The Relationship of Nucleotide Polymorphism, Recombination Rate and Selection in Wild Tomato Species

Abstract

We analyzed the effects of mating system and recombination rate on single nucleotide polymorphisms using 14 single-copy nuclear loci from single populations of five species of wild tomatoes (Solanum section Lycopersicon). The taxa investigated comprise two self-compatible (SC) and three self-incompatible (SI) species. The observed reduction in nucleotide diversity in the SC populations compared to the SI populations is much stronger than expected under the neutral effects of the mating system on effective population size. Importantly, outgroup sequences available for 11 of the 14 loci yield strong positive correlations between silent nucleotide diversity and silent divergence, indicative of marked among-locus differences in mutation rates and/or selective constraints. Furthermore, using a physical estimate of local recombination rates, we find that silent nucleotide diversity (but not divergence) is positively correlated with recombination rate in two of the SI species. However, this correlation is not nearly as strong as in other well-characterized species (in particular, Drosophila). We propose that nucleotide diversity in Lycopersicon is dominated mainly by differences in neutral mutation rates and/or selective constraints among loci, demographic processes (such as population subdivision), and background selection. In addition, we hypothesize that the soil seed bank plays an important role in the maintenance of the large genetic diversity in the SI species (in particular L. peruvianum).

WHETHER diversity levels in the genomes of animals and plants are shaped mainly by neutral processes or selection is a recurrent theme in population genetics. In many organisms, genes in regions of low recombination exhibit reduced levels of DNA polymorphism, while divergence among species remains relatively unaffected. These observations may be explained by genetic hitchhiking, involving advantageous mutations sweeping through a population (Maynard Smith and Haigh 1974) and/or by background selection eliminating unconditionally deleterious mutations (Charlesworth et al. 1993, 1995). Depending on the effective recombination rate in the regions subjected to either of the above processes and on the strength of natural selection, linked neutral variation is reduced under both selection scenarios, while interspecific divergence is not influenced.

A positive correlation between DNA diversity and recombination rate has been found in many organisms, both animals (Aguadé et al. 1989; Stephan and Langley 1989; Begun and Aquadro 1992; Nachman 1997; Nachman et al. 1998) and plants (Dvorak et al. 1998; Kraft et al. 1998; Stephan and Langley 1998). Such patterns were usually attributed to the action of natural selection. However, neutral explanations for an observed positive correlation between recombination and diversity have also been proposed, when, in addition to intraspecific variation, divergence also scaled with recombination rate (e.g., Wang et al. 1997; Hellmann et al. 2003).

Among plants, maize (Zea mays ssp. mays L.) and its wild ancestor were the focus of recent studies examining the interplay between nucleotide variability, recombination, and selection (Tenaillon et al. 2001, 2002, 2004). In maize, single nucleotide polymorphism (SNP) and recombination rate, as measured by a quantitative recombination nodule map, were found to be uncorrelated. However, a positive correlation was found for one of the sequence-based estimators of the population recombination rate, which estimates recombination rate on the basis of linkage disequilibrium (Hudson 1987). Clearly, a thorough understanding of the evolutionary factors underlying the relationship between recombination rates and levels of nucleotide diversity in plants—and even ascertainment of such an association—requires much more sequence data obtained from genomic regions spanning a range of known recombination rates.

Here we report our findings from a multilocus survey of another well-studied plant model system, the clade Lycopersicon, encompassing the cultivated tomato and its progenitor, as well as several other wild species. Native to western South America, the wild relatives of the cultivated tomato represent suitable model organisms to study genealogical footprints of speciation and demographic processes (Städler et al. 2005). The species complex appears to be relatively young yet ecologically well differentiated (Rick and Lamm 1955; Rick 1979; Taylor 1986), and the genetics of the cultivated tomato have long been under intense investigation (Rick and Yoder 1988; Tanksley et al. 1992; Ganal et al. 1998). Collectively, the wild taxa represent a wide spectrum of mating system variation, including both self-compatible, partially self-fertilizing and self-incompatible, obligately outcrossing species (Rick et al. 1976, 1977, 1979; Rick 1983; Igic et al. 2004).

In an exploratory study based on five nuclear genes, Baudry et al. (2001) found very strong effects of the mating system on levels of silent nucleotide diversity among five wild tomato species, while levels of recombination did not have significant effects. A shortcoming of this study, however (shared by many other such studies), was the lack of sufficiently diverged outgroup sequences and hence the inability to distinguish between intra- and interspecific effects on sequence evolution. We have obtained sequence data for nine additional nuclear genes and outgroup sequences for 11 of the 14 loci. Using this much-expanded data set, we are now able to explore the interrelationships among nucleotide diversity, interspecific divergence, genomic recombination rates, and the possible role of natural selection in more depth and with much more confidence. With the addition of 9 more loci, we are also able to investigate the effect of the mating system on levels of DNA diversity in more detail.

Our study addresses the following questions: (i) What is the effect of the mating system on levels of silent nucleotide diversity?, (ii) Are nucleotide diversity and interspecific divergence correlated with each other, as expected under neutral evolution?, (iii) Are recombination rate and nucleotide diversity and/or interspecific divergence correlated with each other?, and (iv) What are the relative strengths of natural selection vs. demographic (and possibly other, neutral) processes in shaping genetic variation in wild tomatoes, and how do these compare to other model organisms for which the data are currently much more complete (in particular Drosophila)?

