Insulin resistance, dyslipidaemia, inflammation and endothelial function in nephrotic syndrome

Transcription

1 Nephrol Dial Transplant (2002) 17: Brief Report Insulin resistance, dyslipidaemia, inflammation and endothelial function in nephrotic syndrome Gursharan K. Dogra, Susan Herrmann, Ashley B. Irish 1, Mark A. B. Thomas 1 and Gerald F. Watts Department of Medicine and Western Australian Heart Research Institute, University of Western Australia and 1 Department of Nephrology, Royal Perth Hospital, Perth, Western Australia, Australia Abstract Background. Nephrotic syndrome (NS) is associated with an increased risk of cardiovascular disease (CVD). We have shown previously that endothelial function, measured by post-ischaemic flow-mediated dilatation (FMD) of the brachial artery, is impaired in NS. In this study our aim was to assess the potential roles of insulin resistance, plasma non-esterified fatty acids (NEFAs) and inflammation in endothelial dysfunction in NS patients. Methods. FMD was compared between NS patients (ns19) and controls (CS, ns19). Plasma glucose, insulin and NEFAs were measured. Insulin resistance was calculated using the Homeostasis Model Assessment (HOMA) score. C-reactive protein (CRP), interleukin-6 (IL-6), tumour necrosis factor a (TNFa) and fibrinogen were measured as markers of inflammation. Results. FMD was significantly lower in the NS group (mean"standard error, NS 5.1"0.7%, CS 7.3"0.7%, Ps0.02). Fasting insulin (NS 12.5"1.5 muul, CS 6.8"0.7 muul, P-0.01), fasting glucose (NS 5.3"0.2, CS 4.8"0.1, Ps0.02) and the HOMA score (NS 3.0"0.4, CS 1.5"0.2, Ps0.001) were significantly higher in NS. These differences persisted after adjusting for waist circumference. Of the inflammatory markers, only fibrinogen (P-0.01) and IL-6 (Ps0.01) were significantly increased in NS. Despite significantly lower plasma NEFAs in NS, the NEFA:albumin ratio showed a non-significant trend to higher levels in NS (NS 10.7"0.1 mmolug, CS 8.7"0.1 mmolug, Ps0.06). Within the NS group, multivariate backward regression analysis showed that NEFAs (P-0.01) and low-density lipoprotein (LDL) cholesterol (Ps0.05) were significant negative independent predictors of FMD. Conclusion. Endothelial function in NS is inversely correlated with plasma concentrations of NEFAs and Correspondence and offprint requests to: Dr G. K. Dogra, University Department of Medicine, Royal Perth Hospital, Box X2213 GPO Perth, Western Australia 6847, Australia. sdogra@cyllene.uwa.edu.au LDL cholesterol. Dyslipoproteinaemia and NEFAs probably contribute to the increased risk of CVD seen in NS. We also postulate that in NS, hypoalbuminaemia increases the delivery of NEFAs to endothelial cells thereby impairing the synthesis and release of nitric oxide. Keywords: arterial function; fatty acids; hyperlipidaemia; inflammation; insulin sensitivity; nephrotic syndrome Introduction The increased risk of coronary morbidity and mortality associated with nephrotic syndrome (NS) may be a consequence of dyslipoproteinaemia, hypoalbuminaemia and hyperoxidative stress, hypercoaguable state anduor inflammation [1,2]. Endothelial dysfunction is an early phase of atherogenesis associated with impaired nitric oxide (NO) bioavailability and can be detected non-invasively using high-resolution ultrasonography to measure post-ischaemic flow-mediated dilatation (FMD) of conduit arteries [3]. Conduit artery endothelial dysfunction is associated with conventional cardiac risk factors such as hypercholesterolaemia and smoking [3], and has been shown to predict coronary events [4]. We hypothesized previously that the increased risk of cardiovascular disease (CVD) in NS is primarily a consequence of dyslipidaemia, and have shown that patients with NS exhibit endothelial dysfunction to the same degree as patients with primary hyperlipidaemia [2]. Although this suggests an aetiological role for dyslipidaemia, serum lipids and lipoproteins were not significant predictors of endothelial dysfunction within the nephrotic patient group [2]. Furthermore, although studies have shown that L-arginine improves NO-mediated vasodilatation in patients with familial hypercholesterolaemia, it has no effect on NO-mediated vasodilatation in NS suggesting that alternative # 2002 European Renal Association European Dialysis and Transplant Association

