A previous study has shown that incompatibility between a S. bayanus nuclear gene, AEP2, and S. cerevisiae mitochondria causes interspecific F2 hybrid sterility [14]. To examine whether nuclear-mitochondrial incompatibility represents a general mechanism of reproductive isolation in yeast, a systematic screen for such genes was conducted in three closely related yeast species: S. cerevisiae (Sc), S. paradoxus (Sp), and S. bayanus (Sb). Hybrid diploid strains between S. cerevisiae and S. paradoxus, or between S. cerevisiae and S. bayanus were induced to generate haploid spores containing different combinations of chromosomes from their parental species. These spores were then assayed for cytonuclear incompatibility. In hybrid diploids between species A and B, incompatibility could occur in two directions, between A-nucleus and B-mitochondria or between B-nucleus and A-mitochondria. When generating the yeast hybrids, we deliberately removed one parental type of mitochondria (by using ρ ⁰ mutants, which lack mitochondrial DNA) so that we could unambiguously assign the direction of incompatibility. After sporulation of hybrids, viable spores were measured for their ability to grow on glycerol, a non-fermentable carbon source (Figure 1). The results showed strong cytonuclear incompatibility in most of the interspecific hybrids; we observed 66%±5%, 78%±7%, and 32%±16% of respiration-proficient spores in the interspecific crosses between Sc-ρ ⁰ and Sp, Sb-ρ ⁰ and Sc, and Sc-ρ ⁰ and Sb, respectively, while the intraspecific crosses generated almost 100% of respiration-proficient spores (Figure 2).

The percentage of viable spores that could grow on glycerol plates was used to estimate the number of strongly incompatible loci (see Materials and Methods). The results of this growth assay suggest that there are only one or few nuclear genetic loci strongly incompatible with mitochondria in the Sc-nucleus and Sp-mitochondria, the Sb-nucleus and Sc-mitochondria, and the Sc-nucleus and Sb-mitochondria pairs. Although no strong incompatibility was detected for the Sp-nucleus and Sc-mitochondria pair, slow spore growth (on glycerol plates) was commonly observed, suggesting that some incompatibility may exist in this pair as well. Because the incompatibility between the Sb-nucleus and Sc-mitochondria has been described earlier [14], we focus here on the remaining two pairs.

In our experimental design, we constructed the hybrid diploids using spo11Δ mutants to prevent meiotic homologous recombination. We expected that hybrid F1 haploids unable to respire (see Figure 1) should carry a specific set of chromosomes containing the incompatible genes. To identify the genes responsible for the observed cytonuclear incompatibility, the respiration-deficient clones were first examined for their chromosome contents using species-specific PCR (see Materials and Methods). In the cross between Sc-ρ ⁰ and Sb, the haploids carrying Sb-mitochondria but unable to respire lacked Sb-Chromosome 6, 9, or both (Table S1A). When these hybrid clones were transformed with genomic DNA libraries to screen for those genes capable of rescuing the respiratory defect, Sb-MRS1 encoded on Sb-Chromosome 9 and Sb-AIM22 encoded on Sb-Chromosome 6 were isolated. The results from the cross between Sc-ρ ⁰ and Sp are more complex, as all the respiration-deficient clones did not contain the two Sp-Chromosomes 4 and 9 (Table S1B), but were fully rescued by Sp-MRS1 alone, which is encoded on Sp-Chromosome 9. The reason why all the hybrid clones are missing Sp-Chromosome 4 is unclear. One possibility is that Sp-Chromosome 4 is also incompatible with Sc-Chromosome 9, an issue requiring further investigation.

To rule out the possibility that the respiration defects caused by Sc-MRS1 and Sc-AIM22 were simply due to nuclear-nuclear incompatibility in the hybrid haploid clones, we crossed Sp-mrs1Δ, Sb-mrs1Δ, and Sb-aim22Δ with Sc-ρ ⁰ and examined the respiratory ability of the diploid cells. All the diploid cells were still respiration-deficient, even though they contained a complete set of the S. cerevisiae genome (Figure 3). This result demonstrated that Sc-MRS1 is incompatible with Sp-mitochondria and that both Sc-MRS1 and Sc-AIM22 are incompatible with Sb-mitochondria.