MATERIALS AND METHODS

Plant material:

The clade Lycopersicon encompasses the cultivated tomato Lycopersicon esculentum and its wild relatives. Although the former genus Lycopersicon has been taxonomically reassigned to Solanum section Lycopersicon (Spooner et al. 1993; Peralta and Spooner 2001), we will retain the old nomenclature for consistency with our and other authors' previous work. Native to western South America, wild tomatoes are herbaceous perennials, with individual plants often bearing receptive, insect-pollinated flowers and fruits simultaneously. Annual recruitment of seedlings may occur in more mesic environments, but seedling establishment might be restricted to favorable years in more xeric habitats (e.g., southern Peru and northern Chile; Rick 1979; T. Städler, personal observations).

Our sampling scheme encompasses a wide spectrum of mating systems in the clade Lycopersicon, from self-compatible (SC) to self-incompatible (SI) species (Igic et al. 2004). The species in our study include L. peruvianum (= Solanum peruvianum; SI; accession no. LA2744) from Tarapaca and L. chilense (= S. chilense; SI; LA2884) from Antofagasta, both located in northern Chile. Our sample of L. hirsutum (= S. habrochaites; SI; LA1775) is from Ancash in central Peru. For geographical details on the distribution and locations of the SI samples, see Städler et al. (2005). We also sampled the following SC species: L. chmielewskii (= S. chmielewskii; SC; LA3653) from Apurimac, south-central Peru, and L. pimpinellifolium (= S. pimpinellifolium; SC; LA1583) from Lambayeque, northern Peru. L. pimpinellifolium is distributed along the coast of Peru and Ecuador and restricted to Andean river valleys (Taylor 1986; Caicedo and Schaal 2004), whereas L. chmielewskii prefers more mesic habitats in interior Peru with a fairly narrow geographic distribution (Rick et al. 1976). Although no estimates of outcrossing rates are available for our two SC source populations, morphological data on flower size and the degree of stigma exsertion indicate facultative outcrossing (http://tgrc.ucdavis.edu). Some populations of L. pimpinellifolium are thought to have outcrossing rates up to 37% (Rick et al. 1977, 1978; Georgiady et al. 2002).

From each species, five individuals (10 alleles) were sampled from one population, except for L. hirsutum where only three or four plants were included, depending on the locus. We studied the same individual plants from one population per species as in previous work (Baudry et al. 2001; Städler et al. 2005), representing one accession per species from the Tomato Genetics Resource Center at the University of California at Davis (http://tgrc.ucdavis.edu).

DNA sequencing and haplotyping:

We sequenced the same 8 loci (CT093, CT114, CT099, CT066, CT166, CT179, CT148, and CT198) in the two SC species previously sequenced in the three SI species by Städler et al. (2005). One additional locus (CT189) was added for all five taxa. Furthermore, we used previously published sequence data for five loci (CT143, CT208, CT251, CT268, and Sucr) (Baudry et al. 2001). For some of these loci, we slightly modified the alignment and/or the exon-intron boundary annotation (see Städler et al. 2005). Moreover, we obtained homologous sequences from one of two outgroup species (S. lycopersicoides or S. ochranthum) for 11 of the 14 loci in this study (see below). Repeated attempts to amplify the loci CT114, CT148, and CT208 from the outgroup taxa failed. Summed across all genes, this amounts to ∼20.5 kb of sequence per “allele” or 41 kb per individual plant. The loci are single-copy nuclear genes and are distributed across 9 of the 12 tomato chromosomes (Table 1). They were chosen from regions representing low, intermediate, and high recombination rates (R_N) according to a recombination nodule map (Stephan and Langley 1998). PCR primers, designed on the basis of cDNA sequences for the cultivated tomato L. esculentum (= S. lycopersicum), are available from the Tomato Gene Index at The Institute for Genomic Research (TIGR; http://www.tigr.org/tdb/lgi/). PCR primers and reaction conditions can be accessed at http://www.zi.biologie.uni-muenchen.de/evol/index.html.

TABLE 1.

Chromosome location, putative function, and recombination rate (R_N) of sequenced loci

Locus	Chromosome	Length (bp)	Putative encoded protein	RNb (×10−8)
Sucra	3	1575	Vacuolar invertase	0.00
CT208a	9	1767	Alcohol dehydrogenase, class III	0.00
CT093	5	1415	S-adenosylmethionine decarboxylase proenzyme	0.00
CT114	7	1169	Phospho-glycerate kinase	0.00
CT189	12	1463	40S ribosomal protein S19	0.35
CT251a	2	1779	At5g37260 gene	0.46
CT099	12	1354	Unknown	0.88
CT066	10	1346	Arginine decarboxylase	0.93
CT166	2	2673	Ferredoxin-NADP reductase	1.61
CT179	3	995	Tonoplast intrinsic protein Δ-type	1.97
CT148	8	1497	Copper/zinc superoxide dismutase	2.00
CT198	9	779	Submergence induced protein 2-like	2.10
CT268a	1	1887	Receptor-like protein kinase	2.33
CT143a	9	1821	Sterol C-14 reductase	2.73

Except for Sucr (sucrose accumulator gene), locus designations refer to particular EST sequences that have been integrated into longer “tentative contigs” in the TIGR Tomato Gene Index (http://www.tigr.org/tdb/lgi/). The length per locus is given across the total alignment of all five tomato species (without outgroup), including indels.

^aFrom Baudry et al. (2001).

^bDerived from Stephan and Langley (1998); recombination rate per site per generation.

Haplotype phase was fully resolved for the newly sequenced nine loci, but not for L. peruvianum in the previously published five loci (Baudry et al. 2001). PCR products were sequenced directly on both strands with a MegaBACE 1000 automated sequencer (Amersham Pharmacia, Freiburg, Germany). Distinct haplotypes within heterozygous individuals were resolved by applying a suite of haplotype-specific sequencing primers. In most cases, we exploited putative or confirmed SNPs to anchor the 3′-end of 18-bp sequencing primers that were intended to resolve the heterogeneous PCR products. This approach enabled us to verify SNPs (and indel variation) and establish haplotype phase on the basis of overlapping information (usually) supported by multiple, differential primer pairs. In a few technically challenging cases (caused by extensive indel polymorphism), we cloned the PCR product (Invitrogen, Karlsruhe, Germany) and then sequenced at least 10 clones per plant. Sequences were edited and initially aligned in Sequence Navigator (Applied Biosystems, Darmstadt, Germany). Interspecific alignments were performed with ClustalW (Thompson et al. 1994) and adjusted manually in MacClade, version 3.07 (Maddison and Maddison 1992).

Estimation of recombination rates and levels of diversity and divergence:

The genomic recombination rate R_N was estimated from a genetic linkage map and a frequency map of recombination nodules based on a cross between L. esculentum and L. pennellii (= S. pennellii) (Stephan and Langley 1998); this sequence-independent estimator of recombination is given as the recombination fraction per site per generation, c. From our sequence data, we also estimated the population recombination parameter (C = 4N_ec, where N_e denotes the effective population size) by the following methods: (i) The estimator ρ was calculated with the composite-likelihood coalescent method of Hudson (2001), as implemented in the program LDhat (McVean et al. 2002), and (ii) γ, a maximum-likelihood estimator that uses subsets of four sequences, was calculated with the program Sites (Hey and Wakeley 1997). All reported estimates are per-site values.

Due to the observed haplotype structure in L. chilense and L. hirsutum (see Städler et al. 2005) and too few nucleotide polymorphisms in L. chmielewskii and L. pimpinellifolium, we estimated ρ and γ only from L. peruvianum sequences. This approach is justified because L. peruvianum is the most polymorphic species and the studied population appears to be close to demographic equilibrium (Städler et al. 2005). Sites with observed multiple hits were excluded from the estimation of γ and ρ.

As an estimator of nucleotide diversity (θ = 4N_eμ, where μ is the mutation rate per site per generation), we calculated Watterson's (1975) θ_W for silent sites (denoted as θ_sil). Divergence at silent sites (K_sil) was measured as the average number of nucleotide substitutions per silent site between species, using Jukes-Cantor correction (Nei 1987) as implemented in DnaSP version 4.0 (Rozas et al. 2003). This analysis was performed for the 11 loci for which we obtained outgroup sequences from either S. ochranthum (Sucr, CT251, CT189, CT099, CT066, CT166, CT179, CT198, and CT143) or S. lycopersicoides (CT093 and CT268). All 11 K_sil estimates were used for correlation analyses, despite the fact that we used different outgroups.

Taxonomic studies based on a variety of molecular markers have identified either S. lycopersicoides or S. ochranthum as the sister group to the tomato clade (Spooner et al. 1993, 2005; Peralta and Spooner 2001). The estimated divergence time between S. ochranthum and Lycopersicon is between 5.8 and 18.6 million years (Rose 2002), while the divergence time between S. lycopersicoides and Lycopersicon has not been estimated. However, S. lycopersicoides is a close relative of the tomato clade and extensive synteny between the chromosomes is observed (Rick 1979; Chetelat et al. 2000). We explored the possible error introduced by using two outgroup species and found that in correlation analyses both always indicated the same trends.

Tests of neutrality:

To test for deviations from the standard neutral model of evolution (Kimura 1983), we performed the following tests using the programs DnaSP and HKA (http://lifesci.rutgers.edu/∼heylab/): (i) the McDonald-Kreitman test (McDonald and Kreitman 1991), which compares the pattern of within-species polymorphism and between-species divergence at synonymous and nonsynonymous sites in coding regions of a gene; (ii) the Hudson-Kreitman-Aguadé (HKA) test (Hudson et al. 1987), which tests for heterogeneity in the ratio of polymorphism to divergence among loci; (iii) Tajima's D (Tajima 1989) for all sites (D_all) and for silent sites (D_sil), which tests the neutral prediction that the estimators π and θ_W should measure the same quantity, θ; and (iv) Fu and Li's D test (Fu and Li 1993), which is based on the differences between the total number of mutations in external branches of the genealogy and the total number of mutations. Fu and Li's D is constructed in a similar way to Tajima's D, but takes into account whether mutations are ancestral or derived. For both tests, negative values indicate an excess of polymorphisms with low frequency, whereas positive values indicate an excess of polymorphisms with an intermediate frequency.

To explore the ratio of nonsynonymous to synonymous substitutions, we calculated intraspecific nucleotide diversity at nonsynonymous (π_a) and synonymous sites (π_s) (Nei 1987) and nucleotide divergence to the outgroup at nonsynonymous (K_a) and synonymous sites (K_s); the latter estimates were obtained with Jukes-Cantor correction (Nei 1987).

