A critical update on the leptin- melanocortin system

hydroxytryptamine


| THE RIS E OF THE CL A SS I C LEP TIN-MEL ANOCORTIN MODEL
Obesity is a complex chronic disease essentially resulting from an imbalance between calories consumed and burned (WHO, 2021).
However, in healthy individuals, mismatches in energy balance often occur and various biological mechanisms exist to correct these discrepancies and maintain body weight. Therefore, a deficient regulation of energy homeostasis could be a major mechanism contributing to the development of obesity, promoting chronic energy surplus and metabolic alterations characteristic of obesity.
Research on the biological mechanisms regulating energy balance took off with the discovery of two distinct autosomal recessive mutations, obese (ob) and diabetes (db), that resulted in the development of obese mice (ob/ob and db/db), that are still routinely used as genetic models of obesity (Hummel et al., 1966;Ingalls et al., 1950). In addition to being obese, ob/ob and db/db mice exhibit hyperphagia, hyperglycemia, hyperinsulinemia, and are sterile, phenotypes that mimic common characteristics of severe obesity in humans (Coleman & Hummel, 1967, 1973Genuth et al., 1971;Hummel et al., 1966). In the mid-1990s, the gene products mutated in the ob/ob and db/db obese mice were identified. Discovery of the OB protein, now known as leptin (Zhang et al., 1994), and its receptor (DB, now known as LEPR) (Chen et al., 1996;Lee et al., 1996;Tartaglia et al., 1995) resulted in the "molecular era" of obesity research, a long and exciting scientific journey that has allowed a better understanding of the mechanisms regulating energy balance.
Many research groups have since contributed to the current knowledge on leptin's functions. Early on, it was shown that intravenous (iv) injection of radioactive leptin led to its accumulation in specific brain regions, suggesting that leptin, produced peripherally, could enter the brain parenchyma to mediate some of its effects (Banks et al., 1996). Multiple studies then cemented the idea that the brain was a primary target of leptin. In particular, iv injection of leptin was shown to activate neuronal groups predominantly located in brainstem and hypothalamic nuclei known to be involved in the regulation of energy balance (Elmquist et al., 1997). Further studies also demonstrated that leptin could influence the expression of specific genes in the arcuate nucleus of the hypothalamus (ARC), a critical region for appetite regulation. For instance, intracerebroventricular (icv) or intraperitoneal (ip) injections of leptin were shown to regulate the mRNA expression of neuropeptide Y (Npy) and pro-opiomelanocortin (Pomc) in the ARC, encoding two peptides well known for their implication in energy metabolism (Schwartz et al., 1996(Schwartz et al., , 1997Thornton et al., 1997). This reinforced the idea that leptin, a satiety factor produced peripherally by the adipose tissue, could induce important changes in the brain.
Although the effects of leptin in the brain were clear, the mechanisms underlying its effects on different neuronal populations were still unknown. Part of the answer was elucidated when research groups began mapping the expression of leptin receptor (Lepr) isoforms in the brain. While most Lepr isoforms have a short intracellular domain, the long (Lepr-b) isoform differs from the others by its long intracellular domain, which contains motifs important for initiating a variety of intracellular cascades (Chen et al., 1996;Lee et al., 1996). It was shown that Lepr-b was expressed in the ARC, retrochiasmatic area (RCA), lateral hypothalamus (LH), ventral premammillary nucleus (PMV), ventromedial nucleus (VMH), paraventricular nucleus (PVH), and the dorsomedial nucleus (DMH) of the hypothalamus Mercer, Hoggard, Williams, Lawrence, Hannah, Morgan, et al., 1996;.
Interestingly, leptin-induced C-FOS immunoreactivity (IR) colocalized with Lepr-b in most hypothalamic nuclei, suggesting that leptin could directly activate cell populations in those regions (Elias et al., 2000;Elmquist et al., 1997). Abundant expression of Lepr-b was also detected in other brain regions such as the cerebellum, multiple thalamic nuclei, the cortex, hippocampus, and the ventral tegmental area (VTA) Elmquist et al., 1998;Figlewicz et al., 2003;Mercer, Hoggard, Williams, Lawrence, Hannah, Morgan, et al., 1996;. Highlighting the contribution of such extrahypothalamic neurons in energy homeostasis, recent reports have identified neuronal populations in the cerebellum regulating satiation (Low et al., 2021), neuronal circuits including the thalamus being partly involved in leptin's anorexigenic effects (Zhang et al., 2020), and LEPR-expressing neurons in the VTA and substantia nigra involved in food reward, energy intake, and locomotion Fernandes et al., 2015;Hommel et al., 2006).
Following these discoveries, a major focus was to characterize leptin-activated neurons. Double labeling experiments were performed and revealed that many cell populations activated by leptin expressed neuropeptides that have potent effects on food intake (Elias et al., 2000). These findings ultimately led to a prevalent model of energy balance regulation in which leptin acts directly on specific hypothalamic neurons to regulate food intake and satiety ( Figure 1) (Schwartz et al., 2000). This model proposed the ARC, where the highest density of Lepr-b was found, as the center of the leptin-melanocortin system. However, recent data have challenged the completeness of this homeostatic model which, as stated by Stephen C. Woods, "has proven to be overly simplistic and misleading" (Stephen & Denovan, 2015). In this review, we highlight findings from recent studies challenging the classical model of the hypothalamic leptin-melanocortin system with a particular focus on ARC melanocortin neurons. We discuss how prenatal models used to study the regulation of energy balance have led to many assumptions concerning leptin's direct effects on POMC neurons. We also provide evidence that non-POMC neurons are more likely to play a direct role in mediating leptin's anorexigenic effects. We hope that this update on the leptin-melanocortin system will help the field to move forward beyond assumptions.

