Relationship of dietary fat, protein, cholesterol, and ...

Relationship of dietary fat, protein, and fiber intake to atherogenic lipoproteins in men13 Paul

T Williams,

Darlene

PhD,

M Dreon,

RD.

Ronald MS,

M Krauss,

MPH,

Karen

MD,

Stacy

cholesterol,

Kindel-Joyce,

M Vranizan,

MA,

RD,

and Peter

MS, D Wood,

DSc

KEY WORDS

Lipoproteins,

cholesterol,

Introduction Studies ofdietary fat and cholesterol intake (1) and their associations with plasma cholesterol concentrations and coronary heart disease (CHD) have received more attention than have experimental studies and cross-cultural comparisons indicating a role of dietaryprotein consumption in the atherosclerotic process. This tendency extends back to the first convincing demonstration of a relationship between diet and CHD when, in 1909, Ignatowski (2) concluded that animal protein was potentially atherogenic after rabbits fed meat, eggs, and animal products developed atherosclerosis. Later, when Anitschkow and Chalatow (3) found that dietary cholesterol induced atherosclerosis in rabbits, Ignatowski’s results were attributed to the effects of cholesterol feeding because the diet he had used had been high in cholesterol as well as animal protein (1). Concomitant increases in plasma-cholesterol levels and CHD in the United States since 1900 and differences in plasma-cholesterol levels and CHD observed cross-culturally (1) 788

Am J Cuin Nuir

l986;44:788-97.

population

studies,

diet, animal

protein,

plant protein

are also assumed to be due to shifts and differences in fat and cholesterol consumption despite changes, often parallel, in the relative proportions of plant and animal protein consumed. Subsequent cross-cultural comparisons (4-6) and metabolic ward studies (7-9) provided further evidence that plasma concentrations of low-density (LDL) and very lowdensity (VLDL) lipoproteins are influenced by diet. Support ofdiet-lipoprotein hypotheses is dampened, however, by the failure of most I From the Stanford Center for Research in Disease Prevention (PTW, SK, DD, KMV, PDW), Stanford University School of Medicine, Stanford, CA and the Donner Laboratory of Medical Physics (PTW, RMK), The University of California, Berkeley, CA. 2Supported by grants HL-24462, HL-18574, and HL30856 from the National Heart, Lung, and Blood Institute. Computer equipment donated by the Macintosh division of Apple Computer, Cupertino, CA. 3Address reprint requests to: Dr Paul Williams, Stanford Center for Research in Disease Prevention, Suite B, 730 Welch Road, Stanford, CA 94305. Received September 23, 1985. Accepted for publication April 22, 1986.

Printed in USA. ©

1986

American

Society for Clinical

Nutrition

Downloaded from www.ajcn.org by guest on July 10, 2011

ABSTRACT Nutritional components (g/l000 kcal)obtained from 3-day diet records are compared to triglyceride, total cholesterol, low-density (LDL), intermediate-density (IDL), and very low-density (VLDL) lipoprotein concentrations of 77 free-living men. Polyunsaturated-fatty acid consumption correlated negatively with concentrations oftriglycerides, total cholesterol, LDL- and VLDL-cholesterol, and total-lipoprotein mass of smaller-LDL particles (S#{176} 0-7), IDL (S#{176} 12-20), and VLDL (Sf0 20-400) in serum and plasma. Animal-protein consumption correlated positively and plant-protein consumption correlated negatively with triglycerides, smaller-LDL mass, VLDL-cholesterol, and VLDL-mass levels. Serum concentrations of smaller-LDL particles were also positively correlated with dietary-cholesterol intake and negatively correlated with crude-fiber consumption. Thus, dietlipoprotein relationships observed cross-culturally and experimentally are further supported when detailed dietary measurements from 3-day diet records and lipoprotein measurements from repeated blood samplings are correlated in free-living men. Am J Gun Nutr l986;44:788-97.

