Article pubs.acs.org/jpr
Metabolic Diversity of Progressive Kidney Disease in 325 Patients with Type 1 Diabetes (the FinnDiane Study) Ville-Petteri Mak̈ inen,*,†,‡,§,∥ Tuulia Tynkkynen,⊥ Pasi Soininen,†,⊥ Tomi Peltola,# Antti J. Kangas,†,‡ Carol Forsblom,§,∥ Lena M. Thorn,§,∥ Kimmo Kaski,# Reino Laatikainen,⊥ Mika Ala-Korpela,†,‡,⊥ and Per-Henrik Groop§,∥,▽ †
Computational Medicine Research Group, Institute of Clinical Medicine, Faculty of Medicine, University of Oulu and Biocenter Oulu, Finland ‡ Department of Internal Medicine and Biocenter Oulu, Clinical Research Center, University of Oulu, Finland § Folkhälsan Research Center, Folkhälsan Institute of Genetics, Biomedicum Helsinki, Finland ∥ Division of Nephrology, Department of Medicine, Helsinki University Central Hospital, Finland ⊥ NMR Metabonomics Laboratory, Department of Biosciences, University of Eastern Finland, Kuopio, Finland # Department of Biomedical Engineering and Computational Science, School of Science and Technology, Aalto University, Finland ▽ Baker IDI Heart and Diabetes Institute, Melbourne, Australia S Supporting Information *
ABSTRACT: Type 1 diabetic patients with varying severity of kidney disease were investigated to create multimetabolite models of the disease process. Urinary albumin excretion rate was measured for 3358 patients with type 1 diabetes. Prospective records were available for 1051 patients, of whom 163 showed progression of albuminuria (8.3-year follow-up), and 162 were selected as stable controls. At baseline, serum lipids, lipoprotein subclasses, and low-molecular weight metabolites were quantified by NMR spectroscopy (325 samples). The data were analyzed by the self-organizing map. In cross-sectional analyses, patients with no complications had low serum lipids, less inflammation, and better glycemic control, whereas patients with advanced kidney disease had high serum cystatin-C and sphingomyelin. These phenotype extremes shared low unsaturated fatty acids (UFAs) and phospholipids. Prospectively, progressive albuminuria was associated with high UFAs, phospholipids, and IDL and LDL lipids. Progression at longer duration was associated with high HDL lipids, whereas earlier progression was associated with poor glycemic control, increased saturated fatty acids (SFAs), and inflammation. Diabetic kidney disease consists of diverse metabolic phenotypes: UFAs, phospholipids, IDL, and LDL may be important in the subclinical phase, high SFAs and low HDL suggest accelerated progression, and the sphingolipid pathway in advanced kidney injury deserves further research. KEYWORDS: nephropathy, NMR metabonomics, sphingomyelin, phospholipids, fatty acids
■
may be important initiators of end-organ damage.8,9 However, most studies on the temporal patterns of diabetic kidney disease have focused on blood glucose, clinical symptoms or specific candidate molecules. Comprehensive information on lipid metabolism and the metabolic phenotypes that underpin insulin resistance is limited. A contiguous follow-up study of the metabolome from diabetes onset to vascular end-points is difficult to organize due to (i) the relatively low prevalence of type 1 diabetes (10% of Finnish diabetic patients), (ii) gradual disease progression that spans several decades, and (iii) changes in treatment policies,
INTRODUCTION Chronic diabetic complications have a significant impact on the quality of life and longevity in type 1 diabetes. Almost all patients are affected by visible changes in the retinal vasculature, and a third of individuals develop clinical signs of kidney disease. The latter, in particular, is associated with a dramatically shortened life-expectancy.1−5 The first indications of kidney injury appear within two decades of diabetes duration. In the affected individuals, urinary albumin excretion begins to increase and blood pressure to rise. Simultaneously, the risk of heart attack and stroke due to atherosclerosis increases substantially.5−7 Those kidney disease patients who avoid or survive the macrovascular diseases will ultimately progress to kidney failure. Insulin resistance and lipotoxicity © 2011 American Chemical Society
Received: October 17, 2011 Published: December 28, 2011 1782
dx.doi.org/10.1021/pr201036j | J. Proteome Res. 2012, 11, 1782−1790
Journal of Proteome Research
Article
nearest-neighbor algorithm to match the baseline kidney status, diabetes duration and gender of the progressors. Patients with a short follow-up ( 75 nm), very large VLDL (average diameter of 64.0 nm), large VLDL (53.6 nm), medium VLDL (44.5 nm), small VLDL (36.8 nm), very small VLDL (31.3 nm), IDL (28.6 nm), large LDL (25.5 nm), medium LDL (23.0 nm), small LDL (18.7 nm), very large HDL (14.3 nm), large HDL (12.1 nm), medium HDL (10.9 nm), and small HDL (8.7 nm). Calibration was performed by linear regression modeling of the spectral line-shapes. Low-molecularweight metabolites and lipid species were quantified by linefitting. Further technical details are available in Supporting Information 1: Page 1, and the NMR data are available in Supporting Information 2. Statistical Analyses
