Articles |
From the Life Sciences Division, Ernest Orlando Lawrence Berkeley National Laboratory, Berkeley, Calif.
| Abstract |
|---|
|
|
|---|
Key Words: gradient gel electrophoresis HDL
| Introduction |
|---|
|
|
|---|
| Methods |
|---|
|
|
|---|
1.20 kg/L was
obtained after a single-spin ultracentrifugation
(114 000g, 24 hours, 15°C; Beckman rotor, Beckman
Instruments). Electrophoresis of HDL was performed on Pharmacia
electrophoresis apparatus (GE 4-II) by using slab gradient
gels (PAA 4/30, Pharmacia).1 14 Time for electrophoresis
of HDL was generally 24 hours. A mixture of four globular proteins (HMW
Calibration Kit) was run on the central lane to calibrate particle
size. Thyroglobulin was added to each plasma sample to identify a
common reference on all lanes of the gel. The thyroglobulin peak was
identified for each sample and assigned a migration distance of 0. The
migration peak of BSA, one of the protein standards of the calibration
lane, was assigned a migration distance of 1 (the assigned values are
arbitrary and do not affect the calculations that follow). The HDL
migration distances (Rf's) were measured as the
fraction of the migration distance between the BSA and thyroglobulin
peaks. The gradient gels were then stained for protein with Coomassie
Brilliant Blue G-250 and scanned with a densitometer (model RFT,
Transidyne Corp) at a wavelength of 603 nm. A computer file of
absorbance versus Rf was obtained for 1000
equidistant points along the gradient gel. Originally, we proposed
using the known concentration of the thyroglobulin added to each HDL
sample to correct for differences in sample volume and stain
uptake.10 In subsequent studies,5 6 7 8 9 11 12 we
made no adjustment for the absorbance of HDL for the area of the
thyroglobulin peak because this adjustment was not found to decrease
the variance of the HDL measurements. Standard least-squares regression was used to fit a quadratic equation with the Rf's of the thyroglobulin, ferritin, lactic acid dehydrogenase, and BSA as the independent variables and the natural logarithm of their hydrated molecular diameters (17.0, 12.2, 8.16, and 7.1 nm, respectively) as the dependent variables. Calculus (transformation of variables) was then used to transform the HDL distribution from the Rf to the particle diameter scale.10 Specifically, in addition to finding the corresponding diameter for each Rf, it was necessary to multiply the height of the distribution by the Jacobian of the transformation.10 The height of the distribution curve (absorbance) at each diameter value was determined by interpolation for each 0.01-nm increment between 7.2 and 12 nm.10
Total variation (biological plus methodological) was assessed from correlations between repeated HDL subclass measurements in the control groups of three longitudinal clinical trials: Study 1, which included 42 men, aged 30 to 59 years, 20% to 60% over Metropolitan Life Insurance Table ideal weight,15 who were sampled at baseline, 7 months, and one year5 ; Study 2, which included 38 men, aged 25 to 49 years, with a body mass index (weight in kilograms divided by the square of height in meters) between 28 and 34 kg/m2 and 36 premenopausal women, all of whom were sampled 12 months apart6 ; and Study 3, which included 125 men and 12 women with angiographically defined coronary atherosclerosis who received usual physician care and who were sampled at baseline and at 1 year.16
Laboratory measurement error was assessed from frozen aliquots of the 1.20 g/mL>d>1.063 g/mL plasma, which was drawn annually from a 60-year-old man over a 5-year period. These samples were included on approximately every fourth gradient gel run at our laboratory, yielding 288 replications. These were used to calculate the SD for the laboratory measurement error (square root of the average variance for the five frozen aliquots) and coefficient of variation (square root of the average squared coefficient of variations for the five frozen aliquots). The laboratory error includes day-to-day variation in staining and destaining and calibration errors for assigning diameter values to Rf's.
