Gene Signaling Pathways Mediating the Opposite Effects of Prepubertal Low-Fat and High-Fat n-3 Polyunsaturated Fatty Acid Diets on Mammary Cancer Risk

A high dietary intake of n-3 polyunsaturated fatty acids (PUFA), as present in fish and some vegetable oils (canola and linseed), can reduce the risk of developing breast cancer (1–3) and inhibit metastasis (4) in human studies. However, some studies report either no change or a significant increase in breast cancer risk among women who consumed high levels of n-3 PUFAs (5–8). Data from animal studies also have generated conflicting data: Some studies show that a high dietary intake of n-3 PUFAs inhibits mammary tumorigenesis (9, 10), whereas this effect is not seen in other studies (11, 12). Perhaps reflecting these inconsistencies, we found that prepubertal exposures to a low-fat or a high-fat n-3 PUFA diet had opposing effects on later breast cancer risk (13). A low-fat n-3 PUFA diet reduced later susceptibility to develop carcinogen-induced mammary tumors, but a high-fat n-3 PUFA diet increased cancer risk (13).

We have extended our previous study to identify changes in gene signaling pathways in the mammary glands that could explain the opposing effects of prepubertal low-fat and high-fat n-3 PUFA exposures on mammary tumorigenesis. Our earlier study showed that these diets differentially affected cell proliferation and apoptosis in the mammary gland, and therefore, we were particularly interested in studying genes involved in these pathways (13). Because prepubertal exposure to both low-fat and high-fat n-3 PUFA diet induced lipid peroxidation (13), genes regulating oxidative damage were also of interest. Therefore, we performed a gene microarray analysis, focusing on antioxidant genes and genes associated with proliferation and apoptosis. The analysis, done using tissues from our previous study, used novel analytic approaches to find functionally relevant gene expression pathways in the context of nutrigenomic animal studies. We also determined protein levels of two key genes linked to cell proliferation: cyclin D1, which is downstream of multiple pathways leading to increased cell proliferation (14, 15), and peroxisome proliferator–activated receptor γ (PPARγ), which inhibits cell proliferation and induces differentiation (16). The latter gene was of particular interest because n-3 PUFAs serve as ligands for PPARγ.

Rats were exposed to low-fat and high-fat n-3 and n-6 PUFA diets during prepuberty as previously described (13). Briefly, timed pregnant Sprague-Dawley rats were obtained from Charles River on gestation day 10 and fed AIN93G diet. After delivery, 10 female pups from three to four litters were housed per a nursing dam. Nursing dams were either kept on a semipurified AIN93G diet or switched to one of the three experimental diets when the offspring were 5 d old. The dietary groups were (a) low-fat n-6 PUFA reference diet (AIN93G diet); (b) high-fat n-6 PUFA diet; (c) low-fat n-3 PUFA diet; and (d) high-fat n-3 PUFA diet. The low-fat diets contained 16% energy from fat and the high-fat diets contained 39% energy from fat. Corn oil was the source of n-6 PUFAs (contains 60% n-6 PUFAs and 1% n-3 PUFA), and menhaden oil of n-3 PUFAs (contains 25% of n-3 PUFAs and 2% n-6 PUFAs). The low-fat n-3 PUFA diet consisted of 35 g/kg of menhaden oil and 35 g/kg of corn oil, whereas the high-fat n-3 PUFA diet consisted of 70 g/kg of menhaden oil and 120 g/kg of corn oil. The low-fat n-6 PUFA diet consisted of 5 g/kg menhaden oil and 65 g/kg of corn oil, and the high-fat n-6 PUFA diet consisted of 15 g/kg of menhaden oil and 175 g/kg of corn oil. Although the absolute levels of n-3 and n-6 PUFAs varied, the n-6 PUFA/n-3 PUFA ratio was kept similar in the n-6 PUFA diets (13:1 in the low-fat n-6 PUFA diet and 12:1 in the high-fat n-6 PUFA diet). Similarly, the n-3 PUFA/n-6 PUFA ratio was also kept similar in the n-3 PUFA diets (1:1 for the low-fat n-3 PUFA diet and 1:2 in the high-fat n-3 PUFA diet). This eliminated the possibility of differences between the low-fat and high-fat n-3 PUFA diet being caused by a change in the n-3/n-6 PUFA ratio.

