Liver tumorigenesis is promoted by a high saturated fat diet specifically in male mice and is associated with hepatic expression of the proto-oncogene Agap2 and enrichment of the intestinal microbiome with Coprococcus

Abstract

Liver cancer results in a high degree of mortality, especially among men. As fatty liver disease is a risk factor for development of hepatocellular carcinoma, we investigated the role of dietary fat type in tumor promotion by high-fat diets in mice after initiation with the chemical carcinogen diethyl nitrosamine. Tumor incidence and multiplicity were significantly greater in males than those in females. In males, fat type had complex effects on tumorigenesis. Preneoplastic foci were most prevalent in mice fed a polyunsaturated fat diet enriched in docosahexaenoic acid, whereas carcinomas and large visible liver tumors were significantly greater in mice fed a saturated fat diet made with cocoa butter relative to mice fed mono- or polyunsaturated fats. Different mechanisms thus seemed involved in early and late tumor promotion. The hepatic transcriptome and gut microbiome were assessed for traits associated with tumorigenesis. Hepatic expression of more than 20% of all genes was affected by sex, whereas fat type affected fewer genes. In males, the saturated fat diet induced expression of the proto-oncogene Agap2 and affected the expression of several cytochrome P450 genes, and genes involved in lipid, bile acid and fatty acid metabolism. The gut microbiome had a higher level of genus Akkermansia and a lower level of Firmicutes in females than in males. Males fed saturated fat had an altered microbiome, including an enrichment of the genus Coprococcus. In conclusion, sex and the dietary fat type affect the gut microbiome, the hepatic transcriptome and ultimately hepatic tumor growth.

In a mouse model of liver tumor promotion, a diet high in saturated fat promoted tumorigenesis more than a low-fat diet or diets high in mono- or polyunsaturated fats specifically in males.

Introduction

Hepatocellular carcinoma (HCC) is one of the leading causes of cancer-related deaths worldwide. Among risk factors for developing HCC are hepatitis viruses, alcohol consumption and obesity (1–3). There is a marked sex difference in the propensity to develop liver cancer, as its prevalence in men is at least twice as high as in women (4). Obesity rates have risen dramatically in the USA in recent decades and rates of liver cancer have risen significantly between 2005 and 2014 with an estimated 5% increase in risk for each unit of body mass index (4). It has been suggested that the increase in HCC in obesity is associated with development of metabolic syndrome and in particular is a consequence of increased incidence of non-alcoholic fatty liver disease (NAFLD) (5,6). The modern Western diet, with a high content of simple sugars, fat and cholesterol is thought to be an important component of the recent obesity epidemic as well as fatty liver disease (7,8). Animal studies have provided data implicating such diets in development of NAFLD and progression of liver pathology to non-alcoholic steatohepatitis and HCC (8). The molecular mechanisms underlying development of NAFLD and the lipid profile of NAFLD livers appear to differ depending on the level of caloric intake, the fat/carbohydrate ratio and the type of dietary fat (9–11). This in turn may affect the risk of NAFLD progression to HCC.

Although the effects of saturated fat diets on human health remain controversial (12), a number of epidemiological studies point to the Mediterranean diet in which the major fat is olive oil as giving a reduced risk for development of liver cancer (13,14). In agreement with a potential protective effect of monounsaturated fats, increased spontaneous liver tumor incidence was reported in C3H/He mice fed safflower oil with a high content of polyunsaturated fatty acids (PUFAs) compared with mice fed olive oil (15).

It has been suggested that dietary intake of ω-3 fats may exert a protective effect on development of NAFLD, including inhibition of hepatic de novo lipogenesis, increased expression of enzymes involved in fatty acid oxidation and inhibition of inflammation (7). A recent study further found that the ω-3 PUFA eicosapentaenoic acid attenuated HCC in mice fed a high-fat diet (16).

Our aim in the current study was to investigate the influence of dietary fat type on tumor promotion in a mouse model of hepatocarcinogenesis under isocaloric conditions. We hypothesized that diets with a high content of polyunsaturated ω-6 fats would lead to greater tumor promotion than diets enriched with saturated and monounsaturated fats, because such diets have been shown to produce greater liver pathology associated with elevations in lipid peroxidation (8–10). We further hypothesized that supplementation of a PUFA diet with the ω-3 fatty acid docosahexaenoic acid (DHA) would protect from tumor progression as the result of prevention of inflammation (7). To assess the model’s applicability to the sex dimorphism of human liver carcinogenesis, we included both sexes in the study. Because mechanisms of tumor promotion by dietary fat types are largely unexplored, we further investigated the effects of the diets on the liver transcriptome and the gut bacterial microbiome in order to identify traits being associated with the cancer phenotype in animals fed a high-fat diet.

