The NOD mouse develops spontaneous type 1 diabetes, with some features of disease that are very similar to the human disease. However, a proportion of NOD mice are naturally protected from developing diabetes, and currently, studies characterizing this cohort are very limited. Here, using both immunofluorescence and multiparameter flow cytometry, we focus on the pancreatic islet morphology and immune infiltrate observed in naturally protected NOD mice. We show that naturally protected NOD mice are characterized by an increased frequency of insulin-containing, smaller-sized, pancreatic islets. Although mice remain diabetes free, florid immune infiltrate remains. However, this immune infiltrate is skewed toward a regulatory phenotype in both T- and B-cell compartments. Pancreatic islets have an increased frequency of IL-10–producing B cells and associated cell surface markers. Resident memory CD69+CD8+ T cells show a significant shift toward reduced CD103 expression, while CD4+ T cells have increased FoxP3+CTLA4+ expression. These data indicate that naturally protected NOD mice have a unique islet signature and provide new insight into regulatory mechanisms within pancreatic islets.
Type 1 diabetes is an organ-specific autoimmune disease characterized by immune-mediated β-cell destruction in pancreatic islets, which results in deficient insulin production. Similar to humans, NOD mice develop spontaneous type 1 diabetes. However, in NOD mouse colonies worldwide, ∼20% (or more) of NOD mice remain normoglycemic and “protected” from diabetes, despite their genetic predisposition (1). Few studies have been done to discover the mechanism of this natural protection. Recently, we have dissected the B-cell functionality in naturally protected NOD mice, highlighting an increased interleukin-10 (IL-10)–producing B-cell frequency and enhanced response to dendritic cells, compared with NOD mice that have developed diabetes (2). Furthermore, it has been suggested that B cells, specifically anergic CD40+IL-10–producing B cells, found in the pancreatic islets of long-term normoglycemic mice (protected) (3), may confer this natural protection. Currently, the phenotype and function of CD4+ and CD8+ T-cell populations in naturally protected NOD mice are unexplored.
Pancreatic islets have a dynamic tissue microenvironment in which immune cells communicate to drive β-cell destruction. This is complicated by cellular and kinetic heterogeneity in both mouse and human pancreatic islets, including the rate of β-cell destruction. The aim of this study was to investigate the characteristics and heterogeneity of the islets in naturally protected NOD mice, including the immune infiltrate. This knowledge may provide insight into disease heterogeneity in humans because not all at-risk individuals develop type 1 diabetes.
Research Design and Methods
Female NOD/Caj mice, originally from Yale University, were bred in-house at Cardiff University. All mice received water and irradiated food ad libitum and were housed in specific-pathogen-free isolators or scantainers, with a 12-h dark/light cycle, at Cardiff University. All animal experiments were approved by the Cardiff University ethical review process and conducted under U.K. Home Office license in accordance with the U.K. Animals (Scientific Procedures) Act 1986 and associated guidelines. Diabetes conversion rates for Cardiff University can be found in Chen et al. (4), with ∼80% incidence in females by 30 weeks and a median incidence at 19 weeks old. NOD female incidence at other institutions or companies can vary; for example, at The Jackson Laboratory, incidence is ∼90% by 30 weeks of age, with a median female onset at 18 weeks.
Mice were monitored weekly for glycosuria (Bayer Diastix) from 12 weeks of age. After two positive glycosuria measurements, blood glucose levels were tested, and if >13.9 mmol/L, mice were diagnosed as diabetic. NOD mice that were ≥35 weeks of age and had never tested positive for glycosuria were considered to be protected from diabetes because the incidence of diabetes after this age is very low.
Pancreata were inflated with collagenase P solution (1.1 mg/mL) (Roche, Welwyn Garden City, U.K.) in Hanks’ balanced salt solution (with Ca2+ and Mg2+) through the common bile duct, followed by collagenase digestion with shaking at 37°C for 10 min. Islets were isolated by Histopaque density centrifugation (Sigma-Aldrich, Dorset, U.K.) and hand-picked under a dissecting microscope. For flow cytometric analysis, islets were then trypsinized to generate a single-cell suspension. Islet cells were rested at 37°C in 5% CO2 in Iscove’s modified Dulbecco’s media (supplemented with 5% FBS, 2 mmol/L L-glutamine, 100 units/mL penicillin, 100 μg/mL streptomycin, and 50 μmol/L 2-mercaptoethanol) overnight before multiparameter staining.
