Opioids depress breathing through two small brainstem sites

The rates of opioid overdose in the United States quadrupled between 1999 and 2017, reaching a staggering 130 deaths per day. This health epidemic demands innovative solutions that require uncovering the key brain areas and cell types mediating the cause of overdose— opioid-induced respiratory depression. Here, we identify two primary changes to murine breathing after administering opioids. These changes implicate the brainstem’s breathing circuitry which we confirm by locally eliminating the µ-Opioid receptor. We find the critical brain site is the preBötzinger Complex, where the breathing rhythm originates, and use genetic tools to reveal that just 70–140 neurons in this region are responsible for its sensitivity to opioids. Future characterization of these neurons may lead to novel therapies that prevent respiratory depression while sparing analgesia.


Introduction
Nearly 400,000 people in the United States died from a drug overdose involving a prescription or illicit opioid between 1999 and 2017 (Scholl et al., 2018). This epidemic is not unique to the United States and with the increasing distribution of highly potent synthetic opioids like fentanyl, it has become a global public health emergency (Rudd et al., 2016). Death from opioid overdose results from slow and shallow breathing, also known as opioid induced respiratory depression (OIRD, Pattinson, 2008). Like humans, breathing in mice is severely depressed by opioids and this response is eliminated when the m-Opioid receptor (Oprm1) is globally deleted (Dahan et al., 2001). Oprm1 is broadly expressed, in both the central and peripheral nervous systems, including sites that could modulate breathing such as: the cerebral cortex, brainstem respiratory control centers, primary motor neurons, solitary nucleus, and oxygen sensing afferents (Mansour et al., 1994;Kirby and McQueen, 1986). Therefore, either one or multiple sites could be mediating the depressive effects of opioids on breathing.
Indeed, multiple brain regions have been shown to independently slow breathing after local injection of opioid agonists (Kirby and McQueen, 1986;Montandon et al., 2011;Mustapic et al., 2010;Prkic et al., 2012). Although informative, doubts remain for which of these sites are necessary and sufficient to induce OIRD from systemic opioids for three reasons. First, injection of opioid agonists or antagonists into candidate areas modulates m-opioid receptors on the cell body (post-synaptic) as well as receptors on incoming terminals (pre-synaptic). Second, these studies necessitate anesthetized and reduced animal preparations, which alter brain activity in many of the candidate Oprm1 expressing sites. And third, there is not a standard and quantitative definition for how breathing changes in OIRD, and this makes comparing studies that use different breathing metrics measured in different experimental paradigms challenging.
To address these limitations, we conducted a detailed quantitative analysis of OIRD in awake animals and identify two key changes to the breath that drive the depressive effects of opioids. These two metrics thereafter define OIRD in our study and can serve as a rubric for others. We then locally eliminate the m-opioid receptor in awake mice, disambiguating pre and post-synaptic effects, and use these metrics to define two key brain sites that mediate OIRD. Recently, a similar approach demonstrated some role for these sites in OIRD (Varga et al., 2020). Among these two sites in our study, we find that one is dominant and driven by just 140 critical neurons in vitro and, importantly, these neurons are not required for opioid-induced analgesia, suggesting a neutral target for developing safer opioids or rescue strategies for opioid overdose.

Results
Up to now, OIRD has generally been described as a slowing and shallowing of breath (Pattinson, 2008). We therefore felt it was important to more precisely, quantitatively describe the changes in breathing in hopes of elucidating potential mechanisms of respiratory depression. We began by asking whether specific parameters of the breath are affected by opioids. We monitored breathing in awake, behaving mice by whole body plethysmography after intraperitoneal injection (IP) of saline for control and then 20 mg/kg morphine at least 24 hr later ( Figure 1A). Compared to saline, breathing after morphine administration (in normoxia) became much slower and inspiratory airflow decreased, each by 60% ( Figure 1B,C). This culminated in~50% decrease in overall minute ventilation (MV = approximated tidal volume x respiratory rate, Figure 1C), demonstrating that 20 mg/kg morphine is, indeed, a suitable dose to model OIRD.