| What prenatal models taught us?
Although the initial model proposed in 2000 (Figure 1) was key in inspiring many important discoveries on the central control of energy balance, this model has been proven overly simplistic, yet is still depicted as a dogma in many recent reviews. In a landmark article published in 2004, Balthasar and colleagues investigated whether the LEPR was required for POMC neurons to regulate energy balance (Balthasar et al., 2004). To do so, they used the Cre/ LoxP system to prenatally delete motifs essential for LEPR signaling in Pomc-expressing cells. They reported that mice lacking the cytoplasmic tail of Lepr-b (Pomc-Cre::Lepr flox/flox ) had modest increase in body weight and fat mass. Importantly, neither food intake nor oxygen consumption could statistically explain this mild phenotype.
Notably, the weight gain observed in Pomc-Cre::Lepr flox/flox mice was only 18% of mice globally lacking Lepr-b (db/db), suggesting that most of leptin's effects on energy balance are mediated through other Lepr-b-expressing neurons (Balthasar et al., 2004). Along the same lines, selective re-expression of Lepr-b in POMC neurons of mice otherwise deficient for Lepr-b (Pomc-Cre::Lepr TB/TB ) was shown to result only in a 15% weight loss with no effects on food intake (Berglund et al., 2012). Following these studies, it is surprising that POMC neurons kept being considered a main site of action for leptin's control of energy balance.
Cumulating evidence suggests that the mild body weight variations described in these studies are most likely the consequence of Lepr deletion/re-expression from non-POMC neurons, including a proportion of antagonistic NPY neurons. In the developing brain, Pomc is broadly expressed in various distinct immature hypothalamic neuronal populations (Padilla et al., 2010;Yu et al., 2022). As brain maturation progresses, Pomc expression is turned off in more than 50% of ventricular hypothalamic cells. Unfortunately, in prenatal Pomc-Cre models, this transient expression of Pomc during development is sufficient to induce Cre-recombination in a decent proportion of future non-POMC cells. For instance, it was shown that Pomc-Cre models could induce recombination in 25% of NPY neurons (Padilla et al., 2010). This observation was also confirmed using single-cell RNA sequencing in Pomc-eGFP mice, showing that 27% of mature POMC neurons in the ARC express low Pomc and high Npy/agouti-related peptide (Agrp) levels, with a gene expression profile very similar to AGRP neurons (Lam et al., 2017). Moreover, in Pomc-eGFP::Pomc-Cre::tdTomato mice, although 83% of ARC leptinresponsive cells were labeled with tdTomato, only 13% of these cells were found to be authentic POMC cells based on eGFP expression (Padilla et al., 2012). Therefore, data resulting from the use of prenatal Pomc-Cre models must be interpreted with extreme caution, even though such models were essential in establishing many important bases in the field. It is also important to note that while GFP models appear better at labeling authentic POMC cells, important caveats need to be considered, including the fact that they can label extra-hypothalamic sites such as the dentate gyrus that does not normally express Pomc (Overstreet, 2004).
Results from prenatal models targeting Lepr must also be interpreted carefully as leptin plays a pivotal role in the development of the hypothalamus. Leptin and Lepr-b are required for the proper development of ARC projections to key regions involved in the regulation of energy homeostasis (Bouret et al., 2004(Bouret et al., , 2012. For

F I G U R E 1
The classic leptin-melanocortin model. Leptin is secreted predominantly by the adipose tissue and reaches the brain parenchyma through the bloodstream. The arcuate nucleus of the hypothalamus (ARC) hosts AGRP/NPY and POMC neurons, two distinct antagonistic neuronal populations that have been shown to express Lepr-b Mercer, Hoggard, Williams, Lawrence, Hannah, Morgan, et al., 1996;. This model proposed that leptin acts directly on those neurons to induce opposite effects. Thus, high leptin levels, signaling a positive energy state, would activate POMC neurons and inhibit AGRP/NPY neurons to promote satiety. On the other hand, low leptin levels would activate AGRP/ NPY neurons and inhibit POMC neurons, promoting food intake. Once activated, the first-order AGRP/NPY and POMC neurons would affect second-order neurons, located outside of the ARC, by releasing NPY, AGRP, or POMC-derived α-MSH. Via their action on second-order neurons, these peptides would then induce broad effects on energy homeostasis (Schwartz et al., 2000). AGRP, agoutirelated peptide; LEPR, leptin receptor; NPY, neuropeptide Y; POMC, pro-opiomelanocortin; α-MSH, α-melanocyte-stimulating hormone.
instance, prenatal disruption of Lepr-b STAT3 signaling alters the axonal density from POMC neurons to the PVH, an effect that persists in adult animals (Bouret et al., 2012). Leptin also seems involved in astrogenesis (Rottkamp et al., 2015), and Lepr-b could have a role in the differentiation of hypothalamic neural precursor cells (Ren et al., 2020). Of note, impairing leptin signaling exclusively during specific developmental periods can lead to long-term metabolic defects in adulthood, such as leptin resistance, increased susceptibility to diet-induced obesity (DIO), alterations in energy homeostasis, and impaired Pomc levels (Attig et al., 2008;Ramos-Lobo et al., 2019).

