Study system:

The yellow monkey flowers of the M. guttatus species complex are a morphologically diverse but largely interfertile group of wildflowers with their center of diversity in western North America (Vickery 1978). M. guttatus, the most common species in the complex, has large, insect-pollinated flowers and is predominantly outcrossing (Willis 1993). Routine selfing appears to have evolved several times in the group (Wu et al. 2008). The most widespread of the selfing taxa, M. nasutus, produces reduced, often cleistogamous flowers. Where M. guttatus and M. nasutus co-occur, potential premating barriers to hybridization include differences in microhabitat and flowering time (Martin and Willis 2007) as well as differences in floral morphology (Ritland and Ritland 1989), pollen production (Fenster and Carr 1997), and pollen tube growth (Diaz and MacNair 1999) associated with their different mating systems. Despite these barriers to mating, hybrids are observed at low frequency at many sympatric sites, and there is evidence of recent local introgression at nuclear loci (Sweigart and Willis 2003).

Specific populations and lines:

Previously, we identified a cytonuclear incompatibility in hybrids between an inbred line of M. guttatus from Iron Mountain, Oregon (IM62) and a M. nasutus line from Sherar's Falls, Oregon (SF5) (Fishman and Willis 2006). In F₂ hybrids with the M. guttatus cytoplasm (F₂G), anther sterility segregates at ∼25%. The reciprocal cross shows no anther sterility, although there is an independent nuclear–nuclear incompatibility causing pollen inviability (Sweigart et al. 2006). This pattern is consistent with a single Mendelian restorer locus with a dominant M. guttatus Rf allele. In a first generation backcross to M. nasutus (CSB1 population), the Rf locus was mapped to a region at the end of linkage group 7 bounded by the marker MgSTS574a (Fishman and Willis 2006).

Genome scan for PPRs:

To target our fine mapping of Rf, we identified regions of the genome with high PPR density by running a hidden Markov model using programs in the HMMER package (http://hmmer.janelia.org/) and the consensus PPR model developed by Lurin et al. (2004) for Arabidopsis thaliana. Because PPR genes are numerous and highly diverse in both sequence and numbers of motifs, simple BLAST algorithms are not sufficient for their complete identification in a genome. Programs in the HMMER package allow the development of a profile of the likelihood of each amino acid in each position of the PPR motif and the searching of databases using this profile. The PPR profile used was developed by running the program hmmbuild on an alignment of 2357 A. thaliana PPR motifs (Lurin et al. 2004). We then ran the program hmmsearch to search our Mimulus sequence database (originally on the 4X genome build to develop markers for fine mapping and then later on the 7X build for comparative purposes) for similarity to this PPR profile. At the time of this work, the genome consisted of a large number of internally contiguous, but unassembled, tracts of sequence called “scaffolds.” To identify PPR genes (each of which consists of a number of PPR motifs), we used a protocol similar to that of Lurin et al. (2004), identifying and removing hits that were 200 nucleotides away from any other motif (orphan motifs).

Marker development and genotyping:

To genetically map PPR-rich scaffolds that were not already anchored to existing M. guttatus complex maps with gene-based MgSTS markers (http://www.mimulusevolution.org), we developed new intron-spanning markers. For a target region, we identified putative intron-containing genes and likely intron positions by aligning the genomic sequence to a database of dicot cDNA using the program GeneSeqer (Usuka et al. 2000) (http://www.plantgdb.org). Primers were designed in exon sequence to span putative introns, tested for amplification in both parental species, and screened for informative differences in amplicon length between the parental lines.

Coarse mapping:

We used a bulk segregant approach to screen all informative MgSTS markers (http://www.mimulusevolution.org), as well as newly designed markers from PPR-rich scaffolds, for association with fertility restoration. The screening panel consisted of six pools of three sterile CSB1 individuals, presumed to be M. nasutus homozygotes (rf/rf), each. Potentially linked markers were then mapped in the full CSB1 population (N = 192). Marker amplification and genotyping protocols followed those used previously (Fishman and Willis 2005, 2006).

Fine mapping:

For fine mapping of Rf, we introgressed the M. guttatus Rf allele, along with the IM62 CMS mitochondrion, into a M. nasutus nuclear background, by backcrossing two fertile CSB1 hybrids to the SF parental line. This selection and backcross process was repeated for two more generations to form CSB4 lines carrying the IM62 cytoplasm, heterozygous at loci involved with male fertility restoration, and 93.75% homozygous for M. nasutus alleles elsewhere. Representatives of the two independent CSB4 lines were selfed to form a large segregating population (CSBG; N = 8448) for fine mapping. At maturity, we phenotyped all individuals, collected leaf and bud tissue, and extracted genomic DNA using a standard CTAB/chloroform extraction protocol (see Fishman and Willis 2005 for details).

The entire CSBG population (N = 6153) was genotyped at the two flanking MgSTS markers and three markers from the target scaffold, and then a subset of informative recombinants was genotyped at three additional markers designed by walking through the Rf region scaffolds. Genotypes were scored automatically using GeneMapper 3.7 (Applied Biosystems), with additional hand scoring where necessary. Primer sequences for all mapping markers are in supporting information, Table S1.

Phylogenetic analysis of PPRs:

To examine the homology of the Mimulus Rf candidate loci with other Rf loci, we compared the phylogenetic relationships of the amino acid sequences of select PPRs from Mimulus, Arabidopsis, petunia, rice, radish, and sorghum. To compare the sequences of the whole PPR gene, which includes the N- and C-terminal sequences, we identified ORFs using the probabilistic gene-identification program Genscan with the Arabidopsis parameter matrix (http://genes.mit.edu/GENSCAN.html, Burge and Karlin 1997) and with the ORF prediction program found in the M. guttatus genome GBrowse (www.mimulusevolution.org); we confirmed that these ORFs were PPRs by using BLAST (Altschul et al. 1990). We included 16 of the 17 PPRs in the regions spanning Rf1 and Rf2 (one of the PPRs was too small to include in this alignment) as well as 2–4 PPRs from the 4 LG7-unlinked scaffolds with the top HMMER-predicted PPRs (scaffolds 4, 8, 11, and 16). Three of these were 13–16 motif PPRs, and we included one additional 13–16 mer from the M. guttatus genome. We also included the Rf sequences from petunia (Bentolila et al. 2002), rice (Kazama and Toriyama 2003; Komori et al. 2004), radish (Desloire et al. 2003), and sorghum (Klein et al. 2005), five Arabidopsis Rf homologs on chromosome 1 (Geddy and Brown 2007), and 5 non-Rf-homologous PPR sequences from A. thaliana (www.arabidopsis.org). Because we are focusing on the relationship between M. guttatus Rf candidates and Rf genes from other taxa, which vary between 13 and 16 motifs, to generate a data set with a sufficient amount of information for accurate tree prediction, we restricted our phylogeny to PPRs that had 7–20 motifs. We aligned all the PPRs with ClustalW.

For tree building, two independent analyses of MCMC were run as implemented in Mr. Bayes assuming a WAG + G + F model of sequence evolution as selected by ProtTest (Abascal et al. 2005). Analyses were metropolis coupled, with three heated chains and one cold chain (from which trees were sampled). Analyses were run for 1 × 10⁶ generations, with trees sampled every 100 generations. Convergence of the analyses on stationarity (i.e., when the chain has become independent of its starting position and is sampling trees in a manner that reflects the posterior probability distribution) was evaluated by the average standard deviation of split frequencies of the two analyses, and the first 25% of trees sampled were discarded as burn-in.

Identification of PPR-rich genomic regions:

We identified 5635 PPR motifs that, after omitting orphan motifs, clustered into 789 groups representing putative PPR genes. The actual number of PPR-coding genes is likely smaller (and potentially much smaller) (see discussion). The distribution of PPRs among genome scaffolds was highly skewed (Figure 1), with a few scaffolds containing large clusters of PPRs and most (regardless of scaffold size) containing none. We were able to genetically map all scaffolds with >10 PPRs by identifying previously mapped MgSTS markers (www.mimulusevolution.org) on each scaffold or by designing and mapping new gene-based markers (Table S1). The distribution of PPRs with restorer-like motif numbers (13–16 motifs) was even more skewed, with a third (29/95) occurring on just three scaffolds (Table 1). Two of those scaffolds (s14 and s97) mapped to LG7 near markers associated with the M. guttatus Rf locus (Fishman and Willis 2006) and became the focus of our fine-mapping efforts.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.—

Distribution of PPR genes on scaffolds of the Mimulus guttatus genome, with a very large majority of scaffolds containing no PPRs and a small number containing many. The actual number of PPR-coding genes in the M. guttatus genome is likely smaller than reported here due to spurious motif identification by HMMER and artificial splitting of single genes; however, we closely examined most of the PPRs on scaffold 14 (the rightmost bar) and it is clear that this scaffold is extremely PPR rich. While variation in number of PPRs among scaffolds is partially explained by scaffold size (r² = 0.42, P < 0.000), the majority of the variation observed in PPR number is not explained by scaffold size, and scaffold 14 falls as a clear outlier.

