Resveratrol prevents renal lipotoxicity and inhibits mesangial cell glucotoxicity in a manner dependent on the AMPK–SIRT1–PGC1α axis in db/db mice
- First Online:
- Cite this article as:
- Kim, M.Y., Lim, J.H., Youn, H.H. et al. Diabetologia (2013) 56: 204. doi:10.1007/s00125-012-2747-2
- 4.1k Downloads
Many of the effects of resveratrol are consistent with the activation of AMP-activated protein kinase (AMPK), silent information regulator T1 (SIRT1) and peroxisome proliferator-activated receptor (PPAR)γ co-activator 1α (PGC-1α), which play key roles in the regulation of lipid and glucose homeostasis, and in the control of oxidative stress. We investigated whether resveratrol has protective effects on the kidney in type 2 diabetes.
Four groups of male C57BLKS/J db/m and db/db mice were used in this study. Resveratrol was administered via gavage to diabetic and non-diabetic mice, starting at 8 weeks of age, for 12 weeks.
The db/db mice treated with resveratrol had decreased albuminuria. Resveratrol ameliorated glomerular matrix expansion and inflammation. Resveratrol also lowered the NEFA and triacylglycerol content of the kidney, and this action was related to increases in the phosphorylation of AMPK and the activation of SIRT1–PGC-1α signalling and of the key downstream effectors, the PPARα–oestrogen-related receptor (ERR)-1α–sterol regulatory element-binding protein 1 (SREBP1). Furthermore, resveratrol decreased the activity of phosphatidylinositol-3 kinase (PI3K)–Akt phosphorylation and class O forkhead box (FOXO)3a phosphorylation, which resulted in a decrease in B cell leukaemia/lymphoma 2 (BCL-2)-associated X protein (BAX) and increases in BCL-2, superoxide dismutase (SOD)1 and SOD2 production. Consequently, resveratrol reversed the increase in renal apoptotic cells and oxidative stress, as reflected by renal 8-hydroxy-deoxyguanosine (8-OH-dG), urinary 8-OH-dG and isoprostane concentrations. Resveratrol prevented high-glucose-induced oxidative stress and apoptosis in cultured mesangial cells through the phosphorylation of AMPK and activation of SIRT1–PGC-1α signalling and the downstream effectors, PPARα–ERR-1α–SREBP1.
The results suggest that resveratrol prevents diabetic nephropathy in db/db mice by the phosphorylation of AMPK and activation of SIRT1–PGC-1α signalling, which appear to prevent lipotoxicity-related apoptosis and oxidative stress in the kidney.
KeywordsApoptosisDiabetic nephropathyOxidative stressResveratrol
AMP-activated protein kinase
BCL-2-associated X protein
B cell leukaemia/lymphoma 2
Cellulose sodium salt
- db/db Res
Diabetic mice receiving resveratrol
- db/m Res
Non-diabetic mice receiving resveratrol
Endothelial nitric oxide synthase
Cell surface glycoprotein F4/80
Class O forkhead box
Nuclear factor κB
PPARγ co-activator 1α
Peroxisome proliferator-activated receptor
Reactive oxygen species
Silent information regulator T1
Sterol regulatory element-binding protein 1
Diabetic nephropathy is the most prevalent cause of end-stage renal disease in developed countries, including Korea . Despite progress in pharmacological strategies to modulate diabetes, the number of patients entering renal failure remains extremely high and the development of new classes of therapeutic agents is eagerly anticipated. Lipotoxicity is one of the causal factors of the progression of diabetic and obesity-induced renal damage , but the underlying molecular mechanism remains elusive.
