Cadherin Engagement Improves Insulin Secretion of Single Human β-Cells

Insulin release by pancreatic β-cells is primarily regulated by metabolites and hormones circulating in the blood. In addition, islet hormones released locally play a role in regulating insulin secretion of β-cells as paracrine modulators. Furthermore, direct cell-to-cell interactions between the neighboring islet cells also modulate the activity of β-cells, mainly by increasing β-cell response to insulin secretagogues (1,2). In this regard, it has been known that insulin release from intact islets or aggregated islet cells is increased compared with that of isolated islet cells (3–5). Using a reverse hemolytic plaque assay (RHPA) to analyze insulin secretion at the single-cell level, it has been shown that only one contact between homologous rat β-cells was sufficient to markedly increase glucose-induced insulin secretion (6). More recently, similar results were obtained with isolated human β-cells (7). Furthermore, it was shown that heterologous contacts between human β-cells and human α-cells were also able to increase insulin secretion (7). The mechanisms by which insulin secretion of contacting β-cells is increased remain to be elucidated. Establishment of intercellular contacts is likely to induce several cellular changes mediated by signaling pathways activated by the formation of junctional complexes and the engagement of cell-adhesion molecules. By different approaches, it has been shown that cell-to-cell communication mediated by gap junctions plays a major role in insulin secretion (8–11), and this type of junction certainly explains, at least in part, the effect of cell-to-cell contacts on insulin secretion. However, this does not preclude involvement of other types of junctions and/or adhesion molecules in this effect. Particular attention should be paid to the members of the cadherin family and especially to E-cadherin. This molecule was found to be downregulated in secretory defective β-cells (2,12,13) and the intensity of expression at the surface of isolated β-cells correlated with their secretory ability (6). Furthermore, treatment of β-cells with an anti–E-cadherin blocking antibody affected intracellular Ca²⁺ levels and insulin secretion (14). Altogether, these results suggest that expression of E-cadherin in β-cells is necessary for optimal insulin secretion. A real proof of involvement of E-cadherin in insulin secretion is difficult to obtain due to the fact that when one attempts to block or affect cadherin function molecules in aggregated-contacting cells, other junctions, including gap junctions, are affected as well in the absence of disaggregation. To overcome this difficulty, we used a strategy consisting of the activation of one specific cadherin on single cells (deprived of intercellular contacts) by its ligation to homologous cadherin peptide attached to an inert substrate (15).

Using this approach, we investigated the role of E-cadherin in regulating cell spreading and insulin secretion in human β-cells in presence of different insulin secretagogues including glucose.

The use of human islets for research was approved by our local institutional ethical committee. Islet isolation was performed as previously described (16,17). Islets were cultured in CMRL 1066-medium containing 5.6 mmol/L glucose and supplemented with antibiotics, HEPES, and 10% FCS. To dissociate islets into single cells, 5,000 islet equivalents were rinsed twice with PBS, resuspended in 1 mL Accutase (Innovative Cell Technologies, San Diego, CA), and incubated at 37°C with gentle pipetting every 30 s. When dissociation was considered to be complete, cells were diluted with cold complete CMRL. Then, islet cells were incubated at 37°C in nonadherent Petri dishes in complete CMRL for 18–20 h.

Recombinant human cadherin–Fc chimeric proteins containing ectodomains for E-cadherin, N-cadherin, and P-cadherin (hereafter referred to as E-cad/Fc, N-cad/Fc, and P-cad/Fc, respectively) were purchased from R&D Systems (Abingdon, U.K.). Glass coverslips or multiwell-printed microscope slides (Thermo Scientific, Braunschweig, Germany) were coated or not with 25 μg/mL cadherin–Fc chimera, diluted in H₂O, and incubated 18–20 h at 4°C. They were then rinsed with H₂O and air dried. Islet cells were seeded on these substrates in appropriate serum-free medium containing 1% albumin as indicated and incubated at 37°C for 3 or 24 h. Preparations were rinsed to remove unattached cells and analyzed and photographed using a Leica microscope (Leica Microsystems, Renens, Switzerland). For experiments with blocking E-cadherin antibody, coated glass and cells were incubated for 1 h at 4°C with E-cadherin antibody (Zymed Laboratories, Lucerne, Switzerland); islet cells were then incubated 18–20 h at 37°C in complete CMRL containing 22.2 mmol/L glucose and 1% albumin. For experiments with inhibitors, islet cells were seeded on coated glass in the presence or not of different inhibitors (FAK inhibitor Y15 [Fluka, Buchs, Switzerland], Rho-associated kinase [ROCK] inhibitor Y27632, or Rac inhibitor [Calbiochem, Zug, Switzerland]) and incubated as described above.

