Most recombination takes place in numerous, localized regions called hotspots. However, empirical evidence indicates that nascent hotspots are susceptible to removal due to biased gene conversion, so it is paradoxical that they should be so widespread. Previous modeling work has shown that hotspots can evolve due to genetic drift overpowering their intrinsic disadvantage. Here we synthesize recent theoretical and empirical results to show how natural selection can favor hotspots. We propose that hotspots are part of a cycle of antagonistic coevolution between two tightly linked chromosomal regions: an inducer region that initiates recombination during meiosis by cutting within a nearby region of DNA and the cut region itself, which can evolve to be resistant to cutting. Antagonistic coevolution between inducers and their cut sites is driven by recurrent episodes of Hill–Robertson interference, genetic hitchhiking, and biased gene conversion.

---

RECOMBINATIONAL hotspots are small chromosomal regions (0.1–3 kb) in which most recombination occurs (reviewed in PETES 2001). They are numerous and highly dispersed throughout the genomes of all species that have been studied, with >25,000 hotspots estimated to occur in humans (MYERS et al. 2006). The locations of well-characterized hotspots have diverged between humans and chimps, as has the recombinational landscape of closely related Drosophila species, indicating that hotspots may arise and go extinct on a timescale of less than several million years (reviewed in NISHANT and RAO 2006).

Recombination is initiated by a double-strand break (DSB) in the DNA of one homologous chromosome, followed by the removal of some of the cut strand's genetic code by exonucleases, making the DNA around the cut site single stranded (Figure 1). The single-stranded DNA flanking the cut site facilitates the damaged DNA's ability to recognize its uncut homologous allele, which it invades to be repaired (Figure 1). The removed bases of the cut strand, and flanking nearby regions, are later replaced by those of the uncut, homologous chromosome using a long-patch DNA repair pathway, which leads to biased gene conversion against base pairs located on the cut chromosome that are near the DSB (Figure 1). This unidirectional exchange of hereditary information from the uncut to the cut chromosome produces an evolutionary disadvantage that is functionally equivalent to meiotic drive (segregation distortion) against the cut strand; yet, paradoxically, they are nonetheless abundant in all studied eukaryotic genomes.

View larger version (33K): In this window In a new window Download PPT slide   FIGURE 1.— The sequence of steps in the double-strand-break (DSB)-induced recombination that leads to biased gene conversion.

 
The molecular process by which recombinational hotspots are created is not known (NISHANT and RAO 2006). Nonetheless, some associations have been established. Hotspots in yeast are more common when the flanking regions have high GC content and open chromatin structure (DNase-1 hypersensitivity) and they are sometimes associated with the binding of transcription factors at upstream locations (NISHANT and RAO 2006). Computational studies based on the extensive human HapMap indicate the 11% of hotspots contain, or are close to, the motif CCTCCCT (MYERS et al. 2005). Studies with mice indicate that changing a single SNP is sufficient to completely shut down some hotspots (JEFFREYS and NEUMANN 2002). Despite these associations, the specific structural features that cause a region of DNA to become a recombinational hotspot are not well understood.

NICOLAS et al. (1989) and BOULTON et al. (1997) were the first to recognize the conundrum that recombination hotspots are frequently selected against but nonetheless accumulate in genomes (the "hotspot paradox"). Computer simulation studies by BOULTON et al. (1997), and later by PINEDA-KRCH and REDFIELD (2005), examined several mechanisms by which natural selection might counter biased gene conversion and thereby favor the accumulation of hotspots. They found no feasible selective mechanism to explain the hotspot paradox. Although hotspots might be selectively favored when they ensure proper segregation of chromosomes during meiosis, they are far too numerous to make this a feasible selective factor explaining their prevalence (PINEDA-KRCH and REDFIELD 2005; COOP and MYERS 2007). Recently both CALABRESE (2007) and COOP and MYERS (2007) used diffusion analysis to study the interplay between genetic drift and biased gene conversion. They extended previous diffusion models of biased gene conversion (GUTZ and LESLIE 1976; NAGYLAKI 1983) to show—with feasible parameter values—that at least some, less extreme, hotspots can drift to fixation. ARCHETTI (2003) proposed an alternative way to solve the hotspot paradox. If the DSB in the cut allele was somehow induced by the homologous allele on the uncut chromosome, then the hotspot paradox vanishes because DSB-inducing alleles would gain from biased gene conversion, rather than be harmed by it.

Here we extend previous work on the hotspot paradox by integrating recent results from two areas: (i) theoretical and empirical studies of the use of preferred codons and (ii) theoretical work on neutral modifiers of localized recombination rate. We first briefly describe these two developments in separate sections below and then integrate them into extant hotspot theory to motivate a new hypothesis to solve the hotspot paradox that is based on natural selection in the context of intragenomic conflict.

Some synonymous codons can be selectively favored over others for a variety of reasons, for example, if they increase the speed and/or accuracy of translation, make splice sites less ambiguous, or increase stability of mRNA secondary structure (reviewed in CHAMARY et al. 2006). In some cases different codons may be preferred in these different contexts (WARNECKE and HURST 2007). The distribution of preferred codons offers a unique ability to evaluate the efficacy of natural selection in different chromosomal regions because, all else being equal, an increased prevalence of preferred codons indicates an increased effectiveness of natural selection.

COMERON et al. (1999) and MCVEAN and CHARLESWORTH (2000) showed that, in theory, the prevalence of preferred codons should be decreased with the length of exons. This theoretical result is a consequence of the increased level of Hill–Robertson interference (HILL and ROBERTSON 1966; FELSENSTEIN 1974) and the associated reduction in the efficacy of selection that occurs when longer stretches of codons are selected simultaneously. To illustrate Hill–Robertson interference intuitively, consider the simplest case of two SNPs (A⁺/A^– and B⁺/B^–, where the "+" superscript denotes the nucleotide favored by selection) that are tightly linked so that they rarely recombine. When sampling error in a finite population causes the two favored SNPs to occur only in coupling (i.e., only A⁺B⁺ and A^–B^– haplotypes are present; positive linkage disequilibrium, LD), selection on one SNP reinforces that on the other, and fixation of the favored haplotype will be relatively fast. But when the SNPs occur only in repulsion (A⁺B^– and A^–B⁺; negative LD), selection on one SNP interferes with that on the other, fixation of the favored haplotype will be relatively slow, and the nonneutral polymorphism is expected to persist longer. Hence the joint operation of sampling error and natural selection causes negative disequilibrium to predominate because it persists longer (Hill–Robertson interfering disequilibrium, or Hill–Robertson interference), which reduces the efficacy of natural selection. The lowered efficiency of selection can be expressed as a lowered effective population size (N_e), since the strength of drift relative to selection increases as effective population size declines. Returning to the expected relationship between exon length and prevalence of preferred codons, longer, contiguous groups of codons have more opportunity for Hill–Robertson interference, and thus there will be a lower efficiency of natural selection on each codon and hence a lower prevalence of preferred codons.

COMERON and KREITMAN (2002) used computer simulation to expand the earlier work on exon length to the context of clusters of exons separated by introns. The presence of introns increases the chance of recombination occurring between the separated parts of the coding sequence and thereby reduces Hill–Robertson interference between the separated parts of the gene. More generally, any time that recombination is increased between blocks of selected sequences, the efficacy of selection is predicted to be increased. As predicted by theory, COMERON and KREITMAN (2002) and COMERON and GUTHRIE (2005) found that the prevalence of preferred codons in Drosophila increased when introns were present and also in those regions of longer exons that were closer to the introns/exon border. LOEWE and CHARLESWORTH (2007) expanded the theory by showing that stronger selection on rare, disfavored, nonsynonymous SNPs (background selection) can also contribute substantially to Hill–Robertson interference. CIRULLI et al. (2007) related all of the above studies to hotspots by showing that there is a positive correlation between the prevalence of preferred codons and the level of recombination that a hotspot produced. Collectively, these studies indicate that localized enhancers of recombination (introns or hotspots) lower the amount of Hill–Robertson interference at neighboring sequences and thereby increase the average fitness of the genetic "neighborhood."

Before closing this section, we need to address the possibility that biased gene conversion, rather than selection, has led to the observed patterns of preferred codon use (reviewed in MARAIS 2003). Recent work (COMERON and KREITMAN 2002; KLIMAN and HEY 2003; COMERON and GUTHRIE 2005; CHAMARY et al. 2006; RESCH et al. 2007) provides strong evidence that, irrespective of a role for biased gene conversion, selection is a substantial factor leading to the prevalence of preferred codons, at least in Drosophila and some mammals.

The studies reviewed in the above section demonstrate that elements, such as hotspots and introns, that increase local recombination rate can increase the efficacy of selection at nearby sites, but not that these elements will themselves be selectively favored and accumulate in the gene pool. However, in a recent series of articles, N. Barton and S. Otto have extended the earlier work by FELSENSTEIN and YOKOYAMA (1976) on neutral modifiers of recombination (BARTON 1995a,b; OTTO and BARTON 1997, 2001; BARTON and OTTO 2005). These authors made a pivotal advance in our understanding of the ramifications of the Hill–Robertson effect by showing that neutral modifiers of recombination will, on average, become nonrandomly associated with the higher-fitness genetic backgrounds that they produce—as long as the linkage is sufficiently tight and the strength of selection is not too strong, or too weak, relative to random genetic drift. This nonrandom association between neutral modifiers of recombination and the high-fitness genetic backgrounds that they generate can be expressed as an "effective selection coefficient" (s_e) on the neutral modifiers of recombination (BARTON and OTTO 2005).

Determining a general solution for the magnitude of s_e is difficult because of the simplifying assumptions needed to model evolution of neural modifiers of recombination. Nonetheless, OTTO and BARTON (2001) and BARTON and OTTO (2005) have shown that s_e can be large enough to overpower drift in species with small N_e, and MARTIN et al. (2006) showed that this result can be extended to species with larger N_e when the species is divided into smaller demes with limited gene flow. ROZE and BARTON (2006) further showed that even in very large populations without subdivision, s_e can be as large as 10% of the value of the average selection coefficient of nearby segregating sites. For pragmatic reasons, nearly all of the recent studies of neutral modifiers of recombination have assumed three loci (one neutral modifier of recombination and two selected loci). ILES et al. (2003) showed that increasing the number of nearby selected sites markedly increases the potential value of s_e. One limitation with all of the previously described studies is that they focus only on beneficial mutations, which are far less numerous then harmful mutations. KEIGHTLEY and OTTO (2006) used numerical analysis to examine the case of genomewide deleterious mutations and neutral modifiers of recombination. They showed that this context led to relatively strong s_e, with the strength of effective selection on the neutral modifiers of recombination increasing with population size, owing to more segregating polymorphism with increasing population size and hence more Hill–Robertson interference. PALSSON (2002) showed that a recombination modifier can also be favored by background selection on deleterious mutations in the context of very tightly linked groups of selected sites. Collectively this body of theory indicates that neutral modifiers of recombination can hitch a ride to higher gene frequency by reducing Hill–Robertson interference in their genetic neighborhood and that this is feasible in populations with large or small N_e. We refer to this "selection by association" for neutral recombination modifiers as the "hitchhiking effect."

In the following section we integrate into the established models of recombinational hotspots the findings described above concerning preferred codons and genetic hitchhiking of neutral modifiers of recombination. This integration leads to a new theoretical property of hotspots: antagonistic coevolution between DSB inducers and DSB-cut sites.