Resveratrol is a natural plant polyphenol that may target ageing and obesity-related chronic diseases by preventing inflammation and oxidative stress [3–6]. Resveratrol activates silent information regulator T1 (SIRT1), an NAD+-dependent protein deacetylase, and AMP-activated protein kinase (AMPK). It subsequently augments peroxisome proliferator-activated receptor (PPAR)γ co-activator 1α (PGC-1α), endothelial nitric oxide synthase (eNOS) and class O forkhead box (FOXO) activation, which is followed by catabolic metabolism, mitochondrial activation, angiogenesis and enhanced cell survival [3–6]. Some of the best-characterised effects of SIRT1 include increased stress resistance and altered metabolism mediated by changes in the activity of the transcriptional factor FOXO , suppression of nuclear factor κ B (NF-κB)-dependent inflammatory responses  and the promotion of gluconeogenesis, fatty acid oxidation and mitochondrial biogenesis, through PGC-1α [5, 6]. The activation of SIRT1 by resveratrol is substrate dependent and many of its effects are consistent with the modulation of its target genes [7, 8]. Hepatocyte-specific deletion of Sirt1 impairs PPARα signalling and decreases fatty acid β-oxidation, whereas the overexpression of Sirt1 induces the expression of PPARα targets. SIRT1 interacts with PPARα, and this is required to activate the PPARα co-activator PGC-1α. It has been reported that SIRT1 and PGC-1α form a stable complex and that SIRT1 regulates the activity and acetylation status of PGC-1α . AMPK is a fuel-sensing enzyme that is activated by decreases in the cellular energy state, as reflected by an increased AMP/ATP ratio . When activated, AMPK initiates metabolic and genetic processes that generate ATP (fatty acid oxidation), and inhibits other processes that consume ATP. The former occurs via its effects on various transcriptional activators and co-activators, including PGC-1α and PPARα . The collective results favour the view that SIRT1 and AMPK play a vital role in the regulation of organic lipid homeostasis, and that pharmacological activation of SIRT1 and AMPK may be important for the prevention of obesity-associated metabolic diseases [8, 10, 11].
Accumulated experimental evidence suggests that PPARα activation attenuates or inhibits several mediators of vascular injury, including lipotoxicty, inflammation, reactive oxygen species (ROS) generation, endothelial dysfunction, angiogenesis and thrombosis in type 2 diabetes and high-fat-diet-induced renal damage [12, 13], which are all associated with AMPK activation and eNOS production [14–16]. PPARs activate PGC-1α/β and its key downstream effector oestrogen-related receptor (ERR)-1α, which induces mitochondrial biogenesis and enhanced mitochondrial antioxidative capacity to provide relief from oxidative stress [17, 18]. Additionally, PPARα activation can suppress the sterol regulatory element-binding protein 1 (SREBP1) pathway through the reduction of liver X receptor (LXR)/retinoid X receptor (RXR) formation in the liver, which plays a crucial role in the regulation of fatty acid metabolism . Notably, FOXO3a is a direct transcriptional regulator of a group of oxidative protection genes and this regulation requires PGC-1α. FOXO3A and PGC-1α interact directly and cooperatively and their interaction regulates mitochondrial oxidative stress .
We hypothesised that resveratrol can potentially prevent renal cell apoptosis and oxidative stress, which are the main causes of renal damage in diabetic nephropathy, via activation of AMPK–SIRT1–PGC-1α and the consequent effects on its target molecules PPARα–ERR-1α and the phosphatidylinositol-3 kinase (PI3K)–Akt–FOXO3a pathway.
Male 6-week-old C57BLKS/J db/m and db/db mice were purchased from Jackson Laboratories (Bar Harbor, ME, USA). Four groups of male C57BLKS/J db/m and db/db mice were used in this study. Resveratrol (Sigma-Aldrich, St Louis, MO, USA) dissolved with 0.5% carboxymethyl cellulose sodium salt (CMC), 20 mg kg−1 day−1, was administered via gavage to diabetic mice (db/db Res, n = 8) and non-diabetic mice (db/m Res, n = 8) starting at 8 weeks of age, for 12 weeks . The diabetic db/db group (n = 8) and the non-diabetic db/m control group (n = 8) received only 0.5% CMC. The mice were placed in individual metabolism cages (Nalgene, Rochester, NY, USA) with access to water and food ad libitum. After 12 weeks of treatment, the systolic BP was determined by the non-invasive tail-cuff system in conscious mice (IITC Life Science, Woodland Hill, CA, USA) after a 5 day accommodation period. At week 20, all animals were anaesthetised by intraperitoneal injection of a mixture of Rompun, 10 mg/kg (Bayer Korea, Ansan, Gyeonggi-Do, Korea) and Zoletil, 30 mg/kg (Virbac, Carros, France). The mice were killed and the kidneys removed. Immediately following removal, each kidney was divided into the cortex and medullar. All our experiments, including western blots, were performed with renal cortex. The kidneys were rapidly dissected and stored in buffered formalin (10%) for subsequent immunohistochemical analyses. Blood was collected from the left ventricle and the plasma was stored at –70°C for subsequent analyses. HbA1c was determined from the red cell lysates by HPLC (Bio-Rad, Richmond, CA, USA). The total cholesterol and triacylglycerol concentrations were measured by an autoanalyser (Hitachi 917, Tokyo, Japan) using commercial kits (Wako, Osaka, Japan). NEFA levels were measured with a JCA-BM1250 automatic analyser (JEOL, Tokyo, Japan). The experiments were performed in accordance with our institutional animal care guidelines, and all the procedures complied with the Guide for the Careand Use of LaboratoryAnimals (National Institutes of Health Publication No. 85–23, revised 1996).
