Abstract
Current evidence suggests that lipid homeostasis in the hippocampus is affected by different genetic, dietary, and hormonal factors, and that its deregulation may be associated with the onset and progression of Alzheimer’s disease (AD). However, the precise levels of influence of each of these factors and their potential interactions remain largely unknown, particularly during neurodegenerative processes. In the present study, we have performed multifactorial analyses of the combined effects of diets containing different doses of docosahexaenoic acid (DHA), estrogen status (ovariectomized animals receiving vehicle or 17β-estradiol), and genotype (wild-type or transgenic APP/PS1 mice) in hippocampal lipid profiles. We have observed that the three factors affect lipid classes and fatty acid composition to different extents, and that strong interactions between these factors exist. The most aberrant lipid profiles were observed in APP/PS1 animals receiving DHA-poor diets and deprived of estrogens. Conversely, wild-type animals under a high-DHA diet and receiving estradiol exhibited a lipid profile that closely resembled that of the hippocampus of control animals. Interestingly, though the lipid signatures of APP/PS1 hippocampi markedly differed from wild-type, administration of a high-DHA diet in the presence of estrogens gave rise to a lipid profile that approached that of control animals. Paralleling changes in lipid composition, patterns of gene expression of enzymes involved in lipid biosynthesis were also altered and affected by combination of experimental factors. Overall, these results indicate that hippocampal lipid homeostasis is strongly affected by hormonal and dietary conditions, and that manipulation of these factors might be incorporated in AD therapeutics.
ABBREVIATIONS
cholesterol phosphatidylethanolamine phosphatidylserine sulfatides phosphatidylcholine cerebrosides phosphatidylinositol phosphatidylglycerol sphingomyelin lysophosphatidylcholines steryl/cholesteryl esters diacylglycerols docosahexaenoic acid arachidonic acid eicosapentaenoic acid long-chain polyunsaturated fatty acids peroxidability index unsaturation index estradiol wild-type mice Alzheimer’s disease transgenic mice model principal component analysis principal component 1 principal component 2 one-way analyses of variance two-way analyses of variance multiple analyses of variance APP/PS1-free-DHA-vehicle animals APP/PS1-high-DHA-vehicle animals APP/PS1-free-DHA-estradiol animals APP/PS1-high-DHA-estradiol animals wild-type-free-DHA-vehicle animals wild-type-high-DHA-vehicle animals wild-type-free-DHA-estradiol animals wild-type-high-DHA-estradiol animals
INTRODUCTION
Evidence accumulated over the last three decades have pointed to brain lipids as critical players involved in the pathogenesis of Alzheimer’s disease (AD). Alterations in different brain lipids, including cholesterol and long-chain polyunsaturated fatty acids (LCPUFA), have been observed by different research groups, including ours, in postmortem human brain even at the earliest stages of AD [1–4]. In particular, docosahexaenoic acid (DHA, 22:6n-3), the main n-3 LCPUFA in nerve cell phospholipids (representing about 20% of total fatty acids), appears to be consistently reduced in different brain areas of AD patients, including the hippocampus, one of the main areas affected in AD [1, 4–6]. Furthermore, data from epidemiological and experimental animal studies have highlighted the beneficial influence of DHA on the preservation of synaptic function and memory capacities in aged individuals or upon amyloid-β exposure [6–9]. On the contrary, DHA dietary deficiency increases the risk of developing AD [5, 10–12].
On the other hand, recent discoveries have shown that, beyond a reproductive hormone, estradiol is also a brain-derived factor and that estrogen receptors (ERs) coordinate multiple signaling mechanisms that protectthe brain from neurodegenerative diseases in both males and females (reviewed in [13–15]). Moreover, though still controversial, epidemiological studies and clinical trial studies suggest that estrogen treatment may prevent age-related cognitive decline and reduce the risk of dementia and AD [15, 16].
In spite of substantial differences in the physiology and biochemistry of estrogens and LCPUFA, there appear to exist common aspects whereby these two factors modulate nerve cell biology and promote neuroprotection. For instance, both factors are associated to neuritogenesis and neuronal plasticity, neuronal excitability, modulation of common signaling pathways (i.e., PI3K-Akt) involved in neuronal survival, protection against oxidative stress, and in the control of pro-inflammatory mediators and neuroinflammation, among other effects associated with cytoprotection against neuronal degeneration [12, 17–24].
Although these potential neuroprotective factors have been extensively studied in different cellular and animal models, the potential cooperation, interaction or even synergism, has never been analyzed in depth in the literature. Only recently, some reports have put some emphasis on the potential role of estrogens as modulators of neuronal signalosomes and brain lipid homeostasis related to protection against neurodegeneration [14]. In the present study, we have undertaken a combinatorial analysis of the effects of different hormonal and dietary DHA conditions, on the lipid composition and expression of lipogenic genes in the hippocampus of wild-type and APPswe/PS1ΔE9 transgenic mice, an established model of familial AD. We have performed different statistical multivariate approaches and analyses to thoroughly assess the contribution of dietary DHA, circulating estrogens, and the existence of an AD-like genotype, as individual factors, as well as their potential interactions, in setting the lipid fingerprint of the hippocampus. The results indicate that hormonal and dietary conditions are key synergistic players in maintaining hippocampal lipid homeostasis, and more importantly, that they might be taken into account in the development of specific nutraceutical strategies for AD, especially at early stages of the disease.