Breath morphology in normoxia after IP saline versus morphine cannot be directly compared since activity of the mouse is different (exploring vs. sedated, Supplementary file 1), which significantly influences the types of breaths taken. This prevented a precise characterization of breath parameters that dictate OIRD. To overcome this, we measured breathing in hypercapnic air (21% O 2 , 5% CO 2 ) which normalizes behavior and thus breathing ( Figure 1D, Supplementary file 1). As in normoxia, morphine depressed respiratory rate (by 50%, Figure 1E,F), peak inspiratory airflow (by 60%, Figure 1E,F), and minute ventilation (by 60%, Figure 1F). Hypercapnic breaths after saline exhibited two phases, inspiration and expiration, each lasting about 50 msec. (Figure 1G,H). After morphine, only the inspiratory phase (measured as inspiratory time, Ti) became substantially longer ( Figure 1G,H). Additionally, hypercapnic breaths showed a new, third phase after the initial expiration (measured as expiratory time, Te, Figure 1G,H) that was characterized by prolonged little to no airflow (<0.5 mL/sec.) preceding hypercapnia induced active expiration (Pisanski and Pagliardini, eLife digest Opioids such as morphine or fentanyl are powerful substances used to relieve pain in medical settings. However, taken in too high a dose they can depress breathing -in other words, they can lead to slow, shallow breaths that cannot sustain life. In the United States, where the misuse of these drugs has been soaring in the past decades, about 130 people die each day from opioid overdose. Pinpointing the exact brain areas and neurons that opioids act on to depress breathing could help to create safer painkillers that do not have this deadly effect. While previous studies have proposed several brain regions that could be involved, they have not been able to confirm these results, or determine which area plays the biggest role. Opioids influence the brain of animals (including humans) by attaching to proteins known as opioid receptors that are present at the surface of neurons. Here, Bachmutsky et al. genetically engineered mice that lack these receptors in specific brain regions that control breathing. The animals were then exposed to opioids, and their breathing was closely monitored.
The experiments showed that two small brain areas were responsible for breathing becoming depressed under the influence of opioids. The region with the most critical impact also happens to be where the breathing rhythms originate. There, a small group of 50 to 140 neurons were used by opioids to depress breathing. Crucially, these cells were not necessary for the drugs' ability to relieve pain.
Overall, the work by Bachmutsky et al. highlights a group of neurons whose role in creating breathing rhythms deserves further attention. It also opens the possibility that targeting these neurons would help to create safer painkillers.   Representative examples of the breathing airflow (mL/sec.) in normoxia (21% O 2 ) measured by whole body plethysmography. 15 min before recordings, animals are intraperitoneally (IP) injected with either saline (black) or 20 mg/kg morphine (red). Morphine recordings are captured 1-7 days after saline. (B) Scatterplot of instantaneous breathing rate (Hz) versus airflow (mL/sec.) for each breath (dot) taken during the 40 min recordings. Morphine causes breathing to become slow and less forceful. (C) Ratio of average breathing parameters after IP injection of morphine-to-saline. Respiratory rate, peak inspiratory airflow, and minute ventilation (MV = approximated tidal volume*rate) for n = 29 animals in normoxia. Red diamond, single animal average. Black diamond, average of all animals. Error bar, standard error of mean (SEM). (D-E) Representative example of breathing airflow and instantaneous scatter plot (rate vs. airflow) from a 10 min whole body plethysmography recording of breathing in hypercapnia (21% O 2 , 5% CO 2 ) to minimize changes in breathing due to differences in behavior after morphine injection. (F) Ratio of respiratory rate (p-value=1Â10 À19 , Cohen's d = 5.96), peak inspiratory airflow (p-value=1Â10 À22 , Cohen's d = 6.18), and minute ventilation (p-value=1Â10 À20 , Cohen's d = 5.31) after IP injection of morphine-to-saline for n = 29 animals in hypercapnia. (G) Representative single breath airflow trace for breaths in hypercapnia after saline (black) or morphine (red) IP injection. Hypercapnic saline breaths can be divided into two phases whose durations (msec.) can be measured: inspiration (Ti) and expiration (Te). Hypercapnic morphine breaths have a third phase after expiration where airflow is nearly 0 mL/sec., which we call a pause. (H) Bar graph of the average length ± standard deviation of Ti and Te for a single representative animal. (I) Probability density function plot of the pause length in breathing during hypercapnia after saline (black) or morphine (red) IP injection on a numerical (left) and logarithmic scale (right). Note, morphine selectively increases Ti and pause length. (J) Percent of the Figure 1 continued on next page 2019). We define this new phase as a pause (low airflow + active expiration, Figure 1G, Figure 1figure supplement 1). Such pauses lasted up to several hundred milliseconds ( Figure 1I), accounting for about one-third of the average breath length ( Figure 1J). Thus, the 50% decrease in respiratory rate after morphine administration is primarily due to prolonging of Ti and pause phases, and the increased prevalence of time spent in pause significantly contributes to the decrease in minute ventilation.