Coarse mapping:

Using a bulk-segregant approach to screen all previously unmapped MgSTS markers, we found one (MgSTS331) associated with Rf in the CSB1 population. A single crossover in this population (N =192) placed MgSTS331 distal to Rf on LG7.

Fine mapping:

The CSBG fine-mapping population segregated, as expected, ∼1/4 (23%; 1428/6153) anther-sterile individuals. All individuals were genotyped at markers known to flank the Rf locus from coarse mapping in the CSB1 population (MgSTS331, this study; MgSTS574a, Fishman and Willis 2006), as well as three newly designed markers from the two LG7 scaffolds rich in restorer-like PPRs (s97: 191_27; s14: 191_45 and 1_36). All sterile individuals were SF M. nasutus (rf/rf) homozygotes at both 191_45 and 191_27, but not at the flanking markers MgSTS331 and 1_36, narrowing the Rf region to 2.15 cM (265 crossovers/12,340 meioses) between the latter markers (Figure 2). These recombinants were genotyped at three additional markers on the two target scaffolds (s14: 14_172, 271_89; s97: 97_329) to orient the scaffolds and more finely map Rf within each scaffold.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.—

Genetic and physical maps of the Rf loci and candidate PPR genes in Mimulus. (A) Genetic map showing informative recombinants. Horizontal bars represent regions of heterozygosity for CSBG individuals. Shaded bars indicate male-fertile plants and solid bars indicate male-sterile plants. Restoration of fertility maps to a region of ∼1.3 cM. We refer to these loci as Rf1 and Rf2. (B) Physical map of the region containing Rf1 and Rf2 showing the location of markers (triangles) and PPR genes (bars). Solid bars indicate those PPRs that are mitochondrially targeted as identified in the program PREDOTAR (Small et al. 2004), and open bars indicate those that are not mitochondrially targeted.

As shown in Figure 2, recombination within this region suggests that there are two very tightly linked Rf loci, each with a dominant M. guttatus allele capable of restoring fertility. First, we found that all individuals heterozygous at either 191_45 or 191_27 were fertile. This alone could be due to a single locus in the >1-cM interval between these markers. However, there were also some individuals that were M. nasutus homozygotes at both 191_45 and 191_27, but were nonetheless fertile. To test whether this was due to the action of multiple restorer loci or to incomplete penetrance, we examined segregation ratios in the selfed progeny of representative lines. Lines homozygous M. nasutus at 191_45, but with heterozygous introgressions only to the left or right both segregated ∼1/4 steriles (total N = 8, Table 2). These data are consistent with the existence of two Rf loci, one (Rf1) in the 0.32-cM interval bounded by 191_45 and 14_172 and the other (Rf2) in the 0.75-cM interval between 97_329 and 191_45. Hybrid anther sterility occurs only when both loci are homozygous for M. nasutus alleles.

In our original screen of the CSBG population, we also found 68 individuals that were M. nasutus homozygotes across the region containing Rf1 and Rf2 but were fertile. This represents <5% of the expected sterile genotypes. Barring double recombinants in the interval between 191_45 and 1_36, possible explanations for this are a M. guttatus introgression at a third locus sufficient for fertility restoration or incomplete penetrance of the CMS. Because segregation ratios in the CSBG and previous mapping populations did not differ significantly from single-locus Mendelian expectations, any third locus would have to be tightly linked to Rf1 and Rf2 and thus would segregate in the selfed progeny of these individuals. However, we saw no steriles in the progeny of three tested fertile individuals that were M. nasutus homozygotes across the entire region between markers MgSTS331 and 1_36. This result is intriguing and worthy of further investigation.

Molecular content of Rf1 and Rf2 regions:

In the final draft of the M. guttatus (IM62) genome assembly, the region containing Rf1 and Rf2 spans two genome scaffolds. Rf1 is bounded by two markers on s14, whereas Rf2 is bounded by markers on s14 and s97 (Figure 2). To assess the occurrence of PPRs and other potential candidate genes, we examined the gene content of both regions using the gene prediction program GENSCAN (Burge and Karlin 1997). In the 133.4-kb region containing Rf1, we found 12 PPR proteins, 11 of which have 13–16 PPR motifs, and 8 of which had mitochondrially targeted transit peptides as identified with the program PREDOTAR (Small et al. 2004) and thus are likely candidates for Rf1 (Figure 2).

