Clinical Chemistry AACC Online Job Center
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Clinical Chemistry 51: 1457-1461, 2005. First published June 16, 2005; 10.1373/clinchem.2004.046748
This Article
Right arrow Abstract Freely available
Right arrow Full Text (PDF)
Right arrow All Versions of this Article:
clinchem.2004.046748v1
51/8/1457    most recent
Right arrow Submit an electronic Letter to
the Editor about this paper
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via ISI Web of Science (6)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Petersen, M.
Right arrow Articles by Dyerberg, J.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Petersen, M.
Right arrow Articles by Dyerberg, J.
Related Collections
Right arrow Lipids, Lipoproteins, and Cardiovascular Risk Factors
(Clinical Chemistry. 2005;51:1457-1461.)
© 2005 American Association for Clinical Chemistry, Inc.

Lipids, Lipoproteins, and Cardiovascular Risk Factors

Quantification of Lipoprotein Subclasses by Proton Nuclear Magnetic Resonance–Based Partial Least-Squares Regression Models

Martin Petersen1,a, Marianne Dyrby3, Søren Toubro1, Søren Balling Engelsen2, Lars Nørgaard2, Henrik Toft Pedersen4 and Jørn Dyerberg1

1 Institute of Human Nutrition, Centre for Advanced Food Studies, and 2 Centre for Advanced Food Studies, Department of Food Science, Quality and Technology, The Royal Veterinary and Agricultural University, Frederiksberg, Denmark.
3 Umetrics AB, Malmö, Sweden.
4 Novo Nordisk A/S, Virology & Molecular Toxicology, Måløv, Denmark.

aAddress correspondence to this author at: Institute of Human Nutrition, Rolighedsvej 30, DK-1958 Frederiksberg C, Denmark. Fax 45-3528-2483; e-mail mpe{at}kvl.dk.


   Abstract
Top
 Abstract
Introduction
 Materials and Methods
 Results
 Discussion
 References
 
Background: Cardiovascular disease risk can be estimated in part on the basis of the plasma lipoprotein profile. Analysis of lipoprotein subclasses improves the risk evaluation, but the traditional methods are very time-consuming. Novel, rapid, and productive methods are therefore needed.

Methods: We obtained plasma samples from 103 fasting people and determined the plasma lipoprotein subclass profiles by an established ultracentrifugation-based method. Proton nuclear magnetic resonance (NMR) spectra were obtained from replicate samples on a 600 MHz NMR spectrometer. From the ultracentrifugation-based reference data and the NMR spectra, we developed partial least-squares (PLS) regression models to predict cholesterol and triglyceride (TG) concentrations in plasma as well as in VLDL, intermediate-density lipoprotein (IDL), LDL, 3 LDL fractions, HDL, and 3 HDL subclasses.

Results: The correlation coefficients (r) between the plasma TG and cholesterol concentrations measured by the 2 methods were 0.98 and 0.91, respectively. For LDL- and HDL-cholesterol concentrations, r = 0.90 and 0.94, respectively. For cholesterol concentrations in the LDL-1, LDL-2, and LDL-3 fractions, r = 0.74, 0.78, and 0.69, respectively, and for HDL subclasses HDL2b, HDL2a, and HDL3, cholesterol concentrations were predicted with r = 0.92, 0.94, and 0.75, respectively. TG concentrations in VLDL, IDL, LDL, and HDL were predicted with correlations of 0.98, 0.85, 0.77, and 0.74, respectively. The cholesterol and TG concentrations in the main lipoprotein fractions and in LDL fractions and HDL subclasses predicted by the PLS models were 94%–100% of the concentrations obtained by ultracentrifugation.

Conclusion: NMR-based PLS regression models are appropriate for use in research in which analyses of the plasma lipoprotein profile, including LDL and HDL subclasses, are required in large numbers of samples.