2 Risk factors for endothelial dysfunction in nephrotic syndrome mechanisms for endothelial dysfunction may exist [5]. Additionally, we did not find that elevated blood pressure and homocysteine were significantly associated with endothelial dysfunction in NS [2,6]. NS is a pro-inflammatory state associated with elevated levels of tumour necrosis factor a (TNFa) [7], and may also be associated with insulin resistance and elevated free fatty acids [8]. Inflammation is associated with increased risk for CVD, both independently and as part of an insulin-resistant state [9]. Insulin resistance is also associated with an increased risk of CVD [10]. Potential mechanisms include increased lipolysis resulting in elevated levels of non-esterified fatty acids (NEFAs) [11], which have been shown to be associated with impaired endothelial function [12,13]. In light of these findings, and the conflicting evidence for the role of dyslipidaemia in the increased risk of CVD in NS, the role of inflammation, insulin resistance and NEFAs warrants further examination in NS subjects. Thus, in an extension to our previous findings, we undertook a study in NS and control (CS) subjects comparing lipids and lipoproteins, indices of insulin resistance, plasma NEFAs and markers of inflammation, and the association of these with endothelial dysfunction. Subjects and methods The study was a cross-sectional comparison of NS and CS subjects matched for age, sex and body mass index (BMI). Nineteen patients with nephrotic-range proteinuria (3.5 guday) and a primary glomerulopathy and 19 healthy, normolipidaemic, non-proteinuric control subjects were recruited from renal clinics and from the community, respectively. Normolipidaemia was defined as total cholesterol -5.5 mmolul, triglycerides -1.8 mmolul and HDL cholesterol )0.9 mmolul in men and 1.1 molul in women. Subjects on immunosuppressive therapy or with a serum creatinine )150 mmolul, secondary cause for proteinuria, CVD, diabetes mellitus, hypothyroidism, liver disease, alcoholism, postural hypotension, intercurrent illnesses and significant psychiatric disorders were excluded. NS patients were studied off lipid lowering drugs, non-steroidal antiinflammatory drugs and aspirin, and were not excluded for taking angiotensin converting enzyme inhibitors (ACE-I) or a stable dose of other antihypertensive drugs. The study received approval from the Royal Perth Hospital Ethics Committee and all volunteers gave informed, written consent. The research was carried out in accordance with the Declaration of Helsinki (1989) of the World Medical Association. Laboratory methods Venous blood samples were obtained after a 12-h fast for the following variables, which were measured by standard laboratory methods unless otherwise stated: serum creatinine and albumin, urinary creatinine and protein, serum glucose, serum insulin, fibrinogen, total cholesterol, high-density lipoprotein (HDL) cholesterol, direct LDL cholesterol and triglycerides. C-reactive protein (CRP) was assayed using a high-sensitivity immunonephelometric method (Dade Behring Marburg GmbH, Marburg, Germany). TNFa was measured using an immunometric assay (Immulite 1 TNFa; Diagnostic Products Corporation, Los Angeles, CA). Interleukin-6 (IL-6) was measured using a high-sensitivity quantitative enzyme immunoassay (Quantikine 1 HS; R&D Systems Inc., Minneapolis, MN). Plasma NEFAs were measured using an enzymatic colorimetric assay (Boehringer Mannheim GmBH, Mannheim, Germany). Insulin sensitivity was calculated using the Homeostasis Model Assessment score [HOMA scoresfasting insulin (muul) 3serum glucose (mmolul)u22.5] [14], where a higher score signifies increasing insulin resistance. The inter-assay coefficient of variation (CV) of all analytical assays was -6%, except IL-6 (CV-6.5%) and TNFa (CV-10%). Brachial artery ultrasonography Endothelium-dependent post-ischaemic FMD and endotheliumindependent glyceryl trinitrate-mediated dilatation (GTNMD) were measured using brachial artery ultrasonography (Acuson Aspen 128 ultrasound device, Acuson Corporation, Mountainview, CA) as described elsewhere, and analysed using edge-detection software validated within our department [2]. Maximal FMD and GTNMD responses were calculated as the percentage change in brachial artery diameter from baseline. The intra-observer CV of the computerized technique is in the order of 6.7%. The CV for repeated within-subject measurement was 14.7% (ns24) with a mean (SD) difference in FMD of 1.6 (1.0)%. Statistical methods and analysis Sample size calculations indicated that 15 subjects per group would be sufficient to show a minimum 20% difference in the HOMA score and a 100% difference in inflammatory markers (CRP, TNFa, IL-6 and fibrinogen) with 80% power and a of Predicted differences in inflammatory markers were a conservative estimate from previous studies showing a nearly 400% increase in TNFa in NS [7]. Results are expressed as mean"se. An independent t-test was used for between-group comparisons (Statistical Package for Social Sciences version 10, SPSS v.10, Chicago, IL). Categorical variables were compared using x 2 and Fisher s exact tests. Correlations are described using Pearson s r correlation coefficient or univariate regression methods. Multiple backward regression methods, excluding variables that correlated with P)0.15, were used to examine determinants of FMD in NS patients. Skewed data are described as geometric mean (95% confidence interval) and log transformed prior to parametric analyses. Results 2221 The NS and CS groups had similar age, male to female ratio, BMI, waist circumference and serum creatinine (Table 1). As expected, the NS group had statistically higher urinary protein to creatinine ratio, lipids and lipoproteins, and a significantly lower serum albumin (all P-0.001). Systolic blood pressure (Ps0.01) and diastolic blood pressure (P-0.01) were also significantly higher in the NS group. Serum creatinine

3 2222 G. K. Dogra et al. Table 1. Clinical and biochemical characteristics of the NS and CS subjects NS (ns19) CS (ns19) P-value Age (years) 45.2 (3.6) 47.4 (3.2) 0.65 Sex (MuF) 8u11 9u BMI (kgum 2 ) 27.0 (1.3) 24.8 (0.9) 0.15 Waist circumference (cm) 87.3 (2.6) 82.4 (2.7) 0.21 SBP (mmhg) 123 (3) 112 (3) 0.01 DBP (mmhg) 75 (2) 67 (2) MAP (mmhg) 90 (2) 82 (2) Serum creatinine (mmolul) 91.1 (7.8) 83.3 (2.6) 0.35 egfr (mlumin) (10.9) 89.8 (4.5) 0.37 Serum albumin (gul) 26.8 (1.6) 43.3 (0.7) U p :U c ratio (mgummol creatinine) a ( ) 2.7 ( ) Serum cholesterol (mmolul) a 9.8 ( ) 4.7 ( ) LDL cholesterol (mmolul) a 6.4 ( ) 2.8 ( ) HDL cholesterol (mmolul) a 1.5 ( ) 1.5 ( ) 0.62 Triglyceride (mmolul) a 2.7 ( ) 0.8 ( ) Results expressed as mean (SE), except a expressed as geometric mean (95% confidence interval). SBP, systolic blood pressure; DBP, diastolic blood pressure; MAP, mean arterial pressure; egfr, estimated glomerular filtration rate using the modified Cockcroft and Gault formula; U p :U c, urinary protein to creatinine ratio. and estimated glomerular filtration rate (modified Cockcroft and Gault formula) were not significantly different between groups. There were two smokers in the NS group and one smoker in the CS group. Median disease duration in the NS patients was 6 months (range months). Of the 19 NS patients, renal histology showed minimal change disease (in five patients), membranous glomerulonephritis (six patients), focal and segmental glomerulonephritis (three patients), IgA glomerulonephritis (one patient), IgA-negative mesangioproliferative glomerulonephritis (one patient), mesangiocapillary glomerulonephritis type 1 (one patient), and two patients had not had a renal biopsy. Eleven of the 19 NS patients were on an ACE-I, two were on angiotensin II receptor antagonists and eight patients received diuretics. Twelve of the 19 NS patients had symptomatic disease with pitting peripheral oedema at the time of this study. Brachial artery vascular function, inflammatory markers and indices of insulin resistance are shown in Table 2. In view of the mean 2.2-kgum 2 difference in Table 2. Vascular function, markers of inflammation and insulin resistance in NS and CS subjects BMI between NS and CS groups, all statistical comparisons are adjusted for difference in BMI. NS patients had significantly lower FMD compared with CS (P-0.01). GTN-mediated vasodilatation was not different between groups. Fibrinogen (P-0.01) and IL-6 (Ps0.01), but not CRP and TNFa, were significantly higher in NS patients. NS patients had significantly higher fasting serum glucose (Ps0.02), and fasting serum insulin (P-0.01). As expected, waist circumference was significantly correlated with the HOMA score in both groups (rs0.58, P-0.001). The HOMA score was significantly higher in NS patients (Ps0.001), even after adjusting for waist circumference (P-0.01). Plasma NEFA levels were significantly lower in NS patients (Ps0.03), and were significantly correlated with serum albumin in combined group analysis (rs0.41, Ps0.02). However, the NEFA:albumin ratio showed a trend to be higher in the NS group (Ps0.06). Within the NS group, only plasma NEFAs were significantly negatively correlated with FMD (univariate NS CS P-value a Resting brachial artery diameter (mm) 3.4 (0.1) 3.3 (0.1) 0.86 Brachial FMD (%) 5.1 (0.7) 7.3 (0.7) 0.02 Brachial GTNMD (%) 20.4 (1.4) 22.4 (2.0) 0.66 IL-6 (pguml) b 3.5 ( ) 1.8 ( ) 0.01 CRP (mgul) b 1.5 ( ) 1.2 ( ) 0.96 Fibrinogen (gul) 4.2 (0.4) 2.6 (0.2) TNFa (pguml) b 4.1 ( ) 3.2 ( ) 0.24 Serum glucose (mmolul) 5.3 (0.2) 4.8 (0.1) 0.02 Serum insulin (muul) 12.5 (1.5) 6.8 (0.7) HOMA score 3.0 (0.4) 1.5 (0.2) NEFAs (mmolul) 0.28 (0.02) 0.37 (0.04) 0.03 NEFA:albumin (mmolug) 10.7 (0.1) 8.7 (0.1) 0.06 Results expressed as mean (SE), except b expressed as geometric mean (95% confidence interval). a P-value adjusted for differences in body mass index.

4 Risk factors for endothelial dysfunction in nephrotic syndrome regression coefficient B"SEs 18.2"7.2, Ps0.02). The NEFA:albumin ratio was also significantly negatively correlated with FMD in the NS group (Bs 4.5"1.5, Ps0.01) (Figure 1). In NS patients, inflammatory markers and the HOMA score were not significantly correlated with FMD; HOMA score was not correlated with serum triglycerides, inflammatory markers, plasma NEFA or severity of nephrosis; cytokine levels were not correlated with duration or severity of nephrosis. In multivariate backward regression analysis, plasma NEFA (b 0.83, Ps0.002) and LDL cholesterol (b 0.44, Ps0.05) were significant independent predictors of impaired FMD in NS patients only, model adjusted R 2 s0.42, Ps0.01 (Table 3). The NEFA:albumin ratio was not entered in the multiple regression model due to confounding inter-correlation with LDL cholesterol as would be expected due to the significant expected correlation between serum albumin and LDL cholesterol in NS (rs 0.64, P-0.01). Serum albumin, serum creatinine, cytokines and CRP, HOMA and blood pressure were not significantly correlated with FMD in NS subjects. Fig. 1. Scatter plot showing association between NEFA:albumin ratio and FMD in nephrotic patients. Table 3. Multiple backward regression model showing association between plasma NEFA, LDL cholesterol and brachial FMD in nephrotic patients Variable Regression coefficient (SE) Standardized b coefficient P-value NEFA (mmolul) 29.2 (7.6) LDL cholesterol 3.5 (1.6) (mmolul) IL 6 (pguml) 1.4 (0.8) Dependent variable: brachial FMD, model adjusted R 2 s0.42, Ps0.01. Discussion 2223 This study suggests that plasma NEFAs are negatively associated with endothelial function in NS independent of insulin resistance. We have also extended our previous finding of impaired endothelial function, as measured by brachial FMD, to a larger number of NS patients and have shown that LDL cholesterol is an independent negative predictor of endothelial function, supporting the hypothesis that dyslipoproteinaemia contributes to the increased cardiovascular risk seen in NS [1,2]. Additional findings of this study were that the NS patients were insulin resistant independent of visceral adiposity, and had elevated levels of IL-6. Plasma NEFA levels were significantly negatively correlated with endothelial function measured by brachial artery FMD. As NEFAs are predominantly bound to albumin, circulating levels were significantly lower in the NS group, but the NEFA:albumin ratio, a marker of cardiovascular risk [15], showed a trend to be higher in NS patients suggesting an increase in nonalbumin-bound NEFA. Elevated levels of circulating NEFAs have been shown to impair endotheliumdependent vasodilatation in resistance arteries of noninsulin resistant subjects [12,13]. Our study suggests that this effect of plasma NEFAs also extends to the conduit arteries of NS patients. Mechanisms involved may include direct endothelial toxicity [13] and decreased endothelial NO synthase (enos) activity [16]. We postulate that nephrotic hypoalbuminaemia allows increased substrate delivery of NEFAs to the endothelium where they induce endothelial dysfunction by the aforementioned mechanisms. This hypothesis could be further tested with studies employing serum albumin replacement as a method of reducing bioavailability of non-protein bound NEFAs to the endothelium, and examining FMD responses. Unlike our previous report [2], LDL cholesterol was found to be a significant independent predictor of endothelial dysfunction. It is likely that the larger sample size, by increasing the power of this study, has allowed us to better define the association between LDL cholesterol and endothelial dysfunction. Native LDL cholesterol can induce endothelial dysfunction via pro-oxidant effects [17] or by associated increases in asymmetric dimethylarginine (ADMA), an endogenous inhibitor of enos [18]. Other mechanisms include impaired expression of enos and inhibition of NO release by oxidized LDL cholesterol [19]. In addition, nephrotic hypoalbuminaemia is associated with increased lipoprotein-bound NEFAs, which facilitate activation of cholesterol ester transfer protein, thereby increasing the atherogenicity of LDL cholesterol [20] and providing a mechanism by which both NEFAs and LDL cholesterol may impair endothelial function. Increased lysophosphatidylcholine modification of LDL cholesterol and increased oxidant stress in hypoalbuminaemic states may also contribute to LDL mediated endothelial dysfunction [5]. Thus, there appears to be a complex interaction between hypoalbuminaemia, NEFAs and LDL cholesterol, all contributing to

5 2224 G. K. Dogra et al. Fig. 2. Hypothesis: multi-factorial risk for CVD in NS. endothelial toxicity and impaired endothelial function (Figure 2). In our study, NS patients were insulin resistant, in spite of preserved glomerular filtration rates and independent of visceral adiposity, as measured by waist circumference. While one previous study has shown normal fasting insulin and glucose levels in NS subjects with impaired renal function, metabolic clearance of glucose was impaired supporting insulin resistance to carbohydrate metabolism [8]. NS patients had a mean 2-fold higher HOMA score compared with CS subjects, which is similar to HOMA scores reported in diabetic subjects [14]. Although insulin resistance is characteristically associated with elevation of NEFAs [11], plasma NEFAs and the HOMA score were not correlated in our study. Other possible explanations for the association between insulin resistance and nephrosis include insulin receptor inhibition by increased levels of inflammatory markers [9]. While we showed significantly elevated IL-6 levels in NS, IL-6 was not significantly correlated with the HOMA score. The HOMA score is a valuable tool for population screening for insulin resistance but has recognized imprecision when used in smaller numbers of subjects, particularly if they are non-diabetic [14]. Hence, the aetiology and role of insulin resistance in NS warrants further investigation with studies employing a hyperinsulinaemic euglycaemic clamp technique both during nephrosis and after disease remission. Our cross-sectional study design is limited in defining causality. The regression model described explains only 42% of the variation in brachial artery FMD and as such we cannot fully exclude a role for insulin resistance, triglycerides and inflammatory cytokines in atherogenesis in NS. Other risk factors for endothelial dysfunction such as ADMA and oxidant stress need to be considered, as do the differences in age and sex, as well as heterogeneity in the genes regulating the release of biogenic molecules such as NO by endothelial cells. NS is associated with complex pathophysiological changes, and the increased risk for CVD is likely to be multi-factorial, and not solely related to dyslipoproteinaemia (Figure 2). In conclusion, we have identified plasma NEFAs as a potential cardiovascular risk factor in addition to LDL cholesterol in patients with NS. Interventions with specific LDL cholesterol lowering agents and lowering of non-albumin-bound NEFAs with serum albumin repletion would test the significance of LDL cholesterol and plasma NEFAs as risk factors for endothelial dysfunction and CVD in NS more rigorously. In the clinical setting, treatment with statins, due to their range of pleiotropic effects, may broadly target the pro-atherogenic changes associated with NS and lead to an improvement in endothelial function. Acknowledgements. We acknowledge the financial support of the National Health and Medical Research Council, Australian Kidney Foundation and Medical Research Foundation of Royal Perth Hospital. We are grateful to all the renal physicians who referred patients for the study. We appreciate technical assistance with ultrasonography provided by Mrs Lisa Rich and Dr David Playford. References 1. Ordonez JD, Hiatt RA, Killebrew EJ, Fireman BH. The increased risk of coronary heart disease associated with nephrotic syndrome. Kidney Int 1993; 44: Watts GF, Herrmann S, Dogra GK et al. Vascular function of the peripheral circulation in patients with nephrosis. Kidney Int 2001; 60:

6 Risk factors for endothelial dysfunction in nephrotic syndrome 3. Celermajer DS, Sorensen KE, Gooch VM et al. Non-invasive detection of endothelial dysfunction in children and adults at risk of atherosclerosis. Lancet 1992; 340: Neunteufl T, Heher S, Katzenschlager R et al. Late prognostic value of flow-mediated dilation in the brachial artery of patients with chest pain. Am J Cardiol 2000; 86: Joles JA, Stroes ES, Rabelink TJ. Endothelial function in proteinuric renal disease. Kidney Int 1999; 56 (Suppl 71): S57 S61 6. Dogra G, Irish AB, Watts GF. Homocysteine and nephrotic syndrome. Nephrol Dial Transplant 2001; 16: Suranyi MG, Guasch A, Hall BM, Myers BD. Elevated levels of tumor necrosis factor-alpha in the nephrotic syndrome in humans. Am J Kidney Dis 1993; 21: Stenvinkel P, Ottosson-Seeberger A, Alvestrand A. Renal hemodynamics and sodium handling in moderate renal insufficiency: the role of insulin resistance and dyslipidemia. J Am Soc Nephrol 1995; 5: Yudkin JS, Kumari M, Humphries SE, Mohamed-Ali V. Inflammation, obesity, stress and coronary heart disease: is interleukin-6 the link? Atherosclerosis 2000; 148: Howard G, O Leary DH, Zaccaro D et al. Insulin sensitivity and atherosclerosis. The Insulin Resistance Atherosclerosis Study (IRAS) Investigators. Circulation 1996; 93: Playford DA, Watts GF. Endothelial dysfunction, insulin resistance and diabetes: exploring the web of causality. Aust NZ J Med 1999; 29: Steinberg HO, Tarshoby M, Monestel R et al. Elevated circulating free fatty acid levels impair endothelium-dependent vasodilation. J Clin Invest 1997; 100: Steinberg HO, Paradisi G, Hook G et al. Free fatty acid elevation impairs insulin-mediated vasodilation and nitric oxide production. Diabetes 2000; 49: Bonora E, Targher G, Alberiche M et al. Homeostasis model assessment closely mirrors the glucose clamp technique in the assessment of insulin sensitivity. Diabetes Care 2000; 23: Pickart L. Increased ratio of plasma free fatty acids to albumin during normal aging and in patients with coronary heart disease. Atherosclerosis 1983; 46: Davda RK, Stepniakowski KT, Lu G et al. Oleic acid inhibits endothelial nitric oxide synthase by a protein kinase C-independent mechanism. Hypertension 1995; 26: Pritchard KA Jr, Groszek L, Smalley DM et al. Native lowdensity lipoprotein increases endothelial cell nitric oxide synthase generation of superoxide anion. Circ Res 1995; 77: Boger RH, Bode-Boger SM, Szuba A et al. Asymmetric dimethylarginine (ADMA): a novel risk factor for endothelial dysfunction: its role in hypercholesterolemia. Circulation 1998; 98: Liao JK, Shin WS, Lee WY, Clark SL. Oxidized low-density lipoprotein decreases the expression of endothelial nitric oxide synthase. J Biol Chem 1995; 270: Braschi S, Masson D, Rostoker G et al. Role of lipoproteinbound NEFAs in enhancing the specific activity of plasma CETP in the nephrotic syndrome. Arterioscler Thromb Vasc Biol 1997; 17: Received for publication: Accepted in revised form: