and palmitic. (16:0) acids exert disparate effects on cholesterol metabolism, whereas the ability of linoleic acid (18:2) to decrease total plasma cholesterol.
Dietary
fatty
acid
thresholds
K. C. HAYES1 AND PRAMOD Foster Biomedical
Research
and
KUOSLA Brandeis Utrk%
Laboratory,
ABSTRACT Results obtained with cebus monkeys indicate that dietary myristic (14:0) and palmitic (16:0) acids exert disparate effects on cholesterol metabolism, whereas the ability of linoleic acid (18:2) to decrease total plasma cholesterol displays an upper limit or threshold. Reanalysis of published data suggests a similar situation pertains in humans. In agreement with an earlier human study, 14:0 appears to be the principal saturated fatty acid that raises plasma cholesterol whereas 18:2 lowers it. Oleic acid (18:1) appears neutral. The effect of 16:0 may vary. In normocholesterolemic subjects consuming diets containing 300 mg/day of cholesterol, 16:0 appears to be without effect on plasma cholesterol. However, in hypercholesterolemic subjects (>225 mg/dl) and especially those consuming diets providing cholesterol intakes of 400 mg/day, dietary 16:0 may expand the plasma cholesterol pool.-Hayes, K. C.; Khosla, P. Dietary fatty acid thresholds and cholesterolemia. FASEB J. 6: 2600-2607; 1992. Key Words: fatty acids acid . cebus monkeys
plasma cholesterol
myrislic acid
linoleic
THE KEYS (I) AND HEGSTED (2) regression equations define a simple, linear relationship between dietary fat saturation and cholesterolemia, empirical observation of the plasma cholesterol response to a variety of dietary fats suggests that this relationship is not always satisfactory and may be nonlinear, i.e., thresholds may exist that modulate the response at low or high intakes of either saturates, polyunsaturates, or both (3-5). In addition, a recent summary of the world literature (227 reports) by Hegsted (6) identified two unexplained dietary interactions, one involving saturated fatty acids x dietary cholesterol, the other saturated x polyunsaturated fatty acids. Examination of these relationships in normocholesterolemic nonhuman primates and humans has revealed three relevant points. First, lauric plus myristic acids (12:0 + 14:0) were more cholesterolemic than palmitic acid (16:0), increasing low density lipoprotein (LDL) more than high density lipoprotein (HDL) (4, 7). Second, the partial exchange of 16:0 for 18:1 in normocholesterolemic humans had no effect on plasma cholesterol or the LDL/HDL ratio (8). Third, 16:0, oleic acid (18:1) and linoleic acid (18:2)-rich oils exerted equal effects on total LDL in normocholesterolemic monkeys when the 18:2 threshold represented at least 4% energy and the diet contained minimal 14:0 and no cholesterol (9). By contrast, when the same three oils were fed to cholesterolsensitive monkeys accompanied by a high level of dietary cholesterol (10) or were fed to hyper-cholesterolemic humans (mean plasma cholesterol 263 mg/dl) (11), the 16:0-rich oil appeared to elicit a higher LDL and total plasma cholesterol than either the monoene or polyene-rich oils. It may seem paradoxical that the same fatty acid (e.g., 16:0) would exert different effects on the plasma cholesterol ALTHOUGH
2600
ch#{243}Iestemiemia ,MMsachmet
02254,
USA
pool, primarily LDL, under different dietary or host circumstances. The modified circumstances appear to involve the inherent LDL receptor (LDLr) activity of the host and/or the effect of key dietary fatty acids directly on LDLr activity (12-14). For example, in certain rat (13), hamster (12), and monkey (10) experiments a generalized saturated fat effect (elevated cholesterol) typically occurs only if cholesterol is fed to partially down-regulate the LDLr, whereas a high intake of coconut oil (14:0 + 12:0) is the only saturated fat demonstrated to consistently elevate cholesterol while specifically reducing LDLr activity in the absence of dietary cholesterol (13, 14). It has also been reported that diets high in either 18:2 or 18:1 enhance LDLr activity in situations where LDL receptors have been down-regulated by dietary cholesterol (12), but the possibility was not excluded that the 18:2 supplied by the 18:1-rich diet exceeded the limiting threshold of 18:2 needed to lower the LDL cholesterol under those circumstances. To gain further insight into these relationships we have analyzed data accumulated in our laboratory from 16 dietary fat feeding trials in cebus monkeys during the last 6 years (1986-1991). Using the approach originally used by Hegsted et al. (2) to quantitate the effects of dietary fat on serum cholesterol in humans, we subjected our data to multiple regression analysis to ascertain the ability of specific dietary fatty acids to predict the plasma cholesterol concentration. The cebus has been used as the primate of choice because its plasma cholesterol is extremely sensitive to variations in dietary fat saturation. Although more sensitive than humans in the magnitude of their response, cebus monkeys respond in the same manner (4).