Consider a small chromosomal region that is located in an area with low recombination (i.e., a "coldspot"). Because of the low recombination, negative linkage disequilibrium accrues that interferes with the efficacy of natural selection (Hill–Robertson interference) in the coldspot. Next consider a mutation that creates a nascent hotspot by inducing a DSB in the coldspot (we refer to the new mutation as a "DSB inducer" and the region that it cuts as the DSB-cut region; Figure 2A). The fate of the DSB-inducing mutation will be affected by drift and selection on its genetic background and also by biased gene conversion depending on the position of the DSB inducer in relation to the DSB-cut region (Figure 2A). As described above, BARTON and OTTO (2005) and ROZE and BARTON (2006) solved for the effective selection coefficient (s_e) for a neutral modifier of recombination that is due to its nonrandom association with higher-fitness genetic backgrounds [which, to avoid ambiguity, we refer to as s_e(GB)]. A comparable effective selection coefficient [s_e(BGC)] can be calculated for the effect of biased gene conversion against the DSB-cut region (NAGYLAKI 1983; CALABRESE 2007; COOP and MYERS 2007) by using diffusion analysis: s_e(BGC) = –2r_H( – B), where r_H is the elevation in the rate of DSB for the "hot" allele that is susceptible to being cut, compared to the "cold" allele, and B is the probability that a cut allele is transmitted to a gamete during meiosis. Let the joint effects, on the DSB inducer, of both biased gene conversion and selection on genetic backgrounds be denoted by S*. Because biased gene conversion and selection will act sequentially and independently, the combined effects of biased gene conversion and selection on genetic backgrounds will be multiplicative, so that fitness = (1 + s_e(BGC))(1 + s_e(GB)) 1 + s_e(BGC) + s_e(GB) = 1 + S*, when both effective selection coefficients are assumed to be small. The joint action of drift and selection acting on the mutation can be approximated with the diffusion equation method (CROW and KIMURA 1970; EWENS 2004). Assuming constant selection coefficients, the probability of fixation of a new DSB-inducer mutation is described by

(1)

However, the effective selection coefficient on the DSB inducer due to genetic backgrounds [s_e(GB)] is a consequence of transient genetic hitchhiking (BARTON and OTTO 2005; ROZE and BARTON 2006), which will become trivially small after the hotspot destroys most of the nearby linkage disequilibrium. This causes the probability of fixation shown in Equation 1 to be an upper bound to the true value. Nonetheless, even modest levels of genetic hitchhiking can be important. For example, suppose that a nascent DSB inducer is generated in a population of size 10,000 and that it hitchhikes to a frequency of 1% before becoming disassociated with higher-fitness genetic backgrounds. Its subsequent probability of fixation by drift is its frequency after becoming selectively neutral (0.01), which is 200 times higher than the probability of fixation of a new neutral mutation [1/(2N)]. As described in a later section, multiple hitchhiking events can further increase the probability of fixation of a DSB inducer. Below we discuss the opportunities for a nascent DSB inducer to rise to high frequency to create a hotspot, in relation to its position in the DSB-cut region.

View larger version (20K): In this window In a new window Download PPT slide   FIGURE 2.— (A) The possible locations of a DSB inducer relative to the DSB-cut site, which is located in the center of the recombinational hotspot. Also shown is the relative frequency of biased gene conversion within the hotspot. (B) The effective selection coefficient (S*) experienced by a DSB inducer in relation to its chromosomal location (cis or trans to the cut chromosome) and position relative to the DSB-cut site (within or outside of the cut site).

 

Suicidal hotspots:

When the DSB inducer is located in cis and in the center of a very narrow DSB-cut region, then the exonuclease activity around the DSB-cut site will always destroy the DNA coding for the DSB inducer (Figure 1). As a consequence, such a hotspot cannot evolve due to a selective advantage based on the hitchhiking effect because the DSB inducer must commit "suicide" to carry out its function. Nonetheless, hotspots of this type can evolve via genetic drift (CALABRESE 2007; COOP and MYERS 2007). Selection due to biased gene conversion [s_e(BGC)] is always expected to be weak because empirical studies of hotspots (reviewed in DE MASSY 2003) indicate that the recombination rates of hotspots are at most 0.1 cM, so at least 99.9% of the time a hotspot is protected from biased gene conversion. In diffusion models of natural selection, the interplay between drift and selection is quantified by the parameter = 4N_es, where is the strength of selection scaled by the effective population size (N_e) and the genic selection coefficient (s). When is small ( <1), genetic drift can overpower selection and lead to the fixation of a mutation that has an evolutionary disadvantage. Because biased gene conversion at colder hotspots can be quite infrequent, = 4N_es_e(BGC) can be small enough to lead to the fixation of "suicidal" hotspots in species like humans with N_e 10,000 (CALABRESE 2007; COOP and MYERS 2007).

Russian roulette hotspots:

When the DSB inducer and the DSB-cut region are in cis but not guaranteed to be overlapping, the DSB inducer will at least sometimes escape being destroyed by biased gene conversion. In this case both biased gene conversion and selection on genetic backgrounds influence the fate of the mutation, and S* = s_e(GB) + s_e(BGC) = s_e(GB) – 2r_H(). Because of the hitchhiking effect, the term s_e(GB) will be positive, on average, and because of biased gene conversion near the DSB-cut region, the term s_e(BGC) will always be negative. The effective selection coefficients due to biased gene conversion and genetic background are therefore opposing.