Statistical Methods
Correlations between repeated samples in individuals were
determined by Pearson correlation and by ANOVA for estimating the
reliability of a measurement.17 The effect of the
laboratory measurement error on the correlation coefficient was
assessed by the attenuation coefficient.13 18
Specifically, when HDL is correlated with another variable
y, then the laboratory error in measuring HDL will attenuate
the true correlation and cause its value to be biased toward 0. Assume
that the observed value of HDL (HDLobs) is equal to its
true value (HDLtrue) plus its laboratory measurement error
(ie, HDLobs=HDLtrue+measurement error). Let
2HDLtrue represent the variance for
HDLtrue,
2HDLobs the variance
for HDLobs, and
2error the
laboratory measurement error variance of HDL. Then
2HDLobs=
2HDLtrue+
2error.
Thus, for
Corr (HDLtrue,y)=True Correlation Between HDLtrue and y and
Corr (HDLobs,y)=Observed Correlation Between
HDLobs and y, the attenuation coefficient is
![]() |
![]() |
2HDLtrue,
2HDLobs, and
2error are estimated by their sample moments
(ie, the squared SDs s2HDLtrue,
s2HDLobs, and s2error,
respectively). | Results |
|---|
|
|
|---|
|
The attenuation coefficient is the ratio of the observed correlation to
the true correlation, and the percent reduction in the true correlation
brought about by laboratory error in measuring HDL is
100x(1-attenuation coefficient). For example, Fig 1
shows that at 8.0
nm (within HDL3b), the observed correlation for HDL versus
other variables is expected to be 97% of the true correlation (ie,
laboratory measurement error decreases the correlation between HDL and
other variables by 3%). The figure also suggests that percent
reduction in the correlation coefficient is <15% for all diameter
values between 7.2 and 11.5 nm. In addition, (1) overall, correlations
involving larger HDL3c and HDL3b are the least
affected by measurement error; (2) percent reduction within the
HDL2a range becomes larger with increasing diameter; and
(3) correlations within the HDL2b interval show the least
reduction at
10.5 nm.
The bottom panel displays the coefficient of variation, which compares the SD of the measurement error relative to a sample mean. The coefficient of variation is lowest for HDL2a and HDL2b and becomes progressively larger for HDL3a, HDL3b, and HDL3c.
Variation in Subclass Levels in Persons Sampled Over Time
The top panel of Fig 2
displays the correlation of
HDL subclasses sampled twice or three times over 1 year in three
samples of men. The graph suggests that the replicate samples were
correlated most strongly within HDL2b, somewhat less within
HDL2a, and more weakly within HDL3a,
HDL3b, and HDL3c. The bottom panel presents
the pooled correlation19 for the three studies of men. The
pooled correlations showed that HDL2b was the least
variable, HDL2a somewhat more variable, and
HDL3 subclasses (particularly HDL3c and
HDL3a) the most variable. The bottom panel also
compares the pooled correlation for men with the pooled correlation for
women.20 As in men, repeated samples from women were
significantly correlated within the HDL2b,
HDL2a, and HDL3b intervals. Between
HDL3b and HDL2a, there was an interval of
weaker correlation, presumably of HDL3a, which may be due
to the intrinsic variation in the HDL3a subclass or the
accumulated variation from three overlapping distributions within this
region (ie, HDL3b, HDL3a, and
HDL2a). Plasma HDL2b levels were significantly
more stable in men than women.
