Summary points

• Non-coding polymorphisms near the CDKN2A/B locus influence risk of several type 2 diabetes-related forms of diabetes

• The CDKN2A/B locus encodes p16INK4A, p15INK4B, p14ARF and ANRIL, which regulate proliferation, oncogenesis, senescence and ageing

• p16INK4A mediates an ageing-related decline in beta cell proliferation in mice, and may regulate insulin secretion independently of beta cell mass

• CDKN2A/B and related genes also regulate adipocyte differentiation, inflammation, hepatic insulin clearance, fasted–fed transition and muscle metabolism

• Polymorphisms at CDKN2A/B have not yet been clearly linked to local gene expression in any metabolic tissue

• The mechanisms linking the CDKN2A/B locus to type 2 diabetes risk in human populations remain unknown

The CDKN2A/B locus impacts human diabetes risk

The CDKN2A/B locus, at chromosome 9p21, influences diabetes risk across varied ethnicities and geography [1–3], including people of European [4–9], Asian [10–24], Indian [25], Pakistani [26], Mexican [27–29] and Arab [30–32] descent. CDKN2A/B polymorphisms impact risk of gestational diabetes [33–35], early progression to type 2 diabetes after gestational diabetes [36], post-transplant diabetes [37, 38] and cystic fibrosis-related diabetes [39]. CDKN2A/B is not associated with type 1 diabetes risk [40–42], but is associated with rapid decline in beta cell function [43], and progression to diabetic nephropathy, in individuals with type 1 diabetes [44]. CDKN2A/B polymorphisms contribute only a fraction of observed heritable type 2 diabetes risk [2]. How CDKN2A/B influences diabetes risk remains uncertain [1–3].

CDKN2A/B is a hotspot influencing genetic risk for many diseases

The CDKN2A/B locus also influences risk for vascular conditions [1], including coronary artery disease [45–49], aneurysm and ischaemic stroke [50, 51], as well as glaucoma [52, 53], Alzheimer’s disease [54], endometriosis [55], periodontitis [56], ageing-related diseases [3, 57] and numerous cancers [58]. Intriguingly, the region of CDKN2A/B influencing type 2 diabetes risk is physically separated from regions contributing risk for other diseases, even for type 2 diabetes-related disorders such as cancer and cardiovascular disease [59–61].

General biology of CDKN2A/B locus genes

In order to consider how single nucleotide polymorphisms (SNPs) in the CDKN2A/B region may impact type 2 diabetes risk, an important starting point is the known biology of local genes.

The CDKN2A/B locus encodes three coding transcripts and a long non-coding RNA

The human CDKN2A/B locus (Fig. 1) encodes three proteins: p14 alternate reading frame (p14ARF) (p19ARF in mice), p15 and p16 inhibitors of cyclin dependent kinase 4 (p15INK4B and p16INK4A), and a long non-coding RNA (lncRNA) called ANRIL (also known as CDKN2B-AS) [1, 62, 63]. p16INK4A and p14ARF, encoded by the CDKN2A gene, share common second and third exons but have different first exons and promoters about 20 kb apart [64]. p16INK4A and p14ARF are in alternate reading frames, resulting in unrelated peptide sequences despite the common mRNA sequence. CDKN2B, about 30 kb from CDKN2A, encodes p15INK4B [65, 66]. Smaller splice variants of both p16INK4A (p12) and p15INK4B (p10) have been described [64, 67, 68]. ANRIL, which overlaps the p14ARF promoter and two exons of p15INK4B [69], transcribed by RNA polymerase II [70] in the antisense direction to CDKN2B, is spliced into linear or circular isoforms [69, 71, 72]. The mouse Cdkn2a/b locus, on chromosome 4, encodes p16INK4A, p19ARF and p15INK4B in a similar arrangement to the human locus, but with a different lncRNA called AK148321 in a position similar to ANRIL [73].

Fig. 1
figure 1

The human CDKN2A/B locus at 9p21 contains genes encoding p16INK4A, p14ARF, p15INK4B and the lncRNA ANRIL. CDKN2A encodes both p16INK4A and p14ARF, which share exons 2 and 3 but in different reading frames, producing unrelated peptides. Polymorphisms influencing type 2 diabetes risk are physically separate from the coronary artery disease risk interval. Diagram not to scale. CAD, coronary artery disease; T2D, type 2 diabetes