   Introduction
Top
 Abstract
 Introduction
Materials and Methods
 Results
 Discussion
 References
 
The risk of cardiovascular disease (CVD) 1 can be estimated in part on the basis of the total plasma cholesterol and triglyceride (TG) concentrations (1). However, if total cholesterol is divided into its 2 major constituents, LDL- and HDL-cholesterol, the estimation of CVD risk is improved (2). On the basis of size and density, LDL particles can be divided into large, buoyant particles and small, dense particles (3). The latter have been demonstrated to correlate strongly with the risk of CVD (4)(5)(6)(7). Risk assessment can also be improved by dividing the HDL particles into subclasses (8). Thus, patients with similar concentrations of LDL- and HDL-cholesterol but with different subclass distributions may have different CVD risks. Consequently, to reveal this difference and improve risk assessment, analysis of lipoprotein subclasses must be performed.

Lipoprotein subclasses are usually measured by gradient gel electrophoresis or by ultracentrifugation. After gradient gel electrophoresis analysis of LDL particles, samples can be classified as pattern A, in which the large, buoyant particles are the most frequent, an intermediate pattern, or pattern B, in which the atherogenic, small, dense particles are the predominant. Gradient gel electrophoresis analysis also allows the peak particle diameter to be estimated (9). Ultracentrifugation can be used to determine the concentration of small, dense LDL particles, which is probably the most clinically relevant risk marker (9)(10). However, ultracentrifugation analysis of both LDL and HDL subclasses is time-consuming and not suitable for studies with a large number of samples.

A novel approach to quantify lipoproteins is high-resolution proton nuclear magnetic resonance (NMR) spectroscopy, followed by a mathematical procedure that translates the NMR signal into lipoprotein concentrations. The advantage of this method is that a plasma sample can be analyzed within a few minutes (11). Several studies have attempted to use this principle to quantify cholesterol and TGs in plasma, as well as in VLDL, LDL, and HDL, with high correlations to established methods (12)(13)(14). Only a few studies have attempted to quantify LDL and HDL subfractions, however, and they used either calibration curve resolution and integration (15) or a chemometric approach (16)(17).

In this study, we used chemometric partial least-squares (PLS) regression to calibrate 600 MHz proton-NMR spectra to measurements from an ultracentrifugation-based technique (18), thus enabling the quantification of cholesterol and TGs in plasma and in lipoproteins, including LDL and HDL subclasses.


   Materials and Methods
Top
 Abstract
 Introduction
 Materials and Methods
Results
 Discussion
 References
 
samples
Participants were recruited from May 2001 until September 2002. The study was approved by the Ethical Committee of København and Frederiksberg. Volunteers were informed about the nature of the study, and written consent was obtained before participation.

Blood was drawn after a 12-h fast from an antecubital vein into EDTA tubes, and plasma was separated by centrifugation at 3000g for 15 min. Each plasma sample was gently mixed, and aliquots for NMR analysis and ultracentrifugation were stored at –80 °C until further analysis.

The samples were obtained from obese participants (initial body mass index >28 kg/m2) of both sexes (36% males) before (n = 34) and after (n = 36) a weight loss of at least 8% body weight, and from nonobese volunteers (n = 33). The age range of the participants was 19–74 years.

ultracentrifugation
Lipoproteins were separated by serial ultracentrifugation as described previously (18). The main fractions—VLDL, intermediate-density lipoprotein (IDL), LDL, and HDL—were separated from plasma one at a time by sequential ultracentrifugation. The LDL fraction was dialyzed against a NaCl–NaBr solution with a density of 1.035 kg/L and subjected to ultracentrifugation. Six LDL fractions were obtained by pipetting from the top of the ultracentrifugation tube. However, for the NMR-based PLS regression models, the 4 intermediate fractions were pooled into a single fraction (LDL-2). HDL subclasses were also isolated by ultracentrifugation and obtained by pipetting. The densities of the lipoprotein fractions were as follows: VLDL, <1.006 kg/L; IDL, 1.006–1.019 kg/L; LDL, 1.019–1.063 kg/L; and HDL, 1.063–1.210 kg/L. LDL was separated into 3 fractions (LDL-1, 1.019–1.031 kg/L; LDL-2, 1.031–1.044 kg/L; LDL-3, 1.044–1.063 kg/L) and HDL into 3 subclasses (HDL2b, 1.063–1.100 kg/L; HDL2a, 1.100–1.125 kg/L; and HDL3, 1.125–1.210 kg/L), giving a total of 10 lipoprotein fractions. In the plasma and the 10 fractions, cholesterol and TGs were quantified by enzymatic assays from Roche Diagnostics on a Cobas Mira Plus instrument (Roche).