|
| Discussion |
|---|
|
|
|---|
HDL subclasses may show different degrees of stability within individuals sampled over time because of differences in genetic and metabolic determinants of subclasses. Between 40% and 60% of plasma HDL cholesterol variability has been attributed to genetic variation,21 22 23 with 25% and 22% attributed to allelic variations in genes that encode hepatic lipase and the apolipoprotein AI/CIII/AIV cluster, respectively.23 The importance of low hepatic lipase levels to the formation of large HDL particles is suggested by studies of mice with inactivated hepatic lipase genes,24 transgenic rabbits with overexpressed hepatic lipase,25 and patients with hepatic lipase deficiency.26 We have reported elsewhere that among all subfractions, HDL2b exhibits the strongest correlation between parents and offspring and that the offspring's HDL2b levels are more strongly correlated with their father's than their mother's HDL2b level.12 This could explain the consistency of HDL2b levels over repeated samples in men (the genetic or enduring effects of family environment remaining constant) and the stronger consistency of the HDL2b measurement in men than women (the inheritance of HDL2b being stronger in men than women).12 It is also true that a better estimate of the paternal than the maternal phenotype may account for the stronger correlation of the offspring's HDL2b level with their father's rather than their mother's.12 Our previous analyses also showed that plasma HDL3b levels were correlated among siblings and between parents and offspring, suggesting that inheritance (genetic or family environment) may also contribute to the stability of HDL3b over time.12
Subclasses with a more rapid turnover may be more sensitive to perturbations than are subclasses that have a longer residence time in plasma. For particles that contain both apoA-I and apoA-II (HDL3b, HDL3a, and HDL2a)27 and particles that lack apoA-II (HDL3c, HDL3a, and HDL2b),28 the HDL3 species are catabolized more rapidly than are the HDL2. This could explain the greater stability of HDL2b vis-á-vis HDL3c and HDL3a and the greater stability of HDL2a vis-á-vis HDL3b and HDL3a. Differences in HDL2a and HDL2b stability could also reflect differing degrees of interactions with other lipoproteins or lipases. For example, when incubated with hepatic lipase, lipolysis of triglycerides and hydrolysis of phosphatidycholine in HDL2 (A-I with A-II) is reported to be substantially greater for HDL2 (A-I with A-II) than HDL2 (A-I without A-II)29 (ie, presumably greater for HDL2a than HDL2b).
The statistical analyses of protein-stained HDL could show
differences between subclasses that are due to laboratory measurement
error and biological variability rather than
physiological differences. Analyses of Figs 1
and 2
show the extent that these two factors could affect statistical
results. Measurement error will bias the correlation coefficient toward
0 (ie, attenuate the correlation so that it is less likely to reach
significance). For example, we have shown that adiposity levels in
cross-sectional samples and changes in weight in longitudinal studies
correlate significantly with HDL2b but not
HDL2a.5 7 8 We have also described significant
correlations among siblings and as noted above between father and
offspring for HDL2b but not HDL2a. However, the
differences in significance between HDL2b and
HDL2a are unlikely to be artifacts of differences in
measurement precision, since the degree of attenuation is at least as
great for HDL2b as for HDL2a (Fig 1
). The power
to detect significant group differences in longitudinal studies is a
function of the within-subject and between-subject variances, which can
be derived from the correlation of two or more measurements in
individuals sampled over time.17 Fig 2
shows that HDL
subclasses exhibit different degrees of variability over time. In men,
correlations between replicate samples over time generally increase
with particle diameter. This results in less statistical power to
detect differences or correlations in HDL3 compared with
HDL2 subclasses. For example, reductions in
HDL3b concentrations during exercise-induced weight
loss5 and increased dietary fat intake30 are
larger in magnitude than are the increases in HDL2b;
however, the significances of the HDL3b decrease is less,
presumably because the variability over time is greater for
HDL3b than for HDL2b.
Measurement error and within-person variability does not affect the probability of a type I error (false-positive) but increases the probability of a type II error, which could lead to misinterpretation. Our analyses suggest that HDL subclasses exhibit only minor differences in laboratory precision but major differences in the stability of the measurements within individuals over time. Thus, observed differences in subclasses are unlikely to arise as artifacts of laboratory imprecision but could reflect biological variability in subclass levels over time.
| Acknowledgments |
|---|
| Footnotes |
|---|
Received March 4, 1996; accepted July 14, 1996.
| References |
|---|
|
|
|---|
2. Wilson HM, Patel JC, Skinner ER. The distribution of high-density lipoproteins subfractions in coronary survivors. Biochem Soc Trans.. 1990;18:1175-1176. [Medline] [Order article via Infotrieve]
3.
Johansson J, Carlson LA, Landou C, Hamsten A.
High density lipoproteins and coronary
atherosclerosis. Arterioscler
Thromb.. 1991;11:174-182.
4. Cheung MC, Brown BG, Wolf AC, Albers JJ. Altered particle size distribution of apolipoprotein A-I-containing lipoproteins in subjects with coronary artery disease. J Lipid Res.. 1991;32:383-394. [Abstract]
5. Williams PT, Krauss RM, Vranizan KM, Albers JJ, Wood PDS. Effects of weight loss by exercise and by diet on apolipoprotein A-I and A-II and the particle-size distribution of high-density lipoproteins in men. Metabolism.. 1992;41:441-449. [Medline] [Order article via Infotrieve]
6. Williams PT, Krauss RM, Stefanick ML, Vranizan KM, Wood PD. Effects of low-fat diet, calorie restriction, and running on lipoprotein subfraction concentrations in moderately overweight men. Metabolism.. 1994;43:655-663. [Medline] [Order article via Infotrieve]
7. Williams PT, Haskell WL, Vranizan KM, Krauss RM. The associations of high-density lipoprotein subclasses with insulin and glucose levels, physical activity, resting heart rate, and regional adiposity in men with coronary artery disease: the Stanford Coronary Risk Intervention Project (SCRIP) baseline survey. Metabolism.. 1995;44:106-114. [Medline] [Order article via Infotrieve]
8.
Williams PT, Vranizan KM, Austin MA, Krauss RM.
Associations of age, adiposity, alcohol intake, menstrual status
and estrogen therapy with high-density lipoprotein subclasses.
Arterioscler Thromb.. 1993;13:1654-1661.
9. Williams PT, Austin MA, Krauss RM. Variations in high-density lipoprotein subclasses during the menstrual cycle. Metabolism.. 1994;43:1438-1441. [Medline] [Order article via Infotrieve]
10. Williams PT, Krauss RM, Nichols AV, Vranizan KM, Wood PDS. Identifying the predominant peak diameter of high-density (HDL) and low-density (LDL) lipoproteins by electrophoresis. J Lipid Res.. 1990;31:1131-1139. [Abstract]
11.
Williams PT, Krauss RM, Vranizan KM, Stefanick ML, Wood
PDS, Lindgren FT. Associations of lipoproteins and
apolipoproteins with gradient gel electrophoresis estimates of
high-density lipoprotein subfractions in men and women.
Arterioscler Thromb.. 1992;12:332-340.
12.
Williams PT, Vranizan KM, Austin MA, Krauss RM.
Familial correlations of HDL-subclasses based on gradient gel
electrophoresis. Arterioscler Thromb.. 1992;12:1467-1474.
13. Fuller WA. Measurement Error Models. New York, NY: John Wiley and Sons; 1987:4.
14. Nichols AV, Krauss RM, Musliner TA. Nondenaturing polyacrylamide gradient gel electrophoresis. Methods Enzymol.. 1986;128:417-431. [Medline] [Order article via Infotrieve]
15. New weight standards for men and women. Stat Bull Metro Life Insur Co.. 1959;40:1-4.
16.
Haskell WL, Alderman EL, Fair JM, Maron DJ, Mackey SF,
Superko HR, Williams PT, Johnstone IM, Champagn MA, Krauss RM, Farquhar
JW. The effects of intensive multiple risk factor reduction on
coronary atherosclerosis and clinical cardiac
events in men and women with coronary artery disease: the
Stanford Coronary Risk Intervention Project
(SCRIP). Circulation.. 1994;89:975-990.
17. Winer BJ, Brown DR, Michels KM. Statistical Principles in Experimental Design, 3rd edition. New York, NY: McGraw-Hill Book Co Inc; 1991:1011-1021.
18. Lord FM, Novick MR. Statistical Theories of Mental Test Scores. New York, NY: Addison-Wesley Publishing Co; 1968:55-57.
19. Snidecor GW, Cochran WG. Statistical Methods, 6th ed. Ames, Iowa: Iowa State University Press; 1972:186.
20. Wood PD, Stefanick ML, Williams PT, Haskell WL. The effects on plasma lipoproteins of a prudent weight-reducing diet, with or without exercise, in overweight men and women. N Engl J Med.. 1991;325:461-466. [Abstract]
21.
Heller DA, De Faire U, Pedersen NL, Dahlen G, McClearn
GE. Genetic and environmental influences on serum lipid levels
in twins. N Engl J Med.. 1993;328:1150-1156.
22. Steinmetz J, Boerwinkle E, Gueguen R, Visvikis S, Henny J, Siest G. Multivariate genetic analysis of high density lipoprotein particles. Atherosclerosis.. 1992;92:219-227. [Medline] [Order article via Infotrieve]
23. Cohen JC, Wang Z, Grundy SM, Stoesz MR, Guerra R. Variation at the hepatic lipase and apolipoprotein AI/CIII/AIV loci is a major cause of genetically determined variation in plasma HDL cholesterol levels. J Clin Invest.. 1994;94:2377-2384.
24.
Homanics GE, de Silva HV, Osada J, Zhang SH, Wong
H, Borensztajn J, Maeda N. Mild dyslipidemia in mice
following targeted inactivation of the hepatic lipase gene.
J Biol Chem.. 1995;270:2974-2980.
25.
Fan J, Wang J, Bensadoun A, Lauer SJ, Dang Q, Mahley
RW, Taylor JM. Overexpression of hepatic lipase in transgenic
rabbits leads to a marked reduction of plasma high density lipoproteins
and intermediate density lipoproteins. Proc Natl Acad Sci
U S A.. 1994;91:8724-8728.
26.
Hegele RA, Little JA, Vezina C, Maguire GF, Tu L,
Wolever TS, Jenkins DJ, Connelly PW. Hepatic lipase deficiency:
clinical, biochemical, and molecular genetic characteristics.
Arterioscler Thromb.. 1993;13:720-728.
27. Nichols AV, Blanche PJ, Shore VG, Gong EL. Conversion of apolipoprotein-specific high-density lipoprotein populations during incubation of human plasma. Biochim Biophys Acta. 1989;1001:325-337. [Medline] [Order article via Infotrieve]
28.
Cheung MC, Albers JJ. Characterization of
lipoprotein particles isolated by immunoaffinity
chromatography: particles containing A-I and A-II and
particles containing A-I but no A-II. J Biol
Chem.. 1984;259:12201-12209.
29. Mowri H-O, Patsch W, Smith LC, Gotto AM Jr, Patsch JR. Different reactivities of high density lipoprotein2 subfractions with hepatic lipase. J Lipid Res.. 1992;33:1269-1279.[Abstract]
30.
Williams PT, Dreon DM, Krauss RM. The effects of
dietary fat on high-density lipoprotein subclasses are influenced by
both apolipoprotein E isoforms and low-density lipoprotein subclass
pattern. Am J Clin Nutr.. 1995;61:1234-1240.
This article has been cited by other articles:
![]() |
V. Y. Fujimoto, J. P. Kane, B. Y. Ishida, M. S. Bloom, and R. W. Browne High-density lipoprotein metabolism and the human embryo Hum. Reprod. Update, August 28, 2009; (2009) dmp029v2. [Abstract] [Full Text] [PDF] |
||||
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
ATVB Home | Subscriptions | Archives | Feedback | Authors | Help | AHA Journals Home | Search Copyright © 1997 American Heart Association, Inc. All rights reserved. Unauthorized use prohibited. |