Estimation of effective population size:

Effective population size (N_e) for the five Lycopersicon species was estimated using the following formulas: θ_sil = 4N_eμ and μ = K_sil/2t, with t being the divergence time in generations. We assume that the split between Lycopersicon and S. ochranthum occurred between 5.8 and 18.6 million years ago (Rose 2002) and calculate θ_sil as the weighted average silent diversity per species and K_sil as the weighted average silent divergence. To account for uncertainties regarding generation time (see above), we assume two rather extreme cases: one generation per year and one generation every seven years. The latter estimate is based on the assumption that the natural habitats of Lycopersicon and the outgroup species have been affected by the El Niño southern oscillation (ENSO), which determines moisture availability and consequently seed germination and plant establishment (see discussion).

RESULTS

Levels of diversity and mating system:

Table 2 summarizes levels of silent diversity (θ_sil) for all 14 loci across the five study species. We observed only very little variation in the two SC species L. chmielewskii and L. pimpinellifolium (mean θ_sil = 0.027% and 0.160%, respectively), whereas we found the three SI species to be highly polymorphic (L. peruvianum mean θ_sil = 2.233%, L. chilense mean θ_sil = 0.642%, and L. hirsutum mean θ_sil = 0.577%). Thus, average levels of variation in the SI species are 3.6- to 14-fold higher than those in the partially selfing L. pimpinellifolium, and 21- to 83-fold higher than those in L. chmielewskii.

TABLE 2.

Levels of silent diversity at 14 loci in five Lycopersicon species

Locus	L. peruvianum	L. chilense	L. hirsutum	L. pimpinellifolium	L. chmielewskii
Sucra	3.189	0.944	0.475	0.103	0.000
CT208a	1.259	0.118	0.000	0.060	0.000
CT093	1.224	0.406	0.063	0.058	0.000
CT114	1.407	0.488	0.354	0.054	0.055
CT189	1.866	0.186	0.000	0.094	0.031
CT251a	2.631	0.197	0.440	0.000	0.000
CT099	3.452	1.632	2.722	0.429	0.000
CT066	3.513	1.812	0.465	0.106	0.000
CT166	1.982	0.861	0.572	0.169	0.000
CT179	2.001	1.663	1.692	0.413	0.149
CT148	2.299	0.778	0.898	0.000	0.000
CT198	4.292	0.481	1.145	0.997	0.207
CT268a	2.840	1.787	1.150	0.163	0.000
CT143a	2.244	0.000	0.153	0.123	0.049
Means	2.233	0.642	0.577	0.160	0.027

All estimates are per-site θ_sil values, expressed in percentages. Silent diversity in the three SI species (the first three species) was reported by Städler et al. (2005), except for locus CT189. The bottom row reports the weighted mean θ_sil across all 14 loci.

^aEstimates deviating from those given in Baudry et al. (2001) are due to minor alignment and/or annotation changes.

Estimates of effective population size also indicate a 3- to 83-fold reduction between SI and SC species (Table 3). The observed differences in neutral diversity between SC and SI populations cannot be attributed to the difference in mating system alone. Additional forces such as natural selection and/or forces affecting the entire genome, such as demographic factors or life-history traits influencing N_e (e.g., metapopulation dynamics or a soil seed bank) must be invoked to explain these results (see discussion).

TABLE 3.

Estimates of effective population size in five Lycopersicon species

	Assumed divergence time
Species	5.8 MYA	18.6 MYA
L. peruvianum	1.165	3.736
L. chilense	0.335	1.073
L. hirsutum	0.308	0.987
L. pimpinellifolium	0.097	0.310
L. chmielewskii	0.014	0.046

The range of estimated divergence times between the clade Lycopersicon and the outgroup S. ochranthum is based on Rose (2002). Estimates of effective population size (×10⁶) assume one generation per year and would be only one-seventh of these estimates under a generation time of 7 years (but see text).

Silent diversity and silent divergence:

Figure 1 shows that for all three SI species, θ_sil and K_sil are positively correlated (L. peruvianum: Spearman's R = 0.718, P = 0.013; L. chilense: R = 0.655, P = 0.029; L. hirsutum: R = 0.682, P = 0.021). This positive correlation between θ_sil and K_sil conforms to the pattern expected under neutrality and clearly suggests that neutral mutation rates vary among loci. Nevertheless, the multilocus HKA test, which assesses the neutral prediction that the ratio of polymorphism to divergence across loci should be constant, is significant for L. chilense, L. hirsutum, and L. pimpinellifolium (Table 4). Loci with the greatest contribution to the overall test statistic differ among species. In L. chilense, CT143 contributes greatly to the significant multilocus HKA statistic; diversity at this locus is strongly reduced in both L. chilense and L. hirsutum, and CT189 has a similar effect in L. hirsutum. For L. pimpinellifolium, there is an excess of polymorphism at locus CT198. Removal of these single loci from the multilocus HKA test causes the test statistic to drop below the critical value in each of these species.

Figure 1.

Scatter plots of levels of silent divergence and silent diversity in the three SI species (n = 11 loci). Regression lines are also shown for each species (Figures 1–4). L. peruvianum: Spearman's R = 0.718, P = 0.013; L. chilense: R = 0.655, P = 0.029; L. hirsutum: R = 0.682, P = 0.021.

TABLE 4.