Assessing candidates for Rf2 was more complicated, as we cannot determine the gene content of the interval between scaffolds. However, in the previous 6X draft M. guttatus genome assembly, markers 191_27 and 191_45 were separated by only 108.7 kb on a single scaffold. Thus, we infer that the current gap between s14 and s97 most likely reflects difficulty in assembling this repetitive genomic region rather than a large intervening genomic region. Consistent with this interpretation, the genetic distance between 191_27 and 191_45 (0.18 cM) is near the expectation for markers ∼100 kb apart, given a local estimate of 320–765 kb/cM in the CSBG mapping population. Assembly issues were also indicated by the fact that markers on the far end of scaffold 97 (past 191_27, opposite 97_329) consistently mapped to another linkage group (data not shown). Therefore, for Rf2, we examined the gene content of the 544.7-kb region of s97 from 97_329 to 71.7 kb past 191_27 (where homology to the contiguous scaffold in the 6X assembly was lost) and the 37 kb of s14 distal to 191_45. In this region, we found six PPR proteins, three of which had PPR motif numbers between 13 and 16, and only one of which had a mitochondrially targeted transit peptide (Figure 2).

In addition to PPRs, we located 16 and 84 other ORFs in the Rf1 and Rf2 regions, respectively. Most of these genes were transposable elements, and none were obvious functional candidates for cytoplasm-dependent anther sterility.

We used a Bayesian tree-building algorithm to examine phylogenetic relationships among Mimulus PPR genes from the Rf1/Rf2 region, Mimulus PPRs (some with 13–16 PPR domains) from elsewhere in the genome, as well as Arabidopsis PPRs from both categories and known Rf alleles from diverse crops. The 16 PPR candidate loci for Rf1 and Rf2 formed a clear clade, which is sister to the Rf loci of petunia and rice as well as the radish Rf0 and Arabidopsis Rf homologs (Figure 3). Most other Mimulus PPRs group with non-Rf PPRs from other taxa, along with the Rf for sorghum. This pattern suggests a history of duplication in the Mimulus Rf region, but also indicates that Rf1 and Rf2 are likely to share significant amino acid sequence similarity with restorer loci in distantly related systems.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.—

Bayesian unrooted tree showing the relationships between M. guttatus Rf-linked PPRs, known Rf from crop taxa, Arabidopsis Rf homologs, and non-Rf PPRs from M. guttatus and other taxa. At, Arabidopsis thaliana. Numbers following M. guttatus PPRs refer to the scaffold followed by a PPR-specific number. Parentheses for M. guttatus PPRs indicate that the PPR has 13–16 motifs, and the number refers to the number of motifs. The inset expands the M. guttatus clade that contains all 16 (of the 17 total) PPRs used in this analysis within the region containing Rf1/Rf2.

Evidence that PPRs restore CMS in Mimulus hybrids:

We began fine mapping with a strong hypothesis that Mimulus Rf was a PPR gene, given known molecular mechanisms of restoration in other systems and the nature of the Mimulus mitochondrial CMS gene (Case and Willis 2008). Four of the five Rf genes discovered to date have been PPRs and, in three taxa (petunia Rf, radish/Brassica Rfo, and rice Rf1a and Rf1b), the Rf PPRs are homologous despite spanning the plant kingdom and the monocot/dicot split (Hanson and Bentolila 2004; Lurin et al. 2004; Klein et al. 2005). Very preliminary evidence suggests that the cotton Rf2 restorer may also be a PPR, adding gametophytic restoration to the above sporophytic systems (Wang et al. 2007). Only a few members of this recently discovered and very large family of genes have been functionally characterized (Lurin et al. 2004), but, in general, PPR genes appear to function in the modification of organellar mRNA transcripts (Andres et al. 2007). Restoration of CMS, which occurs by modification of chimeric mitochondrial transcripts in all known cases (Bentolila et al. 2002), including Mimulus (Case and Willis 2008), falls within this broad functional role.