Islets cells were lysed in sample buffer (62 mmol/L Tris-HCl [pH 6.8], 2% SDS, 5% glycerol, and 1% 2-mercaptoethanol). Protein content was measured using a protein assay kit (Bio-Rad, Glattbrugg, Switzerland). After separation on an SDS-polyacrylamide gel, samples were electroblotted onto polyvinylidene fluoride membranes (Millipore, Billerica, MA) and immunoblotted with the appropriate antibody: anti–E-cadherin antibody (Cell Signaling Technology, Danvers, MA) or anti–N-cadherin antibody (BD Transduction Laboratories, San Diego, CA). An ECL protein detection kit (Amersham Biosciences) and a Molecular Imager ChemiDoc XRS+ System (Bio-Rad) were used for visualization of the bands.

For immunostaining, islet cells were fixed in 10% methanol-free formalin, permeabilized with 0.1% Triton X-100 in PBS, rinsed, incubated in 0.5% BSA in PBS, and then exposed for 2 h to a combination of primary antibodies as indicated in the 9Results. Primary antibodies against cadherins were a mouse anti–E-cadherin purchased from Zymed Laboratories, a rabbit anti–E-cadherin from Cell Signaling Technology, and a mouse anti–N-cadherin from BD Transduction Laboratories. The other antibodies used were a guinea pig anti-insulin, a rabbit antiglucagon from DakoCytomation (Baar, Switzerland), a rabbit antitubulin from Cell Signaling Technology, and a mouse antiglucagon from Sigma-Aldrich. After rinsing in PBS, islet cells were exposed for 1 h to an adequate combination of fluorescence-labeled secondary antibodies (Jackson ImmunoResearch Laboratories, Rheinfelden, Switzerland). Actin was detected by staining with Alexa Fluor 546 phalloidin (Invitrogen). Islet cells were rinsed in PBS and coverslipped before being observed and photographed using a Leica microscope (Leica Microsystems) or a confocal laser scanning microscope, LSM510 META (Zeiss, Feldbach, Germany). Images acquired from the confocal microscope were analyzed for pixel intensity using MetaMorph imaging software (Universal Imaging Corporation, West Chester, PA). Pixel intensity was quantified separately in manually selected regions of the cells.

Insulin secretion of β-cells was assessed by RHPA, as previously described (18,19). Glass microscope slides were treated or not for 18–20 h at 4°C with 25 μg/mL cadherin–Fc diluted in H₂O, rinsed with H₂O, and air dried. β-Cells were diluted in Krebs-Ringer bicarbonate (KRB) buffer supplemented with 0.1% BSA and 2.8 mmol/L glucose. Five percent of packed sheep red blood cells (Behring Institute, Marburg, Germany) previously coated with protein A were mixed with β-cells, and 50–60 μL of this preparation was introduced into Cunningham chambers. After 1-h incubation at 37°C, the chambers were rinsed with KRB containing either 2.8 or 16.7 mmol/L glucose and then filled with the same buffer, supplemented with anti-insulin guinea pig antibody (20). After 1-h incubation at 37°C, chambers were rinsed with KRB containing 2.8 mmol/L glucose, filled with the same buffer containing guinea pig complement (Behring Institute), and incubated at 37°C for 1 h. Chambers were then filled with solution of trypan blue, rinsed, and filled with 10% methanol-free formalin. β-Cells were labeled by immunofluorescence for insulin. Only cells with one Hoechst-labeled nucleus (single cells), labeled for insulin (β-cells) and not labeled for trypan blue (viable cells), were analyzed. Results are expressed as total plaque development representing the total plaque area around 100 single β-cells (18).

Data were expressed as mean ± SEM of n different experiments. Differences between means were assessed either by the Student t test or by one-way ANOVA when appropriate. Where ANOVA was applied, Tukey post hoc analysis was used to identify significant differences.

