Relaxin/insulin-like family peptide receptor 4 (Rxfp4) expressing hypothalamic neurons modulate food intake and preference in mice

Relaxin/insulin-like-family peptide receptor-4 (RXFP4), the cognate receptor for insulin-like peptide 5 (INSL5), has been implicated in feeding behaviour as Rxfp4 knockout mice display shorter meal durations and reduced high fat diet (HFD) intake. Here, we generated transgenic Rxfp4-Cre mice to explore Rxfp4 expression and physiology. Using this model, we identified Rxfp4 expression in the central nervous system, including in the ventromedial hypothalamus (VMH). Intra-VMH infusion of INSL5 increased HFD and highly palatable liquid meal intake (HPM) of ad libitum fed wildtype mice. Single-cell RNA-sequencing of VMH Rxfp4-expressing cells (RXFP4VMH) defined a cluster of Rxfp4-labelled neurons expressing Esr1, Tac1 and Oxtr, alongside known appetite-modulating neuropeptide receptors (Mc4r, Cckar and Nmur2). Viral tracing demonstrated RXFP4VMH neural projections to the bed nucleus of the stria terminalis, paraventricular hypothalamus, paraventricular thalamus and central nucleus of the amygdala. Utilising designer receptors exclusively activated by designer drugs (DREADDs), we found that whole body chemogenetic inhibition (Di) of Rxfp4-expressing cells, mimicking native INSL5-RXFP4 signalling, increased intake of HFD and HPM, whilst activation (Dq), either at whole body level or specifically within the VMH, reduced HFD and HPM intake and altered food preference. Ablating VMH Rxfp4-expressing cells recapitulated the lower HFD intake phenotype of Rxfp4 knockout mice, resulting in reduced body weight. These findings identify a discrete Rxfp4-expressing neuronal population as a key regulator of food intake and preference and reveal hypothalamic RXFP4 signalling as a target for feeding behaviour manipulation.


Introduction 42
Relaxin/insulin-like-family peptide receptor-4 (RXFP4) is the cognate receptor for insulin-like peptide 5 43 (INSL5), a member of the relaxin/insulin-like peptide family (1)(2)(3). We previously reported an orexigenic 44 effect of exogenously applied INSL5 in mice(4). More recently, we observed a transient increase in food 45 intake when Insl5 expressing cells were chemogenetically activated (5). This was only revealed when the 46 dominant anorexigenic action of PYY, co-secreted from the same cell population in the distal large 47 intestine (6), was blocked with a NPY2R antagonist (5). Reports that Insl5 knockout (Insl5 -/-) mice do not 48 display an observable feeding phenotype and that, in some studies, pharmacological administration of 49 INSL5 (both native and PEGylated forms) failed to increase food intake in lean and obese mice (7,8), have 50 shed some doubt on whether INSL5 plays a physiologically relevant role in the control of food intake. 51 However, Rxfp4 knockout (Rxfp4 -/-) mice exhibit shorter meal durations, particularly when fed a high fat 52 diet (HFD), and lack the normal preference for HFD over standard chow diet observed in wildtype mice 53 (4). We therefore consider RXFP4 to be a potential target receptor for the manipulation of feeding 54 behaviour. 55 RXFP4 is a Gαi/o-coupled receptor identified in 2003 through its homology to relaxin/insulin-like peptide 56 receptor 3, RXFP3 (9, 10). Binding of INSL5 activates downstream signalling pathways including 57 phosphorylation ERK1/2, Akt, p38MAPK and S6RP and reduces cytosolic cAMP levels through inhibition of 58 adenylate cyclase (2). In addition to the selective ligand INSL5, Relaxin-3 can also activate RXFP4, 59 stimulating comparable signalling pathways (2, 11). Rxfp4 mRNA expression has been reported in the 60 colon, kidney, heart, liver, testes and ovary of both mice and humans (1,10,11). To identify and 61 manipulate Rxfp4-expressing cells, we developed a new transgenic mouse model in which Cre-62 recombinase expression is driven by the Rxfp4 promoter (Rxfp4-Cre). Utilising this model, we found clear 63 Cre-dependent reporter expression within the CNS, including the ventromedial hypothalamus (VMH). This 64 contrasts with our previous study, in which we failed to detect Rxfp4 expression in the central nervous 65 system by qPCR, and in which intracerebroventricular administration of INSL5 failed to increase food 66 intake significantly (4). Revisiting the potential central role of RXFP4 in food intake regulation, we 67 observed that infusion of INSL5 directly into the VMH induced a significant orexigenic response when mice 68 were offered a HFD or a highly palatable liquid Ensure test meal (HPM), but not when offered a standard 69 chow diet. We therefore used the Rxfp4-Cre mouse model to explore the role of Rxfp4-expressing cells in 70 shaping food intake and preference. 