Empirical studies of hotspots indicate that gene conversion tracts (heteroduplex DNA) tend to be (i) short relative to the hotspot's total length, (ii) stochastically distributed about the center of the hotspot with steep declines in occurrence as distance from the center increases, and (iii) nearly always coconverted (via the long-patch nucleotide-excision-repair pathway, NER) to the sequence of the uncut allele, including heteroduplex regions outside the location of exonuclease activity (reviewed in DE MASSY 2003). These properties cause the probability of biased gene conversion against a sequence located within a hotspot to drop rapidly with its distance from the hotspot's center (Figure 2A). As a consequence, most DSB inducers located within a hotspot will not commit suicide when they induce a DSB, but will instead play "Russian roulette," with the probability that there is a bullet in the pistol's active chamber declining with the distance separating the DSB inducer and the center of the hotspot. Even when the DSB inducer is located in the center of a hotspot, it will be partially protected from biased gene conversion when the DSB-cut region is broad. In this case the DSB inducer can sometimes escape biased gene conversion because some DSBs will occur sufficiently far away so that the DSB inducer is not included in the heteroduplex DNA formed during recombination. In sum, the biased gene conversion cost [s_e(BGC)] to a DSB inducer that is located within a hotspot depends importantly on its position: its accumulation is most strongly opposed by biased gene conversion when it is centrally located within a narrow hotspot (A in Figure 2, A and B), but only weakly opposed when located on the edge of a broad hotspot (A_edge in Figure 2A). Whenever a DSB inducer sometimes escapes biased gene conversion (i.e., when it cuts a neighboring sequence sufficiently far away), it can be potentially favored by selection when the hitchhiking effect more than counterbalances its disadvantage due to biased gene conversion.

Unlike the relatively simple calculations for the effective selection coefficient associated with biased gene conversion [s_e(BGC), which depends on only two parameters, r_H and B; NAGYLAKI 1983; CALABRESE 2007; COOP and MYERS 2007], the effective selection coefficient due to the hitchhiking effect [s_e(GB)] is dynamic and therefore more difficult to calculate empirically (BARTON and OTTO 2005; ROZE and BARTON 2006). Nonetheless, we can use data to evaluate the feasibility that the hitchhiking effect is important in determining the fate of a nascent DSB inducer. To operate, the hitchhiking effect requires that there be ample, tightly linked polymorphisms that influence fitness and create Hill–Robertson interference (which is sensitive to moderate changes in effective population size). Fine-scale mapping of >25,000 hotspots in the human genome indicated that most regions outside of hotspots recombine at a very low level (except those located in heterochromatin) and that hotspots account for most recombination (80%). A hotspot will break down interfering disequilibrium that occurs only between chromosomal regions located on its two sides (flank right and flank left), each of which is expected to accumulate Hill–Robertson interfering linkage disequilibrium due to its low recombination rate. By shuffling the flanking sides of a hotspot, recombination can potentially produce a higher-fitness haplotype (spanning the hotspot) that was formerly absent because of interfering linkage disequilibrium between flank right and flank left and on which a DSB inducer can potentially hitch a ride to higher frequency.

Within the two flanks, polymorphism for mutations influencing fitness will be highest when the selected sites are not easily fixed by selection or drift, i.e., when = 4N_es is not too large or too small. When < 0.25, genetic drift overpowers selection, and when > 2.5, selection predominates over drift, so mutations between these limits (0.25 < < 2.5) will be maintained in mutation–selection–drift balance and provide most of the variation for genetic hitchhiking of a nascent DSB inducer (OHTA and KIMURA 1971; OHTA 1972; LI 1987). This constraint implies that the type of genetic variation (e.g., preferred codons, nonsynonymous SNPs, etc.) providing an advantage to hotspots should differ between species with high vs. low N_e.

The constraint that (0.25 < < 2.5) also implies that hotspots should be able to evolve in species with very different N_e. As N_e changes, the requisite fitness variation between the flanks of a hotspot (that is needed to fuel hitchhiking of nascent DSB inducers) can be achieved by changing the interhotspot interval (causing more or fewer selected sites within each flanking region) or by changing the fitness effects of the constituent sites (e.g., preferred codons, weakly selected nonsynonymous SNPs, or more strongly selected nonsynonymous SNPs, etc.). Although both beneficial mutations and harmful mutations can contribute to the requisite fitness variation of hotspot flanks, the higher abundance of harmful mutations (LOEWE et al. 2006; EYRE-WALKER and KEIGHTLEY 2007) would seem to make background selection on harmful mutations the stronger factor. LOEWE and CHARLESWORTH (2007) combined theory and data from African populations of Drosophila melanogaster to show that harmful nonsynonymous mutations can be sufficiently polymorphic to contribute substantially to Hill–Robertson interference within individual genes, or groups of tightly linked genes, even when >> 1 (as described in a later section).

An interesting prediction of the hypothesis that hotspots evolve due to the hitchhiking effect is that hotspots should be more evenly spaced than predicted by chance. This prediction is a simple consequence of the fact that a nascent DSB-inducer mutation that occurs too close to another hotspot will not have sufficient Hill–Robertson disequilibrium between its two flanks to gain a substantial hitch to higher frequency. Finally, a DSB inducer can experience multiple rounds of hitchhiking accumulation. Both flanking regions of a hotspot are expected to continue to experience low levels of recombination due to the rarer recombination that occurs outside of hotspots. This within-flanks recombination (as opposed to the between-flanks recombination produced by the DSB inducer) will generate new fitness variation within each flanking region that can fuel recurrent rounds of hitchhiking by the DSB inducer.

A second prediction concerning the location of hotspots is that they should share characteristics among species with similar N_e. In species like yeast with very large N_e, = 4N_es may be too large for most nonneutral mutations to be maintained in mutation–selection–drift balance, causing hotspots to be less common within clusters of strongly selected coding genes, as reported by PAL and HURST (2003). So species with similar N_e are predicted to have hotspots in similar genetic regions, as measured by the strength of selection. Also, within a species the density of hotspots should be correlated with the density of sites segregating for nearly neutral mutations (0.25 < < 2.5).

DSB inducer outside the DSB-cut region:

A DSB inducer can also occur in a region that is not within the DSB-cut region. There are three possibilities. The simplest case is a flanking, cis-acting DSB inducer that is outside the DSB-cut region (B in Figure 2, A and B). In this case, any benefits of a new mutation (creating a DSB inducer) that are due to reduced Hill–Robertson interference [s_e(GB)] are unopposed by the cost of biased gene conversion [s_e(BGC)], and the corresponding probability of fixation of the mutation, all else being equal, will be higher. However, there is a cost to being outside the DSB-cut region, because in this case the DSB inducer can recombine only with the flanking site in which it is not located. The other two possibilities involve trans-acting DSB inducers that cut the other homologous chromosome. When the DSB inducer is located in the cold allele and outside the DSB-cut region (D in Figure 2, A and B), the cis vs. trans action has no consequence and the effective selection coefficient [s_e(GB)] on the DSB inducer is the same as case B in Figure 2, A and B. However, when the DSB inducer acts in trans and is located on the cold allele within the DSB-cut region, biased gene conversion favors the DSB inducer (C in Figure 2B). ARCHETTI (2003) considered this case outside the context of the hitchhiking effect and showed that biased gene conversion to the uncut allele can lead to the accumulation of a trans-acting DSB inducer. The influence of the hitchhiking effect will reinforce the accumulation of a trans-acting DSB inducer that cuts its homologous allele. Recent evidence indicates that a trans-acting DSB inducer is possible (BAUDAT and DE MASSY 2007; see also references cited in ARCHETTI 2003), but it is still unclear whether such trans action is common.

Death of a hotspot:

Once a DSB inducer accumulates, it will select for a DSB-cut region that is resistant to the prevailing DSB inducer, due to its advantage via biased gene conversion (Figure 3). COOP and MYERS (2007) showed that if such a resistance allele is initially absent in a large population, then feasible parameter values (for humans) indicate that the waiting time until a resistant DSB-cut region becomes established can be many thousands to hundreds of thousands of generations. The evolutionary advantage of a DSB-cutting region that is resistant to cutting is small because of the low frequency of recombination at all hotspots (<10^–3). Nonetheless, eventually a cold allele for the active DSB-cut region is expected to accumulate due to drift or its meiotic-drive-like advantage due to biased gene conversion of the DSB-cut region (Figure 3). Once this occurs, the chromosomal region surrounding the silenced hotspot will again experience very low levels of recombination, so negative linkage disequilibrium causing Hill–Robertson interference is expected to accrue (BARTON and OTTO 2005).

View larger version (19K): In this window In a new window Download PPT slide   FIGURE 3.— (A) The cycle of antagonistic coevolution that can occur when the DSB inducer flanks the DSB-cut site, rather than the two being coincident. (B) A schematic of the series of events that occur during antagonistic coevolution between a DSB inducer and its flanking DSB-cut site. When recombination is rare, Hill–Robertson interfering linkage disequilibrium (LDinterfering) accumulates (black solid line). The LDinterfering creates a lag load favoring the evolution of a DSB inducer that cuts at a DSB-cut site located at a nearby location—sufficiently far away that the DSB inducer does not always experience biased gene conversion due to the repair of the DSB. The DSB inducer rises to higher frequency via genetic hitchhiking (hitchhiking effect) before the LDinterfering is largely ameliorated (solid red curve). The DSB inducer subsequently drifts (dotted red curve). Low-level recombination within the flanking regions of the hotspot can lead to additional rounds of accumulation of the DSB inducer via genetic hitchhiking, sometimes leading all the way to fixation. The accumulation of the DSB inducer creates a lag load (due to biased gene conversion against the cut DNA strand) favoring a "cold" DSB-cut-site sequence that is resistant to being cut by the DSB inducer. The cold DSB-cut-site sequence (assumed to be initially absent in the finite population) eventually accumulates (solid blue curve), which causes LDinterfering to begin to accumulate, creating a lag load for a new DSB inducer that can cut the extant cold DSB allele, which starts a new cycle of antagonistic coevolution.

 

Hotspot birth and death cycle:

When a DSB inducer is absent in a chromosomal region, recombination occurs at a lower background rate (compared to hotspots) and Hill–Robertson interfering linkage disequilibrium is expected to accumulate (Figure 3). This negative LD creates a lag load (MAYNARD SMITH 1976) that provides for a selective advantage to a mutation with a DSB-inducer phenotype—the further it is from the center of the DSB-cut region, the greater the advantage. The DSB inducer can hitch a ride to higher frequency so long as it sometimes cuts neighboring DNA (or cuts the homologous chromosome) and thereby avoids its guaranteed self-destruction. Once a DSB inducer accumulates to substantial frequency, it selects for a DSB-cut region that is resistant to cutting, due to its advantage from biased gene conversion of the uncut chromosome. After a resistant DSB-cut region accumulates, it leads to lower recombination and increased Hill–Robertson interference between its flanks, which sets the stage for a new DSB inducer to invade that can cut the extant, resistant DSB-cut region—and the cycle of antagonistic coevolution begins anew (Figure 3).

Requirements for the "cut-thy-neighbor" model to operate:

Our hypothesis for the rise and fall of hotspots via antagonistic coevolution between DSB inducers and DSB-cut regions (Figure 3) has several prerequisites that must be met. First, there must be ample polymorphism for selection to act upon in the chromosomal regions surrounding a hotspot. One way to maintain selected polymorphisms is strong negative selection ( = 4N_es >> 1) against nonsynonymous mutations (NORDBORG et al. 1996). LOEWE and CHARLESWORTH (2007) estimated that an average gene in D. melanogaster carries about two nonsynonymous mutations and that 87% of these are strongly selected ( = 4N_es >> 1). They estimated that background selection on this variation would produce substantial opportunity to produce Hill–Robertson interference by reducing N_e by at least 63%.