MATERIALS AND METHODS
Animals
APPswe/PS1ΔE9 transgenic mice and wild-type (WT) littermates were bred and maintained under standard housing conditions in a 12-h dark–light cycle at the Animal Facilities Service from University of La Laguna (Spain). Untreated WT animals were used as controls (CTRL). After weaning, genomic DNA was isolated from tails and genotyped by PCR using the conditions recommended by Jackson Laboratory (Maine, USA). Females carrying the APP/PS1 genotype and their WT littermates were housed separately and randomly assigned to the different groups in the present study. Mice were initially fed with standard diets (AAIN93) and had free access to water. One month after birth, standard diets were gradually replaced with Free-DHA or High-DHA at a rate of 25% per week so that at the age of two months animalswere fed with 100% experimental diets. Free- and High-DHA diets were prepared by Harlan Laboratories (Barcelona, Spain), and their composition is shown in Supplementary Table 1. At postnatal day 90, animals were ovariectomized, and 48 h later implanted subcutaneously with synthetic capsules (Innovative Research of America, Sarasota, USA) containing either estradiol (0.05 mg estradiol acetate/pellet) or vehicle (olive oil) following the manufacturer’s indications. At the age of six months, APP/PS1 transgenic mice and WT littermates were sacrificed by cervical dislocation, and the brains isolated to dissect the hippocampus from both hemispheres. This age was chosen because it is when differentiation in brain lipids between WT and APP/PS1 animals is first observed [25], and when amyloid plaque burden and increased levels of Aβ1 - 42 and Aβ1 - 40 are evident in APP/PS1 animals [26]. Right hemisphere was used for lipid analyses and left hemisphere for the isolation of mRNA. Groups of 5 animals were randomly assigned to each combination of diet (Free-DHA or High-DHA), hormonal (17β-estradiol or Vehicle as placebo) and genotype (WT or APP/PS1) conditions following a 2×2×2 factorial design. All experimental manipulations were performed following the procedures authorized by the Ethics Committee for manipulation of laboratory animals at University of La Laguna (Spain) following the guidelines of the European Community Council (Directive 86/609/EEC).
Lipid analyses
Lipid analyses were performed as described previously [25, 27]. Briefly, total lipids from hippocampal tissues were extracted with chloroform/methanol (2:1 v/v) containing 0.01% of butylated hydroxytoluene as antioxidant. Lipid classes were separated by one-dimensional double development high performance thin layer chromatography using methyl acetate/isopropanol/chloroform/methanol/0.25% KCl (5:5:5:2:1.8 volume basis) as the developing solvent system for the polar lipid classes, and hexane/diethyl ether/acetic acid (22.5:2.5:0.25 volume basis) as the developing solvent system for the neutral lipid classes. Lipid classes were quantified by scanning densitometry after charring plates with 3% (w/v) aqueous cupric acetate containing 8% (v/v) phosphoric acid, using a Shimadzu CS-9001PC dual wavelength spot scanner.
Fatty acids composition was determined from total lipids upon acid-catalyzed transmethylation for 16 h at 50°C, using 1 ml of toluene and 2 mL of 1% sulfuric acid (v/v) in methanol. The resultant fatty acid methyl esters (FAME) and dimethyl acetals (DMA) which originate from the 1-alkenyl chain of plasmalogens, were purified by thin layer chromatography. FAMEs and DMAs were separated and determined in a TRACE GC Ultra (THERMO) gas chromatograph equipped with a flame ionization detector, and quantified by referring to authentic standards.
Determination of mRNA levels by Real Time RT-PCR
Relative mRNA expression levels were performed as described previously [28]. Briefly, total RNA waspurified from hippocampal tissues preserved in RNAlaterTM (Invitrogen), using the RNeasy® Lipid Tissue Kit Mini Kit (Qiagen), followed by DNase I digestion, acid phenol:chloroform extraction, and ethanol precipitation. cDNA samples were obtained with the Transcriptor First Strand Synthesis Kit (Roche) using anchored oligo(dT)18 primers and 6μg of total RNA as template. In order to control for gDNA contamination, amplification primers (Supplementary Table 2) were targeted to different exons and, when possible, spanning an exon/exon boundary. RT-minus controls were incorporated in those cases where the corresponding primers do not discriminate gDNA sequences After cDNA synthesis, RNA integrity was confirmed in all samples through the 3’:5’ assay [29] using two primer-pairs targeted to the TATA binding protein (Tbp) mRNA (Supplementary Table 2). Real-time amplifications were performed in triplicate using SYBR Green detection on the LightCycler 480 platform (Roche). Relative quantities of the targeted mRNAs were calculated from Cq data following an efficiency-correction model implemented in the REST software [30]. The PCR efficiency of each amplification reaction was estimated with the DART-PCR workbook [31]. The normalization factor for each cDNA sample was calculated as the geometric mean of the expression values of reference genes Hprt1, Tuba1, and Tbp genes.
Statistics
Hippocampal lipid and gene expression variables were initially assessed by one-way analyses of variance (ANOVA-I) followed by Tukey’s or Games-Howell post hoc tests, where appropriate. Kruskal-Wallis and Mann-Whitney U tests were used in cases where normality was not achieved. Data were afterwards submitted to two-way or multiple analyses of variance (ANOVA-II or MANOVA, respectively) in order to determine main effects between the different factors, and to assess the existence and degree of their interaction. Lipid classes and main fatty acids were additionally submitted to multivariate analyses by means of principal components analysis (PCA), in order to obtain the extraction coefficient matrixes of lipid components and their contributions to overall variance and weights in group segregation. Factor scores from principal component 1 (PC1) were used to obtain the lipid profile of each experimental group. Factor scores were further analyzed by two-way ANOVA to evaluate the main effects of diet, hormonal status, and genotype, as well as their interactions, in the lipid signatures of the different groups. Regression, total correlation, and partial correlation analyses were performed in order to assess the significance of linear relationships between different lipids or gene expression variables, and to explore for the effects of individual factors in setting the bivariate relationships. Further analyses using (S)MATR software [32] were carried out to seek for statistical differences in regression models of lipid or gene expressionrelationships.