Assessment of renal function and the lipid contents
At week 20, the animals were housed in metabolism cages (Nalgene) for 24 h to collect urine for subsequent measurements of the albumin concentrations by an immunoassay (Bayer, Elkhart, IN, USA). Plasma and urine creatinine concentrations were measured, using HPLC (Beckman Instruments, Fullerton, CA, USA). The kidney lipids were extracted using the method of Bligh and Dyer with slight modifications, as previously described .
Light microscopy study
The mesangial matrix and glomerular tuft areas were quantified for each glomerular cross-section, using sections stained with periodic acid–Schiff's reagent. More than 30 glomeruli, cut through the vascular pole, were counted per kidney and the average was used for analysis.
Immunohistochemistry for TGF-β1, type IV collagen, PPARα, cell surface glycoprotein F4/80, 8-hydroxy-deoxyguanosine and TUNEL assay
We performed immunohistochemistry for type IV collagen, TGF-β1, PPARα, cell surface glycoprotein F4/80 (F4/80), 8-hydroxy-deoxyguanosine (8-OH-dG) and active caspase-3. We also performed a TUNEL assay. See the electronic supplementary material (ESM) Methods for further details.
The total protein of the renal cortical tissues was extracted with a Pro-Prep Protein Extraction Solution (Intron Biotechnology, Gyeonggi-Do, Korea) according to the manufacturer's instructions. Western blot analysis was performed to further confirm the responses using epitope-specific antibodies. Blots were carried out for SIRT1, PPARα, PGC-1α, ERR-1α, SREBP1, carnitine palmitoyltransferase-1A (CPT-1A), total AMPK, phosphorylated (phospho)-AMPK Thr172, PI3K, total Akt, phospho-Akt Ser473, total FOXO3a, phospho-FOXO3a Ser253, active caspase-3, B cell leukaemia/lymphoma 2 (BCL-2), superoxide dismutase (SOD)1 and SOD2. We also measured PI3K activity. See ESM Methods for further details.
Immunoprecipitation for acetylated PGC-1α assay
Acetylated-lysine PGC-1α was analysed by immunoprecipitation of PGC-1α from the renal cortex protein lysates with anti-PGC-1α (Santa Cruz Biotechnology, Santa Cruz, CA, USA) followed by western blotting using an acetyl-lysine antibody (Cell Signaling, Denvers, MA, USA).
Assessment of renal oxidative stress
To evaluate the oxidative DNA damage and lipid peroxidation, the 24 h urinary 8-OH-dG (OXIS Health Products, Portland, OR, USA) and 24 h urinary 8-epi-prostaglandin F2α (8-epi-PGF2α; OxisResearch, Foster City, CA, USA) concentrations were measured, respectively.
To find the effects of high glucose and resveratrol on NMS2 mesangial cells  with regard to the activation of the AMPK–SIRT1–PGC-1α pathway and apoptosis, mesangial cells were grown in 5 mmol/l d-glucose (low glucose), 30 mmol/l d-glucose (high glucose) or 5 mmol/l d-glucose plus 25 mmol/l mannitol (as an osmotic control for 30 mmol/l d-glucose). After reaching 80% confluency, the mesangial cells were exposed to low glucose, high glucose or mannitol control, with or without the additional 48 h application of resveratrol (1, 10 or 50 ng/ml), 5-aminoimidazole-4-carboxamide-1β-d-ribofuranoside (AICAR), 0.5 mmol/l, or metformin (1 mmol/l). The small interfering (si)RNA targeted to Ampkα1 (also known as Prkaa1), Ampkα2 (also known as Prkaa1), Sirt1 and scrambled siRNA (siRNA control) were complexed with transfection reagent (G-Fectin; Genolution, Seoul, Korea), according to the manufacturer's instructions. We also performed a TUNEL assay to evaluate the effects of siRNAs on mesangial cell apoptosis. See ESM Methods for further details.