DIET

AND

ATHEROGENIC

lipoprotein diovascular

IN

profiles disease

789

MEN

indicative risk.

of reduced

car-

Methods Subjects Our report focuses on baseline dietary records and lipoprotein measurements of 77 sedentary but otherwise healthy men, 30-55 yr old, who later participated in a I-yr exercise trial (20). The subjects were selected to be free of known cardiovascular disease and abnormalities, acute illness or active chronic systemic disease, and medication use likely to interfere with lipid metabolism. We also required that all participants had resting blood pressure < 160/100 mmHg, body weight < 140% of Metropolitan ideal weight (21), plasma-total cholesterol < 300 mg/dL, and plasma triglyceride concentrations < 500 mg/dL. Laboratory

procedures

Subjects reported to our clinic in the morning having abstained for 12-16 h from all food and from vigorous activity. While the subject remained in a sitting position, venous blood samples were drawn in Vacutainer tubes providing 1.5 mg/dL disodium EDTA and in empty serum tubes. Plasma and serum were prepared from blood within 2 h, and the blood, serum, and plasma were all kept at 4#{176}C. Plasma-lipid and lipoprotein-cholesterol concentrations were determined by the methods of the Lipid Research Clinics (22). The instrument remained standardized accordingto Lipid Research Clinic criteria duringall analyses (22). The concentrations for low- to very low-density lipoprotein subfractions in serum (as total mass) were determined by the Donner Laboratory of Medical Physics, University ofCalifornia at Berkeley, by computer analysis of the results of analytical ultracentrifugation (23). This technique generates a schuieren curve, which describes the serum concentrations oflipoprotein particles with flotation rates (5?) that range from 0 to 400 (Fig 1). One classification scheme (23) identifies particles of Sf00-12 as lowdensity lipoprotein (LDL), Sf#{176} 12-20 as intermediate density lipoprotein (IDL), and Sf#{176} 20-400 as very low-density lipoprotein (VLDL). Traditionally, lipoprotein concentrations are estimated in epidemiologic and clinical studies by the cholesterol content of the entire Sf#{176} 0-20 region (designated LDL-cholesterol) and ofthe Sf#{176} 20-400 region (VLDL-cholesterol). Although four or more major LDL subclasses probably exist within the Sr#{176} 0-12 region (24-26), the more simple division ofLDL particles into smaller LDL ofS? 0-7 and larger LDL of 50 7-12 (Fig 1) often conveniently summarizes relationships ofLDL subclasses to other variables (26). For example, serum concentrations ofsmaller LDLparticles (S#{176} 0-7) generally are lower in women than in men (26), are decreased by aerobic exercise training (27), and correlate positively with heart rate at rest (28) whereas serum concentrations of larger LDL-particles (Sf#{176} 7-12) are usually higher in women than in men and show no significant relationship to either exercise training or to resting heart rate in men. Moreover, serum concentrations of HDL-cholesterol and HDL2 lipoprotein mass correlate negatively with smaller LDL and positively with larger LDL particle concentrations (26).


cross-sectional studies to show even modest correlations between diet and lipoprotein concentrations within free-living populations. These include the cross-sectional comparisons ofthe Lipid Research Clinic populations (10), Japanese living in Japan, Hawaii, and Callfornia (1 1), the Framingham population (12), the Jerusalem Ischemic Heart Disease Project population (13), the Western Electric Study population (14), and others (15, 16). Loworder nutrient-lipoprotein correlations reported in these studies may be due to 1) the inability to assess accurately the usual food intake ofindividuals (17) with 24-h diet recalls (10) or food-frequency questionnaires (11-15), 2) the inability of single plasma-totalcholesterol or lipoprotein-cholesterol measurements to detect diet-lipoprotein relationships, particularly if a nutrient were to affect only specific lipoprotein subfractions or noncholesterol lipoprotein components (18, 19), and 3) the condensation of nutritional information into generalized food categories (eg, total protein without regard to plant or animal origin) (1). To test the hypothesis that detailed dietary assessment and refined lipoprotein measurements would reveal clear relationships between plasma lipoprotein concentrations and protein and other nutrient intakes, we compare nutrient intakes from 3-day diet records with lipid and lipoprotein cholesterol and lipoprotein subfraction concentrations in a cross-sectional sample of 77 free-living men. We describe elsewhere significant concordant relationships in these men between alcohol intake and high-density lipoprotein (HDL)cholesterol (18), HDL3 mass and apolipoproteins (apo) A-I and A-Il concentrations (18), significant inverse relationships between carbohydrate intake and HDL-cholesterol, HDL3 mass, and apo A-I and A-Il (18), and between coffee intake in excess of 2-3 cups/day and elevated apo B, LDL-cholesterol, and total cholesterol (19). This report describes the relationship of diet to total cholesterol and triglyceride levels and to lipoprotein cholesterol and mass concentrations of subfractions within the low to very low-density lipoprotein spectrum. The results that follow show that those men who ate diets low in animal protein and cholesterol and high in plant protein, fiber, and polyunsaturated fatty acids tended to have