Four types of statistical models were used: (i) hypothesis testing to compare cases and controls, (ii) pruned correlation network analysis, (iii) the self-organizing map (SOM), and (iv) linear regression modeling of prospective albuminuria status. To eliminate confounding effects from gender, the data were first split into men and women, then all continuous variables were converted into ranks within the genders respectively, and finally the two groups were pooled back together for the network and SOM analyses. However, a single rank transformation of the full data set was used for regression modeling, where gender was considered an explanatory factor. The network of continuous variables was based on pairwise Spearman’s correlation coefficients. Specifically, each variable is considered a node and the nodes are connected by links, the weights of which are quantified by the correlation coefficient. The full networks are too dense aznd have to be pruned in order to highlight the relevant patterns.20 We chose a two-stage strategy: first, cliques of variables with r > 0.9 between every pair of members were merged into a single score by estimating the first principal component. This was done to prevent highly intercorrelated variables from obscuring the network structure. The refined network was then pruned by spanning trees for the final visualization (Supporting Information 1: Page 5). This graph theoretic approach was preferred over statistical significance cutoffs due to the inherent preservation of network connectivity. All derived variables such as BMI or the metabolic syndrome were excluded to avoid artificial links. The total number of correlation tests was 2,016; for this number of independent tests the Bonferroni correction yields a single test threshold of P < 2.5 × 10−5 at 5% type 1 error rate. The self-organizing map is an unsupervised data mining method, and we used it here for the characterization of multivariate biochemical profiles.15 The resulting map layout of patients becomes such that those who have similar profiles are as close to each other as possible, whereas those who have
■
MATERIALS AND METHODS At baseline, type 1 diabetic patients were recruited by the Finnish Diabetic Nephropathy Study Group, and 325 patients were included in this study. The initial data collection was cross-sectional (serum and urine samples), but with longitudinal records of albuminuria and clinical history. Type 1 diabetes mellitus was defined as an age of onset below 35 years and transition to insulin treatment within a year of onset. The study protocol was approved by the local ethics committee. The classification of renal status was made according to urinary albumin excretion rate (AER) in at least two out of three consecutive overnight or 24 h urine samples (medical records from local hospitals). Macroalbuminuria or overt kidney disease was defined as AER ≥ 200 μg/min or ≥300 mg/24 h. The intermediary range was defined as microalbuminuria (20 ≤ AER < 200 μg/min or 30 ≤ AER < 300 mg/24 h). Patients on renal replacement therapy at baseline were excluded. Progression of kidney disease was defined as a transition from normal AER to micro- or macroalbuminuria, a transition from microalbuminuria to macroalbuminuria, or a transition from macroalbuminuria to end-stage renal disease. In addition to the clinical classification of albuminuria, a continuous value of AER measured from a single 24 h urine collection at baseline was available. Other variables, including references for the biochemical assays, have been described previously.12 The study design is a matched case-control setting of kidney disease progression (but not for baseline kidney disease itself). At the time of the analyses, prospective data were available for 1051 patients, of which 163 patients who showed progression of kidney disease were selected retrospectively for the NMR measurements (baseline serum was analyzed). In addition, stable patients (162 individuals) were selected by a least-squares 1783
dx.doi.org/10.1021/pr201036j | J. Proteome Res. 2012, 11, 1782−1790
Journal of Proteome Research
Article
Table 1. Patient Characteristicsa normal AER at baseline matching criterion
stable
men | women age (years) type 1 diabetes duration (years) follow-up time (years)
58 | 34 33.6 15.7 8.1
microalb. at baseline
progressed 55 | 33 33.5 14.3 7.5** normal AER at baseline
stable 20 | 6 31.4 20.0 8.4
progressed 26 | 6 34.3 25.2 7.5† microalb. at baseline
macroalb. at baseline stable
progressed
31 | 13 27 | 16 39.8 41.6 27.7 28.8 9.5 8.7† macroalb. at baseline
variable
stable
progressed
stable
progressed
stable
progressed
estimated GFR (mL/min per 1.73 m2) serum creatinine (μmol/L) urinary albumin (mg/24 h) omega-9 and saturated FA (cu) exercise (METh/week) hemoglobin A1c (%) body mass index (kg/m2) sphingomyelin (cu) choline phospholipids (cu) C-reactive protein (mg/L)
97 86 9 7.5 22 8.0 24.5 0.58 2.10 1.54
99 85 16* 8.7* 12† 9.3* 25.5† 0.63† 2.33† 2.14†
98 87 42 8.0 43 8.8 25.8 0.63 2.18 1.50
101 88 139† 10.7† 6† 10.3† 25.1 0.66† 2.39† 2.77†
83 99 352 9.3 8 8.6 25.8 0.70 2.36 2.27
49** 163** 1259† 10.5 2 8.9 24.0† 0.80† 2.43 1.83
NMR measures are in absolute concentration units (cu) that are based on the TSP reference signal. Symbols: †P < 0.05, *P < 0.00018, **P < 10−8 for progressed vs. stable. Bonferroni-corrected multiple test significance is at P < 0.00018. a
Information 1: Page 10). The study was designed for prospective analyses: age and diabetes duration were used for matching the case-control pairs, and they were not different between the stable and progressed patients within the baseline kidney disease groups. Estimated glomerular filtration rate (P < 10−10) indicates reduced kidney function already at baseline for those patients that progressed from macroalbuminuria to end-stage renal disease and, as expected, AER is significant in every comparison (P < 0.0026). Hemoglobin A1c indicates progression from normal AER to micro- or macroalbuminuria (P < 10−6). Chronically high blood glucose (reflected by increased glycated hemoglobin) is considered a prerequisite for diabetic kidney disease,4,21−23 and the findings in Table 1 support that. Progression is also related to saturated lipids, inflammation (C-reactive protein), and body mass. It is, however, unclear if the higher A1c is caused by poor treatment, or by insulin resistance that makes it harder to maintain low blood glucose even with regular insulin injections.
