Hepatic gene expression changes in mouse models with liver-specific deletion or global suppression of the NADPH-cytochrome P450 reductase gene. Mechanistic implications for the regulation of microsomal cytochrome P450 and the fatty liver phenotype.

NADPH-cytochrome P450 reductase (CPR) is an essential component for the function of many enzymes, including microsomal cytochrome P450 (P450) monooxygenases and heme oxygenases. In liver-Cpr-null (with liver-specific Cpr deletion) and Cpr-low (with reduced CPR expression in all organs examined) mouse models, a reduced serum cholesterol level and an induction of hepatic P450s were observed, whereas hepatomegaly and fatty liver were only observed in the liver-Cpr-null model. Our goal was to identify hepatic gene expression changes related to these phenotypes. Cpr-lox mice (with a floxed Cpr gene and normal CPR expression) were used as the control. Through microarray analysis, we identified many genes that were differentially expressed among the three groups of mice. We also recognized the 12 gene ontology terms that contained the most significantly changed gene expression in at least one of the two mouse models. We further uncovered potential mechanisms, such as an increased activation of constitutive androstane receptor and a decreased activation of peroxisomal proliferator-activated receptor-alpha by precursors of cholesterol biosynthesis, that underlie common changes (e.g. induction of multiple P450s and suppression of genes for fatty acid metabolism) in response to CPR loss in the two mouse models. Additionally, we observed model-specific gene expression changes, such as the induction of a fatty-acid translocase (Cd36 antigen) and the suppression of carnitine O-palmitoyltransferase 1 (Cpt1a) and acyl-CoA synthetase long chain family member 1 (Acsl1), that are potentially responsible for the severe hepatic lipidosis and an altered fatty acid profile observed in liver-Cpr-null mice.

mice, are less clear. A better understanding of the mechanisms underlying these phenotypes will shed light on possible metabolic and associated functional consequences in patients carrying defective CPR alleles.
In the present study, genomic analyses of gene expression changes in the livers of the two mouse models were performed. Our goal was to identify mechanisms potentially responsible for the observed hepatic phenotypes in liver-Cprnull and Cpr-low mice. Hepatic gene expression was analyzed with the Affymetrix Mouse Expression Set 430A GeneChip arrays. The Cpr-lox mouse, which was derived from the same embryonic stem cell that was used to generate the Cpr-low and the liver-Cpr-null mice (10), was treated as the wild-type control. Genes with expression levels that differed between liver-Cpr-null and Cpr-lox or that differed between Cpr-low and Cpr-lox groups were identified using the criteria of Ն2.0 or Յ0.5 in fold change (change-fold), and p Ͻ 0.01. Gene ontology (GO) terms that contained the most significantly changed gene expression were identified through pathway analysis using Gene Map Annotator and Pathway Profiler (www.genmapp.org). A detailed analysis of the gene expression changes in the lipid metabolism and transport pathways led us to propose mechanistic schemes that explain the altered gene expression of multiple P450 enzymes in both mouse models, as well as the severe hepatic lipidosis seen in liver-Cpr-null mice. Additional studies of hepatic fatty acid profiles provided evidence that substantiates the genomic changes in the liver-Cpr-null mice.

RNA Preparation and Microarray
Hybridization-Protocols for animal breeding and genotyping were reported previously (10,11,13). Animal use protocols were approved by the Institutional Animal Care and Use Committee of the Wadsworth Center. Animals were maintained at 22°C with a 12-h on, 12-h off light cycle and were allowed free access to water and a standard laboratory diet. Mice were sacrificed at the age of 2 months. Liver was dissected and was stored at Ϫ80°C until use. Total RNA was prepared from individual liver samples using TRIzol (Invitrogen) and was further purified using Qiagen RNeasy mini columns. The integrity of the RNA preparations was determined by spectrometry and by electrophoretic analysis on agarose gels.
Affymetrix Mouse Expression Set 430A GeneChip arrays were used for microarray analysis. Each array contains 22,690 probe sets, representing ϳ14,870 distinct genes. Each probe set consists of 11 pairs of 25-mer oligonucleotides. A probe pair is composed of a perfectly matched sequence, which is complementary to the target sequence, and a mismatched sequence, which contains a single nucleotide mismatch at the central base pair position. The mismatched sequence is used as internal control for nonspecific hybridization.
Three independent RNA samples were analyzed for each mouse strain (liver-Cpr-null, Cpr-low, and Cpr-lox). Each sample was prepared by pooling equal amounts of total RNA from 2 to 3 mice of the same strain. Five micrograms of total RNA were used for synthesis of biotinylated antisense RNA (aRNA) with MESSAGEAMP TM aRNA kit from Ambion. The labeled aRNA was then fragmented and stored at Ϫ20°C until hybridization. GeneChip array hybridization, staining, and washing were performed according to the Affymetrix GeneChip® Expression Analysis Technical Manual at the Microarray Core Facility of the Wadsworth Center. Briefly, 10 g of the fragmented aRNA was hybridized to each array at 45°C for 16 h in a GeneChip® hybridization oven with constant rotation (60 rpm). The arrays were then stained and washed, using the antibody amplification washing and staining protocol (Affymetrix), in a GeneChip Fluidics Station 400. The expression data were collected with a gene array scanner using Affymetrix Microarray Suite version 5.0 (MAS 5.0).
Data Analysis-The experimental data sets were normalized using the Robust Multichip Analysis program of the Genetraffic UNO 3.2 software package (Iobion). The hybridizations for control mice were used as the base line. Analysis for significance was performed using the unpaired t test in Genetraffic UNO 3.2 (Iobion). The ratios of averaged values were used to calculate change-fold between two groups. Genes with significantly changed expression were tabulated, along with gene symbol, gene name, transcript identification number, and change-fold values, and were further examined for reproducibility among multiple probe sets for a given gene, where available.