| How modern technologies have improved knowledge
Because of limitations inherent to prenatal models, different approaches have been developed to avoid possible effects resulting from transient Cre expression in non-POMC neuronal populations.
Using a tamoxifen-inducible Pomc-CreER T2 model allowing the targeting of ~96% of authentic POMC neurons, it was clearly shown that Lepr-expressing POMC neurons are dispensable for the regulation of energy balance. Indeed, selective deletion of Lepr in POMC neurons of adult mice had no effect on food intake, oxygen consumption, locomotor activity, or body weight . The failure of Lepr-expressing POMC neurons to regulate food intake and energy expenditure was further supported by follow-up findings. One group used CRISPR-Cas9-based models to selectively delete Lepr in ARC AGRP or POMC neurons (Xu et al., 2018). Importantly, Lepr deletion in POMC neurons was achieved through viral injections, circumventing possible unspecific deletion events that could have happened with the Pomc-Cre model used. While deleting Lepr in ARC AGRP neurons induced a robust increase in food intake and body weight and a reduction in oxygen consumption, deletion of Lepr in ARC POMC neurons had no effect (Xu et al., 2018). Another group injected adeno-associated viruses (AAV) into the ARC of Agrp-Cre or Pomc-Cre mice to allow selective expression of designer receptors exclusively activated by designer drugs (DREADD) (Üner et al., 2019). While chronic activation or inhibition of AGRP neurons led to a robust increase or decrease in food intake, respectively, only a small increase in food intake was noted following POMC inhibition and no effect on food intake was reported following POMC activation (Üner et al., 2019). As POMC neurons are widely recognized as a very heterogenous population, the slight alteration in food intake reported could be explained by the inhibition of POMC subpopulations that do not express the LEPR, some of which have been shown to regulate feeding behavior (see Section 4). Supporting these observations, a recent study from Jens Brüning's lab put the last nail in the coffin by convincingly showing, using modern techniques, that POMC cells expressing Lepr do not regulate food intake directly (Biglari et al., 2021). By utilizing a novel Cre-Dre intersectional recombinase system, they were able to differentiate POMC cells expressing Lepr from POMC cells expressing the glucagon-like peptide 1 receptor (Glp1r). Of note, the later onset of Dre-dependent recombination allowed them to label almost exclusively authentic POMC neurons, thus circumventing the limitations of prenatal Pomc-Cre models. They found that POMC Lepr and POMC Glp1r were two mostly non-overlapping subpopulations in the ARC with distinct anatomical distribution and unique electrophysiological properties.
By combining the Cre-Dre system with DREADD expression, they activated exclusively either the POMC Lepr or the POMC Glp1r neurons. Strikingly, the activation of POMC Lepr cells had no effect on food intake (Biglari et al., 2021).
Taken collectively, the evidence from recent studies is clear that the effects of leptin on food intake and body weight do not involve direct actions on POMC neurons, in contrast to the model still commonly portrayed as a dogma (Figure 1). An important lesson from this discussion is that prenatal models must be avoided in future studies on the central regulation of energy homeostasis considering that such models may cause unspecific labeling and impair developmental processes, both of which may lead to inaccurate and misleading conclusions. Recent technological advances have led to the development of models allowing spatiotemporal, intersectional, chemogenetic and optogenetic control of gene expression, pathway activation, and neuronal firing. Are these models perfect? Probably not. This is why, as a community, we need to avoid building definitive conclusions from study to study while ensuring that no result is overextended. Instead, efforts should be made in combining tools and (imperfect) models to help us understand and better define the central leptin-melanocortin system.

| Many approaches, same conclusions
The ability of hypothalamic Lepr-b to regulate glucose homeostasis was demonstrated early on by studies using stereotaxic injections of viruses to restore Lepr expression in the ARC. Coppari and colleagues cleverly used viral vectors to allow FLPe-mediated re-activation of Lepr in the ARC of mice otherwise lacking the intracellular domain of the receptor globally (Coppari et al., 2005).
Unilateral restoration of a fully functional Lepr was sufficient to normalize blood glucose and markedly reduced circulating insulin levels that were both high in mice bearing the non-functional Lepr allele (Coppari et al., 2005). Subsequently, Morton and colleagues used a similar approach to restore Lepr-b expression in the ARC of rats globally lacking Lepr (Koletsky rats) (Morton et al., 2005). Koletsky rats are obese and severely diabetic, and re-expression of Lepr-b in the ARC considerably improved insulin sensitivity and glucose tolerance, effects that seemed independent of food intake and body weight, as pair-fed Koletsky rats still had altered glucose homeostasis (Morton et al., 2005). Later studies complemented these findings by showing that restoring LEPR signaling in the ARC improved the insulin-mediated suppression of hepatic glucose production and reduced the expression of key gluconeogenic enzymes (German et al., 2009). These effects were prevented by hepatic vagotomy, suggesting an ARC-liver circuit involved in glucose homeostasis (German et al., 2009). It should be noted that leptin also seemed to regulate hepatic glucose production independently of insulin (German et al., 2011).
While these observations hinted that some ARC neurons might be directly controlled by leptin to regulate glucose homeostasis, the techniques used in these studies prevented the precise identification of the neurons at cause for these effects. On that matter, another set of studies provided strong evidence that POMC cells expressing Lepr are directly involved in the regulation of glucose homeostasis. Selective re-expression of Lepr-b in POMC cells of db/db mice completely normalized blood glucose levels and partially restored blood insulin levels independently of food intake and body weight (Huo et al., 2009). Similar results were also achieved using a different genetic approach. Selective re-expression of Lepr in POMC neurons normalized glycemia to control levels and improved hepatic insulin sensitivity (Berglund et al., 2012). Conversely, inactivating Lepr and insulin receptor (Insr) signaling in POMC neurons lead to hyperinsulinemia, insulin resistance, and impaired suppression of hepatic glucose production (Hill et al., 2010). Collectively, these studies confirmed that direct action of leptin on POMC neurons is crucial for the maintenance of normal glucose homeostasis. It is important to keep in mind that by the nature of the genetic models used to delete or re-express Lepr in POMC neurons, Lepr-expressing POMC cells located outside of the ARC cannot be ruled out and could, therefore, contribute to the altered glucose homeostasis that was reported. Indeed, such neuronal populations have been identified in the nucleus tractus solitarius (NTS), and evidence suggests that they could also be involved in the regulation of energy balance (Bronstein et al., 1992;Cheng et al., 2020;Ellacott et al., 2006;Garfield et al., 2012;Georgescu et al., 2020). Nevertheless, to our knowledge, Lepr-expressing POMC cells in the NTS have not been reported to be involved directly in glucose homeostasis. Moreover, arguing for a predominant role of ARC Lepr-expressing POMC neurons in glucose homeostasis, the observations made with genetic models faithfully reproduced the phenotypes reported following stereotaxic injections of viral vectors into the ARC.