different profiles are placed far apart on the map (Supporting Information 1: Pages 2 and 6). The map is divided into regions, and each region can be colored according to the average value that is observed for the local patients placed in the region. Confidence intervals (i.e., model variance) for the regional averages were estimated by bootstrap analysis. The SOM training set includes the lipoprotein subclass lipids, low-molecular weight metabolites, lipid species, clinical biomarkers, age and diabetes duration, blood pressure, anthropometric measures, smoking and alcohol use, kidney and retinopathy status, and insulin dose. Derived variables such as estimated glomerular filtration rate, and the metabolic syndrome were not included. Prospective data on mortality or vascular complications were not included. The influence of each training variable on the final layout was estimated by permutation analysis. A map coloring can be expressed as a numerical set of regional averages. The statistic δRAW is the variance of these regional values. For each random permutation, the statistic is recalculated; this yields the null distribution of δRAW. The null distribution is close to standard Gaussian after a power transform, and can be used to estimate the P-value for nontraining data. The reported δ is the transformed and standardized δRAW. Detailed description of the procedure can be found on Page 7 in Supporting Information 1 and as Supporting Information in a previous work.15 For supervised modeling, we used the projections to latent traits (PLS components) to reduce the training data into a limited set of scores. We then used logistic regression for binary outcomes and a linear discriminant classifier for multinomial traits. Cross-validation with repeated folds was performed over all the steps (including the dimension reduction) to assess predictive performance (Supporting Information 1: Page 8). The training set was the same as in the SOM analysis with two exceptions: gender was included as an explanatory factor, and clinical kidney status was excluded since it was the target of prediction.
Correlation Network
The comparisons between cases and controls suggest that several variables are affected simultaneously. To understand the underlying physiological links, we investigated the correlation structure of the data set (Figure 1). We successively calculated two maximal spanning trees to prune the network (see Materials and Methods for details) and thus the ratio of edges to vertices was set to 2 in the final figure. We also examined the network with different settings and checked the full list of correlations to verify that the relevant structures were represented (Supporting Information 3). Overall, many of the clinical variables are connected with each other and to the biochemical measures that are the bases of diagnostic classifications (Figure 1A), but there are no strong connections to lipids or other metabolites, except via urinary albumin. In particular, AER is directly connected to triglycerides (r = 0.40, P < 10−17), and to cholesterol-rich lipoproteins via sphingomyelin (r = 0.49, P < 10−17). Hemoglobin A1c is positioned between glucose (r = 0.30, P < 10−8) and triglycerides (r = 0.27, P < 10−5), as expected.
■
RESULTS AND DISCUSSION Table 1 presents selected results from comparative analysis of cases and controls (the full list is available in Supporting 1784
dx.doi.org/10.1021/pr201036j | J. Proteome Res. 2012, 11, 1782−1790
Journal of Proteome Research
Article
Figure 1. Correlation network of continuous variables. The data were corrected for gender and the link weights were quantified by the Spearman correlation coefficient. The metabolite groups A−D are added as visual cues, they were not used in building the network. Only the top 6% topologically signficant links are depicted. Topological significance was determined by constructing successive spanning trees of the full network20 and other levels of pruning were tested to ensure that the figure contained the most important correlation structures. Red color indicates positive correlation, blue negative correlation, and the thin gray lines indicate links that were not statistically significant, but included to ensure connectivity.
protein synthesis,25,30 although the precise impact on glucose homeostasis is unclear. Patients with type 1 diabetes experience severe excursions of daily blood glucose,31 and they cannot be required to fast before sample collection (they were instructed to have a light breakfast). Hence, Figure 1E is probably produced by a combination of acute and chronic physiological responses. Food ingestion can also influence Figure 1C; however, we did not see significant concentrations of chylomicrons.
Lipoprotein lipids comprise the right side of the network: the cholesterol-rich IDL and LDL particles are the primary contributors to total cholesterol concentration (Figure 1B), whereas the triglyceride-rich VLDLs form a tight group at the center (Figure 1C). HDL particles are inversely connected to VLDLs (Figure 1D). These connections match well with previous knowledge.24 Interestingly, very large HDL lipids, phospholipids, and polyunsaturated fatty acids are interconnected (Figure 1C, right-most part). Branched-chain amino acids have been associated with insulin resistance and the development of type 2 diabetes.25,26 Our results are compatible: alanine, valine, and isoleucine are intercorrelated and close to glucose and hemoglobin A1c in the network (Figure 1E). On the other hand, alanine is inversely correlated with 3-hydroxybutyrate and acetoacetate, but not with glucose. Serum lactate and alanine are both involved in gluconeogenesis, and the ketone bodies are typically detected in insulin deprivation.27,28 The positive correlations with glucose are thus likely products of (acute) hyperglycemia. Similarly, the inverse link between serum sodium and glucose may be explained by acute osmolar stress from severe hyperglycemia.29 Leucine and glutamine are linked to HDL lipids and leucine is inversely correlated to very large VLDL lipids. However, leucine is not correlated with urinary albumin (P > 0.05), so the links cannot be attributed to kidney disease. Of note, a diet rich in leucine may increase insulin sensitivity and promote muscle
Self-Organizing Map of Baseline Data
Correlation analysis is able to detect linear associations between variables, but it does not provide information on the metabolic features of individual patients. To complement the network approach, a self-organizing map (SOM) was constructed. We used permutation analysis to determine the influence of each training variable on the map structure, as indicated by the δ-statistic and color intensity (higher value and intensity means higher influence, δ ≥ 1.7 is considered the threshold for nonnegligible influence). Prolonged exposure to diabetes causes the vascular injuries, hence diabetes duration (Figure 2A) and age (Supporting Information 1: Page 18) produce identical patterns on the SOM, and both are strong determinants of the map structure (δ ≥ 5.4). The study was designed to cover maximal range of kidney disease and, accordingly, AER and other kidney markers also produce intense patterns (δ ≥ 4.7) on the map (Figure 2B−D). 1785
dx.doi.org/10.1021/pr201036j | J. Proteome Res. 2012, 11, 1782−1790
Journal of Proteome Research
Article
Figure 2. Selected colorings of the self-organizing map (SOM). The SOM is a two-dimensional organization of the patients, where the position on the canvas is dependent on the measured metabolic profile (Supporting Information 1). Each of the plots shows a coloring of the same map; this is analogous to the colorings of the world map based on people’s education, income, and age, for instance. The interpretation is also similar: one can view the SOM areas as if they were the countries of the world, except in this case they represent metabolically different subgroups of patients instead of national populations. The SOM was constructed from data and serum samples collected in the beginning of the study and the data were corrected for gender differences before analysis. The numbers and colors on the circles indicate regional values in the original measurement unit.