Pancreatic tissues were frozen in optimal cutting temperature medium and sectioned at 7-μm thickness. For wholemounts, pancreatic islets were fixed overnight at 4°C in 1% paraformaldehyde. For pancreatic sections, sections were fixed in 1% paraformaldehyde for 1 h at room temperature. Following fixation, tissue was permeabilized with 0.2% Triton X-100 and blocked with 5% FBS before the addition of a rat anti-mouse CD45 (BioLegend, London, U.K.) and a biotinylated anti-insulin (clone D6C4; Abcam) antibody mix. Secondary labeling was performed with both Alexa Fluor 633–conjugated goat anti-rat antibody (Invitrogen, Waltham, MA) and a streptavidin-conjugated Alexa Fluor 488 antibody (Invitrogen) and mounted with VECTASHIELD mounting medium with DAPI (Vector Laboratories). Islet whole-mounts were centrifuged at 300g for 3 min and then resuspended in mounting medium, with DAPI, before mounting to the slide. All sections and whole-mounts were imaged on a Leica SP5 confocal microscope.
All analyses were performed using Fiji (ImageJ) software (5). Islet area, perimeter, circularity, CD45, and insulin intensity were measured by using a region of interest on individual channels using Fiji’s measurement tool. Islets with insulin remaining were considered to be insulin-containing islets (ICIs) when three or more insulin-positive β-cells were present.
Cells were incubated with TruStain (anti-mouse CD16/32; BioLegend) for 10 min at 4°C followed by fluorochrome-conjugated monoclonal antibodies (mAbs) against cell surface markers for 30 min at 4°C. B-cell phenotyping multiparameter flow cytometry was carried out using the following mAbs: CD3 (145-2C11), B220 (RA3-6B2), CD138 (281-2), CD86 (PO3), CD80 (16-10A1), CD11c (N418), CD11b (M1/70), CD19 (6D5), CD44 (IM7), BAFFR (7H22-E16), MHC class II (10-3-6), and Ki67 (11F6), all purchased from BioLegend. IL-10 (JES5-16E3), IgD (11-26c.2a), and CD40 (3/23) were purchased from BD Biosciences. Galectin-1 antibody was purchased from R&D Systems. T-cell phenotyping multiparameter flow cytometry was carried out using the following mAbs: CD3 (145-2C11), CD8 (53-6.7), CD4 (GK1.5), CD103 (2E7), CD69 (H1-2F3), PD-1 (29F.1A12), interferon-γ (IFN-γ) (XMG1.2), CTLA4 (UC10-4B9), and FoxP3 (MF-14), all purchased from BioLegend. CD25 (PC61) and IL-10 (JES5-16E3) were purchased from BD Biosciences. Dead cells were excluded from analysis by live/dead exclusion dye (Invitrogen). IFN-γ, IL-10, CD107a, and galectin-1 were detected by intracellular cytokine staining after 3 h of stimulation with phorbol myristate acetate (50 ng/mL), ionomycin (500 ng/mL), and monensin (3 µg/mL) (all from Sigma-Aldrich). After extracellular staining, cells were fixed using a fixation/permeabilization kit (BD Biosciences) according to the manufacturer’s instructions and subsequently stained for mAbs against intracellular cytokines or appropriate isotype controls. For FoxP3, CTLA4, and Ki67 staining, cells were fixed/permeabilized using eBioscience nuclear transcription kit. Cell suspensions were acquired on an LSRFortessa (FACSDiva software; BD Biosciences) and analyzed using FlowJo version 10.1 software (Tree Star, Ashland, OR).
Statistical analyses were performed using GraphPad Prism (GraphPad Software, San Diego, CA). Comparison between groups was determined by Mann-Whitney U test or Kolmogorov-Smirnov test. For correlations, Pearson correlation coefficient was calculated. Data were considered significant at P < 0.05.
Data and Resource Availability
The data sets generated or analyzed during the current study are available upon reasonable request.
