Insulin release from pancreatic β-cells is required to maintain normal glucose homoeostasis in man and many other animals. Defective insulin secretion underlies all forms of diabetes mellitus, a disease currently reaching epidemic proportions worldwide. Although the destruction of β-cells is responsible for Type 1 diabetes (T1D), both lowered β-cell mass and loss of secretory function are implicated in Type 2 diabetes (T2D). Emerging results suggest that a functional deficiency, involving de-differentiation of the mature β-cell towards a more progenitor-like state, may be an important driver for impaired secretion in T2D. Conversely, at least in rodents, reprogramming of islet non-β to β-cells appears to occur spontaneously in models of T1D, and may occur in man. In the present paper, we summarize the biochemical properties which define the ‘identity’ of the mature β-cell as a glucose sensor par excellence. In particular, we discuss the importance of suppressing a group of 11 ‘disallowed’ housekeeping genes, including Ldha and the monocarboxylate transporter Mct1 (Slc16a1), for normal nutrient sensing. We then survey the changes in the expression and/or activity of β-cell-enriched transcription factors, including FOXO1, PDX1, NKX6.1, MAFA and RFX6, as well as non-coding RNAs, which may contribute to β-cell de-differentiation and functional impairment in T2D. The relevance of these observations for the development of new approaches to treat T1D and T2D is considered.
Diabetes mellitus ultimately results from the absolute [TiD (Type 1 diabetes)] or relative [T2D (Type 2 diabetes)] loss of insulin secretory capacity, and affects more than 8% of the adult population worldwide . This figure is expected to double in the next ~20 years to affect more than 500 million individuals, largely due to increasingly sedentary lifestyles and the mounting incidence of obesity . The attendant burden on health care systems required to treat the complications of the disease–which range from stroke, cardiovascular disease, blindness to kidney failure–currently exceeds 10% of total annual healthcare spend of most developed nations ($245 billion per year in the United States alone) .
β-Cells are the predominant cell type within the pancreatic islet in mammals and are the sole source of circulating insulin. Scattered throughout the pancreas, and typically making up ~1% of the total volume of the gland , each micro-organ (diameter 50–500 μm) comprises 50–3000 cells. In rodents, β-cells form the core of the islet , with other endocrine cells, notably glucagon-secreting α-cells and somatostatin-secreting δ-cells, along with smaller numbers of polypeptide P (PP) and ghrelin-expressing ε-cells, arranged towards the periphery . In human islets, there is a greater intermingling of cell types. In this case, the final compact structure of the islet probably results from the folding of a trilaminar sheet comprising α-cells (outer layers) and β-cells (inner layers) . The latter topology may afford greater interaction between different cell types within the islet, with possible physiological consequences for the regulation of insulin secretion . Although changes in the function of other islet cells, in particular the α-cell complement, are likely to contribute to the pathology of the disease , in the present review, we focus on better understood, as well as emerging, changes that affect β-cells.
In the face of lowered, but constant, insulin sensitivity , progressive loss of β-cell mass (whose extent is contested) [11–13], and dysfunction of the remaining cells [14,15], lead to ‘decompensation’ and the emergence of frank T2D . Both ‘environmental’ changes (e.g. hyperglycaemia and hyperlipidaemia, and possibly inflammation ), as well as genetic contributions  appear to be involved in β-cell failure. The reader is referred elsewhere [18,19] for detailed reviews on this topic.
Interestingly, assessments of β-cell age in man suggest that there is little expansion in numbers in adults , despite dramatic increases in secretory capacity under some circumstances–notably during pregnancy or in response to obesity-induced insulin resistance (‘compensation’). However, and although increases in β-cell mass during pregnancy have been demonstrated in rodents , evidence for large changes in man is limited [22,23]. Conversely, in the context of β-cell failure, early studies from Maclean and Ogilvie in 1955 , and more recent findings , suggest that β-cell mass, assessed post mortem, is lower by ~50% in T2D (and by >60% in obese diabetics) versus healthy control pancreata. However, these conclusions have been questioned [13,25]. Indeed, work from Rahier et al.  suggests that the difference in β-cell number between the healthy and T2D population is on average only ~25% within 5 years of diagnosis–a change insufficient to cause the symptoms of the disease. In the absence of an ability to monitor β-cell mass over time, it is impossible to be certain that these changes do actually reflect ‘loss’ of β-cells in T2D, rather than an increase in susceptibility to the disease in those born with fewer. In any case, β-cell numbers vary considerably even between normoglycemic individuals (from 0.3 to 2.0% of pancreatic mass) , with substantial overlap between these values and those found in T2D donors.
Given the absence of marked differences in β-cell mass, despite clear deficiencies in glucose (but not other secretagogue)-stimulated insulin secretion in T2D both in vivo [26,27] and in vitro [14,15], the focus now is on understanding the molecular basis of changes in β-cell function and, in particular, in glucose sensing. Moreover, models of β-cell ablation in rodents suggest that non-β cells have the potential to undergo changes in identity, beginning to express insulin and to respond to glucose with the secretion of the hormone [28,29]. Such changes might provide exciting new avenues to replenish the β-cell pool as a treatment for T1D.
