Effects of Leptin Deficiency and Short-Term Repletion on Hepatic Gene Expression in Genetically Obese Mice

Libraries of Unigene murine cDNAs were obtained from Research Genetics (Huntsville, AL) as bacterial colonies in a 96-well microtiter plate format. Plasmids that represented each clone were isolated with the use of a high-throughput purification system (Turbo Miniprep; Qiagen, Valencia, CA). The insert from each clone was amplified in a 75-μl polymerase chain reaction (PCR), using vector-derived primers (T7–5′-AATTAACCCTCACTAAAGGG-3′ and T3–5′-TAATACGACTCACTATAGGG-3′). To ensure that successful and specific amplification was achieved, 3 μl of each product was electrophoresed on a 1% agarose/Tris-borate-EDTA gel. The remainder (∼72 μl) of reaction mixes that yielded single products of the appropriate size and concentration were ethanol precipitated and resuspended in 3× sodium chloride–sodium citrate (SSC) with a final DNA concentration of 0.2–0.6 μg/μl. The DNA was transferred to new 96-well plates, sealed, and stored at −20 until arrayed. Polylysine-coated glass 75 × 25-mm microscope slides were obtained commercially (Sigma, St. Louis, MO).

All arraying was performed with a four-pin-head GMS 417 Arrayer (Affymetrix, Santa Clara, CA) according to the manufacturer’s instructions at a density of 1,300 spots/cm². The robotics systems are self-contained within a Plexiglas enclosure and maintained at a relative humidity of 40–50% and temperature of 23–26°C during arraying. The microarrays were stored in a low-humidity environment for >2 weeks before use.

For ensuring covalent linkage of PCR products to the polylysine surface, microarrays were preprocessed on the day of hybridization. Each microarray was moistened over a bath of heated water (95°C) and exposed to 60 mJ ultraviolet energy. After cross-linking, the microarrays were washed serially in 0.1% SDS, 0.2× SSC, and 100% ethanol and dried by centrifugation in a tabletop centrifuge adapted for this purpose. Dried microarrays were prehybridized immediately.

Total RNA from 50 mg liver was extracted with the use of an acid-phenol method according to the manufacturer’s instructions (Trizol; BRL Gibco, Gaithersburg, MD). From each sample, 0.5–1.0 μg polyA RNA was isolated using a polydeoxythymidine resin (Oligotex; Qiagen). PolyA RNA was transcribed into fluorescently labeled cDNA using an anchored oligo-dT primer (dT₁₈MN), unlabeled deoxynucleotide triphosphates (dNTPs), reverse transcriptase, and deoxyuridine-triphosphate (dUTP) labeled with one of two fluorophores, Cy3 or Cy5 (17). Unincorporated dNTPs were removed, and the sample was concentrated with the use of a minifiltration system (YM-30; Microcon, Bedford, MA).

Microarrays were prehybridized with 15 μl of a formamide-based hybridization solution (0.75 mol/l NaCl, 0.15 mol/l Tris-HCl [pH 7.5], 0.1 mol/l Na₂HPO₄, 0.1% Na₄HP₂O₇, 2% SDS, 5× Denhardt’s solution, 100 μg/ml sheared salmon sperm DNA, and 50% formamide) for 1 h at 42°C in a humidified hybridization chamber. For each microarray, Cy5- and Cy3-labeled cDNA from two separate mouse liver RNA extractions were combined with 10 μl hybridization buffer, heated at 95°C for 10 min, and centrifuged for 5 min in a microfuge at maximum speed to remove any particulate matter. The combined sample was placed on a prehybridized microarray and covered with a coverslip. The coverslip was held in place by rubber cement, and the microarray was hybridized for 12–14 h at 42°C in a humidified chamber. After hybridization, the coverslip was removed and the microarray was washed at room temperature in serial 3-min washes of 0.1× SSC/0.1% SDS and two washes of 0.1× SSC. The slides were dried by centrifugation (1,000g) in a tabletop centrifuge.

