon illustrating viral strategy for selective LC expression of GCaMP8f.

(B) Left: 40× differential interference contrast (DIC) image showing spatial information of extracted ROIs (scale bar: 50 μm). Middle and right: DIC and fluorescence images showing simultaneous electrophysiological and imaging recordings, respectively (scale bars: 10 μm).

(C) Distribution of cell-attached LC-NE firing rate in typical slice preparation. Central tendency line indicates the mean.

(D) Example aligned simultaneous cell-attached and calcium-imaging recordings.

(E) Cartoon depicting machine-learning-based network training and spike deconvolution from trained model.

(F) Example simultaneous recording with perfect action potential inference.

(G) Predictive accuracy for spike deconvolution. Proportions of correct and failed predictions are denoted by differently colored circles (correct, black; miss, blue; incorrect, red).

Pharmacological scan of LC-NE neurons

If afferent control through GPCRs underlies some aspect of LC efferent modularity, we must first identify how neurons respond to receptor activation (Figure 2A). With the ability to monitor LC action potential activity across most of the structure established, we next sought to determine the response of individual LC neurons to various GPCR ligands. Following transcriptomic and translatomic results for LC-NE neurons from previous studies,^19–22 we tested 40 agonists targeting different LC-expressed GPCRs and selected 18 based on an initial screen for activity changes (see STAR Methods). We further divided these agonists into three groups according to their G-protein coupling or functional implications (Table 1). We then applied subsaturating doses of agonists from these three groups in a mostly randomized order with complete functional washouts between ligands (Figures 2B, S2A, and S2B). A few exceptions to randomization were experimentally necessary: one agonist in each group caused irreversible effects and was therefore always applied last (Figure S2B; Table 1). Additionally, LC-NE neurons co-express muscarinic acetylcholine receptors 1 and 3 (mAChR1/3), but not 2 and 4 (mAChR2/4). However, there is no selective mAChR3 agonist.⁷² To obtain selective activation of mAChR3, we applied pirenzepine, an antagonist of mAChR1, along with the non-selective mAChR agonist muscarine. We necessarily always applied this combination after mAChR1 activation. In preliminary cell-attached studies, we identified doses for all agonists that caused detectable, but washable, effects (Figures S2C–S2F). We then performed slice calcium imaging during baseline and agonist application for each agonist and calculated the pharmacological change in firing rate from each ligand. In a subset of trials, we bath applied clonidine (5 μM), a potent alpha2-adrenergic receptor agonist, after the entire pharmacological scan to functionally verify the cellular identity of LC-NE neurons beyond the genetic constraint of GCaMP8f expression (Figure 2B). Only 1 out of 54 ROIs was not inhibited by clonidine and excluded (Figures S2G and S2H). Cell-attached and calcium-imaging recordings showed strikingly similar means and standard deviations in response to the MOR agonist [D-Ala², N-MePhe⁴, Gly-ol]-enkephalin (DAMGO; 200 nM), further demonstrating the reliability of this approach. Our slice imaging approach substantially increased experimental throughput compared to cell-attached recordings (Figures 2C and 2D, 33 cells from 18 slices vs. 108 ROIs from 5 slices) and enabled identification of LC neurons with distinct DAMGO-mediated responses. The increased sampling from calcium imaging revealed more LC neurons completely silenced by DAMGO as well as a few that increased firing during DAMGO application, consistent with prior literature^73–75 (Figures 2C, 2D, and S2I). Generally, DAMGO inhibited most LC-NE neurons, but some variability could be caused by differential expression of functional MOR, intracellular signaling molecules, and/or a G-protein-gated inwardly rectifying K⁺ (GIRK) channel across the LC^76,77 and showing a weak relationship (R² = 0.2165) with baseline neural activity (Figure S2I). Using this approach, we further tested a comprehensive profile of pharmacological responses to agonists targeting different LC GPCRs (Figure S3A). As with MOR agonism, many agonists’ responses were non-homogeneous, offering direct evidence for GPCR-mediated modules in the LC (Figures 2E–2G). Although it is possible the pharmacological response could be partially underestimated due to the greater predicted misses in cells with higher firing rate, the variability of agonist-induced responses was still capable of identifying increased firing rates. Table S1 lists the detailed statistics for results of the serial pharmacological scan in LC-NE neurons. Most responses appear to be cell autonomous, as repeating the scan for group I agonists in the presence of synaptic blockers (5 mM kynurenic acid, 1 μM strychnine, and 100 μM picrotoxin) left most results unaltered. Preadministration of synaptic blockers did, however, abolish the excitatory effect of the alpha1-adrenergic receptor agonist phenylephrine. This blunting suggests that alpha1-adrenergic receptor-mediated excitation of the LC is largely presynaptic (Figures 2H and S3B), consistent with recent electron microscopy results.⁷⁸

Figure 2. Ex vivo serial pharmacological scan of LC GPCRs.

(A) Cartoon displaying differential expression of multiple GPCRs in LC-NE.

(B) Top: diagram illustrating the pharmacological scan. Bottom: representative calcium traces showing pharmacological effects of DAMGO and substance P.

(C) Firing rate plots showing DAMGO effects using cell-attached (left) and imaging (right) recordings. Cell-attached: paired-t test, t = 11.99, ****p < 0.0001. Imaging: Wilcoxon matched-pairs signed rank test, W = −5,747, ****p < 0.0001. Data are represented as the mean ± SD.

(D) Plots showing the percentile of changes in firing rate from data shown in (C).Data are represented as the mean ± SD.

(E–G) Summarized results of GPCR activation-induced firing rate changes across three groups of GPCR agonists. Repeated-measures two-way ANOVA followed by Bonferroni test, ****p < 0.0001; ns, not significant; see Table S1 for detailed statistics. Plots show lines for the median, first quartile, and third quartile.

(H) Plot of presynaptic modulation from group I agonists. Repeated-measures two-way ANOVA, ****p < 0.0001; ns, not significant; see Data S1 for detailed statistics. Plot shows lines for the median, first quartile, and third quartile.

Abbreviations: Phe, phenylephrine; McN, McN-A-343; Mus, muscarine; Pir, pirenzepine; Calc, calcitonin; U50, U50488; SNC, SNC-162; N/OFQ, nociceptin; SubP, substance P; Aden, adenosine; Bac, baclofen; OXA, orexin A; 8OH, 8OH-DPAT; Ana, anandamide; SST, somatostatin; SB, synaptic blockers.

Table 1.

Agonist concentrations targeting different GPCRs for ex vivo pharmalogical scan

Agonist	Target	Concentration
Group I (Gq-coupled GPCRs)
DOI	5HTR2a/c	10 μM
Phenylephrine	α1-AR	20 μM
DHPG	type I mGluR	50 μM
McN-A-343	mAChR1	10 μM
Muscarine	mAChRs	5 μM
Calcitonina	calcitonin receptor	20 nM
Group II (pain- and stress-related)
U50	kappa-OR	10 μM
DAMGO	mu-OR	200 nM
SNC162	delta-OR	20 μM
N/O FQ	NOPR	2.5 nM
CRF	CRFR	1 μM
Substance Pa	NK1R	200 nM
Group III (other)
Adenosine	adenosine receptors	100 μM
Baclofen	GABAB receptor	20 μM
Orexin A	OXR1/2	50 nM
8OH-DAPT	5HTR1a	10 μM
Anandimide	CBR1/2	10 μM
Somatostatina	somatostatin receptor	20 nM

^aAgonists with irreversible effects were applied last.