Pathway Analysis Using Gene Map Annotator and Pathway Profiler-Two programs, MAPPFinder (14) and GenMAPP (version 2.0) (15), were used to group genes having significantly changed expression according to the GO hierarchy at the level of biological processes (GO process), cellular components (GO component), and molecular functions (GO function). The relative extent of gene expression changes in each GO node was compared using the "z score" (15), a standardized difference score, and the number (and percentage) of the genes measured in a GO term that meet user-defined criteria for "significant" changes (e.g.Ϯ25% in change-fold and p Ͻ 0.01). Redundant GO terms were removed according to the following rules (assuming that term 1 contains more genes than does term 2): if term 1 contains more genes with expression changes than does term 2, then term 2 is removed; however, if the two terms contain the same set of genes with expression changes, then term 1 is removed.
Real Time RNA-PCR and Immunoblot Analysis-Real time PCRs were performed according to a protocol described previously (11), with gene-specific PCR primers for CYP2A4/5, CYP2B10, CYP4A10, CD36 antigen (CD36), and ␤-actin. The PCR primers used for CYP2A4/5, CYP2B10, and ␤-actin were the same as reported (11). The other primers used were as follows: for CD36, 5Ј-agtatgtcgtcatgttcc-3Ј and 5Јcactataacagctctccaag-3Ј; for CYP4A10, 5Ј-agtgtctctgctctaagcc-3Ј and 5Јcccaaagaaccagtgaaaag-3Ј. These primers were designed using Seqweb version 2.1 (Accelrys Inc.). Identities of PCR products were confirmed by electrophoretic analysis on agarose gels (for CYP4A10 and CD36) and sequencing (for CYP4A10). Immunoblot analysis was carried out as described (11), with use of rabbit antibodies to CYP2A5 (16) or rat CPR (BD Biosciences), and goat antibodies to rat CYP1A1/2, CYP2B1, or CYP3A2 (DaiiChi), and monoclonal anti-mouse CD36 (BD Biosciences). Microsomes were prepared according to Coon et al. (17) but without the pyrophosphate washing step. The plasma membrane fraction was prepared according to a protocol described by Zhang and Menon (18). Protein concentration was determined by the bicinchoninic acid method (Pierce) with bovine serum albumin as the standard.
Lipid Isolation and Analysis-Livers were excised from 8-week-old mice that had been fed a standard chow diet ad libitum (Prolab® RMH3500). Excised livers were cut into three pieces to provided triplicate samples for analysis and then were frozen at Ϫ80°C until use. For lipid extraction, the Folch method was used (19). Briefly, tissue sections were homogenized in a Polytron homogenizer (model PT3100) for 30 s at 15,000 -20,000 rpm in chloroform/methanol (2:1) at 20 ml/g liver. The homogenate was shaken for an additional 2 h to fully extract the remaining lipids. After centrifugation and filtration, the extracts were dried down under a stream of nitrogen, and the lipids were resuspended in chloroform. The final lipid samples were split into three equal portions for analysis of complex lipids by high performance liquid chromatography (HPLC), fatty acids by gas chromatography-mass spectroscopy, and total phosphorous determination (20).

RESULTS
By using relatively conservative criteria in change-fold (Ն2.0 or Յ0.5), and a p value of Ͻ0.01, we identified a number of mouse hepatic genes that were differentially expressed between the liver-Cpr-null and control (45 up and 18 down) groups or between the Cpr-low and control (22 up and 26 down) groups. Comparisons of expression levels of these genes among the liver-Cpr-null, Cpr-low, and Cpr-lox (control) groups are shown in Table I, in which the identified genes are grouped in seven functional categories based on GenMAPP and additional literature searches. A number of unknown genes were also identified that have differential expression among the comparison groups.
Total P450 content in liver microsomes was elevated in both liver-Cpr-null (11,12) and Cpr-low mice (13). In the liver-Cpr-TABLE I Genes that were differentially expressed in livers of the liver-Cpr-null, Cpr-low, and Cpr-lox mice Genes with significantly changed expression (p Ͻ 0.01) and change-fold Ն2.0 or Յ0.5 in the liver-Cpr-null/Cpr-lox or Cpr-low/Cpr-lox comparisons, in at least one probe set and one set of comparisons, are shown. An arbitrary cut-off value (Ͻ60) was used to filter genes with very low expression value; probe sets for which none of the three groups has averaged expression values Ն60 were excluded. For genes represented by multiple probe sets, the results for all probe sets are included even though all probe sets may not meet the selection criteria, and the results are arranged in ascending order according to the list of Affymetrix probe set ID numbers. A change-fold value is shown when p Ͻ 0.05; otherwise, "no significant change" is assigned. Results from probe sets having Ͻ90% homology to target sequence are excluded. For each entry, the reference sequence transcript identification number (RefSeq transcript ID) is given along with the gene symbol and gene name (according to Affymetrix    null mouse model, induction of several CYP forms was reported previously, including CYP1A2, CYP2A5, CYP2B10, CYP3A11, and CYP3A13 mRNAs (11) and CYP1A, -2A, -2B, -2C, -2E1, -3A, and -4A proteins (11,12). The individual CYPs induced in the Cpr-low model have not been reported previously. In the present study, we found that not all CYPs are induced by the decrease in CPR expression in either of the two mouse models. Thus, although the levels of CYP2A4/5, CYP2B10, CYP2C55, CYP51, CYP7A1, and CYP26A1 mRNAs were significantly increased, the levels of CYP4A10 and CYP7B1 mRNAs were significantly decreased in both mutant models (Table I). Numerous other CYPs were also found to be up-or down-regulated in the liver-Cpr-null or Cpr-low mice, albeit with smaller change-folds. These are included in the data set submitted to the NCBI (accession number GSE2362). It should be noted that the change-fold values were most likely underestimated in the microarray analysis for most genes, as indicated by the results for Cyp2a4/5 and Cyp2b10, for which the average change-fold between liver-Cpr-null and Cpr-lox mice was 4.4 and Ͻ15, respectively, by microarray analysis, whereas it was 30 and 68, respectively, by RNA-PCR (11). A decrease in CYP4A10 expression was further demonstrated by RNA-PCR (78 Ϯ 8% and 60 Ϯ 8% decreases in the Cpr-low and liver-Cpr-null mice, respectively, compared with Cpr-lox mice; n ϭ 3, p Ͻ 0.01). The induction of CYP2A5 and CYP2B10 mRNAs (data not shown), and of CYP1A, CYP2A, CYP2B, and CYP3A proteins ( Fig. 1), was also confirmed. For the CYPs shown in Fig. 1, a lesser extent of induction was seen in the Cpr-low than in the liver-Cpr-null mice, correlating with the differing extents of CPR loss with these strains. Additionally, as shown in Table I, the level of CYP2C39 mRNA was reduced only in the Cpr-low mice, indicating a model-specific regulatory event.