The microarrays were scanned according to the manufacturer’s instructions with the use of an Affymetrix 428 Array Scanner. The scan of each microarray generated two 16-bit TIFF image files, one representing the fluorescence emission from each fluorophore.

Paired image files, generated from scans of microarrays, were analyzed with the use of Imagene software (BioDiscovery, Los Angeles, CA). This software package identifies spots within an array image and calculates mean, total, and median signal intensity values for the pixels that constitute each spot. In addition, the software calculates local background intensities and allows for outlier pixels to be excluded from calculations. In combination with CloneTracker 2.0 (BioDiscovery), the Imagene software correlates each spot with a specific cDNA clone and generates a spreadsheet output file containing mean signal and local background for each spot.

After the raw data from a single microarray were extracted from two TIFF images, they were downloaded into a single spreadsheet. Subtracting the mean local background from the mean signal of each spot generated a net signal for each represented cDNA. When the net signal for any spot was both greater than the average background for the entire array and greater than the local background, that spot was used for subsequent calculations. When both of these criteria were not met, a spot and the cDNA that it represents were not likely to give reliable data and, therefore, were not considered in further calculations. The sum of the net signal for each spot on both images was calculated, and the ratio between the sums for each image was defined as the normalization factor. Individual normalized expression ratios were calculated by dividing the net signal of a given spot i from image A by the net signal from the same spot on image B, multiplied by the normalization factor:

Mean normalized ratio for cDNA_i = (net signal A)_i/(net signal B)_i × (Σ net signal B)/(Σ net signal A)

Each microarray experiment was performed in triplicate using RNA extracted from a total of six animals. Thus, in comparing wild-type (Lep⁺/Lep⁺ C57BL/6J) with obese (Lep^ob/Lep^ob C57BL/6J) expression profiles, RNA was extracted separately from the three mice of each genotype to generate labeled cDNA. The labeled cDNA from a single wild-type mouse was combined with cDNA from a single obese mouse and hybridized to a microarray. Normalized mean ratio data were calculated for each cDNA on each microarray, and those sets of genes that showed a consistent 30% or greater increase or decrease in expression between two genotype/treatment groups were identified. The normalized mean ratio data were then used to calculate mean ratios for genes that demonstrated consistent changes.

All mice were females purchased from Jackson Laboratory (Bar Harbor, ME) at 4–6 weeks of age and acclimatized for at least 1 week before experimental manipulation or killing. The mice were kept individually in Plexiglas, ventilated cages with cob bedding and except where noted had free access to water and a standard pellet diet, which derives 10% of calories from fat (PicoLab Diets). The temperature was kept at 22 ± 1°C, and a 12-h light-dark cycle was maintained. All animals were killed by CO₂ asphyxiation 1 h into the light cycle. Columbia’s Institutional Animal Care and Use Committee guidelines were followed at all times.

Recombinant murine leptin was purchased from Peprotech (Princeton, NJ) and resuspended according to the manufacturer’s instructions in sterile phosphate-buffered saline, pH 7.4 (1 mg/ml), and sterile filtered (0.22-μm cellulose acetate filter). Resuspended leptin was stored at −20°C for no more than 2 weeks. Leptin-treated animals were weighed and received an intraperitoneal injection once daily 1 h before the dark cycle with 0.5 μg/g. Vehicle-treated animals were handled identically to the leptin-treated animals except that they received an injection of phosphate-buffered saline.

Ten micrograms of total RNA was extracted from the livers of Lep^ob/Lep^ob or wild-type C57BL/6J mice, electrophoresed through denaturing formaldehyde agarose gels, blotted to nylon membranes, and probed with ³²P-labeled, randomly primed DNA probes using standard protocols (18). Probes were generated from purified PCR products used in arraying and subjected to cycle sequencing with T7 and T3 primers using recommended protocols and an ABI3700 automated sequencing apparatus.

An important measure of any test’s utility is its specificity. Specificity is a function of the rate of positive results in a defined population. Knowing the rate and factors that influence the proportion of false-positive results is essential for interpreting any microarray expression data. Unfortunately, the specificity of microarray expression analysis has not been studied rigorously, and there is little consensus on the number of replicates needed or on criteria used to identify differentially regulated genes. In lieu of documenting specificity, investigators typically set a “cutoff,” or threshold, criterion above or below which changes in expression are deemed meaningful. This approach necessitates performance of second confirmatory tests, by either Northern or quantitative PCR analysis. Such confirmation, however, usually involves only a subset of the genes whose expression is believed to be increased or decreased. In addition, specificity is unlikely to be uniform for all microarray expression experiments and is almost certain to vary as a function of differences in array production and in biologic systems.

To assess the specificity of our arrays and protocols in the study of liver expression, we measured the false-positive rate as a function of threshold criteria and experimental repetition. We produced a set of arrays (M1.6K) that represented >1,600 murine genes. The cDNAs were taken randomly from a commercially obtained, sequence-verified, Unigene plasmid library. No effort was made to enrich the pool of cDNAs for liver-specific transcripts or for genes known or suspected to be involved in energy homeostasis. One microgram of polyA RNA was prepared from the liver of an 8-week-old C57BL/6J female mouse. The sample was divided equally; one half was used to generate Cy3-labeled cDNA, the other half was used to generate Cy5-labeled cDNA. The two differentially labeled samples were hybridized simultaneously to a M1.6K array, washed, and scanned with a fluorescent scanning confocal microscope. Using different lasers to excite each of the dyes, we obtained raw fluorescence data for each spot on the array. The pixels (10 μm × 10 μm) that define each spot were determined in a two-step process. Initially, a map that contained the location and size of each spot of DNA was loaded into an analysis program (Imagene, BioDiscovery) and subsequently confirmed visually by us. Thus, for each spot, total, median, mean, and background pixel fluorescent intensity were obtained for the two identical samples that had been labeled with two different fluorophores. From these intensity data, a ratio that represented the relative signal from each dye was calculated for every spot. Each ratio was multiplied by a normalization factor intended to account for systematic bias in the incorporation or scanning of the fluorescent dyes. For these calculations, any ratio <1 was transformed by taking its multiplicative inverse.

If the specificity of the expression arrays were ideal, then the normalized fluorescence ratio calculated for each spot in this experiment would equal 1. Any variation from unity, therefore, represents the technical specificity limits of our arrays and protocols. The distribution of ratios from this experiment was very tight (Fig. 1A). Fewer than 0.5% of spots had a normalized fluorescence ratio of >2. Thus, for a threshold of 2, a value used by many workers, the specificity of our array and protocol is >99%. As would be expected, the specificity of microarray expression analysis is a function of threshold ratio—as the threshold is increased, the specificity increases with a concomitant decrease in sensitivity. However, independently repeating the experiment and combining data across array experiments also would be expected to increase the specificity but with less reduction in sensitivity. Repeating this experiment two more times, with polyA RNA prepared from a second and third liver, demonstrated a similar distribution of ratios for each individual array (data not shown). Averaging the expression ratios across all three arrays, while keeping the threshold ratio at 2, increases the specificity to 99.9% (Fig. 1B). If there is no gene-specific bias or systematic error, then another and more conservative method for increasing specificity is to set a minimum ratio threshold across a set of arrays. In our test arrays, for example, there were no spots for which the expression ratio was 1.2 or greater on each of the three microarrays (Fig. 1B). Thus, if we set a minimum expression ratio threshold of 1.3, then we will detect differences with a specificity of >99.9%. Or, stated more directly, a difference in expression of 30% or more across three microarrays will identify genes with true differences in expression, with a false-positive rate of <0.1%.

To interrogate the gene expression profile from livers of wild-type and obese mice, we produced microarrays that contained 5,182 unique cDNAs (M5KA). So that we could monitor the internal reproducibility of our expression analysis, we spotted 384 genes in duplicate in the initial set of microarrays used in these studies. Livers were obtained from three 8-week-old female Lep^ob/Lep^ob mice and paired wild-type (Lep⁺/Lep⁺) control mice. PolyA RNA was extracted from each liver and used separately to generate fluorescently labeled cDNA. For one pair of RNA samples, Cy5 and Cy3 were switched so that on one array the Lep^ob/Lep^ob cDNA was labeled with Cy5, whereas on the other two it was labeled with Cy3. This was done to overcome any gene-specific dye bias. Differentially labeled cDNA from single control and single Lep^ob/Lep^ob mice were hybridized to a microarray. After washing and scanning, normalized expression ratios for each gene were calculated. Guided by our specificity calculations, we selected the subset of genes that demonstrated a consistent 30% alteration across all three microarray experiments. Among this group were 91 genes that were consistently upregulated and 71 that were consistently downregulated in livers of obese mice (Table 1). For several of the genes, our data confirmed previous reports of altered expression in livers of obese mice, e.g., fatty acid synthase and insulin binding proteins. However, the majority of the genes were not known previously to be differentially regulated in Lep^ob/Lep^ob mice.

We calculated confidence intervals for the normalized expression ratio of each gene (α = 0.05). Among the 162 genes that were consistently changed by 30% or more, only six have confidence intervals that cross unity, providing support that our expression profiling is robust and that our method identifies genes whose expression is significantly altered. To provide further confidence that the changes measured by our microarray expression experiments accurately reflected gene expression patterns within the livers of obese mice, we selected seven genes, six of which were differentially expressed on our microarrays, and performed Northern analysis. The RNA used for the Northern analysis was extracted from different mice than those used in the microarray expression studies but showed the same patterns of altered expression (Fig. 2). In addition, we resequenced these seven clones from our microarrays to confirm their identity.

The set of genes with altered expression within the livers of obese mice represent a broad spectrum of functional classes. The majority of the genes with altered expression are expressed sequence tags about which only sequence information is known. However, among the rest, we could identify nine intracellular signaling molecules, five cytochrome genes, three genes involved in glutathione metabolism, eight secreted proteins, and five transcription factors. The largest group of genes participate in intermediary metabolism, in particular fatty acid metabolism. (Striking is that the microarrays detect increases in expression of genes required for substrates necessary for fatty acid synthesis and fatty acid synthesis [fatty acid synthase, pyruvate dehydrogenase E1-α subunit, fructose bisphosphate aldolase] as well as increases in expression of genes involved in fatty acid oxidation [sterol carrier protein 2, hydroxymethylglutaryl (HMG)-CoA synthase 2, HMG-CoA lyase, 3-hydroxyacyl-CoA dehydrogenase, methylmalonyl-CoA mutase, medium chain acyl-CoA dehydrogenase].) On our array, no genes that are known to be involved in fatty acid metabolism were suppressed in livers of obese mice. These data suggest that in obese mice, leptin deficiency upregulates the transcription—or suppresses the mRNA turnover—of genes that regulate both fatty acid synthesis and oxidation. The transcription factors peroxisomal proliferating activated receptor-α (19) and sterol regulatory element-binding protein 1 (SREBP-1, also known as ADD1) (20) positively regulate the expression of genes involved in the oxidation and synthesis of fatty acids, respectively. The expression of both are elevated in Lep^ob/Lep^ob mice and thereby provide a transcriptional mechanism for the upregulation of a broad spectrum of fatty acid metabolism genes.

In human obesity, there is an increased risk of cardiovascular disease and mortality (21,22). However, Lep^ob/Lep^ob are relatively resistant to atherogenesis and have elevated concentrations of serum HDL particles (23,24). The basis of these two subphenotypes remains poorly understood. The elevation of HDL and its reduction by low doses of leptin result from altered hepatic metabolism of HDL (25). Among the genes altered in livers Lep^ob/Lep^ob mice, we identified four with potential roles in modulating serum cholesterol and thereby atherogenic susceptibility: HDL-binding protein, HMG-CoA synthase 2, HMG-CoA lyase, and apolipoprotein A-IV (Apo A-IV). All show increased expression in livers of obese mice. If the increased expression of any of these four genes mediates the atherogenic-resistant and elevated HDL subphenotype in obese mice, then their expression would be expected to normalize with short-term leptin treatment as HDL levels decrease.

Treating Lep^ob/Lep^ob mice with leptin rapidly reverses many aspects of the obese phenotype (26). Within 1 h of peripheral or intracerebroventricular leptin administration, serum glucose and insulin decline. Within 24 h, food intake decreases more than 50%, and within 3 days, total body weight is significantly lower. To identify genes whose expression is modulated by leptin, we compared the expression profiles of obese mice that were treated for 1 week with leptin with those treated with vehicle. After 1 week of daily intraperitoneal leptin treatment, obese mice lost 17% of their body mass, whereas the vehicle-treated obese mice continued to gain weight (4%). As expected, Lep^ob/Lep^ob mice that were pair-fed with leptin-treated mice lost less body mass (13%) than leptin-treated mice (17%). Liver RNA was extracted from three mice in each treatment group and converted individually to fluorescently labeled cDNA. cDNA from each leptin-treated mouse was paired with a vehicle-treated or pair-fed cDNA sample and hybridized to a microarray. Data from the six microarrays identified sets of genes whose expression was altered between obese mice and obese mice that were treated with leptin and between obese mice that were treated with leptin and pair-fed obese mice.

The profiles generated from arrays that compared leptin- and vehicle-treated mice identified 29 genes whose expression was consistently altered 30% or more (Table 2); the comparison of leptin-treated and pair-fed profiles consistently detected such differences in expression of 35 genes (Table 2). Of the 158 genes whose expression differed between livers of obese and wild-type type, 11 showed correction or partial correction with short-term leptin treatment. The number of genes with corrected expression is a small fraction of the total and likely reflects several inadequacies of short-term treatment in mice with long-term leptin deficiency. Specifically, long-term deficiency is likely to have developmental consequences that are not correctable with 1 week of leptin treatment. In addition, once-daily intraperitoneal injections do not mimic the complex responses of ambient leptin to feeding and circadian rhythms (27,28). Finally, the very stringent criteria that we set for identifying differentially expressed genes likely will overlook genes whose expression is only partially corrected by leptin.

Taken together, the expression data from all experiments can be used to identify seven groups of genes whose hepatic expression is modulated directly or indirectly by leptin (Table 3). The largest group (D) contains 146 genes that are only differentially expressed between livers of Lep^ob/Lep^ob and wild-type mice. These are genes whose expression is altered by long-term leptin deficiency and not restored by short-term treatment or caloric restriction. The group with the fewest regulated genes contains three transcripts that are specifically regulated by short-term effects of leptin. In the absence of leptin and independent of energy balance, these genes are downregulated. (Hence, in comparison with leptin-treated or wild-type mice, leptin-deficient obese, mice whether gaining weight (fed ad libitum) or losing weight (pair-fed), express lower levels of these three genes.) For two of the leptin-dependent transcripts, there is little functional information, only sequence similarity data. One is similar to the glycolytic enzyme glyceraldehyde 3-phosphate dehydrogenase, and the other is highly similar to an insulin-regulated transcript. The third leptin-dependent transcript encodes Apo A-IV, a gene implicated in cholesterol trafficking, feeding behavior, and resistance to atherosclerosis.

Another small group of genes (B) is altered in the livers of Lep^ob/Lep^ob mice and partially corrected by leptin treatment and pair feeding. Hence, the dysregulation of these genes is reversible by decreased food intake or negative energy balance and is not dependent on leptin per se. They likely are part of pathways that mediate signals and responses to negative energy balance, e.g., increased fatty acid metabolism, and are not likely to mediate the increased metabolic efficiency, insulin resistance, or atherosclerosis resistance of the obese phenotype.

Although hyperphagic and obese, leptin-deficient Lep^ob/Lep^ob mice are postulated to exist in a state of perceived starvation (29). Thus, with free access to food, they behave as though they have been calorie-restricted, eating voraciously. However, their hyperphagic response can be enhanced further with fasting (30). This suggests that there is a set of nutrient-sensing pathways that are activated further in the fasted or calorie-restricted obese mice (31). The set of genes (C) that increased in expression in obese and pair-fed obese mice, which are not correctable with short-term leptin treatment, may represent genes that are part of the augmentable hyperphagic response of Lep^ob/Lep^ob mice.

The phenotype of calorie-restricted Lep^ob/Lep^ob mice has not been explored fully, but in addition to an augmented hyperphagic response, obese mice survive longer than wild-type mice under severe calorie restriction (32). The subset of genes (G)—only induced or repressed in calorie-restricted obese mice when compared with leptin-treated mice—are candidates for genes in metabolic and signaling pathways that mediate the increased survival phenotype of obese mice. Finally, the direct comparison of expression profiles between mice that gained (obese vehicle) and lost weight (obese leptin) detected 18 genes with consistently altered expression. The expression of these genes (F) is independent of leptin and inversely regulated in negative and positive energy balance states.

The emergence of new technologies to analyze variations of sequence and expression of genes provides the tools to identify the complex interactions that underlie the physiology of both normal and pathologic states. In the studies reported here, spotted cDNA microarrays were produced and used to monitor the gene expression patterns from wild-type, leptin-deficient Lep^ob/Lep^ob mice, and Lep^ob/Lep^ob mice that were treated with leptin and calorie restriction. Differences in expression were measured with high specificity (>99.9%). The assessment of sensitivity requires a standard that identifies all true positives and all false negatives. Unfortunately, there is no simple way to identify these data with an expression array. However, Pollack et al. (33) provided an estimate of spotted array sensitivity. Using single spotted cDNA microarrays probed with labeled genomic DNA, they detected differences in copy number with a sensitivity of 55% when a threshold ratio of 1.5 was used. That is, genes for which there was a known two-fold difference in copy number were detected 55% of the time when spots with normalized fluorescence ratios of 1.5 or greater were included. Given these data, our identification of differentially regulated hepatic genes probably is conservative and the sensitivity of our arrays likely is much lower than our specificity. The sensitivity of these experiments could have been increased by using a mean threshold ratio rather than stipulating a minimum consistent change across all microarrays. As is often the case, increasing the sensitivity would have required decreasing the specificity. In the context of the studies reported here, smaller data sets with fewer false-positive data points were much more useful than larger data sets with larger numbers of false-positive data points.

Leptin plays a central role in the regulation of feeding behavior, basal metabolism, glucose utilization, insulin sensitivity, and energy expenditure and thus overall in the regulation of energy homeostasis. In addition, leptin can modulate fertility, lipoprotein metabolism, bone turnover, and immune responses (34). Using microarrays, we identified more than 160 genes whose expression is altered in the livers of leptin-deficient Lep^ob/Lep^ob mice. By further comparing expression among leptin-treated and calorie-restricted Lep^ob/Lep^ob mice, we correlated the expression of subsets of genes with known phenotypic responses to pharmacologic and dietary interventions. The goal of these studies was to identify candidate genes that mediate specific metabolic subphenotypes of leptin deficiency.

Three genes specifically modulated by leptin (group A) were identified. In obese mice, these genes are upregulated and are corrected by short-term leptin treatment. In comparison with leptin treatment, however, the expression of these genes was not altered by calorie restriction. This result underscores the observation that leptin’s peripheral effects are not mediated solely by leptin’s ability to modulate feeding behavior. Indeed, most of leptin’s pleiotropic effects, including those that regulate insulin resistance, basal energy metabolism, fertility, lipoprotein metabolism, and immune responses, must be mediated, at least in part, by pathways that are independent of food intake and energy balance. Of the genes that we identified that are dependent on leptin but not on nutrient intake, two have not yet been characterized but do share sequence similarity to other genes. The third is an apolipoprotein.

Normally, Apo A-IV is secreted primarily into the portal circulation from the small intestine and rises in response to high-fat meals (35). A lesser amount is produced by the liver. Apo A-IV has been postulated to modulate feeding behavior negatively (36), regulate HDL metabolism (37), and impart atherosclerosis resistance (38,39). Our microarray profiles detected a 10-fold increase in hepatic expression of Apo A-IV in Lep^ob/Lep^ob mice. This increase is partially normalized after a short course of daily leptin injections. C57BL/6J mice that overexpress Apo A-IV in the liver have elevated serum HDL and are resistant to developing atherosclerosis, even when fed an atherogenic diet or when made genetically deficient in Apo E (38,39). Conversely, mice homozygous for a null mutation in the Apo A-IV gene have lower serum HDL (37). These data make Apo A-IV an excellent candidate gene for mediating the elevated HDL and atherosclerosis-resistant phenotypes of obese mice. The role of Apo A-IV in mediating these phenotypes can be tested directly. If, in obese mice, Apo A-IV is necessary for the elevated HDL levels and resistance to atherogenesis, then mice that are homozygous for null mutations in leptin and Apo A-IV, <Lep^ob/Lep^ob>;<ApoA4⁻/ApoA4⁻>, will be more susceptible to atherosclerosis than the Lep^ob/Lep^ob mice. In addition, mice that carry the lethal yellow agouti allele are resistant to both exogenous leptin and atherogenesis (23), and we would predict that levels of liver expression of Apo A-IV will be elevated in <A^y/a⁺> animals.

Another of the leptin-regulated transcripts (Unigene cluster Mm.27136) is highly similar (94% nucleotide sequence identity) to a human gene, insulin-induced protein 2 (IIP2). IIP2 is expressed in regenerating livers (40) and is upregulated in hepatocytes treated with insulin (41). These studies show that in hyperinsulinemic obese mice, the IIP2-like transcript is upregulated approximately four-fold. Leptin treatment, which lowers the expression of this gene, also normalizes insulin sensitivity, resulting in decreases in both serum glucose and insulin concentrations. Calorie restriction of obese mice can improve hyperglycemia and modestly reduce insulin concentration, but such restriction does not reverse the underlying insulin resistance and does not normalize IIP2-like expression. Shimomura et al. (42) recently argued that in leptin-deficient states, insulin is unable to downregulate gluconeogenesis and glycogenolysis appropriately, whereas its ability to stimulate lipogenesis remains intact. Specifically, they suggest that within the liver of leptin-deficient mice, insulin signaling diverges along two paths, one that remains insulin-sensitive and stimulates excess fatty acid production and one that is resistant to insulin downregulation and leads to excessive glucose synthesis. If correct, the IIP2-like transcript is a candidate for mediating the insulin-sensitive responses in leptin-deficient mice. An initial test of this hypothesis would measure the transcriptional response of the IIP2-like gene in obese mice treated with insulin. If the IIP2-like gene is part of a pathway that remains insulin-sensitive in the leptin-deficient mice, then its expression should increase further in tandem with the transcription factor SREBP1-c, which remains insulin-responsive in obese mice.

In conclusion, the complex pathways that regulate short- and long-term energy homeostasis remain poorly defined. By defining transcriptional perturbations that occur in the livers of Lep^ob/Lep^ob mice, we identified pathways and candidate genes whose hepatic function is altered by leptin deficiency. Using microarrays, we demonstrated upregulation of genes required for lipid metabolism and identified candidate genes that may mediate the metabolic subphenotypes of resistance to atherosclerosis and insulin-responsive fatty acid overproduction. These results suggest that such studies could be expanded fruitfully to include a larger set of genes, to interrogate other energy-regulating tissues (hypothalami, adipose tissue, and skeletal muscle), and to include other murine models of obesity (Ay/a⁺, Cpe^fat/Cpe^fat, diet-induced). Through performance of more comprehensive expression analyses, additional candidate genes that contribute to obesity-related phenotypes and participate in regulating feeding behavior, energy partitioning, and energy expenditure could be identified.

Supported in part by National Institutes of Health Grants R01-DK-52431 and T32-DK-07559, an American Diabetes Association Training Grant, the Lucille P. Markey Foundation, and the Columbia Innovation Enterprise.