Yeast cells utilize non-fermentable carbon sources to induce meiosis. Previous studies have shown that respiration-deficient cells were unable to sporulate [34]. To confirm that indeed the cytonuclear incompatibility observed in our experiments contributes to reproductive isolation, the aforementioned diploid cells (Sp-mrs1Δ×Sc-ρ ⁰, Sb-mrs1Δ×Sc-ρ ⁰, and Sb-aim22Δ×Sc-ρ ⁰) were grown on sporulation medium and examined for their sporulation efficiency. No ascus was observed in these cultures, while the control cultures sporulated efficiently. Thus, cytonuclear incompatibility caused by Sc-MRS1 or Sc-AIM22 results in reproductive isolation between these yeast species.

Mrs1 is a mitochondrial protein required for excision of the aI5β intron in COX1 and the bI3 intron in COB [35],[36]. A previous study has shown that S. douglasii Mrs1 (S. douglasii is a synonym of S. paradoxus) is required to splice a S. douglasii-specific COX1 intron not existing in the S. cerevisiae COX1 [37]. We observed a similar result in S. bayanus. Sc-MRS1 could not complement the respiratory defect of the Sp-mrs1Δ or Sb-mrs1Δ mutants (Figure S1). When Sb-Mrs1 or Sp-Mrs1 were replaced by Sc-Mrs1, the level of mature COX1 mRNA was drastically reduced but the COB mRNA was not affected (Figure 4A). We further analyzed the translation products from purified mitochondria and confirmed that only the Cox1 protein was missing (Figure 4B). By contrast, Sc-MRS1 transcription and protein transport into mitochondria appear to be normal in both S. paradoxus and S. bayanus (Figure 5). Thus the incompatibility between the mitochondrial and nuclear genomes is most likely due to a change in the splicing specificity of Mrs1 that occurred after S. cerevisiae diverged from the common ancestor of S. cerevisiae and S. paradoxus.

In order to understand how the COX1 introns evolved in yeast, we compared COX1 intron patterns between S. cerevisiae, S. paradoxus, S. bayanus, S. servazzii, Candida glabrata, and Kluyveromyces thermotolerans. S. servazzii and C. glabrata are species outside the sensu stricto complex and K. thermotolerans is a pre-WGD (whole-genome duplication) species. The comparison indicates that the intron in the Sp- or Sb-COX1 gene incompatible with Sc-Mrs1 is an ancient intron (Figure 4C). Since this intron was eliminated only in the S. cerevisiae lineage, it is likely that the Sc-MRS1 gene product lost the ability to splice this intron after the intron loss event (by adaptation or drift).

The fact that S. cerevisiae and S. paradoxus share a high degree of nucleotide sequence identity allowed us to determine the key mutations underlying the functional change of the MRS1 gene and to reconstruct the process of Mrs1 evolution. To this end, chimeric proteins with regions from Sc-Mrs1 and Sp-Mrs1 were constructed and assayed for their ability to complement the respiratory defect of the Sp-mrs1Δ mutants. We found that the functional difference between Sc-Mrs1 and Sp-Mrs1 is mainly determined by a region comprising 63 amino acids (a.a. sites 179–241; Figure 6A). Nine nonsynonymous changes have accumulated in this region since Sc-MRS1 and Sp-MRS1 diverged (Figure S2A). To determine which of these mutations led to the altered activity of Mrs1, we introduced the S. paradoxus version of each of these sites into Sc-MRS1 and assayed their ability to rescue the Sp-mrs1Δ respiratory defect. Analogous experiments were performed in the reverse direction by introducing Sc-specific amino acids into Sp-MRS1. Only mutations in three amino acids (Sc to Sp: T201A, V211A, and M227I) had obvious contributions (Figure 6B and Figure S2). When all three mutations were combined together in a single mutant clone, it explained most of the effect. The growth rate of cells carrying the mutant plasmid (Sc-MRS1-n123) in a glycerol-containing medium is about 75% of that of wild type cells. Interestingly, these three amino acids are all conserved in S. paradoxus, S. kudriavzevii, and S. bayanus but are changed in S. cerevisiae. Among the other six nonsynonymous changes in this region, only two of them (Sc to Sp: K186E and R223Q) share the same pattern. This observation is consistent with our hypothesis that the functional change of MRS1 occurred only after an ancestral COX1 intron was lost in S. cerevisiae. Our results clearly demonstrate that the observed nuclear-mitochondrial incompatibility results from cumulative effects of multiple mutations. On the other hand, they also suggest that only a small fraction of the nonsynonymous changes between species contributes to the incompatibility.

