1H NMR metabonomics approach to the disease continuum of diabetic complications and premature death
Ville-Petteri Mäkinen1,2,3, Pasi Soininen4, Carol Forsblom2,3, Maija Parkkonen2,3, Petri Ingman5, Kimmo Kaski1, Per-Henrik Groop2,3 & Mika Ala-Korpela1 for the FinnDiane Study Group
- Computational Medicine Research Group, Laboratory of Computational Engineering, Systems Biology and Bioinformation Technology, Helsinki University of Technology, Finland
- FinnDiane Study Group, Folkhälsan Research Center, Folkhälsan Insititute of Genetics, Biomedicum Helsinki, University of Helsinki, Finland
- Division of Nephrology, Department of Medicine, Helsinki University Hospital, Finland
- Laboratory of Chemistry, Department of Biosciences, University of Kuopio, Finland
- Instrument Centre, Department of Chemistry, University of Turku, Finland
Correspondence to: Per-Henrik Groop2,3 FinnDiane Study Group, Folkhälsan Research Center, Folkhälsan Insititute of Genetics, Biomedicum Helsinki, University of Helsinki, PO Box 63, Helsinki FI-00014, Finland. Tel.: +358919125459; Fax: +358919125452; Email: per-henrik.groop@helsinki.fi
Correspondence to: Computational Medicine Research Group, Laboratory of Computational Engineering, Systems Biology and Bioinformation Technology, Helsinki University of Technology, PO Box 9203, Helsinki FI-02015 HUT, Finland. Tel.: +358503535457; Fax: +35894514833; Email: mika.ala-korpela@hut.fi
Received 8 October 2007; Accepted 5 December 2007; Published online 12 February 2008
Article highlights
- The combined application of 1H NMR metabonomics of serum and self-organizing maps resulted in a new holistic framework to visualize and interpret data and to link metabolic phenotypes to underlying diagnostics, biochemical variables and premature death
- Based solely on 1H NMR data of serum, up to a 7.8-fold relative risk of premature death was observed for type 1 diabetic patients with an adverse metabolic profile.
- The metabolic phenotype with the highest mortality combined biochemical features from the metabolic syndrome (high triglycerides, low HDL2 cholesterol), insulin resistance (high lactate) and kidney disease (high creatinine, low albumin).
- The diffuse nature of micro- and macrovascular diseases was illustrated by subtle multimetabolite differences between the metabolic syndrome and diabetic kidney disease; a demonstration of the enhanced detection power of the metabonomics approach beyond single biomarkers and univariate statistics.
Synopsis
People with diabetes are at high risk of dying from heart disease and stroke, and many patients also suffer from severe degradation of the kidneys, retina and nervous system. Diabetes-related diseases reflect the imbalance of glucose metabolism: patients with type I diabetes completely lack the normal insulin response that makes glucose available for cellular processes. With insulin replacement therapy the acute symptoms can be cured, but the natural metabolic balance is nevertheless disturbed, which leads to chronic systemic stress.
Diabetic kidney disease (DKD) is an important predictor of premature death in type I diabetes. Its absence does not, however, preclude other risk factors for heart disease. Furthermore, the clinical diagnosis is based on a single biomarker (excess protein in urine) and subject to large individual variation that makes the early stages difficult to detect. We are therefore developing new cost-effective analytical and computational approaches that can augment the existing biomarkers and provide a quantitative multidimensional disease characterization.
In this study, we measured the 1H NMR spectra of blood serum for 613 patients with type I diabetes from the Finnish Diabetic Nephropathy Study. We chose 1H NMR spectroscopy, as it can detect many of the important risk markers (such as cholesterol, triglycerides, glucose and creatinine) with a single standardized experimental procedure (Figure 1). Our starting point was exploratory—we did not try to predict urine protein excretion, but rather to identify the diverse and diffuse systemic metabolic states of the diabetic condition, as seen in serum.
