Feeding is primarily a response to habitually entrained rhythms, including circadian rhythms. Feeding behaviour is regulated by a system with the hypothalamus at the centre, where how much we eat is a response to an internal energy status 12. There are complex but integrated interconnections between the hypothalamic nuclei that maintain energy homeostasis through regulation of food intake and energy expenditure 13141516. One of the key components of the hypothalamic system is the POMC neurones of the arcuate nucleus 1011. One of the main longer term peripheral signals is the hormone leptin 1011.

The site of leptin's action is the mediobasal hypothalamus, principally the hypothalamic arcuate nucleus, via the long form of the leptin receptor (LEPRB) that influences the activities of two separate groups of neurones with opposing feeding functions. These include POMC neurones that co-express CART and AgRP neurones that co-express NPY. These arcuate POMC neurones have processes close to the fenestrated capillaries of the median eminence and can thus be targeted by hormones such as leptin in the circulation 17. POMC neurones in the arcuate nucleus, which express LEPRB, are activated by leptin. They project to the DMH, PVN and lateral hypothalamus 1011. Leptin may thus be the signal linking peripheral energy stores with POMC signalling activity in the hypothalamus. LEPRB is also expressed in many other hypothalamic nuclei that may have a role in appetite. These include the DMH, VMH, PVN, lateral hypothalamic area, periventricular nucleus and SON 1819. It must be remembered that not all mammalian hypothalamic POMC neurones express leptin receptors 2021, suggesting the existence of a leptin-unrelated melanocortin signalling system too.

POMC deficiency in mice and humans and PC1 homozygous mutations in humans are characterised by adrenal failure, early-onset obesity and, with POMC mutations, altered pigmentation and tall stature 7938. Similarly, cpe ko mice show obesity, as well as a number of other endocrine defects 27. Subjects with the human genetic obesity disorder, Prader-Willi syndrome, have reduced levels of PC2 in their hypothalami on immunocytochemical assessment 39.

A point mutation in the cleavage site between β-MSH and β-endorphin (Figure 1) forms an aberrant fusion protein, which causes obesity in humans, perhaps by altering central melanocortin signalling in the CNS 40. Heterozygous mutations in MC4R also cause obesity and tall stature in humans 7. Recent data suggests that β-MSH may be the important POMC product in inhibiting feeding centrally in humans 41, but not rodents, as rodents do not produce β-MSH physiologically 742.

Mendelian human POMC disorders are very rare 43, but there is linkage between POMC and other obesity related traits, suggesting that it may form part of a common genetic obesity predisposition 4445. These findings suggest a clinically relevant role for POMC within the feeding circuits in the CNS.

There are two main neuronal populations located in the arcuate nucleus regulating appetite. There is evidence from both gene and protein expression studies that the satiety neurones produce both POMC and cocaine and amphetamine regulated transcript (CART) 48. Food restriction reduces hypothalamic pomc mRNA expression 49, whereas hypothalamic pomc mRNA expression is increased in overfed rats 50. The orexigenic neurones contain NPY and agouti-related peptide (AgRP). Fasting activates mRNA expression in hypothalamic NPY neurones 51. NPY is thought to be one of the strongest stimuli to feeding 51 perhaps partly acting by inhibiting arcuate nucleus pomc mRNA expression via the Y₂ receptor 52. The major function of AgRP is to stimulate feeding by antagonising melanocortins at the MC3R and MC4R in the hypothalamus 53. Selective ablation of NPY/AgRP-expressing neurones in adult mice results in acute reduction of feeding 54. AgRP polymorphisms are associated with inherited leanness in humans 55. In the parvocellular division of the paraventricular nucleus, close apposition of the perikarya of POMC/CART neurones and NPY/AgRP-containing terminals is seen. CART arcuate nucleus neurones markedly inhibit NPY-induced feeding in fasted and normal rats 56.

