Engineering recombination between diverged yeast species reveals genetic incompatibilities

The major cause of the sterility of F1 hybrids formed between Saccharomyces cerevisiae and Saccharomyces paradoxus is anti-recombination. The failure of homologous chromosomes from the different species to recombine causes them to mis-segregate, resulting in aneuploid gametes, most of which are inviable. These effects of anti-recombination have previously impeded the search for other forms of incompatibility, such as negative genetic interactions (Bateson-Dobzhoansky-Muller incompatibilities). By suppressing the meiotic expression of MSH2 and SGS1, we could increase recombination and improve hybrid fertility seventy-fold. This allowed us to recover meiotic tetrads in which all four gametes were viable, ensuring that segregation had occurred properly to produce perfectly haploid, not aneuploid, recombinant hybrid gametes. We sequenced the genomes of 84 such tetrads, and discovered that some combinations of alleles from different species were significantly under-represented, indicating that there are incompatible genes contributing to reproductive isolation.


39
Willis, 2007, and anti-recombination) and incompatibilities between individual 48 genes from the diverging populations (Presgraves, 2010). There is particular 49 interest in the latter class of genic incompatibilities, which are often referred to as 50 "Bateson-Dobzhansky-Muller incompatibilities" (BDMIs) or "speciation genes" (Orr, 51 1996). As we don't know whether these incompatibilities themselves are the cause 52 of speciation or have developed post-speciation, we will refer to them as BDMIs 53 throughout. 54 55 BDMIs represent a case where alleles (at two or more loci) that have evolved 56 to work well together within a species perform poorly when combined in a hybrid 57 individual with alleles from another species, whose alleles have evolved 58 independently (Coyne and Orr, 2004). Since BDMIs offer a universal mechanism for 59 speciation, they have been studied intensely, both theoretically and empirically, yet 60 only a handful have been discovered and characterized at the molecular level 61 In principle, chromosome mis-segregation alone is capable of explaining 92 yeast hybrid sterility without invoking any role for BDMIs. We recently quantified 93 the precise rates at which each chromosome segregates in F1 hybrids (Rogers et al., 94 2018). The average rate of correct distribution for each chromosome in hybrids 95 formed between S. cerevisiae and S. paradoxus is 59.7%, so we expect only 0.03% of 96 gametes to receive exactly one copy of each chromosome (0.597 for each 97 chromosome, raised to the power of 16 to account for all sixteen chromosomes). 98 However, gametes carrying more than one copy of a chromosome can also be viable, 99 as shown by the high rates of aneuploidy detected in viable hybrid gametes. In the 100 40.3% of hybrid meioses in which a chromosome does not segregate properly, half 101 of the resulting spores (20.15%) will receive two copies of the chromosome and 102 might therefore be viable, whilst the remaining 20.15% will receive no copies and 103 will certainly be inviable. Therefore 2.7% of gametes (0.597 plus 0.2015, raised to 104 the power of 16) will receive at least one copy of each essential chromosome, and 105 could be viable, depending on the effect of the additional chromosomes that they 106 carry (Boynton et al., 2018). Thus chromosome mis-segregation due to anti-107 recombination accounts for at least 97.3%, and potentially all, of the observed 108 hybrid sterility. However, there is little direct evidence that extra chromosomes 109 contribute to spore inviability (Rogers et al., 2018), so the smaller figure is more To date, no such BDMIs have been detected in yeast. BDMIs have been 115 detected between mitochondrial genes from one yeast species and nuclear genes 116 from another (Lee et al., 2008;Chou et al., 2010; also see Xu and He, 2011), but these 117 act earlier by reducing F1 mitotic viability and preventing F1 meiosis from even 118 occurring, not by causing inviability of the gametes produced by hybrid meiosis. We 119 have previously shown that most S. paradoxus chromosomes can successfully 120 replace their homologues in S. cerevisiae haploid gametes when substituted one at a 121 time, indicating that they do not contain always-lethal incompatibilities (Greig, 122 2007 The explanation for such under-representation would be that they are incompatible 128 and cause gamete inviability. This method has been modelled by Li et al. (2013), and 129 has been implemented by Kao et al. (2010). Whilst the distribution of genotypes 130 differed significantly from what was expected by chance, the additional aneuploid 131 chromosomes carried by the genotyped gametes confounded analysis to an extent 132 that the effective sample size was too low to identify individual pairs of 133 incompatible loci (Kao et al., 2010). 134