Materials and methods

Animals

Pregnant C57Bl/6J mice were purchased from Jackson Laboratories. On postnatal day 13–14, pups were intraperitoneally injected with 10 mg/kg of diethyl nitrosamine (DEN). Mice were fed semi-purified AIN-93G-based diets ad libitum beginning at 4 weeks of age until killing after additional 15 or 30 weeks (Figure 1A). Untreated control mice were never treated with DEN and were housed on standard rodent chow. Mice were killed by CO₂. The major liver lobes were fixed in 10% formalin for histopathology. A minor liver lobe was flash frozen in liquid nitrogen for RNA extraction. The cecum with its contents was flash frozen in liquid nitrogen for later microbiome analysis. Abdominal fat pads were weighed. Frozen tissues were stored at −80°C. The study was approved by the Institutional Animal Care and Use Committee of LSU Health Sciences Center, New Orleans.

Figure 1.

Experimental design. (A) DEN-treated mice were fed semi-purified AIN-93G diets with different fat content and fat type. Control mice on regular chow were not treated with DEN. The number of males and females killed after 30 weeks on diet are indicated. (B) For each diet, the energy content (% of kcal) coming from fat is indicated together with the amount of energy coming from different classes of fatty acids. (C) Weekly intakes for each dietary group were recorded five times between mouse ages of 12 and 32 weeks. Columns with different letters indicate group means that were significantly different at P < 0.05. DEN, diethyl nitrosamine; DHA, docosahexaenoic acid.

Diets

Pelleted AIN-93G-based diets with 20% protein (casein) and either low 10% fat/70% carbohydrate (starch) or high 35% fat/45% carbohydrate (as % of total calories) were purchased from Research Diets Inc. Detailed diet compositions are listed in Supplementary Table 1, available at Carcinogenesis Online. The high-fat diets were isocaloric and the fats were of four types: cocoa butter, olive oil, corn oil and corn oil supplemented with DHA. Olive oil and corn oil were chosen as major dietary mono- and polyunsaturated fats for human consumption. Cocoa butter was chosen due to its high content of saturated palmitic acid combined with the ease of incorporating it into food pellets. The lipid composition is shown in Figure 1B. Cocoa butter, olive oil and corn oil diets were enriched to more than 50% with saturated, monounsaturated and PUFAs, respectively. The PUFA fraction in corn oil was predominantly the ω-6 fatty acid linoleic acid (17). The corn oil + DHA diet contained 8% ω-3 fatty acids. Diets with olive oil and corn oil + DHA were only provided to males. The other diets were fed to both sexes.

Macroscopical and histopathology liver examination and scoring

Visible liver tumors (labeled ‘macrotumors’) were recorded upon dissection. Tumor multiplicity was the number of macrotumors per animal. Tumor incidence was the percentage of animals with macrotumors. Lesions in formalin-fixed liver, an index of infiltration of inflammatory cells, and indices of steatosis were recorded by certified veterinary pathologists (F.D.P., T.W.) that were blinded to the experimental treatments. Lesions were classified as preneoplastic foci, adenomas or carcinomas. The lesion multiplicity was calculated as the number of observed lesions per number of sections analyzed for each animal. The incidence of lesions was the percentage of animals in which any lesions were observed in five slides of formalin-fixed liver sections. Serum alanine transaminase as an indicator of liver damage was measured as described previously(18).

Hepatic transcriptome analysis

Total RNA was isolated from the smallest liver lobe, none of which had visible tumors, using TRI Reagent. Global hepatic gene expression profiles were assessed via directional RNA sequencing (RNA-Seq). Expression of selected genes were quantified by one-step quantitative RT-PCR (qRT-PCR). Details are described in Supplementary Methodology, available at Carcinogenesis Online.

Analysis of gut microbiota

Changes in gut microbial ecology were assessed using 16S rRNA amplicon sequencing of cecal contents. Details are described in Supplementary Methodology, available at Carcinogenesis Online.

Statistical analysis

Incidence data were analyzed by pairwise comparison of groups with Fisher’s exact test. Other parameters were assessed with one-way analysis of variance with post hoc Student–Newman–Keul’s procedure for comparing group means. Post hoc contrasts for assessing sex differences and differences due to different classes of diets were assessed for statistical significance. Details are described in Supplementary Methodology, available at Carcinogenesis Online.