Both the POMC/CART and NPY/AgRP pathways express LEPRB and project to the leptin-dependent regions in the DMH, PVN and lateral hypothalamus (Table 1). Leptin has opposite effects on each pathway with regard to feeding, facilitating transmission in the POMC/CART system and inhibiting the NPY/AgRP neurones, when measure by a system using fluorescent protein (GFP)-tagged POMC (red) and NPY (green) neurones 57. Surgical disruption of the arcuate nucleus to PVN connection in rats results in increased food intake and weight gain, including body fat, implying that the balance of the background anorexigenic effects of POMC/CART neurones is greater than the orexigenic effects of the NPY/AgRP neurones 58. It must be born in mind that lesioning experiments are likely to disrupt many neuronal connections and their interpretation requires some caution.

Using ob/ob mice expressing GFP on both POMC and NPY neurones in the arcuate nucleus, leptin-deficiency has been shown to increase both the number of excitatory synapses to NPY/AgRP neurones and the number of inhibitory synapses to POMC/CART neurones. When leptin is given to the ob/ob mice, synaptic numbers revert to the wild type pattern within 6 hours 57. Fasting leads to an increase in the action potential of the arcuate nucleus NPY/AgRP neurones in normal mice, but this is not seen in ob/ob or db/db strains lacking central leptin signalling 59. This leptin-induced neuronal plasticity may relate to its role as a signal of fat stores. Melanotan II (MTII), an MC3R/MC4R agonist reduces the orexigenic and adipogenic effects of NPY, but does not alter NPY-induced suppression of the reproductive and growth neuroendocrine axes 60. A large proportion of NPY neurones in the rat hypothalamus express mc3r mRNA while a much lower number of NPY neurones express mc4r mRNA, suggesting that POMC neurones may directly modulate the activity of the hypothalamic NPY system too, mainly through activation of MC3R 61.

A subpopulation of arcuate nucleus NPY/AgRP/GABA-producing neurones, identified immunocytochemically, project to the PVN and send inhibitory GABA collaterals to arcuate nucleus POMC neurones that express GFP 62. GABA blocks the anorexic effect of icv α-MSH, whereas a GABA antagonist increases the anorexia 63. Thus, GABA facilitates the feeding effect of NPY at target sites in the PVN, by blocking opposing POMC transmission. However, in ob/ob mice (who congenitally lack leptin, leading to obesity), leptin deficiency exaggerates the direct inhibitory effect of NPY neurones on POMC neurones in the arcuate nucleus, an effect independent of the actions of GABA 64. Also, approximately one-third of POMC/CART neurones express glutamic acid decarboxylase (GAD) mRNA 47. GAD is the enzyme necessary to produce GABA, although GABA is not thought to inhibit the NPY/AgRP neurones and its exact function is unknown at present. It may be involved more in the extra-hypothalamic projections of POMC 65. There is recent immunocytochemical evidence that hypothalamic POMC neurones express cholinergic fibres too 66. Studies utilising immuno-electron microscopy, immunocytochemistry and in-situ hybridisation show that both the NPY/AgRP nerve fibres and the CART/POMC neurones are innervated by glutaminergic fibres too 6768. This links the control of feeding into wider feedback systems, for example glutamate neurones connect to the VMH, DMH and lateral hypothalamus, the other main feeding areas in the hypothalamus 6768.

The melanocortins have two endogenous antagonists, agouti and AgRP, that show some subtype selectivity as MCR antagonists. AgRP only acts on the CNS receptors, MC4R and MC3R (Table 2; 69). Both AgRP and the MSHs can bind to MC3R and MC4R either presynaptically or postsynaptically 1011. Like POMC, AgRP may undergo post-translational processing 70. Both agouti and AgRP function as inverse agonists in vitro, so they may regulate their respective MCRs in vivo, even in the absence of melanocortins 71.