The data are expressed as the means±SD. Differences between the groups were tested for statistical significance by ANOVA with Bonferroni correction, using SPSS version 11.5 (SPSS, Chicago, IL, USA). A p value <0.05 was considered to show a statistically significant difference.
Physical and biochemical characteristics of mice
Biochemical and physical characteristics of study groups
Body weight (g)
30 ± 2
30 ± 2
40 ± 5***
41 ± 4***
Kidney weight (g)
0.18 ± 0.01
0.18 ± 0.03
0.22 ± 0.03*
0.22 ± 0.02*
9.38 ± 2.94
9.99 ± 1.94
26.92 ± 2.72***
25.36 ± 5.05***
4.2 ± 0.2
4.1 ± 0.1
12.2 ± 2.1***
13.4 ± 0.9***
(22.4 ± 0.7)
(21.3 ± 0.6)
(109.8 ± 5.3***)
(123.0 ± 2.7***)
24 h albumin (μg)
8.6 ± 4.2
4.5 ± 1.2
117.8 ± 54.2***
25.7 ± 17.2
Serum Cr (mmol/l)
6.98 ± 0.70
7.16 ± 0.88
7.24 ± 0.97
7.07 ± 0.79
Cr clearance (ml/min)
0.34 ± 0.21
0.30 ± 0.33
0.67 ± 0.29***
0.41 ± 0.21
Systolic BP (mmHg)
98.6 ± 4.5
96.6 ± 5.2
101.1 ± 5.2
99.3 ± 4.5
Food intake (g)
12.0 ± 0.3
11.7 ± 0.7
23.6 ± 4.9***
28.6 ± 1.1***
Effects of resveratrol on renal phenotype, TGF-β1, type IV collagen and F4/80
Renal levels of phospho-AMPK Thr172and total AMPK, SIRT1, total and acetylated-Lys-PGC-1α, PPARα–ERR-1α–PGC-1α signalling, and intra-renal triacylglycerol and NEFA
Renal levels of PI3K, phospho-Akt Ser473 and pFOXO3a
Renal levels of pro-apoptotic BAX, anti-apoptotic BCL-2, cleaved caspase-3 and TUNEL-positive cells
Effects of resveratrol on SOD1, SOD2, and renal and 24 h urinary 8-OH-dG and isoprostane concentrations
In vitro studies
Pharmacologically targeting transcriptional networks to regulate or modulate the gene-expression programmes that favour energy expenditure is an attractive concept to combat metabolic diseases, particularly the complications of diabetes. Both AMPK and SIRT1 have emerged as interesting targets as they are heavily involved in catabolic metabolism, mitochondrial activation, angiogenesis and enhanced cell survival [3–6]. It is well known that the effects of resveratrol are mediated by both SIRT1 and AMPK . However, the effects of resveratrol in the kidneys of an animal model of type 2 diabetes are not well known. The results of the present study demonstrate that diabetic nephropathy in db/db mice is associated with increases in renal lipid accumulation, apoptotic renal cell injury and oxidative stress that relate to the inactivation of AMPK–SIRT1–PGC-1α signalling and the deregulation of their target molecules, PPARα–ERR-1α and PI3K–Akt–FOXO3a. Diabetic nephropathy was ameliorated by resveratrol treatment via the activation of AMPK–SIRT1–PGC-1α signalling and the subsequent activation of PPARα–ERR-1α and FOXO3a, which reversed renal lipid accumulation, apoptotic renal cell injury and oxidative stress.