135
In order to identify BDMIs involved in hybrid spore inviability, it is therefore 136 necessary to overcome the primary effect of anti-recombination, in order to produce 137 haploid spores without additional aneuploid chromosomes for genotyping. Hunter 138 et al. (1996) previously showed that knocking out genes involved in monitoring the 139 fidelity of recombination increases both the rate of recombination and the 140 proportion of viable gametes produced by hybrid meioses. By deleting the mismatch 141 repair gene MSH2, they increased crossing over in hybrids on average 13-fold, 142 resulting in a nearly 9-fold increase in hybrid spore viability. Kao et al. (2010) 143 therefore used msh2∆ knock-out mutants in their search for BDMIs, but the 144 improvement in chromosome segregation was insufficient to relieve the extensive 145 aneuploidy of the hybrid gametes. Here we employed two additional tools in order 146 to produce perfectly euploid hybrid gametes for genotyping. First, we repressed the 147 expression of both MSH2 and a second anti-recombination factor, DNA-helicase 148 SGS1, specifically in meiosis, thereby retaining their normal function during mitosis, 149 which reduces the mutagenic effects of knocking them out entirely. Secondly, we 150 dissected hybrid gametes out of their meiotic tetrads and genotyped only those that 151 came from tetrads in which all four spores were viable. Our sample therefore 152 excluded not only those gametes containing lethal combinations of the parent 153 species' alleles, but also aneuploid gametes, since any chromosome mis-segregation 154 will kill some of the gametes in a tetrad. 155

156
We sequenced all 336 haploid gametes from 84 F1 hybrid meioses and tested 157 statistically for pairs of alleles for which parental combinations were over-158 represented. We were able to map four broad pairs of genomic regions that show 159 evidence of incompatibility. Thus, for the first time, we find evidence of naturally-160 occurring nuclear BDMIs causing sterility of hybrids between two species of yeast. 161 162 163

165
Restoration of hybrid fertility 166 167 We constructed strains of S. cerevisiae and S. paradoxus in which the native promoters of 168 MSH2 and SGS1 were replaced with the CLB2 promoter, which is specifically repressed during  avoid any unwanted effects such as an increased recessive-lethal mutation rate, which would 173 actually reduce fertility (Hunter et al., 1996). In a previous study, we found that suppressing 174 meiotic expression of SGS1 alone improved the rate of correct segregation by almost half in 175 hybrid meioses (Rogers et al., 2018). Here, we find that spore viability is also dramatically 176 improved. Suppression of SGS1 alone increased hybrid spore viability from 0.46% to 20.8%; 177 and in combination with suppression of MSH2, spore viability was further improved to 32.6% 178 ( Figure 1, Source Data 1). Significantly more of the double mutant spores were viable than in 179 the wild-type hybrid (chi-squared contingency test: X 2 = 479.91, df = 1, p-value < 2.2x10 -16 ). 180 The restoration of hybrid fertility vastly increased the production of hybrid tetrads in which 181 all four spores were viable, which were specifically selected for genotyping and further 182 analysis. All spores from such tetrads are necessarily euploid, as mis-segregation of even a 183 single chromosome would result in at least one dead spore (lacking that chromosome). By 184 Figure 1: Restoration of hybrid fertility by meiotic repression of MSH2 and SGS1. Percentages are spore viabilities of the indicated hybrid strains. In the PCLB2-MSH2 PCLB2-SGS1 strain, a significant 32.14% increase in spore viability was observed (double mutant when compared with the wild type: X 2 = 479.91, df = 1, p-value < 2.2x10 -16 ). Numbers in parentheses indicate the total number of dissected spores checked for viability. Full data, including other strains, can be found in Source Data 1.
analyzing only euploid spores, we ensured that recessive BDMIs were not masked by 185 aneuploidy. 186