Cells dissociated from human islets by Accutase were incubated 3 or 24 h at 37°C in the presence of 5.6, 22.2, or 22.2 mmol/L glucose supplemented with IMBX and phorbol myristate acetate (PMA). As expected from our previous study (21), Western blotting showed that expressions of N- and E-cadherins were low immediately after Accutase treatment and time-dependently increased after incubation (Fig. 1). Compared with 5.6 mmol/L glucose, expressions of cadherins were similar at 22.2 mmol/L glucose and 22.2 mmol/L glucose supplemented with IMBX and PMA. Therefore, glucose or IMBX and PMA stimulations did not increase cadherin expressions. By immunofluorescence, we observed that labeling of cadherins at the surface of freshly isolated cells was virtually absent immediately after Accutase treatment. After incubation, labeling at the cell surface was time dependently recovered in a secretagogue-independent manner (data not shown).

Isolated islet cells, including α- and β-cells, maintained a round shape when attached to control glass and incubated for 3 or 24 h at 37°C. Under this control condition, the shape of islet cells was not modified either by high glucose (22.2 mmol/L) or high glucose supplemented with IMBX and PMA (Fig. 2). When attached to N-cad/Fc or E-cad/Fc, islet cells tended to flatten and to alter their shape, a phenomenon hereafter referred to as cell spreading. The percentage of spreading cells was significantly higher after 24 h compared with 3 h of culture (Fig. 2A vs. Fig. 2B; P < 0.0001). In response to glucose (22.2 vs. 5.6 mmol/L), after both 3- and 24-h cultures, the percentage of spreading cells increased in β-cells (P < 0.05) and not in α-cells (Fig. 2A and B), suggesting that cadherin-mediated adhesion should be related to some insulinotropic secretagogue mechanisms. These effects were observed in islet cells attached to both N-cad/Fc and E-cad/Fc. When stimulated with 22.2 mmol/L glucose supplemented with IMBX and PMA, the percentage of spreading β-cells on N-cad/Fc or E-cad/Fc increased compared with glucose alone (fold increase: 2.69 ± 0.65 and 2.62 ± 0.35, respectively; P < 0.01). This increase was also observed in α-cells (fold increase: 2.41 ± 0.47 and 1.57 ± 0.27, respectively; P < 0.01). Spreading of islet cells on E-cad/Fc was prevented when cells were treated with blocking cadherin antibodies (Fig. 3C and D) and when medium was depleted in calcium (Fig. 3E and F). These effects were observed in both α- and β-cells.

We then tried to understand whether exposition to secretagogues induced a sustained cell-spreading ability according to timing of cadherin engagement. To this end, isolated cells were preincubated at 37°C in suspension for 24 h in the presence of 5.6, 22.2, or 22.2 mmol/L glucose supplemented with IMBX and PMA, then attached to cadherin peptides, and incubated 3 h at 37°C in the presence of 5.6 mmol/L glucose. We observed that under these conditions, cells were unable to spread regardless of previous secretagogue stimulation. This indicates that, in order to induce cell spreading, acute metabolic stimulation of islet cells must occur after attachment to cadherin peptides.

The aim in this study was to assess the effect of adhesion of β-cells to cadherin peptides on insulin secretion. Most cells rapidly reaggregate following Accutase dissociation, and, when attached to cadherin peptides, only single cells are activated by immobilized cadherin peptides. Therefore, RHPA was used to specifically analyze insulin secretion from single β-cells. We observed that under basal condition (2.8 mmol/L glucose) insulin secretion of single β-cells was not affected by either P-cad/Fc, N-cad/Fc, or E-cad/Fc. At high glucose (16.7 mmol/L), insulin secretion from single β-cells was increased by both N- and E-cad/Fc compared with control (Fig. 4; P = 0.03 and 0.002, respectively).

In the presence of E-cad/Fc and after glucose stimulation, 5–10% single β-cells exhibited a spreading shape. As compared with 52 ± 8% round β-cells, virtually all spreading β-cells (96 ± 5%) were surrounded by a hemolytic plaque (Fig. 5A; P < 0.0001). The mean plaque area of spreading cells was about three times higher compared with that of round cells (Fig. 5B; P = 0.05). As a consequence, total insulin secretion was six times higher in spreading β-cells compared with round β-cells (Fig. 5C; P = 0.04).

