Obesity and its comorbidities, particularly type 2 diabetes, have become serious public health problems over the past few decades. Although the current pandemic is largely caused by societal environmental changes in diet, variation in response to these changes have, in part, a genetic basis. Here we address the genetic basis for both obesity- and diabetes-related traits themselves and dietary fat responses for these traits in a set of recombinant inbred mouse strains formed from the cross of LG/J with SM/J (LGXSM lines) fed a standard low-fat (15% calories from fat) or high-fat (42% calories from fat) diet. We found substantial genetic variation for most of the traits studied. Weight at time of death, liver weight, and weight of the reproductive fat pad had especially high heritabilities, whereas heart weight and serum levels of free fatty acids and triglycerides had low heritabilities. Genetic correlations were very high among fat pad weights and serum leptin, indicating shared genetic variation between fat levels and hormonal appetite control. These obesity traits were moderately correlated with adult growth, liver weight, and serum insulin and cholesterol levels. A majority of traits also displayed genetic variation in response to a high-fat diet, especially the weight of the reproductive and renal fat pads as well as the liver. Genetic correlations in dietary response followed a pattern similar to that found for the traits themselves. Several strains manifested discordant responses for obesity, glucose, and insulin, consistent with the presence of genotypes protective for diabetes in the presence of obesity. These recombinant inbred strains represent potentially valuable new models for dissecting the complex physiological relationships among obesity and diabetes.
Obesity is an escalating threat to human health worldwide. Its prevalence has increased throughout the past century so that by 2001, nearly one-fifth of U.S. adults were classified as clinically obese (1). Obesity is also associated with other clinical problems, including cardiovascular disease, high blood pressure, and type 2 diabetes (2). Both obesity and diabetes are complex diseases, with variation caused by many interacting genes of individually small effect, environmental factors, and interactions between genes and aspects of the environment, such as diet. Although many individual alleles of large effect have been identified as causing obesity in both human populations and animal models (3), they account for very little of the observed variation in obesity and diabetes among the general population because these variants are quite rare.
Evolutionary genetic changes are not implicated as a source of the recent secular trend for increased obesity (4). Instead, the increased prevalence of obesity is linked to changes in environmental factors such as diet and activity levels over the past few decades. However, there is still substantial genetic variation in individual response to changing environments. Some individuals respond to an increase in dietary fat by becoming obese, whereas others do not, remaining relatively lean (5).
Because of the complexity underlying variation in obesity and diabetes in human populations, it can be difficult to systematically study the causes of these traits in human populations (5,6). Rodent models represent an important supplement to human studies because both genetics and environment may be more closely controlled and monitored (7). The use of previously characterized inbred lines as parental strains, coupled with the precise environmental control that is possible with laboratory rodents, makes them amenable to the rigorous statistical analysis necessary for the dissection of complex traits and their environmental interactions.
Although many substantial advances have been gained from studying single gene mutations of major effect in mice (3), early hopes for the molecular characterization of human obesity through analysis of major mouse mutations and transgenics have faded (8), and in fact no novel single gene has been reported recently in either human or murine model systems (3). It is therefore increasingly important to study mouse models that correspond more closely to human populations in that their genetic architecture is caused by many interacting pleiotropic genes of relatively small effect. We have studied the genetic architecture of body size and obesity in the cross of LG/J and SM/J inbred mouse strains by mapping quantitative trait loci (QTL). We found that body size and composition variation in the F2 intercross between these lines has a genetic architecture like that found in human populations in that variation is caused by many interacting genes of individually small effect (9–13). In addition, our other studies of LG/J and SM/J mice have shown that the strains differ in their response to increased levels of dietary fat, with SM/J mice responding more strongly to a high-fat diet than the LG/J strain (12,13). Genetic variation in response to a high-fat diet was confirmed by quantitative genetic analyses of the F16 generation of an advanced intercross line (14) formed from the F2 intercross between LG/J and SM/J (T.H.E.,J.P.K., L.S.P., J.M.C., unpublished observations). Traits showing significant genetic variation in dietary response in the advanced intercross line include liver weight, adult growth (between 10 and 20 weeks), and fat pad weight.
Here, we report genetic variation in a series of obesity- and diabetes-related traits in a set of recombinant inbred strains (16) formed by close inbreeding from the F2 generation of the LG/J by SM/J intercross (T. Hrbek, L.S.P., R. Alves de Brito, J.M.C., unpublished observations). Each strain represents a unique recombination of the parental genome. The strains have been segregated for both genetic markers and obesity- and diabetes-related phenotypes. In addition to considering genetic variation in the traits themselves, here we consider genetic variation among these strains in their response to a high-fat diet. We expect to find strains that respond strongly to a high-fat diet and others that are not responsive to this environmental change. If our expectations are fulfilled, this set of strains, or individual strains from the set, can serve as mouse models for diet-induced obesity and diabetes that mimic the genetic architecture of these features in human populations.