Module-selective pharmacological profiling of LC-NE neurons

The non-homogeneous response of LC-NE neurons to GPCR activation suggests that these receptors may underlie endogenous control of LC modules. GPCR-mediated regulation across anatomically defined LC efferent modules may enable rapid real-location of noradrenergic resources across brain circuits during distinct behaviors.^{20,21,27–29,42,43,62} To investigate the pharmacological profile of anatomically defined LC modules, we combined our ex vivo calcium imaging approach with fluorescent retrograde tracing to identify LC neurons by their efferent targets. To do so, we injected the common retrograde neuronal tracer cholera toxin subunit b conjugated to a CF 594 fluorescent tag (Ctb594) into the mPFC (Figures 3A and 3B). In the dorsal pons, almost all Ctb594 was co-localized with tyrosine hydroxylase (TH; the rate-limiting enzyme for catecholamine synthesis). These LC-mPFC neurons were scattered across the whole LC except for sparse distribution in the rostroventral LC (Figures 3C and S4A–S4E), where many spinally projecting LC-NE neurons are located.^28,47 We next repeated the ex vivo pharmacological scan to determine the pharmacological response of LC-mPFC neurons. To ensure accuracy of the Ctb594 signal, we reconstructed recorded slices using confocal z-stack scanning in concert with post hoc immunohistochemistry of GCaMP8f and Ctb594. The extracted ROIs were then registered to their modular category (LC-mPFC or non-LC-mPFC) based on the presence of Ctb594 (Figures 3C–3F). Surprisingly, some agonists drove differential effects between mPFC- and non-mPFC-projecting LC-NE modules. We observed stronger inhibition of LC-mPFC neurons by DAMGO and the 5HTR1a agonist 8OH-DPAT. For the non-LC-mPFC module, McN-A-343, a selective mAChR1 agonist, preferentially drove excitation, while baclofen, a GABAB receptor (GABABR) agonist, preferentially inhibited the non-mPFC-projecting LC module (Figures 3G–3J, S3C, and S3D). For calcium imaging, an important concern with G_αq-coupled GPCR activation, such as with McN-A-343, is the potential contribution of action-potential-independent intracellular calcium release.⁷⁹ Compared to conventional quantification of calcium signals (i.e., ΔF/F),^57,58,80,81 our calcium-fluctuation-based deconvolution of neural activity⁵⁶ is less vulnerable to disturbance from action-potential-independent intracellular calcium changes when testing GPCR ligands.^60,61 In particular, the spike prediction accuracy following G_αq-coupled GPCR activation-mediated intracellular calcium release is maintained, suggesting the deconvolved signal represents the absolute neuronal firing rate (Figures S5A and S5B). Additionally, the performance of our model remains excellent across LC modular architecture, with no differences observed between mPFC-projecting and non-mPFC-projecting LC modules (Figure S5A). Detailed statistical information is listed in Tables S2 and S3. To eliminate the possibility that these effects could arise from differential spontaneous firing at baseline, we also compared baseline activity between modules and found no significant differences between any group of agonists (Figures S5C–S5F). Together these findings demonstrate pharmacological dissection of LC efferent modules.

Figure 3. Pharmacological scan of GPCRs targeting the LC-mPFC module.

(A) Cartoon of viral and retrograde strategies targeting the LC-mPFC.

(B) Cartoon demonstrating the recording setup.

(C) Representative fluorescence image showing distribution of LC-mPFC neurons indicated by co-localization of TH (cyan) and Ctb (magenta) (scale bars: inset, 25 μm; main, 200 μm).

(D–F) Seven representative extracted ROIs showing their spatial information (scale bar: 10 μm) (D), modular identity by co-localization of GCaMP8f (cyan) and Ctb (magenta) (scale bar: 5 μm) (E), and responses to orexin A (pink denotes LC-mPFC ROI (F).

(G–I) DAMGO-, 8OH-DPAT-, and McN-A-343-induced changes in firing rates across mPFC- and non-mPFC-projecting LC-NE neurons. DAMGO: Student’s t test, t = 3.260, **p < 0.01. 8OH-DPAT: Mann-Whitney test, U = 106, *p < 0.05. McN-A-343: Mann-Whitney test, U = 209, *p < 0.05. Data are represented as the mean ± SD.

(J) Boxplot summarizing modular pharmacological scan. Boxes denote the mean and 10th–90th percentile of distribution. See Tables S2 and S3 for statistical details. Data are represented as the mean ± SD. Abbreviations: 5HTR2a/c, serotonin receptor 2a/c; α1-AR, alpha1-adrenergic receptor; mAChR1/3, muscarinic receptor 1/3; CalcitoninR, calcitonin receptor; KOR, kappa opioid receptor; DOR, delta opioid receptor; MOR, mu opioid receptor; NOPR, nociceptin receptor; CRFR, corticotropin-releasing factor receptor; NK1R, neurokinin 1 receptor; AdenosineRs, adenosine receptors; GABABR, GABAB receptor; Orexin1/2R, orexin receptors ½; 5HTR1a, serotonin receptor 1a; CB1/2R, cannabinoid receptor ½; SomatostatinR, somatostatin receptor.

Synergistic antinociceptive effects from modular-selective pharmacology

There is clear modularity in LC-mediated pain modulation. Activation of the LC-mPFC projection is thought to drive pronociception, while noradrenergic projections to the spinal cord exert descending antinociception.^28,44 Furthermore, LC efferents targeting other brain regions engage in different aspects of aversive and pain-related behaviors.^{27,28,30,42,43,45,46,49,51,52,54,62,82} Accordingly, to drive antinociception, an ideal LC modulation strategy is to selectively dampen the pronociceptive activity of LC-mPFC projection, reallocating noradrenergic modulation across the central nervous system. We next sought to test whether the agonists we identified as shifting neural activity away from the LC-mPFC module ex vivo (Figures 3G–3J) could generate in vivo antinociception. Here, we selected DAMGO, 8OH-DPAT, and McN-A-343 for in vivo testing. We implanted a bilateral cannula above the LC in C57BL/6J mice and began with single compound infusions. We tested thermal paw-withdrawal latencies using the Hargreaves test (Figures 4A and 4B) and calculated ratios between baseline measurements and pharmacological conditions (Figures 4A, 4B, and S6A–S6F) to construct dose-response curves from single-compound infusions. We found that DAMGO and 8OH-DPAT produced significant antinociception (Figures 4C and 4D). McN-A-343 elicited a smaller, subtly antinociceptive effect (Figure 4E). Following these dose-response curves, we then designed two pharmacological cocktails of these agonists based on their in vivo EC₅₀ values (DAMGO, 0.46 μM; 8OH-DPAT, 1.96 μM; and McN-A-343, 2.27 μM), namely, cocktail A (DAMGO + 8OH-DPAT, 1:4) and cocktail B (DAMGO + 8OH-DPAT + McN-A-343, 1:4:5). To further quantify synergy from pharmacological combinations, we used isobolographic analysis, a method often used to assess whether biological responses to pharmacological combinations are equal to, greater than, or less than expected from individual ligands. The “isobole” is the line representing the theoretical simple additive effects along the dose-response curves for each ligand. Results above the isobole are considered subadditive or antagonistic, and results under the isobole are considered superadditive or synergistic. Interestingly, when we generated isobolograms from the dose-response curves of these two pharmacological cocktails, there was significant synergistic antinociception driven by cocktail A, but not cocktail B (EC₅₀ 0.46 μM for cocktail A and 2.1 μM for cocktail B) (Figures 4F–4J and S6G–S6I). Surprisingly, cocktail B caused more antinociception than any other ligand or cocktail (Figures 4G, 4H, and 4J). However, three-dimensional isobolographic analysis⁸³ shows an additive agonist interaction where cocktail B merely enhances antinociception past the other drug conditions once cocktail B reaches a minimum dose threshold (Figures S5H and S5I). Because these tests were performed within subject over many weeks, we considered the possibility that baseline nociception could drift across the course of pharmacological treatments and behavioral examination; however, we found no significant differences across baseline measurements in seven trials (Figures S6A–S6F). Furthermore, the thermal plantar assay relies on intact locomotor responses such that any sedative effect could lead to fictive antinociception. Cocktail B, which provides the greatest paw-withdrawal latency, did not alter locomotion in the open field test (Figure S6J), suggesting the increased paw-withdrawal latency is indeed driven by reduced noxious stimulus detection. We then returned to ex vivo slice imaging to determine whether responses across LC modules during co-application of these three agonists drive activity away from the LC-mPFC. As hypothesized, application of cocktail B preferentially inhibited the LC-mPFC module (Figure 4K). Taken together, our ex vivo calcium imaging approach identified a pharmacological strategy for LC-mediated antinociception.

Figure 4. Multiple pharmacological approaches drive antinociception by shifting activity away from the LC-mPFC module.

(A) Experimental design of Hargreaves thermal plantar assay in concert with local LC infusions.

(B) Representative bilateral LC cannula implantation. Dashes indicate trajectory of cannula, TH is in red (scale bar: 0.5 mm).

(C–G) Dose-response curves of pharmacological antinociception from intra-LC DAMGO, 8OH-DPAT, McN-A-343, cocktail A, and cocktail B. Paw-withdrawal latency ratios were calculated between baseline and pharmacological conditions. DAMGO: repeated-measures one-way ANOVA followed by Dunnett’s test, F = 20.77, **p < 0.01, ****p < 0.0001. 8OH-DPAT: repeated-measures one-way ANOVA followed by Dunnett’s test, F = 11.90, **p < 0.01. McN-A-343: Friedman test followed by Dunn’s test, Friedman statistic = 22.82, *p < 0.05. Cocktail A: repeated-measures one-way ANOVA followed by Dunnett’s test, F = 22.07, **p < 0.01, ***p < 0.001. Cocktail B: repeated-measures one-way ANOVA followed by Dunnett’s test, F = 33.86, ***p < 0.001, ****p < 0.0001. Data are represented as the mean ± SD.

(H) Full dose-response curves from data shown in (C)–(G). Rectangle denotes 10 μM pharmacological response.Data are represented as the mean ± SD.

(I) Isobologram showing synergistic effect of cocktail A. Experimental and theoretical EC₅₀ of cocktail A and EC_50s of DAMGO and 8OH-DPAT are indicated by black dots. Experimental EC_50s of individual mice are indicated by red circles, and black line denotes additive isobole. Inset shows statistical evaluation of synergism. One-sample Wilcoxon signed-rank test, *p < 0.05.

(J) Comparison of antinociceptive effects from different pharmacological approaches at 10 μM. One-way ANOVA followed by Holm-Šídák’s test, F = 13.52, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. Data are represented as the mean ± SD.

(K) Module-selective cocktail B-induced firing rate inhibition. Student’s t test, t = 2.150, *p < 0.05. Data are represented as the mean ± SD.

Modular deletion of LC-MOR disrupts cocktail-induced antinociception

One possibility for how LC modules drive discrete behavioral functions is through interaction, either competitive or cooperative, between two or more efferent modules. To test whether such modular interaction is required for the intra-LC cocktail-mediated antinociception, we created an LC-mPFC module-selective conditional knockout of oprm1^fl/fl (mKO). We bilaterally injected a noradrenergic-selective, retrogradely trafficked canine adenovirus to drive Cre recombinase expression under the synthetic PRSx8 promoter (CAV-PRS-Cre-V5) into the mPFC of C57BL/6J and oprm1^fl/fl homozygous mice^47,84,85 (Figures 5A and 5B). Oprm1^fl/fl mice have loxP sites flanking the second and third exons of the oprm1 gene.⁸⁶ Consequently, selective Cre recombinase expression genetically deletes MOR. To visualize successful Cre recombination, in a subset of mice, we also introduced Cre-dependent expression of the red fluorophore, mCherry, into the LC by local delivery of AAV8-DIO-mCherry (Figure 5C). LC-mPFC mCherry expression was consistent with Ctb594 retrograde tracing (Figures 5C and S4). Further, we quantified LC-mPFC neurons by mCherry expression showing that only a subset of LC neurons project to the mPFC and undergo Cre-mediated recombination (27.7% ± 7.1%, Figure S4J). Having used this approach previously,⁴² we corroborated functional modular MOR exclusion using DAMGO-induced inhibition during cell-attached recordings (Figures 5D and 5E). We next injected another cohort of oprm1^fl/fl mice with either CAV-PRS-Cre-V5 or CAV-mCherry in the mPFC and implanted bilateral cannula above the LC (Figure 5F). Following deletion of LC-mPFC MOR, we tested thermal paw-withdrawal thresholds using a fixed concentration of ligands or cocktails (10 μM, Figure 5G). Consistent with our electrophysiological findings, we found that LC-mPFC mKO mice lost LC MOR-mediated antinociception. LC-mPFC mKO also surprisingly disrupted the antinociceptive action of cocktail B (Figure 5H; Table S4). This suppression of cocktail B-mediated antinociception could be caused by MOR activation on non-LC-mPFC-projecting neurons. To test this hypothesis, we designed cocktail C as a pharmacological combination of 8OH-DPAT and McN-A-343 in a fixed ratio (4:5). Cocktail C elicited robust antinociception that was not affected by LC-mPFC MOR mKO (Figure 5H). This finding suggests that MOR activation of non-LC-mPFC-projecting neurons could yield a pronociceptive response that competes against the antinociception driven by the MOR-insensitive portion of cocktail B.^{26–28,47,62} Further, these results show that both LC-mPFC and non-LC-mPFC neurons are actively involved in pharmacologically mediated antinociception, providing a net outcome from the relatively stronger inhibition of the LC-mPFC innervating module. This differential, pharmacologically mediated modulation was outflanked by modular impairment of MOR-mediated regulation. This intermodule interaction in nociception is likely the consequence of integration across brain and spinal regions that receive noradrenergic inputs from distinct LC efferent modules. A recent study also demonstrated that alpha2-adrenergic receptor-mediated lateral inhibition underlies some of the modular organization of LC.⁷¹ This cross-modular inhibition is theoretically scaled by the concurrent excitation of LC modules; however, the in vivo dynamics of cross-modular integration in response to nociceptive stimuli remains elusive. Furthermore, the intra- and intermodular electrical coupling between LC neurons can also contribute to cross-module interactions.⁶⁸ Our slice imaging approach would provide an ideal platform to test discrete hypotheses regarding cross-module interactions.

Figure 5. Modular, conditional LC-mPFC MOR knockout disrupts cocktail B-mediated antinociception.

(A) Cartoon illustrating viral approach for the modular deletion of MOR with fluorescent labeling in LC.

(B) Injection site of CAV-mCherry into mPFC (scale bar: 1 mm).

(C) Distribution of LC-mPFC neurons by the co-localization of TH (cyan) and mCherry (magenta) (scale bars: inset, 20 μm; main, 200 μm).

(D) DIC and fluorescence images during cell-attached recording (scale bars: left and middle, 10 μm; right, 0.5 mm).