It has been proposed that cortical actin behaves as a barrier, and its dissolution facilitates granule mobilization to the membrane and exocytosis (22). In this study, we questioned whether actin is located differentially between the free and the E-cadherin–contacting membranes. In control conditions, single β-cells displayed a round shape when observed by confocal microscopy, in both horizontal (x- to y-axis) and vertical (x- to z-axis) sections, and actin staining appeared evenly distributed on the cell circumference (Fig. 6A). When β-cells were attached to E-cad/Fc, their shape changed, so that most cells displayed a hemispheric profile in confocal vertical (x- to z-axis) sections (Fig. 6B). In some β-cells (30%), the shape change was more pronounced, and their profile appeared more flattened in confocal vertical (x- to z-axis) sections (Fig. 6C). These flattened cells certainly correspond to the cells with a spreading form observed by light microscopy. In all cases, when cells were attached to E-cad/Fc, cortical actin was highly expressed at the free cell membrane and minimally expressed or absent at the E-cad/Fc–contacting membrane, whatever the shape of the cells. By contrast, another cytoskeleton protein, tubulin, was more evenly distributed throughout the cytoplasm and did not display the asymmetrical distribution seen for actin. Quantification of staining in basal (E-cad/Fc–contacting area) and opposed apical regions of β-cells confirmed that actin intensity was lower in E-cad/Fc–contacting area, whereas tubulin was similarly distributed between these two regions (Fig. 6D).

We then studied human β-cell pairs in order to assess cortical actin distribution in free and cell-to-cell–contacting membranes. When β-cell pairs were attached for 2 h to E-cad/Fc at normal glucose concentration (5.6 mmol/L), we observed that actin was highly expressed at the free cell membrane and minimally expressed at the cell-to-cell–contacting membranes and at the E-cad/Fc–contacting cell membrane where cell-surface cadherins are expected to be engaged (Fig. 7A). When cells were attached for 2 h to E-cad/Fc at high glucose (22.2 mmol/L), similar staining was observed (Fig. 7B), indicating that glucose stimulation did not affect intensity and location of actin staining. Furthermore, this asymmetrical cortical actin distribution was also observed after 24 h at both low and high glucose.

To further confirm that asymmetrical cortical actin distribution was caused by E-cadherin engagement, E-cad/Fc–coated slides and islet cells were treated with a blocking anti–E-cadherin antibody before the adhesion experiment. We showed that under this condition, most β-cells kept a round profile, and cortical actin was located on the whole cell circumference (Fig. 7E), similarly to cells attached to uncoated glass (Fig. 7C).

To determine the importance of the Rho-GTPase pathway on E-cad–mediated adhesion, we tested the effect of two agents (the ROCK inhibitor Y27632 and the Rac inhibitor) known to act on this pathway. Treatment of cells with Y27632 for 24 h significantly increased the percentage of spreading β-cells plated on E-cad/Fc–coated slides (Fig. 8A). This effect was not observed on α-cells (Fig. 8A). Furthermore, in the presence of ROCK inhibitor, β-cell morphology of cells plated on E-cad/Fc–coated slides was changed: cells appeared larger and more spread (Fig. 8E). This effect was specific to E-cadherin engagement because ROCK inhibitor had no impact on cell morphology when cells were plated on the control slide (Fig. 8C). Rac and FAK inhibitors had no impact on islet cells spreading (Fig. 8A). The same results were obtained with N-cad/Fc–coated slides (data not shown).

A substantial body of evidence demonstrates that cell-to-cell adhesion is important for insulin secretion. Most works were performed in rodents (2,13,23), and few data were obtained from human tissues (7). Whatever the species, the mechanisms underlying the effect of cell-to-cell adhesion on insulin secretion are not yet elucidated. This is rendered difficult by the fact that many kinds of molecules and junctions are engaged in aggregated β-cells. To overcome this difficulty, we worked with isolated β-cells obliged to engage one type of adhesion molecule, immobilized on an inert substrate. Using this strategy, we have previously shown that E-cadherin and N-cadherin engagements were able to reduce apoptosis in isolated β-cells (21). These two cadherin types were shown to be expressed at the surface of human β-cells (21). In this study, we used a similar strategy aimed to show that stimulation of insulin secretion was able to promote adhesion of β-cells to cadherins and that cadherin engagement was able to affect insulin secretion.