Results

High-fat diets promote hepatic steatosis in males

Semi-purified diets with different lipid composition were fed to DEN-treated mice ad libitum starting at 4 weeks of age (Figure 1A). Mice consumed about 23% more of the cocoa butter diet than of the other diets (Figure 1C). However, this did not translate into increased weight gain (Figure 2A). A time course of body weights is shown in Supplementary Figure 1, available at Carcinogenesis Online. After 30 weeks on diets, the males were significantly (P < 1 × 10^–6) heavier than females (Figure 2A). Males fed the corn oil and corn oil + DHA diets were significantly heavier than other animals. The relative weight of abdominal fat was likewise higher (P < 0.001) in males, with the highest levels reached in animals fed the corn-oil-containing diets (Figure 2B). By contrast, the relative liver weights did not differ significantly between the sexes (Figure 2C). Liver necrosis as assessed by serum alanine transaminase, was low with no significant group differences (Figure 2D). There was little hepatic steatosis in males fed the low-fat diet and in females, whereas males on the high-fat diets had higher steatosis scores (Figure 2E). The greater degree of hepatic steatosis in males than in females was statistically significant (P = 2 × 10^–6). Fibrosis was not observed in any of the mice. Infiltration with inflammatory cells was generally mild with the highest scores in males fed the low-fat diet (Figure 2F).

Figure 2.

Physiological parameters after feeding with semi-purified diets for 30 weeks. The parameters measured were body weight (A), weight of abdominal fat expressed in percent of body weight (B), liver weight expressed in percent of body weight (C), serum alanine transaminase (ALT) as in indicator of liver injury (D), an index of hepatic steatosis (E) and an index of infiltration in the liver by inflammatory cells (F). Columns with no letters in common indicated means that were significantly different at P < 0.05. There was no significant group difference for serum ALT.

The saturated fat diet promotes late stages of tumorigenesis in males

Subsets of three animals were killed after 15 weeks on each diet. A few foci of cellular alterations were observed in males, but no macrotumors were present (data not shown). After 30 weeks on diets, there was clear evidence of tumorigenic progression. The multiplicity of total liver lesions was significantly (P = 1.4 × 10^–9) higher in males than in females (Figure 3A). The incidences of lesions were also significantly higher in males than females (Figure 3B). Among males, the highest multiplicities of lesions were observed for the saturated cocoa butter and the polyunsaturated corn oil + DHA diets. However, the distribution of lesion types differed for the two diets. Preneoplastic foci were present at a significantly higher frequency in males fed the corn oil + DHA diet than in any other group (Figure 3C). By contrast there were significantly more adenomas for males fed cocoa butter than in any other group except males fed corn oil + DHA (Figure 3D). The multiplicity of carcinomas was significantly higher in males fed the cocoa butter diet than in any other group (Figure 3E). Multiplicities of preneoplastic foci, adenomas and carcinomas were significantly higher in males than in females (P = 1.3 × 10^–7, P = 2.0 × 10^–6 and P = 5.5 × 10^–5, respectively, for the main sex effects).

Figure 3.

Cancer parameters after feeding with semi-purified diets for 30 weeks. Precancerous and cancerous lesions were enumerated and classified by microscopy of liver sections from animals fed special diets for 30 weeks. The panels show the multiplicity of total lesions (A), the incidence of total lesions (B), the multiplicity of preneoplastic foci (C), the multiplicity of adenomas (D) and the multiplicity of carcinomas (E). A liver from a male fed the cocoa butter diet shows macrotumors, i.e. liver tumors that are visible to the naked eye. The image was sharpened to emphasize the tumors (F). Panels show the multiplicity of macrotumors per mouse (G) and the incidence of animals with macrotumors (H). Groups means with no letters in common are significantly different at P < 0.05.

The most severe cancer phenotype in this study was the presence of HCCs large enough to be visible to the naked eye (Figure 3F). The males fed the saturated fat diet had a significantly higher frequency of macrotumors than any other group (Figure 3G). Macrotumors were present in every single male on the cocoa butter diet, but in less than half of the mice in any other group (Figure 3H).

We conclude that hepatic tumor promotion was more pronounced in males than in females. Dietary DHA stimulated formation of preneoplastic foci, whereas the saturated fat diet preferentially promoted later phases of tumorigenesis.

The expression of the proto-oncogene Agap2 is associated with development of hepatic tumors

Altered hepatic gene regulation is a potential mechanism in carcinogenesis. For an unbiased approach, we assessed the whole-liver transcriptome by RNA-Seq. Means and standard errors of log₂-transformed values of reads per million (RPM) values for 26 074 genes are listed in Supplementary File 1, available at Carcinogenesis Online. For each gene, we performed analysis of variance with post hoc tests of contrasts for main effects of sex, chow versus semi-purified diets, low-fat versus high-fat diets, saturated fat versus unsaturated fats, monounsaturated fats versus polyunsaturated fats, and the effect of added DHA. The number of significant genes at P < 0.05 determined without corrections for the number of genes (least significant difference) as well as the number of significant genes at a false-discovery rate (FDR) less than 0.05 are listed in Supplementary Table 2, available at Carcinogenesis Online.