| Unclear mechanisms
An important consideration regarding the models used in previous studies is that these were prenatal models that could thus cause unspecific effects and have important developmental repercussions (as discussed above). Recent studies using inducible genetic models were able to confirm the role of Lepr-expressing POMC cells in glucose homeostasis in adulthood. Using tamoxifen-inducible Pomc-CreER T2 mice, we previously deleted Lepr from POMC cells in adult mice and observed that mice developed insulin resistance and altered insulin-mediated suppression of hepatic glucose production as early as 1 week post recombination . Interestingly, glycemia was not altered until 2-3 weeks after treatment with tamoxifen, suggesting that Lepr deletion from POMC neurons first impairs liver insulin sensitivity, which then leads to hyperglycemia. Importantly, these effects were independent of food intake, fat mass, and body weight . Providing mechanistic insights, a subsequent study showed that leptin modulates the conductance of R-type voltage-dependent calcium channels (Ca v ) in leptin-responsive POMC neurons (Smith et al., 2018). Short-hairpin RNA (ShRNA)-mediated knockdown of the subunit forming the pore of the R-type Ca v (Ca v 2.3) in POMC cells impaired the ability of leptin to induce long-lasting depolarization. Interestingly, in vivo, this selective knockdown in POMC cells significantly increased hepatic glucose production and blood insulin levels without affecting body weight nor food intake. This work suggests an important role for R-type Ca v in leptin-mediated, sustained POMC depolarization and regulation of liver metabolism (Smith et al., 2018). To further investigate whether POMC neurons' activity is important for the regulation of glucose metabolism, Üner and colleagues performed stereotaxic injections of Cre-dependent DREADD viruses into the ARC of Pomc-Cre mice (Üner et al., 2019). In line with the idea that arcuate POMC neurons are involved in glucose homeostasis, chronic inhibition of ARC POMC cells significantly altered blood glucose levels of normoglycemic mice, independently of food intake. However, in opposition to the latter studies, this chronic inhibition decreased glycemia (Üner et al., 2019). This discrepancy could be due to the fact that this approach would inhibit all subpopulations of POMC neurons at once, rather than specifically targeting the ones that express LEPR. Therefore, it is still unclear whether some subsets of POMC neurons positively, while other negatively, affect glucose homeostasis, and the specific mechanisms by which they do so. This potential bidirectional effect could involve the innervation and control of different organ functions, such as hepatic glucose production or glucose uptake in adipose tissue, which were previously linked to melanocortin effects (Labbé et al., 2015;Manceau et al., 2020;Mountjoy, 2010;Ruud et al., 2017;Shin et al., 2017). Reinforcing the idea that ARC POMC cells can affect glucose homeostasis, a recent study characterized a circuit linking hypothalamic POMC cells to the liver. Using retrograde Cre-recombinase encoding viruses injected into the liver of reporter mice (Kwon et al., 2020) Moreover, optogenetic stimulation of those ARC POMC neurons projecting to the DMV increased expression of gluconeogenic enzymes and elevated glycemia. Further investigation suggested that this ARC POMC → DMV ACh → Liver circuit is particularly sensitive to hypoglycemia and stimulates hepatic glucose production independently of insulin or glucagon levels through mechanisms involving the vagus nerve (Kwon et al., 2020). However, it should be noted that recent studies have questioned the existence of direct parasympathetic (vagus) innervation of hepatocytes. In particular, two independent studies used immunolabeling-enabled three-dimensional imaging of solvent-cleared organs plus (iDisco+), and convincingly concluded that neural innervations within the liver are exclusively of sympathetic nature (Adori et al., 2021;Liu et al., 2021). This suggests that the vagus control of liver metabolism may occur through indirect effects, such as alterations of portal blood flow or by modulating the sympathetic nervous system (Bockx et al., 2012;Martinez-Sanchez et al., 2022). Nevertheless, the work from Kwon and colleagues is consistent with the study from Üner and colleagues, reporting that chronic inhibition of ARC POMC neurons lowers blood glucose (Üner et al., 2019). However, the observations from Kwon, Üner and colleagues (Kwon et al., 2020;Üner et al., 2019) contrast with previously mentioned studies that predominantly targeted Lepr-expressing POMC cells. It is not clear whether ARC POMC neurons projecting to the DMV express Lepr. Considering the heterogeneity of POMC cells Chen et al., 2017;Lam et al., 2017;Quarta et al., 2021;Steuernagel et al., 2022), it is likely that ARC POMC → DMV Ach and ARC Lepr-expressing POMC cells represent two distinct neuronal populations regulating different aspects of glucose metabolism. In sum, even though the mechanisms by which Lepr-expressing POMC neurons regulate glucose homeostasis are still unclear, the data support an important role for these neurons in mediating the direct effects of leptin on glucose, but not energy homeostasis (Figure 2a).

| POMC is still important in the control of energy balance
The previous sections emphasized the evidence that leptin regulates glucose metabolism but not food intake through direct actions on POMC neurons. However, this does not invalidate the importance of POMC in controlling energy balance. In fact, POMC has a pivotal role in the control of energy balance . In humans, cases of early-onset, severe obesity have been linked to congenital mutations in the POMC gene (Biebermann et al., 2006;Farooqi et al., 2006;Krude et al., 1998Krude et al., , 2003Lee et al., 2006). Moreover, mice lacking Pomc Yaswen et al., 1999) Lam et al., 2017;Steuernagel et al., 2022). The majority of POMC neurons could, therefore, still have a predominant role in the regulation of energy balance. In this regard, diphtheria toxin-mediated deletion of POMC cells using a Pomc-Cre model increased food intake and body weight (Gropp et al., 2005). Although part of this effect could result from unspecific targeting, this observation aligns with the obese phenotype reported in POMC-deficient individuals and underscores the involvement of POMC cells in energy balance. Since Lepr-expressing POMC neurons are predominantly involved in controlling glucose homeostasis, the reported obese phenotype of POMC DTR mice is likely to be mediated by non-Lepr-expressing POMC cells, as discussed below (Gropp et al., 2005).
Of note, ARC POMC neurons that were responsive to the 5-HT(2C) agonist m-chlorophenylpiperazine (mCPP) were not responsive to leptin, suggesting that Lepr-expressing and Htr2c-expressing POMC neurons were distinct subpopulations (Sohn et al., 2011). This was later confirmed by single-cell sequencing Lam et al., 2017). In parallel with promising clinical studies on the use of 5-HT(2C) agonists for weight management (Brashier et al., 2014;Smith et al., 2010), many studies cemented the involvement of

