Evaluation of Established Coronary Heart Disease on the Basis of HDL and Non-HDL NMR Lipid Profiling Christina E. Kostara,† Athanasios Papathanasiou,‡ Manh Thong Cung,§ Moses S. Elisaf,‡ John Goudevenos,‡ and Eleni T. Bairaktari*,† Laboratory of Clinical Chemistry, and Department of Internal Medicine, Medical School, University of Ioannina, 451 10, Ioannina, Greece, and Laboratoire de Chimie-Physique Macromole´culaire, UMR 7568 CNRS-INPL, Nancy-Universite´, 1 Rue Grandville, B.P. 20451, 54001 Nancy Cedex, France Received September 3, 2009
A 1H NMR-based lipid profiling approach was used to investigate the prediction of coronary heart disease (CHD) and examine the confounding effect of factors such as gender, triglycerides, HDL-cholesterol and age levels on the prediction of disease. The HDL and non-HDL lipid profiles in 47 patients with triple vessel disease (TVD) and 41 patients with normal coronary arteries (NCA) both documented angiographically were generated. The presence of CHD was predicted with a sensitivity and specificity of 52% and 75% for HDL model and 78% and 80% for non-HDL, respectively. The lipid constituents of HDL lipoproteins which contributed to the separation between the two groups were the saturated fatty acids, cholesterol, total ω-3 fatty acids, degree of unsaturation, diallylic protons from polyunsaturated fatty acids, linoleic acid and, to a lesser extent, the number of fatty acids, triglycerides, unsaturated fatty acids and phosphatidylcholine. Respectively, for non-HDL, lipoproteins were the saturated fatty acids, number of fatty acids, cholesterol, unsaturated fatty acids and phosphatidylcholine. Gender, triglycerides, HDL-cholesterol and age influenced the lipid constituents of HDL and non-HDL lipoproteins that contributed to the separation between subgroups and confounded the predictive power of the models. NMR-based lipid profiling analysis could contribute to the identification of noninvasive markers for the presence and the development of the disease. Keywords: coronary heart disease • lipoprotein composition • lipid extracts • 1H NMR spectroscopy • lipid profiling
Introduction The causative relationship between plasma lipoproteins and atherosclerosis is widely recognized. Lipid and apolipoprotein constituents of lipoprotein particles have become increasingly important in characterizing the risk of cardiovascular diseases and in the diagnosis and management of disorders of lipoprotein metabolism.1,2 In clinical practice, the degree of the atherogenic or antiatherogenic properties of the plasma lipoprotein particles is commonly estimated by their cholesterol content. However, experimental studies have suggested that cholesterol content explains only partially these properties and that the composition and overall structure of the particles may play important roles in regulating metabolism and cholesterol homeostasis.3-8 In the emerging field of metabonomics, that is defined as “the quantitative measurement of the dynamic multiparametric metabolic response of living systems to pathophysiological * To whom correspondence should be addressed. Eleni T. Bairaktari, Ph.D., Eur Clin Chem; Head of Biochemistry Laboratory, University Hospital of Ioannina; Associate Professor of Clinical Chemistry, Laboratory of Clinical Chemistry, Medical School University of Ioannina, 451 10 Ioannina, Greece. Phone: +30-2651007620. Fax: +30-2651007871. E-mail:
[email protected]. † Laboratory of Clinical Chemistry, University of Ioannina. ‡ Department of Internal Medicine, University of Ioannina. § Nancy-Universite´. 10.1021/pr900783x
2010 American Chemical Society
