Neuron-enriched RNA-binding Proteins Regulate Pancreatic Beta Cell Function and Survival

Abstract

Pancreatic beta cell failure is the central event leading to diabetes. Beta cells share many phenotypic traits with neurons, and proper beta cell function relies on the activation of several neuron-like transcription programs. Regulation of gene expression by alternative splicing plays a pivotal role in brain, where it affects neuronal development, function, and disease. The role of alternative splicing in beta cells remains unclear, but recent data indicate that splicing alterations modulated by both inflammation and susceptibility genes for diabetes contribute to beta cell dysfunction and death. Here we used RNA sequencing to compare the expression of splicing-regulatory RNA-binding proteins in human islets, brain, and other human tissues, and we identified a cluster of splicing regulators that are expressed in both beta cells and brain. Four of them, namely Elavl4, Nova2, Rbox1, and Rbfox2, were selected for subsequent functional studies in insulin-producing rat INS-1E, human EndoC-βH1 cells, and in primary rat beta cells. Silencing of Elavl4 and Nova2 increased beta cell apoptosis, whereas silencing of Rbfox1 and Rbfox2 increased insulin content and secretion. Interestingly, Rbfox1 silencing modulates the splicing of the actin-remodeling protein gelsolin, increasing gelsolin expression and leading to faster glucose-induced actin depolymerization and increased insulin release. Taken together, these findings indicate that beta cells share common splicing regulators and programs with neurons. These splicing regulators play key roles in insulin release and beta cell survival, and their dysfunction may contribute to the loss of functional beta cell mass in diabetes.

Introduction

Insulin-secreting pancreatic beta cells share many phenotypic traits with neurons. Similarities range from an analogous physiology and function to similar development and gene expression (1). Beta cells release insulin using a similar exocytotic machinery as used by neurons to release neurotransmitters. Indeed, insulin is stored and secreted using scaffolding proteins and synaptic-like vesicles similar to neuronal cells (2–4), and like neurons, beta cells are able to generate action potentials in response to different stimuli (5). Despite having different embryonic origins (6, 7), global gene expression and active chromatin marks of beta cells are closer to neural tissues than to any other tissue, including other pancreatic cells (8). Neurons are phylogenetically older than beta cells, and in some primitive organisms neurons express insulin and regulate circulating glucose levels (9, 10). Taken together, these findings suggest that beta cells have evolved into specialized insulin-secretory cells by adopting, at least in part, neuronal transcription programs (1). Supporting this hypothesis, both neurons and beta cells lose the expression of the transcriptional repressor element-1 silencing transcription factor (REST)⁵ during differentiation (11). REST is expressed in most non-neuronal cells and acts as a master negative regulator of neurogenesis by repressing a subset of neuron-specific genes that play a pivotal role in the development of the neuronal phenotype (12). In beta cells, the absence of REST allows the expressions of transcription factors and genes of the exocytotic machinery that are crucial for beta cell differentiation, survival, and secretory function (11–15).

Understanding how beta cells acquire neuron-like features has focused mainly on transcription, but it remains unclear whether beta cells and neurons share similar post-transcriptional regulation. Alternative splicing (AS) is a key post-transcriptional mechanism that allows cells to fine-tune their transcriptome and generate proteome diversity to gain functional specialization. By regulating splice site selection and inclusion of alternative exons into mature mRNAs, AS enables individual genes to produce multiple protein isoforms that often display different properties or functions (16). AS is particularly widespread in the brain, playing a central role in neuronal development, complexity, and physiology (17, 18). Indeed, defects in splicing regulation are associated with a wide spectrum of neuropsychiatric and neurological disorders (19, 20). Neuron splicing programs are controlled by specific RNA-binding proteins (RBPs) that regulate the splicing of genes involved in key neuronal functions such differentiation, morphology, migration, electrophysiological activity, and synapse formation (21–23).

Beta cell failure is the central event leading to diabetes. Although the role of AS in the development of diabetes remains to be clarified, recent studies from our group indicate that AS alterations induced by both inflammation and disease susceptibility genes contribute to beta cell dysfunction and death (24–26). Furthermore, we found that the “neuron-specific” splicing factor Nova1 is expressed in beta cells (24) and regulates AS of genes involved in exocytosis and apoptosis (27).

In this study, we examined the expression of splicing-regulatory RBPs in pancreatic islets, brain, and other human tissues and detected a group of splicing regulators that are preferentially expressed in human islets and brain. We next focused on four of them, namely Elavl4 (also known as HuD), Nova2, Rbfox1, and Rbfox2. We confirmed their expression in pancreatic beta cells and identified their roles in beta cell function and survival. The findings obtained suggest that beta cells and neurons share common splicing programs and indicate that defects in splicing regulation of common brain-beta cells RBPs may contribute to beta cell dysfunction and death.

Results

Pancreatic Beta Cells Express Neuron-enriched RNA-binding Proteins

We used a previous RNA sequencing dataset of human islets (24) and the Illumina BodyMap 2.0 (GEO: GSE30611) to compare the expression of 118 splicing-regulatory RBPs between human islets and 16 other tissues. Some brain-expressed RBPs clustered together with the five human islet preparations evaluated, and we identified a group of RBPs that are relatively highly expressed in both human islets and brain (Fig. 1A, yellow square). This includes members of the ELAVL, RBFOX, and NOVA families of splicing-regulatory RBPs that have been reported previously to regulate key splicing programs in brain (18, 19). The expression of ELAVL4, NOVA2, RBFOX1, and RBFOX2 was confirmed in human islets and in the human insulin-producing EndoC-βH1 cell line by real time qPCR (Fig. 1, B–E). Of note, NOVA2, RBFOX1, and RBFOX2, but not ELAVL4, were also well expressed in adipose tissue. To understand the reasons for the similar RBPs gene expression pattern observed between human islets and brain, we assessed whether their expression was under the control of the transcriptional repressor REST. Gene expression of RBPs in rat INS-1E cells was analyzed after infection with an adenovirus encoding REST (Fig. 2A). REST overexpression down-regulated the expression of Snap25 (Fig. 2B), a known target of REST, but it did not modify the expression of Elavl4, Nova2, or Rbfox2, suggesting that these genes are not targets of REST in beta cells (Fig. 2, C, D, and F). REST overexpression down-regulated the expression of Rbfox1 (Fig. 2E), but the finding that Rbfox1 is expressed in other non-neuronal tissues (Fig. 1D) argues against a direct transcriptional regulation by REST. Compensatory cross-regulation between RBP paralogs was observed for Elavl4 and Nova2. Elavl4 knockdown (KD) by small interfering RNA (siRNA) increased the levels of paralog Elavl1 (Fig. 3, A and B). Similar observations were made for Nova1 expression under Nova2 KD (Fig. 3, C and D). However, no compensatory cross-regulation between Rbfox1 and Rbfox2 was observed (Fig. 3, E–H).