Pups initially consumed the diets through the dam's milk from postnatal day 5 through postnatal day 15. It has been shown that the dam's milk closely reflects what is taken in via the diet in terms of fatty acids (17). Although still nursing, pups begin to consume food pellets at about postnatal day 15, and therefore at that age onward, they obtained n-3 and n-6 PUFAs both through milk and through consuming food pellets. At day 26, all pups were weaned and switched to the reference AIN93G diet.

When the rats were 50 d of age, some were administered 10 mg of the mammary carcinogen 7,12-dimethylbenz[a]anthracene (Sigma Chemical Co.) by oral gavage. The low-fat n-6 PUFA control group was composed of 23 rats; the high-fat n-6 PUFA group had 25 rats; the low-fat n-3 PUFA group had 25 rats; and the high-fat n-3 PUFA group had 24 rats. Carcinogen was dissolved in peanut oil and given in a volume of 1 mL. Animals were checked weekly for mammary tumors by palpation. Tumor growth was measured with a caliper, and the length, width, and height of each tumor were recorded. Animals were sacrificed when the tumor burden was ∼10% of total body weight. All remaining animals, including those that did not develop tumors, were sacrificed 18 wk after 7,12-dimethylbenz[a]anthracene administration. All animal procedures were approved by the Georgetown University Animal Care and Use Committee, and the experiments were done following the NIH guidelines for the proper and humane use of rats in biomedical research.

Tumor incidence was calculated by the methods developed by Kaplan and Meier, followed by the log-rank test. Differences in final tumor incidence among groups were compared using χ² test. The differences were considered significant at P < 0.05. All probabilities were two-tailed.

To identify changes induced by prepubertal dietary PUFA exposures that might have mediated the effects of these diets on mammary tumorigenesis, we obtained serum and removed the 3rd thoracic and 4th abdominal mammary glands from 26- and 50-d-old rats that were fed different PUFA diets during prepuberty (but not exposed to 7,12-dimethylbenz[a]anthracene). The 3rd thoracic and 4th abdominal mammary glands from each side were removed because these are the easiest to locate and remove anatomically. The glands were snap frozen in liquid nitrogen and stored at −80°C, except for one of the 4th glands, which was processed for immunohistochemistry. Total RNA from the other 4th mammary gland was used in the microarray analysis, whereas total RNA from the 3rd mammary gland was used for real-time PCR studies. Therefore, there was total RNA from each animal used in both the microarray studies and the real-time PCR studies. RNA from the animals was not pooled. An additional set of rats was used to measure BRCA1 expression, cyclin D1 protein, PPARγ protein, and DNA damage levels.

Total RNA was extracted from the mammary glands of rats exposed to low-fat or high-fat n-3 PUFA and n-6 PUFA diets. Frozen tissue samples were placed in 1 × 1-in. plastic bags, pulverized on dry ice, transferred to 35-mL conical Oakridge tubes (Nalgene), and weighed. Tissues were homogenized in 1 mL of TRIzol reagent (Invitrogen Corporation) per 50 mg of tissue using a PowerGen 35 handheld homogenizer (Fisher Scientific) with RNase-free disposable OMNI-Tips (Fisher Scientific) for 30 s. From this point, procedures were followed according to the manufacturer's instructions for use of the TRIzol reagent. The quantity and quality of RNA were measured by comparing the absorbance ratios (A₂₆₀/A₂₈₀) obtained using a Beckman DU640 Spectrophotometer. There were a total of six rats fed the low-fat n-3 PUFA diet, six fed the low-fat n-6 PUFA diet, five fed the high-fat n-6 PUFA diet, and five fed the high-fat n-3 PUFA diet.

We used GF300DS rat filters (Research Genetics, Inc.) that contain 5,531 known genes, 192 “housekeeping” genes, and 192 control genes on each filter. To synthesize the labeled cDNA probe, 2 μg of total RNA were incubated at 70°C for 10 min with 2 mg of oligo dT and then chilled on ice for 2 min. The primed DNA was incubated at 37°C for 90 min in a solution containing 1× first strand, 3 mmol/L DTT, 1 mmol/L dGTP/dTTP, 300 units of reverse transcriptase, 50 mCi of [³³P]dCTP, and 50 mCi of [³³P]dATP. The second strand was synthesized by adding 1× reaction buffer, 100 units of DNA polymerase I, 500 ng of random primers, 1 mmol/L dGTP/dTTP, 50 mCi of [³³P]dCTP, and 50 mCi of [³³P]dATP. The reaction was incubated for 2 h at 16°C. Radiolabeled probe was purified using a BioSpin-6 chromatography column (Bio-Rad) and denatured by boiling for 3 min.

Filters were prehybridized in Microhyb solution (ResGen) for 2 h at 42°C. Purified probe was added to the hybridization roller tube containing the prehybridized GeneFilter and incubated for 12 to 18 h at 42°C in a Robin Scientific Roller Oven. Each hybridized GeneFilter was washed twice in 2× SSC, 1% SDS at 50°C for 20 min and once at 55°C in 0.5× SSC, 1% SDS for 15 min. Hybridization signals were detected by phosphorimage analysis using a Molecular Dynamics Storm phosphorimager.

All data processing and analysis was implemented in the Mathworks MATLAB programming software environment (Fig. 1A).

The raw intensity of each spot on a filter was imported into the Pathways 4.0 software (Research Genetics), which was used to correct the local nonspecific binding of the probe to filter for each spot (background correction). Raw intensity data were normalized such that every raw intensity value from a filter was divided by the mean intensity values of that array. For some genes, the use of radiolabeled probes can produce signal bleeding into adjacent spots. We used an algorithm that iteratively identifies signal bleed effects and includes genes found to be free of bleeding effects in at least 70% of all arrays.⁵

Following preprocessing, dimensionality in the remaining data was reduced by eliminating those genes least likely to contain relevant discriminant and/or biological information. We first applied, in a supervised manner, a series of simple univariate statistical filters: Student's t test, t test for unequal variances (assumes normal distribution of the data), and the Wilcoxon test (nonparametric distribution-free) were applied without correction for multiple comparisons. Because the distribution of the data among and within replicate experiments and for individual genes cannot be determined accurately in such high dimensions (18), logarithm-transformed and nontransformed data also were compared. The level of significance was set at P < 0.05, using the inflated type I error (many false positives) and reduced type II error (few false negatives) to exclude those genes least likely to be truly differentially expressed. A further filter of relative fold changes ≤0.56 for down-regulation and ≥1.8-fold for up-regulation was applied to identify those genes most likely to have biologically relevant changes in expression among groups. Following this dimensionality reduction, all spots in all arrays that remained within the reduced dimensional data set and were determined to be free of signal bleeding were visually inspected to further assess whether or not these signals should be included for subsequent data analysis. This approach generated a final reduced dimensional data set of 282 genes (dimensions) that was used for data analysis.

Whereas dimensionality reduction applied univariate criteria, for gene selection we used an approach designed to preserve the joint discriminant power of genes within the reduced dimensional data set. Thus, the individual gene selection algorithm excludes genes based on weak joint discriminant power, whereas the multilayer perceptron identifies genes where the joint discriminant power is high (classification accuracy of >70%). The profile selection algorithm first eliminates genes by their lack of contribution to the Fisher's discriminant components of the data set, and further eliminates those genes that least change the trace of a weighted Fisher's scatter matrix (19). Comparisons were made between the low n-3 versus high n-3 PUFA array filters using the mean intensity normalized results obtained from each filter for the genes in the 282-dimensional data set.

