The incidence and severity of obesity and type 2 diabetes are increasing in Western societies. The progression of obesity to type 2 diabetes is gradual with overlapping symptoms of insulin resistance, hyperinsulinemia, hyperglycemia, dyslipidemias, ion imbalance, and inflammation; this complex syndrome has been called diabesity. We describe here comparisons of gene expression in livers of A/a (agouti) vs. Avy/A (obese yellow) segregants (i.e., littermates) from BALB/cStCrlfC3H/Nctr × VYWffC3Hf/Nctr-Avy/a matings in response to 70% and 100% of ad libitum caloric intakes of a reproducible diet. Twenty-eight (28) genes regulated by diet, genotype, or diet × genotype interactions mapped to diabesity quantitative trait loci. A subset of the identified genes is linked to abnormal physiological signs observed in obesity and diabetes.
- gene regulation
- quantitative trait locus
the incidence and severity of obesity and type 2 diabetes mellitus (T2DM) are increasing in Western societies (63). Obesity often induces or is associated with subphenotypes including insulin resistance, hyperinsulinemia, hyperglycemia, dyslipidemias, ion imbalance, and inflammation that lead to non-insulin-dependent DM (reviewed in Ref. 5). The progression with obesity to type 2 diabetes is gradual with overlapping symptoms, and this complex syndrome has been called diabesity (57).
Cell culture systems have revealed details of signal transduction, biochemical pathways, nuclear receptors, transcription factors, and cell type-specific responses that are associated with diabesity. In vitro studies provide important links with disease processes but require integration with other in vivo processes and pathways as well as analyses in humans. Hansen (32) provided a detailed review of the genes, their proteins, and their pathways in the pancreas that are suspected to be candidate genes for T2DM. Such integrative analyses are needed for all of the organ systems involved in diabesity.
An approach for discovering causative genes for diabesity are association studies that identify chromosomal regions and the genes encoded therein that contribute to complex phenotypes. Quantitative trait loci (QTL) mapping and candidate gene association studies provide evidence for several chromosomal regions or genes that play a role in diabesity (reviewed in Refs. 8, 32). Excluding mutations that cause mature onset diabetes of the young (MODY), variants of ∼25 genes have been linked with type 2 diabetes in certain human populations (see Ref. 32), although only 6 genes (sulfonylurea receptor, glucagon receptor, glucokinase, potassium inward rectifier channel Kir6.2, peroxisome proliferator activated receptor-γ, and GLUT1 glucose transporter) have shown significant associations in three (P < 0.05), two (P < 0.01), or one (P < 0.001) study (23).
The expression of genetic information depends upon a genotype’s interactions with its environment, since the genome is constantly exposed to naturally occurring bioactive compounds. Diet may be the most important environmental variable in altering genetic expression due to the daily exposure to different types and concentrations of chemicals in food. For example, food contains ligands for retinoic acid receptors (RAR), retinoid X receptors (RXR), peroxisome proliferator activated receptors (PPARs), vitamin D receptors, and other nuclear transcription factor receptors (10, 14, 24, 40; and reviewed in Refs. 42, 43, 65). Epidemiological studies have consistently shown that high intakes of certain foods are linked to the incidence and severity of obesity, diabesity, cardiovascular diseases, certain cancers, and other chronic diseases (reviewed in Ref. 98). We previously proposed that certain diet-regulated genes would cause or contribute to initiation, development, or progression of chronic diseases (44, 69; and reviewed in Ref. 42).
We have conducted studies to identify diet-regulated genes that are likely to influence or participate in disease development or severity in different genotypes (18, 44, 68, 69, 86). Gene expression was compared between inbred strains of mice and between genotypes with different susceptibilities to a complex phenotype of chronic disease influenced by diet. Genes differentially expressed, based upon diet, genotype, or their interactions, participate in producing the difference in phenotype. Subsets of regulated genes that play a role in the disease can be identified when they map to chromosomal regions associated with that complex phenotype, i.e., within independently derived QTL or within loci identified by other genetic methods. Our approach can be considered a variation of the common variant/common disease (CV/CD) hypothesis (12, 51), with the added proviso that diet influences expression or activity of variants of common genes. Simply stated, the CV/CD hypothesis is that combinations of naturally occurring alleles of unlinked genes rather than deleterious mutations produce chronic diseases. Leiter and coworkers have demonstrated that diet (11) and maternal environment (75) interact with genetic loci to influence development of type 1 diabetes (11) or specific subphenotypes of diabesity (75).