LIPOPROTEINS

790

WILLIAMS

ET AL

Spearman’s correlations provide a nonparametric test for significant association, have high efficiency when the data are in fact normal, and are robust to outliers. Scatterplotsmoothing procedures offer considerable flexibility over

was 0.52 ± 0.23 for these men; mean total caloric intake was 2485.6 ± 586. 1 kcal/day. t Protein from unspecified sources was 2.7 ± 1.9 g/l000 kcal.


linear regression since they do not assume a simple straightline fit between nutrient levels and lipoprotein concentrations and may, thus, reveal important nonlinear relationships. The smoothing procedure applied to these data is described by Cleveland (30) and uses one-halfofthe data for fitting each point. Partial correlation coefficients are used to adjust proteinlipoprotein relationships for the potentially confounding effects of age, cigarette smoking, adiposity, and other dietary factors. Because the assumption ofmultivariate normality may not hold for our data, the standard estimates of significance and confidence intervals for partial correlations were verified by permutation test(3l)and bootstrap resampling procedures (32), respectively. First, adjusted nutrient intake and adjusted lipoprotein concentration were obtained from the residuals of separate regression SMALLER LDL LARGER LDL PDL equations that predicted nutrient intake and lipoprotein concentrations from the adjustment variables. To deterI I I I I I I I I #{149}#{149}_j_#{149}_#{149}_#{149}__#{149}J 0 2 4 6 8 10 12 14 16 18 20 mine a nonparametric significance level we randomly permuted, 1000 times, the adjusted nutrient values among FLOTATION RATE the adjusted lipoprotein concentrations to obtain 1000 partial correlation coefficients under the null hypothesis FIG I. The mean distribution of lipoprotein mass by of zero-partial correlation. A two-tailed nonparametric flotation rate for low to intermediate density lipoproteins significance level for testing whether the observed partial in 77 men. Particles of flotation rates of Sr#{176} 20-400 (verylow-density lipoproteins) are not displayed. correlation was different from zero was then obtained by doubling the proportion oftimes that the correlations for the permuted data exceeded the partial correlations of the original (unpermuted) data. Plasma lipid and lipoprotein cholesterol are the average To determine a nonparametric 95% confidence interval of two blood samples drawn at Visits 2 and 3, and lipofor a partial correlation coefficient, we constructed 1000 protein mass measurements were made from a single bootstrap data sets by sampling 77 vector-valued obsample at Visit 3. servations (ie, consisting of nutrient intake, lipoprotein Percent body fat was estimated by hydrostatic weighing, concentrations, and the adjustment variables) with reand maximal oxygen uptake was calculated from a graded placement from the original set of77 observations. Partial treadmill exercisetest(20). correlations were calculated for the 1000 bootstrapped samples, which were then arranged in ascending order. Nutritionau assessment The 25th and 975th largest correlations define a nonparametric 95% confidence interval after correction for bias Three-day food records were completed on consecutive by the percentile method (32). days The starting date was assigned randomly so as to ensure a proportional number of weekdays and weekend days. The starting date occurred(mean ± SD) 1 1 ± 8 days TABLE 1 after Visit 2 and 5 ± 10 days prior to Visit 3. A trained Average daily nutrien t intake for 77 mi ddle-aged nutritionist reviewed the written records and, if unclear, men* verified them with the participants. Records were coded g/l000kcai a using the Nutritional Coding Center (Minneapolis, MN) Nutrient mean ± SD mean ± SD code book and rules. Mean total calorie and nutrient intakes over the 3 days were determined using the computTotal proteint 39.6 ± 6.2 97.3 ± 22.9 erized food-composition tables (version 6 plus) of the Animalprotein 26.8 ± 6.9 66.2 ±21.4 Nutrition Coding Center and an analysis program. The Plant protein 10.2 ± 3.1 24.9 ± 8.8 validity ofthe 3-day record to assess usual food intake has Saturated fatty been reported by St Jean (29). In this report, dietary vanacids 16.2 ± 3.3 40.8 ± 13.6 ables are expressed in g/l000 kcal except for dietary choMonounsaturated lesterol, which is analyzed as mg/l000 kcal. Mean nutrient fatty acids 17.4 ± 3.9 43.7 ± 15.5 intakes for the sample of 77 men appear in Table 1. Polyunsaturated fattyacids 8.1 ± 2.6 20.0 ± 8.0 Statistical calculations Total carbohydrates 95.2 ± 20.2 236.3 ± 74.3 Crudefiber 1.9 ± 0.7 4.5 ± 1.7 The strengths ofthe relationships between lipoprotein Alcohol 8.0 ± 8.4 19.5 ± 20.9 concentrations and nutrient levels are measured by SpearCholesterol 0. 19 ± 0.08 0.46 ± 0.18 man’s rho correlation coefficients (r,) and their forms are * The polyunsaturated to saturated fatty acid (PS) ratio graphically displayed by data-smoothing procedures.

DIET

AND

ATHEROGENIC

LIPOPROTEINS

The 77 men were healthy, sedentary, middle-aged (mean ± SD = 45.8 ± 5.9 yr) mostly nonsmokers (60 nonsmokers, 17 smokers who smoked an average of 15 cigarettes/day), slightly overweight (percent body fat = 21.6 ± 5.6), who ate typical American diets (2486 ± 586 kcal; 40.5 ± 6.7% from fat, 38.1 ±8.l%fromcarbohydrates, 15.8 ±2.5% from protein, and the remaining 5.6 ± 5.9% from alcohol). monounsaturated, acids

Dietary

cholesterol

ofdietary fat intake (g/l000 to lipoprotein concentrations

Spearman’s

Mean

Total cholesterol Triglycerides LDL-cholesterol VLDL-cholesterol Smaller LDL-mass (Sf#{176} 0-7) Larger LDL-mass (Sr#{176} 7-12) IDL mass (S#{176} 12-20) VLDL mass (Sf#{176} 20-400) * Significance levels (p):

Saturated fatty acids

± SD

Monounsaturated fatty

acids 12

214.2

±

30.8

-0.05

-0.

1 19.6 146.0 19.4 226.6 133.6 42.7 101.9 0.05; b p

± ± ± ±

56.7 27.3

0.10 -0.01 0.07 0.09 0.04 -0.07 0.08 0.001.

-0.10 -0.02 -0. 12 0.09 0.09 -0.21 -0.1 1

1 1.5

63.6 ±41.7 ± 18.9 ± 69.3 0.01;

C

p

kcal) and the polyunsaturated (mg/dL)

in 77 sedentary,

correlations

Polyunsaturated fatty

acids

0.29”

P5 ratio -0.19

-0.18 -0.31” 0.18 0.29”

0.10 -0.19 -0.19


The supposition (7, 33) that total-, LDL-, and VLDL-cholesterol concentrations are more strongly related to saturated-fatty acid intake than to polyunsaturated-fatty acid intake is not supported by the findings of Table 2. Our data show that polyunsaturated-fatty acid intake correlates negatively with plasma concentrations of triglycerides, total cholesterol, LDL- and VLDL-cholesterol, and serum mass concentrations ofsmaller LDL, IDL, and VLDL particles. When polyunsaturated-fatty acid intake and lipoprotein concentrations are adjusted for age, percent body fat, and cigarette use, polyunsaturated fatty acids remain significantly correlated (p 0.05) with triglycerides, total cholesterol, LDL- and VLDLcholesterol, and total mass of smaller LDL and IDL. In contrast (Table 2), saturated fat and monounsaturated fat exhibit only weak and nonsignificant correlations with these lipoprotein measurements. Moreover, lipoprotein

