Targeting the NADPH Oxidase-4 and Liver X Receptor Pathway Preserves Schwann Cell Integrity in Diabetic Mice

Diabetic peripheral neuropathy (DPN) is one of the most common and debilitating complications of diabetes, characterized by myelin defects, impaired nerve conduction velocity (NCV), abnormal sensation and proprioception, accompanied by axonal atrophy with blunted regenerative potential, leading to the loss of nerve fibers, pain, poor quality of life, foot ulcers, and ultimately amputations (1). Until now, no effective treatment for DPN has been described (1,2).

Emerging evidence suggests that among the different types of cells that are affected by DPN, Schwann cells (SCs) play a significant role in the pathology of the disease (3). In the peripheral nervous system, myelinating SCs maintain myelin sheath integrity by expressing structural proteins, like Myelin Protein Zero (MPZ) and Peripheral Myelin Protein 22 (PMP22), that are involved in the compaction and the stabilization of myelin sheaths around the axons. Alterations in the expression of these proteins have been shown to provoke deleterious changes in the peripheral nerve (4). Despite the efforts to understand DPN onset and progression, the mechanisms underlying demyelination still remain elusive.

In recent years, our group and others showed the importance of the liver X receptors (LXRs) in re/myelination processes (5–7). LXRs are oxysterol-activated transcription factors consisting of two isoforms, LXRα and LXRβ, with the latter being the major isoform in the nerve (7). Nonetheless, the pathophysiological role exerted by LXR on SC integrity and function in DPN remains unclear.

Alongside, experimental and clinical data show that oxidative stress is a common feature in the pathogenesis of DPN (8,9); however, antioxidant therapy has been insufficient in halting the progression of this complication (10,11). Subsequently, targeted therapy that is based on identifying specific cellular sources of reactive oxygen species (ROS) should be implemented. The sources of ROS that emerged within an inflammatory context leading to DPN include the cyclooxygenase COX-2 (12,13) and 12,15-lipooxygenases (14,15). Inflammation occurs early in the development of diabetes, and in the presence of additional risk factors, it results in DPN. Therefore, targeting inflammation in diabetes-induced peripheral injury is a mechanism-based strategy that needs further investigation. Furthermore, aldose reductase activation, known to enhance ROS production as a result of the elevated polyol flux, has been shown to contribute to DPN. Conversely, the beneficial effect of aldose reductase inhibitors is limited (16). Likewise, advanced glycated end products and their receptor enhance ROS production in primary sensory neurons through an NADPH-dependent mechanism (17). Still, targeting this pathway for DPN therapy has not held promising results. Moreover, organelle-associated ROS production has been shown to be implicated in DPN pathologies as well (18,19). For instance, impaired mitochondrial processes, electron transfer, and shuttling lead to an overload in the mitochondrial electron transfer chain and a reduced mitochondrial action potential. These, in turn, induce redox changes associated with the NADPH oxidases and endothelial nitric oxide synthase systems, leading to DPN (18,20). Conjointly, and among the various sources of ROS (8,12), these observations attest to the notable contribution of NADPH family of enzymes in the direct and/or indirect pathogenesis of DPN. The NADPH oxidases family (Noxes) emerged as major enzymes responsible for the production of superoxide and hydrogen peroxide (H₂O₂) (21). Noxes includes five members each with tissue and disease-specific regulation, including diabetes-induced complications (21–25). In DPN, few studies tackle the role of Noxes in DPN (26,27). To our knowledge, no investigations have been conducted on the direct involvement of the NADPH oxidases, specifically NADPH oxidase-4 (Nox4), in DPN.

In this study, we provide the first evidence that Nox4-derived ROS contribute to diabetes-induced neuropathy and SC injury in cultured SCs, diabetic animals, and human skin biopsies. This effect is mediated by inactivation of the LXR pathway. We further demonstrate that the activation of LXR or inhibition of Nox4 blocks the effect of high-glucose/hyperglycemia-induced NADPH-dependent ROS generation. Importantly, we show that this effect is associated with neuroprotection, repair of the myelin profile, and reduction in SC injury. These findings open new perspectives in the treatment of DPN.