(E) Representative LC-mPFC cell-attached recordings demonstrating lack of DAMGO response.

(F) Cartoon illustrating viral approach for modular LC-mPFC MOR deletion and cannula implantation.

(G) Timeline for sensory testing with pharmacological infusions.

(H) Impact of modular knockout of MOR in LC on pharmacological antinociception. Data are represented as the mean ± SD.

(I) Systemic cocktail enhances antinociception.

Fen, fentanyl; 8OH, 8OH-DPAT; Cev, cevimeline. Repeated-measures two-way ANOVA followed by Bonferroni’s test, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. See Table S4 for detailed statistics. Data are represented as the mean ± SD.

Slice-imaging-derived systemic cocktails enhance antinociception

Although local LC infusion of synergistic cocktails demonstrates that findings from ex vivo slice imaging can translate to in vivo conditions, local pharmacology is unlikely to be a clinically relevant strategy. To test the translational potential of our slice imaging and local infusion findings, we tested potential antinociceptive effects of systemic injection of blood-brain barrier-permeative ligands for MOR, 5HT1a, and mAChR1 (fentanyl, 8OH-DAPT, and cevimeline, respectively) together and in combination. To avoid the inherent antinociceptive effects of systemic MOR and 5HT1a activation, we selected doses close to the established EC₂₅ for fentanyl (0.25 μg/kg, intraperitoneally [i.p.]) and 8OH-DAPT (0.6 mg/kg, i.p.).^87–90 Cevimeline is typically used to treat dry mucous membranes and potentially enhance cognition in neurodegenerative disease.^91–93 As it is not known to have antinociceptive properties, we used a commonly used dose (1 mg/kg, i.p.).^91–93 As expected, no single drug was antinociceptive (Figure 5I). Surprisingly, however, drug combinations analogous to cocktail A (fentanyl + 8OH-DAPT), cocktail B (fentanyl + 8OH-DAPT + cevimeline), and cocktail C (8OH-DAPT + cevimeline) all enhanced thermal antinociception, suggesting synergistic interaction between these ligands at the systemic level as well (Figure 5I). These findings demonstrate the translational potential of a pipeline that begins with ex vivo neural-circuit-selective imaging of ligand-mediated calcium responses, but further study will be required to fully understand the mechanistic complexity arising from systemic cocktail administration. Altogether, we introduce an approach to identify neural-circuit-selective pharmacological strategies as a means to drive GPCR-mediated antinociception across LC modules. As hypothesized by us and others,^28,42,43 this neural-circuit-selective antinociception requires an active interaction between different efferent LC modules.

DISCUSSION

Technical potential of pharmacological scan and individual spike deconvolution

Here, we demonstrate an ex vivo pharmacological scan to quickly test multiple responses across many LC-NE neurons. We were able to deconvolute individual spikes during drug application using a machine-learning-based model based on spontaneous fluctuating calcium signals. This technical combination has broad compatibility with cell types in different brain regions. For example, it would be straightforward to implement the same strategy in striatal cholinergic interneurons, which also spontaneously fire around 0.5–5 Hz in normal slice preparation.⁹⁴ The firing-frequency-dependent deconvolution allows us to dissect neural activity in compactly arranged nuclei without bias from bleaching issues. This approach also decreases potential confounding from G_αq-mediated intracellular calcium release (Figures S5A and S5B). In a typical slice electrophysiology experiment, only a few simultaneous recordings are possible with traditional cell-attached recordings. With substantial equipment and expertise, up to eight LC neurons have been recorded simultaneously,⁶⁸ but our imaging approach enables recording more than twice that using only standard slice electrophysiology equipment. Here, we tested simultaneous pharmacological responses across 15–30 ROIs. This advance could be further expanded in larger brain nuclei and/or with more advanced optical setups. Furthermore, this approach enables functional study of relationships between cells, such as spike synchronicity (Figures S1F and S1G), and reveals neural ensembles by pharmacological profiling (Figures 2 and 3).^67,80,95,96 Together, this fast, efficient, and largely affordable survey of neural responses during pharmacological application provides a useful toolbox to understand cell-type-selective pharmacological action in physiological or pathological conditions.

Interaction between GPCRs

To screen agonists for several receptors, we consecutively applied agonists targeting various GPCRs with sufficient washouts for firing to return to baseline (Table 1). Despite our best efforts, this approach likely gives rise to inherent interactions between GPCR signaling cascades. We found that the application of some compounds caused robust effects that were resistant to prolonged washing (Figures S2C–S2F). These instances could arise from saturating doses, high ligand affinity, or persistence of cellular mechanisms underpinning firing rate effects. Generally, GPCRs activating Gi/o-coupled G-protein pathways were more difficult to wash out than Gq-coupled GPCRs (Figures S2C–S2F). The sustained potassium conductance through GIRK channels may underestimate responses from receptors using a similar mechanism.^97,98 The weak correlation between baseline neural activity, DAMGO, and clonidine responses (Figures S2G–S2I) at least partially alleviates this possibility. It is also possible that signaling pathways might interact with subsequent GPCR activation.^99,100 For instance, MOR activation is known to cross-desensitize the alpha2-adrenergic receptor and somatostatin receptor,¹⁰¹ and consecutive carbachol, a potent cholinergic agonist, applications desensitize mAChR1/3-mediated excitation in A7 noradrenergic neurons.¹⁰² Phosphorylation of various second messengers like CREB, beta-arrestin, GRK, and ERK in the LC could result in a complicated interaction between different receptors.^{64,77,103,104} This issue was exaggerated when receptor subtypes with similar cellular mechanisms, such as opioid receptors and mAChRs, were in the same imaging group. Furthermore, heteromerization can occur with discrete pairs of GPCRs, and different ligands can be dependent on such heteromerization.^105–109 Whether such heteromers endogenously form in LC neurons is not yet clear, but our approach could be modified to identify such functions. To minimize bias from crosstalk between receptors, we used randomized subsaturating concentrations of agonists. Irreversible agonists (i.e., calcitonin, substance P, and somatostatin) were preidentified by cell-attached recordings and always applied last, prior to clonidine, to minimize impact on interpretation.

Comparisons between optogenetics, chemogenetics, and neural-circuit-selective pharmacology

Optogenetics and chemogenetics are powerful tools widely used to dissect specific cell populations during discrete behaviors. Optogenetic tools such photoactivatable ion channels, pumps, and receptors with various selectivity in light wave-length, provide a toolbox for precise control of neural activity with excellent spatiotemporal resolution.^3,4,110,111 Chemogenetic tools such as designer receptor exclusively activated by designer drugs (DREADDs) and Pharmacologically Selective Actuator Modules (PSAMs) designed from mAChRs and ligand-gated channels, respectively, offer easy, long-term control of specific neural circuits.^1,2,112,113 Bridging these tools with photosensitive GPCRs enables spatiotemporal control of intracellular signaling cascades.^114–116 Nevertheless, a few concerns using optogenetics and chemogenetics have been reported. Proton pumps are limited in presynaptic inhibition, while chloride pumps and channels are often subject to rebound excitation.^117,118 While these issues are mitigated by photoactivatable Gi/o-coupled GPCRs, these tools are often more challenging to validate.^114–116 Additionally, highly synchronous neural activity under optogenetic excitation might also interfere with, or otherwise non-physiologically entrain, brain oscillations. Ultrastructural evidence suggests that the cellular locations of chemogenetic constructs are largely affected by tagged proteins.¹¹⁹ Occupancy of intracellular signaling pathways due to overexpressed receptors could also disrupt transmission through endogenous GPCRs. Further research is needed to evaluate the cross-desensitization of endogenous GPCRs during the activation of chemogenetic receptors and photoactivatable GPCRs. In contrast, our neural-circuit-selective pharmacology approach relies merely on intact cellular machinery triggered by endogenous GPCRs. Therefore, this approach is largely constrained by the nature of GPCR expression in given cell populations. Furthermore, presynaptic modulation can also be included in the pharmacological response during agonist applications.^72,120–123 Given that, in normal slice preparations, distal afferents remain mostly silent without axonal action potentials, we could be underestimating the net effects caused by presynaptic action in vivo. Our approach likely yields less selective control of neural activity compared to clear excitation or suppression using optogenetics and chemogenetics. However, several critical benefits still arise from recruiting endogenous GPCRs: (1) it helps to alleviate off-target effects from overwhelming optogenetic or chemogenetic manipulations while providing adequate, more physiological control on neural activity. (2) It sidesteps the need for genetic editing, lending itself to more translational applications. (3) It reduces unexpected effects from overexpression and off-target expression of foreign genes.^{9–12,124–126} Nevertheless, substantial efforts will be required to build our understanding of circuit-selective pharmacological profiles to expand this approach for non-genetic circuit-selective manipulations in other brain regions.