In the present paper, we describe firstly some of the key biochemical features of the β-cell, in terms of its ability to synthesize, store and release insulin when required, i.e. as blood glucose concentrations rise. In particular, we survey current understanding of stimulus-secretion coupling in these cells before discussing mechanisms through which signal generation may become defective. The molecular mechanisms which may also allow non-β-cells to transdifferentiate into β-cells are also discussed.
ACUTE REGULATION OF INSULIN SECRETION: THE β-CELL GLUCOSE SENSOR
Glucose is the most important physiological secretagogue for insulin. The β-cell is thus poised to convert small fluctuations in blood glucose concentration (typically from 4.5 to 8 mM in man) into large changes in insulin secretion within minutes. Within β-cells, newly synthesized insulin is first produced as the prohormone proinsulin, and converted into mature insulin through the action of prohormone convertases (PC1, PC2, encoded by Pcsk1 and Pcsk2, respectively)  during trafficking through the secretory pathway. Active insulin is then stored in dense core secretory granules (5–10000 per cell) [31,32], each containing 300000 or more molecules of insulin. The tightly-regulated release of only a fraction of the granules through exocytosis (~2% per hour at maximal glucose concentrations)  is sufficient to achieve regulation of blood glucose levels within the above narrow physiological range. This tight regulation is important not only to prevent hyperglycemia, but equally to suppress the potentially lethal hypoglycemia which would accompany over-secretion of insulin.
Central to glucose sensing by β-cells is the stimulation of glycolytic and oxidative metabolism of the sugar [34,35], ultimately causing enhanced mitochondrial ATP synthesis. Thus, flux through alternative pathways including the pentose phosphate shunt  is usually small, though it may be increased under some circumstances . Increases in total , as well as free cytosolic  mitochondrial matrix and sub-plasma membrane ATP/ADP ratios  then lead to the closure of ATP-sensitive K+ (KATP) channels [41,42]. The unbalanced influx of positively charged ions, notably Na+, then leads to plasma membrane depolarization , the firing of action potentials and the opening of voltage-gated Ca2+ channels . This, in turn, prompts the activation of secretory granule-associated small N-ethylmaleimide-sensitive factor receptor (SNARE) proteins  and granule fusion with the plasma membrane . Highly localized changes in free Ca2+ , for example at the inner surface of the plasma membrane , at the mouth of voltage-gated calcium channels , and at the surface of secretory granules , are also believed to be important in controlling this process. Calcium release from intracellular organelles including the ER (endoplasmic reticulum)  and Golgi  [mediated via IP3 (inositol 1,4,5-trisphosphate)] as well as secretory granules [53,54] and other acidic stores including lysosomes  (via the generation of nicotinic acid–adenine dinucleotide phosphate, NAADP), may also be involved, although this is a contested area (see  for a recent comprehensive review of Ca2+ signaling in β-cells). Nonetheless, the above description summarizes the essentials of the ‘canonical’ pathway for glucose-stimulated insulin secretion (Figures 1 and 2). In addition, a constellation of further intracellular signalling events, independent of KATP channel closure, is also likely to be important for normal glucose sensing and will be discussed further below.
Overview of canonical signalling mechanisms involved in β-cell glucose sensing, and responses to secretory potentiators or inhibitors
Besides glucose, a range of other fuel secretagogues including amino acids such as leucine, and the α-ketoacid ketoisocaproate are also ‘primary’ metabolic stimulators of insulin release, and are likely, at least in part, to engage the same metabolic signalling pathways activated by glucose (Figure 1 and see below). Forming a distinct group from the above, a range of physiologically-important secretory ‘potentiators’ also exists (Figure 1). These enhance insulin release only at permissive (i.e. stimulatory; usually above 6 mM) glucose concentrations. The latter group includes the incretin hormones glucagon-like peptide-1 (GLP-1) and glucose-dependent insulinotropic peptide (GIP), as well as cholesystokinin (CCK), peptide YY (PYY) and oxyntomodulin, released from the gut in response to food transit . These are responsible for the augmentation of insulin release in response to food intake versus an identical change in glycemia imposed by intravenous injection of the sugar. The incretins act via specific G [Gs] protein-coupled receptors (GPCRs, e.g. GLP1R and GIPR) on the β-cell surface to increase intracellular cAMP concentrations , activating protein kinase A (PKA), as well as exchange protein activated by cAMP 2 (EPAC2) [58,59] and other signalling pathways (mediated, for example, by β arrestin and MAPK) . Acetyl-choline, acting through muscarinic M3 receptors , and ATP (via P2X and P2Y purinoreceptors) , as well as fatty acids (via GPR40/FFAR1) [63,64], and vasopressin, also enhance secretion triggered by nutrients by increasing cytosolic Ca2+, whereas vasoactive intestinal peptide (VIP), PYY and oxyntomodulin probably act via cAMP (Figure 1). Inhibitors of secretion include somatostatin, which acts in part via an inhibitory G protein , adrenaline (epinephrine) and noradrenaline (norepinephrine), the latter acting through α2 receptors to open KATP channels and hyperpolarize the cell  (Figure 1). The actions of epinephrine (and norepinephrine) are important to suppress insulin release during exercise .