Increased Frequency of Insulin-Containing Small Pancreatic Islets in Naturally Protected NOD Mice
To investigate the features of the pancreatic islets in naturally protected NOD mice (not diabetic by 35 weeks of age and hereafter referred to as protected), we used immunofluorescence histochemistry, including both pancreatic islet whole-mounts and sections. First, we analyzed size, by area measured in pixels/μm, of pancreatic islets in wholemounts from protected NOD mice (Fig. 1A and B). Representative images (Fig. 1A) and a summary graph (Fig. 1B) demonstrated a range of sizes in the remaining islets in protected NOD mice. Next, we determined the size, by area, of pancreatic islet sections, from both protected and diabetic NOD mice (Fig. 1C and D). Smaller pancreatic islets were significantly more frequent in NOD mice that were protected from diabetes compared with mice that had developed diabetes (P < 0.001) (Fig. 1D). To analyze these islet data further, we used a frequency distribution graph to show the relative contribution, in percentages, of each islet to set “bins” according to islet area. Islet size distribution analysis (in percentages) showed that the relative frequency of islets with an area <50,000 pixels/μm, was greater in protected (85%) than in diabetic (51%) NOD mice (Fig. 1E). Further analysis in both protected and diabetic NOD mice revealed that ICIs with detectable insulin-positive β-cells were significantly smaller in size and more frequent in the protected NOD mouse pancreata compared with diabetic NOD mice (P < 0.01) (Fig. 1F). Interestingly, in the diabetic NOD mice, the very few ICIs detected were larger in area (Fig. 1G). Crucially, a comparison between ICIs and insulin-deficient islets (IDIs) in both NOD groups revealed that ICIs were significantly smaller in islet area (Fig. 1H) and were more frequent (Fig. 1I) compared with IDIs in protected NOD mice. However, this feature was lost in diabetic NOD mice, with no significant difference found in islet area (Fig. 1H) or frequency (Fig. 1I) in the few ICIs identified.
Alongside islet area, islet perimeter was analyzed, and the features observed in protected NOD mice were further confirmed (Supplementary Fig. 1A–C). In addition to islet size, we investigated whether islet circularity was different in protected NOD mice compared with mice that developed diabetes; however, no difference was observed in total islets analyzed (Supplementary Fig. 1D) or between ICIs and IDIs in either the protected or the diabetic NOD mouse cohort (Supplementary Fig. 1E).
Morphological Characterization of Prediabetic NOD Pancreatic Islets
Since protected mice had an increased frequency of smaller islets than diabetic mice, we determined whether smaller islets were less affected by immune cell infiltrate when pancreatic insulitis is established and diabetes begins to manifest (mice aged 12–18 weeks). Whole-mount pancreatic islets were stained with insulin and CD45 (a marker of immune cells) to identify islets undergoing attack, and the size was measured by area. Figure 2A demonstrates heterogeneity of pancreatic islets, shown by representative pictures from two different NOD mice. Z-stacks taken from two individual islets are also shown in Supplementary Fig. 2A. As expected, size, shape, and quantity of the CD45 immune infiltrate were variable in each individual islet. However, we observed a clear pattern between size and quantity of CD45 immune infiltrate. We confirmed this with correlation analysis comparing islet area (Fig. 2B) and islet perimeter (Supplementary Fig. 2B) with fluorescence intensity of CD45 (left) and insulin (right). A significant positive correlation was observed (P < 0.001) between islet area and CD45 immune infiltrate alongside a significant negative correlation (P < 0.01) between islet area and insulin-positive β-cells.
Regulatory B Cells Are Increased in the Islets of Naturally Protected NOD Mice
Because the pancreatic islets of naturally protected NOD mice have considerable immune infiltration, although remaining normoglycemic, we analyzed the B-cell infiltrate to investigate the frequency of regulatory cells by flow cytometry (Fig. 3). First, we compared the percentage of total B cells in groups of pooled protected NOD mice (>35 weeks) compared with groups of pooled younger prediabetic NOD mice (6–8 weeks old) (Fig. 3A and B) and observed a significant increase in CD19+ B cells (P < 0.001). Second, complementing our previous observation that naturally protected NOD mice have increased splenic IL-10–producing B cells (2), we showed a significant increase in pancreatic islet IL-10–producing B cells (Fig. 3C and Supplementary Fig. 3A). Furthermore, we demonstrated a population of galectin-1–positive B cells (Supplementary Fig. 3B), which encompassed the majority of the IL-10–producing B cells and were increased in >35-week-old mice (P = 0.06) (Fig. 3D and Supplementary Fig. 3C). Interestingly, we also observed a significant increase in CD40+- (Fig. 3E) and CD80+-expressing B cells (Fig. 3F) (for representative gating, see Supplementary Fig. 3D), both markers associated with regulatory B-cell function (6,7). It should be noted that no significant differences were observed in B-cell expression of MHC class II, CD86, or BAFFR (Supplementary Fig. 3E and F).