In hamsters, central administration of AgRP increases food hoarding more than food intake 72. Does this suggest species differences in the response to AgRP, or perhaps actions other than pure antagonism of the melanocortin anorexic signal? AgRP_83–132 has been shown to have effects in the mc4r ko mouse, which could be attributable to MC3R antagonism, or perhaps as yet unknown effects on a different receptor system 73.

Perhaps AgRP neuronal activity is more directly linked to metabolic change than POMC neuronal activity? In the ob/ob mouse, agrp gene expression is increased 5–10 fold compared with wild type 7475, whereas the absence of leptin only reduces pomc gene expression by approximately two-fold 76. Others have found that there is virtually no change in hypothalamic pomc expression in either food restricted, fasted or leptin resistant (Zucker) rats versus controls 70. Overall this gives a picture of a rather tonic anorexic signal from the POMC/CART neurones, perhaps with a more finely tuned and responsive orexigenic signal from the AgRP/NPY system? AgRP, its interactions with the melanocortin system and other roles has been well reviewed recently 70.

Peptide YY (PYY) and NPY are part of the same peptide family, acting via the Y-group of receptors. After a meal, the gut releases PYY_3–36 into the circulation, which crosses the BBB and reduces food intake by suppressing NPY release and increasing POMC release in the arcuate nucleus 77787980. PYY_3–36 inhibits the action potential firing activity of arcuate POMC neurones, acting through postsynaptic Y2 receptors 81. However, mc4r ko, pomc ko and normal mice have similar anorexia with PYY, suggesting it can act independently of melanocortins too 982. This may seem paradoxical, but could relate to differing development of the hypothalamic NPY/PYY system in the absence of pomc or mc4r from birth.

VGF (non-acronymic) is a hypothalamic neuropeptide coexpressed in POMC and NPY neurones, identified by immunocytochemistry, in the arcuate nucleus 83. Lethal yellow is an inbred obese mouse strain, resulting from overexpression of agouti, which then has excess AgRP-like actions in the CNS 69. vgf ko mice are lean with normal food intake and increased oxygen consumption and locomotor activity at rest 83. Crossing the vgf ko mouse with ob/ob and lethal yellow mutants, improves the obesity of the ob/ob mouse and prevents the lethal yellow obese phenotype. This suggests that the actions of VGF are downstream of AgRP and the melanocortins 83. Arcuate nucleus vgf mRNA expression is increased in hamsters in experimental short photoperiod situations mimicking winter, suggesting a feeding-related hibernation function for VGF 84.

Cells in the arcuate nucleus, DMH and PVN show galanin immunocytochemical staining 85. When injected icv, or directly into the PVN, VMH and lateral hypothalamus, galanin is mildly orexigenic, perhaps by facilitating NPY's actions 8586. Galanin blocks arcuate neuronal firing in those neurones expressing gal-r1 receptor mRNA, perhaps via direct contact with arcuate nucleus POMC neurones 87.

LEPRB is found in several brainstem nuclei involved in the control of food intake, such as the DMX, area postrema, NTS, parabrachial, hypoglossal, trigeminal, lateral reticular and cochlear nuclei, locus coeruleus and inferior olive 94. Leptin injected into either the fourth or lateral ventricles, or into the dorsal vagal complex (DMX, AP and NTS), suppresses feeding 94. This suggests that the brainstem neurones are just as effective at mediating anorexia as the hypothalamic neurones and, like POMC, are direct targets for the action of leptin in the control of energy homeostasis 94. However, part of this effect may be being mediated by hypothalamic arcuate nucleus POMC cells, as retrograde tracing experiments showed that a small percentage of these hypothalamic neurones project to the dorsal vagal complex 91.

Also, when leptin is administered centrally it activates the arcuate POMC neurones, but not the NTS neurones 95. Recently, leptin has been shown to modulate taste sensation directly, via the taste receptors, as well as via the brainstem 96. Leptin could have a refining influence on feeding, via its actions on taste receptors and brainstem nuclei, but this is unlikely to involve brainstem POMC neurones in the process.

In contrast to the restricted distribution of the POMC neurones, MC4R and MC3R are widely present throughout the brain, although with different patterns of distribution (Table 3). The evidence, either genetic or neuropharmacological, for the key role of MC4R in feeding and the pathogenesis of obesity is as follows.

The most common monogenic forms of human obesity are MC4R mutations 7. Murine and human MC4R homozygous mutants are obese and hyperphagic 979899100. Murine and human MC4R heterozygous mutants are obese to a lesser extent. This shows sensitivity to quantitative variation in MC4R expression 97100, with mechanisms such as poor cell-surface expression or intracellular retention of the mutant receptors 101102. It is also possible that genetic defects in the intracellular trafficking mechanisms, required to present MC4R on the cell surface, could also lead to human obesity 103. Interestingly, human MC4R gene variants have also been associated with a lack of physical activity 104.

mc4r mRNA and MC4R protein is concentrated in the feeding areas of the PVN, DMH and lateral hypothalamus 90105. The obesity of the lethal yellow and viable yellow strains of mice is due to over-expressed agouti (with AgRP-like actions) in the hypothalamus, having an antagonistic effect at the MC3R and MC4R receptor 106. MC4R agonists reduce feeding in rodents, with antagonists having the opposite effect 107108. Finally, MC4R agonists administered intranasally decrease bodyweight in humans 109. Also simple behavioural change, such as scheduling meal times, can significantly reduce the anorexic effect of MC3R/MC4R agonists in rats 110. Perhaps the MC4R receptor does integrate the anorexigenic signal from leptin via the POMC neurones of the arcuate nucleus and brainstem. However, it is unlikely to be the only pathway. Double mutant ob/ob mc4r/mc4r ko mice show partial leptin resistance when compared with ob/ob strains, suggesting an alternate central anorexic signalling system for leptin exists 111.

POMC neurones in the arcuate nucleus, acting on other feeding areas such as the PVN via MC4R activation, may be the principal CNS conduit for leptin's peripheral satiety signal.

Similarly one can review the evidence, either genetic or neuropharmacological, for the function of MC3R in the pathogenesis of obesity.

mc3r ko mice are obese with increased fat mass and decreased lean body mass, but without hyperphagia, in contrast to mc4r ko mice. However, mice lacking both mc3r and mc4r are more obese than mc4r ko mice alone 112113. Also, the obesity of mc3r ko mice is more dependent on fat intake than that of the mc4r ko mice 114. Diet induced obesity in these two ko strains affects insulin-sensitivity more adversely in the mc4r ko mice 114. mc4r ko mice do not respond to the anorectic action of MTII 73. MC3R gene variants are common in humans, but they are not associated with obesity 115. However, MC3R may mediate different responses to leptin than MC4R. While leptin administration reduces food intake in mc4r ko mice, mc3r ko mice do not show an anorexic response to leptin. This suggests that the ability of leptin to reduce food consumption depends more upon MC3R, rather than MC4R 116.

The MC3R is particularly expressed on the arcuate nucleus, including on the POMC/CART neurones 117118 and the AgRP/NPY neurones 61 and is also more generally expressed in the CNS than the MC4R 90118. α-MSH, β-MSH and γ₂-MSH all activate different neuronal targets within this nucleus 42. However, it should be noted that MC4R is found on the arcuate nucleus too 90105. There are intra-arcuate POMC connections, suggesting that MC3R may mediate an autofeedback mechanism in the arcuate nucleus (Tables 1 and 3; 61118119). Administration of a specific MC3R agonist reduces the frequency of action potentials in POMC-containing neurones in the arcuate nucleus, which supports this hypothesis 62. It is not yet known if any of these observations are important with regard to feeding behaviour, but they may be important with regard to overall control of POMC neural projections. Overall, the role of MC3R in feeding behaviour and obesity is less clear than for MC4R.

α-MSH suppresses feeding in free-feeding or fasted rodents, when administered centrally, as do its synthetic analogues containing the core MCR binding sequence 42106123124. Desacetyl-α-MSH, in which the N-terminal serine remains unacetylated, is a major precursor of α-MSH and is found widely in the brain 125. Whilst not being shown to activate hypothalamic neurones in vivo 42, it binds to and activates both MC3R and MC4R in vitro and has an anorectic effect at high doses in vivo 123. Leptin facilitates the acetylation of desacetyl-α-MSH to the more active melanocortin, α-MSH 126. Perhaps a failure to acetylate desacetyl-α-MSH in mammals could lead to obesity? This has not yet, however, been demonstrated.

In radioligand binding studies, β-MSH has a higher affinity for the MC4R than α-MSH, with γ-MSH having the lowest affinity. In contrast, γ-MSH binds to the MC3R with higher affinity than either α-MSH or β-MSH 127. It should be noted that binding of a ligand to a receptor does not necessarily correlate with biological activation of the cell expressing the receptor. Two independent studies have shown that β-MSH can have a suppressive effect on feeding in rats free-feeding or fasted for 24 hours 123124, but this has not been found in one study looking at rats fasted for 48 hours 42. Perhaps this is because β-MSH is not produced physiologically in rodents 7120 and so any observed effects could be through pharmacological activation of the MC4R? 123124127 In a recent study in ko mice lacking pomc, central administration of α-MSH, β-MSH and γ-MSH can all reduce food intake, but only α-MSH actually reduces weight gain and significantly reverses the obese phenotype of the ko mouse 120. A selective β-MSH-derived peptide agonist has been shown to decrease food intake and weight gain in diet-induced obese rats, but not MC4R-deficient mice 128. This suggests that the anorectic pharmacological action of this β-MSH agonist in vivo is via the MC4R 128. Recently, several missense mutations have been identified in human β-MSH, associated with early-onset human obesity 41129, as well as a cleavage site mutation preventing processing of POMC to β-MSH, which is also associated with human obesity 40. γ₂-MSH has a delayed anorectic effect in fasted rats 42. These data are consistent with MC4R being the principal anorectic receptor in mammals, with either α-MSH or β-MSH being the principal ligands depending upon species 741120. Perhaps the delayed feeding effect of γ₂-MSH reflects autostimulation of the POMC neurones projecting to the PVN, where POMC products bind to MC4R? Alternatively, it could reflect an inhibitory action via MC3R on NPY neurones in the arcuate nucleus 61. It is possible that individual arcuate POMC projections in mammals release different combinations of peptides (MSHs or β-endorphin) at the different hypothalamic and extra-hypothalamic sites with which they synapse, as part of the complex integration of feeding behaviour in the CNS 42.

β-endorphin is another post-translational product of POMC and, together with the opioid peptides, the enkephalins and dynorphin, acts at μ- and κ-receptors to stimulate feeding 130131132133. Opioid antagonists block NPY induced feeding 134. β-endorphin ko mice gain 10–15% more body weight than wild type after puberty 135. Thus, there may be a more complementary interaction between the various POMC peptides in the regulation of feeding 136. For example, it is possible that secretory vesicles could contain both MSHs and β-endorphin, which would have antagonistic effects on feeding. Immunocytochemistry has been used to show that there are synaptic connections between POMC and enkephalin neurones in the arcuate nucleus 137. Using techniques already described in this article, the nucleus accumbens has been shown to have a POMC projection from the arcuate nucleus (Table 1; 93). It also contains endogenous opioids which mediate the positive emotional response to palatable foods such as sugar and fat. This may be their main role in appetite control, as opioid-evoked feeding is generally short lived 138.