An important feature of glucose-stimulated insulin secretion is its ‘phasicity’. Thus, two defined phases follow a square wave increase in glucose both in vivo in man  and when measured from the perfused rat pancreas . The first phase, lasting 3–10 min, is followed by a nadir and a subsequent, gradually-increasing second phase which may last 60 min or more. Many theories have been proposed historically to explain this behaviour [33,70,71], though none provides a complete explanation. Importantly, the first phase is preferentially impaired in T2D .
It is important to distinguish the first phase, as described above, from the ‘cephalic’ phase of insulin release. The latter occurs in vivo in response to the sight and smell of food, and before any change in glycaemia, both in rodents  and man . This results in relatively modest increases in insulin levels (~25% above basal), and reflects an autonomic response, mediated by parasympathetic hormone release from nerve terminals within the islet. Both cholinergic and non-cholinergic responses are implicated , whereas gut-derived incretin hormones (GLP-1 etc.) are not involved (though the actions of the latter have been proposed at least in part to involve GLP-1 receptors in the brain) . Despite the modest degree of insulin release, studies mimicking these changes suggest that the cephalic phase does contribute to normal glucose homoeostasis .
Insulin release is also usually pulsatile in vivo in humans  and other species including dogs . In man, this property is lost in first degree relatives of T2D patients . Pulsatility, which has a period of 10–15 min, may represent the integral of the electrical ‘bursting’ observed at the level of the whole islet [43,78] and in single β-cells . Whatever their provenance, how metabolic, electrical or secretory oscillations, apparent at the level of individual β-cells and islets, are relayed throughout the full complement of islets within the pancreas (~1000 in rat or up to one million in human) is largely unknown . Although the muscarinic signalling phase sets activity in isolated rodent islets , supporting a neuronal mode of coordination, it is more difficult to envisage how this may occur in the human pancreas, which is relatively devoid of cholinergic fibres .
ENZYMOLOGY OF β-CELL NUTRIENT SENSING
Glucose transport and phosphorylation
Work by Matschinsky and Ellerman , as well as Randle et al. , first established the cardinal importance of glucose uptake and subsequent metabolism, as opposed to the binding of the sugar to a receptor (but see below), in triggering insulin secretion (reviewed by [82,83]) (Figure 1). Thus, glucose metabolism, which occurs prior to action potential firing (Figure 2), closely parallels the stimulation of insulin release, with half maximal values for both close to 10 mM. Moreover, only efficiently-metabolized sugars serve as secretagogues. On the other hand, fructose appears to be able to prompt insulin secretion in vitro and in vivo via the sweet taste receptor T1R2 , an effect which may potentiate the effects of glucose. Moreover, very recent evidence  suggests that the T1R3 sweet taste receptor may mediate some of the effects of glucose by further enhancing the metabolism of the sugar. Importantly, the affinity of these receptors (K0.5 ~8 mM for glucose)  would appear to be compatible with a physiological contribution to the control of insulin secretion.
Glucose-stimulated changes in plasma membrane potential, intracellular ATP/ADP and Ca2+ in a single pancreatic β-cell
In normal circumstances, i.e. in the healthy β-cell, the rate of glucose transport across the plasma membrane is thought to exert minor control over insulin secretion. Although β-cells, in common with the liver, express the low affinity glucose transporter GLUT2 (encoded by Slc2a2), the capacity for transport is believed to exceed that of glucose phosphorylation, such that glucose concentrations quickly equilibrate across the plasma membrane . The relative importance of GLUT2 and other members of this family are in any case contested in human islets , with evidence existing for a predominant role for GLUT1. Further complicating the debate, GLUT1 (encoded by Slc2a1) is highly inducible by hypoxia . Since the latter frequently occurs during human islet isolation, it is possible that levels of GLUT1 mRNA in healthy human islets have been overestimated. If this is the case, then lowered GLUT2 expression may contribute to impaired glucose sensing in the T2D β-cell .
The identification in islets of a high (~10 mM) Michaelis constant (KM), and cooperative (Hill coefficient ~1.5)  HK (hexokinase) type IV, GK (glucokinase), encoded by Gck, present alongside higher affinity hexokinases, suggests that glucose phosphorylation is the key flux-generating step. Subsequent studies  revealed that the higher affinity HKs are largely absent from highly purified β-cells. Importantly, and providing direct evidence for the importance of GK in controlling insulin secretion in man, mutations in the human GCK gene lead to monogenic diabetes (maturity onset diabetes of the young-2; MODY2) .
The role of mitochondria
Demonstrating the importance of mitochondrial metabolism in the regulation of insulin secretion in man, abnormalities in the expression of the mitochondrial genome are associated with impaired insulin secretion in maternally-inherited diabetes and deafness (tRNALeu mutation) [93,94]. Likewise, T2D-associated variants in the human transcription factor B1 mitochondria (TFB1M) gene lead to impaired mitochondrial oxidative phosphorylation .