Distribution

and fiber

Metabolic ward studies show that total cholesterol and LDL cholesterol are often elevated when dietary cholesterol intake is increased (7-9), but these relationships usually have not been seen cross-sectionally within populations (10-16). Although we also do not find cholesterol intake to correlate significantly with either total cholesterol (r = 0.09) or LDL cholesterol levels (r = 0. 14), there is a significant correlation of dietary cholesterol versus smaller LDL mass (r = 0.30; p 0.01) suggesting that the total-cholesterol and LDLcholesterol measurements may be insensitive to a relationship that is apparently specific to the smaller (Sf#{176} 0-7) LDL particle region. Figure 2a displays the distribution ofsmaller LDL levels versus dietary cholesterol intake and the functional relationship between these variables as revealed by the scatterplot-smoothing procedure of robust locally weighted regression (30). Higher serum concentrations of smaller LDL mass appear to be related linearly to increasing intakes of dietary cholesterol. Serum concentrations of smaller LDL also appear to be related linearly (but in opposite direction) to crude-fiber consumption (Fig 2b). A relationship between plasma-total-cholesterol concentrations and dietary (soluble) fiber intake has been reported by others (34). The association we observe between crude fiber

and

TABLE 2 Lipoprotein distributions and Spearman’s correlations fatty acid to saturated fatty acid (P:S) ratio compared middle-aged men

791

MEN

concentrations are correlated more weakly with the polyunsaturated to saturated fatty acid (P:S) ratio than with grams of polyunsaturated fatty acid per thousand calories consumed.

Results

Polyunsaturated, saturatedfatty

IN

WILLIAMS

792 I

I

I

I

(a)

I

#{149}

r0.30

#{149}

#{149} #{149}

350

#{149}#{149}

300

#{149}#{149}#{149}#{149}#{149}#{149} #{149}#{149}#{149} I

250

200

I

#{149}I.. 1.. I

150

#{149} #{149}

I

I

S S S

#{149}

100-

S I

I

100 DIETARY

LU -J -J

I

I

I

300

CHOLESTEROL

40C-

I

(b)

400 KcaI)

(mg/1000

I

I

#{149}

r-0.29 S

350 S

300

S

S

#{149}#{149}#{149}

#{149}

S

S

S

250

200 #{149}#{149}#{149}

5#{149}

150

S S

S

#{149}5

100

#{149}

S

0

#{149} S

1.0 CRUDE

2.0

3.0

FIBER (g/1000 KcaI)

FIG 2. Robust locally weighted regression serum concentrations of smaller LDL particles etary cholesterol and crude fiber intake.

curves of versus di-

(largely insoluble) and smaller LDL may be a rough approximation ofan effect due to a soluble fiber (soluble fiber was not available from our food composition tables and its relationship to lipoproteins could not be assessed directly). Animalprotein

and plant

protein

Analysis ofour cross-sectional data suggests that there are strong associations between lowand very low-density lipoprotein levels and animal and plant protein consumption that may be obscured when total protein intake is analyzed without regard to its source. Consistent with other studies (10-16), we find that