A contrast for the main sex difference in hepatic gene expression was significant at FDR < 0.05 for 6288 genes, meaning that the expression of more than 20% of all hepatic genes was affected by sex. In Figure 4A, the t-test statistic for sex differences is plotted against the analysis of variance F test statistic for any group differences. Most genes with group differences also show a clear sex difference in expression. This demonstrates that sex was the major factor affecting the liver transcriptome. Gene ontology (GO) analyses are reported in Supplementary File 2, available at Carcinogenesis Online. The sex-regulated genes affect a multitude of biological processes. GO analysis of the top 100 genes in terms of statistical significance (i.e. with the lowest P values) indicate enrichment of genes involved in e.g. lipid metabolism, fatty acid metabolism, redox reactions and the epoxygenase P450 pathway (Figure 4B). Cytochrome P450 genes were among the enriched genes with 17 of the 98 cytochrome P450 genes among the top 100 sex-regulated genes.

Figure 4.

Associations of the liver transcriptome with cancer parameters. (A) Sex disparity in hepatic gene expression accounts for most of the overall variation in the RNA sequencing (RNA-Seq) data set. For each gene, the t-test statistic for the main sex difference is plotted against the analysis of variance F test statistic for any groups differences. A point to the right of the vertical red line represents a gene with significant group differences at FDR < 0.05. A point above the upper horizontal red line or below the lower horizontal red line is a gene with significant sex difference at FDR < 0.05. Data points between the two horizontal red lines represent genes with no major sex difference in expression. (B) Gene ontology analysis of the top 100 genes with lowest P-values for a test of a main sex effect was conducted. Pathways with overrepresentation of genes relative to the reference genome are listed. (C) Exchanging a saturated fat diet with unsaturated fat diets affects hepatic expression of relatively few (70) genes. For each gene in the RNA-Seq data set, the t-test statistic for the effect of the cocoa butter diet in males relative to the unsaturated high-fat diets in males is plotted against the analysis of variance (ANOVA) F test statistic for any groups differences. A point to the right of the vertical red line represents a gene with significant group differences at an FDR < 0.05. A point above the upper horizontal red line or below the lower horizontal red line represents a gene with significant cocoa butter versus unsaturated fats difference at an FDR < 0.05. (D) Gene ontology analysis of the 70 genes with a significant effect of cocoa butter versus unsaturated high-fat diets in males at an FDR < 0.05 was conducted. Pathways with overrepresentation of genes relative to the reference genome are listed. (E) Venn diagram of genes that were regulated by the cocoa butter diet in males relative the three other high-fat diets in males at least 2-fold and significantly (P < 0.05 in a two-sample t-test). The seven genes that were regulated by the cocoa butter diet relative to all three other diets are listed with arrows indicating whether they are up- or downregulated by the cocoa butter diet. (F) Expression of Agap2 mRNA of RNA pools measured by RNA-Seq. (G) Expression of Agap2 mRNA of individual RNA samples measured by qRT-PCR. (H) Group means of Agap2 mRNA measured by qRT-PCR and RNA-Seq methods were strongly correlated. The correlation coefficient r is negative as C_T values decrease with increasing Agap2 concentrations. (I) The expression of the gene, Lipe, for the hormone-sensitive lipase as measured by RNA-Seq was correlated to the multiplicity of preneoplastic foci. (J) The expression of the gene, Tgfbr2, for TGFβ receptor type 2 as measured by RNA-Seq was correlated to the multiplicity of preneoplastic foci. (K) The expression of the gene Agap2 as measured by qRT-PCR was correlated to the multiplicity of hepatic macrotumors.

A contrast for the main effect of chow versus semi-purified diets was significant at FDR < 0.05 for 2305 genes. The contrast expresses not merely a diet-mediated difference, but also any difference caused by the DEN treatment. Among the top 100 most significantly affected genes were glucuronidation genes and 10 cytochrome P450 genes that all were upregulated in the chow-fed animals compared with animals on defined diets, probably due to the more complex chow diet upregulating genes involved in detoxification reactions.

Differences among the semi-purified diets affected smaller numbers of genes (e.g. Figure 4C). Contrasts of low versus high-fat diets, the saturated fat diet versus the unsaturated high-fat diets and the monounsaturated olive oil diet versus the polyunsaturated diets were significant at FDR < 0.05 for 397, 115 and 173 genes, respectively. A common theme of GO analyses of these gene lists were significant enrichment of genes involved in pathways of lipid and carboxylic acid metabolism (Supplementary File 2, available at Carcinogenesis Online). Inclusion of DHA in the corn oil diet did not lead to any marked changes in hepatic gene expression.