71 an intra-VMH infusion of INSL5 after a 2h fast. During the light phase, INSL5 had no effect on standard 96 (std) chow intake ( Fig.1j) but in mice which were habituated to the appearance of a HFD or HPM for 1 97 hour, treatment with INSL5 significantly increased HFD and HPM intake compared to the vehicle control 98 treatment (Fig.1k,l). 99 As intra-VMH INSL5 significantly modulated feeding behaviour, we further characterised the Rxfp4-100 expressing cells in this region. Initially, we generated a single cell resolution transcriptomic profile of 101 feeding-neuromodulators, like glucagon-like peptide-1 receptor (Glp1r) and cholecystokinin receptor B 116 (Cckbr), were preferentially expressed in cluster 6 ( Fig. 2d). 117 We subsequently aimed to establish the neuronal circuitry around RXFP4 VMH cells. Anterograde 118 projections were mapped by stereotactically injecting Cre-dependent rAAV8-ChR2-mCherry into the VMH 119 of RXFP4 GCaMP3 mice (17) (Fig.3a,b). Axonal transport of the ChR2-mCherry fusion protein revealed 120 RXFP4 VMH projections to multiple regions including the bed nucleus of the stria terminalis (BNST), preoptic 121 area (POA), anteroventral periventricular nucleus (AVPV), arcuate nucleus (ARC), paraventricular 122 hypothalamus (PVH), central nucleus of the amygdala (CeA), periaqueductal grey (PAG, dorsomedial and 123 lateral) and parabrachial nucleus (PBN, lateral and medial) (Fig. 3c, Suppl. Fig. 3). Retrograde projections 124 were assessed after AVV2-TVAeGFP-oG injection followed by Rab-ΔG-EnvA-mCherry (18) injection 21 days 125 later into the VMH of Rxfp4-Cre mice (Fig. 3d). The retrograde monosynaptic transport of Rab-mCherry 126 labelled inputs from several nuclei established in feeding regulation, including the ARC, DMH and PVH 127 ( Fig. 3d-f). The neuronal circuitry surrounding RXFP4 VMH cells is summarised in Fig. 3g. 128 Due to the previously reported altered feeding patterns and macronutrient preferences of Rxfp4 knock-129 out mice (4), we further explored the physiology of the RXFP4 VMH population by ablating Rxfp4-expressing 130 cells using rAAV-DTA injected into the VMH of RXFP4 GCaMP3 mice (Fig. 4a). Postmortem histological analysis 131 revealed a >90% reduction in RXFP4 VMH cells (identified by GFP immunohistochemistry) across the VMH. 132 In the 8 weeks following the surgery, the RXFP4 VMHKO mice gained less body weight compared to the 133 control rAAV-mCherry injected mice (Fig. 4c), when fed with a choice of standard chow and HFD in parallel. 134 Five weeks post-surgery the mice were studied in metabolic cages. RXFP4 VMHKO mice had a lower 135 respiratory exchange ratio (RER, Fig. 4d), whilst energy expenditure and ambulatory activity (Fig. 4e,f) 136 were unaffected by ablation. Average 24hr food intake was significantly reduced, a consequence of 137 reduced HFD intake (Fig. 4g). Meal duration was unaffected by treatment, however the interval between 138 meals ( Fig. 4i) was significantly increased. These data demonstrate the importance of this neuronal 139 population in governing long-term feeding behaviour. 140 Given that RXFP4 VHMKO mice ate less HFD, yet central Insl5 infusion, which we would predict to inhibit 141 RXFP4-expressing neurons, increased HFD intake, we investigated the effects of acute activation or 142 inhibition of Rxfp4-expressing cells. Initially we used a whole-body hM4Di DREADD Cre-reporter (RXFP4 wb-143 Di ) to mimic the established RXFP4-signalling via pertussis-toxin sensitive Gi pathways (2) (Fig. 5a). During 144 the light phase, activation of Di in Rxfp4-expressing cells using CNO had no measurable effect on food 145 intake in AL fed mice (Fig. 5b). However, when animals were habituated to the appearance of a HFD or 146 HPM for 1 hour during the light phase, CNO application resulted in increased food intake (Fig. 5c,d). These 147 results were consistent with the response to infusion of INSL5 into the VMH. To investigate this further, 148 we gave mice housed in metabolic cages the choice between standard chow and HFD. CNO injection 149 during the light phase significantly increased HFD but not chow intake in RXFP4 wb-Di mice (Fig. 5e). This 150 effect was transient (Fig. 5f), consistent with the pharmacokinetics of CNO (19), and occurred without any 151 significant differences in ambulatory activity or energy expenditure compared to the vehicle cross-over 152 control (Suppl. Fig. 4a,b). 