Full size image

p14ARF, p15INK4B and p16INK4A are cell cycle inhibitors

CDKN2A/B proteins block cell cycle progression and influence tumorigenesis, senescence and ageing [62, 63]. p16INK4A and p15INK4B are cyclin-dependent kinase (CDK) inhibitors that prevent activation of CDK4/6 by D-cyclins (Fig. 2). CDK4/6 phosphorylates retinoblastoma (Rb); hypophosphorylated Rb represses early region 2 transcription factor (E2F) to prevent cell cycle entry [74]. p14ARF, also anti-proliferative, acts by stabilising the tumour suppressor p53 by sequestering its negative regulator, mouse double minute 2 homologue (MDM2) [75]. p53 reduces cell cycle entry through the CDK-interacting protein/kinase inhibitory protein (CIP/KIP) family inhibitor p21. The p10 and p12 splice variants of p16INK4A and p15INK4B also inhibit the cell cycle, but through slightly different mechanisms [64, 67]. ANRIL has widespread influences on gene expression, and impacts the cell cycle by regulating the expression of p14ARF, p15INK4B and p16INK4A [73].

Fig. 2
figure 2

INK family inhibitors (p15INK4B, p16INK4A, p18INK4C and p19INK4D) bind CDK4/6 and prevent CDK binding and activation by D-cyclins. Hypophosphorylated Rb sequesters E2Fs; active (cyclin-bound) CDK phosphorylates Rb, de-repressing transcriptional activity of activating E2Fs to promote cell cycle entry

Full size image

CDKN2A/B and tumorigenesis, senescence and ageing

The CDKN2A/B locus is inactivated by deletion, methylation or mutation in many cancers; CDKN2A/B aberrations correlate with advanced tumour stage and reduced overall and disease-free survival [58, 76]. Conversely, restoring p16INK4A suppresses tumour growth [77]. CDKN2A was the first familial melanoma gene identified [78]; germline loss-of-function CDKN2A mutations are the most frequent genetic events underlying familial melanoma susceptibility [79]. Activating CDK4 mutations (R24C and R24H) that prevent inhibition by p16INK4A increase melanoma risk [80]. Intriguingly, loss of p16INK4A also contributes to pancreatic neuroendocrine tumours, which include islet tumours such as insulinomas [81–85]. Paradoxically, when not deleted, p16INK4A may be overexpressed in tumours and transformed cells [86, 87]. p14ARF loss is oncogenic as well, via MDM2/p53 and other mechanisms [62]. p15INK4B may be co-deleted with p16INK4A in haematological malignancies [62], and plays a role in TGF-β-induced pancreatic cancer [88, 89]. ANRIL is also implicated in many cancer pathways, downstream of oncogenic Ras, phospholipase D and specificity protein 1 (SP1), and upstream of Kruppel-like factor 2, p21 and CDKN2A/2B [90–93].

p16INK4A is an effector of senescence, an irreversible growth arrest that occurs when a cell reaches the end of its replicative lifespan [63, 94]. p16INK4A expression increases with age [62], triggered by ageing-dependent gene demethylation, telomere shortening and other senescence inducers such as oncogenic activation, oxidative stress, nutrient deprivation and DNA damage [95, 96]. Mice and humans with the R24C activating mutation of CDK4 are resistant to senescence [97]. A recent study has shown that ablating senescent cells, as defined by p16INK4A expression, extends longevity and forestalls ageing-related tissue functional decline [98]. p14ARF is also linked to senescence, independent of p16INK4A [99].

Evidence that CDKN2A/B influences type 2 diabetes risk via beta cell mass and proliferation

The best-known role of CDKN2A/B gene products in a metabolic tissue is the ageing-related negative impact of p16INK4A on beta cell proliferation and regeneration [100, 101]. The critical dependence of mouse beta cell mass on CDK4 further supports the importance of this pathway in islet biology. Here, we will review the known roles of CDKN2A/B genes in islets.

Effects of CDKN2A/B -related proteins on metabolic tissues

Tissue

Protein

Effects

Islets

p16INK4A

Restricts beta cell proliferation in ageing, restricts beta cell regeneration, mediates overnutrition-related senescence, reduces insulin secretory function