We determined the CVs of the lipid analyses by analyzing 6 duplicate samples. The CVs and between-run errors were between 3% and 9% for the ultracentrifugation-based measurements of cholesterol and TGs in plasma, VLDL, IDL, LDL, and HDL. For HDL subclassification, there was an additional error of 4%. For LDL fractions, we expected a similar error but could not assess it because of a dialysis step in the preparation of samples for classification of LDL fractions.

proton nmr
Samples were prepared in well plates with each well containing 100 µL of plasma, 350 µL of 9 g/L NaCl in H2O, and 50 µL of D2O to provide a deuterium field/frequency lock. Proton spectra were measured at 30 °C on a Bruker DRX600 operating at 14.1 T with a 120-µL flow probe. A 1-dimensional longitudinal eddy current bipolar gradient pulse presaturation with 2 stimulated echoes, diffusion-edited NMR experiment was performed with a diffusion time of 0.1 s to suppress signals from small molecules. Signals from water were suppressed by use of the nuclear Overhauser effect spectroscopy (NOESY)-presaturation sequence, and a spectral window of 12 019 Hz was accumulated in an acquisition time of 1.36 s. The relaxation delay was 2.0 s, the free induction decay was collected into 16 000 complex data points, and 128 scans were accumulated for each sample.

After acquisition, the free induction decays were Fourier-transformed by application of 2 times zerofilling and an exponential window function, giving 1-Hz line broadening. The spectra were automatically phase-corrected by use of an in-house routine written in Matlab (The MathWorks) followed by a manual check, referenced to the EDTA singlet located at 2.56 ppm and corrected for baseline offset errors by use of the flat regions at both ends of the spectrum. The spectral area chosen for multivariate analysis was 5.7–0.2 ppm, excluding 5.25–5.23, 5.12–4.36, 3.96–3.24, 3.18–3.07, and 2.72–2.36 ppm, which covers the residual water region and the regions containing peaks from EDTA as well as some minor peak residuals from small molecules (Fig. 1 ).



View larger version (9K):
[in this window]
[in a new window]
  		Figure 1. Diffusion-edited proton-NMR spectrum of a plasma sample.

Water, EDTA, and lipid peaks are indicated by arrows.

chemometric analysis and statistics
PLS regression models were developed by use of the acquired NMR spectra and the concentrations of cholesterol and TGs in each of the lipoprotein subfractions. For each model, outliers attributable to instrumental artifacts or very extreme samples were removed, and to determine the correct number of PLS components, full cross-validation was applied. To evaluate the quality of the prediction models, we calculated the root mean square error of cross-validation (RMSECV) and correlation coefficients (r) between measured and predicted concentrations. The RMSECV was corrected for degrees of freedom and then calculated according to the equation:

where n is the number of samples included in the model, and A is the number of components used in the model. Correction for degrees of freedom was performed to obtain a reasonable but conservative estimate of the prediction error. All calculations were performed with the PLS_Toolbox, Ver. 3.0 (Eigenvector Research) running under Matlab, Ver. 7.0.1 (The MathWorks).


   Results
Top
 Abstract
 Introduction
 Materials and Methods
 Results
Discussion
 References
 
The cholesterol and TG concentrations in lipoprotein subfractions as measured by the ultracentrifugation method are presented in Table 1 .


View this table:
[in this window]
[in a new window]
  		Table 1. Summary of cholesterol and TG concentrations (in mmol/L) in 103 samples assayed by the ultracentrifugation-based method.

The results from the calibrations are seen in Table 2 . The correlation coefficients (r) between the concentrations of plasma TG and cholesterol measured by ultracentrifugation and the values predicted by the NMR-based PLS models were 0.98 and 0.91, respectively. The correlation coefficients for the LDL- and HDL-cholesterol concentrations predicted by the PLS regression models vs the ultracentrifugation method were 0.90 and 0.94, respectively. The cholesterol concentrations in LDL-1, LDL-2, and LDL-3 were predicted with correlation coefficients of 0.74, 0.78, and 0.69, respectively. For HDL subclasses HDL2b, HDL2a, and HDL3, the cholesterol concentrations were predicted with correlation coefficients of 0.92, 0.94, and 0.75, respectively. The TG concentrations in VLDL, IDL, LDL, and HDL were predicted with correlation coefficients of 0.98, 0.85, 0.77, and 0.74, respectively.


View this table:
[in this window]
[in a new window]
  		Table 2. Prediction results (from cross-validation) of lipoprotein subclasses by NMR, with ultracentrifugation as comparison method (n = 103).

The cholesterol and TG concentrations in the lipoprotein fractions, including LDL fractions and HDL subclasses, predicted by the NMR-based PLS regression model were 95%–100% of the values obtained by ultracentrifugation.


   Discussion
Top
 Abstract
 Introduction
 Materials and Methods
 Results
 Discussion
References
 
In the present study we established NMR-based PLS regression models that estimate blood lipid concentrations, including LDL fractions and HDL subclasses. The estimates were calculated with high agreement with an established ultracentrifugation method. The PLS-predicted cholesterol and TG concentrations in the main lipoprotein and LDL fractions and HDL subclasses were 95%–100% of the results obtained by ultracentrifugation. Pairwise results obtained by the ultracentrifugation-based method and the NMR-based PLS regression model were within 1%, indicating very good agreement between the methods.

The agreement between the concentrations of TG in plasma and VLDL measured by the ultracentrifugation method and predicted by PLS was good (r = 0.98 for both). The prediction of TG concentrations in IDL, LDL, and HDL had a lower correlation with the ultracentrifugation method, which is probably a consequence of lower TG concentrations in these fractions. The number of PLS components were only 2 and 3 for TG concentrations in plasma and VLDL and 4–5 for TG concentrations in IDL, LDL, and HDL, indicating that the models for the latter are based on more subtle differences in the NMR spectra. Correlations were generally acceptable for the TG concentrations in the LDL fractions and HDL subclasses, except for LDL-3, for which r was 0.57.

For cholesterol, the correlation coefficients between the ultracentrifugation measurements and the PLS models were 0.91, 0.94, 0.90, and 0.94 for plasma, VLDL, LDL, and HDL, respectively. These correlations indicate that the NMR-based PLS models are able to predict cholesterol concentrations with adequate accuracy. The correlation coefficients for the prediction of cholesterol concentrations in LDL fractions and HDL subclasses ranged from 0.69 to 0.94. This is probably attributable to the high cholesterol concentrations in these subclasses.

Bathen et al. (12) also used NMR-based PLS models to predict cholesterol and TG concentrations in plasma, LDL, and HDL. Their correlations were slightly lower than those found in our study, which could be attributable to the use of fewer samples in their models (n = 44). Another explanation might be the broader spectral range allowed in the present study compared with the study by Bathen et al. (12), who used only the CH3 and CH2 peaks. Furthermore, in the present study, diffusion-edited NMR spectra were used to suppress the lactate signal at 1.33 ppm, which overlaps with the CH2 signal. The suppression of this interference signal possibly improved our model.