In the experiments described here, gene expression was compared in livers of A/a (agouti) vs. Avy/A (obese yellow) segregants (i.e., littermates) from BALB/cStCrlfC3H/Nctr × VYWffC3Hf/Nctr-Avy/a matings in response to 70% (caloric restriction, CR) and 100% (AL) of ad libitum caloric intakes. The Avy mutation resulted from insertion of an intracisternal A particle genome in the promoter of the agoutigene (7). The insertion causes ectopic expression of the 131-amino acid containing agouti gene product (agouti signaling protein, ASP). Avy/− mice are mottled yellow in coat color, more metabolically efficient, obese, and display subphenotypes of diabetes including mild hyperglycemia and hyperinsulinemia, with increased risk of spontaneous and chemically induced cancers (99, 101; and reviewed in Refs. 64, 93, 103, 107). In contrast, A/a mice remain disease- and symptom-free for most of their lives. We have reported analyses of blood glucose with body and brain weights in mice from this experiment (102). The results indicated that caloric restriction decreased metabolic efficiency associated with continuous ectopic expression of agouti.
Comparisons of gene expression between A/a and Avy/A mice in response to defined diets identify genes regulated by diet (caloric restriction), genotype (A/a cf. Avy/A), and diet × genotype interactions. Genes regulated by genotype alone under one or more dietary conditions yield molecular information about physiological differences between agouti and obese yellow mice.
Twenty-eight genes mapped to diabesity loci. Some are expected to contribute to characteristics expressed in obese yellow mice. An additional 41 genes mapped to obesity and weight gain QTL. A subset of the identified genes has been associated with diabetic subphenotypes, and others can be linked to abnormal physiological conditions observed in obesity and diabetes.
MATERIALS AND METHODS
Mottled yellow Avy/A and agouti A/a F1 offspring were produced by mating BALB/cStCrlfC3H/Nctr dams with VYWffC3Hf/Nctr-Avy/a sires. Animal husbandry conformed to regulations for animal care at the National Center for Toxicological Research/FDA (NCTR, Jefferson, AR) as described (102). Fifty-two females of each genotype were fed ad libitum, and 52 females of each genotype were fed 70% (CR) of the average intake of the ad libitum (AL) group for 11 wk beginning at age 13 wk. Diets were supplied by Research Diets (New Brunswick, NJ), and the composition of the diets ensured that the CR mice received 100% of the protein, fat, fiber, salts, and vitamins supplied to the AL mice (102). The fast-refeed regimen described in Refs. 68 and 69 was followed to assure that mice were in the same nutritional state when euthanized.
RNA isolation from tissue.
Livers were immediately removed at euthanasia and immersed in liquid N2. The frozen livers were stored at −80°C until isolation of RNA. Immediately prior to RNA isolation, the frozen livers were ground with a mortar and pestle under liquid N2, and 100–200 mg was processed using TRIzol reagent (Life Technologies, catalog no. 15596-026). mRNA was isolated from total RNA with the MessageMaker mRNA Isolation System (Life Technologies, catalog no. 10551-018). Yields for RNA and mRNA were determined by spectrophotometry. Quality of total RNA was assessed using gel electrophoresis by integrity of the 18S and 28S rRNA.
RNA from two mice of each genotype-diet combination (n = 2, 8 mice total) were analyzed. Then 2.5 μg mRNA was hybridized with oligo-dT (Genome Systems, St. Louis, MO) and lyophilized. Samples were rehydrated in a 7-μl volume containing MMLV reverse transcriptase (Promega catalog no. M1701) and 10:1 [α33P]dCTP:dATP,dGTP,dTTP (Amersham Pharmacia Biotech, catalog no. AH9905) in buffer and incubated for 2 h at 42°C. Remaining RNA was degraded by alkali treatment. Radiolabeled probe was acid-precipitated and purified using ProbeQuant columns (Amersham Pharmacia Biotech, catalog no. 27-5335-01). Label incorporation was determined as %(cpm of purified probe/cpm of probe prior to RNA degradation). Only probes with incorporation >20% were used.
