The MafA Transcription Factor Becomes Essential to Islet β-Cells Soon After Birth

The pancreatic β-cell is one of several hormone-producing cell types critical to glucose homeostasis present within the islet of Langerhans. However, hyperglycemia is specifically caused by the loss or inability of β-cells to synthesize and secrete the insulin hormone, causing diabetes mellitus, a life-threating condition associated with millions of people worldwide. Current therapies for diabetes involve external insulin treatment either through injection or a pump, but this still often results in long-term medical complications, including cardiovascular disease, chronic renal failure, and retinopathy. These complications are likely due to the inability of exogenously supplied insulin to regulate blood glucose levels in the same facile manner as endogenous β-cells. As a consequence, efforts have been focused on understanding the molecular mechanisms underlying normal β-cell development and function, with the hope that such knowledge will provide insight into therapeutic ways of either sustaining endogenous β-cell function or engineering replacement cells.

Genetic and biochemical studies have demonstrated that islet-enriched transcription factors play a central role in orchestrating the intricate series of events required in the biosynthesis of islet hormone-producing α (i.e., glucagon), β, δ (somatostatin), ε (ghrelin), and pancreatic polypeptide cells (reviewed elsewhere [1–3]). The functional significance of these proteins was reinforced upon observing that six of nine genes associated with a monoallelic form of diabetes, maturity-onset diabetes of the young, encode for transcription factors essential to islet cells (reviewed elsewhere [4,5]). Moreover, altering the expression of individual islet-enriched transcription factors often influences the presence and/or activity of a particular islet cell type (e.g., Arx [6], Pax4 [7], Pdx1 [maturity-onset diabetes of the young, type 4] [8]). In addition, the combinatorial actions of these proteins are capable of reprogramming nonislet cells to β-like cells, exemplified by misexpression of MafA, Pdx1, and Ngn3 in pancreatic acinar cells (9).

Interestingly, members of the same transcription factor family often contribute to β-cell formation, including winged-helix/forkhead-containing FoxA1 and FoxA2 (10), zinc finger Gata4 and Gata6 (11,12), NK6 homeodomain Nkx6.1 and Nkx6.2 (13), paired box homeodomain Pax4 and Pax6 (7,14), as well as basic leucine-zipper MafA and MafB (15). FoxA1/2, Nkx6.1/6.2, Gata4/6, and Pax4/6 are expressed broadly in early pancreatic epithelial cells and then become principally confined to islet cells (e.g., Nkx6.1 [β], FoxA1/A2 [all], Gata6 [all], Pax6 [all]) or cease to be expressed after birth (Gata4, Pax4, Nkx6.2) (2,3). However, MafA and MafB are distinct in being produced relatively late during development and primarily (i.e., MafB [16]: α, β, Ngn3⁺ islet cell progenitors), if not exclusively in hormone⁺ cells (β only, MafA [17]). Consequently, analysis of Mafa and Mafb mutant mice has revealed more specialized roles in islet cell maturation than other islet-enriched transcription factors. Thus islet cell identity and/or numbers are often profoundly affected in islet-enriched transcription factor knockout mice (1–3), whereas only terminal β- and α-cell maturation is defective in Mafb^−/− embryos (16,18). In contrast, islet cell development was unchanged in Mafa^−/− (19) or pancreas-specific knockout Mafa^∆panc (20) mice, although glucose-regulated insulin secretion and islet cell architecture was compromised in adults. Remarkably, human embryonic-stem-derived insulin⁺ cells do not become glucose responsive until the appearance of MafA (21,22), with induction currently requiring transplantation of in vitro programmed cells into mice (22).

Rodent β-cells switch from producing primarily MafB during development to exclusively MafA postnatally (18,20). Hence, MafB is expressed in all embryonic glucagon⁺ cells and insulin⁺ cells produced during the first and secondary wave of hormone⁺ cell formation and then disappears from β-cells roughly 2 weeks after birth (16,18). In contrast, MafA is produced exclusively in the second or principal wave of insulin⁺ cells arising from embryonic (E) day 13.5 (17), which become the β-cells of the islet. Notably, MafA controls many embryonically activated MafB genes in adults (20). Here we have more precisely defined the functional interrelationship between these closely related large Maf proteins by answering the following questions: 1) When does MafA become important to postnatal β-cells? 2) Is transient expression of MafB postnatally significant to islet β-cell activity? 3) How does MafA regulate islet β-cells? 4) Are there genes regulated by MafA in adults that are unaffected by MafB during development?