RESULTS
Lipid classes
Results in Table 1 show that polar lipids are more abundant than neutral lipids in all groups, though their relative proportions varied depending on the combination of factors considered. Among neutral lipids, cholesterol (CHO) is, by far, the most abundant lipid class, while in polar lipids phosphatidylethanolamine (PE) is predominant, followed by phosphatidylserine (PS), sulfatides (SUL), phosphatidylcholine (PC), cerebrosides (CER), phosphatidylinositol (PI), phosphatidylglycerol (PG), and sphingomyelin (SM). ANOVA I analyses of hippocampal lipid classes revealed statistical differences between groups, and affecting most lipid classes (Table 1). The three factors, i.e., dietary DHA, presence of estradiol, and genotype, appear to have significant influences on the distribution of hippocampal lipids, though many alterations appear to depend on the type of lipid considered. For instance, lysophosphatidylcholines (LPC) were only observed in the APP/PS1 vehicle group, but were totally absent in the rest of groups. This same group displayed the lowest levels of total neutral lipid and the highest contents of total polar lipids in the whole set of groups.
The high complexity of datasets and multiple comparisons encouraged us to tackle a multivariate approach using principal component analysis (PCA) (Fig. 1A). PCA revealed that differences were mostly attributed to cholesterol (CHO), phosphatidylethanolamine (PE), lysophosphatidylcholine (LPC), steryl/cholesteryl esters (SE), and diacylglycerols (DAG), with components 1 and 2 (PC1 and PC2, respectively) explaining over 60.2% (PC1 43.88% + PC2 16.36%) of overall variance (Fig. 1A). Two-way ANOVA applied to factor scores from PC1 showed a dramatic effect of genotype (F1,30 = 79.59 p = 0.000) and hormonal treatment (F1,30 = 48.51 p = 0.000) on the hippocampal lipid classes composition, which were marked by a strong interaction (F1,30 = 158.08 p = 0.000). Diet displayed a significant interaction only with both factors combined (F1,30 = 19.51 p = 0.000) for factor score 1. The effect of diet is better represented in factor score 2, being the most important factor in explaining PC2 (F1,30 = 20.04 p = 0.000).
As most of the variance was explained by PC1 (43.88%), factor scores for the whole dataset were plotted according to this component, and the different experimental factors were identified across groups (Fig. 1B). As can be seen for genotype and hormonal status factors, WT (Fig. 1Ba) and estradiol (E2) (Fig. 1Bb) groups were clustered as single groups, whereas for APP/PS1 and vehicle groups, a clear separation into two clusters was observed. The cluster segregated on the left side corresponded to the intersection APP/PS1-Vehicle, which reflect the high interaction between both factors. Diet factor, on its own, did not allow a clear declustering of groups (Fig. 1Bc). However, identification of individual groups indicate that within the segregated cluster, Free- and High-DHA APP/PS1 groups (AFV and AHV) are located along factor score 2 with the High-DHA group occupying the higher scores (Fig. 1Bd), which agrees with the statistical observation that diet influences hippocampal lipid classes only when overlapped with the other two factors combined.
Based on these results, we tracked the effects of sequential incorporation of each factor on APP/PS1 animals (Fig. 2A). We observed that, for APP/PS1-High DHA-Vehicle animals (AHV), incorporation of estradiol displaces the cluster (AHE) towards the “normal” profile observed in WT-High DHA-E2 animals (WHE), which, in turn, overlaps with the signature observed in control CTRL untreated animals (Fig. 2A). Moreover, for APP/PS1-Vehicle-Free DHA animals (AFV), which exhibit the most aberrant lipid profile, sequential incorporation of estradiol (AFE) followed by dietary DHA (AHE), results in the partial restoration of the normal lipid profile (Fig. 2A).
The significant effect of genotype on hormonal status was particularly evident in lipid classes highly co-related either negatively (CHO and DAG) or positively (PE and LPC) to PC1 (Fig. 1A). These lipid classes exhibited a high association with hormonal status whose direction depended on the genotype, being the maximal differences observed between the APP/PS1-vehicle and WT-vehicle groups. Moreover, administration of estradiol largely reversed thealteration.
The significant interaction between diet on one side, and hormonal status and genotype on the other, is clearly observed in lipid classes such as SE (Fig. 2C), where the three factors significantly interact two-by-two but only their full combination restored the values observed in WT-High-DHA-E2 animals (Table 1). Finally some lipid classes, like PC and PS, appear to be mostly under the influence of diet factor and diet * genotype interaction (Fig. 2D). In these cases, the High-DHA diet increases (PC and PS) or reduces (CER) the lipid class contents in hippocampus, especially in the WT genotype (Fig. 2D, Table 1).