Probes were hybridized to GDA (ver. 1.1) arrays (Genome Systems, St. Louis, MO). These are 22 × 22-cm nylon filters arrayed with duplicate spots of 18,376 separate cDNAs and expressed sequence tags (ESTs) from the IMAGE mouse library, as well as 32 additional control sequences.
Array filters were prewetted and prehybridized with NorthernMax hybridization buffer (Ambion, catalog no. 8677) for 2 h at 42°C in rotating tubes in a hybridization oven. Four filters were used for comparison of one complete set: one filter was hybridized with cDNA from Avy/A mice fed AL, one filter hybridized with cDNA from Avy/A fed CR, etc. A second set of filters was used for the replicates. Hybridization was carried out with purified probes and labeled control markers in 10 ml NorthernMax hybridization buffer at 42°C for 16–20 h. Posthybridization, filters were rinsed with 2× SSC, 1% SDS, then washed thoroughly under stringent conditions (0.6× SSC, 1% SDS at 68°C).
Washed filters were exposed to a PhosphorImager screen (Molecular Dynamics) for 48 h and were scanned on a Molecular Dynamics STORM 860 imaging system at 100-μm resolution.
Analysis of images.
PhosphorImager data were submitted to Genome Systems for scanning and analysis using their internal software. All numerical data generated were returned to us. Background correction was calculated for each spot by subtracting the pixel density value of neighboring pixels from the pixel density within the spot. Spot intensity values were the background-corrected mean pixel density for each spot multiplied by the number of pixels comprising the spot. For each pair of filters (A and B), spot intensity values were normalized using a ratio of background-corrected global mean intensities. Global means were calculated for all spot intensity values on each filter (meanA and meanB). All spot intensity values on filter B were multiplied by a ratio of meanA/meanB. For each clone, spots were arrayed in duplicate on each filter. The intensity value for calculating the expression ratio was the value of those normalized for each pair. Genome Systems performed these calculations.
Data from replicates were averaged and then analyzed using the method described in study of gene expression in obese and diabetic mice (50) as modified by Lin et al. (60). This procedure detects genes whose expression does not change from genes whose expressions change by assessing their differential expression relative to the intrinsic noise found in the nonchanging genes. We report the results of these statistical analyses in this paper.
Accession numbers of each clone were converted to an NCBI unique identifier number, GI. FASTA sequences of the 354 genes were blasted against the GenBank database (Supplemental Table S1, available at the Physiological Genomics web site).1 The accession or GI numbers and the function of each gene were then used to search the Mouse Genome Database (MGD) at Jackson Laboratory (http://www.jax.org) for chromosomal map positions. Links within MGD were used to find additional functional information in LocusLink (NCBI, http://www.ncbi.nlm.nih.gov) and the Online Mendelian Inheritance in Man (OMIM; http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=OMIM), since some genes have different names in different databases. Primary literature reports were found for genes mapping to diabesity QTL (Supplemental Table S2). Genes mapping to obesity QTL are provided in Supplemental Table S3. Some information regarding gene function therefore comes from diverse cells, tissues, and organisms.
The genotypes (A/a vs. Avy/A) of each of the 8 mice (2 from each group) were confirmed by the presence (Avy) or absence (A) of an 870-bp product indicative of the Avy allele. Genomic DNA was isolated from the kidneys of test mice using standard procedures. Control DNAs were genomic DNAs from the livers of mice of known genotypes. About 1 μg of genomic DNA (without restriction digestion) was amplified with a primer set spanning the proximal long terminal repeat (LTR) of the Avy allele as described previously (as in Ref. 13, with minor modifications). PCR products were analyzed on a 1% agarose in 0.5× Tris-borate-EDTA buffer, stained with ethidium bromide, and imaged under ultraviolet illumination with a digital camera (UV Products, Upland, CA).