As in other genomes (Geddy and Brown 2007; O'Toole et al. 2008), restorer-like PPRs appear to cluster in Mimulus (Figure 1; Table 1). Most notably, the adjacent genome scaffolds (s14 and s97) containing Rf1 and Rf2 have very high densities of restorer-like PPRs (Table 1), with s14 containing the highest number of PPRs of any scaffold in the Mimulus genome (Figure 1, Table S2). Although it is remotely possible that some other gene(s) in this region encodes Rf1 and Rf2, this coincidence strongly suggests that Mimulus Rf loci are also restorer-like PPRs. The Mimulus Rf candidates on s14 and s97 have the P-type PPR motif shared by all known Rf PPRs (Lurin et al. 2004; Geddy and Brown 2007), and most also have 13–16 motifs (Figure 2). While PPR motif number in Arabidopsis ranges from 2 to 26 (Lurin et al. 2004), the known Rf loci all range from 13 to 16 motifs: petunia (14 motifs), rice (16 motifs), sorghum (14 motifs), and radish/Brassica (16 motifs). The significance of motif number for fertility restoration is unknown; it is likely that similar motif numbers reflect homology and not direct effects on restoration. Sequence polymorphisms, and not motif number, differentiate fertility restoring Rf and nonrestoring rf in sorghum and radish (Koizuka et al. 2003; Klein et al. 2005), and deletions on the promoter region differentiate Rf and rf in petunia and rice (Bentolila et al. 2002; Kazama and Toriyama 2003; Komori et al. 2004). If we restrict our candidate genes to mitochondrially targeted PPR genes with 13–16 repeat motifs in the relevant regions of s14 and s97, there is one remaining strong candidate gene for Rf2 and six for Rf1 (Figure 2). Given their sequence similarity (see below) and difficulties with the genome assembly in this chromosomal region, positionally cloning and functional confirmation of the individual Rf1 and Rf2 genes will likely be difficult. However, we are confident that these few candidate PPRs accurately represent the functional category of Mimulus Rf genes.

Commonalities with CMS restoration in other systems:

In addition to the features used in their identification, the candidate PPRs for Mimulus Rf loci share several features with molecularly characterized Rf loci in other systems. First, all of the PPRs in these intervals (both mitochondrially and nonmitochondrially targeted), cluster together in a single distinct clade including the rice Rf locus (Kazama and Toriyama 2003; Komori et al. 2004), with petunia and sorghum Rf genes as the closest sisters (Bentolila et al. 2002; Klein et al. 2005). Because PPRs appear to evolve very rapidly and sequence alignments are complicated by variation in motif number (Rivals et al. 2006), we do not want to overinterpret this finding. However, it is clear that our candidates share close amino acid sequence homology with restorers in several other systems.

Second, it appears that the PPRs in the Rf region expanded via tandem duplication and/or transposition, as they are more closely related to each other than to PPRs elsewhere in the Mimulus genome. This appears to be a common pattern for PPR genes at multiple evolutionary scales. PPR proteins are present in low copy numbers in other eukaryotes (five in yeast, six in humans, and two in Drosophila), but have undergone a massive expansion in land plants, with >450 in Arabidopsis and rice (Lurin et al. 2004; Andres et al. 2007). This primary expansion is thought to have occurred early in land plant evolution, through whole-genome duplications and retrotransposition (Geddy and Brown 2007; Kato et al. 2007; O'Toole et al. 2008). The Rf-like PPRs appear to have expanded more recently than the rest of the gene family and show particularly strong evidence for duplication, transposition, and diversifying selection (Geddy and Brown 2007; O'Toole et al. 2008). Unlike other PPRs, the Rf PPRs in rice, radish, and petunia, as well as the Rf-homolog PPRs in Arabidopsis, exist in clusters suggesting relatively recent local duplication (Hanson and Bentolila 2004; Lurin et al. 2004; Kato et al. 2007). This pattern, which is also evident with the Rf-associated PPRs in Mimulus, is consistent with ongoing selection for new (or more) PPRs by selfish evolution of mitochondrial CMS (Geddy and Brown 2007).