CDK4

Required for postnatal beta cell mass expansion

Cyclin D1, D2

Required for postnatal beta cell mass expansion

Adipose

p15INK4B

Inhibits adipocyte differentiation

p16INK4A

Modulates adipose macrophage activation and polarisation

CDK4

Promotes adipocyte differentiation

Liver

p16INK4A

Restrains hepatic gluconeogenesis

Cyclin D1/CDK4

Regulates fasted–fed transition

Muscle

CDK4

Impacts mitochondrial oxidative metabolism via E2F1

Beta cell proliferation decreases with age

Type 2 diabetes incidence increases with age, related to declining beta cell proliferation, beta cell function and/or insulin sensitivity [102–105]. Human autopsy studies report an age-dependent reduction in beta cell proliferation [102, 106, 107]. The importance of this has been contested on the basis that proliferation is rare even in young humans, such that a loss with ageing is not meaningful [108]. The demonstration that some measures of proliferation on autopsy specimens do not accurately reflect in vivo proliferation raises questions about the validity of postmortem analyses [109]. Beta cell mass is reported to be unchanged, or only modestly reduced, in ageing Japanese [108, 110]. As human beta cells may be extremely long lived, maintenance of beta cell mass may not require much ongoing proliferation [103]. Adult beta cell mass is thought to depend on early postnatal beta cell proliferation [106, 107]. In rodents, an age-associated decline in proliferation is clear [105]. Intriguingly, old islets exposed to young circulation via parabiosis or transplantation recovered a youthful proliferation frequency, suggesting the loss of proliferation is due to a circulating factor [111]. However, the rate of human beta cell proliferation is lower than that of rodents, and engrafting human islets into mice does not increase human beta cell proliferation to the rodent frequency [112, 113].

p16INK4A expression in rodent and human islets increases with age

Given the low beta cell replication frequency, cell cycle inhibitors have been a focus in this field. p16INK4A is expressed in mouse islets [101, 114–126]. Telomere shortening impacts islet p16INK4A expression as it does in other tissues [121]. The abundance of p16INK4A in young healthy islets is low, however, at or below the limit of detection in our hands by quantitative PCR using primers that do not amplify p14ARF (data not shown). Mouse islet p16INK4A abundance increases with ageing [101, 116, 117, 119, 121, 124–126]. p16INK4A is expressed in human islets [110, 127–132] in an age-dependent fashion [128], possibly related to progressive CDKN2A locus demethylation with age [96]. In human pancreas, nuclear p16INK4A staining was occasionally present in fetal beta cells, but was evident in 63% of adult beta cells [127]; p16INK4A staining was significantly lower in younger (age 0–9 years) than older (age 10–59 and 60–79 years) samples [110].

Regulation of islet p16INK4A expression and subcellular localisation

Islet p16INK4A expression is regulated at the RNA level via epigenetic histone modification by the polycomb repressors enhancer of zeste 2 (EZH2) and B cell-specific Moloney murine leukaemia virus insertion site 1 (BMI1). BMI1 and EZH2 bind the p16INK4A promoter to increase H3K27 trimethylation and reduce histone 2A (H2A) ubiquitination and H3K4 trimethylation via trithorax group (TrxG) mixed lineage leukaemia protein 1 (MLL1) [116, 119]. BMI1 and EZH2 occupancy of the p16INK4A promoter is reduced with ageing, resulting in de-repression of gene expression. Knockdown of EZH2 reduced BMI1 presence at the p16INK4A promoter, suggesting cooperative recruitment [116]. Overexpression of EZH2 suppressed p16INK4A expression in young but not old mice; in aged islets, MLL1 prevented EZH2 recruitment to the p16INK4A promoter [133]. EZH2 expression in human islets is inversely related to age [119]. These observations may apply to p16INK4A and p14ARF, but not p15INK4B [119].

Insulin signalling pathways impact islet p16INK4A expression. BMI1 is regulated by p38 mitogen-activated protein kinase (p38MAPK). Mice expressing an activation-resistant allele of p38Mapk (also known as Mapk14) had reduced abundance of p15INK4B, p16INK4A and p14ARF in aged islets; age-dependent BMI1 repressor loss at the p16INK4A promoter was reduced, possibly through p38MAPK target MAPK-activated protein kinase 3 (MK3) phosphorylation of BMI1 [134]. The age dependence of p38 activation was due to reduction of the wild-type p53-induced phosphatase 1 (WIP1) p38 phosphatase with age. Phosphoinositide 3-kinase (PI3K) signalling also influences p16INK4A expression; mice with phosphatase and tensin homologue (PTEN) deficiency had increased p16INK4A expression via a pathway involving cyclin D1, E2Fs and EZH2 [124, 125].

The protein subcellular localisation of cell cycle regulators likely impacts activity. In human beta cells, directly driving the cell cycle by overexpressing cyclin D3 and CDK6 caused p16INK4A, but not p15INK4B, to shift from the cytoplasm to the nucleus [130]. Nuclear relocalisation of p16INK4A depended on culture duration; at shorter time points after transduction (24–48 h) p16INK4A was less likely to be nuclear than in control cells, but after 72 h p16INK4A was more likely to be nuclear. The proportion of beta cells with nuclear p16INK4A expression increased with culture duration in control islets as well. Human islet cells undergo growth arrest after 10–15 divisions, with shortened telomeres and increased p16INK4A expression [135]. Taken together, this suggests that driving the cell cycle by activating CDK6 in human beta cells may hasten culture-related senescence.

