Alzheimer's disease (AD) is a neurodegenerative disease of the central nervous system characterized by an accumulation of extracellular and cerebrovascular amyloid and intracellular aggregated tau protein 1. Amyloid deposits are composed of amyloid-β peptide (Aβ), a 4 kD fragment generated by proteolytic processing of the transmembrane amyloid precursor protein (APP) 2. The "amyloid cascade hypothesis" is one popular model of AD pathogenesis, and amyloidogenic processing by γ-secretases is enhanced by mutations in either APP, presenilin 1, or presenilin 2. Among the two major species of Aβ, the C-terminally extended Aβ42 is more highly amyloidogenic than is the shorter but more abundant Aβ40. The ratio of Aβ42/Aβ40 determines the ratio of Aβ clearance vs accumulation, and Aβ aggregates or oligomers are believed to lead eventually to neuronal and synaptic dysfunction and neuronal death 2. In addition to the deleterious amyloidogenic pathway, an additional non-amyloidogenic pathway has also been described. In this pathway, APP is first cleaved by one of several α-secretases (e.g., ADAM10, ADAM17) between positions 16 and 17 of the potential Aβ fragment, thereby precluding amyloidogenesis 2.

Extensive data support the notion that diets low in calories, cholesterol, and saturated fatty acids, but rich in vegetables, fruits, and fish, may delay the onset and/or slow the progression of clinical AD. The regulation of various secretases by dietary components is also well-documented (e.g., cholesterol regulation of γ-secretase; 3). However, there have been no head-to-head studies employing diets high with various specified sources of calories to determine whether dietary composition has importance beyond the known effect of high caloric intake to exacerbate amyloid pathology 4. Therefore, in the current study we evaluated the effects of 4 different diets in a murine model of AD (TgCRND8) that harbors 3 mutations in the APP gene (K670M/N671L and V717F)5. Using this mouse model of AD, we set out to determine whether different dietary compositions could modulate Aβ levels, brain mass, neuronal density, and/or tissue volume in 3 subregions of the hippocampus.

The goal of this study was to evaluate the effect on Alzheimer's type neuropathology associated with one of four different diets: regular or reference commercial chow (58 kcal% carbohydrate/29 kcal% protein/13 kcal% fat), RC; high carbohydrate custom chow (60 kcal% carbohydrate/30 kcal% protein/10 kcal% fat), HC/LF; high fat custom chow (60 kcal% fat/30 kcal% protein/10 kcal% carbohydrate), HF/LC; high protein custom chow (60 kcal% protein/30 kcal% fat/10 kcal% carbohydrate), HP/LC; on solubilizable Aβ levels and neuronal loss in TgCRND8 plaque-forming mice. According to the two-way ANOVA, diet (as a general variable) played a role in determining both solubilizable Aβ40 levels (F_(3,43) = 7.298, p < 0.0005) and solubilizable Aβ42 levels (F_(3,43) = 13.978, p < 0.0005). Sex also played a role in solubilizable Aβ40 and Aβ42 levels with levels being higher in females than males (Aβ40, F_(1,43) = 7.685, p = 0.008; Aβ42, F_(1,43) = 11.835, p = 0.001). Interestingly, there was no interaction between diet and sex for either solubilizable Aβ40 (F_(3,43) = 0.770, p = 0.517) or solubilizable Aβ42 (F_(3,43) = 0.920, p = 0.439) demonstrating that the sex effect was similar within each diet, and, correspondingly, that the diet effect was similar within each sex.

Pairwise comparison of each diet to RC revealed that solubilizable Aβ42 levels were significantly increased in mice fed with HF/LC diet (RC vs F-P 147.89 vs 227.03, p = 0.014), but no difference was found between solubilizable Aβ42 levels of mice fed with HP/LC or HC/LF diets as compared to RC (HP/LC 130.89, p = 0.862; HC/LF 152.48, p = 1.000) (Figure 1). For solubilizable Aβ40, results approached statistical significance for the HF/LC vs RC comparison (RC vs HF/LC 26.77 vs 47.94, p = 0.064), while neither HP/LC nor HC/LF diets affect solubilizable Aβ40 levels as compared to RC (HP/LC 29.83, p = 0.966; HC/LF 29.19, p = 0.971).