FIGURE 1.

Pancreatic beta cells express neuron-enriched RNA-binding proteins. A, heat map representing the expression of RBPs in human islets and in 16 other human tissues. Gene expression was assessed by RNA-sequencing using a previously published dataset consisting of five different human islets preparations (24) and the Illumina BodyMap 2.0. Expression values were hierarchically clustered using Gene Pattern modules. Blue and red colors indicate low and high expressed genes, respectively. RBPs showing high expression in brain and in human islets are highlighted by a yellow square. B–E, mRNA expression of four RBPs assessed by qRT-PCR in human islets (n = 3), insulin-producing EndoC-βH1 cells (n = 3), and in a panel of normal human tissues (n = 1). B, ELAVL4; C, NOVA2; D, RBFOX1; and E, RBFOX2.

FIGURE 2.

Expression of neuron-enriched RBPs after REST overexpression in INS-1E cells. INS-1E cells were infected with adenovirus overexpressing Renilla luciferase (AdLuc) or REST (AdREST) for 24 h at m.o.i. of 10 or left untreated (NT, non-treated). Expression of the following was measured by qRT-PCR and normalized by the housekeeping gene Gapdh: A, REST; B, Snap25; C, Elavl4; D, Nova2; E, Rbfox1; and F, Rbfox2. Results are mean ± S.E. of four to six independent experiments. *, p < 0.05; **, p < 0.01; and ***, p < 0.001 versus AdLuc; paired t test.

FIGURE 3.

Compensatory regulation within RBPs families. INS-1E cells were transfected with siCTR or siRNAs targeting different RBPs for 48 h. The expression of the different RBPs was measured by qRT-PCR and normalized by the housekeeping gene Gapdh. Expression of the following was evaluated after Nova2 KD: A, Elavl4; B, Elavl1. Expression of Nova2 (C) and Nova1 (D) is shown. Expression of Rbfox1 (E) and Rbfox2 (F) was evaluated after Rbfox1 KD. Expression of Rbfox2 (G) and Rbfox1 (H) was evaluated after Rbfox2 KD. mRNA expression values were normalized by the highest value of each experiment, considered as 1. Results are from 3 to 5 independent experiments. *, p < 0.05; **, p < 0.01 and ***, p < 0.001 versus siCTR; paired t test.

Elavl4 Modulates Beta Cell Death

To elucidate the function of Elavl4 in pancreatic beta cells, we used siRNAs to knock down Elavl4 in INS-1E, FACS-purified primary rat beta cells, and EndoC-βH1 cells (Fig. 4, B, F, and H). Elavl4 KD increased apoptosis under basal conditions but protected against cytokine-induced apoptosis in the three different cell models studied (Fig. 4, C, G, and I). Elavl4 KD affected neither glucose-stimulated insulin secretion nor insulin content (data not shown). Modulation of beta cell death following Elavl4 KD was secondary to activation of the mitochondrial pathway of apoptosis, as indicated by changes in the expression of cleaved caspase-3 and -9 measured by Western blotting (Fig. 4, A, D, and E). To identify the mechanisms underlying Elavl4 KD-induced apoptosis, we assessed the expression of several apoptotic regulators of the Bcl-2 family, as well as markers of endoplasmic reticulum stress and oxidative stress, but no changes were detected (data not shown).

FIGURE 4.

Elavl4 modulates pancreatic beta cells death. INS-1E cells (A–E), FACS-purified primary rat beta cells (F and G), and EndoC-βH1 cells (H and I) were transfected with siCTR or independent siRNAs targeting Elavl4 for 48 h and then exposed to the pro-inflammatory cytokines IL-1β + IFN-γ. Cytokine exposure was 24 h for INS-1E cells and 48 h for primary rat beta cells and EndoC-βH1 cells. A, two representative Western blottings showing Elavl4, cleaved caspase-9 and -3, and α-tubulin (used as loading control) after Elavl4 knockdown in INS-1E cells. B, Western blotting densitometric measurements of Elavl4. C, apoptosis in INS-1E cells was evaluated by propidium iodide staining. D, Western blotting densitometric measurements of cleaved caspase-9; E, cleaved caspase-3. F, mRNA expression of Elavl4 in FACS-purified primary rat beta cells measured by qRT-PCR and normalized by the housekeeping gene Gapdh; G, apoptosis evaluated by propidium iodide staining. H, protein expression of ELAVL4 and α-tubulin (used as loading control) in EndoC-βH1 cells measured by Western blotting. One representative Western blotting and the densitometric measurements are shown. I, apoptosis in EndoC-βH1 cells evaluated by propidium iodide staining. mRNA and protein expression values were normalized by the highest value of each experiment, considered as 1. Results are mean ± S.E. of three to five independent experiments. *, p < 0.05, **, p < 0.01, and ***, p < 0.001 versus untreated siCTR; #, p < 0.05 and ##, p < 0.001, versus cytokine-treated siCTR; paired t test.

Nova2 KD Increases Basal and Cytokine-induced Cell Death via the Mitochondrial Pathway of Apoptosis

Nova2 was silenced in INS-1E, EndoC-βH1, and FACS-purified primary rat beta cells (Fig. 5, A, F, and H). Nova2 KD increased INS-1E cell apoptosis under basal conditions and sensitized these cells to cytokine-induced apoptosis (Fig. 5B). This was accompanied by cleavage of caspase-9 and -3, indicating that cell death is secondary to the activation of the mitochondrial pathway of apoptosis (Fig. 5, C–E). Nova2 KD also increased apoptosis in FACS-purified primary rat beta cells and EndoC-βH1 cells (Fig. 5, G and I). Nova2 KD neither affected glucose-stimulated insulin release nor insulin content in INS-1E cells (data not shown).

FIGURE 5.

