Brain regulation of food intake and appetite: molecules and networks


  • This paper builds partly on presentations made at a Nobel Conference on ‘Brain Control of Feeding Behaviour’ organized at the Karolinska Institute, Stockholm, Sweden, in September 9–11, 2004.

Christian Broberger MD, PhD, Department of Neuroscience, Karolinska Institutet, 171 77 Stockholm, Sweden.
(fax: +468 33 16 92; e-mail:


In the clinic, obesity and anorexia constitute prevalent problems whose manifestations are encountered in virtually every field of medicine. However, as the command centre for regulating food intake and energy metabolism is located in the brain, the basic neuroscientist sees in the same disorders malfunctions of a model network for how integration of diverse sensory inputs leads to a coordinated behavioural, endocrine and autonomic response. The two approaches are not mutually exclusive; rather, much can be gained by combining both perspectives to understand the pathophysiology of over- and underweight. The present review summarizes recent advances in this field including the characterization of peripheral metabolic signals to the brain such as leptin, insulin, peptide YY, ghrelin and lipid mediators as well as the vagus nerve; signalling of the metabolic sensors in the brainstem and hypothalamus via, e.g. neuropeptide Y and melanocortin peptides; integration and coordination of brain-mediated responses to nutritional challenges; the organization of food intake in simple model organisms; the mechanisms underlying food reward and processing of the sensory and metabolic properties of food in the cerebral cortex; and the development of the central metabolic system, as well as its pathological regulation in cancer and infections. Finally, recent findings on the genetics of human obesity are summarized, as well as the potential for novel treatments of body weight disorders.


Few tasks executed by the brain hold greater survival value than keeping us fed and in adequate nutritional state. It is not surprising then that the central nervous system has developed a meticulously interconnected circuitry to meet this challenge. A consequence of this organization is that an energy-dense environment favours the development of obesity, whilst overcompensation may shut down the drive to feed. In today's society where evolving disease demographics and lifestyle allow for a greater diversity of metabolic phenotypes than perhaps ever before [1] disorders of both extremes of energy intake are common in health care. Obesity is increasing at an alarming rate in industrialized and developing countries alike [2] and is associated with a wealth of conditions afflicting virtually all organ systems [3, 4]. Examples diverge widely to include cholelithiasis [5], osteoarthritis [6], infertility [7], stroke [8], cutaneous infections [9], wound healing deficiencies [10], as well as a general increase in mortality [11] and social and professional stigmatization [12]. The urgency of the problem is illustrated dramatically by the previous rarity of paediatric obesity-associated type 2 diabetes, which is increasing to the point of taking over as the leading cause of childhood diabetes [13]. The opposite extreme of anorexia and hypophagia includes not only anorexia nervosa [14] but is also a common and potentially fatal complication of infections [15], malignancies [16] and ageing [17].

Unlike many other common diseases, these disorders have an obvious solution: adjusting food intake and exercise until normal body weight has been restored. However, it is no great revelation that this solution is as simple as it has repeatedly proved elusive [18]. Experimental studies confirm the common knowledge that weight loss almost always is followed by a rapid return to initial weight once the anorexigenic regimen is terminated [19]. Notably, the same applies to humans subjected to voluntary overfeeding [20], supporting the concept of a tightly regulated set-point for body weight. Treatment of eating disorders has been remarkably unsuccessful – a consequence possibly of that we are battling ancient systems maintained by ‘thrifty genes’ that favour the preservation of energy stores [21]. Available options for pharmacological therapy leave much to be desired, and compounds that have been introduced for obesity management have subsequently often been withdrawn due to intolerable side-effects [22]. The most effective obesity treatment at present is bariatric surgery, but this is a complicated procedure not without adverse effects [23]. Preventive measures have frustratingly limited effect. It has proved even more difficult to devise strategies for increasing food intake in cases of anorexia. Although some success has been reported with behavioural approaches for anorexia nervosa [24], the more common forms of cancer- and inflammation-associated anorexia remain a major therapeutic challenge. Novel treatments are greatly needed.

But what systems should such treatments target? Early clinical observations that patients with pituitary tumours and accompanying injury to the base of the brain develop obesity [25–28], inspired experimental lesion studies [29–33], which demonstrated that damage to particular regions of the hypothalamus and brainstem lead to profound, often fatal, alterations of feeding behaviour. It also became apparent that signals from the peripheral energy stores [34] and gastrointestinal canal [35] provide essential cues for appetite and satiety. Based on these and other findings, Stellar [36] half a century ago proposed a dual centre hypothesis for the initiation of motivated behaviour. The hypothesis included both mechanisms for sensing peripheral cues, separate nuclei (i.e. the ‘dual centres’) for stimulating and inhibiting behaviour, and connections between the hypothalamus and higher brain regions to allow for internal states to determine the threshold for initiating behaviour. Of all motivated behaviours, the model is perhaps most applicable for food intake. Yet, for all its heuristic value, the dual centre hypothesis was as low on specifics as it was laden with theory. Research dating in particular from the last decade has changed this. Today, we have an understanding of the circuitry and neuropharmacology of feeding behaviour that can be probed for therapeutic targets. The present article is not an exhaustive review of the central control of energy metabolism [37, 38], but summarizes recent advances, which have brought new insight into the peripheral signals describing the metabolic state to the brain; the input stations in the hypothalamus and brainstem sensitive to these signals, the organization of feeding behaviour in simple and complex organisms; the link between food intake and energy expenditure; the neural framework for integrating metabolic cues and reward properties; the mechanisms of infection- and cancer-associated anorexia; developmental and genetic causes of obesity and novel therapeutic strategies.

A central framework for sensing and orchestrating energy metabolism

The regulation of energy metabolism presents a prototypical homeostatic system, with the brain acting as the central coordinator and rectifier (Fig. 1). It is one of the great wonders of the brain that body weight stays remarkably fixed (as a ‘set-point’) most of the time in most people [39]. The first step in maintaining this homeostasis is for the brain to inform itself of the metabolic status of the individual, which it does through two main channels. First, hormonal signals reflecting the availability and demand for metabolic fuel is relayed via neurones in the hypothalamus. The receptors for these signals are primarily expressed on two neurochemically distinct sets of neurones located in the arcuate nucleus (Arc) in the mediobasal hypothalamus, alongside the third ventricle [40]. One neurone group expresses neuropeptide Y (NPY); increasing NPY release or activation of these neurones by a variety of approaches results in increased food intake and decreased energy expenditure. The other group expresses the neuropeptide precursor pro-opiomelanocortin (POMC), which is processed to melanocortin peptides; activation of these neurones has the opposite effect of triggering the NPY cells, i.e. decreased food intake and increased energy expenditure. The yin–yang relationship between the two Arc groups is further underscored by their opposite regulation by leptin and insulin, hormones signalling metabolic affluence, which decrease the expression of NPY, whilst they increase the expression of POMC. The second main input for information pertaining to energy balance is the brainstem, classically viewed as a channel for visceroceptive information as cranial nerves, in particular the vagus nerve, carrying information from the alimentary organs enter the brain here. Vagal afferents synapse onto [41, 42] and excite [43] neurones in the brainstem nucleus tractus solitarii (nTS). From both the hypothalamus and the brainstem, projections fan further into the brain to engage other brain regions in the initiation and organization of food intake. As in all homeostatic systems, the brain has at its disposal three effector pathways to activate when the controlled variable (i.e. body weight) needs to be adjusted: behaviour (i.e. food intake), the endocrine system and the autonomic nervous system [44]. Importantly, all three of these systems are engaged downstream of the Arc and nTS stations to provide a synchronized response to fluctuations in energy balance; the first primarily in the volitional control of intake, the latter two regulate energy expenditure.

Figure 1.

The central metabolic circuitry is regulated by numerous endocrine and neural inputs. Schematic illustration of how brain networks regulating ingestive behaviour communicate with peripheral organs. Hormones supplying information about the peripheral metabolic state to the brain include the gastrointestinal peptides ghrelin and PYY(3-36), insulin from the pancreas and leptin from adipose tissue. Ghrelin and leptin act both on the hypothalamus (Arc) and the brainstem (nTS). The afferent portion of the vagus nerve innervates most of the gastrointestinal tract where it collects information about the immediate alimentary state, and terminates in the nTS. The lipid mediator OEA is produced in the duodenum and activates the brainstem, possibly via the vagus nerve. Both the Arc (via antagonistic NPY- and POMC-expressing cells) and the nTS project further into the brain in parallel pathways to engage higher brain regions into ingestive behaviour. Outputs from the brain regulating energy expenditure include both branches of the autonomic nervous system; the sympathetic system whose preganglionic neurones are located in the intermediolateral cell column (IML), which is directly innervated by POMC neurones from the Arc, as well as the parasympathetic system with preganglionic neurones for the efferent portion of the vagus nerve located in the dorsal motor nucleus of the vagus (DMX). The efferent autonomic innervation regulates, e.g. glucose homeostasis via actions in liver and skeletal muscle. See text for details.

Peripheral control of feeding behaviour

Metabolic state is reflected in a diverse array of signals of the brain. Recent investigations have shed light on some of the key hormones and the vagal mechanisms that shape the central response to nutritional challenges (Fig. 1).

The vagus nerve

The gastrointestinal canal is equipped with a myriad of sensory receptors along its full crown-rump extension [45]. Thus, the taste, texture and mechanic stress of food is reported to the brain to provide an online description of the immediate alimentary state. This information is routed to the nTS primarily via the afferent portion of the vagus nerve, so that vagal activation causes satiation and meal termination. (Parallel neural feedback is also supplied by sensory neurones innervating the oral cavity mediating, e.g. taste, and lesser-studied splanchnic nerves [46].) Vagal afferents are broadly sensitive to gastrointestinal signals, including gastroduodenal distension, the presence of chemically distinct nutrients within the gastrointestinal tract as well as peptides produced by endocrine cells in the gut wall, most prominently cholecystokinin (CCK), a well-characterized satiety signal [47, 48]. Importantly, these signals are integrated within the individual vagal sensory neurone prior to the signal being relayed in the nTS [49, 50]. The neurochemical identity of viscerosensory vagal neurones has remained enigmatic, but these cells likely signal via glutamate [51] and the neuropeptide cocaine- and amphetamine-regulated transcript [52], which inhibits feeding upon brainstem administration [53].


An appetite-regulating endocrine signal from fat tissue maintaining energy homeostasis had been hypothesized already with parabiosis experiments in the 1950s [34]. The seminal discovery of leptin, the adipocyte-derived protein hormone providing this signal, by Friedman and collaborators in 1994 [54] was a decisive catalyst for much of the current investigation on peripheral modulation of central networks. A little more than a decade later, leptin has been shown to modulate several aspects of energy balance through several different mechanisms and across a wide spectrum of timeframes, alerting the brain to the state of body adiposity [55] and acting as a ‘fat-o-stat’. It is now well established that the pronounced obesity in genetically leptin-deficient ob/ob mice is due to the loss of a centrally active feeding-inhibitory messenger, as restitution of the leptin signal in these animals normalizes food intake and body weight [56] (Fig. 2). Serum leptin correlates well to the size of the body fat deposit, and falls with weight loss [57]. This relationship is seen also in obesity, where the combination of hyperleptinaemia and hyperphagia has led to the suggestion that overweight is characterized by leptin resistance [58]. Central actions underlie both leptin-mediated feeding suppression as well as the extensive peripheral metabolic effects of this hormone; thus, e.g. replacement of leptin receptors selectively in the brain of ob/ob mice is sufficient to prevent hepatic steatosis [59].

Figure 2.

Genetically leptin-deficient ob/ob mice treated subcutaneously (s.c.) with saline (left) or with leptin (right). The severe obesity in these animals is abolished with leptin replacement therapy. Figure generously provided by Dr Jeffrey M. Friedman.


Insulin is well recognized as the key glucostatic regulator. Recent data demonstrate that in addition to the control of peripheral glucose uptake, this role also encompasses powerful central effects, in synergism with leptin. First, intracerebroventricular (i.c.v.) administration of insulin decreases food intake [60] via insulin receptors expressed on Arc neurones [61]. The role of insulin in feeding is complicated by the fact that the hypoglycaemia resulting from elevations in serum insulin in itself stimulates food intake. However, when blood glucose changes are compensated for, hypophagia is seen also with increases in peripheral insulin [62, 63], suggesting that the central effects of insulin are anabolic. (This secondary hypoglycaemia may also explain why the insulin secretion triggered already at the sight of a palatable-looking meal stimulates appetite (‘the cephalic phase’ [64]), an indication that direct sensory input relayed via the cortex can set off powerful appetitive mechanisms.) Secondly, and again similar to leptin, insulin also modulates peripheral energy metabolism via central effects by inhibiting liver gluconeogenesis [65, 66]. Thus, whilst the brain does not depend on insulin for glucose uptake, it is very much interested in what insulin has to say about the metabolic state of the body.

Peptide YY (3-36)

In addition to CCK, several gut-derived peptides provide alimentary feedback to the central metabolic circuits [67]. Peptide tyrosine-tyrosine (3-36) [PYY(3-36)], a member of the NPY peptide family produced by enteroendocrine cells [68], has recently been shown to act as an important feedback signal from the gut to the hypothalamus. Following a meal, PYY(3-36) is released into the circulation [69], specifically stimulated by the presence of lipids and carbohydrates in the lumen of the distal ileum and colon [70, 71]. Peripheral administration of this hormone inhibits food intake and causes weight loss [72, 73]. While some laboratories initially were unable to replicate this effect [74], this may partly be due to discrepancies in technique [75] and the results have since been independently confirmed [76, 77]. The satiety effect of PYY(3-36) is comparatively modest but, importantly, is observed also in humans, including obese patients [73, 78]. Plasma levels of PYY(3-36) rise markedly following ileal resection [79, 80], an observation that has been linked to the weight loss observed in patients undergoing this procedure (S.R. Bloom and C. Le Roux, personal communication).


Thus, the gastrointestinal-brain axis has long been viewed as a key channel subserving meal termination with CCK and PYY(3-36) as prime mediator examples. Novel findings on the hormone ghrelin (produced in stomach and small intestine epithelia [81], see [82]) are challenging this doctrine. Ghrelin is unique as the first gut hormone shown to stimulate food intake [67]. Both peripheral and central injections of ghrelin result in increased feeding as well as fat mass [83–86]. Ghrelin levels peak sharply in anticipation of a meal in humans as well as experimental animals [87], resulting in stimulation of both feeding and gastric emptying [88] through actions possibly involving the vagus nerve [89], and may thus provide a meal initiation signal. In the hypothalamus, peripherally administered ghrelin mainly activates the NPY neurones [85, 90] and antagonizing the actions of these cells inhibits the orexigenic effect of ghrelin administration [85, 91]. In contrast, the melanocortin pathway does not appear to be primarily involved [90]. Recent reports have proposed that ghrelin is synthesized also in hypothalamic neurones, but this issue remains controversial, in part due to the contradictory claims of the site of central ghrelinergic neurones and the failure to demonstrate ghrelin mRNA in the brain (cf. [92] and [93]).