A time course study performed with mice lacking MafA in the pancreas (termed Mafa^∆panc) revealed that loss of MafA first significantly affected β-cell mass, β-cell function, and islet architecture by 3 weeks of age. Large-Maf-regulated gene expression in islet β-cells was compromised even earlier in Mafa^∆panc;Mafb^+/− than Mafa^∆panc mice. Gene profiling analysis revealed that MafA regulates many key aspects of the glucose-stimulated insulin secretion pathway, including glucose metabolism, insulin production, and insulin granule docking. These findings substantially expand our knowledge of how MafA affects islet β-cell activity and provide a greater awareness into why this factor is so critical to the production of functional β-cells.

Pancreas-wide MafA deletion mutant mice (Mafa^Δpanc) were generated by breeding Mafa^fl/fl with Pdx1^5.5-Cre mice (20). Mafa^fl/fl mice served as experimental wild-type controls. MafB^+/− mice have been described previously (20). Compound Mafa^fl/fl;Mafb^+/−;Pdx1^5.5-Cre mutant mice were referred to as Mafa^Δpanc;Mafb^+/−. All mice used are on a mixed genetic background (C57BL/6J, I129, Balb/c). The Vanderbilt University Institutional Animal Care and Use Committee approved all of these studies.

Mice of 2, 3, or 4 weeks of age were anesthetized and then killed via cervical dislocation. The abdominal cavity was exposed and collagenase P (Roche) in PBS was injected into the common bile duct with the end connecting to the duodenum blocked. The pancreas was then isolated and digested further with collagenase P and islets isolated by hand picking under 10× magnification. Islet RNA was isolated using TRIzol (Invitrogen) and further treated with the DNA-Free RNA Kit (Zymo Research). RNA quality was determined using the Agilent 2100 Bioanalyzer in the Vanderbilt Functional Genomics Shared Resource Core Laboratory; RNA was only used if the integrity value was 7 or greater. Quantitative real-time PCR (qPCR) was conducted as previously described on a Light Cycler 480 II (Roche) (20); primer sequences are available upon request. The RNA data presented in Figs. 4 and 6 were performed with islet preparations free of acinar cell contamination, as determined by amylase analysis.

The conditions used for paraffin embedding pancreatic sections and immunofluorescence labeling were described previously (20). The following primary antibodies were used for immunostaining: guinea pig α-insulin (1:2,000; Linco Research), guinea pig α-glucagon (1:2,000; Linco Research), rabbit α-glucagon (1:2,000; Linco Research), rabbit α-MafB (1:10,000; Bethyl Laboratories), rabbit α-MafA (1:1,000, Bethyl Laboratories), rabbit α-Slc30a8 (1:1,000, Mellitech), and rabbit α-Glut2 (1:1,000; Chemicon). Species-matched secondary antibodies were used for immune detection (Cy5-conjugated donkey anti–guinea pig, Cy5-conjugated donkey anti-sheep, Cy2-conjugated donkey anti–guinea pig, Cy2-conjugated donkey anti-rabbit, Cy3-conjugated donkey anti-rabbit [all 1:500; Jackson ImmunoResearch Laboratories]). Cy3-conjugated tyramide signal amplification (1:400, PerkinElmer) was used for detecting MafB labeling with the biotin-conjugated donkey anti-rabbit secondary antibody (1:500; Jackson ImmunoResearch Laboratories). Nuclear counterstaining was performed using YoPro1 or DAPI (Invitrogen). Immunofluorescence images were acquired with a Carl Zeiss LSM510 confocal microscope.

Evenly spaced pancreatic sections at 150 and 240 μm apart were used from 1- or 2- and 3-week-old mice, respectively. Insulin⁺ cells and glucagon⁺ cell images were counted at 400× magnification. The β- to α-cell ratio was calculated by dividing the total number of insulin⁺ cells by the total number of glucagon⁺ cells. The relative β-cell mass was determined in pancreatic sections where insulin⁺ cells were visualized by DAB labeling, and pancreatic tissue was counterstained with eosin. Images of pancreatic sections were captured at 20× magnification using a ScanScope CS (Aperio), and the relative β-cell area was calculated by dividing the area of insulin⁺ cells by that of the whole pancreas.