Although our models predict the plasma lipid profile better than those used in previous reports, the correlation coefficient for the prediction of TG in LDL-3 is only 0.57. One possible source of error, which can partly explain the low correlation observed, is that the subclasses of lipoproteins separated by ultracentrifugation do not represent particles with discrete density but rather continuous ranges of densities. The data therefore do not comply with the basic assumption for applying PLS regression: that the data are low-rank bilinear. Another explanation for poor predictions is that severe spectral overlaps make it difficult for the calibration models to distinguish the nearly identical NMR signals arising from neighboring subclasses. In addition, certain response variables, e.g., TG in LDL-3, are found in low concentrations and had a small variation in the calibration set.

The pioneer NMR method for prediction of lipoproteins, introduced by Otvos et al. (11)(15), is not a model-free PLS method, but a so-called hard model. This means that the signals from the NMR spectrometer are related to a particular lipid profile measured by a comparison method through a multiparametric curve-resolution technique. Nevertheless, the correlations between established methods and the concentrations predicted by the NMR LipoProfile® (the method patented by Otvos et al.) are high.

Prediction of lipoprotein concentrations in unknown samples can be correctly performed only if the concentrations fall between the minimum and maximum values indicated in Table 1Up . In addition, when the plasma TG concentration exceeds 5 mmol/L, the VLDL particles stick to the ultracentrifugation tubes and the quantities are consequently underestimated, particularly VLDL-TG and VLDL-cholesterol concentrations. Because we used ultracentrifugation as the comparison method, this underestimation will also be reflected in the NMR method. However, both LDL fractions and HDL subclasses are heavily skewed toward an excess of small, dense particles if the plasma TG concentration is >5 mmol/L (19), and with respect to measurement of subclasses, the practical relevance of this bias is of less importance.

In conclusion, the present NMR-based PLS models for plasma lipoprotein subclass analysis are suitable for research projects including a large number of participants.


   Footnotes
 
1 Nonstandard abbreviations: CVD, cardiovascular disease; TG, triglyceride; NMR, nuclear magnetic resonance; PLS, partial least-squares; IDL, intermediate-density lipoprotein; and RMSECV, root mean square error of cross-validation.


   References
Top
 Abstract
 Introduction
 Materials and Methods
 Results
 Discussion
 References
 