RESEARCH DESIGN AND METHODS
The LGXSM recombinant inbred line set was produced from the F2 intercross of LG/J females with SM/J males obtained from The Jackson Laboratories. Details of the line formation and history are presented in the report by Hrbek et al. (T. Hrbek, L.S.P., R. Alves de Brito, J.M.C., unpublished observations). The genetic characteristics of individual strains scored at 506 microsatellite markers are available online from the authors at http://thalamus.wustl.edu/cheverudlab. The parental strains were the outcome of separate selection experiments for large (LG/J) (18) and small (SM/J) (19) adult body weight, respectively. Chai (20) analyzed the genetics of body size in the cross of these strains and found that differences in adult size were caused by many genes of relatively small effect. Because of their history, we felt this cross would produce a genetic architecture suitable for a model of complex human diseases such as obesity and diabetes. QTL gene mapping for body size, growth, and morphology have detected a large number of chromosomal regions involved in effects on these complex traits (9–13,21).
We include data on 16 of the recombinant inbred lines in this report. Data from the parental strains was previously reported (13) and are listed in Table A1 in the online appendix (available at http://diabetes.diabetesjournals.org). Care should be taken when comparing the data from the parental strains to those of the recombinant inbred strains because of differences in housing conditions and procedures at the times these data were collected. The target design was to rear eight males and eight females of each strain on standard low- and high-fat diets. Sample sizes at individual levels of this design vary because of differential success in producing experimental animals from our breeding colony (see Table A1 in the online appendix) but average approximately eight animals per sex, diet, and strain. Experimental animals remained with their dam, who was fed standard mouse chow (PicoLab Rodent Chow 20, no. 5053), until weaning at 3 weeks of age. At this time litters were divided, and animals were randomly placed into single-sex and dietary treatment groups. Multiple animals were housed in a single cage, with no more than five animals per cage. Animals were placed on either the high- or low-fat diet continuously from 3 to 20 weeks. The high-fat diet (Harlan Teklad catalog no. TD88137) and low-fat diet (Research Diets catalog no. D12284) compositions are presented in Table 1. All animals were fed ad libitum, and the animal facility was maintained at a constant temperature of 21°C, with 12-h light and dark cycles. All procedures followed the guidelines for the care of laboratory animals at Washington University School of Medicine (assurance no. A-3381-01).
Measurements.
Animals were weighed weekly from 1 to 20 weeks of age, at which time they were killed and necropsied. Growth rates were calculated for three growth periods: infant growth from 1 to 3 weeks of age, subadult growth from 3 to 10 weeks, and adult growth from 10 to 20 weeks. Growth rates were transformed using the base 10 logarithm before analysis. We do not expect dietary effects on infant growth because animals were not subject to dietary treatment at that age. The subadult growth period is a period of general growth in body size. Most skeletal growth is accomplished in this period. Finally, the adult growth period begins after skeletal growth is complete, and most weight added is in soft tissue, including fat (13).
After 20 weeks of age, animals were subjected to an intraperitoneal glucose tolerance test. Animals were fasted for 4 h, after which a basal blood glucose level was obtained using a Glucometer Dex blood glucose meter (Bayer). Animals were then injected with a 10% glucose solution at 0.01 ml/g body wt i.p. Additional glucose readings were obtained 15, 30, 60, and 120 min after the initial injection. The graph of the glucose levels over the period of the test are summarized by the area under the curve (AUC), which is used as a general measure of an animal’s response to the glucose challenge. Lower values indicate a more robust response to glucose.
At a later date, animals were fasted for 4 h and anesthetized with sodium pentobarbital, and a terminal blood sample was obtained via cardiac puncture. Blood plasma was separated through centrifugation and analyzed for free fatty acids, cholesterol, triglycerides, leptin, and insulin. Internal organs (liver, spleen, heart, and kidneys) and fat pads (reproductive, renal, mesenteric, and inguinal) were removed and weighed.
Statistical analysis.
The data are analyzed first using a three-way mixed model ANOVA:
where an individual’s phenotype (Yijkl) is modeled by the overall mean (μ); the deviation due to the ith sex, the jth diet, the kth strain, and the lth individual from the specified sex-diet-strain cohort; the interactions among these factors; and the residual (eijkl). Sex and diet are treated as fixed factors, whereas strain is a random factor (22).
Variance due to strain (σ2str) represents the total genetic variance across these strains for the traits themselves, including the additive, dominance, and epistatic genetic variance components (23). This variance can be considered as pooled across the four sex and diet cohorts. It is calculated as
where MS indicates mean square, subscript r refers to residual, and n is the number of animals in each sex-diet-strain class. Because sample sizes vary slightly from class to class, we apply the adjustment recommended by Sokal and Rohlf (22). The broad-sense heritability (H2) (23) for the traits is estimated as
where σ2r (= MSr) is the residual variance.
The interaction variance components in this model have a more complex and interesting interpretation. The variance caused by the sex-by-strain interaction (σ2sxstr) can be interpreted as the total genetic variance in sexual dimorphism among these strains:
If this variance is statistically significant, it indicates that sex differences are greater in some strains than in others and therefore can be used to obtain a broad-sense heritability for sexual dimorphism pooled across the diets:
The most interesting of these variance components in the present context is the diet-by-strain interaction variance
This variance measures total genetic variance in dietary response for the traits. Some strains will respond strongly to the high-fat diet and others not at all. These strain differences are genetic in origin. Thus, we can consider the crucial question of the heritability of dietary response, pooled across the sexes, measured as
Finally, we can consider the variance due to the three-way interaction between sex, diet, and strain:
This variance component can be interpreted as representing the total genetic variance in the sexual dimorphism of dietary response. It is significant when there are detectable differences in the genetic basis for dietary response in males and females. Its heritability is given by
When this heritability is significant, it may indicate that dietary response is significantly heritable in one sex but not the other. To determine the cause of these interaction variances, it is necessary to examine the heritability of traits and dietary response within the individual sample stratifications by making separate calculations for each sex using the following model:
Genetic correlations between traits and between their dietary responses will be calculated from the strain-specific means and dietary differences in strain-specific means, respectively, using Pearson product moment correlations. These genetic correlations indicate the extent to which various traits or dietary responses are affected by the same pleiotropic genes or by different genes in linkage disequilibrium among the strains. Genetic correlations between the same phenotype expressed in the two dietary environments will also be calculated (23). These correlations are 1.0 when the genetic basis for trait variation is the same in both diets and decreases to 0.0 when the genetic basis for trait variation is random across environments.
RESULTS
Effects of sex and diet.
Sex-, diet-, and strain-specific mean values; standard deviations; and sample sizes are presented for each trait in the Table A1 in the online appendix. This information allows the identification of specific recombinant inbred strains that are relative outliers for obesity- and diabetes-related traits and allows genetic relationships to be established between these traits and others measured in the future on this recombinant inbred strain set. The three-way, mixed-model ANOVA was significant for every trait. The high-fat diet increased trait value for nearly all traits, with the exceptions of infant growth and adult tail length.
Most traits are significantly different between the sexes. Exceptions include infant growth; renal, mesenteric, and inguinal fat pad weights; and serum levels of free fatty acids, triglycerides, and leptin. Male values are higher for all traits except adult growth and reproductive fat pad weight. There are many strong interactions between sex and the other factors, as indicated in Table A2 in the online appendix. Significant sex-by-diet interactions were observed for adult growth (P = 1.45 × 10−4), weight at necropsy (P = 0.031), kidney weight (P = 1.74 × 10−4), and reproductive fat pad weight (P = 6.68 × 10−5). In each case, females responded more strongly to the high-fat diet than males. Nearly all traits showed a significant sex-by-strain interaction, indicating substantial genetic variation for sexual dimorphism in the LG/J–by–SM/J cross, including adult growth (P = 1.0 × 10−8); basal glucose (P = 2.1 × 10−4); area under the glucose tolerance curve (P = 5.5 × 10−5); body weight at necropsy (P = 1.4 × 10−11); heart (P = 0.029) and liver (P = 9.5 × 10−6) weight; tail length (P = 0.006); the reproductive (P = 1.3 × 10−11), renal (P = 1.8 × 10−8), mesenteric (P = 2.6 × 10−5), and inguinal (P = 9.0 × 10−9) fat pad weights; and serum insulin (P = 1.5 × 10−8), cholesterol (P = 1.0 × 10−9), free fatty acid (P = 8.7 × 10−5), and leptin (P = 1.4 × 10−11) levels. The only traits lacking genetic variation for sex differences are infant growth, subadult growth, kidney weight, spleen weight, and triglyceride level. Further analysis treats the sexes separately because of the ubiquity of sex interactions for these traits.
Trait heritability and genetic correlation.