Expression levels of 18,000+ genes were compared among livers of each genotype fed AL or CR and between genotypes (A/a cf. Avy/A) fed either AL or CR. Genotypes were confirmed by amplification of an Avy-specific fragment of 870 bp (Fig. 1). These comparisons are designated Aal:Yal [A/a fed AL divided by Avy/A fed ad libitum (AL) with caloric restriction (CR)] and are reported in the expression ratio columns in Supplemental Tables S1–S3. We analyzed expression of only those genes whose expression levels showed statistical significance based upon the algorithm of Lin et al. (50, 60). As calculated, the ratios do not show up- or downregulated genes but rather refer to the comparison between the two genotype:diet cf. genotype:diet combinations. For these criteria, the four two-way comparisons between A/a and Avy/A fed 70% or 100% of ad libitum caloric intakes revealed 388 genes whose regulation differed among various genotypes and diet combinations. Of these 388 genes, the functions of 223 are of known function, are named genes (see Supplemental Table S1), or whose map position in mouse chromosomes are known. The genes were sorted based upon gene ontology into metabolic enzymes, signal transduction, structural, transcription, splice components, immune function, protease, and unknown function (Fig. 2 and Table 1).
Genes regulated by genotype.
The sum total of gene expression of all genotype-regulated genes contributes to physiological differences between Avy/A and A/a mice. Ten genes were regulated by genotype regardless of caloric intake: A/a fed CR relative to Avy/A fed CR, and A/a fed AL relative to Avy/A fed AL (Table 1, genotype in both diet rows). Ectopic expression of agouti altered abundance of these genes in a predictable manner: seven genes more abundant in Avy/A and A/a mice, and the other three were less abundant in both genotypes. An additional 76 genes were regulated by genotype [comparing abundance of mRNA in A/a vs. Avy/A in either AL-fed mice (Table 1, genotype AL, 39 genes) or CR-fed mice (Table 1, genotype CR, 37 genes)]. These genes might be regulated in the same fashion regardless of calories eaten, but they did not pass the statistical cut off.
Two notable examples (Aes and Mark4) illustrate how ectopic expression of agouti (in Avy/A) alters expression and phenotype.
Aes, amino-terminal enhancer of split, is more abundant in Avy/A relative to A/a mice regardless of the calories consumed. Aes is a co-repressor of NFκB (87), which is activated in insulin-resistant tissues. Aes is less abundant in muscle tissue of humans with a family history of T2DM relative to those with no family history (table 3 from Ref. 71).
Mark4, MAP/microtubule affinity-regulating kinase, is more abundant in Avy/A relative to A/a mice regardless of the calories consumed. Mark4 participates in the Wnt/β-catenin signaling pathway (45), which, when misregulated, may result in cancer. GSK-3β (glycogen synthase kinase 3β) is a negative regulator of this pathway and is downregulated when cells are exposed to Wnts (reviewed in Ref. 17). GSK-3β is also a suppressor of glycogen synthase and insulin receptor substrate 1 (41).
Genes regulated by genotype in one dietary condition (Supplemental Table S1, columns 3 and 4) also contribute to the overall expression of Avy/A phenotypes.
More genes were differentially regulated by diet in the Avy/A genotype than in the A/a genotype (Table 1, rows 1 and 2). We previously showed that CR abolished metabolic efficiency in obese yellow (Avy/A) mice (103). Gene products with differential abundance in Avy/A fed AL vs. CR (Supplemental Tables S1–S3, Yal:Ycr column) may alter energy metabolism during caloric restriction. Among the genes regulated by diet are Pdtgs, Pparbp, and Fabp.
Pdtgs, prostaglandin D synthase, was more abundant in Avy/A fed AL relative to Avy/A fed CR. This enzyme produces prostaglandin G2 which is a precursor to prostaglandin J2 (PGJ2). PGJ2, which is in turn a precursor to 15-deoxy-Δ12–14-PGJ2, the primary ligand of PPAR-γ (reviewed in Ref. 66).
Pparbp, peroxisome proliferator activated receptor binding protein, is also more abundant in Avy/A fed AL relative to Avy/A fed CR. Pparbp is a coactivator of PPAR-α and -γ, RAR-α, RXR, estrogen receptor (ER), and thyroid receptor β1 (TR-β-1) (108). PPAR-γ appears to be one of the key regulators of glucose and lipid homeostasis (30).
Fabp, fatty acid binding protein, is more abundant in A/a mice fed AL relative to A/a mice fed CR. Fabp may function by targeting its ligand to the nucleus and may participate in regulation of gene expression by binding to PPAR-γ (104).
Supplemental Table S1 lists the other genes of known or predicted function, whose transcription is modulated by diet and genotype, that may also contribute to differences in energy metabolism between AL and CR Avy/A. Integrating these genes into a coherent explanation for those differences may require analyses of gene expression with more mice of each group to improve statistical significance.