Suggesting that mitochondrial metabolism of glycolytically-derived pyruvate and NADH are central to normal glucose sensing, fuel secretagogues other than glucose, such as leucine and ketoisocaproate , are largely or exclusively metabolized by mitochondria . Moreover, the cell permeant mitochondrial substrates methylsuccinate and methylpyruvate are potent stimulators of insulin secretion  in both rodent and human islets . Conversely, respiratory chain inhibitors and uncouplers strongly inhibit insulin release . Of note, stimulation of intramitochondrial dehydrogenases by Ca2+ , flowing into mitochondria in response to glucose-induced rises in cytosolic Ca2+ concentration [102,103], and increased mitochondrial membrane potential, are likely to further enhance oxidative metabolism in response to high glucose. This may maintain ATP levels (and hence the closure of KATP channels) in the face of increased cytosolic energy consumption (for ion pumping, protein synthesis, etc.). Correspondingly, chelation of mitochondrial matrix calcium  or silencing of the MCU (mitochondrial Ca2+ uniporter) [39,105], impairs the second phase of insulin secretion, whereas inhibition of the mitochondrial Na+–Ca2+ exchanger NCLX increases secretion .
The cellular origins of pulsatile insulin secretion, as discussed above, are poorly defined, but may involve repolarization due to the closure of Ca2+-sensitive K+ (KCa) channels . Changes in metabolic flux, involving oscillations in ATP [108,109], have also been proposed to play a role and may be driven both by an intrinsic ‘glycolytic oscillator’  or by changes in mitochondrial oxidative activity due to the stimulation of intramitochondrial dehydrogenases by Ca2+  (see below for further discussion). Conversely, increased ATP consumption, for example to power Ca2+ extrusion, is likely to drive ATP concentrations lower , setting up a complex interplay between metabolic and Ca2+ oscillations during glucose stimulation. Although cause and effect is still debated, we have noted (Tarasov, A.I., G.A.R., unpublished) sustained oscillations in electrical activity in mouse β-cells against a background of otherwise unchanging cytosolic ATP/ADP ratio , arguing against the importance of metabolic oscillations in driving changes in electrical activity. Nonetheless, highly localized changes in ATP concentration beneath the plasma membrane  cannot be excluded. Suggesting that [ATP] oscillations are unlikely to be driven by pulses in cytosolic Ca2+, fluctuations in electrical activity are still observed in β-cells from mice null for either KATP channel subunit [112,113], as well as following application of diazoxide .
‘KATP CHANNEL-INDEPENDENT’ INSULIN SECRETION
Recent reviews  describe in detail the key features of glucose-regulated insulin secretion which can occur in the absence of changes in KATP channel activity. The existence of such mechanisms was first inferred from the observations that (i) elevated glucose concentrations are still able to enhance insulin secretion either in the absence of external calcium  or, more effectively, in the presence of elevated, but unchanging, intracellular Ca2+ concentrations ; (ii) stimulation with incretins such as GLP-1, whereas prompting elevation in cytosolic Ca2+ (more profoundly in human than mouse islets)  and ATP, exert little effect on insulin secretion unless glucose concentrations are elevated .
Although the role of cAMP in the control of β-cell glucose metabolism is still an area of some controversy, the observation that the adenylate cyclase ADCY5, and cAMP generation , are required for normal glucose-induced ATP increases in human islets points to a mutualism in the actions of glucose and incretins on the β-cell, as first pointed out by Holz et al. . Possible mechanisms through which increases in cAMP may modulate glucose metabolism include activation of GK (via EPAC2) , pyruvate carboxylase  or phosphofructokinase (PFK1) . It should be stressed that these metabolic actions of cAMP are separate from distal effects of the messenger on e.g. Ca2+ channels  or the secretory machinery .
Over the years a long, and still-growing, list of coupling factors and mechanisms has been implicated in the KATP channel-independent actions of glucose, of which a selection are discussed below. For comprehensive reviews, see [126,127].
Anaplerosis and the generation of NAD(P)H, reduced glutathione. Although gluconeogenesis in β-cells is precluded by low phosphoenolcarboxykinase (Pck1) and fructose 1,6-bisphosphatase (Fbp1) expression (www.biogps.org), pyruvate carboxylase levels are relatively high in these cells and 40–50% of pyruvate-derived carbon enters the tricarboxylate cycle after conversion into oxaloacetate [128,129], the remainder through decarboxylation to acetyl-CoA via pyruvate dehydrogenase (PDH) (Figure 3). The former process ‘tops up’ mitochondrial intermediates (‘anaplerosis’) and may allow for their re-export from mitochondria in a signalling role. Although release from mitochondria as citrate or glutamate (see below) represents one possible fate for pyruvate, the regeneration in the cytosol of six carbon metabolites including isocitrate, and the action of cytosolic isocitrate dehydrogenase, raises the possibility of generating cytosolic NADPH via isocitrate/2-oxoglutarate and pyruvate/malate cycles  (Figure 3). Studies by Renström and co-workers  suggest that such an increase in NAD(P)H/NADP ratio may reduce the cytosolic redox sensors glutaredoxin and thioredoxin, in turn regulating the Ca2+-dependent exocytosis of secretory granules. Carbon flux from glucose or glutamine into glutamate may also enhance the synthesis of reduced glutathione , contributing to both the acute stimulation of insulin secretion and to the resistance of the β-cell to oxidative stress.
