Polyploidization, a state resulting from doubling of a genome within a species (autopolyploidy) or the merging between different species (allopolyploidy) (Campbell et al., 2016), is an important evolutionary force which shapes eukaryotic genomes (Albertin and Marullo, 2012), triggers speciation, and can result in phenotypic changes driving adaptation (Ohno, 1970). A whole-genome duplication (WGD) event occurred approximately 100–200 Mya in the common ancestor of six yeast genera in the family Saccharomycetaceae, including Saccharomyces cerevisiae (as reviewed by Wolfe et al., 2015). WGD was recently proposed to be a direct consequence of an ancient hybridization between two ancestral species (Marcet-Houben and Gabaldón, 2015), followed by genome doubling of initially sterile hybrid to regain fertility, i.e., the ability to undergo meiosis and produce viable spore (Wolfe, 2015).

Different mechanisms can contribute to hybrid infertility, such as chromosomal missegregation caused by meiosis I non-disjunction (Boynton et al., 2018), chromosomal rearrangements (Liti et al., 2006; Rajeh et al., 2018), and Dobzhansky–Muller gene incompatibilities either between nuclear genes (Bizzarri et al., 2016) or between mitochondrial and nuclear genes (Lee et al., 2008). Specialized loci, called the mating-type (MAT)-like (MTL) cassettes, regulate mating between haploid cells with opposite MATa and MATα idiomorphs, as well as meiosis in diploid a/α cells. In diplontic yeast S. cerevisiae MAT locus on chromosome III contains either the a1 or the α1 and α2 genes in Ya and Yα segments, respectively, surrounded by X and Z regions at the left and right sides. In haploid α cells, α1 activates the α-specific genes (αsgs), while α2 represses a cohort of a-specific genes (asgs), which a cells transcribe by default (Haber, 2012). Finally, diploid a/α cells are meiosis but not mating-competent, because the a1-α2 heterodimer positively regulates IME1 (Inducer of Meiosis) gene expression and represses the transcription of RME1, a haploid-specific gene (hsg) that inhibits entry into meiosis, and of other hsgs required for mating responses. S. cerevisiae cells also have extra copies of MAT genes at the HMRa and HMLα loci located close to telomeres of chromosome III and silenced by a combination of the Sir1–4 proteins (Hickman et al., 2011). These extra copies serve as donors during the mating-type switching which enables MATa cells to convert into MATα cells, or vice versa, and to mate each other. This autodiploidization event is triggered by a site-specific endonuclease called HO which induces double-strand break at Z region of the MAT locus. In Saccharomyces interspecies hybrids, experimental deletion of one MAT locus or elimination of the entire chromosome carrying one MAT locus yielded fertile allotetraploids (Greig et al., 2002; Pfliegler et al., 2012; Karanyicz et al., 2017). More recently, the MAT locus damage was proposed to be the most plausible evolutionary route which enables natural interspecies hybrids of the Zygosaccharomyces bailii complex to rescue mating and meiosis (Ortiz-Merino et al., 2017; Braun-Galleani et al., 2018).

In the Saccharomycetaceae lineage, Z. rouxii stands on the crossroad where different and relevant evolutionary events take their way (Dujon and Louis, 2017). This evolutionary route involves ancient allopolyploidization between two parental lineages, one of which was close to Z. rouxii and Torulaspora delbrueckii (ZT) clade (Marcet-Houben and Gabaldón, 2015). Z. rouxii represents the early branching species before WGD that recruits HO from a LAGLIDADG intein to catalyze the first step of mating-type switching (Fabre et al., 2005). Furthermore, Z. rouxii exhibits the triplication of MTL loci, which is a genomic landmark of the Saccharomycetaceae family, but, in contrast to S. cerevisiae, it lacks of MAT-HMR linkage. Whereas the route of αsg regulation appears to be conserved, the regulatory circuit of asgs has been extensively rewired across the Saccharomycotina clade. Instead of the negative regulatory circuit widespread in post-WGD species, several pre-WGD species activate asgs by an HMG-domain protein (a2) that is encoded by MATa (Tsong et al., 2003). Conventionally, Z. rouxii displays haplontic life style, where heterothallic haploid cells with opposite mating-type mate each other or, alternatively, homothallic haploid cells switch mating-type and subsequently undergo mating between mother and daughter cells. In both cases, the transient diploid zygote should sporulate to restore the haploid state. Alternatively, stable allodiploid strains arose from mating between divergent haploid parents. One parental haplotype (called T-subgenome) resembles Z. rouxii and was 15% different from the other parental haplotype (called P-subgenome) (Gordon and Wolfe, 2008; Bizzarri et al., 2016, 2018; Watanabe et al., 2017).