METHODS The database used for the analyses described herein represent a summary of results from various feeding studies during the last 6 years in which cebus monkeys were fed 16 cholesterol-free purified diets (Table 1: diets 1 and 2 from ref 15; diets 3-7 from ref 4; diets 8-10 from ref 9; diets 11_13;2 diets 14-16 from ref 16). The diets provided either 31 or 40% of the energy as fat with the range in % en (or % energy) from the most predominant fatty acids as follows: 12:0 (0-19%); 14:0 (0-7.5%); 16:0 (2-i7%); 18:0 (0.7-1.9%); 18:1 (3-30%); 18:2 (1-29%); 18:3 (0-1.1%). In all cases total cholesterol was determined enzymatically on fasting plasma samples. The final data set comprised 129 cholesterol values generated from a group of 16 monkeys fed a total of 16 different diets. The composition of the diets has been detailed previously (4, 9, 15). The dietary protein source was either lactalbumin (diets 1-7, 11-16) or lactalbumin and casein (diets
‘To whom correspondence should be addressed. Khosla and K. C. Hayes (unpublished observations).
2p
0892-6638/92/0006-2600/$o1.50.
© FASEB
TABLE
1. Diets, percentageenergyfrom dietasyfatty acids, and the observedplasma cholesterol Dietary fattyacids
Diets
12:0
14:0
16:0
18:0
18:1
18:2
18:3
Plasma cholesterol’
(n)b 1(4) 2 (4) 3 (8) 4 (8) 5 (8) 6 (8) 7 (8) 8 (9) 9 (9) 10 (9) 11(10) 12 (10) 13 (10) 14 (6) 15 (12) 16 (6)
0.00 14.73 14.82 7.38 4.15 0.06 0.12 0.64 0.92 0.60 19.12 0.08 0.16 0.00 0.00 0.00
0.06 3.69 0.68 7.78 18.57 0.19 6.88 4.00 1.27 3.35 0.78 0.00 5.83 3.32 1.02 2.91 2.64 0.28 2.98 2.67 0.93 11.47 4.96 0.37 1.80 7.78 1.12 11.53 4.12 0.25 0.31 12.49 1.27 11.47 4.77 0.31 0.22 7.25 1.21 12.74 8.43 0.84 0.52 2.08 1.00 29.64 5.76 0.08 0.56 16.28 1.92 15.64 3.92 0.16 0.52 2.52 0.12 5.48 29.12 0.16 7.52 4.28 1.32 3.76 3.40 0.36 0.40 16.12 1.64 14.80 6.16 0.40 0.28 9.36 1.56 16.44 10.88 1.08 0.47 16.96 1.52 8.84 3.22 0.00 0.31 11.19 1.27 13.80 3.44 0.00 0.25 7.41 1.21 17.95 3.63 0.00 aThe composition of the purifieddiets has been described elsewhere (4, 9, 15). Diets were fed with fat contributing either 31%
152 263 246 191 186 161 151 142 145 118 233 155 145 183 177 176 energy
± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
10 25 17 8 13 11 9 6 6 5 10 8 8
11 5 10
(diets
1-7
and 14-16),or 40% energy (diets8-13). All dietswere cholesterol-free. The fattyacidcomposition of each dietwas determined by GLC. Dietary fats were formulated (using either a single oil or blends of oils) as follows: 1, corn oil; 2, coconut oil; 3 and 11, 90% coconut oil/10% soybean oil; 4, 45% coconut oi/40% high-oleic safflower oil/15% soybean oil,5, 45% palm oil/22% coconut oilJ2O% high-oleicsaffloweroil/13% soybean oil; 6 and 12, 90% palm oil/lO% soybean oil; 7 and 13, 45% palm oil/40% soybean oil/15% high-oleic safflower oil; 8, high-olcic safflower oil; 9, palm oil;10, high-linoleic safflower oil; 14, 95% palm stearinl5% high-linoleic safflower oil; 15, 54% palm stearinl43 % olive oilI3% high-linoleic safflower oil; 16, 24% palm steanin/75% olive oil/I % high-linoleic safflower oil. 6Number of monkeys. ‘mg/dl plasma. Mean ± SEM. 8-10). The fat source fed was either a single oil (diets 1, 2 and 8-10) or a blend of oils (diets, 3-7 and 