  1. Castelli WP. Lipids, risk factors and ischaemic heart disease. Atherosclerosis 1996;124(Suppl):S1-S9.
  2. Gordon T, Kannel WB, Castelli WP, Dawber TR. Lipoproteins, cardiovascular disease, and death. The Framingham study. Arch Intern Med 1981;141:1128-1131.[Abstract]
  3. Krauss RM, Burke DJ. Identification of multiple subclasses of plasma low density lipoproteins in normal humans. J Lipid Res 1982;23:97-104.[Abstract]
  4. Austin MA, Breslow JL, Hennekens CH, Buring JE, Willett WC, Krauss RM. Low-density lipoprotein subclass patterns and risk of myocardial infarction. JAMA 1988;260:1917-1921.[Abstract]
  5. Lamarche B, Tchernof A, Moorjani S, Cantin B, Dagenais GR, Lupien PJ, et al. Small, dense low-density lipoprotein particles as a predictor of the risk of ischemic heart disease in men. Prospective results from the Quebec Cardiovascular Study. Circulation 1997;95:69-75.[Abstract/Free Full Text]
  6. Stampfer MJ, Krauss RM, Ma J, Blanche PJ, Holl LG, Sacks FM, et al. A prospective study of triglyceride level, low-density lipoprotein particle diameter, and risk of myocardial infarction. JAMA 1996;276:882-888.[Abstract]
  7. Gardner CD, Fortmann SP, Krauss RM. Association of small low-density lipoprotein particles with the incidence of coronary artery disease in men and women. JAMA 1996;276:875-881.[Abstract]
  8. Laakso M, Lehto S, Penttila I, Pyorala K. Lipids and lipoproteins predicting coronary heart disease mortality and morbidity in patients with non-insulin-dependent diabetes. Circulation 1993;88:1421-1430.[Abstract/Free Full Text]
  9. Austin MA, Hokanson JE, Brunzell JD. Characterization of low-density lipoprotein subclasses: methodologic approaches and clinical relevance. Curr Opin Lipidol 1994;5:395-403.[Medline] [Order article via Infotrieve]
  10. Lamarche B, Lemieux I, Despres JP. The small, dense LDL phenotype and the risk of coronary heart disease: epidemiology, patho-physiology and therapeutic aspects. Diabetes Metab 1999;25:199-211.[ISI][Medline] [Order article via Infotrieve]
  11. Otvos JD, Jeyarajah EJ, Bennett DW. Quantification of plasma lipoproteins by proton nuclear magnetic resonance spectroscopy. Clin Chem 1991;37:377-386.[Abstract/Free Full Text]
  12. Bathen TF, Krane J, Engan T, Bjerve KS, Axelson D. Quantification of plasma lipids and apolipoproteins by use of proton NMR spectroscopy, multivariate and neural network analysis. NMR Biomed 2000;13:271-288.[CrossRef][ISI][Medline] [Order article via Infotrieve]
  13. Serrai H, Nadal L, Leray G, Leroy B, Delplanque B, de Certaines JD. Quantification of plasma lipoprotein fractions by wavelet transform time-domain data processing of the proton nuclear magnetic resonance methylene spectral region. NMR Biomed 1998;11:273-280.[CrossRef][ISI][Medline] [Order article via Infotrieve]
  14. Ala-Korpela M, Hiltunen Y, Bell JD. Quantification of biomedical NMR data using artificial neural network analysis: lipoprotein lipid profiles from 1H NMR data of human plasma. NMR Biomed 1995;8:235-244.[ISI][Medline] [Order article via Infotrieve]
  15. Otvos JD, Jeyarajah EJ, Bennett DW, Krauss RM. Development of a proton nuclear magnetic resonance spectroscopic method for determining plasma lipoprotein concentrations and subspecies distributions from a single, rapid measurement. Clin Chem 1992;38:1632-1638.[Abstract/Free Full Text]
  16. Dyrby M, Petersen M, Engelsen SB, Nørgaard L, Sidelmann UG. Magnetic resonance in food science. Webb GA Belton PS Gil AM Rutledge DN eds. Latest developments 2002:101-107 The Royal Society of Chemistry Cambridge. .
  17. Dyrby M, Petersen M, Whittaker AK, Lambert L, Nørgaard L, Bro R, et al. Analysis of lipoproteins using 2D diffusion-edited NMR spectroscopy and multi-way chemometrics. Anal Chim Acta;in press..
  18. Baumstark MW, Kreutz W, Berg A, Frey I, Keul J. Structure of human low-density lipoprotein subfractions, determined by X-ray small-angle scattering. Biochim Biophys Acta 1990;1037:48-57.[CrossRef][Medline] [Order article via Infotrieve]
  19. Griffin BA, Minihane AM, Furlonger N, Chapman C, Murphy M, Williams D, et al. Inter-relationships between small, dense low-density lipoprotein (LDL), plasma triacylglycerol and LDL apoprotein B in an atherogenic lipoprotein phenotype in free-living subjects. Clin Sci (Lond) 1999;97:269-276.[Medline] [Order article via Infotrieve]




This Article
Right arrow Abstract Freely available
Right arrow Full Text (PDF)
Right arrow All Versions of this Article:
clinchem.2004.046748v1
51/8/1457    most recent
Right arrow Submit an electronic Letter to
the Editor about this paper
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via ISI Web of Science (6)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Petersen, M.
Right arrow Articles by Dyerberg, J.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Petersen, M.
Right arrow Articles by Dyerberg, J.
Related Collections
Right arrow Lipids, Lipoproteins, and Cardiovascular Risk Factors

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS