Functions and Transport Mechanisms'

Lipids are insoluble in water and circulate in plasma in association with certain specific proteins called apolipoproteins. The lipoproteins are large, macromolecular complexes of apolipoproteins and lipids in varying proportions. The four classes of specific lipoproteins that circulate in plasma are called chylomicrons, very-low-density lipoproteins (VLDL), low-density lipoproteins (LDL), and high-density lipoproteins (HDL).

The primary function of plasma lipoproteins is lipid transport. The major lipid transported in lipoproteins—triglyceride—is only slightly soluble in water, yet up to several hundred grams must be transported through the blood daily. Hence, transport mechanisms have evolved to permit the packaging of thousands of triglyceride molecules in individual lipoprotein particles, which deliver the transported lipid to specific cells. Fatty acids esterified to glycerol constitute approximately 90% of the mass and about 95% of the potential energy of the triglyceride molecule. Free fatty acids are transported in noncovalent linkage as albumin-fatty acid complexes. This latter mode of transport does not permit the high degree of selective targeting of fatty acids to specific sites that is permitted by transport in lipoproteins, but these two modes of fatty acid transport together provide a more versatile system for bulk movement of a major substrate for energy metabolism.

Cholesterol is the other major lipid transported in lipoproteins. It is not used for energy; it is the precursor of steroid hormones and bile acids and is a structural component of cellular membranes. In higher animals, including all mammals, it is transported mainly in the form of cholesteryl esters, which are synthesized in cells or in the plasma compartment itself. As with triglycerides, the transport of cholesteryl esters in lipoproteins permits specific targeting of cholesterol to tissues that require it for structural purposes or for making its metabolic products.

Two of the lipoprotein classes, chylomicrons and VLDL, are composed primarily of triglyceride.

Chylomicrons transport exogenous (dietary) triglyceride, and VLDLs transport endogenous triglyceride. Chylomicrons are not normally present in postabsorptive plasma after an overnight fast. The VLDLs normally contain 10 to 15% of the plasma total cholesterol. LDLs contain cholesterol as their major component and normally contain most (60 to 70%) of the plasma cholesterol. HDLs are approximately half protein and half lipid and usually contain 20 to 30% of the total plasma cholesterol. Lipoproteins are lighter than the other plasma proteins because of their high lipid content. This characteristic permits both the operational classification and the ultracentrifugal separation of the different classes of lipoproteins.

Each lipoprotein class is heterogeneous in its protein constituents. Nine distinct apolipoproteins have been separated and described. Most investigators group the apolipoproteins into five families (designated apo A, apo B, apo C, apo D, and apo E) on the basis of their chemical, immunologic, and metabolic characteristics. Apo A refers to the apolipoproteins (apo A-I and apo A-II) that are primarily, but not exclusively, found in HDL. A third member of the apo A family, apo A-IV, is a minor component of chylomicrons. Apo B is the major apoprotein of LDL but also comprises about 35% of VLDL protein. There are two forms of apo B: a large form called apo B-100 and found in LDL and a smaller form called apo B-48 and produced mainly in the intestine. Apo C represents a group of apoproteins (apo C-I, apo C-II, apo C-III) that were originally described as major components of VLDL but that are also present as minor components in HDL. Apo D is a minor component of HDL. Apo E is a major component of VLDL and a minor one of HDL. The apolipoproteins serve both structural and functional roles. Some apoproteins are ligands for specific cell surface receptors, e.g., apo B-100 and apo E for the LDL (or apo B/E) receptor (Brown and Goldstein, 1986); others are cofactors for enzymes, e.g., apo C-II is a necessary activating cofactor for lipoprotein lipase. For reviews of lipoprotein structure and metabolism, see Havel (1987) and Stanbury et al. (1983).

Epidemiologic Evidence for CHD

Most major epidemiologic studies have focused on white men, but a few have provided information about women and nonwhites of both sexes.

Total Cholesterol (TC)

TC is used in this chapter as an abbreviation for the total cholesterol in either serum or plasma. TC concentration is usually expressed as milligrams of cholesterol per 100 ml of serum or plasma (mg/dl). TC concentrations in serum are about 2% higher than those measured in corresponding plasma (Folsom et al., 1983). Although this difference should be considered in comparing the results of studies with one another when numbers of subjects are large and small systematic biases might affect the comparison, it does not affect the major results or conclusions of studies discussed in this report in which serum or plasma is used in analyses of cholesterol. Thus, TC is used interchangeably for both serum and plasma total cholesterol.

Until the past decade, TC, rather than lipoprotein cholesterol, was measured in most epidemiologic studies because reliable methods for measuring lipoprotein cholesterol in large numbers of people were not available. Therefore, most data on disease risk are based on TC level.

Variation in Mean TC Among Populations

Meanlevels of TC vary widely among populations. In the Seven Countries Study, investigators studied 16 populations of middle-aged men residing in seven countries: Finland, the Netherlands, Italy, Yugoslavia, Greece, the United States, and Japan (Keys, 1970, 1980b). Examination methods, laboratory procedures, and quality control procedures were standardized. Mean TC for men ages 50 to 54 years varied from 157 mg/dl in a Japanese population to 262 mg/dl in eastern Finland (Keys, 1980b). In the Ni-Hon-San Study, three population-based samples of men of Japanese ancestry were compared. Mean TC levels for men ages 50 to 54 were 182, 219, and 228 mg/dl in Japan, Hawaii, and California, respectively (Nichaman et al., 1975). In the Israel Ischemic Heart Disease Study, the age-adjusted mean TC level in male civil servants age 40 and older varied from 195 mg/dl for those born in Africa to 219 mg/dl for those born in Central Europe (Kahn et al., 1969). Similar differences were found some 15 years later for male and female adolescents and adults in the Jerusalem Lipid Research Clinics Prevalence Study (Halfon et al., 1982a,b).

Other differences among populations have been observed for men in Puerto Rico, Hawaii, and Framingham, Massachusetts (Gordon et al., 1974), and for men and women in London, Naples, Uppsala, and Geneva (Lewis et al., 1978). Some of this evidence is reviewed in the report of a Conference on Health Effects of Blood Lipoproteins (1979). The results of these various studies, particularly the studies of migrants, indicate that the differences in mean TC levels among populations are due largely to environmental factors, principally diet, rather than to constitutional factors.

Large population differences in mean TC levels have also been observed among children and adolescents; the pattern of variation in these means closely parallels that of the adult values, but at lower absolute values (Conference on Blood Lipids in Children, 1983).

Variation in CHD Rates Among Populations

Large differences also exist among populations in the incidence of and mortality from CHD and in the prevalence and severity of atherosclerosis. For example, in the Seven Countries Study, age-standardized, 10-year incidence of first major CHD events (myocardial infarction and coronary death) among men free of CHD at entry varied from 3 in 1,000 on Crete to 107 in 1,000 in eastern Finland (Keys, 1980b). Corresponding figures for 10-year CHD mortality were 0 and 68 in 1,000, respectively. In the Ni-Hon-San Study, relative risks of first major CHD event were 0.46, 1.00, and 1.54 for the cohorts in Japan, Hawaii, and California, respectively (Kagan et al., 1981; Marmot et al., 1975; Robertson et al., 1977). Incidence of first major CHD events among middle-aged men in Framingham was twice that in Puerto Rico and Hawaii (Gordon et al., 1974).

Variation in Atherosclerosis Among Populations

In the International Atherosclerosis Project, the extent of atherosclerosis in the coronary arteries and aortas was measured in 23, 207 autopsied people from 19 populations in 14 countries (McGill, 1968b). The mean percentage of intimal surface with raised lesions varied from 6% in Durban Bantu to 18% in New Orleans whites. Differences among populations were noticeable at ages 15 to 24 and marked at ages 25 to 34. With few exceptions, ranking the populations according to extent of raised lesions corresponded closely to ranking them by CHD mortality rate.

Correlations Between Mean TC and CHD Rates Among Populations

Variation in mean TC levels among populations is highly correlated with variation in CHD incidence and extent of atherosclerosis. The correlation coefficient for median level of TC with age-standardized, 10-year CHD death rates for 16 cohorts in the Seven Countries Study was .82 (Keys, 1980b). The correlations between median TC and national CHD death rates for these seven countries at 0, 5, 10, and 15 years after TC was measured were .86, .90, .93, and .96, respectively (Rose, 1982). In the International Atherosclerosis Project, there was a correlation of .76 between the extent of atherosclerosis and mean TC concentration in 19 populations (Scrimshaw and Guzman, 1968). Populations with mean TC levels less than 180 mg/dl are largely free of atherosclerosis and CHD, whereas those with mean TC levels above 220 mg/dl are characterized by high rates of CHD (Conference on Health Effects of Blood Lipoproteins, 1979). These results support the conclusion that variation in CHD rates among populations is determined predominantly by differences in levels of TC.

CHD Incidence and Mortality Among Individuals Within Populations

In prospective studies of middle-aged men, TC levels above 200-220 mg/dl are positively associated with risk of CHD in the United States (Pooling Project Research Group, 1978; Stamler et al., 1986) as well as in Norway (Holme et al., 1981), France (Ducimetiere et al., 1980), Japan (Johnson et al., 1968), Israel (Goldbourt et al., 1985), England (Rose and Shipley, 1980), Italy (Italian National Research Council, 1982), Finland, the Netherlands, Greece, and Serbia (Keys, 1980b). The association may be weak or absent in some populations with low mean levels of TC and low absolute risk of CHD, e.g., in rural areas of Bosnia and Croatia (Keys, 1980b; Kozarevic et al., 1976, 1981).

Results have been less consistent regarding the association of TC levels below 200 mg/dl with the risk of CHD. In fact, questions have been raised as to whether the association of serum TC with CHD risk is continuous or whether there is some level of serum TC below which it is not related to risk of CHD (e.g., Goldbourt, 1987). In four of the eight studies in the U.S. Pooling Project, age-standardized incidence of CHD for men ages 40 to 64 years was lower in the second quintile of serum TC (194 to 218 mg/dl) than in the first quintile (8194 mg/dl) (Pooling Project Research Group, 1978). In the Israel Ischemic Heart Disease Study of 9,902 male civil servants 40 or more years old, age-standardized 15-year CHD mortality rates according to quintile of serum TC (<176, 176 to 197, 198 to 216, 217 to 241, and >241 mg/dl) were 4.5, 4.9, 4.4, 6.7, and 10.2 per 100, respectively (Goldbourt et al., 1985). In the same study, however, the corresponding 7-year CHD mortality rates previously showed a steadily increasing pattern: 10, 12, 16, 17, and 30 per 1,000, respectively (Yaari et al., 1981). Also, in that study, the 5-year incidence rates for myocardial infarction were 29, 39, and 60 per 1,000 for men in the serum TC tertiles of 77 to 189, 190 to 219, and 220 to 500 mg/dl, respectively (Medalie et al., 1973).

Many other large prospective studies have also shown a clear monotonic association of CHD with TC levels below 200 mg/dl. In the Hiroshima Adult Health Study of 4,256 men age 40 and older, 6-year age-standardized CHD morbidity ratios associated with three levels of TC (<180, 180 to 219, and >220 mg/dl) were 72, 162, and 333, respectively (100 representing risk for the whole group) (Johnson et al., 1968). Ten-year CHD mortality rates among 17,718 British civil servants ages 40 to 64 according to quintile of TC (<159, 159 to 183, 184 to 203, 204 to 233, and >233) were 28, 34, 36, 44, and 54 per 1,000, respectively (Rose and Shipley, 1980). In the Framingham Study, the 20-year CHD incidence for men ages 33 to 49 according to level of TC (114 to 193, 194 to 213, 214 to 230, 231 to 255, and 256 to 514 mg/dl) was 86, 153, 220, 268, and 306 per 1,000, respectively (Kannel and Gordon, 1982). Among 356,222 men ages 35 to 57 who were initially screened in the Multiple Risk Factor Intervention Trial, age-standardized 6-year CHD mortality increased steadily according to decile of TC from 3 per 1,000 for TC <168 mg/dl to 13 per 1,000 for TC >263 mg/dl (Stamler et al., 1986). The data for that trial are shown in Figure 7-1.

FIGURE 7-1

Relationship of serum cholesterol to CHD death in 361,662 men ages 35 to 57 during an average followup of 6 years. From Martin et al. (1986). Each point represents the median value for 5% of the population.

In that trial the 6-year mortality rate doubled between 153 and 226 mg/dl (3.16 to 6.94 per 1,000) and doubled again between 226 and 290 mg/dl (6.94 to 13.05 per 1,000). The weight of evidence supports the idea that TC level, at least from 150 mg/dl upward, is positively associated in a continuous fashion with CHD risk. Because the incidence of CHD is low at TC levels under 200 mg/dl, occasional exceptions to this rule are more likely due to statistical artifact than to biologic diversity. Results from the observations of screenees in the Multiple Risk Factor Intervention Trial also indicated that the association between TC and 5-year risk of CHD death for 23,490 black men was similar to that for 325,384 white men (Neaton et al., 1984).

The association between TC and the relative risk of CHD declines with age. In the pooled results of five U.S. studies, the relative risk of CHD in the highest quintile of TC (>256 mg/dl) compared to the lowest quintile (<194 mg/dl) (defined as the ratio of the risks in these two groups) was 3.6, 1.9, 1.8, and 1.5 for men ages 45 to 49, 50 to 54, 55 to 59, and 60 to 64, respectively (Pooling Project Research Group, 1978). This observation has sometimes been misinterpreted to mean that the level of TC is relatively unimportant in elderly people. The committee believes that this misinterpretation may arise from failure to distinguish between the concepts of relative risk (the ratio between two risks) and attributable risk (the difference between two risks). The former is commonly used to evaluate the magnitude of an epidemiologic association; the latter is commonly used to evaluate its public health importance. In fact, in the set of studies cited above, the attributable risk (calculated as the difference in risk between the highest and lowest quintiles) did not vary consistently with age. Thus, the corresponding attributable risks were 46, 28, 40, and 42 per 1,000 in 8 years for men ages 45 to 49, 50 to 54, 55 to 59, and 60 to 64, respectively.

Atherosclerosis at Autopsy

TC measured by standardized procedures in apparently healthy men was strongly associated with extent of atherosclerosis at autopsy in the Hiroshima Adult Health Study (Rickert et al., 1968), Honolulu Heart Program (Reed et al., 1987; Rhoads et al., 1978), the Framingham Study (Feinleib et al., 1979), the Oslo Study (Holme et al., 1981), and the Puerto Rico Heart Study (Sorlie et al., 1981). There are few data on women.

In the Framingham Study, the extent of coronary atherosclerosis in men was positively correlated with TC measured 1, 5, and 9 years before death; in women, only TC measured 9 years before death but not TC measured 5 and 1 years before death correlated significantly (Feinleib et al., 1979). In the Bogalusa Heart Study, TC measured at 3 to 18 years of age was correlated (R = .67) with percentage of aortic intimal surface involved with fatty streaks as measured during the autopsies of 35 people who died between 7 and 24 years of age (Newman et al., 1986; Strong et al., 1986). Correlations with raised lesions and with lesions in the coronary arteries could not be reliably determined in this series because only 6 people had fibrous plaques and the extent of coronary intimal surface involved in fatty streaks was small, varying only from 0 to 6% (mean, 1%). In a follow-up of the initial report, Freedman et al. (1988) increased the number of cases to 44 and examined the relationship within races. As anticipated, blacks had more extensive fatty streaks than did whites, but there also was a strong positive association between the extent of aortic fatty streaks and the LDL cholesterol concentration within each race group.

In summary, epidemiologic findings among populations and for individuals within populations consistently indicate a strong, continuous, and positive relationship between TC levels and the prevalence and incidence of, as well as mortality from, atherosclerotic CHD. This relationship has been confirmed in autopsy studies.

LDL and HDL Cholesterol

There is less epidemiologic evidence on lipoprotein cholesterol than on TC. Few early studies included measurements of lipoprotein cholesterol because of technical difficulties and cost.

Differences Among Populations

Lipoprotein cholesterol was not measured in the Seven Countries Study; thus, variation among populations is less well characterized for LDL cholesterol (LDL-C) and HDL cholesterol (HDL-C) than for TC. Apparently, average values of LDL-C vary widely among populations, but the evidence is less clear for HDL-C. In the Four European Communities Study, mean LDL-C varied from 116 to 168 mg/dl for men and from 114 to 146 mg/dl for women; in contrast, mean HDL-C varied only from 52 to 55 mg/dl for men and from 60 to 64 mg/dl for women (Lewis et al., 1978). In the 1950s, studies of three groups of sedentary Japanese men matched for fatness showed that mean β-lipoprotein cholesterol (LDL-C) was 120, 183, and 213 mg/dl for those residing in Shime, Honolulu, and Los Angeles, respectively; corresponding mean values for a-lipoprotein cholesterol (HDL-C) were 40, 40, and 35 mg/dl (Keys et al., 1958b). Results from the Jerusalem Lipid Research Clinic Study showed that differences in mean TC according to place of birth (Asia, North Africa, Europe, and Israel) were due mainly to differences in LDL-C (Harlap et al., 1982). On the other hand, a preliminary report of a survey of men in 13 countries indicated that mean HDL-C varied from 27 mg/dl in an African population to 58 mg/dl in eastern Finland (Knuiman and West, 1981). In a study of schoolboys 6 to 7 years old selected from 26 rural and urban populations in 16 countries (Knuiman et al., 1980), mean HDL-C varied from 31 mg/dl in rural Nigeria to 65 mg/dl in rural Finland and correlated (.90) with mean TC. The basis for these different results is not clear.

There is no firm evidence regarding the ecological association of mean HDL-C to risk of CHD. In one study, a correlation of -.57 was found between mean HDL-C and CHD mortality for 19 countries (Simons, 1986), but these results were based on nonstandardized measurements of undefined samples of people. Knuiman and West (1981) obtained standardized measurements of HDL-C in a survey of small samples of men in 13 countries and found a correlation of .57 with national CHD mortality in those countries. One of the lowest population mean values for HDL-C was found in Mexico in Tarahumara Indians, who also had very low values of LDL-C and TC (Connor et al., 1978). Thus, the extent to which variation among populations in CHD rates may be affected by (associated with) variations in mean HDL-C levels is uncertain. Clearly, the association is substantially less than that between population CHD rates and mean LDL-C levels.

Differences Among Individuals

Baseline levels of LDL-C were positively associated with risk of CHD in men and women ages 50 to 79 years in the Framingham Study (Gordon et al., 1977), in Israeli male civil servants age 40 and older (Medalie et al., 1973), and in men ages 35 to 57, black and white, separately identified, who were assigned to the Usual-Care Group in the Multiple Risk Factor Intervention Trial (Watkins et al., 1986). Change in LDL-C level has also been related directly to change in CHD risk; among hypercholesterolemic men treated with cholestyramine, reduction of 11% in LDL-C was associated with a decrease of 19% in risk of CHD (Lipid Research Clinics Program, 1984b).

In the Framingham Study, HDL-C was inversely related to incidence of CHD in men and women ages 49 to 82 years (Castelli et al., 1986; Gordon et al., 1977); an increment in HDL-C of 25 mg/dl was associated with nearly a 50% decrement in risk of CHD for men and women after adjustment for age, body mass index (see Chapter 21 for definition), cigarette smoking, systolic blood pressure, and TC (Castelli et al., 1986). Among the hypercholesterolemic men treated with cholestyramine in the Lipid Research Clinics Coronary Primary Prevention Trial, 1 mg/dl increments in HDL-C at baseline and afterward were associated with decrements of 5.5 and 4.4% in risk of CHD, respectively (Gordon et al., 1986). Baseline level of HDL-C was inversely associated with risk of CHD among men assigned to the Usual-Care Group in the Multiple Risk Factor Intervention Trial (Watkins et al., 1986), but the association was weaker than that observed in the Framingham Study or the Lipid Research Clinics Program. These findings of an inverse relationship between HDL-C and CHD have been supported by observations in the Israel Ischemic Heart Disease Study (Goldbourt et al., 1985; Medalie et al., 1973) and in prospective case-comparison studies of frozen serum samples in Oslo (Enger et al., 1979) and in Tromsö, Norway (Miller et al., 1977).

HDL-C was not significantly associated with the 24-year CHD death rate in a cohort of 526 Finnish men ages 36 to 61 (Keys et al., 1984) and was only weakly associated with CHD death rate in a cohort of 284 Minneapolis business and professional men ages 45 to 55 at entry into the study (Keys, 1980a; Keys et al., 1963). HDL-C was weakly and not significantly (inversely) associated with the incidence of CHD after adjustment for age, body mass index, blood pressure, cigarette smoking, and non-HDL cholesterol in the British Regional Heart Study (Pocock et al., 1986). A study of men in the USSR also failed to find an inverse association between HDL-C and CHD (Levy and Klimov, 1987).

In the Oslo Study of men ages 40 to 49 years at entry in 1972 and 1973, antemortem measurements of plasma lipids and other characteristics were available for 129 men for whom there were also postmortem measurements of the extent of atherosclerosis. The percentage of the coronary intimal surface involved with raised lesions was positively correlated with TC (.32) and negatively correlated with HDL-C (-.25) (Holme et al., 1981).

Although changes in HDL-C are related to changes in CHD risk in most populations, the benefit that can be expected from raising HDL as a preventive strategy in itself is not entirely clear. Added benefit, over that derived from lowering LDL, is suggested, however, by the long-term results of the Coronary Drug Project (Canner et al., 1986), by the results of the Cholesterol Lowering Atherosclerosis Study (Blankenhorn et al., 1987), and by those of the Helsinki Heart Study (Frick et al., 1987; Manninen at al., 1988), in which HDL-C was raised and TC, LDL-C, and VLDL-C were lowered.

Among people ages 50 to 79 years in the Framingham Study, the ratio of TC to HDL-C was strongly associated with risk of CHD in men and women. The strength of the association was not significantly improved by adding TC or HDL-C to the equation containing this ratio (Castelli et al., 1983). A similar result was obtained with the ratio of LDL-C to HDL-C for women, but information about the level of LDL-C did add to the strength of the association in men. The authors concluded that ratios can be useful predictors of risk, but warned that they may not always be as informative as the joint use of the two individual figures used to calculate the ratio. It seems reasonable to expect that increasing knowledge about the various classes and subclasses of lipoproteins will lead to improved ability to predict risk of atherosclerotic diseases. Whether these prediction formulas will take the form of ratios or more detailed specifications of lipoprotein levels is uncertain.

Although the ratio of TC or LDL-C to HDL-C may be a very good predictor of CHD risk in the U.S. population, it is probably not the best target for clinical management or therapy. The recent report of the National Cholesterol Education Program (1988) identified the absolute level of LDL-C as the key index for clinical decision-making about cholesterol-lowering therapy and as the specific target for therapy. The authors of the report stated, ''Reliance on a ratio of either total or LDL-cholesterol to HDL-cholesterol as a key factor in decisions regarding treatment is not a practice recommended in this report. Blood pressure and smoking are not combined into a single number because the clinician needs to know both facts separately in order to recommend an intervention. Similarly, HDL-cholesterol and LDL-cholesterol are independent risk factors with different determinants, and combining them into a single number conceals information that may be useful to the clinician" (National Cholesterol Education Program, 1988).

In summary, of the lipoprotein fractions, LDL has the strongest and most consistent relationship to individual and population risk of CHD. HDL has generally been found to be inversely associated with risk of CHD in individuals within a population, but in at least three long-term population studies, this inverse association was not seen (Keys et al., 1984; Levy and Klimov, 1987; Pocock et al., 1986). Population rates of CHD are much more strongly related to average TC and LDL-C values than to HDL-C.

These findings (together with the results of animal experiments and clinical research reviewed later in this chapter) strongly support the conclusions that LDL-C is centrally and causally important in the pathogenic chain leading to atherosclerosis and CHD. Variation in LDL-C levels explains a large part of individual risk within high-risk cultures and explains almost all the differences in CHD rates among populations. HDL-C is associated inversely with individual susceptibility to CHD within populations where there is widespread elevation of LDL-C and TC. Data are inadequate to characterize ratios of LDL to HDL as a major determinant of the atherosclerotic disease burden among populations; however, the ratio provides improved individual prediction, again within high-risk, high-LDL cultures.

Apolipoproteins

Apolipoproteins play key roles in both the structure and function of plasma lipoproteins. Research on the molecular structure, genetic variability, and metabolism of plasma apolipoproteins has progressed rapidly in recent years, particularly with the application of the new techniques of molecular biology. Knowledge about apolipoproteins has added greatly to our understanding of lipoprotein metabolism and how it is related to atherosclerosis. In a number of instances, genetically controlled variations in apolipoproteins affect lipoprotein structure, composition, and metabolism. For example, polymorphic forms of apolipoprotein E (apo E) interact with dietary fats to influence plasma lipoprotein concentrations, and assessment of apo E phenotypes is an essential procedure in the diagnosis of familial dysbetalipoproteinemia. However, the evidence currently available does not clearly show that plasma apolipoprotein levels are better predictors of CHD than are the plasma levels of cholesterol in the major lipoprotein classes.

Apo E Polymorphism and Hyperlipidemia

In 1977, Utermann et al. (1977, 1979) demonstrated that genetic polymorphisms in apo E were associated with different plasma cholesterol, LDL-C, and β-VLDL concentrations. The apo E phenotypes were shown to be due to segregation of three alleles at a single locus (Zannis and Breslow, 1981) and the major isoforms to be determined by substitution of the amino acid cysteine for arginine (Weisgraber et al., 1981). The three major isoforms are called apo E2, E3, and E4. If the most common phenotype, designated E3/3, is considered to be normal, women who are heterozygous, with phenotype E3/2, have more than 40% higher plasma VLDL-C and intermediate-density-lipoprotein (IDL)-C levels, and 12% lower LDL-C levels (Robertson and Cumming, 1985). Men with the E3/2 phenotype have an approximately 15% increase in plasma VLDL-C and IDL-C concentrations and 20% lower LDL-C concentrations. The E2/2 phenotype is uncommon but is associated with extreme variations in plasma lipoprotein levels. Patients with familial dysbetalipoproteinemia (type 3 hyperlipoproteinemia) generally have the E2/2 phenotype. Most people with this phenotype do not have familial dysbetalipoproteinemia, however, and in fact have lower plasma cholesterol levels than the general population (see Davignon et al., 1988). In contrast, the presence of the E4 allele (and the E4/3 or 4/4 phenotype) appears to be associated with an increased level of LDL and an increase in coronary risk. Thus, a single amino acid substitution in one apolipoprotein can have a substantial effect on plasma cholesterol concentrations and on the plasma lipoprotein profile. This topic has been reviewed thoroughly by Davignon et al. (1988).

The Apolipoproteins and Atherosclerosis

In the 1980s, the availability of better methods for fractionating lipoproteins, for measuring serum apolipoproteins, and for detecting apolipoprotein variants made possible a new series of studies that sought relationships among lipoproteins, apolipoproteins, and atherosclerosis. This topic was reviewed by Brunzell et al. (1984) and by Wallace and Anderson (1987).

Ishikawa et al. (1978) were among the first to examine the relationship of plasma apolipoprotein concentrations to CHD. In the Tromsö Heart Study, apo A-I levels were associated inversely with CHD, but were less powerful predictors than HDL-C levels. A number of investigators subsequently compared CHD patients with controls and generally found apo B levels higher and apo A-I and apo A-II levels lower in diseased subjects (DeBacker et al., 1982; Fager et al., 1980, 1981; Franceschini et al., 1982; Maciejko et al., 1983; Naito, 1985; Onitiri and Jover, 1980; Riesin et al., 1980; Sniderman et al., 1980; Whayne et al., 1981). However, these investigators were not in agreement about whether apolipoprotein levels or lipoprotein cholesterol levels were better indicators of CHD. Pilger et al. (1983) used stepwise discriminant analysis to determine which measures best differentiated subjects with and without peripheral vascular disease. The best model included 14 variables, among which, in addition to HDL-C and LDL-C, were apo B, apo A-I, and apo A-II concentrations.

Menzel et al. (1983) examined the relationship of six different apo E phenotypes to coronary lesions assessed by angiography in 1,000 patients. Heterozygotes (E3/2) were more frequent in patients with little or no coronary atherosclerosis, but E2/2 homozygosity, despite the presence of β-VLDL in the plasma, was not associated with more severe coronary lesions.

Wallace and Anderson (1987) concluded their review by stating that apolipoprotein levels appeared promising as predictors of CHD, but that large cohort studies would be required to determine whether they are better predictors of atherosclerotic disease than plasma lipoprotein lipid levels and whether they are useful clinically.

Lp(a) and Human Atherosclerosis

The role of lipoprotein(a) [Lp(a)] in atherosclerosis and atherosclerotic heart disease has been controversial ever since Lp(a) was discovered by Berg (1963). The apoprotein of Lp(a) has been shown to consist of two peptides—apo(a) and apo B—linked by one or more disulfide bonds (Gaubatz et al., 1983). The recent sequencing of a cloned apo(a) complementary DNA (cDNA) showed that apo(a) is very similar to plasminogen (McLean et al., 1987).

Several studies have shown a relationship between serum Lp(a) levels and CHD (Berg et al., 1974; Dahlén et al., 1975). High scores for coronary arteriosclerosis on angiography were correlated with serum Lp(a) levels (Dahlén et al., 1976). Walton et al. (1974) demonstrated Lp(a) in arterial lesions by immunofluorescence. Overall, the evidence concerning the association of Lp(a) with atherosclerosis is rather scanty, but highly suggestive. Lp(a) levels in plasma are generally believed not to be affected by diet.

Lipoproteins and Apolipoproteins in Atherosclerotic Lesions

Kao and Wissler (1965) demonstrated the presence of LDL in human lesions by immunofluorescence with rabbit antisera to LDL. Hoff and associates have demonstrated the presence of apo B, apo A-I, and apo C in the arteries of humans and several animal species (Hoff and Gaubatz, 1975; Hoff et al., 1974, 1975a,b,c; Yomantas et al., 1984).

The same investigators quantified apo B in lesions (Hoff et al., 1977a,b, 1978, 1979b; reviewed by Hoff and Morton, 1985) and showed that LDL in lesions differed from plasma LDL (Hoff and Gaubatz, 1982; Hoff et al., 1979a). LDL in lesions stimulated the production of cholesteryl esters by macrophages (Clevidence et al., 1984; Goldstein et al., 1981). Apo A-I-containing lipoprotein extracted from arteries also differed in composition from plasma HDL (Heideman and Hoff, 1982).

These studies clearly show that apo B-containing lipoproteins accumulated in atherosclerotic lesions. LDL from atherosclerotic lesions contained particles larger than those found in plasma LDL, was more electronegative, and stimulated cholesterol esterification in mouse peritoneal macrophages. The extent to which these alterations in LDL occurred in the arterial intima, after the LDL had entered the vessel wall, is not clear. This combined evidence implicates apo B-containing lipoproteins in the pathogenesis of atherosclerosis.

Apolipoproteins and Dietary Responsiveness

In view of the critical roles of apolipoproteins in lipoprotein metabolism, genetic variants in addition to those of apo E probably influence the lipemic responses of individuals to dietary fats and thereby affect the risk of CHD. This research area is quite active, and new findings are likely to be available in the near future.

Summary

Apolipoproteins show promise of helping us to understand the mechanisms linking diet to atherosclerosis and cardiovascular disease, but as yet they have not provided predictive power for atherosclerotic diseases beyond that provided by plasma lipid and lipoprotein cholesterol concentrations. They are not now useful in assessing the relationship of dietary fats to atherogenesis or atherosclerotic cardiovascular diseases on a population-wide basis.

VLDL CHOLESTEROL (VLDL-C) AND TRIGLYCERIDES

Differences Among Populations

Fasting plasma triglyceride or VLDL-C levels have not been systematically ascertained in standardized, population-based surveys designed to compare cultures. Nonfasting triglyceride levels were measured in the Framingham, Honolulu, and Puerto Rico studies. Results indicated that relatively high mean nonfasting triglyceride levels are compatible with low-population CHD rates when the mean TC is also low (Gordon et al., 1974). However, in most populations found to have low average TC levels, triglyceride levels are also low (Conference on Health Effects of Blood Lipoproteins, 1979). The correlation was small (R = .26) and not statistically significant between 1980 CHD death rates in 19 countries and mean triglyceride levels measured in undefined population samples in those countries during the 1970s (Simons, 1986).

Differences Among Individuals

A fairly large number of epidemiologic studies have explored relationships between plasma triglyceride levels and the risk of CHD. The studies conducted before 1980 were reviewed in full by Hulley et al. (1980). In almost all reported studies, a significant positive association between triglyceride levels and CHD was found upon univariate analysis. However, triglyceride has not held up as an independent risk factor in most of the studies that have provided multivariate adjustment for a full set of major risk variables. It is well established that a reciprocal relationship generally exists, both within populations and within individuals, between plasma triglyceride (and, therefore, VLDL) and HDL levels. Thus, individuals with high triglyceride levels tend to have lower HDL levels, and an increase or decrease in triglyceride level in a patient is generally accompanied by an opposite change in HDL-C level. On the other hand, our understanding of the association of VLDL and VLDL remnants to atherosclerosis is still incomplete, and there is some evidence that the latter particles may have an important role in atherogenesis. Hence, the answer to the question of whether or not elevated triglyceride (and VLDL) levels are causally linked to atherosclerotic disease in some patients, or groups of people, is not clear. What is clear is that, as a group, people with high triglyceride levels have elevated coronary risk. Overall, however, the potential atherogenicity of the triglyceride-rich lipoproteins has not been determined. (For reviews and reports of recent studies, see Aberg et al., 1985; Carlson and Bottiger, 1981; Hulley et al., 1980; NIH, 1984; Wallace and Anderson, 1987.)

In the Framingham Heart Study, the plasma triglyceride level was found to be an independent predictor of CHD in women (Castelli, 1986). Several studies have observed that triglycerides (and VLDL) may have independent predictive power for CHD in women and a greater association with peripheral arterial disease than with CHD, but none of the associations found was strong (Aberg et al., 1985; Gordon et al., 1977, 1981b).

The National Institutes of Health (NIH) Consensus Conference on Treatment of Hypertriglyceridemia concluded, "Careful evaluation of existing data indicates that in the presence of normal cholesterol levels mild elevations of plasma triglycerides do not necessarily increase the risk of CVD. When triglycerides are below 250 mg/dl, risk generally does not exceed that of other Americans, and changes in lifestyle are unnecessary beyond those recommended for the general public. The same can be said for many normocholesterolemic individuals with borderline hypertriglyceridemia (defined as triglyceride levels ranging from 250 to 500 mg/dl) who have no risk factors or family history for cardiovascular disease. However, triglyceride levels in the range of 250 to 500 mg/dl can be a marker for secondary disorders or for a subset of patients with genetic forms of hyperlipoproteinemia who are at increased risk and who need specific therapy. Dietary intervention is the primary approach to therapy in these patients, but drugs have a role in selected individuals not responding to dietary management. Finally, the danger of pancreatitis is present in frank hypertriglyceridemia (triglycerides >500 mg/dl), and triglyceride lowering by diet and, if necessary, by drugs is indicated" (NIH, 1984). The Committee on Diet and Health endorses these conclusions.

Summary

The relationship between plasma triglyceride levels and cardiovascular diseases is somewhat controversial and unclear. In most within-population studies, plasma triglyceride levels were positively associated with increased risk for cardiovascular diseases but they were not independently predictive for CHD after statistical adjustment for closely associated -attributes such as TC, HDL-C, hypertension, cigarette smoking, and obesity. As indicated above, however, in the Framingham Heart Study the plasma triglyceride level was found to be an independent predictor of CHD in women. Even so, the plasma total triglyceride level, rather than being a direct cause of atherosclerotic disease, probably reflects the presence of certain atherogenic lipoproteins. It is well known that VLDLs are highly heterogeneous; a number of different kinds of lipoprotein particles are contained within this density class. Many disease entities that elevate triglyceride levels, such as diabetes mellitus, nephrotic syndrome, chronic renal disease, and certain primary hyperlipidemias, carry an increased risk of CHD. In these situations, the high triglyceride level may be a clue to the presence of other lipoprotein abnormalities that are more directly associated with CHD, such as low HDL-C, elevated apo B or LDL-C, or atherogenic remnant triglyceride-rich lipoprotein particles that have not been well defined. Thus, whether or not VLDL and triglycerides are directly involved in the atherogenic process, elevated levels can be helpful in identifying people at increased risk of cardiovascular diseases.

Clinical-Genetic Evidence of CHD

A number of genetic disorders of lipoprotein metabolism have been identified and characterized (Brown and Ginsberg, 1987; Stanbury et al., 1983). The study of these disorders has provided many insights into the structure, metabolism, and regulation of the plasma lipoproteins and apolipoproteins. Several of these disorders are characterized by severe hypercholesterolemia and premature atherosclerosis and CHD. Such disorders include familial hypercholesterolemia (FH), familial combined hyperlipidemia, and familial dysbetalipoproteinemia (type 3 hyperlipoproteinemia).

FH provides strong evidence that a high LDL level per se is a cause of accelerated atherosclerosis and premature CHD. FH is caused by a defect in the gene encoding for the LDL receptor—the cell surface receptor that normally removes LDL from the circulation (Brown and Goldstein, 1986). The defective gene results in the receptor being either absent or nonfunctional. One gene for the LDL receptor normally is inherited from each parent. In the heterozygous form of FH, the patient has one normal and one abnormal gene for LDL receptors, and LDL-C levels are approximately doubled to levels exceeding 200 mg/dl.

Heterozygous FH occurs in the general population at a frequency of about 1 in 500, making FH one of the most common single-gene-determined diseases in humans. FH heterozygotes have TC levels that usually exceed 300 mg/dl and frequently have tendon xanthomas, corneal arcus, premature CHD, and a strong family history of hypercholesterolemia. Approximately 5% of patients with myocardial infarction before age 60 will have heterozygous FH. Affected men often develop CHD in their 30s and 40s; in women with FH, CHD often occurs in their 50s and 60s.

Rarely, about once per million people, individuals inherit two abnormal genes for LDL receptors and hence are homozygous for FH. Homozygous FH patients have TC levels that range from 600 to 1,000 mg/dl, and they usually have planar and tuberous xanthomas as well as tendon xanthomas. Severe and often fatal coronary disease frequently develops while these people are in their teens. One homozygote had an acute myocardial infarction as early as 18 months of age; another died of an acute myocardial infarction at 3 years of age. Very few FH homozygotes survive past age 30 (Goldstein and Brown, 1983).

An animal counterpart of FH has been identified in a strain of rabbits called the Watanabe heritable hyperlipidemic (WHHL) rabbit (Goldstein et al., 1983). These rabbits have very high TC and LDL-C levels and develop severe atherosclerosis that is similar to that seen in humans. This genetic animal model provides further strong evidence of the causal relationship between high LDL-C levels and atherosclerosis. Elevated LDL-C levels due to a defect in the LDL receptor have also been observed in a rhesus monkey family by Scanu et al. (1988).

The research on FH and on the LDL receptor has provided insights into the mechanisms that might be involved in the more common problem of the mild to moderately elevated blood cholesterol levels that are seen in the general population. Potential relationships between LDL receptors, diet, and atherosclerosis have been discussed by Brown and Goldstein (1986). It is likely that the dietary components that raise plasma TC and LDL-C levels (specifically, dietary SFAs and cholesterol, as discussed in detail later in this chapter) act, at least in part, by suppressing hepatic LDL receptor activity. Recent studies (Spady and Dietschy, 1988) in hamsters powerfully demonstrate the role of dietary saturated triglycerides in augmenting the effect of dietary cholesterol in suppressing hepatic LDL receptor activity and elevating plasma LDL-C levels.

Clinical Trials of CHD

Small-Scale Trials (Angiographic End Points)

Evidence has been accumulating and a number of studies are under way to determine the effect of lowering TC by diet, drugs, or ileal bypass surgery on the progression or regression of atherosclerotic lesions of coronary and large peripheral arteries as determined by angiography and ultrasound. As shown angiographically, in the earliest reported randomized controlled trial, patients with intermittent claudication had less angiographic progression and more regression of plaques in the diet- and drug-treated group than did a control group over a 19-month period. This was directly related to change in plasma LDL-C concentration (Duffield et al., 1983). In a recently completed randomized controlled trial in which angiographic techniques were used, patients treated with diet and several TC-lowering drugs had less progression and more actual regression of plaques in natural and bypass graft coronary vessels compared to controls over a 2-year period (Blankenhorn et al., 1987). These results and those of other systematic studies cited below are generally consistent with the conclusion that reducing LDL-C in hyperlipidemic patients reduces the rate of progression (and can induce regression) of atherosclerosis according to the degree of blood lipid reduction (Arntzenius et al., 1985; Brensike et al., 1984; Cohn et al., 1975; Kuo et al., 1979; Nash et al., 1982; Nikkila et al., 1984).

Large-Scale Preventive Trials (CHD End Points)

Nine randomized controlled trials on the primary prevention of CHD have been conducted. Key elements of their designs are summarized in Table 7-1. In five trials, various methods were used to change the plasma lipid concentrations as the only intervention; in two, diet was used; and in three, drugs. The other four studies, in which diet was also used, were multifactorial trials in which effects of change in plasma lipids were confounded with effects of change in cigarette smoking and blood pressure. Only one of the six trials involving change in diet—the Los Angeles Veterans Administration Domiciliary Study—was conducted under double-blind conditions (Dayton et al., 1968). In two studies—the World Health Organization Multifactor Trial (WHO European Collaborative Group, 1983) and the Finnish Mental Hospital Study (Hjermann et al., 1981)—groups of participants were randomized. In the others, individuals were randomized. The number of participants in these studies varied from approximately 800 to nearly 50,000.

TABLE 7-1

Design of Randomized Controlled Trials on Primary Prevention of Coronary Heart Disease.

Results of the trials summarized in Table 7-2 have been listed according to the percentage difference in mean serum cholesterol that was created between the treatment and comparison groups—from 0% in the Göteborg Multifactor Trial (Wilhelmsen et al., 1986) to -15% in the Finnish Mental Hospital Study (Hjermann et al., 1981; Miettinen et al., 1972). (A negative sign here indicates that the experimental group had a lower mean level than the control group.) These results indicate that the magnitude of the difference in risk of CHD varied directly with the magnitude of the difference in mean TC. Studies that achieved little or no difference in mean TC had little or no difference in rate of CHD, whereas studies that achieved larger difference in mean TC experienced larger differences in CHD rates.

TABLE 7-2

Results of Randomized Controlled Trials on Primary Prevention of Coronary Heart Disease.

As indicated in Table 7-2, differences in CHD rates were statistically significant in five of the seven studies that achieved a difference of 8% or more in mean TC. The number of cardiac events in the other two studies—in the entire cohort of men in the Los Angeles Veterans Administration Domiciliary Study and in the sample of women in the Finnish Mental Hospital Study—were relatively small, and the percentage differences in CHD rates, although comparable in magnitude to those that were statistically significant, did not reach the 5% criterion for statistical significance.

Combined with the evidence from epidemiologic, clinical, and laboratory animal studies, this body of evidence establishes unequivocally that lowering serum TC, particularly LDL-C, reduces the incidence of CHD in middle-aged, hypercholesterolemic men. High-risk populations were used in most of these studies for quite practical reasons; the cost and problems of conducting such a trial in the general population would be very large and the results might well be inconclusive. However, in light of the total body of evidence, it seems reasonable to extend this conclusion to the general population of men and women with serum TC of 200 mg/dl or higher. Results of the Los Angeles Veterans Administration Domiciliary Trial support the inference that the conclusion may extend to people over 65 years of age as well, perhaps with a somewhat smaller effect, but the evidence bearing on this important question is still weak.

Most of the trials listed in Tables 7-1 and 7-2 showed an excess of non-CHD deaths in the intervention group as compared to the comparison group. How much of this can be attributed to chance and whether any of it reflects adverse effects of intervention are complex questions that may never be completely resolved. An analysis of 20 randomized controlled clinical trials that involved change in serum TC as the only systematic intervention indicates no statistically significant effect on risk of non-CHD death (Richard Peto, University of Oxford, personal communication, 1987). In the trials shown in Table 7-2, none of the percentage differences in risk of non-CHD death or total mortality was statistically significant except for the WHO Clofibrate Trial. In the final mortality follow-up, however, the Committee of Principal Investigators (1978) of that study found that the excess mortality in the clofibrate-treated group did not continue after the end of treatment, and they could not find a reasonable explanation for the excess mortality that did occur during the period of active treatment. The excess deaths were due to a variety of non-CHD causes and had no apparent association either with the extent of reduction in TC or with duration of treatment with clofibrate. These results do not rule out the possibility of a toxic effect, but seem more consistent with the hypothesis that the excess of deaths in the treated group was the result of random sampling variation.

The possible adverse effects of the drugs used in these trials are not strictly relevant to a report on diet and health and are not discussed further. The safety of dietary changes to lower serum TC is discussed in Chapter 28 of this report. There is no firm evidence to indicate that any of the dietary interventions used in the trials described above increased the risk of non-CHD deaths. In the Los Angeles Veterans Administration Domiciliary Trial, an excess of cancer deaths was observed in the experimental group that ate a diet high in PUFAs (on average, 15% of calories as PUFAs), but the difference was larger in the subgroup that adhered poorly to the prescribed diets than in the subgroup that adhered well (Pearce and Dayton, 1971). Also, results from other clinical trials did not indicate an excess of cancer deaths in groups assigned to a high-PUFA diet (Ederer et al., 1971). These considerations support the conclusion that the excess of cancer deaths observed in the Los Angeles study did not result from the high-PUFA diet.

Rose et al. (1974) analyzed the pooled data from six prospective studies of CHD and found that the mean serum TC of men who subsequently developed colon cancer was significantly lower than expected. The extensive literature that has since developed on this topic has been reviewed by McMichael et al. (1984). In some studies, an inverse association was found over a protracted period; in others, an inverse association was observed during the first few years of follow-up but disappeared during continued follow-up; in yet others, no association was observed between initial serum TC level and subsequent risk of cancer. Some of the results may have reflected a preclinical effect of cancer on serum TC and some may have been due to metabolic characteristics (e.g., increased secretion of bile) of people who maintain low serum TC despite a diet high in SFAs and cholesterol. There is no evidence from studies in humans that low consumption of SFAs and cholesterol are positively correlated with rates of mortality from colon cancer (Liu et al., 1979). This evidence is discussed in more detail later in this chapter and in Chapter 28.

Richard Peto (Oxford University, personal communication, 1987) contends that the primary information to be obtained from the randomized clinical trials is the time required for changes in TC to have an effect on CHD rates. He has obtained information on 18 published and 2 unpublished randomized trials that involved changes in TC as the only systematic intervention, regardless of the mode of intervention (diet or drugs) and regardless of whether they were trials of disease prevention or disease treatment. Analysis of all data from these trials indicates that a 10% reduction in TC was associated with an average reduction of 16% in CHD risk (95% confidence interval of 10.6 to 20.4%) in trials lasting 4 years. Peto also noted that the reduction in risk varied according to the duration of treatment: 11% (±5%) in 13 shorter trials and 21% (±4%) in 7 longer trials, for a 10% reduction in TC. There have been no trials lasting a decade or longer, but results from the observational studies suggest that a reduction of 10% in TC over decades would be associated with a 33% reduction in rate of fatal CHD.

Other Atherosclerotic Cardiovascular Diseases

Peripheral Vascular Diseases (PVD)

There are few systematic data on the relationship of blood lipids and lipoproteins to the prevalence or risk of PVD or large-vessel peripheral arterial disease (PAD). In the Framingham Study, the 26-year cumulative incidence of intermittent claudication showed a concentration of cases in the upper part of the distribution for TC values in men but not in women (i.e., 11.6% in the upper quintile versus 5.3% in the lower) (Kannel and McGee, 1985). The results are, however, inconsistent among studies. For example, serum cholesterol was strongly related in some studies (Bothig et al., 1976; Reunanen et al., 1982) but only weakly or not related in others (Cavallo-Perin et al., 1984; Davignon et al., 1977; Greenhalgh et al., 1971; Hughson et al., 1978; Isacsson, 1972; Sirtori et al., 1974).

Triglycerides (and VLDL) may be more highly related to PAD than either HDL or LDL. In a British in-patient case-control study, there was a suggestion that prebetalipoprotein (i.e., VLDL) and serum triglyceride levels were disproportionately elevated in PVD patients, leading the authors to speculate about differences in the time course or severity of atherosclerosis with this blood lipid pattern, compared to that in coronary disease (Greenhalgh et al., 1971). At any rate, the overlap was considerable, in that a third of their PVD patients also had CHD. Clustering of intermittent claudication with other atherosclerotic diseases was also very strong in the Framingham Study. Overt coronary disease, cerebrovascular disease, or congestive heart failure were found in one out of three cases of intermittent claudication at the time of diagnosis, indicating the strong likelihood of common risk characteristics (Kannel and McGee, 1985).

The linkage in the basic processes leading to manifestations of PVD and CHD is further suggested by the finding of a strong curvilinear relationship between a Framingham risk index involving multiple risk characteristics for coronary disease and the incidence over time of manifest intermittent claudication, primarily due to an excess risk of intermittent claudication in the upper quintile of the coronary index (Kannel and McGee, 1985). A similar predominance of hypertriglyceridemia was observed in older men and in women among a series of Swedish patients with PVD (Leren and Haabrekke, 1971).

Because the underlying pathogenesis of PVD and CHD is atherothrombotic disease, it is likely, but not established, that the same evidence outlined for the relationship of diet to CHD holds for PVD as well. This is also suggested by the epidemiologic evidence about PVD, i.e., the positive relationship of intermittent claudication to the coronary disease risk index in the Framingham population. Finally, although the pathogenesis of atherosclerosis is similar in the different arterial beds, the relative importance of different risk factors in causing the disease process clearly differs among different anatomic sites. This issue is discussed further in Chapter 19.

Stroke

There are few systematic data relating blood lipids and lipoproteins to stroke incidence. A U-shaped relationship between lipids and stroke was suggested in the Chicago Stroke Study, in which a lower stroke risk was found in the central part of the TC distribution (Ostfeld et al., 1974). This was more striking in the 20-year Framingham follow-up of the older population, ages 65 to 74, in which the highest risk of stroke and acute brain infarction was observed at TC levels below 190 and above 295 mg/dl. This U-shaped relationship was stronger for acute brain infarction than for all strokes. In the same series, there was a negative relationship between LDL-C and overall stroke as well as for HDL-C and overall stroke (Kannel and Wolf, 1983). A similar U-shaped relationship was also observed in the Honolulu Heart Program, both for TC and for LDL-C, whereas HDL-C was negatively related and VLDL was positively but not significantly related to stroke risk (D. Reed et al., 1986). The left side of the U-shaped relationship to stroke in the Honolulu Study was due primarily to the inverse relationship between TC and hemorrhagic stroke (Kagan et al., 1980).

In later studies involving improved ability to discriminate between intracerebral hemorrhage and thrombosis by the use of computed tomography (CT) scanning, a clearer epidemiologic picture emerges. For example, in recent data from the Multiple Risk Factor Intervention Trial on more than 360,000 screenees observed for 10 years, there is a clear, positive, monotonic, linear relationship between TC level and death ascribed to cerebral thrombosis (Iso et al., 1988). This explains the right side of the U-shaped distribution of TC to overall stroke risk. A strong linear negative relationship was found between TC level and deaths coded to cerebral hemorrhage, the major explanation for the left side of the U-shaped TC-stroke risk curve. However, the negative relationship between TC and cerebral hemorrhage was limited to people with TC levels under 160 mg/dl and diastolic pressures greater than 90 mm Hg (Iso et al., 1988).

Animal Studies of Atherosclerosis

Hypercholesterolemia has been the common denominator of experimental atherosclerosis in a variety of animal species since shortly after the turn of the century (reviewed by Anitschkow, 1967; Duff, 1935; Jokinen et al., 1985; Katz and Stamler, 1953; Roberts and Straus, 1965; Strong, 1976; Wissler and Vesselinovitch, 1987a,b). In most species, hypercholesterolemia is induced by a diet enriched in cholesterol, SFAs, or both. Some animal species, such as the rat and the dog, are resistant to diet-induced hypercholesterolemia and develop atherosclerotic lesions when hormonal or other manipulations are used in conjunction with diet to induce hypercholesterolemia. When hypercholesterolemia is prolonged over several years, these animals develop all the complications seen in advanced atherosclerosis in humans and, eventually, myocardial infarction and other clinical manifestations of atherosclerosis in humans (Taylor et al., 1963).