The Mrs1 protein does not contain any specific functional domain. However, a recent study has used computer modeling to predict the Mrs1 protein structure [38]. We examined the relative positions of these residues using the predicted structure. All three residues (a.a. 201, 211, and 227) were found to localize on the RNA-binding surface (Figure S3). It is possible that these amino acid changes have altered the substrate specificity of Mrs1 that leads to the incompatibility.

S. cerevisiae nuclei can only support S. bayanus mitochondria if the cells contain a S. bayanus AIM22 gene. AIM22 encodes a lipoate-protein ligase homologous to the bacterial lplA protein [39],[40]. In eukaryotic cells, lipoic acid has been shown to be an essential cofactor to a variety of mitochondrial proteins and lipoate-protein ligase (together with other enzymes) is required to lipoylate these mitochondrial targets [39],[41]. We have not investigated which mitochondrial protein or enzyme in Sb-mitochondria is incompatible with Sc-Aim22. However, our data indicate that the incompatibility is not caused by misregulation of the Sc-AIM22 transcription or failure to transport the Sc-Aim22 to Sb-mitochondria (Figure 5).

To investigate whether the AIM22 gene in the S. cerevisiae-S. paradoxus branch has been under positive selection during evolution, we measured the ratio of nonsynonymous (Ka) to synonymous (Ks) nucleotide substitution rates between these species. The Ka/Ks values of AIM22 in the S. cerevisiae–S. paradoxus, S. cerevisiae–S. bayanus, and S. paradoxus–S. bayanus pairs are 0.13, 0.12, and 0.14, showing no sign of positive selection (Ka/Ks>1). We also ran a PAML's branch-model analysis on the AIM22 gene [42],[43] but could not detect any signature of significant positive selection (Figure S4).

Previous studies of AEP2 have shown that the nuclear-mitochondrial incompatibility is asymmetrical. While Sb-Aep2 is completely incompatible with Sc-mitochondria, Sc-Aep2 retains partial compatibility with Sb-mitochondria [14]. We also found that incompatibility caused by AIM22 and MRS1 only occurred in one direction. Although Sc-AIM22 and Sc-MRS1 are not compatible with Sb-mitochondria, Sb-AIM22 could complement the Sc-aim22Δ mutant and both Sb- and Sp-MRS1 could rescue the Sc-mrs1Δ mutant.

Nuclear-mitochondrial incompatibility has been shown to occur commonly between different yeast species [32],[33]. However, it is unclear whether nuclear-mitochondrial incompatibility between different pairs of species has evolved at different periods of time in different species lineages. To address this issue, we tested the compatibility between different orthologues of these incompatible genes and mitochondria from each species. Different orthologous alleles of MRS1, AIM22, or AEP2 were transformed into the S. cerevisiae, S. paradoxus, or S. bayanus mutants in which the wild-type copy had been deleted. The transformants were then tested for their ability to grow on glycerol plates (Figure 7A and Figure S1). Information from these assays was used to deduce the time of occurrence of the functional change leading to incompatibility. A clear correlation between the emergence of cytonuclear incompatibility and the phylogeny is observed (Figure 7B). Sc-MRS1 is incompatible with Sp-mitochondria or Sb-mitochondria, indicating that the functional change of Sc-MRS1 occurred only in the S. cerevisiae lineage. On the other hand, the functional change of AIM22 represents a more ancient event in the common ancestor of S. cerevisiae and S. paradoxus, because Sc-AIM22 and Sp-AIM22 are exchangeable but neither of them is compatible with Sb-mitochondria. Finally, the data suggest that Sb-AEP2 diverged in function only in the S. bayanus lineage since Sb-AEP2 is incompatible with either Sc-mitochondria or Sp-mitochondria. These results provide evidence that nuclear-mitochondrial incompatibility has repeatedly arisen during the history of yeast evolution and probably represents an important reproductive isolation mechanism in yeast species.