Insights into multimodal analgesia

From a clinical perspective, combinations of multiple analgesic agents like opiates/opioids, acetaminophen, non-steroidal anti-inflammatory drugs, gabapentinoids, alpha2 adrenoreceptor agonists, NMDA receptor antagonists, and serotonin-norepinephrine reuptake inhibitors are commonly used to treat postoperative and chronic pain conditions.^127–129 Taking advantage of additive or synergistic pharmacology can alleviate undesirable opioid-induced side effects such as nausea, vomiting, and respiratory depression.^130–132 However, most multimodal analgesia efforts are conceptually derived from a combination of known analgesics. Here, we demonstrate a circuit-driven perspective to derive antinociception from multiple pharmacological approaches that may not directly link to antinociception alone (i.e., McN-A-343). The synergistic and enhanced antinociception driven by cocktail A and cocktail B, respectively, each containing a MOR agonist, provides a route for reduced opioid use by leveraging the modular interaction in the LC-NE system with other receptor ligands. Furthermore, the results from cocktail C and its systemic analog (8OH-DAPT + cevimeline) offer a non-opioid strategy for pain control. Considering the prevalence of co-morbid anxiodepressive disorders and sleep dysfunction during the chronification and maintenance of chronic pain,^133–135 the LC-NE system could be an excellent target for pain therapeutics due to its contribution to cognitive function, stress processing, and pain regulation.^{27–30,38,43,51,62,135} Previous studies have shown decreased mRNA expression and faster desensitization of MOR in LC during the development of neuropathic pain,^136,137 and selective antagonism of alpha1/2-adrenergic receptors reversed neuropathic pain-induced depression.⁴⁵ Accordingly, one interesting future direction would be to test whether the neural-circuit-selective agonist cocktails designed in this study could alleviate the co-morbid hypersensitivity and anxiodepressive symptoms in chronic pain.

GPCR-mediated control of functional LC modules

The rich expression of various GPCRs regulates LC-NE neurons to flexibly tune responses to diverse environmental stimuli.^{19,20,22–25} Historically, the LC was considered homogeneous, with nearly universal efferent innervation of most parts of the CNS.¹³⁸ Recently, substantial multidisciplinary efforts have revealed extensive LC heterogeneity arising from distinct anatomical, genetic, and physiological properties.^{19,22,26–30,62,67,139,140} These properties, when ascribed function, have been used to define distinct LC modules. Due to the operational definition of LC modules shifting across different functional domains (i.e., anatomical, genetic, and physiological) it can be difficult to determine how many modules exist, and it remains important to be clear how modularity is being considered and quantified in each study. A growing number of studies have interrogated the functional contribution of modular LC efferent architecture,^{27–30,42,54,62} but relatively little effort has been made to determine how afferent input rapidly shifts activity between LC functional modules. To address this issue, our work demonstrates differential modulation of LC modules by GPCR activation. We found that subsaturating concentrations of many GPCR agonists cause heterogeneous effects across the LC, raising the possibility of defining LC modularity by pharmacological responses.⁹⁵ This cellular response variability might also occur after release of endogenous ligands in vivo, something difficult to precisely mimic exogenously. The lack of comprehensive understanding of presynaptic endogenous ligands and postsynaptic locations of many GPCRs presents challenges in studying the physiological role of LC GPCRs. We recently showed that conditional knockout of MOR in noradrenergic neurons increased baseline mechanical and thermal sensitivity, and this phenomenon was replicated by module-selective deletion of MOR in LC-mPFC neurons, demonstrating a modular role of LC MOR in pain regulation.⁴² Here, interestingly, the same LC-mPFC modular deletion of MOR disrupted cocktail B-mediated antinociception. This was not the case for the non-opioid cocktail C. This latter observation suggests active interaction between LC modules during pharmacological activation. Although the maximum antinociception was not increased, we also found a synergistic interaction between DAMGO and 8OH-DPAT from cocktail A. This potentially opioid-sparing phenomenon might be due to rapid activation of shared downstream signaling pathways targeting Gi/o-coupled GPCRs and suggests full recruitment of GIRK-mediated hyperpolarization.^76,77,100 In contrast, the increased antinociception from cocktail B might reflect competition between Gq- and Gi/o-coupled GPCRs on individual cells at lower doses, while higher doses unlock important modular interactions to enhance antinociception.¹⁰¹ In summary, we present an ex vivo imaging approach to rapidly guide the development of neural-circuit-selective pharmacological approaches. Our work leveraged GPCR-mediated regulation of LC-mPFC to provide a clear target for synergistic pharmacological antinociception.

Limitations of the study

Despite advances in precision and efficiency, the necessity of stable, spontaneous action potential firing limits the application of machine-learning-based spike deconvolution, presenting challenges when applying this approach to cells that do not fire under typical ex vivo conditions. This issue can be partially mitigated in other ex vivo or in vivo preparations that allow repeated drug applications, such as superficial layers of spinal cord.⁹⁵ Although spontaneous activity can be generated by adjusting recording buffers,^79,141,142 technical difficulties remain in cells with discrete firing modes at different membrane potentials, like thalamic relay neurons,^143,144 or cells with adaptive spiking profiles.^145,146 Moreover, the inherent spatiotemporal resolution constraints of imaging data acquisition restrict compatibility with faster-firing neurons. We translated the pharmacological approach in living animals through intra-LC infusions. However, one should consider the inherent limitations of these local dose-response curves, such as variability in drug delivery dynamics, reduced receptor selectivity at higher concentrations, and potential effects from activating GPCRs in surrounding brain areas. These limitations are mitigated by: (1) cell-type- and projection-selective genetic deletion of LC MOR, which eliminates DAMGO- and cocktail B-induced antinociception (Figure 5H), and (2) the relatively high Oprm1 and Chrm1 mRNA expression in the LC compared to the rest of the dorsal pons,^147,148 which likely leads to activation of these target receptors on LC cells. However, the broader expression of Htr1a across the dorsal pons could enable receptor activation in neighboring brain regions,¹⁴⁹ although we used stringent inclusion criteria for cannula placements (Figure S6K). Despite these limitations, we introduce a useful framework to identify genetic-tool-free, pharmacological strategies for neural-circuit-selective manipulation with broad translational potential.

RESOURCE AVAILABILITY

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Jordan G. McCall (jordangmccall@wustl.edu).