Not only has hypercholesterolemia been the sine qua non of experimental atherosclerosis, but the relationship of the specific plasma lipoproteins to atherosclerosis in animals is similar to that in humans. For example, in humans and nonhuman primates, LDL is positively, and HDL is negatively, associated with atherosclerosis. In multiple regression analyses, plasma LDL and HDL concentrations account for as much as 50% of the variance in the extent of atherosclerosis (McGill et al., 1981a, 1985; Rudel, 1980). In some experiments, physical characteristics of LDL, such as molecular size, are also associated with atherosclerosis (Rudel et al., 1985). When diet-induced hypercholesterolemia is combined with other conditions known to augment atherosclerosis in humans, such as hypertension, the experimental lesions also are increased in extent and severity (McGill et al., 1985). Thus, the association of altered plasma lipoprotein concentrations in laboratory animals is not only strong and consistent, but the pattern of the association is similar to that in humans.

Furthermore, many studies have demonstrated that reduction of plasma cholesterol concentrations by withdrawal of cholesterol or fat, or both, from the diet, or by administration of drugs, leads to regression of experimental atherosclerosis (reviewed by Malinow, 1983). These regression studies, which were conducted in nonhuman primates, gave considerable support to the concept that the progression of atherosclerosis in humans can be retarded and possibly that lesions can be reversed by treating hyperlipidemia with diet and drugs.

In Vitro Studies of Atherogenesis

The strong relationship between high levels of plasma LDL and atherosclerosis has provided an important background for in vitro studies dealing with the cellular and molecular mechanisms of atherogenesis. These in vitro studies have sought to determine how LDL and other atherogenic lipoproteins could lead to the formation of cholesteryl ester-filled macrophages (foam cells) and other elements of the atherosclerotic lesion. The results of such studies have, in turn, supported data on humans and animals that suggest a primary etiologic role for LDL and related lipoproteins in the development of atherosclerosis.

A prominent and early feature of atherosclerotic lesions is the foam cell. In order to understand the mechanisms of foam cell formation, investigators have studied the interaction of plasma lipoproteins with various types of cultured macrophages. Native LDL does not lead to the accumulation of cholesteryl ester in many types of macrophages (reviewed in Brown and Goldstein, 1983). This lack of native LDL-induced cholesteryl ester accumulation is due, at least in part, to down-regulation of the LDL receptor on these cells by small amounts of excess cellular cholesterol, thus preventing a large influx of LDL-C. However, modified forms of LDL, such as acetyl-LDL, oxidized LDL, and malondialdehyde-LDL, do lead to massive accumulations of cholesteryl ester in cultured macrophages. These modified forms of LDL enter the cell by a receptor (called the scavenger receptor) that is distinct from the LDL receptor and that is not subject to down-regulation (Brown and Goldstein, 1983, 1986). Thus, cellular influx of these modifled forms of LDL continues at a high rate, leading to significant accumulation of cholesteryl ester.

These in vitro observations have spawned a widespread search for evidence of LDL modification in vivo. Several recent studies have shed light on this topic. Pitas et al. (1983) demonstrated that foam cell macrophages from explants of rabbit aorta atheroma have scavenger receptors capable of binding modified LDL. Kita et al. (1987) and Carew et al. (1987) took advantage of the in vitro finding that the drug probucol, which is an antioxidant transported in LDL, prevents oxidative modification of LDL. These investigators demonstrated that probucol treatment of LDL receptor-deficient WHHL rabbits decreased LDL uptake in foam cell lesions and reduced the rate of development of these lesions independently of plasma cholesterol levels. These results are consistent with the hypothesis that oxidative modification of LDL may be important in the development of foam cell lesions.

Haberland et al. (1988) found immunochemical evidence for the existence of malondialdehyde-LDL in atherosclerotic aortas. Malondialdehyde is a byproduct of arachidonic acid metabolism, which is an active process in the arterial wall. Whether malondialdehyde-LDL is involved in foam cell formation in vivo has yet to be determined.

A potential role for native LDL in foam cell formation has recently been suggested by findings that another cultured macrophage, the J774 macrophage, accumulates large amounts of cholesteryl ester in the presence of native LDL (Tabas et al., 1985). In this case, the LDL receptor is poorly down-regulated by LDL-C because of intracellular shunting of cholesterol away from the regulatory pathway into a hyperactive cholesterol esterification pathway. Since LDL receptors are abundant on foam cells in vivo (see below), indicative of poor LDL receptor down-regulation, this mechanism of foam cell formation may help explain the atherogenic effects of native LDL.

Foam cell formation has also been demonstrated in vitro by lipoproteins other than native and modified forms of LDL. One such lipoprotein, β-VLDL, is a cholesterol-enriched lipoprotein in the plasma of cholesterol-fed animals and in the plasma of humans with the genetic disease familial dysbetalipoproteinemia. β-VLDL causes massive deposition of cholesteryl ester in macrophages in vitro (Brown and Goldstein, 1983; Koo et al., 1986) and enters the macrophage by binding to the same receptor as does native LDL, and this receptor demonstrates poor down-regulation (Koo et al., 1986). Receptors for β-VLDL (and thus for native LDL) are located on foam cells from rabbit aorta atheroma (Pitas et al., 1983). These observations, together with the strong correlation between b-VLDL plasma levels and atherosclerosis in cholesterol-fed animals and dysbetalipoproteinemic humans, indicate that β-VLDL is atherogenic.

β-VLDL has been found only in humans with familial dysbetalipoproteinemia. However, chylomicron remnants, which occur commonly in humans, interact with macrophages as does β-VLDL and cause the accumulation of cholesteryl ester as well as a large amount of triglyceride (Van Lentern et al., 1985). These and other findings have led some investigators to speculate that postprandial lipoproteins, particularly cholesterol-enriched particles such as chylomicron remnants, may be atherogenic in humans.

Foam cell research is an active area of in vitro investigation exploring the relationship between LDL and atherogenesis. However, other areas of study have also focused on this relationship (reviewed in Steinberg, 1983). For instance, LDL, either native or oxidized, causes endothelial cell injury in vitro. Endothelial injury, in turn, has been hypothesized to initiate atherogenesis by causing platelet adherence to the vessel wall and release of growth factors. In addition, LDL has been reported to directly augment platelet aggregation and to stimulate the growth of smooth muscle cells—two important features of the atherosclerotic lesion. It is clear, moreover, that other components of cells and vessel walls, not yet well defined, are also important factors in atherogenesis. The potential roles of growth factors, endothelial injury, and the different cells in arterial walls in the pathogenesis of atherosclerosis have been reviewed by Ross (1986).

In summary, numerous in vitro studies have demonstrated the ability of LDL and related lipoproteins to cause the formation of foam cells and other elements of atherosclerosis. Although there may be other important metabolic, cellular, and vessel wall factors in atherogenesis, in vitro evidence strongly suggests an important causative role for LDL and related lipoproteins.

Summary

There are large interpopulation differences in mean plasma lipid and lipoprotein levels and CHD rates. Average population TC levels also differ between adults and children; levels in children parallel those in adults at lower absolute values. Population correlations between CHD risk and mean TC level are strong but are weak for VLDL, triglyceride, and HDL-C. TC means and CHD rates among migrants change toward those of the adopted country, whether higher or lower than the country of origin.

Individual correlations of TC and LDL-C levels with CHD risk within populations are strong, positive, and continuous. Individual HDL-C level is inversely related to CHD risk in studies in the United States, Norway, and Israel, but not in Finland and the USSR. HDL-C population means are weakly correlated, if at all, with population TC levels and CHD rates.

Randomized clinical trials show a consistent relationship between decreases in TC and LDL-C levels resulting from altered diets and drugs and decreases in CHD incidence; the magnitude of the difference in risk varies directly with the magnitude of the difference in mean TC between the experimental and control groups. Differences in CHD rates between the experimental and control groups were statistically significant in five of the seven trials that achieved a difference of 8% or more in mean TC. The magnitude of the benefit that can be expected from lowering TC and LDL levels can be estimated from observational epidemiologic studies and from clinical trials. The estimates from these different kinds of studies are consistent in projecting that for individuals with TC levels initially in the 250- to 300-mg/dl range, each 1% reduction in TC level reduces CHD rates by approximately 2%. Thus, observational studies and clinical trials indicate that relatively small differences or changes in average TC level have a relatively large population effect or public health impact on disease. In addition, the trials indicate that risk can be reduced in individuals at moderate to high risk and that much of the effect of a lower TC on risk is seen after only a very few years.

Epidemiologic Evidence

Studies of Populations

Before World War II, reports from many parts of the world indicated that the prevalence of atherosclerotic cardiovascular diseases varied according to intake of fat and cholesterol (Rosenthal, 1934a,b,c). In the decade following World War II, further studies provided evidence that populations with low mean intakes of SFAs and cholesterol had low mean levels of serum TC in comparison to populations with diets high in SFAs and cholesterol (Bronte-Stewart et al., 1955; Keys and Anderson, 1954; Scrimshaw et al., 1957). [Only representative publications have been cited here. This large body of literature was reviewed by Katz et al. (1958) and by Keys (1963).] Although consistent with the hypothesis that dietary factors play an important role in determining the distribution of TC in populations, these studies often suffered from inadequate sampling and lack of standardized procedures.

In the Seven Countries Study (Keys, 1970, 1980b), standardized procedures (including 7-day weighed food records, a central laboratory, and chemical analysis of duplicate meals) were used. Median TC levels varied from 156 to 264 mg/dl. The mean percentage of calories from fat varied from 9 to 40% of total calories, with 3 to 22% from SFAs, 3 to 29% from MUFAs, and 3 to 7% from PUFAs. The mean concentration of TC correlated with mean percentages of calories from total fats (R = .67) and from SFAs (R = .87), and with estimated mean serum TC (R = .90) calculated from the formula, 164 + 1.35 (2S - P), where S and P represent the percentage of calories from SFAs and PUFAs, respectively (Keys, 1980b). Neither mean intake of PUFAs nor mean relative body weight was significantly correlated with mean serum TC. Dietary cholesterol was not measured in this study, but mean intake of SFAs probably had a strong correlation with mean intake of cholesterol, and results for SFAs probably apply to cholesterol as well.

The strong correlation between mean serum TC and mean percentage of calories from SFAs has been observed in other studies in which groups have formed the unit of analysis. As noted by Keys (1980b), the correlation coefficient was .94 for participants in the Israeli Ischemic Heart Disease Study (Kahn et al., 1969) when categorized according to region of birth: Israel, Asia, Africa, and eastern, central, and southern Europe. Stallones (1983) reviewed a large array of data from population surveys and found that the correlations between mean percentage calories from fats and mean serum TC was .86 for 76 groups of men and .79 for 27 groups of women. Stallones found the magnitude of these correlations to be remarkable since the studies were conducted with different procedures at different times; measurements of diet and of serum TC were not always on the same people; dietary data were not always presented separately by gender; and serum TC and fat intake were subject to appreciable errors of measurement and are surrogates for the variables of real interest—serum LDL-C, dietary SFAs, and cholesterol.

These observations indicate that 60 to 80% of the variance among populations in mean serum TC can be accounted for on a statistical basis by variation in mean percentage of calories from SFAs. (Dietary cholesterol probably contributes to this result as well, but the evidence from cross-population studies on dietary cholesterol is much less than the evidence on SFAs). Many feeding experiments (reviewed below) have established that isocaloric changes in the lipid composition of the diet—i.e., in the percentage of calories from SFAs with 12 to 16 carbon atoms, the percentage of calories from PUFAs, and the amount of dietary cholesterol—affects the concentration of serum TC. However, these studies focused on the assessment of short-term effects; they did not attempt to determine whether long-term changes in the lipid composition of the diet would change the mean serum TC of populations characterized by a high incidence of CHD to levels (<200 mg/dl) commonly seen in populations with a low incidence of CHD. Evidence relating to this important question must come from a variety of sources, e.g., surveys of migrants, comparisons of socioeconomic classes that differ in access to meat and dairy products, analyses of secular trends, and feeding experiments. Some of this literature is reviewed below. Other reviews of this extensive literature have been done by Katz and Stamler (1953), Katz et al. (1958), Keys (1963), and Stamler (1967, 1979).

South Africa

In a survey of employed, middle-aged men in Cape Town, South Africa, mean serum TC and mean consumption of fats were found to be lowest in Bantus, intermediate in "Cape Coloureds," and highest in "European Whites" (Bronte-Stewart et al., 1955). Mean TC levels were 166, 204, and 234 mg/dl, respectively; means for β-lipoprotein cholesterol were 122, 160, and 194 mg/dl; and means for fat intake were approximately 42, 62, and 88 g/day. (Given the differences among these groups in the sources of fats, it seems reasonable to infer that the groups differed substantially in intake of SFAs and cholesterol.) When each racial group was stratified according to income, differences in mean serum TC were closely paralleled by corresponding differences in mean consumption of fats, and racial groups similar in economic status also had similar intakes of fat and mean serum TC. The authors interpreted this evidence as supporting the hypothesis that the observed differences in serum TC could be accounted for by the observed differences in diet and that invoking genetic factors was unnecessary. This interpretation was supported by the results of feeding experiments in humans (Antonis and Bersohn, 1962b; Bronte-Stewart et al., 1955).

Southern Italy

In the early 1950s, working-class inhabitants of Naples consumed 20% of their calories from fats (Keys et al., 1955). The great bulk of their diet consisted of bread and pasta. Cheese and meat were consumed in small amounts, olive oil was used sparingly, and butter was not used at all. Mean serum TC was 172 mg/dl in a sample of 230 clinically healthy male steel workers, firemen, and clerks 35 to 60 years of age (Keys et al., 1955).

In Boston, a survey was conducted on 189 clinically healthy male factory workers ages 20 to 50 who had lived all their adult lives in the United States but whose parents had been born near Naples. Investigators found that 43% of their calories came from fats (approximately 32% from animal fats) and that their mean serum TC was 239 mg/dl (Miller et al., 1958). It seems unlikely that the difference in mean serum TC between Naples and Boston was due to differences in body fatness; when the Boston sample was stratified according to relative weight, mean serum TC was 223, 243, and 232 mg/dl for men with relative weights less than 100, between 100 and 119, and 120% of standard or greater.

Ferro-Luzzi et al. (1984) partially substituted butter and other foods rich in SFAs and cholesterol for the customary olive oil in the diets of healthy middle-aged men and women who resided in a rural area south of Naples and who habitually consumed the traditional diet. Mean serum TC changed from 214 to 245 mg/dl in men and from 179 to 203 mg/dl in women. These differences were due almost entirely to changes in LDL-C, which were 148 to 182 mg/dl for men and 119 to 142 mg/dl for women.

The Japanese

A survey of Japanese men in Japan, Hawaii, and Los Angeles indicated that mean serum TC was directly correlated with mean percentage of calories from fats, and that the mean serum TC of Japanese men in Los Angeles (about 240 mg/dl) was similar to that of American men generally (Keys et al., 1958b). A feeding experiment conducted in Japanese men indicated that the change in mean serum TC associated with change in lipid composition of the diet was comparable in magnitude to that of white subjects in Minnesota (Keys et al., 1957b).

In the Ni-Hon-San Study, standardized methods were used beginning in 1965 to investigate the three cohorts of middle-aged men of Japanese ancestry in Hiroshima and Nagasaki, in Honolulu, and in San Francisco (Kagan et al., 1974; Kato et al., 1973; Tillotson et al., 1973). In the latter two areas, more than 85% of the participants were Nisei who had been born in the United States to parents who had emigrated from Japan; the remainder were Issei who had migrated to the United States before the Exclusion Act of 1924. Sources of migration for more than 70% of the Issei and Nisei were the areas around Hiroshima and Nagasaki in the southwestern region of Japan. All three cohorts had similar distributions of the A, B, and O blood groups.

Data on diet were obtained by a 24-hour diet recall interview and a self-administered dietary acculturation questionnaire designed to ascertain adherence to a traditional Japanese diet (Tillotson et al., 1973). Information on fatty acid composition of foods in Japan was not available, and a food grouping system was used in order to categorize fats as primarily saturated or primarily unsaturated.

Mean values for serum TC for all cohorts in Japan, Honolulu, and San Francisco were 181, 218, and 228 mg/dl, respectively. Corresponding mean values for total energy intake were 2,164, 2,275, and 2,262 kcal/day; for percentage of calories from fats, 15, 33, and 38; for percentage of calories from saturated fats, 7, 23, and 26; for intake of cholesterol, 464, 545, and 533 mg/day; and for subscapular skinfold thickness, 10, 16, and 16 mm.

Although data on fatty acid composition of the diet were not available for the cohort in Japan, such data have been published for the cohorts in Hawaii and California (Tillotson et al., 1973). These data indicate that SFAs accounted for 54% of saturated fats in both Hawaii and California and that PUFAs accounted for 58% of unsaturated fats in Hawaii and 52% in California. The mean percentages of calories from SFAs and PUFAs in Japan were estimated, based on the assumption that the proportions were the same as those in Hawaii. The age-adjusted data for men ages 45 to 59 in Japan, Hawaii, and California are as follows: mean percentage of calories from SFAs was 3.5, 12.4, and 14.1% and from PUFAS, 5.3, 6.1, and 6.1%; mean intake of cholesterol was 217, 237, and 243 mg/1,000 kcal; and serum TC was 175.9, 219.2, and 226.1 mg/dl, respectively.

The effect of the differences in lipid composition of these diets on mean serum TC can be estimated by applying the equations of Keys et al. (1965c) and of Hegsted et al. (1965) (discussed in the next section of this chapter)—and subsequently modified by Hegsted (1986) for diets with 0 to 400 mg of dietary cholesterol per 1,000 kcal of intake. For men ages 45 to 59, the Keys equation accounts for more than 50% of the observed differences between the three groups. The Hegsted equation yields slightly lower values: 47% of the difference between Hawaii and Japan, 54% of the difference between California and Japan, and 57% of the difference between California and Hawaii. The predictive equations were based on short-term feeding experiments lasting for several weeks, and it is reasonable to expect that effects of long-term exposure to different diets would be greater. Therefore, these results support the concept that the major portion of the differences in mean serum TC between the cohorts in Japan, Hawaii, and California can be accounted for by differences in lipid composition of the diet.

The Yemenite Jews

Toor et al. (1960) studied three groups of people who migrated from Yemen to Israel and had lived in Israel for approximately 5 years, 9 to 10 years, and more than 20 years, respectively. The economic status of the most recent migrants was very poor during 1953 and 1954. A survey showed that their diet was low in calories (mean, 1,750 kcal/day) and contained 12 g of animal protein and 30 g of total fat per day—almost all from vegetable sources. The economic status of the earliest migrants was better. Their mean intakes were: energy, 2,500 kcal/day; animal protein, 29 g/day; and total fat, 65 g/day (28% from animal sources). For men ages 45 to 54, mean values for serum TC from the most recent to the earliest migrants were 158, 180, and 197 mg/dl, respectively. Corresponding values for women of the same age group were 172, 189, and 206 mg/dl. These differences in mean serum TC were due primarily to differences in β-lipoprotein (LDL) cholesterol; differences in the α-lipoprotein (HDL) fraction were small and not statistically significant. Mean weight-to-height ratios [weight in kg x 100/ (height in cm - 100)] were 92, 98, and 96 for men and 95, 94, and 108 for women.

The Irish

Results from the Ireland-Boston Diet-Heart Study (Kushi et al., 1985) indicate that migration without substantial change in lipid composition of the diet is not associated with substantial change in mean serum TC. Comparison of middle-aged Irishmen living in Ireland with their brothers and a sample of unrelated Irishmen living in Boston showed that the age-adjusted means for serum TC were similar—216, 218, and 215 mg/dl, respectively. Corresponding mean values for energy intake were 4,033, 3,099, and 2,946 kcal/day. The proportions of calories from SFAs were 18, 17, and 16% and from PUFAs, 2, 3, and 3%; the intakes of cholesterol were 233, 273, and 240 mg/1,000 kcal. Mean values for a diet score intended to estimate the effect of dietary lipids on serum TC were 60, 59, and 54 when calculated according to the equation of Keys et al. (1965c) and 89, 96, and 84 when calculated according to a modification of the equation of Hegsted et al. (1965). Mean relative weights were 100, 107, and 106% of standard.

These results indicate that the mean serum TC level of populations is strongly correlated with mean intake of SFAs (and probably of cholesterol as well, although data on this point are not adequate). Correlations of TC level are weaker with mean intake of PUFAs and mean relative weight. In short-term feeding experiments in small groups of people, mean TC changes predictably when intakes of SFAs, PUFAs, and cholesterol are changed under controlled conditions. Most of the differences among populations in mean TC can be accounted for by these short-term effects, and it seems reasonable to believe that long-term effects would be greater. Although populations differ in genetic factors, in body fatness, and in other factors affecting TC, it is not necessary to invoke such differences in order to account for the major portion of the differences among populations in mean TC levels. In view of the evidence, it seems reasonable to infer that variation among populations in the lipid composition of the diet, principally in mean intake of SFAs (and cholesterol), is an important factor, perhaps the major factor, in determining differences in mean serum TC.

Studies of Individuals

Cross-sectional studies of individuals within populations have generally shown little or no association between the fatty acid composition of the diet and serum TC. [See reviews of the extensive literature by Stallones (1983) and Stamler (1979).] Initially, this observation was difficult to reconcile with the clear experimental evidence for groups of people that changes in the lipid composition of the diet predictably affect mean LDL-C and TC in the blood and with the evidence from observational studies that mean SFA intake is strongly correlated with mean TC across populations. However, subsequent investigations have shown that these results are not inconsistent and, in fact, are expected.

Levels of LDL-C (and consequently of TC) are determined by a variety of intrinsic and extrinsic factors—the lipid composition of the diet being a key extrinsic factor (Hegsted et al., 1965; Keys et al., 1965a). However, people differ substantially in their TC levels, even when maintained on the same diet under controlled conditions. Jacobs et al. (1979) showed that weak cross-sectional correlations between the individual intake of SFAs and level of TC are a consequence of the large variation in intrinsic TC levels among individuals. Moreover, random error in determining the usual intake of nutrients by individuals biases these already weak correlations further toward zero (Beaton et al., 1979; Keys, 1965; Liu et al., 1978; Rush and Kristal, 1982). Knowledge of hypercholesterolemia can lead people to change their diets, thereby producing a negative cross-sectional correlation between SFA intake and level of TC (Shekelle et al., 1981b). Although the correlations were weak, after adjustment for intraindividual variability Shekelle et al. (1981b) showed that the magnitude of the regression of TC on lipid composition of the diet was consistent with the effect demonstrated in controlled feeding experiments.

In summary, the epidemiologic evidence obtained between and within populations indicates that increased SFA intake is associated with increased plasma TC and that SFA intake is a major extrinsic determinant of TC levels.

Clinical and Metabolic Studies

Effects Of Saturation Of Dietary Fatty Acids On Plasma Cholesterol And Lipoprotein Concentrations

In 1952, Kinsell et al. (1952) studied the effects of formula diets on the responses of patients to various hormones and found that diets high in vegetable fat dramatically lowered TC concentrations. In the same year, Groen et al. (1952) observed that a diet rich in vegetable fat fed to volunteers resulted in much lower TC levels than a diet containing an equal amount of animal fat, regardless of cholesterol intake. Despite the skepticism typical of new findings, these initial observations were soon confirmed (Ahrens et al., 1954, 1957; Beveridge et al., 1955, 1957; Bronte-Stewart et al., 1955; Keys et al., 1957a,b; Kinsell et al., 1953; Malmros and Wigand, 1957). It was established that the serum cholesterol concentration was more responsive to the saturation of dietary fatty acids than to the total amount of fats or to the amount of dietary cholesterol. After many comparisons of different fats and oils, several investigators (Ahrens et al., 1957; Malmros and Wigand, 1957; Keys et al., 1957a) proposed that the presence of PUFAs in vegetable oils was responsible for the hypocholesterolemic effect. The effect was observed principally or exclusively on LDL-C.

Using the results of a series of experiments, Keys et al. (1957b, 1959) developed a formula predicting the change in serum cholesterol concentration from the change in the amounts of SFAs and PUFAs in the diet:

Δ Chol = 2.74 S - 1.31 P,

where ΔChol is the change in serum cholesterol in mg/dl and S and P are the changes in the percentage of calories derived from saturated (S) and polyunsaturated (P) fatty acids. At that time, the prevailing opinion was that dietary cholesterol did not affect the serum cholesterol concentration. Therefore, it was not part of the equation.

After additional experiments with dietary fats and cholesterol, Keys et al. (1965c) modified the equation and incorporated the effect of dietary cholesterol as follows:

ΔChol = 1.35 (2 S - P) + 1.5 Z,

where Z is the difference between the square root of the initial intake of cholesterol and the square root of the subsequent intake of cholesterol, both expressed as mg per 1,000 kcal.

Hegsted et al. (1965) conducted a comprehensive experiment in which they examined the effects of different types of fats (safflower, olive, and coconut oils), amounts of fat (22 or 40% of calories), and amounts of cholesterol. The amounts of dietary fats within these limits did not influence serum cholesterol concentration; the proportion of saturated and unsaturated fatty acids was a major determinant. Dietary cholesterol produced a small but statistically significant effect independently of the amount or type of fat. These investigators also developed regression equations that predicted the responses of plasma cholesterol concentrations to changes in dietary fats. The most practical equation was the following:

ΔChol = 2.16 S - 1.65 P + 6.77 C - 0.53,

where C is a change of 100 mg dietary cholesterol per day (or 100 mg per 2,600 kcal, or approximately 38 mg per 1,000 kcal) and S and P have the same meaning as above. This formula was subsequently modified by Hegsted (1986) to:

ΔChol = 2.16 S - 1.65 P + 0.097 C,

where C is the difference in cholesterol intake between the two diets in mg/1,000 kcal. In these studies, the SFAs alone accounted for 72% of the variation in serum cholesterol concentrations.

Keys and Parlin (1966) compared the predictive power of the equation of Hegsted et al. (1965) with those he and his associates had proposed previously (Keys et al., 1957b, 1959, 1965c) and found them to be quite similar. The principal difference resided in the computation of the dietary cholesterol effect; the Keys equation treated it as a square-root function, whereas the Hegsted equation treated it as a linear function. Both equations predicted approximately 90% of the effect of the dietary fat changes on the observed serum cholesterol concentrations.

Equations of the type developed by Keys, Hegsted, and their associates in the publications cited above, and similar ones devised by other investigators, were intended to indicate an average effect of dietary modifications on plasma cholesterol concentrations in a group of individuals—not to represent precise physiological or metabolic mathematical models. The equations were useful in subsequent studies to compare the results of experiments that varied in proportions of dietary fatty acids and cholesterol. Most such experiments have confirmed the validity of these equations as predictors of group averages of responses to dietary manipulations, but at the same time many investigators have commented on the large individual variability in responses (Keys et al., 1959, 1965b). This individual variability was more prominent in experiments involving dietary cholesterol, but it also appeared in experiments involving types of dietary fats. For example, Grundy and Vega (1988) reviewed the results of four experiments on 46 subjects to determine the effects of PUFAs and SFAs on plasma cholesterol and LDL-C. Six (about 13%) of the 46 subjects did not respond to a change in dietary fatty acid saturation with any substantial change in plasma cholesterol concentration.

Katan et al. (1988) characterized the responsiveness of people to dietary cholesterol by subjecting them to repeated dietary challenges and then compared the responses of these same people to high-SFA and high-PUFA diets. Average egg eaters (who consumed an average of approximately 500 mg of cholesterol per day) showed a high degree of congruence between responsiveness to dietary cholesterol and responsiveness to dietary SFAs (R = .62, p <.01), whereas eaters of large amounts of eggs (approximately 1,130 mg of cholesterol per day) showed a low degree of congruence (R = .15, not significant). There was no ready explanation for the difference between the two groups. It will be important in future research to determine the extent to which the responsiveness to dietary fatty acids is a genetically determined characteristic, whether it is influenced by previous diets, whether it is dependent on an interaction between genetics and diet, and whether it involves the same metabolic processes as does responsiveness to dietary cholesterol.

One of the most convincing demonstrations of the effects of dietary fatty acid saturation was the Lipid Research Clinics Coronary Primary Prevention Trial, in which more than 6,000 hypercholesterolemic men were instructed to adopt a diet lower in cholesterol and higher in the relative proportion of PUFAs to SFAs. The change in dietary SFAs was correlated directly, and the change in PUFA intake was correlated inversely, with lowered plasma cholesterol concentration—mainly LDL-C (Gordon et al., 1982).

Within a few years of the initial observations, the weight of the evidence led to the clear conclusion that the effect was real and consistent, and research efforts thereafter were directed toward identifying the specific fatty acids responsible for the effect, the mechanisms of the effect, and the role of SFAs relative to that of cholesterol in causing hypercholesterolemia in humans. These topics have been reviewed by Ahrens (1957), Goodnight et al. (1982), Grundy (1979, 1987), Jackson et al. (1978), McNamara (1987), and Nestel (1987).

Saturated Fatty Acids

In early experiments concerned with the effects of fatty acid saturation on serum cholesterol concentrations, cocoa butter, which is approximately 35% stearic acid (C18:0), did not elevate serum cholesterol as did most other natural foods rich in SFAs (Ahrens et al., 1957; Erickson et al., 1964). Keys et al. (1965c) found that the predictive equations were more accurate when stearic acid was eliminated from the computation of the percentage of calories derived from SFAs. The SFAs with chain lengths shorter than 12 carbon atoms likewise appeared not to influence serum cholesterol levels, and the equations were more predictive when they were eliminated from consideration. Grande et al. (1970) demonstrated that stearic acid did not affect serum cholesterol concentrations. This finding was reconfirmed by Reiser et al. (1985), who compared the effects of beef fat, which is about 20% stearic acid, with those of coconut oil and safflower oil on serum cholesterol concentrations, and more recently by Bonanome and Grundy (1988), who compared the effects of formula diets high in palmitic, stearic, and oleic acids.

In early experiments, butter and coconut oil were strongly hypercholesterolemic. Because these fats have a high proportion of saturated medium- and short-chain fatty acids (C6 to C12), it was suspected that their effects might be due to these fatty acids. Beveridge et al. (1959) and Hashim et al. (1960) demonstrated that medium-chain triglycerides, which were 99% SFAs with chain lengths from C6 to C12, produced little or no elevations in serum cholesterol concentrations over those produced by corn oil diets. Keys et al. (1965c) found that SFAs with fewer than 12 carbon atoms did not contribute to the predictive accuracy of their equation for estimating serum cholesterol changes due to fatty acid intake, but included lauric acid (C12) in their equation. Hegsted et al. (1965) found that C10 and C12 SFAs did not contribute significantly to the predictive accuracy of their equation for estimating the cholesterol-raising effects of dietary fats. There has been little or no investigation of the metabolic effects of these fatty acids since 1965, and the role of lauric acid (C12) remains uncertain.

The average effects of the various fatty acids commonly encountered in human diets are summarized graphically in Figure 7-2 (Grundy, 1981). This figure compares the predictions of the Keys and Hegsted equations for serum cholesterol concentration changes when the fatty acid content of the diet is changed.

FIGURE 7-2

Responses to changes in dietary fatty acids predicted from the studies of Keys et al. (1965c) and Hegsted et al. (1965). Figure adapted from Grundy (1981).

In conclusion, palmitic (C16) and myristic (C14) acids are the SFAs most effective in elevating serum cholesterol concentrations. Stearic acid (C18:0) and SFAs with chain lengths of 10 or less have no effect. Lauric acid (C12) also raises the serum cholesterol level but the quantitative extent of its effects, compared to those of C14 and C16 acids, requires further investigation.

Monounsaturated Fatty Acids

Early investigators noted that olive oil, which is rich in oleic acid (C 18:1), lowered serum cholesterol concentrations when substituted for SFAs, as did the PUFAs (Ahrens et al., 1957; Bronte-Stewart et al., 1955; Keys 1957, 1958a). Keys et al. (1958a) found that substituting oleic acid for carbohydrates in isocaloric quantities did not affect the serum cholesterol level. Both Keys et al. (1965a) and Hegsted et al. (1965) reported that including MUFAs in regression equations did not increase the predictive value of the equation.

The effects of dietary MUFAs on serum lipoprotein levels were reexamined by Baggio et al. (1988), Becker et al. (1983), Grundy (1986), Grundy et al. (1986, 1988), Mattson and Grundy (1985), and Mensink and Katan (1987), all of whom confirmed that MUFAs did not elevate serum cholesterol concentrations as did SFAs. Furthermore, they found that high-MUFA diets did not lower HDL-C concentrations as did replacement of SFAs by carbohydrates.

The effects of MUFAs on serum lipoprotein levels are important, because olive oil or other oils rich in these fatty acids could be used, along with or as an alternative to carbohydrates, as replacement for SFAs in diets designed to lower serum cholesterol and, particularly, LDL-C levels. If the early and recent findings regarding the effects of MUFAs on LDL-C and HDL-C levels are consistently confirmed, the use of oils rich in these fatty acids makes possible the design of diets that would lower LDL-C levels, maintain HDL-C levels, and be more palatable than very-low-fat diets.

For patients with high serum cholesterol levels, it remains to be determined whether attempts to minimize HDL-C lowering (in the course of dietary treatment to lower LDL-C) by partial substitution of SFAs with MUFAs might be more beneficial than substitution with carbohydrates.

ω-6 PUFAs

The major dietary PUFA with the double bond in the ω-6 position is linoleic acid (C18:2). The proportion of linoleic acid in plant oils, expressed as percent of all fatty acids, ranges from 1 to 2% in coconut and palm kernel oils to 50 to 60% in corn, cottonseed, and soybean oils and to about 75% in safflower oil. Corn oil, cottonseed oil, and safflower oil are the natural sources most frequently used in the dietary experiments in humans cited in this report. The other important PUFA found in plants is linolenic acid (C18:3), which has a double bond in the ω-3 position.

Keys et al. (1957b, 1965a) and Hegsted et al. (1965) showed that isocaloric exchange of PUFAs for carbohydrates or MUFAs lowered the serum cholesterol concentration in men consuming mixed-food diets that provided 10 to 40% of calories from fat. Predictive equations developed by linear regression analysis of data from those experiments, described in the previous section, indicated that the isocaloric exchange of PUFAs for 10% of calories derived from MUFAs or from carbohydrates decreased serum cholesterol by 13 mg/dl (Keys et al., 1965c) or by 16.5 mg/dl (Hegsted et al., 1965) when other factors were held constant. This effect was observed directly in experiments in which PUFAs were exchanged for carbohydrates or for MUFAs with little change in intake of other nutrients.

Although Keys et al. (1957b, 1965a, 1965c) and Hegsted et al. (1965) published results only for serum cholesterol, both measured β-lipoprotein (the term then used for LDL) and stated that the changes observed in that fraction paralleled the changes in serum cholesterol. The observed differences in serum cholesterol were larger than would be accounted for by changes in HDL-C. These results suggested that PUFAs had a specific serum-cholesterol- (particularly, an LDL-C-) lowering effect when substituted isocalorically for carbohydrates as well as when substituted for SFAs.

Between 1959 and 1986, several studies in humans addressed the effects of PUFAs substituted for carbohydrates and suggested that dietary PUFAs did not lower serum TC or LDL-C levels when substituted for carbohydrates (Brussaard et al., 1980, 1982; Grundy et al., 1986; Horlick, 1959, 1960; Kuusi et al., 1985). Some of these experiments involved women, who may not have responded as did men, and the observations were not controlled for possible effects of the menstrual cycle on serum lipids and lipoproteins (Oliver and Boyd, 1953); some involved small numbers of subjects and had limited power; some were confounded by variation in other dietary components, particularly cholesterol; and in one, formula diets were used. These reports have, however, raised the question of whether there is a specific LDL-C-lowering effect of PUFAs when substituted for carbohydrates.

There have been no recent reports of experiments designed to test this question directly and rigorously in people eating mixed-food diets. The early experiments conducted by Keys, Hegsted, and their associates (described above) indicated that there was such an effect in men consuming mixed-food diets. Furthermore, the analysis of effects of dietary changes in the Lipid Research Clinics Coronary Primary Prevention Trial showed an association of increased intake of PUFAs with decreased serum cholesterol concentration (mainly in LDL-C). This association was independent of changes in body weight and intake of SFAs and cholesterol (Gordon et al., 1982).

This issue is not important in the design of practical diets for lowering serum cholesterol levels, because for other reasons, diets containing more than approximately 7% of calories as PUFAs are not recommended. The issue is of interest as part of the overall question of the mechanism by which dietary PUFAs lower LDL-C concentrations. The evidence bearing on this mechanism is discussed below.

A consistent finding in all comparisons of isocaloric diets in which fats of any type were substituted for carbohydrates is that serum triglyceride levels were lowered. These studies have been of short duration. The effect may well be temporary, since populations that habitually consume a low-fat, high-carbohydrate diet are not characterized by a high prevalence of hypertriglyceridemia. The comparative effects of carbohydrates versus fats in both short- and long-term studies are discussed in more detail in Chapter 9.

ω-3 PUFAs

Soon after the hypocholesterolemic effects of PUFAs were discovered in 1952, some investigators found that oils derived from fish and marine mammals were as effective as vegetable oils in lowering serum cholesterol levels in humans (Ahrens et al., 1959; Bronte-Stewart et al., 1955; Keys et al., 1957a). This observation received little attention for nearly 25 years until interest was stimulated by fragmented and poorly documented reports of low rates of CHD among fish-eating Greenland Eskimos. The recent literature on the effects of fish oils on plasma lipids, plasma lipoproteins, and lipoprotein metabolism has been reviewed by Carroll (1986a), Goodnight et al. (1982), Herold and Kinsella (1986), Klasing and Pilch (1986), Leaf and Weber (1988), Nestel (1987), von Schacky (1987), and Carroll and Woodward (1989).

Fish oils consistently reduce serum triglyceride concentrations, particularly in hypertriglyceridemic subjects (Harris et al., 1983, 1984; Illingworth et al., 1984; Nestel et al., 1984; Phillipson et al., 1985; Sanders et al., 1985; Sullivan et al., 1986). This effect appears to result from their inhibition of the synthesis of VLDL-triglycerides, leading to a reduced total secretion of VLDL-triglycerides (Connor, 1986; Nestel et al., 1984). Substitution of ω-3 PUFAs for SFAs in the diet results in a lowering of LDL-C. This response appears to be the result of removing SFAs from the diet. Thus, supplementation of the diet with ω-3 PUFAs without altering SFA intake has been reported not to cause a lowering of LDL-C levels (Rogers et al., 1986). In hypertriglyceridemic patients who are treated with ω-3 PUFAs, the lowering of plasma triglycerides is often associated with an increase in LDL-C levels (Connor, 1986). This response is typical of other triglyceride-lowering agents as well.

Fish oils also affect the hemostatic system (described below in the section on Effects of Dietary Fats and Other Lipids on Hemostasis and Eicosanoid Metabolism and reviewed in depth by Herold and Kinsella, 1986). The effects of the ω-3 PUFAs are currently being investigated intensively, and a considerable amount of new information regarding their effects on lipoprotein metabolism may be anticipated in the near future.

Hydrogenated Fatty Acids

The hydrogenation of vegetable oils with high proportions of unsaturated fatty acids was developed as a process to make the oils more palatable and useful as substitutes for animal fats. This process not only reduces the degree of unsaturation of the oils, but also generates geometric isomers—the trans fatty acids (see section on Trans Fatty Acids below for a discussion of the effects of these isomers). In early experiments shortly after the discovery of the serum-cholesterol-lowering effects of the PUFAs, hydrogenation was found to raise serum cholesterol levels. Malmros and Wigand (1957) found that hydrogenated coconut oil elevated serum cholesterol more than other plant oils rich in PUFAs, but they did not test the effects of hydrogenation alone.

Anderson et al. (1961) reported that hydrogenated corn and safflower oils, compared to their native (naturally occurring) forms, consistently raised serum cholesterol concentrations, but the elevation was less than that predicted by the equations developed by Keys et al. (1957b). They attributed the less-than-predicted effect to the presence of trans fatty acids, which they tested in another experiment. Horlick (1960) observed that hydrogenated corn oil, as compared to a low-fat diet, elevated serum cholesterol in four subjects but not to the same degree as did butter. McOsker et al. (1962) found that the effects of hydrogenated cottonseed and soybean oils did not differ from effects of the native oils on serum cholesterol concentrations.

There is a widespread belief that hydrogenated vegetable oils, as consumed in margarines, shortenings, and salad oils, elevate serum cholesterol levels because of a reduction in the number of double bonds, but no controlled experiments have been conducted since 1962 to test this. The experiments with hydrogenated oils have focused on the trans fatty acids produced in the hydrogenation process and have attempted to eliminate saturation of fatty acids as a variable by using fat blends as controls.

Hydrogenated coconut oil was frequently used in the early experiments cited above. This oil was already rich in the cholesterol-raising SFAs (C12 to C16) and thus may not have been representative of the other vegetable oils. When oils rich in oleic and linoleic acids are hydrogenated, stearic and oleic acids, both of which do not elevate serum cholesterol, are produced; almost no myristic or palmitic acids (the main fatty acids responsible for raising serum cholesterol) are produced because the oils contain very few C14 or C16 unsaturated fatty acids. Furthermore, some PUFAs may remain. For example, analyses of more than 80 different margarines prepared by hydrogenation of vegetable oils showed that the palmitic acid content ranged from 8.5 to 13.2% of all fatty acids, with a mean of approximately 10% (Slover et al., 1985). Palmitic acid constituted 24% of one margarine made from lard. In the vegetable oil-based margarines, the linoleic acid content ranged from 6 to 46% (average, about 27%), and the linolenic acid content ranged from 0.2 to 3.6% (average, about 2%). Each hydrogenated oil preparation must be examined with regard to the spectrum of fatty acids it contains, and hydrogenated fats cannot be classified in general as cholesterol raising. This conclusion, based principally on the application of the predictive equations described above, needs verification by experimentation in humans.

The formation of trans fatty acids during hydrogenation is another matter that confounded many of the experiments with hydrogenated oils. This issue is discussed in the following section.

Trans Fatty Acids

The trans fatty acids are geometric isomers of the unsaturated fatty acids (mostly oleic, linoleic, and linolenic acids) occurring naturally in the fats of ruminants and formed in vegetable oils during hydrogenation. From 2 to 7% of beef fat and butterfat, and from 10 to 30% of margarines, shortenings, and salad oils, are trans fatty acids. Between 1965 and 1983, the daily per-capita consumption was about 8 g/day or 6% of the total fat consumed in the United States. The estimates of daily consumption have been confirmed by analyses of trans fatty acids in the adipose tissue of humans (reviewed by Senti, 1985). The biologic properties of the trans fatty acids have been studied extensively, and the original reports have been reviewed by Applewhite (1981), Beare-Rogers (1983), Emken (1984), Hunter and Applewhite (1986), Kinsella et al. (1981), and Senti (1985). A recent survey of more than 80 samples of commercial margarines produced and sold in the United States showed that trans-unsaturated octadecanoic acids ranged from 10.7 to 30.1% of all fatty acids and that small amounts of tras, trans, trans, cis, and cis, trans isomers of linoleate were present (Slover et al., 1985).

Anderson et al. (1961) compared the effects of native and hydrogenated oils similar in their proportions of SFAs, MUFAS, and PUFAS, but varying in their proportions of trans isomers. The oils with high proportions of trans isomers produced higher serum levels of cholesterol and triglycerides than did the native oils. However, the hydrogenated corn oil contained half as much of the essential fatty acids as did the native corn oil. Beveridge and Connell (1962) compared the effects of margarines, corn oil, and butter in formula diets and found that margarines produced higher serum TC levels than did corn oil diets, but not as high as did butter. Grasso et al. (1962) found that the trans fatty acid composition of a hydrogenated oil lowered serum cholesterol levels as did an oil with the same distribution of SFAs, MUFAs, and PUFAs.

McOsker et al. (1962) reported no difference in serum cholesterol concentrations in six subjects fed hydrogenated oils containing 15 to 20% trans fatty acids. Erickson et al. (1964) compared the effects of partially hydrogenated oils with blends of native oils having a similar distribution of SFAs, MUFAs, and PUFAs, but up to 10% trans isomers, and found no difference in serum cholesterol levels and no interaction with dietary cholesterol. After a series of experiments, de Iongh et al. (1965) also concluded that trans fatty acid isomers had no effect on serum cholesterol concentrations; the levels attained followed the predictive equation of Keys et al. (1965c) and were not influenced by the trans fatty acid content. Mishkel and Spritz (1969) tested the effects of high levels of trans isomers of linoleic acid in six patients and found elevations of serum triglycerides but inconsistent changes in serum cholesterol concentrations.

Mattson et al. (1975) conducted the most thorough and controlled test of trans fatty acid effects in 33 men who consumed formula diets containing similar distributions of SFAs, MUFAs, and PUFAs, but differing to an extreme in trans fatty acid isomers. More than 60% of the MUFAs and approximately 50% of the PUFAs in the contrasting diet were in the trans configuration. There were no differences in serum cholesterol or triglyceride levels after 4 weeks on the diets.

In conclusion, some effects of the trans fatty acid isomers on lipid or lipoprotein metabolism may remain undetected. However, most evidence indicates that these isomers, in the quantities usually consumed in the U.S. diet, do not influence serum cholesterol concentrations. There may be other effects unrelated to lipid and lipoprotein metabolism. These possibilities deserve careful attention and additional investigation.

Peanut Oil

Peanut oil, which is rich in MUFAs and PUFAs, was found to be unusually atherogenic in some experiments in animals (see following section on Animal Studies). Very few tests of the effects of peanut oil on serum lipoproteins have been conducted in humans. In a few experiments in humans (Ahrens et al., 1957; Baudet et al., 1984; Bronte-Stewart et al., 1955), peanut oil resulted in serum cholesterol and VLDL-C, LDL-C, and HDL-C concentrations similar to those produced by other unsaturated vegetable oils.

Fatty Acid Effects on Plasma Lipoproteins

There have been many attempts to determine the mechanisms of action of the SFAs in raising serum cholesterol concentrations or of the PUFAs in lowering them. The unsaturated dietary fatty acids were incorporated into the lipoprotein lipids within 14 days (Spritz and Mishkel, 1969). In studies of humans fed high-PUFA diets, Shepherd et al. (1980a) found decreases in the cholesterol-to-protein ratio in LDL, and Pownall et al. (1980) observed a number of differences in the thermal properties of plasma LDL related to the fatty acid composition of the diet. Kuksis et al. (1982) found many changes in lipoprotein composition and decreased numbers of LDL particles, but no changes in the sizes of the particles, in people consuming a high-PUFA diet. Zanni et al. (1987) also found decreased numbers of LDL and HDL particles related to PUFA intake.

PUFA increased the fractional catabolic rate of LDL apo B (Shepherd et al., 1980a) and in some studies decreased the production of LDL (Cortese et al., 1983; Turner et al., 1981). However, these changes were small and did not occur consistently.

As yet, the only evidence regarding the type of effects exerted by fats on LDL receptor activity are the result of experiments in animals. Spady and Dietschy (1985) found that receptor-mediated uptake of LDL in hamsters was suppressed about 30% by dietary cholesterol when the animals were consuming PUFAs (safflower oil) and was suppressed about 90% when they were consuming SFAs (coconut oil). In more recent studies in hamsters, Spady and Dietschy (1988) observed that dietary cholesterol suppressed hepatic LDL receptor activity and raised plasma LDL levels in a dose-dependent manner. At each level of cholesterol intake, polyunsaturated triglycerides diminished and saturated triglycerides accentuated the effects of dietary cholesterol. In baboons, hepatic LDL receptor messenger RNA (mRNA) was decreased when the animals were fed SFAs with cholesterol as compared to those fed either MUFAs or PUFAs (Fox et al., 1987). In all these studies, the effects on LDL receptor activity were consistent with the observed fat-induced changes in plasma lipoprotein concentrations.

Plasma triglyceride levels, representing VLDL and other triglyceride-rich lipoproteins, were lower in humans on PUFA diets than on SFA diets. The difference was usually proportionally greater than for plasma cholesterol (Chait et al., 1974).

Initially, plasma HDL levels were believed to be insensitive to dietary factors (Keys et al., 1959). Between 1958 and 1980, 6 of 15 observational studies and experiments suggested an inverse relationship between HDL-C concentration and intake of PUFAs (reviewed by McGill et al., 1981c). PUFAs also decreased the synthesis of apo A-I but did not change its fractional catabolic rate (Shepherd et al., 1978).

Since 1981, a number of experiments in humans have been conducted to examine the effects of PUFAs on HDL. Brussaard et al. (1982), Ehnholm et al. (1984), Jackson et al. (1984), Kuusi et al. (1985), Mattson and Grundy (1985), Schonfeld et al. (1982), Shepherd et al. (1980b), and Vega et al. (1982) found that dietary PUFAs decreased HDL-C, but Baudet et al. (1984), Becker et al. (1983), Kraemer et al. (1982), McNamara et al. (1987), Oh and Miller (1985), and Weisweiler et al. (1985, 1986) reported no change in HDL-C in subjects on similar diets. Both Brussaard et al. (1982) and Kuusi et al. (1985) observed that the decrease in HDL was due primarily to a drop in HDL₂-C, whereas HDL₃-C was unchanged (HDL₂ and HDL₃ are the two major subfractions of HDL). Jones et al. (1987b) found that the reduction in HDL-C associated with an increased intake of PUFAs occurred only when total fat intake was low (20% of calories).

In conclusion, dietary PUFAs probably lower plasma HDL-C concentrations, particularly HDL₂ levels, to some degree, but individual variability in the response probably leads to varying results among experiments involving small numbers of subjects.

Effects of Fat on Cholesterol Absorption, Synthesis, and Excretion

There is little evidence that dietary fat saturation influences cholesterol absorption in humans (Grundy and Ahrens, 1970; Nestel et al., 1975). However, there are several reports of enhanced cholesterol absorption by PUFAs in squirrel monkeys (Tanaka and Portman, 1977). Measurements by sterol balance methods in humans (Grundy and Ahrens, 1970) and by in vitro methods in rat liver (Wilson and Siperstein, 1959) indicated that cholesterol synthesis was not influenced by saturation of dietary fat.

Although there is little support for the concept that fatty acid saturation influences cholesterol absorption or synthesis in humans, some results have shown a transient enhanced sterol excretion when PUFAs were substituted for SFAs in the diet (Connor et al., 1969; Moore et al., 1968; Nestel et al., 1973, 1975; Wood et al., 1966). The initial increase in sterol excretion may account for the reduction in plasma cholesterol resulting from dietary PUFAs. However, if cholesterol absorption and synthesis were not increased, increased sterol excretion would not be expected upon attainment of a new steady state. This may explain the failure to detect consistently enhanced sterol excretion in humans fed PUFAs as reported by Avigan and Steinberg (1965), Shepherd et al. (1980a), and Spritz et al. (1965) and in patients with hypertriglyceridemia (Grundy, 1975) or hypercholesterolemia (Grundy and Ahrens, 1970).

Grundy and Ahrens (1970) suggested that PUFAs cause a shift in cholesterol from plasma to liver. Hepatic cholesterol accumulation upon PUFA feeding has been reported in several animal experiments (Avigan and Steinberg, 1958; Bloomfield, 1964; Tidwell et al., 1962). However, in a group of 12 hypercholesterolemic patients, polyunsaturated fats, compared to saturated fats, decreased the liver cholesterol content (Frantz and Carey, 1961).

In summary, transient increased sterol excretion may explain the decrement in plasma cholesterol among normolipidemic subjects, but not among those with hypercholesterolemia. A plausible, but not firmly established, mechanism is that PUFAs accelerate the transport of lipoprotein cholesterol to the liver and promote higher sterol excretion until a new steady state is achieved. Precisely how PUFAs mediate the decrease in LDL production and LDL plasma concentration is not known. However, effects on LDL receptor activity in the liver are likely to play an important role in the mechanism of the effects of dietary SFAs and PUFAs. SFA intake may decrease LDL receptor activity, thereby leading to an increase in LDL and TC levels.

Summary

Some dietary SFAs (palmitic, myristic, and lauric acids) elevate serum LDL-C and HDL-C concentrations. ω-6 PUFAs lower serum TC and LDL-C levels when substituted for SFAs and are also likely to do so to a small extent when isocalorically exchanged for MUFAs or carbohydrates in mixed-food diets that provide 10 to 40% of calories from fats. ω-6 PUFAs also lower HDL-C levels. MUFAs do not affect serum TC or LDL-C levels when isocalorically substituted for carbohydrates in mixed-food diets, but they do lower serum TC and LDL-C levels, without lowering HDL-C, when substituted for SFAs. The average effect on TC levels brought about by variations in the proportions of the various dietary fatty acids can be predicted largely from equations based on experimental observations. However, the mechanism of the effect is not clearly defined, despite many proposed hypothetical mechanisms. Recent results from experiments in animals suggest that the effects of dietary SFAs and PUFAs on hepatic LDL receptor activity play a role in this phenomenon. There is considerable interindividual variability in response of plasma lipoproteins to dietary fatty acid saturation. This is probably controlled genetically but has not been studied as intensively as has the response to dietary cholesterol.

Animal Studies

After the initial discovery that dietary cholesterol produced hypercholesterolemia and atherosclerosis in rabbits, a variety of species were found also to be sensitive to dietary cholesterol. Fat in the experimental diet was considered simply as a vehicle for the added cholesterol until the effects of SFAs and PUFAs on serum cholesterol levels were discovered in humans in 1952. Most studies of the effects of various types of fats on plasma lipid and lipoprotein levels and on lipoprotein metabolism have been conducted in humans, but a few investigators have examined these effects in animal models.

SFAs Compared To PUFAs

In 1954, Kritchevsky et al. (1954) demonstrated that dietary cholesterol elevated serum cholesterol to a greater extent when fed to rabbits with hydrogenated vegetable oil than when fed with corn oil. This effect was associated with saturation of dietary fatty acids (Kritchevsky et al., 1956).

Similar observations were made in nonhuman primates. SFAs usually produced higher serum cholesterol levels than did PUFAs when fed with low or moderate levels of dietary cholesterol to squirrel monkeys (Corey et al., 1974; Lofland et al., 1970; Robison et al., 1971), cebus monkeys (Corey et al., 1974; Portman and Andrus, 1965; Wissler et al., 1962), vervet monkeys (Mendelsohn et al., 1980), and rhesus monkeys, but not when fed to baboons (Strong and McGill, 1967). In some experiments, two or more types of fats differing in origin and in fatty acid chain length were compared. For example, in cebus monkeys (Wissler et al., 1962), coconut oil produced higher serum cholesterol levels than did butterfat. However, only a few such experiments were conducted, and it is difficult to reach a firm conclusion other than the generalization that SFAs elevate serum cholesterol and cause more severe atherosclerosis in most nonhuman primates than do PUFAs.

More recent experiments have been conducted to examine the effects of various types of fatty acids on the major serum lipoprotein classes. In baboons, for example, SFAs, as compared to unsaturated fatty acids, elevated serum LDL-C and HDL-C concentrations to approximately the same extent. (McGill et al., 1981b; Mott et al., 1982).

In rhesus monkeys, coconut oil, unlike corn oil, elevated plasma LDL-C proportionately more than it did HDL-C in a short-term experiment (Chong et al., 1987), but coconut oil elevated LDL-C only when fed with cholesterol in a long-term experiment (Ershow et al., 1981). Coconut oil also elevated serum cholesterol in cebus and squirrel monkeys (Corey et al., 1974). Although there are some variations among the results in nonhuman primates, the preponderance of evidence indicates that these animals respond to saturated and unsaturated fatty acids as do humans, i.e., with an elevation in the plasma concentrations of both LDL-C and HDL-C in response to SFAs.

A number of studies in animals have been conducted in an attempt to identify the mechanisms responsible for these effects of saturated and unsaturated fatty acids on plasma lipoproteins (Chong et al., 1987; Ershow et al., 1981; Johnson et al., 1985; Parks and Rudel, 1982; Rudel et al., 1981, 1983, 1986; Spady and Dietschy, 1985). The observation promising the most insight into the process (and discussed above) is that dietary SFAs fed with cholesterol suppress LDL receptor activity in the hamster liver to a much greater extent than does dietary cholesterol alone (Spady and Dietschy, 1985, 1988)—a change that would be expected to elevate plasma LDL-C levels.

Effects of Dietary Fats and Other Lipids on Hemostasis and Eicosanoid Metabolism

There is no conclusive evidence that hemostatic variables such as platelet function and blood coagulation as measured in vitro are influenced by dietary intake of SFAs, MUFAs, or ω-6 PUFAs, such as linoleic, dihomogamma linolenic, or arachidonic acids (von Schacky, 1987). However, some reports have suggested that long-term differences in the intake of SFAs or of ω-6 PUFAs may influence platelet function or thrombogenesis (Renaud et al., 1970, 1985). Additional definitive studies would be useful in this area.

More definite evidence about such effects is available for the ω-3 PUFAs, such as eicosapentaenoic acid (EPA, C20:5, ω-3) and docosahexaenoic acid (DHA; C22:6, ω-3). These ω-3 PUFAs appear to have inhibitory effects on platelet function and prolong bleeding time—phenomena somewhat similar to those elicited by aspirin (Knapp et al., 1986; Leaf and Weber, 1988). However, no studies have as yet determined whether ω-3 PUFAs reduce morbidity or mortality in cardiovascular disease (for full reviews and references, see Leaf and Weber, 1988, and von Schacky, 1987).

Shortened coagulation times measurable in hyperlipemic plasma were originally believed to represent hypercoagulability, but this was later shown to be an artifact induced by platelets trapped in the lipemic plasma (Marcus, 1966). With the possible exception of ω-3 PUFAs, dietary lipids do not influence eicosanoid metabolism as presently understood (Mahadevappa and Holub, 1987).

Of greatest interest concerning effects of dietary fats and other lipids on hemostasis and eicosanoid metabolism are the ω-3 PUFAs of marine origin (von Schacky et al., 1985b). Some investigators have reported that Eskimo populations in Greenland appear to have reduced mortality from occlusive vascular disease, putatively related to consumption of fish and fish oils containing ω-3 fatty acids (Bang and Dyerberg, 1980). However, the epidemiologic evidence on this assertion is quite incomplete. Furthermore, there is no evidence that dietary supplementation with ω-3 fish oils will reduce the risk of cardiovascular disease (Carroll, 1986a).

ω-3 Polyunsaturated Fatty Acids

The ω-3 group of PUFAs includes not only EPA and DHA, but also a-linolenic acid (C18:3, ω-3). These PUFAs are cyclooxygenated and lipoxygenated in a manner similar to ω-6 fatty acids, but the end products may differ functionally from those of the ω-6 group.

The ω-6 PUFAs include linoleic acid (C18:2, ω-6), dihomogamma linolenic acid (C20:3, ω-6), and arachidonic acid (C20:4, ω-6). Arachidonic acid is cyclooxygenated to prostaglandins E₂, D₂, F_2α, prostacyclin (PGI₂, measured as 6-keto-PGF_1α), thromboxane (TX) A₂, HHT, and malondialdehyde (Marcus, 1978, 1984), whereas EPA is metabolized to PGE₃, PGF_3α, PGI₃, and TXA₃. EPA is also lipoxygenated in stimulated leukocytes to the leukotrienes LTB₅, LTC₅, LTD₅, and LTE₅ and to 5-hydroxyeicosapentaenoic acid (5-HEPE). This is in contrast to leukocyte arachidonate, which is lipoxygenated to LTB₄, LTC₄, LTD₄, LTE₄, and 5-hydroxyeicosatetraenoic acid (5-HETE). In activated platelets, arachidonate is lipoxygenated to 12-HETE, and EPA to 12-HEPE (von Schacky, 1987).

Some studies indicate that TXA₃ does not possess the same proaggregatory or vasoconstrictory properties as TXA₂, whereas PGI₃ is reported to retain the vasodilatory and platelet inhibitory properties of PGI₂. More definitive studies are necessary to confirm and examine further the spectrum of biologic properties of PGI₃ and TXA₃.

Terminology

The widespread interdisciplinary interest in dietary lipids as possible modulators of hemostasis, eicosanoid metabolism, and cardiovascular disease has led to the inadvertent conceptualization of these different processes as synonymous events (Marcus, 1988b). Thrombogenesis, atherosclerosis, hypercoagulability, and platelet hyperreactivity are frequently referred to as phenomena occurring in parallel. Atherosclerosis, however, is not clearly related to blood coagulation or to platelet function, until it reaches the advanced occlusive stages in which thrombosis occurs.

Patients with clinically manifest coronary artery disease may have circulating hyperactive platelets, but these could result from increasing cell-to-cell encounters between platelets and atherosclerotic plaques. Platelet hyperreactivity may represent a secondary event such as a response to injury (Marcus, 1988a).

A direct relationship between diet and hemostasis or eicosanoid metabolism has not been established. The only possible exception may be an association with ω-3 PUFAs (Mahadevappa and Holub, 1987).

Effects Of ω-3 Fatty Acids On Platelet Function

Studies have been conducted in volunteers and patients with vascular disease who have ingested cold water marine fish such as mackerel or salmon or fish oils per se. Cod liver oil, which contains 10% EPA and 15% DHA, has been administered in doses ranging from 10 to 40 ml daily for varying periods as has the commercial preparation known as MaxEPA—a blend of fish body oil containing 20% EPA and 10% DHA. Purified EPA and DHA also have been administered separately in a short-term study. Bleeding time, a major indicator of platelet function, was prolonged to some degree in almost all study participants (von Schacky, 1987). In Greenland Eskimos, bleeding time averaged 8 minutes; in a Danish control population, it was 5 minutes. Bleeding time in patients with atherosclerosis extended from a control of 3.3 minutes to 5.6 minutes during the third week in which their diet was supplemented with MaxEPA. Although these findings are suggestive, it is difficult to know with certainty whether such lengthening of the bleeding time indeed signifies protection against vascular disease. Nevertheless, the prolongation is consistent and is comparable to changes in bleeding time observed following the ingestion of aspirin (Knapp et al., 1986).

In addition to lengthening bleeding time, the ingestion of cod liver oil reduces ex vivo platelet aggregation in response to adenosine diphosphate (ADP) and collagen. Responsiveness to agonists would be more adequately interpretable if each stimulus were studied over a concentration curve, because high doses of agonists such as thrombin and collagen may mask mild inhibitory effects of fish oils.

When EPA and DHA were administered separately for 6 days at 6 g/day, EPA reduced platelet responsiveness to collagen. In contrast, DHA reduced platelet aggregation to both ADP and collagen. Thus, DHA may be an important antithrombotic agent in its own right (von Schacky et al., 1985b).

A Comparison Of Arachidonate And Epa Metabolism In Humans

Early in vitro studies on human or animal platelets demonstrated that added EPA was incorporated into all phospholipid subclasses whether or not arachidonate was also present. In contrast, EPA ingested by volunteers was incorporated mainly into platelet phosphatidylcholine and phosphatidylethanolamine, but not phosphatidylinositol or phosphatidylserine. Upon stimulation of washed platelets obtained from these volunteers, EPA was released by way of the phospholipase A₂ mechanism, acting on phosphatidylcholine and phosphatidylethanolamine. Moreover, the dietary EPA released was demonstrably converted to TXA₃ (von Schacky et al., 1985a).

When arachidonic acid is added to platelets in vitro, it is rapidly metabolized to TXA₂, HHT, and 12-HETE, and platelet aggregation ensues via release of ADP (Marcus, 1987). In contrast, the addition of EPA does not induce platelet aggregation. EPA will compete with free arachidonate for platelet cyclooxygenase; however, compared to arachidonate, EPA is a poor platelet agonist. More detailed research on the biologic action of TXA₃ in vitro is desirable (von Schacky et al., 1985b).

DHA is incorporated into platelets in vivo in a manner similar to EPA. At high concentrations (150 µM in vitro), DHA appears to inhibit platelet aggregation, but at low concentrations (2 to 10 µM) platelet aggregation is not inhibited. Following dietary supplementation, DHA is not released from stimulated human platelets or neutrophils. In contrast, dietary DHA is partially retroconverted to EPA (C20:5, ω-3). Further studies on the effects of DHA on platelets ex vivo should be of great interest (Mori et al., 1987; von Schacky et al., 1985b).

During ω-3 PUFA dietary supplementation of healthy volunteers or patients with atherosclerosis, the quantity of TXA₂ formed by stimulated platelets ex vivo is reduced. This could be due to preferential incorporation and may lead to less arachidonate being available for cyclooxygenation. There may also be competition for cyclooxygenase between arachidonate and EPA (which is oxygenated less efficiently than arachidonate), resulting in less TXA₂ formation. This may represent part of the suggested protective action of dietary ω-3 PUFAs.

Metabolism Of EPA In Endothelial Cells

EPA added to endothelial cells of human umbilical cord origin in the presence of arachidonate results in formation of PGI₃, whose biologic activity has been reported to be similar to that of PGI₂. During dietary supplementation with EPA, the urinary metabolite of PGI₃ is detectable. Metabolism of ω-3 PUFAs acids in various types of cultured cells should be a fruitful subject for future investigation. Results obtained thus far suggest that dietary lipid modification or supplementation may benefit blood vessel function (von Schacky, 1987).

Summary

There are many compelling reasons for pursuing basic and applied research on the effects of dietary ω-3 PUFAs in vascular, inflammatory, and allergy diseases. Among these are the broad spectrum of possible beneficial effects of two natural compounds (EPA and DHA), avoidance of pharmacologic intervention with its inherent side effects and problems of compliance, and the long-term economic advantages of successful prophylaxis in coronary artery disease and cerebrovascular disease. The effects of these ω-3 PUFAs can include reductions in platelet responsiveness, as well as favorable changes in blood lipids, blood pressure, blood rheology, and eicosanoid biosynthesis. More detailed information on the effects of ω-3 PUFAs on platelet function, coagulation, and eicosanoid metabolism is desirable. In studies of each disease entity possibly amenable to improvement by the ingestion of ω-3 PUFAs, problems of patient selection, precise dosages, criteria for evaluation of results, and definition of end points remain to be resolved.

Consumption of ω-3 PUFAs should be differentiated from that of fish. Epidemiologic data suggest that consumption of fish is associated with a reduced risk of CHD (Kromhout et al., 1985; Shekelle et al., 1981a). In contrast, it has not been established that intake of ω-3 fish oils per se will reduce the risk of CHD. Furthermore, it is not known whether long-term ingestion of these PUFAs will lead to undesirable side effects. The information available does not support a recommendation to use fish oil supplements to reduce the risk of CHD.

Epidemiologic Evidence

Self-reported survey data from longitudinal population studies indicate that specific dietary fatty acids are related to CHD risk. Data from national sources on food consumption and vital statistics on causes of death also suggest an association.

Studies Of Differences Among Populations

In the Seven Countries Study, correlations between mean intake of SFAs and 10-year CHD mortality and incidence were .84 and .73, respectively (Keys, 1980b). CHD was not significantly associated with mean intake of PUFAs, which varied little among the cohorts. Another salient finding from these comparisons was the observation on Crete that a high intake of total fat (40% of daily calories) was compatible with very low risk of CHD, along with a very low total death rate, when the mean intake of SFA was low (8% of daily calories, while monounsaturates accounted for 29%).

Data from national food balance sheets and vital statistics show positive correlations of CHD deaths with eggs, meat, milk, sugar, and coffee, and negative correlations for flour (Armstrong et al., 1975; Stamler, 1979). However, such data have severe limitations, and the results should be interpreted with great caution, even when appropriate adjustments have been made for changes in the International Classification of Diseases and when data are used only from countries having the more reliable mortality statistics. These correlations require assumptions about lag time between measurement of diet and manifestation of disease. Moreover, confounders cannot usually be measured or accounted for in analysis. The diet and disease data bases are usually different, and the dietary data may not be representative for the populations of interest (e.g., middle-aged men). Nevertheless, these data suggest that more systematically acquired data, as from the Seven Countries Study, can be generalized.

Greenland Eskimos have been cited as an example of a population with a high intake of fats and oils from marine sources and very low rates of CHD (Bang and Dyerberg, 1980), but the epidemiologic data are insufficient to confirm this assertion. In fact, the risk of CHD is unknown in this small population, whose members live under harsh conditions and usually die before middle age.

The ecological associations between composition of the diet and CHD rates, based on systematic and standardized survey comparisons, indicate that low CHD rates may indeed occur in the presence of moderately high total (and monounsaturated) fatty acid intake—when intake of dietary SFAs and cholesterol is low. A low SFA intake, i.e., <10% of daily energy intake, along with a low intake of cholesterol, is common to all populations found to have truly low CHD rates.

Studies Of Individuals Within Populations

Studies in which intake of nutrients was not adjusted for energy intake and dietary information was not obtained at the initial examination (when participants were unaware of their CHD risk status) have not been considered here. The committee considers their results to be uninterpretable with regard to associations between diet and CHD risk (Shekelle et al., 1981b; Willett and Stampfer, 1986).

Even studies that met these conditions have produced inconsistent results. For instance, the percentage of calories from SFAs was positively associated with risk of CHD in the rural sample of the Puerto Rico Study (Garcia-Palmieri et al., 1980), the Honolulu Heart Program (McGee et al., 1984), and the Ireland-Boston Diet-Heart Study (Kushi et al, 1985), but was not significantly associated in the Israel Ischemic Heart Disease Study (Medalie et al., 1973), the urban sample in Puerto Rico (Garcia-Palmieri et al., 1980), the Western Electric Study (Shekelle et al., 1981a), or the Zutphen Study (Kromhout and de Lezenne Coulander, 1984). The percentage of calories from PUFAs was inversely associated with risk of CHD in the Western Electric Study (Shekelle et al., 1981a), but not in the other studies; in fact, it was positively associated with CHD in the urban sample in Puerto Rico (Garcia-Palmieri et al., 1980). The ratio of PUFAs to SFAs (the P:S ratio) was inversely associated with risk of CHD in the small cohort of British men studied by Morris et al. (1977), but not in Israeli civil servants (Medalie et al., 1973).

In two studies, the Western Electric Study and the Ireland-Boston Diet-Heart Study, the lipid composition of the diet was characterized by scores based on the formulas developed by Keys et al. (1965c) and by Hegsted et al. (1965) to summarize the effect of change in dietary SFAs, PUFAs, and cholesterol on change in TC. In both studies, the scores were significantly associated with risk of CHD.

Animal foods in general were related to CHD risk and total deaths in Seventh-Day Adventists (Phillips et al., 1978). Fish-eating habits of men in the Zutphen Study and in the Western Electric Study were related to subsequent individual CHD risk. In these studies, men who rarely or never ate fish had a higher risk of CHD in comparison to men who usually consumed fish one or more times per week (Kromhout et al., 1985; Shekelle et al., 1985). In other population studies, however, fish intake was not significantly related to risk of death from CHD (Curb and Reed, 1985; Snowdon, 1981; Vollset et al., 1985).

The results of studies of individuals within populations have in general been somewhat inconsistent and inconclusive. Investigation of the basis for these inconsistencies has led to greater awareness of the problems inherent in using cohort and case-comparison studies to investigate a diet-disease hypothesis. These problems include the limited variation in characteristic intake of dietary lipids among individuals within populations; the large misclassification of persons according to characteristic intake of dietary lipids; and the substantial misclassification of individuals according to severity of coronary atherosclerosis (the dependent variable of real interest) when classification is based on signs and symptoms of CHD. These factors will produce weak and inconsistent associations in case-comparison and cohort studies of dietary lipids and CHD, even when important causal associations exist. When combined with results from metabolic ward studies, data from carefully conducted ecological studies, such as the Seven Countries Study (Keys, 1980b), provide a more accurate estimate of the role of dietary lipids in producing widespread atherosclerosis and CHD.

Trans Fatty Acids

Thomas et al. (1981, 1983a,b) compared the trans fatty acid content of adipose tissues from men dying of CHD and from other causes. CHD subjects had a higher proportion of shorter chain trans fatty acids than did non-CHD subjects. The investigators concluded that the CHD subjects had consumed more hydrogenated fats than had the controls; however, they did not adjust their findings for other risk factors. In contrast, Heckers et al. (1977) found no differences in tissue trans fatty acids related to atherosclerosis in a few cases. No other systematic epidemiologic data regarding the possible association of trans fatty acid consumption and atherosclerotic disease are available.

Atherosclerosis at Autopsy

In an autopsy study of 253 men in New Orleans (Moore et al., 1976), data on diet were obtained retrospectively from surviving spouses. The extent of raised lesions in the coronary arteries was positively associated with reported intake (expressed as percentage of total calories) of fat and animal protein and was inversely associated with intake of vegetable protein and starch. In contrast, however, antemortem data on diet in the prospective Honolulu Heart Program (Reed et al., 1987) were not significantly associated with the extent of coronary atherosclerosis. Results from the prospective Puerto Rico Heart Health Program (Sorlie et al., 1981) showed small negative correlations between extent of coronary atherosclerosis and reported intake of calories, starch, and fat, but the results are difficult to interpret because the data on intake of nutrients were not adjusted for intake of calories.

Diet and Other Cardiovascular Disease Manifestations

Even though data may be collected systematically, they are insufficient for comparing habitual diet and rates of brain infarction and peripheral vascular disease among populations. This is partly because of the difficulty of obtaining accurate diagnoses and because rates of stroke and peripheral vascular disease rates are much lower than those for CHD, thus giving less power for analysis.

Animal fat, saturated fat, and total fat have been positively related to the risk of thrombotic brain infarction (D. Reed et al., 1986). However, where hemorrhage accounts for most cases of stroke, as in Japan, animal protein and saturated fat are inversely related to cerebral hemorrhage and overall incidence (Kagan et al., 1980, 1985; Kato et al., 1973; Takeya et al., 1984). The components of diets consumed by omnivorous Seventh-Day Adventists, including meat, eggs, milk, and cheese, were shown to be unrelated to risk of stroke (Snowden, 1988).

In studies of Japanese populations, investigators have hypothesized that a Western-type diet (i.e., with a fat content higher than the usual Japanese diet), which is associated with higher TC levels, is protective against stroke in Japan (Komachi 1977). This idea arose not only from the findings of the negative association of TC level to cerebral hemorrhage, but also from the decline in overall stroke rates in Japan with acculturation to Western ways in the last 20 years (Shimamoto et al., 1989). This hypothesis has been supported in Western countries by the negative association of TC with cerebral hemorrhage deaths among hypertensives in the Multiple Risk Factor Intervention Trial screening population and by the positive association of TC with cerebral thrombotic deaths cited above. However, it is still questionable whether the Western diet with its higher saturated and total fat intake and resultant higher TC levels is causally related to decreased risk of cerebral hemorrhage, because obvious confounders, such as the traditional Japanese diet of high salt and relatively low fat and protein intake, have not been sufficiently considered. Another potential confounder, alcohol and alcohol-related bleeding diseases, has also not been excluded.

Randomized Trials of Diet and Coronary Heart Disease

The randomized controlled trials on primary prevention of CHD are summarized above in Tables 7-1 and 7-2. Results indicated that the percentage difference in risk of CHD between the experimental and control groups varied directly with the percentage difference in mean serum TC: the larger the difference in mean TC, the larger the difference in risk of CHD.

Only two small studies included changes in diet as the only intervention. The double-blind Los Angeles Veterans Administration Domiciliary Study (Dayton et al., 1968) of 846 men ages 55 to 89 years was not strictly focused on primary prevention, since 20% of the participants had evidence of old myocardial infarction at entry. In this study, investigators compared the effects of two diets, each providing approximately 40% of calories as fat, but with a much higher proportion of PUFAs and a lower proportion of SFAs in the experimental diet than in the control diet. The experimental diet (which contained linoleic and oleic acid each at 35 to 40% of total fatty acids) induced a prompt and sustained difference in mean TC level between the experimental and control groups, amounting to 12.7% of the starting level. Incidence of the primary end point, sudden death and definite myocardial infarction, was 24% lower in the experimental group than in the control group (54 compared to 71 cases, respectively), but the difference was not statistically significant. However, when these end points were pooled with cerebral infarction and other secondary end points, the totals were 66 in the experimental group and 96 in the control group (31% reduction; p = .01). Fatal atherosclerotic events numbered 48 in the experimental group and 70 in the control group (31% reduction; p <.05).

Initially, there was some concern that the experimental diet might have increased the risk of cancer, since 31 of 174 deaths in the experimental group were due to cancer as compared to 17 of 178 deaths in the control group (p = .06) (Pearce and Dayton, 1971). However, analysis of results from four other clinical trials in which high-PUFA diets were used to lower serum TC (Ederer et al., 1971) showed that the incidence of and mortality from all cancers were lower in the experimental group than in the control group; the relative risks were .75 for incidence and .62 for death. None of the relative risks were significantly different from unity, and the combined results from all five studies were consistent with the conclusion that the cholesterol-lowering diets did not influence risk of cancer. However, these studies could not rule out the possibility of small effects in either direction (increasing or decreasing the risk of cancer).

Further examination of the results of the Los Angeles Veterans Administration Domiciliary Study showed an association between the experimental diet and cholelithiasis (Sturdevant et al., 1973). Autopsied men who had eaten more than 33% of the experimental meals served from entry into the trial to death were more likely to have gallstones than were control subjects; the prevalence rates were 34 and 14%, respectively (p <.01). The prevalence of gallstones at autopsy was positively correlated with number of trial meals eaten by men in the experimental group (p <.05) but not by men in the control group.

High-PUFA diets (i.e., more than 10% of total calories from PUFAs) are no longer recommended in the United States for lowering serum TC, in part because of concerns about the long-term safety of such diets.

In the Finnish Mental Hospital Study (Turpenien et al., 1979), participants were 450 men ages 34 to 64 who were hospitalized in one of two mental hospitals near Helsinki. At the beginning of the study, one hospital received a diet low in SFA and cholesterol and relatively high in PUFA, while the other hospital continued with the usual hospital diet. Six years later, the diets were reversed, and the study was continued for another 6 years. In both hospitals, the special (high PUFA) diet was associated with markedly lower mean TC. The difference in one hospital was -50.9 mg/dl (-19%) and in the other, -32.0 mg/dl (-12%). The mean difference between the special-diet and usual-diet periods in the two hospitals was -41.4 mg/dl (-15%). The incidence of coronary events in the special-diet and usual-diet periods for both hospitals combined was 2.3 and 4.3 per 1,000 person-years, respectively, for coronary death (p = .10); 4.2 and 12.7 per 1,000 person-years for coronary death plus major electrocardiogram (ECG) change (p = .001); and 13.5 and 24.3 per 1,000 person-years for coronary death plus intermediate or major ECG changes (p = .008).