Many B cells in pancreatic islets express CD138 (8,9), a marker that identifies plasmablasts and plasma cells. We investigated our previously described populations identified by CD138 and IgD expression (9), examining them further (Fig. 3G–I) using markers for murine plasmablasts (10,11). We assessed CD19 (Fig. 3H), CD44, and Ki67 (Fig. 3I) in each CD138+/− population, revealing that CD138hiIgD− cells (red gate) contained a CD44hiKi67+ highly proliferative population that also expressed intermediate levels of CD19 (CD19int). With this further analysis, we propose that CD138hiIgD− cells (red gate) are a subpopulation of dividing plasmablasts. Few B cells that remained IgD+ were proliferating, and so it is probable that they represent classical B cells (CD138−IgD+) (blue gate) and an intermediate stage of plasmablast (CD138+IgD+) (orange gate). CD138+IgD− cells (gray gate) were a mixture of both CD19+/− cells and increased Ki67 expression, compared with CD138−IgD+ classical B cells, and are likely to represent both nondividing intermediate-stage plasmablasts and dividing plasmablasts.
We demonstrated that protected NOD mice displayed significant increases in both CD138+IgD+ B cells (Fig. 3J) and CD138−IgD+ B cells (Fig. 3K) compared with young, 6–8-week-old mice. This observation indicated that CD138+IgD+ cells were not selectively recruited over CD138−IgD+ B cells (Fig. 3L). However, the frequency of CD138+IgD− B cells in both young and older protected NOD mouse groups was not altered (Fig. 3M). Thus, the ratio of the CD138+IgD− to CD138−IgD+ B cells was increased in the younger NOD mice (Fig. 3N). This may suggest that these B cells arrive early to the pancreatic islets. Analysis of the small population of dividing plasmablasts (CD138hiIgD−) showed a significant increase in protected NOD mice compared with younger NOD mice (Fig. 3O).
Enrichment of CD4+FoxP3+CTLA4+ Regulatory T Cells in Naturally Protected NOD Mice
To determine the characteristics of CD4+ T cells in the pancreatic islets of NOD mice, that are naturally protected from diabetes, we studied the islet-infiltrating T cells of groups of pooled protected NOD mice by multiparameter flow cytometry and compared these to islet-infiltrating T cells from groups of pooled mice 6–8 weeks of age (mice with early-stage insulitis). CD4+ T-cell frequency and expression were increased in protected NOD mice, although nonsignificantly because of the variability in younger NOD mice (Fig. 4A and B). To ascertain whether CD4+ regulatory T cells (Tregs) contributed to the protection seen in protected NOD mice, we investigated the presence of CD4+FoxP3+ Tregs in the pancreatic islets and revealed a significant increase in CD4+FoxP3+ Tregs in the pancreatic islets of protected NOD mice compared with 6–8-week-old mice (Fig. 4C and D). Further analysis of the CD4+FoxP3+ T cells revealed a significant increase in the frequency of CTLA4+CD4+FoxP3+ T cells in protected NOD mice compared with 6–8-week-old mice (P < 0.01) (Fig. 4E and F). No significant differences were observed in the percentage of CD4+FoxP3+ T cells expressing CD25, PD-1, or CD103. Interestingly, intracellular cytokine analysis of both CD4+FoxP3−/+ T cells showed a significant increase in IFN-γ–producing CD4+FoxP3− T cells in protected NOD mice compared with mice aged 6–8 weeks (Fig. 4G and H). No differences in IL-10 in FoxP3+CD4+ T cells were observed. This increase in IFN-γ production from restimulated CD4+ T cells may reflect enhanced antigen experience.