Third, the presence of more than one Rf allele in M. guttatus parallels other taxa. In the CMS-BT strain of rice, fertility restoration is achieved by at least two tightly linked Rf alleles, Rf1a and Rf1b, either of which is sufficient to restore male fertility (Wang et al. 2006). A similar situation exists in the Ogura CMS of radish, in which Rfo and Rfo2 are tightly linked and both sufficient to restore male fertility (Wang et al. 2008), along with at least several more loci with complicated effects (Nieuwhof 1990; Bett and Lydiate 2004). Recent work with the CMS-C system in maize suggests two tightly linked Rf loci with dominant effects, Rf4 and Rf5, only one of which is sufficient to restore fertility over all CMS-C types (Hu et al. 2006). In cotton, the sporophytic Rf₁ and gametophytic Rf₂ are also tightly linked (Zhang and Stewart 2001; Wang et al. 2007). It is not yet clear what evolutionary processes contribute to this pattern, but it is intriguing, particularly given population genetic theory about the cospread of CMS and Rf alleles (see below).

Evolutionary implications:

The genetic and molecular characterization of Mimulus Rf sets the stage for understanding the evolutionary dynamics of a major source of plant hybrid incompatibility. CMS genes are largely generated through mitochondrial genome rearrangements, which are thought to be common in plants (Palmer and Herbon 1988). This mutational input, along with strong selection favoring any mitochondrial variant that can achieve increased female fertility through decreases in male function (Frank 1989; Atlan et al. 1992; Ashman 1994), generates the conditions for repeated selfish mitochondrial evolution. This favors nuclear Rf alleles, which under some conditions can be polymorphic and maintain gynodioecy, but are more commonly expected to spread to fixation along with the CMS gene (Charlesworth and Ganders 1979; Hurst et al. 1996; Burt and Trivers 2006). This would leave populations with no phenotypic evidence of past conflict, but with genotypic traces of rapid divergence in the cytoplasmic and nuclear loci involved, and potential hybrid incompatibilities with non-Rf populations (Hurst et al. 1996). Although high frequencies of anther sterility have never been reported in numerous field studies of this genus (Wu et al. 2008), it is likely that Iron Mountain M. guttatus (like hermaphroditic crop plants with hybrid CMS) has experienced the history of selfish mitochondrial evolution and nuclear coevolution predicted by these models. However, unlike crop plants with a selective and demographic history complicated by domestication, M. guttatus provides the opportunity to explicitly test for the molecular population genetic signature of the coevolutionary process underlying this extremely common phenomenon (Kaul 1988). As in the few animal systems in which D-M incompatibility genes have been characterized (Wittbrodt et al. 1989; Ting et al. 1998; Barbash et al. 2003; Presgraves et al. 2003; Brideau et al. 2006; Masly et al. 2006; Phadnis and Orr 2009; Tang and Presgraves 2009), it may be that selfish evolution and compensatory coevolution are major drivers of hybrid breakdown in plants.

Although we cannot yet say whether the Mimulus cytonuclear incompatibility evolved via selfish mitochondrial evolution and nuclear coevolution, the finding that the Mimulus Rf loci map to a cluster of PPR genes provides circumstantial evidence for such a scenario. PPRs function by modifying organellar transcripts, and precise transcript modification is necessary for removal of the chimeric ORF associated with NAD6 in Iron Mountain M. guttatus (Case and Willis 2008). The finding that cytonuclear hybrid male sterility in plants commonly involves interactions between rearranged respiratory proteins and PPRs (Hanson and Bentolila 2004) argues for a common, and predictable, evolutionary process. The theory of selfish CMS evolution provides such a process (Hurst et al. 1996; Burt and Trivers 2006; Geddy and Brown 2007) and, unlike other scenarios for the evolution of hybrid sterility, predicts that male sterility should have no negative effect on female fitness. Of course, in the absence of strong evidence for selection on Mimulus CMS and/or restorer loci, it remains possible that one or both restorers predates the CMS, such that male sterility was never expressed. However, the chance fixation of a novel neutral mitochondrial mutation in the very large (census size >10,000 annually), predominantly outcrossing Iron Mountain M. guttatus population is difficult to envision.

The particular genetics of nuclear restoration in Mimulus raises additional questions about the origins of cytonuclear hybrid incompatibilities and mating system variation in plants. Traditional discussions of CMS assume a matching-alleles model with the introduction and increase of a novel CMS-inducing mitochondrial gene driving the selection of a unique and matching nuclear restorer (Frank 1989). Loss of the particular CMS is predicted to lead to loss of the matching restorer, likely through a negative fitness cost of the restorer (Bailey 2002). The presence of two linked loci in Mimulus, each capable of effecting fertility restoration, as well as more complicated genetic details known from other systems, challenges the simplicity of this model. These details underscore the importance of incorporating the increasingly readily available molecular and genetic information about hybrid incompatibilities into future theoretical models.