Genetic Dissection of Complex Traits with Chromosome Substitution Strains of Mice

See allHide authors and affiliations

Science  16 Apr 2004:
Vol. 304, Issue 5669, pp. 445-448
DOI: 10.1126/science.1093139


Chromosome substitution strains (CSSs) have been proposed as a simple and powerful way to identify quantitative trait loci (QTLs) affecting developmental, physiological, and behavioral processes. Here, we report the construction of a complete CSS panel for a vertebrate species. The CSS panel consists of 22 mouse strains, each of which carries a single chromosome substituted from a donor strain (A/J) onto a common host background (C57BL/6J). A survey of 53 traits revealed evidence for 150 QTLs affecting serum levels of sterols and amino acids, diet-induced obesity, and anxiety. These results demonstrate that CSSs greatly facilitate the detection and identification of genes that control the wide diversity of naturally occurring phenotypic variation in the A/J and C57BL/6J inbred strains.

Most traits show substantial genetic variation in natural populations and among inbred strains, reflecting the segregation of quantitative trait loci (QTLs). Genetic identification of QTLs can provide insights into molecular mechanisms of development and physiology, but these studies remain tedious and time consuming (1).

The traditional approach for QTL analysis includes two challenging steps. The first step requires arranging a large cross between at least two strains in which hundreds or thousands of progeny are assayed for relevant phenotypes and genotyped for polymorphic markers spanning the genome (2, 3). Because such crosses involve the simultaneous segregation of multiple QTLs, the resulting “phenotypic noise” limits both the power to detect individual QTLs to those with large effect and the precision to localize QTLs to large chromosomal regions. The second step involves molecular identification of the genetic variants that are responsible for each QTL. This step typically requires studying individual QTLs in isolation by performing 5 to 10 generations of backcrosses to construct congenic strains with chromosomal segments carrying alternative alleles of the QTL on an otherwise isogenic background and then interbreeding the congenic strains to carry out fine-structure genetic mapping and cloning. Detection of QTLs and their molecular identification have proven to be serious bottle-necks in studies of complex traits (1, 4, 5).

We recently proposed (6) an approach for QTL analysis that involves prior construction of a panel of chromosome substitution strains (CSSs) between a donor strain (A) and a host strain (B). Strain CSS-i carries both copies of chromosome i from the donor strain, but all other chromosomes from the host strain are intact and homozygous. A CSS panel partitions the variation between two strains and provides a permanent resource for studying the genetic control of phenotypic variation. Investigators can test individuals from each CSS for any phenotype of interest and immediately infer that a phenotypic difference between the CSS and the host strain implies that at least one QTL resides on the substituted chromosome. Fine-structure mapping of the QTL can then be performed with crosses between the CSS and the host parental strain (7, 8) or with a panel of congenic strains derived directly from the CSS (9).

Construction of CSS panels involves successive backcrosses to the host strain, in which progeny carrying a nonrecombinant copy of the desired chromosome are identified in each generation and used as parents for the next backcross to eventually produce progeny heterosomic (A/B) for the desired chromosome on an otherwise host (B/B) background (6). These progeny are then intercrossed to produce progeny homosomic (A/A) for the desired chromosome. Although the concept is straightforward, CSS construction has only become feasible with the availability of complete genetic maps that can be used to trace inheritance throughout the genome. [The exception is Drosophila melanogaster, in which special balancer chromosomes that suppress recombination can be used for chromosome substitution (10)].

We created a complete CSS panel in a vertebrate species, using the inbred mouse strains A/J and C57BL/6J as donor and host, respectively. These strains (henceforth referred to as A/J and B6, respectively) were chosen because they differ for many physiological, morphological, behavioral, immunological, and oncological traits, including many important models of birth defects and adult diseases in humans (4, 5). The program required more than 17,000 mice and took about 7 years. The methodology for selective breeding and genotyping has become more efficient over time, and we estimate that such a program could now be compressed to 4 years.

Each chromosome was followed with genetic markers along its length to ensure substitution of the entire region between the most proximal and distal markers (Fig. 1) (11). The small regions outside these flanking markers (typically ∼2 to 5 cM) could not be followed, and a fraction of the distal sequence may not have been substituted from the donor strain in some cases. (The only known anomaly in this panel involved the mislocalization of the most distal marker on chromosome 1, resulting in the failure to substitute the most distal ∼19 cM of the chromosome.) In each cross, the proportion of heterosomic backcross progeny (∼20%) and homosomic intercross progeny (∼4%) largely agreed with expectation, with the precise proportion being inversely related to chromosome length. The only notable difficulty involved the homozygosis of chromosome 5, which was overcome through using larger numbers of progeny.

Fig. 1.

