Necdin shapes serotonergic development and SERT activity modulating breathing in a mouse model for Prader-Willi syndrome

Prader-Willi syndrome (PWS) is a genetic neurodevelopmental disorder that presents with hypotonia and respiratory distress in neonates. The Necdin-deficient mouse is the only model that reproduces the respiratory phenotype of PWS (central apnea and blunted response to respiratory challenges). Here, we report that Necdin deletion disturbs the migration of serotonin (5-HT) neuronal precursors, leading to altered global serotonergic neuroarchitecture and increased spontaneous firing of 5-HT neurons. We show an increased expression and activity of 5-HT Transporter (SERT/Slc6a4) in 5-HT neurons leading to an increase of 5-HT uptake. In Necdin-KO pups, the genetic deletion of Slc6a4 or treatment with Fluoxetine, a 5-HT reuptake inhibitor, restored normal breathing. Unexpectedly, Fluoxetine administration was associated with respiratory side effects in wild-type animals. Overall, our results demonstrate that an increase of SERT activity is sufficient to cause the apneas in Necdin-KO pups, and that fluoxetine may offer therapeutic benefits to PWS patients with respiratory complications.


Introduction
Respiration is a complex function controlled in large part by raphe serotonergic (5-HT) neurons (Teran et al., 2014). Central 5-HT depletion induces severe apneas during the early postnatal period (Barrett et al., 2016;Trowbridge et al., 2011) and serotonopathy is implicated in the genesis of breathing disorders in human pathologies including neurodevelopmental diseases such as Sudden Infant Death Syndrome (Duncan et al., 2010;Hilaire et al., 2010;Kinney et al., 2011;Paterson et al., 2009), Rett syndrome (Abdala et al., 2010;Toward et al., 2013) and Prader-Willi Syndrome (PWS) (Zanella et al., 2008). However, the cellular and molecular events that underlie serotonopathy, and the causal link between serotonopathy and respiratory dysfunction in these pathologies are poorly understood.
A role for serotonergic transmission in the genesis of respiratory dysfunction in the Necdin-KO model is supported by neuroanatomical studies: Pagliardini and colleagues report abnormal morphology and orientation of axonal fibers that contain large 5-HT/Substance P varicosities in the developing Ndn tm1-Stw -KO medulla (Pagliardini et al., 2005;Pagliardini et al., 2008). Similarly, we eLife digest Prader-Willi syndrome results from the disruption of a cluster of neighboring genes, including one called Necdin. Symptoms begin in early infancy and worsen with age. Affected children tend to develop an insatiable appetite, which often leads to obesity. They also experience serious problems with their breathing. Chest infections, high altitude and intense physical activity can be dangerous for children with Prader-Willi syndrome. This is because a slight shortage of oxygen may trigger breathing difficulties that could prove fatal. The brain cells that produce a chemical messenger called serotonin help to control breathing. Several lines of evidence suggest that loss of Necdin may trigger breathing difficulties in Prader-Willi syndrome via effects on the serotonin system. First, serotonin neurons produce the Necdin protein.
Second, laboratory mice that lack the gene for Necdin have abnormally shaped serotonin neurons. Third, these mice show breathing difficulties like those of individuals with Prader-Willi syndrome. But while this implies a connection between serotonin, Necdin and breathing difficulties, it falls short of establishing a causal link.
Matarazzo et al. now reveal an increase in the quantity and activity of a protein called the serotonin transporter in mutant mice that lacked the gene for Necdin compared to normal mice. Serotonin transporter proteins mop up the serotonin that neurons release when they signal to one another. Neurons in the mutant mice take up more serotonin than their counterparts in normal mice; this means they have less serotonin available for signaling. This may make it harder for the mutant mice to regulate their breathing.
Drugs called selective serotonin-reuptake inhibitors (or SSRIs for short) can block the serotonin transporter. These drugs, which include Fluoxetine (also called Prozac), are antidepressants. Matarazzo et al. show that SSRIs temporarily restore normal breathing in young mice that lack the gene for Necdin. However, these drugs have harmful long-term effects on breathing in non-mutant mice. Further studies should test whether short-term use of SSRIs could offer immediate relief for breathing difficulties in infants and children with Prader-Willi syndrome.
have also previously found that 5-HT fibers contained 'swollen 5-HT varicosities' in the Ndn tm1-Mus -KO model, and that Necdin is expressed in virtually all 5-HT neurons (Zanella et al., 2008).