poor glycemic control are typically accompanied by overproduction of triglyceride-rich VLDL particles and depletion of cholesterol-rich HDL particles,33,34 and the same phenomenon is also visible in the type 1 diabetic patients (Figure 2E,I,L). Patients with the lowest hemoglobin A1c have decreased IDL and LDL lipids (Figure 2E,J,K, top and bottom areas). Sphingomyelin is increased in patients with albuminuria (Figure 2B,M). Choline phospholipids, phosphatidylcholine, and phosphoglycerides share the bimodal pattern of IDL and LDL lipids (Figure 2J,K,N and Supporting Information 1: Page 18). Surprisingly, omega-3 and other unsaturated fatty acids are also increased both on the left and right sides (Figure 2O and Supporting Information 1: Page 18). Triglycerides and omega-9 and saturated fatty acids are increased on the right side of the map (Figure 2P and Supporting Information 1: Page 18).
Crucially, there is only partial overlap with aging and kidney disease, so the development of microvascular complications is likely to be modulated by additional factors. Persistent high blood glucose is the classical risk factor for complications.21 Hemoglobin A1c is an important but imperfect measure of persistent hyperglycemia 32, which can explain the clear but weaker than expected pattern for glycemic control in our study (δ = 2.7, Figure 2E). Interestingly, both glucose and lactate behave the same way (Figure 2F and Supporting Information 1: Page 18). These patients have average or increased insulin doses (Figure 2G), which precludes insulin deprivation as the main cause for the observed hyperglycemia. Total lipids in large VLDL, IDL, medium LDL particles and cholesterol in HDL particles reveal the main features of the lipoprotein continuum (Figure 2I-L). Insulin resistance and 1786
dx.doi.org/10.1021/pr201036j | J. Proteome Res. 2012, 11, 1782−1790
Journal of Proteome Research
Article
Progression from normal AER to microalbuminuria involves minor structural changes in the nephrons and the kidney function remains within normal range.35 However, the transition from macroalbuminuria to end-stage renal disease is defined by the loss of glomerular filtration and leads to the initiation of hemodialysis. Simultaneously, nutritional balance is affected and overall metabolic profiles may change considerably.36,37 Figure 3C shows the progression rates from macroalbuminuria to end-stage renal disease. Phenotype IV that already has direct indications of reduced kidney function at baseline is the most likely to progress (78%). As before, the progression rates are the same for phenotypes II and III (33% vs 34%) in spite of the difference in diabetes duration. Phenotype I was not observed in patients with baseline macroalbuminuria.
The SOM training set comprised numerous lipid variables, which may lead to overemphasis of lipids. To counter this, we merged the most correlated lipids before training; the strong patterns observed for aging, kidney disease, and other clinical features indicate that the SOM provides a balanced view of the metabolic profiles. Baseline Phenotypes and Disease Progression
We used linear projections to latent traits (PLS) to reduce the dimensionality of the training data and then used logisitic regression to classify the samples. The baseline categories of macroalbuminuria and normal AER are clearly separated (96% sensitivity and 93% specificity), but within the albuminuria groups the stable and progressed individuals overlap heavily (Table 2). For those with normal AER at baseline, 70% of stable and 61% of progressed patients were correctly classified by the logistic PLS model at optimal cutoff point. In the macroalbuminuria group, 86% of stable and 67% of progressed patients were correctly classified. Results from linear discriminant analyses are listed on Page 19 in Supporting Information 1. Briefly, the three baseline kidney disease categories could be classified with 74% overall accuracy (33% for the null model), and a combined baseline and prospective kidney disease model with six categories achieved 45% overall accuracy (16% for the null model). The SOM enables the modeling of the categorical data as continuous progression rates, while simultaneously preserving the easy interpretation of the model phenotypes. Importantly, prospective data was not used for model training to prevent overfitting. We chose four established metabolic indicators as the basis for phenotypic risk assessment: Phenotype I was chosen by estimated glucose disposal rate and is characterized by favorable clinical picture and low overall lipids (Figure 2H); phenotype II has the highest triglycerides and shows signs of metabolic stress from insulin resistance and saturated fatty acids (Figure 2P); phenotype III has the highest HDL cholesterol (Figure 2L); phenotype IV has advanced kidney disease with the associated secondary effects (Figure 2D). Figure 3A illustrates the progression from normal AER for the four model phenotypes. Phenotypes I and II are close in terms of diabetes duration (15 vs 18 years), but phenotype II has a higher progression rate (31% vs 71%). Put differently, the complication risk for phenotype II is more than double that of Phenotype I at the same length of diabetes exposure. Interestingly, phenotype II and III have comparable progression rates (71% vs 66%) despite a larger difference in diabetes exposure (18 vs 29 years). Similar pattern of progression rates can also be seen for patients with baseline microalbuminuria (Figure 3B)
Summary and Hypotheses
Improvement in glycemic control produces long-lasting beneficial effects on a type 1 diabetic patient’s health.21 However, glycated hemoglobin, the established marker of persistent hyperglycemia, provides only limited information on the disease processes and it alone cannot explain the development and progression of diabetic complications. For this reason, we investigated a large number of metabolic traits at different clinical stages to better undestand the course of diabetic kidney disease. Although we did not have multiple time points available, the combination of baseline and prospective clinical data lets us hypothesize the probable metabolic course of diabetic kidney disease. Younger patients can be characterized by two phenotype models: Phenotype I represents a favorable state with adequate glycemic control and low overall lipids, whereas phenotype II shows signs of insulin resistance and lipotoxicity from excess saturated fatty acids. Crucially, the patients with phenotype II have double the risk to develop kidney disease within the next eight years (Figure 3). Insulin deprivation cannot explain the differences (Figure 2G), but insulin resistance,38 aggravated by the modern sedentary life style,39,40 could be the underlying factor. Phenotype III represents older patients with type 1 diabetes who do not have a lipotoxic and insulin resistant metabolic profile, but nevertheless show an increase in hemoglobin A1c, cholesterol-rich lipoproteins, phospholipids and unsaturated fatty acids. These features are shared with phenotype II, and the progression rates are comparable between the two models. Polyunsaturated fatty acids are considered beneficial,41,42 so our results are surprising. On the other hand, increased LDL cholesterol can be a marker of nutritional excess or high-energy diet,43 which would also explain the mild hyperglycemia and elevated progression risk in phenotype III.