A large-scale trial on primary prevention of CHD with change in diet as the only intervention was never conducted because of concerns about feasibility (NHLBI, 1971). Thus, the remaining diet-heart trials are either 2-factor or 3-factor trials in which any effect of dietary change was confounded with effects of changes in the other factors. Three of those trials (the Göteborg Multifactor Trial, the WHO Multifactor Trial, and Multiple Risk Factor Intervention Trial) achieved differences in mean TC between the intervention and comparison groups of 0, -1, and -2%, respectively; the differences in coronary rates were correspondingly small (0, -4, and -7%, respectively).

The Oslo Study (Hjermann et al., 1981), which began with 1,232 hypercholesterolemic men ages 40 to 49, achieved a 13% reduction in mean TC levels in the experimental group as compared to the control group. At the end of the follow-up period, 7.5 years later, the incidence of myocardial infarction (fatal and nonfatal) plus sudden death was 47% lower in the experimental group than in the control group (p = .028). Total mortality in the experimental and control groups was 26/1,000 and 38/1,000, respectively (p = .25). From their own analysis of the data in this study, the authors concluded that changes in cigarette smoking accounted for at most approximately one-fourth of the observed difference in risk of CHD and that most of the difference in CHD was apparently due to diet-induced changes in serum cholesterol. Although this quantitative conclusion is subject to some question, it is clear that the diet-induced lowering of TC played a major role in the observed significant reduction in CHD.

Summary

Epidemiologic evidence shows that serum TC levels are positively associated with risk of CHD, experimental studies in animals and clinical research have elucidated in large part the mechanisms for this association, and results of clinical trials strongly support the concept that lowering TC levels reduces incidence of CHD. Experimental studies in humans have demonstrated that intake of dietary SFAs, PUFAs, and cholesterol affects the level of serum TC. The mechanisms of these effects are now being elucidated. Epidemiologic studies support the concept that variation among populations in mean intake of SFAs (and probably cholesterol) is a major factor accounting for variation in mean TC levels. Results of dietary intervention clinical trials are consistent with the conclusion that changing the lipid composition of the diet in ways expected to lower the TC levels (i.e., decreasing SFA and cholesterol intake, and increasing PUFA intake) in fact does lower mean levels of serum TC and reduces incidence of CHD.

Animal Studies

SFAs and PUFAs

The relationship of dietary fatty acid saturation to experimental atherosclerosis parallels closely the fatty acid effects on levels of serum cholesterol, particularly LDL-C. Cholesterol fed with hydrogenated vegetable oil (source unspecified) produced more severe atherosclerosis in rabbits than cholesterol fed with corn oil (Kritchevsky et al., 1954). Subsequently, SFAs were found to be more atherogenic than unsaturated fatty acids in several species of monkeys (Lofland et al., 1970; Mendelsohn et al., 1980; Robison et al., 1971; Wissler et al., 1962).

Effects Of Dietary Fats With No Cholesterol

Hypercholesterolemia and atherosclerosis were produced in rabbits by feeding them semisynthetic diets enriched with hydrogenated coconut oil, butterfat, or margarines and no added cholesterol (Carroll, 1971; Funch et al., 1960, 1962; Lambert et al., 1958; Malmros and Wigand, 1959; Wigand, 1959). It was later shown, however, that these diets were deficient in essential fatty acids, and a small amount of PUFAs prevented hypercholesterolemia (Diller et al., 1961). Finally, an improved semisynthetic diet eliminated the essential fatty acid deficiency, and it was shown that hypercholesterolemia and atherosclerosis were related to fatty acid saturation when fed with a low dose of cholesterol (Wilson et al., 1973). These results were confirmed by other investigators (Kritchevsky and Tepper, 1965a,b; McMillan et al., 1960). Cholesterol-free semisynthetic diets also produced hypercholesterolemia and atherosclerosis in vervet monkeys and in baboons (Kritchevsky et al., 1974a,b).

Peanut Oil

Gresham and Howard (1960) reported that peanut oil, which has a high proportion of PUFAs, had the unanticipated effect of being more atherogenic than other fats when fed with cholesterol to rats. This finding was confirmed in rats (Scott et al., 1964) and extended to rabbits (Imai et al., 1966; Kritchevsky et al., 1971) and rhesus monkeys (Scott et al., 1967a,b; Vesselinovitch et al., 1974, 1980). Later experiments suggest that the unusual atherogenicity may be related to triglyceride structure (Kritchevsky et al., 1973, 1981; Myher et al., 1977), but there are no studies that explain this effect in terms of an effect on lipoprotein metabolism.

In cynomolgus monkeys, peanut oil fed with a concentration of dietary cholesterol equivalent to that consumed by humans affected serum lipoproteins and atherosclerosis as did other unsaturated vegetable oils (Alderson et al., 1986).

The contradictory results have not been explained. Kritchevsky (1988) suggested that because positional randomization of the triglyceride fatty acids of peanut oil reduced its atherogenicity, the unusual effect might be due to its triglyceride structure or that a lectin in peanut oil might be responsible. Further investigation is needed to resolve these discrepant results.

Trans Fatty Acids

Effects of trans fatty acid isomers have been studied in laboratory animals (reviewed by Kritchevsky, 1982; Senti, 1985). These isomers are not metabolized in the same way as the cis isomers. However, at dietary intakes similar to those consumed by humans, and with a small proportion of linoleic acid in the diets, the trans fatty acid isomers appear to have no deleterious health effects (Jackson et al., 1977; Kritchevsky, 1983; Royce et al., 1984).

Summary And Conclusions

In most animal species, SFA intake is associated with more severe atherosclerosis than that achieved with similar diets without (or with little) SFA. Some types of SFAs, especially coconut oil, are more atherogenic than indicated by their effects on plasma lipoproteins alone. The observations in animals support the putative role of SFAs in causing hypercholesterolemia and atherosclerosis in humans.

Epidemiologic Studies

Studies Of Populations

Stamler et al. (1970, 1972) examined the relationship of mean per-capita consumption of dietary cholesterol to arteriosclerotic disease mortality rates in 20 developed countries. The correlation coefficient for daily cholesterol intake and mortality from CHD in 1964 was .617 (p <.01) for 45- to 54-year-old men and .685 (p <.01) for 55- to 64-year-old men. These mortality rates were also correlated with national per-capita income, cigarette smoking, total protein consumption, total fat consumption, and other environmental variables.

Armstrong et al. (1975) compared the consumption of various commodities with mortality from ischemic heart disease in nine regions of the United Kingdom and in 30 countries. They found strong correlations between egg consumption and ischemic heart disease. These results are consistent with the hypothesis that dietary cholesterol is causally associated with incidence of CHD, but evidence from other types of studies must be considered in order to assess alternative hypotheses and to reach a conclusion.

STUDIES OF INDIVIDUALS

The Effect of Dietary Cholesterol on Plasma Lipids

With one exception (Connor et al., 1978), cross-sectional studies of individuals within populations have shown little or no association between dietary cholesterol and concentration of serum TC (Frank et al., 1978; Garcia-Palmieri et al., 1977; Harlan et al., 1983; Kahn et al., 1969; Nichols et al., 1976; Shekelle et al., 1981a; U.S.-U.S.S.R. Steering Committee, 1984; Yano et al., 1986). Jacobs et al. (1979) reported that correlations close to zero are a necessary result of cross-sectional studies because of the large differences in intrinsic TC levels among individuals. They concluded that cross-sectional studies are inappropriate for studying the effects of dietary cholesterol on serum TC.

In contrast, longitudinal observational studies have generally shown a positive association between changes in dietary cholesterol and changes in TC concentrations. For example, in the Lipid Research Clinics Coronary Primary Prevention Trial, 6,000 hypercholesterolemic men were advised to lower their intake of cholesterol and SFAs (Gordon et al., 1982). Multiple regression analysis showed that the reduction in dietary cholesterol was independently associated with reduction in serum TC concentrations. The reduction in TC was due primarily to a reduction in LDL-C. A similar association of changes in dietary cholesterol with changes in serum TC was found in the Zutphen Study (Kromhout, 1983) and in the Western Electric Study (Shekelle et al., 1981a). The magnitude of the association found in observational studies has been less than that found in experimental studies due to the well-known effects of day-to-day intraindividual variation in intake of nutrients and in serum TC levels (Beaton et al., 1979; Keys, 1965; Liu et al., 1978).

Dietary Cholesterol and Coronary Heart Disease

Before 1981, epidemiologic studies of individuals within populations showed little or no association between dietary cholesterol and risk of CHD (Garcia-Palmieri et al., 1980; Gertler et al., 1950a; Gordon et al., 1981a; Paul et al., 1963; Yano et al., 1978). However, in these studies, investigators did not adequately consider one or more of three important methodological issues: adjustment for the confounding effect of caloric intake (Willett and Stampfer, 1986); limitations in the reliability of dietary measurements (Beaton et al., 1979; Keys, 1965; Liu et al., 1978); and bias resulting from systematic dietary changes made by hypercholesterolemic persons after learning of their condition (Shekelle et al., 1981b). Since then, four long-term cohort studies—the Western Electric Study (Shekelle et al., 1981a), the Zutphen Study (Kromhout and de Lezenne Coulander, 1984; Kromhout et al., 1985), the Ireland-Boston Diet-Heart Study (Kushi et al., 1985), and the Honolulu Heart Program (McGee et al., 1984)—have shown that dietary cholesterol, assessed reliably at entry and expressed as mg/1,000 kcal, was positively associated with risk of CHD after adjustment for age, blood pressure, cigarette smoking, and serum cholesterol level. Combining the results of these studies, Stamler and Shekelle (1988) calculated that a dietary cholesterol increment of 200 mg/1,000 kcal daily was associated with a 30% greater risk of CHD.

Considering the difficulties in measuring cholesterol intake, the many etiologic factors involved in CHD, the long natural history of the underlying atherosclerosis, and the variability among individuals in response to diet, the committee interprets these observations as supporting an etiologic role of dietary cholesterol in CHD. Since the association between dietary cholesterol and CHD persisted after adjustment for serum TC, the results suggest that dietary cholesterol may affect atherogenesis and CHD by other mechanisms in addition to its effect on LDL-C concentration (Zilversmit, 1979). The possible contribution of dietary cholesterol to cerebral vascular disease and peripheral vascular disease was not addressed in these studies.

Clinical and Metabolic Studies

Response Of Serum Total Cholesterol

Many experiments have been conducted to measure the response of serum or lipoprotein cholesterol concentrations in humans to different amounts of dietary cholesterol. Cholesterol intake was usually modulated by adding or removing eggs from the diet. These experiments have varied widely in the degree of control of dietary intake among the subjects. They range from experiments in which the subjects were confined to a metabolic ward and fed laboratory-prepared formula diets to those in which noninstitutionalized subjects were advised to add eggs to or remove eggs from their habitual diets.

Examples of rigorously controlled experiments include those reported by Connor et al. (1961a,b, 1964). In each of three experiments, six men were placed on cholesterol-free and high-cholesterol diets that were either formula diets or carefully selected natural foods. Cholesterol was added as egg yolk. In all three experiments, substantial increases in serum TC resulted from the addition of egg yolk to the diet.

Beveridge et al. (1960) fed formula diets containing varying amounts of added crystalline cholesterol for 8 days to 67 noninstitutionalized university students. There was a linear increase in serum TC with increases in daily intake up to about 600 mg. A similar study was conducted by Mattson et al. (1972) in 56 healthy prison inmates who consumed formula diets with varying amounts of added egg yolk. There was an approximately linear relationship between cholesterol intake and elevation of serum TC levels.

An unusual opportunity to examine the effects of changes in habitual diet on plasma lipoprotein cholesterol levels was provided by the Lipid Research Clinics Coronary Primary Prevention Trial (Gordon et al., 1982). As in other cross-sectional studies, there was no association of any dietary component with either LDL-C or HDL-C before subjects were instructed to change their diets. However, the change in cholesterol intake (of about 27 mg/1,000 kcal per day) was associated with a change in plasma cholesterol concentration in a multiple regression model. The association was independent of the effects of changes in saturated and unsaturated fat and of weight loss, and there was no interaction between the type of fat and cholesterol intake. The change in plasma cholesterol was almost entirely due to a change in LDL-C.

In a follow-up analysis of data from the same trial after it was terminated, serum TC and LDL-C levels were less strongly and less consistently correlated with decreases in cholesterol intake up to 7 years after the initial dietary instruction period (Glueck et al., 1986). The partial regression coefficients were statistically significant at 4 years, but not at 0.8 and 7 years. In contrast, the coefficients for SFA intake were significant at all times.

In another cross-sectional dietary study, the investigators focused on subfractions of the lower-density lipoproteins measured by analytic ultracentrifugation (Williams et al., 1986). Daily cholesterol intake averaged 190 mg/1,000 kcal. Dietary cholesterol was not significantly correlated with serum TC or LDL-C, but was significantly correlated with the mass of small LDL in the S_f0-7 range (Rs = .30, p <.01). Although there is considerable interest in the relative atherogenicity of lipoprotein subclasses, we do not know whether S_f0-7 LDL has any special significance.

Predictive Equations Relating Dietary Cholesterol To Serum Cholesterol

The first report of an equation to describe the effect of dietary cholesterol or serum cholesterol was made by Grande et al. (1965) in 1965. Keys (1984) reviewed this equation in light of 39 subsequent reports and concluded that the same equation, shown below, remained the best description:

ΔChol = 1.5 (C₂ ^1/2 - C₁ ^1/2),

where ΔChol is the change in serum TC concentration in mg/dl, C₂ is the new intake of cholesterol in mg/1,000 kcal per day, and C₁ is the initial intake.

Hegsted (1986) reviewed the same 39 reports plus several others and concluded that the association was essentially linear from 0 to 400 mg per 1,000 kcal and could be described by the following equation:

ΔChol = 0.0974 C,

where ΔChol is the change in serum cholesterol in mg/dl and C is the difference in cholesterol intake between the two diets in mg/1,000 kcal.

Variability In Response To Dietary Cholesterol

Most reports of experiments on dietary cholesterol in humans mention the high variability of responses of serum cholesterol and LDL-C among individuals. Some people do not respond at all; many have small responses; and a few have large responses. Indeed, as discussed above, the variability among individuals is so great that it obscures correlations in cross-sectional studies.

Reports after 1980 began to focus more attention on this variability. A group of investigators in the Netherlands conducted an intensive study of individual variability in responsiveness to dietary cholesterol. Katan et al. (1986) tested 94 people on diets providing 121 and 625 mg of cholesterol per day in egg yolk. The highest quartile and the lowest quintile of responders were challenged 1 month and again 7 months after the initial challenge. The high and low responders were significantly different from one another in the repeated challenges. Standardized correlation coefficients of the responses to the three challenges, taken in pairs, ranged from .34 to .53 (all p <.05). The mean of each subject's responses during the three challenges, taken as the best estimate of the response, showed a normal unimodal distribution from -12 to +46 mg/dl. Approximately 90% of the change in serum TC was accounted for by changes in LDL-C. McNamara et al. (1987) found that people who responded to dietary cholesterol with an increase in TC were also those in whom endogenous cholesterol synthesis was not suppressed.

There now appears to be convincing evidence that consistent differences in responsiveness of serum cholesterol to dietary cholesterol do exist in humans. When these responses are measured in terms of specific lipoprotein classes, the differences among individuals become even more prominent. No simple marker for sensitivity is available. A test for responsiveness involves feeding a controlled diet for 6 weeks or longer and analyzing several blood samples. The procedure is laborious and cannot be performed on large numbers of people. The definition of a responder, or a high responder, is arbitrary, but is the basis for determining the proportion affected, since the distribution of responses is continuous. However, the results indicate that approximately 20% of adults are likely to have a response more than 10% higher than that predicted by the Hegsted (1986) or Keys (1984) equations.

Effects Of Dietary Cholesterol On Lipoprotein Metabolism

In most studies up to about 1980, the effects of dietary cholesterol were measured only as changes in serum TC. As new analytical methods became available, investigators began to measure the effects of dietary cholesterol on the major lipoprotein classes or the apolipoproteins or on variables such as lipoprotein secretion or degradation. Only in the late 1970s did reliable techniques become available for measuring cholesterol in major lipoprotein fractions and apolipoproteins in a large number of plasma samples. Many cholesterol feeding experiments were repeated to take advantage of these new techniques. For example, Applebaum-Bowden et al. (1979) found that elevations in serum cholesterol were due principally to an elevation of LDL-C and that plasma apolipoprotein B concentrations were also increased. Similar results were obtained by Mistry et al. (1981), Packard et al. (1983), and Schonfeld et al. (1982).

Beynen and Katan (1985) and Schonfeld et al. (1982) reported that most of the change in serum cholesterol was accounted for by a change in LDL-C. Zanni et al. (1987) demonstrated that dietary cholesterol affected HDL-C and found considerable heterogeneity in response among individuals. Applebaum-Bowden et al. (1984) observed that LDL receptor activity in mononuclear cells, as measured by ability to degrade LDL, was decreased by dietary cholesterol proportionally to the increase in serum LDL-C levels. Lin and Connor (1980) found that the percentage of dietary cholesterol absorbed in two subjects did not change within the range of intake from 45 mg to 1,000 mg per day. Ginsberg et al. (1981) reported no change in the rates of production or clearance of VLDL or LDL with changes in dietary cholesterol in five subjects. In all the studies in which the characteristics of the elevated serum LDL were examined, the LDL particles were essentially normal in size and other characteristics; there were simply more particles.

Katan and Beynen (1987) looked for differences in metabolic characteristics between high and low responders and found that only habitual cholesterol intake and HDL₂ cholesterol levels were associated with responsiveness. McNamara et al. (1987) observed that two-thirds of a group of 75 men responded to a 600 mg/day increase in dietary cholesterol with no change in plasma cholesterol level, whereas approximately one-third of the subjects responded with an increase greater than 5%. The nonresponders suppressed cholesterol biosynthesis in blood mononuclear cells, but endogenous biosynthesis was not decreased in the responders.

Zilversmit (1979) suggested that absorbed dietary cholesterol is concentrated in the postprandial lipoproteins, which are thereby made more atherogenic. Investigation of the postprandial lipoproteins on a larger scale has been hampered by limitations in the methods for separating and quantitating these lipoproteins. Therefore, this mechanism remains an attractive but speculative possibility.

In summary, the effect of dietary cholesterol on serum lipoprotein concentrations is principally an increase in the concentration of normal serum LDL particles. There may be other effects on the characteristics of VLDL or HDL that are related to the increased LDL, but these have not been clearly identified and linked to the change in LDL.

Interactions Of Dietary Cholesterol With Other Nutrients

Many investigators have looked for an interaction between dietary cholesterol and type of fat with regard to their effects on serum lipoproteins. A positive interaction with SFAs was found by the National Diet-Heart Study Research Group in the Faribault Study (1968). Bronsgeest-Schoute et al. (1979a) observed no significant interaction in one experiment, but did find one in a later experiment (1979b). Neither Chenoweth et al. (1981) nor Oh and Monaco (1985) reported an interaction with any type of fat.

In summary, despite the widespread belief that there is a positive interaction between dietary cholesterol and SFAs in their effects on serum lipid or lipoprotein concentrations, the evidence from studies in humans is inconsistent, perhaps due to varying proportions of responders and nonresponders in the different studies. Further research is needed to resolve this question.

Animal Studies

Since the initial demonstration of dietary cholesterol-induced atherosclerosis in rabbits in 1913 (Anitschkow, 1967; Anitschkow and Chalatow, 1913), a variety of animal species have been shown to be sensitive to cholesterol. Guinea pigs (Bailey, 1916), mice (Löwenthal, 1925), goats (Chalatow, 1929), swine (Bragdon et al., 1957), chickens (Dauber and Katz, 1942), and most species of nonhuman primates (Clarkson et al., 1976a,b) are quite sensitive; dogs (Steiner and Kendall, 1946) and rats (Wissler et al., 1954) are relatively resistant. Among the nonhuman primates, rhesus, cynomolgus, squirrel, and vervet monkeys are moderately sensitive, whereas baboons are only slightly sensitive (Corey et al., 1974; Eggen, 1974; McGill et al., 1976; Portman and Andrus, 1965; Srinivasan et al., 1976). In each species, as in humans, there is considerable interindividual variability, and it has been shown in several species that much of this variability is under genetic control (Clarkson et al., 1971, 1985; Flow et al., 1981). The metabolic basis of the genetic variability has not been identified.

In animal models, as in humans, the elevated serum cholesterol resulting from dietary cholesterol is carried primarily in LDL (McGill et al., 1981b). Effects on HDL vary considerably from one species to another. In some, dietary cholesterol increases HDL; in others, it decreases HDL.

The increases in serum lipids and lipoproteins caused by dietary cholesterol (usually accompanied by a high level of dietary SFAs) result in arterial lesions similar to those found in humans with atherosclerosis. A few experiments with prolonged feeding of a cholesterol-supplemented diet showed that arterial lesions in animals can undergo complications typical of those seen in advanced atherosclerosis in humans and occasionally result in thrombotic occlusion of the coronary artery and myocardial infarction or occlusion of the iliac or femoral arteries and peripheral vascular disease (Taylor et al., 1963).

In summary, dietary cholesterol is the dominant feature of experimental hyperlipidemia and atherosclerosis in animal models. Humans resemble the more resistant species of nonhuman primates in their sensitivity to dietary cholesterol. The results of studies in which cholesterol is fed to laboratory animals are of particular value in examining physiological mechanisms. The role of dietary cholesterol as an etiologic agent in atherosclerosis in humans should be interpreted with caution, because the relative amount of cholesterol fed to animals usually is much greater than that consumed by human populations. Experimental diets often contain 1,000 to 2,000 mg of cholesterol per 1,000 kcal, whereas a high intake for humans is 300 to 500 mg per 1,000 kcal. These very high cholesterol diets are used to produce high serum cholesterol levels and atherosclerotic lesions within a limited time. Lower dietary intakes, well within the range of human intakes, fed over longer periods to susceptible species have produced atherosclerotic lesions. Thus, the observations in laboratory animals are consistent with the causative role of dietary cholesterol in human disease.

Cholesterol Autoxidation Products

There has been much speculation about the possible toxicity of cholesterol derivatives. Several reports, all from one laboratory (Imai et al., 1976, 1980; Janakidevi et al., 1984; K.T. Lee et al., 1985; Murray et al., 1981; Peng et al., 1978, 1979; Taylor et al., 1979) indicate that 25-hydroxycholesterol and cholestane triol cause necrosis of arterial smooth muscle cells and influence the growth of these cells in culture. None of these effects has been confirmed by other investigators. There is no epidemiologic or clinical evidence suggesting that these derivatives cause hyperlipidemia or atherosclerosis in humans.

Plant Sterols

The intake of plant sterol by humans ranges from 150 to 400 mg/day—about the same daily intake as cholesterol. Most plant sterols are poorly absorbed by the intestinal tract, and those that are absorbed are rapidly excreted. Sitosterol is the most abundant plant sterol; others include stigmasterol, campesterol, brassicasterol, and 24-methylene cholesterol. The amounts that are absorbed are inversely proportional to the molecular weight of the sterols.

Long ago, plant sterols (β-sitosterol and related compounds) were found to prevent the absorption of dietary cholesterol (Best et al., 1955; Farquhar and Sokolow, 1958; Farquhar et al., 1956; Lees et al., 1977; Peterson et al., 1959), apparently by blocking absorption of cholesterol in the intestine (Davis, 1955; Grundy and Mok, 1977; Jandacek et al., 1977; Mattson et al., 1977). More recent reports indicate that these compounds may be more effective in small doses than previously believed (Mattson et al., 1982). No deleterious side effects have been identified, but β-sitosterol has not been widely used in a lipid-lowering diet.

A rare genetic disorder causes people to absorb a large proportion of dietary plant sterols (Bhattacharyya and Connor, 1974; Lin et al., 1983; McArthur et al., 1986; Miettinen, 1980; Shulman et al., 1976; Tilvis and Miettinen, 1986). In this condition, known as phytosterolemia, the affected subjects have high plasma levels of plant sterols and xanthomas. The patients appear to be at an increased risk for atherosclerotic cardiovascular disease and should therefore greatly limit their intake of plant sterols.

Summary

Dietary cholesterol causes hypercholesterolemia and atherosclerosis in a variety of animal species. In humans, within the usual range of intakes (300 to 600 mg/day, or about 150 to 300 mg/1,000 kcal), dietary cholesterol influences the plasma cholesterol concentration, primarily LDL-C. On average, 100 mg of dietary cholesterol per 1,000 kcal elevates LDL-C by 8 to 10 mg/dl. Individual responses vary considerably on both sides of this average response in a continuous distribution. Some people are resistant to the cholesterol-raising effects of dietary cholesterol; others are quite sensitive. For those in the upper range of this distribution—at least one-third of the population—dietary cholesterol is a substantial contributor to serum TC and LDL-C concentration and thus to the progression of atherosclerosis and to the risk of clinically manifest atherosclerotic disease. The only way to identify a susceptible person is by repeated dietary challenges—a procedure that is not feasible on a large scale. For the overall population, the evidence indicates that dietary cholesterol contributes to the development of atherosclerosis and the risk of CHD.

Dietary Fat

The evidence from epidemiologic studies has been based in part on geographical comparisons correlating differences in incidence of and mortality from various cancers with estimated fat intake by populations. Strong associations were found for cancer of the breast, colorectum, endometrium, and prostate, and, to a lesser extent, kidney and testicles (Armstrong and Doll, 1975; Carroll, 1975; Hems, 1980; Knox, 1977). In studies based on data from personal interviews in Hawaii, Kolonel et al. (1981) found significant associations of total fat, animal fat, and saturated and unsaturated fat intakes with ethnic patterns of breast, prostate, and corpus uteri cancer incidence but not of colon cancer.

Other studies have suggested an association of dietary fat with breast cancer (Lubin et al., 1981; Miller et al., 1978; Nomura et al., 1978; Phillips, 1975), prostate cancer (Hirayama, 1977; Rotkin, 1977; Schuman et al., 1982), and colorectal cancer (Dales et al., 1979; Jain et al., 1980; Martinez et al. 1979; Phillips, 1975). Negative studies were also reported, especially for colorectal cancer (Bjelke, 1978; Graham et al., 1978; Haenzel et al., 1980). As noted by the Committee on Diet, Nutrition, and Cancer (NRC, 1982), however, some of these studies were conducted on groups of relatively homogeneous populations, and some were not specifically designed to test the hypothesis that fat consumption is associated with colorectal cancer. The studies specifically designed to test the dietary fat hypothesis (e.g., Dales et al., 1979; Jain et al., 1980) tend to show the most striking direct associations.