First, we showed that cadherin expression in human islet cells was not affected by insulin secretagogues, including glucose, in contrast to what is seen in rat islet cells (6). This discrepancy is not understood, but is not surprising given the increasing published evidence of structural differences between rodent and human islet cells (24). Even if cadherin expression in human islet cells was not affected by insulin secretagogues, adhesion of these cells to E- and N-cadherin, but not to P-cadherin, was increased in the presence of insulin secretagogues. Increased cell adhesion was revealed by cell spreading, allowing cells to enlarge their contact area with the substrate. Cell spreading in response to E-cadherin engagement was inhibited with blocking anti–E-cadherin antibodies or calcium deprivation, indicating that, at least for E-cadherin, spreading did not result from nonspecific interactions. β-Cell spreading on laminin-332 has been described in rodents (25,26) and humans (27). We have demonstrated for the first time that β-cells have the ability to spread when attached to cadherins, suggesting that cadherins and integrins share similar outside-in signaling pathways involved in cytoskeleton remodeling and other cell processes related to cell-shape modification. Cytoplasmic domains of both cadherins and integrins interact with diverse intracellular proteins, including signaling molecules and cytoskeletal linker proteins, and by this way regulate diverse key cellular processes.

One can argue that cell spreading is a phenomenon only occurring in vitro and that cells are unable to spread within the islets. Instead, we propose that upon stimulation, cadherins and associated cytoskeleton proteins in islet cells are likely to undergo subtle reorganizations or adjustments conducting to cellular and molecular modifications that could be beneficial for insulin secretion. These events are accountable for cell spreading in vitro but induce no apparent morphological modification in vivo. With regard to cell spreading in vivo, it is interesting to remember the special association between β-cells and α-cells, previously described both in vitro in isolated human islet cells and in vivo within human pancreases (28). Indeed, most β-cells were shown to wrap around adjacent α-cells, and this is rendered possible by a supposed plasticity of β-cells that may involve cadherins and associated cytoskeleton proteins, similar to cells that spread upon engagement by cadherin peptides. It will be of interest to analyze in the future whether glucose and/or other insulin secretagogues modify the wrapping of β-cells on α-cells.

We observed that cadherin-mediated cell spreading was glucose dependent in β-cells but not in α-cells. Glucose activates metabolic activity and secretion of β-cells and not of α-cells. Therefore, the mechanisms involved in cadherin-mediated cell spreading are likely related to glucose metabolic and/or secretory processes. In rodents, glucose-induced spreading of β-cells on laminin-332 requires an increase in intracellular calcium concentration and is not a consequence of secretory processes that follow elevation of Ca²⁺ concentration (26). Calcium mobilization could be also involved in cadherin-mediated cell spreading of human β-cells. This hypothesis is consistent with recent studies indicating that cadherin-mediated cell–cell adhesion regulates calcium entry dynamics (14,29).

For the first time, we show that specific ligation of cadherins is able to increase insulin secretion of β-cells in response to glucose. By using the E-cad/Fc peptide approach we can exclude the engagement of other cell surface molecules, as well as the formation of functional gap junctions, known to control the secretory activity of β-cells (30). Therefore, the effect observed on insulin secretion can be attributed to the sole engagement of specific cadherins. Under the studied conditions, E-cad/Fc had more effect that N-cad/Fc in increasing glucose-induced insulin secretion. The same difference was previously observed when similar experiments were conducted to assess the effect of cadherin engagement on apoptosis of β-cells (21). The reason for the differential effects of N- and E-cadherins is not yet understood. One possible explanation could be the differential levels of cadherin-type expressions at the surface of cells. The lower level of N-cadherin compared with E-cadherin observed by immunofluorescence seems to confirm this hypothesis; however, variable affinity/specificity of antibodies possibly account for such a difference.