The observation that the commercial diet (58 kcal% carbohydrate/29 kcal% protein/13 kcal% fat) and the custom HC/LF diet (60 kcal% carbohydrate/30 kcal% protein/10 kcal% fat) yielded identical results tends to exclude the possibility that any variations were attributable to the custom formulation per se. The elevation in solubilizable Aβ levels in the brains of HF/LC diet-fed TgCRND8 mice was in accordance with data obtained in other laboratories that directly connected high fat/high cholesterol diets with amyloidosis and conventional measures of soluble and insoluble Aβ 678. However, in our experiment, cholesterol was added to the RC, HC/LF, and HP/LC diets in order to bring those custom diets up to an identical level of cholesterol content to that present in the HF/LC diet. Therefore, elevations in solubilizable Aβ levels in HF/LC diet-fed TgCRND8 mice were not attributable cholesterol but to dietary fats other than cholesterol.

A two-way ANOVA for differences in plaque density showed no statistically significant effect of diet (F_{(3, 23)} = 1.039, p = 0.394), sex (F_{(1, 23)} = 0.277, p = 0.603), or diet*sex interaction (F_(3,23) = 0.237, p = 0.870). This suggests that the higher levels of solubilizable Aβ that we observed in our HF/LC diet-fed mice might involve primarily non-plaque forms of Aβ (e.g., oligomeric Aβ).

We then evaluated the effect of diets on the weight of the mice. At baseline, at the age of 4 weeks, there was no difference in body weight of the mice according to sex or diet (Table 2; Figure 2). By week 6 (i.e., only two weeks into the 4-diet protocol), we observed an effect of diet on body weight (highest to lowest: HF/LC > HC/LF = RC > HP/LC) and an effect of sex (higher weights of males than of females), but no interaction between these two factors. Pairwise comparison showed the HF/LC and HC/LF diets were significantly different from RC, while HP/LC diet was not.

At week 17, body weights of the mice differed by diet, but overall, there was no effect of sex (Table 2; Figure 2). A significant diet*sex interaction indicated that the weights of the male mice were higher than those of females in the RC, HP/LC, and HC/LF diets (M vs F, RC 24.7 g vs 20.8 g, HP/LC 21.4 g vs 17.8 g, HC/LF 29.6 g vs 27.1 g), while mice fed with HF/LC diet showed the opposite pattern with females being heavier than males (M vs F, HF/LC 33.5 g vs 43.7 g).

We performed a two-way ANOVA by diet and sex for repeated measures of body weight from week 6 through week 17. As shown (Table 3; Figure 2), as expected, the body weight of the mice was affected by age. There were significant age*diet and age*diet*sex interactions, while the age*sex interaction was only a trend that approached significance. ANOVA of linear trends from the repeated measures analysis showed comparable results, demonstrating that the age interactions primarily reflected linear trends. The age*diet interaction showed a greater increase in the HF/LC group than in the other groups. The age*diet*sex interaction in the HF/LC diet group showed a greater increase among females than males, while increases for males were at least as large as for females in the other diets.

We then evaluated the weights of the brains at time of death and compared them by two-way ANOVA. While diet affected the weight of the brain (F_(3,34) = 5.413, p = 0.004), neither sex (F_(1,34) = 0.0025, p = 0.972) nor diet*sex interaction (F_(3,34) = 0.381, p = 0.997) affected the brain weight. A pairwise analysis showed that mice fed with HF/LC and HC/LF diets had brain weights equal to mice fed the RC diet (RC vs F-P 0.535 g vs 0.540 g, p = 0.972; vs C-P 0.538 g, p = 0.997), while mice fed the HP/LC diet had a significantly lower brain weight compared to RC (RC vs P-P 0.535 g vs 0.514 g, p = 0.032) (Figure 3). The brain/total weight ratios (brain weight as % of total weight) were [HC/LF] 1.929 ± 0.227 (SD) (n = 10); [HP/LC] 2.668 ± 0.397 (n = 13); [HF/LC] 1.413 ± 0.307 (n = 10); [RC] 2.418 ± 0.321 (n = 9). Therefore, the brain weight as % of total weight tended to be highest in the HP/LC group and lowest in the HF/LC group.

