DISCUSSION

In theory, the relative contributions of background and hitchhiking selection can be determined by comparing recombination rates to genetic diversity on the basis of markers that evolve with different rates (Slatkin 1995). However, demography can obscure the relationship between recombination and diversity. Here we have measured genetic diversity for three types of molecular markers in the hope that comparisons among markers would provide insight into the forces shaping maize genetic diversity. Sampling was identical for the three marker types, and hence information among marker types is directly comparable. Recombination was measured both by the population-recombination parameter C and by a quantitative cytogenetic map of chromosome 1.

The quantitative cytogenetic map provides estimates of recombination rate (R) that are related to physical distance along SC1; R is the first quantitative measure of recombination in maize on a chromosomal, rather than a genic, scale. The distribution of R along SC1 indicates that the frequency of exchange per physical unit is reduced in centromeric regions relative to distal chromosomal regions (Figure 2), similar to centromeric suppression observed in other organisms (see Jones 1984 and Resnick 1987 for reviews), including Drosophila (Hudson and Kaplan 1995) and cultivated tomato (Sherman and Stack 1995). The pattern observed in maize chromosome 1 is similar to that reported for large grass genomes such as wheat and barley, in which recombination primarily occurs along the distal half of the chromosomal arm (Gillet al. 1996; Kunzelet al. 2000). In contrast, the region of centromeric repression is relatively small in rice (Chenget al. 2001).

The distribution of R also suggests substantial heterogeneity in recombination along chromosomal arms (Figure 2). Although the magnitude and scale of recombination needs to be characterized further, heterogeneity in R is consistent with the observation in barley that recombination is mainly confined to a few small areas spaced by large segments in which recombination is severely suppressed (Kunzelet al. 2000). Similar recombination hotspots have been previously described in maize (Dooner 1986; Civardiet al. 1994; Okagaki and Weil 1997; Fuet al. 2002).

The correlation between SNP diversity and recombination estimates: One striking result is thatĈ correlates with θ^ for two of three C estimators. It is unclear why the third estimator, based on Hudson’s (2001) method, behaves differently than the first two, but the positive correlation betweenĈ and θ^ in two cases indicates that the correlation is not solely an artifact of the estimator (Ĉ_Hud87) used in the previous study (Tenaillonet al. 2001). In contrast toĈ and θ^ _, we detect no correlation between θ^ and R or betweenĈ and R (Figure 3).

Given these results, it is important first to consider differences between C and R. One obvious difference is that the two parameters differ in spatial scale. R is estimated on a chromosomal scale and therefore reflects an “average” recombination rate over large chromosomal regions. In contrast, C is estimated for a particular genetic locus. Maize contains recombination hotspots, particularly in genic regions (Civardiet al. 1994; Fuet al. 2002), and it is therefore possible that C more accurately incorporates information about recombination on the “local” spatial scale at which θ is measured.

More importantly, R and C measure different quantities. Both R and C describe recombination to some extent, but R measures only the recombination rate per physical distance; it is unaffected by population history, selection, and demography. In contrast, C is scaled by population size N, and it is inversely related to LD. Like LD, C is affected by population admixture, population subdivision, fluctuations in population size, and selection, in addition to recombination (reviewed in Pritchard and Przeworski 2001). The lack of correlation between R andĈ may indicate that selection or demographic factors contribute to an uncoupling between LD and recombination.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

—Correlations between two estimates of recombination (R andĈ_Hud87) and between estimates of recombination and diversity. b-e are based on 18 genetic loci, but f-i contain data from all genetic loci (see results). Regression lines, Pearson correlation coefficients (r), and P values are given.

What could be the evolutionary forces causing a correlation betweenĈ and θ^ _? The correlation could be primarily a function of selection. Under this scenario, hitchhiking or background selection that decreases θ acts in a similar fashion on C. It is clear that C can be affected by selection. For example, balancing selection decreases C (Pritchard and Przeworski 2001); C also decreases briefly after a selective sweep, but some estimators may not detect this effect (Przeworski 2002). However, if background selection or selective sweeps are prominent enough to cause, in some unknown fashion, a correlation betweenĈ and θ^ _, one expects a correlation between R and θ, as documented in several other systems (Begun and Aquadro 1992; Nachman 1997; Nachmanet al. 1998). Although we cannot be certain that our estimates of R accurately reflect recombination in the local regions of the 21 genes, we do not observe a correlation between R and θ^ in maize (Figure 3). Furthermore, under a hitchhiking model Tajima’s D should correlate with recombination rate (Bravermanet al. 1995; Andolfatto and Przeworski 2001), but Tajima’s D for the 21 genetic loci is not correlated with either R (r =-0.06; P = 0.60) orĈ (r =-0.38; P = 0.93). Altogether, there is no convincing evidence that hitchhiking or background selection contributes to the correlation betweenĈ and θ^ _. It is important to note, however, that these results do not imply that background and hitchhiking selection are not acting to shape maize SNP diversity. The signature of background or hitchhiking selection could be overridden by other factors.

Both C and θ (= 4Nμ) contain historical information about population size. Both are also estimated from SNPs that evolve at an estimated rate of ∼10^-9 substitutions per site per year (Gautet al. 1996) and therefore encompass relatively long time frames. Some SNPs have been retained in Zea populations for 1 million years or more (Gaut and Clegg 1993). As a result, the time frame encompassed by both C and θ exceeds maize domestication ∼7500 (Iltis 1983) to ∼9000 years ago (Matsuokaet al. 2002b). Domestication was associated with a bottleneck that decreased SNP diversity in maize ∼20% on average relative to its wild ancestor (Zhanget al. 2002). There have also been substantial demographic events since domestication, such as the geographic patterning of extant maize races (e.g., Matsuokaet al. 2002b).

It is not yet clear, however, how demographic events, like a domestication bottleneck or geographic subdivision, affect C and θ jointly. One possibility is that population size N varies among loci because gene flow and other demographic factors vary from locus to locus, both within maize and among its wild relatives (as in the D. pseudoobscura complex; Wanget al. 1997; Machadoet al. 2002). If demographic effects vary among loci, they could contribute substantially to a correlation between C and θ through N. Variation in N among loci can establish strong positive correlations between C and θ, even in the absence of correlations between θ and c. Simulations with 21 loci suggest that N can vary <10-fold and establish a correlation between C and θ (data not shown). To explore the effect of demography more fully, it will be helpful to have some knowledge of diversity in maize prior to domestication and also of divergence population genetics (Klimanet al. 2000) in the genus Zea. We are in the process of gathering empirical data from wild relatives and will address the effect of demography on C and θ more thoroughly in future work.

Microsatellite diversity: We identified 47 microsatellite loci in our data, and 33 of these loci were polymorphic. Levels of diversity in these loci, as measured by H_microsat, are positively correlated with R. There are at least three possible explanations for this correlation.

The first explanation is based on sampling—i.e., our sample may contain rapidly evolving loci in regions of high recombination by chance alone. This scenario is particularly plausible because the microsatellites in this study likely evolve with different mutation rates. Microsatellite mutation rates (μ) vary considerably by repeat motif (Chakrabortyet al. 1997; Schug et al. 1997, 1998), length (Schlottereret al. 1998; Schug et al. 1998b; Udupa and Baum 2001), and base composition (Schlotterer and Tautz 1992; Glennet al. 1996; Bachtroget al. 2000). For example, μ is estimated to be ∼7.7 × 10^-4 mutations per generation for dinucleotide repeats in maize but <5 × 10^-5 for longer repeat motifs (Vigourouxet al. 2002).

To examine whether any particular class of microsatellite is driving the correlation between H_microsat and R, we partitioned microsatellite loci into different classes by repeat type, including perfect mono-, di-, tri-, and hexanucleotide repeats, as well as compound + imperfect repeats (Figure 4). The only class exhibiting a positive and significant correlation between H_microsat and R was the compound + imperfect class (Figure 4), but this correlation was not significant after multiple test correction. However, four of the five classes exhibited a positive correlation between H_microsat and R (Figure 4), suggesting that a positive correlation with R may be a general property of the microsatellite loci in our sample. The mononucleotide repeat class is particularly interesting, both because these loci may evolve rapidly and because they are primarily located in regions with high R (Tables 1 and 3). The mononucleotide class is positively but not significantly correlated with R (r = 0.46; P = 0.14), but the overall correlation between H_microsat and R remains when this class is removed from analysis (r = 0.42; P = 0.02). Thus, it does not appear that the overall correlation between H_microsat and R is driven either by one particular class of microsatellite or by the chance location of rapidly evolving microsatellites (like mononucleotide repeats) in high R regions.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

—Correlations between microsatellite diversity (H_microsat) and R for each microsatellite repeat class. Only two polymorphic loci were available from the tetranucleotide class and none from the pentanucleotide class. Compound and imperfect loci were combined because there is little a priori information as to their mutation rates. The regression line, Pearson correlation coefficient (r), and P value are given for each microsatellite class.

A second possibility for the correlation between H_microsat and R is that recombination is itself mutagenic, thereby causing microsatellite polymorphisms. For example, human data suggest that recombination can lead to the contraction and expansion of trinucleotide repeats (Richard and Paques 2000). The effect of recombination on microsatellite diversity needs to be investigated further, but mutagenic effects of recombination could underlie the correlation between H_microsat and R.