Yeast strain genotypes are listed in Table S2. The parental S. cerevisiae strains (JYL1127 and JYL1128) are isogenic with W303 (MATa ura3-1 his3-11,15 leu2-3,112 trp1-1 ade2-1 can1-100). The parental S. paradoxus strains (JYL1137 and JYL1138) are derived from YDG 197 and are a gift from Dr. Duncan Greig (University College London, UK). The parental S. bayanus strains (JYL1030 and JYL1031) were derived from a strain (S. bayanus #180) collected by Dr. Duccio Cavalieri (University of Florence, Italy). The strains JYL1157, 917, and 1256 were used for measuring hybrid fertility. Substitutive and integrative transformations were carried out by the lithium acetate procedure [64]. Media, microbial, and genetic techniques were as described [65].

a and α cells of one species (S. cerevisiae, S. paradoxus, or S. bayanus) were crossed to α and a ρ ⁰ cells of another species to generate F1 hybrid cells. In both species, the SPO11 genes were deleted to prevent meiotic recombination between homologous chromosomes. Strains from the first species also had a URA3 marker inserted near the centromere of a chromosome so that haploid spores could be efficiently selected on 5-FOA plates. After the hybrid diploids were induced to sporulate, viable spores were tested for their respiratory ability. Respiration-deficient clones were further crossed with ρ ⁰ mutants of the parental species to confirm that the defect was caused by cytonuclear genomic incompatibility (Figure 1). A genetic analysis was used to estimate the number of the nuclear genes that are incompatible with mitochondrial DNA from another species: in a cross between two spo11Δ mutants, meiotic recombination does not occur and homologous chromosomes segregate randomly [66],[67]. For a specific chromosome A, 25% of the spores will carry two A chromosomes (of both parental types), 50% of them will carry one A chromosome, and 25% of them will have no A chromosome (these cells will not survive so they will not be counted in our later analysis). If only one gene on chromosome A is recessively incompatible with mitochondrial DNA (in our case, all spores from a single cross carry the same parental type of mitochondrial DNA), we expect to see that 66% of the viable spores are Gly+ (1/3+2/3×1/2 = 66%). The spores carrying two A chromosomes should be Gly+ because the incompatibility is recessive. Half of the spores that carry only one A chromosome should be Gly+ if their A chromosome is from the same parent of the mitochondrial DNA. If two genes on different chromosomes are involved, we expect to see that 43% of the viable spores are Gly+ (66%×66% = 43%). Epistatic effects are not taken into account in this analysis because considering such effects would make it impossible to estimate the involved locus number.

We examined the chromosome composition of respiration-deficient hybrid lines by PCR using species-specific primers for all 16 chromosomes. Yeast genomic DNA libraries constructed from S. bayanus or S. paradoxus genomes were transformed into the Gly− clones to screen for incompatible genes.

Yeast strains were grown in 3 ml YPD liquid cultures at 30°C to stationary phase and total RNA was isolated using Qiagen RNeasy Midi Kits (Qiagen, Valencia, CA). Ten µg of total RNA was separated on a 1.3% agarose-formaldehyde gel and then transferred to a nylon membrane (Millipore, Billerica, MA). Northern blotting was performed as described [68]. Because both COX1 and COB signals were difficult to be completely washed out and the background was high after the second hybridization, we used the same RNA samples to load three repeats on the same gel, cut the membrane after transferring, and hybridized each repeat with one specific probe. The gene-specific primers for the DIG-labeled probes were described in Rodeheffer et al. [69] except for the probe of COX2. The primers for COX2 were 5′-TTAATGATAGTGGTGAAACTGTTG-3′ and 5′-CCAAAGAATCAAAATAAATGCTCG-3′. The probe generation and hybridization were as described in the Genius System User's Guide (Roche, Indianapolis, IN).