Typically the length of inspiratory time is determined by a stretch-evoked feedback signal from the lung which terminates inspiration (West, 2005). This reflex is represented by the correlation observed between Ti and peak inspiratory airflow ( Figure 1K). Breaths in morphine still maintain this correlation despite having a longer Ti and decreased inspiratory airflow ( Figure 1K). As a result, morphine breaths have a similar approximated tidal volume (TV) compared to saline control ( Figure 1L,M). In other words, as opioids decrease inspiratory airflow, Ti displays a compensatory increase to preserve TV (Figures 1N and Hill et al., 2018). In summary, opioids cause only two primary changes to the breath, namely, 1) decreased inspiratory airflow and 2) addition of a pause phase that delays initiation of subsequent breaths ( Figure 1N). These two parameters can both be controlled by the breathing central pattern generator, the preBö tzinger Complex (preBö tC), in the brainstem and suggest that this may be a key locus affected during OIRD (Smith et al., 1991;Feldman et al., 2013;Cui et al., 2016).
Indeed, the preBö tC has been proposed to play a key role in OIRD since localized injection of opioids results in respiratory depression and localized naloxone reverses decreased breathing after administration of systemic opioids (Montandon et al., 2011;Montandon and Horner, 2014). However, such experiments fail to distinguish between the action of opioids on presynaptic terminals (Mudge et al., 1979) of distant neurons projecting into the preBö tC versus direct action on preBö tC neurons themselves (Figure 2A; Montandon et al., 2011;Dobbins and Feldman, 1994;Gray et al., 1999) To overcome this, we genetically eliminated the m-Opioid receptor (Oprm1) from preBö tC cells exclusively, sparing projecting inputs, by stereotaxic injection of adenoassociated virus constitutively expressing Cre (AAV-Cre-GFP) into the preBö tC of Oprm1 flox/flox (Oprm1 f/f ) adult mice ( Figure 2B). Injection site specificity was confirmed by the restricted expression of Cre-GFP (Figure 2-figure supplement 1), and subsequent Oprm1 deletion, was inferred. To establish a baseline, we first measured breathing after administration of saline and morphine in normoxia and hypercapnia in intact animals, as described above. At least one month after bilateral injection of virus into the preBö tC, we then re-analyzed breathing ( Figure 2C). With this protocol, each animal's unique breathing and OIRD response serves as its own internal control, which is necessary due to the variability in OIRD severity between mice ( Figure 1F). Deletion of Oprm1 in the pre-Bö tC did not affect breathing observed after saline injection ( Figure 2D,E), suggesting that in this context, opioids do not exert an endogenous effect. In contrast, breathing was markedly less depressed by morphine administration ( Figure 2D,E) compared to the intact control state: breaths average breath period spent in inspiration, expiration, or pause for hypercapnic breaths after saline or morphine injection. (K) Scatterplot of inspiratory time (msec.) vs. peak inspiratory airflow (mL/sec.) for 10 min of hypercapnic breaths after saline (black) or morphine (red) IP injection. As inspiratory time increases, peak inspiratory flow decreases. (L) Scatterplot of tidal volume (mL.) vs. peak inspiratory airflow (mL/sec.) for 10 min of hypercapnic breaths after saline (black) or morphine (red) IP injection. Even though peak inspiratory airflow decreases after morphine, tidal volume is preserved due to prolonged Ti. (M) Ratio of tidal volume after IP injection of morphineto-saline for n = 29 animals in hypercapnia (p-value=3Â10 À6 , Cohen's d = 1.13). (N) Schematic of the two key morphine induced changes to the breath: decreased inspiratory airflow and pause. Decreased inspiratory airflow prolongs Ti since negative feedback from the lung reflecting breath volume is slower. We interpret the pause as a delay in initiation of the subsequent inspiration. The online version of this article includes the following source data and figure supplement(s) for figure 1: Source data 1. Average frequency, peak flow, minute ventilation, tidal volume and pause duration for 31 animals after intraperitoneal saline or morphine in normoxia and hypercapnia.    (Oprm1) is expressed by a subset of preBö tC neurons and is also expressed on presynaptic terminals of some neurons projecting to the preBö tC. This confounds the effects observed after localized preBö tC injection of opioid or naloxone to investigate its role in OIRD. (B) To overcome this, we eliminate Oprm1 only from the preBö tC, and not the presynaptic inputs, to define the role of preBö tC neurons in OIRD. (C) Experimental time-course. Breathing is measured 15 min after IP injection of saline or 20 mg/kg morphine in both normoxia and hypercapnia. Then Oprm1 f/f animals are injected with a constitutive-Cre AAV into the preBö tC bilaterally. After several weeks breathing is assayed again as above. In this way, each animal's breathing before viral injection can serve as its own control breathing. Response to pain is also measured with a tail-flick assay before and after viral injection to ensure that analgesic response is unaffected. (D-E) Representative examples of the breathing airflow (mL/sec.) in hypercapnia after saline (black) or morphine (red) before (D) and after (E) Cre-virus injection. (F) Scatter plot of instantaneous respiratory frequency vs. airflow (ml/ sec), as well as probability density function plots of both parameters for a representative animal during hypercapnia after saline (black) or morphine (red, blue) IP injection, before (red) and after (blue) Cre-injection.
(G) Probability density function plot of pause length (msec.) for a representative animal during hypercapnia after IP morphine before (red) and after ( Although key features of OIRD (inspiratory airflow and pause) were attenuated by preBö tC AAV-Cre injection, rescue was incomplete. This could be explained by incomplete Oprm1 deletion within the preBö tC, or participation of another brain site in OIRD. Injection of opioids into the parabrachial (PBN)/Kolliker-Fuse (KF) nucleus can also slow breathing, making it a candidate second site (Mustapic et al., 2010;Prkic et al., 2012). In fact, the PBN/KF has been proposed to be the key site mediating OIRD (Eguchi et al., 1987;Lalley et al., 2014). We therefore took a similar approach to test the role of the PBN/KF in OIRD ( Figure 3A Figure 3F,G), but had a more moderate effect than injection into the preBö tC.
To determine if the preBö tC and PBN/KF can completely account for OIRD ( Figure 3A), we genetically deleted Oprm1 from the preBö tC and then from the PBN/KF (Cohort 1) or vice versa (Cohort 2, Figure 3B). In either cohort, double deletion breathing after morphine administration looked nearly identical to that of saline control animals ( Figure 3C), with breathing rate and inspiratory airflow depressed by only~20% compared to saline ( Figure 3D Figure 3E,G). Breathing in double-deleted animals was even resilient to super-saturating doses of opioid that severely slow breathing in control animals (150 mg/kg fentanyl, Figure 3H,I). Taken together, our data are consistent with a model in which both the preBö tC and PBN/KF contribute to opioid respiratory depression, with the former being predominant, and together account for OIRD.
Given the relative importance of the preBö tC to OIRD, we sought to identify which Oprm1 expressing cells within this region depress breathing. Single cell transcriptome profiling of the ventral lateral brainstem of P0 mice ( Figure 4A) showed that Oprm1 (mRNA) is expressed almost exclusively by neurons (Figure 4-figure supplement 1) and is remarkably restricted to just 8% of presumed preBö tC neurons ( Figure 4B). This alone is interesting, as it suggests that modulation of only a small subset of neurons with the preBö tC is enough to significantly impact its ability to generate a rhythm. We also determined that within the preBö tC, Oprm1 (mRNA) was expressed by representing post viral injection conditions in experimental and control animals, respectively. Diamond and square, single animal average. Mixed repeated two-way ANOVA comparing Cre vs. Sham injected was statistically significant for respiratory rate (F(1,11)=5.5, p-value=0.04) and pause (F(1,11)=6.2, p-value=0.03), but not for peak inspiratory airflow (F(1,11)=2.1, p-value=0.17) Black diamond, average of all animals. Error bar, standard error of mean (SEM). * indicates p-value<0.05. ns indicates p-value>0.05. The online version of this article includes the following source data and figure supplement(s) for figure 2: Source data 1. Average frequency, peak flow and pause duration after intraperitoneal saline or morphine in hypercapnia at baseline or after Oprm1 deletion from the preBö tC. Source data 2. Average frequency, peak flow and pause duration after intraperitoneal saline or morphine in hypercapnia at baseline or after sham viral injection into the preBö tC.     glycinergic (Slc6a5 expressing), gabaergic (Gad2/Slc32a1 expressing), and glutamatergic (Slc17a6 expressing) neural types alike ( Figure 4C) and therefore Oprm1 (mRNA) expression is not exclusive to any known rhythmogenic preBö tC subpopulation. Slices containing the preBö tC autonomously generate respiratory-like rhythmic activity in vitro which is depressed in both rate and amplitude by bath administration of opioid agonists (Gray et al., 1999;Lorier et al., 2010), similar to opioid effects we observed on breathing in vivo.
Next, we dissected the glutamatergic Oprm1 preBö tC neurons by two developmental transcription factors, Dbx1 or Foxp2 (Gray et al., 2010;Bouvier et al., 2010;Gray, 2013), to determine if a subset can rescue rhythm depression ( Figure 4G). Triple-labeling of Dbx1-YFP, OPRM1 fused to mCherry (OPRM1-mCherry), and FOXP2 protein quantified within a single preBö tC revealed three molecular subtypes of P0 Oprm1 glutamatergic neurons: 92 ± 9 Dbx1, 50 ± 2 Dbx1/FOXP2, and 20 ± 4 FOXP2 ( Figure 4H,I). We selectively eliminated the m-Opioid receptor in these two lineages (Dbx1-Cre;Oprm1 f/f or Foxp2-Cre;Oprm1 f/f ) and measured preBö tC slice activity at increasing concentrations of DAMGO, exceeding the dose necessary to silence the control rhythm (500 nM vs. 50 nM). Elimination of Oprm1 from both genotypes was sufficient to rescue the frequency and amplitude of preBö tC bursting in DAMGO, and the Dbx1 rescue was comparable to elimination of Oprm1 from all glutamatergic neurons, while the Foxp2 rescue was substantial, but partial (~50-60%, Figure 4J). This shows that opioids silence a small cohort (~140) of glutamatergic neurons to depress preBö tC activity, and that a molecularly defined subpopulation, about half, can be targeted to rescue these effects.

Discussion
Here we show that two small brainstem sites are sufficient to rescue opioid induced respiratory depression in vivo. Between them, the preBö tC is the critical site and we molecularly define~140 Oprm1 glutamatergic neurons within that are responsible for this effect in vitro. Future study of these neurons will provide the first example of endogenous opioid modulation of breathing. Furthermore, characterization of these neurons and their molecular response to opioids may extend existing strategies (Manzke et al., 2003) or reveal a novel strategy for separating respiratory depression Source data 1. Average frequency, peak flow and pause duration after intraperitoneal saline or morphine in hypercapnia at baseline or after Oprm1 deletion from the preBö tC or PBN/KF and then the preBö tC and PBN/KF.      The preBö tC has input to the hypoglossal motor neurons which form the CN12 rootlet, relaying an inspiratory motor command to the tongue in intact animals. Due to its input from the preBö tC, extracellular recording from this rootlet display autonomous rhythmic activity corresponding to in vitro respiration (Mansour et al., 1994). A rubric for studying opioid and other respiratory depressants We find that although multiple breathing parameters are impacted by opioids, decreased inspiratory airflow and delayed breath initiation, which we term pause, represent the primary changes that result in OIRD. Pauses occur during expiration and account for tens to hundreds of milliseconds of low airflow per breath. In hypercapnia, these pause periods terminate with an active expiration. The force, timing of inspiration, and expression of active expiration are ultimately determined by the inspiratory rhythm generator, the preBö tC, and focused our initial studies to this site (Smith et al., 1991;Feldman et al., 2013;Cui et al., 2016;Huckstepp et al., 2016). Additionally, correlates of these two changes manifest as decreased burst size and frequency in the activity of the preBö tC slice in vitro. These two key changes in the breath can guide future OIRD studies and efforts to characterize and test novel opioid drugs. Additionally, this workflow can be applied to the analysis of other respiratory depressants.