These findings suggest a potential role for abnormalities in 5-HT metabolism and release as a potential mediator of respiratory dysfunction in the Necdin-KO model of PWS, but fall short of proving causality. Here, we demonstrate a causal link between the perturbed development of the 5-HT system in Ndn tm1-Mus -KO mice (referred to hereafter as Ndn-KO) and their observed respiratory phenotype (central apnea and hypercapnia). Our data implicate increased activity of serotonin transporter (SERT) as a key mediator of central apnea in this model, and that its inhibition restores normal breathing in Ndn-KO mice.

Results and discussion
Lack of Necdin affects the development and function of 5-HT neurons Pet-EYFP mice expressing YFP under Pet1-promoter control, an early marker of developing 5-HT neurons (Hawthorne et al., 2010), were used to show that Necdin is expressed from E10.5 in early post-mitotic 5-HT precursors and later on in all 5-HT neurons until adulthood ( Figure 1A, Figure 1figure supplement 1A-I).
We then assessed whether Necdin deficiency could induce alterations of 5-HT neuronal development. In wild-type mice rostral hindbrain 5-HT neurons project to the mesencephalon at E12.5, and we observed a decrease in those ascending 5-HT projections in Ndn-KO embryos (Figure 1-figure supplement 1J), confirming previous work (Pagliardini et al., 2008). At E16.5, when the 5-HT raphe nuclei reach their mature configuration, we observed misplaced 5-HT neurons in Ndn-KO embryos ( Figure 1B), with~30% reduction in the total number of 5-HT neurons in the B1-B2 caudal raphe nuclei at birth ( Figure 1C).
Our observations suggested a defect in 5-HT neuronal migration; which was tested using the Pet-EYFP model. In E10.5 WT embryos, Pet-EYFP neurons displayed typical bipolar morphology with oval-shaped somata aligned with two primitive processes attached to the ventricular and pial surfaces, required for somal translocation and involved in migration processes (Hawthorne et al., 2010) ( Figure 1D). In contrast cells were not correctly aligned and process orientation was significantly disturbed in Pet-EYFP/Ndn-KO embryos ( Figure 1D-E). Cell migration was also defective in organotypic slice cultures prepared from E12.5 embryos. Two-photon time-lapse imaging indicated that migratory behavior, based on somal translocation, was altered in Ndn-KO mice ( Figure 1F-H, Figure 1-video 1 and 2) with tracked cells exhibiting increased tortuosity ( Figure 1G) and decreased velocity ( Figure 1H) of their growth trajectories. Interestingly, a comparable migration defect has been described in primary cultures of Ndn tm1-Stw -KO cortical neurons (Bush and Wevrick, 2010) Here we revealed an alteration of cell migration of 5-HT precursors leading to misplaced 5-HT raphe nuclei in Ndn-KO mice.
The acquisition of specific firing properties is considered a critical marker of 5-HT neuronal and circuit maturation (Rood et al., 2014). Using visually guided patch-clamp recordings on brain slices (P15), we demonstrated a significant increase of spontaneous firing in Pet-EYFP/Ndn-KO cells ( Figure 1I-K) suggesting a decreased availability of extracellular 5-HT (Maejima et al., 2013). Overall, our results show that Necdin is responsible for the normal migration of 5-HT precursor neurons during development and exerts effects on their electrophysiological properties in post-natal life.