Another way to maintain selected polymorphisms occurs when selection is at an intermediate level relative to population size, i.e., when the N_e-scaled strength of selection ( = 4N_es) is not too small or too large (i.e., 0.25 < < 2.5) (OHTA and KIMURA 1971; OHTA 1972, 1995; LI 1987). In D. melanogaster, COMERON and GUTHRIE (2005) estimated for preferred codons to be between 1.3 and 3.1, with the small estimates associated with larger coding sequences that are uninterrupted by introns. In D. pseudoobscura and D. miranda, LOEWE and CHARLESWORTH (2007) estimated that a substantial proportion (13%) of nonsynonymous mutations also had 1. In humans EYRE-WALKER and KEIGHTLEY (2007) estimated than about a third of nonsynonymous mutations have 0 < < 10, indicating that a substantial proportion would be expected to be in mutation–selection–drift balance. These data suggest that in species ranging from Drosophila with very large N_e (10⁶) to humans with relatively small N_e (10⁴), there is abundant variation from harmful mutations that can provide the localized variation needed for the hitchhiking of DSB inducers. The finding by M. Noor's group (CIRULLI et al. 2007)—that the heat of hotspots is strongly positively correlated with the prevalence of tightly linked preferred codons in D. pseudoobscura—supports the conclusion that sufficient Hill–Robertson interference is present in the genome of D. pseudoobscurra to make selection at tightly linked sites sufficiently strong that a realized response to selection is manifest. This is the best evidence to date that the hitchhiking effect may be able to operate over the short distances surrounding hotspots.

There is also recent evidence that selection is effective for preferred codons in species with smaller N_e, like small mammals. RESCH et al. (2007) detected purifying selection acting on synonymous sites in 28% of the 1562 genes that were compared between mice and rats and directional selection in 12% of the same group of genes. KONDRASHOV et al. (2007) and PARMLEY and HURST (2007) provide preliminary evidence that will at least sometimes be in the required range (0.25–2.5) for preferred codons to contribute to the hitchhiking effect on hotspots in humans. Given the much smaller N_e of larger organisms like humans, it seems likely that more strongly selected sites, such as indels or SNPs affecting nonsynonymous codons and regulatory regions, rather than preferred codons, will be the primary variation contributing to the hitchhiking effect on hotspots. It remains to be demonstrated, however, whether synonymous codon variation or small-effect mutations like SNPs in regulatory regions and nonsynonymous codons are generally of suitable density and selective strength to promote the antagonistic coevolution between DSB inducers and DSB-cut sites that we have described here—but the preliminary evidence suggests that this may be so.

Second, there must be a buildup of sufficient Hill–Robertson interfering disequilibrium around coldspots. Theoretical work predicts that this prerequisite will be met [i.e., that reduced recombination among blocks of weakly selected sites leads to increased Hill–Robertson-interfering disequilibrium (MCVEAN and CHARLESWORTH 2000; COMERON and KREITMAN 2002; LOEWE and CHARLESWORTH 2007] in species like Drosophila, but additional empirical work testing this prerequisite in other species is still needed. Nonetheless, compliance with these prerequisites is supported by the recent findings that enhancers of local recombination rate [hotspots (CIRULLI et al. 2007) and introns (KLIMAN et al. 2003; COMERON and GUTHRIE 2005)] were associated with increased prevalence of preferred codons.

The third prerequisite is that DSB inducers be closely linked to the DNA-cut region but not always located within the cut DNA. When a DSB inducer is trans acting (cuts the other chromosome), this prerequisite is necessarily achieved. Although available evidence indicates that trans-acting DSB inducers are possible (e.g., YAO and SCHNABLE 2005; BAUDAT and DE MASSY 2007), most empirical work suggests that cis-acting factors, such as binding of transcription factors, and GC content are major factors influencing the operation of a DSB inducer (NISHANT and RAO 2006). There is substantial evidence that open chromatin structure is a key feature for hotspot activity since most hotspots are hypersensitive to DNase 1 (NISHANT and RAO 2006). However, open chromatin structure appears to be a necessary, but not a sufficient, condition for hotspot activity (e.g., BORDE et al. 1999), and in both yeast and mice a single SNP can convert a hotspot into a coldspot. Also, in yeast it was found that an 8.5-kb recombinational reporter was strongly influenced by its chromosomal location, indicating that a DSB-cut region can be strongly influenced by flanking sequences (BORDE et al. 1999). Collectively these observations suggest that DSB inducers that promote cuts at nearby DSB-cut regions (that do not cause the DSB inducer itself to be lost via biased gene conversion each time a DSB is made) may be common and therefore that there is substantial potential for antagonistic coevolution between them.

Hotspots vs. introns as modifiers of local recombination rate:

There is now substantial evidence that introns act as recombinational enhancers within coding sequences that reduce Hill–Robertson interference and thereby increase the prevalence of favored genetic variation, such as preferred codons in Drosophila (MCVEAN and CHARLESWORTH 2000; COMERON and KREITMAN 2002; LOEWE and CHARLESWORTH 2007). The interplay between indels, genetic drift, and Hill–Robertson interference suggests that there may be cyclic variation in the size of introns, at least in species with large N_e. If a deletion that markedly reduced the length of an intron drifted to fixation, then Hill–Robertson interference would be expected to accrue, which would lead to selection [i.e., s_e(GB) > 0 via the hitchhiking effect] favoring the invasion of an insertion mutation that created a longer intron. Recombinational hotspots represent an analogous process on a larger scale, but with the caveat that the hotspots simultaneously reduce Hill–Robertson interference and lead to biased gene conversion against the cut DNA strand. So unlike the case of introns, the cost (via biased gene conversion) to inducing recombination creates the context for antagonistic coevolution between the DSB inducer and its DSB-cut regions, so long as the DSB inducer cuts neighboring DNA and thereby escapes biased gene conversion at least some of the time.

Conclusions and future directions:

Past models of the evolution of recombinational hotspots have either not incorporated natural selection as a factor contributing to their evolution (ARCHETTI 2003; CALABRESE 2007) or concluded that the inclusion of selection did not help explain their evolution (BOULTON et al. 1997; PINEDA-KRCH and REDFIELD 2005; COOP and MYERS 2007). BOULTON et al. (1997) and PINEDA-KRCH and REDFIELD (2005) specifically showed that hotspots were unlikely to persist at equilibrium due to selection on their flanking sites. Our hypothesis evaluates a nonequilibrium solution to the hotspot paradox by showing that hotspots can, in principle, achieve a dynamically steady state in the genome due to a birth and death cycle that is driven by antagonistic coevolution between neighboring genomic regions, rather than by individual hotspots persisting at equilibrium. By integrating the hitchhiking effect in response to Hill–Robertson interference at closely linked sites, as well as recent advances in our understanding of selection on preferred codons and nonsynonymous codons, a plausible case can be made that natural selection is contributing importantly to the evolution of hotspots. Although natural selection can be important in offsetting the cost of biased gene conversion when DSB inducers are located in the center of the DSB-cut region, the highest potential for natural selection to contribute to the birth and death of hotspots is via antagonistic coevolution between distinct, but tightly linked, DSB inducers and DSB-cut regions. The critical next step in evaluating our cut-thy-neighbor model of hotspot evolution is the fuller characterization of the molecular properties that influence DSB induction and cut regions and to better empirically quantify the strength of selection on nascent hotspots that is generated by the hitchhiking effect.