Genotype × diet interactions genes.
A subset of gene products (19 of 223) identified by comparing Avy/A vs. A/a and calories eaten were regulated by more complex genotype × diet interactions: i.e., they were regulated by different genotypes fed CR or AL and by different diets in A/a vs. Avy/A mice (Table 1, row 6). The majority of these genes (16) were regulated in the same manner in two different diet-genotype conditions: they were more abundant in A/a mice fed AL vs. CR and in Avy/A mice fed CR relative to A/a mice fed CR. Increased calories or the presence of the Avy allele may independently contribute to increased transcription of these genes.
Summary gene regulation.
Table 2 shows genes regulated in each comparison calculated by summing the genes in each column of Supplemental Table S1 followed by sorting into up- or downregulation by diet or genotype (e.g., Aal:Acr and Acr:Aal; Table 1). Ad libitum feeding increases mRNA abundance of more genes in agouti and obese yellow comparisons (e.g., row 1 vs. 2 and row 3 vs. 4, in Table 2). The Avy/A phenotype also increased abundance of more genes relative to A/a in CR mice but not AL-fed mice (row 5 vs. 6 and row 7 vs. 8, Table 2). No consistent patterns among types of regulation and gene functions were obvious in this sample set of 223 genes.
Genes mapping to diabesity QTL.
The map positions of the 223 genes of known functions were determined and compared with QTL associated with various subphenotypes of diabetes (Supplemental Table S2). Twenty-eight of the diet-, genotype-, and diet × genotype-regulated genes in liver mapped to diabesity QTL (Supplemental Table S2). An interval distance of ±10 cM from a given QTL was used, a distance consistent with the marker density employed in most QTL association studies. Five murine QTL involved in diabesity, insulin levels, or regulation of insulin-like growth factor levels (labeled with footnotes 4–8 in Supplemental Table S2) overlapped with T2DM QTL in humans (23).
Several of the gene products regulated by diet, genotype, or their interactions are associated with diabetes and/or are in pathways that alter phenotypes consistent with one of the conditions of diabetes (Supplemental Table S2, diabesity association). Genes mapping to diabesity QTL and differently regulated between AL-fed Avy/A and A/a mice (Aal:Yal column in Supplemental Table S2) are likely to cause differences in diabetes subphenotypes observed in this model: e.g., continuous ectopic expression of agouti may “override” normal regulation of these genes (e.g., Aes, Pdgfra, and Grb2).
Aes (43 cM) maps near an insulin-like growth factor binding protein (Igftbp3q2) QTL (46 cM) on chromosome 10. IGF and its binding proteins, particularly IGFBP3, are thought to be involved in glucose homeostasis (reviewed in Ref. 37). Aes (amino-terminal enhancer of split) is a co-repressor of NFκB, which is activated in insulin-resistant tissues. Aes expression is increased in Avy/A mice relative to A/a mice, regardless of the diet (Table 2, columns 3 and 4). Aes expression in muscle was decreased significantly (P = 0.0178) in individuals with a family history of T2DM relative to individuals with no family history (71). However, its abundance was elevated (not statistically) in patients with DM. Evans et al. (19) suggested that oxidative and stress-activated signaling pathways (e.g., NFκB) underlie the development of complications in T2DM.
Pdgfra, platelet-derived growth factor receptor-α, was more abundant in Avy/A mice fed AL relative to A/a mice fed AL (Supplemental Table S2) and maps within a diabesity locus (Dbsty2) on chromosome 5. Dbsty2 is associated with increased adiposity. Pdgfra interacts with the Hedgehog signaling pathway. Changes in the hedgehog (Hh) pathway affected insulin production in the pancreas (90). Expression of the receptor’s ligand, Pdgfa, is under the control of Kruppel-like factor 5 (KLF5) and is cooperatively activated by the NFκB p50 subunit (1). A similar Kruppel-like gene, Klf1, is more abundant in Avy/A mice relative to A/a mice, regardless of the diet (Supplemental Table S1).