Htr2c-expressing POMC cells in energy homeostasis. For instance,
Pomc-Cre-mediated re-expression of Htr2c in POMC cells normalized the hyperphagic and obese phenotype of Htr2c null mice . Conversely, mice with Pomc-CreER T2 -mediated deletion of Htr2c in POMC cells had increased food intake and were more susceptible to DIO . Subsequent studies then aimed to assess the importance of POMC in mediating the effects Htr2c-expressing POMC cells also seem to be involved in glucose homeostasis (Figure 2b). Expression of Htr2c only in POMC cells is sufficient to restore normal insulin-mediated suppression of hepatic glucose production , and its deletion from POMC cells has the opposite effect . Moreover, Pomc expression in ARC (and NTS) Htr2c-expressing POMC cells is F I G U R E 2 Neuronal diversity of energy homeostasis in the arcuate nucleus of the hypothalamus. (a) LEPR-expressing POMC neurons do not mediate the direct effects of leptin on food intake and body weight. These neurons play important roles in mediating leptin's direct effects on glucose homeostasis. Other POMC neuron populations exist and mediate various effects on energy and glucose homeostasis. (b) POMC 5-HT(2C)+ can modulate food intake and body weight, as well as glucose homeostasis. (c) POMC GLP1R+ neurons regulate food intake and body weight, especially in the context of treatment with GLP1R agonists. It is still unclear whether these neurons mediate effects on glucose homeostasis. (d) AGRP neurons regulate energy balance and glucose homeostasis. (e) LEPRexpressing GABAergic neurons potently regulate food intake, energy expenditure, body weight, and glucose homeostasis. However, the precise identity of these neurons is still unknown. (f) RIP-Cre LEPR+ neurons are involved in controlling energy expenditure, body weight, and glucose homeostasis, whereas (g) NOS1 LEPR+ neurons regulate food intake, energy expenditure, body weight, and glucose homeostasis. (h) PNOC neurons regulate food intake and body weight. (i) ACBD7 neurons regulate food intake, energy expenditure, and body weight. It should be noted that GABAergic LEPR+, RIP-Cre, NOS1, PNOC, and ACBD7 are also found outside of the ARC. 5-HT(2C), 5-hydroxytryptamine receptor 2C; ACBD7, acyl-CoA-binding domain-containing 7; AGRP, agoutirelated peptide; LEPR, leptin receptor; GLP1R, glucagon-like peptide receptor 1; NOS1, nitric oxide synthase1; PNOC, prepronociceptin; POMC, pro-opiomelanocortin; RIP, rat insulin promoter; α-MSH, α-melanocytestimulating hormone. sufficient to mediate this glucoregulatory effect, along with MC4R expression in cholinergic neurons (Burke et al., 2017;Rossi et al., 2011;Sohn et al., 2013). Interestingly, 5-HT(2C) agonists can partly restore glucose sensitivity and blood insulin levels of hyperinsulinemic and obese ob/ob and DIO mice without affecting food intake or body weight (Zhou et al., 2007), suggesting that the effects of Htr2cexpressing cells on glucose metabolism can happen independently of their effects on energy balance.

| Glp1r-expressing POMC neurons
Glucagon-like peptide 1 (GLP1) is an incretin hormone that can act via its receptor (GLP1R) to induce potent glucoregulatory effects, notably by suppressing glucagon and stimulating insulin secretion in response to the ingestion of glucose (Baggio & Drucker, 2007).

Consequently, multiple GLP1R agonists have been developed
and are now used for the treatment of type 2 diabetes (T2D) (Drucker, 2018;Drucker et al., 2017;Drucker & Nauck, 2006;Le Roux et al., 2017). GLP1 and long-acting GLP1R agonists were also shown to promote satiety, reduce food intake, and reduce body weight, leading to the use of GLP1R agonists for the treatment of obesity (Astrup et al., 2009;Blundell et al., 2017;Flint et al., 1998;Jastreboff et al., 2022;O'Neil et al., 2018;Pratley et al., 2018;Van Can et al., 2014;Verdich et al., 2001). Glp1r is densely expressed in brain regions involved in the control of energy homeostasis, including the ARC (Cork et al., 2015;Farkas et al., 2021;Merchenthaler et al., 1999;Shughrue et al., 1996), and studies have confirmed its expression in POMC neurons (Péterfi et al., 2021;Secher et al., 2014). However, these inputs do not seem sufficient to significantly alter the activity of POMC neurons (Dong et al., 2021), as opposed to the direct postsynaptic activation of GLP1R (Dong et al., 2021). Studies have shown that GLP1R-mediated depolarization of POMC cells was dependent on a mixed-cation conductance mediated by transient receptor potential cation channel C5 (TRPC5) subunits (Dong et al., 2021;He et al., 2019). Deletion of Trpc5 in POMC cells using a Pomc-CreER T2 -inducible model altered chronic, but not acute, GLP1R agonist-induced reduction of food intake and body weight, suggesting that Glp1r-expressing POMC neurons are required to mediate the full anti-obesity effects of GLP1R agonists (He et al., 2019).
In line with these observations, antagonizing ARC GLP1R blunted the weight loss induced by peripheral injections of liraglutide, a GLP1R agonist (Secher et al., 2014). Physiologically, high fat-diet fed mice with prenatal knockdown of Glp1r in POMC cells had increased weight gain, but no effects were reported when mice were fed a normal diet (Burmeister et al., 2017).
One thing that remains unclear is the proportion of Glp1rexpressing POMC neurons that co-express Lepr. While He and colleagues suggested that all POMC neurons responsive to GLP1R agonists also express Lepr (He et al., 2019), other recent studies using different approaches reported that the majority of ARC POMC cells express either Lepr or Glp1r (Biglari et al., 2021;Lam et al., 2017), with only a minority of cells co-expressing these receptors (~10%) (Biglari et al., 2021). Using a novel Cre-Dre intersectional recombinase system, a recent study allowed the selective expression of DREADDs in Glp1r-expressing POMC cells (Biglari et al., 2021). Strikingly, chemogenetic activation of Glp1r-expressing POMC cells generated a robust feeding suppression in male but not female mice, while the activation of Lepr-expressing POMC cells had no effect (Biglari et al., 2021). POMC Glp1r and POMC Lepr cells were also shown to have distinct distributions in the ARC, with POMC Glp1r being more caudal and POMC Lepr more rostral (Biglari et al., 2021). Reconciliating with the study from He and colleagues, Lepr-expressing neurons activated by GLP1R agonists were localized between the RCA and the rostral ARC (bregma −0.94 mm to −1.70 mm) (He et al., 2019), where Lepr-expressing POMC neurons were shown to predominantly localize (Biglari et al., 2021;Williams et al., 2010). In sum, Glp1r-expressing POMC neurons seem to be required to mediate the full effects of GLP1R agonists on body weight ( Figure 2c). More research is needed to precisely characterize ARC Glp1r-expressing POMC neurons and their physiological significance.