AL

the proportion of calories from total protein fails to correlate with low- to very low-density lipoprotein levels. Animal protein, however, correlates positively (Table 3) with concentrations oftotal cholesterol, triglycerides, VLDLcholesterol, smaller-LDL mass, and VLDL mass while plant protein appears to be inversely related to triglycerides, smaller LDLmass, and both the cholesterol content and the total mass of VLDL particles. Figure 3 suggests that serum concentrations of smaller LDL increase linearly as the proportion of total calories from animal protein is increased and the proportion from plant protein is decreased. Figure 4 shows that plasma VLDLcholesterol concentrations are generally unrelated to animal protein for low-intake levels and that the positive correlation between VLDL-cholesterol and animal protein is due to their concordant relationship above protein intake in excess of 30 g/1000 kcal. The regression curves for IDL mass, VLDL mass, and triglyceride concentrations plotted against animal protein show similar threshold effects (results not displayed). Since dietary-protein effects on plasma lipoproteins have been attributed historically to concomitant intakes of other nutrients, partial correlations were computed to establish the significance of the protein-lipoprotein associations while controlling for the potentially confounding influences of fat and other nutrients intake. Partial correlations (Table 4) suggest that the positive correlation of animal protein intake with triglycerides, VLDLcholesterol, smaller-LDL mass, and VLDL

TABLE

3

Spearman’s correlations (r,) ofdietary protein intake 1000 kcal) versus lipoprotein concentrations (mg/dL) 77 sedentary, middle-aged men*

Total cholesterol Triglycerides LDL-cholesterol VLDL-cholesterol Smaller LDL mass (S#{176} 0-7) Larger LDL mass (50 7-12) IDL mass (S#{176} 12-20) VLDL mass (50 20-400) S

Significance

0.01.

levels

(p)

are

Total

Animal

protein

protein

0.13 0.19 0.13

0.22a 0.31” 0.21

0.21

0.29”

0.21 0.04 0.15 0.20

0.341) -0.01 0.20 0.23a

coded:

a

(g/ in

Plant protein

-0.18 032b

-0.15 035” 0.07 -0.13 0.05,

b


I

I

200

ET

DIET

AND

ATHEROGENIC

LIPOPROTEINS

IN

bution-free

significance

mutations

of the data

Multivariate

.D

E N.

a 0

ANIMAL I

I

KcaI)

I

I

r

S

-J

I

I

I

(b)

0

(g/1000

I

-0.35

=

S

350

S

LU -J -J

300 S

S

#{149} S

250

-

200

-

S S S

S S S S S S

150

S

#{149}Sj S S

:#{149} #{149} S

-

S

S

S

55 S

S

100

_________________ I

2

I

6 PLANT

I

I

10 PROTEIN

I

(results

on per-

not displayed).

analysis

Whereas the limited nutritional information that is usually gathered in large cross-sectional studies often explains only a small proportion of the plasma-lipoprotein variation within a population, we find that the more precise dietary intake and lipoprotein data from 3-day diet records and repeated lipoprotein measurements may contribute substantially to understanding lipoprotein differences among men. Multiple regression analysis was used to assess the increase in the proportion of the variance explained when four or fewer nutrients are added to regression equations that already include effects due to age, percent body fat, and cigarettes/day. The addition of dietary variables increased the proportion ofthe total variance explained (100 X R2, where R2 is the multiple correlation coefficient) by 23.7% for total cholesterol (from 13.7% explained by age, percent body fat, and cigarettes/day to 37.4% explained when dietary factors are added to the model); by 19.9% for triglycerides (from 7.7%); by 15. 1% for LDL-cholesterol (from 1 1.6%); by 13.5% for VLDL-cholesterol (from 8.0%); by 22.7% for smaller LDL-mass (from

I

14 (g/1000

based

10.9%); by 19.4% for VLDL mass (from 6.3%), and by 14.6% for larger LDL and IDL mass (from 3.4%).

S

S

S

793

levels

18 KcaI)

‘,‘-

-I-

50

-

40

-

-

-

I

r

i S

FIG 3. Robust locally weighted regression serum concentrations ofsmaller LDL particles imal and plant protein consumption.

curves of versus an-

S S

mass and the negative correlations of plant protein consumption with triglyceride and VLDL-cholesterol concentrations are not simply the result of their linear relationships with age, percent body fat, cigarettes/day, dietary cholesterol, saturated fatty acids, polyunsaturated fatty acids, fiber, and sucrose. The significance of the partial correlations is reinforced by the distribution-free confidence intervals based on bootstrap resampling procedures (ie, the confidence interval does not include zero whenever the traditional significance level is