All animal procedures were conducted in accordance with the U.S. Public Health Service’s Policy on Humane Care and Use of Laboratory Animals and were approved by the institutional animal care and use committee at the American University of Beirut and Paris Descartes University. Animal protocols and phenotyping were used as per the Animal Models of Diabetic Complications Consortium procedures. At 6 weeks of age, Swiss Webster mice were made type 1 diabetic (T1DM) using consecutive intraperitoneal streptozotocin (STZ) injections (85, 70, and 55 mg/kg; dissolved in citrate buffer, pH 4.5). This T1DM animal model is widely used in DPN because it displays a neuropathic phenotype as well as alterations in neurological functions resembling human neuropathic alterations (28). Control mice were injected with citrate buffer alone. Diabetes (defined as fasting glucose ≥250 mg/dL) was confirmed 7–10 days after the first injection. At the onset of diabetes, control and T1DM mice were treated with either 40 mg/kg body weight of GKT137831 (GKT), an orally bioavailable low-toxicity compound acting as a dual inhibitor of NADPH oxidase isoforms Nox4 and, to a lesser extent, Nox1 and generously provided by Genkyotex SA (Geneva, Switzerland) or a low dose of 5 mg/kg body weight of the LXR agonist T0901317 (T0). This dose of T0 was used to minimize its potential side effect. Both drugs were administered three times a week by oral gavage for 10 weeks. In parallel experiments, normoglycemic age-matched wild-type (WT) male and LXRα/β^−/− mice on a mixed-strain background (C57BL/6:129Sv) were also used in this study.

Sensorimotor dysfunction was assessed using the raised beam walking and the hind paw withdrawal (thermal algesia) tests (6,7,29). The magnitude of nerve dysfunction was assessed using the NCV test (30).

Mouse SCs (MSC80), preserving the characteristics of normal SCs (31), were cultured in 5 mmol/L glucose (normal glucose [NG]) or treated with 25 mmol/L glucose (high glucose [HG]) for 48 h in the presence or absence of 20 μmol/L GKT or in the presence or absence of 5 μmol/L T0. For the RNA interference experiments, a SMARTpool consisting of siRNA duplexes specific for mouse Nox4 or LXRβ were purchased from Dharmacon, and the transfection with the siRNA of interest (100 nmol/L) or with the control nontargeting siRNA (scramble [Scr]) were performed as previously described (32).

Fourteen paraffin-embedded human skin biopsies obtained from healthy individuals and patients with type 2 diabetes (T2DM) between 2013 and 2015 were used. The institutional review board of Cochin Hospital (Paris, France) approved the experiments, and all participants signed a consent form. The clinical characteristics were determined on the basis of the physician/hospital-saved records (Table 1). To assess neuropathy in these patients, the Michigan Diabetic Neuropathy Score (MDNS) was used. This test allows the diagnosis and staging of diabetic neuropathy (33).

mRNA was analyzed by quantitative real-time PCR using the ΔΔC_t method (22,23,32). mRNA expression was quantified using a CFX96 Touch thermal cycler (Bio-Rad, Hercules, CA) with SYBR Green dye and mouse and human RT² qPCR Primers (QIAGEN, Germantown, MD) of the corresponding gene of interest.

Homogenates from sciatic nerves or MSC80 were prepared, and Western blot (WB) analysis was performed as previously described (22,23,32) using rabbit polyclonal antibodies against Nox4 (1:500; Santa Cruz Biotechnology), P0 (1:1,000; Abcam), and PMP22 (1:1,000; Sigma). Densitometric analysis was performed using National Institutes of Health ImageJ software.

NADPH oxidase activity was measured in MSC80 or in sciatic nerve homogenates as previously described (22,23,32).

The peroxide-sensitive fluorescent probe 2′,7′-dichlorodihydrofluorescin (DCF) diacetate (Molecular Probes) was used to measure intracellular ROS in MSC80 (22), and cellular H₂O₂ in sciatic nerves was assessed by high-performance liquid chromatography (HPLC) analysis of dihydroethidium (DHE)-derived oxidation products (23).

Protein misfolding is a common contributor to cellular injury. PMP22 is one of the aggregate-prone proteins identified to contribute to neurodegeneration and SC injury (34). PMP22 aggregation was performed on lysates prepared from MSC80 and incubated with the PMP22 antibody (Sigma) as previously described (6).

Immunohistochemistry for S100 protein (antibody dilution determined by pretitration; Abcam) was done on paraffin-embedded tissue sections as previously described (35).