All traits show significant genetic variation pooled across sexes and within each sex among the LGXSM recombinant inbred strains (see Table A2 in the online appendix), including infant (P = 3.8 × 10−4), subadult (P = 1.0 × 10−11), and adult (P = 0.006) growth; basal glucose level (P = 1.5 × 10−6) and AUC (P = 0.009) at 20 weeks; body weight at necropsy (P = 2.0 × 10−7); tail length (P = 9.0 × 10−12); heart (P = 1.3 × 10−11), kidney (P = 9.5 × 10−12), spleen (P = 8.1 × 10−12), and liver (P = 7.1 × 10−12) weights; the weights of the reproductive (P = 8.6 × 10−12), renal (P = 9.7 × 10−12), mesenteric (P = 1.1 × 10−11), and inguinal (P = 1.0 × 10−11) fat pads; and serum insulin (P = 1.1 × 10−11), cholesterol (P = 8.1 × 10−12), leptin (P = 1.1 × 10−11), free fatty acid (P = 8.2 × 10−5), and triglyceride (P = 3.9 × 10−5) levels. Multivariate analysis confirms these results over all traits jointly, with a probability of obtaining the observed results given no strain effect of <1.0 × 10−16. Although all heritabilities are significant, they vary substantially in magnitude, from a low of 5% to a high of 60%. Infant growth and triglyceride levels have very low heritabilities, whereas weight at time of necropsy, liver weight, and reproductive fat pad weight have heritabilities >50%. Some traits differ in heritability between the sexes, particularly subadult growth (51% ♀, 14% ♂), reproductive fat pad weight (59% ♀, 39% ♂), and serum insulin levels (55% ♀, 19% ♂).
Genetic correlations between traits are provided in Table A3 in the online appendix. Correlations >0.50 are significant at the 5% level. Some features vary independently of most other traits, including infant and subadult growth; tail length; heart, kidney, and spleen weight; and serum free fatty acid and triglyceride levels. All of the remaining traits are genetically correlated, centering on the core measures of obesity. The four fat pads are all very highly correlated, with pairwise correlations ranging from 0.7 to 0.9. Thus, they have 50–80% of their genetic effects in common. Genetic correlations between the collective weight of the fat pads and the individual pads are all >0.90. Because of these high correlations, fat pads will be combined using the sum of their weights in some analyses below. The collective fat pads are very highly correlated with necropsy weight (r = 0.91) and serum leptin levels (r = 0.87) and moderately correlated with adult growth (r = 0.66), insulin (r = 0.54), and cholesterol (r = 0.60) levels. However, the fat pad weights are not significantly correlated with either fasting glucose or AUC at 20 weeks. Basal glucose levels and AUC at 20 weeks are strongly correlated with each other and have moderate correlations with liver weight as well as leptin and insulin levels. Serum insulin, cholesterol, and leptin levels are moderately intercorrelated with each other. Liver weight is correlated with a wide variety of factors, including moderate correlations with necropsy weight, the fat depots, and cholesterol and leptin levels, and it has a high correlation with insulin level. Necropsy weight is also correlated with most traits, being moderately correlated with liver weight as well as insulin and cholesterol levels, and it is highly correlated with fat depot weights and leptin levels.
Dietary response heritability and genetic correlation.
Dietary response heritabilities for these traits are provided in Table A2 in the online appendix, along with the genetic correlation between environments. Many of the traits show significant heritability for dietary response, including adult growth (P = 0.006), body weight at necropsy (P = 2.0 × 10−7), kidney weight in males (P = 0.020), spleen (P = 1.2 × 10−4) and liver weights (P = 9.2 × 10−12), reproductive (P = 1.2 × 10−11) and renal (P = 1.4 × 10−11) fat pad weights, mesenteric (P = 0.017) fat pad weight in males and inguinal (P = 1.8 × 10−4) fat pad weight in females, serum cholesterol (P = 1.4 × 10−6) and leptin (P = 5.6 × 10−6) levels, and serum insulin (P = 0.043) and fasting glucose (P = 1.2 × 10−5) levels, and AUC (P = 2.5 × 10−4) in males. Traits for which there is no significant genetic variation in dietary response include infant and subadult growth, tail length, heart weight, and serum free fatty acid and triglyceride levels. Significant dietary response heritabilities range from 10 to 60% and are nearly always lower than their trait value counterparts. Dietary response heritability is especially high for liver weight and reproductive and renal fat pad weights. Most other traits show low-to-moderate heritability for dietary response. There are some striking differences in dietary response heritability between the sexes. Dietary response heritability tends to be much higher in females for dietary obesity (dietary response of fat pads) and much higher in males for dietary hyperglycemia (dietary response for basal glucose level and AUC), suggesting discordance between the sexes in the relationship between dietary obesity and dietary hyperglycemia.
The genetic basis for differences in dietary response can also be evaluated using the genetic correlation between expressions of the same trait in contrasting dietary environments. Genetic correlations across environments vary from 0.25 to ∼0.80 for traits with significant genetic variation in dietary response. Low cross-diet correlations were observed for glucose and insulin levels in males, whereas moderate correlations were observed for the spleen, liver, various fat pads, and cholesterol and leptin levels.