Significant changes in expression were also found for a variety of other genes (Table I), including those for (non-CYP) biotransformation, antioxidant and stress response, steroid metabolism and transport, triglyceride and fatty acid metabolism and transport, carbohydrate and amino acid metabolism, and growth and signal transduction. Among the antioxidant and stress-response genes, an elevated expression of heme oxygenase 1 had been demonstrated earlier by immunoblot and functional analyses (11). Alterations in the steroid and fatty acid metabolism and transport pathways were anticipated, because of the known changes in cholesterol homeostasis in both mutant strains, and the known fatty liver phenotype seen in the liver-Cpr-null mice (11)(12)(13). Although numerous other genes in these pathways were changed, these are not listed in Table I because they did not meet the criterion of a Ն2-fold change.
To gain a view of the metabolic changes in the livers of these two mouse models that is more comprehensive than the data presented in Table I, we used a less stringent criterion for significant changes (i.e. Ͼ25% in change-fold and p Ͻ 0.01) than that used in Table I. The selection of this smaller change-fold was according to published methods (14), and it was also based on the consideration that accumulation of small changes in a biological pathway might cause observable biological effects. By using this criterion, we identified 283 significantly induced and 171 significantly suppressed genes in the liver-Cpr-null mice and 110 significantly induced and 234 significantly suppressed genes in the Cpr-low mice. We analyzed the expression data for these genes using MAPPFinder, to identify pathways that contain the most significantly changed gene expression.
Metabolic pathways (GO terms) that contain the most significantly altered gene expression were identified using the criteria of z score Ն4.0, change-percent Ն10%, and changenumber Ն3 for GO process, and z score Ն4.0, change-percent Ն10%, and change-number Ն5 for GO molecular function or cellular component. A less stringent criterion was applied to the GO terms of biological processes because there are ratelimiting enzymes in many metabolic pathways. Nonredundant GO terms that meet these criteria in at least one of the two comparisons (liver-Cpr-null/Cpr-lox and Cpr-low/Cpr-lox) are shown in Table II. Genes meeting the criteria for changed expression in each of these pathways, filtered further by the removal of genes with very low expression value, are shown in the Supplemental Material.
As shown in Table II, three pathways under the heading of GO biological process, vitamin metabolism, coenzyme metabolism (including NADPH regeneration), and sulfur metabolism, were found to contain the most significantly increased gene expression in the liver-Cpr-null mice. Less extensive, but significant, increases (p Ͻ 0.05 for the z score) in vitamin metabolism and coenzyme metabolism were also found in the Cpr-low mice. In addition, significant increases were found in the liver-Cpr-null mice in the pathway of steroid metabolism. In this pathway, most of the changed genes are for cholesterol metabolism (not shown). Of interest, gene expression changes in steroid metabolism appeared to be more extensive in the Cprlow mice than in the liver-Cpr-null mice.
Significant increases in gene expression were found in the liver-Cpr-null mice in two cellular components: microsome and lysosome. Small, but significant, increases were also observed in the Cpr-low mice in microsome but not in lysosome. Under molecular function, glutathione transferase activity is topranked, containing the most significantly increased gene expression in both models. Significant changes in gene expression were also detected in monooxygenase activity and oxidoreductase activities in both mouse models, with greater increases in the liver-Cpr-null mice. A much smaller number of GO terms Cpr-low ("Low"; lanes 4 -6), or liver-Cpr-null mice ("Null"; lanes 7-9) was analyzed on immunoblots. A single band was detected with the anti-CYP1A or the anti-CYP2A antibody, whereas multiple CYP-related bands were detected with the anti-CYP2B and anti-CYP3A antibodies. B, quantitative immunoblot analysis was performed to compare the relative abundance of CYP1A, -2A, -2B, and -3A proteins in the three mouse groups. For CYP3A, the upper band and the two lower bands were quantitated separately. For CYP2B, the two CYP2B bands were combined for quantification. When the two CYP2B bands were quantitated separately, the intensity of the upper band was about 40 and 200 times greater in the Cpr-low and liver-Cpr-null mice, respectively, than in the Cpr-lox mice, whereas the intensity of the lower band was about twice as great in the Cpr-low and liver-Cpr-null mice as in the Cpr-lox mice. The data reported (fold-change over the Cpr-lox group) are means Ϯ S.D. (n ϭ 3). had significant decreases in gene expression. GO biological processes of fatty acid metabolism and gluconeogenesis were significantly reduced in both models, with greater decreases in the liver-Cpr-null mice. Notably, much overlap exists among multiple pathways, such as coenzyme metabolism and vitamin metabolism, coenzyme metabolism and sulfur metabolism, or microsome, monooxygenase, and oxidoreductase activities (see Supplemental Material).
CPR-dependent enzymes are directly involved in cholesterol and bile acid biosynthesis and metabolism, as well as fatty acid oxidation, and many CYPs are regulated by endogenous ligands produced in these lipid metabolic pathways. Therefore, a detailed analysis of the changes in lipid metabolism and transport pathways was performed, to provide clues for what might have triggered the hepatic CYP gene expression changes and why the liver-Cpr-null mouse alone developed fatty liver. Fig. 2 summarizes all gene expression changes detected in the pathways of cholesterol biosynthesis, bile acid and sterol synthesis, fatty acid synthesis and oxidation, and lipid transport. These pathways are derived from information in GenMAPP (www. genmapp.org) and KEGG (www.genome.jp/kegg), with additional data from a recent work of Horton et al. (23) on cholesterol and fatty acid synthesis pathways.