Table 2. Classification Results for Binary Traits from 7-Fold Cross-Validation with 1000 Repeatsa binary trait
controls
cases
PLS components
AUC
AUC, null model
cutoff
sensitivity
specificity
baseline macroalbuminuria baseline retinopathy progression from normal AER progression from microalb. progression from macroalb. deceased
180 190 92 26 44 303
87 134 88 32 43 22
7 5 4 3 3 3
0.98 0.85 0.65 0.72 0.84 0.76
0.50 0.49 0.49 0.48 0.49 0.50
0.19 0.25 0.47 0.38 0.56 0.06
96% 86% 61% 60% 67% 70%
93% 73% 70% 65% 86% 74%
a
The optimal cutoff points for the logistic models were determined by maximizing the sum of sensitivity and specificity for each model, and then averaging over the cross-validation repeats. Similar averaging was done also for the areas under the receiver operator characteristic curves (AUC). The clinical trait was randomly permuted for each repeat of the null modeling (Supporting Information 1). The control group for baseline macroalbuminuria had normal AER. 1787
dx.doi.org/10.1021/pr201036j | J. Proteome Res. 2012, 11, 1782−1790
Journal of Proteome Research
Article
Figure 3. Model phenotypes from the SOM analysis and their associations with kidney disease progression. The phenotypes were defined based on baseline data (Figure 2D,H,L,P). To prevent overfitting, prospective data were not used for the training or selection of the models. The plots depict the average value and 95% interval from bootstrap analysis. The red and blue arrows depict the concentration differences with respect to the average value in the full data set.
■
Serum sphingomyelin is correlated with albuminuria at baseline,44 and the same trend was observed also for progression (Table 1). Epidemiological information on sphingolipids in human type 1 diabetes is limited; genetic influences have been observed in the general population45 and an animal model of type 1 diabetes exhibits abnormal sphingolipid metabolism.46 Interestingly, a study in high-fat fed mice showed that manipulating the sphingolipid pathway changed kidney injury outcome,47 and another report implicated sphingolipids in nondiabetic human kidney disease.48 Figure 1 shows a related phenomenon: sphingomyelin has a pivotal connecting role between albuminuria and cholesterol transport. Furthermore, the SOM colorings show how sphingomyelin peaks between phenotypes II and IV (Figure 2D,M,P), which supports the theory that excess availability of substrate (saturated fatty acid) leads to increased ceramide accumulation via the sphingolipid pathway,47,49 and subsequent tissue injury. Hence, the sphingolipid pathway could be a potential drug target also for diabetic complications.
ASSOCIATED CONTENT
S Supporting Information *
Additional results and experimental details. This material is available free of charge via the Internet at http://pubs.acs.org.
■
AUTHOR INFORMATION
Corresponding Author
*Telephone: +358 40 5696 117. Fax: +358 9 191 25452. E-mail:
[email protected]. Author Contributions
V.-P.M., M.A.-K. and P.-H.G. conceived the study and wrote the manuscript. V.-P.M. also participated in the data collection and analyzed the data. T.T., P.S., T.P., K.K., A.J.K., R.L., and M.A.-K. produced and/or preprocessed the metabonomics data. C.F., L.M.T. and P.-H.G. collected the clinical material. All authors reviewed and discussed the manuscript. No external editorial assistance was used in this manuscript.
■
■
CONCLUSIONS Diabetic kidney disease is the combined product of long-term exposure to hyperglycemia, genetic predisposition, and the ratemodulating effects of treatment and lifestyle. Our statistical models suggest that phospholipids, IDL and LDL lipids, and unsaturated fatty acids are increased in the early phase of the disease, but not in the late phase. Simultaneously, the patients exhibit extensive metabolic diversity with respect to saturated fatty acids, inflammation, and HDL metabolism, and this diversity may be one explanation to the variability in kidney disease onset. In the final phase, the sphingolipid pathway could be a mediator of lipotoxicity from saturated fatty acids on the path to end-stage renal disease and/or premature death.