stimuli or genetic modification” a large number of low molecular weight (LMW) from body fluids or tissues are detected quantitatively in a single step, promising immense potential for early diagnosis, therapy monitoring and for understanding the pathogenesis of many diseases.9,101H NMR spectroscopy is a valid tool in metabonomic studies and combined with advanced chemometric approaches provides a powerful platform for clinical research and diagnostic applications.11-13 Previous studies have examined the ability of NMR-based metabonomic analysis of serum constituents to predict the presence of coronary heart disease (CHD).14,15 Brindle et al. first used 600 MHz 1H NMR analysis of total serum to differentiate patients with severe coronary heart disease from angiographically normal subjects.14 Following this study, Kirschenlohr et al. examined the predictive power of the same methodology using 400 MHz spectra and plasma samples and suggested that the method is influenced by confounding factors such as gender and cholesterol-lowering medication and therefore lacks sufficient diagnostic accuracy.15 In both studies, the analysis was based on the regions of the 1H NMR spectrum mainly attributing to the major lipid classes. In the present study, compositional analysis of plasma lipids with proton NMR-based metabolic profiling was carried out to assess their contribution to the prediction of CHD and the Journal of Proteome Research 2010, 9, 897–911 897 Published on Web 12/20/2009
research articles confounding effect of factors such as gender, triglycerides, HDL-cholesterol and age levels on the prediction of the disease was examined. To better investigate the contribution of the atheroprotective or potentially atherogenic lipoproteins to the diagnosis of the disease, we performed the analysis separately for HDL and non-HDL lipid extracts. The atheroprotective biological properties of HDL, including cellular cholesterol efflux capacity, and antioxidative and anti-inflammatory activities are well-recognized.16 Recent observational and intervention studies suggest that the predictive value of non-HDL-C, a surrogate measure of atherogenic particle concentration for cardiovascular risk is as good, or better than, that of LDL-C and is comparable to that of apoB.17 Moreover, the composition of these atherogenic particles has been related with coronary heart disease.18,19
Materials and Methods Subjects. Forty-seven patients who were admitted to the Intensive Care Unit of University Hospital of Ioannina with a confirmed diagnosis of acute coronary syndrome, without persistent elevation of the ST segment in the electrocardiograph (NSTE-ACS) and had demonstrated angiographically threevessel disease, TVD, (defined as the presence of g50% diameter luminal narrowing in all 3 major epicardial vessel systems) participated in the study. The control group comprised 41 consecutive patients who were admitted to the hospital because of atypical episodes of chest pain (i.e., pain, pressure or discomfort in the chest, neck or arms not clearly exertional or not otherwise consistent with pain or discomfort of myocardial ischemic origin) without any increase in biochemical markers, met the inclusion and exclusion criteria (as defined below) of the study and had angiographically normal coronary arteries, NCA, (i.e., completely smooth coronary lumen with normal caliber). All patients underwent diagnostic coronary angiography within 7-9 days after the onset of the symptoms. The diagnosis of NSTE-ACS was based on (i) the presence of typical angina (time interval between onset of the symptoms and arrival to the emergency department no more than 4 h), (ii) the electrocardiographic changes (ST-segment depression of at least 0.5 mm to 0.05 mV in 2 or more contiguous leads, including reciprocal changes and/or inverted T waves of at least 1 mm to 0.1 mV always in the absence of any ST-segment elevation), and (iii) the increased biochemical markers (i.e., values of CK-MB mass or CK-MB fraction more than double those of the reference range, or troponin I >0.2 ng/mL within 48 h from admission), according to the AHA/ACC guidelines.20 Patients who did not meet the criteria for NSTE-ACS after the initial evaluation were excluded from the study. In addition, patients with history of CHD, chronic renal disease (glomureral filtration rate 3 times the upper normal limit or higher, cirrhosis), chronic obstructive pulmonary disease, overt hyper-/hypo-thyroidism or rheumatic diseases, patients on lipid-lowering (including statins, fibric acid derivatives, nicotinic acid, cholestyramine, ezetimibe or ω-3 fatty acids), or patients with history of diabetes mellitus (fasting glucose levels >126 mg/dL) were excluded from the study. Samples. In the patients who had a confirmed acute coronary episode, blood samples were obtained within the first 12 h from the onset of symptoms, while in the patients of the control group, in the morning before angiography. Serum was separated by centrifugation at 1500g for 15 min and an aliquot was stored at -80 °C until NMR analysis. 898
Journal of Proteome Research • Vol. 9, No. 2, 2010
Kostara et al. For all participants in the study, demographic characteristics, smoking habits, body mass index, as well as personal history with regard to the presence of hypertension (i.e., history of hypertension diagnosed and treated with medication, diet and/ or exercise, or blood pressure greater than 140 mmHg systolic or 90 mmHg diastolic on at least 2 occasions, or current use of antihypertensive pharmacological therapy), and family history of premature cardiovascular disease (i.e., any direct blood relativessparents, siblings, childrenswho have had angina, myocardial infarction or sudden cardiac death at age of less than 55 years old for male and 65 years for female relatives) were recorded. The Ethics Committee of the University Hospital of Ioannina approved the study and written informed consent was obtained from each participant. Clinical Chemistry. Analysis of clinical chemistry parameters of serum was carried out on an Olympus AU600 analyzer (Olympus Diagnostica, Hamburg, Germany). Total cholesterol and triglycerides were determined enzymatically and HDLcholesterol by a direct assay (Olympus Diagnostica, Hamburg, Germany). LDL-C was calculated by the Friedewald formula and non-HDL-C was calculated as total - HDL-C. Serum apolipoproteins AI, B and Lp(a) were measured on a Behring Nephelometer BN ProSpec (Dade-Behring, Lieberbach, Germany). Lipid Extraction. HDL lipoprotein particles were isolated from the non-HDL lipoproteins by a well-established precipitation technique, the AACC (American Association of Clinical Chemistry) Selected Method.21 The principle of the method is based on the selective precipitation of the apo-B containing lipoproteins (non-HDL lipoproteins) by mixing serum (1.5 mL) with polyanions and divalent cations [150 µL of a solution containing 10 g/L Dextran Sulfate (50 000 Da) and 500 mmol/L MgCl2]. The precipitated lipoproteins were then sedimented by low-speed centrifugation. HDL lipoproteins were collected from the supernate and non-HDL lipoproteins in the sediment were redisolved in Phosphate Buffer Solution (PBS). Lipid content of the two subfractions was extracted with methanol/ chloroform (2:l) according to a modification of the Bligh and Dyer method,22 dried in a stream of nitrogen and stored at -80 °C to avoid oxidative degradation. The samples were redissolved in deuterated mixture of methanol/chloroform (2:l) and bubbled with nitrogen in order to remove oxygen just prior to recording spectrum. To minimize the standing time of the sample in the liquid phase, 3-4 samples were redissolved each time. 1 H NMR Spectroscopy. The 1H NMR spectra were recorded at 298 K on a 600 MHz Bruker Avance DRX NMR spectrometer (Laboratoire de Chimie-Physique Macromolecuraire, CNRSINPL, ENSIC, Nancy Cedex, France) operating at a field strength of 14.7 T and running on TopSpin 1.2 (Bruker Biospin Ltd.). Spectra were acquired in the Fourier transform (FT) mode with 32K data points, 128 free induction decays (FID), 90° pulses and relaxation delay of 3 s. A spectral width of 7184 Hz resulted in an FID acquisition time of 4.56 s for a total recycle time of 7.56 s. The residual HOD signal at about 4.75 ppm was suppressed by the application of a continuous and selective secondary irradiation during the relaxation delay. All FIDs were multiplied by an exponential weighting function corresponding to a 0.3 Hz line-broadening factor prior to Fourier transformation. The acquired NMR spectra were manually corrected for phase and baseline distortions (by applying a simple polynomical curve fit) with TopSpin 1.2 (Bruker Biospin Ltd.) and
research articles
Evaluation of Established Coronary Heart Disease
Figure 1. 1H NMR (600 MHz) spectrum of an HDL (down) and non-HDL (up) lipid extract. Peak assignments are summarized in Table 1. (a) Partial spectrum of cholesterol C-18 methyl group. (b) Partial spectrum of phospholipids headgroups.