As shown in Fig. 2, 10 cholesterol biosynthetic enzymes were induced in the liver-Cpr-null and Cpr-low mice. However, because of the loss or suppression of the activities of several CPRdependent enzymes, e.g. squalene epoxidase, CYP51, and 7-dehydrocholesterol reductase, the induction of the other enzymes in this pathway was insufficient to support a normal level of cholesterol production. The mRNA levels for the rate-limiting en-zyme in the classical bile acid synthesis pathway, CYP7A1, and those for CYP8B1, also of the classical bile acid synthesis pathway, were significantly induced in the liver-Cpr-null mice. On the other hand, the mRNA levels of CYP7B1, of the alternative bile acid synthesis pathway, and those of CYP17A1, of the steroid hormone biosynthesis pathway, were decreased in both mouse models. Additional signs of alterations in hepatic steroid hormone metabolism include a suppressed expression of Cyp39a1 (of the alternative bile acid synthesis pathway) and Hsd3b4/5 in the liver-Cpr-null mice, an induction of Hsd17b12 in the liver-Cprnull mice, and an induction of Hsd3b2 in the Cpr-low mice.
The expression of several fatty acid synthesis enzymes was similarly altered in the two models (Fig. 2), implicating changed composition of hepatic fatty acids in these mice. These gene expression changes included increases for stearoyl-coenzyme A desaturase 1 (Scd1) and a gene related to elongation of very long chain fatty acids (Elovl6) and decreases for fatty acid desaturase 2 and another two ELOVL family members (Elovl2 and Elovl5). Additional changes in Elovl1 (increase) and Elovl3 (decrease) were seen only in the liver-Cpr-null mice. As a consequence of the altered expression of genes encoding fatty acid metabolic enzymes, levels of total fatty acids and specific fatty acid species were altered in the liver-Cpr-null mice, compared with Cpr-lox mice (Fig. 3, Table III, and Table IV). As noted previously, liver-specific disruption of Cpr results in steatosis. Triglyceride levels in the Cpr-null livers were 5-fold higher than in Cpr-lox livers (Fig. 3A). Because expression of the major ⌬9-fatty acyl-CoA desaturase (Scd1) gene was elevated, it was not surprising that there was also a very large increase in the long chain monounsaturated fatty acid species.

TABLE II
Biological processes, molecular functions, and cellular components that showed the most significant gene expression alterations in the livers of the liver-Cpr-null and Cpr-low mice Pathway analysis was performed using MAPPFinder 2.0 and the Mm-Std_20040824.gdb data base (www.genmapp.com). Of the 22,690 probes in this data set, 21,865 are linked to a Mouse Genome Informatics ID. The criteria for genes with significantly increased or decreased expression were change-fold Ͼ25% (Ͼ1.25 or Ͻ0.80) and p Ͻ 0.01. GO terms are sorted into three types as follows: biological process, molecular function, and cellular component. For each GO term, the number of genes that meet the criteria for a significant increase or decrease was determined (No. changed column). This number was compared with the number of genes in the GO term that are measured by the MOE 430A chip (No. of genes measured column) for the calculation of the percentage of genes measured in the GO term that meet the criteria for a significant increase or decrease (% changed, in parentheses). z score, a standardized difference score for comparison of the relative extents of gene expression changes in various GO nodes, was also shown. The permuted p values for z score were Ͻ0.01 unless indicated otherwise. The pathways shown were filtered using the criteria of percent changed Ն10%, z score Ն4.0, and number changed Ն3 (for biological processes) or 5 (for molecular functions and cellular components) in at least one of the two comparisons (liver-Cpr-null/Cpr-lox and Cpr-low/Cpr-lox). The results for both comparisons are shown even if only one meets the selection criteria (shown in boldface). Redundant pathways were excluded. The specific genes meeting the criteria for changed expression in each GO term are shown in the Supplemental Material. GO  In particular, we observed 10-and 7-fold increases in C16:1 and C18:1, respectively (Table III). The essential fatty acids linoleic acid (C18:2) and linolenic acid (C18:3) cannot be synthesized in mammals and must be obtained from the diet. Because these animals were maintained on standard laboratory chow, they received ample C18:2 and C18:3 from this source. As shown in Table III, the liver-Cpr-null mice accumulated 3.4-and 7.3-fold higher levels of C18:2 and C18:3, respectively, than the Cpr-lox animals. This suggests that fatty acid uptake was not diminished and might have been increased as a result of disruption of Cpr, perhaps through CD36, a long chain fatty acid translocase (24,25), which was also elevated (Table I and Fig. 4).
Alterations in the amounts of saturated and unsaturated very long chain fatty acids were also observed. The levels of C20:1, C20:2, and C20:3 were 14.9-, 6.4-, and 4.7-fold higher in the liver-Cpr-null mice than in the Cpr-lox mice. However, arachidonic acid (C20:4) levels were similar in the two mouse models (Table III), and relative to the levels of total C12-C24 fatty acids, the relative amounts of C20:4 were actually lower in the liver-Cpr-null mice than in the Cpr-lox mice (Table IV). This was likely because of the reduced expression of Fads2 (Fig. 2), which encodes the ⌬6-fatty acyl-CoA desaturase, the rate-limiting enzyme for the synthesis of the essential polyunsaturated fatty acids C20:5 and C22:6, in addition to C20:4 (26). Indeed, a significant reduction in relative amounts was also found for C22:6 in the liver-Cpr-null mice, although a significant change was not found for C20:5 (Table IV). The altered pattern of polyunsaturated fatty acids in the liver-Cprnull mice has features in common with animals treated with a

FIG. 2. Scheme of altered lipid metabolism and CYP expression in the livers of liver-Cpr-null and Cpr-low mice.