ACKNOWLEDGMENTS The authors are grateful for the contribution of health care professionals who collected information on the type 1 diabetic patients in the local hospitals (Supporting Information The study was supported by grants from the Folkhälsan Research Foundation, the Wilhelm and Else Stockmann Foundation, the Liv och Hälsa Foundation and the CEED3 partnership within the Seventh Framework Programme of the European Union (Project 223211). The work was also supported by the OrionFarmos Research Foundation (V.-P.M.), the National Graduate School of Organic Chemistry and Chemical Biology in University of Eastern Finland (T.T.), the Jenny and Antti Wihuri Foundation (A.J.K.), the Responding to Public Health 1788
dx.doi.org/10.1021/pr201036j | J. Proteome Res. 2012, 11, 1782−1790
Journal of Proteome Research
Article
(17) Tukiainen, T.; Tynkkynen, T.; Mäkinen, V.-P.; Jylänki, P.; Kangas, A.; Hokkanen, J.; Vehtari, A.; Gröhn, O.; Hallikainen, M.; Soininen, H.; Kivipelto, M.; Groop, P.-H.; Kaski, K.; Laatikainen, R.; Soininen, P.; Pirttilä, T.; Ala-Korpela, M. A multi-metabolite analysis of serum by 1H NMR spectroscopy: early systemic signs of Alzheimer’s disease. Biochem. Biophys. Res. Commun. 2008, 375, 356−361. (18) Vehtari, A.; Mäkinen, V.-P.; Soininen, P.; Ingman, P.; Mäkelä, S. M.; Savolainen, M. J.; Hannuksela, M. L.; Kaski, K.; Ala-Korpela, M. A novel bayesian approach to quantify clinical variables and to determine their spectroscopic counterparts in 1H NMR metabonomic data. BMC Bioinf. 2007, 8 (Suppl 2), S8. (19) Okazaki, M.; Usui, S.; Ishigami, M.; Sakai, N.; Nakamura, T.; Matsuzawa, Y.; Yamashita, S. Identification of unique lipoprotein subclasses for visceral obesity by component analysis of cholesterol profile in high-performance liquid chromatography. Arterioscler., Thromb., Vasc. Biol. 2005, 25, 578−584. (20) Mäkinen, V.-P.; Forsblom, C.; Thorn, L. M.; Wadén, J.; Kaski, K.; Ala-Korpela, M.; Groop, P.-H. Network of vascular diseases, death and biochemical characteristics in a set of 4,197 patients with type 1 diabetes (the Finndiane Study). Cardiovasc. Diabetol. 2009, 8, 54. (21) DCCT Writing Team Sustained effect of intensive treatment of type 1 diabetes mellitus on development and progression of diabetic nephropathy. J. Am. Med. Assoc., 2003, 290, 2159−2167. (22) Svensson, M.; Eriksson, J. W.; Dahlquist, G. Early glycemic control, age at onset, and development of microvascular complications in childhood-onset type 1 diabetes: a population-based study in Northern Sweden. Diabetes Care 2004, 27, 955−962. (23) Raile, K.; Galler, A.; Hofer, S.; Herbst, A.; Dunstheimer, D.; Busch, P.; Holl, R. W. Diabetic nephropathy in 27,805 children, adolescents, and adults with type 1 diabetes: effect of diabetes duration, A1c, hypertension, dyslipidemia, diabetes onset, and sex. Diabetes Care 2007, 30, 2523−2528. (24) Vergès, B. Lipid disorders in type 1 diabetes. Diabetes Metab. 2009, 35, 353−360. (25) Timmerman, K. L.; Volpi, E. Amino acid metabolism and regulatory effects in aging. Curr. Opin. Clin. Nutr. Metab. Care 2008, 11, 45−49. (26) Wang, T. J.; Larson, M. G.; Vasan, R. S.; Cheng, S.; Rhee, E. P.; McCabe, E.; Lewis, G. D.; Fox, C. S.; Jacques, P. F.; Fernandez, C.; O’Donnell, C. J.; Carr, S. A.; Mootha, V. K.; Florez, J. C.; Souza, A.; Melander, O.; Clish, C. B.; Gerszten, R. E. Metabolite profiles and the risk of developing diabetes. Nat. Med. 2011, 17, 448−453. (27) Shulman, G. I.; Lacy, W. W.; Liljenquist, J. E.; Keller, U.; Williams, P. E.; Cherrington, A. D. Effect of glucose, independent of changes in insulin and glucagon secretion, on alanine metabolism in the conscious dog. J. Clin. Invest. 1980, 65, 496−505. (28) Lanza, I. R.; Zhang, S.; Ward, L. E.; Karakelides, H.; Raftery, D.; Nair, K. S. Quantitative metabolomics by H-NMR and LC-MS/MS confirms altered metabolic pathways in diabetes. PLoS One 2010, 5, e10538. (29) Roscoe, J. M.; Halperin, M. L.; Rolleston, F. S.; Goldstein, M. B. Hyperglycemia-induced hyponatremia: metabolic considerations in calculation of serum sodium depression. Can. Med. Assoc. J. 1975, 112, 452−453. (30) Layman, D. K.; Walker, D. A. Potential importance of leucine in treatment of obesity and the metabolic syndrome. J. Nutr. 2006, 136, 319S−23S. (31) Service, F. J.; Molnar, G. D.; Rosevear, J. W.; Ackerman, E.; Gatewood, L. C.; Taylor, W. F. Mean amplitude of glycemic excursions, a measure of diabetic instability. Diabetes 1970, 19, 644− 655. (32) Sacks, D. B. A1c versus glucose testing: a comparison. Diabetes Care 2011, 34, 518−523. (33) Adiels, M.; Olofsson, S.; Taskinen, M.; Borén, J. Overproduction of very low-density lipoproteins is the hallmark of the dyslipidemia in the metabolic syndrome. Arterioscler., Thromb., Vasc. Biol. 2008, 28, 1225−1236. (34) Musunuru, K. Atherogenic dyslipidemia: cardiovascular risk and dietary intervention. Lipids 2010, 45, 907−914.