Importantly, a loss of the hunger message relayed by ghrelin has been suggested as the mechanism behind the weight-reducing effects of bariatric surgery [94]. The initial rationale for gastric bypass [95] was that the procedure would produce weight loss by means of malabsorption. However, this turns out to be a transient effect due to the considerable compensatory potential of the digestive system. Nevertheless, weight loss persists, caused instead by a loss of appetite and hypophagia [96]. Concomitant with this effect, a fall in plasma ghrelin is observed following the bypass procedure, in contrast to the ghrelin increase associated with nonsurgical weight reduction, where weight relapse is common [86, 97]. (Note, however, findings that argue against such a relationship, see [67].) Furthermore, clinical data tie the hyperphagia observed in Prader–Willi syndrome to strikingly high plasma ghrelin levels [98]. These results, coupled with the discovery that elevated plasma ghrelin is a marker for future weight gain (D.E. Cummings and J. Krakoff, personal communication) indicate that interfering with ghrelin signalling offers a clinically promising approach to treating eating disorders.


A role for endogenous cannabinoids in appetite regulation has long been suspected from the carbohydrate craving observed in marijuana smoking [99]; indeed, increased appetite is a diagnostic criterion for cannabis intoxication [100]. Neuronal production of cannabinoids is widespread and these mediators play an important and general role in the modulation of synaptic transmission [101], with the orexigenic effects likely mediated via central cannabinoid CB1 receptors [102, 103]. However, the lipid family to which the cannabinoids belong also includes other members with opposite and peripheral effects on energy metabolism. Piomelli and colleagues have accumulated evidence that the fatty acid oleoylethanolamide (OEA), chemically but not pharmacologically similar to the cannabinoids, is produced in the duodenum and acts via the vagus nerve to decrease body weight through activation of the nTS [104]. OEA increases the inter-meal latency, an effect distinct from that of, e.g. CCK, which primarily decreases meal size [105]. However, changes in energy expenditure also underlie the OEA-mediated weight reduction and are especially pronounced in models of obesity, involving in particular increased fat utilization, whereas glucose homeostasis is relatively unaffected [106]. The catabolic effects are most noticeable as a slowing of body weight gain in growing rats, with OEA synthesis reduced by food deprivation and stimulated in response to increased demands on energy availability such as cold exposure [104]. The metabolic actions of OEA depend selectively and critically on genomic as well as nongenomic actions of the ubiquitous nuclear peroxisome proliferator-activated receptor-alpha (PPAR-α) [107]. These results add obesity to the growing list of potential therapeutic applications for nuclear receptor pharmaceuticals. Notably, drugs that target PPAR-α, e.g. gemfibrozil, are already in clinical use to treat hypercholesterolaemia [108].

Integration of peripheral signals in the Arc

The peripheral signals described above thus act upon the Arc (and nTS, see below) to influence the central pathways regulating energy balance. In the Arc, receptors for leptin and insulin found on NPY and POMC neurones serve to inhibit transcription of NPY [109, 55] and increase POMC mRNA levels [110–112] via differential second messenger systems [113]. It is becoming evident that insulin, leptin and other metabolically relevant hormones eventually converge not only on a common set of neurones, but indeed also on the same molecules. Recent reports highlight the role of the ATP-dependent potassium current, IK(ATP), as a molecular target mediating rapid, electrophysiological effects, of peripheral hormones. This K+ conductance is a priori sensitive to the availability of metabolic fuel as a fall in intracellular levels of the energy donor ATP causes the channel to open, leading to K+ influx and hyperpolarization; this mechanism enables neurones expressing IK(ATP) to vary their excitability in response to changes in glucose concentration [114]. Leptin and insulin both hyperpolarize Arc neurones by enhancing IK(ATP) [115, 116], by activating a common enzyme, phosphoinositide 3 (PI3) kinase [116, 117]. It should be emphasized that the transmitter phenotype of Arc neurones expressing IK(ATP) is a controversial issue which remains to be conclusively resolved [118–120]. Additional signals likely weigh in on IK(ATP); this current is augmented when the concentration of fatty acid derivatives is increased locally within the Arc by inhibition of lipid oxidation, a message of energy surplus that also decreases food intake [66, 121]. This convergence of nutrient information makes the PI3-kinase/IK(ATP) a key integration node within the metabolic signalling chain, attractive as a therapeutic target.

Modulation of the membrane potential of Arc neurones has recently been demonstrated to control glucose homeostasis. Opening of Arc K(ATP) channels via either hyperinsulinaemia or central inhibition of lipid oxidation inhibits vagal efferent (i.e. parasympathetic) gluconeogenic signals to the liver, promoting the use of fat as metabolic fuel [66, 122]. The Arc is also the site of central leptin regulation of glucose homeostasis as selective restoration of Arc leptin receptor expression in otherwise leptin receptor-deficient mice is sufficient to correct their hyperglycaemia [123]. These results show that insulin modulates glucose homeostasis by independent peripheral and central mechanisms and emphasize that interconnectivity within brain metabolic regions serve to switch the body between different fuel sources, in parallel to controlling food intake. Interestingly, in obese rats, hypothalamic IK(ATP) channels fail to respond to leptin and insulin [115, 116]. Whether similar defects underlie insulin and/or leptin resistance in human diabetes and obesity is an interesting possibility, which remains to be investigated.

Output from the Arc

NPY neurones.  Neuropeptide Y is one of the most potent stimulators of feeding known [124], an effect that has been confirmed by various approaches [40]. While there is conflicting data on whether deletion of the NPY gene produces hypophagia (cf. [125] and [126]), the obesity of ob/ob mice is attenuated when combined with an NPY−/− genotype [127], suggesting that NPY is an important mediator of central leptin signalling. Stimulation of feeding appears to be transduced predominantly via postsynaptic NPY Y1 receptors, as determined from pharmacological and genetic engineering studies (reviewed in [128], Fig. 3a). However, the synergistic actions of multiple NPY receptor subtypes participate to produce orexigenic effects in vivo [129]. Detailed behavioural analysis of those effects suggests that NPY primarily stimulates appetitive rather than consummatory behaviour [130].

Figure 3.

Expression of NPY and melanocortin receptors in the mouse brain. In situ hybridization histochemistry (a) shows the distribution of NPY Y1 receptor mRNA detected as silver grains in a coronal section, revealing dense expression in the cerebral cortex and nuclei in the amygdala, thalamus and hypothalamus. In panel b, green fluorescent protein (GFP) is expressed in a neurone under control of the melanocortin (MC) 4 receptor promoter; note strong immunoreactivity throughout cell soma and dendrites. A Nissl-stained coronal section (c) shows neurones clustered to form the paraventricular hypothalamic nucleus (PVH) alongside the third ventricle. The PVH constitutes a central integrative ‘hub’ within the metabolic circuitry. (d) Immunohistochemical for GFP (indicating the presence of the MC4 receptor) and in situ hybridization for NPY Y1 receptor mRNA have been combined in a section from the amygdala, revealing the coexistence of these receptors in neurones downstream of the Arc. Figure produced by Drs Toshiro Kishi and Joel K. Elmquist. Reprinted with permission from Macmillan Publishers Ltd.; Molecular Psychiatry 2005;10:132–146.

POMC neurones.  Pro-opiomelanocortin is a large precursor protein which gives rise to several bioactive peptides. Among these, the melanocortin peptides, in particular α- and γ-melanocyte-stimulating hormone, have been shown to exert potent anorexigenic effects when administered i.c.v. [131, 132]. Central melanocortin effects are mediated by the melanocortin 3 and 4 receptors (MC3R and MC4R, respectively; Fig. 3b). Deletion of the genes for either POMC, MC3R or MC4R result in obesity in mice, suggesting that the melanocortin system is crucial in maintaining body weight [133–135] – as supported by similar findings in humans (see below). MC4R−/− mice also increase their feeding in response to a high fat diet, in contrast to wild-type littermates where a reduction is seen and ob/ob mice, which maintain the same intake as with regular chow [136], underlining the importance of the melanocortin system for adjusting food intake in response to caloric variations. In addition, the hallmark hypophagia seen in disease models as diverse as renal failure, immunological challenge with lipopolysaccharide (LPS) and tumour implants is ablated. The obesity in MC4R-deficient animals is partly due to changes in energy expenditure, such as deficient diet-induced thermogenesis [136]. The anatomical substrate for this effect may be a direct projection from the POMC neurones in the Arc to the preganglionic sympathetic neurones in the spinal cord [137–139] constituting a link between the metabolic integrator and the autonomic effector system. Interestingly, the spinal projection sets the POMC neurones apart from the neighbouring NPY neurones which otherwise exhibit very parallel innervation patterns. However, it should be pointed out, that in humans, the melanocortin system appears to be more geared towards regulating feeding behaviour, with a proportionately smaller role in peripheral metabolism [140].

