EphrinB1 modulates glutamatergic inputs into POMC neurons and controls glucose homeostasis

Proopiomelanocortin (POMC) neurons are major regulators of energy balance and glucose homeostasis. In addition to being regulated by hormones and nutrients, POMC neurons are controlled by glutamatergic input originating from multiple brain regions. However, the factors involved in the formation of glutamatergic inputs and how they contribute to bodily functions remain largely unknown. Here, we show that during the development of glutamatergic inputs, POMC neurons exhibit enriched expression of the Efnb1 (EphrinB1) and Efnb2 (EphrinB2) genes, which are known to control excitatory synapse formation. In vitro silencing and in vivo loss of Efnb1 or Efnb2 in POMC neurons decreases the amount of glutamatergic inputs into these neurons. We found that mice lacking Efnb1 in POMC neurons display impaired glucose tolerance due to blunted vagus nerve activity and decreased insulin secretion. However, mice lacking Efnb2 in POMC neurons showed no deregulation of insulin secretion and only mild alterations in feeding behavior and gluconeogenesis. Collectively, our data demonstrate the role of ephrins in controlling excitatory input amount into POMC neurons and show an isotype-specific role of ephrins on the regulation of glucose homeostasis and feeding.


Proopiomelanocortin (POMC) neurons are major regulators of energy balance and 26
glucose homeostasis. In addition to being regulated by hormones and nutrients, 27 POMC neurons are controlled by glutamatergic input originating from multiple brain 28 regions. However, the factors involved in the formation of glutamatergic inputs and 29 how they contribute to bodily functions remain largely unknown. Here, we show that 30 during the development of glutamatergic inputs, POMC neurons exhibit enriched 31 expression of the Efnb1 (EphrinB1) and Efnb2 (EphrinB2) genes, which are known to 32 control excitatory synapse formation. In vitro silencing and in vivo loss of Efnb1 or 33 Efnb2 in POMC neurons decreases the amount of glutamatergic inputs into these 34 neurons. We found that mice lacking Efnb1 in POMC neurons display impaired 35 glucose tolerance due to blunted vagus nerve activity and decreased insulin 36 secretion. However, mice lacking Efnb2 in POMC neurons showed no deregulation of 37 insulin secretion and only mild alterations in feeding behavior and gluconeogenesis. 38 Collectively, our data demonstrate the role of ephrins in controlling excitatory input 39 amount into POMC neurons and show an isotype-specific role of ephrins on the 40 regulation of glucose homeostasis and feeding. 41