Challenges Research Programme of the Academy of Finland (M.A.-K.) and the Finnish Foundation for Cardiovascular Research (M.A.-K.).
■
REFERENCES
(1) Allen, K. V.; Walker, J. D. Microalbuminuria and mortality in long-duration type 1 diabetes. Diabetes Care 2003, 26, 2389−2391. (2) Klein, B. E. K.; Klein, R.; McBride, P. E.; Cruickshanks, K. J.; Palta, M.; Knudtson, M. D.; Moss, S. E.; Reinke, J. O. Cardiovascular disease, mortality, and retinal microvascular characteristics in type 1 diabetes: Wisconsin Epidemiologic Study of Diabetic Retinopathy. Arch. Intern. Med. 2004, 164, 1917−1924. (3) Gross, J. L.; de Azevedo, M. J.; Silveiro, S. P.; Canani, L. H.; Caramori, M. L.; Zelmanovitz, T. Diabetic nephropathy: diagnosis, prevention, and treatment. Diabetes Care 2005, 28, 164−176. (4) Stadler, M.; Auinger, M.; Anderwald, C.; Kästenbauer, T.; Kramar, R.; Feinböck, C.; Irsigler, K.; Kronenberg, F.; Prager, R. Longterm mortality and incidence of renal dialysis and transplantation in type 1 diabetes mellitus. J. Clin. Endocrinol. Metab. 2006, 91, 3814− 3820. (5) Secrest, A. M.; Becker, D. J.; Kelsey, S. F.; Laporte, R. E.; Orchard, T. J. Cause-specific mortality trends in a large populationbased cohort with long-standing childhood-onset type 1 diabetes. Diabetes 2010, 59, 3216−3222. (6) Torffvit, O.; Lövestam-Adrian, M.; Agardh, E.; Agardh, C. Nephropathy, but not retinopathy, is associated with the development of heart disease in type 1 diabetes: a 12-year observation study of 462 patients. Diabetic Med. 2005, 22, 723−729. (7) Kim, W. Y.; Astrup, A. S.; Stuber, M.; Tarnow, L.; Falk, E.; Botnar, R. M.; Simonsen, C.; Pietraszek, L.; Hansen, P. R.; Manning, W. J.; Andersen, N. T.; Parving, H. Subclinical coronary and aortic atherosclerosis detected by magnetic resonance imaging in type 1 diabetes with and without diabetic nephropathy. Circulation 2007, 115, 228−235. (8) Pang, T. T. L.; Narendran, P. Addressing insulin resistance in type 1 diabetes. Diabetic Med. 2008, 25, 1015−1024. (9) Schauer, I. E.; Snell-Bergeon, J. K.; Bergman, B. C.; Maahs, D. M.; Kretowski, A.; Eckel, R. H.; Rewers, M. Insulin resistance, defective insulin-mediated fatty acid suppression, and coronary artery calcification in subjects with and without type 1 diabetes: the CACTI Study. Diabetes 2011, 60, 306−314. (10) Newman, D. J.; Mattock, M. B.; Dawnay, A. B. S.; Kerry, S.; McGuire, A.; Yaqoob, M.; Hitman, G. A.; Hawke, C. Systematic review on urine albumin testing for early detection of diabetic complications. Health Technol. Assess. 2005, 9, iii-vi−xiii-163. (11) Caramori, M. L.; Fioretto, P.; Mauer, M. Enhancing the predictive value of urinary albumin for diabetic nephropathy. J. Am. Soc. Nephrol. 2006, 17, 339−352. (12) Mäkinen, V.-P.; Forsblom, C.; Thorn, L. M.; Wadén, J.; Gordin, D.; Heikkilä, O.; Hietala, K.; Kyllönen, L.; Kytö, J.; RosengårdBärlund, M.; Saraheimo, M.; Tolonen, N.; Parkkonen, M.; Kaski, K.; Ala-Korpela, M.; Groop, P.-H. Metabolic phenotypes, vascular complications, and premature deaths in a population of 4,197 patients with type 1 diabetes. Diabetes 2008, 57, 2480−2487. (13) Ala-Korpela, M. Critical evaluation of 1H NMR metabonomics of serum as a methodology for disease risk assessment and diagnostics. Clin. Chem. Lab. Med. 2008, 46, 27−42. (14) Mierisová, S.; Ala-Korpela, M. MR spectroscopy quantitation: a review of frequency domain methods. NMR Biomed. 2001, 14, 247− 259. (15) Mäkinen, V.-P.; Soininen, P.; Forsblom, C.; Parkkonen, M.; Ingman, P.; Kaski, K.; Groop, P.-H-; Ala-Korpela, M. 1H NMR metabonomics approach to the disease continuum of diabetic complications and premature death. Mol. Syst. Biol. 2008, 4, 167. (16) Soininen, P.; Haarala, J.; Vepsäläinen, J.; Niemitz, M.; Laatikainen, R. Strategies for organic impurity quantification by 1H NMR spectroscopy: constrained total-line-shape fitting. Anal. Chim. Acta 2005, 542, 178−185. 1789
dx.doi.org/10.1021/pr201036j | J. Proteome Res. 2012, 11, 1782−1790
Journal of Proteome Research
Article