Genetic correlations between dietary responses are presented in Table A4 in the online appendix for traits that displayed significant dietary response heritability. The dietary obesity traits (fat pad weights, leptin levels, and necropsy weight) have very highly intercorrelated dietary responses. These responses are also moderately correlated with dietary responses for adult growth, liver and kidney weights, and insulin and cholesterol levels. Although dietary responses for basal glucose levels and AUC are highly intercorrelated, dietary hyperglycemia is largely independent from dietary obesity. Finally, dietary responses for liver weight and insulin and cholesterol levels are moderately correlated.
Strain characteristics.
Although the results presented above describe the collective qualities of the LGXSM recombinant inbred strain set as a model for the study of complex obesity- and diabetes-related traits, it is also instructive to look directly at the characteristics of individual strains that may be useful because of their particular constellation of characteristics.
Figure 1 presents a plot of average basal glucose values and insulin levels for the strains, separated by sex and diet. Different positions on this graph can be used to classify progression toward a diabetic state through a series of stages. Baseline glucose and insulin levels (stage 0) are represented by the box in the lower-left portion of the graphs, with fasting glucose level <150 mg/dl and insulin level <5 ng/ml. These specific glucose and insulin thresholds are proposed for purposes of illustration with this population of strains and should not be considered absolute because there are no undisputed norms for these traits in mice. Progression toward diabetes can be related to positions on the glucose-insulin graph. The first pre-diabetic stage (stage 1) involves isolated hyperinsulinemia, represented by the upper-left portion of the glucose-insulin graph. During this period insulin resistance develops, and higher insulin levels are required to maintain normal glucose levels. During the next stage (stage 2), hyperglycemia also develops because insulin resistance becomes severe enough that glucose levels are not controlled even with high levels of insulin. This hyperinsulinemic and hyperglycemic state is represented by the upper-right portion of the glucose-insulin graph. Finally, full-blown diabetes, hyperglycemia combined with hypoinsulinemia (stage 3), is reached with the failure of insulin-producing β-cells in the pancreatic islets. This stage of diabetes is represented in the lower-right portion of the graph.
Inspection of Fig. 1 indicates that various recombinant inbred strains are located in different segments of the glucose-insulin graph. Females on a low-fat diet (see Fig. 1B) are nearly all in the normal range. Only one strain, LGXSM-22, has progressed to the pre-diabetic stage, with normal glucose levels and hyperinsulinemia. Most female strains raised on a high-fat diet (see Fig. 1A) are still located in the lower-left portion of the graph, although LGXSM-31 has joined LGXSM-22 in the normal glucose–hyperinsulinemia region. Females from both of these strains show a strong response of insulin to dietary fat.
Levels of both glucose and insulin are higher in males than in females. Hence, whereas males from most strains fed a low-fat diet (see Fig. 1D) are located in the lower-left segment of the graph, several strains are located in the normal glucose–hyperinsulinemia region, including males from strains LGXSM-19, -20, -23, and -35. One strain (LGXSM-33) displays hyperinsulinemia and hyperglycemia, indicating pre-diabetic and early-diabetic states, even on a low-fat diet. Males from most strains fed a high-fat diet are located outside the normal region of the glucose-insulin graph. LGXSM-4, -15, -18, -19, -23, -31, -45, and -46 lie just outside the normal range, showing a moderate hyperinsulinemic response to dietary fat, whereas LGXSM-48 has near-normal glucose levels with strong hyperinsulinemia. LGXSM-33 remains in the hyperinsulinemic and hyperglycemic state, not responding substantially to dietary fat. LGXSM-20, -22, and -35 join LGXSM-33 as hyperinsulinemic and hyperglycemic when fed a high-fat diet. None of the strains reached a fully diabetic, hyperglycemic and hypoinsulinemic, state during our 20-week experiment, although insulin levels fell with high-fat feeding in LGXSM-33. Glucose tolerance curves for each strain separated by sex and diet are presented in Fig. 2.
The relationship between obesity and diabetes among strains can be evaluated by examining the level of obesity (sum of the four fat pad weights) and the diabetes stage of specific strains. Females exhibit limited variation in diabetes stage because only stages 0 and 1 are present even on the high-fat diet. Females from strains in stage 1 are more obese than normal strains, including the hyperobese strain 31. The relationship of obesity to diabetes stage in males is illustrated in Fig. 3. The range of obesity does not differ between the diabetes stages for males fed a low-fat diet, although, on average, strains at stages 1 and 2 have heavier fat pads than those in the normal range. This pattern results in normal and diabetic strains with the same level of obesity. For example, on a low-fat diet, normal strains LGXSM-2, -18, -31, -46, and -48 have the same level of obesity as stage 1 and 2 strains LGXSM-19, -23, -33, and -35. High-fat-fed males display an interesting pattern of relationship between obesity and diabetes in that stage 1 strains are, on average, more obese than stage 0 strains, but stage 2 strains have lower levels of obesity than stage 1 strains. Obesity declines as diabetes progresses from isolated hyperinsulinemia to combined hyperinsulinemia and hyperglycemia. Even under this general trend, there are strains at different diabetic stages with the same level of obesity (LGXSM-5 and -10 at stage 0; LGXSM-4, -15, -18, -23, and -45 at stage 1; and LGXSM-22, -33, and -35 at stage 2), illustrating the dissociability of obesity from diabetes in these strains.