Nova2 knockdown increases apoptosis in pancreatic beta cells. INS-1E cells (A–E), FACS-purified primary rat beta cells (F and G), and EndoC-βH1 cells (H and I) were transfected with siCTR or independent siRNAs targeting Nova2 for 48 h and then exposed to the pro-inflammatory cytokines IL-1β + IFN-γ. Cytokine exposure was 24 h for INS-1E cells and 48 h for primary rat beta cells and EndoC-βH1 cells. A, protein expression of Nova2 and α-tubulin (used as loading control) in INS-1E cells was measured by Western blotting. One representative blot and densitometric measurements are shown. Apoptosis in INS-1E cells was evaluated by propidium iodide staining (B) and by Western blotting for cleaved caspase-9 and cleaved caspase-3 (C). Densitometric measurements of cleaved caspase-9 (D) and 3 (E) are shown. mRNA expression of Nova2 in FACS-purified primary rat beta cells was measured by qRT-PCR and normalized by the housekeeping gene Gapdh (F), and apoptosis was evaluated by propidium iodide staining (G). Protein expression of NOVA2 and α-tubulin (used as loading control) in EndoC-βH1 was measured by Western blotting (H). One representative blot and the densitometric measurements are shown. Apoptosis in EndoC-βH1 cells was evaluated by propidium iodide staining (I). mRNA and protein expression values were normalized by the highest value of each experiment, considered as 1. Results are mean ± S.E. of three to four independent experiments. *, p < 0.05; **, p < 0.01; and ***, p < 0.001 versus untreated siCTR; #, p < 0.05; ##, p < 0.01; and ###, p < 0.001 versus cytokine-treated siCTR. A and E–H, paired t test. B, D, and E, paired t test with Bonferroni's correction.

Silencing of Rbfox1 and Rbfox2 Increases Insulin Secretion and Content

Rbfox1 and Rbfox2 were independently silenced in INS1-E cells (Figs. 6, A and B, and 7A). Rbfox1 silencing did not affect basal or cytokine-induced apoptosis (data not shown) but induced a significant increase in insulin release after glucose or glucose plus forskolin stimulation (Fig. 6, C and D). Interestingly, there was a dose-response relation between the induced Rbfox1 knockdown by the different siRNAs and the observed increase in insulin secretion. A similar phenotype was observed following Rbfox2 KD (Fig. 7B). In addition, increased K_ATP-channel independent insulin release was observed in Rbfox1 KD cells following stimulation with KCl (Fig. 6E). Moreover, insulin cellular content was significantly increased following Rbfox1 and Rbfox2 silencing (Figs. 6F and 7C). Insulin mRNA expression was significantly increased following Rbfox1 KD by siFOX1#2 (the siRNA inducing more marked Rbfox1 inhibition) but not following KD with the less effective siFOX1#1 (Fig. 6G). To determine whether the observed increase in insulin secretion was secondary to increased glucose metabolism and ATP generation or due to augmented exocytosis, we measured glucose oxidation and mitochondrial function, and we analyzed calcium currents through patch clamp. No differences in glucose oxidation were observed between control and Rbfox1 KD cells (Fig. 6H). In line with this, there were no differences in mitochondrial function parameters such as glucose-induced respiration, ATP-linked respiration (response to oligomycin), and maximal respiration (response to FCCP following oligomycin), as evaluated by oxygen consumption rate measurements (Fig. 6I). Calcium currents were evoked by depolarizing the cell membrane from a resting potential of −70 mV to voltages ranging from −50 to +50 mV, and the data obtained were analyzed for charge (Q). Rbfox1 KD cells showed a tendency to increase charge (Fig. 6, J and K). These findings suggest that mild electrophysiological alterations may underlie Rbfox1 KD, but its contribution to the observed phenotype is unclear.

FIGURE 6.

Rbfox1 knockdown increases insulin secretion. INS-1E was transfected with siCTR or three independent siRNAs against Rbfox1 (siFOX1#1, siFOX1#2, and siFOX1#3) for 48 h. A, mRNA expression of Rbfox1 measured by qRT-PCR and normalized by the housekeeping gene Gapdh. B, protein expression of Rbfox1 and α-tubulin (used as loading control) measured by Western blotting. One representative blot and the densitometric measurements are shown. C, insulin secretion following Rbfox1 KD evaluated by ELISA after 30 min of incubation with 1.7 mm glucose, 17 mm glucose, or 17 mm glucose plus forskolin (20 μm). D, scatter plots showing individual insulin secretion experiments shown in C. Individual paired experiments are indicated by the same color. E, insulin secretion after 30 min of incubation with 1.7 mm glucose or 1.7 mm glucose plus 30 mm KCl. Insulin secretion values are expressed as fold increase as compared with siCTR exposed to 1.7 mm glucose. F, insulin content fold increase as compared with siCTR. G, mRNA expression of Ins2 measured by qRT-PCR and normalized by the housekeeping gene Gapdh. H, glucose metabolism following exposure to 1.7 or 17 mm glucose after Rbfox1 KD. Values are expressed as fold increase as compared with siCTR exposed to 1.7 mm glucose. I, oxygen consumption rates (OCR) relative to basal (1.7 mm glucose) following sequential stimulation with glucose (17 mm), oligomycin (5 μm), FCCP (4 μm), and rotenone (1 μm). J and K, electrophysiological characterization of voltage-gated Ca²⁺ channels following Rbfox1 KD. J, example trace of currents evoked by a depolarization from −70 to 0 mV in a single siFOX1#2- (lower trace) or siCTR (upper trace)-transfected cells. K, charge (Q)-voltage (V) relationship in siFOX1#2- (black squares) and siCTR (white circles)-transfected cells. Charge is measured as the area enclosed by the curve in J. mRNA and protein expression values were normalized by the highest value of each experiment, considered as 1. Results are mean ± S.E. of three to eight independent experiments (A–I). K, results are mean ± S.E. of 17–20 cells. A, B, F, G, I, and K: *, p < 0.05; **, p < 0.01, and ***, p < 0.001 versus siCTR. C and D: ***, p < 0.001 versus siCTR exposed to 1.7 mm glucose; #, p < 0.05, and ###, p < 0.001 versus siCTR exposed to 17 mm glucose; &, p < 0.05; &&, p < 0.01, and &&&, p < 0.001 versus siCTR exposed to 17 mm glucose plus forskolin. F, ***, p < 0.001 versus siCTR exposed to 1.7 mm glucose; ###, p < 0.001 versus siCTR exposed to 30 mm KCl. A, B, F, G, I, and K: paired t test. C–E and H, paired t test with Bonferroni's correction.

FIGURE 7.

Rbfox2 knockdown increases insulin secretion and content. INS-1E cells were transfected with siCTR or siFOX2#1 siRNA for 48 h. A, mRNA expression of Rbfox2 measured by qRT-PCR and normalized by the housekeeping gene Gapdh. mRNA expression values were normalized by the highest value of each experiment, considered as 1. B, insulin secretion evaluated by ELISA after 30 min of incubation with 1.7 mm glucose, 17 mm glucose, or 17 mm glucose plus forskolin (20 μm) following Rbfox2 KD. Values are expressed as fold increase as compared with siCTR exposed to 1.7 mm glucose. Individual paired experiments are indicated by the same color. C, insulin content was evaluated by ELISA. A and C: *, p < 0.05, and ***, p < 0.001 versus siCTR; paired t test. B: ###, p < 0.001 versus siCTR exposed to 17 mm glucose, and &&, p < 0.01 versus siCTR exposed to 17 mm glucose plus forskolin; paired t test with Bonferroni's correction.

Rbfox1 Regulates Alternative Splicing of Key Genes Controlling Insulin Secretion

To identify Rbfox target genes that might affect insulin release, we re-analyzed a previously published dataset that identified 1,059 Rbfox-regulated AS events in mouse brain (28). Pathway enrichment analysis using IPA (Ingenuity Systems) and DAVID (david.abcc.ncifcrf.gov) software showed that Rbfox-regulated transcripts in brain are enriched in pathways that can affect insulin secretion in beta cells, including cytoskeleton organization, vesicle-mediated transport, and calcium signaling. Three AS events regulated by Rbfox proteins in brain were selected for evaluation in beta cells (Fig. 8). Using two independent siRNAs, we confirmed Rbfox1-dependent splicing modulation for the actin-remodeling protein gelsolin (Gsn) and the voltage-gated calcium channel 1C (Cacna1c) (Fig. 8, A and C). Splicing modulation of the insulin receptor (Insr), however, was observed only with one siRNA (Fig. 8B). Interestingly, Rbfox1 KD reduced the inclusion of a frameshift cassette exon in Gsn, reducing the expression of an isoform predicted to be a target for nonsense-mediated mRNA decay (Fig. 8A).

FIGURE 8.

Rbfox1 regulates the alternative splicing of key genes related to pancreatic beta cell function. A–C, representative images of agarose gels showing that Rbfox1 controls alternative splicing of gelsolin (Gsn) (A), insulin receptor (Insr) (B), and voltage-gated calcium channel 1C (Cacna1c) (C) in INS-1E cells. The percentage of inclusion/exclusion of each exon was quantified by densitometry and is shown under the respective gels. Results are mean ± S.E. of five to ten independent experiments. *, p < 0.05, and **, p < 0.01 versus siCTR; paired t test.

Rbfox1 KD-induced Insulin Secretion Is Mediated by the Actin Remodeling Protein Gelsolin

Gelsolin is a calcium-activated actin-severing protein that plays a positive role in insulin secretion. Gelsolin mediates glucose-dependent actin cytoskeleton depolymerization and interacts with the t-SNARE protein syntaxin 4, facilitating the trafficking, docking, and fusion of insulin granules (29, 30). To evaluate the role of gelsolin in the observed increase in insulin release following Rbfox1 KD, we knocked down in parallel both Rbfox1 and gelsolin in INS-1E cells (Fig. 9). The observed alternative splicing shift following Rbfox1 KD was associated with an increase in gelsolin mRNA expression (Fig. 9B), probably secondary to the decrease in the nonsense-mediated mRNA decay-targeted isoform observed in Fig. 8A. Knockdown of gelsolin was associated with increased Rbfox1 expression (Fig. 9A), indicating a cross-regulation between both genes. Double KD of Rbfox1 and gelsolin prevented the increase in insulin secretion detected after Rbfox1 KD alone (Fig. 9C), indicating that the observed phenotype is probably secondary to alterations in cytoskeleton remodeling.

FIGURE 9.

Gelsolin silencing prevents the insulin secretion increase produced by Rbfox1 knockdown. INS-1E cells were transfected with siCTR, siFOX1#2, siGSN, or siFOX1#2 + siGSN for 48 h. mRNA expression of Rbfox1 (A) and gelsolin (B) was measured by qRT-PCR and normalized by the housekeeping gene Gapdh. mRNA expression values were normalized by the highest value of each experiment, considered as 1. C, insulin secretion was evaluated by ELISA after 30 min of incubation with 1.7 mm glucose, 17 mm glucose, or 17 mm glucose plus forskolin (20 μm). Values are expressed as fold increase as compared with siCTR exposed to 1.7 mm glucose. Results are mean ± S.E. of six independent experiments. A and B, *, p < 0.05; **, p < 0.01, and ***, p < 0.001 versus siCTR; paired t test. C, ###, p < 0.001 versus siCTR exposed to 17 mm glucose; &&, p < 0.01 versus siCTR exposed to 17 mm glucose plus forskolin; and **, p < 0.01 as indicated by bars. Paired t test with Bonferroni's correction.

Rbfox1 KD Accelerates Actin Depolymerization Kinetics after Glucose Stimulation

To study whether Rbfox1 KD indeed affects cytoskeleton organization, we evaluated the dynamics of depolymerization of actin filaments following stimulation with glucose using confocal and fluorescence microscopy. Rbfox1 KD did not induce major differences in actin cytoskeleton organization in cells maintained at 0 mm glucose (Fig. 10A, left panel), but Rbfox1 KD cells showed less actin stress fibers (F-actin) after stimulation with 17 mm glucose when compared with control cells (Fig. 10A, right panel). Actin depolymerization was quantified by calculating the ratio between filamentous and globular actin (F/G-actin). Rbfox1-KD cells displayed a faster glucose-induced actin depolymerization when compared with control cells (Fig. 10, B and C). These differences were more pronounced with siFOX1#2, the siRNA inducing higher levels of Rbfox1 KD and a more extreme phenotype, and were prevented by gelsolin silencing (Fig. 10, A–C). These findings indicate that the observed increase in insulin secretion following Rbfox KD is secondary to increased gelsolin expression and consequent enhanced actin depolymerization.

FIGURE 10.

Rbfox1 knockdown accelerates actin depolymerization dynamics after glucose stimulation, a phenomenon prevented by gelsolin silencing. A, confocal microscopy images of actin cytoskeleton in INS-1E cells following incubation at 0 or 17 mm glucose during 10 min. Cells were transfected with siCTR, siFOX1#1, siFOX1#2, or siFOX1#2 + siGSN for 48 h. B and C, quantification of actin depolymerization following glucose stimulation by the F/G ratio in INS-1E cells. Atto 550-phalloidin (red) staining actin filaments and DNase I-Alexa 488 (green) staining globular actin were used to quantify the emitted fluorescence, and calculation of the ratio is shown in C. ###, p < 0.001 versus siCTR exposed to 17 mm glucose during 5 min; &&&, p < 0.001 versus siCTR exposed to 17 mm glucose during 10 min; @@@, p < 0.001 versus siCTR exposed to 17 mm glucose during 15 min; *, p < 0.05; **, p < 0.01, and ***, p < 0.001 as indicated by bars. Paired t test with Bonferroni's correction. Scale bars, 10 μm (A) and 20 μm (B).

Discussion

Beta cells and neurons share many similarities in gene transcription, but whether both cell types present similar AS regulation remains an open question. We presently show that some neuron-enriched splicing regulators are also expressed in pancreatic beta cells where they play a major role in regulating beta cell survival and function. Our data analysis identified a cluster of RBPs that are expressed in both brain and in human islets. This cluster includes members of the Elavl, Nova, and Rbfox families, which have been implicated in neuronal physiology and disease (28, 31–35). We found that Elavl4 and Nova2 are required for beta cell survival, and their depletion leads to activation of the intrinsic pathway of apoptosis. In contrast, Rbfox1 and Rbfox2 mostly regulate beta cell function, affecting the insulin secretory capacity but without impact on beta cell viability.

Elavl4 has been implicated in several aspects of mRNA turnover and function, including AS, stability, and translation. Elavl4 acts as a master regulator of neurogenesis, regulating neuronal survival, function, and plasticity (34, 36–38). In pancreatic beta cells Elavl4 has been shown to regulate insulin translation (39), autophagy (40), and lipid synthesis (41). Here we show that Elavl4 also regulates cell survival in beta cells. Intriguingly, Elavl4 silencing increases basal apoptosis but protects against cytokine-induced apoptosis in the three different models studied. This suggests that Elavl4 may regulate different transcripts and pathways depending on the functional state of the beta cells. Supporting this hypothesis, previous findings indicate that pro-inflammatory cytokines induce changes in the expression of key gene networks for beta cell function and phenotype (42–44) that may be altered by Elavl4 KD. Nevertheless, additional experiments are required to clarify the mechanisms underlying Elavl4 KD.

Nova proteins have been extensively studied in brain where they regulate splicing networks involved in survival, migration, synaptic function, and axon guidance (45–48). We have previously shown that Nova1 controls beta cell function and survival by regulating genes involved in exocytosis, apoptosis, insulin receptor signaling, splicing, and transcription (27). Here we provide additional data showing that Nova2 depletion impairs beta cell survival but not insulin secretion. Increased apoptosis by Nova2 KD may be secondary to up-regulation of the FoxO3A transcription factor and pro-apoptotic protein Bim, as observed previously in Nova1 KD (27). The absence of insulin secretion defects may be explained by differences in targets genes between the Nova paralogs. Alternatively, this may be related to the observed Nova1 up-regulation upon Nova2 silencing, revealing a compensatory mechanism that could decrease physiological perturbations due to changes in splicing pathways. Reciprocal compensatory regulation between RBP paralogs has been previously observed in the Nova and other RBP families (49, 50).

Rbfox1 and Rbfox2 are splicing regulators preferentially expressed in neurons and skeletal muscle and, as shown here, are also found in pancreatic beta cells. In brain, Rbfox proteins regulate a wide range of developmental and synaptic functions. Studies in knock-out mice and in human neural stem cells showed that Rbfox1 regulates neuronal development and excitability (32, 51), whereas Rbfox2 affects neuron migration in the cerebellum (52). In addition, mutations in Rbfox1 are found in patients with epilepsy, spinocerebellar ataxia, and autism spectrum disorders (53–56). In skeletal muscle, the absence of Rbfox1 leads to abnormal myofibrillar structure and impaired muscle function (57). Our data indicate that Rbfox proteins also play an important role in regulating insulin exocytosis in beta cells. Proper beta cell function relies on efficient insulin storage and secretion in response to increased blood glucose levels. Glucose-stimulated insulin secretion is modulated by a number of factors, including glucose metabolism, calcium signaling, and cytoskeleton remodeling (58–61). We observed that both Rbfox1 and Rbfox2 silencing increase insulin secretion and content. The similar phenotype observed for Rbfox1 and Rbfox2 KD suggests that both RBPs regulate overlapping sets of transcripts in beta cells. This is in agreement with previous studies showing that Rbfox proteins bind to the same UGCAUG motif due to the presence of a common RNA binding domain (28, 32, 62). Functional analysis of the role of Rbfox proteins in beta cells (present data) indicates that Rbfox depletion increases the insulin secretory capacity by affecting actin depolymerization kinetics following glucose stimulation. Thus, Rbfox1 regulates the splicing of the actin remodeling protein gelsolin, modulating the expression of a non-productive isoform and thus affecting overall gelsolin expression. Inhibition of gelsolin expression prevented the Rbfox1 KD-induced insulin secretion increase and actin cytoskeleton alterations. Gelsolin enhances insulin secretion by mediating glucose-dependent actin cytoskeleton remodeling and by interacting with t-SNARE proteins, thus favoring insulin granule exocytosis (29, 30). In line with the present findings, Rbfox1 targets in brain and muscle are enriched in genes involved in cytoskeleton organization and actin filament-based processes (28, 57). We cannot exclude, however, that other Rbfox targets (such as calcium channels) also contribute to the observed phenotype.

In conclusion, these data indicate that pancreatic beta cells express neuron-enriched RBPs that control beta cell function and survival, suggesting that neurons and beta cells share common splicing regulatory programs. Splicing programs provide a highly interconnected regulatory layer to control gene expression and promote cell specialization in differentiated tissues. These findings provide a better understanding on how AS regulates beta cell survival and function, and they point the way to identify the role for AS in beta cell dysfunction in diabetes.

Experimental Procedures

RNA-sequencing Data Visualization

To analyze the expression pattern of splicing factors in human islets and in 16 other human tissues, we used a previously published RNA-sequencing dataset (24) and the Illumina BodyMap 2.0 dataset (GEO: GSE30611). A list of human splicing factors was manually curated by combining the SpliceAid-F database (63) with other RBPs that have been described in the literature or in other databases as splicing regulators. Log transformed reads per kb/million mapped reads (RPKM) gene expression values were used to generate a two-way hierarchical clustering heat map using Gene Pattern modules (64) with default parameters (not centered, not normalized).