Glutamate. Originally proposed as a coupling factor by Maechler and co-workers [132,133], increases in glutamate levels following challenge with high glucose were shown in permeabilized β-cells to prompt exocytosis. However, others have challenged this hypothesis, citing small changes in cytosolic glutamate in response to elevated glucose  and the failure of glutamine to prompt insulin secretion despite producing large increases in glutamate . Note that in the presence of leucine, which allosterically activates glutamate dehydrogenase (GDH), glutamine efficiently stimulates secretion, by furnishing glutamate-derived carbon atoms to the TCA (tricarboxcylic acid) cycle (Figure 3). Defining a molecular mechanism through which glutamate acts directly on exocytosis has been challenging, with uptake into secretory granules, the activation of granule-localized metabotropic glutamate receptors (leading to highly localized changes in calcium close to or within the granule), or granule alkalinization all mooted as possibilities . Nonetheless, recent studies by Seino and co-workers  have shown that glutamate is likely to be the elusive factor that mediates the glucose-dependency of incretin action.
Acyl-CoA, acyl-glycerol. Findings including the fact that perturbation of fatty acid oxidation, achieved by modulating malonyl-CoA decarboxylase expression, interferes with insulin release  have led Prentki and co-workers to place lipid species in a central position in the control of glucose-stimulated insulin secretion . Supporting this view is the high ratio of acetyl-CoA carboxylase (ACC) versus fatty acid synthetase activity in β-cells , and relatively low levels of acyl-CoA thioesterase 7 (ACOT7) versus other tissues (Figure 3 and see below under Disallowed genes).
Reactive oxygen species (ROS). Evidence for a role for ROS , and specifically for H2O2 [142,143], has been provided by a number of studies. However, the interpretation of these results is complicated by the challenges inherent to studying the role of these small and highly reactive molecules. Interestingly, overexpression of catalase, a β-cell disallowed gene [144,145] (see below), exerted no effect on glucose-simulated insulin secretion in a recent report .
AMP-activated protein kinase (AMPK). Given that changes in ATP, ADP and AMP, and notably in adenylate charge [i.e. ATP/(ADP + AMP)], follow glucose stimulation, we and others  have speculated that AMPK, an evolutionarily-conserved fuel sensor , may also generate signals that regulate secretion independently of Ca2+ changes. Supporting this view, AICA-riboside (a cell permeable precursor of ZMP, an AMP analogue) stimulates insulin secretion at low but not high glucose . Furthermore, a constitutively-active form of the enzyme (AMPK CA) blocked insulin release at high glucose . In vivo, mice expressing AMPK CA selectively in the β-cell displayed impaired glucose tolerance . Correspondingly, dense core granules were depleted from the plasma membrane by AMPK activation . These effects may involve impaired granule mobility  mediated by phosphorylation of kinesin light chain 1 (KLC1)  (Figure 3).
Conversely, a dominant-negative form of AMPK (α1.D157A) stimulated insulin release at low glucose and improved the performance of transplanted islets in streptozotocin-induced diabetic mice . In contrast with in vitro findings, Cre recombinase-mediated deletion of both isoforms of the catalytic subunits of AMPK (α1 and α2) selectively in β-cells led, unexpectedly, to impaired secretion in vivo [151,155]. Nevertheless, glucose-stimulated release of insulin from islets isolated from these animals was strongly potentiated by the absence of AMPK ex vivo [151,156]. This difference may reflect changes in the expression of receptors for secretory potentiators, including those for glucagon, incretins and somatostatin , which impair β-cell glucose responsiveness in vivo.
Importantly, a role for AMPK inhibition is entirely in line with the ‘acyl-CoA hypothesis’ described above: release of the inhibition of acetyl-CoA carboxylase by loss of AMPK-mediated phosphorylation at Ser79 is expected to enhance malonyl-CoA synthesis, thus decreasing fatty acid oxidation (Figure 3). Alternatively, it is conceivable that non-esterified (free) fatty acids (FFA) are released from the β-cell and agonize FFAR1 (GPR40) receptors, which, via Gq-coupling to phospholipase C may then mobilize intracellular Ca2+; weak expression of the acyl-CoA thioesterase ACOT7 (see below) may, however, suppress this (Figure 3). It should be stressed, however, that others have questioned the above model: Newgard and co-workers  were unable to detect any changes in insulin release upon treatment of INS1(832/13) cells or islets with AICA riboside.
Anaplerosis and lipid-derived factors in stimulus-secretion coupling
INTRA-ISLET CONTROL OF INSULIN SECRETION
This topic has recently been reviewed elsewhere [158,159] and will be discussed only briefly here. β-Cells within the islet display more robust secretory responses than dissociated cells , a feature attributed to transmission of poorly-defined signals between β-cells. Moreover, the preserved expression of gap junction proteins, notably Cx36, is essential for normal stimulus-secretion coupling  and insulin pulsatility in vivo , most probably through promotion of electrotonic coupling . Recent data, applying network theory previously used to study neuronal circuits [163,164], has revealed the importance of more long-range interactions between β-cells , and suggests a hierarchical arrangement of ‘hub’ (pacemaker-like) and follower cells within the islet (Hodson, Rutter and Johnston, unpublished). The molecular and cellular basis of this connectivity is influenced by a range of diabetogenic insults  and by diabetes-associated genes (ADCY5  and TCF7L2 [165,166]), and is currently a topic of active investigation.