Results are expressed as mean ± SEM. Statistical significance, determined as P < 0.05, was assessed by one-way ANOVA, followed by Tukey posttest when more than two variables were analyzed. Two-group comparisons were performed by Student t test. All the observations are truly representative of the underlying phenomena.

All data generated or analyzed during this study are included in the published article. For the complete set of figures generated, these are available from the corresponding author upon request. No novel resources were generated/analyzed during the study.

To assess the role of LXR in DPN, we initially examined the mRNA level of LXRβ in the sciatic nerves isolated from STZ-induced T1DM mice after 10 weeks of diabetes onset. LXRβ was highly expressed in the nerves (7). Our results show that hyperglycemia reduces the mRNA levels of LXRβ (Fig. 1A). Ligand-dependent activation of LXRs by oxysterols induces the expression of specific target genes involved in cholesterol and lipid metabolism, including the ATP-binding cassette (ABC) transporters, especially ABCA1. To determine whether LXRs are active in the sciatic nerves of the T1DM mice, we treated the animals with the nonsteroidal synthetic LXR agonist T0 and found that it significantly induced the expression of ABCA1 (Fig. 1B). Of note, treatment with T0 did not alter glycemic levels (550 ± 45 vs. 537 ± 50; STZ-induced T1DM vs. STZ-induced T1DM + T0). Further, mouse body weight was not significantly changed in the different groups (40.5 ± 3.2 g vs. 36.8 ± 6.3 g STZ-induced T1DM vs. 37.2 ± 4.3 g STZ-induced T1DM + T0). We next examined the impact of LXRβ alteration on nerve dysfunction and locomotor and sensory impairments brought about by DPN (Fig. 1C–I). Our data show an overall reduction in NCV of STZ-induced T1DM mice relative to their controls (Fig. 1C and D). Treatment with T0 for 10 weeks ceased the neurophysiological defects seen in STZ-induced T1DM (Fig. 1C and D). Interestingly, the results of the hind paw withdrawal test show that the T1DM mice took a significantly longer time to sense the heat of the beam and withdraw their paws relative to the controls (Fig. 1E). T0-treated animals had a significantly lower latency, suggesting that LXR activation restored thermal hypoalgesia (Fig. 1E). In parallel, fine motor coordination and balance (Fig. 1F–I) show a relatively longer period of time for the LXRdKO animals to cross the beam (Fig. 1F and G), with an increased tendency to slip (Fig. 1H) and stop (Fig. 1I) in comparison with the controls that crossed the beam with minimal setbacks. It should be noted that T0 treatment in control mice did not have any effect on the behavioral function of the animals (Fig. 1C–I). Collectively, these results demonstrate that activating LXR pathway ameliorates diabetes-induced neurophysiological defects and sensorimotor abnormalities.

To correlate the behavioral changes brought about by diabetes with cellular and molecular alteration, we examined the role of LXR in myelin injury. Our findings suggest that T1DM-induced myelin-related abnormalities, evident by increased mRNA and protein expression of MPZ and PMP22, was halted by T0 treatment (Fig. 2A–D). To correlate whether the effects seen in T1DM animals could be due, at least in part, to SC injury, MSC80 cells were exposed to HG for 48 h in the absence or presence of T0. HG induces a significant upregulation of MPZ and PMP22 mRNA and protein levels (Fig. 2E–H), an effect significantly diminished upon LXR activation (Fig. 2E–H). Likewise, PMP22 aggregation markedly observed in SCs exposed to HG was discontinued upon T0 treatment (Fig. 2I). To further validate our observations, silencing of LXRβ in MSC80 using siRNA targeting LXRβ in the absence of glucose treatment mimicked the injurious effect of HG treatment (Fig. 2J–N). Taken together, our results show that LXR activation restores myelin homeostasis in SCs. It should be noted that in all our in vitro SC cultured experiments, equimolar concentrations of mannitol were used as osmotic control and did not have any effect on the performed experiments when compared with NG-treated SCs (Fig. 2O and P).

Next, we determined whether LXR regulates Nox4-induced ROS production in the sciatic nerves of the STZ-induced T1DM mice. This was evident by the decrease in superoxide generation, NADPH oxidase activity, and Nox4 mRNA levels and protein expression upon T0 treatment (Fig. 3A–D). These observations were also confirmed in MSC80 cells (Fig. 3E–H).