Introduction
Results 87

Onset of glutamatergic inputs into POMC neurons 88
To visualize the development of excitatory presynaptic terminals in POMC 89 neurons, we performed immunohistochemical labeling of presynaptic glutamatergic 90 vesicular transporter (vGLUT2) in POMC neurons expressing a tdTomato reporter 91 (Pomc-Cre;tdTomato) in postnatal days P6, P14 and P22 male mice ( Figure 1A). At 92 P6, glutamatergic inputs were already observed to be in contact with POMC neurons,93 and the amount of inputs increased gradually until P22 ( Figures 1A-C). 94 Based on this developmental observation, we next performed transcript 95 profiling of POMC-expressing (POMC + ) and NPY + cells at P14 to identify putative 96 genes involved in glutamatergic synapse formation (Figures 1D and 1E). POMC 97 progenitors can give rise to NPY neurons or POMC neurons. We thus performed  Next, we examined whether the lack of Efnb1 in POMC neurons caused 177 disturbances in body weight and food intake. The postnatal growth curves (body 178 weights) and body composition of Pomc-Cre;Efnb1 loxP/0 mice were undistinguishable 179 from Efnb1 loxP/0 control male mice (Figures 4D and 4E). Consistent with these data, 180 daily food intake was similar between the groups ( Figure 4F). In addition to their 181 fundamental role in energy balance, POMC neurons have been shown to be involved 182 in the regulation of peripheral glucose homeostasis (5,6). Accordingly, we also 183 investigated the effect of genetically deleting Efnb1 in POMC neurons on peripheral 184 glucose homeostasis and insulin sensitivity. The basal glycemia and insulinemia 185 were unchanged in Pomc-Cre;Efnb1 loxP/0 mice and Efnb1 loxP/0 mice ( Figures 5A and  186 5B). Pomc-Cre;Efnb1 loxP/0 mice displayed significantly elevated glycemia 15 to 45 187 minutes after a glucose challenge ( Figures 5C and 5D). After 15 but not 30 minutes, 188 glucose-stimulated insulin secretion was impaired in mutant mice, suggesting that 189 only the cephalic phase of insulin secretion (first phase) was impacted (Figures 5E 190 and 5F). Activation of the cholinergic parasympathetic innervation of the pancreatic 191 islets and inhibition of the sympathetic nervous system control insulin secretion in 192 response to hyperglycemia (24). We thus assessed cholinergic (parasympathetic) 193 innervation of pancreatic islets in Pomc-Cre;Efnb1 loxP/0 and Efnb1 loxP/0 male mice 194 (Figures 5G and 5H). No difference in the density of cholinergic fibers was found in 195 islets of Pomc-Cre;Efnb1 loxP/0 mice compared to those of Efnb1 loxP/0 male mice ( Figure  196 5H). We then measured parasympathetic nerve (vagus) activity upon glucose 197 challenge. We observed no difference in basal firing activity between  Cre;Efnb1 loxP/0 male mice and Efnb1 loxP/0 control littermates (Figures 5I and 5J). 199 However, whereas a glucose challenge increased firing activity by 2.5-fold in 200 Efnb1 loxP/0 control mice, no response was detected in the vagus nerve of Pomc-201 Cre;Efnb1 loxP/0 mice ( Figure 5J). We also performed pyruvate and insulin tolerance 202 tests, and the results were identical in control and mutant mice (Figures S4D and 203 S4E). Notably, similar metabolic disturbances were found in Pomc-Cre;Efnb1 loxP/loxP 204 female mice ( Figure S4). Together, these data show that mice lacking Efnb1 in 205 POMC neurons develop glucose intolerance that is associated with impaired insulin 206 secretion and impaired parasympathetic nerve activity. 207 208 Lack of Efnb2 in POMC neurons impairs feeding and gluconeogenesis in a sex-209 specific manner 210 To study the role of Efnb2 in the development of glutamatergic synapses in 211 POMC neurons, we crossed Efnb2 loxP mice (25) with Pomc-Cre mice. As expected, 212 the level of Efnb2 mRNA was significantly reduced in the ARH of Pomc-213 Cre;Efnb2 loxP/loxP mice, whereas the level of Efnb1 mRNA was unchanged between 214 mutant and control mice ( Figures 6A and 6B). Efnb2 mRNA was also detected in 215 adeno-pituitary during late fetal and adult life ( Figure S4A and S5A). However, no 216 change was observed in Efnb2 mRNA expression in the pituitaries of mice that lack 217 Efnb2 in their POMC cells when compared to control mice ( Figure S5A).   Together, these results suggest that the lack of Efnb2 in POMC neurons 232 impairs gluconeogenesis in males and it impairs food intake in a refeeding paradigm 233 in females. 234