We thank Aneil Agrawal, Sergey Gavrilets, Laurence Hurst, Tristan Long, Kathryn Schoenrock, and Robert Trivers for constructive comments on a previous version of this manuscript. U.F. was funded by a Wenner-Gren Foundations postdoctoral scholarship and W.R.R. was supported by the National Science Foundation (DEB-0128780 and DEB-0111613).

ARCHETTI, M., 2003 A selfish origin for recombination. J. Theor. Biol. 223: 335–346.[CrossRef][Medline]

BARTON, N. H., 1995a A general-model for the evolution of recombination. Genet. Res. 65: 123–144.[Medline]

BARTON, N. H., 1995b Linkage and the limits to natural-selection. Genetics 140: 821–841.[Abstract]

BARTON, N. H., and S. P. OTTO, 2005 Evolution of recombination due to random dirift. Genetics 169: 2353–2370.[Abstract/Free Full Text]

BAUDAT, F., and B. DE MASSY, 2007 Cis- and trans-acting elements regulate the mouse Psmb9 meiotic recombination hotspot. PLoS Genet. 3: 1029–1039.

BORDE, V., T. C. WU and M. LICHTEN, 1999 Use of a recombination reporter insert to define meiotic recombination domains on chromosome III of Saccharomyces cerevisiae. Mol. Cell. Biol. 19: 4832–4842.[Abstract/Free Full Text]

BOULTON, A., R. S. MYERS and R. J. REDFIELD, 1997 The hotspot conversion paradox and the evolution of meiotic recombination. Proc. Natl. Acad. Sci. USA 94: 8058–8063.[Abstract/Free Full Text]

CALABRESE, P., 2007 A population genetics model with recombination hotspots that are heterogeneous across the population. Proc. Natl. Acad. Sci. USA 104: 4748–4752.[Abstract/Free Full Text]

CHAMARY, J. V., J. L. PARMLEY and L. D. HURST, 2006 Hearing silence: non-neutral evolution at synonymous sites in mammals. Nat. Rev. Genet. 7: 98–108.[CrossRef][Medline]

CIRULLI, E. T., R. M. KLIMAN and M. A. F. NOOR, 2007 Fine-scale crossover rate heterogeneity in Drosophila pseudoobscura. J. Mol. Evol. 64: 129–135.[CrossRef][Medline]

COMERON, J. M., and T. B. GUTHRIE, 2005 Intragenic Hill-Robertson interference influences selection intensity on synonymous mutations in Drosophila. Mol. Biol. Evol. 22: 2519–2530.[Abstract/Free Full Text]

COMERON, J. M., and M. KREITMAN, 2002 Population, evolutionary and genomic consequences of interference selection. Genetics 161: 389–410.[Abstract/Free Full Text]

COMERON, J. M., M. KREITMAN and M. AGUADE, 1999 Natural selection on synonymous sites is correlated with gene length and recombination in Drosophila. Genetics 151: 239–249.[Abstract/Free Full Text]

COOP, G., and S. R. MYERS, 2007 Live hot, die young: transmission distortion in recombination hotspots. PLoS Genet. 3: 377–386.

CROW, J. F., and M. KIMURA, 1970 An Introduction to Population Genetics Theory. Harper & Row, New York.

DE MASSY, B., 2003 Distribution of meiotic recombination sites. Trends Genet. 19: 514–522.[CrossRef][Medline]

EWENS, W. J., 2004 Mathematical Population Genetics. Springer-Verlag, New York.

EYRE-WALKER, A., and P. D. KEIGHTLEY, 2007 The distribution of fitness effects of new mutations. Nat. Rev. Genet. 8: 610–618.[CrossRef][Medline]

FELSENSTEIN, J., 1974 The evolutionary advantage of recombination. Genetics 78: 737–756.[Abstract/Free Full Text]

FELSENSTEIN, J., and S. YOKOYAMA, 1976 The evolutionary advantage of recombination. II. Individual selection for recombination. Genetics 83: 845–859.[Abstract/Free Full Text]

GUTZ, H., and J. F. LESLIE, 1976 Gene conversion—hitherto overlooked parameter in population-genetics. Genetics 83: 861–866.[Abstract/Free Full Text]

HILL, W. G., and A. ROBERTSON, 1966 The effect of linkage on limits to artificial selection. Genet. Res. 8: 269–294.[Medline]

ILES, M. M., K. WALTERS and C. CANNINGS, 2003 Recombination can evolve in large finite populations given selection on sufficient loci. Genetics 165: 2249–2258.[Abstract/Free Full Text]

JEFFREYS, A. J., and R. NEUMANN, 2002 Reciprocal crossover asymmetry and meiotic drive in a human recombination hot spot. Nat. Genet. 31: 267–271.[CrossRef][Medline]

KEIGHTLEY, P. D., and S. P. OTTO, 2006 Interference among deleterious mutations favours sex and recombination in finite populations. Nature 443: 89–92.[CrossRef][Medline]

KLIMAN, R. M., and J. HEY, 2003 Hill-Robertson interference in Drosophila melanogaster: reply to Marais, Mouchiroud and Duret. Genet. Res. 81: 89–90.[CrossRef][Medline]