Just two small brainstem sites mediate OIRD
Our experimental design allowed us to determine that both the preBö tC and PBN/KF have independent and additive rescue of OIRD. Of these two, the preBö tC has the larger magnitude rescue of our two core breathing parameters. The combined deletion of m-Opioid receptor from both sites essentially eliminates OIRD, even to extremely high doses of the potent opioid fentanyl. This suggests that targeting just these two sites is sufficient to rescue opioid respiratory depression. We interpret the small remaining effect of opioids we observe to be due to incomplete transduction of these brain areas but cannot rule out other minor contributing sites. It is also possible, given the challenge of restricting viral transduction, that some of the demonstrated effects are already due to spillover deletion of m-Opioid receptor in neighboring brain areas, such as the bulbospinal rostral ventral respiratory group. Our studies were limited to two opioids (morphine and fentanyl) at a specific dose and it will be important in future work to determine if these two brain sites are also critical for OIRD caused by these opioids at different doses or other opioids altogether.

Depression of preBö tC rhythm by silencing a small glutamatergic subpopulation
The two hallmark changes during OIRD, decreased inspiratory airflow and delayed initiation, perfectly match the opioid induced depression of amplitude and frequency in the preBö tC slice. We show that~140 Oprm1 glutamatergic preBö tC neurons mediate this effect. And surprisingly, half this number, just~70 glutamatergic neurons are sufficient to rescue opioid depression of the pre-Bö tC rhythm (50 Dbx1/FOXP2 and 20 FOXP2 in Foxp2-Cre;Oprm1 f/f , Figure 4-figure supplement  3). Further, given the importance of Dbx1 neurons in respiratory rhythm generation (Gray et al., 2010;Bouvier et al., 2010), rescuing Oprm1 in just~50/140 Dbx1 neurons (the Foxp2+ subset) may be sufficient to prevent preBö tC depression, the smallest number of neurons we propose. This small number is remarkably consistent with the number of Dbx1 neurons that must be lesioned to arrest preBö tC activity (Wang et al., 2014). Given the similarity of these effects, we hypothesize that opioids are primarily acting by silencing presynaptic release, effectively removing these neurons from the network. Alternatively, it is proposed that just a small subset of preBö tC excitatory neurons may generate the inspiratory rhythm (Phillips et al., 2019) and perhaps these key Oprm1 neurons are enriched within this subgroup. In this instance, since hyperpolarization of preBö tC rhythmogenic neurons slows and silences breathing (Koizumi et al., 2016), opioids may act postsynaptically as proposed by others (Montandon et al., 2011;Johnson et al., 1996). Regardless, it is profound that such a small number can abruptly halt the respiratory rhythm in a network of more than 1000 neurons and suggests that either these neurons act as a key population for rhythmogenesis, or that recurrent excitatory networks are exquisitely sensitive to the number of participating cells. Important future work will need to use the Dbx1/Oprm1 and Foxp2/Oprm1 molecular codes to selectively eliminate Oprm1 from these neurons to test if truly so few neurons profoundly control or modulate breathing in vivo.

Materials and methods
Key resources

Recombinant viruses
All viral procedures followed the Biosafety Guidelines approved by the University of California, San Francisco (UCSF) Institutional Animal Care and Use Program (IACUC) and Institutional Biosafety Committee (IBC). The following viruses were used: AAV5-CMV-Cre-GFP (4.7 Â 10 19 particles/mL, The Vector Core at the University of North Carolina at Chapel Hill), AAV5-CAG-GFP (1.0 Â 10 13 particles/mL, The Vector Core at the University of North Carolina at Chapel Hill) or AAV5-CAG-tdtomato (4.3 Â 10 12 particles/mL, The Vector Core at the University of North Carolina at Chapel Hill).