The relative characteristics of individual strains can be examined using the data in Table A1 in the online appendix. A few strains are nearly always either positively or negatively extreme over obesity- and diabetes-related traits (basal glucose level, insulin level, cholesterol level, total weight of fat pads, liver weight, and leptin level), such as the positively deviant strains LGXSM-22 and -19 and the negatively deviant strains LGXSM-18, -38, and -45. Many strains are intermediate for most traits, including LGXSM-4, -5, -15, -23, -31, -33, -35, and -46. Perhaps most interesting are strains that show discordant phenotypes or phenotype distributions that run against the genetic correlation observed between traits. Of the various strains, ∼30% have discordant values for some trait pair in at least one sex on one diet. For example, male LGXSM-46 strains fed a high-fat diet have a relatively low basal glucose level but high levels of obesity, leptin, and cholesterol.
Table A1 in the online appendix can also be used to detect outliers for the dietary response of glucose, insulin, obesity, leptin and cholesterol levels, and liver weight. LGXSM-22, -31, and -48 are strongly responsive to dietary fat for a number of traits, whereas LGXSM-18, -20, and -38 tend to be generally unresponsive, and LGXSM-4, -5, -10, -15, -18, -35, and -45 show overall moderate responsiveness. Nearly one-half of the sex-specific values of the strains show discordant phenotypes for some combination of dietary responses. The physiological bases of obesity- and diabetes-related traits may be especially interesting in these strains. For example, LGXSM-48 is strongly responsive for glucose levels but is unresponsive for insulin levels.
DISCUSSION
We discovered heritability for all obesity- and diabetes-related traits analyzed in the LGXSM recombinant inbred lines, supporting this set of strains as a model for these complex human diseases. Liver weight and reproductive fat pad weight in particular were strongly heritable in this cross. Other traits, such as infant growth and free fatty acid and triglyceride levels, were only slightly heritable, indicating that these strains are not genetically variable for these traits, although they could be heritable in other mouse strain crosses.
Genetic correlations among traits indicate the extent to which traits are coordinately affected by genes. A large number of traits did not show strong correlations with the others, including infant and subadult growth, tail length, heart and kidney weight, and free fatty acid and triglyceride levels. These traits are likely affected by different sets of genes distinct from those affecting obesity- and diabetes-related traits. The remaining traits are all intercorrelated to varying degrees. The central traits in the network of intercorrelations are the fat depot weights. They are all very highly correlated with each other in these strains, indicating that genes that affect one fat pad probably affect the other fat pads as well. The sum of the fat depot weights can be used to summarize nearly all of the genetic variation in fat depot weights because each individual fat depot shares between 80 and 92% of its genetic variation with the sum of the fat pad weights. There is correspondingly little genetic variation in fat distribution among the strains.
Fat depot weights are genetically correlated with both necropsy weight and adult growth between 10 and 20 weeks. The correlation of fat depot weights, necropsy weight, and adult growth shows that much of the variation in final body and fat depot size is caused by fat added during the 10- to 20-week period, after skeletal growth is largely complete. Fat depot weights are also very highly correlated with leptin levels. This correlation represents the direct physiological mechanism through which adipocytes produce leptin (24). Fat depot weights are also correlated with cholesterol and insulin levels. Liver weight seems to be in an intermediate position in the network of correlations in that it is most strongly correlated with insulin levels and is the only trait other than insulin and leptin level that is genetically correlated with basal glucose levels and AUC. There is a general lack of correlation between glucose levels and response and obesity, indicating that different sets of genes are likely to be affecting these traits in the LGXSM strains.
Ehrich et al. (13) showed that caloric intake from the high- and low-fat diets were similar across the parental strains when fed ad libitum, so that the same amount of energy was available from the two experimental diets. This indicates that the diet-based differences found in this study are caused by differences in dietary composition rather than differences of energy intake.
Tail length was included in the analysis as a representative skeletal size trait, in contrast to the fat pad weights, which represent obesity. Although tail length is heritable, it is one of only two traits not to show a dietary effect. The lack of dietary effect on tail length is an indication that increased dietary fat itself does not influence skeletal growth, but has its body composition effects restricted to soft tissues such as organ size and fat pad weights.