Summary of multilocus neutrality tests

Species	Tajima's Dall	Tajima's Dsil	Fu and Li's D	HKAall	HKAsil
L. peruvianum	−0.37 (−0.35)	−0.39 (−0.34)	−0.53	0.99	0.98
L. chilense	0.58* (0.84)***	0.60* (0.88)***	0.83*	0.02*	0.01*
L. hirsutum	0.24 (0.35)	0.32 (0.42)*	0.56*	0.04*	0.02*
L. pimpinellifolium	0.18 (0.08)	−0.01 (−0.07)	0.25	0.004**	0.004**
L. chmielewskii	−0.15 (−0.13)	−0.003 (0.002)	−0.044	—	—

Tajima's D_all (based on all sites) and D_sil (based on silent sites) values are averages across the 9 new loci, followed by the mean values using all 14 loci in parentheses (i.e., including Sucr, CT208, CT251, CT268, and CT143 from Baudry et al. 2001). Average Tajima's D_sil values based on 13 loci (without CT189) in the three SI species were presented by Städler et al. (2005). Analyses in the other three columns are based on the 11 loci for which an outgroup sequence is available (Sucr, CT093, CT189, CT251, CT099, CT066, CT166, CT179, CT198, CT268, and CT143). Numbers in the HKA columns are probability values from the χ² distribution with 10 d.f.; L. chmielewskii harbors too few polymorphisms for these analyses. *P < 0.05; **P < 0.01; ***P < 0.001 (one-tailed tests).

Other neutrality tests:

We applied several other neutrality tests to probe for signatures of selection and/or demographic processes in our data. In L. peruvianum, CT198 displays the highest silent nucleotide diversity among all loci (θ_sil = 4.29%) and shows a significant McDonald-Kreitman test (Fisher's exact test, P = 0.033). However, this result rests on only four fixed differences among species, three of which are replacement substitutions. Tajima's D and Fu and Li's D are significantly positive for this locus in L. pimpinellifolium (CT 198; Tajima's D_all = 2.42, P < 0.01; Fu and Li's D = 1.76, P < 0.02). CT099 displays the greatest apparent deviation in L. hirsutum, with a significantly positive Tajima's D (D_all = 1.48, P = 0.039) and Fu and Li's D (D = 1.78, P = 0.023) and unusually high polymorphism for this species (θ_sil = 2.72%; Table 2). The average values of Tajima's D and Fu and Li's D in the three SI species are consistent with parameter estimates obtained under the simple “isolation” model of speciation; both L. chilense and L. hirsutum are inferred to have undergone population-size contractions compared to the common, ancestral species (Städler et al. 2005).

Although several loci yield significant results for various tests in one or more species, the above results were obtained without correcting for multiple testing; hence we see no conclusive evidence that one of the loci has recently been subject to natural selection. The lack of nucleotide diversity at some loci may simply be due to our limited sample size or due to demographic effects. A further aspect to consider is that mean values for Tajima's D for all sites and silent sites across 14 loci are not significantly different (Mann-Whitney U-test: L. peruvianum, P = 0.98; L. chilense, P = 1.00; L. hirsutum, P = 0.95; L. chmielewskii, P = 0.93; L. pimpinellifolium, P = 0.85). Thus we find no indication that the frequency spectrum of nonsynonymous polymorphisms is significantly different from that of silent polymorphisms.

Recombination rate and silent genetic diversity:

We analyzed the effect of recombination rate on levels of DNA diversity for 14 genes in three self-incompatible wild tomato species using a physical estimate of the genomic recombination rate (R_N), which is based on detailed recombination nodule maps of all 12 tomato chromosomes (Stephan and Langley 1998). In addition, we compared our results for R_N estimates with sequence-based estimators of the population recombination parameter, ρ (Hudson 2001) and γ (Hey and Wakeley 1997).

On the basis of 14 loci, Figure 2 presents correlations between θ_sil and R_N for the three outcrossing species. In both L. peruvianum and L. hirsutum, θ_sil tends to be positively correlated with R_N (Spearman's R = 0.471, P = 0.089 and R = 0.528, P = 0.052, respectively). In L. chilense, however, we find no positive correlation (R = 0.198, P = 0.498; Table 5).

Figure 2.

Scatter plots of recombination rate per site and silent nucleotide diversity in the three SI species (n = 14 loci). L. peruvianum: Spearman's R = 0.471, P = 0.089; L. chilense: R = 0.198, P = 0.498; L. hirsutum: R = 0.528, P = 0.052.

TABLE 5.

Correlations between diversity levels (θ_sil) and estimates of recombination rate

Species	θsilvs. RN	θsilvs. γ	θsilvs. ρ
L. peruvianum	0.471 (0.089)	0.617 (0.077)	0.600 (0.088)
L. chilense	0.198 (0.498)	0.050 (0.898)	0.267 (0.488)
L. hirsutum	0.528 (0.052)	0.567 (0.112)	0.617 (0.077)

All values represent Spearman's coefficient of rank correlation; P-values (two-tailed) follow in parentheses. All 14 loci were used for the analyses involving the sequence-independent estimate of recombination, R_N. Our ρ and γ estimates were obtained from L. peruvianum sequences, and the correlation analyses shown are based on the 9 loci with experimentally established haplotype phase (see text).

