Selection on heterozygosity drives adaptation in intra- and interspecific hybrids

Hybridization is often considered maladaptive, but sometimes hybrids can invade new ecological niches and adapt to novel or stressful environments better than their parents. However, the genomic changes that occur following hybridization and facilitate genome resolution and/or adaptation are not well understood. Here, we address these questions using experimental evolution of de novo interspecific hybrid yeast Saccharomyces cerevisiae x Saccharomyces uvarum and their parentals. We evolved these strains in nutrient limited conditions for hundreds of generations and sequenced the resulting cultures to identify genomic changes. Analysis of 16 hybrid clones and 16 parental clones identified numerous point mutations, copy number changes, and loss of heterozygosity events, including a number of nuclear-mitochondrial mutations and species biased amplification of nutrient transporters. We focused on a particularly interesting example, in which we saw repeated loss of heterozygosity at the high affinity phosphate transporter gene PHO84 in both intra- and interspecific hybrids. Using allele replacement methods, we tested the fitness of different alleles in hybrid and S. cerevisiae strain backgrounds and found that the loss of heterozygosity is indeed the result of selection on one allele over the other in both S. cerevisiae and the hybrids. This illuminates an example where hybrid genome resolution is driven by positive selection on existing heterozygosity, and generally demonstrates that outcrossing need not be frequent to have long lasting impacts on adaptation.


INTRODUCTION
Hybridization is now recognized as a common phenomenon across the tree of life.
Historically however, the detection of hybrids has been difficult, and its incidence may be underreported for both plants and animals, and almost certainly for certain eukaryotes like insects and fungi (Albertin and Marullo 2012;Bullini 1994). Its importance as an evolutionary force has thus been maligned, as hybrids appeared both rare, and typically at a reduced fitness.
In addition to potential post-reproductive barriers, the hybrid is theorized to be ill-adapted to its environment and will also suffer minority cytotype disadvantage because other hybrids are uncommon and backcrosses to parental species may be unfit (Mallet 2007). However, hybrids can have a variety of advantages over their parents, including heterozygote advantage, extreme phenotypic traits, and reproductive isolation (usually resulting from polyploidy), and can thus facilitate adaptation to novel or stressful conditions, invade unoccupied ecological niches, and even increase biodiversity.
Some hybridization events lead to new hybrid species (Mavarez, et al. 2006;Meyer, et al. 2006;Nolte, et al. 2005;Rieseberg 1997;Schumer, et al. 2014;Soltis and Soltis 2009), while most result in introgression from hybrid backcrosses to the more abundant parental species (Dasmahapatra, et al. 2012;Dowling, et al. 1989;Grant, et al. 2005;Taylor and Hebert 1993;Wayne 1993). This introduces genetic variation into a population at orders of magnitude greater than what mutation alone can achieve, in a sense operating as a multi-locus macro-mutation (Abbott, et al. 2013;Barton 2001;Grant and Grant 1994;Mallet 2007). Therefore, hybridization via introgression, polyploidy, or homopoloid hybrid speciation, may offer a rapid strategy for adaptation to changing environmental conditions. For example, in Darwin's finches, adaptive introgression supplied the morphological variation which allowed the species to survive following an El Niño event Grant 2010, 2002), while in ancient humans, introgression allowed adaptation to high altitudes (Huerta- Sanchez and Casey 2015), among other traits (Racimo, et al. 2015). The most iconic example comes from the hybrid sunflower species Helianthus anomalus, Helianthus deserticola, and Helianthus paradoxus, from the parents Helianthus annuus and Helianthus petiolaris. These three hybrid species are locally adapted to extreme desert, salt marsh, and dune habitats respectively, and show traits such as increased drought or salt tolerance relative to their parents (Heiser 1954;Rieseberg 1991;Rosenthal, et al. 2002;Schwarzbach, et al. 2001).
Agriculture and industry use both intra-and interspecific hybrids as a tool to increase yield or robustness, introduce resistance to pests, and create novel phenotype or flavor profiles.
For example, plant breeders have crossed domesticated species to wild species to introduce resistance to a variety of pathogens in wheat, potato, and canola (Mason and Batley 2015), and almost all maize grown in the United States is grown from intraspecific hybrid seeds, which has increased yield and provided improved resistance to biotic and abiotic factors (Crow 1998).
Vintners and brewers have created interspecific hybrids to select for traits such as lower acetic acid concentration (Bellon, et al. 2015), and many incidental fungal hybrids have been discovered in brewing and industry, including Pichia sorbitophila (Louis, et al. 2012b), and various hybrids across the Saccharomyces clade (Bellon, et al. 2015;Gonzalez, et al. 2006;Gonzalez, et al. 2008;Hittinger 2013;Muller and McCusker 2009), most notably the lagerbrewing yeast, Saccharomyces pastorianus (Baker, et al. 2015;Dunn and Sherlock 2008;Gibson and Liti 2015;Peris, et al. 2016;Tamai, et al. 1998;Walther, et al. 2014). It is presumed that the severe selection pressures exerted during industrial processes have selected for interspecific hybrid genomes that may be more able to cope with the extreme environments.
At the genomic level, hybridization induces chromosome loss/aneuploidy, chromosomal rearrangements, gene loss, changes in gene expression, changes in epigenetic modifications, transposable element mobilization, and large scale loss of heterozygosity, in which the allele of one species is lost and the allele of the other species is retained via gene conversion or break induced replication (Abbott, et al. 2013;Ainouche and Jenczewski 2010;Albertin and Marullo 2012;Borneman, et al. 2014;Doyle, et al. 2008;Landry, et al. 2007;Masly, et al. 2006;Michalak 2009;Soltis, et al. 2014;Soltis 2013). These extensive changes can result in a chimeric, stabilized hybrid, although the period of time for genome stabilization to occur can range dramatically (Soltis, et al. 2014). There is some speculation that genetic distance between parental hybridizing species influences genome stabilization and bias in genome resolution, but this remains an open question. It is also unknown whether there are structural and functional biases in the ways in which genes/alleles are lost or modified. Both drift and selection influence the resolution of the hybrid genome, but their contributions are difficult to untangle.
To further explore these types of questions, an experimental approach is helpful.
Researchers have long been exploring the genetics of hybrid traits in the lab, particularly in agricultural crops, although this is often slowed by infertility and reduced viability in many interspecific hybrids (Hajjar and Hodgkin 2007;Ouyang, et al. 2010;Perez-Prat and van Lookeren Campagne 2002). The genus of the budding yeast Saccharomyces cerevisiae lends itself particularly well to experimental study. Many hybrids of this genus have been discovered in brewing, industrial, and natural environments; indeed, the genus itself is speculated to be a product of an ancient hybridization event (Barbosa, et al. 2016;Hittinger 2013;Marcet-Houben and Gabaldon 2015). Viable interspecific hybrids can be created de novo in the lab (Greig, et al. 2002;Marinoni, et al. 1999), and their ability to grow mitotically means that the catastrophic postzygotic barriers to speciation that generally doom other obligate sexually reproducing hybrids can be avoided. This experimental system allows us to observe evolution in real time in the laboratory environment, and the genetic and genomic tools available in this model genus facilitate characterization of the connection between genotype and phenotype, including fitness.
Previous work in our lab group has utilized experimental evolution to investigate adaptive events in homozygous haploid and diploid S. cerevisiae (Gresham, et al. 2008;Payen, et al. 2014;Sunshine, et al. 2015). To investigate genome evolution post hybridization, we utilize an interspecific hybrid, S. cerevisiae x Saccharomyces uvarum, and its parentals: a homozygous diploid S. uvarum and an intraspecific hybrid S. cerevisiae GRF167 x S. cerevisiae S288C. This allows us to understand the impact of varying levels of heterozygosity on adaptation and genome evolution, ranging from none (S. uvarum and previous S. cerevisiae experiments), to intraspecific heterozygosity (S. cerevisiae GRF167 x S. cerevisiae S288C), to the most extreme case of interspecific hybrids. S. uvarum is one of the most distantly related species of S. cerevisiae in the Saccharomyces clade, separated by 20 my and 20% sequence divergence at coding sites (Cliften, et al. 2006;Kellis, et al. 2003). Despite this extensive divergence, S. cerevisiae and S. uvarum are largely syntenic and create hybrids, though less than 1% of zygotes are viable (Greig 2009). The two species differ in their stress tolerances, for example, S. cerevisiae being more thermotolerant, S. uvarum being cryotolerant (Almeida, et al. 2014). Previous evolution experiments using lab derived hybrids has revealed novel and/or transgressive phenotypes for ammonium limitation, ethanol tolerance, and growth on xylose (Belloch, et al. 2008;Dunn, et al. 2013;Piotrowski, et al. 2012;Wenger, et al. 2010). Notably, Dunn et al. (2013) reveal several loss of heterozygosity events and a repeatable reciprocal translocation that produces a gene fusion at the high-affinity ammonium permease MEP2 after selection in ammonium limitation, offering insight into potential mutational events in the adaptation and/or stabilization of S. cerevisiae x S. uvarum hybrids.
Here, we evolved these hybrids and diploids in replicate in three nutrient limited conditions for hundreds of generations. Using whole genome sequencing, we found whole chromosome aneuploidy, genome rearrangements, copy number variants, de novo point mutations, and loss of heterozygosity. We sought to determine how initial heterozygosity impacts adaptation to novel conditions, and explore whether neutral or selective forces are influencing the resolution of the hybrid genome over time. In particular, we investigated a reoccurring loss of heterozygosity event observed in both intra-and interspecific hybrids, and found support for the hypothesis that loss of heterozygosity at this locus is due to selection.