Both haploid and allodiploid strains show highly variable gene arrangements around MTL, suggesting that these loci are recombination hotspot during error-prone mating-type switching events (Watanabe et al., 2013; Solieri et al., 2014). Structural rearrangements are so rampant in these regions that different stock cultures of the same haploid (Watanabe et al., 2013) or allodiploid (Bizzarri et al., 2016; Watanabe et al., 2017) strains can display distinct MTL repertoires. For instance, differences in MTL loci were recently found between two sub-cultures of the allodiploid strain ATCC42981. In our previous work, we found 7 MTL loci in in-house stock of ATCC42981 (termed ATCC42981_R for convenience) (Bizzarri et al., 2016), while Watanabe et al. (2017) detected 6 MTL loci in strain JCM22060, the Japanese stock of ATCC42981. Ectopic recombination between MTL-flanking regions from divergent parental haplotypes yields chimeric arrangements hardly to resolve both by targeted long PCR approaches (Bizzarri et al., 2016) and by genome sequencing technologies based on short reads (Watanabe et al., 2017).

In 2014, the MinION sequencing device (Oxford Nanopore Technology, ONT) was released and initially exploited to sequence and assemble PCR products or microbial genomes (Jain et al., 2016). Recent improvements in protein pore (a laboratory-evolved Escherichia coli CsgG mutant named R9.4), library preparation techniques (1D ligation and 1D rapid), sequencing speed (450 bases/s), and control software enabled the usage of Nanopore sequence data, in combination with other sequencing technologies, for assembling eukaryotic genomes including yeasts, nematodes and human (Istace et al., 2017; Jansen et al., 2017; Yue et al., 2017; Jain et al., 2018). The main advantage of ONT is that reads can reach tens of kilobases (Jain et al., 2016), making more easy to resolve repeat regions and to detect structural variation. Recently, the genome of allodiploid strain ATCC42981_R was sequenced and assembled through a de novo hybrid strategy which combined MinION long and Illumina short reads (Bizzarri et al., 2018).

Here, we took advantage from the newly released genome of ATCC42981_R (Bizzarri et al., 2018), in order to resolve incongruences in the highly dynamic MTL loci. Furthermore, we deleted the expressed MATα^P locus in ATCC42981_R to test whether the loss of MAT heterozygosity can induce genome doubling and rescue fertility in allodiploid cells of the ZT clade.

Our study is the first to combine the Nanopore whole-genome sequencing to conventional PCR-based methods in order to survey MTL loci in a Z. rouxii allodiploid genome. This yeast is particularly prone to outbreeding and provides a particularly appealing platform to study genome re-shaping after the merger of two parental subgenomes. Recombination and introgression between subgenomes have been rampant in hybrid yeasts, resulting in loss of heterozygosity and gradual genome reduction (Sipiczki, 2008). In Z. rouxii MTL loci markedly contribute to this genomic plasticity (Watanabe et al., 2013; Solieri et al., 2014). As consequence, this species frequently undergoes chromosomal translocations at the MTL loci, which make hard the understanding of true cell identity by simple MTL genotyping. For example, haploid Z. rouxii strain CBS732^T switched mating-type at the CHA1-MAT-SLA2 locus (Bizzarri et al., 2018), suggesting that CHA1 gene flanks the actively transcribed MAT locus instead of DIC1. Several assortments of different flanking gene variants and distinct idiomorph encoding genes make challenging and laborious to resolve the complex genetic MTL architecture by PCR targeted approaches. For these reasons, we generated a high-quality genome assembly in order to dissect complex rearrangements at the MTL loci that were not fully resolvable from the earlier survey based only on long-range PCR amplification (Bizzarri et al., 2016). One of the major advantages of the ONT is the possibility of sequencing very long DNA fragments, which span the entire MTL cassettes. This strategy assures to accurately reconstruct gene order around different MTLs. On the other hand, using noisy ultra-long reads for self-correction and assembling of highly heterozygous genomes can affect the consensus sequence accuracy and the parental haplotypes sorting. In case of ATCC42981_R, distinguishing between homeologous sequences is further challenging as only the Z. rouxii parental genome is available to guide homeologous scaffold assembly. Error rate made necessary to polish MinION reads with Illumina-derived reads, resulting into DBG2OLC-driven hybrid de novo genome assembly (Bizzarri et al., 2018). However, our result showed that a single “best assembler” does not exist to resolve highly heterozygous and highly repeated MTL regions. DBG2OLC assembly suffers from poor performance in certain sequence contexts, such as in regions with low coverage or regions that contain short repeats. Besides, the new assembly generated with MaSuRCA showed higher sequencing accuracy compared to DBG2OLC, but loses some MTL cassettes. As bottom-end validation step, PCR approach was used to discard artificial MTL arrangements arisen from flawed contig assemblies. This strategy resolves controversies over MTL loci in ATCC42981_R genome derived from the analysis of the Japanese stock JCM22060 (Watanabe et al., 2017).