Culture of Rat INS-1E Cells, FACS-purified Primary Rat Beta Cells, Human EndoC-βH1 Cells, and Human Islets

Rat insulin-producing INS-1E cells, kindly provided by Dr. C. Wollheim (University of Geneva, Geneva, Switzerland), were cultured in RPMI 1640 GlutaMAX-I medium (Invitrogen) as described previously (65).

Primary beta cells were obtained from male Wistar rats (Charles River Laboratories, Brussels, Belgium). Rats were housed and used according to the guidelines of the Belgian Regulations for Animal Care, with the approval by the local Ethical Committee. Pancreatic islets were isolated by collagenase digestion, handpicked, and dissociated into single cells by mechanical and enzymatic dispersion, and beta cells were purified by FACS sorting as described previously (66). Purified beta cells were cultured in Ham's F-10 medium containing 10 mm glucose, 2 mm GlutaMAX, 0.5% BSA, 50 μm isobutylmethylxanthine, 50 units/ml penicillin and 50 μg/ml streptomycin, and 5% heat-inactivated fetal bovine serum (FBS, Gibco Life Technologies, Inc.).

Human insulin-producing EndoC-βH1 cells, kindly provided by Dr. R. Sharfmann (Institut Cochin, Université Paris Descartes, Paris, France), were grown on plates coated with Matrigel/fibronectin (100 and 2 μg/ml, respectively, Sigma), and cultured in DMEM as described previously (67, 68).

Human islets from eight non-diabetic donors were isolated in Pisa, Italy, using collagenase digestion and density gradient purification (69). The donors (four men and four women) were age 64 ± 18 years and had a body mass index of 26 ± 2 kg/m². Cause of death was cardiovascular disease in four cases, trauma in two cases, cerebral hemorrhage in one case, and post-anoxic encephalopathy in one case. Beta cell purity, as evaluated by immunofluorescence for insulin using a specific anti-insulin antibody, was 56 ± 7% (Table 1). Islets were cultured at 6.1 mmol/liter glucose as described previously (24, 70).

TABLE 1.

Characteristics of the organ donors and human islet preparations used for RNA-sequencing and independent confirmation

The abbreviations used are as follows: F, female; M, male; BMI, body mass index; CVD, cardiovascular disease; CH, cerebral hemorrhage; PAE, post-anoxic encephalopathy. Purity indicates the percentage of beta cells in the human islet preparations as determined by immunostaining for insulin.

	Gender	Age	BMI	Cause of death	Purity
		years	kg/m2		%
Islets for RNA-sequencing
ID1	F	77	24	Trauma	45
ID2	F	46	23	CVD	60
ID3	F	79	28	Trauma	61
ID4	M	36	26	CVD	62
ID5	M	77	25	CVD	62
Islets for qPCR
ID6	F	75	29	CVD	49
ID7	M	79	25	CH	48
ID8	M	44	28	PAE	59

Cell Treatment

INS-1E and EndoC-βH1 cells were exposed to the following cytokine concentrations, based on previous dose-response experiments performed by our group (65, 68): recombinant human IL-1β (R&D Systems, Abingdon, UK) 10 units/ml for INS-1E cells or primary rat beta cells and 50 units/ml for EndoC-βH1 cells; recombinant rat IFN-γ (R&D Systems) 100 or 500 units/ml for INS-1E cells or primary rat beta cells, respectively; and human IFN-γ (PeproTech, London, UK) 1000 units/ml for EndoC-βH1 cells. Lower cytokine concentrations and shorter time intervals were used in INS-1E and FACS-purified rat beta cells experiments because rat beta cells are more sensitive to cytokine-induced damage than human beta cells (71).

Infection with Recombinant Adenoviruses

To overexpress REST in INS-1E cells, we used a recombinant adenovirus containing the full-length cDNA of human REST kindly provided by Dr. F. Allagnant (University of Lausanne, Lausanne, Switzerland) (13). After 48 h of pre-culture, INS-1E cells were infected with the recombinant adenovirus encoding REST or with a control adenovirus encoding Renilla luciferase (72). Cells were infected for 24 h at 37 °C at an m.o.i. of 10 and collected for mRNA extraction.

RNA Interference

The small interfering RNAs (siRNAs, Life Technologies, Inc.) targeting rat and human genes used in this study are described in Table 2. Allstar Negative Control siRNA (siCTR) (Qiagen) was used as negative control in all experiments. This siCTR does not affect beta cell gene expression, function, or viability (73). Transient transfection was performed using 30 nm siRNA and Lipofectamine RNAiMAX lipid reagent (Invitrogen-Life Technologies, Inc.) as described previously (73). After 16 h of transfection, cells were cultured for a 48-h recovery period before exposure to cytokines.

TABLE 2.