referenced to the methanol peak (δ1 H 3.30). Chemical shifts were identified as described elsewhere23,24 and by obtaining 1D and 2D COSY spectra of authentic lipid samples. 1 H NMR Analysis of HDL and Non-HDL Total Lipids. A typical 1H NMR spectrum of the total lipid extract of HDL lipoprotein particles is shown in Figure 1. Signals attributed to cholesterol, in free and esterified form, triglycerides, phospholipids and fatty acid residues comprise the total lipid extract NMR fingerprint. Each lipid molecule appears in the spectrum with more than one signal, the chemical shift of which depends on the chemically different proton groups in its structure, as it is described below for each class of lipids. Certain signals, having similar chemical nature with signals from other molecules, occur superimposed, whereas others appear wellresolved having a unique chemical shift and are therefore suitable for identification (Table 1). Cholesterol. The signals from all proton groups of cholesterol are extended in the spectrum region from 0.68 to 5.36 ppm. The prominent cholesterol signals are the methyl groups on the carbons 18, 19, 21, 26, 27 of the molecule (Figure 1). Other cholesterol signals observed are the C-3 proton multiplet at 3.40 ppm and the C-6 vinyl proton at 5.36 ppm. The most wellresolved signal of cholesterol is that of C-18 at 0.68 ppm and it is further split into its two forms, free and esterified (Figure 1a). Triglycerides. The proton groups of the glycerol backbone of triglycerides appear with 3 signals in the region 4.16-5.22 ppm with the best resolved of them to be the signal of C1Hd and C3Hd at 4.32 ppm. Phospholipids. Phospholipids can be distinguished in the NMR spectrum by unique characteristic signals of their headgroups. Choline containing phospholipids, phosphatidylcholine and sphingomyelin accounts for approximately 93% of total
Table 1.
1
H NMR Resonances of HDL Lipids
lipid Cholesterol
Triglycerides
Phospholipids
Fatty acid residues
numbera
assignments
1 3 5 16 23 18 20 21 15 17 19
C18H3 C26H3, C27H3, C21H3 C19H3 C3H C6H C1Hu and C3Hu of glycerol C1Hd and C3Hd of glycerol C2H of glycerol N+(CH3)3 (PC and SM) CH2-N+(CH3)3 (PC and SM) -O-CH2-CH2-N+(CH3)3 (PC and SM) CH2-NH2 of PE ω-CH3 ω-CH3 of total omega-3 fatty acids (CH2)n CO-CH2-CH2 β-CH2 of ARA+EPA CH2-CHd -CO-CH2 R and β CH2 of DHA -CHdCH-CH2-CHdCH- of linoleic acid (CHdCH-CH2-CHdCH)n, n > 1 CH)CH
14 2 4 6 7 8 9 10 11 12 13 22
chemical shift (ppm) 0.68 0.87 1.00 3.40 5.36 4.16 4.32 5.22 3.20 3.59 4.24 3.10 0.88 0.95 1.30 1.59 1.67 2.04 2.30 2.38 2.75 2.80 5.36
a Number in the spectrum of Figure 1. u, up-field; d, down-field; ARA, arachidonic acid; DHA, docosahexaenoic acid; EPA, eicosapentaenoic acid; PC, phosphatidylcholine; PE, phosphatidylethanolamine; SM, sphingomyelin.