Reconstruction of MTL structure indicates that ATCC42981_R resembles CBS4837 for the exception of an additional scaffold containing DIC1^T-MTLα^P-SLA2^P linked to CHA1_L^T-MTLα^T-ZYRO0F18634g^T (Figure 3). This assessment was congruent with previous PFGE-Southern blotting which showed two signals for MATα-specific probe (Bizzarri et al., 2016). The most significant difference between ATCC42981_R and JCM22060 is that ATCC42981_R harbors the transcriptionally active MATa^N cassette in addition to the expected MATα^P. Differently from Z. parabailii (Ortiz-Merino et al., 2017), MATa^N cassette of ATCC42981_R contains MATa1 gene. This means that Z. rouxii retains the ancestral regulatory circuit based on a1–α2 heterodimer as diploid cell sensor (Booth et al., 2010). Watanabe et al. (2017) showed that strain JCM22060, which contains only MATα^P, mates the tester strain a in a medium containing Shoyu-koji extract. By contrast, we did not find any evidence of meiosis or mating in ATCC42981_R (Bizzarri et al., 2016), when grown on the media reported in literature to promote Z. rouxii mating and sporulation (James and Stratford, 2011). Watanabe et al. (2017) argued that difference in medium composition could account for the phenotypic discrepancy between ATCC42981_R and the sister stock JCM22060. As the Shoyu-koji extract is difficult to gain in western countries, we cannot rule out this hypothesis. Otherwise, heterozygosity at the MAT locus could significantly contribute to the allodiploid infertility. In particular, the hybrid heterodimer with divergent a1 and α2 subunits brings the cell in an ‘haploid-diploid intermediate’ functional state which hamper both the meiosis commitment and the responsiveness to mating stimuli (Bizzarri et al., 2016).

In Saccharomyces clade, experimental deletion of one MAT locus leads to allotetraploids suitable to undergo meiosis (Greig et al., 2002; Pfliegler et al., 2012). Similarly, Z. parabailii and Z. pseudobailii hybrid strains ATCC60483 and MT15 were recently supposed to be fertile due to the accidental breakage of 1 of the 2 homeologous copies of the MAT locus (Ortiz-Merino et al., 2017; Braun-Galleani et al., 2018). A prediction of this model is that artificial deletion of one MAT locus in Zygosaccharomyces cells should override the arrest in mating commitment. In our model, ATCC42981_R cells did not behave as haploids with idiomorph a, when the MATα^P locus was deleted. This suggests that mechanism underpinning the cell identity in Z. rouxii hybrids could be different from those involved in cell identity regulation of the sister species Z. parabailii and Z. pseudobailii.

Gene deletion of transcriptionally active MATα^P locus did not rescue the ability to produce conjugated asci in ATCC42981_R, while the persistence of α1 and α2 transcripts suggests that HMLα silencing was leaky in ATCC42981_R. Consequently, α^P genes either from CHA1_L^T-MTLα^P-SLA2^P or CHA1_L^P-MTLα^P-ZYRO0F18634g^P are transcriptionally active in MATα^PΔ mutants. Strain NBRC110957, which does not have the CHA1_L^T-MTLα^P-SLA2^P cassette, uses CHA1_L^P-MTLα^P-ZYRO0F18634g^P as donor during switching from a^P to α^P (Watanabe et al., 2017). This suggests that CHA1_L^P-MTLα^P-ZYRO0F18634g^P cassette is most likely silenced and that α^P could be expressed by the CHA1_L^T-MTLα^P-SLA2^P in ATCC42981_R. Congruently, strain CBS4837 actively transcribed genes from CHA1_L^T-MTLa^P-SLA2^P cassette. These findings make less probable the alternative hypothesis that MATα^P deletion induces HMLα cassette de-silencing. Abnormal expression of cryptic HMR/HML loci has been described in Vanderwaltozyma polyspora, the Z. rouxii closest relative that branched after WGD (Roberts and Van der Walt, 1959). Consequently, V. polyspora haploid cells behave as a/α diploid and appear mating-incompetent for many generations only to subsequently restore silencing. Significantly, V. polyspora lacks of Sir1 histone deacetilase, which mediates the HM loci silencing in S. cerevisiae together with the SIR complex (Sir2/Sir3/Sir4). In S. cerevisiae failure to recruit Sir1 is thought to account for the instability of subtelomeric silencing relative to HM loci (Chien et al., 1993). Like V. polyspora, Candida glabrata is another species close to Z. rouxii, which lacks of a SIR1 ortholog (Gabaldón et al., 2013). A defective silencing system leads to the expression of MATa genes in C. glabrata MATα cells (Muller et al., 2008) and makes HML more prone to HO cleavage at the Y/Z junctions (Boisnard et al., 2015). Z. rouxii has the archetypal member of the SIR1 family, KOS3 (Kin of Sir1 3) (Gallagher et al., 2009). In pre-WGD species Torulaspora delbrueckii KOS3 located ∼1 kb away from HMR and plays a key role in HML/HMR silencing (Ellahi and Rine, 2016). Strikingly, in ATCC42981_R we also found two KOS3 copies, KOS3^T and KOS3^P, upstream of HMRa^T and HMRa^P loci, respectively. In addition, Sir1 and the components of SIR complex have been reported to rapidly evolve in the Saccharomycetaceae family. This could potentially jeopardize the efficiency of the silencing machinery in interspecific hybrids. For example, Sir1, Sir4 and the cis-acting silencer sequences are incompatible in S. cerevisiae × S. uvarum hybrids (Zill et al., 2010, 2012). In ATCC42981_R, heterochromatin formation across silent loci could be less effective due to the incompatibility in the silencing machinery between the T- and P-subgenomes. Watanabe et al. (2017) suggest that chimeric MTL cassettes could display epigenetic expression control when only E silencer sequence is maintained around MTL locus. This could produce allodiploid single cells which undergo epigenetic silencing at one of MAT loci and restore fertility. In ATCC42981_R two DIC1-MAT-SLA2 cassettes assure active transcription of opposite idiomorphs, while the presence of E silencer only at the right side of HMLα^P locus could unlock the silencing and mask the loss of heterozygosity at the MAT locus induced by MATα locus deletion.