We first ran the gene selection algorithm on all samples to select a 20-gene data set for visualization, to assess the likelihood that we would retain sample separability. We obtained three-dimensional projections by both principal component analysis and discriminant component analysis (19, 20). Principal component analysis is an unsupervised method that maximizes the capture of variance within the data set. Discriminant component analysis is a supervised method based on a Fisher separability matrix that uses class information to maximize separation among treated groups while minimizing variability within groups (19, 20). For each projection, the reconstruction error was calculated by dividing the sum of the variances captured by the three top principal or discriminant components by the total variance spanned by all gene dimensions.

We then used a leave-three-out method and ran the algorithm through 100 iterations to select 100 gene profiles, each profile containing 20 genes (Fig. 1B). At each gene selection iteration, (a) three random samples from each group are excluded; (b) a 20-gene profile is obtained; (c) this profile is used as the input data to train a multilayer perceptron classifier; and (d) each classifier is trained using a leave-two-out method (100 training iterations) and (e) validated against the four samples left out before gene selection at the first step in the process. Twenty-gene profiles that produced multilayer perceptrons with a classification accuracy of >70% for the independent data set were collated into a final gene list. For the classifier step, multilayer perceptron prediction models were built using three layers, one output node, and three hidden nodes. Weighted inputs were fed into the hidden layer and then transferred to the outer layer, with both layers using a tan-sigmoid transfer function. The multilayer perceptron was trained using a Quasi-Newton numerical optimization technique (“trainbfg” Matlab routine; a back-propagation method).

To ensure that validation was independent of the microarrays, we used RNA from mammary glands not used in the microarray studies, as described above. RNA was extracted from each individual mammary gland and cDNA was reverse transcribed from 100 μg/mL of total input RNA using TaqMan reverse transcription reagents as described by the manufacturer (Applied Biosystems). Next, PCR products were generated from the cDNA samples using the TaqMan Universal PCR Master Mix (Applied Biosystems) and Assays-on-Demand (Applied Biosystems) for the appropriate target gene. The 18S Assay-on-Demand (Applied Biosystems) was used as an internal control. All assays were run on 384-well plates so that the cDNA sample from each mammary gland was run in triplicate for the target gene and the endogenous control. Real-time PCR was done on an ABI Prism 7900 Sequence Detection System and the results were assessed by relative quantitation of gene expression using the ΔΔC_T method.

Formalin-fixed tissue sections (5 μm) obtained from the 3rd thoracic mammary glands of five 50-d-old rats per group were deparaffinized in xylene, hydrated through graded alcohol, and incubated with 3% H₂O₂ for 10 min to block endogenous peroxidases. Nonspecific binding was blocked with normal rabbit serum from the Vectastain Elite ABC Kit (Vector Laboratories, Inc.) for 20 min., blocked, incubated with cyclin D1 antibody (1:700; #RB-212, Lab Vision Corporation) or PPARγ antibody (1:100; #H-100, Santa Cruz Biotechnology), washed, treated with biotinylated goat antiserum to mouse IgG, and then incubated with streptavidin-peroxidase conjugate (ARK kit, DakoCytomation). Antigen-antibody complexes were visualized by 3′3-diaminobezidine and counterstained with hematoxylin stain, dehydrated, and mounted. Control slide was incubated with normal mouse serum. For cyclin D1, the percentage of positive cells was determined by calculating the number of cells that had positive staining (only darkly stained cells were counted) per 1,000 cells per mammary gland structure (terminal end buds, lobules, and ducts). PPARγ expression, in turn, was assessed in terminal end buds, lobules, and ducts using a scale of 0 to 5 for percentage staining and 0 to 3 for staining intensity. The combined values were used for statistical analysis. Slides were blindly evaluated with the Metamorph software.

DNA damage was measured in the serum using an 8-hydroxy-2′-deoxyguanosine (8-OHdG) enzyme immunoassay (Oxis Health Products, Inc.). Serum was centrifuged at 2,000 × g for 90 min and the supernatant collected. The level of 8-OHdG was measured in the supernatant as described by the manufacturer. Each sample was run in triplicate and the mean value was calculated.