Effect of cadherin engagement on insulin secretion occurred regardless of a β-cell shape change, since most β-cells retained an apparent spherical morphology after attachment to cadherin peptides and insulin secretion. When cell shape on E-cad/Fc was reassessed by confocal microscopy, we observed in vertical sections that a majority of β-cells attached to E-cad/Fc had a clear hemispheric shape that was undetectable by conventional microscopy. This indicated that a large membrane area was engaged in the contact with the substrate. By conventional microscopy, only a small percentage of cells attached to E-cad/Fc displayed a recognizable spreading form, corresponding to the cells with a higher membrane area engaged with E-cad/Fc. Using RHPA, we were able to study the level of insulin secretion according to the round and spreading appearance of β-cells. We observed that insulin secretion was six times higher in spreading β-cells compared with round β-cells. In rat β-cells, we previously observed that spreading cells secreted more insulin in response to glucose compared with round cells (25). This effect was attributed to outside-in signaling through β1 integrin, which induces phosphorylation of focal adhesion kinase (FAK), a protein that localizes at focal contacts (26,31). Despite distinct molecular structures, some crossroads exist between integrins and cadherins (32). For instance, FAK was shown to localize at sites of cell–cell contacts in several cell types (33,34), and cell–cell contact formation was shown to regulate FAK phosphorylation (35). FAK signaling has the ability to modulate the activity of the Rho-GTPases (35), which are so far the best understood mode of integration of integrin- and cadherin-mediated signals (36–38). Interestingly, Rho-GTPases are known to play a central role in the regulation of actin cytoskeleton dynamics (39). Because the cytoplasmic domain of cadherins establishes a linkage to the actin cytoskeleton by binding to multiple actin-binding proteins including α-catenin and vinculin (40), we investigated whether changes in the actin cytoskeleton occur upon cadherin-mediated adhesion. We demonstrated that actin is redistributed in β-cells engaged in adhesion processes implicating cadherins. Indeed, cells attached to E-cad/Fc exhibited a cortical actin staining at the free membrane and not at the membrane attached to E-cad/Fc and neither to the cell–cell-contacting areas. Our results are similar to those reported by Yamada and Nelson (41), who showed in simple epithelial (Madin-Darby canine kidney) cells that actin filaments were prominent at the edges of the expanding E-cadherin zone and were greatly reduced at the cell-to-cell contact. Furthermore, Rho-GTPases are involved in the initiation, expansion, and completion of cell–cell adhesion. In this study, we demonstrated that inhibition of ROCK, a major effector of RhoA, increased spreading of β-cells specifically on E-cad/Fc, suggesting that ROCK regulate cadherin function. Indeed, it has been shown that constitutive activation of ROCK disrupts adherens junctions, while pharmacological inhibition appears to promote adherens junction stability (42). Another study further revealed that RhoA activity at adherent junctions was cadherin dependent and, in particular, associated with dynamic physical interaction between p120 and multiple factors acting up- and downstream of RhoA to coordinate its activity at the actin–cadherin interface (43). Actin cytoskeleton remodeling was previously reported in rat β-cells spreading on extracellular matrix (44), a situation that also improves glucose-stimulated insulin secretion (25). Furthermore, a relationship between actin cytoskeleton polymerization levels and insulin secretory properties of MIN6 cells has been established (22). In β-cells, filamentous actin is organized as a dense web beneath the plasma membrane (45) and is transiently depolymerized after glucose stimulation of insulin secretion (22,46). Indeed, final access of secretory granules to the plasma membrane requires dissolution of a defined cortical actin web. Recent reports have underlined a possible role for the Rho-GTPases (44,47,48) in cortical F-actin reorganization during regulated insulin secretion.

Taken together, our results demonstrate that adhesion of β-cells to specific cadherins is regulated by glucose and that cadherin engagement promotes insulin secretion. From the current study, we propose that cadherin activation acts as a node for transmission of signals emanating from cell-to-cell contact and targets the coupling from stimuli to secretion via regulation of actin cytoskeleton remodeling.

Acknowledgments. The authors thank Corinne Sinigaglia and Caroline Rouget (Cell Isolation and Transplantation Center, Geneva University Hospitals and University of Geneva, Geneva, Switzerland) for excellent technical assistance.

Funding. This work was supported by grants from the Swiss National Science Foundationhttp://dx.doi.org/10.13039/501100001711 (310030-143462/1), the Juvenile Diabetes Research Foundationhttp://dx.doi.org/10.13039/100000901 (31-2012-783), and the Insuleman Foundation.

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

Author Contributions. G.P. researched data and wrote the manuscript. V.L. and B.B. reviewed the manuscript. D.M.-D. researched data. P.M. and T.B. contributed to discussion and reviewed the manuscript. D.B. designed the study, researched data, and reviewed the manuscript. D.B. 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.