‘DISALLOWED’ (‘FORBIDDEN’) GENES IN THE β-CELL
Consistent with an important role for mitochondrial oxidative metabolism in the stimulation of insulin secretion (see above), more than 80% of glucose carbon is converted into CO2 and H2O in β-cells [91,167]. This feat is achieved in few if any other mammalian cells, and matched rarely in other species, an exception being highly oxidative insect flight muscle . Moreover, β-cells show neither a Crabtree effect, i.e. suppressed oxidative metabolism with high glycoloytic flux [169,170], nor a Pasteur effect (enhanced glycolytic production of lactate during anaerobosis) . Thus, and possibly uniquely in β-cells, glycolytic flux is tightly coupled to mitochondrial oxidation of substrates. How is this unusual metabolic configuration achieved?
An early clue was provided by a study from Hellman and Taljedal  who showed that lactate dehydrogenase activity in pancreatic islets was lower than in surrounding acinar tissue. We  subsequently showed that both lactate dehydrogenase and lactate transport activity across the plasma membrane were indeed lower in clonal and purified primary β-cells than in other islet or non-islet cells, results subsequently confirmed by others  [findings of relatively high LDH (lactate dehydrogenase) levels in islets  may reflect contamination with other pancreatic cells] (Figure 1). On the other hand, glycerol phosphate dehydrogenase (GPDH) activity is remarkably high in the β-cell , such that a highly active glycerol phosphate shunt, as well as the malate/aspartate shuttle [137,174], regenerate the NAD+ required for continued flux through glyceraldehyde-3-phosphate dehydrogenase, and hence glycolysis. Importantly, the stimulation of GPDH (located on the inner mitochondrial membrane) by Ca2+ ions , is likely to further stimulate glycolytic flux, complementing the stimulation of intramitochondrial dehydrogenases (see above).
Subsequent studies [144,145] have identified ~66 genes which have relatively low levels of expression in islets compared with all other tissues. Of these, 11 were common to the two studies  and form a core of ‘disallowed’ β-cell genes. The direct relevance of the suppression in the β-cell of the founder member of the group, Slc16a1, encoding the monocarboxylate transporter MCT-1 , was demonstrated firstly by the identification of promoter variants in the human gene which are hyper-activated in vitro, and thus likely to cause overexpression of the protein in the β-cell, and consequently ‘exercise-induced hypoglycaemia’ (EIHI) . Thus, in EIHI patients, but not healthy individuals, pyruvate generated alongside lactate by muscle during exercise is taken up by the β-cell, triggering enhanced mitochondrial metabolism and hence insulin release. Providing direct evidence for this model , forced overexpression of MCT-1 selectively in the β-cell in mice (Figure 4A) recapitulates the exaggerated release of insulin seen during exercise in man  and allows pyruvate to stimulate insulin secretion from isolated islets (Figures 4B and 4C). This likely reflects an unmasking of the ability of pyruvate to stimulate mitochondrial oxidative metabolism, as observed in isolated islets overexpressing MCT-1 .
Impact of MCT-1/Slc16a1 overexpression on glucose- and pyruvate-stimulated secretion
What of the other disallowed genes? Although studies of their role are still at an early stage, in unpublished work we have shown that overexpression of another disallowed gene, Acot7, leads to defective secretion selectively in the presence of fatty acids, possibly by preventing the accumulation of key lipid-linked signalling intermediates  (and see above) (Figure 3). This finding lends support to the importance of lipid signalling in glucose-stimulated insulin secretion . Likewise, low levels of anti-oxidative enzymes including catalase and glutathione peroxidase may contribute to β-cell vulnerability in T2D .
IS THERE A β-CELL ‘IDENTITY CRISIS’ IN TYPE 2 DIABETES?
Might loss of β-cell differentiation, and a change in the metabolic configuration described above, contribute to defective insulin secretion in T2D?
The notion that β-cell identity may be compromised was first invoked by the work of Jonas et al. , who found that hyperglycaemia provoked by partial pancreatectomy in the rat led to changes in the expression of the weakly expressed genes Ldha, as well as hexokinase 1 (Hk1), in rat islets. It was later suggested that this may involve de-differentiation towards a precursor state, based on the finding that, in the face of stresses such as multiple pregnancy or aging, loss of FOXO1 provoked the induction of progenitor-like genes including Ngn3, Oct4, Nanog and other de-differentiation markers . However, there is no evidence to date that such changes occur in human T2D islets .
β-Cell identity also appears to be regulated in the adult by transcription factors including NEUROD1 , NKX6.1 , PAX6  and PDX1 . Perhaps most dramatically, PDX1 suppresses the trans-differentiation of β-cells towards an α-cell fate , whereas Nkx6.1-deleted β-cells assume characteristics of δ-cells. Importantly, in the latter model, glucose metabolism, as assessed through phosphoAMPK staining or measurements of ATP, was severely compromised, a change likely to contribute to the observed impairment in basal and glucose-stimulated insulin secretion (see above). Since NKX6.1 expression is reportedly reduced in T2D versus healthy human islets , it is plausible that reduced levels of this factor contribute to secretory dysfunction in the disease.