Total RNA was isolated using Qiagen RNeasy Midi Kit (Qiagen). First-strand cDNA was synthesized using High Capacity cDNA Reverse Transcriptase Kit (Applied Biosystems, Foster City, CA) at 37°C for 2 h. A 25-fold dilution of the reaction products was then subjected to real-time quantitative PCR analysis using gene-specific primers, the SYBR Green PCR master mix, and an ABI-7000 sequence detection system (Applied Biosystems). Data were analyzed using the built-in analysis program.

Yeast cells were cultured in selection medium at 30°C to the mid-exponential phase. Mitochondria fractions were prepared as described [70]. The post-mitochondrial supernatant (PMS) fractions were collected immediately following centrifugation of mitochondria. The PMS was precipitated with ice-cold 10% trichloroacetic acid (TCA) and centrifuged at 14K rpm for 15 min at 4°C, followed by an ice-cold acetone wash using the same centrifugation conditions. Laemmli sample buffer was added to the mitochondrial pellets resuspended in TE buffer plus protease inhibitors, and to the PMS fractions. For Western blot analyses, mitochondrial protein (40 µg) was separated by SDS-PAGE using the GE Healthcare Life Science system (Mini-Vertical Units SE260) and then transferred to a BioRad Immun-Blot PVDF membrane in transfer buffer (25 mM Tris base, 200 mM glycine, 20% methanol, 0.01% SDS). The immunopositive bands were visualized by using Western Lightning chemiluminescence reagent (PerkinElmer, Waltham, MA). Anti-G6PDH polyclonal antibody was purchased from Sigma (St. Louis, MO). Anti-c-Myc polyclonal antibody (A-14) was from Santa Cruz Biotechnology (Santa Cruz, CA). Anti-Hsc82 and Anti-Atp2 polyclonal antibodies were obtained from Dr. Chung Wang (Academia Sinica, Taipei).

Cells in the early log phase were inoculated in 2 ml standard minimal medium and grown for 30 min. Cycloheximide stock solution (10 ml/mg in dH₂O) was added to a final concentration of 100 µg/ml. Cells were incubated for 5 min prior to addition of 0.1 mCi of [³⁵S]-methionine. The reaction was terminated after 1 h by adding 2 ml of chase solution (1% casamino acid, 2 mg/ml Na₂SO₄). Mitochondria were prepared as described [65]. The radiolabeled proteins were separated on a 17.5% polyacrylamide gel.

The ancestral AIM22 sequence of S. cerevisiae and S. paradoxus was constructed using a maximum likelihood procedure (free-ratio model) as implemented in the PAML package. The Ka/Ks ratios were calculated using DNAsp 5.0 [71].

The Ka/Ks ratios were estimated using the free-ratio branch model of PAML. This method allows the Ka/Ks ratios to vary among branches in a given phylogeny and is useful in detecting positive selection acting on particular lineages [42]. The AIM22 sequences from five Saccharomyces sensu stricto species (S. cerevisiae, S. paradoxus, S. mikatae, S. bayanus, and S. kudriavzevii) were used in our analysis. Results of the branch-model PAML analysis suggested that no particular lineage was subjected to positive selection (i.e., Ka/Ks>1) (Figure S4). Because both Sc-AIM22 and Sp-AIM22 are incompatible to S. bayanus mitochondria, it is possible that there has been an accelerated sequence evolution in either the lineage leading to S. cerevisiae and S. paradoxus, or in the lineage leading to S. bayanus. To test this hypothesis, we performed an analysis as described in Yang and Nielsen [43]. One ω-ratio and two ω-ratio branch models were implemented. The first model assumes that only one ω-ratio leads to whole phylogeny branches and the second model assumes that one ω-ratio leads to the branch that we are interested in and another ω-ratio leads to the rest of the other branches. Twice the difference of their likelihood ratio between any two models (likelihood ratio test; LRT) was then compared against a chi-square distribution. The degree of freedom (d.f.) was obtained based on the difference of parameters used in two different models. From our analysis, no accelerated evolution was observed in any lineage along the phylogeny of AIM22.