Results for the data obtained on 8-OHdG enzyme and real-time PCR were analyzed with SigmaStat software using one-way ANOVA, separately at 26 and 50 d of age. Where appropriate, between-group comparisons were done using Tukey's multiple comparisons test. The differences were considered significant at P < 0.05. All probabilities were two-tailed.

As previously reported (13), rats fed a low-fat n-3 PUFA diet during prepuberty developed significantly fewer mammary tumors than rats fed a high-fat n-3 PUFA diet (Fig. 2). Additionally, mammary tumor incidence was significantly lower in the rats fed a prepubertal low-fat n-3 PUFA diet compared with the reference diet (P = 0.0327). Conversely, rats fed a prepubertal high-fat n-3 PUFA diet exhibited significantly increased mammary tumor incidence, compared with the reference diet (P = 0.0198). No changes in mammary tumor latency or the number of tumors per rat (tumor multiplicity) were noted between the two groups.

Our first approach to data analysis was to determine whether n-3 and n-6 PUFA diets produce different patterns of gene expression. This comparison would establish our ability to find genetic changes associated with exposure to two different types of PUFAs. Because n-3 and n-6 PUFAs induce significantly different changes in gland morphology (13), we expected this comparison to also produce marked changes in mammary gene expression. Indeed, if we could not separate these two groups, it would be unlikely that we could separate any other groups. Therefore, the initial goal was to identify a subset of genes differentially expressed between all n-3 PUFA–exposed (low + high) and all n-6 PUFA–exposed (low + high) mammary glands (Table 1).

To determine if these genes are truly differentially regulated in a meaningful pattern, subgroups of samples were used for gene selection, neural network training, and evaluation, as described in Materials and Methods. The remaining samples (those not used for gene selection or neural network training/evaluation) were used as an independent data set for testing the neural network classifier.

Figure 3A shows discriminant component analysis visualization of the top 20 potentially discriminant genes from the n-3 and n-6 PUFA-exposed glands, capturing 97% of the cumulative variance in the data. To confirm the discriminant power of this data set, a multilayer perceptron (nonlinear neural network classifier) was built to predict in an independent sample set whether they are from an n-3 or n-6 PUFA-exposed mammary gland. We achieved 100% accuracy (no misclassifications) during training, 79 ± 3% accuracy with the evaluation data set, and 73 ± 2% accuracy for predicting the dietary exposure in the independent data set.

We then determined whether some genes are differentially expressed in mammary glands exposed to a high-fat versus a low-fat diet irrespective of whether the fat is n-3 or n-6 PUFA (Table 2). We used the same data and data analysis approaches but combined the gene expression profiles for the analysis such that we compared all low-fat (n-3 + n-6 PUFA) with all high-fat (all n-3 + all n-6 PUFA) exposed rats. The data in Fig. 3B represent the multidimensional scaling from 20 dimensions to 3 dimensions and capture 80% of the cumulative variance in the data. We used three samples from each group that were not used for either gene selection or network training/evaluation as the independent data test set. We achieved 100% accuracy (no misclassifications) during training, 78 ± 3% accuracy with the evaluation data set, and 80 ± 4% accuracy for predicting the dietary exposure in the independent data sets. The data in Fig. 3B provide compelling evidence that low-fat and high-fat diets differentially affect mammary gland gene expression.

Having established the ability of gene expression profiles to discriminate between dietary exposures to n-3 and n-6 PUFAs and between exposures to low-fat and high-fat diets, we asked whether we could identify a gene set that discriminates between the effects of diet on susceptibility to mammary carcinogenesis. Thus, we wanted to determine whether we could identify a gene subset that would discriminate between mammary glands with a low cancer susceptibility (low n-3 PUFA diet) and those with a “normal/high” susceptibility (high n-3, low n-6, and high n-6 PUFA diets; Table 3).

The data in Fig. 3C represent the multidimensional scaling from 20 dimensions to 3 dimensions and capture 96% of the cumulative variance in the data. The projection shows that the samples from the two cancer susceptibilities are linearly separable in gene expression data space. Because we have a limited number of replicates in the low n-3 PUFA exposed group, we built and trained a neural network classifier using genes selected from all 6 of 6 low n-3 PUFA-exposed samples and 8 of 11 n-6 PUFA-exposed samples. This approach left all high n-3 PUFA and 3 of 11 n-6 PUFA samples as independent data for classifier testing. We achieved 100% accuracy (no misclassifications) during training and evaluation. Whereas the high accuracy of predicting the low cancer susceptibility (during evaluation), in part, reflects the use of all these data for gene selection, the data evaluation is a “leave-two-out” method (two samples are randomly excluded from each iteration during evaluation of the training set). The independent test data are exclusively from high susceptibility profiles, but we achieve a robust 92 ± 1% accuracy for predicting this phenotype.

In the final analysis, a total of 91 genes (excluding ESTs) were identified as differentially expressed in the mammary glands of rats prepubertally fed a low-fat n-3 PUFA diet when compared with a high-fat n-3 PUFA group (Table 4), using the criteria established above. Four separate statistical tests were used to identify these genes. Because PUFAs induce lipid peroxidation that can potentially increase oxidative stress (21–23), and we previously showed an increase in lipid peroxidation in these animals (13), we looked at the list of 91 genes specifically for genes with these implied or known functions in the gene subset. Four genes that were found to be up-regulated in the mammary glands of low-fat n-3 PUFA-fed rats were directly and/or indirectly associated with protection from oxidative stress: thioredoxin (24), heme oxygenase (25), NADP-dependent isocitrate dehydrogenase (26), and metallothionein III (27). Differential expression of all four genes was confirmed by real-time PCR (Fig. 4). All four genes were expressed at a significantly higher level in the 50-day-old prepubertally low-fat n-3 PUFA fed rats than in the control rats fed low-fat n-6 PUFA diet. Only heme oxygenase was expressed at a lower level in the high-fat n-3 PUFA group than in the controls. These changes were not seen in the mammary glands of 26-day-old rats. Further, at 50 days of age, none of these genes were differentially expressed between the low-fat and high-fat n-6 PUFA groups, but in the microarray analysis they were significantly down-regulated in the mammary glands of high-fat n-3 PUFA groups when compared with all the other three groups (data not shown). To ensure validation, the glands used were obtained from different rats than those from which the RNA was extracted for the microarray analysis.

We also expected the gene expression microarray experiments to implicate other key signaling pathways involved in the effect of dietary exposures on tumorigenicity, such as apoptosis. From within this data set, our analyses identified cardiotrophin-1 as a candidate gene for further evaluation. Cardiotrophin-1 phosphorylates, and thereby activates, the Akt protein (28), a signaling molecule implicated in several key functions in the mammary gland, including promoting cell survival (29, 30). In our earlier study, we found that prepubertal exposure to a high-fat n-3 PUFA diet elevates Akt phosphorylation (13). We confirmed differential expression of cardiotrophin-1 mRNA by real-time PCR in the high-fat n-3 PUFA group (P < 0.001; Fig. 5). This increased expression of cardiotrophin-1 in the 50-day-old mammary glands of the high-fat n-3 PUFA group was not observed in the 26-day-old mammary glands of the high-fat n-3 PUFA group.

Some of the genes in Table 1 are those that have been linked to cell proliferation and differentiation. Instead of confirming their expression by reverse transcription-PCR, we decided to assess protein levels of two genes closely linked to cell proliferation: cyclin D1, which increases proliferation, and PPARγ, which induces differentiation and reduces cell proliferation. The data indicated that the levels of cyclin D1 were elevated in the mammary glands of rats exposed to a high-fat n-3 PUFA diet during prepuberty (P < 0.001; Fig. 6A), consistent with increased cell proliferation reported in these rats (13). PPARγ protein levels were significantly elevated in the mammary glands of prepubertally low-fat n-3 PUFA diet fed rats (P = 0.035; Fig. 6B), and this is in agreement with reduced cell proliferation noted in their mammary glands (13).

The changes in genes repairing oxidative damage led us to hypothesize that DNA damage repair pathways may be altered. BRCA1 is a tumor suppressor gene that affects DNA damage repair and is strongly implicated in a high proportion of inherited breast cancers (31, 32). We measured Brca1 mRNA expression in the mammary glands of 26- and 50-day-old rats fed low-fat or high-fat n-3 PUFA or n-6 PUFA diets during prepuberty by real-time PCR.

Twenty-six-day-old rats fed the high-fat n-3 PUFA diet exhibited a significant decrease in Brca1 expression compared with the reference diet and the low-fat n-3 PUFA diet (P < 0.001; Fig. 7A). At 50 days of age, there was a significant 30% reduction in Brca1 expression in the rats fed the high-fat n-3 PUFA diet compared with the rats fed the low-fat n-3 PUFA diet (P = 0.026), but no significant differences in Brca1 expression were observed between the rats fed either of the n-3 PUFA diets and the reference diet (Fig. 7B).

DNA damage has been associated with increased breast cancer risk (33, 34). We measured DNA damage caused by oxidative stress by determining 8-OHdG levels in the serum of low-fat and high-fat n-3 PUFA fed rats and rats fed the reference diet. Rats fed the high-fat n-3 PUFA diet had the highest level of DNA damage at both 26 days (P < 0.001) and 50 days of age (P < 0.001; Fig. 8). Furthermore, rats fed the low-fat n-3 PUFA diet had the lowest amount of DNA damage at these two ages when compared with the reference diet–fed rats and the rats fed the high-fat n-3 PUFA diet (P < 0.05).

In our earlier study, prepubertal dietary exposure to a low-fat n-3 PUFA diet reduced, whereas a high-fat n-3 PUFA diet increased, susceptibility to develop carcinogen-induced mammary tumors in rats (13). These observations were confirmed in the present study. The increase in mammary tumorigenesis in rats fed the high-fat n-3 PUFA diet during prepuberty was unexpected, mostly because prepubertal exposure to a high-fat n-6 PUFA had no effect on mammary tumorigenesis and n-3 PUFAs are generally considered protective against breast cancer. For example, findings in athymic nude mice show that n-3 PUFAs inhibit the growth of human breast cancer cells (35–37). However, a diet containing menhaden oil, when given after carcinogen administration, did not modify the growth of N-nitrosomethylurea–induced mammary tumors in rats (11). Further, when the amount of n-3 PUFAs relative to n-6 PUFAs in the diet was increased, 7,12-dimethylbenz[a]anthracene–induced mammary tumor incidence and tumor weights were increased in rats (12).

To identify molecular pathways that may have mediated the opposing effects of the low and high prepubertal n-3 PUFA exposures on mammary cancer risk, we performed gene microarray analyses. The ability of our microarray analysis to accurately identify independent samples as those obtained from rats fed n-3 PUFA diets versus n-6 PUFA diets, and those obtained from rats fed low versus high-fat diets, regardless of the source of fat, shows that the genes selected are expressed or repressed in both patterns and at levels consistent with the model. This is an appropriate and rigorous test of the approach because the initial goal was simply to determine if the molecular profiles are consistently different. Building a multigene predictor is a more efficient and less resource intensive test of the selected genes than would be obtained by confirming expression of multiple genes via real-time PCR or Western assays.

Several genes involved in protecting the mammary gland from oxidative DNA damage were up-regulated in the low-fat n-3 PUFA fed rats, supporting the findings that this diet reduced DNA damage and mammary tumorigenesis. The altered genes included thioredoxin, heme oxygenase, metallothionein III, and NADP-dependent isocitrate dehydrogenase. Each of these genes has an antioxidant function and can protect cells against DNA damage due to oxidative stress and reactive oxygen species (27, 38–42). Changes in gene expression were seen only in 50-day-old rats but not in younger rats (i.e., at the time when they were still on the special diets). These results suggest that the changes in antioxidant genes and genes regulating cell survival may reflect functional, long-term changes in the cell membrane fatty acid composition. It also is possible that the changes in gene expression are manifested in the context of mature mammary gland, but not during pubertal development.