As males fed the saturated fat diet showed the strongest promotion of HCC, we determined distinguishing expression characteristics of males fed this diet. A male-specific contrast between the saturated cocoa butter diet and unsaturated fat diets was significant for 70 genes at FDR 0.05 (Figure 4C and Supplementary File 1, available at Carcinogenesis Online). Pathways of lipid metabolism, steroid metabolism and redox reactions were enriched (Figure 4D). We also performed pairwise comparisons of males on the saturated fat diet with males on any of the other high-fat diets. According to criteria of gene expression changes that were significant at P < 0.05 and corresponded to at least a 2-fold change in expression, there were a total of 51, 87 and 87 genes whose gene expression differed in cocoa-butter-fed males compared with males on the olive oil, corn oil and corn oil + DHA diets, respectively. Seven genes were common for all three comparisons (Figure 4E). For these seven genes, the expression in all dietary groups was assessed by both RNA-Seq of the RNA pools and qRT-PCR of all individual RNA samples (Figure 4F and G, and Supplementary Figure 2, available at Carcinogenesis Online). The RNA-Seq and qRT-PCR methods gave similar profiles with significant correlations (P < 0.05) between group means measured by the two methods (Figure 4H and Supplementary Figure 2, available at Carcinogenesis Online). Only Agap2 clearly and significantly distinguished the male saturated fat group from other groups. It is of interest that Agap2 (ArfGAP with GTPase domain, ankyrin repeat and PH domain) is a proto-oncogene (19).

To further assess whether gene expression levels were associated with development of liver lesions, we calculated the correlation coefficient between the expression level of each gene quantified as the log₂RPM value and the average multiplicity of preneoplastic foci, adenomas, carcinomas and visible tumors for the two mice whose liver RNA was included in each pool for the RNA-Seq data set (Supplementary File 1, available at Carcinogenesis Online). The frequency of preneoplastic foci was significantly correlated at an FDR < 0.05 to the expression of 761 genes. GO analysis shows enrichment of genes involved in e.g. triglyceride sequestering, lipid particle organization, fatty acid oxidation and metabolism, epoxygenase P450 pathway and nucleotide metabolism. The associations between two of the top five genes, Tgfbr2 and Lipe, and the multiplicity of preneoplastic foci are illustrated in Figure 4I and J. As indicated by these figures, the correlations are largely due to the sex difference in hepatic gene expression combined with the sex difference in the multiplicity of preneoplastic foci (Figure 3C). Determining correlations to preneoplastic foci in males only, results in just a single gene, Naf1, by the FDR < 0.05 criterion. Naf1 is involved in assembly of small nucleolar ribonucleoprotein complexes and telomere maintenance (20). GO analysis of the top 100 genes in males does not indicate enrichment for genes in lipid metabolism, but rather for genes involved in cell motility, signal transduction and protein phosphorylation, such as Src, Tgfbr2, Rhob, Bcr, Fyn, Naf1, Axin2 and Gadd45a (Supplementary Files 1 and 2, available at Carcinogenesis Online).

Surprisingly, no genes had expression levels that correlated sufficiently strongly with the frequency of adenomas and carcinomas to satisfy an FDR < 0.05, and only a single gene with very low expression, Sal4, fulfilled this criterion for macrotumors. This does not preclude associations between cancer and hepatic gene expression. The RNA-Seq data set may simply not have sufficient statistical power to detect them as significant by the FDR < 0.05 criterion. For example, the correlation between macrotumor frequency and Agap2 expression of individual samples measured by qRT-PCR was highly significant (P = 2.8 × 10^–6) (Figure 4K).

To determine whether the expression levels of Agap2 after feeding for 30 weeks with semi-purified diets were due to dietary effects on normal liver tissue or a consequence of presence of transformed cells, we also measured the hepatic Agap2 mRNA concentration after 15 weeks of special diet feeding when neoplasms were absent. The gene expression profile is similar to the 30 weeks profile with significant correlation between the group means at 15 and 30 weeks (Supplementary Figure 3, available at Carcinogenesis Online). The cocoa butter diet thus directly enhanced hepatic Agap2 expression.

The saturated fat cocoa butter diet leads to enrichment of Coprococcus in the gut microbiome

The gut microbiome has increasingly been recognized to exert effects on human health, including carcinogenesis (21,22). We analyzed the bacterial microbiome from the cecum after feeding semi-purified diets for 30 weeks. Similarities between bacterial cladograms from individual samples were assessed by the unweighted Unifrac distance measure (23). A subsequent principal coordinate analysis (Figure 5A) showed a mixture of mouse-specific, cage-specific and diet-specific clustering. Two corn-oil-fed females were distant from all other samples, probably due to antibiotic treatment to treat dermatitis in these females. The microbiomes from these two outliers also exhibited much lower taxon abundance (alpha diversity) than other samples (Figure 5B). These outliers were excluded from further analysis. The microbiome composition of chow-fed animals was more similar to each other than to animals fed the semi-purified diets with, e.g. an increased content of phylum Tenericutes and a marked depletion of phylum Verrucomicrobia (Figure 5C). Principal coordinate 2 separated subgroups of males fed corn oil and cocoa butter diets due to cage-specific enrichment of Proteobacteria and the genus Bacteroides. Clustering of all animals fed the cocoa butter diet was apparent by the third versus fourth principal coordinate, reflecting particular similarities in taxonomic composition in this group. The cocoa butter diet also resulted in lower alpha diversity than the other semi-purified diets (Figure 5B).