153 We next investigated the effects of whole-body hM3Dq Cre-reporter activation in Rxfp4-expressing cells 154 (RXFP4 wb-Dq ) (Fig. 6a). After a 2 h brief fast, to enhance feeding motivation, activation of Dq in Rxfp4-155 expressing cells at the onset of the dark phase had no measurable effect on the food intake of mice offered 156 only standard chow (Fig. 6b). However, when RXFP4 wb-Dq animals were habituated to the appearance of 157 HFD or a HPM for 1 hour at the onset of the dark phase, activation of Dq expressing cells with CNO resulted 158 in a marked reduction in HFD or HPM intake (Fig. 6c,d). These results were consolidated in AL fed animals 159 tested in metabolic cages with parallel access to standard chow and HFD. RXFP4 wb-Dq activation in AL fed 160 animals at the onset of the dark phase had no effect on standard chow intake, but significantly and 161 transiently reduced HFD consumption (Fig. 6e,f). RXFP4 wb-Dq activation also attenuated the increase in 162 energy expenditure associated with the onset of the dark phase, however, there was no effect on 163 ambulatory activity (Suppl. Fig. 4c,d). 164 To probe whether Rxfp4-expressing cells play a role in the motivational aspects of feeding, we calorically 165 restricted male RXFP4 wb-Dq animals to 95% body weight and placed them in operant chambers. Mice were 166 tested with a fixed ratio (FR) schedule, requiring 5 nose pokes to release a food reward (liquid Ensure), or 167 a progressive ratio (PR) schedule requiring increasing number of nose pokes for each subsequently earned 168 reward (in this case, +4, i.e. 1, 5, 9, 13, etc). RXFP4 wb-Dq mice treated with CNO completed fewer attempts 169 under FR to earn individual Ensure rewards (Fig. 6g). Under a PR schedule, they exhibited a reduced 170 breakpoint, i.e. CNO-treated mice stopped working for the HPM-reward at lower ratios than when 171 receiving vehicle treatment (Fig. 6h). In an effort related choice (ERC) paradigm, where animals had the 172 choice of working for a HPM (FR8, liquid Ensure) or consuming freely available standard chow, CNO 173 treatment reduced HPM consumption (Fig. 6i). However, animals consumed similar amounts of standard 174 chow and displayed otherwise normal behaviour (supplementary video 1), suggesting that activation of 175 Rxfp4-expressing cells reduced motivation for the HPM rather than inducing generalised malaise. 176 To assess whether the Rxfp4-expressing population in the VMH is involved in the feeding phenotype 177 observed in RXFP4 wb-Dq mice, the effect of acute chemogenetic manipulation of RXFP4 VMH neuron activity 178 on food intake was investigated. Male and female Rxfp4 GCAMP3 mice received bilateral VMH injections of 179 Cre-dependent hM3Dq-expressing rAAVs (rAAV-hSyn-DIO-hM3D(G)q-mCherry) designed to preferentially 180 target neurons, to produce RXFP4 VMHDq mice (Fig. 7a). Targeting efficiency was subsequently determined 181 by immunohistochemistry (Fig. 7b), or in a subset by fluorescent microscopy in live slices (see below). All 182 mice demonstrated robust transduction that was limited to the target region. To confirm the functional 183 activation of these RXFP4 VMHDq neurons, we generated ex vivo brain slices from a subset of RXFP4 VMHDq 184 mice (Fig. 7c). Calcium imaging demonstrated that ex vivo treatment with CNO activated 185 mCherry/GCaMP3 positive RXFP4 VMHDq , stimulating an increase in the frequency of calcium oscillations 186 and an increase in GCaMP3 fluorescence in Rxfp4-expressing somas (Fig. 7c,d). 