Strains vary considerably in skeletal size, but the genetic correlation between tail length and body, organ, and fat depot sizes is low (r = 0.42). If we control for body weight, the partial correlation between tail length and fat depot sizes is reduced to zero, as in the F3 generation of this intercross (25). Skeletal size and obesity are genetically independent in this cross, variation being affected by different sets of genes (10). However, the genetic correlation between body size and the total weight of the fat pads is very high (r = 0.94). Although variation in final body weight among these mice is primarily attributable to variation in obesity, and this variation is independent of skeletal size, because of their close part-whole relationship, it would be inappropriate to study fat pad size after controlling for body weight in this population.
In addition to genetic effects on the traits themselves, we found moderate levels of genetic variation in dietary response of the traits. Heritabilities of dietary response are nearly always lower than heritabilities of the traits themselves, but still substantial. These results indicate that some genotypes respond to a high-fat diet by becoming obese, hyperleptinemic, hyperglycemic, and hyperinsulinemic, whereas other genotypes are not responsive. Unresponsive genotypes remain lean and maintain normal leptin, glucose, and insulin levels despite the high-fat diet. This is the kind of genetic variation that is most important for understanding and treating the developing obesity pandemic in human populations.
Genetic correlations between dietary responses follow a similar pattern as described above for the traits themselves. The various fat depots, leptin levels, and body weight responses to the high-fat diet are highly intercorrelated with each other, again forming a core of closely related responses affected by nearly the same set of genes. Fat depot response is also correlated with responses in adult growth, leptin, cholesterol and insulin levels, and liver weights. Glucose responses to the high-fat diet are largely independent of these traits but are moderately correlated with the responses of liver weight and insulin and leptin levels. Thus, it appears that variation in the dietary responses of the fat pads and of glucose levels are caused by different genetic factors.
Genetic correlations between the same traits expressed in the high- and low-fat dietary environments provide an indication of whether the genetic basis for trait variation is different in the two environments. A low genetic correlation indicates that either different genes are involved in variation in the different environments or that the same genes have opposite effects in those environments. Overall, genetic correlations between environments were moderate, averaging 0.5 in females and 0.6 in males, indicating that ∼30% of the variance is shared across environments. This result suggests that some genes are responsible for genetic effects in both environments but that there should also be diet-specific gene effects.
Our results indicate that there is sufficient genetic variation among the LGXSM recombinant inbred lines to map QTL both for the traits themselves and for dietary responses to a high-fat environment. This set of strains can serve as a model both for obesity- and diabetes-related traits and for their responses to excess levels of dietary fat, mimicking the kinds of genetic effects responsible for human variation in obesity and diabetes. Interestingly, there is a relative lack of genetic association between obesity and diabetes in this cross.
There was an interesting relationship between obesity and progress through the stages of diabetes. In both males and females, strains in stage 1, showing isolated hyperinsulinemia, were more obese than strains with normal glucose and insulin levels (stage 0), although the ranges of obesity did overlap between strains in the different stages. However, males in strains that have reached stage 2, hyperinsulinemia and hyperglycemia, have lighter fat pads than those in stage 1. They also show slower, or no, growth in body weight over the 10- to 20-week period. It is quite possible that these animals lost fat and body weight after entering stage 2 in the diabetes progression because of resistance to incorporating serum glucose into tissues.
The diabetic status of the strains was examined using relative glucose and insulin levels. On a low-fat diet, females showed little evidence for diabetic or pre-diabetic phenotypes. Only a single strain expressed hyperinsulinemia (LGXSM-22), and it maintained normal glucose levels. Although both insulin and glucose levels increased when females were fed a high-fat diet, most strains remained in or near the normal range. The high-fat diet resulted in LGXSM-22 becoming even more hyperinsulinemic, but it still maintained high-normal glucose levels. The high-fat diet moved LGXSM-31 into the pre-diabetic hyperinsulinemic, normal-glucose range from the normal range it inhabited on a low-fat diet. However, overall there was no significant heritability for dietary response in either glucose or insulin levels in females.
Males had higher insulin and glucose levels and provide more evidence for strains with diabetic characteristics. In human populations, type 2 diabetes is more strongly associated with obesity in male than in female subjects (2), again demonstrating the relevance of this model system. They also displayed significant heritability for dietary response for these traits. Most strains are normal for both insulin and glucose levels on a low-fat diet. Four strains display mild hyperinsulinemia and normal glucose, and one strain (LGXSM-33) is both hyperinsulinemic and hyperglycemic on the low-fat diet. Both insulin and glucose levels increase in males fed a high-fat diet, and only two strains remain in the normal insulin-glucose level range (LGXSM-5 and -35). A majority of the strains show mild hyperinsulinemia (55%) combined with high-normal glucose levels. These strains can be considered to be entering a pre-diabetic state at 20 weeks. Two strains, LGXSM-35 and -48, have high-normal glucose levels combined with hyperinsulinemia and can be considered pre-diabetic. Finally, three strains (LGXSM-20, -22, and -33) are hyperinsulinemic and hyperglycemic and can be classified as entering the early stages of diabetes. Interestingly, LGXSM-33 did not respond to the high-fat diet by becoming more diabetic than it was on the low-fat diet.