p16INK4A mediates ageing-related loss in beta cell proliferation

Overexpression of p16INK4A reduced beta cell proliferation in young mice, and deletion of p16INK4A rescued the age-related loss of proliferation [101]. Mice lacking BMI1 or EZH2, such that p16INK4A is prematurely induced, have reduced beta cell proliferation; in the case of EZH2, dependence of proliferation loss on p16INK4A was confirmed [117, 119]. Overexpression of EZH2 repressed p16INK4A and increased proliferation in young mice; in older mice the same effects were seen but only if MLL1 was also reduced [133]. Ex vivo, knockdown of p16INK4A using small interfering (si)RNA rescued the loss of proliferation induced by NEFA exposure [123]. Alterations in the WIP1/p38MAPK/BMI1 or PTEN/E2F/EZH2 pathways that reduced p16INK4A expression increased beta cell proliferation in ageing mice [124, 134]. Loss of p16INK4A and a related INK family inhibitor, p18INK4c, synergistically increased beta cell proliferation in a CDK4-dependent manner [136]. p15INK4B also impacts beta cell proliferation; transgenic overexpression of TGF-β in alpha cells resulted in pancreas and islet hypoplasia associated with increased islet expression of p15INK4B [137].

p16INK4A restricts beta cell regeneration

The impact of ageing on beta cell regeneration is an area of controversy. Beta cell proliferation after partial pancreatectomy, exendin-4 treatment, high-fat feeding or streptozotocin ablation is lower in ageing mice than in young controls [101, 104, 117]. On the other hand, after diphtheria toxin ablation or glucokinase activation mice even beyond 2 years of age retained capacity for regenerative proliferation [138]. p16INK4A mediates age-related loss of islet regenerative capacity [101]. Loss of BMI1 and EZH2 impair beta cell proliferation in response to regenerative stimuli [117, 119]. BMI1 may also promote exocrine pancreas regeneration [139, 140]. A synthetic antagonist of hepatocyte nuclear factor (HNF)4α increased beta cell proliferation by suppressing the expression of CDK inhibitors including p16INK4A; inhibiting HNF4α induced alpha, beta and delta cell proliferation after beta cell ablation [141]. Beta cell proliferation in response to parathyroid hormone-related peptide (PTHrP) was associated with decreased expression of p16INK4A [122]. On the other hand, increased islet expression of p16INK4A may have prevented the expected increase in beta cell proliferation with overexpression of activated S6 kinase [120].

CDK4: critical for postnatal mouse beta cell mass

INK family inhibitors, which include p15INK4B and p16INK4A as well as p18INK4c and p19INK4d, bind to and inhibit CDK4/6 (Fig. 2) [74]. The essential role for CDK4 in mouse islet biology supports the importance of INK family inhibitors in islets. Mice lacking CDK4 develop severe insulin-deficient diabetes because of hypoplastic islets [142]. Intriguingly, CDK4 deletion also impacts other endocrine systems: male and female infertility, and poor growth, are possibly related to pituitary defects. Islet morphology is normal at birth, suggesting CDK4 is not required for pancreatic development, but islets fail to expand during postnatal growth. Conversely, mice expressing the INK-resistant mutant CDK4 ‘R24C’ have hyperplastic islets due to increased postnatal beta cell proliferation [142, 143]. Mutation of the arginine at CDK4 position 24 is oncogenic in many tissues, despite also having impaired binding to D-cyclins and reduced efficacy as an Rb kinase [144]. Rodent insulinoma cell lines have increased expression of CDKs including 4 and 6; overexpressing CDK4/6 in rat islets increased proliferation [86].

D-cyclins and mouse beta cell proliferation

Cyclins D1, D2 and D3 activate CDK4/6 kinases, linking external growth signals with cell cycle regulation. INK family inhibitors bind CDKs near the D-cyclin binding site, preventing D-cyclin binding and kinase activation. Consistent with the importance of CDK4 in islets, D-cyclins also regulate beta cell proliferation in mice. Cyclin D2 and, to a lesser extent, cyclin D1, are required for postnatal beta cell proliferation [145, 146]. Like CDK4, D-cyclins are not required for prenatal pancreatic development. Upstream signalling events that promote beta cell proliferation, such as nutrient excess, insulin signalling and Wnt signalling, increase D-cyclin abundance in islets [122, 123, 147–155]. Islet overexpression of a stabilised cyclin D2 increased islet mass, although proliferation was not increased at the time points tested [151].