| Other subsets of POMC neurons
As discussed throughout this review, ARC POMC neurons are highly heterogeneous. Recent transcriptomic studies revealed that the functional and electrophysiological heterogeneity of POMC neurons extends to differences in gene expression profile. In particular, Campbell and colleagues identified three distinct subtypes of POMC neurons based on differences in enriched transcripts , which were recently confirmed in another study (Steuernagel et al., 2022). Of these three transcriptionally distinct populations, Lepr was predominantly restricted to POMC cells expressing high levels of annexin A2 (Anxa2), while Htr2c was highly expressed in POMC populations enriched in transthyretin (Ttr) and GLI pathogenesis-related 1 (Glipr1). Interestingly, transcriptional responses to fasting were different between POMC populations. While Anxa2-enriched and Ttr-enriched had somewhat similar changes, the Glipr1-enriched subset had a markedly different transcriptional changes , suggesting heterogeneity in the response of POMC neurons to energy availability. Using similar approaches, another study identified four clusters of POMC neurons with many overlapping genes, suggesting that it is the combined expression of many genes, rather than a single gene, that characterize the heterogeneity of POMC cells (Lam et al., 2017). Therefore, additional work is needed to better understand the transcriptional segregation of POMC neurons, particularly from an energy balance point of view.  (Vong et al., 2011). Based on these results, a recent study took a complementary approach to determine whether Lepr re-expression only in GABAergic neurons (Slc32a1-Cre::Lepr TB/TB ) was sufficient to normalize the severe obesity phenotype of global Lepr KO mice (Quaresma et al., 2021). Lepr expression exclusively in GABAergic cells induced a drastic reduction in body weight and fat mass compared to mice globally deficient for Lepr-b (Lepr TB/TB ) but could not quite normalize these parameters to control levels. In fact, Lepr expression in GABAergic neurons was sufficient to completely normalize food intake, but only partially restored the reduced oxygen consumption of Lepr TB/TB mice. Plasma leptin, insulin, and glucose levels were also not completely rescued (Quaresma et al., 2021).

| GABAergic neurons mediate leptin's actions
This partial recovery suggests the involvement of many neuronal populations mediating specific effects of leptin, rather than one leptin-responsive neuronal population mediating all of leptin's effects. Alternatively, mice with disrupted GABA release from Leprexpressing neurons developed mild obesity, resulting from increased food intake and reduced oxygen consumption (Xu et al., 2012). Moreover, these mice had blunted response to leptin administration and greater sensitivity to DIO (Xu et al., 2012). Together, these studies highlight the importance of GABAergic neurons in mediating leptin's metabolic actions.

| AGRP neurons: Only one piece of the puzzle
One logical candidate to mediate the potent effects on energy balance underlined in the latter studies are the ARC AGRP neurons ( Figure 2d). Indeed, AGRP neurons are GABAergic (Horvath et al., 1997), have potent effects on energy homeostasis (Aponte et al., 2011;Bewick et al., 2005;Gropp et al., 2005;Krashes et al., 2011), express Lepr (Håkansson et al., 1998;Mercer, Hoggard, Williams, Lawrence, Hannah, Morgan, et al., 1996;, and are inhibited by leptin (Baver et al., 2014;Takahashi & Cone, 2005). CRISPR-Cas9mediated genetic ablation of Lepr in AGRP neurons caused hyperphagia and resulted in substantial weight gain, to about 80% of that observed in db/db mice (Xu et al., 2018). Contrastingly, prenatal ablation of Lepr in AGRP neurons only resulted in a very modest weight gain (Van De Wall et al., 2008). While this mild phenotype might be the result of developmental compensation, it remains clear, in both studies, that leptin's actions on AGRP neurons cannot solely explain the weight gain seen in db/db animals. This suggests the implication of other leptin-responsive non-AGRP neurons. The importance of these GABAergic non-AGRP neurons in obesity development was highlighted in a recent, brilliant study led by Qinchung Tong's group (Zhu et al., 2020). Using selective expression of the bacterial sodium channel (NachBac) or the inward rectifying potassium channel (Kir2.1), the authors induced chronic activation, or inhibition, of specific neuronal populations in the ARC. Interestingly, chronic activation of either ARC AGRP or non-AGRP GABAergic neurons both led to massive obesity, suggesting that AGRP neurons are not required for the development of obesity. Reinforcing this idea, diphtheria toxin-mediated deletion of AGRP neurons in ob/ob mice had no effect on obesity development and did not impair leptin's anorexigenic effects (Zhu et al., 2020). Strikingly, while chronic inhibition of ARC AGRP neurons had no effect in ob/ob mice, inhibition of all ARC GABAergic neurons almost completely reversed obesity development through reduced food intake and increased oxygen consumption and locomotor activity (Zhu et al., 2020). In sum, while AGRP neurons are definitely important for energy balance regulation, this study suggests that other yet unknown ARC GABAergic neurons can robustly drive obesity development and mediate leptin's anorexigenic effects (Figure 2e).

| A quest to identify the unknown Leprexpressing GABAergic neuronal populations
While evidence presented above implicates non-AGRP ARC GABAergic neurons in the central control of energy balance, the identity of the GABAergic populations responsible for these effects is still elusive. One interesting candidate is the GABAergic rat insulin-2 promotor (RIP)-Cre neurons. RIP-Cre mice express the Cre-recombinase under the control of the rat insulin-2 (Ins2) promoter (RIP) which drives recombination in pancreatic β-cells and in specific brain regions such as the ARC (Gannon et al., 2000;Postic et al., 1999;Song et al., 2010;Wicksteed et al., 2010). In the hypothalamus, leptin-responsive RIP-Cre neurons are located predominantly in the ARC and are distinct from POMC and AGRP neurons Choudhury et al., 2005;Kong et al., 2012).
Early on, it was shown that deleting motifs essential for Lepr signaling in RIP-Cre expressing cells led to alterations in glucose metabolism, increased body weight, and decreased energy expenditure, but had no effect on food intake (Covey et al., 2006). To re-assess the potential role of RIP-Cre neurons in glucose homeostasis, a recent study selectively deleted or re-expressed Lepr in RIP-Cre cells (Singha et al., 2020). Authors showed that LEPR in RIP-Cre cells was required for the glucose-lowering effect of leptin in insulindeficient animals (Singha et al., 2020). However, this study could not rule out the possibility that the effects could result in part from β-cells, which express RIP-Cre and Lepr, although deletion of Lepr in these cells does not seem to induce drastic effects on glucose metabolism (Marroqui et al., 2012;Soedling et al., 2015). Bradford Lowell's group then selectively deleted Vgat from RIP-Cre cells in order to investigate the physiologic importance of GABA release from these cells (Kong et al., 2012). Although this model is thought to selectively affect GABA release from RIP-Cre neurons, reports have shown that VGAT is also minimally expressed in a small subset of pancreatic β-cells (Chessler et al., 2002;Gammelsaeter et al., 2004;Menegaz et al., 2019). RIP-Cre::Vgat flox/flox mice had increased body weight and fat mass, a marked reduction in oxygen consumption, but no change in food intake was observed. Moreover, disrupting GABA release from RIP-Cre cells had no effect on leptin-mediated reduction in food intake, but it altered leptin's ability to stimulate energy expenditure through brown adipose tissue thermogenesis (Kong et al., 2012). Therefore, GABAergic RIP-Cre neurons appear to play roles in mediating leptin's actions on glucose and energy metabolism, but they do not seem to regulate food intake or leptin's anorexigenic effect (Figure 2f).
Other potential candidates for mediating leptin's effects on energy homeostasis are the nitric oxide synthase 1 (NOS1) neurons.