No strains reached the final stage of diabetes involving hyperglycemia and hypoinsulinemia. It is not common for rodent models of diabetes to enter this final stage. It remains uncertain whether the LGXSM recombinant inbred strains contain members that could reach this final stage on a high-fat diet because the experiment was terminated at 20 weeks after 17 weeks of dietary treatment. Extended time on the treatment could lead more female strains into the pre-diabetic stage, more male strains into the early diabetic state with hyperglycemia and hyperinsulinemia, and perhaps some strains into end-stage diabetes with hyperglycemia and hypoinsulinemia.
What is most intriguing about this data are the variation in progress toward diabetes among the LGXSM recombinant inbred strains. This genetic variation allows studies of genetic factors affecting obesity and diabetes. However, these strains are also a valuable resource for testing therapeutic agents on well-defined animals with levels of diabetic progression. In addition, crossing strains discordant for obesity- and diabetes-related traits with other mice known to be at risk for diabetes represents an attractive strategy for identifying genes that decrease diabetes risk in the context of obesity.
The 20-week insulin levels (ng/ml) plotted against fasting glucose levels (mg/dl) for females fed a high-fat diet (A), females fed a low-fat diet (B), males fed a high-fat diet (C), and males fed a low-fat diet (D). Strain numbers are used as symbols. The box in the lower-left quadrant indicates the range of normal glucose and insulin levels.
The 20-week insulin levels (ng/ml) plotted against fasting glucose levels (mg/dl) for females fed a high-fat diet (A), females fed a low-fat diet (B), males fed a high-fat diet (C), and males fed a low-fat diet (D). Strain numbers are used as symbols. The box in the lower-left quadrant indicates the range of normal glucose and insulin levels.
Responses of individual strains to the intraperitoneal glucose tolerance test (IPGTT; mg/dl) for females fed a high-fat diet (A), females fed a low-fat diet (B), males fed a high-fat diet (C), and males fed a low-fat diet (D).
Responses of individual strains to the intraperitoneal glucose tolerance test (IPGTT; mg/dl) for females fed a high-fat diet (A), females fed a low-fat diet (B), males fed a high-fat diet (C), and males fed a low-fat diet (D).
The relationship between diabetic stages and levels of obesity in males fed a low- and high-fat diet. Strain numbers are indicated next to their respective symbols. Stage 0 is normal insulin and fasting glucose levels, stage 1 is isolated hyperinsulinemia, and stage 2 is hyperinsulinemia combined with hyperglycemia. No strains reached stage 3, hypoinsulinemia with hyperglycemia.
The relationship between diabetic stages and levels of obesity in males fed a low- and high-fat diet. Strain numbers are indicated next to their respective symbols. Stage 0 is normal insulin and fasting glucose levels, stage 1 is isolated hyperinsulinemia, and stage 2 is hyperinsulinemia combined with hyperglycemia. No strains reached stage 3, hypoinsulinemia with hyperglycemia.
Composition of high- and low-fat diets
| Component | High fat | Low fat |
|---|---|---|
| Energy from fat (%) | 42 | 15 |
| Casein (g/kg) | 195 | 197 |
| Sugars (g/kg) | 341 | 307 |
| Corn starch (g/kg) | 150 | 313 |
| Cellulose (g/kg) | 50 | 30 |
| Corn oil (g/kg) | — | 58 |
| Hydrogenated coconut oil (g/kg) | — | 7 |
| Anhydrous milk fat (g/kg) | 210 | — |
| Cholesterol (g/kg) | 1.5 | — |
| Kilojoules per gram | 18.95 | 16.99 |
| Component | High fat | Low fat |
|---|---|---|
| Energy from fat (%) | 42 | 15 |
| Casein (g/kg) | 195 | 197 |
| Sugars (g/kg) | 341 | 307 |
| Corn starch (g/kg) | 150 | 313 |
| Cellulose (g/kg) | 50 | 30 |
| Corn oil (g/kg) | — | 58 |
| Hydrogenated coconut oil (g/kg) | — | 7 |
| Anhydrous milk fat (g/kg) | 210 | — |
| Cholesterol (g/kg) | 1.5 | — |
| Kilojoules per gram | 18.95 | 16.99 |
Additional information for this article can be found in an online appendix at http://diabetes.diabetesjournals.org.
Article Information
This work was supported by National Institutes of Health Grants DK55736, DK52514, HL58427, and RR15116 and Washington University’s Clinical Nutrition Research Unit (DK56341) and its Diabetes Research Training Center (2 P60 DK20579).