Parental composition of the CSS genomes. For each chromosome, the ideogram shows the substituted chromosome in the corresponding CSS. By convention, the centromeric end of the chromosome is on the left. Black denotes regions known to have been substituted by A/J, because their inheritance has been traced with flanking genetic markers; these regions comprise ∼83% of the genome. Gray indicates regions that are beyond the farthest markers studied but that are highly likely to have been substituted by virtue of close linkage to the nearest marker; these regions comprise 16% of the genome, of which 11% is expected to be A/J. White indicates a single region in which an error in map location of a genetic marker resulted in a segment that was not substituted, corresponding to ∼1% of the genome.

The complete CSS panel consists of 22 strains, one for each of the 19 autosomes, the two sex chromosomes, and the mitochondria. The panel has been provided to the Jackson Laboratory for preservation and distribution as a research resource (12).

We then sought to demonstrate the ability of CSSs to dissect genetic factors affecting 53 complex traits, 24 in females and 29 in males, related to sterol levels, diet-induced obesity, anxiety, and amino acid levels. Phenotypic differences between the CSSs and the host B6 strain were measured by a statistical test that accounts for testing multiple hypotheses. Significance levels were chosen so that, for each trait tested, the expected number of false positives for any given trait (that is, the total number of strains in the CSS panel showing a significant result by chance) would be 0.05, and the total number of false positives expected across the 53 traits would thus be less than 3. In fact, one or more CSSs showed significance for nearly all traits studied. Overall, we found evidence for 150 QTLs, far exceeding chance expectation. Results for four representative traits are shown in fig. S1, and the full set of significant results for all traits is shown in Fig. 2.

Fig. 2.

Results for phenotype screening in the CSS panel. The upper half of each box indicates results in males, the lower half in females. Significant results are indicated in red. Dashes indicate that CSSs were not tested for a particular trait. Not shown are results for CSS-mito; this CSS gained significantly less weight than did B6 on the high-fat, high-sucrose diet, bringing to 17 the total number of CSSs that showed resistance to diet-induced obesity.

We measured levels of three sterols— cholesterol, campesterol, and sitosterol—in male (ntotal = 225) and female (ntotal = 210) mice from the two parental strains and 19 CSSs (fig. S1a). Significant variation was found for each sterol, with males generally showing more variation across the parental strains and CSSs. These results indicate the presence of QTLs affecting cholesterol on at least 8 chromosomes, campesterol on 6 chromosomes, and sitosterol on 14 chromosomes.

We studied weight gain in response to two well-established models of diet-induced obesity: a high-fat, low-sucrose diet and a high-fat, high-sucrose diet (13) (fig. S1b). In both cases, mice were weaned at 4 weeks, fed ad libitum, and weighed at weekly intervals. B6 mice tended to be slightly larger at weaning and gained weight much more rapidly on the high-fat diets. For the high-fat, high-sucrose diet, we studied 22 CSSs and found that at least 17 deviated from the pattern of diet-induced obesity seen in B6. In all 17 cases, the weight gain was intermediate between the parental strains. These results also show that certain kinds of nonadditive effects are readily detected in CSSs; the cumulative effect of each QTL on resistance to weight gain in the panel of 22 CSSs was greater than the difference in weight gain for the B6 and A/J progenitor strains (fig. S1b). For the high-fat, low-sucrose diet, we studied 14 CSSs and found one that showed significant resistance to the diet-induced obesity seen in B6. Together, these results revealed QTLs on at least 17 chromosomes, most of which were not previously reported (14).

We studied models of anxiety (15, 16) by subjecting the parental strains and the CSSs to two standard tests: an open-field test and a light-dark transition test (fig. S1c). In the first test, A/J mice were reluctant to move in the open field and typically deposited about six fecal pellets, whereas B6 mice energetically explored the test area and rarely left fecal pellets. In the second test, A/J mice were typically slow to move from a dark to a light chamber and spent much less time in the lit area than did B6 mice. Significant differences relative to B6 were found in three CSSs for open-field ambulation and five CSSs for light-dark transition.

Finally, we surveyed the serum levels of 21 amino acids in male and female mice (fig. S1d). Such measurements are commonly used to diagnose various metabolic, congenital, and neurological abnormalities in humans (17). Significant differences from the B6 phenotypes were found for 15 of the 21 amino acids. The median number of CSSs showing significant differences for a given amino acid was 3, with a range of 0 to 16.

We next tested whether a phenotypic difference in a CSS demonstrates the presence of a QTL on the corresponding chromosome. Genetic proof requires showing that the QTL can be mapped to a specific location on the implicated chromosome in an appropriate cross between the CSS and the B6 host strain. We selected eight of the inferred QTLs (one related to cholesterol level, three related to obesity, and four related to anxiety) and arranged mapping crosses to test the existence of the QTL and define its location. The crosses typically involved ∼80 progeny, and the significance threshold for linkage was chosen to reflect the fact that a genetic scan was performed across a complete chromosome rather than the entire genome (18).