Materials availability

This study did not generate new unique reagents.

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Both sexes of adult C57BL/6J (JAX:000664), Dbh-Cre+/− (JAX:033951), Oprm1fl/fl (JAX: 030074) mice over 8 weeks-old in age were used in this study. Animals were originally introduced from The Jackson Laboratory (Bar Harbor, ME, USA) then group housed and bred in a barrier facility with an ad libitum access to food pellets and water under a 12:12-hour light/dark cycle (lights on at 7:00 AM). Animals were transferred to the experimental facility and allowed to be inhabited for at least 2 weeks. All experiments and procedures were approved by the Institutional Animal Care and Use Committee of Washington University School of Medicine in accordance with National Institutes of Health guidelines.

METHOD DETAILS

Please provide precise details of all the procedures in the paper (behavioral task, generation of reagents, biological assays, modeling, etc.) such that it is clear how, when, where, and why procedures were performed. We encourage authors to provide information related to the experimental design as suggested by NIH and ARRIVE guidelines (e.g., information about replicates, randomization, blinding, sample size estimation, and the criteria for inclusion and exclusion of any data or subjects).

Stereotaxic surgeries

Mice were anesthetized in an induction chamber (3% isoflurane) then mounted on a stereotaxic apparatus (Model 940, Kopf Instruments, CA, USA) and maintained at 1-2% isoflurane. Small craniotomies were performed and glass pipettes filled with tracers and viruses were slowly advanced to the brain area of interest as the following coordinates (in mm), mPFC : AP +2.0, ML 0.4, DV 0.9 & 1.8 from Bregma; LC: AP −0.9, ML 0.9, DV 2.9 from Lambda. 250nL of viruses were gently delivered through Nanoject III injector (Drummond Scientific Company, PA, USA) at rate of 40nL/mins. Tracers and viruses used included Ctb-CF594 (00072, Biotium, CA, USA), AAV1-syn-FLEX-jGCaMP8f-WPRE (162379-AAV1, Addgene, MA, USA), CAV-mCherry and CAV-PRS-Cre-V5 (PVM, Biocampus, Montpellier, France). Mice were allowed to recover for 2 weeks for neuronal tracing and at least 6 weeks for viral expression. For cannula implantation, 26-gauge bilateral guide cannulas coupled with blind cannulas (Protech Int. Inc) were slowly directed above the LC region (AP −0.9, ML 0.9, DV 2.8 from Lambda), 33-gauge injector cannulas (Protech Int. Inc) were then used in intra-cerebral infusion (Figure S5K). Crowns of implantation were secured using C&B-Metabond (Parkell Inc., NY, USA) and super glue. Postoperative analgesia was made upon administration of carprofen (5 mg/kg) through oral tablets or subcutaneous injection for 3 days. Mice were allowed to have 2 weeks of recovery after surgery prior to behavioral tests. For the genetic deletion of modular MOR by retrograde virus, mice were allowed to have 7 weeks of recovery to complete the genetic knock out.

Slice preparation and electrophysiology

Adult mice were deeply anesthetized with an i.p. injection of cocktail containing ketamine, xylazine & acepromazine then perfused with ice-cold slicing-aCSF consisting of (in mM) 92 N-methyl-d-glucose (NMDG), 2.5 KCl, 1.25 NaH₂PO₄, 10 MgSO₄, 20 HEPES, 30 NaHCO₃, 25 glucose, 0.5 CaCl₂, 5 sodium ascorbate, 3 sodium pyruvate and 2 thiourea, oxygenated with 95% O₂ and 5% CO₂, then adjusted to pH 7.3-7.4 and 315-320 mOsm of osmolarity with sucrose. Brains were dissected rapidly then embedded and mounted with 2% agarose made in slicing-aCSF, coronal brainstem slices containing LC region were cut using vibratome (VF310-0Z, Precisionary Instruments, MA, USA). Slices were incubated in warm (32°C) slicing-aCSF for 30 minutes then transferred to holding-aCSF consisting of (in mM) 92 NaCl, 2.5 KCl, 1.25 NaH₂PO₄, 30 NaHCO₃, 20 HEPES, 25 glucose, 2 MgSO₄, 2 CaCl₂, 5 sodium ascorbate, 3 sodium pyruvate and 2 thiourea, oxygenated with 95% O₂ and 5% CO₂, then adjusted to pH 7.3-7.4 and 310-315 mOsm of osmolarity with sucrose. The slice preparation procedure was modified from our previous study^42,123. Slices were placed in a recording chamber mounted on an upright microscope (BX51WI, Olympus Optical Co., Ltd, Japan) with epifluorescent equipment and continuously perfused with warm (29-31°C) recording-aCSF consisting of (in mM) 124 NaCl, 2.5 KCl, 1.25 NaH₂PO₄, 24 NaHCO₃, 5 HEPES, 12.5 glucose, 2 MgSO₄ and 2 CaCl₂, oxygenated with 95% O₂ and 5% CO₂, then adjusted to pH 7.3-7.4 and 305-310 mOsm of osmolarity with sucrose at a rate around 2.5-3mL/minute. Live tissue images were monitored by a digital CMOS camera (ORCA-Flash4.0LT, Hamamatsu Photonics, Japan) through a 40x water immersion objective lens (LUMPLFLN-40xW, Olympus, Tokyo, Japan) coupled with HC Image program (Hamamatsu Photonics, Japan). For cell-attached recordings, glass pipettes pulled by borosilicate glass capillary (GC150F-10, Warner Instruments, CT, USA) with a resistance around 5 MΩ when filled with recording-aCSF, either 470 or 505nm LED light from the epifluorescent system were used to identify the fluorescent expression in recorded cells. All electrophysiological data were collected by Multiclamp 700B amplifier (Molecular Devices, CA, USA) upon a low pass filtered at 2 kHz and digitalized at 10k Hz by Axon Digidata 1550B interface (Molecular Devices, CA, USA) coupled with Clampex software (Molecular Devices, CA, USA). Recording traces were collected and further analyzed by Clampex software, MATLAB (MathWorks, MA, USA) and GraphPad Prism 9 (GraphPad Software, MA, USA).

Pharmacological scan

For ex vivo calcium imaging recordings, brain slices were transferred in the recording chamber with continuous perfusion of warm recording-aCSF as described above. The excitation and emission illuminations were delivered and collected using the live imaging system and epifluorescent equipment mounted on the Olympus BX51WI microscope. A 470 nm LED light under command of Axon Digidata 1440A interface coupled with Clampex software was used for field illumination of blue light at intensity of 0.5-1mW/mm² based on the expression of calcium indicator. Fluorescent images were taken as 512 x 512-pixel square resolution after 4 x 4 binning covering 350 x 350 μm of FOV at 20 Hz with ≤ 50ms in exposure time (Figure 1B), the focal plane was fixed at typically around 25-75 μm from the upper surface of recorded slices along the entire experiment, a sturdy anchor made of platinum was used to reduce unwanted motions and maximize the number of detectable cells in the FOV at selected focal plane. Upon the pharmacological applications, image stacks with length of 95 seconds was captured during baseline conditions or pharmacological bath perfusions. Simultaneous single or dual cell-attached recordings were made during the entire procedure to obtain raw material for model fitting or act as a live indicator of pharmacological action. The intensity of 470 nm excitation light was mildly raised depending on the bleaching of fluorescent signals along the pharmacological scanning protocol but remained consistent within each recording epoch to maintain the quality of imaging data.