Sequences of siRNAs used to knock down gene/protein expression

Gene	Species	Name	Supplier	Sequence
None		siCTR All-stars Negative control siRNA	Qiagen, Venlo, The Netherlands	Not provided
Elavl4	Rattus norvegicus	siELAVL4#1 silencer select siRNAi	Invitrogen, Paisley, UK	5′ GAGGCAUUGGUGAAAUCGAATT 3′
				5′ UUCGAUUUCACCAAUGCUCCC 3′
Elavl4	R. norvegicus	siELAVL4#2 silencer select siRNAi	Invitrogen, Paisley, UK	5′ GAUUCAGGCUGGACAAUUUTT 3′
				5′ AAAUUGUCCAGCCUGAAUCTT 3′
Nova2	R. norvegicus	siNOVA2#1 stealth select siRNAi	Invitrogen, Paisley, UK	5′ AGGUCCGAGAAAUCCCUCAAGCGAU 3′
				5′ AUCGCUUGAGGGAUUUCUCGGACCU 3′
Nova2	R. norvegicus	siNOVA2#1 stealth select siRNAi	Invitrogen, Paisley, UK	5′ CGGGAGCCACCAUCAAGCUAUCUAA 3′
				5′ UUAGAUAGCUUGAUGGUGGCUCCCG 3′
Rbfox1	R. norvegicus	siFOX1#1 silencer select siRNAi	Invitrogen, Pasley, UK	5′ GAUUUGGUUUCGUAACUUUTT 3′
				5′ AAAGUUACGAAACCAAAUCCC 3′
Rbfox1	R. norvegicus	siFOX1#2 silencer select siRNAi	Invitrogen, Paisley, UK	5′ CGAGGUUAAUAAUGCGACATT 3′
				5′ UGUCGCAUUAUUAACCUCGAT 3′
Rbfox1	R. norvegicus	siFOX1#3 silencer select siRNAi	Invitrogen, Paisley, UK	5′ GCGGUGUUGUUACCAGGATT 3′
				5′ UCCUGGUAAACAACACCGCCA 3′
Rbfox2	R. norvegicus	siFOX2#1 silencer select siRNAi	Invitrogen, Paisley, UK	5′ AGAAGAUGGUCACACCAUATT
				5′ UAUGGUGUGACCAUCUUCUTG 3′
Gsn	R. norvegicus	siGSN FlexiTube siRNA	Qiagen, Hilden, Germany	5′ CAUCACUGUCGUUAGGCAATT 3′
				5′ UUGCCUAACGACAGUGAUGGG 3′
ELAVL4	Homo sapiens	siELAVL4#1h silencer select siRNAi	Invitrogen, Paisley, UK	5′ CAAUUACCAUUGAUGGAUTT 3′
				5′ AUUCCAUCAAUGGUAAUUGGG 3′
NOVA2	H. sapiens	siNOVA2#1h stealth select siRNAi	Invitrogen, Paisley, UK	5′ GACAGAGCCAAGCAGGCCAAGCUGA 3′
				5′ UCAGCUUGGCCUGCUUGGCUCUGUC 3′
NOVA2	H. sapiens	siNOVA2#2h stealth select siRNAi	Invitrogen, Paisley, UK	5′ UGCUGGCCAUCAGCACGGCGCUUAA 3′
				5′ UUAAGCGCCGUGCUGAUGGCCAGCA 3′

Assessment of Cell Viability

The percentage of apoptotic, necrotic, and viable cells was determined by fluorescence microscopy after 15 min of incubation with the DNA-binding dyes propidium iodide (5 mg/ml) and Hoechst 33342 (10 mg/ml) (Sigma). A minimum of 600 cells was counted for each experimental condition by two independent observers, one of them unaware of the sample identity. The agreement between researchers was >90%. Apoptosis was further confirmed in some experiments by cleavage of caspase-3 and -9 by Western blotting (see below).

mRNA Extraction, RT-PCR, and qRT-PCR

Poly(A)⁺-RNA was isolated from INS-1E cells, primary rat β-cells, EndoC-βH1 cells, and human islets using the Dynabeads mRNA DIRECT kit (Invitrogen) and reverse-transcribed as described previously (74). cDNAs from 14 normal human tissues (adipose, brain, colon, heart, kidney, liver, lung, lymph node, pancreas, skeletal muscle, small intestine, spleen, stomach, and testes) were obtained from BioChain Laboratories (San Francisco, CA). The real time PCR amplification reaction was performed using SYBR Green (Bio-Rad, Nazareth Eke, Belgium) and compared with a standard curve as described (75). Expression values were corrected for the housekeeping genes glyceraldehyde-3-phosphate dehydrogenase (Gapdh) for rat beta cells or β-actin for human cells. Expression of these housekeeping genes is not modified under the present experimental conditions (70). The primers used in this study are detailed in Table 3.

TABLE 3.

Primers sequences used for real time and splicing analyses

The abbreviations used are as follows: qRT, primers used for real time qRT-PCR; SPL, primers used to analyze splicing variants.

Target gene	Application	Forward (5′–3′)	Reverse (5′–3′)
Rat
Gapdh	qRT	AGTTCAACGGCACAGTCAAG	TACTCAGCACCAGCATCACC
Elavl4	qRT	CCAAAGGATGCAGAGAAAGC	GGGAAGGCCACTAACGTACA
Nova2	qRT	CTCAATCATCGGCAAAGGTG	GGCTCTGTCGGGGTTCATC
Rbfox1	qRT	GGGCGAGGAGGGGAAGGGAG	GTGCGCACTGTAGCAGGCCA
Rbfox2	qRT	GCAAATGGCTGGAAGTTAAGC	CATTGCCTAGGGACACATCA
Rest	qRT	TGAAAAGTCCGTCAAAGCAG	GCACATCCATCTCTTTCACCT
Elavl1	qRT	TGACAAACGGTCTGAAGCAG	CCTGAATCTCTGTGCCTGGT
Nova1	qRT	GGAGCAGTCAGGGGCTTGGG	TGAGACAGCTGCCACTCTGTGGA
Ins2	qRT	TGACCAGCTACAGTCGGAAA	GTTGCAGTAGTTCTCCAGTTGG
Gsn	qRT	CTTTGTTTGGAAAGGCAAGC	GGGGTACTGCATCTTGGAGA
Cacna1c	SPL	AGTGATCCCTGGAATGTTTTTG	AGGACTTGATGAAGGTCCACAG
Insr	SPL	TGCACAACGTGGTTTTTGTT	TCTCTTCTGGGGAGTCCTGA
Gsn	SPL	ATGGCTCCGTATTGCTCCT	AACTTCTCCACACGCCAGAT
Human
ACTB	qRT	GCGCGGCTACAGCTTCA	CTTAATGTCACGCACGATTTCC
ELAVL4	qRT	GCTACCCAGGTCCACTTCAC	GGATGTTCATTCCCACAAGG
NOVA2	qRT	CCATCAAGCTCTCCAAGTCC	GGGATTTCTCGGACCTTCTC

Insulin Secretion

To assess insulin secretion, INS-1E cells were pre-incubated for 1 h in glucose-free RPMI 1640 Gluta MAX-I medium (Life Technologies, Inc.) followed by incubation with Krebs-Ringer solution for 30 min. Cells were then exposed to 1.7, 17, or 17 mm glucose plus 20 μm forskolin or to 30 mm KCl during 30 min. Insulin release and insulin content were measured using the rat insulin ELISA kit (Mercodia, Uppsala, Sweden) in cell-free supernatants and acid/ethanol-extracted cell lysates, respectively. Results were normalized by total protein content, determined by the Bradford dye method.