Figure 5.

Composition of the bacterial microbiome after 30 weeks with semi-purified diets. (A) Principal coordinate analyses based on the unweighted Unifrac distance measure reveals mouse-specific, cage-specific and diet-specific differences of the bacterial microbiomes. The panels depict principal coordinates 1, 2, 3 and 4 (PC1, PC2, PC3 and PC4, respectively). (B) The alpha diversity measured by the number of operational taxonomic units (OTU) normalized to 7600 sequences per sample is depicted. Females on the corn oil diet are split in two subgroups: one consisting of two outliers with a very low alpha diversity, and one consisting of all the other animals in this group. Columns with no letters in common indicate group means that were significantly different at P < 0.05. (C) The average composition of the microbiome for each group at the bacterial phylum level.

Although bacterial cladograms did not show strong sex-based clustering (Figure 5A), there were sex differences in the relative abundance of bacteria at the phylum level (Figure 5C). The content of Firmicutes was higher in males than females (Figure 6A), whereas phylum Verrucomicrobia represented by the single genus Akkermansia was more abundant in females (Figure 6B). The content of Akkermansia correlated inversely with both the multiplicity of preneoplastic foci and hepatic steatosis due to the clear sexual dimorphism of these traits (Figure 6C and D).

Figure 6.

Sex and diet characteristics of components of the bacterial microbiome. Major taxa that exhibit a significant main sex difference in the relative taxon abundance after 30 weeks on semi-purified diets include phylum Firmicutes (A) and genus Akkermansia of phylum Verrucomicrobia (B). The relative abundance of Akkermansia was negatively correlated with both the multiplicity of preneoplastic foci (C) and hepatic steatosis (D). Taxa that distinguishes males fed the cocoa butter diet for 30 weeks include the Bacteroidetes family S24-7 (E), the Firmicutes family Lachnospiraceae (F), and genus Coprococcus (G) within Lachnospiraceae. (H) The relative abundance of Coprococcus showed positive correlation with the multiplicity of hepatic macrotumors. (I) After 15 weeks on special diets, the bacterial microbiomes of both males and females were enriched with genus Coprococcus relative to the other diets. In column graphs, columns with no letters in common indicate group means that are significantly different at P < 0.05.

Due to the strong late-stage tumor promotion in males fed the cocoa butter diet, we determined clade characteristics distinguishing this group. Males on the cocoa butter diet exhibited a reduction in the content of the Bacteroidales family S24-7 and an enrichment of Lachnospiraceae (Figure 6E and F). Among Lachnospiraceae, bacteria of genus Coprococcus were significantly enriched in males fed the cocoa butter diet (Figure 6G). Every single male fed the cocoa butter diet had a higher content of Coprococcus that any other mouse. Although not reaching statistical significance, the highest mean percentage of Coprococcus in females also occurred for the cocoa butter diet. Taxa whose abundance were characteristic for males fed the cocoa butter diet also exhibited significant correlation to the macrotumor multiplicity, as is illustrated in the case of Coprococcus (Figure 6H). Enrichment of Coprococcus in both sexes also appeared after feeding the semi-purified diets for just 15 weeks (Figure 6I). We conclude that the cocoa butter diet led to a microbiome with enrichment of the genus Coprococcus.

Discussion

Liver cancer incidence in society is increasing at the same time as there is a substantial increase in obesity. There appears to be a clear link between Western lifestyle, high-fat diets, obesity, fatty liver disease and ultimately liver cancer (4–6). Our study demonstrates that a high-fat diet in males has higher tumor promotion potential than a low-fat diet, but that the fat source in isocaloric high-fat diets also has profound effects. The fat type had varying effects on different phases of carcinogenesis. However, the observed effects were different from what we had hypothesized: The polyunsaturated corn oil + DHA diet gave the highest prevalence of preneoplastic foci whereas the saturated cocoa butter diet gave the highest prevalence of carcinomas and macrotumors. This suggests that different mechanisms affect early and late cancer promotion. Hepatic expression levels of different classes of genes were correlated to occurrence of preneoplastic foci and of later stages of tumorigenesis. For correlations between the different classes of lesions, macrotumors furthermore were less correlated to preneoplastic foci than to adenomas and carcinomas (Supplementary Table 3, available at Carcinogenesis Online). We interpret the dietary effects on tumorigenesis to be due to the differences in the contents of saturated, monounsaturated, ω-6 and ω-3 polyunsaturated fats in the diets. As we only used a single fat source for each type of diet, future studies with other fat sources are required to verify whether it is the degree of fatty acid saturation that caused the tumorigenic effects. High saturated fat diets with palm oil, lard or beef tallow could thus determine if tumor promotion and associated gene expression and microbiome changes are directly associated with saturated fatty acids or with other aspects of the cocoa butter diet.