To assess consistency between the different estimators of recombination, we performed correlation analyses and compared the results for the entire data set (14 loci) with results for the 9 newly sequenced loci. This comparison was done to assess the possible error introduced by the lack of reliable haplotype-phase information in L. peruvianum for the 5 previously studied loci. R_N is positively correlated with γ (9 loci: Spearman's R = 0.669, P = 0.049; 14 loci: R = 0.476, P = 0.086), but not with ρ (9 loci: R = 0.326, P = 0.391; 14 loci: R = 0.302, P = 0.294). The values of γ and ρ exhibit strong positive correlations among loci (9 loci: R = 0.733, P = 0.025; 14 loci: R = 0.648, P = 0.012). Results for both analyses (using 14 and 9 loci, respectively) are thus consistent with each other, suggesting that the assignment of “quasi-haplotypes” at 5 L. peruvianum loci does not introduce marked biases in our analyses.

Additionally, results for correlations between R_N and θ_sil (14 loci) and between θ_sil and the sequence-based estimators (9 loci) are consistent; γ and ρ show a strong positive correlation with θ_sil for both L. peruvianum and L. hirsutum, but not for L. chilense (Table 5). Since sequence-based estimators of recombination rate are believed to be biased upward in regions of high genetic diversity (Wall 2000), in the following sections we concentrate on results based on the physical estimate of recombination, R_N.

Recombination rate and interspecific divergence:

If neutral processes explain the positive correlation between nucleotide diversity and recombination rate (e.g., mutagenic effects of the recombination process), as was suggested by the study of Hellmann et al. (2003), divergence between species is also expected to be positively correlated with recombination. We analyzed the correlation between recombination rate (R_N) and levels of silent divergence (K_sil), calculated for 11 loci with outgroup sequences from either S. ochranthum (n = 9 loci) or S. lycopersicoides (n = 2 loci); no significant correlation between R_N and K_sil was detected (L. peruvianum: Spearman's R = 0.159, P = 0.640; L. chilense: R = 0.064, P = 0.852; L. hirsutum: R = 0.036, P = 0.915) (Figure 3).

Figure 3.

Scatter plots of recombination rate per site and silent divergence for the three SI species (n = 11 loci). L. peruvianum: Spearman's R = 0.159, P = 0.640; L. chilense: R = 0.064, P = 0.852; L. hirsutum: R = 0.036, P = 0.915.

θ_sil/K_sil and recombination rate:

If the weakly positive correlation between recombination rate and neutral genetic diversity is in fact a signature of diversity-reducing selection in regions of low recombination, we would also expect a positive trend between recombination rate and diversity corrected for interspecific divergence (i.e., θ_sil/K_sil). Although a positive trend remains, θ_sil/K_sil is not significantly correlated with R_N in any of the three SI species (L. peruvianum: Spearman's R = 0.123, P = 0.719; L. chilense: R = 0.064, P = 0.852; L. hirsutum: R = 0.405, P = 0.216) (Figure 4). This is consistent with our findings of significant positive correlations between θ_sil and K_sil (Figure 1). Thus, selection does not seem to play a significant role in shaping levels of genetic diversity in the genomes of wild tomato species. In the case of positive selection, we would expect that regions with very low recombination harbor very little silent genetic diversity (Innan and Stephan 2003). Yet in both L. peruvianum and L. chilense we detected notable levels of θ_sil in regions of low recombination (Figure 2). L. hirsutum also retains appreciable amounts of genetic diversity at some loci with zero recombination; only locus CT208 (R_N = 0) displays zero silent genetic diversity. Since this observation for CT208 is based on only six alleles in L. hirsutum, sampling effects could account for this result. The steepest slope for the regression between θ_sil/K_sil and R_N was seen in L. hirsutum (Figure 4); the same trend was previously observed by Stephan and Langley (1998).

Figure 4.

Scatter plots of recombination rate per site and silent diversity corrected for divergence (n = 11 loci). L. peruvianum: Spearman's R = 0.123, P = 0.719; L. chilense: R = 0.064, P = 0.852; L. hirsutum: R = 0.405, P = 0.216.

DISCUSSION

We examined intraspecific nucleotide diversity at 14 loci in five species of wild tomatoes with different mating systems and interspecific divergence to an outgroup at 11 of these 14 loci. In comparison to the previous study by Baudry et al. (2001), this study considerably increases the amount of available sequence data in Lycopersicon and permits a more detailed examination of the effect of recombination rate differences on levels of silent nucleotide diversity. Importantly, the outgroup sequences for 11 of our loci (not previously available) enable us to distinguish between intraspecific and interspecific effects influencing sequence evolution in wild tomatoes.

Mating system differences:

The mating systems of our study species vary from partial self-fertilization in the SC taxa (i.e., L. chmielewskii and L. pimpinellifolium) to obligate outcrossing in the SI populations (i.e., L. hirsutum, L. peruvianum, and L. chilense). Evolutionary change in mating system from SI to SC is common and irreversible in the Solanaceae and self-incompatibility is thought to be ancestral (Igic et al. 2004). Previous studies in wild tomatoes have shown that genetic variation within species decreases with an increasing degree of selfing (Rick et al. 1976, 1977, 1979; Caicedo and Schaal 2004). Given that our study encompasses both SI and SC populations, we can show directly how partial selfing and/or demographic processes manifest themselves at the population level. All other things being equal, population-genetic theory predicts a twofold reduction in levels of neutral variation between outcrossing and completely selfing populations (Pollak 1987; Nordborg and Donnelly 1997; Charlesworth 2003; and references therein) and slighter differences for partially selfing populations. Clearly, average levels of variation in the SC (but not necessarily highly selfing) wild tomato populations are lower than expected under effects of the mating system on effective population size per se.