A third possibility is that the correlation is a property of the relationship between recombination and selection. To discuss this possibility, it is first important to note that microsatellite mutation rates have been measured in many organisms, including humans (Weber and Wong 1993; Xuet al. 2000), Drosophila (Vazquezet al. 2000), chickpea (Udupa and Baum 2001), and maize (Vigourouxet al. 2002). In all of these organisms, microsatellites mutate at a rate μ> 10^-6 mutations per generation. Thus, the microsatellites in this study probably mutate at least three orders of magnitude more rapidly than SNPs. The consequence of high mutation rates is profound. Microsatellites are expected to quickly approach an equilibrium between mutation and drift (Slatkin 1995), and they recover rapidly from demographic and selective events. For example, microsatellites in Drosophila, which mutate relatively slowly at μ= 5.1 × 10^-6 (Vazquezet al. 2000) compared to plant microsatellites, are estimated to recover from selective sweeps in <1000 years (Nurminsky 2001). Although the number of Drosophila generations in 1000 years likely exceeds the number of maize generations since domestication, it is possible that some maize microsatellites may have recovered, at least partially, from the effect of the domestication bottleneck ∼7500 (Iltis 1983) to ∼9000 years ago (Matsuokaet al. 2002a). If this is true, it is possible that the signature of ongoing hitchhiking or background selection is no longer dominated by a past demographic event (i.e., a domestication bottleneck) in microsatellite loci as it may be in SNPs.

Finally, we note that comparisons between V and R do not yield a positive correlation (Figure 3). However, when V is based on repeat number, rather than allele size, results with H_microsat and V are more comparable. It is desirable to use repeat number, as opposed to allele size, because V based on repeat number is not biased by repeat length. However, V based on repeat number cannot be calculated for several microsatellite loci in our sample because the repeats were imperfect, were compound, or did not evolve in stepwise fashion. For the 21 polymorphic perfect loci that evolve in stepwise fashion (Table 1), V based on repeat number is positively, but not significantly, correlated with R (r = 0.19; P = 0.23). When tb1 is dropped from consideration, the correlation is significantly positive (r = 0.47; P = 0.024), and this result is comparable to that we obtained with H_microsat. All of our analyses with V—whether based on allele size or repeat number—were heavily influenced by the outlying tb1 microsatellite locus. Altogether, the reliance on tb1, the bias due to repeat length for V based on allele size and the dependence on stepwise mutations for V based on repeat number, diminish the value of V as a measure of microsatellite diversity for these data.

Indel diversity: Indel diversity in maize is marked by a size distribution that is heavily skewed toward small indels (1-5 bp), with a few large (>100 bp) indels marking the extreme tail of the distribution (Figure 1). Similar distributions have been reported for mammalian and Drosophila nuclear DNA (Gu and Li 1995; Bergman and Kreitman 2001), and hence maize is not unique in having a preponderance of small indels. Indel polymorphism was not correlated withĈ or R (Figure 3). Because little is known about indel mutation rates and how μ varies among different indel sizes, it is difficult to interpret the lack of correlation.

It is perhaps more interesting that the population frequency of indels is skewed by size. In our sample, large indels are on average less frequent in the population sample than small indels, suggesting that large indels are slightly deleterious. The 10 large indels also have a lower Tajima’s D value than the small indels. Of these 10, only 2 are clearly associated with coding DNA (adh1 and csu381; Table 2); the rest are located in anonymous RFLP marker regions. These results raise an interesting paradox. Greater than 50% of the maize genome consists of retrotransposons (SanMiguelet al. 1996). Given the preponderance of transposable elements in the maize genome, it seems unlikely that large indels are usually strongly deleterious, yet these population data suggest they are measurably deleterious. The resolution to this problem consists of two components. First, the vast majority of the maize genome consists of retrotransposons that insert into one another (SanMiguelet al. 1996); presumably the targeting of retrotransposons into nonessential genic regions is evolutionarily favorable for element proliferation. Because of this targeting, retrotransposons may be under different evolutionary dynamics from the indels in our sample, none of which are retrotransposons. Second, MITEs and Ds elements, the only identifiable elements in our study, preferentially insert into transcribed regions (Bennetzen 2000), suggesting some of our genetic loci are near coding regions where insertions are more likely to be deleterious.

The forces affecting genetic diversity in maize: This study offers several insights into the forces contributing to genetic diversity in maize. First, there is no evidence that R and θ^ are positively correlated, as expected under hitchhiking and background selection models. Assuming R provides reasonable estimates of recombination, it thus appears likely that other effects—perhaps demography—drive the correlation betweenĈ and θ^ _. Second, the correlation between H_microsat and Rˆ suggests either that recombination is mutagenic for microsatellite loci or that a pattern of hitchhiking or background selection is evident in markers that may be partially recovered from some historical events. However, one must question whether the maize genome has sufficient gene density to permit extensive background selection, because the strength of background selection depends on gene density (Payseur and Nachman 2000). The lower the gene density, the lower the deleterious mutation rate and hence the lower the strength of background selection. In maize, the gene space is restricted to only 20% of the genome (Carelset al. 1995; Barakatet al. 1997) and consequently the low recombination regions of maize, where the effects of background selection are most pronounced, may contain very few genes.