Strikingly, the depletion of α^P1 and α^P2 genes switched off the a2 but not the a1 gene transcription. Moreover in both deleted and wild type strains two a1 alternative spliced isoforms are present, one of them compatible with the retention of first intron. In S. cerevisiae exon–intron structure is conserved and the retention of first intron resulted in a functional a1 transcriptional factor that prevents mating (Ner and Smith, 1989). Since α1 activates the αsgs in the ancestral circuit of yeast cell identity (Baker et al., 2011), we rule out the possibility that α1 is involved in a2 gene repression. In S. cerevisiae, α2 represses asgs by binding asgs cis-regulatory sequences cooperatively with a MADS-box transcription regulator, Mcm1 (Tsong et al., 2003). Z. rouxii, which branched from the S. cerevisiae lineage prior to the loss of a2 gene, should maintain both the a2 activation and the α2 repression of asgs (Tsong et al., 2006; Baker et al., 2012). In Lachancea kluyveri haploid cells, α2 deletion induces the transcription of the asgs AGA1 and AGA2, while a2 deletion decreases the asgs transcript levels (Baker et al., 2012). However, to the best of our knowledge, no evidence has been provided until now about the consequences of α2 gene deletion in diploid cells which retain a2 gene. As a1 is still expressed in Z. rouxii MATαΔ/MATa hemizygous cells, we speculate that a2 silencing could be a promoter-driven event directly or indirectly regulated by α2. Furthermore, in our MATαΔ/MATa model, the asgs were expressed even when a2 was switched off by the MATα2 deletion, suggesting the existence of a different asgs regulatory network in the ATCC42981_R hybrid compared to Z. rouxii.

This study revised the pattern of MTL loci in allodiploid strain ATCC42981_R. By taking advantage from ONT technology, we captured a novel MATa cassette which did not correspond to the expected T and P counterparts, providing preliminary evidences that a third haplotype contributes to this genome. The differences between ATCC42981_R and JCM22060 support that MTLs are a root source of genetic variation, leading to novel chimeric MTL cassettes, different cell identities and, consequently, distinct phenotypic behaviors. While further researches are required to investigate mechanisms responsible of this extensive MTL reshaping, our results confirm that these yeast stocks are genetically unstable (Watanabe et al., 2013; Bizzarri et al., 2018). We also demonstrated how HMR/HML silencing is crucial to establish the cell identity, as leakage in HML silencing prevents allodiploid MATα^PΔ cells to behave like haploids. How allodiploid cell modulates a2 expression via α2 transcriptional factor represents an unexplored regulatory circuit that has to be investigated in future.

The whole genome sequence datasets generated for this study can be found under the NCBI BioProject number PRJEB26771.

SC and LS contributed conception and design of the study. MB conducted the experiments described in this study. LB contributed to in vitro PCR validation and asg expression. HS and MD contributed to deletion mutant construction. SC and LP performed bioinformatic analysis of the whole genome sequence data. LS wrote the manuscript. SC and MB contributed to draft revision. All authors read and approved the final manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.