Plant mating systems are well known to have significant impacts on the levels and distribution of genetic diversity, either directly via differences in effective population size due to (partial) selfing or indirectly via life-history features that are associated with mating-system differences, such as colonization potential (Ingvarsson 2002; Barrett 2003). The relationship between mating systems and levels of genetic variation has been studied empirically in animals and plants (e.g., Liu et al. 1999; Graustein et al. 2002; Wright et al. 2002), and until recently was dominated by studies using allozymes to quantify levels of diversity within and among local populations (Hamrick and Godt 1990; Schoen and Brown 1991). In recent studies on closely related angiosperm species with different mating systems, the discrepancies in levels of diversity between inbred and outbred taxa were attributed to asymmetric introgression from selfing to outcrossing species (Sweigart and Willis 2003), demographic effects (Savolainen et al. 2000), and/or natural selection (Baudry et al. 2001). Since selfing reduces the effective recombination rate, the reduction of nucleotide diversity due to selective sweeps or background selection should be more pronounced in partially self-fertilizing plants than in outcrossers. The SC populations in our study show a moderate-to-high reduction in nucleotide diversity compared to the SI species, which exceeds the theoretical twofold prediction despite the fact that our SC populations are not completely selfing (Rick et al. 1976, 1977, 1978).

Various factors could contribute to our observations. Ecological adaptations (e.g., the presence of soil seed banks in adverse environments) might differ among the study species. Since SI species appear to be less prone to extinction due to high diversity and gene flow that reduce genetic isolation despite high local extinction and recolonization rates (Igic et al. 2004), characteristics and longevity of their soil seed banks could differ from those of SC species. Second, stronger population substructure in partially selfing taxa may cause DNA variation to be largely distributed among populations, whereas in outcrossing species to a large extent it is found within local demes (Awadalla and Ritland 1997; Charlesworth 2003; Sweigart and Willis 2003). Since our sequence data are derived from single populations per species, we cannot offer a conclusive assessment of this factor. Moreover, selfing in species with high rates of population turnover may cause reductions in levels of DNA diversity that greatly exceed the magnitude expected without such metapopulation dynamics (Ingvarsson 2002).

Diversity-reducing forces and variation in mutation rates and selective constraints:

We found positive correlations between a physical estimate of the recombination rate (R_N) and silent nucleotide diversity (θ_sil) in two of three SI taxa, L. peruvianum and L. hirsutum. Studies in Drosophila found much stronger positive correlations between levels of nucleotide polymorphism and recombination rate (Aguadé et al. 1989; Stephan and Langley 1989; Begun and Aquadro 1992) and attributed these findings to an interplay between recombination and selection. In Drosophila melanogaster, positive directional selection (i.e., selective sweeps) seems to be the predominant cause for the strong positive correlation between silent genetic diversity and recombination rate. In regions of low recombination, selection removes more neutral genetic variation around the selected site than in regions of high recombination. In particular, levels of variation are very low in regions of very low recombination, as expected under directional selection (Innan and Stephan 2003). In Lycopersicon, the situation appears to be different. In the three SI species, we observed only a weak correlation between R_N and K_sil, which is pronounced enough, however, that the ratio θ_sil/K_sil exhibits no positive correlation with R_N.

Tenaillon et al. (2001, 2002) studied 21 loci in maize (Z. mays ssp. mays L.) and found no correlation between SNP diversity and their physical estimate of recombination rates. Sequence-based estimates of recombination rate, however, suggested a different picture. Hudson's (1987) and Wall's (2000) estimators were correlated with nucleotide diversity, whereas ρ (Hudson 2001) was not correlated with diversity. This appears to be in contrast to our findings in Lycopersicon, where both physical estimates of recombination rate (R_N) and sequence-based estimators (ρ and γ) are positively correlated with silent diversity (in two of three SI species) and are consistent with each other.

An important insight gained through the availability of sufficiently diverged outgroup sequences is that silent nucleotide diversity and silent divergence are strongly correlated, as expected if neutral mutation rates and/or selective constraints vary among loci or genomic regions (Figure 1). In a recent study, Tenaillon et al. (2004) found no correlation between diversity and any measure of recombination in the wild ancestor of maize, teosinte (Z. mays ssp. parviglumis). However, comparable to our findings in wild tomatoes, silent diversity and silent divergence were positively correlated. This suggests that in both of these plant model systems, mutation rate (and/or selective constraint) differences among loci drive this correlation (at least partly). Since silent diversity was not corrected for divergence in the earlier maize studies (Tenaillon et al. 2001, 2002), we assume that the positive correlation between recombination rate and nucleotide diversity would be considerably weaker if corrected for divergence.