In humans, liver cancer is more prevalent in men than women (4). Enhanced liver tumorigenesis in males in mouse DEN models has likewise been described (24–27). We observed a marked sex disparity in carcinogenesis with females largely protected from development of liver cancers. Inhibition by the female sex thus supersedes fat-type-mediated tumor promotion. Sex hormones contributes to the sex dimorphism, as estrogen signaling via the estrogen receptor-α is protective whereas androgen signaling via the androgen receptor promotes hepatocarcinogenesis (25–27).

The large number of genes with sex disparity in hepatic expression is also a well-known phenomenon. A total of 6612 genes were previously found to be more than 1.5-fold differentially expressed between males and females (28). Among the 100 most significantly sex-regulated genes, we observed an enrichment of genes involved in lipid and fatty acid metabolism. These genes may very well contribute to the difference in hepatic steatosis observed between the two sexes. The hepatic steatosis score correlated significantly with the multiplicity of preneoplastic foci (Supplementary Table 3, available at Carcinogenesis Online). However, as there was no correlation when considering only the males, the correlation between the two parameters can be explained by strong sex disparity of both steatosis and multiplicity of preneoplastic foci. Although the higher degree of steatosis in males than females may contribute to enhanced carcinogenesis, it does not explain the difference in carcinogenesis among the male dietary groups.

Hepatic cancer is often shown as developing through steps of increasing liver damage: steatosis → steatosis + inflammation (steatohepatitis) → cirrhosis → liver cancer (7). Involvement of inflammation in hepatocarcinogenesis is suggested by a mouse study in which interleukin-6 knockout males had reduced rate of hepatocarcinoma development compared with wild-type males (25). Surprisingly, we observed the highest infiltration with inflammatory cells in the males fed the low-fat diet, and there was no positive correlation between inflammatory scores and any of the cancer parameters (Supplementary Table 3, available at Carcinogenesis Online). The expression of interleukin-6 was too low to give any reads in the RNA-Seq data set. By more sensitive qRT-PCR, we actually observed a higher level of interleukin-6 mRNA in males fed the low-fat diet than males fed the high-fat diets and higher levels in females than males fed high-fat diets (data not shown). It was further remarkable that we observed no liver fibrosis, suggesting that stellate cell activation is not required for tumor promotion under conditions of NAFLD.

Expression of many genes involved in lipid metabolism correlated with the multiplicity of preneoplastic foci. These correlations were mostly due to the sex dimorphism for both preneoplastic foci and the expression of these genes. Even though correlations were not sufficient to fulfill the FDR < 0.05 criterion, among the 100 genes most correlated to preneoplastic foci for males there was an enrichment for genes involved in cell migration, signaling and protein phosphorylation, which are functions often associated with tumorigenesis.

Not surprisingly, genes involved in lipid metabolism, including cytochrome P450 genes, were among those whose expression were affected by different dietary fats. Among the genes affected by the high saturated fat diet relative to other high-fat diets in males, the only proto-oncogene was Agap2. Its biological function is distinct from other cocoa-butter-regulated genes. This is illustrated in Supplementary Figure 4, available at Carcinogenesis Online for which we built a gene network with genes whose expression was altered by the cocoa butter diet relative to the corn oil diet in males. Many of these genes, including genes affecting lipid and fatty acid metabolism are linked to PPARα and GR nodes, but only indirectly linked to Agap2. The Agap2 protein, also known as phosphoinositide-3-kinase enhancer (PIKE), is a GTP-binding protein that can interact with a wide variety of other proteins. For instance, PIKE can bind to and activate phosphoinositide 3-kinase (PI3K). Other binding partners include focal adhesion kinase, v-akt murine thymoma viral oncogene homolog 1 (AKT) and nuclear factor-kappaB. Through binding to its partners, PIKE affects cellular functions, including cell growth in a posttranscriptional fashion (19). It is overexpressed in several human cancers, including samples of liver cancer (29), and it has been shown to promote tumor progression in a xenograft model with human prostate cancer cells (30). The Agap2 gene also encodes a long non-coding RNA termed AGAP2-AS1 that by itself has tumor promoting effects in human non-small-cell lung cancer and gastric cancer (31,32). Activation of the PI3K-AKT pathway by constitutively active AKT is sufficient to induce liver cancers in mice. The PI3K–AKT pathway further enhances carcinogenesis in the presence of other oncogenes (33). Stimulation of the PI3K–AKT signaling by Agap2 is therefore a conceivable tumor promotion mechanism operating in mice fed the cocoa butter diet. Future studies are planned to address this further.