Four-week-old mice (n ≥ 7) were fasted from 1800 to 0800 h, and blood glucose levels from the sampled saphenous vein were determined using a FreeStyle glucometer (Abbott Diabetes Care). The mice were then weighed, and 1 mg dextrose/g body weight (FisherBiotech) in sterile PBS was injected intraperitoneally. Blood glucose levels were measured at 15, 30, 60, 90, and 120 min postinjection. Fed blood glucose levels were measured prior to fasting.

Freshly isolated islets were washed in Ca²⁺-free Hanks' balanced salt solution and frozen at −80°C in size-matched 20-islet aliquots. Islets were lysed in 1 mL lysis buffer (1% Triton X-100 in 10 mmol/L Tris-HCl, pH = 7.4, made in double-distilled H₂O [18.2 MΩ], avoiding any glassware), and the Zn²⁺ concentration in the lysate was measured using a FluoZin-3 (Invitrogen) fluorescent dye assay (23). The Zn²⁺ concentration per islet (micromoles per islet) was determined by comparing the fluorescent signal at 516 nm to a standard curve generated from serial dilutions of ZnCl₂ in lysis buffer.

Freshly isolated islets were incubated in imaging medium (125 mmol/L NaCl, 5.7 mmol/L KCl, 2.5 mmol/L CaCl₂, 1.2 mmol/L MgCl₂, 10 mmol/L HEPES, 2 mmol/L glucose, 0.1% BSA, pH = 7.4) for 90 min and placed in a polydimethylsiloxane microfluidic flow device (24) on an LSM710 microscope (Zeiss). NAD(P)H autofluorescence was imaged under two-photon excitation using a 710 nm mode-locked Ti:sapphire laser oscillator (Coherent) and a custom 380–500 nm band-pass filter (Chroma) and nondescanned detector. Z-stacks of six images were acquired at 2 μm spacing 10 min after a change in glucose stimulation. All microscope and laser settings were kept constant between measurements.

Total isolated islet RNA was prepared from four independently derived 3-month-old littermate Mafa^Δpanc and wild-type mice. Total RNA (50 ng) was amplified using the Ovation RNA Amplification System V2 (Nugen Inc.), labeled with the BioPrime Array CGH Genomic Labeling System (Invitrogen Life Technologies), purified, and then analyzed on the whole mouse genome G4122A oligo microarray (Agilent Technology). The microarray analysis was performed at the Functional Genomics Core at the University of Pennsylvania.

Islet function was studied in a dynamic cell perifusion system at a perifusate flow rate of 1 mL/min (25,26). The effluent was collected at 3-min intervals using an automatic fraction collector. Insulin concentration in each fraction was measured by radioimmunoassay (RI-13K, Millipore). Insulin content in islet extracts was determined by radioimmunoassay (insulin, RI-13K, Millipore).

Islet β-cell ultrastructure was assessed by transmission electron microscopy as described previously (25). The density of β-cell secretory granules was determined using MetaMorph v7.7 software (Universal Imaging, Downington, PA) of calibrated images captured at 5,600× magnification. The number of docked β-cell granules was expressed per membrane length and β-cell granule size measured using ImageScope software (Aperio, Vista, CA) of calibrated images captured at 5,600×. The granule size was measured using ImageScope software of calibrated images captured at 25,000× magnification. Each analysis was based on 12–15 images per genotype.

All data were presented as mean ± SEM, and mean differences were tested for statistical significance using the Student two-tail t test unless otherwise noted.

Genes important to β-cell function are regulated by MafB at E18.5 and by MafA in 3-month-old islets, including those involved in glucose sensing (i.e., Slc2a2), vesicle maturation (Slc30a8), Ca²⁺ signaling (Camk2b), and glucose metabolism (G6pc2) (20). To more precisely define the postnatal period when MafA becomes essential for regulating β-cells, a time course study was performed. Quantitative PCR analysis revealed that Insulin 1 (Ins1), Insulin 2 (Ins2), Slc2a2, G6pc2, and Slc30a8 mRNA expression levels were reduced in 3-week-old Mafa^Δpanc islets (Fig. 1A), with only Slc30a8 significantly reduced within 2 weeks. Notably, MafB is barely detectable in 2-week-old β-cells (20). In addition, there were clearly fewer insulin⁺ cells producing the Slc2a2 and Slc30a8 proteins in 3-week-old Mafa^Δpanc islets (Fig. 1B), which, respectively, represent the glucose and zinc transporters required for normal β-cell function (27–29). The number of islet β-cells was also diminished in 3-week-old Mafa^Δpanc islets (Fig. 2), whereas the glucagon⁺ α-cell population was unchanged (Supplementary Fig. 1). Notably, this resulted from diminished β-cell replication and not increased apoptosis (Fig. 2B and C).

The presence of α-cells within the islet β-cell-enriched core is commonly observed in pancreatic transcription factor-deficient mice (e.g., Pdx1 [30,31] and Pax6 [14]). Conspicuously, MafA (but not MafB) expression is compromised in many, if not all, of these mutants (e.g., Pdx1 [32], Pax6 [33], data not shown). The previously observed intermingling of glucagon⁺ α-cells within the β-cell enriched islet core (19) was also regulated within the 3-week temporal period in Mafa^Δpanc islets, as was the decrease in the islet insulin⁺ to glucagon⁺ cell ratio (Fig. 3A). Furthermore, glucose clearance was impaired in 4-week-old Mafa^Δpanc mice (Fig. 3B). Collectively, these data highlighted how indispensable MafA is in postnatal islet β-cells after the loss of MafB expression.

Unfortunately, the functional significance of MafB in neonatal β-cells remains unknown since Mafb^−/− mice die at birth from central apnea (34). To obtain a greater awareness of MafB actions during this period, changes in islet morphology and function were compared between Mafa^Δpanc;Mafb^+/− and Mafa^Δpanc littermates. The reduction in the β- to α-cell ratio started 2 weeks after birth in Mafa^Δpanc;Mafb^+/− islets, 1 week earlier than the Mafa^Δpanc mutant (Fig. 3A). Such changes were not observed at any postnatal time point in Mafb^+/− islets (data not shown). Additionally, Mafa^Δpanc;Mafb^+/− mice displayed more profound defects in acute glucose clearance (Fig. 3B). Taken together, these data indicate a role for MafB in postnatal β-cell maturation.

Because blood glucose clearance was compromised after an intraperitoneal glucose challenge in Mafa^Δpanc mice (20), islet perifusion assays were performed to determine whether this resulted from an effect on first- and/or second-phase insulin granule release. In relation to wild-type littermates, first-phase insulin secretion was significantly reduced in Mafa^Δpanc islets, with second-phase trending in this direction (Fig. 4A and B). Islet insulin content was also decreased by ∼40% in Mafa^Δpanc β-cells (Fig. 4C). Analogous findings were also observed in Mafa^Δpanc;Mafb^+/− islets.

Given that biphasic insulin secretion reflects the recruitment and release of geographically distinct pools of insulin-containing granules, the ultrastructure appearance of control and Mafa^Δpanc β-cells were compared by transmission electron microscopy. Mafa^Δpanc β-cells contained 40% less mature insulin granules with a characteristic dark core appearance (Fig. 4E). Granule density and size was also reduced to a similar extent (Fig. 4F). Moreover, the number of docked insulin granules that are exocytosed during the first phase of glucose-stimulated insulin secretion was decreased by 55% in Mafa^Δpanc β-cells. A similar deficiency in mature insulin granule formation was found in Slc30a8^−/− islets (35), circumstances that preclude Zn²⁺ transport into granules for insulin crystallization and the dense-core appearance. Both Slc30a8 expression (Fig. 1B) and Zn²⁺ levels (Fig. 5A) were also significantly reduced in Mafa^Δpanc and Mafa^Δpanc;Mafb^+/− β-cells, with the compound mutant being effected more dramatically. In contrast, NAD(P)H-sensitive metabolic activity in Mafa^Δpanc and Mafa^Δpanc;Mafb^+/− islets was not impaired until high glucose concentrations (Fig. 5B). These data support a role for MafA in regulating processes directly involved in insulin synthesis, storage, and secretion.

To obtain an unbiased and comprehensive perspective on MafA control in mouse islets, Affymetrix whole mouse genome profiling was performed on isolated 3-month-old Mafa^Δpanc and control islets. There was a significant difference in 1,093 array transcripts in the Mafa^Δpanc sample (with a false discovery rate of 10% and a fold change >2), with 741 of these mapped to distinct EntrezGene identifiers (Supplementary Table 1). However, there were several exocrine-enriched genes within the differentially expressed group (e.g., elastase, carboxypeptidase, chymotrypsinogen), indicating possible contamination of the islet RNA preparation. This was verified by analyzing expression for exocrine marker gene levels in other independently isolated Mafa^Δpanc RNA samples (data not shown). Consequentially, 13 candidate genes were selected from the microarray to determine the accuracy of the data. Importantly, qPCR studies performed on 3-week-old islet RNAs confirmed the same changes between wild-type and Mafa^Δpanc (Fig. 6A), with 11 genes altered significantly in the expected manner and 2 showing the same trend. These results indicate that much of the microarray data reflects MafA control of islet β-cell gene expression. This included cyclin D2 (i.e., Ccdn2), a primary cell cycle regulator in islets (36), contributing to the decrease in Mafa^Δpanc β-cell mass (Fig. 2). Notably, these results also revealed early control by MafA in postnatal β-cells, suggesting that MafA is influencing transcription directly.

Gene ontology enrichment analysis performed on the Mafa^Δpanc and control islet microarray data disclosed that protein-binding and ion-binding activity were affected most profoundly, although MafA appears to influence a broad range of cellular functions found in many distinct subcellular compartments (Fig. 6B and C). Importantly, established effectors of granule docking (Sytl4, granuphilin [37]), granule release (Stxbp1, Munc18–1 [38]), and calcium influx (Atp2a2 [39,40]) were identified (Supplementary Table 1), with regulation confirmed by qPCR (Fig. 4). When these data are also considered in the context of the decrease in Ins1, Ins2, Slc30a8, Slc2a2 (the major β-cell GLUT [27]), and G6pc2 (a key gluconeogenic enzyme [41]) levels in Mafa^Δpanc islet β-cells (Fig. 1), these combined deficiencies provide an explanation as to why the first phase of glucose-stimulated insulin secretion was particularly impaired.

Interestingly, while many genes appear to be commonly regulated by MafA and MafB (20), expression-profiling analyses identified genes only affected in 3-month-old Mafa^Δpanc islets and not E18.5 Mafb^−/− pancreata (e.g., Ccnd2, Nqo1, and Atp2a2). Moreover, pathway analysis further supported MafA control in many aspects of β-cell signaling, including those involved with the extracellular matrix receptor (see 4512), chemokines (4062), focal adhesion (4510), cytokine–cytokine receptor (4060), and mitogen-activated protein kinase signaling (4010). (The numbers in the parentheses refer to pathways described in Fig. 6, Supplementary Fig. 2, and Supplementary Table 2.) Collectively, these gene expression studies provide a broader perspective into the many modes of action used by MafA to regulate β-cell function.

Our principal objective here was to precisely define when MafA first became functionally significant to mouse β-cells postnatally and to explore the nature of the dependency. Experiments performed in Mafa^Δpanc mice revealed that MafA was important to β-cells from 3 weeks of age, just after their loss of MafB. This included changes in β-cell mass, islet cell architecture, and glucose-stimulated insulin secretion. The proteins likely mediating these changes were identified by candidate and microarray analysis. These studies provide an in-depth perspective into the many ways MafA influences islet β-cells and mechanistic insight into why it is so necessary in cell maturation and adult activity.

MafA only begins to influence β-cells around weaning time in laboratory rodents. The switch from high-fat-based mother’s milk to a chow diet results in higher MafA expression and significant changes in the metabolic profile of rodent β-cells (42). Notably, MafB⁺ neonatal β-cells were functional immature, characterized by higher basal insulin secretion and little response to glucose. The MafB-to-MafA transition appears to be key to β-cell maturation. This is illustrated not only by the islet β-cell deficiencies of Mafa^−/− and Mafa^Δpanc mice, but also from studies examining the glucose-sensitive regulatory properties of in vitro generated insulin⁺ cells derived during human embryonic stem differentiation (21,22,43). Thus these Ins⁺ cells are only MafB⁺ and do not secrete insulin in a glucose-stimulated manner until after transplantation into mice, when the cells become MafA⁺ (22). The present studies predict many likely inadequacies in the human embryonic stem–derived MafB⁺ Ins⁺ cells due to their lack of MafA, including diminished expression of key effectors of insulin granule secretion (Stxbp1, Sytl4), calcium signaling (Atp2a2), and glucose signaling (G6pc2, Slc2a2).

MafA has a clear effect on insulin granule release. Hence glucose-stimulated insulin secretion, insulin granule density, and the number of predocked insulin granules on the plasma membrane were decreased in Mafa^Δpanc β-cells (Fig. 4). The reduction in the total number and size of mature dense-core insulin granules per cell is presumably attributable to diminished production of both insulin and the Slc30a8 zinc transporter (Figs. 1 and 4). Despite an ∼30% paucity of insulin granules in Mafa^Δpanc β-cells, there was a >50% loss predocked granules. Consistent with the concept that predocked granule release supports first-phase insulin secretion, Mafa^Δpanc β-cells were only significantly decreased during this phase. Presumably, this was due to limited expression of key MafA-regulated docking effectors (e.g., granuphilin [Sytl4] and Munc18–1 [Stxbp1]). Importantly, Munc18–1 was recently found to specifically regulate first-phase insulin secretion (38).

The architecture of the Mafa^Δpanc islet is similar to human (44), with fewer β-cells and α-cells present within the core (Fig. 3). The reduced number of β-cells in Mafa^Δpanc islets reflected a change in replication and not cell survival, likely involving cyclin D2. Potentially, the coexpression of MAFA and MAFB in human islets (26) contributes to the difference in islet architecture and β-cell composition between species. This is supported by the lack of any observable change in endocrine cell numbers, replication rate, or cyclin D2 levels in the embryonic Mafb^−/− pancreas (16) (data not shown). On the other hand, adult MafA controls many genes activated by MafB during mouse embryonic development (20). The more severe defects of the compound Mafa^Δpanc:Mafb^+/− mutant relative to Mafa^Δpanc further indicates that MafB is also beneficial to β-cells prior to its loss postnatally.

The overall mechanistic similarities and differences between MafA and MafB are still unclear. Notably, only 35 genes were identified in the PancChip 6 microarray performed on E18.5 Mafb^−/− pancreata (i.e., ≥1.5-fold; 47 in Mafa^Δpanc;Mafb^−/−; none in Mafa^Δpanc [20]), while 741 mapped to EntrezGene identifiers using the Affymetrix platform on 3-month-old Mafa^Δpanc islets here. Much of the difference is likely due to the relative insensitivity of the PancChip 6 microarray. It is expected that reinspection of Mafb^−/− pancreata will reveal both common and distinct regulatory targets in relation to the Mafa mutant, especially considering that Mafa^Δpanc β-cells are dysfunctional despite the presence of many MafB⁺ Ins⁺ cells (20). Both biochemical and cell-line-based studies also provide evidence supporting functional differences between islet cell expressed MafA and MafB (15,45,46). Consequentially, human MAFA:MAFB heterodimer activity in islet β-cells may contribute to observed differences in glucose-stimulated insulin secretion properties with rodents (26).

Funding. This work was supported by grants from the National Institutes of Health (DK090570 to R.S.; DK66636, DK68854, DK69603, DK72473, DK89572, DK089538, Merit Review Award from the VA Research Service [BX000666], JDRF [26-2008-863] to A.C.P.; DK53434 to D.W.P.; DK85145 to R.K.P.B.) and the Vanderbilt Diabetes Research and Training Center (DK20593). Imaging was performed with National Institutes of Health support from the Vanderbilt University Medical Center Cell Imaging Shared Resource (CA68485, DK20593, DK58404, HD15052, DK59637, EY08126) and Vanderbilt University Medical Center Islet Procurement and Analysis Core (DK20593).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. Y.H. contributed to the conception and design of the study, collected data and contributed to its analysis and interpretation, wrote the original draft of the manuscript, reviewed the penultimate draft of the manuscript, and read and approved the final version of the manuscript. T.Y. contributed to the conception of the study and collected data. R.K.P.B. and M.B. contributed to the conception and design of the study, collected data, contributed to the analysis and interpretation of data, reviewed and revised the original version of the manuscript, and read and approved the final version of the manuscript. M.G. collected data. W.B. contributed to the analysis and interpretation of data. D.W.P., A.C.P., and M.M. contributed to the conception and design of the study, reviewed and revised the original version of the manuscript, and read and approved the final version. D.C.T. contributed to data analysis and interpretation, reviewed and revised the original version of the manuscript, and read and approved the final version. R.S. contributed to the conception and design of the study, contributed to data interpretation, reviewed the penultimate draft of the manuscript, and read and approved the final version. R.S. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.