187 The effect of chemogenetic activation of this cell population in vivo on food intake was studied in a 188 crossover design. In chow fed mice, and in line with RXFP4 wb-Dq animals, CNO treatment of RXFP4 VMHDq 189 mice had no effect on standard chow intake at the onset of the dark phase (Fig. 7e). When animals were 190 habituated to the appearance of a HFD or HPM at the onset of the dark phase, CNO resulted in a significant 191 reduction in food intake (Fig. 7f,g). When offered HFD and chow diet in parallel in metabolic cages, CNO 192 significantly reduced intake of the HFD whilst intake of standard chow was not altered, resulting in an 193 overall reduced caloric intake (Fig. 7h). As seen with RXFP4 wb-Dq animals, this was a transient effect no 194 longer apparent in the 2 nd hour post CNO administration (Fig. 7i). CNO had no effect on energy 195 expenditure or ambulatory activity in these RXFP4 VMHDq animals (Suppl. supporting the conclusion that Rxfp4 neurons in the VMH play a role in the physiological drive to consume 210 highly palatable foods. Overall, these data identify hypothalamic RXFP4 signalling as a key regulator of 211 food intake and preference. 212 Rxfp4 expression has been difficult to localise due to low mRNA expression levels and the lack of suitable 213 verified antibodies. Rxfp4 expression was previously reported in the colon (4) and in enteroendocrine 214 tumor cell lines (21,22). In contrast to our previous report (4), we detected Rxfp4 mRNA in the 215 hypothalamus ( Fig. 1g,h) and found substantial Rxfp4-dependent reporter expression in multiple brain 216 regions, with distinct Rxfp4-expressing cell populations in the accessory olfactory bulb, RSC, VMHvl, and 217 mammillary body (Suppl. Fig. 1). While it could be argued that this reflects lineage tracing from Rxfp4-218 positive precursor cells, the detection of Rxfp4 mRNA by RNAscope, the activation of Cre-dependent 219 rAAV-constructs when stereotactically injected into the adult VMH, the cAMP responsiveness of Rxfp4-220 positive cells to locally-perfused INSL5 in slice preparations, and the effect of stereotactically injected 221 INSL5 on feeding behaviour, confirm active Rxfp4-promoter activity and consolidate active RXFP4 222 expression and function in the adult mouse brain (Fig. 1g,h, Fig. 7b). 223 To characterise the transcriptomic profile of hypothalamic Rxfp4-expressing cells we performed scRNA-224 Seq. Although some cells will have been lost and some genes may have exhibited altered expression 225 during the cell dissociation and sorting process, the results allowed us to cluster Rxfp4-expressing cells 226 into several subpopulations each characterised by a profile of cell-specific marker genes (Fig. 2a,b). Rxfp4 227 was identified in microglia, ependymocytes and endothelial cells that potentially constitute the blood 228 brain barrier, suggesting that INSL5 may additionally exert effects on non-neuronal cells (23) previously associated with food intake and energy expenditure (12,15,16). However, chemogenetic 236 activation of RXFP4 VMHDq neurons did not result in increased energy expenditure or ambulatory activity, 237 contrasting with the previously described Esr1 VMHDq phenotype(16). The role of RXFP4 VMH cells in feeding 238 regulation is further suggested by the co-expression of the neuropeptide receptors Mc4r and Nmur2. 239 MC4R activation has been linked to suppressed food intake through regulation of Bdnf expression in the 240 VMH (26) -Bdnf was co-expressed in cluster 1 neurons in our dataset (Fig. 2d). Acute administration of 241 NMUR2 agonists have been shown to decrease feeding, with one agonist being somewhat selective to 242 HFD intake regulation (27), mirroring RXFP4 VMHDq cell activation (Fig 7). 243 Mapping of the retrograde inputs to and anterograde projections from the Rxfp4-expressing cells in the 244 VMH revealed a distinct neural circuitry surrounding this hypothalamic population. While we aimed to 245 target the VMHvl specifically during stereotactic injections, disparities between current mouse brain 246 atlases make it difficult to distinguish the VMHvl from the adjacent VMHc and TN, which have also been 247 implicated in feeding behaviour (13,30) . It is therefore possible that some Rxfp4-expressing cells in the 248 VMHc and TN were transfected by the viral vectors and acted as starter cells in these tracing experiments. 249 The Rxfp4-expressing population we targeted is therefore best described as RXFP4 VMH . Monosynaptic 250 inputs to RXFP4 VMH cells were labelled predominantly from brain regions involved in homeostatic 251 regulation of food intake such as the ARC, PVH and LHA ( Rxfp4-expressing cells in RXFP4 wb-Dq mice reduced an animal's drive to seek out and work for a highly 263 palatable food reward (Fig 6g-i). Taken together, these data suggest that Rxfp4-expressing cells may 264 influence motivation and reward-related behaviour via regulation of central reward signalling pathways. 