From our data, we can provide several observations that may be interpreted as a signature of background selection removing deleterious mutations (Charlesworth et al. 1993, 1995). This interpretation rests on the assumption that the positive correlation between R_N and θ_sil is real and not spurious, which is consistent with previous RFLP data based on a larger number of nuclear loci (Stephan and Langley 1998). First, the regression lines are relatively flat for all SI species (R_N vs. θ_sil; Figure 2). Second, nucleotide variation in regions of very low recombination does not drop down to zero; this is best seen in L. peruvianum. Third, there is a slight positive trend between θ_sil/K_sil and R_N for all three SI species, in particular for L. hirsutum. Furthermore, the relative paucity of segregating nonsynonymous variation at all sequenced loci is a clear signal of purifying selection. In particular, the ratios π_a/π_s and K_a/K_s are much lower than one for all loci (Figure 5), indicating that selection has removed deleterious nonsynonymous mutations within lineages. Likewise, none of the sequenced alleles in our study appear to be nonfunctional. Although we did not sequence all exons for any of the genes, we would nevertheless expect mutations resulting in frameshifts and premature stop codons to occur under neutral evolution. The background selection model was also favored by Innan and Stephan (2003) who found that in RFLP data (Miller and Tanksley 1990), levels of polymorphism in L. pimpinellifolium and L. chmielewskii are best explained by a concave function of the recombination rate in regions of low recombination. According to their analysis, under hitchhiking associated with positive selection, the shape of this function would be expected to be convex in regions of low recombination, as was seen in data from D. melanogaster.

Figure 5.

Ratios of average pairwise differences per nonsynonymous site to synonymous site in L. peruvianum (π_a/π_s; 14 loci) and of average pairwise divergence to the outgroup (K_a/K_s; 11 loci). The dotted line represents K_a/K_s = 1 and π_a/π_s = 1.

Life-history traits and demographic processes affecting N_e:

Life-history traits such as the presence of a seed bank can have important consequences for effective population size in plants, especially in the presence of population turnover (Nunney 2002). Fluctuations in population size are likely to be common in the natural habitats of our study populations. The west coast of South America is affected by the ENSO, and this recurrent meteorological phenomenon affects not only populations of adult plants but also the replenishment of the seed bank (Gutiérrez and Meserve 2003). The onset, frequency, and strength of the ENSO before historical times are difficult to establish (DeVries 1987; Tudhope and Collins 2003), but there is reason to believe that the current arid climate has shaped the vegetation of coastal western South America for considerable lengths of time (Gregory-Wodzicki 2000; Hartley 2003).

We attempted to estimate N_e for the wild tomato species using levels of silent diversity and silent divergence (see materials and methods). Assuming one generation per year, estimates for N_e in L. peruvianum vary between 1.165 × 10⁶ and 3.736 × 10⁶, depending on the timing of the split between S. ochranthum and the tomato clade. If we assume only one generation every 7 years, our estimates are reduced to one-seventh of the above values (but see below). Estimates for the other species are much lower, in accordance with the lower levels of nucleotide variation in these species (Table 3). Our estimates of effective population size imply that substitution rates between 1.6 × 10⁻⁹ and 5.2 × 10⁻⁹ silent substitutions/year (weighted over all loci), depending on the timing of the split between the tomato clade and the outgroup. These estimates are not substantially different from the previously estimated synonymous substitution rates in several angiosperm genes (Moniz de Sá and Drouin 1996; Dvornyk et al. 2002).

Our N_e estimates seem extraordinarily high for plant species characterized by patchy distribution in temporally fluctuating environments (T. Städler, personal observations). Estimates for African D. melanogaster yielded a similar N_e of ∼10⁶ (Li et al. 1999), yet the census size for Drosophila and wild tomatoes differs dramatically. Likewise, in loblolly pine, an outcrossing gymnosperm that is abundant in the southeastern United States, Brown et al. (2004) calculated an effective population size of only 5.6 × 10⁵ and explained the apparent discrepancy between census population size and N_e by glacial population fluctuations.

For wild tomato species, the existence of soil seed banks could explain the apparent incongruence of small, isolated local populations and our high N_e estimates. Nunney (2002) showed that the effective size of a local population is proportional to the generation time, T, for species with seed banks. Average generation times may very well vary between the more mesic habitats of L. hirsutum and the hyperarid conditions experienced by both L. peruvianum and L. chilense in southern Peru and northern Chile, perhaps approaching the period between successive El Niño events in the latter region. One should also keep in mind that species such as L. peruvianum consist of many subpopulations. Thus, assuming the presence of a soil seed bank buffering against genetic bottlenecks and assuming a large number of subpopulations may help to explain our large θ_sil and N_e estimates.

The above considerations of demographic processes and life-history traits may help to explain why positive directional selection is not a significant force causing reductions of silent diversity in regions of low recombination in wild tomatoes. Even if selective sweeps occur locally in restricted geographic areas, their effects on diversity in the total population are expected to be weak, because the selected alleles may be unable to spread fast through the entire species range.

Conclusions:

Sequence evolution in Lycopersicon seems to be dominated to a large extent by demographic processes, variation in neutral mutation rates, and/or selective constraints among genes, and most likely, background selection. This is in contrast to findings in Drosophila where selective sweeps are likely to be the underlying cause for a strong positive correlation between recombination rate and nucleotide diversity. In wild tomato species, a comparably strong positive correlation is not detected. Despite small census size of local isolated populations, wild outcrossing tomatoes are able to maintain high effective total population sizes. Both observations may be explained by the presence of soil seed banks and extensive population substructure. Demographic processes and life-history traits presumably contribute mostly to the marked differences in diversity levels between partially selfing and obligately outcrossing tomato species. Further studies of additional populations are needed to understand how population substructure, extinction-recolonization processes, and soil seed banks interact to shape levels of diversity in wild tomatoes.