Lack of Necdin increases the expression and activity of serotonin transporter
We hypothesised that reduced availability of extracellular 5HT could have contributed to the excessive electrophysiological activity we observed in Pet-EYFP neurons in Ndn-KO animals and examined potential mechanisms through which extracellular 5-HT could be reduced. We compared the distributions of 5-HT-immunoreactive enLarged Punctiform Axonal stainings (5-HT LPAs, previously named 'swollen large varicosities' [Pagliardini et al., 2005;Zanella et al., 2008]) in Ndn-KO and WT mice. In all regions analyzed we found significantly more 5-HT LPAs in Ndn-KO mice (Figure 2A-B). These 5-HT LPAs could result from (1) an increase of 5-HT synthesis and/or (2) a decrease in 5-HT degradation and/or (3) an increase of 5-HT reuptake. HPLC analyses showed a similar level of L-Trp and 5-HT in Ndn-KO compared with WT mice, but a significant increase of 5HIAA product in Noticeably, transcript levels of Tryptophan hydroxylase 2, the enzyme that converts L-Trp to 5-HT, were similar in Ndn-KO and WT mice ( Figure 2-figure supplement 1E). These results suggest that the increase in 5-HT LPAs found in Ndn-KO brainstems probably result from an accumulation of intracellular 5-HT due to an increased 5-HT reuptake, since there is no increase of 5-HT synthesis but, on the contrary, an increase of 5-HT degradation. We hypothesised that overexpression of serotonin transporter (SERT) represents a plausible mechanism through which 5-HT could be accumulated in Ndn-KO mice, based on the observation that inactivation of Maged1, another member of the Mage gene family, leads to overexpression of SERT (encoded by the Slc6a4 gene) (Mouri et al., 2012). Indeed, we observed a 3.2 fold increase in SERT protein expression in the brainstems of Ndn-KO compared to WT pups ( Figure  Subsequently, in 5-HT neurons of raphe primary cultures, we assessed SERT activity by live single cell uptake assay, using ASP+ (4 (4-(dimethylamino)styryl)-N-methylpyridinium), a fluorescent substrate of SERT (Lau et al., 2015;Oz et al., 2010). Changes in the kinetics and saturation of ASP+ uptake were measured after 8 days in vitro culture in 5-HT neurons from neonatal (P0) WT, Ndn-KO, and Slc6a4-KO mice ( Figure 2E-H, Figure 2-figure supplement 2A-B). As expected, cultures accumulated ASP+ over time in all conditions tested. However, kinetics experiments show that ASP+ accumulation was significantly faster (greater mean velocity v) in Ndn-KO compared to WT raphe neurons ( Figure 2E). Saturation experiments using increasing concentrations of ASP+ confirmed that ASP+ uptake is a saturable process ( Figure 2F) and showed a Vmax ( Figure 2G) and KM ( Figure 2H) significantly higher in Ndn-KO than in WT or Slc6a4-KO neurons. ASP+ uptake was~2 fold increased in Ndn-KO while it was null in Slc6a4-KO cell cultures. We conclude that there is an increase of ASP+ uptake in Ndn-KO neurons, specifically dependent on SERT activity, suggesting a mechanism for 5-HT LPAs accumulation in vivo. To determine whether in vivo deletion of Slc6a4 could suppress the 5-HT LPAs in Ndn-KO, we compared the number of 5-HT LPAs in Ndn-KO, Slc6a4-KO and Ndn/Slc6a4-double KO (Ndn/ Slc6a4-DKO) neonates in various brain structures. The number of 5-HT LPAs was similar in brains of Ndn/Slc6a4-DKO and WT mice ( Figure 2A-B), indicating that the absence of Ndn is functionally compensated for by the lack of Slc6a4.  11.8 ± 0.8, p=0.003; Raphe B3: WT: 2.6 ± 0.8; Ndn-KO: 9.3 ± 1.6, p=0.01; Raphe B7: WT: 6.5 ± 1.2; Ndn-KO: 15.5 ± 2.6, p=0.07; Raphe B9: WT: 5.1 ± 2.72; Ndn-KO: 14.6 ± 2.9, p=0.0001; Cortex: WT: 5.9 ± 2.0; Ndn-KO: 13.4 ± 0.8, p=0.01; Hippocampus: WT: 4.5 ± 1.2; Ndn-KO: 11.8 ± 2.5, p=0.01. Together, our data show that increased SERT expression in Ndn-KO mice underlies an increase of 5-HT reuptake, which accumulates in 5-HT LPAs. In the absence of any increase in 5-HT synthesis (and in fact increased 5-HT degradation), this sequence of events could be sufficient to cause a physiologically relevant decrease extracellular 5-HT.
Genetic ablation or pharmacological inhibition of SERT uptake restores normal breathing in Ndn-KO mice As exogenous 5-HT application stabilized respiratory rhythm of Ndn-KO mice in vitro, (Zanella et al., 2008), we hypothesized that SERT dysregulation observed in Ndn-KO mice might underlie their respiratory phenotype. To further investigate this causal link, we compared breathing parameters in WT, Ndn-KO, Ndn/Slc6a4-DKO and in Ndn-KO pups treated with Fluoxetine, a selective 5-HT reuptake inhibitor (SSRI) used clinically to increase extracellular 5-HT ( Figure 3A-B). First, we confirmed that respiratory deficits, quantified as the percentage of mice exhibiting apnea ( Figure 3C), the number of apneas per hour ( Figure 3D), or the accumulated apnea duration ( Figure 3E), were significantly increased in Ndn-KO compared to WT mice. These deficits were suppressed by reducing SERT function either by constitutive genetic inactivation (Ndn/Slc6a4-DKO pups) or by 10 days of Fluoxetine treatment (P5-P15; 10 mg/kg/day) in Ndn-KO pups ( Figure 3C-E). Other basic respiratory parameters (minute ventilation, frequency of breathing, tidal volume) were unchanged between all genotypes ( Figure 3F-H). Therefore, our results show that increasing extracellular 5-HT is sufficient to suppress apneas in juvenile Ndn-KO mice.