Discussion 235
Energy and glucose homeostasis are tightly controlled by the brain. POMC 236 and AgRP/NPY neurons in the ARH are key regulators of these functions and 237 respond to peripheral signals through projections to second-order neurons controlling 238 endocrine and autonomic nervous systems. However, POMC and AgRP/NPY 239 neurons also receive extensive inputs from a plethora of areas of the brain (18) and 240 are thus integrated in a complex neuronal network. Although our understanding of 241 the control of feeding behavior and glucose homeostasis has improved over the last 242 few decades, it is still largely unknown how central circuits that regulate POMC 243 activity and their associated functions are being assembled. 244 Here, we showed that there is an enrichment of EphrinB members in POMC 245 neurons when glutamatergic inputs develop, and we described the role of ephrin 246 signaling in the control of the amount of excitatory inputs. EphrinB1 and EphrinB2 247 appear to both control the number of glutamatergic terminals on POMC neurons; 248 however, they play a distinct role in controlling energy and glucose homeostasis. 249 These are interesting findings given that glutamatergic input pattern is impaired both 250 in mice lacking EphrinB1 and those lacking EphrinB2 in POMC neurons, suggesting 251 functional heterogeneity in POMC neuronal circuits. 252 The present study is in agreement with previous work performed in rats (20) 253 and it shows that in mice, the amount of glutamatergic inputs into arcuate POMC 254 neurons increases gradually after birth until weaning. During this important period of 255 neuronal connectivity formation, we showed that POMC neurons were enriched with 256 EphrinB1 and EphrinB2, two members of the ephrin family. These proteins enable 257 cell to cell contacts through interaction with EphA and EphB receptors to control axon 258 growth, synaptogenesis or synaptic plasticity. We focused our study on Ephb1 and 259 Ephb2 receptors because their interactions with EphrinB1 and EphrinB2 are well 260 described (21) and EphrinB members are known to play a key role in AMPA and 261 NMDA glutamate receptor recruitment, stabilization of glutamatergic synapses, and 262 control of the number of excitatory synapses (26,27). 263 Here, we used a developmental approach that interfered with excitatory 264 synaptic input formation to specifically assess the role of glutamatergic inputs in 265 POMC neurons in the control of energy and glucose homeostasis. We showed that 266 lacking EphrinB1 or EphrinB2 in POMC neurons reduces the amount of 267 glutamatergic input into these neurons. Interestingly, these two mouse models do not 268 have similar physiological outcomes, suggesting specificity in the establishment of 269 glutamatergic input patterns. We first hypothesized that these differences arose from 270 POMC heterogeneity, as distinct subsets of leptin receptor-, insulin receptor-and 271 serotonin receptor-expressing POMC neurons are linked to functional differences; 272 (28,29) however, our data showed that every detected POMC neuron expressed both 273 Efnb1 and Efnb2. Thus, the differential effects of Efnb1 and Efnb2 inactivation on 274 energy and glucose metabolism could stem from EphrinB favoring the formation of 275 presynaptic inputs arising from distinct areas. Indeed, several Eph receptors can 276 interact with EphrinB1 and EphrinB2 (21) with different affinities (30) and can also be 277 differentially expressed in areas known to project to POMC neurons. These aspects 278 of the study require further analysis. 279 The loss of Efnb1 in POMC neurons is associated with alterations in glucose 280 tolerance and losses of parasympathetic nerve activity and insulin secretion. These another NMDA subunit and through AMPA receptors (36);, both cases are in 295 agreement with our data, in that they cannot distinguish which glutamatergic 296 receptors are predominantly involved. In some cases, POMC functions require long 297 or chronic chemogenetic activation (9,12), which could reflect the recruitment of 298 NMDA receptors alongside AMPA receptors, since Ca 2+ entry through AMPA 299 receptors precedes full activation of NMDA receptors (37). 300 POMC neurons are well known glucose-excited neurons that drive the 301 response in maintaining normal glycemia (16,38). Our findings may also suggest that 302 glutamatergic innervation of POMC neurons is required for their ability to sense 303 extracellular levels of glucose. This has been shown for NTS catecholamine neurons, 304 which sense glucose through a presynaptic mechanism that is dependent upon, 305 among other factors, glutamate release (39). Moreover, glucose sensing in POMC 306 neurons requires ATP-sensitive potassium (K ATP ) channels (16) and Kir6.2 KO mice 307 (preventing ATP-mediated closure of KATP channels) display an absence of 308 glutamatergic AMPA receptors (40). Finally, glucose induces glutamatergic synaptic 309 remodeling of POMC neurons, potentially regulating the sensitivity of melanocortin 310 system to hormonal and neuronal signals (41). Collectively these data link glucose 311 sensing and glutamatergic release. 312 The loss of Efnb2 in POMC neurons impaired gluconeogenesis in males and 313 food intake in females in a refeeding paradigm. These findings are surprising, as 314 chemogenetic activation and inhibition of arcuate POMC neurons repress and 315 increase hepatic glucose production, respectively. Interestingly, a lack of ephrin 316 expression has been shown to induce local reorganization of glutamatergic synaptic 317 inputs (42). The lack of EphrinB2 in POMC neurons could therefore affect the 318 number of excitatory terminals connecting to nearby AgRP neurons, which are known 319 to suppress hepatic glucose production through insulin action (7,8) and to promote 320 feeding (43). 321 Our findings also suggest sexual dimorphism in the melanocortin system, as 322 the lack of Efnb2 in POMC neurons does not lead to impaired gluconeogenesis in 323 females, but it does result in changes to refeeding after an overnight fast. The ARH 324 has not been primarily viewed as a dimorphic structure, but recent studies showed 325 differences between males and females in the number of ARH POMC neurons, their 326 firing rate, the development of diet-induced obesity and the activation of STAT3 in 327 POMC neurons (44)(45)(46). 328 The phenotypes we observed in mice lacking Efnb1 or Efnb2 can also be due 329 to synaptic plasticity, as cells with EphrinB have been shown to control this function 330 (27) and are found in the brains of adult animals (Allen Mouse Brain Atlas, 2004;(47). 331 Moreover, hormones (8,17), metabolic status, physical activity (48) and the age (49) 332 can directly modulate the amount of glutamatergic and GABAergic synaptic inputs 333 into POMC neurons. 334 In conclusion, our data show that distinct Ephrin members control the 335 glutamatergic innervation of POMC neurons and specific functions such as glucose 336 homeostasis or feeding. This supports the idea that POMC neuronal network is 337 heterogeneous and that POMC neurons should not be considered as first-order 338 neurons but have to be thought predominantly as integrators of multiple kinds of 339 complex peripheral and central information to control energy and glucose 340 homeostasis. 341