Nesfatin-1 is a novel satiety molecule that is processed from the nucleobindin2/nesfatin precursor molecule 143144 and its mRNA is distributed in hypothalamic nuclei including the arcuate nucleus and PVN 145. Nesfatin-1 concentrations are reduced in the PVN with fasting. Central administration of leptin does not alter nucleobindin2/nesfatin mRNA expression. Also, the satiety effect of nesfatin-1 is not altered in Zucker rats with an leprb mutation and prior administration of an anti-nesfatin-1 antibody does not block leptin-induced anorexia. This implies that leptin signalling is independent of nesfatin-1 signalling. In contrast, central administration of α-MSH markedly stimulates nucleobindin2/nesfatin gene expression in the PVN 145. Prior administration of the MC3R/MC4R antagonist SHU9119 abolishes nesfatin-1-induced feeding suppression, but nesfatin-1 does not show any direct agonistic action on MC3R or MC4R. These observations suggest that nesfatin-1 signalling is involved in a leptin-independent melanocortin signalling pathway in the hypothalamus.

It is clear that the arcuate nucleus to PVN circuit is fundamental in the transmission of the satiety message in the hypothalamus.

Significant proportions of neurones activated by α-MSH, β-MSH and γ₂-MSH project outside the BBB and are thus presumed to be neuroendocrine in origin 42, such as CRF and TRH. Fasting suppresses CRF release, which is blocked by α-MSH 146 and melanocortins can induce CRF gene transcription in some PVN CRF nerves expressing MC4R 147. This suggests that CRF acts downstream of the melanocortin system. TRH release is elicited by α-MSH, whereas γ-MSH and AgRP inhibit TRH release. MC4R and MC3R agonists mimic the effect of α-MSH and γ-MSH respectively 148. This implies that hypothalamic TRH release is controlled by POMC and AgRP neurones. In the POMC ko mouse, there is reduced pituitary TSH and hypothalamic TRH production, but the thyroid gland compensates, with elevated plasma T₃ and T₄ 149. These central effects of the melanocortins on the stress and thyroid axes suggest that POMC neurones may influence peripheral metabolism, which will in turn indirectly affect feeding behaviour.

MCH causes hyperphagia, whether injected icv or directly into the arcuate nucleus, PVN and DMH. This is achieved in part by altering the balance of neuronal activity, favouring NPY/AgRP release and reducing α-MSH/CART release from arcuate neurones 152153154. For example, in hypothalamic explant experiments injection of MCH into the hypothalamus increased the production of NPY and AgRP and decreased the production of MSH and CART, as measured by radioimmunoassay 153.

Orexin expressing cells are located in the lateral, dorsal and perifornical nuclei. They innervate the arcuate nucleus, preoptic area, paraventricular nucleus of the thalamus, septal nuclei, locus coeruleus and DMX in the brainstem, as measured immunocytochemically. There is also immunocytochemical evidence that orexin neurones synapse with NPY/AgRP and POMC/CART neurones in the arcuate nucleus. Microinjection of orexins into the arcuate nucleus, PVN and lateral hypothalamus stimulates feeding 155156157. It is possible to measure the action potential of individual living arcuate POMC neurones in mouse brain slices, identified by GFP transgenic tagging. Using whole cell patch clamp recordings, orexin suppresses the spontaneous firing in these neurones, suggesting that its appetite enhancing effects include an effect in suppressing hypothalamic POMC neuronal activity 158.

Brain-derived neurotrophic factor (BDNF) and its receptor, TrkB, control neurodevelopment and synaptic plasticity. TrkB-deficient humans show learning difficulties and severe obesity 168. bdnf-deficient heterozygous mice are obese, with high leptin and insulin levels which correct with diet alone 169. bdnf mRNA or protein is expressed in the VMH, DMH and lateral hypothalamus, but not the arcuate nucleus 169170. Its expression is reduced in the VMH of lethal yellow mice, whereas administration of MTII increases BDNF expression in the VMH of wild type mice. BDNF suppresses feeding and weight gain in MC4R ko mice, so its anorexic actions lie downstream of POMC. It is unlikely that BDNF modulates either CART/POMC or NPY/AgRP neurones in the arcuate nucleus as there is no trkb receptor mRNA expressed in this nucleus 170.