Strikingly, loss of NeuroD1 caused a marked increase in Ldha expression , though whether changes in NEUROD1 expression are observed in the human β-cell in T2D is unclear . Recent findings  also demonstrate that Rfx6 is a critical regulator of the disallowed genes. Although mutations in the RFX6 gene in man, or inactivation by homologous recombination in mice , lead to a failure in endocrine pancreas development and neonatal diabetes, inactivation of Rfx6 in adult mouse β-cells also leads to secretory dysfunction and abnormal glucose sensing. RNA-Seq analysis reveals that the latter changes are likely to be due, at least in part, to the ability of RFX6 to act both positively as a regulator of β-cell signature genes including Gck, Glut2/Slc2a2, Abcc8 and Kcnj11 (as well as a constellation of Ca2+ channel subunits) and to bind directly to, and repress, a substantial fraction (32/54) of the Rfx6-regulated disallowed genes . Such findings suggest that inhibitors of the disallowed genes may be useful not only for treatments of common forms of T2D, but also in cases of Rfx6 mutation.
MECHANISMS RESPONSIBLE FOR PRESERVING β-CELL IDENTITY: DISALLOWED GENE SILENCING
At present, knowledge of how a subset of disallowed genes is selectively suppressed in β-cells is limited, with studies restricted to only a few of the core group, i.e. MCT-1/Slc16a1  and Acot7 . For those that have been examined, both transcriptional (epigenetic) mechanisms and control of mRNA stability are involved (Figure 5).
Putative mechanisms for suppression of disallowed genes
Many of the disallowed genes are located in transcriptionally silent areas characterized by histone trimethylation at H3K27, whereas active marks (H3K4me3) are absent . The polychrome repressor complex (PRC1) is also an important contributor to the suppression of the disallowed gene family [14,193]. Thus, mice inactivated selectively in the β-cell for RING1B, a component of the PRC1 complex, show an enhancement (by gene set enrichment analysis) of these genes, as well as of neural genes.
Interestingly, impairment of AMPK signalling in the β-cell  leads to up-regulation of many of the same genes, suggesting that AMPK targets are involved in the suppression of these genes (Figure 5). Thus, when AMPK deletion in β-cells in vivo was followed by RNA-Seq, gene sets enriched included both neuronal genes (usually expressed at relatively low levels in β-cells but not formally ‘disallowed’) as well as several of the disallowed genes, were significantly up-regulated. In silico analysis of transcription factor binding revealed strong enrichment for the neuronal factor ZFP206, but notably also HIF1α. The latter is a known regulator of glycolytic genes in the β-cell  whose up-regulation, after the inactivation of von Hippel Lindau factor, leads to the induction of Ldha and other genes, impairing insulin secretion. It is tempting to speculate, therefore, that maintaining high levels of AMPK activity in the β-cell between meals (i.e. during effective nutrient deprivation for these cells) is important to maintain their identity: sustained hyperglycaemia, on the other hand, may entrain a vicious cycle in which not only are the flux-generating glucose sensors including Glut2 and Gck suppressed , but disallowed gene expression is increased. Such changes may contribute to fasting hyperinsulinemia which characterizes T2D, conceivably by permitting the inappropriate stimulation of secretion by fuels such as pyruvate , and also attenuated secretion at high glucose levels. Indeed, AMPK activators such as metformin , which improve glycaemic profiles in T2D  largely by suppressing hepatic glucose output , may also maintain normal insulin secretion in the long term by preserving the differentiated state of β-cells.
Until recently thought of as of little more than transcriptional ‘biproducts’, non-coding RNAs have begun to assume an important place in the pantheon of regulatory molecules in the β-cell. Thus, both microRNAs (miRNAs) [198–200] and lncRNAs (long non-coding RNAs)  are now implicated in β-cell differentiation and function.
Several studies have shown that miRNAs are essential for pancreas development [202,203], for the conversion of stem cells towards a β-cell fate  and for normal β-cell function . Thus, depletion of Dicer (therefore disrupting miRNA maturation) early in pancreas development resulted in gross defects in all pancreatic lineages and pancreas agenesis , whereas disruption only in β-cells during embryonic progression led to defective insulin secretion, β-cell mass reduction and overt diabetes mellitus [204,205].
Supporting a role for miRNAs in suppressing β-cell dysfunction and the development of diabetes, variations in miRNA expression have been observed upon exposure of β-cells to gluco(lipo)toxic conditions in vitro or during the development of T1D and T2D . Several miRNAs are differentially expressed in islets from non-diabetic and T2D organ donors, and target genes associated with β-cell dysfunction and death [207,208]. These two independent studies [207,208] each report large (~10-fold) increases in miR-187 levels, whereas a cluster of miRNAs is down-regulated . Changes in the expression of almost 50 miRNAs also occur in prediabetic NOD mice islets, a well-established model of T1D  and in islets from Goto-Kakizaki rats, a non-obese T2D animal model . More than 60 miRNAs are differentially expressed in islets from the db/db mouse model for T2D and in mice fed a high-fat diet .
miRNAs are thus interesting candidates as controllers of β-cell identity . Correspondingly miR-29 family members are required for the suppression of Slc16a1 expression in β-cells , and may suppress neuronal genes  and transcriptional repressors . Moreover, inactivation selectively in β-cells of DICER, required for miRNA maturation, leads to the mis-expression of a number of disallowed genes (A.M.-S., T.J.P and G.A.R., unpublished). A survey of these mechanisms is presented in Figure 5.