Early life Fluoxetine-treatment has deleterious long-term respiratory consequences in WT mice Although Fluoxetine had beneficial but transient effects on apnea incidence in Ndn-KO mice, we observed deleterious and long-lasting effects on respiratory function in WT controls. Early life Fluoxetine-treatment induced a significant increase in the number of apneic mice, the frequency of apneas, and the cumulative distribution of apneas at all timepoints measured (0, 15 and 45 DAT, affected resting PBf of WT en bloc preparations but instead abolished the normal increases in PBf responses to acidosis. Thus, we confirm that Fluoxetine treatment abolishes the capacity of WT mice to respond to acidosis , and we propose a role for 5-HT1A-R activity in this response. We show here, for the first time, adverse effects of Fluoxetine on breathing outcomes.

Conclusion
Previously, a pleiotropic function of Necdin has been reported in different neuronal populations and at different developmental stages. Concerning the 5-HT system, an expression of Necdin was observed in virtually all 5-HT neurons (Zanella et al., 2008) and an alteration of the 5-HT system in  (Lee et al., 2005;Pagliardini et al., 2005) and 5-HT fibers containing swollen 5-HT 'varicosities' (Pagliardini et al., 2005;Zanella et al., 2008). Furthermore, an alteration of 5-HT metabolism (Zanella et al., 2008) was observed in mutant neonates suggesting that it might alter 5-HT modulation of the Respiratory Rhythm Generator. Finally, an in vitro exogenous application of 5-HT on brainstem-spinal cord preparations of Ndn mutant mice alleviates the incidence of apneas (Pagliardini et al., 2005;Zanella et al., 2008). Despite those observations, the pathological mechanism responsible for the serotonopathy in Ndn-KO mice and the causal link between this serotonopathy and the breathing alterations were not investigated. Here, we aimed to answer those questions.
Noticeably, all previous studies have been performed on heterozygous Ndn-deficient mice, with a deletion of the Ndn paternal allele only (Ndn+m/-p), the maternal allele being normally silent. However, we have shown that, due to a faint and variable expression of the Ndn maternal allele (+m), Ndn+m/-p mice present a variability in the severity of respiratory phenotype compared with the Ndn-/-mice (here named Ndn-KO) (Rieusset et al., 2013). For instance, reduction of 5-HT neurons was not previously found significant in the Ndn + m/-p mice (Zanella et al., 2008) but has been found significantly reduced in the Ndn-/-mice. In order to avoid such variability and to get consistent results, we chose here to study Ndn-/-mice. air or (C) at 5th min upon hypercapnic respiratory challenge. (D) Respiratory frequency (Rf) in WT and Ndn-KO pups when subjected to hypercapnic stress: WT Air: 91 ± 8; WT hypercapnia: 163 ± 16; n = 8, p=0.004; Ndn-KO Air: 95 ± 12; Ndn-KO hypercapnia: 123 ± 11; n = 8, p=0.31, N.S. p-values determined by two-way-ANOVA test followed by Bonferroni post-hoc test. Bar graphs represent mean ±SEM; **p<0.01; N.S.: non-significant. (E) Tidal Volume (VT) (ml.g À1 ) in WT neonates: WT Air: 9 ± 0.7; WT hypercapnia: 12 ± 1; n = 8, p=0.12 N.S.; in Ndn-KO: Ndn-KO Air: 7.5 ± 0.9; Ndn-KO hypercapnia: 10.9 ± 1.1; n = 8, p=0.055. p-values determined by two-way ANOVA test followed by Bonferroni post-hoc test. (F) Minute Ventilation (VE) (ml.min À1 .g À1 ) in WT neonates: WT Air: 0.8 ± 0.1; WT hypercapnia: 2.1 ± 0.1; n = 8, p=0.01; in Ndn-KO: Ndn-KO Air: 0.7 ± 0.1; Ndn-KO hypercapnia: 1.3 ± 0.2; n = 8, p=0.3 N.S. p-values determined by two-way ANOVA test followed by Bonferroni post-hoc test. Bar graphs represent mean ±SEM. Here, we have shown that Necdin plays a pleiotropic role in the development of 5-HT neuronal precursors that guides the development of central serotonergic circuits and the physiological activity of mature 5-HT neurons. Our results suggest that Necdin controls the level of SERT expression in 5-HT neurons and that lack of Necdin increases the quantity and activity of SERT leading to an increased reuptake and intra-cellular accumulation of 5-HT, as visualized by 5-HT LPAs, leading to a reduction in available extracellular 5-HT. Importantly, in vivo inhibition of SERT activity, genetically or pharmacologically (Fluoxetine treatment), is sufficient to prevent the formation of those 5-HT LPAs and suppresses the apnea observed in Ndn-KO mice. We also demonstrate, using an ex vivo approach, that the altered chemosensitivity to CO 2 /acidosis is caused by a central 5-HT deficit and is rescued by Fluoxetine-treatment. We conclude that an increase of 5-HT reuptake is the main cause of breathing deficits (central apnea and hypercapnia response) in Ndn-KO mice.