Metabolic insults may cause premature islet ageing

Metabolic stress influences ageing-related markers in general, and p16INK4A expression specifically. Failed human islet grafts in diabetic nude mice showed widespread p16INK4A immunostaining [129]. Energy restriction, an intervention that reduces metabolic load and delays ageing, decreased ageing-related p16INK4A expression in many tissues [126]. On the other hand, intrauterine protein-energy malnutrition increases type 2 diabetes risk. In rats, in utero nutrient deprivation reduced postnatal beta cell proliferation and led to premature ageing, with telomere shortening and induction of senescence markers including p16INK4A [156]. Maternal islets are also sensitive; postpartum, but not never-pregnant, female mice treated with the endocrine disruptor bisphenol-A developed glucose intolerance related to lost beta cell mass and insulin secretory function, with increased p16INK4A and p53 and reduced cyclin D2 and CDK4 [118].

The impact of lipid excess on beta cell proliferation remains controversial [123, 157, 158]. In vivo or ex vivo exposure to glucolipotoxicity reduced proliferation and induced islet p16INK4A expression in young mice [123]. Senescence markers in islets heralded metabolic decompensation after long-term high-fat feeding [159]. In contrast, short-term intravenous infusion with glucose and lipid in rats did not induce p16INK4A expression under conditions where beta cell proliferation was high [158]. Nutrition deprivation followed by nutrient excess induced islet senescence markers in rats [156]. The impact of nutrients on islet p16INK4A expression and senescence may depend on other factors related to development, genetics, duration of insult, overall stress load and stress tolerance.

Differences between cell cycle regulation in mouse and human beta cells: do D-cyclins, CDK4/6 or p16INK4A influence human beta cell proliferation?

Surveys of cell cycle regulator expression report similar patterns for rodent and human islets; of 34 proteins tested, only CDK6, E2F3, E2F7 and E2F2 show differences, with CDK6, E2F3 and E2F7 present in human islets but absent in mouse islets, and the converse for E2F2 [132, 160, 161]. Some reports suggest cyclin D2 may be less abundant than D1 and D3 in human islets [128, 132, 162]; however, recent reports link the CCND2 genomic locus, which encodes cyclin D2, with human type 2 diabetes risk [163–165]. A low-frequency non-coding variant at CCND2 reduced type 2 diabetes risk by about half in European populations; risk reduction was due to increased insulin secretion [164]. The impact of CCND2 was sex dependent, with a larger effect in men than women [166]. In ex vivo human islet cultures D-cyclins and CDK4/6 were detected in the cytoplasm rather than the nuclear compartment [131]; the same was true in rat islet cells for those cell cycle activators tested (cyclins D1/D2 and CDK2) [130]. Overexpression of D-cyclins was not as effective as CDKs in increasing human beta cell proliferation, this was perhaps related to the cytoplasmic location of CDKs, the abundance of INK family inhibitors [131, 162] or synergy with nutrient signals [167]. Data linking p16INK4A to reduced proliferation in human beta cells are somewhat limited. In autopsy material, p16INK4A was never detected in actively proliferating beta cells [127], and p16INK4A staining inversely correlated with beta cell mass [110]. Ex vivo human beta cells with nuclear p16INK4A after overexpression of cyclin D3 and CDK6 did not stain for the proliferation marker Ki67, despite 10–20% of beta cells in these cultures staining for Ki67 [130]. Knockdown of p16INK4A in the transformed EndoC-bH1 human beta cell line did not impact proliferation; however, transformation bypasses normal cell cycle regulation [168].

CDKN2A/B gene products may impact type 2 diabetes risk via other mechanisms

Evidence suggesting p16INK4A does not increase type 2 diabetes risk by reducing islet mass

Several observations contradict the concept that p16INK4A impacts diabetes incidence via effects on beta cell mass. The combined loss of p16INK4A and p14ARF was not sufficient to restore the glucose induction of mitotic genes lost in aged islets [169]. Old islets that recovered proliferation when exposed to young mouse circulation did not have reduced p16INK4A expression [111]. Perhaps most puzzling from an islet-centric viewpoint are observations from the Super-Ink4/ARF mouse, which contains an extra copy of the entire Cdkn2a/b locus [170]. Despite modestly increased expression of p15INK4B, p16INK4A and p14ARF in several tissues, these mice have improved glucose tolerance with ageing from enhanced insulin sensitivity in liver and muscle. Beta cell proliferation and islet number are unaffected. Reduced insulin secretion is observed, but is likely to be secondary to the improved insulin sensitivity. Also curious, a survey of islet gene expression at type 2 diabetes-linked loci in diabetes-prone New Zealand Obese (NZO) mice vs diabetes-resistant ob/ob mice revealed increased p15INK4B and p16INK4A expression in the diabetes-resistant mice [115].