Might other species of non-coding RNAs, including long non-coding RNAS (lncRNAs) also be involved? This issue has recently been reviewed . In brief, GWA studies have implicated several lncRNAs in T2D susceptibility (e.g. ANRIL and KCNQ1OT1). Combined epigenomic and transcriptomic profiling of human β cells has also identified >1100 lncRNAs expressed in these cells, of which around half were islet-specific . The potential for these lncRNAs to influence β-cell identity is underlined by the demonstration that one of these lncRNAs regulates the monogenic diabetes gene, GLIS3 .
COULD OTHER ISLET CELLS BE REPROGRAMMED TOWARDS A β-CELL IDENTITY IN VIVO?
There is growing excitement in the field at the prospect that islet non-β-cells may be ‘reprogrammed’ under some circumstances to replenish a depleted (or functionally-deficient) β-cell pool in diabetes (Figure 6). For instance, destruction of β-cells in mice, through the selective activation of a β-cell restricted diphtheria toxin, leads to repopulation with cells identified through lineage tracing to be of an α-cell origin . Very recent studies have shown that δ-cells can also serve as the pool of progenitor cells, and these are the predominant source in prepubertal mice . Likewise, the group of Patrick Collombat has shown that deletion of Arx, normally expressed in the α-cell and restricting cell fate to a glucagon-expressing phenotype, leads to their conversion into functional β-like cells . Likewise, duct cell conversion into β-cells has recently been described  (Figure 6).
β-Cell generation by reprogamming of non β-cells
RECENT ADVANCES: CONVERSION OF HUMAN iPS CELLS INTO β-CELLS FOR TRANSPLANTATION
Cells whose fate can be switched in vitro towards that of the β-cell provide, at least in theory, a near exhaustible supply of donor or patient-derived β-cells for transplantation and thus a ‘cure’ of T1D (provided that means are found to suppress or abrogate the action of the autoantibodies which led to disease onset in the first place). Examples include liver , intestinal K , human embryonic stem (ES) and induced pluripotent stem cells (iPSCs)  (Figure 6). Although it has been possible for a number of years to convert human ES cells into β-(like) cells–provided an in vivo stage using mice was included –only very recently have robust protocols been described through which the entire process can be achieved in vitro [221,222]. Importantly, although these cells express respectable levels of insulin and respond to glucose with important signalling events including increases in cytosolic Ca2+, it remains to be seen both how close these are to ‘genuine’ β-cells–i.e. with an ability to respond in vivo to the myriad complex signals described here, with complex patterns of insulin release (phases, pulses, etc.). Furthermore, this approach is costly, and likely to provide significant benefits for T1D but not T2D patients.
SUMMARY AND PERSPECTIVES
We provide here a description of some of the features which define the ‘gold standard’ β-cell, the processes which control the maintenance of this phenotype and its loss in disease states.
Might changes in β-cells identity contribute to the heritability of T2D? Although very few of the genes regulating β-cell identity described above have so far been identified in human genomic loci associated with T2D risk by genome-wide association studies (GWAS) , two genes close to such regions control secretory granule maturation (SLC30A8)  and GLP-1 receptor expression (TCF7L2) . Both of these characteristics define the mature β-cell.
Are there therapeutic ramifications of our increasing awareness of the importance of β-cell identity for normal secretory function? Certainly reversing the de-differentiation, and hence dysfunction, of β-cells in T2D would seem likely to provide a means of increasing insulin output, and might even be tailored to the needs (or genotype?) of individual patients: a method for quantifying β-cell mass, thus allowing such a treatment to focus on those where this is largely maintained, would be helpful. On the other hand, in the context of T1D, agents which drive non-β-cells towards a β-cell identity in vivo or in vitro may allow the β-cell complement to be replenished to restore insulin production.
acyl-CoA thioesterase 7
AMP-activated protein kinase
carnitine palmitoly transferase-1
non-esterified (free) fatty acids
ATP-sensitive K+ channel
T2D, Type 1 and Type 2 diabetes, respectively
We thank present and former colleagues for invaluable discussion, particularly Drs Andrei Tarasov, Isabelle Leclerc and Gabriela da Silva Xavier.
This work is supported by the Wellcome Trust Senior Investigator [grant number WT098424AIA]; the MRC Programme [grant number MR/J0003042/1]; the Diabetes UK Project Grant [grant number 11/0004210]; the Royal Society Wolfson Research Merit Awards Diabetes Research and Wellness Foundation Non-clinical Fellowship [grant number SCA/01/F/12 (to T.J.P.)]; Diabetes UK R.D. Lawrence Fellow [grant number BDA 12/0004431 (to D.J.H.)]; the Innovative Medicines Initiative Joint Undertaking [grant number 155005 (to IMIDIA)]; the European Union's Seventh Framework Programme [grant number FP7/2007-2013]; and EFPIA companies’ in kind contribution.