There was a significant Pearson correlation of neuronal count and volume estimate (r = 0.420, p < 0.001) among the subset of mice studied. There were no significant effects of diet, sex, or diet*sex interaction for neuronal count or volume in any of the three hippocampal subregions: dentate gyrus, CA1, and CA3. The mean ± standard error of the mean (SEM) for each subregion is presented in Figure 4. In CA3, for neuronal count, the HP/LC mean was only 52% of the RC mean, and the HP/LC volume was only 70% of the RC volume, but the comparisons were not significant (p = 0.308 and p = 0.282, respectively). On the other hand, when HP/LC vs RC for the CA3 region were compared without adjustment for multiple comparisons, the approximate t-tests assuming unequal variances yielded p values that approached statistical significance (neuronal count: t = 2.092, df = 7.37, p = 0.073; neuronal volume: t = 2.223, df = 5.84, p = 0.069). This raises the possibility that hippocampal CA3 was involved in the loss of brain mass reflected in the diminished brain mass in HP/LC diet-fed TgCRND8 mice. However, in the absence of a cohort of HP/LC diet-fed nontransgenic mice, we cannot be certain what role, if any, was played by diet alone vs the interaction of cerebral amyloid with diet.

Effects of various diets on clinical and experimental AD pathology have been reported in the literature, including (a) high carbohydrate diets, (b) restricted calorie diets, (c) the ketogenic diet, and (d) diets high or low in cholesterol and other fats. Under certain circumstances, high carbohydrate diets have been reported to exacerbate the pathology of AD in experimental animals 9. Mice given 10% sucrose-sweetened water in order to induce glucose intolerance, hyperinsulinemia, and hypercholesterolemia, showed a 2-3-fold increase in levels of insoluble Aβ in their brains 9. In the current study, we did not observe any association of brain weight loss, Aβ levels, or hippocampal integrity with either of two diets (one commercial, one custom) in which most calories were derived from carbohydrates 9. A growing body of work implicates insulin resistance as a key feature of carbohydrate-related disturbances in Aβ metabolism 10, possibly explaining the apparent discrepancy between our study and that of Cao et al. 9, since our mice were not diabetic. A caloric restriction (CR) diet has been shown to attenuate amyloidosis in murine and monkey models 511.

A ketogenic diet (low in carbohydrates, high in fat) has been reported to reduce brain Aβ levels in a mouse model 12. This is somewhat surprising since several studies in the literature show a robust association of HF/LC diet with increased solubilizable brain Aβ levels 678. Here, we confirmed the generally-held association of high fat diet with increased Aβ accumulation, but, since all our diets contained identical levels of cholesterol, the solubilizable Aβ levels as measured in our HF/LC diet-fed mice must be due to fats other than cholesterol.

A typical "Western diet," containing 40% saturated fatty acids and 1% cholesterol, was reported to be associated with increased brain Aβ levels in a murine model of AD when compared with a diet enriched in docosahexaenoic acid (DHA) 13. An omega-3 fatty acid-enriched diet was reported to reduce the amyloid burden in an AD mouse model 14. In a clinical study of the possible associations between intake of specific types of fat and AD in a cohort of 815 65-year-old patients, intake of saturated fat and trans-unsaturated fat was correlated with an increased risk for clinical AD, while high intake of unsaturated, unhydrogenated fats was correlated with decreased risk 15. Therefore, it is possible that it is specifically the saturated fat content of a diet--and not its cholesterol content -- that modulates Aβ metabolism. Some fats (e.g., DHA) are reported to be relatively protective, and it is worth noting that mice fed HF/LC diets showed no loss of brain weight or hippocampal integrity despite the higher levels of solubilizable Aβ. These data are consistent with observations that clinical dementia in humans correlates best with atrophy and not with neuropathological burden 161718192021.