To further validate the cross talk between LXR and Nox4, sciatic nerves isolated from age-matched adult normoglycemic LXRα/β^−/− (LXRdKO) mice were used. These mice exhibit a reduced NCV, severe locomotor deficits, muscular weakness, and peripheral myelin injury (6). Our results show that knocking down LXR significantly increased NADPH-dependent superoxide generation (Fig. 3I) and enhanced Nox4 mRNA levels and protein expression in the sciatic nerves of the LXRdKO animals (Fig. 3J and K). Consistent with the in vivo observations, silencing of LXRβ in cultured SCs exposed to NG significantly induced ROS production, enhanced NADPH oxidase activity, and increased Nox4 mRNA levels and protein expression (Fig. 3L–O). Taken together, these results are the first, to our knowledge, to suggest a possible functional link between LXR and Nox4 in the pathogenesis of DPN.

To evaluate whether the alteration of Nox4 plays a role in nerve injury, STZ-induced T1DM mice were treated with GKT for 10 weeks. Treatment with GKT did not affect glucose levels (545 ± 42 vs. 513 ± 40; STZ-induced T1DM vs. STZ-induced T1DM + GKT). Furthermore, mouse weight was not significantly changed in the different groups (44.7 ± 4.3 vs. 38.6 ± 7.2 STZ-induced T1DM vs. 39.4 ± 5.6 STZ-induced T1DM + GKT). Nevertheless, treatment with GKT improves the attenuated motor NCV (MNCV) and sensory NCV (SNCV) in the STZ-induced T1DM mice to levels similar to the controls (Fig. 4A and B). In addition, GKT treatment prevents thermal hypoalgesia (Fig. 4C) and reduces sensorimotor dysfunction evoked by diabetes (Fig. 4D–G). These results indicate that hyperglycemia-induced Nox4 upregulation mediates neurophysiological and behavioral defects in T1DM.

Next, we assessed the involvement of Nox4 in demyelination and SC injury in STZ-induced T1DM. Our results show that GKT treatment significantly reinstated mRNA and protein levels of MPZ and PMP22 to physiological control levels (Fig. 4H–K). These results were paralleled with a decrease in ROS production and NADPH oxidase activity in the T1DM-treated mice (Fig. 4L and M).

Similarly, our findings correlated with our in vitro data showing a significant decrease in the upregulation of MPZ and PMP22 mRNA levels and protein expression and a decrease in the induction of PMP22 aggregation in SCs exposed to HG for 48 h in the presence of GKT (Fig. 5A–E). These results were paralleled with a decrease in ROS production and NADPH oxidase activity (Fig. 5F and G). Similar results were observed in cells transfected with small interfering Nox4 (siNox4) in the presence of HG (Fig. 5H–P). These results suggest that Nox4 inhibition is a candidate target to prevent demyelination-prompted SC injury in diabetes.

To cross-correlate our animal observations in a clinical setting, we examined the gene expression of LXRβ, ABCA1, and Nox4 in skin biopsy samples of patients with long-standing T2DM with or without peripheral neuropathy. The clinical characteristics of the control and T2DM groups are shown in Table 1. Briefly, in both control and T2DM groups, 57.1% were male and 42.9% were female. Mean BMI, dyslipidemia, and arterial hypertension were not significantly different between groups. HbA_1c was high in the T2DM group and significantly superior compared with the control group. Peripheral neuropathy was found in 42.9% of the patients with T2DM and was nonexistent among the control individuals. Antidiabetic drug usage was only recorded in the T2DM group, while comedications with antihypertensive (cardiovascular drugs) or lipid-lowering drugs (including statins) were prescribed to approximately 57.1% of the control individuals and to 85.7% of patients with T2DM.

Immunohistochemical staining of paraffin sections shows S100 protein staining in the skin biopsies (Fig. 6A). S100 alteration in skin biopsies is suggested to identify signs of SC alteration in peripheral neuropathy and is used as a potential biomarker of distal myelin and mechanoreceptor integrity (36). RNA extracted from these biopsies shows a SOX10 mRNA expression, highlighting the existence of SCs in the used biopsy model (Fig. 6B). Additionally, LXRβ and ABCA1 mRNA levels were significantly reduced (Fig. 6C and D), while Nox4 levels were considerably increased (Fig. 6E) in the skin samples of the patients with T2DM. More importantly, LXRβ and ABCA1 gene expressions were significantly lower and Nox4 gene significantly more expressed in the patients with T2DM with clinical signs of neuropathy compared with the patients with T2DM without neuropathy (Fig. 6F–H).