It is notable that AMPK and SIRT1 regulated each other and share many common target molecules, such as PGC-1α, PPARs, FOXOs, and NF-κB. Moreover, AMPK and SIRT1 have clinical relevance with regard to metabolic syndrome and type 2 diabetes, as their effects on target molecules lead to insulin resistance, mitochondrial dysfunction, oxidative and endoplasmic reticulum stress, and lipotoxicity associated with ectopic lipid accumulation . SIRT1 also responds to changes in the availability of nutrients, such as in conditions of energy-intake restriction or starvation, and energy expenditure, through a forkhead-dependent pathway  and PGC-1α . Thus, it has been suggested that the activation of AMPK and SIRT1 allows for the concurrent deacetylation and phosphorylation of their target molecules and decreases the susceptibility to diabetes-associated disorders. The current study demonstrates that the activation of AMPK and SIRT1 by resveratrol may prevent renal lipid accumulation and cell injury related to the activation of PGC-1α. This in vitro study demonstrated that the high-glucose-induced suppression of AMPK–SIRT1–PGC-1α signalling was activated by AMPK activators, such as resveratrol, AICAR and metformin. Additionally, increased AMPK–SIRT1–PGC-1α production mediated by resveratrol in high-glucose media was suppressed with the introduction of Ampkα1, Ampkα2 and Sirt1 siRNAs. When any of one of AMPK, SIRT1 and PGC-1α was knocked down in mesangial cells, levels of the others were also reduced. These results suggest that AMPK and SIRT1 activate each other and jointly regulate the downstream PGC-1α in the mesangial cells, as in the skeletal muscles [11, 26]. Consistent with our results, several recent studies have demonstrated that AMPK activation by adiponectin  or AICAR  protected renal podocytes and diminished albuminuria in diabetic animals by decreasing NADPH oxidase 4 (NOX4)-dependent oxidative stress and apoptosis. However, Kitada et al demonstrated that resveratrol protects against diabetic nephropathy through the direct scavenging of ROS, normalisation of the Mn-SOD function and glucose–lipid metabolism via AMPK/SIRT1-independent mechanisms . The discrepancy between our results and previous studies regarding AMPK/SIRT1 signalling might reflect differences in the dose of resveratrol administered (0.3% vs 20 mg kg−1 day−1), the part of the renal tissue examined (whole kidney vs renal cortex) and the renal cell type studied (proximal epithelial cell vs mesangial cell).
PGC-1α exerts a wide array of effects and it directly co-activates multiple transcription factors, including nuclear receptors such as the PPARs [30, 31], glucocorticoid receptors , oestrogen receptors and ERRs [32, 33], and non-nuclear receptor transcription factors, such as myocyte enhancer factor-2  and the family of FOXO transcription factors . By simultaneously co-activating these transcription factors, PGC-1α can quickly and coordinately modulate a transcriptional programme that governs the energy metabolism. Although PGC-1α co-activation can change in response to different stimuli or in a tissue-specific manner, in the current study, PGC-1 exerted, in part, various renal effects by enhancing the production of PPARα and its target molecule ERR-1α, in addition to co-activating FOXO3a in relation to the PI3K–Akt axis.
We and others have demonstrated that PPARα activation by PPARα agonist prevents and improves diabetes- or high-fat-diet-induced renal injury by increasing the production of lipolytic enzymes and reducing lipid accumulation, oxidative stress and renal cell apoptosis, while inhibiting the development of albuminuria and glomerular fibrosis [12, 13, 36]. ERR-1α activation by PGC-1α and PPARα induces genes with roles in lipid transport , oxidative phosphorylation [17, 37], fatty acid oxidation [38, 39], the tricarboxylic acid cycle [30, 34, 35], mitochondrial biogenesis [17, 39] and oxidative stress defence . In the current study, resveratrol increased PPARα–ERR-1α production and decreased SREBP1, and this was closely associated with decreased lipid content in the kidney. We also showed that high-glucose treatment of mesangial cells inhibits the expression of genes involved in lipolysis, such as those encoding PPARα and CPT-1A. Additionally, lipogenic genes, such as that encoding SREBP1, were induced by high-glucose treatment. Resveratrol treatment in the presence of high glucose reversed these changes, suggesting that the renoprotective impact of resveratrol may be mediated by shifting the gene-expression profile of the cells to a state that favours lipolysis.
The db/db mouse, which lacks signalling through the leptin receptor, is an excellent model of type 2 diabetes because it develops hyperphagia, obesity, hyperleptinaemia and overt hyperglycaemia . In this study, db/db mice, with or without resveratrol treatment, ate about twice the amount eaten by the db/m mice. Leptin increases glucose uptake and type 1 collagen in db/db mesangial cells through a TGF-β–PI3K-dependent pathway . In addition, transgenic mice with leptin overexpression demonstrated an increase in collagen type IV and fibronectin mRNA in the kidney . Recently, it has been demonstrated that the reduced AMPK and SIRT1 function in db/db mice causes dysregulation of adipogenic, lipogenic and lipo-oxidative genes, including leptin, to favour an obese phenotype . These results suggest that resveratrol treatment might partially improve leptin-related TGF-β–PI3K and lipid dysregulation through the activation of AMPK–SIRT1–PGC-1α in the db/db mouse kidney.
In the current study, we found that resveratrol treatment increased the production of AMPK and SIRT1 and decreased PI3K–Akt signalling in the kidneys of db/db mice and the mesangial cells in high-glucose media. In contrast, Schenk et al demonstrated that restriction of energy intake induced SIRT1 activation in mice, resulting in enhanced skeletal muscle insulin sensitivity, via PI3K modulation, and suggested that insulin-stimulated PI3K signalling is essential for the functional improvement of glucose transport in the skeletal muscle, and that SIRT1 is a key orchestrator of adaptation mediated by restriction of energy intake . The discrepancy in effects in this study and the previous energy-restriction study might be because of differences in plasma insulin concentration and target organ. Restriction of energy intake augments insulin-stimulated PI3K activity in response to a physiological insulin concentration (60 μU/ml) . In contrast, at the supraphysiological concentration of insulin, such as in db/db (more than 200 μU/ml) mice, PI3K would be activated in the kidney .
Regarding FOXO signalling in diabetic nephropathy, a novel mechanism of diabetic nephropathy has been suggested . Diabetic conditions increase TGF-β abundance in the mesangial cells and TGF-β activates p38 mitogen-activated protein kinase (MAPK), extracellular signal-regulated kinases (ERKs), PI3K–Akt and Smads . Among these, PI3K–Akt activation by TGF-β induces FOXO3a phosphorylation that results in decreased BCL-2 interacting mediator of cell death (BIM) and Mn-SOD. In contrast, another study has demonstrated that the PI3K–Akt pathway acts as a survival and anti-apoptotic signal in the renal cells . Overall, the data from the present study indicate that diabetes increases PI3K–Akt phosphorylation and FOXO3a phosphorylation, and this is accompanied by significant changes in the expression of key FOXO3a target genes, as reflected by the decreases in anti-apoptotic BCL-2 and the antioxidants SOD1 and SOD2, and the increase in the expression of the pro-apoptotic gene Bax. Consequently, FOXO3a inactivation resulted in apoptotic renal cell death and oxidative stress. However, it is interesting that 12 weeks of treatment with resveratrol led to a protective increase in FOXO3a in the kidney, where it contributed to protection against apoptotic cell death and oxidative stress.
In conclusion, our results demonstrate that activation of AMPK and SIRT1 was associated with PCG-1α-ameliorated lipotoxicity in the kidney, which could be related to apoptosis and oxidative stress in diabetic nephropathy. Resveratrol, as an activator of AMPK–SIRT1–PGC-1α signalling, affected at least two pathways in the development of diabetic nephropathy. One pathway enhanced the PPARα–ERR-1α–SREBP1-mediated removal of lipids that had accumulated in the kidney. The other pathway decreased apoptotic renal cell death and oxidative stress related to FOXO3a activation. These signalling events may improve glomerular matrix accumulation, inflammation and albuminuria. Our data further substantiate the suggestion that high-glucose-induced mesangial cell damage translates into lower AMPK–SIRT1–PGC-1α activity and exacerbated oxidative stress and apoptotic cell damage in the setting of diabetes. All of the changes in AMPK–SIRT1–PGC-1α signalling were reversed by resveratrol treatment. These results suggest that resveratrol may be a novel therapeutic agent for type 2 diabetic nephropathy.
The authors would like to thank S. W. Kim, Division of Urology, The Catholic University of Korea, and B. S. Choi, Seogang University, for their valuable discussion.
This research was supported by the Basic Science Research Program through the National Research Foundation of Korea funded by the Minister of Education, Science and Technology (to C. W. Park; A111055).
Duality of interest
The authors declare that there is no duality of interest associated with this manuscript.
MYK, JHL, HHY, YAH, KSY, HSP, SC, SHK, SJS, BSC, HWK, YSK, JHL, YSC and CWP designed and performed the studies and analysed data. CWP directed the study, interpreted the data and wrote the paper. All authors critically revised and approved the final version.