Studies in mice 344
All experimental procedures were approved by the Veterinary Office of Canton de 345 Vaud. Mice were group housed in individual cages and maintained in a temperature-346 controlled room with a 12hr light/dark cycle and provided ad libitum access to water 347 and standard laboratory chow (Kliba Nafag). Mice were single housed only for food 348 intake experiments. 349 All mice used in this study have been previously described: Pomc-Cre (23) Institute) using the same coordinates. Control mice were injected with helper virus or 368 EnvA-SADdG-mcherry alone. One week later, mice were anesthetized and perfused 369 with 4% PFA and frozen for brain cryosectioning.

Images analyses 478
To quantitatively analyze cholinergic (VAChT-positive) fibers in pancreatic cells, 479 between 16 and 27 pancreatic islets per animal were imaged using a Zeiss LSM 710 480 confocal system equipped with a 20X objective. Each image was binarized to isolate 481 labeled fibers from the background and to compensate for differences in 482 fluorescence intensity. The integrated intensity, which reflects the total number of 483 pixels in the binarized image, was then calculated for each islet. This procedure was 484 conducted for each image. Image analysis was performed using Image J analysis 485 software (NIH). 486 To quantitatively analyze glutamatergic innervation of POMC neurons, adjacent 487 image planes were collected in lateral part of the ARH through the z-axis using a 488 Zeiss LSM 710 confocal system at a frequency of 0.25 µm through the entire 489 thickness of the ARH. Three-dimensional reconstructions of the image volumes were 490 then prepared using Imaris 9.3.1 visualization software. The number of glutamatergic 491 inputs into POMC neurons was quantified. Each putative glutamatergic input was 492 defined as a spot, and we quantified the number of glutamatergic spots that 493 contacted Pomc-Cre;tdtomato+. 494

Vagus nerve activity recording 528
The firing rate of the thoracic branch of the vagal nerve along the carotid artery was 529 All values were represented as the mean ± SEM. Numbers for every experiment are 542 found in the relevant part of the STAR Methods. Statistical analyses were conducted 543 using GraphPad Prism (version 7). Statistical significance was determined using 544 unpaired 2-tailed Student's t test, 1-way ANOVA followed by Tukey's post hoc test, 2-545 way ANOVA followed by Sidak's post hoc test when appropriate. P ≤ 0.05 was 546 considered statistically significant. We thank the CIG Genomic Technologies Facility (GTF) for RNA-sequencing 555 experiments and analyses, the EPFL Flow Cytometry Core, the UNIL Flow Cytometry 556 Facilities for cell sorting, and the CIG Animal Facility for their assistance with animal 557 husbandry. We are also grateful to Marc Lanzillo with his assistance with animal 558 genotyping. This work was supported by the Swiss National Science Foundation 559 grant (PZ00P3_167934/1) and Novartis grant (19B145).