Plethysmography, respiratory and behavioral analysis
Adult (8-20 weeks) Oprm1 f/f mice were first administered either IP 100-200 mL of saline or morphine (20 mg/kg, Henry Schein 057202) and placed in an isolated recovery cage for 15 min to allow full onset of action of the drug. Individual mice were then monitored in a 450 mL whole animal plethysmography chamber at room temperature (22˚C) in 21% O 2 balanced with N 2 (normoxia) or 21% O 2 , 5% CO 2 balanced with N 2 (hypercapnia). For fentanyl (150 mg/kg, Sigma F3886) onset of action was so fast (<10 s) that animals were placed directly in the plethysmography chamber after administration of drug. Each session (combination of drug and oxygen condition) was separated by at least 24 hr to allow full recovery. Breathing was monitored by plethysmography, and other activity in the chamber monitored by video recording, for 40 min periods in normoxia and 10 min periods in hypercapnia. In cases where mice were subject to single or double site AAV injection to delete Oprm1 or sham controls, breathing was recorded first before viral injection and then again after deletion (or sham) more than 4 weeks later. Breathing traces were collected using EMKA iOX2 software and exported to Matlab for analysis. Each breath was automatically segmented based on airflow crossing zero as well as quality control metrics. Respiratory parameters (e.g. peak inspiratory flow, instantaneous frequency, pause length, tidal volume, etc) for each breath, as well as averages across states, were then calculated. Instantaneous frequency was defined as the inverse of breath duration. Pause length was defined as the expiratory period after airflow dropped below 0.5 mL/sec. The pause period is initially a prolonged airflow around or just above 0 mL/sec. and terminates with an increase in expiratory airflow, likely the active expiration induced by hypercapnia (11, Figure 1-figure supplement 1). The 0.5 mL/sec. threshold was chosen since it identifies low airflow pauses that are just above 0 mL/sec. which rarely occur in control hypercapnic breaths (Figure 1-figure supplement 2).
Although pauses in length 50-100msec. could be considered false positives because they do not have a considerable low airflow period (see Figure 1-figure supplement 1 and , panel 2), they occur at a low rate in saline (1.53%) and we see an increased distribution of pauses in morphine that last hundreds of milliseconds ( Figure 1I, see Figure 1-figure supplement 1 and , panel 3 and 4).
Other respiratory parameters were defined by when airflow crosses the value of 0, with positive to negative being inspiration onset and negative to positive being expiration onset. Note, reported airflow in mL/sec. and tidal volume in mL are approximates of the true volumes. Whole body plethysmography imperfectly measures these parameters without corrections for humidity and temperature. However, since humidity and temperature are largely stable between recordings, because they are conducted in a temperature and humidity stable mouse facility, the estimated airflow (mL/sec.) and tidal volume (mL) can be compared in saline vs. morphine or pre and post-Cre virus injection studies. Additionally, in some instances respiratory parameters are appropriately normalized to animal weight in order to accurately compare between animals. However, this normalization is not appropriate for our study since lung volume in mice does not change in adulthood (Mitzner et al., 2001), and all respiratory measurements are compared statistically as the ratio of saline to morphine injections within the same animal. The analysis was performed with custom Matlab code available on Github with a sample dataset (Bachmutsky, 2020, https://github.com/Yackle-Lab/Opioids-depress-breathing-through-two-small-brainstem-sites; copy archived at https://github. com/elifesciences-publications/Opioids-depress-breathing-through-two-small-brainstem-sites). Due to limitations in breeding, a power calculation was not explicitly performed before our studies. Studies were conducted on all mice generated; six cohorts of animals. After respiration was measured, mice were sacrificed and injection sites were validated before inclusion of the data for further statistical analysis. We first conducted a Shapiro-Wilk normality on the average values (averaged across breaths) of the pre-and post-morphine respiratory parameters (e.g., peak inspiratory flow, instantaneous frequency) from n = 29 animals. We then used either paired Student's t-test (if normal) or Wilcox Rank Sum test (if not normal) to evaluate statistical significance in comparing the distribution of these values. In comparisons of Oprm1-deleted vs. Sham conditions a mixed-repeated measure two-way ANOVA was performed to determine if these two groups were significantly different. Post-hoc Student's t or Wilcox Rank Sum tests were then used to evaluate statistical significance between normalized (morphine/saline, or morphine-saline) respiratory parameters for intact vs. Oprm1 deleted or intact vs. Sham conditions. Normality in this case was determined by Shapiro-Wilk test on the distribution of normalized respiratory parameters from n = 29 animals. All the above statistics were performed using the publicly available Excel package 'Real Statistics Functions' and SPSS.