Islet CD8 Resident Memory T Cells in Naturally Protected NOD Mice Switch to a CD103− Phenotype
Additional analysis of T cells was performed. Similar to CD4+ T cells, CD8+ T cells were modestly increased in frequency in >35-week-old mice (Fig. 5A and B). Because CD8+ T cells are found from an early age in the pancreatic tissue, we investigated markers of tissue residency. CD8+ tissue resident memory cells (TRM) can be distinguished by the surface markers CD69 and CD103 (12). CD8+ TRM populations have now been widely studied in various tissues and play a crucial role in immunosurveillance and protect against secondary viral infections (13). We demonstrated three key populations of CD8+ T cells in the pancreas of NOD mice: CD103−CD69− recirculating cells, CD103−CD69+ TRM, and CD103+CD69+ TRM (Fig. 5D–G). Protected >35-week-old NOD mice and younger NOD mice had similar frequencies of recirculating CD8+CD103−CD69− T cells (Fig. 5D). Strikingly, CD8+ TRM were significantly different between protected NOD mice and mice aged 6–8 weeks, with a shift toward greater prominence of CD103− TRM in protected NOD mice (Fig. 5E and F). Furthermore, both CD8+CD103−/+ TRM populations expressed CD107a, IFN-γ, and PD-1, with the CD103− TRM becoming more activated overall, after stimulation (Fig. 5G). There was no enrichment of PD-1 (a marker for T-cell exhaustion ) in either TRM population (Fig. 5G). Further analysis of the CD107a+ cells (a marker for recent degranulation) in the TRM populations revealed that CD107a+CD103+ TRM had fewer IFN-γ+PD-1−-expressing cells than the CD107a+CD103− TRM (P < 0.05) (Fig. 5H, summary graph) but, overall, a greater proportion of PD-1 expression (Fig. 5I, pie charts). Evaluation of IFN-γ and PD-1 subpopulations in the younger as well as the protected NOD mice showed a significant increase in IFN-γ+PD-1+ T cells in both CD103+/− TRM populations (Fig. 5J and K). Overall, naturally protected NOD mice showed a shift toward a CD8+CD103− TRM population, which has a more activated phenotype after stimulation, alongside an increase in IFN-γ in the TRM subsets.
In this report, we demonstrate key characteristics of pancreatic islets in a group of mice that are naturally protected from developing spontaneous diabetes. First, we discovered that smaller islets remain in these protected mice, with a clear correlation between islet size and immune infiltrate. Furthermore, insulin-positive β-cells are still present in pancreatic islets despite florid immune infiltrate. This immune infiltrate has a high frequency of B and T cells; however, the compositional signature was notably different in both immune cell compartments.
For the first time, we show that protected NOD mice have an increased frequency of smaller islets remaining in the pancreas, with ICIs smaller in size compared with IDIs. NOD mice that developed diabetes did not display this pattern. Moreover, we show in prediabetic NOD mice that larger islets have larger immune infiltrates. Previously, islet size has been shown to decrease as the duration of disease progresses in the NOD mouse (15); however, more sophisticated imaging identified that the smaller islets, located peripherally, are destroyed earlier in the disease process (16). However, surprisingly, the CD3+ immune infiltrate was not localized to the smaller islets or within a particular islet region (16). In humans, individuals with recent-onset type 1 diabetes have larger islets compared with people with long-standing type 1 diabetes (17).
Islets smaller in size than those found in the 12–16-week-old NOD mice may have been destroyed; however, we noted that islet size and infiltrate were correlated, but curiously, the very few remaining islets with insulin-positive β-cells were not smaller in size. Certainly, no correlation between β-cell mass and insulitis was observed in human pancreatic sections from donors with type 1 diabetes, but islet size was not addressed (18). This dichotomy requires further study to ascertain why this is the case. An explanation for this correlation between islet size and immune infiltrate could be explained by an increased capillary density (19), providing more immune cell access.
IL-10–producing B cells diminish the inflammatory response (20). Building on previous work (2), we now show that naturally protected NOD mice have a regulatory B-cell bias in the pancreatic islets. We demonstrate that galectin-1–positive B cells in pancreatic islets, a marker shown to be necessary for the function of regulatory B cells (21), and production of this protein, from activated B cells, can influence T-cell responses, including inducing T-cell apoptosis (22). Furthermore, the infiltrating B cells have increased levels of other cell surface markers associated with regulation. First, the CD80 molecule is known to preferentially bind CTLA4 (23), which we also found to be significantly increased on CD4+FoxP3+ Tregs in these protected NOD islets. Second, IL-10–producing B cells require CD40 to suppress effector T cells and autoimmunity (6). Interestingly, we describe here a significant increase in dividing CD138hiCD44hi plasmablasts in protected NOD mice previously reported to be an IL-10–producing population (10).