p16INK4A may influence insulin secretory function, insulin clearance and insulin sensitivity

Some observations suggest an impact on insulin secretory function independent of the effects on beta cell mass (see the text box ‘Effects of CDKN2A/B-related proteins on metabolic tissues’). Knockdown of p16INK4A in the EndoC-bH1 human beta cell line increased insulin secretion [168]. In mice, haploinsufficiency for telomerase increased p16INK4A expression and impaired insulin secretion via altered regulation of exocytosis. These mice were glucose intolerant even though beta cell mass was normal, suggesting a primary effect on beta cell function rather than proliferation [121]. CDK4 also regulates insulin secretion, through Rb-dependent transcriptional regulation of potassium inward rectifying channel 6.2 (Kir6.2) [171]. In human studies, individuals from familial melanoma kindreds with heterozygous loss of function of CDKN2A had increased insulin secretion, impaired insulin sensitivity and reduced hepatic insulin clearance [168].

CDKN2A/B locus genes impact adipose, liver and muscle

CDKN2A/B locus genes and their CDK targets influence adipose, liver and muscle biology. Deletion of a region of mouse chromosome 4, orthologous to the human 9p21 cardiovascular disease risk interval, reduced local gene expression and increased body weight, linking the region with obesity and metabolic risk [172]. ANRIL regulates genes involved in glucose and fatty acid metabolism [173]. p15INK4B is highly expressed in subcutaneous adipose tissue and may inhibit expandability of the subcutaneous fat depot, a critical protection against overnutrition toxicity [174, 175]. Mice lacking CDK4 in all tissues are smaller, with less fat mass, and have insulin resistance in addition to insulin deficiency [176]. CDK4 regulates adipogenesis, via both Rb-dependent and Rb-independent mechanisms, including direct phosphorylation of insulin signalling intermediates [177–179]. CDKN2A/B locus genes also impact adipose tissue inflammation; p16INK4A modulates the activation and polarisation of adipose-associated macrophages [180], although deleting p16INK4A in bone marrow-derived cells did not impact glucose metabolism, even under obese conditions [181]. In liver, p16INK4A regulates hepatic gluconeogenesis independently of the cell cycle [182]. p16INK4A deficiency increased liver glucose production via a protein kinase A (PKA)-mediated induction of gluconeogenic gene expression. CDK4 participates in the hepatocyte fasting–fed transition by phosphorylating and activating the general control of amino acid synthesis protein 5-like 2 (GCN5) histone acetyltransferase, which regulates PPARG coactivator 1 α (PGC-1A) and hepatic glucose metabolism [183]. CDK4 also impacts muscle mitochondrial oxidative metabolism [184]. These observations, collectively, greatly broaden the potential mechanism(s) by which CDKN2A/B polymorphisms might impact diabetes risk.

How might CDKN2A/B polymorphisms influence type 2 diabetes risk?

Type 2 diabetes SNPs at CDKN2A/B are non-coding, and high-resolution mapping of the CDKN2A/B locus did not reveal SNPs with greater disease association than known SNPs [185]. Attention has thus turned to the regulation of local genes.

Do CDKN2A/B polymorphisms influence local gene expression?

Testing whether genomic polymorphisms alter mRNA abundance is performed by expression quantitative trait locus (eQTL) analysis. Thus far, however, eQTL analysis for the CDKN2A/B locus has failed to produce major mechanistic breakthroughs. Type 2 diabetes SNPs at CDKN2A/B were not associated with expression of CDKN2A or CDKN2B in pancreas, liver or colon [186], nor in pancreatic islets themselves [187]. CDKN2A/B SNPs impacting cardiovascular risk, outside the type 2 diabetes region, influence ANRIL expression [71, 188]. Both type 2 diabetes and non-type 2 diabetes SNPs at this locus appear to have a stronger effect on ANRIL expression than CDKN2A or CDKN2B expression [189]. Expression of ANRIL is reported to correlate with expression of CDKN2A and CDKN2B in many tissues, suggesting coordinated local regulation [69, 71, 76, 189], although independent regulation is also reported [76, 126]. Deletion of a large region including the type 2 diabetes-risk locus plus part of ANRIL reduced CDKN2A and CDKN2B expression in several vascular-relevant cell types in mice [172]. ANRIL regulates gene expression at the CDKN2A/B locus by recruiting polycomb proteins chromobox homologue 7 (CBX7 [PRC1]) and suppressor of zeste 12 (SUZ12 [PRC2]) to modulate epigenetic repression by H3K27 methylation [70, 72, 73, 90, 190]. ANRIL can also regulate distant genes with retrotransposon Alu repeats in their promoters [191]. In sum, existing eQTL analyses do not identify a mechanism by which polymorphisms at CDKN2A/B influence type 2 diabetes risk through local gene expression, but studies may have been performed in the wrong cell type, developmental, environmental or nutritional state to identify the point of activity. The requirement to perform these studies in human tissues, the difficulty in obtaining human samples, and caveats introduced through postmortem state, tissue collection and ex vivo culture are barriers to this type of study.