11-16) designed to isolate specific fatty acid effects. To ensure that the diets were cholesterol-free, only vegetable oils were used. These included coconut oil, corn oil, soybean oil, hi-oleic safflower oil, hi-linoleic safflower oil, palm oil, and olive oil. Except for two diets (each fed to four different animals), all diets were fed to 6-12 monkeys for 6- to 12-wk periods. For all diets the fatty acid composition was determined by GLC (4). In an attempt to define the plasma cholesterol response in terms of its dietary fatty acid descriptor (or descriptors), the observed plasma cholesterol (mean for a given diet) was regressed against the dietary energy (% of total) contributed by a specific fatty acid (or acids) to generate the appropriate multiple regression equations. With seven dietary variables (the seven major fatty acids), a total of 127 possible regression equations resulted.3 Calculations were carried out on a Macintosh Plus computer (Apple Systems Inc., Cupertino, Calif.) using the Statview 512k (Brain Power Inc., Calabasas, Calif.) and Cricket Graph (Cricket Software Inc., Philadelphia, Pa.) statistical packages.
sidered (Table variables was: PC
Cebus
cholesterol
profile
129 individual cebus plasma cholesterol responses 16 cholesterol-free diets averaged 174 ± 4 mg/dl (mean ± SEM) with a range of 96-355 mg/dl, indicating that the inherent cholesterol level for this group of monkeys was essentially normal. to
The all
Response to total fatty acids
saturated
For the initial calculation, (SATS) and polyunsaturated
and
polyunsaturated
137
+
multiple
3.28 ESATS
regression
-
1.19
using
EPOLYS
these two
(Cl)4
The multiple regression coefficient was 0.822 (P = 0.0007). This equation differs somewhat from those described previ-
ously for humans (1, 2) in that it assigns a greater cholesterol-raising property to the SATS, and a lesser cholesterol-lowering power to POLYS. Equation Cl, however, only accounted for 68% of the observed variation in plasma cholesterol (r2 = 0.676). Equation C2 through equation C6 (Table 2) describe the characteristics of the various regression equations based on total SATS and/or POLYS or these in combination with monounsaturated oleic acid (18:1). None of the relationships described in Table 2 were considered satisfactory either because of their low predictability or (in the case of Eq. C4 and Eq. C6) the fact that the role of the SATS was highly improbable and in contrast to that pertaining to humans (1, 2). Multiple
RESULTS
=
2). The
regression
analysis
Using the % energy from individual fatty acids as variable, a total of 127 multiple regression equations were generated. The most predictive (biologically meaningful) regressions had the following characteristics: 1) They predicted at least 80% of the variation in the total plasma cholesterol response, i.e., exceeded that attributable to myristic acid alone (see Table 3); 2) the intercept term, I, approximated the overall inherent cholesterol value attributed to this group
3For seven variables the total number of possible regression equations is given by 2” 1 where n = 7. 4Where PC denotes plasma cholesterol (mg/dl), denotes % energy from total saturated or polyunsaturated fatty acids.