To assess whether Agap2 expression is induced in other tumor promotion models, we have determined the Agap 2 mRNA concentration in archival samples of RNA from studies of hepatic tumor promotion in male mice fed liquid high-fat Lieber DeCarli diets with or without ethanol to produce non-alcoholic and alcoholic steatohepatitis (18,34). In these studies, we showed that replacement of the casein in Lieber DeCarli diets with a soy protein isolate dramatically reduced tumorigenesis. As shown in Supplementary Figure 5, available at Carcinogenesis Online, the Agap2 mRNA concentration was significantly increased in livers from DEN-treated mice fed the liquid high-fat diets compared with DEN-untreated control mice on regular solid chow (P = 0.0004). There was an overall higher level of Agap2 in mice receiving casein than mice receiving soy protein isolate (P = 0.02). In addition, elevated levels of Agap2 in mice receiving ethanol were not statistically significant (P = 0.17). Although the Agap2 mRNA concentration is not by itself sufficient to explain the degree of tumorigenesis, we can conclude that an elevated level is a trait associated with several dietary conditions resulting in fatty liver disease and supporting hepatic tumor promotion.

The gut microbiota can affect liver physiology by mechanisms such as chemical modification of bile acids, production of short chain fatty acids (acetic, butyric and propionic acid) and ceramides, and by disruption of the intestinal epithelial barrier and subsequent lipopolysaccharide-mediated inflammatory responses (22,35). Obesity-associated liver cancer can involve a complex interplay between obesity, bacterial metabolites like lipoteichoic acid, modified bile acids such as deoxycholic acid and stellate cell activation (36,37).

As we observed some cage-specific effects on the cecal microbiota, a limitation of our microbiota analysis is that statistical inferences were due to the combination of cage-specific and group-specific effects. Yet, we observed major differences due to sex and diet that can not simply be explained by cage-specific effects. The most striking sex disparity in the microbiota was a significantly higher Akkermansia content in females than in males. High levels of Akkermansia muciniphila (the single known species of Akkermansia) is linked to protection from adverse effects of a Western diet and obesity (38–40). Oral administration of Akkermansia protected mice from acute liver injury induced by Concanavalin A (41). One mechanism for the beneficial effects of Akkermansia is strengthening of the intestinal barrier (39,41). Males fed the saturated cocoa butter diet had an increased relative abundance of taxa Lachnospiraceae and Coprococcus. These taxa have been implicated in intestinal dysbiosis associated with liver morbidities, showing either increases or decreases dependent on the model or disease. Although Lachnospiraceae abundance was elevated in patients with NAFLD (42), it was decreased in human patients with liver cirrhosis (43). A high level of Coprococcus was associated with inflammation in fructose-fed rats (44). On the other hand, lower abundance of Coprococcus were found in humans with both simple steatosis and non-alcoholic steatohepatitis compared with healthy controls (45), and Coprococcus protected the liver from inflammation in farnesoid X receptor knockout mice (46). The abundance of these taxa is clearly highly dependent on experimental conditions.

In summary, the enhanced tumor promotion in males fed the saturated fat diet was associated with increased hepatic Agap2 expression and intestinal enrichment with Coprococcus. Further studies are required to determine whether Agap2 and Coprococcus play mechanistic roles in hepatic carcinogenesis.

The major conclusion from this study is that a high-fat diet does not have tumor-promoting effects simply due to the quantity of lipid. The fat source is also a crucial factor. We previously reported that the protein source likewise can affect tumor promotion (18,34). It demonstrates that liver cancer is a disease whose progression can be markedly influenced by common nutritional factors.

Funding

Start-up fund from Louisiana State University Health Sciences Center-New Orleans (to M.R.); National Institutes of Health (R21 CA169389, R37 AA09300 and R37 AA18282 to M.R., R25AA021304 to A.P.); the United States Department of Agriculture (Agricultural Research Service Project 6026-51000-010-05S to K.S. and U.D.W.).

Conflict of Interest Statement: None declared.

Abbreviations

DEN

diethyl nitrosamine

DHA

docosahexaenoic acid

FDR

false-discovery rate

GO

Gene ontology

HCC

Hepatocellular carcinoma

NAFLD

non-alcoholic fatty liver disease

PIKE

phosphoinositide-3-kinase enhancer

PI3K

phosphoinositide 3-kinase

PUFAs

polyunsaturated fatty acids

qRT-PCR

quantitative RT-PCR

RNA-Seq

RNA sequencing