Steps in the metabolic pathways are connected by black arrows. Genes with significantly changed gene expression (p Ͻ 0.05, and change-fold Ͼ25% in at least one of the comparisons) are included, but those with very low expression value (Ͻ60) were excluded. The probe set identification number for each gene (without the "_at" extension) is shown in brackets, and change-folds (liver-Cpr-null/Cpr-lox and Cpr-low/Cpr-lox, respectively) are shown in parentheses, with red indicating an increase, green indicating a decrease, and NC indicating no significant change. For genes represented by multiple probe sets, the change-fold represents the value of the first probe set in the list sorted according to probe set ID numbers, if the results from all probe sets are consistent; otherwise, the majority results are shown, and the change-fold represents the value of the first majority probe set in the sorted list. For genes represented by two or four probe sets that yielded inconsistent results (50% significant, 50% not), the significant results are shown, but the values are denoted with *. Boxed items are key metabolites involved in the pathways. CPR-dependent enzymes are underlined. Modulation of nuclear receptor function by endogenous metabolites and regulation of target gene expression by nuclear receptors are indicated by red (activation) or green (suppression) arrows. Abbreviations and gene symbols not already described in Table I include Abcc3, ATP-binding cassette subfamily C (CFTR/MRP) member 3; Acadl, acyl-CoA dehydrogenase, long chain-specific; Acadvl, acyl-CoA dehydrogenase, very long chain; Acsl1, acyl-CoA synthetase long chain family member 1; Ahr, aryl-hydrocarbon receptor; Cpt, carnitine O-palmitoyltransferase; Dci, dodecenoyl-coenzyme A ␦ isomerase; Fads, fatty-acid desaturase; Fdft1, farnesyl-diphosphate farnesyltransferase 1; Hadhb, hydroxyacylcoenzyme A dehydrogenase; Hmgcr, 3-hydroxy-3-methylglutaryl-coenzyme A reductase; Hsd17b12, hydroxysteroid (17-␤) dehydrogenase 12; Nr0b2, nuclear receptor subfamily 0, group b, member 2; Nsdhl, NAD(P)-dependent steroid dehydrogenase-like. specific ⌬6-desaturase inhibitor, SC-26196 (27,28).
The observed reduction in C22:6, docosahexaenoic acid, which is essential for brain and retinal structure and function, was also likely because of reduced Elovl2 and the peroxisomal fatty acid oxidation enzyme encoded within Ehhadh, because the synthesis of C22:6 from C18:3 requires one round of peroxisomal ␤-oxidation in addition to elongation and desaturation (29). The reduced levels of several elongases also resulted in significant changes in the percentage of fatty acids greater than C20 in length (Table IV).
We noted that the severity of the fatty liver phenotype varied among the liver-Cpr-null mice. Thus, for the experiments described in Tables III and IV, one liver-Cpr-null mouse was found to have levels of hepatic triglyceride and fatty acids that were similar to or only slightly higher than those found in wild-type mice. The data obtained from that mouse was not included in Tables III and IV, because the animal did not present the typical fatty liver phenotype for the liver-Cpr-null mice. The causes of the variable penetrance of the fatty liver phenotype remain to be determined.
Six genes of the fatty acid oxidation pathway (Cpt2, Acadvl, Acadl, Hadhb, Ehhadh, and Dci) were similarly suppressed in liver-Cpr-null and in Cpr-low mice (Fig. 2), indicating reduced fatty acid utilization in the two models. Suppressed expression was also found, in both models, for Cyp4a10, which is involved in the -oxidation of fatty acid (30), and fatty acid-binding protein 2 (Fabp2), an intracellular fatty acid carrier (31). However, expression of another two genes that are involved in fatty acid utilization (Acsl1 and Cpt1a) was suppressed only in the liver-Cpr-null mice. In contrast, lipoprotein lipase (Lpl), a tri-glyceride hydrolase (32), was induced in both models (Table I  and Fig. 2), whereas Cd36 was induced only in the liver-Cprnull mice (Fig. 2). The differential induction of CD36 in the two mouse models was confirmed by RNA-PCR and immunoblotting (Fig. 4). DISCUSSION Gene expression changes in specific metabolic pathways will likely reflect changes in the flux of pertinent metabolites, because the enzymes catalyzing the metabolism of endogenous substances are often regulated by feedback or feed-forward mechanisms. CPR can transfer electrons to a number of redox partners, although not all partners necessarily depend on CPR for function in vivo. If CPR-dependent enzymes play an essential role in a metabolic pathway, then the loss of CPR (or a significant decrease) would cause an accumulation of substrates upstream of the CPR-dependent enzymes and a depletion of products downstream of the CPR-dependent enzymes. In this study, we uncovered, using the criteria of Ն2.0 or Յ0.5 in change-fold and p Ͻ 0.01, numerous genes that are differentially expressed between liver-Cpr-null and Cpr-lox or between Cpr-low and Cpr-lox mice. We also identified, through pathway analysis, many GO terms that contained the most significantly changed gene expression in at least one of the two mouse models. Although it is impossible to consider all gene expression changes here, we have performed a focused analysis of the gene expression changes found in lipid metabolism and transport pathways. Below we will provide a detailed discussion of the potential mechanisms that explain the altered gene expression of multiple P450 enzymes in both mouse models and the severe hepatic lipidosis seen in liver-Cpr-null but not Cprlow mice.