Evidence for hybrid incompatibility 187 188
A fertility-reducing BDMI between a pair of loci would result in fewer gametes 189 containing hybrid combinations of alleles at these loci. Reasoning that we could not 190 map such loci at a resolution higher than the linkage groups produced by the 191 crossovers that occurred within the 84 tetrads in our sample, we divided the In addition, we calculated the 99% confidence interval (CI) for the odds ratios. An 203 odds ratio of 1 indicated that the parental and hybrid types were present in equal 204 frequencies. An odds ratio greater than 1 would be observed for a bias towards 205 parental types; and an odds ratio less than 1 would indicate that hybrid types were 206 preferentially observed. We found all pairwise comparisons for which the calculated 207 99% CI did not encompass the value of 1 (lower bound of CI > 1 or upper bound of 208 CI < 1). 1.9% of all comparisons (13,082/676,294) had CIs that indicated a hybrid 209 preference and 2.6% (17,492/676,294) had CIs that indicated a parental preference. 210 Parental types were not only over-represented more often, but were also more 211 highly favoured. Parental types were over-represented by 3/4 in 190 cases 212 (parent/hybrid ratio ≧ 1.75) while hybrid types were over-represented by 3/4 in 213 only 22 cases (hybrid/parent ratio ≧ 1.75). 214 Individual significant interactions were determined as described in Li et al. 215 (2013) and in the Methods. Briefly, a null distribution of top ORs was produced by 216 randomly re-sampling the observed data 100 times (see Source Code 1). The 5th 217 largest OR from this set of the top 100 ORs was used as the critical value from which 218 we judged significance. All observed pairs with a higher OR than the critical value 219 Anti-recombination as a barrier between species 238 239 By reducing the expression of just two genes, SGS1 and MSH2, during meiosis we 240 were able to rescue the fertility of a sterile hybrid between S. cerevisiae and S. 241 paradoxus, increasing its ability to produce viable gametes 70-fold, from 0.46% to 242 32.6%. The fertility of our rescued inter-species hybrid was around one third that of 243 its non-hybrid parents, which is about the same as intra-species crosses formed 244 between diverged populations of a single species, S. paradoxus (e.g., Greig  where a simple BDMI is one that kills a certain hybrid genotype. They found several 336 pairs of segments with distributions that were statistically significantly different 337 than what would be expected by chance, but they attributed these to more complex 338 interactions, likely involving multiple loci with weak effects. They found some 339 evidence of three-way interactions, but lacked confidence due to limited statistical obtaining complete tetrads, we avoided this problem. Moreover, we also reduced 359 potential genotyping errors because each recombination event is supported in two 360 separate, reciprocal samples. As well as these improvements, by using OR instead of 361 ChiSq, as recommended by Li et al. (2013), we could focus solely on the case in 362 which there is a depletion of hybrid types (OR higher than expected). 363 364 Using these improved methods, we found six major regions of the genome 365 that appear to define four putative two-locus BDMIs (Figure 2). These regions were 366 found on only four chromosomes (chr IX, X, XII and XV). Many genes map to these 367 regions, and fine-scale mapping will be necessary to determine the causative loci. 368 Among the known interacting genes in BioGRID, there are none identified between 369 genes found in regions A and E or B and D (Oughtred et al., 2018). Among the genes 370 in regions C and E, there is one known interaction; a negative genetic interaction 371 between IMA2 (an isomaltase) and CDC6 (an essential protein required for DNA 372 replication), which was found in a large-scale genetic interaction study (Costanzo et 373 al., 2016). Regions D and F harbour many known interacting genes, but this is 374 unsurprising because together they encompass the largest number of genes. Despite

417
Strains 418 419 We used as a template a previously constructed S. cerevisiae strain NHY 2039, in 420 which the promotor of SGS1 had been replaced by the CLB2 promotor (Oh et al.,421 2008) using the pFA6a-KANMX6-pCLB2-3HA created by Lee and Amon (2003). We 422 amplified the CLB2 promotor and the KANMX4 drug resistance marker out of 423 NHY2039 (i.e. YDG832) using primer pairs (see Supplementary File 5) that allowed 424 us to transform it in place of the natural promoters of MSH2 and SGS1 in both S. 425 cerevisiae (W303 background) and S. paradoxus (N17 background). The resulting S. 426 cerevisiae and S. paradoxus haploid strains YDG968 and YDG969 (see Table 1 contained reads mapping to a given ORF of one or the other or both species, the two 503 spores with the highest proportion of reads mapping to one species ORF would have 504 it assigned to that species and the other two would have the ORF assigned to the 505 other species. If the four copies of an ORF within a tetrad did not all contain reads 506 mapping to either or both species then the ORF would be assigned to the same 507 species as the neighbouring ORF. These genotyping rules produced a recombination 508 map (see Supplementary File 6 for an example, full data available in Bozdag et  Pairs of segments on the same chromosome were excluded because physical linkage 527 would skew the numbers towards parental combinations, the same effect that we 528 expect to see due to incompatibility between loci, thus making the results difficult to 529 interpret. 530

531
To control for type I error, we produced a null distribution against which to 532 test our observed ORs (Source Code 1). First, a new set of 84 tetrads was simulated 533 by pulling each chromosome for each tetrad from the pool of chromosomes of that 534 type (with replacement). This is similar to, but not exactly the same as, the method 535 used by Li et al. (2013). They, in contrast, shuffle the chromosomes without 536 replacement instead of sampling with replacement, and they do so among random 537 spores instead of whole tetrads. ORs for the segment pairs were then calculated as 538 before but on the simulated set of chromosomes. This process was repeated 99 539 more times, and the top OR from each set of simulated tetrads was recorded. The 540 5th largest OR from this set of 100 top ORs was chosen as the critical value from 541 which to judge significance. All observed pairs with a higher OR than this critical interaction with the same segment in another chromosome, they were part of the 551 same block. In one case, one significant interaction was between a non-adjacent 552 segment and a segment that interacted with many other nearby segments. Because 553 this interaction was so close to the others (within nine segments), we arbitrarily 554 decided to treat it as part of the same interaction (Source Data 2, row 37 as 555 compared to surrounding rows). Similarly, region B was found to interact with a 556 segment adjacent to those interacting with region F. In this case, we collapsed both 557 interacting regions into one (region D), as we consider it most likely that the two 558 regions are interacting with a single gene in the region. 559 560 561

563
Numerous people have contributed to this long-running project over the years, and we 564 apologise if we have neglected to name you. We are particularly grateful to colleagues at the 565