phospholipids in HDL lipoproteins and are well-discerned from three signals: the very intense -N+(CH3)3 signal at 3.20 ppm, the -CH2-N+(CH3)3 at 3.59 ppm and the -O-CH2-CH2-N+(CH3)3 at 4.24 ppm. Furthermore, the signal at 3.20 ppm is split in Journal of Proteome Research • Vol. 9, No. 2, 2010 899
research articles the N-trimethyl groups of phosphatidylcholine and sphingomyelin (Figure 1b). Phosphatidylethanolamine is identified by the characterictic resonance of the headgroup -CH2NH2 methylene proton at 3.10 ppm. Fatty Acids. All fatty acid chains show for the ω-CH3 protons signals at 0.88 ppm and specifically for the ω-CH3 protons of total omega-3 fatty acids at 0.95 ppm, for the R- and βmethylene protons at 2.30 and 1.59 ppm, respectively, and for the rest of the methylene protons [(CH2)n] at 1.30 ppm. Allylic and diallylic methylene protons appear at 2.04 and 2.80 ppm, respectively, and olefinic protons at 5.36 ppm. Individual polyunsaturated fatty acids that were resolved and were quantified from the NMR spectrum are: fatty acids containing only two double bonds (i.e., linoleic acid; 18:2∆9,12) from their structure specific diallylic proton resonances at 2.75 ppm, docosahexaenoic (DHA) from the isolated methylene group of resonances at 2.38 ppm characterizing the ∆4 double bond, and collectively the fatty acids arachidonic and eicosapentaenoic (ARA + EPA) from the characteristic signal at 1.67 ppm characterizing the ∆5 double bond. The same lipid constituents appear in the spectrum of nonHDL lipid extracts (Figure 1). Statistical Analysis. Statistical analysis was performed with Statistica Ver. 6.0 (StatSoft Inc. Tulsa, OK). Values were expressed as mean value ( standard deviation (SD) and compared by using t test, while Lp(a) parameter was presented as median (range). Significance levels were set at 0.05. NMR Data Reduction and Pattern Recognition (PR). The 1 H NMR spectra were automatically reduced by using the AMIX (Analysis of MIXtures) software package (version 3.2.4, Bruker Analytik, Rheinstetten, Germany) to 179 continuous integral segments (variables or bins) of equal width of 0.03 ppm corresponding to the chemical shift range δ1H, 0.49-5.98. The bin size selected was appropriate for the correct estimation of the independent contribution of most of the signals and especially that of phosphatidylcholine and sphingomyelin at 3.2 ppm. The area between 4.58 and 5.00 ppm was excluded to remove any effect of variation from the suppression of the water resonance and the area between 3.24 and 3.50 ppm containing deuterated methanol solvent (MEOD). All data was normalized by dividing each integral segment by the total area of the spectrum in order to compensate for the differences in overall concentration between individual HDL and non-HDL lipid extracts samples. The resulting data matrix, consisting of 145 NMR integral segments, was exported to the SIMCA-P software package (version 10.5, UMETRICS AB, Box 7960, SE 90719, Umea˚, Sweden) for the PR analysis. Prior to the analysis, the NMR data were centered scaled, where the average value of each variable is calculated and then subtracted from the data. PCA was used for the overview of the data set and the spotting of outliers, and then for the detection of any grouping or separation trend.25 The PCA scores plot was used to reveal observations lying outside the 0.95 Hotteling’s T2 ellipse (strong outliers) and the loadings plot to interpret the patterns seen in the scores plot. With Partial Least Squares Discriminant Analysis (PLS-DA), a relationship was sought between the matrix of variables X (NMR spectral bins) and a matrix of dependent variables Y (dummy variables encoding the class membership, i.e., patients with TVD or those with NCA). The method was used to find the best possible discriminant function (model) that separates patients with TVD from those with NCA on the basis of their X 900
Journal of Proteome Research • Vol. 9, No. 2, 2010
Kostara et al. 25