NPY–POMC interactions.  Interactions between the Arc populations allow the NPY neurones to control the activity of the POMC cells via two mechanisms. First, NPY neurones coexpress agouti gene-related peptide (AGRP), an endogenous melanocortin antagonist [141–143]. Thus, at the axon terminal, melanocortin action can be blocked by simultaneous release of AGRP, and in agreement with such an arrangement, a single i.c.v. administration of AGRP causes an impressively long-lasting (one week) suppression of food intake [144]. Secondly, at the cell body level, POMC neurones are innervated by NPY-ergic terminals [145] and express the Y1 receptor [146], through which NPY causes a powerful membrane potential hyperpolarization (i.e. inhibition) [147]. Surprisingly, no reciprocal innervation has yet been described and Roseberry et al. [147] did not detect any changes in the electrical properties of NPY neurones using a melanocortin analogue. Thus, there may exist an asymmetrical interaction in the Arc favouring the orexigenic NPY/AGRP message over anorexigenic melanocortin signalling. However, an inhibitory influence over the NPY neurones may be provided by PYY (3-36), which is a selective agonist of the inhibitory Y2 autoreceptors [148] expressed by these cells [146]. Such gastrointestinal negative feedback has been proposed as the mechanism whereby PYY(3-36) inhibits feeding as no such effect is observed in mice genetically deficient for the Y2 receptor and application of the peptide inhibits the electrical activity of Arc NPY terminals [73]. This effect is relatively selective as disruption of other relevant metabolic pathways does not affect the satiety effect; the PYY(3-36) effect persists both after vagotomy and in MC4−/− mice [76], suggesting that neither the nTS nor the Arc POMC neurones are directly involved.

Classical transmitters: glutamate and GABA.  While much of the current research on the central regulation of energy balance focuses on the role of peptides, it should be emphasized that in the hypothalamus, as in the rest of the brain, the key chemical mode of communication between neurones is via amino acid transmitters, i.e. excitatory glutamate and inhibitory γ-amino butyric acid (GABA). Indeed, in the absence of glutamate and GABA-mediated transmission, little remains of hypothalamic synaptic activity [149, 150]. The major function of peptides, in addition to their genomic effects, is likely to modulate the synaptic transmission of ‘classical’ transmitters [151] with which they coexist [152]. Interestingly, glutamate N-methyl-d-aspartate (NMDA) receptors have been found to stimulate feeding with remarkable anatomical specificity within the lateral hypothalamic area (LHA), in comparison with other hypothalamic regions tested and the amygdala [153]. Infusion of NMDA antagonists locally within the LHA blocks both agonist-induced and deprivation-induced food intake, indicating the involvement of endogenous glutamatergic tone in natural feeding [154]. Histochemical studies suggest that within the Arc, NPY neurones largely contain GABA, whereas POMC neurones signal via glutamate [155, 156].

Downstream targets of the Arc.  The downstream cellular effects of NPY are still mysterious. It was initially assumed that ‘feeding-promoting neurones’ in loci sensitive to NPY orexigenesis were excited by NPY. However, all known members of the NPY receptor family couple to inhibitory second messenger systems [128]. Electrical excitation has been proposed to come about in the form of disinhibition via NPY-mediated suppression of GABA-dependent inhibitory postsynaptic currents [157, 158], with melanocortin stimulation producing the opposite result, i.e. inhibition via stimulation of GABA release [157]. However, that does not explain the role of postsynaptic Y1 receptors, which exist throughout the hypothalamus [146, 159, 160] (Fig. 3a). The most potent orexigenic effects of NPY are seen within the perifornical region/LHA [124], where NPY/AGRP-ergic terminals appear to target two separate populations of neurones expressing the neuropeptides hypocretin (Hcrt; also known as orexin) and melanin-concentrating hormone (MCH; Fig. 4, [142, 161, 162]). This pathway is of interest as Hcrt and MCH potently modulate wakefulness [163–167], providing a means for metabolic signals to control arousal state. Surprisingly, in a recent investigation of the electrophysiology of the LHA neurones, melanocortin stimulation did not affect the electrical properties of MCH-expressing cells, whereas both these and Hcrt-expressing cells were inhibited by NPY [168, 169]. Furthermore, microinjection of NPY into the LHA appears to activate a group of neurones distinct from those expressing Hcrt or MCH [170]. As with all neural interactions, it is important to bear in mind that the activity of neurones can be influenced via several independent mechanisms, including (but not exclusively) electrical and transcriptional effects, and that different changes proceed along different temporal scales. Thus, a functional Arc-LHA pathway cannot be excluded. Nevertheless, these data invite a re-evaluation of the role of Hcrt and MCH as downstream mediators of the NPY and POMC neurones.

Figure 4.

Integration in higher brain regions determines the central response to changes in peripheral metabolic state. Schematic illustration of connections between brain regions responsible for coordinating the behavioural somatomotor (i.e. food intake), autonomic and endocrine (the latter two regulating energy expenditure) responses that together constitute the motivated ingestive behaviour used by the nervous system to meet nutritional challenges. The antagonistic orexigenic NPY and anorexigenic POMC neurones in the Arc project in parallel paths to numerous subcortical nuclei [including the bed nucleus of the stria terminals (BST), the medial preoptic area (MPO), the paraventricular nucleus of the thalamus (PVT), several hypothalamic nuclei, e.g. the dorsomedial nucleus (DMH), the amygdala (Amgdl), the serotonin-containing system in the raphe nuclei, the periacqueductal grey area (PAG) and the parabrachial nucleus (PBN)] distributed throughout the brain. A projection to neurones expressing melanin-concentrating hormone (MCH) or hypocretin (Hcrt) in the lateral hypothalamic area (LHA) provides an indirect pathway to the cerebral cortex for metabolic signals relayed via the Arc. The cortex in turn projects back heavily to both the LHA and other feeding-regulatory regions. In addition, the LHA also receives an inhibitory input from the shell of the nucleus accumbens (AcbSh), which in turn is modulated via prominent excitatory inputs from the prefrontal cortex (PFCx). Thus, the LHA is positioned to integrate both homeostatic and reward-related signals in the gating of food intake. Energy expenditure is modulated via outputs from the Arc to neuroendocrine neurones in the paraventricular hypothalamic nucleus (PVH), which control release of, e.g. thyrotropin-releasing hormone and adrenocorticotropic hormone from the pituitary gland. Energy expenditure is also regulated by projections from POMC neurones in the Arc and descending pathways from the PVH to autonomic preganglionic neurones in, e.g. the dorsal motor nucleus of the vagus (DMX; parasympathetic) and spinal cord intermediolateral cell column (IML; sympathetic). Note that ascending projections from the brainstem, which provide parallel important metabolic inputs to the brain, have not been included in the figure. See text for details.