Taken together, the human skin biopsy data correlate with our cultured cells and experimental T1DM animal model observations. Altogether, they suggest that diabetes pathologically affects SCs and triggers myelin injury in the peripheral nervous system by activating the ROS producer Nox4 and inhibiting a cellular safeguard, the LXR pathway (Fig. 7).

DPN is the most prevalent complication of diabetes, characterized by functional and structural changes in peripheral nerves (37,38), and correlates with impaired myelination. SCs have been implicated in the pathogenesis of DPN (11), yet the mechanisms of SC injury in diabetes are incompletely characterized. In this study, we provide evidence that sciatic nerves of STZ-induced T1DM mice exhibit major morphological and biochemical injury of the sciatic nerve, a phenomenon that was also observed in cultured SCs in the presence of HG. Importantly, our results were also observed in diabetic human biopsies, where LXR and Nox4 showed a similar alteration to the one observed in cultured SCs and T1DM mice. To the best of our knowledge, this is the first study that highlights the involvement and the cross talk of LXR and Nox4 in diabetes/HG-induced sciatic nerves and SC injuries, thus suggesting new therapeutic avenues for the treatment of DPN.

In the first part of our study, we evaluated the physiological changes induced by hyperglycemia/HG on sciatic nerves isolated from T1DM mice as well as on cultured SCs. Our data show an alteration in MPZ and PMP22, paralleled by an induction of PMP22 aggregation. These data corroborate findings from other demyelinating diseases (e.g., Charcot-Marie-Tooth) highlighting the crucial role of MPZ and PMP22 in maintaining myelin sheath functionality and nerve integrity (6,7) where injury has been linked to the inability of SCs to effectively activate an appropriate transcriptional repair response, an effect paralleled by myelin dysfunction and upregulated myelin protein levels central to the proper physiology and survival of SCs (39). Relatedly, Cermenati et al. (5) have reported an alteration in major myelin proteins in diabetes-induced peripheral injuries, thus highlighting the role of myelin in DPN. However, in their report, myelin alteration was correlated to a reduced MPZ expression (5). While it is noteworthy to advance that any alteration in the myelin protein expression correlates with nerve injury (5–7), the discrepancy in MPZ expression between our study and that of Cermenati et al. could be explained either by the difference in the used animal models (mice vs. rats) or by the duration of diabetes (early vs. advanced) (40).

The pathobiochemical changes that we observed were associated with reduced nerve function highlighted by reduced thermal sensitivity, declined motor function and balance, and reduced NCV. While allodynia and hyperalgesia represent two common phenotypes occurring in diabetes (41), reduced NCV is considered the hallmark of clinical manifestations of DPN, and it affects sensorimotor coordination, leading to loss of proprioception and locomotor defects (42). Our behavioral results show a clear correlation between diabetes-induced peripheral nerve dysfunction and demyelination.

Additionally, our results suggest that activation of LXR ameliorates DPN and blocks SC injury. We identified the LXR/ABCA1 signaling pathway as one of the mechanisms through which hyperglycemia/HG lead to SC damage and nerve injury, an observation that is manifested in cultured cells, T1DM animals, and human skin biopsies. Although limited information exists on the role of LXR in DPN, our results are in line with other published work showing the favorable effect of LXR activation on other diabetic complications. For instance in diabetic nephropathy, expression levels of both LXRs were significantly decreased in animal models of T1DM (43) and in patients with diabetic nephropathy (44). Moreover, LXR synthetic activation preserved renal glomerular integrity in the diabetic kidney by reducing endoplasmic reticulum stress (45). Similarly, LXR activation in diabetic animals exerted cardioprotective effects against diabetic cardiomyopathy by modulating myocardial apoptosis (46).