265 RXFP4 VMH cells also send projections to the PAG and PBN, two integration sites responsible for relaying 266 sensory information between the forebrain and hindbrain and coordinating behaviour in response to 267 various stimuli including metabolic, gustatory and nociceptive inputs (36-38). This RXFP4 VMH projection 268 map aligns with previously identified projection regions from the VMHvl and SST-expressing cells in the 269 TN (13,39). Interestingly, all retrograde-labelled input regions also received projections from RXFP4 VMH 270 cells suggesting a high level of bidirectional connectivity within the RXFP4 VMH signalling network (Fig. 3g). 271 Similar bidirectional connectivity has been shown for an Esr1+ve VMHvl population (39). The RXFP4 VMH 272 neural network established in the present study suggests these cells may integrate metabolic and 273 nutritional cues either directly or via other hypothalamic regions and regulate the reward system to 274 influence ingestive behaviours. The low number of RXFP4 VMH input regions compared to projection regions 275 suggests these cells may comprise an early node in this network. 276

Inhibition of Rxfp4-expressing cells via direct intra-VMH infusion of INSL5 or whole body Di acutely 277
increased intake of both a HFD and a HPM in the home cage and when animals were offered a choice of 278 standard laboratory chow and HFD in metabolic cages, without altering energy expenditure and activity 279 in both male and female mice (Fig 1j,k, Fig 5, Suppl. Fig 4a,b). This is consistent with our previous 280 demonstration that INSL5 administration increased food intake (4, 5). Activation of Di receptors should, 281 at least in part, mimic the Gαi/o coupling of RXFP4 (2). By contrast, activation of Rxfp4-expressing cells with 282 Dq produced a robust suppression of HFD and HPM intake (Fig 6). Targeting the RXFP4 VMH population with 283 an rAAV-Dq reporter recapitulated the findings with the whole body-Dq approach (Fig. 7). Whilst any of 284 the identified Rxfp4-expression sites could participate in the feeding phenotype of the global DREADD 285 reporter mice, the VMH-specific rAAV-Dq reporter phenotype indicates that this hypothalamic population 286 either underlies or at least contributes to the observed anorexigenic effects in the RXFP4 wb-Dq model. 287 Our data suggest that activation of Rxfp4-expressing cells in the VMH suppresses the consumption and 288 drive to work for calorie dense HFD and HPM. The importance of the VMH in the regulation of feeding 289 and metabolism has been disputed (reviewed in (40)). Initial studies suggested the VMH might be a 290 "satiety centre", as VMH-lesioned rats, particularly females, over-consumed a HFD when fed AL (41), 291 despite being seemingly less willing to work for food on a fixed ratio lever-pressing paradigm (42). The 292 observed hyperphagia has subsequently been linked to additional damage to adjacent hypothalamic 293 structures (see King 2006(40) for discussion), while the reduced motivation was not observed when rats 294 were trained pre-operatively, suggesting that the VMH lesion may have altered "trainability" rather than 295 feeding motivation (40,43). Lesioning studies, whilst informative, lack cellular precision and damage 296 neural connections to other feeding centres within the brain. More recent work employing 297 immunohistochemistry, RNA sequencing, chemogenetics and neuronal projection mapping, has 298 demonstrated that the VMH consists of anatomical subdivisions made up of distinct cell populations (15, 299 44, 45). Functional studies suggest neurons in the central and dorsomedial VMH regulate feeding, energy 300 expenditure and glucose homeostasis (30, 46), while the VMHvl is more frequently implicated in the 301 control of social and sexual behaviours (45,47,48). However, several studies have demonstrated the 302 involvement of VMHvl neurons in energy expenditure and feeding behaviour (12, 49). Increases in physical 303 activity have been observed following chemogenetic activation of NK2 homeobox transcription factor 1 304 (Nkx2-1)-expressing neurons in the VMHvl of female rats, while knockout of Nkx2-1 in the VMHvl leads to 305 decreased physical activity and thermogenesis (50). Furthermore, chemogenetic activation of Esr1-306 expressing VMHvl neurons was found to stimulate physical activity and thermogenesis in both sexes (16). 