Consistent with the increase in expression of antioxidant genes, rats exposed prepubertally to the low-fat n-3 PUFA diet incurred lower levels of DNA damage, as indicated by the reduced levels of 8-OHdG. This observation could also reflect several other differentially altered functions besides a change in the antioxidant genes. For example, the high-fat n-3 PUFA diet might have induced high levels of DNA damage as a consequence of the increasing lipid peroxidation (43, 44) and a decrease in apoptosis (13). The failure of cells to undergo apoptosis in the prepubertally high-fat n-3 PUFA diet fed rats may be related to the observation that cardiotrophin-1 was up-regulated in their mammary glands. Cardiothropin-1 phosphorylates Akt, leading to promotion of cell survival (28). We confirmed the increase in cardiotrophin-1 expression by real-time PCR. Previously, we have shown that phosphorylated Akt is increased in the mammary glands of rats fed the high-fat n-3 PUFA diet during prepuberty (13). Increased activation of Akt in the prepubertally high-fat n-3 PUFA-fed rats implies that signaling through this pathway could be inhibiting apoptosis in cells with damaged DNA, with a consequent increase in tumorigenesis.

The effect of prepubertal exposure to a high-fat n-3 PUFA diets on DNA damage repair pathways was also assessed by measuring the level of Brca1 mRNA expression. Germ-line BRCA1 mutations are linked to familial breast cancers (45, 46), reflecting its function in DNA damage repair and recombination, processes related to maintenance of genomic integrity, control of cell proliferation, and regulation of gene transcription (31, 47). A previous study has reported up-regulation of BRCA1 expression by n-3 PUFAs in breast cancer cell lines, but a reduced expression in normal mammary cells (48). We observed a decrease in Brca1 mRNA expression in the mammary glands of rats exposed prepubertally to the high-fat n-3 PUFA diet, and this could have further contributed to the increased levels of DNA damage in these rats. Thus, both the down-regulation of antioxidant genes and the impairment in the DNA repair mechanisms in the prepubertally high-fat n-3 PUFA diet fed rats might have contributed to the increase in their mammary tumorigenesis. Up-regulation of cyclin D1 expression in these rats might also be involved in increasing their mammary tumorigenesis, and perhaps explains the increase in cell proliferation noted in these rats (13).

The results generated in this study also shed light on why low-fat n-3 PUFA diet, when consumed before puberty onset, reduces mammary cancer risk, although it increases lipid peroxidation (13). Cells have multiple ways to defend themselves from the toxic effects of free radicals, including increased activation of heme oxygenase, which in turn generates antioxidant products. We found that during adulthood, rats fed a low-fat n-3 PUFA diet during prepuberty exhibited increased expression of heme oxygenase and several antioxidants, suggesting that this dietary exposure can cause a permanent up-regulation of antioxidant genes. Further, low-fat n-3 PUFA groups exhibited reduced expression of cardiotrophin-1, promoting apoptosis, and perhaps explaining why rats exposed to a low-fat n-3 PUFA diet during prepuberty exhibited reduced levels of DNA adducts. Our results also implicated up-regulation of PPARγ expression in the mammary glands of these rats. n-3 PUFAs act as ligands for this receptor (16), which is known to inhibit cell proliferation and induce differentiation (16). Thus, elevated PPARγ levels in the mammary glands of rats fed a low-fat n-3 PUFA diet during pregnancy may be linked to their reduced mammary cancer risk.

In conclusion, we studied plausible mechanisms explaining why rats fed prepubertally a low-fat n-3 PUFA diet are at reduced mammary cancer risk and why those fed a high-fat n-3 PUFA diet are at an increased risk. Based on the findings, increased antioxidant gene expression is strongly implicated in the reduction in mammary cancer risk. The increase in risk, in turn, may be caused by increased cell proliferation and survival and DNA damage. We further show that the levels of Brca1 expression are reduced in these rats. Together these changes could explain why rats prepubertally fed the high-fat n-3 PUFA diet are subsequently at an increased risk of developing mammary cancer.

No potential conflicts of interest were disclosed.