NOS1 neurons are a subset of hypothalamic neurons that express
Lepr-b and are localized primarily in the ARC, DMH, and PMV (Leshan et al., 2012). Notably, a small number of NOS1 neurons, predominantly located in the ARC, are GABAergic (Chachlaki et al., 2017;Marshall et al., 2017). Lepr ablation from NOS1 neurons resulted in reduced oxygen consumption, increased food intake, and an obese phenotype similar to that of Lepr KO mice. Furthermore, mice with Lepr deletion selectively in NOS1 neurons had serum leptin, insulin, and glucose levels as high as db/db mice (Leshan et al., 2012).
Overall, this makes the NOS1 neurons a promising population for leptin-mediated control of energy homeostasis that warrants further investigation (Figure 2g).
In addition to RIP-Cre and NOS1 neurons, a substantial number of studies have allowed the identification of other leptin-sensitive GABAergic neurons, located outside of the ARC, and mediating leptin's effects on the mesolimbic dopamine system. Early studies have shown that LEPR-expressing LH neurons are GABAergic and send projections to multiple brain regions, including the VTA (Leinninger et al., 2009). Interestingly, intra-LH infusion of leptin decreased food intake and body weight in ob/ob mice (Leinninger et al., 2009), while chemogenetic activation of LEPR-expressing LH neurons decreased body weight and food intake, and increased locomotor activity and body temperature (De Vrind et al., 2019).
Leptin was also shown to regulate the activity of VTA dopaminergic neurons by acting on different neuronal populations, including VTA GABAergic neurons and the above-mentioned LEPR expressing LH neurons, ultimately leading to effects on reward processing, motivated behavior, and affecting the incentive value of food (Nieh et al., 2016;Omrani et al., 2021;Schiffino et al., 2019). Together, these studies underscore the pivotal role of multiple populations of GABAergic neurons in mediating leptin's wide range of effects on energy homeostasis. However, such neuronal populations located outside of the ARC still cannot explain the massive obesity resulting from the chronic activation of ARC GABAergic neurons (Zhu et al., 2020). Other non-POMC LEPR-expressing neuronal populations were also reported to play a role in the control of energy balance in response to leptin. These populations include, but are not limited to, steroidogenic factor-1 (Sf-1) neurons (Dhillon et al., 2006), neurotensin (Nts) neurons (Leinninger et al., 2011), prodynorphin (Pdyn) neurons (Allison et al., 2015), and growth-hormone-releasing hormone (Ghrh) neurons (Rupp et al., 2018). Lepr-b deletion in these neuronal populations induces various, relatively mild effects on energy balance, and thus, these neurons most likely do not play a central and pivotal role in mediating leptin's effects on energy homeostasis (Rupp et al., 2018).

| Indirect control of POMC neurons
While the evidence described above indicates that leptin does not act directly on POMC neurons to regulate energy homeostasis, a growing body of evidence suggests that leptin's anorexigenic effects could still be indirectly mediated by POMC neurons through LEPR-expressing GABAergic populations . This is supported by reports that have shown that ARC POMC neurons receive substantial GABA inputs and that blockade of GABA A receptors can significantly alter their calcium activity (Cowley et al., 2001;Rau & Hentges, 2019). Potential candidates involved in these effects are the ARC GABAergic AGRP neurons. Early on, AGRP neurons were shown to send GABAergic projections to POMC perikarya in the ARC (Cowley et al., 2001;Horvath et al., 1992). Importantly, leptin was shown to decrease the frequency of GABAergic inhibitory postsynaptic currents (IPSCs) arriving onto POMC cells, suggesting that GABA release from AGRP neurons could indirectly mediate leptin's effects on POMC neurons (Cowley et al., 2001).
This connection between AGRP and POMC neurons was later confirmed by modern techniques. Optogenetic stimulation of AGRP neurons induced a strong GABA-dependent inhibition of POMC neurons, but this acute inhibition did not result in any change in food intake (Atasoy et al., 2012). Conversely, hyperpolarization of AGRP neurons instantly elevated the firing frequency of POMC neurons . Brief activation of AGRP neurons also induced long-term effects on ARC POMC neurons, reducing their responsiveness to excitatory postsynaptic currents (EPSCs) long after AGRP stimulation and involved the release of GABA, AGRP, and NPY. Interestingly, these long-term effects on POMC neurons were robustly attenuated by leptin . On the other hand, AGRP neurons may not be as pivotal as it seems in mediating leptin's effects on POMC neurons. Indeed, while leptin was shown to reduce IPSC frequency in POMC neurons, Lepr deletion in AGRP neurons only minimally impaired this effect (Vong et al., 2011).
Strikingly, Lepr deletion in all GABAergic neurons completely normalized IPSC frequency in POMC neurons, suggesting a major indi-