Grb2, growth factor receptor bound protein 2, was less abundant in Avy/A mice fed AL relative to A/a mice fed AL. Grb2 is a signal transduction adaptor protein that is recruited to caveolae-localized receptor complexes (including the insulin receptor) by increased levels of IRS-1 and insulin (4). Grb2 participates with Ash to reorganize the cytoskeleton in response to insulin (91). Grb2 maps to chromosome 11, 72 cM near Nidd4 at 68 cM.
Several other genes (e.g., Smo and Flnc, below) had more complex regulatory patterns but may play a role in causing differences in subphenotypes of diabesity.
Smo, smoothened homolog, is a member of the Indian Hedgehog (IHH) signaling pathway. Smo maps (chromosome 6, 7.2 cM) near Fglu (chromosome 6, 16 cM) which is associated with increased fasting plasma glucose levels. IHH may be involved in chronic pancreatitis and insulin production (46). Smo mRNA was more abundant in A/a mice fed AL relative to A/a mice fed CR and was more abundant in Avy/A mice fed AL relative to A/a mice fed reduced calories, an example of diet × genotype interaction.
Flnc, filamin also mapped (8.5 cM) near the Fglu QTL on chromosome 6. Filamin had the same complex expression pattern as Smo. Insulin causes changes in cytoskeleton architecture (91), and filamin may bind to the insulin receptor (33).
Some individual genes in Supplemental Table S2 have not been studied for a role in processes affected by diabetes, but members of their functional family have been linked to processes that are altered in that disease. Other genes and their products mapping to diabesity QTL have only a tentative association to conditions in diabetes: Madl2 (mitotic arrest deficient), debrin, and disheveled 2. Nevertheless, these genes are candidates for diabetes subphenotypes in Avy/A mice by virtue of their regulation by genotype or diet and their map position near QTL associated with diabetes symptoms.
Genes mapping to obesity QTL.
Forty-one genes mapped to weight gain or obesity QTL (Supplemental Table S3), and 12 of these genes (Idh1, Pdgfra, Flnc, Dvl2, Nhp2, Sept8, Skpa1, Mpdu1, Grb2, Gsc, Gpt1, H1fo) mapped to overlapping diabesity QTL. Genes mapping to obesity loci may contribute to the subphenotypes expressed in this model (Supplemental Table S3). Other QTL for adiposity at various sites and total carcass lipid levels (for a comprehensive review, see Ref. 6, or see http://www.jax.org) were not included in our analyses, since these parameters were not measured in this study. Many diabesity QTL overlap obesity QTL (Supplemental S2 and S3) as would be expected for the diabesity phenotype (57). Associations with many specific molecular pathways influencing or involved in obesity and/or weight gain can be made for each of the other diet-, genotype-, and diet × genotype-regulated genes mapping to obesity and weight gain QTL (Supplemental Table S3, Obesity Association).
Diabesity is a complex trait resulting from interactions between multiple genes and environmental factors. In humans, chronic exposure to excessive calories, deficiencies of micronutrients, and certain types of macronutrients induce obesity and diabetes in individuals, presumably without deleterious mutations in participating genes. These diseases therefore fit the CV/CD hypothesis proposed by Lander (51) and Collins and colleagues (12). We have proposed that one or more of the gene products participating in development of chronic diseases will be regulated at least in part by diet (44, 69; and reviewed in Ref. 42), since different macronutrients and excess calories are associated with almost all chronic diseases (e.g., 97, 98).
Although chronic diseases are multigenic in nature, much information regarding the pathways involved in disease development has been discovered by the study of rodent models with single gene defects or induced mutations (knockouts and transgenics) that mimic diabetes and/or obesity. Comprehensive reviews of the mouse models for insulin resistance (38) and obesity have recently been published (6). The general conclusion from these reviews echoes the conclusion of Wolff (100) that similar if not identical phenotypic expressions of a disease state can be reached by different metabolic routes. That is, alterations in many pathways can produce the same phenotype. The specific genes and their transcriptional regulation reported herein, therefore, are most applicable to obesity and subphenotypes of diabetes produced by the dominant mutation (Avy) in the agouti gene. Nevertheless, these genes and their variants may identify sets of pathways that collectively produce the specific diabetes subphenotypes and obesity pattern in Avy/A mice or other genetically susceptible mice. That is, some of these pathways may also be involved in other models of diabesity if gene variants in the key regulatory or structural genes collectively produce expression changes similar to those observed in this specific mouse model. Genes identified in the Avy/A and A/a comparison may contribute to obesity or diabesity in humans if their regulation is altered in a similar manner. Five of the genes found in our analyses (fatty acid synthase, malate dehydrogenase, sterol C5 desaturase, dynein, and epidermal growth factor receptor pathway 15) were also differently regulated in livers of BTBR-ob/ob (obese and diabetes susceptible) compared with C57BL/6-ob/ob mice (obese and diabetes resistant) fed a chow-based diet ad libitum. Although we and others identify candidate disease genes through gene expression analyses, changes in the activity of proteins encoded by coding single nucleotide polymorphisms (SNPs) may also be associated with disease development (51).