-
the role of the total saturated (POLYS) fatty acids was con-
DIETARY FATTYACIDS AND CHOLESTEROLEMIA
2601
TABLE 2. Characteristics of regressionequations in cebusmonkeysfed 16 dietary fats basedon total saturated and polyunsaturated fatty acids with or without inclusion of monoenes Variables E”
SATSb
Cl C2 C3 C4 C5 C6
3.28 3.90
18:1
-
-
-1.19
-
-
-3.75 - 1.01 -3.93
-
137 118 202 245 137 252
-3.45 -3.74
-
0.23 3.44
1d
POLYS’
-
-3.92
SE
P
24.68 24.83 32.78 19.60 25.05 18.86
0.0007 0.0002 0.0105 0.0001 0.0008 0.0001
r2 0.676 0.646 0.384 0.811 0.666 0.811
bTotal saturated fatty acids (sum of 12:0, 14:0, 16:0, ‘Total polyunsaturated fatty acids (sum of 18:2 and 18:3). dlntercept of the regression equation (basal plasma cholesterol concentration). r2 is a measure of the total variance explained by the regression equation. SE is the standard error about the regression line and corresponds to what Hegsted et al. (2) refer to as the ‘SD from the regression.” The regression in Cl is PC - 3.28 ESA 1.19 EPOLYS + 137, where PC = plasma cholesterol (mg/dl), ESATS is the % energy from the saturated fatty acids and EPOLYS is the % energy from polyunsaturated fatty acids. Therefore it follows that (on the basis of Eq. Cl) for any two diets, the change in plasma cholesterol (PC) will be described by PC 3.28 ESATh 1.19 EPOLYS, where ESATs is the change in % energy from the saturated fatty acids and EPOLYS is the change in % energy from the polyunsaturated fatty acids.
for palmitic acid was described by 32 equations. Similarly, the coefficient for stearic acid was positive in 23 of the 54 equations and negative in 31, but the range of intake was too low to develop a reliable coefficient. Of the 64 equations including lauric acid, the coefficient was positive (n = 32) or negative (n = 32), depending on the inclusion (negative) or exclusion (positive) of myristic acid in the regression equation. In addition, the lauric acid response was biased by its absence from 4 of the 16 diets. On the basis of individual fatty acids (Table 3), myristic acid alone explained 80% of the variation in plasma cholesterol (Eq. C8) whereas linoleic acid accounted for 66% of the observed variation (Eq. C12).
‘Equation.
and
18:0).