Circadian regulation of metabolic processes.  In addition to the various controls summarized above, metabolic processes also follow strict circadian variations – as recently underscored by the demonstration that inactivation of key genes maintaining circadian rhythmicity results in manifest metabolic syndrome in mice [171]. Thus, for example, in rats the active period of the day is immediately preceded by coordinated peaks in hepatic glucose output (via the sympathetic nervous system) and glucose uptake in striated muscle (a parasympathetic effect; see [172]). Buijs et al. [173] have investigated which brain regions are responsible for this synchronization. Using anatomical tracing they find that the chains of neurones innervating liver and muscle are separated all the way through brainstem and hypothalamus to distinct populations of preautonomic master neurones in the suprachiasmatic nucleus [173], the brain region maintaining circadian rhythmicity and entrained by direct retinal input [174, 175]. The distinct pathways originating in the suprachiasmatic nucleus are particularly interesting in conjunction with the discovery that the autonomic inputs to intra-abdominal and subcutaneous fat stores are also separate [176]. In humans, shift work [177] and sleep deprivation [178] are associated with increased adiposity, findings that have been linked to the sleep-associated peak in leptin secretion [179]. However, this anatomically separate innervation indicates that loss of periodicity in the circadian input to adipose tissue may disrupt the balance between different fat compartments leading, in turn, to manifestations of the metabolic syndrome, which is correlated to abdominal but not subcutaneous fat accumulation [180].

Contributions of hind- and forebrain to feeding regulation

As mentioned above, the brainstem provides a port for vagal and other neural sensory signals into the brain. Classical accounts of brain regulation of feeding described two systems balancing each other: the hypothalamus, monitoring the periphery for signals alerting central circuits to diminishing energy stores, and the brainstem, receiving oral and gastrointestinal information as an online signal of the amounts and qualities of the food that was being ingested. This arrangement would allow the hypothalamus to function as a long-term control orchestrating meal initiation and the brainstem served as a short-term control for meal termination. Much of our knowledge on the different contributions of the fore- and hindbrain in meal regulation comes from a lesion model developed by Grill and Norgren [33]. Disconnecting the forebrain (which includes the hypothalamus) produces a rat incapable of the motor activation necessary for normal feeding. However, if this animal – whose brainstem remains intact – is provided sucrose solution via an intraoral cannula, intake can be measured as the solution consumed until the meal is terminated as the animal lets solution drip out of the mouth. These decerebrated rats maintain the ability to terminate their meal in response to changes in gastrointestinal feedback, but are unable to compensate for variations in the caloric value of the fed solution, resulting in anorexia if the sucrose concentration is reduced. Similarly, removal of the post-oral feedback (by e.g. vagus nerve transection or gastric drainage) leads to increases in meal size as well as duration [181, 182], although there is a compensatory delay in the latency to meal initiation, possibly mediated by the hypothalamus. These results underscore the role of the Arc as a metabolic sensor. It should be pointed out that an intact Arc is not necessary for meal initiation – humans and animals with selective lesion of this region not only eat, they eat copiously [29, 183–185].

It is now becoming evident that the brainstem can integrate much the same signals as have been shown to modulate hypothalamic activity. Leptin receptors are expressed at several strategically located brainstem sites, and selective stimulation of these receptors suppresses food intake at doses comparable with those used in forebrain injections [186, 187]. Here, leptin activates the same medial region of the nTS that is stimulated by gastric distension [188], suggesting an anatomical site of integration of long- and short-term feeding controls. Likewise, melanocortin agonists can reduce feeding and body weight by brainstem mechanisms [189]. Interestingly, the neurones in the nTS mediating the viscerosensory signal may also be POMC-encoded [43, 190], a finding that puts our understanding of melanocortin-mediated meal suppression in a new light. Orexigenic effects of ghrelin are also seen with selective local administration both in the hypothalamus [84] and in the brainstem [191] (Fig. 5a,b). Finally, glucosensitive neurones have been recorded in the nTS [192]. Thus, the nTS is in no way a passive transducer of viscerosensory signals, but serves to integrate numerous indices of the animal's metabolic state. This conclusion is supported by the observation that the hyperphagia of rats that lack leptin receptors is caused by larger meal size (i.e. meal termination delay) rather than an increased number of feeding bouts [193], suggesting an action localized in the hindbrain. However, the response to CCK (i.e. brainstem satiety signalling) as well as normal meal size is restored in these animals following selective re-establishment of leptin receptor expression in the Arc. Weighed together, these data emphasize that actions of metabolically relevant hormone take place at a few, but distributed, sites in the brain contributing to a coordinated feeding response.

Figure 5.

Ghrelin increases food intake following brainstem administration. Unilateral injection of 10 pmol (but not 5) of ghrelin (black bars) into the dorsal vagal complex, including the nucleus tractus solitarii, results in a significant increase of food intake both 1.5 and 3 h after drug administration compared with vehicle (white bars), in an experiment by Faulconbridge et al. [191]. Reprinted with permission from the American Diabetes Association; Diabetes 2003;52:2260–2265.

Beyond the primary sensors: CNS integration of hunger and satiety signals

As in all matters involving the brain, behaviour begs the question: What are the pathways? For the metabolic signals to produce behaviour they need to proceed further into the brain beyond the primary sensors in the Arc and nTS and ultimately engage regions that initiate and organize behavioural, autonomic and endocrine response patterns. Histochemical studies have revealed that (i) the Arc projections diverge widely throughout the brain [194–197] including, via indirect pathways, a massive cortical innervation [198] (Fig. 4), (ii) the NPY and POMC populations project in remarkably parallel paths [196] and may converge on the same cells as supported by the widespread coexistence of Y1 and MC4 receptors [199] (Fig. 3d), and (iii) the nTS innervates largely the same nuclei as the ascending projections from the Arc [196, 200–202]. This circuitry indicates that integration between the primary metabolic sensors in the Arc and nTS is a distributed phenomenon, and it has been suggested that this arrangement allows for motivational state to be weighed into the network before reconvergence and the final decision for a proper metabolic response is made [40]. In addition to this scheme, extensive reciprocal projections connect the Arc with its target regions [203]. The functional implications of this arrangement are not clear at present. There may exist a sequential arrangement hidden within the network that has so far eluded the techniques used to investigate the system. Another alternative is that the reciprocity may produce a reverberating signal, such as the large-scale oscillations described in, e.g. thalamocortical systems [204]. Possibly, such persistent activity may be important in the triggering and maintenance of an anabolic response. The hierarchical organization of the metabolic circuits and its relationship to behaviour presents a major future scientific challenge. (For further discussion of the systems organization of energy balance regulation, the reader is referred to recent exhaustive reviews [37, 38].)

Coordinating the metabolic output

It also remains to explain how the divergence–convergence organization of the Arc/nTS projections interdigitates with the efferent networks underlying the final metabolic response. As already mentioned, motivated behaviours (classically divided into ingestive, reproductive and defensive) involve three distinct outlets: components of the autonomic and neuroendocrine systems as well as coordinating the overall behaviour of the animal [44]. Activity within these three effector systems is hypothesized to be organized by a collection of cell groups within the medial hypothalamus area, collectively termed the hypothalamic visceromotor pattern generator network [205], which subsequently recruits elements within a ‘behavioural control column’, spanning the mes- and diencephalic midline [206]. Separate groups of control column nuclei produce ingestive, reproductive and defensive behaviours, but connections between these networks allow them to interact for purposes of mutual exclusion so that only one behaviour is expressed at once [207]. The paraventricular nucleus (PVH; Fig. 3c) is a crucial control column module for ingestive behaviour. The PVH collects metabolic information from oropharyngeo- and viscerosensory receptors and humoral signals (both directly and via the Arc), and is regulated by biological rhythms via the suprachiasmatic nucleus and by the overall state of the animal as reported from the cerebral hemispheres via relays in the septum [206]. Output from the PVH, in turn, employs all three outlets described above: endocrine via neuroendocrine neurones controlling pituitary hormone release, autonomic via direct projections to preganglionic neurones in the spinal cord intermediolateral cell column (sympathetic) and the dorsal motor nucleus of the vagus (parasympathetic), and behavioural (i.e. somatomotor) via several pathways innervating the brainstem [137, 208–210] (Figs 1 and 4). The hierarchical arrangement upstream of these motor nuclei bears some similarity to the basal ganglia organization for the control of conscious movement [206]. The details of the underlying anatomy remain mysterious, but it is clear that, rather than a simple sequential organization where information flows neatly from one collection of neurones to another, we are faced with an interconnected series of ‘hubs’ that collect, integrate and disseminate information.