Laboratory animal studies have consistently shown that reducing caloric intake is the most effective means to reduce the incidence and severity of chronic diseases, retard the effects of aging, and increase genetic fidelity (reviewed in Refs. 92, 96). Caloric restriction may produce its largest effects by increasing respiration (59) with the concomitant increase in the amount of NAD+ (58). NAD also is a cofactor for Sir2, a histone deacetylase involved in chromatin silencing of nucleolar rDNA and the telomere-located mating type locus (29, 62). Several genes involved in NAD+ metabolism were found in our screen, (e.g., NADH-ubiquinone oxidoreductase 1α subcomplex, lactate dehydrogenase, malate dehydrogenase, aldehyde dehydrogenase, and isocitrate dehydrogenase) but these were not regulated consistently within any one genotype × diet condition (Supplemental Table S1).
Several laboratories examined the effect of caloric restriction on gene expression in individual mouse strains in relation to their age. Genes involved in many different pathways were regulated by CR and/or aging in livers of C3B19RF1 (a long-lived F1 hybrid mouse; 16) and C3B10RF1 mice (9), in muscle of C57BL/6 mice (54, 55, 95), in heart tissue from B6C3F1 mice (53), and in livers of the long-lived Snell dwarf (dw/dw) stock. A few genes identified in these studies (Aes, Fasn, Fabp; Ref. 9) matched genes found in Supplemental Table S1, although the regulation by CR was not always consistent with our results.
We found 388 hepatic genes or ESTs regulated in the same manner in replicate experiments with 223 genes having a known function. We believe that the designations of genotype, diet, and genotype × diet interactions will be specific to the experimental model used in this study. Agouti protein has been shown to regulate gene expression in cell culture systems (26, 27, 83). Therefore, in the Avy genotype, the constant ectopic expression of agouti signaling protein may “override” normal genotype-specific or diet-regulated gene regulation. Less obvious will be those genes regulated by genotype × diet (i.e., environmental interactions). Many promoters are regulated by multiple receptors and by accessory factors. For example, HNF3γ is regulated differently in rats fed protein-free, casein, or gluten diets (39). Hence, depending upon the diet, transcription of genes regulated by this receptor may show differential regulation by non-diet influenced factors in mice fed diets that decrease the expression of Hnf3γ. A confounding variable that is likely to alter gene regulation will be variants (SNPs) within each regulatory gene and the promoters that interact with them (106). Although this complexity is noteworthy, recent reports from the human genome project suggest a limited number of haplotypes in the human population (28), and it is likely that mice also have a limited genetic diversity (94).
The sum total of the expression differences between Avy/A and A/a mice fed AL identify the hepatic genes that contribute to the obese yellow phenotype (Aal:Yal; expression ratios in Supplemental Tables S1–S3). Genes of all functional classes and types of regulation were differently expressed in this genotype comparison. No apparent pattern was discernable within this set, and new analytical tools will be needed to identify key regulatory and expression patterns among the many genes, their pathways, and their type of regulation.
QTL are used to associate chromosomal regions with complex traits. There are now over 1,700 QTL for disease, subphenotypes of disease, enzyme or protein levels, behavior, and other complex traits in mice (see http://www.jax.org). The limitations of using QTL data are that these 1) may be specific to the inbred strains analyzed, 2) identify 20- to 30-cM regions of DNA, and 3) often cannot detect interactions with other loci (22). In addition, few mapping studies rigorously control or report diets; environment is known to have a large influence on the identification of QTL affecting complex traits, at least in plants (70). Our approach combines the strength of array technology with the power of genetics to identify potential causative genes. A key additional component of our approach is the rigorous control of diet composition and a timed feeding regimen (68, 69) that will allow for replication of the experiments.