The remainder of this section presents the evidence by cancer site, the way most of it is reported.

Breast Cancer

Since the 1982 report of the Committee on Diet, Nutrition, and Cancer (NRC, 1982), data from an earlier case-control study in Canada (Miller et al., 1978) have been reanalyzed with more modem methods and in light of the contribution made by different dietary factors (Howe, 1985). As a result of this analysis, the strongest association was found for SFA intake, and the dose-response relationship was particularly significant in premenopausal women. In this group, the risk for those in the highest SFA consumption tertile (>35.4 g of SFAs per day) relative to the lowest (<23.4 g of SFAs per day) was 5.9. Hirohata et al. (1987) studied 183 case-control sets of Japanese and 161 case-control sets of Caucasian subjects in Hawaii. Each case was matched to one hospital and one neighborhood control, and all were between the ages of 45 and 74. No statistically significant differences were found between cases and controls in their mean intake of total fat, SFAs, oleic acid, linoleic acid, animal protein, and cholesterol. However, there was a suggestion that cases consumed more SFAs and oleic acid than did neighborhood controls. As the authors noted, one difficulty with this investigation may have been the use of a relatively limited dietary questionnaire, which was designed to collect information on a selected list of 43 food items believed to represent the major sources of total and saturated fat, cholesterol, and animal protein.

Lubin et al. (1986) reported the findings of a large case-control study in Israel in which neighborhood controls were also used. When consumption levels were considered in quartiles, an increased risk of breast cancer was found among high fat, high animal protein, and low fiber consumers compared to those with the lowest consumption of fat and animal protein and highest consumption of fiber. For women under the age of 50, the relative risk was 1.6; for women age 50 or more, it was 2.0. A significant trend (p <.02) was found for the latter group.

In British Columbia, Canada, Hislop et al. (1986) administered a food frequency questionnaire on current diet and dietary practices in childhood. Although unable to determine nutrient intake, they found positive associations of breast cancer with current intake of fat-associated foods, especially whole milk and beef, and an increased risk among those who consumed visible fat on meat, both recently and in childhood. In northern Italy, risk estimates for indices of fat intake based on selected food items with a high fat content were found to be moderately elevated (Talamini et al., 1984). In a large case-control study in France involving 1,010 breast cancer cases and 1,950 controls with nonmalignant diseases, increased risk was found for consumption of dairy products but not for butter or yogurt; for the latter, there was a suggestion of a protective effect (Lê et al., 1986). For daily cheese consumption relative to never, the risk was 1.5, and for full cream milk consumption relative to none, the risk was 1.8.

At least three recent studies, however, were negative. In a case-control study in Greece involving 120 patients with breast cancer and 120 controls from an orthopedic hospital, Katsouyanni et al. (1986) found no association with consumption of fats and oils, but a protective effect for the consumption of vegetables. However, the authors pointed out that animal fat consumption in Greece is less than one-third of that in the United States or Canada and all study subjects may have therefore been at low risk of breast cancer, whereas vegetable consumption in Greece is twice as high. In a case-control study in Australia involving 451 cases and 451 controls, Rohan et al. (1988) found no association between dietary fat, protein, or caloric intake, but a protective effect of β-carotene.

A cohort study of nurses who completed a self-administered questionnaire also showed no association between dietary fat and risk of breast cancer (Willett et al., 1987). Among the possible reasons for discrepancies between this large cohort study (89,538 nurses ages 34 to 59 who had 601 breast cancers during 4 years' follow-up) and most case-control studies are recall bias in the case-control studies, and in the cohort study, less than complete estimates of fat intake in the self-administered questionnaire compared to a detailed dietary history (Jain et al., 1982), a short follow-up period (4 years), a narrow range of fat intakes in the study sample, and measurement error combined with a weak association. Another cohort study based on NHANES I data showed an inverse association between dietary fat and breast cancer occurrence that was no longer significant when adjusted for total energy intake. However, the findings may have been influenced by methodologic problems with the dietary assessment; indeed, the design of the NHANES study has been considered unsuitable for analyses based on individual intakes (Jones et al., 1987a) (see Chapter 2).

In addition to the limitations noted above, part of the inconsistencies in the epidemiologic studies could be that fat intake early in life may have a greater influence on breast cancer risk than intake later in life. This could account for the slow change in breast cancer rates in some migrant groups and could be reflected in the failure to observe reduced breast cancer incidence in population groups who as adults have made major dietary changes resulting in lower total fat intake. This applies to groups of nuns who adopted a low-fat diet in adult life (Kinlen, 1982). Furthermore, measurement errors in dietary recall in case-control and cohort studies can account for weak or absent associations of dietary fat with breast cancer, even if the relationship is real (Freudenheim and Marshall, 1988; Prentice et al., 1988).

In summary, the overall evidence is inconsistent. Nevertheless, it tends to support a weak association between dietary fat and breast cancer risk in humans.

Colorectal Cancer

Miller et al. (1983) compared the intake of food groups and various nutrients, particularly SFAs, in relation to risk of colorectal cancer. This was a further analysis of the case-control study first reported by Jain et al. (1980). The risk associated with a higher consumption of SFAs relative to lower consumption, considered in tertiles, for colon and rectal cancer cases combined relative to the combined neighborhood and surgical control series was 2.4 for males and 2.6 for females. In females, increased intake of SFAs consistently increased risk for colon cancer and also had an important independent effect over and above an apparent effect of some meats in increasing risk of rectal cancers. In males, different meat products appeared to increase risk and intake of SFAs did not appear to have an independent effect. Nevertheless, it seems possible that imprecision in consumption estimates, particularly for men, may have contributed to this result.

Willett and Stampfer (1986), in a commentary on differences in total energy intake in dietary studies, recalculated nutrient densities for SFA intake (SFAs divided by total calories) from the study of Jain et al. (1980) and found that the association with colorectal cancer largely disappeared. They noted, however, that inverse associations (as observed for crude fiber intake) can be produced by this process even when such nutrients are not independently associated with disease. This caused Howe et al. (1986) to reanalyze the data from the Jain et al. (1980) study once more, constructing logistic regression models that included variables representing the effect of SFAs and calories from sources other than SFAs. In this approach, the risk for highest, relative to low, consumption levels for SFAs was found to be 2.0 for males and 2.3 for females. No effects were found for calories or for fiber.

Manousos et al. (1983) in a case-control study conducted in Greece based on a food frequency questionnaire observed a positive association between intake of beef and lamb and colorectal cancer risk and a negative association with intake of cabbage, lettuce, spinach, and beets.

Stemmermann et al. (1984) reported the results of a 14- to 17-year follow-up of 8,006 men of Japanese ancestry whose intake of total fat was assessed by a 24-hour recall administered at baseline. The estimated mean intake of fat by the 99 cases was 69 g/day. For the 378 controls, it was 77 g/day. Thus this study did not support the dietary fat hypothesis.

In the Marseilles region of southern France, 399 cases and a corresponding number of surgical controls were studied by Macquart-Moulin et al. (1986), who used a food frequency dietary questionnaire and estimated consumption of fat. There was no evidence of any difference in the consumption of total fat between cases and controls. Indeed, a reduced risk appeared to be associated with increasing consumption of oil, largely monounsaturated oil. Vegetables, particularly vegetables with low fiber content, appeared to be protective in this study.

In Belgium, a study conducted with a similar questionnaire showed no effect of fat intake, but there was a clear dose-response effect for intake of mono- and disaccharides. A protection effect was found for linoleic acid and dietary fiber (Tuyns et al., 1987).

In a case-control study conducted in Adelaide, Australia, Potter and McMichael (1986) found evidence of increasing risk with increasing consumption of a number of nutrients and a pattern of food intake involving high dietary fat and possibly SFAs. Although risk associated with SFAs and various components of fat was increased, it was the impression of the authors that the major contribution to increased risk came from dietary protein.

A case-control study of colorectal cancer in Melbourne, Australia, included 715 cases and 727 age- and sex-matched community controls. Fat intake was found to be a risk factor for both males and females (Kune et al., 1987); however, the relative risk associated with high fat intake was not large (1.4) and became apparent only after adjustment for meat and vegetable intake. The effect was similar, however, for colon and rectal cancer.

In summary, the weight of the evidence points to an association between colorectal cancer and a pattern of food intake involving high dietary fat. The negative outcome of some studies may be partly attributable to methodological difficulties, but it is possible that different cultures differ in the extent to which factors operate either in increasing risk or in providing protection against colorectal cancer.

Prostate Cancer

Results from at least four case-control studies have been published since the 1982 report Diet, Nutrition, and Cancer (NRC, 1982). Heshmat et al. (1985) reported a case-control study of 181 blacks with histologically confirmed prostate cancer and 181 controls from Washington, D.C. Risk enhancement was associated with increased intake of proteins, total fat, SFAs, and oleic acid. These items were related to intake by people ages 30 to 49 years, but not for those age 50 or more.

Graham et al. (1983) compared 262 patients from Buffalo, New York, with two different control series. Risk of prostate cancer increased with increasing ingestion of animal fats. Risk ratios of approximately 2.0 were found for fat consumption greater than 2,100 g/month relative to fat consumption of less than 1,200 g/month.

In a study of 452 prostate cancer cases and 899 age-matched population controls in Hawaii, Kolonel et al. (1988) found a dose-response gradient in risk of prostate cancer associated with intake of SFAs among men 70 years of age and older. The risk ratio was 1.7 for the highest relative to the lowest quartile of SFA intake.

In a follow-up study of 6,763 Seventh-Day Adventists who completed a dietary questionnaire in 1960, Snowdon et al. (1984) found positive associations between fatal prostate cancer and the consumption of milk, cheese, eggs, and meat. There was a graded dose-response between each of the four animal products and risk. The predicted relative risk of fatal prostate cancer was 3.6 for those who consumed all four animal products heavily. Although the dietary instrument used could not provide an estimate of fat consumption, it does seem possible that dietary fat at least contributed to the increased risk of fatal prostate cancer in this study.

In summary, though not entirely consistent, there is evidence of an association of dietary fat, especially from animal sources, with prostate cancer. In one study, there was an association with incidence of prostate cancer, particularly in younger men, and in another, with fatal prostate cancer.

Ovarian Cancer

Two case-control studies on dietary fat and risk of ovarian cancer have produced contradictory results. In one, based on data collected between 1957 and 1965, 274 white women ages 30 to 79 with epithelial carcinoma of the ovary were compared with 1,034 hospital controls (Byers et al., 1983). The investigators found no risk associated with consumption of fat at any age. However, Cramer et al. (1984), who studied 215 white women with epithelial ovarian cancer and 215 control women matched by age, race, and residence, found that the women with ovarian cancer favored, and consumed greater amounts of, foods high in animal fats and significantly less vegetable fat compared with control subjects. There was a significant trend toward an increased risk of ovarian cancer with increasing animal fat consumption. The risk controlling for Quetelet's Index for those with a high intake of animal fat (intake score of 225 or more) was 1.83 compared to those with a low intake (intake score of less than 125).

Details of the methods of dietary assessment used by Cramer et al. (1984) were not given. It is known that in the method used by Byers et al. (1983), relatively limited information was collected on food items contributing to dietary fat.

The geographic correlations for dietary fat and ovarian cancer are somewhat lower than those found for breast, colorectal, and prostate cancer (Armstrong and Doll, 1975). The positive results of Cramer et al. (1984) suggest that further study of this reported association is warranted.

Conclusions

In summary, evidence suggests that the risk of breast cancer and, to a possibly greater extent, of colon, prostate, and ovarian cancers, is associated with dietary fat. Although studies on diet and endometrial cancer have not been reported since 1982, the incidence of endometrial cancer has also been associated geographically with high total fat intake and with the incidence of these other cancers. The data suggest that the endocrine-related cancers of the breast, prostate, ovary, and endometrium, and also colorectal cancer, which itself has some features of an endocrine-related disease, are all associated with diet and possibly with total fat or SFA intake. This evidence supports the evidence from animals in studies reviewed below in the section on Animal Studies.

Dietary Cholesterol

Cholesterol intake was greater in premenopausal cases than in controls in a case-control study of diet and breast cancer (Miller et al., 1978). However, this association was not statistically significant and was quantitatively weaker than an association with dietary fat. In a reanalysis of the same data, dietary cholesterol did not have an effect independent of SFA intake (Howe, 1985). Cholesterol intake by cases was greater than that by controls in a case-control study of colorectal cancer, but the risk ratios were lower than for SFA intake (Jain et al., 1980). Risk ratios were also elevated for dietary cholesterol and colon cancer in a second case-control study, but they were lower for rectal cancer where risk was more strongly associated with dietary protein (Potter and McMichael, 1986). In a study comparing cholesterol intake by Seventh-Day Adventist lacto-ovovegetarians and nonvegetarians, differences were not "striking" (Turjman et al., 1984b). However, Liu et al. (1979), using food disappearance data from 20 countries, found that when they controlled for dietary cholesterol, the associations of dietary fat and fiber with colon cancer mortality that they previously observed were no longer significant.

Dietary cholesterol was associated with an increased risk of lung cancer in a case-control study by Hinds et al. (1983). The association was found in cigarette smokers and in males but not in females. Dietary cholesterol has also been associated with an increased risk of bladder cancer in a case-control study by Risch et al. (1988). In these studies, however, relatively restricted dietary questionnaires were used and it was not possible to determine whether the association with dietary cholesterol was due to other dietary factors associated with cholesterol intake.

No significant association was found between dietary cholesterol and cancer mortality in a 10-year follow-up of the Honolulu Heart Program (McGee et al., 1985), but data for individual cancer sites were not reported. In a cohort study of 89,538 U.S. nurses, no association between dietary cholesterol and breast cancer was found (Willett et al., 1987).

Serum Cholesterol

Some studies suggest that low serum TC is associated with an increased risk of cancer. These are discussed in the following paragraphs. The topic of competing risks is discussed further in Chapter 28.

Data on TC and cancer collected up to 1983 were reviewed by McMichael et al. (1984), who concluded that observational studies provide substantial evidence that preclinical cancer causes a lowering of TC and that limited but biologically plausible evidence suggests that males with low TC despite a diet high in SFAs and cholesterol are at increased risk of colon cancer. Since the studies reviewed by McMichael and colleagues (i.e., from 1983 on), there have been reports of seven studies primarily related to follow-up of cohorts established for the study of cardiovascular disease. In a study of 4,035 residents of California ages 40 to 89, Wingard et al. (1984) found no association between serum TC and cancer morbidity or mortality over a 7-year period for either men or women for any cancer site.

In a 5-year follow-up of 10,940 participants in the Hypertension Detection and Follow-up Program, a small but statistically significant inverse relationship was found between baseline TC and cancer incidence (Morris et al., 1983). When cases diagnosed during the first 2 years were excluded, the association was similar in magnitude but no longer statistically significant. The number of cases did not permit analysis by cancer site. In up to 6 years of follow-up (mean, 3 years) of 10,000 middle-aged men in the Malmo Prevention Program (Peterson and Trell, 1983), TC was inversely related to cancer mortality (44 deaths)—a relationship also observed for the 25 cancer deaths that occurred more than 2.5 years after screening. Serum urate levels at screening in this study were correlated with early but not late (more than 2.5 years after screening) cancer mortality (Peterson and Trell, 1983). As urate levels might indicate proliferation of cancer cells, the association of elevated TC with the late deaths may be due to a factor other than cancer operating at the time of screening.

In the Busselton Community Study in western Australia, 1,564 subjects ages 40 to 74 were followed for 13 years (Cullen et al., 1983). In men ages 60 to 74, but not in men ages 40 to 59 or in women, a negative association was found between serum TC and cancer mortality. In this study, however, it was not indicated whether the association persisted if early cancer deaths were excluded. In New Zealand, 630 Maoris ages 25 to 74 were followed for more than 17 years (Salmond et al., 1985). A significant inverse relationship between cancer mortality and TC was found for men and women considered together. The relative risk derived by comparing cancer mortality in men and women in the 10th and 90th percentiles of TC concentration decreased from 3.0 to 2.4 after excluding deaths during the first 5 years.

In the Seven Countries Study, 11,325 healthy men ages 40 to 59 were followed for 15 years (Keys et al., 1985). Among 477 cancer deaths 5 or more years after TC measurement, there was a significant excess of deaths from lung cancer in the lowest 20% of the TC distribution. Nevertheless, regional comparisons of cancer mortality showed that the highest cancer rates occurred in northern Europe where the TC levels were highest. In contrast, in a cohort study of 92,000 subjects less than 75 years old examined during 1963-1965 and followed by linkage to the Swedish Cancer Registry to 1975, there was a positive association between TC levels and risk of rectal cancer in men (Tornberg et al., 1986). The risk for men with elevated levels of both serum TC (³250 mg per deciliter) and β-lipoprotein (³2.2 g/liter), relative to those with lower levels of both, was 1.62 for colon cancer (95% confidence interval [CI], 1.18-2.22) and 1.70 for rectal cancer (95% CI, 1.18-2.44).

Cancer incidence was determined in 160,135 men and women who were members of a prepaid health plan and whose TC was determined as part of a multiphasic health examination (Hiatt and Fireman, 1986). These subjects were followed from 8 to 16 years. No consistent association of low TC with cancer incidence was found, but cancer incidence was highest among those in the lowest quintile of TC in the first 2 years after the measurement. In the largest study so far reported, a follow-up of the 361,662 men screened for the MRFIT study, there was a significant excess of cancer in those in the lowest decile of TC (<168 mg/dl), which attenuated over time (Sherwin et al., 1987).

Serum TC was assessed in five case-control studies. In one, elevated levels of TC were found in 37 cases of primary brain tumors compared to 74 controls (Abramson and Kark, 1985). The difference was not reduced by controlling for potential confounders (including weight). In another case-control study, TC was measured in 244 patients with adenomatous polyps of the colon, 182 patients with Dukes' A or B colon cancer, and 688 hospital controls (Neugat et al., 1986). The mean TC levels were lower in the cancer cases than in the controls, the Dukes' B cases accounting for most of the difference. There was no difference between the mean levels for those with adenomatous polyps and the controls. After adjustment for nutritional status based on serum albumin level, however, there were no differences among any of the groups. In a nested case-control study within a cohort of members of a prepaid health care plan, 245 newly diagnosed cases of large bowel cancer and 5 matched controls for each case were compared on the basis of TC levels measured during a multiphasic health examination before diagnosis of the cases (Sidney et al., 1986). No direct or inverse relationship between TC and large bowel cancer was found. A fourth case-control study was based on a cohort of 18,995 people examined at a health center between 1970 and 1973 (Gerhardsson et al., 1986). The investigators found medical records for 100 of 176 cancer cases who died by 1979, for 393 of 900 controls still alive in 1979, and for 69 of 153 people who died of cardiovascular disease by 1979. Serum TC levels in the cancer cases were significantly lower than those of controls only in the 2-year period before death.

In a fifth study, a positive association was found between TC levels and the prevalence of adenomatous polyps found during colonoscopies of 842 patients (Mannes et al., 1986). For large bowel adenoma, the odds ratio between the highest and lowest quintiles of serum cholesterol was 1.9 (95% CI, 1.1-3.5) after adjustment for age and was 2.0 (95% CI, 1.1-3.6) after adjustment for body mass index. In a study by Tartter et al. (1984), serum TC was assessed in relation to disease-free survival of 279 colon cancer patients. There was an 11% but nonsignificant lower cumulative disease-free survival at 5 years in those whose TC levels were below the median than in those with levels above the median. In another study, a family history of cancer was found to be positively associated with TC levels in young adults (T. Reed et al., 1986).

Thus, although some epidemiologic studies of cohorts assembled to study cardiovascular disease risk show associations of low serum TC with cancer incidence and mortality, the studies designed specifically to assess the relationship have not in general found an association. When site-specific data are available, they are not consistent.

Mechanism of Action of Dietary Fat

In animals treated with a single dose of a chemical carcinogen to induce mammary tumors, the predominant effect of dietary fat is exerted after exposure to the carcinogen, is positively correlated with the duration of feeding the high-fat diet, and is inversely related to the time between exposure to carcinogen and introduction of the high-fat diet (Dao and Chan, 1983; Rogers and Lee, 1986). This indicates that dietary fat is acting primarily as a promoter (Carroll and Khor, 1975).

As indicated above, some dietary fats, e.g., lard and beef tallow (Rogers et al., 1986; Sylvester et al., 1986a), enhance mammary tumorigenesis when fed only at or before exposure to a carcinogen (Cohen, 1986; Rogers and Lee, 1986; Welsch, 1986). The mechanism of action for this is not known (Rogers and Lee, 1986) but could involve alteration of the rates of formation and inactivation of carcinogens or an influence on their binding to DNA (Hopkins and West, 1976; Wade and Dhardwadkar, 1986).

Many of the hypotheses advanced to account for the promoting action of dietary fat have been concerned specifically with mammary cancer (Carroll, 1986a; Cohen, 1986; Kidwell, 1986; Rogers et al., 1986; Welsch, 1986; Welsch and Aylsworth, 1983). One of the first and most intensively studied hypotheses is that dietary fat perturbs the endocrine system, especially prolactin (Cohen, 1986). After reviewing this literature, Welsch (1986) concluded that there is no consistent support for the concept that dietary fat enhances mammary tumorigenesis by increasing blood levels of pituitary or ovarian hormones. This concept is also weakened by evidence that high-fat diets promote the development of tumors that both are and are not dependent on hormones (Sylvester et al., 1986b: Welsch, 1986). Other studies have shown little difference between target tissue receptor levels and responsiveness to estrogen and prolactin in animals fed moderate and high levels of fat. Animals fed low levels of fat have reduced hormone receptor levels, which may be related to essential fatty acid deficiency (Welsch, 1986).

Other proposed mechanisms are related to the observation that dietary PUFAs promote mammary carcinogenesis in animals more effectively than SFAs (Carroll et al., 1986). As indicated above, this difference appears to be related to a requirement for ω-6 PUFAs, which are precursors of a variety of biologically active eicosanoids, including prostaglandins, thromboxanes, and leucotrienes. Evidence that the promoting effect of dietary fat can be prevented by inhibitors of prostaglandin synthesis (Carter et al., 1983; Hillyard and Abraham, 1979) suggests that the effect of fat may be mediated by eicosanoids. Further support for this idea is provided by the failure of dietary fish oils to promote mammary carcinogenesis (see the section on Breast Cancer). The ω-3 PUFAs in fish oils competitively inhibit the formation of eicosanoids from ω-6 PUFAs, and eicosanoids formed from ω-3 PUFAs differ in structure and biologic properties from those derived from ω-6 PUFAs. However, the prostaglandin synthesis inhibitor carprofen marginally enhanced the induction of mammary tumors induced by dimethylbenzanthracene in rats fed diets containing 5 or 20% corn oil, although the average weight of tumors was reduced significantly (Carter et al., 1987).

Although this is an attractive hypothesis, it is based mainly on speculation. The highly unsaturated fatty acids in fish oils are very susceptible to oxidation. Furthermore, while products of lipoxidation may initiate or promote carcinogenesis (Carroll, 1986a), there is also the alternative possibility that such compounds may inhibit carcinogenesis by exerting a toxic effect on tumor cells.

PUFAs of the ω-6 and ω-3 families can be incorporated into cellular membrane lipids. Since these fatty acids cannot be synthesized by animals, the amounts in tissues are influenced by the dietary supply (Holman, 1986). A model in which dietary unsaturated fatty acids are first taken up by adipose tissue and then transferred to epithelial tissue of the mammary gland has been described by Kidwell (1986).

The effects of dietary fat on the composition and properties of cellular membranes could influence carcinogenesis in a number of different ways. One possibility is an alteration in immune responses, which are known to be affected by PUFAs. These effects may be mediated by alterations in structure and function of cellular membranes or by modulation of prostaglandin biosynthesis (Erickson, 1986). Alterations in membrane structure may affect gap junction-mediated intercellular communication and thus serve as a mechanism by which dietary fat may promote carcinogenesis (Aylsworth, 1986).

These and other studies have led to an interest in diglyceride and protein kinase C as potential mediators of the effects of dietary fat on carcinogenesis. Protein kinase C is activated by diglyceride and also by phorbol ester—a known tumor promoter. Activation of protein kinase C has in turn been associated with cellular proliferation. Since unsaturated diglycerides activate protein kinase C more effectively than saturated diglycerides, it is possible that these compounds could mediate effects of dietary fat on carcinogenesis (Welsch, 1986).

Most proposed mechanisms involve the possible effects of dietary fat on cellular proliferation. Evidence presented by Hillyard and Abraham (1979) suggests, however, that dietary PUFAs may enhance tumor growth by prolonging the life of tumor cells and thus altering the balance between accretion and loss of cells by the tumor.

In epidemiologic studies, total dietary fat shows the strongest correlation with breast cancer incidence and mortality, whereas dietary PUFAs show little or no correlation (Carroll et al., 1986). This has been explained on the basis that human diets normally contain enough PUFAs to satisfy the threshold requirement for tumor promotion observed in animal studies. The positive correlations observed in epidemiologic studies are consistent with the promotional effect of dietary fat observed in animal models (Carroll, 1986a). Therefore, mechanisms by which high-fat diets per se may promote carcinogenesis should be considered.

There is a strong positive correlation between caloric intake and level of dietary fat (Carroll and Khor, 1975), and it is well known that carcinogenesis can be inhibited in animals by caloric restriction (Beth et al., 1987; Cohen et al., 1988; Klurfeld et al., 1987; Kritchevsky, 1986; Reddy et al., 1987; Sylvester, 1986; Tannenbaum, 1959). It is therefore possible that promotion of tumorigenesis by high-fat diets is due to excess caloric intake, but this has not been demonstrated experimentally. The metabolism of animals on high-fat diets is substantially different from that of animals on low-fat diets (Clarke, 1986); some of these differences may favor carcinogenesis.

The mechanisms by which dietary fat affects mammary carcinogenesis are not necessarily the same as those by which it affects other sites. In colon cancer, excretion of secondary bile acids has been positively correlated with the promotion of colon cancer by different types of dietary fat (Reddy, 1986a,b). No specific mechanisms have yet been proposed for effects of dietary fat on pancreatic cancer (Birt and Roebuck, 1986).