The most unexpected result of our study was the loss of overall brain mass associated with a HP/LC diet that, notably, did not exacerbate solubilizable Aβ levels or amyloid pathology. It is worth noting that a major confound in the field of Alzheimer's disease mouse modeling has been the conspicuous absence of neuronal loss in these transgenic mouse models. Indeed, this widely confirmed precedent led us not to include nontransgenic mice in our studies of the effects of custom diets. TgCRND8 mice, because of their aggressive amyloid pathology, are relatively insensitive to modulators of amyloidosis, and this may explain the minimal effect that we observed for dietary fat. However, the malignant amyloid pathology may well increase the sensitivity for detecting modulators of amyloid toxicity. This may explain why we observed the loss of brain mass with the HP/LC diet, but this conclusion cannot be drawn until we examine the effect of this HP/LC diet on the brains of nontransgenic mice.

The "encephalization quotient" (EQ; brain weight/body weight) is modulated by both genetic and dietary factors during brain development 2223, but the effect of diet on EI in adult animals is less well-studied. The EQ in our study was highest among the HP/LC mice and lowest among the HF/LC group. Typically, protein deprivation is associated with low brain weight and low EQ 22. Conversely, high fat diet is believed to have facilitated the evolution of higher EQ 23. On the surface, then, our observations run counter to other reported effects on EQ.

Interestingly, protein-carbohydrate ratio has been demonstrated to modulate brain excitotoxicity differentially as a function of age. A high carbohydrate/low protein (HC/LP) diet has been reported to increase excitotoxicity in the hippocampus and hypothalamus of young rats, while, conversely an HC/LP diet has been associated with decreased excitotoxicity in the same brain regions in older rats. However, potentially relevant to our result, a high protein-low carbohydrate (HP/LC) diet has been reported to associate with enhanced excitotoxicity specifically in the aged brain 24. Since Aβ is known to lower the threshold for glutamatergic excitoxicity in cultured neurons 25, it is conceivable that high protein diet could play a role in Aβ- related neurodegeneration via aging-dependent sensitization to glutamatergic excitotoxicity. Further experiments (including studies of nontransgenic mice) will be required to determine whether the trend toward lower neuronal counts and decreased volume in the hippocampal CA3 subregion might be related to Aβ toxicity and might reach statistical significance using larger cohorts. These data, if specifically associated with the presence of cerebral Aβ amyloid, suggest that a diet rich in protein might enhance neuronal vulnerability to amyloidosis.

Of course, regardless of what happens in the mouse models, the more important question is whether these data have implications for the human aging brain and/or the human AD brain. Given the association of high protein diet with aging-related neurotoxicity 24, one wonders whether particular diets, if ingested at particular ages, might increase susceptibility to incidence or progression of AD. This can only be established by prospective randomized double blind clinical trials of various diets. This would be a challenging undertaking but potentially worthwhile if there is a serious possibility for modifying the course of AD by via manipulating dietary composition.

S.G. holds an investigator initiated grant from Forest Research Institute, and he also holds consultancy positions with Diagenic and with Amicus Therapeutics. S.G. serves on the Safety Monitoring Board for the Wyeth/Elan Aβ active vaccination program.

SP, CT, and HB performed the experiments and statistical analysis, and JS assisted with the statistical analysis. LH, RM, GP, and SG designed the custom diets. PF, DW, and PSH created and characterized the TgCRND8 transgenic mice. JDB provided reagents and advice for the Aβ analyses. DLD and PRH supervised the morphometric analyses. ME and SG provided the overall design, supervised the collection and analysis of data, secured funding for the project, and supervised the preparation of the manuscript.