Unexpectedly, we reveal an adverse and long-term effect of early life administration of Fluoxetine on the breathing (apneas, chemosensitivity to CO 2 /acidosis) of healthy mice. Previous adverse effects have been observed on anxiety and depression (Glover and Clinton, 2016;Millard et al., 2017) after an early postnatal administration of Fluoxetine but the respiratory deficits are reported here for the first time and should be further investigated in another study.
Respiratory failure in patients with PWS constitute a challenging issue since it is the most common cause of death for 73% of infants and 49% of children, (Butler et al., 2017). Death is often linked to respiratory infection or respiratory disorder and may be sudden, with some reported cases of sudden death occurring at night (Gillett and Perez, 2016). In PWS patients, any environmental acute respiratory challenge caused by, for instance, a respiratory tract infection, high altitude or intense physical activity further exacerbates their inherent disability (blunted response to hypoxima/hypercapnia) to adapt an respiratory response. Until now, the underlying pathology for respiratory failure remained elusive and did not appear to be impacted by recent advancements in treatment modalities (Butler et al., 2017). Although oxygen treatment is efficient in preventing the hypoxemia induced by central apneas (Urquhart et al., 2013), such treatment is physically constraining. Within the context of PWS, the current study points towards a critical link between Necdin, serotonopathy, and chemosensing, a function in which brainstem serotonergic circuits play a critical role. Since our study shows that Fluoxetine can suppress apnea and restore chemosensitivity, we propose that Fluoxetine might be an appropriate 'acute' treatment that could be considered for Prader-Willi infants/ children when they present the first signs of any breathing difficulties.

Animals
Mice were handled and cared for in accordance with the Guide for the Care and Use of Laboratory Animals (N. R.C., 1996)  3). In order to avoid a variability in our results due to a stochastic and faint expression of the maternal allele (Rieusset et al., 2013), we worked with the Ndn tm1-Mus strain and decided to study Ndn-/mice (named here Ndn-KO), instead of Ndn+m/-p mice as it has been done previously.
Fluoxetine was obtained from Sigma (Saint-Quentin Fallavier, France) for cell culture and en bloc medullary experiments and from Mylan pharma for in vivo experiments.
Brainstem structures were sampled by selecting the raphe obscurus area and counting was performed on three sagittal sections/animal of 100 mm which represent the entire PET1-YFP positive cell population of the raphe obscurus (ROb/B2) and pallidus (RPa/B1), both nuclei being difficult to separate. For each section, a Z-stack composed of 10 confocal images (8 mm focal spacing) was acquired. For quantification, stereological method has been applied on each Z-stack image using the eCELLence software developed by Glance Vision Technologies (Italy). The total cell number/per animal was obtained by summing the sub-total of cells counted for the 3 Z-stacks.
Images of 5-HT LPAs were acquired using a confocal microscope (Olympus). Between 4 and 8 fibers/brain region for each animal (3WT and 3 KO) were analyzed for the presence of 5-HT LPAs (>1.8 mm 2 ) on 100 mm long fiber. The size of 5-HT LPAs was quantified using Image J. 5-HT LPA diameter has been defined ad arbitrium as the size of the largest 5-HT punctiform labelling found in the WT fibers.