307 Esr1 signalling in the VMHvl was previously demonstrated to influence food intake, energy expenditure 308 and glucose tolerance as VMHvl-restricted knockdown of Esr1 resulted in increased food intake, 309 decreased physical activity and thermogenesis, reduced glucose tolerance and obesity in female rats (12). 310 Activation of RXFP4 VMHDq neurons reduced HFD and HPM intake but had no effect on chow intake, energy 311 expenditure or ambulatory activity (Fig. 7, Suppl. Fig. 4e,f). This, alongside the transcriptomic profile and 312 neural circuitry of RXFP4 VMH neurons, suggests that these neurons comprise a distinct VMH population 313 modulating food intake and preference based on the rewarding aspects of food rather than the 314 homeostatic food intake or energy expenditure responses observed during chemogenetic manipulation 315 of other VMH populations. 316 We recognise several limitations to this study. First, the exact classification of Rxfp4-expressing brain 317 regions is difficult given the disparities between current brain atlases. Whilst we aimed to target Rxfp4-318 expressing cells in the VMHvl, the inexact nature of stereotactic injections may have resulted in additional 319 targeting of neurons in adjacent VMHc, LHA and TN regions, which may have contributed to the food 320 intake phenotype of RXFP4 VMHDq mice. The fact that only a subset of Rxfp4 CAMPER neurons responded to 321 INSL5 in acute slices likely reflects technical limitations of this preparation. The similarities of the feeding 322 outcomes of intraparenchymal INSL5 in WT mice and CNO application in Rxfp4 VMHDq mice together with 323 RNAscope confirmation of Rxfp4-Cre labelled VMH neurons makes the alternative explanation of 324 substantial additional off target Cre-expression unlikely, but we cannot fully rule this out. It is also 325 uncertain whether INSL5 is the endogenous ligand acting on Rxfp4-expressing cells in the brain. We have 326 previously been unable to identify Insl5-expressing cells in the mouse brain (5) and there is no evidence 327 that INSL5 can cross the blood brain barrier. In addition, relaxin-3 also activates RXFP4 (2), is expressed in 328 the mouse brain and is orexigenic (51), hence it is possible that relaxin-3 is involved in central RXFP4 329 action. The viral tracing techniques used to decipher the RXFP4 VMH neuronal network also have some 330 limitations. In the retrograde tracing experiments, it was difficult to detect rAAV-GFP immunoreactive 331 cells, making it hard to confirm the exact starter cells in this experiment. Furthermore, while the ChR2-332 mCherry construct is preferentially targeted to axon terminals, it is possible that some of the mCherry-333 labelled fibres in adjacent regions, such as the ARC, are in fact dendrites (52) which may underlie our 334 inability to detect retrograde-only labelled regions. Nevertheless, we have been able to identify distinct 335 regions that project onto and receive projections from Rxfp4-expressing cells in the VMH that connect 336 these cells to known feeding-related neural networks. Finally, it is difficult to explain apparently similar 337 outcomes of acute chemogenetic activation and chronic ablation of RXFP4 VMH neurons, both resulting in 338 reduced HFD consumption. The chronic ablation outcome would be compatible with a positive valence of 339 Rxfp4 VMH neurons, whilst the acute Dq experiment would be compatible with a negative valence of the 340 same neurons on HFD consumption, and both interpretations would be compatible with the response to 341 acute parenchymal Insl5 and the phenotype of Rxfp4-knock-out mice. Whilst we are currently unable to 342 distinguish these two interpretations, we nevertheless establish Rxfp4 VMH neurons as integral players in a 343 complex neural network consuming highly palatable food consumption. 344 In summary, we have characterised a previously unrecognised population of ventromedial hypothalamic 345 cells that express Rxfp4 in mice, identified projections in homeostatic and hedonic feeding centres in the 346 CNS, and demonstrated that acute manipulation of these cells modulates HFD/HPM intake without 347 affecting chow intake or energy expenditure. Together, these findings suggest Rxfp4-expressing neurons 348 in the VMH are key regulators of food preference and represent a target for the modulation of feeding 349 behaviour.