Responses to Decreased Cholesterol and Bile Acid Synthesis-Steroid metabolism, particularly cholesterol metabolism, was among the most significantly changed pathways in both mouse models; this is consistent with the known function of CPR-dependent enzymes in both cholesterol biosynthesis and degradation and with the phenotypes observed. Serum cholesterol level was only 18% of wild-type value in liver-Cpr-null mice (11) and about 50 -75% of wild-type values in Cpr-low mice (13). As shown in Fig. 2, expression of 10 genes in the cholesterol biosynthesis pathway was significantly increased in both mouse strains, in response to the lowered cholesterol level. However, this feedback regulation should have little effect on the rate of cholesterol synthesis in the liver-Cpr-null mice, because the loss of hepatic CPR expression abolishes activities of several critical, CPR-dependent, enzymes in this pathway, i.e. squalene epoxidase, CYP51, and 7-dehydrocholesterol reductase. In the Cpr-low mice, which still have ϳ25% CPR expression in hepatocytes, the up-regulation of the 10 biosynthetic genes, combined with the residual activities of the three CPR-dependent enzymes, should support a higher level of cholesterol synthesis than that in the liver-Cpr-null mice.
In mammals, cholesterol homeostasis is tightly regulated by feedback mechanisms mediated by sterol-response elementbinding proteins (SREBPs). The SREBPs, which remain in inactive form when cellular oxysterol level is high, stimulate the expression of genes involved in cholesterol biosynthesis when the cellular oxysterol level becomes low (33). Thus, the observed induction of the 10 cholesterol synthesis genes in the liver-Cpr-null and Cpr-low mice is likely mediated through the activities of SREBPs. Indeed, gene expression changes that are similar to the ones observed in the present study were reported in transgenic mice overexpressing activated forms of SREBPs (23), although the magnitude of the changes that we see in the liver-Cpr-null and Cpr-low mice is lower than in that study. The homeostasis of cholesterol is also regulated by the rate of biosynthesis of bile acid; this is the major pathway of cholesterol clearance from the body in mammals (34,35). In the liver-Cpr-null mice, bile acid synthesis in the liver would be abolished, because of the loss of four CPR-dependent enzymes, CYP7A1 and CYP8B1 of the classic pathway and CYP7B1 and CYP39A1 of the alternative pathway (34,35). Indeed, bile volume was reported to be reduced to only 10% of wild-type values in the liver-Cpr-null mice (12). Bile acid synthesis in extrahepatic tissues proceeds via the alternative pathway, because CYP7A1 and CYP8B1 of the classic pathway are both liver-specific (36,37). In the liver-Cpr-null mice, extrahepatic bile acid synthesis should be intact, although the yield may be decreased because of the much-reduced circulating cholesterol levels. A preliminary study on serum bile acid levels revealed a significant increase in taurochenodeoxycholic acid, a product of the alternative pathway, and a decrease in taurocholic acid, of the classic pathway, in the liver-Cpr-null mice 2 ; this confirms an increased contribution of extrahepatic bile acid synthesis to the total bile acid pool.
Although the four CPR-dependent microsomal P450 enzymes in the bile acid synthesis pathways are not functional in the liver of the liver-Cpr-null mice, their gene expression changes reflect the cellular responses to the lowered bile output and to changes in serum bile components. As shown in Fig. 2, expression of Cyp7a1 and Cyp8b1 was induced, whereas that of Cyp7b1 and Cyp39a1 was reduced in the liver-Cpr-null mice. CYP7A1, which is the rate-limiting enzyme in classic pathway, is a major regulation point for cholesterol homeostasis. Expression of Cyp7a1 is positively regulated by cholesterol and oxysterol, through the liver X receptor-␣, and is negatively regulated by bile acid, through the farnesoid X receptor (FXR) and the short heterodimer partner (SHP or NR0B2, nuclear recep-2 H. Cui and X. Ding, unpublished data. a FAME, fatty acid methyl ester. b S.E., standard error of the mean, n ϭ 6 (triplicate tissue samples from two independent animals per group). c Significance defined using a paired t test. d Fold-change, the mean value for liver-Cpr-null mice divided by the mean for Cpr-lox mice. a FAME, fatty acid methyl ester. b S.E., standard error of the mean, n ϭ 6 (triplicate tissue samples from two independent animals per group). c Significance defined using a paired t test. c Fold-change, the mean value for liver-Cpr-null mice divided by the mean for Cpr-lox mice.
tor subfamily 0, group b, member 2) (38 -41). CYP8B1, which controls the level of cholic acid in the bile acid pool, is also negatively regulated by FXR and NR0B2 (40 -42). It appears that the induction of Cyp7a1 and Cyp8b1 in the liver-Cpr-null mice was mainly because of an inhibition of the negative regulation by FXR and NR0B2 of these genes. The reduced activity of FXR was also reflected by the significantly reduced expression of Nr0b2, which is another FXR target gene. CYP7B1 and CYP39A1 of the alternative bile acid synthesis pathway are both oxysterol 7␣-hydroxylases. Their decreased hepatic expression in the liver-Cpr-null mice may reflect cellular responses to an increased level of chenodeoxycholic acid in the bile acid pool. However, little is known about the transcriptional regulation of Cyp7b1 or Cyp39a1. A recent report (43) showed that Cyp7b1 might be negatively regulated by SREBP, an idea consistent with the predicted activation of SREBP in the liver-Cpr-null mice.
The rates of bile production have not been determined in the Cpr-low mice. However, because of the residual CPR expression in the liver (about 25% of wild-type levels), and the relatively small decreases in cholesterol levels, bile acid synthesis is expected to be less strongly affected in the Cpr-low mice than in the liver-Cpr-null mice. Thus, the magnitude of increase in Cyp7a1 expression was lower in Cpr-low than in liver-Cpr-null mice. Furthermore, although a decreased expression was found for Cyp7b1, an alteration consistent with the predicted activation of SREBP, the expression of Cyp8b1 and Cyp39a1, as well as Nr0b2, was not significantly changed in the Cpr-low mice.
In both mouse models, the adaptive response to reduced bile acid production may also include the induction of Aqp8 and Slco1a4 (Table I). AQP8 is important in bile secretion (44), whereas SLCO1A4 (also called SLC21A5 or OATP2) is involved in the uptake of bile salts (45).
