Local Rates of Recombination Are Positively Correlated with GC Content in the Human Genome

Abstract

The GC content of human DNA varies widely across the genome, ranging from 30% to 60%, and regions of hundreds of kilobases (often referred to as isochores [Bernardi 2000]) may have relatively homogeneous base compositions. This compositional heterogeneity appears to be very widespread in eukaryotes (Nekrutenko and Li 2000) and may represent an important level of genome organization, insofar as gene density (Zouback, Clay, and Bernardi 1996 ), gene length (Duret, Mouchiroud, and Gautier 1995 ), and patterns of codon usage (Sharp et al. 1995 ), as well as the distribution of different classes of repetitive elements (Soriano, Meunier-Rotival, and Bernardi 1983 ; Duret, Mouchiroud, and Gautier 1995 ), are all correlated with GC content. Despite intensive investigation, the underlying cause(s) of the observed heterogeneity remains contested, with two major hypotheses competing: Bernardi has suggested that selection is primarily responsible for maintaining the observed patterns (Bernardi and Bernardi 1986 ; Bernardi 2000 ), a view supported by recent analysis of polymorphism in the human MHC cluster (Eyre-Walker 1999 ), whereas the balance of opinion has favored systematic mutational bias as the ultimate cause (Filipski 1987 ; Suoeka 1988; Wolfe, Sharp, and Li 1989 ; Francino and Ochman 1999 ). Ultimately, however, it is not clear why either selection for increased GC content or mutation bias should promote such marked local variation in genomic nucleotide content. A third hypothesis, namely, that recombination may explain, or at least contribute to, the observed compositional differences (Holmquist 1992 ; Eyre-Walker 1993 ; Charlesworth 1994 ), has received less attention. The idea that recombination may be involved in determining variation in GC content arose from the observation of an association between recombination and GC-rich chromosomal regions. The clustering of chromosomal rearrangements in isochores with high GC contents was described first by Bernardi (1989) and subsequently by Holmquist (1992) . Eyre-Walker (1993) demonstrated a statistically significant positive correlation between overall chromosomal GC content and chiasmata density, which accorded both with Ikemura and Wada's (1991) finding that third-position GC content was positively correlated with chiasmata density and with Holmquist's (1992) observation that GC-rich chromosomal bands were chiasmata-dense. Although the coarse level at which the association was identified prevented determination of the causality underlying the relationship, GC-biased repair of gene conversion was suggested as the most likely explanation (Holmquist 1992 ; Eyre-Walker 1993 ). Another author, commenting on the same data, suggested instead the possible indirect effects of recombination via its impact on the efficacy of natural selection (Charlesworth 1994 ). This evolutionary explanation posited that changes in GC content promoted by natural selection would be confined to those regions of the genome with a recombination rate high enough to overcome the effects of random genetic drift (the so-called “Hill-Robertson effect”; Hill and Robertson 1966 ; Felsenstein 1974 ), thereby explaining the correlation as selectively rather than mutationally determined. There were not enough data available at the time to resolve the issue, however, so the apparent association has remained only incompletely described and largely unexplained. In particular, the scale of the association has remained unknown in the absence of data on the relationship between GC content and recombination rate within chromosomes. With the human genome project nearly complete and detailed sequence and genetic map data readily available, we chose to revisit the reported association, both to confirm that the rough correlation described originally still holds and to determine whether, with more information, it is possible to disentangle the forces likely to contribute to the inferred relationship. To examine the association of recombination and GC content in the human genome, we used intron sequences for genes in GenBank for which genetic map positions were known, and band-specific estimates of recombination rate derived using the Genetic Location Database (LDB) (Collins et al. 1996 ; http://cedar.genetics.soton.ac.uk/public_html/ldb.html). The GC content of intronic sequences has been shown to be correlated both with the third-codon-position GC content of associated genes and with the GC content of the surrounding isochore (Clay et al. 1996 ); thus, intron sequences provide a convenient measure of chromosomal GC content. Only genes with defined map information were considered, and redundant sequences were eliminated prior to analysis with the program GPURGE from Falling Rain Genomics (http://www.fallingrain.com/publicserver/index.html). Recombination rates (in cM/Mb) were estimated from the integrated genetic and physical map data contained in the LDB. The ultimate source of the genetic map in the LDB is th