CDKN2A/B polymorphisms are located in regulatory enhancers

Cis-regulatory elements are found near the type 2 diabetes-risk interval [192–194]. Disease-associated SNPs are predicted to disrupt transcription factor binding sites, including some factors with known roles in beta cell development or survival, including, v-maf avian musculoaponeurotic fibrosarcoma oncogene homologue B (MAFB), NK homeobox protein 6.1 (NKX6.1), nuclear factor of activated T lymphocytes (NFAT), FOXA2, forkhead box A2 (FOXA2), nuclear factor κβ (NFκB), hepatocyte nuclear factor 1 (HNF1) and CCAAT-enhancer-binding protein homologous protein (CHOP) [90, 192, 194]. Polymorphisms may also impact microRNA regulation of transcription or translation [195], although most SNPs at the type 2 diabetes locus are non-coding. An enhancer identified in tumours, RD INK4/ARF, interacts with oncoproteins to silence the CDKN2A/B locus [196, 197]. Deletions of RD INK4/ARF have been detected in pancreatic neuroendocrine tumours, implying activity in islet cells [198].

Do polymorphisms influence epigenetic regulation at the CDKN2A/B locus?

Epigenetic modification regulates gene expression at this locus in islets [116, 119]. The type 2 diabetes-associated rs564398 removes a DNA methylation CpG site, reducing methylation of other local CpG sites and decreasing insulin content in human islets, although without impacting local gene expression [199]. Histone modifications can be influenced by polymorphisms as well; many epigenetic modifiers act at the CDKN2A/B locus [200]. Metabolic inputs, such as overfeeding or energy restriction, influence CpG methylation and histone modification at CDKN2A/B in humans [201, 202]. Whether this is influenced by type 2 diabetes-risk polymorphisms remains unknown. Taken together, although many mechanisms exist by which polymorphisms at CDKN2A/B might influence local gene expression, none has yet been proved.

Evidence that human CDKN2A/B-related type 2 diabetes risk involves beta cells

CDKN2A/B genotype influences insulin secretory capacity

Many type 2 diabetes risk polymorphisms, including at CDKN2A/B, may impact human diabetes risk by altering insulin secretory capacity [203]. Age influences the association between CDKN2A/B polymorphism rs10811661 and type 2 diabetes, consistent with the known interaction between age and p16INK4A activity in islets [204]. The rs10811661 ‘T’ (risk) allele is associated with reduced insulin secretion after both oral and intravenous glucose challenge [205, 206]. In Europeans undergoing hyperglycaemic clamp, although CDKN2A/B was not independently associated with beta cell function, a composite including CDKN2A/B among eight loci predicted reduced first-phase insulin secretion [207]. In this cohort, each additional risk allele lowered glucose-stimulated insulin secretion (GSIS) by 5%, but no effect was seen on insulin sensitivity. The detrimental effect on insulin secretion was of similar magnitude in individuals with normal glucose tolerance and impaired glucose tolerance, suggesting CDNK2A/B impact pre-dates metabolic decompensation. In a large meta-analysis including more than 58,000 individuals, CDKN2A/B was associated with insulinogenic index, acute insulin response, fasting glucose or HOMA-B, but not with insulin sensitivity index, fasting insulin, fasting proinsulin, or HOMA-IR [208]. A parallel study of whether CDKN2A/B genotype impacts islet function in vivo and ex vivo revealed an in vivo association with reduced insulin secretion and disposition index, but not insulin sensitivity or glucagon secretion, but no striking impact on ex vivo function [209]. Individuals with reduced p16INK4A or p16INK4A/p14ARF activity have increased basal and stimulated insulin secretion [168].

Little evidence for or against a role in CDKN2A/B in human beta cell mass accrual

The observed reduction in insulin secretory capacity with CDKN2A/B polymorphisms could be due to reduced beta cell function (glucose sensing, insulin production, stimulus-secretion coupling) or reduced beta cell mass. Despite the data linking p16INK4A to beta cell mass in mice, few studies have assessed whether CDKN2A/B genotype impacts human beta cell proliferation or mass. A limitation is that human beta cell mass and proliferation cannot easily be measured in living people. In a single study, CDKN2A/B polymorphisms were not associated with AIRmax, a surrogate for beta cell mass [207]. On the other hand, the contribution of p16INK4A loss to pancreatic neuroendocrine tumour risk may imply a role in human islet cell proliferation [81–85].