Mouse models 363
To express Cre recombinase under the control of the Rxfp4 promoter, we replaced the sequence between 364 the start codon and the stop codon in the single coding exon of Rxfp4 in the murine-based BAC RP24-72I4 365 (Children's Hospital Oakland Research Institute) with iCre (53) sequence using Red/ET recombination 366 technology (GeneBridges) (Fig 1A). The resulting BAC was used to create BAC-transgenic mice -of four 367 initial founders, two passed the transgene to their offspring; both resulting lines showed similar Cre-368 reporter expression and one line, Rxfp4-73, was used throughout this manuscript (see Suppl. CNO at 11:00 (± 30mins) following a 2h fast. Food was weighed at 1h post-injection. For dark phase 407 activation, animals were injected with vehicle or CNO at 19:00 at the onset of the dark phase following a 408 2h fast. Food was weighed at 1h post-injection. In trials with a high fat diet (HFD) and a highly palatable 409 meal (HPM, liquid Ensure, Abbott Laboratories, 353-3601), mice were habituated to the appearance of 410 the test meal for 1 hour (5 days per week) for two weeks prior to testing. 411 412

Operant chambers 413
Twelve male RXFP4 wb-Dq mice (weighed 3 times weekly) were food restricted to maintain 95% body weight 414 for two weeks prior to training and testing in standard mouse Bussey-Saksida touchscreen chambers 415 (Campden Instruments Ltd, Loughborough, UK). Training and testing procedures were conducted as 416 previously described (58). Briefly, mice were trained to touch a screen for a reward (the HPM, 20 µL) under 417 a fixed ratio (FR) schedule for 2 weeks, progressing from FR1 to FR5 (training deemed successful when the 418 animal earnt 30 rewards within 1 hour), followed by testing. Mice then progressed to progressive ratio 419 (PR, increment +4 i.e. 1, 5, 9, 13, etc), where the breakpoint was defined as the last reward earned before 420 5 minutes elapsed without operant response. Following testing of the breakpoint, mice progressed to the 421 effort related choice schedule (ERC) -mice were trained on FR8, with the addition of standard chow to 422 the operant arena. Once animals successfully earned 30 rewards within 60 minutes, testing was 423 undertaken. The 60-minute training and testing sessions took place at the same time each day (between 424 10:00-13:00). 425 426

Metabolic cages 427
Animals were acclimated to metabolic cages prior to study and data collection. Oxygen consumption and 428 carbon dioxide production were determined using an indirect calorimetry system (Promethion, Sable 429 Systems, Las Vegas, NV). The system consisted of 8 metabolic cages (similar to home cages), equipped 430 with water bottles and food hoppers connected to load cells for continuous monitoring, in a temperature 431 and humidity-controlled cabinet. The respiratory exchange ratio (RER) and energy expenditure (via the 432 Weir equation) were calculated, whilst ambulatory activity was determined simultaneously. Raw data 433 were processed using ExpeData (Sable Systems). Animals were exposed to standard chow and a HFD 434 during metabolic assessment or standard chow. 435

CAMPER and calcium imaging 437