Selection of pharmacological agonists and dosages

All GPCR targeting agonists and the dosages were screened and selected through the following criteria prior to the main examination:

1. Selection of specific agonists based on the washout efficiency of ligand-induced changes in activity. Figure S2A shows examples of the initial pharmacological survey for mu opioid receptor using two selective agonists, DAMGO and fentanyl. Application of each agonist (1 μM) caused complete suppression of spontaneous action potential firing. However, only 200nM DAMGO-mediated effect showed sufficient variability and was successfully restored after 30 minutes of wash session. Accordingly, 200nM DAMGO was selected as the agonist and its working concentration for mu opioid receptor for the study.

2. Using agonists with irreversible effects as the final application in each pharmacological group. The middle traces in Figure S2B display the pharmacological test of calcitonin, the endogenous ligand for calcitonin receptor, that does not have known exogenous ligands. In line with the robust expression of calcitonin receptor in LC-NE neurons, application of 20 nM calcitonin caused strong, irreversible activation that was not restored at all following 30 minutes of aCSF perfusion. Therefore, this ligand, and others with similar effects, are the last to be tested in their respective pharmacological groups (Table 1).

3. Discarding agonists with no obvious effect in initial screens. The bottom traces in Figure S2B show that vasopressin (500 nM), the endogenous ligand for vasopressin receptors, did not cause a detectable effect following bath application. We therefore did not include this ligand, and others with similar results, in this study.

The excluded ligands and doses used in the initial screens are listed as the following: 2-Phenethylamine (10 μM), 4MH (20 μM), A11-DL15-OXB (1 μM), Alpha-MSH (500 nM), Amylin (20nM), Bradykinin (1 μM), BW723C86 (20 μM), CCK8 (500nM), Dimaprit (100 μM), EMD386088 (10 μM), Galanin (500 nM), Isoproterenol (10 μM), L-AP4 (200 μM), LY354740 (2.5 μM), Neuropeptide Y (100 nM), Neurotensin (500 nM), OXA 17-33 (500 nM), Oxytocin (500 nM), Quinpirole (10 μM), RAMH (20 μM), SKF38393 (10 μM), Vasopressin (500 nM).

Imaging data processing and spike deconvolution

All image stacks collected from the same slice were applied with 4 x 4 binning into 128 x 128 pixel-square then concatenated along the order of imaging recordings following the contrast adjustment to the same level across stacks by build-in functions of FIJI program¹⁵⁰. The non-rigid motion correction was processed onto the image stacks by using the Non-Rigid Motion Correction (NoRM-Corre) algorithm¹⁵¹ operated in MATLAB, followed by the baseline subtraction adopted a floating 2 second baseline along the whole stack then the first 5 seconds of each epoch were discarded due to the initial bleaching of fluorescent signals. The spatiotemporal information of each ROI was automatically extracted by extended constrained nonnegative matrix factorization (CNMF-E), an open-source algorithm for analysis of microendoscopic calcium imaging data¹⁵², with the criteria as following: minimum local correlation coefficient and peak-to-noise ratio for a seeding pixel = 0.7-0.8 and 3-4, respectively; average and maximum size of recorded neurons = 4 and 6 pixels, respectively. Typically, 15-30 ROIs were extracted per slice, temporal traces across ROIs were further separated by recording epochs for spike deconvolution (Figure S1A). 644 ROIs were extracted, including 489 (73.7%) used in this study; 62 (9.3%) found as repeated ROI and 113 (17.0%) discarded due to the lower signal-to-noise ratio. For the training of deconvolution model, ROIs represented the somata of recorded LC-NE neurons were aligned with simultaneous cell-attached recordings and formatted into algorithm-digestible files using MATLAB. CASCADE, a machine-learning-based algorithm for spike inference from calcium signals using Python⁵⁶, was used to fit the model for spike deconvolution based on training materials including around 6 hours (29 cells in 16 slices from 13 mice) of simultaneous recordings. For model evaluation, deconvoluted traces showing likelihood of spiking were calculated by the trained model upon the import of a pre-preserved set (15% of total materials, around 1 hour in length) of simultaneous recordings. Note that the recordings with relatively high firing frequency were intentionally collected for model training and evaluation due to the smaller cell population in normal brain slice preparation (Figure 1C). Spike inferences were made based on peaks and waveform of traces from both spike likelihood and raw calcium signal, then aligned to the ground truth representing the concomitant spikes by custom MATLAB scripts (Figures 1F and S1D). Upon the alignment, pairs between ground truth and predicted spike with temporal difference < 2 image frames (±125ms) were considered as a correct prediction (Figure S1E). For the spike deconvolution for pharmacological scan, calcium signals under baseline and pharmacological conditions were converted into traces of spike likelihood then the individual spike inferences were made as described above. To verify the cell identity by applications of clonidine, comparisons were made between inter-spike-intervals under baseline and clonidine application, simple judgements upon spike number were used when the firing rate was lower than 0.1 Hz. In a small subset of ROIs, spontaneous burst activities represented by the transient increase of bulk calcium signal as well as a short period of firing inhibition following the burst¹²³, were manually removed prior to the deconvolution process.

Immunohistochemistry

Mice were deeply anesthetized with an i.p. injection of cocktail containing ketamine, xylazine & acepromazine then perfused with icecold 4% paraformaldehyde in 0.1M PB. Brains were dissected and postfixed for 24 hours at 4°C with the same fixative. Upon the infiltration of 30% sucrose in 0.05M PB, frozen sections with 50μm thickness were cut through a sliding microtome (SM2000R, Leica, Germany). Sections were rinsed 3 times with PBS then incubated in blocking solution containing 2% bovine serum albumin (BSA) plus 5% normal goat serum (NGS) in PBST for 1 hour before transferred to PBST solutions containing primary antibodies plus 10% blocking solution for 24 hours at 4°C. Sections were then washed by PBS for 3 times and incubated in PBST solutions with secondary antibodies for 2-3 hours at room temperature followed by a rinse with PBS and wash with PB for 3 times. Sections were then mounted on glass slides with Vectashield mounting medium (Vector Labs, CA, USA). For post-hoc morphological reconstruction of slices used in the ex vivo calcium imaging recording, slices were fixed by 4% paraformaldehyde in 0.1M PB for few days and rinsed with PBS for 3 times. Slices were incubated in blocking solution consisting of 2% bovine serum albumin (BSA) plus 10% normal goat serum (NGS) in PBST for 1 hour followed by PBST solutions with primary antibodies plus 10% blocking solution for 72 hours at 4°C. Then slices were washed with PBS for 3 times and transferred to PBST solutions with secondary antibodies for 24 hours at 4°C followed by a rinse with PBS and wash for 3 times with PB. Slices were then incubated in RapiClear 1.47 (SUNJin Lab, Taiwan) for 1 hour and mounted with the same clearing agent for imaging scan. Images were collected using Leica confocal microscope (SP8, Leica, Germany), z-stacks with 2-3 μm intervals were scanned for the reconstruction of tissue morphology of recorded slices after post-hoc immunohistochemistry. The usage of antibodies is listed below:

Antibody	Catalog	Supplier	Dilution	Figure
mouse anti-TH	MAB318	MilliporeSigma, MA, USA	1:500	3C–E, 5C, S4B–E
rabbit anti-TH	AB152	EMD Millipore, MA, USA	1:2000	4B, S4F–I, S6K
chicken anti-GFP	GP1010	Aves Labs, CA, USA	1:1000	3C–E
rabbit anti-Ctb	ab34992	Abcam, Cambridge, UK	1:500	3C–E, S4B–E
rabbit anti-mCherry	600-401-P16	Rockland, NJ, USA	1:1000	5B–C, S4F–I