Food intake in lower organisms: models for the organization of behaviour

Intriguingly, an improved understanding of the organization of feeding behaviour is now coming from studies of lower organisms. Animals such as the nematode Caenorhabditis elegans and the sea slug Aplysia present several technical advantages: behaviour is easily divided into discrete sequences, the nervous system consists of a limited number of neurones whose electrical properties and interconnectivity has been characterized in detail, and, in the case of C. elegans, the genome is highly accessible for molecular manipulation. This knowledge makes it possible to understand how nature organizes physical networks to efficiently initiate, organize and terminate behaviour.

Caenorhabditis elegans

Feeding in C. elegans is polymorphic; depending on genetic background animals will feed either alone or in aggregates [211–213]. Naturally occurring variations in a single amino acid position of the neuropeptide receptor NPR-1 (for neuropeptide receptor resemblance-1) translate into either a solitary or a social feeding phenotype [212]. Activation of NPR-1, by shifting network properties, leads to activation of social feeding behaviour, and the amino acid substitution in NPR-1 determines the response to stimulation with the neuropeptide ligands flp 18 or −21 [214]. Conversely, null mutations in the npr-1 locus alter the balance in favour of solitary feeding [214]. Thus, a single gene is sufficient to redirect behaviour. Intriguingly, the predicted transmembrane domains of NPR-1 display considerable homology to mammalian NPY receptors [212], suggesting that similar molecular components underlie related behaviours throughout evolution. Recent data demonstrate that within this network an important upstream regulator of NPR-1-mediated feeding is oxygen concentration, showing how external cues reset behaviour [215, 216].


In Aplysia, a meticulously dissected network has been highly informative in characterizing the neural mechanisms underlying various behavioural components. Aplysia feeding can be initiated by sensory stimulation of the lip and is consolidated by arousal caused by the exposure to food. In parallel with the feeding central pattern generator (CPG) circuit, a network controlling arousal is triggered, which then feeds back into the CPG [217]. Termination of food intake is achieved by switching ingestion to the opposite behaviour of egestion [218]. Separate motor neurones within the CPG effectuate ingestion and egestion. Switching the balance between these neurones is elegantly accomplished by recruiting a single additional interneurone into the circuitry via an electrical synapse [219]. Feeding patterns in this social mollusc also differ substantially in the absence or presence of other animals; company is an important incentive for feeding. Pheromones secreted from other Aplysia stimulate a neurone directly driving the appetitive phase of feeding, which also excites a control neurone for the consummatory phase so that these behaviours are properly organized, leading to larger and more frequent eating bouts [220]. These studies provide information on how organization within a neural system translates into behaviour, with potentially broad implications considering the surprisingly large overlaps between human and mollusc feeding behaviour. Importantly, they may shed light on the process of selection within the behavioural repertoire and why we eat in apparent violation of homeostatic mechanisms when our fat stores are filled to the brim.

Food reward: role of the nucleus accumbens

However, to elucidate the neural organization of feeding in complex vertebrates, it is necessary to consider that our choice of food is not simply a function of energy supply and demand, but also very much linked to reward value. Homeostatic systems active within the brain operate at the mercy of motivational states. The motivational state is a product of, inter alia, limbic influences relayed in part via the amygdala, and reward factors. Obviously, adding reward experience to food is a means of, e.g. avoiding consumption of foods whose taste indicate the presence of invasive microorganisms, but also promoting those whose taste signals particular nutritional value. As a result, pairing feeding with pleasure may override normal satiety mechanisms, resulting in hyperphagia and obesity. The concept of reward is intimately linked to that of addiction, and it has been suggested that obesity is a consequence of an addiction to food [221]. Drugs of abuse converge upon the mesolimbocortical system to produce reward, specifically by enhancing dopamine release in the nucleus accumbens (Acb) of the forebrain as a final common pathway [222]. There is little doubt that changes in dopaminergic transmission affects food intake. Indeed, animals unable to produce brain dopamine die of starvation unless fed by gavage [223], and a common side-effect of neuroleptics affecting dopamine signalling is obesity [224]. However, novel data suggest that dopamine primarily acts to reinforce behaviours at the initial encounter with a novel reward, but the release of dopamine decreases once the behaviour has been established within the behavioural repertoire [225]. Thus, whilst dopamine modulates learning and locomotion associated with motivational behaviour, it appears not to modulate feeding behaviour per se; blocking Acb dopamine signalling does not alter total food consumption in starved rats, although it suppresses ambulation associated with feeding [226].

However, transmitter systems other than the dopaminergic system connect Acb to components of the metabolic circuitry [227]. Notably, chronic access to a preferred flavour (chocolate-fat solution) produces the same transmitter changes in the Acb as chronic morphine or ethanol, suggesting a common reward mechanism for palatable food and conventional drugs of abuse [228]. These changes include increased transcription of opioid peptides such as enkephalin. In turn, stimulating μ-opioid receptors in the Acb increases intake of fat-enriched foods with high palatability value [229], which may be interpreted as positive reinforcement. This effect is not seen following inhibition of neural transmission in the basolateral amygdala (BLA) or the LHA [230]. The connection between the BLA and forebrain cortical regions has been implicated in gauging food palatability, and an intact BLA is required for determining the reward value embedded in sensory input [231]. In contrast, the central amygdala (CeA) serves more like a general gatekeeper of feeding as inactivation of this subnucleus blocks consumption of all foods [230]. This dichotomy may be explained by the different projections of the BLA (innervating higher forebrain regions such as the prefrontal cortex) and the CeA (aimed towards the postulated hypothalamic and brainstem feeding pattern generators). The connection between the Acb and the LHA has also been shown to modulate food intake (Fig. 4). Kelley and colleagues have shown that blocking excitation of the GABA-encoded Acb results in hyperphagia, an effect contingent upon intact transmission in the LHA [232]. Glutamatergic excitation in the Acb is predominantly supplied by the cortex, so that this cortex-Acb-LHA pathway could offer the prefrontal cortex in particular a channel for inhibiting feeding behaviour in favour of other behaviours [227]. These data indicate that in the regulation of ingestive behaviour, the Acb is indeed at the interface between ‘motivation and action’ as originally postulated for this structure [233].

Studies on the food intake experience in humans and monkeys: relevance for understanding obesity

Defining the cortical involvement in feeding behaviour is likely vital for understanding human eating disorders. Crosstalk between the cerebral cortex and primary sensors is extensive; impressively, the hypothalamus provides the largest external cortical input with the exception of the thalamus [198], and reciprocal connections are plenteous [234]. Investigations of monkeys and humans by Rolls and collaborators have revealed that the sensory properties of food are processed in two steps in the cortex [see 235]. The insular cortex, functionally and neuroanatomically implicated as viscerosensory cortex [236], acts as a primary taste cortex where these individual features, i.e. taste, appearance, smell and texture, are represented to determine which food is being ingested. Individual neurones are sensitive to single stimuli, and do not adapt their firing even when the stimulus (e.g. glucose) has been present for a long time [237]. The orbitofrontal cortex (which receives direct input from the insular cortex), however, acts as a higher-order taste cortex, which determines how pleasant a particular food is. Orbitofrontal neurones are broadly tuned to react to multiple sensory features, so that the firing of these cells result from the combined inputs from several sensory modalities [238], although different cells respond to different quantitative combinations of stimuli. The subjective pleasantness rating is proportional to orbitofrontal activation and this activation drops accordingly when a particular food is eaten to satiety [239] (Fig. 6). Thus, computations within the orbitofrontal cortex confer hedonic qualities upon the feeding circuitry, a role whose importance in human appetite may be underestimated when extrapolating from rodent studies. The orbitofrontal cortex feeds directly into the LHA, which may thus constitute an important nexus for linking the subjective experience of food with homeostatic signals.

Figure 6.

Neurones in the orbitofrontal cortex decrease their firing in response to a particular food as that food is consumed to satiety. Extracellular recording from an odour-responsive neurone in the orbitofrontal cortex of a male macaque. The firing rate of one neurone in response to different odours was recorded whilst the animal consumed a blackcurrant juice. As satiety increases (lower panel), the electrophysiological response to blackcurrant odour diminishes, whereas the response to unrelated odours [banana, mango, phenyl ethanal (pe), citral (ct), onion (on)] is unaffected or elevated. Figure produced by H.D. Critchley and E.T. Rolls and is used with permission from the American Physiological Society; J Neurophysiol 1996;75:1673–1686.

The relationship of the hedonic/pleasure experience to human obesity has begun to be addressed in neuroimaging studies, further emphasizing the cortical parcellation of different feeding-related processing. Whereas during hunger, activation is observed predominantly in regions associated with the regulation of emotions (limbic and paralimbic cortex), satiety is followed by activation in the prefrontal cortex, postulated to play a role in the inhibition of inappropriate behaviours [240]. The presentation of a palatable sweet solution after a day and a half ?hangover 0>of fast resulted in increased signal from the insular cortex [241]. Moreover, almost all observed changes are accentuated in obese compared with lean subjects, i.e. both increases and decreases in activity are of greater amplitude [241] (Fig. 7). These differences are not exclusively accounted for by the hyperglycaemia and hyperinsulinaemia manifest in obese subjects. The functional implication of the intensified patterns of activation and inactivation in obese subjects is not clear at present, but may be of pathophysiological importance. Indeed, these responses persist also after weight loss in postobese subjects [242] suggesting that they are not a consequence of the overweight as such, although it is not known if such accentuated responses are seen also prior to the development of obesity. Elucidating these mechanisms may be significant also for understanding the human response to food advertisements.

Figure 7.

Lean and obese subjects show differences in brain activation during different states of hunger. Statistical parametric maps of significant brain responses (P ≤ 0.005, not corrected for multiple comparisons) to hunger and early satiety in obese (top row) and lean (bottom row) subjects, respectively, at 4 mm above (left images), 4 mm below (middle images), and 16 mm below (right images) a horizontal plane passing through the anterior and posterior commissures (coordinates from the Montreal Neurological Institute). The right hemisphere in each section is on the reader's right. The T-value colour-coded areas were regions of the brain in which significant changes in blood flow (a marker of neural activity) were detected in response to hunger (from yellow to white, in increasing order of T-value), as stimulated by a 36-h fast, or in response to early satiety (from blue to green, in increasing order of T-value), as stimulated by consumption of a satiating liquid meal [304]. The figure is intended for visual inspection only of several brain regions, where significantly greater responses in obese compared with lean individuals were detected, including the middle temporal gyrus (TEMP) insula (INS), dorsolateral prefrontal cortex (DLPFC), hippocampus (HIPP), temporal pole (T.POLE), orbitofrontal cortex (OFC), and ventrolateral prefrontal cortex (VLPFC). Figure generously provided by Dr Angelo DelParigi.

Development of the feeding circuitry

A series of recent studies have shed light both on the normal ontogenetic development of hypothalamic circuits as well as how these processes are influenced by the nutritional state of the young animal, with important implications for the aetiology of obesity. At birth, the hypothalamus is rather sparsely innervated by Arc NPYergic fibres, and in, e.g. the PVH a substantial NPY/AGRP innervation is seen first at postnatal day (P) 15 [197, 243, 244]. Similarly, whilst Arc NPY and AGRP mRNAs are detectable from birth, levels peak at P15, and drop to adult levels by P30 [244] (Fig. 8). This development parallels the maturation of the ability to regulate suckling in response to caloric demands [245], linking changes within the Arc metabolic sensor and adult control of feeding. Notably, failure in the development of the Arc is associated with fatal anorexia in a genetic model [246, 247]. It should be mentioned, that during the early postnatal period, and under certain physiological circumstances, transient expression of NPY in other hypothalamic nuclei can be seen [243, 248]. The role of these transitory NPY projections remains to be determined.

Figure 8.

Arc NPY/AGRP projections to the PVH develop during the postnatal period in the rat. Figure displays confocal micrographs of double-label immunofluorescence for NPY (red) and AGRP (green) in the PVH. Double-labelled fibres are shown in yellow. These images demonstrate that at postnatal day 5 (P5) that there are minimal Arc NPY/AGRP projections to the PVH, whilst there is an abundance of NPY fibres that originate from other sources. By P10 there is a significant concentration of Arc NPY/AGRP projection in the PVH, but they do not reach the adult levels until around P15. Images represent a 10-μm thick collection of optical sections collected at 0.5-μm intervals. Images were captured with a 25× oil objective (0.75 NA) and represent an area of 400 × 400 μm. Figure generously provided by Dr Kevin Grove.

As the Arc provides a main conduit for leptin, it is perhaps not surprising, given these data, that pups are unable to respond to leptin by changing their food intake [249], nor to ghrelin [250]. Yet, there is a pronounced peak in serum leptin prior to weaning [251]. Recent data from Bouret et al. [252] suggest that a postnatal leptin surge is essential for the development of Arc projections. In adult ob/ob mice, there is a distinct paucity of Arc-derived terminals. However, administration of leptin to ob/ob mice during early development, but not in adulthood, results in innervation patterns similar to what is seen in lean littermates, in parallel with a normalization of body weight. These data provide a novel mechanism for how nutritional signals in the early postnatal stage exerts long-lasting effects on the metabolic wiring in the adult. It will be of interest to determine if similar mechanisms are at play in the development of the brainstem-vagal system. In this context, the importance of the prenatal metabolic environment should also be remembered. Gestational diabetes and obesity is associated with obesity in the offspring of both rats [253] and humans [254]. An extensive long-term study is currently in progress to investigate what changes can be seen in central metabolic circuits in nonhuman primates following intrauterine exposure to diabetes [255]. Conversely, changes in hypothalamic circuitry during senescence may contribute to the loss of appetite that often accompanies ageing [17]. Such changes are observed in older rats, who display significantly lower levels of expression of NPY [256] and POMC, as well as POMC-positive neurones [257] and diminished dendritic arbours [258] in the arcuate nucleus compared with younger animals.

Mechanisms of anorexia in infection and cancer

The central systems mediate not only obesity but also anorexia of various aetiologies. Infection-based anorexia is a well-established and clinically highly relevant model [259], which has been instrumental in uncovering the pathways through which microorganisms cause reduced food intake. Bacterial components such as LPS from Gram-negative bacteria and muramyl dipeptide from Gram-positive bacteria activate CD14- and Toll-like receptors on host T-cells to induce production of cytokines such as interferon-γ; these results are corroborated by experiments employing genetic removal of strategic components along the pathway [260]. Cytokines are produced peripherally, but activate vagal afferents [261]. In addition, cytokines stimulate prostaglandin production via the enzyme cyclooxygenase 2 (COX-2) in cerebral endothelial and perivascular cells [262]. This latter pathway is important for the induction of LPS-induced anorexia, and can be blocked with indomethacin and other, more selective, COX-2 inhibitors [263]. Further downstream, the prostaglandins activate neurones producing serotonin (5-HT; [264]), a known anorexigenic transmitter [265, 266], which in turn has recently been shown, via 5-HT2C receptors, to stimulate melanocortin signalling [267]. Thus, the signalling cascade set off by pathogenic bacteria ultimately results in activation of the central anorexigenic system. These results may shed light also on the anorexia accompanying noninfectious inflammatory conditions. In this context it is interesting to note that, based on structure and signalling pathways, leptin itself belongs to the cytokine family [268, 269].

While anorexia is an important component also of wasting in cancer patients, nutritional supplements only alleviate part of the cachectic syndrome [270], which accounts for a fifth of cancer deaths [271]. Cachexia differs from starvation in that both adipose and lean mass is lost, whereas starvation primarily decreases fat stores [272]. A main explanation for this relationship is that many cancer tumours secrete proteins, e.g. proteolysis-inducing factor [273], which suppress protein synthesis in skeletal muscle, via activation of an ubiquitin proteolytic pathway [274]. Parenteral nutrition may thus be of very limited value to the patient if such catabolic mechanisms are not interrupted. However, it turns out that a polyunsaturated fatty acid found in fish, eicosapentaenoic acid (EPA), suppresses the activity of the proteolytic complex whilst simultaneously inhibiting tumour growth [275]. This finding has promising bedside implications; adding EPA to the diet of cancer patients attenuates muscle degradation and stabilizes body weight [276].

From rodent to human: genetic dissection

In few fields of medicine is the question of nature versus nurture more immediate than in the regulation of body weight. While it is evident that the incidence of obesity has accelerated far more rapidly than can be explained by population shifts within the genome, it is also true that individual differences determine how we react in an energy-dense environment. Aptly summed up by Olden and Wilson [277]: genes load the gun, environment pulls the trigger. Studies of monozygotic twins show that the heritable component of obesity equals that of height and surpasses virtually every other major disease studied, e.g. breast cancer, schizophrenia, cardiovascular disease [278–280]; some 40% of obesity can be attributed to genetic causes [281]. In a series of elegant studies, in particular by O'Rahilly, Farooqi and their colleagues, it has been shown that mutations in several components of the anorexigenic signal chain, including leptin [282] (Fig. 9), the leptin receptor [283], POMC [284] and neuropeptide processing enzymes [285] result in severe early-onset obesity in humans. While these monogenic disorders incontrovertibly demonstrate that genetic abnormalities can cause obesity, it cannot automatically be concluded that mutations and sequence variants are common causes of the metabolic syndrome.

Figure 9.

Leptin deficiency in humans responds to leptin treatment. A 3-year-old boy with congenital leptin deficiency with severe obesity (body weight 38 kg; BMI SD = 7.2) (left). On the right, the same patient, after four years of daily subcutaneous administration of recombinant leptin. Leptin treatment results in a dramatic decrease in adiposity (body weight 29 kg; BMI SD = 0.9) and normalization of all metabolic abnormalities including hyperinsulinaemia. Figure generously provided by Drs Sadaf Farooqi and Stephen O'Rahilly.

However, it is now becoming apparent that mutation of a particular key signalling protein, the MC4R [286], accounts for as many as 5% of cases of severe obesity [140, 287–289]. There is a remarkably solid relationship between the severity of the mutation, as revealed in in vitro assays, and the size of a test meal consumed by the patient [140]. Indeed, the heritable component is almost exclusively represented by increased intake of energy, with only a small component accounted for by changes in resting metabolic rate [140]. (In contrast, in the mouse MC4R−/− counterpart, deficient energy expenditure is an important factor underlying overweight [136].) The sometimes modest phenotypes observed in rodent gene knockout experiments (e.g. NPY gene deletion [125]) have been interpreted as evidence for a high degree of redundancy within the system underlying metabolic regulation. However, genetic studies such as these (as well as, notably, observed in nematodes [212]), suggest that there are weak points distributed throughout the system where minute changes in nucleotide sequence can have profound effects on the expression of behaviour. These vulnerable links along the metabolic signalling chain offer promising targets for therapeutic intervention.

Therapeutic prospects

As pointed out in the Introduction, effective treatments for eating disorders are short in supply and urgently needed. While much can be gained simply by increased efforts to educate patients in the benefits of weight loss and exercise [290, 291], it is also clear that such therapies in many cases fail in the absence of adequate pharmacological buttress. The anti-obesity drugs presently used in clinical practice have relatively modest effects, and others still have been withdrawn due to intolerable cardiovascular adverse effects [22]. These drugs have often targeted broadly distributed systems, e.g. serotonin pharmacology, and it is perhaps not surprising that alterations ensue within multiple body functions. Another issue to consider is the timing of administration of feeding-regulatory drugs, which may be more crucial than in any other therapeutic application (T. Bartfai, personal communication). However, based largely on the research summarized above, several novel compounds are now being tested in clinical trials, phases II and III [22, 292]. Many of these compounds target neuropeptide systems. The selective neuroanatomical distribution of many neuropeptides may provide an advantage in attempting to minimize side-effects, and whilst the field of peptide-based pharmaceuticals has not lived up to initial hopes, the clinical experience from opioid drugs such as morphine and naloxone [293] suggest that interfering with peptide signalling can produce powerful neuropsychiatric effects in humans [294]. In particular, drug development initiatives have centred on the melanocortin system [295], encouraged by the human genetic data reviewed above, and several nonpeptide compounds are now being tested in a clinical setting [22].

The cannabinoid system also provides an attractive target, with promising early results with the inverse CB1 receptor agonist, rimonabant [296], currently in phase III clinical trial. Initial evaluations suggest that rimonabant decreases body weight by ca 5–10%, with beneficial effects also on insulin resistance and dyslipidaemia [292]. Interestingly, this drug has the added benefit of facilitating smoke cessation. However, the full safety profile of rimonabant has not yet been released. Conversely, the active component of marijuana, Δ9-tetrahydrocannabinol, is also being evaluated for treatment of AIDS-associated anorexia [297].

Early hopes for a leptin-based obesity regimen were quelled by clinical trials showing only very limited weight loss following leptin administration to obese subjects [298], and there are also indications that such treatment may in itself produce the leptin resistance hypothesized to be a part of obesity [299]. It should be pointed out, however, that in the rare cases of genetic leptin-deficiency, treatment with recombinant leptin has virtually normalized body weight in near-fatally obese children, whilst simultaneously inducing age-appropriate puberty [300] (Fig. 9). The same treatment has also been used to successfully rectify several of the metabolic abnormalities associated with lipodystrophy [301, 302]. Furthermore, pertaining to a not uncommon diagnosis, leptin treatment has recently been demonstrated as a highly promising therapy for functional amenorrhoea [303].


In summary, significant advances in our understanding of feeding behaviour have been achieved using a combination of clinical, behavioural, electrophysiological, anatomical, genetic and imaging techniques. This investigation has defined an underlying circuitry and neuropharmacology, which can now be probed for therapeutic targets. Thus, whilst many questions remain to be answered, in particular regarding the causal mechanisms of obesity and anorexia as well as the central circuitry bridging metabolic input and output, there is reason for hope in offering effective treatments for these exceptionally common and debilitating disorders.

Conflict of interest statement

No conflict of interest was declared.


The author gratefully acknowledges generous financial support from Wenner-Gren Stiftelserna, Vetenskapsrådet, Jeanssons Stiftelser, Hagbergs Stiftelse, Rut och Arvid Wolffs Stiftelse, Thurings Stiftelse, Svenska Läkaresällskapet, Åke Wibergs Stiftelse, Kgl. Vetenskapsakademien, Hedlunds Stiftelse, Magnus Bergvalls Stiftelse, Lars Hiertas Minne, Axel Linders Stiftelse, Längmanska Kulturfonden, Teodor Neranders Fond and internal funds of Karolinska Institutet.