Organotypic slice cultures and time lapse experiments
Slice cultures from E11.5 embryonic mouse brainstems were prepared from Pet-EYFP and Ndn KO/ Pet-EYFP mice. Thick coronal sections (250 mm) brainstem were cut using a tissue chopper and cultured in Neurobasal medium (Thermofisher) containing 2% B27 (Thermofisher), 4% horse serum, 10 mg/ml insulin, 200 mM HEPES, 1% Antibiotic Antimycotic (Thermofisher). For time lapse experiments, the dishes were mounted in a CO 2 incubation chamber (5% CO 2 at 37˚C) fitted onto an inverted confocal microscope (LSM510, Zeiss). Acquisitions of the region containing raphe Pet-EYFP +neurons were performed every 10 min for up to 15 hr. Cell coordinates, velocity, and tortuosity (total length of the track/direct distance from the first to the last point) were calculated using MtrackJ plugin of Image J.

In vitro recordings from en bloc brainstem-spinal cord preparations
As previously reported (Berner et al., 2012), the medulla and cervical cord of P0-P1 neonatal mice were dissected, placed in a 2 ml in vitro recording chamber, bubbled with carbogen, maintained at 27˚C and superfused (3.5-4.5 ml per min) with aCSFcomposed with (mM): 129.0 NaCl, 3.35 KCl, 21 NaHCO 3 , 1.26 CaCl 2 , 1.15 MgCl 2 , 0.58 NaH 2 PO 4 , and 30.0 D-glucose ('Normal aCSF': pH 7.4) or using the same components except with 10 mM NaHCO 3 ('Acidified aCSF': pH 7.1). Inspiratory discharges of respiratory motoneurons were monitored by extracellular recording with glass suction electrodes applied to the proximal cut end of C4 and C3 spinal nerves roots. Axoscope software and Digidata 1320A interface (Axon Instruments, Foster, CA, USA) were used to collect electrophysiological data. Offline analysis was performed with Spike 2 (Cambridge Electronic Design, UK) and Origin 6.0 (Microcal Software, Northampton, MA, USA) software for PC. Burst frequency was analyzed and calculated as the number of C4 bursts per minute. The values of inspiratory burst frequency were calculated as the mean of the last 3 min of any condition: ACSF (7.4) and ACSF (7.1). Standardized experiments in WT and Ndn-KO preparations were repeated on different preparations from different litters. For a given preparation, only one drug was applied and only one trial was performed.

RT-qPCR
For RT-qPCR, mice were sacrificed at P1, the brainstem dissected, and tissues were rapidly collected and frozen in liquid nitrogen prior to RNA isolation using standard conditions. RNA, reverse transcription and real time PCR were conducted as previously described (Rieusset et al., 2013). Sequences of the various primer pairs used for qPCR, as well as the slope of the calibration curve established from 10 to 1 Â 10 9 copies and qPCR efficiency E, were as follow: Tph2: F: 5'-GAGC TTGATGCCGACCAT-3'; R: 5'-TGGCCACATCCACAAAATAC-3'; Slc6a4: F:5'-CATATGCTACCAGAA TGGTGG-3'; R:5'-AAGATGGCCATGATGGTGTAA-3'. For each sample, the number of cDNA copies was normalized according to relative efficiency of RT determined by the standard cDNA quantification. Finally, gene expression was expressed as the cDNA copy number quantified in 5 mL aliquots of RT product.

Western blot
Newborn mice were sacrificed and brainstems were immediately dissected and snap-frozen in liquid nitrogen and stored at À80˚C until protein extraction. Protein extraction was conducted as previously described (Felix et al., 2012). Membranes were blocked with PBS containing 5% BSA for 1 hr, followed by an overnight incubation at 4˚C with the following primary antibodies: guinea pig anti-SERT (1/2000, Frontier Institute), mouse anti-B3 tubulin (1/2000, ThermoFisher Scientific). Membranes were then washed and incubated 2 hr with either anti-guinea pig (1/1000, ThermoFisher Scientific), or anti-mouse (1/2000; DAKO) horseradish peroxidase-conjugated secondary antibodies. Visualisation was performed using the Super signal West-pico chemolumniscent substrate (Pierce, Thermo Scientific, France). Quantification was performed using ImageJ.