Tail flick assays
Mice were injected with saline (control trials) or 20 mg/kg morphine. 15 min later mice were put into a restraining wire mesh with the tail exposed. One-third of the tail was dipped into a 48-50˚C water bath and time was measured for the tail to flick. Immediately after the flick, the tail was removed from the bath. If the tail did not flick within 10 s, then the tail was removed. The procedure was video recorded so time to response could be quantified post-hoc. Each mouse was recorded for two saline and two morphine trials.

Stereotaxic injection
Bilateral stereotaxic injections were performed in mice anesthetized by isoflurane. Coordinates used for the preBo€tC were: À6.75 mm posterior, À5.05 mm ventral from surface, ±1.3 mm lateral from bregma. Coordinates used for the PBN/KF were: À5.05 mm posterior, À3.7 ventral from surface, ±1.7 lateral from bregma. Injection sites specificity was confirmed by the restricted expression of Cre-GFP, GFP, or tdTomato centered in the anatomically defined Parabrachial/Kolliker-Fuse (Levitt et al., 2015) and preBö tC (Smith et al., 1991;Feldman et al., 2013) areas. In the case of preBö tC injections, anatomical location of injection site was also confirmed by localization with Somatostatin antibody staining (Tan et al., 2008). m-Opioid receptor deletion was not explicitly demonstrated by immunohistochemistry. After injection of the virus, mice recovered for at least 3-4 weeks before breathing metrics were recorded again. In a subset of animals, mice were then subject to a second site deletion of the complementary brain area, ie. preBö tC and then from the PBN/KF (Cohort 1) or vice versa (Cohort 2). These mice were then allowed to recover for another period of at least 3-4 weeks, after which a third set of breathing metrics were recorded. A subset of PBN/KF injected mice had only unilateral expression of Cre and their use is acknowledged in the text.
Single cell mRNA sequencing and analysis 650 mm-thick medullary slices containing the preBö tC were prepared from 10 P0 mice C57Bl/6 mice as described above. The preBö tC and surrounding tissue was punched out of each slice with a P200 pipette tip and incubated in bubbled ACSF containing 1 mg/ml pronase for 30 min at 37˚C with intermittent movement. Digested tissue was centrifuged at 800 rpm for 1 min, and the supernatant was discarded and replaced with 1% FBS in bubbled ACSF. The cell suspension was triturated serially with fire-polished pipettes with~600 mm,~300 mm and~150 mm diameter. The cells were filtered using a 40 mm cell strainer (Falcon 352340). DAPI was added to a final concentration of 1 mg/mL. The cell suspension was FACS sorted on a BD FACS AriaII for living (DAPI negative) single cells. The cells were centrifuged at 300 g for 5 min and resuspended in 30 mL 0.04% BSA in PBS. The library was prepared using the 10x Genomics Chromium Single Cell 3' Library and Gel Bead Kit v2 (1206267) and according to manufacturer's instructions by the Gladstone genomics core. The final libraries were sequenced on HiSeq 4000.
For analysis, sequencing reads were processed using the 10x Genomics Cell Ranger v.2.01 pipeline. A total of 1860 cells were sequenced. Further analysis was performed using Seurat v2.3. Cells with less than 200 genes were removed from the dataset. Data was LogNormalized and scaled at 1e4. Highly variable genes were identified and used for principal component analysis. 25 principal components were used for unsupervised clustering using the FindCluster function. 12 clusters were identified at a resolution of 1.0, displayed in Figure 4-figure supplement 1. FindAllMarkers and violin plots of known cell type markers were used to identify each cluster.