Mice were anesthetized using sodium pentobarbital at a dose of 180mg/kg (Dolethal, Vetoquinone). A 438 laparotomy was then performed, the heart exposed and the animals were intracardiacally perfused with 439 Sections were then transferred to standard ACSF (in mM: 3 KCl, 118 NaCl, 25 NaHCO3, 10 Glucose, 1MgCl2, 445 CaCl2; pH7.4) at room temperature for 1h before recording. All solutions were continuously bubbled with 446 95% O2/5% CO2. 447 Imaging was performed using a Zeiss Axioskop2 FS Plus microscope. Slices were immobilised in a glass 448 bottom chamber using a slice anchor (Warner instruments) and imaged using a 40x water immersion 449 objective. For CAMPER imaging, RXFP4 CAMPER neurons of the VMH were excited using a LED light source 450 Fluor® 488 or 555; Invitrogen) diluted 1:500 for 2 hrs at room temperature. Following washing, mounted 503 sections were coverslipped on superfrost slides using Vectashield (Vector Laboratories, H-1400-10). Slides 504 were imaged using an Axio Scan.Z1 slide scanner (Zeiss) and confocal microscope (Leica TCS SP8 X, 505 Wetzlar, Germany) with a 20x or 40x objective as indicated. Images were analysed in Halo Image Analysis 506 Software (Indica Labs) and ImageJ. 507

GFP
Tissue was digested in Hibernate-A medium without calcium (BrainBits) containing 20 U/mL Papain 515 (Worthington) and 1% GlutaMAX for 30 min at 37 o C under agitation (Thermomixer, 500 rpm). After 516 digestion, tissue was triturated in Hibernate-A medium with 3.5 U/mL DNase I (Sigma, D4263). The 517 trituration supernatant was loaded on top of a BSA gradient prepared in Hibernate-A medium, spun for 5 518 min at 300 rcf, and the pellet was resuspended in Hibernate-A medium + 1% BSA. The cell suspension was 519 filtered through a 70 µm cell strainer into a fresh tube. Fluorescence-activated cell sorting was performed 520 using an Influx Cell Sorter (BD Biosciences, San Jose, CA, USA). Cell gating was set according to cell size 521 (FSC), cell granularity (SSC), FSC pulse-width for singlets, fluorescence at 488 nm/532 nm for EYFP and 522 647/670 nm for nuclear stain with DraQ5 (Biostatus, Shepshed, Leicester, UK) to exclude cellular debris, 523 aggregates and dead cells. Cells were sorted directly into individual wells of 96-well plates containing lysis 524 buffer. 384 YFP-positive cells were isolated and processed using a Smart-Seq2 protocol (60). Libraries 525 were prepared from ~150 pg of DNA using the Nextera XT DNA preparation kit (Illumina, FC pentobarbital solution (50 mg/kg in saline) and transcardially perfused with PBS followed by 4% PFA in 543 PBS. Brains were extracted and post-fixed in 4% PFA for 24 hrs before being transferred to 25% sucrose 544 for 24 hrs at 4°C. Brains were embedded in OCT compound, frozen in a Novec 7000 (Sigma)/dry ice slurry 545 and stored at -80 C. 16µm cryosections containing the hypothalamus were prepared on a Leica CM1950 546 cryostat (Wetzlar, Germany) at -12°C and stored at -20°C until required. 547 Slides were thawed at room temperature for 10 min before baking at 60 o C for 45 min. The sections were 548 then post-fixed in pre-chilled 4% PFA for 15 min at 4 o C, washed in 3 changes of PBS for 5 min each before 549 dehydration through 50%, 70%, 100% and 100% ethanol for 5 min each. The slides were air-dried for 5 550 min before loading onto a Bond Rx instrument (Leica Biosystems). Slides were prepared using the frozen 551 slide delay prior to pre-treatments using Epitope Retrieval Solution 2 (Cat No. AR9640, Leica Biosystems) 552 at 95°C for 5 min, and ACD Enzyme from the LS Reagent kit at 40 o C for 10 minutes. Sections were imaged on a Slide Scanner Axio Scan.Z1 microscope (Zeiss) using a 40x air objective. Three 559 z-stack slices spanning 1.5μM were combined into an extended depth of field image (ZEN 2.6, Zeiss). The 560 CZI files were read into Halo Image Analysis Software (Indica Labs). 561 562