Even with limitations of the current experiment (type of array, number of mice, single tissue source), the data presented herein identify potential novel candidate genes in many different functional pathways that may play a role in expression of subphenotypes of diabesity. Several of the genes that were found to be diet-regulated and mapped to diabetes QTL had previously been linked to specific pathways affected by or involved in diabetes. Aes, Grb2, and Pdgt1 are linked to type 2 diabetes or are in pathways directly regulating insulin function. Other genes mapping to QTL from our screen can be associated with various alterations in metabolism found in diabesity. They become candidates for further testing.
Since obesity is an “amorphous” phenotype, with adiposity, weight gain, and overall weight as the key phenotypic markers, it is more difficult to compare candidates identified in this screen with those found in other model organisms or humans (reviewed in Ref. 84). In addition, Wolff (100) reviewed phenotypic and molecular differences between obesity induced by dominant mutations in the agouti gene and by the recessive mutation Lepob in the leptin gene and concluded that many physiological parameters are diametrically opposed in these two obesity models, a conclusion consistent with that of others (6, 72). Genetic analyses support this conclusion, since a large number of overlapping obesity and weight gain QTL have been identified (see Supplemental Table S3 for a subset of these QTL and http://www.jax.org). Nevertheless, genes analyzed in this screen that are regulated by diet, genotype, and genotype × diet that map to obesity QTL may be considered candidates for obesity development or severity.
The strategy described is a means to identify diet-, genotype-, and genotype × diet-regulated genes that cause or promote the development and severity of complex phenotypes. A similar approach compared gene expression patterns in strains of nondiabetic obese mice and diabetic mice but did not systematically alter diet (50). Comparative genetic approaches can be applied to different mutant models and their normal inbred parent or strain and to congenic siblings produced specifically for separating and combining QTL producing a complex phenotype (e.g., Ref. 76). By comparing across the different genotypes fed the same diet, genotype-regulated genes can be identified. Similarly, by feeding two or more diets to mice with different genotypes, diet-regulated and diet × genotype-regulated genes can be identified.
Understanding diabetes and obesity will require integration of knowledge from individual pathways that have been elucidated to date. However, inclusion of diet as a variable in a systems biology approach will also be necessary to fully explain complex phenotypes, almost all of which are influenced by environment, and specifically by dietary variables. This type of scientific study is called “nutrigenomics” or “nutritional genomics” (reviewed in Ref. 43). Knowledge of the interactions of diet and genotype will be needed when testing and treating these diseases in human populations.
This work was supported by American Institute for Cancer Research Grant NCB93B63 and by a CRADA from the National Center for Toxicological Research/United States Federal Drug Administration (NCTR/FDA), by National Institutes of Health Grant PO1-AG-20641 (to C. A. Cooney), by the Illinois-Missouri Biotechnology Alliance (USDA, to J. Kaput and K. G. Klein), and by National Center for Minority Health and Health Disparities Center of Excellence in Nutritional Genomics Grant MD-00222.
The research described in this report was conducted at NCTR/FDA (Jefferson, AR), Northwestern University Medical School (Chicago, IL), the University of Arkansas for Medical Sciences (Little Rock, AR), and NutraGenomics (Chicago, IL).
We thank Heather Mangian and Eureka Wang for contributions to this work.
Present addresses: K. G. Klein, Nanosphere, Inc., Northbrook, IL 60062; E. J. Reyes, Applied Biosystems, Inc., Foster City, CA 94404; W. A. Kibbe, Robert Lurie Cancer Center and Center for Functional Genomics, Northwestern University and Northwestern University Medical School, Evanston and Chicago, IL 60611; B. Jovanovic, Department of Preventive Medicine, Northwestern University Medical School, Evanston, IL 60611.
↵1 The Supplementary Material for this article (Supplemental Tables S1–S3) is available online at http://physiolgenomics.physiology.org/cgi/content/full/00065.2003/DC1.
Article published online before print. See web site for date of publication (http://physiolgenomics.physiology.org).
Address for reprint requests and other correspondence, and present address for J. Kaput: NutraGenomics, Box 32, 2201 West Campbell Park Drive, Chicago, IL 60458 (E-mail:).
- Copyright © 2004 the American Physiological Society