KLIMAN, R. M., N. IRVING and M. SANTIAGO, 2003 Selection conflicts, gene expression, and codon usage trends in yeast. J. Mol. Evol. 57: 98–109.[CrossRef][Medline]

KONDRASHOV, F. A., T. A. GURBICH and P. K. VLASOV, 2007 Selection for functional uniformity of tuf duplicates in gamma-proteobacteria. Trends Genet. 23: 215–218.[CrossRef][Medline]

LI, W. H., 1987 Models of nearly neutral mutations with particular implications for nonrandom usage of synonymous codons. J. Mol. Evol. 24: 337–345.[CrossRef][Medline]

LOEWE, L., and B. CHARLESWORTH, 2007 Background selection in single genes may explain patterns of codon bias. Genetics 175: 1381–1393.[Abstract/Free Full Text]

LOEWE, L., B. CHARLESWORTH, C. BARTOLOME and V. NOEL, 2006 Estimating selection on nonsynonymous mutations. Genetics 172: 1079–1092.[Abstract/Free Full Text]

MARAIS, G., 2003 Biased gene conversion: implications for genome and sex evolution. Trends Genet. 19: 330–338.[CrossRef][Medline]

MARTIN, G., S. P. OTTO and T. LENORMAND, 2006 Selection for recombination in structured populations. Genetics 172: 593–609.[Abstract/Free Full Text]

MAYNARD SMITH, J., 1976 What determines rate of evolution. Am. Nat. 110: 331–338.[CrossRef]

MCVEAN, G. A. T., and B. CHARLESWORTH, 2000 The effects of Hill-Robertson interference between weakly selected mutations on patterns of molecular evolution and variation. Genetics 155: 929–944.[Abstract/Free Full Text]

MYERS, S., L. BOTTOLO, C. FREEMAN, G. MCVEAN and P. DONNELLY, 2005 A fine-scale map of recombination rates and hotspots across the human genome. Science 310: 321–324.[Abstract/Free Full Text]

MYERS, S., C. C. A. SPENCER, A. AUTON, L. BOTTOLO, C. FREEMAN et al., 2006 The distribution and causes of meiotic recombination in the human genome. Biochem. Soc. Trans. 34: 526–530.[CrossRef][Medline]

NAGYLAKI, T., 1983 Evolution of a finite population under gene conversion. Proc. Natl. Acad. Sci. USA Biol. Sci. 80: 6278–6281.[CrossRef]

NICOLAS, A., D. TRECO, N. P. SCHULTES and J. W. SZOSTAK, 1989 An initiation site for meiotic gene conversion in the yeast Saccharomyces cerevisiae. Nature 338: 35–39.[CrossRef][Medline]

NISHANT, K. T., and M. R. S. RAO, 2006 Molecular features of meiotic recombination hot spots. BioEssays 28: 45–56.[CrossRef][Medline]

NORDBORG, M., B. CHARLESWORTH and D. CHARLESWORTH, 1996 The effect of recombination on background selection. Genet. Res. 67: 159–174.[Medline]

OHTA, T., 1972 Fixation probability of a mutant influenced by random fluctuation of selection intensity. Genet. Res. 19: 33–38.[Medline]

OHTA, T., 1995 Synonymous and nonsynonymous substitutions in mammalian genes and the nearly neutral theory. J. Mol. Evol. 40: 56–63.[CrossRef][Medline]

OHTA, T., and M. KIMURA, 1971 Linkage disequilibrium between two segregating nucleotide sites under the steady flux of mutations in a finite population. Genetics 68: 571–580.[Free Full Text]

OTTO, S. P., and N. H. BARTON, 1997 The evolution of recombination: removing the limits to natural selection. Genetics 147: 879–906.[Abstract]

OTTO, S. P., and N. H. BARTON, 2001 Selection for recombination in small populations. Evolution 55: 1921–1931.[CrossRef][Medline]

PAL, C., and L. D. HURST, 2003 Evidence for co-evolution of gene order and recombination rate. Nat. Genet. 33: 392–395.[CrossRef][Medline]

PALSSON, S., 2002 Selection on a modifier of recombination rate due to linked deleterious mutations. J. Hered. 93: 22–26.[Abstract/Free Full Text]

PARMLEY, J. L., and L. D. HURST, 2007 How do synonymous mutations affect fitness? BioEssays 29: 515–519.[CrossRef][Medline]

PETES, T. D., 2001 Meiotic recombination hot spots and cold spots. Nat. Rev. Genet. 2: 360–369.[CrossRef][Medline]

PINEDA-KRCH, M., and R. J. REDFIELD, 2005 Persistence and loss of meiotic recombination hotspots. Genetics 169: 2319–2333.[Abstract/Free Full Text]

RESCH, A. M., L. CARMEL, L. MARINO-RAMIREZ, A. Y. OGURTSOV, S. A. SHABALINA et al., 2007 Widespread positive selection in synonymous sites of mammalian genes. Mol. Biol. Evol. 24: 1821–1831.[Abstract/Free Full Text]

ROZE, D., and N. H. BARTON, 2006 The Hill-Robertson effect and the evolution of recombination. Genetics 173: 1793–1811.[Abstract/Free Full Text]

WARNECKE, T., and L. D. HURST, 2007 Evidence for a trade-off between translational efficiency and splicing regulation in determining synonymous codon usage in Drosophila melanogaster. Mol. Biol. Evol. 24: 2755–2762.[Abstract/Free Full Text]

YAO, H., and P. S. SCHNABLE, 2005 Cis-effects on meiotic recombination across distinct a1-sh2 intervals in a common Zea genetic background. Genetics 170: 1929–1944.[Abstract/Free Full Text]

Communicating editor: K. G. GOLIC

International Access Link

Online ISSN: 1943-2361  Print ISSN: 0016-6731
Copyright 2009 by the Genetics Society of America
phone: 412-268-1812   fax: 412-268-1813   email: genetics-gsa{at}andrew.cmu.edu