Figure 1
1H NMR spectral profile of diabetic kidney disease. (A) The SOM of 613 
2 1H NMR spectra of serum, colored according to the percentage estimate of DKD within a given map region. Each hexagonal map unit defines a specific model spectrum and a corresponding subset of patients, the spectra of which best match the aforementioned model. (B) The low molecular weight metabolites (LMWM) model spectrum and (C) the lipoprotein lipid and albumin (LIPO) model spectrum for a patient subset within the map unit with the lowest percentage of DKD. The colored curve segments indicate the current model, whereas the solid black curve indicates the mean spectrum over all data, thus serving as a constant reference. The colored areas below the model spectra represent the proportional differences of the unit-specific model and the mean model. (D) The LIPO model and (E) the LMWM model spectrum for patients within a map unit of the highest DKD percentage. An interactive presentation of the model spectra is available in Supplementary data 3.
The complex molecular data cannot be used as such; we thus visualized the spectral features with a self-organizing map (SOM). Simply speaking, the SOM is just a layout of patients on a 2D canvas in such a way that patients with similar spectra are placed close to each other. Consequently, the map can be colored according to locally averaged values for a particular variable, which reveals the differences in the metabolic profiles between specific map regions. We also developed a new method to estimate the statistical significance of the observed patterns and to normalize the colorings, so that different sources of information can be easily visualized and reliably compared (Figure 1).
Our results show that in the study group (aged between 30 and 50 years) mortality during the next decade was over three times higher than in the same age group of the entire Finnish population (Figure 6). Most of the premature deaths were attributed to the combination of DKD and adverse serum profile (eightfold relative risk). Note that none of the patients was on dialysis, so they still had adequate kidney function. The spectral features for these patients revealed hallmarks of insulin resistance that are characteristic of additional disturbance in glucose metabolism besides the insulin deprivation. High concentration of triglycerides, elevated total cholesterol and a decrease in high-density lipoprotein particles (HDL2) were observable in the 1H NMR spectra, along with an increase of creatinine, which is associated with reduced filtering capacity of the kidneys. Lactate and acetate were also different between the high- and low-risk groups, which further indicates alterations in cellular glucose metabolism (Figure 1).
Figure 6
Summary of clinical and metabolic characteristics. (A–F) Statistics for a selection of non-NMR variables for patient groups defined by six districts on the SOM. The map was constructed based on the 1H NMR spectra for 613 type I diabetic patients. The percentages of cases with respect to the total number of patients in a given district and for the whole population (ALL) are listed for 10-year mortality (normalized by follow-up time), DKD, the MetS, MVD, DRD and male gender. Relative risk of death (RR) was defined as the ratio of the observed mortality in type I diabetic patients against the entire Finnish population. The MetS was defined as present if the score was three or more. Median values are listed for the continuous variables, with the full statistics available for the non-NMR data in Table 3 in Supplementary data 2.
Full figure and legend (253K)Figures & Tables indexIn addition to the 1H NMR spectra, we also had numerous biochemical measurements and extensive clinical information available for the study subjects. The coalescent nature of kidney disease and insulin resistance was confirmed by overlapping the SOM patterns for urine albumin excretion, weight-adjusted insulin dose, glycosylated hemoglobin (measure of long-term glucose control) and waist circumference. The ability of a single 1H NMR measurement to reveal multiple features of the effects of diabetes was thus validated (Figure 6).
This work is, to our knowledge, the first metabonomics study on premature death and vascular diseases in a large human cohort. We used only serum to characterize the patients, and yet the high-risk metabolic features were easily observable. This is an encouraging result with respect to general applicability as, unlike type I diabetes, urine albumin (or any other single biomarker) does not have an equally critical role in type II diabetes, let alone in the nondiabetic population. Furthermore, our application of 1H NMR metabonomics and statistical visualizations may improve the tracking of patients' progress in the diabetic disease continuum in a way not attainable by traditional approaches. Hence, it may become possible to re-route the multimetabolite path of a vulnerable patient away from adverse clinical endpoints and towards a more favorable phenotype before it is too late.
Acknowledgements
The skilled technical assistance by Antti Niinikoski and Taru Tukiainen is gratefully acknowledged. The study was supported by grants from the Folkhälsan Research Foundation, Samfundet Folkhälsan, the Jenny and Antti Wihuri Foundation and the Graduate School of Electrical and Communications Engineering at Helsinki University of Technology. This work was also supported by the Centre of Excellence Program of the Academy of Finland (KK, MAK). For a complete listing of the FinnDiane Study Group, please see Supplementary data 4.