We observed that CD138+IgD− pancreatic islet B cells are similar in frequency in both younger and protected NOD mice. Early islet B cells have a B1 B-cell phenotype (24), B cells that are preferentially located in the peritoneum and pleural cavities, and are players in the initiation of type 1 diabetes (25). CD138+IgD− cells identified here are a heterogenous pool of dividing and nondividing plasmablasts at various intermediate differentiation stages, which consist of CD19+ and CD19− cells that lack IgM expression and are more proliferative compared with classical B cells. Crucially, this subset also contains antigen-specific B cells (9). Further work on these defined subsets is required to understand whether they are a B1-like cell that has altered as a result of the inflamed tissue environment.
Finally, we identified a significant shift in the CD8+ T-cell compartment. A CD8+CD103+ TRM phenotype is more prevalent in younger NOD mice, whereas a CD8+CD103− TRM phenotype dominates in protected NOD mice, with CD8+CD103− TRM producing increased IFN-γ when restimulated ex vivo. CD8+ TRM respond rapidly to control local immune responses in the tissue and so are central in tissue immunosurveillance. CD8+CD103− TRM have been described in various tissues, with CD8+CD103+ TRM located preferentially in the epidermis while CD8+CD103− TRM are located in the dermis of human skin (26), with distinct functional roles (26,27). CD8+CD103+/− TRM have been defined as separate populations, with different patterns of recall, molecular signature, and migration (27–29). CD8+CD103− TRM are more transient in the tissue compared with CD103+ counterparts (28), alongside an increased expression of sphingosine-1-phosphate receptor 1 (27). It remains unclear whether the enrichment of the CD8+CD103− TRM population in the older protected mice represents a more transient CD8 TRM population, and this would require further interrogation. CD8+ TRM in human pancreatic tissue from adults with recent-onset type 1 diabetes have been identified, but interestingly, only CD8+CD103+ TRM were detected in pancreatic tissue (30). However, in donors without diabetes, a CD8+CD103− TRM phenotype has been observed in ∼20–30% of resident CD8+ T cells in pancreatic islets (31).
E-cadherin, the principal ligand for CD103, is expressed by pancreatic islets (32), and this interaction can result in the release of cytokines and lytic granules from CD8+ T cells, therefore controlling cytotoxic CD8+ T-cell responses (33,34). CD103 is also required for the efficient destruction of pancreatic islet allografts (35). A shift toward CD8+ T cells lacking CD103 in naturally protected NOD mice may result in reduced targeted cell death. However, this CD103− TRM shift may also be a result from the loss of the ligand E-cadherin because of substantial loss of pancreatic islets.
In conclusion, naturally protected NOD mice have a unique pancreatic signature, with remaining islets that contain smaller insulin-producing β-cells and an immune infiltrate (both T and B cells) shifted toward a regulatory phenotype. These results are important to understanding the balance between a destroyed islet and an islet that remains, even partially, intact. Limitations in this study reflect the difficulty of studying both heterogenous islets and immune cell subsets, especially with limited cell numbers. Furthermore, protected NOD mice represent a pool of mice that cannot be detected early in the disease process, and so, comparison with mice that are younger, including mice that have established insulitis, is challenging. These younger mice would encompass both those that will develop diabetes and those that will be naturally protected, and we cannot, as yet, predict which mice will develop diabetes and distinguish them from mice that will be spared. Nevertheless, these observations highlight the need for further investigation into the dynamic process of β-cell destruction.
J.B. is currently affiliated with the Institute of Biomedical & Clinical Science, University of Exeter, Exeter, U.K.
This article contains supplementary material online at https://doi.org/10.2337/figshare.13643393.
Funding. This work was funded by Medical Research Council (U.K.) grant MR/K021141/1 to F.S.W.
Duality of Interest. No potential conflicts of interest relevant to this article were reported.
Author Contributions. J.B. performed the experiments and analyzed the data. J.B. and F.S.W. designed the experiments and wrote the manuscript. T.C.T. and J.D. contributed to the experimental procedures. All authors reviewed the manuscript. F.S.W. 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.