Analyses of gene expression
Given that cholesterol is the most abundant lipid class in the hippocampus, and that the experimental factors affected cholesterol contents and its metabolic relatives such sterol esters, we have analyzed the expression of different genes encoding for cholesterol metabolism. These include HMG-CoAR (encoding for 3-hydroxy-3-methyl-glutaryl-CoA reductase, the rate-controlling enzyme of the mevalonate pathway, that produces cholesterol), Acat1-3 (encoding for Acyl-CoA cholesterol acyltransferase, which forms steryl/cholesteryl esters), and Scd1-2 (encoding for stearoyl-CoA desaturase, or Δ9 desaturase). The results illustrated in Fig. 3 shows that the relative expression of all genes under study was differentially affected, and to different extents, by the combination of the three factors (Fig. 3A). Thus, outcomes of ANOVA II revealed that HMG-CoAR expression was modified by genotype (F1,30 = 6.241 p = 0.02), diet (F1,30 = 22.56 p = 0.000), and hormonal status (F1,30 = 12.27 p = 0.002), with a significant interaction existing between genotype and hormonal status (F1,30 = 3.83 p = 0.026), which is in line with the results described above. For the Acat gene family, the most important changes were observed for Acat1 that exhibits a high dependence on genotype (F1,30 = 37.16 p = 0.000) which, in turn, displays a significant interaction with diet (F1,30 = 9.15 p = 0.006), and hormonal status (F1,30 = 4.89 p = 0.037). Further, Acat1 gene expression exhibited an important interaction between hormonal status and diet (F1,30 = 9.63 p = 0.005), and also between the three factors (F1,30 = 7.03 p = 0.014). These observations in Acat1 gene expression are genetically and metabolically relevant because Acat1 is the Acat paralog mostly expressed in hippocampus [33], and because they paralleled the changes observed for steryl/cholesteryl esters described in the previoussection.
Regarding Δ9 desaturases involved in desaturation of long chain saturates to produce oleic (18:1n-9) and palmitoleic (16:1n-7) acids, which esterify cholesterol to render cholesterol esters and unsaturated phospholipids, the two analyzed Scd genes are expressed in hippocampal tissue, being the Scd2 paralog predominant over Scd1, which agrees with previous studies on the distribution of stearoyl-CoA desaturases in brain tissue [34, 35]. Outcomes of ANOVA II indicates that the two genes are highly dependent on the hormonal status (Scd1: F1,30 = 9.40 p = 0.005; and Scd2: F1,30 = 48.25 p = 0.000), but only Scd1 is affected by the genotype (F1,30 = 5.35 p = 0.029) and diet (F1,30 = 10.86 p = 0.003).
The potential cross-interactions between the expression of the different genes, considered up to now and the experimental factors, were assessed by regression and partial correlation analyses. Results in Fig. 3B revealed that HMG-CoAR expression was negatively correlated to Acat1 (F1,30 = 15.761, r = –0.587, p = 0.000) (Fig. 3Ba) and positively to Scd2 (F1,30 = 3.78, r = +0.335, p = 0.061) (not shown). Moreover, detailed regression analyses of Acat1 over HMG-CoAR in response to the different factors indicates that their overall negative relationship was only present in estradiol-receiving animals (F1,15 = 12.46, r = –0.686, p = 0.03) (Fig. 3Bb), WT animals (F1,15 = 15.59, r = –0.726, p = 0.01) (Fig. 3Bc), and High-DHA animals (F1,15 = 14.89, r = –0.718, p = 0.02) (not shown) but disappeared in the vehicle and APP/PS1 mice (Fig. 3Bb and 3Bc, respectively). Likewise, analyses of Scd2 over HMG-CoAR revealed significant overall positive relationships which were only present in Free-DHA animals (F1,15 = 22.117, r = –0.783, p = 0.00) (Fig. 3Bd), and APP/PS1 animals (F1,15 = 7.89, r = 0.603, p = 0.01) (Fig. 3Be), but was unrelated to hormonal treatment, and disappeared in the WT genotype and in animals receiving High-DHA (data not shown). Partial correlation analyses excluding the influence of genotype and hormonal status interaction showed that the significant relationships of HMG-CoAR versus Acat1 and Scd1 versus Acat1 disappeared, while that for HMG-CoAR versus Scd1 was unaffected and that of Scd2 versus Acat1 became significant (data not shown). Similarly, partial correlations excluding the effects of genotype * diet, or genotype * diet * hormonal status, showed that the correlations between Scd1 or Scd2 and Acat1 were abolished. These findings fit well with the observations that changes in gene expression occurring in APP/PS1 animals modify the relationships between cholesterol and steryl/cholesteryl esters (Fig. 3Bf) and also with the opposing effects of hormonal status and diet in SE levels between APP/PS1 and WT animals (Fig. 2C, Table 1).
Fatty acids
Analyses of hippocampal fatty acids composition from total lipids in the different groups are summarized in Table 2 for WT animals and APP/PS1 mice. The most obvious changes were observed for DHA (22:6n-3) contents, which were, by average, 23% higher in the High-DHA groups compared to Free-DHA diets in all groups. Paralleling these changes, other fatty acids of the n-3 series were also changed, i.e., eicosapentaenoic acid (EPA, 20:5n-3) and docosapentaenoic acid (DPA, 22:5n-3). These fatty acids were totally absent in the Free-DHA groups but appeared in the High-DHA animals as a consequence of DHA to EPA retroconversion [36–38]. Consequently, total n-3 LCPUFA, and peroxidability (PI) and unsaturation indexes (UI) were significantly increased (Table 2 and Supplementary Table 3). Noticeably, although not modified in the administered diets, levels of arachidonic acid (ARA, 20:4n-6), docosatetraenoic acid (22:4n-6), docosapentaenoic acid (DPA, 22:5n-6), and total n-6 LCPUFA, were all significantly reduced in the High-DHA groups, indicating a cross-inhibition of n-6 LCPUFA metabolism by the n-3 LCPUFA load. Further, parallel alterations were observed for monoenes, especially oleic acid (18:1n-9) and saturates (stearic acid 18:0, and palmitic acid 16:0) to different extents in response to the DHA content of diets.
When comparing hippocampal fatty acids in WT and APP/PS1 animals, it became evident that a number of differences were related to genotype and also to hormonal status, being APP/PS1-Free DHA-Vehicle (AFV) group the one displaying most statistical differences compared to the rest of groups (Table 2).
In order to detect these effects, we first performed PCA on hippocampal fatty acids. The results illustrated in Fig. 4Aa indicated that 45.5% of overall variance was explained by PC1 and above 60% by the inclusion of PC2. Fatty acids showing largest absolute extraction values for PC1 were 22:6n-3, 22:5n-3, 22:5n-6, 22:4n-6, 20:5n-3, and 20:4n-6, which indicate the massive alteration of long-chain PUFA levels. Also monoenes 18:1n-7 and 18:1n-9 were extracted with high communalities. PC1 was positively related to the n-3 and monoenoic series (more abundant in the High-DHA diet) and negatively to the n-6 fatty acids (more abundant in the Free-DHA diet). Two-way ANOVA on factor scores from PC1 revealed that inter-subjects variance was largely associated with diet (F1,22 = 994.62 p = 0.000), and no interaction was detected with hormonal treatment or genotype. However, a very significant influence was detected for genotype (F1,22 = 21.87 p = 0.000) and for the interactions genotype * hormonal treatment (F1,22 = 23.14 p = 0.000) and hormonal status * diet (F1,22 = 4.352 p = 0.049) in factor score 2 (Fig. 4Ab).
Plotting factors scores from hippocampal fatty acids revealed the existence of two clusters perfectly separated that correspond to Free- and High-DHA conditions (Fig. 4Ba), which indeed stresses the enormous impact of diet on fatty acid profiles in hippocampus. Moreover, within each major cluster, two subclusters corresponding to each genotype (Fig. 4Bb) and hormonal condition (Fig. 4Bc) could be identified, indicating the existence of interactions between these variables and diet in the factor score 2.
As it happened with lipid classes, each experimental group exhibited a characteristic lipid profile. Based on these results, we tracked the effects of sequential incorporation of each factor on WT and APP/PS1 animals (Fig. 4C). We observed that, for WT-Free DHA-Vehicle (WFV) animals, incorporation of DHA and estradiol (+E2) displaced the cluster towards the “normal” profile observed in CTRL animals (Fig. 4Ca). Moreover, for APP/PS1-Free DHA-Vehicle animals (AFV), which again exhibits the most aberrant lipid profile, incorporation of the high-DHA diet (+DHA) followed by administration of estradiol (+E2), results in the partial restoration of the normal lipid profile (Fig. 4Cb).
It should be emphasized that, in both cases, the most important modifications were induced by incorporation of DHA, with estradiol having a lower effect in restoring the hippocampal fatty acids profile. In fact, partial correlation analyses excluding diet factor showed that significant associations between LCPUFAs (i.e., 22:6n-3 versus 20:5n-3 r = +0.834 p = 0.000; 22:6n-3 versus 20:4n-6 r = –0.520 p = 0.003; 22:6n-3versus 22:5n-6 r = –0.726 p = 0.000, 20:4n-6 versus 20:5n-3 r = –0.870 p = 0.000; 22:5n-3 versus 22:5n-6 r = –0.949 p = 0.000) were all vanished or, alternatively, changed their sign (see Supplementary Table 3).
However, in spite of the enormous effect of diet in hippocampal fatty acids, the significant effect of hormonal status and/or genotype can be demonstrated for different LCPUFA highly correlated with PC1, including DHA itself, by means of MANOVA. As it can be observed in Fig. 5Aa, plots of marginal means for DHA revealed that incorporation of this fatty acid to hippocampal phospholipids was favored by the presence of estradiol (F1,22 = 37.64 p = 0.000), and influenced by the genotype (main factor genotype F1,22 = 18.55 p = 0.000; genotype * hormonal status interaction F1,22 = 5.88 p = 0.024). In turn, arachidonicacid (20:4n-6), which was negatively correlated to DHA in the whole dataset (Fig. 5Ab), displayed mayor effects attributed to genotype (F1,22 = 36.62 p = 0.000), which favored the accumulation of this fatty acid in APP/PS1 animals when treated with estradiol (genotype * hormonal status interaction F1,22 = 10.69 p = 0.004), while in WT animals arachidonic acid levels remained rather constant and higher than in APP/PS1 animals (Fig. 5Ac). These influences determined the significant displacement of regression lines between Free-DHA and High-DHA (Fig. 5Ab), as demonstrated by the analyses intercepts (F1,26 = 79.11, p = 0.000).
Another important group of fatty acids seriously affected by diet and other experimental factors are monoenes, and their saturate precursors. Results summarized in Fig. 5B for oleic (18:1n-9) and stearic (18:0) acids, indicate that the physiologically relevant negative relationship between them disappeared in response to estrogen deprivation (Fig. 5Bb). In general, High-DHA animals exhibit higher amounts of oleic acid and lower stearic acid compared to Free-DHA counterparts (F1,22 = 117.50 p = 0.000 for oleic acid and F1,22 = 35.11 p = 0.000 for stearic acid) (Fig. 5Ba), but their absolute contents are significantly affected by the hormonal status (F1,22 = 38.22 p = 0.000 for oleic acid, and F1,22 = 17.52 p = 0.000 for stearic acid), being always lower the values in the presence of estradiol (Fig. 5Bc). Besides, an effect of genotype is also detectable in both fatty acids leading to increased levels in APP/PS1 animals (F1,22 = 37.47 p = 0.000 for oleic acid and F1,22 = 30.58 p = 0.000 for stearic acid), especially in the vehicle condition (Fig. 5Aa and 5Bc, for oleic and stearic acids, respectively). In the case of oleic acid, the most important interaction was detected for genotype and diet (F1,22 = 24.46 p = 0.000), according to which, APP/PS1 High-DHA animals exhibit increased levels of oleic acid. These discrepancies between interactions suggest a metabolic uncoupling between oleic and stearic acid levels that may explain their lack correlation in vehicle animals. This result agrees with the changes observed in the expression of Scd genes described above.
Expression of genes involved in LCPUFA biosynthesis
We finally analyzed the expression of main genes involved in the biosynthetic pathway for LCPUFA. These include Elovl2-5 (after Elongases of very-long-chain fatty acids, and encoding for elongases 2, 4 and 5), Fads1-2 (encoding for Δ5 and Δ6 desaturases), and Acox1 (encoding for peroxisomal acyl-CoA oxidase 1, that catalyzes the chain-shortening of 24-carbon intermediates through partial β-oxidation in peroxisomes, ultimately yielding C22:5n-6 and 22:6n-3) [39]. The results indicate that all these genes are expressed in the hippocampus and that, in some cases, their relative expression is differentially affected by the combination of the three factors (Fig. 6A). Expression levels of elongases, desaturases, and Acox1 are generally well correlated and covariated positively, not only within gene families (i.e., Elovl2 versus Elovl4, Fads1 versus Fads2, Fig. 6Ba and 6Bb) but also between families (i.e., regressions Acox1 versus Fads1, Elovl2 versus Fads2, Fig. 6Bc and 6Bd). Results from ANOVA I indicated significant differences between groups for Elovl2, Elovl4, Fads2, and Acox1 genes. ANOVA II revealed that, compared to the outcomes from fatty acids analyses, diet itself is less important as main factor, but influences the effect of genotype or hormonal status (Fig. 6C). In particular, Fads2 expression is affected by genotype as main factor (F1,32 = 4.19 p = 0.052), and this factor displays interaction with hormonal status (F1,32 = 6.08 p = 0.021) (Fig. 6Ca and 6Cb). Hormonal status is perhaps a more determinant main factor, as observed for Elovl family of genes (Elovl2: F1,32 = 4.85 p = 0.037; Elovl4: F1,32 = 9.47 p = 0.005; but not for Elovl5) (Fig. 6Cc) Fads1 (F1,32 = 4.41 p = 0.045) and Acox1 (F1,32 = 6.26 p = 0.020) (Fig. 6Cd), and often this effect of hormonal status projected on interactions with diet (i.e., Acox1: F1,32 = 6.18 p = 0.02). Finally, as it was mentioned before, oleic acid levels appear to be higher in the High-DHA condition (inset in Fig. 7a, Table 2). However, a closer examination of data on the effect of diet on oleic acid and DHA levels revealed a very significant negative correlation existing between both fatty acids under each dietary condition (Fig. 7a). Thus, the regression line for the High-DHA condition shifted toward higher values, and was significantly different from the Free-DHA regression (F1,26 = 38.02, p = 0.000). These changes were paralleled by compatible changes in the relationship between relative expression levels of Scd2 and Fads1 or Fads2, using similar regression models (shown in Fig. 7b for Scd2 on Fads1), whereby the slope of their linear relationships was significantly higher for the High-DHA diet. Finally, we also found that the presence of estradiol and the APP/PS1 genotype increased the slopes of regression relationships between Scd2 and Fads1 or Fads2 (shown in Fig. 7c for the effect of genotype on Scd2 versus Fads1, and in Fig. 7d for the effect of hormonal status on Scd2 versus Fads2). These results agree with the analyses of interactions described before. In any case, diet effects on gene expression were much less evident than for changes in fatty acid composition.
DISCUSSION
The present results demonstrate that hippocampal lipids are profoundly altered by changes in the dietary supply of DHA, estrogen hormonal status, and also by the presence of the familial-type AD genotype. In general terms, the results are in agreement with a number of studies from several laboratories which have unequivocally shown, in different animal models, that brain lipids are influenced by dietary DHA, and affected by the presence of AD-like genotypes [3, 41]. Conversely, only few studies have suggested that brain lipids might be differentially modulated by estrogens in AD brains [2, 42]. This is somehow striking since estrogens have been long recognized as key mediators of brain maintenance [13, 18], and because abrupt decline in estrogen levels at menopause have been associated with cognitive deficits and increased risk for AD [43]. In this context, we provide the first evidence that estrogens differentially affect hippocampal lipids, especially lipid classes, in WT and transgenic animals, as indicated by the very strong interaction between genotype and estrogen status. Interestingly, within lipid classes, cholesterol appears to be particularly affected by estradiol deprivation in the AD genotype, being their levels ≈20% lower in the APP/PS1-placebo group than in the rest of experimental groups. The expected effect of this alteration is tremendous given the essential role of cholesterol in the organization and functionality of nerve cell membranes [44–46]. Indeed, cholesterol is associated with the etiopathology of AD [2, 47], and it has been demonstrated that brain cholesterol alteration affects amyloid-β peptide generation [48]. However, conflicting reports exist on the direction cholesterol changes affect amyloidogenic processing of AβPP [44, 50], a discrepancy that requires further clarification. Irrespective of these changes in cholesterol, it is known that most brain cholesterol is synthesized locally, and that brain tissues are endowed with all the genetic machinery for the synthesis of cholesterol from acetyl-CoA [51, 52]. In astrocytes and neurons, cholesterol is synthesized in a complex biochemical pathway being the 3-hydroxy-3-methylglutaryl coenzyme A reductase (HMG-CoAR), the rate-limiting enzyme in cholesterol biosynthesis. Our results demonstrate that the three factors modulate HMG-CoAR expression and, more importantly, that the degree of transcriptional modulation is influenced by the significant interactions between genotype and hormonal status, which agrees with the outcomes of hippocampal cholesterol contents. This led us to conclude that changes in HMG-CoAR expression correlate with local cholesterol synthesis, which is lowest in the estrogen-deprived APP/PS1 group of animals. Paralleling changes in HMG-CoAR expression, we demonstrate significant, and correlated, changes in Acat gene family, in particular for Acat1 gene, which is the paralog predominantly expressed in hippocampus [33]. As for HMG-CoAR gene, Acat1 expression is affected by the three factors, and there exist important interactions between them, which closely mirror variations in HMG-CoAR expression. It is worth mentioning that no previous study has demonstrated a role of DHA in the regulation of Acat1 gene expression in brain. The negative relationship between HMG-CoAR and Acat1 expression projects on, and largely explains, the distribution of cholesterol and steryl/cholesteryl esters contents in hippocampal tissues, especially in the APP/PS1-vehicle animals, which exhibits largest steryl/cholesteryl esters and lowest cholesterol levels. In agreement, we have reported in human brain cortex, increased levels of SE, usually accompanied by reductions in cholesterol levels, in lipid rafts from AD patients even at very early stages of the disease (stages I/II of Braak and Braak staging) [4]. The pathogenic relevance of this aberrant profile is currently unknown but it is likely associated with the severity of amyloid pathology. Indeed, it has been shown that genetically mutant cell lines that overproduce cholesterol but cannot synthesize cholesteryl esters because of deficient ACAT activity, amyloid-β peptide production is almost completely inhibited [53]. Further, recent studies in transgenic AD models have reported that treatment with isotype-nonselective ACAT inhibitors substantially diminished amyloid plaque density [33, 54] have demonstrated that Acat1 gene ablation in triple transgenic (3XTg-AD) mice led to more than 60% reduction in full-length human AβPPswe expression as well as its proteolytic fragments, and ameliorated cognitive deficits. On the other hand, analysis of genetic data in humans suggests that SOAT1 (the gene encoding for Acat1 in humans) is linked with the risk of AD [55] and that a frequent polymorphism that leads to reduced ACAT activity might provide protection against AD [56]. These observations have led to a novel, and promising, strategy based on the manipulation of the CHO-to-SE conversion and the potential use of ACAT inhibitors in the therapeutic treatment or prevention of AD [57].
Another noticeable result observed in the present study for lipid classes is the presence of significant amounts of lysophosphatidylcholines (LPC) in APP/PS1 Vehicle (AFV and AHV) groups, which demonstrates the deleterious effect of estrogen deprivation and DHA-poor diets in the APP/PS1 genotype. This lipid class is normally undetectable in brain lipid extracts but is particularly important because it is generated secondarily (although not exclusively) by free radical-catalyzed oxidation of polyunsaturated phosphatidylcholines, an indicative of oxidative damage of membrane phospholipids [58]. This finding would support the general conception that oxidative stress is an essential part of the pathological process in the progression of AD [26, 59–61]. According to our results, these deleterious oxidative effects observed in APP/PS1 animals, may be mitigated by administration of diets containing DHA, and the incorporation of physiological levels of estradiol (and perhaps other natural compounds endowed with estrogenic activity, i.e., phytoestrogens).
It is largely unknown what the link between estrogen decline and AD is, nor it is the role of systemic and brain estrogen synthesis in the homeostasis in brain lipids. It is known that brain is endowed with the biochemical machinery for the local synthesis of estradiol independent of peripheral estradiol in both male and female [62, 63], and that decline in brain estrogen production (in both sexes) accelerates amyloid plaque formation [64]. At present, no clear evidence exits that local brain E2 synthesis is reduced by age [65], at least in humans. However, it is known that ER, which are expressed in high densities in hippocampus and medial amygdala [63, 65], undergo age-dependent changes as inferred by the number of ERα and ERβ positive neurons, especially in triple transgenic 3xTgAD female mice [65]. These differences might provide a molecular link between the AD genotype and hormonal refractoriness in clinical trial studies [15, 66], because even in the absence of changes in local E2 production, the reduction of hippocampal ER would affect cellular genomic and non-genomic signaling triggered by E2 [13, 18], leading to deficient neuromodulation in hippocampal neurons [62].
Perhaps the most dramatic effect on hippocampal lipids observed in the present study was the marked reduction of DHA in all groups receiving Free-DHA diets, which, by average, are 23% lower than in the High-DHA groups. This depletion of DHA is expected to have a profound impact in hippocampal function, given the pleiotropic actions of DHA in neuronal physiology [12, 67–70]. Indeed, deficits in brain DHA have been demonstrated to be concurrent with neurodegenerative diseases, in particular in AD [1, 4], and in Parkinson’s disease [71]. Accordingly, we report here that the magnitude of DHA depletion is dependent on the genotype and hormonal status. Thus, lowest DHA values are found in APP/PS1 transgenic animals receiving placebo and fed a Free-DHA diet. Furthermore, the outcomes of our multivariate analyses indicate that incorporation of DHA in the presence of estrogens largely restored the normal lipid phenotype as identified in the hippocampus of control WT animals.
There is general consensus that brain is a poor synthesizer of LCPUFA, yet it is accepted that a certain degree of local production is carried out by glial and neuronal cells, which contribute to maintaining the high levels of LCPUFA contained in brain [24, 72–75]. We show here that hippocampal tissue is endowed with the genetic machinery for the elongation and desaturation of n-3 and n-6 LCPUFA precursors, as well as for the chain-shortening of 22–24 carbons intermediates. Surprisingly, despite the important effects on fatty acid composition, our analyses indicate that dietary DHA (specially the free-DHA) has only a minor effect on the expression levels of Elovl, Fads, and Acox families of genes. These data are in stark contrast with the well-documented effects of low-DHA diets in liver, which represents a major source of LCPUFA for the whole body demands, including brain [24, 75]. In agreement, Igarashi et al. [76] have found that dietary n-3 PUFA deprivation for several weeks upregulates elongase and desaturase genes expression in rat liver but not in brain. Reconciling these findings, it has been recently suggested that the endogenous synthesis of LCPUFA (at least n-3 LCPUFA) from α-linolenic acid is regulated more by substrate availability than by changes in gene expression or by modulation of regulatory transcription factors [24, 77].
Our present data also revealed that hippocampallevels of LCPUFA are also dependent on the genotype and hormonal status. For instance, we observed that incorporation of dietary DHA into membrane phospholipids is favored by estradiol, especially in the WT genotype, while arachidonic acid levels were always lower in the APP/PS1 genotype, especially in the absence of estradiol. These changes are also reflected in the expression levels of Fads2, Elovl4, and Acox1, whose degree of variation by dietary DHA appear to be modulated by the genotype and hormonal status in a complex manner. As an example, Fads2 gene expression was downregulated in the High-DHA diet but this only occurred in the presence of estradiol and in the APP/PS1 genotype. These observations indicate an important level of interaction between estrogens and brain lipid biosynthetic pathways that overlaps with dietary and genotypic influences. In agreement, a recent transcriptome analysis of mouse brain gene expression has revealed a complex framework of metabolic pathways being modulated by 17β-estradiol, these including lipid oxidation and biosynthetic pathways [78]. Noticeably, increased expression of lipid regulatory transcription factors PPARδ, RARα, and RXRα in the brain of ovariectomized rats has been demonstrated in response to estradiol [79], which may provide a molecular link for the influence of estrogens in brain lipidhomeostasis.
Other studies in transgenic mouse models of AD have revealed differences in the hippocampal (and frontal cortex) transcriptome between WT and humanized PS1 or AD–linked DE9 hPS1 mutant transgenic mice [80]. The differences affect several metabolic and transduction pathways, including genes involved in brain fatty acids desaturation, such is the case of Scd2, which has been previously associated (upregulated) with AD pathophysiology [80, 81]. Interestingly, we have observed that this gene, as well as its metabolic framework (18:0, 18:1n-9 and their negative relationships) are differentially modulated by the genotype, but also profoundly affected by dietary DHA and by the presence of estradiol, in a complex manner and involving interactions between the three factors.
Several reports have demonstrated that estradiol represses Scd1 gene expression in liver, and that upon ovariectomy or in ER KO mice, there is a profound increase in Scd1 gene expression in liver and adipose tissue [35]. However, no previous reports have analyzed the effects of estradiol or dietary DHA on Scd2 expression in the brain. Nonetheless, our present results revealed increased or equivalent levels of 18:1n-9 in hippocampal tissues of estradiol-treated, which suggest that modulation of Scd2 gene expression by estradiol largely differs from that of Scd1. Further, although the Free-DHA diet used here contains twice as much the content of monoenes than the High-DHA diet (mostly in the form of oleic acid), levels of 18:1n-9 observed in hippocampal tissues remained fairly similar between diets and rather constant irrespective of genotypes. This phenomenon was particularly relevant in the APP/PS1 genotype and in the presence of estradiol, which suggest that these factors are critical in monoenoic fatty acid homeostasis. Therefore, we concluded that dietary DHA has a profound impact on the profile of hippocampal fatty acids, well beyond than the mere presence of DHA in membrane phospholipids, but the final effect on hippocampal lipid fingerprint is conditioned by the relative weight of the other factors analyzed.
In summary, we demonstrate that dietary DHA, circulating estrogens, and the presence of a familial AD-like genotype, all influence the hippocampal lipid structure. To the best of our knowledge, this is the first study integrating the combined influence of all these factors using a multivariate approach. Furthermore, we demonstrate that the influence of these factors in the hippocampal lipid matrix is not entirely independent, but is rather marked by important interactions between factors, which differentially affect hippocampal lipid homeostasis at different levels, from variations in the quantitative composition of lipid classes and fatty acids to changes in the expression levels of lipogenic genes. Finally, we found that dietary incorporation of DHA and administration of physiological doses of estrogens to APP/PS1 animals, largely restores the normal lipid profiles observed in the hippocampus of non-transgenic animals. These observations suggest that, if extrapolated to humans, nutraceutical interventions together with (fito)estrogen supplementation, might provide a novel therapeutic approach for the prevention and progression of AD, particularly at the earliest recognizable symptoms of AD development.
Footnotes
ACKNOWLEDGMENTS
Supported by grants SAF2010-22114-C02-01/02 and SAF2014-52582R from Ministerio de Economía y Competitividad (Spain). VC-S held a research fellowship from Gobierno de Canarias (Spain). DQA holds a pre-doctorate fellowship from Fundación CajaCanarias (Spain). We are grateful to CEI-Canarias, Campus Atlántico Tricontinental (Universidad de La Laguna) and to Biosearch Life-Puleva Biotech for continuous support to our project at the Laboratory of Membrane Physiology and Biophysics.