Diseases of the Nervous System

Administration of choline or PtdCho can benefit patients with tardive dyskinesia (Gelenberg et al., 1979; Zeisel, 1981). This disorder is a side effect of virtually all antipsychotic drugs marketed in the United States and presumably results from an imbalance between dopaminergic and cholinergic neurotransmission.

Cholinergic neurons are essential for memory and other cognitive functions in rats and humans. An age-related diminution of memory in mice was exacerbated by a diet low in choline and was alleviated by a choline-rich diet (Bartus et al., 1980). Administration of choline reportedly improved memory in serial-learning and selective-reminding tasks among humans who were poor initial learners (Sitaram et al., 1978; Zeisel, 1981).

Supplemental choline (usually as PtdCho) has also been tested in the prophylaxis or treatment of memory disorders associated with Alzheimer's disease in which central cholinergic neurotransmission is known to be deficient (Zeisel, 1981). Preliminary nonblinded studies in Alzheimer's patients yielded positive findings, but most well-controlled, double-blind studies have failed to show clear-cut improvement (Corkin et al., 1982). In a single long-term, double-blind study, approximately 20 g of highly purified PtdCho or placebo given daily for 6 months led to a significant improvement in the oldest third of 50 patients (Little et al., 1985).

Supplemental choline has also been used to treat mania, which may reflect excessive activation of adrenergic receptors in the brain. When PtdCho was given to manic patients with lithium, the improvement in behavior was marked but temporary (Cohen et al., 1980).

PtdCho produced mild improvement in a few patients with Friedreich's ataxia, but no improvement in patients with another ataxic disorder, spinocerebellar degeneration (Barbeau, 1978). Administration of choline only occasionally resulted in reproducible improvement in other ataxic disorders (Lawrence et al., 1980). Some improvement was observed in one patient with the myasthenic syndrome (Kranz et al., 1980). Supplemental choline does not appear to be beneficial in myasthenia gravis, but this difference has not been explored extensively.

The Cardiovascular System

Choline increases release of acetylcholine by the heart in vitro (Loffelholz et al., 1979) but has not been found to affect cardiac function in healthy humans. Rats exposed to transient periods of choline deficiency develop irreversible hypertension (Kratzing and Perry, 1971; Michael et al., 1975). This may be related to renal as well as cardiovascular function.

Lipid Transport

In a study on rabbits fed an atherogenic diet, soya lecithin (1 or 5 g daily) was found to decrease serum cholesterol and atherosclerosis (Kesten and Silbowitz, 1942). Other investigators reported reductions in serum cholesterol in hypercholesterolemic patients treated with soya lecithin (Morrison, 1958; Steiner and Domanski, 1944), but subsequent studies had less encouraging results (Cobb et al., 1980; Davies and Murdoch, 1959; ter Welle et al., 1974). Childs et al. (1981) observed small but significant reductions in LDL-C and an increase in HDL-C in normolipidemic and familial hypercholesterolemic subjects given 36 g of an oral soya lecithin daily. Since soy lecithin preparations used for such studies generally contained less than 30% PtdCho, observed effects could have been due to other components.

High-PUFA PtdCho has been given in some studies. Svanberg et al. (1974) reported a decrease in serum triglycerides and VLDL-C and an increase in HDL-C in hypertriglyceridemic subjects treated with lecithin high in linoleate. Greten et al. (1980) observed no change in serum lipid or lipoprotein concentrations in patients with familial hypercholesterolemia or normal controls treated with lecithin with high linoleate content, but fecal stool excretion was increased. PtdCho has been found to interfere with intestinal absorption of cholesterol both in humans and in laboratory animals (Beil and Grudy, 1980; Rampone and Machida, 1981). In some of these studies, care was taken to distinguish between effects of PtdCho and its constituent PUFAs. Disaturated PtdChos adsorb to surfaces more readily than those with unsaturated fatty acids. Elevated levels of disaturated PtdChos have been observed in some patients with atherosclerotic heart disease and may constitute a risk factor (Gershfeld, 1979).

Dietary Cholesterol

Case-control studies have shown no association between cholelithiasis and consumption of eggs (Sarles et al., 1969) or cholesterol (Scragg et al., 1984). In a cohort study, cholesterol intake did not predict the occurrence of cholelithiasis (Friedman et al., 1966). The lithogenicity of bile (bile saturation index) was increased modestly by large intakes of cholesterol as egg yolk in one experiment (DenBesten et al., 1973) and was increased substantially in another experiment (D.W. Lee et al., 1985), especially in patients who already had gallstones. However, dietary cholesterol did not increase the bile saturation index in three other experiments (Andersen and Hellstrom, 1979; Dam et al., 1971; Sarles et al., 1970). Thus, despite the obvious logic relating cholesterol intake to cholesterol gallstones, little evidence from studies in humans supports the relationship. The variability in results from different experiments, each involving small numbers of subjects, suggests that a few persons may be sensitive to the lithogenic effects of dietary cholesterol but that most persons are not.

Polyunsaturated Fats

A number of experiments in humans between 1958 and 1967 showed no effect of dietary PUFAs on the lithogenicity of bile (Dam et al., 1967; Hellstrom and Lindstedt, 1966; Lewis, 1958; Lindstedt et al., 1965; Sodhi et al., 1967; Watanabe et al., 1962). In 1973, attention was redirected to this issue by an observation from a dietary trial to prevent coronary heart disease. The prevalence of cholelithiasis (as detected at autopsy) was about twice as high in men who had consumed a low-cholesterol, high-PUFA diet as in men who had consumed a high-cholesterol, high-SFA diet (Sturdevant et al., 1973). However, no difference related to PUFAs was observed in another, similar dietary trial (Miettinen et al., 1976).

Grundy (1975) identified 4 of 11 patients with hypertriglyceridemia in whom bile cholesterol saturation was increased by PUFA feeding. However, this difference could not be distinguished from random or individual variability, and it could not be extrapolated to the general population because the subjects were highly selected. More recently, Kohlmeier et al. (1985) found extreme variability in the response of bile cholesterol saturation to a high-PUFA, low-cholesterol diet fed to 12 women, 2 of whom developed much higher cholesterol saturation indices, while the rest experienced no significant change.

In summary, the evidence suggests that a few people may be susceptible to the lithogenic effects of PUFAs when the cholesterol saturation index of bile is increased, but that most humans do not seem to be susceptible. It would be important to identify such subjects if they exist.

Caloric Intake

Numerous epidemiologic studies have shown that gallstones are more frequent among obese persons than among persons of appropriate weight for height (for example, Friedman et al., 1966; Scragg et al., 1984). Obese people have higher bile saturation indices than do nonobese persons, and the condition is reversible (Bennion and Grundy, 1975). This is caused by increased hepatic secretion of cholesterol into the bile (Leijd, 1980) (see also Chapter 21, Obesity and Digestive Diseases). A high-calorie diet also increases the bile cholesterol saturation index (Sarles et al., 1971; Scragg et al., 1984).

Obesity is at least one factor involved in the predisposition of Pima Indians to gallstones (Howard, 1986). Thus, it is well established that obesity possibly resulting from high caloric intake is associated with increased risk of cholelithiasis through greater secretion of cholesterol by the liver and higher bile cholesterol saturation.

Other Factors in Cholelithiasis

Recent research indicates that the lithogenicity of bile may be much more complex than the interrelationships among cholesterol, bile salts, and phospholipids (reviewed by Somjen and Gilat, 1986). Vesicles composed mainly of phospholipids and cholesterol have been observed in the bile of humans, and some proteins may be involved in the processes that influence stone formation. Thus, the cholesterol saturation index, which does not take into account other carriers and other mechanisms stabilizing cholesterol in bile, may be limited in its ability to detect the conditions that favor the formation of gallstones. There is no information concerning the effects of dietary cholesterol, fat, or other nutrients on these newly identified properties of bile that are likely to influence gallstone formation. Such information should lead to new insights into the genetic-environmental interactions involved in cholelithiasis.

Early Observations

Scattered reports between 1912 and 1950 describe gallstones induced in rabbits by feeding them excess cholesterol, but the results were inconsistent and many cholesterol-fed rabbits did not develop gallstones (reviewed by Freston and Bouchier, 1968). Interest in animal models was stimulated by the discovery of diet-induced gallstones in hamsters in 1952. Subsequently, cholesterol gallstones were observed in mice, prairie dogs, and several species of monkeys. These animal models appear to involve different mechanisms. Dam (1971), Freston and Bouchier (1968), Gurll and DenBesten (1978), and Holzbach (1984) reviewed animal models of gallstones.

Hamster

While examining the relationship of nutrition to muscular dystrophy in hamsters, Dam and Christensen (1952) found that a high-carbohydrate, low-fat diet produced gallstones. These investigators carried out a number of experiments (reviewed by Dam, 1971) concerned with the etiology and pathogenesis of this phenomenon. It was eventually determined that the nutritional factor was an essential fatty acid deficiency that did not alter the secretion of bile salts but did increase biliary cholesterol secretion several-fold (Robins and Fasulo, 1973). The enhanced biliary cholesterol secretion was not accompanied by increased tissue cholesterol levels. The addition of cholesterol to the diet yielded varying results, but most investigators have found that increased lithogenicity of bile was associated with dietary cholesterol, e.g., as reported by Singhal et al. (1983). The hamster, which appears to be genetically more susceptible than other rodents to the lithogenic effects of dietary cholesterol (Ho, 1976), continues to be the subject of studies to investigate the pathogenesis and prevention of lithogenic bile and gallstones, but this animal model is of little help in identifying nutritional factors contributing to gallstones in humans.

Prairie Dog

A high-cholesterol (1.2% by weight, derived from egg yolk), high-fat (41% of calories from egg yolk) diet for 6 months invariably produced cholesterol gallstones in prairie dogs (Cynomys ludovicianus) (Brenneman et al., 1972). The bile became lithogenic primarily through the increased secretion of cholesterol. The mechanism was interpreted as a massive overloading of the hepatic system for metabolism of cholesterol, apparently because the prairie dog absorbed cholesterol efficiently from the intestine. The overload occurred quite rapidly after beginning the diet, since bile cholesterol was increased within 36 hours, cholesterol crystals were found in the gallbladder within 5 days, and most animals had gallstones within 14 days (DenBesten et al., 1974). A semisynthetic diet with cholesterol was even more effective in inducing gallstones (B.I. Cohen et al., 1986). Susceptibility was not due to lack of responsiveness of 7α-hydroxylase in the liver (Howell et al., 1983). Thus, this animal model also may be useful in studying the pathogenesis of gallstones.

Ground Squirrel

MacPherson et al. (1987) produced cholesterol gallstones rapidly in ground squirrels (Spermphilus richardsonii) by giving them a cholesterol-enriched (2%) rat chow diet. The features of this model appear to be similar to those of the cholesterol-fed prairie dog.

Mouse

Tepperman et al. (1964) produced cholesterol gallstones in mice by feeding them a diet containing 1% cholesterol, 0.5% cholic acid, and 31% lard. There has been no subsequent exploration of this phenomenon in inbred strains of mice.

Nonhuman Primates

Osuga and Portman (1971, 1972) and Osuga et al. (1974) produced lithogenic bile and cholesterol gallstones in squirrel monkeys (Saimiri sciureus ) with a diet containing 0.9 mg of cholesterol per kcal and 45% of total calories from butter. They concluded that dietary cholesterol was the major factor in the production of these effects (Osuga et al., 1976). Armstrong et al. (1982) found that dietary cholesterol increased the cholesterol bile saturation index of squirrel monkeys, whereas dietary fat saturation (corn vs. coconut oil) had no independent effect. The effect of cholesterol was greater when it was fed with PUFAs (Melchior et al., 1978; Tanaka et al., 1976). African green monkeys (Cercopithecus aethiops) also developed a more lithogenic bile when fed cholesterol with PUFAs (St. Clair et al., 1980). In contrast, the bile of Cebus monkeys, which do not develop gallstones from dietary cholesterol, was much less influenced by diet.

Baboons (Papio sp.) occasionally develop cholesterol gallstones spontaneously (Glenn and McSherry, 1970), and their bile is frequently supersaturated with cholesterol (McSherry et al., 1971a). The bile of these animals has a cyclic variation that reflects the enterohepatic circulation of bile salts (McSherry et al., 1971b).

Summary

The animal models of cholelithiasis present a mixed picture. Several species of rodents (but not the rat) are susceptible to the lithogenic effects of dietary cholesterol, but they require large doses of cholesterol and are accompanied by high levels of serum cholesterol. The common denominator of these rodent models is enhanced secretion of cholesterol into the bile, as in humans with cholelithiasis, but the underlying genetically determined metabolic characteristic that makes them susceptible to dietary cholesterol is not known. Among nonhuman primates, the squirrel monkey is similar to humans in its sensitivity to dietary cholesterol. The baboon appears to simulate more closely the human pathophysiology. The extreme susceptibility of some species to dietary cholesterol does not seem to have any parallel in humans with cholelithiasis.

Childhood Atherosclerotic Lesions

Atherosclerosis and its clinical sequelae are found primarily in adults. In rare cases (about 1 in a million), however, children with homozygous familial hypercholesterolemia develop severe atherosclerosis and clinical CHD. Holman—a pioneer in the exploration of atherosclerosis in childhood (Holman et al., 1958)—in 1959 presented a paper posing the question, ''Atherosclerosis—a pediatric nutrition problem?" at the Ninth International Congress of Pediatrics (Holman, 1961). At that time, the question could not be answered with certainty, because there was no evidence supporting the relationship in children between diet and subsequent atherosclerosis, in contrast to the considerable amount of data supporting such an association in adults. Abnormal lipid deposits (fatty streaks), believed by many to be the precursors of raised atherosclerotic lesions, were found in the aortas of all children by 3 years of age in all populations around the world, regardless of dietary fat and cholesterol intake (Strong et al., 1958; Tejada et al., 1968). Furthermore, these simple lipid deposits had not been firmly linked to the fibrous plaques that lead directly to arterial occlusion (see Chapter 19).

Progression of Fatty Streaks to Fibrous Plaques

Long before there was any concern about hyperlipidemia or atherosclerosis in children, Enos et al. (1953) described advanced coronary artery lesions in young American adults in a landmark study of Korean battlefield casualties. In this study, 77% of men averaging 22 years of age had grossly detectable coronary artery lesions, and 15% had coronary artery lesions that obstructed the lumen by 50% or more. This remarkable prevalence of coronary atherosclerosis in young men was confirmed in other studies of young people who died from accidental causes in the United States (Strong and McGill, 1962).

Subsequent intensive examination of large numbers of children and young adults showed that a continuous progression of lesions could be traced from fatty streaks to fibrous plaques (Geer et al., 1968; McGill, 1984; Robertson et al., 1963; Stary, 1987a,b; Strong and McGill, 1969; reviewed by McGill et al., 1963, 1980). These observations and many more left little doubt that many fatty streaks in the coronary arteries and some fatty streaks in the aortas of children are converted to fibrous plaques in adolescents and young adults.

Immediate Effects

Preweaning (suckling) infants are quite sensitive to the plasma cholesterol-raising (or lowering) effects of dietary fat and cholesterol. Approximately 50% of the calories in human breast milk is supplied by fat, half of which is SFA (Lo, 1985; reviewed by Jensen et al., 1978). Its cholesterol content averages 11 mg/dl (range, 3 to 29 mg/dl) (Picciano et al., 1978; reviewed by Jensen et al., 1978). In comparison to human milk, cow's milk has more SFA (65.6 versus 48.2% of total fatty acids), less MUFA (30.3 versus 39.8%), and much less PUFA (4.1 versus 12.0%) (Lo, 1985). In contrast, commercially prepared infant formulas are made with soy, corn, safflower, and coconut oils. Although these formulas provide the same proportion of calories as fat as does breast milk, they contain a higher proportion of PUFAs (up to 50% as linoleic acid) and virtually no cholesterol (Benson, 1981; Jensen et al., 1978). In several studies, as might be anticipated from their higher intake of cholesterol and SFAs, breast-fed infants had higher serum cholesterol concentrations than did formula-fed infants (Carlson et al., 1982; Pomeranze, 1961; reviewed by Wissler and McGill, 1983).

In a study of the nutrient composition of diets given to infants over the age of 6 months, Montaldo and coworkers (1985) compared solid food diets supplemented with whole milk from cows or with formula. Using data from NHANES II, they found that infants fed the cow's milk received less iron and linoleic acid. Manipulation of the fat and cholesterol content of infant formulas showed the anticipated effects (Sweeney et al., 1961). Similar observations have been made in animals, particularly nonhuman primates, whose breast milk is similar in composition to that of humans (Mott et al., 1978). The effect of plasma cholesterol concentrations in the suckling child on the subsequent development of atherosclerosis is totally unknown. It has generally been believed not to be significant because of the limited duration of exposure and the overriding importance of other nutritional considerations during the first year of life.

Deferred Effects

In 1972, Reiser proposed that a high cholesterol intake before weaning, such as that provided by breast milk, would induce a higher level of tolerance for dietary cholesterol in adulthood and presented results from studies in swine and rats to support this hypothesis (Reiser and Sidelman, 1972; Reiser et al., 1979). Tests of this hypothesis in other animal species, including nonhuman primates, have generally been negative (Mott et al., 1982; reviewed by Innis, 1985). However, Mott et al. (1982) found that breast-fed infant baboons had lower plasma HDL-C and apo AI concentrations when they reached adolescence than did formula-fed baboons, regardless of the cholesterol content of the formula. In another experiment, Lewis et al. (1988) also found that breast-fed baboon infants reared on an atherogenic diet had lower HDL-C levels and more extensive atherosclerosis at age 5 than baboons fed a formula as infants and reared on an identical diet. No similar effects have been detected in humans.

Summary

Plasma lipid and lipoprotein concentrations in infants and children under 2 years of age are quite sensitive to dietary fat and cholesterol, but there is no evidence that elevated plasma lipoproteins in this age group are factors in the etiology of atherosclerosis. Other nutritional considerations are more important during this period. Results from experiments in animals indicate that breast feeding, as compared to formula feeding, exerts a long-term deferred effect on lipoprotein metabolism that continues into adulthood, but no such effect has yet been detected in humans. Therefore, the remaining concerns regarding dietary fat and cholesterol during childhood are focused on children over the age of 2 years.

Age and Sex Distributions

Until about 1970, descriptive studies of the distribution of serum cholesterol and lipoprotein cholesterol concentrations focused on adults. In the early 1970s, several research groups began to survey these variables in U.S. children (Berenson et al., 1979; Frerichs et al., 1976, 1978a,b; Friedman and Goldberg, 1973; Lauer et al., 1975; Morrison et al., 1979a,b; NHLBI, 1978; Wilmore and McNamara, 1974). Results of these studies indicate that TC concentrations at birth were approximately 70 mg/dl (half of it as HDL-C) and rose rapidly to approximately double that value at 1 year and increased slowly thereafter to about 160 mg/dl (roughly two-thirds LDL-C and one-third HDL-C) at 10 years of age. Between 12 and 18 years of age, TC in boys declined slightly to about 150 mg/dl. The 10 mg/dl decrease was accounted for primarily by HDL-C and was associated with an increase in plasma testosterone levels (Kirkland et al., 1987). Serum TC concentrations and lipoprotein cholesterol distributions in girls remained at about the same level. At any age, the frequency distribution of TC levels was skewed to the high end similar to that of the adult population. In the 10- to 14-year age group, the fifth percentile levels were approximately 120 mg/dl, and the 95th percentile levels were about 200 mg/dl (Lipid Research Clinics Population Studies Data Book, 1980).

Dietary Intakes by Children

In the Bogalusa Heart Study, which periodically screens a pediatric population of approximately 5,000 children from a semirural community, Berenson et al. (1988) found that the schoolchildren's daily diet contained an average of 38% fat and 300 mg of cholesterol; the P:S ratio was 0.4. These data are similar to the findings of the National Health Survey (Carroll et al., 1983).

Emmons (1986) studied the nutritional adequacy of diets consumed by low-income people participating in the Special Supplemental Food Program for Women, Infants, and Children (WIC Program) and found that both white and black children consumed approximately 35% of their calories as fat. However, black children consumed more animal fat than vegetable fat and thus had a diet with a lower P:S ratio than that of white children. On average, therefore, it appears that children in the United States, especially those in lower socioeconomic groups, consume diets that tend to increase their LDL-C.

The H.J. Heinz Company of Canada conducted a longitudinal survey of the eating habits of infants and children (Leung et al., 1984). The food intake of more than 400 4-year-old children was obtained from data collected from a 4-day dietary record. This study documented that preschool Canadian children obtained 34% of their calories as fat. Their snacks, unlike typical processed food snacks in the United States, had the lowest caloric density of any of the meals, suggesting that they were composed mainly of carbohydrates. These Canadian data contrast strongly with the diet of U.S. children (Carroll et al., 1983).

The Lipid Research Clinics Program Prevalence Study presented the dietary data for white American males and females between the ages of 6 and 25 years. Enrolled within that study were more than 1,100 schoolchildren. The percentage of calories as fat was 39% for the boys and 38% for the girls. The P:S ratio was 0.4 for both sexes (Beaglehole et al., 1980).

Ecological and Familial Associations

Serum cholesterol levels in U.S. children were considerably higher than those in children of other regional groups or countries where CHD was much less frequent (Knuiman et al., 1980; Savage et al., 1976; Sullivan et al., 1987). The families of children with high TC levels had a greater incidence of deaths from CHD than did the families of children with low TC levels (Schrott et al., 1979). Families of children with low HDL-C levels also had an excess of deaths from CHD (Bodurtha et al., 1987). Conversely, approximately one-third of the children selected because their parents had precocious CHD had serum TC or triglyceride levels in the upper percentiles for their age (Chase et al., 1974; Glueck et al., 1974; Tamir et al., 1972).

Tracking

Of importance in studies of childhood serum lipid levels is tracking, that is, whether children maintain similar levels (relative to one another) of serum total and lipoprotein cholesterol as they grow. Tracking is important in determining whether a child in a high percentile of serum cholesterol concentration would remain in the same high percentile several years later when he or she entered adulthood and therefore would be at high risk of CHD. Most studies have demonstrated that serum TC concentrations track to a moderate degree during childhood (Clarke et al., 1978; Frerichs et al., 1979; Laskarzewski et al., 1979; Webber et al., 1983, 1986) and from childhood into young adulthood (Orchard et al., 1983). For example, in the study by Orchard and colleagues, 49% of children with a TC level in the top quintile at 12 years of age were in the top quintile when 21 years old, and the correlation coefficient between the 12- and 21-year values was .52. Children who were obese had both higher TC and blood pressure and increased left ventricular mass (Frerichs et al., 1978a; Schieken et al., 1981).

Walter and Hofman (1987) studied the differences in the distribution of risk factors in children stratified by race and socioeconomic status in the "Know Your Body" program. They found that white children in the lower socioeconomic strata had the highest level of risk as defined by blood pressure, physical conditioning, and serum lipoprotein levels. These children also reportedly consumed the most atherogenic diets of the groups in the study. Thus, the evidence indicates that children with high TC values will be likely to become adults with high serum cholesterol values.

Observational Studies

Observational cross-sectional studies of diet and serum lipids in children have yielded results similar to those obtained from adults—correlations that are either zero or weak (Frank et al., 1978; McGandy, 1971; Morrison et al., 1980; Rasanen et al., 1978; Vartiainen et al., 1982). For example, Frank et al. (1978) studied 185 10-year-old children and found a positive but statistically insignificant association of TC concentration with total fat and SFA intake and no association with dietary cholesterol. Some statistically significant correlations emerged from other surveys, but they were low in magnitude (Frank et al., 1986).

Experimental Results

Like studies in adults, early experiments on the effects of dietary cholesterol on serum TC levels in children were negative (Heymann and Rack, 1943). In later studies, positive results were obtained. In one study conducted in a boys' boarding school, meals were modified to reduce total fat intake from 38 to 33% of calories, SFAs from 16 to 10% of calories, and cholesterol from 720 mg/day to 380 mg/day; PUFA intake was increased from 3 to 13% of calories (Ford et al., 1972; McGandy et al., 1972). The average TC concentration of the 63 13- to 18-year-old boys fell from approximately 190 mg/dl, equivalent to that of other age-matched boys in similar schools (McGandy et al., 1972), to about 170 mg/dl—a reduction of 15%. The boys with the highest TC levels experienced the greatest decreases. Several other dietary interventions in children subsequently showed a similar responsiveness of TC concentrations to reductions in dietary SFAs and cholesterol (Nestel et al., 1979; Puska et al., 1985; Stein et al., 1982; Vartiainen et al., 1986). Thus, adolescent children are as responsive to dietary fat modification as are adults.

Cholesterol as an Essential Nutrient

Since cholesterol is an essential component of mammalian tissues, and since it is a precursor of steroid hormones, dietary cholesterol has been hypothesized to be necessary for optimum growth and development, particularly of the central nervous system. However, there is no experimental or observational evidence to support the concept that cholesterol is an essential nutrient.

The well-known capacity of mammalian cells to synthesize cholesterol suggests that dietary cholesterol is not required for normal growth. However, there have been no experiments testing the effects of cholesterol deprivation on children or on growing animals, probably because the hypothesis is considered to be very unlikely. As in other situations, proving the lack of a requirement is more difficult than proving that a requirement exists.

Total Fat Requirements of Children

Evidence indicates that there is no need for a fat intake of more than 10% of calories, provided the diet contains an adequate amount of essential fatty acids. The requirement for essential fatty acids in infants is met if they ingest 3% of their total calories as linoleic acid. This requirement drops to between 1 and 2% of total calories as linoleic acid in adults (NRC, 1980)—an amount well below the usually recommended 30% of calories from fat or even the 20% of calories from fat sometimes recommended. Energy requirements can readily be met by an increased intake of carbohydrate.

Observational Studies

Friedman and Goldberg (1976) found no differences associated with a low-fat, low-cholesterol diet for a number of developmental and biochemical variables of children. Dwyer (1980) discussed the preparation of diets for children that would fulfill the recommendations for lowering fat and cholesterol intake, but would also meet the Recommended Dietary Allowances for other nutrients. If care is not taken, she pointed out, deficiencies in the intakes of some vitamins and minerals might result. These and other aspects of safety have been reviewed by Kwiterovich (1980, 1986) and by Voller and Strong (1981). It is clear that caloric intake needs to be monitored to ensure its adequacy, if fat intake is decreased, in order to maintain optimal growth and development.