(35) Fioretto, P.; Caramori, M. L.; Mauer, M. The kidney in diabetes: dynamic pathways of injury and repair. the Camillo Golgi Lecture 2007. Diabetologia 2008, 51, 1347−1355. (36) Kalantar-Zadeh, K.; Cano, N. J.; Budde, K.; Chazot, C.; Kovesdy, C. P.; Mak, R. H.; Mehrotra, R.; Raj, D. S.; Sehgal, A. R.; Stenvinkel, P.; Ikizler, T. A. Diets and enteral supplements for improving outcomes in chronic kidney disease. Nat. Rev. Nephrol. 2011, 7, 369−384. (37) Sato, E.; Kohno, M.; Yamamoto, M.; Fujisawa, T.; Fujiwara, K.; Tanaka, N. Metabolomic analysis of human plasma from haemodialysis patients. Eur. J. Clin. Invest. 2011, 41, 241−255. (38) Kilpatrick, E. S.; Rigby, A. S.; Atkin, S. L. Insulin resistance, the metabolic syndrome, and complication risk in type 1 diabetes: ″double diabetes″ in the Diabetes Control and Complications Trial. Diabetes Care 2007, 30, 707−712. (39) Mattsson, N.; Rönnemaa, T.; Juonala, M.; Viikari, J. S. A.; Raitakari, O. T. The prevalence of the metabolic syndrome in young adults. the Cardiovascular Risk in Young Finns Study. J. Intern. Med. 2007, 261, 159−169. (40) Conway, B.; Miller, R. G.; Costacou, T.; Fried, L.; Kelsey, S.; Evans, R. W.; Orchard, T. J. Temporal patterns in overweight and obesity in type 1 diabetes. Diabetic Med. 2010, 27, 398−404. (41) Cárdenas, C.; Bordiu, E.; Bagazgoitia, J.; Calle-Pascual, A. L. Polyunsaturated fatty acid consumption may play a role in the onset and regression of microalbuminuria in well-controlled type 1 and type 2 diabetic people: a 7-year, prospective, population-based, observational multicenter study. Diabetes Care 2004, 27, 1454−1457. (42) Shapiro, H.; Theilla, M.; Attal-Singer, J.; Singer, P. Effects of polyunsaturated fatty acid consumption in diabetic nephropathy. Nat. Rev. Nephrol. 2011, 7, 110−121. (43) O’Keefe, J. H. J.; Cordain, L.; Harris, W. H.; Moe, R. M.; Vogel, R. Optimal low-density lipoprotein is 50 to 70 mg/dl: lower is better and physiologically normal. J. Am. Coll. Cardiol. 2004, 43, 2142−2146. (44) Mäkinen, V.; Tynkkynen, T.; Soininen, P.; Forsblom, C.; Peltola, T.; Kangas, A.; Groop, P.-H.; Ala-Korpela, M. Sphingomyelin is associated with kidney disease in type 1 diabetes (the Finndiane Study). Metabolomics 2011, DOI: 10.1007/s11306-011-0343-y. (45) Hicks, A. A.; Pramstaller, P. P.; Johansson, A.; Vitart, V.; Rudan, I.; Ugocsai, P.; Aulchenko, Y.; Franklin, C. S.; Liebisch, G.; Erdmann, J.; Jonasson, I.; Zorkoltseva, I. V.; Pattaro, C.; Hayward, C.; Isaacs, A.; Hengstenberg, C.; Campbell, S.; Gnewuch, C.; Janssens, A. C. W.; Kirichenko, A. V.; König, I. R.; Marroni, F.; Polasek, O.; Demirkan, A.; Kolcic, I.; Schwienbacher, C.; Igl, W.; Biloglav, Z.; Witteman, J. C. M.; Pichler, I.; Zaboli, G.; Axenovich, T. I.; Peters, A.; Schreiber, S.; Wichmann, H.; Schunkert, H.; Hastie, N.; Oostra, B. A.; Wild, S. H.; Meitinger, T.; Gyllensten, U.; van Duijn, C. M.; Wilson, J. F.; Wright, A.; Schmitz, G.; Campbell, H. Genetic determinants of circulating sphingolipid concentrations in European populations. PLoS Genet. 2009, 5, e1000672. (46) Fox, T. E.; Bewley, M. C.; Unrath, K. A.; Pedersen, M. M.; Anderson, R. E.; Jung, D. Y.; Jefferson, L. S.; Kim, J. K.; Bronson, S. K.; Flanagan, J. M.; Kester, M. Circulating sphingolipid biomarkers in models of type 1 diabetes. J. Lipid Res. 2011, 52, 509−517. (47) Boini, K. M.; Zhang, C.; Xia, M.; Poklis, J. L.; Li, P. Role of sphingolipid mediator ceramide in obesity and renal injury in mice fed a high-fat diet. J. Pharmacol. Exp. Ther. 2010, 334, 839−846. (48) Fornoni, A.; Sageshima, J.; Wei, C.; Merscher-Gomez, S.; Aguillon-Prada, R.; Jauregui, A. N.; Li, J.; Mattiazzi, A.; Ciancio, G.; Chen, L.; Zilleruelo, G.; Abitbol, C.; Chandar, J.; Seeherunvong, W.; Ricordi, C.; Ikehata, M.; Rastaldi, M. P.; Reiser, J.; Burke, G. W. Rituximab targets podocytes in recurrent focal segmental glomerulosclerosis. Sci. Transl. Med. 2011, 3, 85ra46. (49) Yang, G.; Badeanlou, L.; Bielawski, J.; Roberts, A. J.; Hannun, Y. A.; Samad, F. Central role of ceramide biosynthesis in body weight regulation, energy metabolism, and the metabolic syndrome. Am. J. Physiol. Endocrinol. Metab. 2009, 297, E211−24.
1790
dx.doi.org/10.1021/pr201036j | J. Proteome Res. 2012, 11, 1782−1790