Model-specific Development of Fatty Liver in the Liver-Cprnull Mice-Fatty acid metabolism is the top-ranked pathway containing the most significantly decreased gene expressions in both mouse models. The occurrence of hepatic lipidosis and necrosis was observed previously in the liver-Cpr-null mice (11,12). In the Cpr-low mice, hepatic lipidosis was mild and focal and was only seen in some animals, whereas necrosis was never detected (13). Peroxisomal proliferator activated recep-tor-␣ (PPAR␣) is the major regulator of fatty acid uptake and oxidation in the liver (46). As shown in Fig. 2, expression of six genes in the fatty acid ␤-oxidation pathway was decreased in both models; four of these genes (Hadhb, Acad1, Acadv1, and Ehhadh) are identified as target genes of PPAR␣ (47,48). However, the expression of acyl-CoA oxidase, the rate-limiting enzyme in the peroxisomal ␤-oxidation pathway, was not changed (data not shown). Common decreases were also seen in the expression of two other PPAR␣ target genes; these are genes for CYP4A10, a fatty acid -hydroxylase, and Fabp2, which interacts with both fatty acids and nuclear receptors that regulate genes involved in fatty acid metabolism (49,50). In contrast, decreased expression of two other PPAR␣ target genes, Cpt1a and Acsl1 (47,51,52), was seen only in the liver-Cpr-null mice; Cpt1a is critical in mitochondrial fatty acid oxidation, whereas acyl-CoA synthetase is required for activation of fatty acids prior to metabolic utilization.
The mechanisms for the decreased PPAR␣ activity, although not yet clear, are potentially interesting. None of the CPR-dependent enzymes is directly involved in fatty acid synthesis or ␤-oxidation. Although microsomal P450 enzymes are important for fatty acid -hydroxylation, it is not clear whether blockage of -hydroxylation alone can cause sufficient changes in the cellular fatty acid pool to effect a feedback regulation of PPAR␣. In this regard, altered expression of the fatty-acid elongases and desaturases resulted in altered composition of the hepatic fatty acid pool, and this in turn might mediate the suppression of PPAR␣ activity. In particular, eicosanoid metabolites of arachidonic acid are expected to be natural agonists of the PPARs (53). The elevated level of hepatic monounsaturated fatty acids found in the liver-Cpr-null mice is consistent with the observed increase in the expression of Scd1, which is known to be regulated by SREBPs as well as PPAR␣ (23,54,55). On the other hand, the reduced expression of PPAR␣ might be triggered by changes in the cholesterol biosynthesis pathway. Inhibition of hydroxy-3-methylglutaryl-CoA reductase by statins leads to dose-dependent increases in the expression and activity of PPAR␣, through a pathway that is negatively regulated by mevalonate and its nonsteroidal metabolites produced in cholesterol biosynthesis pathway (56 -59). Thus, the expected accumulation of these same metabolites in the livers of our transgenic mouse models may have an outcome that is opposite the effects of the statins, i.e. a reduction of PPAR␣ activity. More importantly, the changes in either fatty acid pool or metabolites of the cholesterol biosynthesis pathway will be greater in the liver-Cpr-null than in the Cpr-low mice, as determined by the degree of loss of CPR-dependent activities in the two models. Accordingly, a significant reduction in the expression of PPAR␣ (0.8-fold; p Ͻ 0.01) was detected in liver-Cpr-null but not in Cpr-low mice. The differing extent of reduction in PPAR␣ activity would explain the modelspecific suppression of selected PPAR␣ target genes, such as Cpt1a and Acsl1.
A model-specific induction of CD36 (a fatty-acid translocase) was found in the liver-Cpr-null mice. CD36 is a membrane protein that plays an important role in fatty acid uptake and storage (24,25,60). Increased expression of Cd36 was associated previously with overexpression of PPAR␥ in mice (61). In our study, the model-specific induction of CD36 was also associated with a significant, albeit small, induction (1.6-fold; p Ͻ 0.01) of PPAR␥ expression in the liver-Cpr-null mice. The model-specific induction of PPAR␥, in turn, might be due to differing extents of changes in the level, as well as the composition, of hepatic fatty acids. The latter might also influence the activity of PPAR␥ through direct ligand-nuclear receptor interactions. Note, how-  1-3), Cpr-lox (lanes 4 -6), or Cpr-low mice (lanes 7-9) were analyzed on immunoblots, with use of an antibody to CD36. B, comparisons of CD36 protein and mRNA levels among liver-Cpr-null, Cpr-lox, and Cpr-low mice. Relative CD36 protein levels were determined on immunoblots. CD36 mRNA levels were quantified using real time RNA-PCR, with total hepatic RNA from individual mice, as described under "Materials and Methods"; the results shown were normalized to the levels of ␤-actin mRNA. Values presented are ratios (fold-change) of Liver-Cpr-null/Cpr-lox (Null/Lox) and Cpr-low/Cpr-lox (Low/Lox) (means Ϯ S.D.; n ϭ 3). ever, that hepatic expression of Lpl, which was previously known to be induced through PPAR␣ in the liver and through PPAR␥ in the adipocytes (62), was increased in both models, suggesting that the increase in hepatic PPAR␥ activity was sufficient to overcome the decrease in PPAR␣ activity, particularly in the liver-Cpr-null mice. Therefore, although a decrease in fatty acid oxidation and an increase in fatty acid uptake appears to account for the increases in hepatic fatty acid levels in both models, the model-specific increase in Cd36 expression and decreases in Cpt1a and Ascl1 expression might be the main causes of the development of fatty liver in the liver-Cpr-null mice but not in the Cpr-low mice.
Regulation Mechanisms of Hepatic CYP1-4 -The expression of many xenobiotic metabolizing P450s in families CYP1-4 is up-regulated in the livers of both liver-Cpr-null (11,12) and Cpr-low mice (13). Increased microsomal P450 expression was also found in the kidney of the Cpr-low mice (13). In the present study, we detected the induction of microsomal P450s in additional Cyp gene families, such as Cyp7, Cyp8, and Cyp51. Despite the overall increases in hepatic microsomal P450 levels, we also found that the expression of some microsomal Cyp genes (e.g. Cyp4a10 and Cyp7b1) was suppressed.