Evidence that CDKN2A/B influences type 2 diabetes risk via non-islet mechanisms

Evidence that CDKN2A/B increases diabetes risk without impacting islet function

CDKN2A/B was not associated with proinsulin conversion to insulin [210] or the effect of ambient glycaemia on insulin secretion [211]. Although CDKN2A/B significantly impacted risk of diabetes in a Han Chinese population, this was not related to reduced HOMA-B, although other loci (CDKAL1, IGF2BP2 and SLC30A8) did impact beta cell function in this population [212]. In a relatively young cohort of European individuals tested by OGTT, CDKN2A/B was not associated with insulin secretion or glucose sensitivity, although CDKAL1 and HHEX were [212]; intriguingly, in this cohort two of three CDKN2A/B risk alleles tested (rs10757283 and rs564398 but not rs10811661) showed a trend towards reduced insulin sensitivity. In a hyperglycaemic clamp study, CDKAL1 and IGF2BP2, but not CDKN2A/B, were associated with reduced first-phase insulin secretion [213]. Complicating matters, the influence of CDKN2A/B type 2 diabetes polymorphisms on insulin secretion may depend on ambient insulin sensitivity [214].

CDKN2A/B impacts beta cell response to particular diabetes therapies

In a post-hoc analysis of the Diabetes Prevention Program (DPP), CDKN2A/B identity was not related to baseline insulin secretory capacity; however, unique among the eight loci tested, CDKN2A/B predicted response to therapy. Insulin secretion, but not insulin sensitivity, was improved in protective-allele carriers, but only in the thiazolidinedione treatment arm [215]. In another pharmaco-genetics study, of 27 diabetes loci tested, only CDKN2A/B rs10811661 predicted response to therapy: the protective allele was associated with greater response to sulfonylurea therapy [216]. The CDKN2A/B locus may [217] or may not [218] influence the metabolic response to exercise.

Evidence that CDKN2A/B impacts diabetes risk through other metabolic tissues

The intriguing relationship between CDKN2A/B genotype and response to thiazolidinedione in the DPP cohort suggests activity in adipose tissue; as lipids negatively impact pancreatic beta cell proliferation and function this may represent an adipocyte–beta cell axis [123, 157, 215]. In a Japanese population, CDKN2A/B was not related to visceral fat accumulation [219], but in an Indian sibling-pair study, CDKN2A/B polymorphisms impacted fasting insulin and HOMA-IR but not HOMA-B, suggesting a primary effect on insulin sensitivity [220]. The melanoma kindred study also found that individuals haploinsufficient for p16INK4A had impaired insulin sensitivity [168]. In the Helsinki Birth Cohort, CDKN2A/B type 2 diabetes-risk polymorphisms were associated with reduced birthweight, suggesting the possibility that CDKN2A/B diabetes risk might be related to developmental impact of this locus, bringing an additional temporal variable to play [221].

Type 2 diabetes CDKN2A/B polymorphisms are not yet clinically useful

Initial attempts to use type 2 diabetes genome-wide association study (GWAS) information to predict disease risk, optimise therapeutic impact and estimate prognosis have been disappointing. Even combining risk alleles only marginally improves type 2 diabetes prediction over clinical factors [222–224]. As described above, some progress has been made using the CDKN2A/B genotype to predict response to therapy [215, 216]. Although CDKN2A/B polymorphisms are not generally associated with longevity [225], cardiovascular mortality was paradoxically reduced in individuals homozygous for the type 2 diabetes-risk allele at rs10811661 [226]. In sum, despite progress, genotype information at the CDKN2A/B locus does not yet have meaningful clinical implications for individual patients.

Future directions and unresolved questions

Great progress has been made in identifying roles played by CDKN2A/B gene products in islets and other metabolic tissues, mostly in rodents but also in humans. p16INK4A and related proteins are critical regulators of rodent beta cell mass, but whether human CDKN2A/B polymorphisms influence type 2 diabetes risk via islet biology remains uncertain. Gene regulation analyses have not yet proved a relationship between type 2 diabetes SNPs and local gene expression, but the analyses to date may not have been performed in the relevant tissue, developmental stage, metabolic milieu and/or human subpopulation. CDKN2A/B genes impact islet, adipose, muscle, liver and immune cell function, at stages ranging from in utero development to ageing. Human biological variation likely influences CDKN2A/B effects. The availability of human samples across tissues and stages is a serious limitation in this field. We are hopeful that, in the future, CDKN2A/B polymorphisms will improve diabetes understanding and inform clinical decisions.