To a further extent, a deeper dissection of the mechanisms altered in DPN led us to identify a cross talk between LXR inactivation and activation of Nox4-induced ROS production. Our results lend support to the hypothesis that hyperglycemia/HG-induced oxidative stress commences through an NADPH-dependent mechanism. To our knowledge, this is the first study to demonstrate that Nox4 induces SC injury through a direct cross talk with LXR, even in the absence of a metabolic dysfunction. We show that activation of LXR by T0 suppresses the activation of the NADPH oxidases/Nox4 mechanism. Moreover, knockout of LXRβ in cultured SCs and in the sciatic nerves of adult LXRα/β^−/− mice enhances ROS production by activating the NADPH oxidases/Nox4 signaling and mimics the structural and functional changes induced by diabetes in vitro and in vivo. Previous reports, including those from our group, have highlighted the critical role of oxidative stress and ROS production in the pathogenesis of diabetic complications (22,23,25). For instance, and with regard to DPN, the reduction of oxidative stress in experimental diabetes may have therapeutic potential (26). In fact antioxidant treatments attenuate the progression of neurovascular deficits in DPN (9). Efforts attempting to inhibit ROS production resulted in the partial restoration of perineurium morphology, NCV, and pain and thermal perception (8,26,47). Despite the controversy regarding the exact response of SCs to diabetes-induced oxidative stress (11), our results support the published findings whereby ROS inhibition ameliorated SC injury (48).

Particular interest in the current research focused on identifying the sources of ROS in a disease-specific manner to introduce more-explicit therapies (9,10,22,32,49). We and others have highlighted the critical role of Nox4-derived ROS in the pathogenesis of other diabetic complications (22,23,25), although, to our knowledge, this study is the first to examine the effect of the specific inhibition of Nox4 on SCs and nerves within the context of DPN. Our results show that administration of GKT to T1DM mice restored peripheral myelin protein expression and resulted in neuroprotection, thereby improving NCV, sensorimotor deficits, and thermal sensitivity. These in vivo observations were also confirmed in vitro, where SCs were protected from HG-induced injury upon inhibition of Nox4 by GKT or siRNA. Likewise, our data corroborate that described in diabetic cardiomyopathy or nephropathy, where pharmacological inhibition of NOX1 and/or NOX4 reduce or reverse disease progression. More importantly, these reports showed that consistent with Nox4^−/− mouse data, GKT shows a similar protective effect (50). Along these lines, other studies showed that Nox4 expression is tightly associated with neuropathic pain signaling in animal models, and it has been shown that thermal hyperalgesia was significantly attenuated in neuropathic pain–induced animals deficient for Nox4 (24).

In brief, our results advance, for the first time, a previously unrecognized link between LXR and NADPH oxidases, especially the Nox4 subunit, in HG-induced nerve and SC injury and suggest that the synthetic activation of LXR attenuates Nox4 expression and improves DPN injury. On a final clinical note, although it is now widely accepted that oxidative stress is the unifying mechanism in the pathogenesis of DPN, the administration of antioxidants does not result in complete protection against nerve dysfunction in human diabetes. Our observations indicate that inhibition of Nox4 and/or the activation of LXR may be of therapeutic relevance in addition to glycemic control to alleviate diabetic peripheral nerve damage.

Acknowledgments. The authors thank the American University of Beirut Animal Care Facility staff for help in taking care of the animals used in this study. The authors also thank Genkyotex for providing the Nox1/Nox4 inhibitor (GKT).

Funding. S.A.E. was funded by a predoctoral fellowship from the Lebanese National Council for Scientific Research (CNRS-L). This work was partially funded by a Medical Practice Plan-American University of Beirut grant to A.A.E. and a CEDRE Program project grant to A.A.E. and C.M. This work was supported by University Paris Descartes and INSERM to C.M.

Duality of Interest. P.W. is currently the executive vice president and chief medical officer at Genkyotex SA. C.S. is employed by Genkyotex SA. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. S.A.E. performed the experiments and wrote the manuscript. M.E.M. helped in performing some of the experiments. M.Hi., M.Ha., and S.B. helped in performing part of the animal experiments. J.G., B.D., and R.B. participated in the discussion of the results. J.C. and S.A. provided the human skin samples. P.W. and C.S. provided the Nox1/Nox4 dual inhibitor (GKT). S.T.A., C.B., and G.Z. provided critical scientific input to the experiments. A.A.E. and C.M. conceived and designed the study. All authors reviewed the results, provided essential reviews of the manuscript, and approved the final version of the manuscript. A.A.E. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for integrity of the data and the accuracy of the data analysis.