Raphe primary neuronal culture and live cell uptake assay Raphe primary cell culture Newborn mice (n = 6 per culture) were decapitated, brainstems extracted, the meninges removed and the medial part of the brainstem dissected. Tissues were enzymatically digested at 37˚C for 30 min with HBSS containing 2 mg/mL of filter-sterilized papain. Cells were resuspended in Neurobasal medium (Thermofisher) containing 2% B27 (Thermofisher), 0.5 mM L-glutamine, glucose (50 mM), 50 ng/ml NGF, 10 ng/ml bFGF, 10 mg/ml insulin. 2 Â 10 5 cells were plated on round 14 mm glass coverslip pre-coated with Polyethyleneimine (20 mg/ml). Cells were cultured during 8 days in presence of 5% of NU serum (Becton Dickinson) during the first 2 days. Immunocytochemistry was performed to verify presence of 5-HT + neurons in the culture.
Live cell imaging of (4-(4-(dimethylamino)styryl)-N-methylpyridinium (ASP+) uptake Cells were placed in a bath chamber on the stage of an inverted microscope (Nikon eclipse TE300) and perfused (2 ml/min) with Krebs medium (mM): 150 NaCl; 2.5 KCl; 2 CaCl 2 ; 2 MgCl 2 ; 2.5 Hepes acide; 2.5 Hepes-Na; pH 7,4. Time-lapse cell acquisition was started when ASP+ (1, 2, 5, 10, 15 or 20 mM) was added to the perfusion. ASP+ was excited at 488 nm and fluorescence was captured at 607 nm every 10 s for 5 min using Metamorph software (MolecularDevices). Each ASP+ concentration was tested on three different cultures for WT and Ndn-KO and one for Ndn/Slc6a4-DKO. Cells placed on the coverslip were replaced for each concentration tested. For each ASP+ cells, an ROI of the same surface was delineated on the soma in order to measure pixel intensity in arbitrary fluorescence units. 6 ROI were determined at each measurement. Data were background subtracted and ASP+ fluorescence intensity was expressed as a function of initial fluorescence intensity.

In vivo recordings of breathing parameters by plethysmography
Breathing of unrestrained, non-anesthetized mice was recorded using constant air flow whole-body plethysmography filled with air or 4% CO 2 in air (EMKA Technologies, Paris, France). Neonatal mice (P0-P1) were recorded in 25 ml chambers (calibrated by injecting 50 ml of air) maintained at neonatal thermoneutral ambient temperature (32 ± 0.5˚C). For adolescent and adult mice (P15-P30-P60), four plethysmography 200 ml chambers containing air or (calibrated by injecting 1 ml of air) maintained at 25 ± 0.5˚C were used to allow simultaneous measurements. Analog signals were obtained using an usbAMP device equipped with four inputs and processed using EMKA technologies IOX software (EMKA Technologies, Paris, France). For neonatal mice, we measured mean respiratory frequency (Rf, expressed in cycles per minute) during quiet periods when mice breathed air or 5 min after breathing hypercapnic air. For adolescent and adult mice respiratory parameters (frequency, tidal volume, minute ventilation) were recorded over 30 min after an initial 30 min period of stabilization in the apparatus.,Apnea was defined as a prolonged expiratory time (four times eupneic expiratory time), which corresponds to a threshold of 1 s.

Statistical analysis
Analyses were performed using two-tailed non-parametric statistical tools due to the size of the samples (GraphPad, Prism software). Values are indicated as following: (Q2 (Q1, Q3), n; statistical test, p-value) where Q2 is the median, Q1 is the first quartile and Q3 is the third quartile and scatter dot plots report Q2 (Q1, Q3). Histograms report the mean ±SEM. The level of significance was set at a p-value less than 0.05. Appropriate tests were conducted depending on the experiment and are indicated in the figure legends. Mann-Whitney (MW) test was performed to compare two unmatched groups: differences between WT and Ndn-KO (Figure 1 and Figure 2-figure supplement 1). Kolmogorov-Smirnov test was performed to compare the cumulative distribution of two unmatched groups: differences between WT and Ndn-KO in apnea accumulation over time ( Figure 3E Figure 2E). *p<0.05; **p<0.01; ***p<0.001; ****p<0.0001.
of September 22th 2010 (2010/63/EU, 74). Experimental protocols were approved by the Institutional Ethical Committee guidelines for animal research with the accreditation no. B13-055-19 from the French Ministry of Agriculture.