Xenobiotic-induced expression of many genes in the Cyp1-4 families is mediated through the activation of one or more nuclear receptors, including PXR, CAR, aryl-hydrocarbon receptor, and PPARs (63). These nuclear receptors also interact with endogenous ligands, such as fatty acid and sterol metabolites, which may mediate the regulation of P450 expression in response to physiological variations, such as food intake and circadian rhythm. We had speculated previously that a feedback regulation by endogenous CAR ligands is involved in the large induction of Cyp2b10 in the liver-Cpr-null mice (11). The results of the present genomic analysis further support the idea that CAR is activated in the livers of the liver-Cpr-null and Cpr-low mice.
Inhibition of squalene epoxidase is known to induce CYP2B in the rat (64), through activation of CAR by upstream endogenous ligands (65). The activity of squalene epoxidase will be blocked in the livers of liver-Cpr-null mice and decreased in the livers of Cpr-low mice. Because squalene epoxidase is the first CPR-dependent enzyme in the cholesterol biosynthesis pathway, the loss of squalene epoxidase activity, together with the induction of five upstream enzymes (Fig. 2), will likely lead to accumulation of squalene and its precursors, and consequently activation of CAR. In the liver-Cpr-null mice, activation of CAR may be further augmented by a decrease in the expression of Nr0b2, a known inhibitor of CAR and other nuclear receptors (34,66). Consistent with activation of CAR, several known CAR target genes (in addition to Cyp2b10), including Abcc3, Ahr, Aldh1a1, Aldh1a7, Cyp1a2, Cyp2a5, and Gstm1 (67,68), as well as Cyp2c55, a likely CAR target gene based on the known activation of human CYP2C9 by CAR (69), were induced in the liver-Cpr-null mice, and all but Abcc3 were also induced in the Cpr-low mice (Fig. 2). The magnitude of induction of these genes was generally lower in the Cpr-low than in the liver-Cprnull mice, as expected from the lack of a decrease in Nr0b2 expression and the partial loss of CPR-dependent enzyme activities in the Cpr-low mice.
In contrast to CAR, PXR does not appear to be activated in either of the mouse models studied. Endogenous PXR ligands include bile acid intermediates, which stimulate expression of Cyp3a via activation of PXR (70,71). However, bile acid production would be reduced in the liver-Cpr-null and Cpr-low mice. Accordingly, the mouse Cyp3a genes, which are known targets of PXR (or SXR) (72), had Ͻ2-fold increases in expression in either mouse model.
Other Changes-In addition to microsomal Cyps, many other biotransformation genes, such as Aldh1a1, Aldh1a7, Ces2, and multiple isoforms of Gst, were induced in both models. Others, however, such as a Ugt gene (AI788959), were induced only in the liver-Cpr-null mice. The induction of these genes would enhance biotransformation of endogenous and xenobiotic compounds through non-P450 pathways, as a compensatory response to the loss of CPR. These changes should be considered in the design of any in vivo metabolic studies that will use these mouse models.
The induction of Gsts in both mouse models and the additional induction of Hmox1 in the liver-Cpr-null mice indicate the occurrence of oxidative stress in the liver, a finding confirmed by decreases in hepatic nonprotein sulfhydryl levels in the liver-Cpr-null mice (73). Oxidative stress was most likely induced by the increased accumulation of lipids, as well as by the overproduction of P450 enzymes. Most interestingly, the expression of several heat shock proteins was reduced in both liver-Cpr-null and Cpr-low mice, a feature that distinguishes the observed stress response from the gene expression changes induced by endoplasmic reticulum stress caused by CYP2C2 overproduction in cultured cells (74). Induction of enzymes in amino acid metabolism and decreases in fatty acid-binding protein were found in both studies, but heat shock proteins were induced by endoplasmic reticulum stress in HepG2 cells overexpressing P450 (74). However, overexpression of CYP2C2 in HepG2 cells would not cause the metabolomic changes associated with CPR loss. The anticipated metabolomic differences might explain the differences in gene expression changes between the two differing systems.
Increased gene expression was found in the coenzyme A and vitamin metabolism pathways for both mouse models. Because many of the altered genes encode lipogenic enzymes, their induction together with reduced expression of genes in the gluconeogenesis pathway would further support an inhibition of fatty acid oxidation and an increase in lipid storage in the liver. The differences between the two mouse models, in the extent of these gene expression changes, might also contribute to the development of fatty liver only in the liver-Cpr-null mice. Furthermore, the model-specific induction of genes included in the GO term lysosome might be related to the observed mild inflammation and hepatic necrosis in the liver-Cpr-null mice (11).
In summary, our findings offer mechanistic insights into the function of CPR-dependent pathways in nuclear receptor signal transduction, CYP gene regulation, hepatic lipid metabolism, and sterol homeostasis. Our results indicate that the loss of CPR-dependent enzyme activities led to metabolomic changes in the sterol biosynthesis and metabolism pathways and that these changes in turn led to activation of SREBP, CAR, and PPAR␥, but not PXR, and repression of PPAR␣ and FXR. Alterations in the expression or activity of these nuclear receptors led to altered expression of numerous target genes, including many Cyps. The altered expression of a number of PPAR target genes eventually led to a decrease in fatty acid oxidation and an increase in fatty acid uptake, changes that account for increases in hepatic fatty acid levels in both models. However, an increase in Cd36 expression and decreases in Cpt1a and Ascl1 expression (all of which are model-specific changes that presumably result from the differing extent of CPR loss and, consequently, the differing extent of changes in PPAR␣ and PPAR␥ activities between the two models) appear to be the main causes of the development of fatty liver in the liver-Cprnull but not the Cpr-low mice. Further experimental confirmation of these hypotheses, including confirmation of gene expression changes at both protein and mRNA levels, and in-depth analysis of additional altered pathways are warranted.