variables. For the interpretation of the scores plot, the regression coefficients plot was used which shows all spectral regions that contribute to the separation between the studied groups. The technique of Orthogonal Signal Correction (OSC) was applied to remove linear combinations of variables X that were orthogonal to the Y vector of the dependent variables, to eliminate the intersubject variability and to describe maximum separation based on class.26 The default method of 7-fold internal cross-validation (CV) of SIMCA-P software was applied and the extracted parameter Q2 was used to provide an estimation of the predictive capability of the PLS-DA models with Q2 > 0.5 considered ‘good’ and Q2 > 0.9 ‘excellent’.25 The parameter R2 describes the explained variation and how well the data can be mathematically reproduced by the training model. In addition, validation was performed using held-back and external data procedures. In held-back data validation, the 100% of the data set was used for the construction of the model and then a portion of the training set, named as test set, was left out of the model and then predicted, whereas in external data validation, a portion of the training set, now named as prediction set, that was not used when the model was built, is left out of the model and then predicted. Held-back and external data validation for the TVD - NCA patients was performed using 80% of the data as the training set and the remaining 20% as the test (prediction) set. All observations were assigned with a class-specific numerical value to form a response Y-matrix. Correct classification was based on a predicted Y cutoff of 1.5 (held-back data validation) and 0.5 (external validation) with a 95% confidence level. Correct and incorrect assignments were used to define True Positives (TP), True Negatives (TN), False Positives (FP) and False Negatives (FN) classification rates and then to estimate as percent sensitivity [TP/(TP + FN) × 100] and specificity [TN/ (TN + FP) × 100].27
Results The main clinical and biochemical data of the two groups studied are shown in Table 2. Patients with TVD were characterized by higher age (67.62 ( 9.72 vs 61.02 ( 9.62 year, p < 0.01) compared to those with NCA. There was no statistically significant difference in risk factors for coronary heart disease such as hypertension, smoking, family history of premature cardiovascular disease and body mass index (BMI). Patients with TVD presented significantly higher levels of triglycerides (p < 0.001) and lower levels of HDL-C (p < 0.01) and apoAI (p < 0.001) compared to those with NCA. Total cholesterol, LDLC, non-HDL-C, apoB and Lp(a) were not significantly different between the two groups. Construction of the Angiographic Status Based Models. The data set consisted of the 1H NMR spectra of HDL and nonHDL lipid extracts from 88 patients: 47 with TVD and 41 with NCA. The models for HDL and non-HDL data sets were constructed separately. In both HDL and non-HDL data sets, PCA and PLS-DA were applied and no distinct grouping was identified between the two groups studied in the scores plots (Figure 2a-d, Table 3) apart from a slight trend in the non-HDL models of the patients with TVD to be placed at the right part of the plot and of those with NCA to spread mostly in the left part. To minimize the possible intrinsic contribution of intersubject variability, the method of OSC filtering was applied from which two orthogo-
research articles
Evaluation of Established Coronary Heart Disease Table 2. Clinical and Biochemical Characteristics of the Patients with and without CHD
Age (years) Males/Females
patients with TVDa
patients with NCA
47 67.62 ( 9.72b 40/7
41 61.02 ( 9.62 25/16
Risk Factors for CHD Hypertension (n, %) 31 (57.4%) 22 (52.4%) Current Smokers (n, %) 19 (35.2%) 15 (35.7%) Family history of 6 (11.1%) 5 (11.9%) premature cardiovascular disease (n, %) 26.9 ( 2.37 27.8 ( 2.24 Body mass index (kg/m2) Lipid Parameters Total Cholesterol (mg/dL) 206.0 ( 53.1 Triglycerides (mg/dL) 168.4 ( 72.4 HDL-C (mg/dL) 41.0 ( 8.7 LDL-C (mg/dL) 131.4 ( 47.4 non-HDL-C (mg/dL) 165.1 ( 47.6 apoAI (mg/dL) 109.6 ( 27.5 apoB (mg/dL) 99.1 ( 29.2 41.7-78.6 Lp(a) (mg/dL)d
212.24 ( 47.6 121.3 ( 43.1 48.8 ( 12.0 139.2 ( 39.0 163.5 ( 39.6 135.4 ( 28.5 100.1 ( 26.0 28.2-45.7
p