Hargreaves thermal paw withdrawal assay and intracerebral infusion

Mice were allowed to daily habituate with Hargreaves apparatus (IITC Life Science, CA, USA) for a week with 1 hour per day. Prior to behavioral examinations, mice were placed in the apparatus for 30 minutes for habituation. Upon thermal stimulation, hind paw withdrawal behavior was identified as the paw being removed from the glass surface of Hargreaves apparatus and the maximum duration was set at 20 seconds. Intensity of stimuli were adjusted between 25 to 40% of setting to obtain a paw withdrawal latency around 9 to 12 seconds for each mouse, measurements were made with this fixed intensity for the entire protocol. For the examinations coupled with intra-cerebral infusions, injector cannulas were connected to 2.0 μL Hamilton syringe (Neuros Model 7002 KH, Hamilton, NV, USA) mounted on a syringe pump (70-4500, Harvard Apparatus, MA, USA) and pre-filled with recording-aCSF based pharmacological solutions. Mice were connected to the injector cannulas before habituation session, 250 nL of solutions were infused at a rate of 50 nL/mins. During baseline measurements for each trial, mice didn’t receive any infusion but still connected to the setup. At least 3 days were allowed to washout the pharmacological effect before next tests. Measurements were made between 5-30 minutes after the end of infusions with 5 minutes intervals, results from both paws were averaged for further analysis.

Isobolographic analysis

The isobolographic analysis was used to evaluate the drug interaction within pharmacological combinations^83,153,154. Shortly, EC₅₀s were first calculated by a non-linear regression adopting the three-parameter dose-response formula as a built-in function in GraphPad Prism 10.0. Isobolograms were drawn by the EC₅₀s of pharmacological effect from single compound infusions, the theoretical EC₅₀s and variance were calculated on the additive isobole depending on the recipe of drug combinations. An additive interaction is considered when the experimental EC₅₀ shows no difference to the theoretical EC₅₀, whereas the significantly smaller or larger experimental EC₅₀ compared to theoretical EC₅₀ denote a synergism or antagonism, respectively.

Open field test

Mice were allowed to habituate to the test room for 1 hour prior to the test then connected to infusion setup and an additional 15 minutes of habituation was made. Mice then received 250 nL solutions at rate of 50 nL/min with either vehicle or 10μM Cocktail B, 5 minutes were allowed for the diffusion of injected solutions. Mice were individually placed into a 50 cm x 50 cm x 30 cm square acrylic box for 15 minutes. The behavioral activity was recorded through a Google Pixel 3 XL and the video. Traveled distances were further analyzed via Ethovision XT 13 (Noldus Information Technology, The Netherlands).

Quantification and statistical analysis

All data are expressed as mean ± SD except the isobologram with SEM demonstrations (Figure 4I). One-sample t-Test, Student’s t test, paired t test, One-way ANOVA or Two-way ANOVA followed by posthoc tests were used to analyze data that pass the Shapiro-Wilk normality test. Otherwise, nonparametric tests were used. Statistical significance was indicated as *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 and ns indicates not significant. Statistical comparisons were conducted in GraphPad Prism 10.0. All statistical tests performed as part of this manuscript are available in Data S1.

Supplementary Material

Supplemental information can be found online at https://doi.org/10.1016/j.celrep.2025.116294.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
mouse anti-TH	MilliporeSigma, MA, USA	RRID: AB_2201528
rabbit anti-TH	MilliporeSigma, MA, USA	RRID: AB_390204
chicken anti-GFP	Aves Labs, CA, USA	RRID: AB_2307313
rabbit anti-Ctb	Abcam, Cambridge, UK	RRID: AB_726859
rabbit anti-mCherry	Rockland, NJ, USA	RRID: AB_2614470
Bacterial and virus strains
AAV1-syn-FLEX-jGCaMP8f-WPRE	Addgene, MA, USA	RRID: Addgene_162379
CAV-mCherry	PVM, Biocampus, Montpellier, France	NA
CAV-PRS-Cre-V5	PVM, Biocampus, Montpellier, France	NA
Chemicals, peptides, and recombinant proteins
Ctb-CF594	Biotium, CA, USA	Cat. Number: 00072
8OH-DAPT	Tocris Bioscience, MN, USA	Cat. Number: 3624
Adenosine	Tocris Bioscience, MN, USA	Cat. Number: 0529
Anandimide	Tocris Bioscience, MN, USA	Cat. Number: 1339
Baclofen	Sigma-Aldrich, MO, USA	Cat. Number: B5399
Calcitonin	Anaspec, CA, USA	Cat. Number: AS-20673
Cevimeline	MedChemExpress, NJ, USA	Cat. Number: HY-76772
Clonidine	Tocris Bioscience, MN, USA	Cat. Number: 0690
CRF	Anaspec, CA, USA	Cat. Number: AS-24254
DOI	Sigma-Aldrich, MO, USA	Cat. Number: D101
DAMGO	Sigma-Aldrich, MO, USA	Cat. Number: E7384
DHPG	Tocris Bioscience, MN, USA	Cat. Number: 0805
Kynurenic acid	Sigma-Aldrich, MO, USA	Cat. Number: K3375
McN-A-343	Tocris Bioscience, MN, USA	Cat. Number: 1384
Muscarine	Tocris Bioscience, MN, USA	Cat. Number: 3074
N/O FQ	Tocris Bioscience, MN, USA	Cat. Number: 0910
Orexin A	Hello Bio, NJ, USA	Cat. Number: HB 2937
Picrotoxin	Hello Bio, NJ, USA	Cat. Number: HB 0506
Phenylephrine	Tocris Bioscience, MN, USA	Cat. Number: 2838
Pirenzepine	Tocris Bioscience, MN, USA	Cat. Number: 1071
U50488	Tocris Bioscience, MN, USA	Cat. Number: 0495
SNC162	Tocris Bioscience, MN, USA	Cat. Number: 1529
Somatostatin	Tocris Bioscience, MN, USA	Cat. Number: 1157
Strychnine	Sigma-Aldrich, MO, USA	Cat. Number: S0532
SubstanceP	Hello Bio, NJ, USA	Cat. Number: HB 2915
Vasopressin	Anaspec, CA, USA	Cat. Number: AS-24289
Experimental models: Organisms/strains
C57BL/6J mouse	The Jackson Laboratory, ME, USA	RRID: IMSR_JAX:000664
Dbh-Cre+/− mouse	The Jackson Laboratory, ME, USA	RRID: IMSR_JAX:033951
Oprm1fl/fl mouse	The Jackson Laboratory, ME, USA	RRID: IMSR_JAX:030074
Software and algorithms
CASCADE	https://github.com/HelmchenLabSoftware/Cascade	NA
CNMF-E	https://github.com/zhoupc/CNMF_E	NA
NoRMCorre	https://github.com/flatironinstitute/NoRMCorre	NA

Highlights.

• Ex vivo calcium imaging enables deconvolution of individual LC action potentials

• This slice imaging approach is used to profile pharmacological responses across 18 LC GPCRs

• MOR, 5HT1aR, mAChR1, and GABABR differentially alter projection-selective LC firing rates

• Agonist cocktails synergistically reduce LC-mPFC activity to drive antinociception

Data and code availability

• All statistical tests performed in this study are available in Data S1.

• All data presented in this article are available in Data S2.

• The custom code used here is included in Data S3.

• Any additional information required to reanalyze the data reported in this work is available from the lead contact upon request.