We have constructed one of the most dense and accurate genetic maps for any organism. Using both outbred animals and recombinant inbred lines, we genetically placed more than 10,000 markers onto the mouse genome, yielding an average genetic inter-marker distance of 0.15 cM. The average resolution is 0.37–0.45 cM/Mb for the sex-averaged maps. Even small numbers of genotype errors will inflate genetic map distances, and this factor has often made it difficult to draw conclusions about the factors that influence local recombination rates. Our high level of genotyping accuracy (~99.99%), the use of large pedigrees, and the construction of separate maps based on two types of populations has result in a highly reliable linkage map and highly accurate estimates of recombination rates.
Our map was constructed using HS mice that were also employed to find QTLs [11
]. We collected 101 phenotypes for 1,904 HS mice (389 parents of these animals were also genotyped but not phenotyped, accounting for the 2,293 HS animals used in map construction). For the QTL analysis, 10,202 markers unambiguously positioned on the HS were used, supplemented by a further 1,910 informative markers whose locations were derived by interpolation from the physical map then available (NCBI build 34). Thus a sex-averaged map of 12,112 markers was used for QTL identification [11
]. We have not yet explored the effect on QTL identification of using sex-specific maps.
We found that chromosome-specific recombination rates do not greatly differ between outbred (HS) and inbred populations (RIs): there are very few regions with discrepant recombination rates. Petkov et al. [28
] suggest that inbreeding results in strong selection for specific allele combinations in mouse inbred lines: selection against recombination occurs when a combination of alleles at linked loci confers a survival advantage during the inbreeding process. Our data do not support the hypothesis of lower recombination events in inbreds or of strong selection forces that favor specific alleles. The inheritance of alleles in the RIs is not significantly different from what is expected.
We do see, however, some differences related to chromosome size. Recombination in the RIs appears to be more constrained by chromosome size, which explains 73% of the variance in the chromosome average recombination rate, whereas in the HS, it explains only 35% of the variance. Higher recombination rates are observed in the HS relative to the RIs in the large chromosomes, and lower rates in the small chromosome (with the exception of the X chromosome). We cannot at this point explain this observation.
We found remarkable similarity in the features that contribute to variation in recombination rates in mouse and human chromosomes. We confirmed that chromosome, position on the chromosome, sex, and sequence composition are common important factors. In both species, differences in recombination rates are correlated with chromosome size (or arm size) and proportional distance from the centromere. It is possible that, as has been suggested in humans, variation in recombination rates between chromosomes occurs because each chromosome arm is constrained to have a single chiasma, and the effect of fixing the lower limit of recombination will affect smaller chromosome arms more than larger [29
]. Unlike human chromosomes, mouse chromosomes are all acrocentric (they lack a fully-formed P arm). Humans therefore have approximately twice as many chromosomal arms as mice, which would explain the difference in average recombination rates: our estimate of 0.6 cM/Mb is about half that found in humans. However, data from rats do not support this notion [31
In both humans and mice, the average recombination rate in females is higher than males. Recombination rates are higher near the centromeres for females and towards the telomeres for males [32
]. The correlation between female and male recombination rates across the whole genome is relatively weak, with many sex-specific peaks and troughs in recombination rates. This suggests that a larger proportion of the variance in recombination can be explained by sex-specific recombination events. A direct measure of recombination using immunohistological methods has shown that sex, and not genotype, influences recombination [34
The main difference between the LD landscapes of the human and mouse genome is the presence of large haplotype blocks of up to several megabases in size in mouse inbred strains. We have shown that the position of these haplotype blocks tends to coincide with recombination deserts. In contrast, recombination jungles coincide with borders of these blocks or regions without evident block structure. An important implication of this observation is that genetic variants (such as those underlying quantitative trait loci) that lie within a recombination desert will be difficult to identify with the SNP maps we have generated.
We exploited the LD landscape of the mouse genome to concentrate on regions of very high-recombination rates (jungles) and very low rates (deserts). We used the selected jungles and deserts, as well as the recombination rates in the whole genome, to study sequence features that were shown to influence global and local recombination rates in humans. A major limitation of this analysis is that the location of recombination hotspots within jungles is unknown. Although deserts are essentially homogenous regions without recombination, jungles are presumably heterogeneous regions with an uneven distribution of recombination events. Therefore, in this study we focused on factors that are known to influence recombination in humans.
In humans, sequence motifs are an important determinant of variation in recombination rate. Analyses of the sequence correlates of recombination based on low resolution genetic maps have been unable to identify motifs that independently predict recombination rate. Most of the correlations have low coefficients and the sequence variables are correlated with each other. High resolution mapping of recombination hotspots in humans found CCTCCCT oligomers to be the strongest signal of recombination hotspots [16
]. The motif is also the strongest predictor of recombination jungles in the mouse, explaining 63% of the deviance. In humans, this motif is found within the long terminal repeats of two retrovirus-like transposons, THE1A and B. The mouse genome does not contain this motif, providing additional support for its independent effect.
Our data support a two-stage model of recombination in which recombination rates are constrained over large scales, but are rapidly evolving on a small scale. The conservation of recombination deserts in inbred lines, shown by haplotype blocks and the presence of some of those deserts also in humans, are also consistent with this model. Comparisons between species using low-resolution sex-averaged maps have previously detected only a slight positive correlation between recombination rates [31
]. The fact that the frequency of the CCTCCCT motif is well correlated across species in the regions that we studied, whereas the location of hotspots is not conserved, suggests that an interaction of several elements, including the CCTCCCT motif, influences recombination. The high degree of similarity between the factors that are correlated with mouse and human recombination rates, together with the correlation in recombination rates itself, implies the existence of a common mechanism that influences recombination rates in mammalian genomes. The genetic maps we have made are available from http://gscan.well.ox.ac.uk/#genetic_map
and as Tables S1