Glucose Oxidation

d-[U-¹⁴C]Glucose (specific activity, 300 mCi/mm; concentration, 1 mCi/ml (PerkinElmer Life Sciences) was used to assess glucose oxidation in control and Rbfox1 KD cells exposed to different glucose concentrations as described (76). The rate of glucose oxidation was expressed as pmol/10⁵ cells·120 min⁻¹ and normalized by the glucose oxidation observed in control cells under low glucose concentrations.

Mitochondrial Respiration

Oxygen consumption rates were measured by the XFp Extracellular Flux Analyzer (Seahorse Bioscience, North Billerica, MA). INS-1E cells were pre-incubated for 1 h in assay medium containing 1.7 mm glucose at 37 °C in air after which respiration was measured following sequential injections of 17 mm glucose, 5 μm oligomycin, 4 μm FCCP, and 1 μm rotenone as described (77).

Electrophysiological Measurements

Experiments were conducted on single INS-1E cells. Patch pipettes were pulled from borosilicate glass capillaries and then coated with sticky wax (Kemdent, Wiltshire, UK) and fire-polished. Patch pipettes had a resistance of 3–6 megohms when filled with the pipette solution specified below. All recordings were conducted at 32–34 °C using an EPC-9 patch clamp amplifier and Pulse software version 8.80 (HEKA Electronics, Lambrecht, Germany) without visualization of the transfected cells. Standard whole-cell configurations of the patch clamp technique were used, and Ca²⁺ currents were evoked by depolarizing the cell membrane from a resting potential of −70 mV to voltages ranging from −50 to +50 mV (78). Standard extracellular solution consisted of the following (in mm): 118 NaCl, 20 tetraethylammonium chloride (to block K⁺ currents), 5.6 KCl, 2.6 CaCl₂, 1.2 MgCl₂, 5 glucose, and 5 HEPES (pH 7.4 using NaOH), and the pipette solution consisted of the following (in mm): 125 cesium glutamate, 10 NaCl, 10 CsCl, 1 MgCl₂, 0.05 EGTA, 3 Mg-ATP, 5 HEPES (pH 7.15 using CsOH) and 0.1 cAMP. To exclude any depolarization-induced Na⁺ currents from the analysis, charge (Q) was measured 1.4 ms after the onset until the end of the depolarization.

Actin Filaments Immunofluorescence

INS-1E cells were cultured on glass slides coated with Matrigel-fibronectin (100 and 2 μg/ml, respectively; Sigma) and transfected with control or Rbfox1 siRNAs for 48 h. Following the same protocol used to analyze insulin secretion, cells were incubated for 1 h in glucose-free RPMI 1640 GlutaMAX-I medium followed by 30-min incubation in glucose-free Krebs-Ringer solution. Cells were then exposed to 17 mm glucose at different time points (5, 10, and 15 min) and immediately after fixed in 4% paraformaldehyde for 10 min. The cells were then washed with PBS and permeabilized in Triton X-100 0.25% for 3 min. Slides were then blocked using 3% BSA and incubated at room temperature in sequential order with deoxyribonuclease I-Alexa Fluor 488 conjugate (Life Technologies, Inc.) and Atto 550-phalloidin (Sigma) to specifically stain G-actin (globular) and F-actin (filaments), respectively. Nuclei were stained with Hoechst 33342. Cells were visualized under a Zeiss Axiovert 200 inverted fluorescence microscope (Carl Zeiss, Zaventem, Belgium) at ×40. Images were acquired and processed using the AxioVision LE64 software (Carl Zeiss). For F/G actin quantification, all samples were processed together to avoid changes in the image intensity. Exposure was locked to 100 ms to obtain all acquisitions under the same conditions. ImageJ software was used to measure the pixel intensity of both channels using the whole field of every image. These intensities were used to calculate the F/G actin ratio.

Confocal Imaging

Cells samples were observed under a Zeiss LSM780 inverted confocal microscope using ×40/1.1 water objective. Z-stacks of images were acquired sequentially and processed using Zen2010 software (Carl Zeiss).

Western Blotting

For Western blotting, cells were washed with cold PBS and lysed using Laemmli Sample Buffer. Total protein was extracted and resolved by 8–12% SDS-PAGE, transferred to a nitrocellulose membrane, and immunoblotted with the specific antibodies for the protein of interest (Table 4) as described (79). Protein signal was visualized using chemiluminescence Supersignal (Pierce). ChemiDoc MP system and Image Lab software version 4.1 (Bio-Rad, Perth, UK) were used for image acquisition and densitometric analysis of the blots. The software allows automatic exposure optimization, ensuring the best use of the dynamic range. Saturation or overexposure is avoided by overriding the images that displayed pixels in red. Densitometric values were corrected by the housekeeping protein α-tubulin as loading control, after background subtraction.

TABLE 4.

Antibodies and fluorescent labels used for Western blotting and immunofluorescence

Antibody	Company	Reference	Dilution
Elavl4 (rabbit)	Abcam, Cambridge, UK	ab96474	1:1000
Nova2 (rabbit)	MBL, Ina, Japan	RN044PW	1:500
Rbfox1 (rabbit)	Abcam, Cambridge, UK	Ab154490	1:1000
Cleaved caspase 3 (rabbit)	Cell Signaling, Danvers, MA	9661S	1:1000
Cleaved caspase 9 (rabbit)	Cell Signaling, Danvers, MA	9507	1:1000
Tubulin (mouse)	Sigma, Bornem, Belgium	T5168	1:10,000
HRP-conjugated anti-rabbit IgG	Jackson ImmunoResearch	711-036-152	1:10,000
HRP-conjugated anti-mouse IgG	Jackson ImmunoResearch	711-036-150	1:10,000
Deoxyribonuclease I-Alexa Fluor 488	Life Technologies, Inc.	D12371	1:500
Atto 550-phalloidin	Sigma, Bornem, Belgium	19083	1:500

Statistical Analysis

Data are showed as means ± S.E. or as independent data points. Significant differences between experimental conditions were assessed by a paired Student's t test. In case of multiple t test comparisons, the Bonferroni correction was applied as indicated (80). p values < 0.05 were considered statistically significant.

Author Contributions

J. J.-M. and T. H. R. designed the study, researched the data, performed analyses, and wrote the manuscript. O. V., E. L-M., A. W., J.-V. T., L. A. B., T. R. N., T. C. N., M. I. A., J. L. S. E., and L. E. researched the data. P. M. contributed with material and critically revised the manuscript. D. L. E. designed the study, analyzed the data, and wrote the manuscript. All co-authors read the final version of the manuscript. D. L. E. and J. J.-M. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analyses.