rect action of leptin on POMC cells through non-AGRP GABAergic
Lepr-expressing cells (Vong et al., 2011). Moreover, another study clearly showed that AGRP neurons do not contribute significantly to the strong spontaneous GABAergic tone onto POMC neurons (Rau & Hentges, 2017). Providing insight on the origin of this GABA input on POMC neurons, recent work showed that altering GABA release from DMH GABAergic neurons significantly reduced spontaneous IPSCs on POMC cells, resulting in increased calcium activity in these cells and a mild, transient decrease in body weight. It is not known, however, whether these DMH GABAergic cells are sensitive to leptin (Rau & Hentges, 2019). In sum, electrophysiological data show robust presynaptic regulation of POMC neurons via non-AGRP GABAergic cells.
Evidence is clear that GABA plays an important role in energy regulation. However, leptin can induce long-lasting effects on energy balance that can hardly be explained considering the fast-acting inhibitory nature of GABA. Thus, it is likely that at least part of the long-lasting effects of leptin are mediated through neuropeptides.
Therefore, leptin-sensitive GABAergic neurons producing neuropeptides are of particular interest and are worth further investigation. One such example is a recently identified population of ARC GABAergic neurons expressing prepronociceptin (Pnoc), which were shown to mediate diet-induced hyperphagia and weight gain in mice (Jais et al., 2020) (Figure 2h). Importantly, PNOC ARC neurons form synapses with POMC neurons and can induce strong spontaneous IPSCs on these cells following optogenetic stimulation. PNOC ARC neurons' activity is robustly regulated by glucose, and a proportion of PNOC ARC neurons are inhibited by leptin. Further research is needed to better understand the importance of PNOC ARC neurons in mediating leptin's metabolic effects. Another neuropeptide worth mentioning is the nonadecaneuropeptide (NDN), resulting from the processing of acyl-CoA-binding domain-containing 7 (ACBD7).
Interestingly, Acbd7 is expressed in a subset of POMC cells as well as in some non-AGRP, GABAergic neurons in the ARC (Lanfray et al., 2016). In mice, icv injection of NDN led to a reduction of food intake and body weight, and an increased oxygen consumption, together with a significant increase in Pomc mRNA levels in the ARC.
In addition, icv injection of leptin led to a robust increase in Acbd7 mRNA and NDN peptide levels in the hypothalamus, suggesting that NDN and Acbd7-expressing neurons could be important to mediate leptin's anorexigenic effects (Lanfray et al., 2016) (Figure 2i).
In this long-lasting quest to identify the neuronal populations that are crucial for mediating leptin's effects on energy homeostasis, evidence converges toward still unknown, GABAergic Lepr-expressing neurons, mediating robust effects on food intake and body weight,

| CON CLUDING REMARK S
The leptin-melanocortin system plays critical roles in energy and glucose homeostasis. Early models suggested that leptin regulated energy balance through direct actions on POMC neurons. However, this view has drastically changed with the evidence that POMC neurons are highly heterogenous, together with the development of new technologies allowing more specific targeting of subsets of neurons. As we discussed in this review, we now appreciate that POMC neurons are direct targets of leptin for the control of glucose metabolism but not food intake (Figures 2 and 3) is it connected to? The development of new technologies allowing spatiotemporal, intersectional, chemogenetic and optogenetic manipulations of genes, pathways, and neurons is greatly improving and reconciling decades of data, but conclusions need to consider limitations and caveats for these new tools. Therefore, more work is needed in order to fully unravel the complexity of neuronal expression profiles, projections, and functions involved in the central regulation of energy homeostasis.
Another important question persists: what is the physiological function of the central leptin-melanocortin system? Although this system was discovered from studies of very obese mice and always viewed as a promising target for the treatment of obesity, evidence accumulates showing that this system plays critical roles in metabolic and neuroendocrine adaptations to starvation aimed at promoting survival (Ahima et al., 1996(Ahima et al., , 1999. As initially hypothesized by Dr. Jeffrey Flier, rising leptin may not be a critical metabolic signal, but rather, falling leptin levels during fasting might represent the true physiologic function of this hormone (Ahima et al., 1996(Ahima et al., , 1999Flier, 1998;Flier & Maratos-Flier, 2017). This concept is nicely depicted by ob/ob and db/db mouse models, where the inability to produce or sense leptin conveys an artificial starvation signal, leading to the development of massive obesity. Conversely, while reduction in leptin levels increases body weight and drives food seeking behaviors (Exner et al., 2000;Verhagen et al., 2011), increased leptin levels generate marginal metabolic adaptations that rapidly reach a plateau (Ravussin et al., 2014;Speakman & Elmquist, 2022). From an evolutionary perspective, obesity is not an acutely dangerous state, but starvation is, and therefore, the lack of leptin may be more critical than its surplus. Supporting this hypothesis, leptin transport through the brain is highly inefficient when circulating leptin levels are elevated. In fact, higher serum leptin levels correspond to even lower CSF/serum ratios (Banks, 2021). This suggests that the leptin transport system in the CNS is not adapted to convey the very high levels of leptin observed in obesity. As such, leptin in higher concentrations could be important for peripheral signaling, but low levels during starvation more important for the central control. Considering leptin as a starvation signal represents an important consideration regarding the experimental design of future studies studying leptin's biology. Indeed, it appears more appropriate to consider experiments where leptin is reduced or removed rather than elevated or added, which is common practice in the field. There is still limited knowledge on how the leptin-melanocortin system is regulated during F I G U R E 4 Summary of the different ways leptin can indirectly affect POMC neurons. POMC neurons of the ARC receive substantial GABA inputs from AGRP neurons and other GABAergic populations that are sensitive to leptin. They could also be influenced by pronociceptin (NOP) and nonadecaneuropeptide (NDN) released from PNOC and ADBD7 neurons, respectively, in response to leptin. ACBD7, acyl-CoA-binding domain-containing 7; AGRP, agouti-related peptide; DMH, dorsomedial hypothalamus; PNOC, prepronociceptin; POMC, pro-opiomelanocortin.
fasting. We hope that this discussion will help lifting many assumptions in the field and serve as a first step in moving forward to better understand the central leptin-melanocortin system.

AUTH O R CO NTR I B UTI O N S
Olivier Lavoie wrote the manuscript and prepared the figures.
Olivier Lavoie, Natalie Jane Michael and Alexandre Caron researched, discussed, reviewed, and edited the manuscript before submission.

ACK N OWLED G M ENTS
The authors apologize to all colleagues whose work could not be cited owing to space limitations. This work was supported by fund-

CO N FLI C T O F I NTE R E S T S TATE M E NT
The authors declare no competing interests.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data sharing not applicable to this article as no datasets were generated or analysed during the current study.