In all eight cases, a QTL was mapped to a specific location on the substituted chromosome (Table 1 and fig. S2). In two cases, we were able to compare our results to previous QTL mapping studies. Several investigators have studied the A/J and B6 strains with respect to anxiety and reported evidence for a QTL on chromosome 1 (1921). The location in our mapping cross involving CSS-1 agrees closely with the reported location for this QTL. The two strains have also been studied for weight gain on a high-fat diet. Suggestive evidence has been reported for a QTL at 24 cM on chromosome 6 (22), which agrees closely with our result for the same chromosome.

Table 1.

Results of eight mapping crosses, showing trait, chromosome, cross, maximum LOD score along chromosome with appropriate significance threshold (18), and location of the peak in cM, with closest flanking markers studied.

Phenotype CSS strain Cross Max LOD [threshold] Location
Cholesterol CSS-4 Intercross (n = 78) 2.5 [1.6] 54.0 cM (D4Mit31 - D4Mit203)
Diet-induced obesity (high-fat, low-sucrose) CSS-6 Intercross (n = 93) 2.7 [1.6] 27.6 cM (D6Mit274 - D6Mit209)
Diet-induced obesity (high-fat, low-sucrose) CSS-10 Intercross (n = 93) 2.3 [1.6] 59.6 cM (D10Mit95 - D10Mit14)
Diet-induced obesity (high-fat, low-sucrose) CSS-16 Intercross (n = 72) 4.3 [1.6] 53.8 cM (D16Mit125 - D16Mit189)
Light-dark testing (first time to exit) CSS-1 Backcross (n = 91) 3.3 [1.7] 93.0 cM (D1Mit151 - D1Mit511)
Light-dark testing (total time in light) CSS-1 Backcross (n = 91) 4.1 [1.7] 103.0 cM (D1Mit151 - D1Mit511)
Open-field testing (defecation) CSS-1 Intercross (n = 91) 6.7 [1.7] 95.5 cM (D1Mit151 - D1Mit511)
Open-field testing (activity) CSS-6 Intercross (n = 82) 3.7 [1.5] 4.9 cM (D6Mit138 - D6Mit27)
Open-field testing (defecation) CSS-6 Intercross (n = 82) 1.6 [1.5] 17.0 cM (D6Mit274 - D6Mit209)
Light-dark testing (total time in light) CSS-17 Intercross (n = 70) 3.6 [1.5] 51.7 cM (D17Mit39 - D17Mit221)

The CSS approach has several merits that facilitate QTL detection and mapping in mice and other organisms. CSS mapping is typically more efficient than traditional crosses in the purely statistical sense of requiring fewer animals to detect a given effect or allowing smaller effects to be detected with a given number of animals (11). The relative power of CSS panels and segregating crosses has also been studied in a recent paper by Belknap (23), who reached similar conclusions.

CSSs are especially advantageous for detecting a given QTL in the presence of many other QTLs. In comparing our results with published mapping studies, we detected as many, and usually substantially more, QTLs with CSSs than with crosses of roughly comparable size (table S1). For example, studies of sitosterol and campesterol levels in an intersubspecific cross revealed only 3 QTLs, whereas we found 20 QTLs, 6 for campesterol and 14 for sitosterol. Similarly, studies of body weight and diet-induced obesity typically reveal 3 to 4 QTLs, whereas we found 17.

CSS panels also greatly simplify the subsequent work of fine-structure mapping and molecular identification of a QTL. To follow up initial QTL mapping in crosses, one must first undertake a lengthy breeding program to create a congenic strain in which the QTL of interest has been isolated from other unlinked loci. In contrast, CSSs allow one to move immediately to fine-structure mapping by crossing any CSS of interest to the host strain. Such crosses provide good resolution with relatively few progeny (∼80 in the experiments above) and can also be used to rapidly produce congenic strains carrying a small interval around the QTL. Molecular identification of QTLs is also aided by recent advances in mouse genomics, including the availability of both genome sequences for the A/J and B6 strains that facilitate gene discovery and bacterial artificial chromosome libraries for both strains that facilitate functional studies in transgenic mice (24).

The CSS approach obviously requires the availability of a CSS panel for the strain combination of interest. At present, the approach is limited to the A/J and B6 strains. Although these strains have abundant physiological differences (4), additional CSS panels of mice and other organisms (25, 26) will be valuable as resources for the scientific community.

Supporting Online Material

Materials and Methods

SOM Text

Figs. S1 and S2

Table S1


References and Notes

View Abstract

Navigate This Article