of cebus (174 ± 4). Otherwise, a strong predictive equation would likely apply to certain types of fatty acid combinations but not to others; 3) addition of any fatty acid to the simplest best fit equation improved r2, I, or both. Role
of individual
fatty
acids
Among the saturated fatty acids (12:0-18:0), myristic acid was by far the most hypercholesterolemic. In fact, myristic acid was the only fatty acid identified as cholesterol-raising in all 65 regressions in which it was included, being assigned a positive coefficient in all cases (range 3.1-36.0). The variation in the observed plasma cholesterol accounted for by regressions that included myristic acid ranged from 80 (myristic acid alone) to 97.5% (all seven fatty acids). By contrast, only 23 of the 55 regression equations that included palmitic acid found it to be hypercholesterolemic (positive coefficient). Furthermore, the cholesterol-raising property of palmitic acid (if any) was significantly less than the cholesterol-raising property of myristic acid (i.e., the maximum value for the coefficient for palmitic acid was 3.44 whereas for myristic acid the minimum value was 3.1). A hypocholesterolemic role (negative coefficient; range -0.12 to -7.65)
TABLE
PC
=
151
+
PC
=
240
-
14E14o 90.6 log E182
r2
=
0.80
(C8)
r2
=
0.66
(C12)
No significant relationship was observed when palmitic, stearic or linolenic acids were considered alone. The regression for oleic acid alone was negative but weak (r2 = 0.32). Consistent with previous reports (1, 2), myristic and linoleic acids had opposite effects, i.e., cholesterol-raising and cholesterol-lowering, respectively (Fig. 1). In addition, the logarithmic nature of the response to 18:2 indicated that a nonlinear relationship existed between increasing 18:2 intake and the observed plasma cholesterol (Fig. lb). Although not a true dose-response curve for 18:2 (because increased % energy from 18:2 was simultaneously coupled with decreased
% energy from other fatty acids), Fig. lb nevertheless serves to illustrate the physiological impact of 18:2 on the nonlinear relationship described by Eq. C12. It is apparent from Fig. lb that increments of 18:2 reach a threshold beyond which further increases exert minimal effect on the plasma cholesterol level. The simplest, most inclusive multiple regression equation obtained by including two or more fatty acids (Eq. C14) revealed a regression coefficient of 0.95 7, and was based on the % energy derived solely from myristic and linoleic acids, which explained 92% of the variation (r2) in plasma cholesterol. The standard error about the regression was 12.6 mg/dl. The constant term (192) represents the baseline cebus plasma cholesterol value independent of any dietary fat effect. This equation is superior to Eq. Cl, in which only total SATS and POLYS were considered. The observed plasma cholesterol values plotted against the plasma cholesterol predicted by Eq. C14 are depicted in Fig. 2.
PC
=
192 + 10 Eit:0
-
48 log E182
T2
=
0.92
(C14)
3. Coefficientsfor individual fatty acid regressionin equationsfor cebus monkeysfed 16 dietary fats Variables
E”
12:0
C7 C8 C9 ClO Cli Cl2 Cl3
5.46
16:0
14:0 -
14.02
-
18:0
18:1
18:2
18:3
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
1.94
-
-
-
-
-
-
-
-
-
-3.39
-
-
-
-
-
-
-
-90.60’
-
-
-
-
-
-
-
-29.93
-
5.81
I’
r2
155 151 192 170 216 240 185
0.752 0.800 0.064 0.004
0.324 0.655 0.050
5E
20.10 18.60 40.40 41.67 34.32 24.51 40.68
“Equation. Regressions C7, C8, Cli, C 12 were significant at P < 0.001. Intercept of the regression equation. r2 is a measure of the total variance explained by the regression equation. SE is the standard error around the regression lines. ‘Indicates a log function.
2602
Vol. 6
May
1992
The FASEBJournal
HAYES AND KHOSLA
Inclusion of one or two additional fatty acids after 14:0 and 18:2 had been considered failed to improve the predictability. Therefore, Eq. C14 was deemed superior to the rest. Again the logarithmic term in Eq. C14 indicates a nonlinear response. The superior simplicity of Eq. C14 is demonstrated by equations that added either 16:0 (Eq. C15) or 18:1 (Eq. C16) to the original Eq. C14. Both revealed intercepts (1) that deviated farther from the actual mean than Eq. C14, and r2 was only minimally improved by Eq. C16 (where the intercept deviated considerably): PC
=
r2
=
199 + 9.4E14:0 - 0.4E16:#{252} 51 log E18.2; 0.89 SE = 15
(C15)
270
a
250
1-.
210
230
1!
190
I
130
170 150
110 110
130
150 Observed
PC
=
r2
=
216 + 7.4E1+:0 - 1.2E18:i 0.94 SE = 12
(C16)
PC = 247 + 4.4E140 - 1.3E16.0 69 log E182; r2 = 0.95 SE = 9
1.6E181
-
(C17)
Figure 3 provides the simplest graphic illustration of the dietary fatty acid-plasma cholesterol relationship described by Eq. C14 and Eq. HI (see below) by plotting the ratio of
300
250
r = 0.894; p