Experimental evolution of hybrid and parental species
An interspecific hybrid was created by crossing S. cerevisiae and S. uvarum (strains in Supplemental Table 1), and evolved in continuous culture in the chemostat (Monod 1949;Novick and Szilard 1950a, b). In parallel, homozygous diploid S. uvarum and heterozygous diploid S. cerevisiae (GRF167xS288C) were also evolved. Each strain was grown in two or more replicate independent cultures under three different nutrient limitations-glucose, phosphate, and sulfate-for 85-557 generations (median 158) at 30°C, except for S. uvarum, which was unable to achieve steady state in all conditions at 30°C and so was evolved at 25°C. The population sizes were largely similar across strains, species, and conditions. Each evolved clone was subsequently competed individually against the appropriate GFP-tagged ancestor to gauge relative fitness. As expected, evolved hybrid and parental clones generally exhibit higher fitness than their unevolved ancestor, with typical relative fitness gains between 20-30% (Tables 1, 2).

Mutations in nuclear encoded mitochondrial genes may be more prevalent in interspecific hybrids
To identify mutations in the evolved hybrids, we generated whole genome sequencing for sixteen clones from the endpoints of the evolution experiments ( Table 1). We thus captured data from a range of nutrient limitations (6-phosphate; 3-glucose; 7-sulfate) and generations (100-285, median 154 generations). Each clone had an average of 2.4 point mutations, a number of which have been previously identified in prior S. cerevisiae evolution experiments. For example, a nonsynonymous mutation in the S. cerevisiae allele of the glucose sensing gene SNF3 has been identified in glucose limited experiments in S. cerevisiae (Kvitek and Sherlock 2013;Selmecki, et al. 2015) . To our knowledge, 20/27 coding point mutations are unique to these experiments .
In evolved parentals, we again sequenced one clone from the endpoint of each population. In total, we sequenced 16 clones, 6 from each of the three nutrients (two S. uvarum diploids, and four S. cerevisiae diploids), except in glucose limitation in which only two S.
cerevisiae populations were sampled. The generations ranged from 234-557 (median 477) in S. uvarum with an average of 2.83 mutations per clone, and from 127-190 (median 166.5) in S. cerevisiae with an average of 0.9 point mutations per clone ( Table 2). This discrepancy in point mutations between S. cerevisiae and S. uvarum may be explained by differences in generation time, or perhaps other mutational events are more prevalent in S. cerevisiae.
With the limited number of samples we have from hybrid and parental clones, it is difficult to draw any conclusions regarding unique point mutations in hybrids. However, one class of mutations that may be of particular interest in hybrids are genomic mutations which may interact with the mitochondria, as previous work has shown that nuclear-mitochondria interactions can underlie hybrid incompatibility (Chou and Leu 2010;Lee, et al. 2008;Meiklejohn, et al. 2013). Other studies have found that only the S. cerevisiae mitochondria are retained in S. cerevisiae x S. uvarum hybrids (Antunovics, et al. 2005), and we recapitulate these findings, potentially setting the stage for conflicting interactions between the S. uvarum nuclear genome and the foreign mitochondria. We observe several mitochondria related mutations in hybrids in both S. cerevisiae and S. uvarum alleles. For example, one point mutation, a nonsynonymous mutation in the S. cerevisiae allele of the mitochondrial ribosomal protein gene MHR1, was seen in two separate clones independently evolved in phosphate limitation. This gene may be of particular interest as it was discovered in a previous screen as being haploproficient (increased fitness of 19%) in hybrids in which the S. cerevisiae allele is missing and the S. uvarum allele is retained (Lancaster S, Dunham MJ, unpublished data), suggesting that this mutation may alter or disable the S. cerevisiae protein in some way. Another example involves the gene IRC3, a helicase responsible for maintenance of the mitochondrial genome, which has a nonsynonymous mutation in the S. uvarum allele in clone Gh3 and is deleted in clone Gh2, potentially suggesting that the uvarum allele is deleterious in the hybrid background.
While our sample size is small, 4/27 point mutations in hybrids are related to mitochondria function compared to 0/26 in parentals, and may represent interesting targets for further exploration.

Copy number variants frequently involve the amplification of nutrient transporters
Yeast in both natural and artificial environments are known to frequently experience changes in copy number, ranging from single genes to whole chromosomes (Dunham, et al. 8 2002;Gresham, et al. 2008;Kvitek and Sherlock 2013;Payen, et al. 2014;Selmecki, et al. 2015;Sunshine, et al. 2015;Zhu, et al. 2016). This holds true in our evolution experiments: we observe copy number changes across all genetic backgrounds ( It therefore does not appear that our interspecific hybrids are more prone to genomic instability, as has previously been suggested in other systems (Chester, et al. 2015;Mason and Batley 2015;Xiong, et al. 2011). The most common event across nutrient limitations in the interspecific hybrids was an amplification of the S. cerevisiae copy of chromosome IV, which occurred in four independent hybrid clones (3 in phosphate limitation, 1 in glucose limitation; Supplemental Figure 3). Several other characteristic rearrangements occurred in the evolved S. cerevisiae clones, including the amplification of the left arm of chromosome 14 accompanied by segmental monosomy of the right arm of chromosome 14, an event seen previously in other evolved populations (Dunham, et al. 2002;Gresham, et al. 2008;Sunshine, et al. 2015). All copy number events in S. cerevisiae had breakpoints at repetitive elements known as Ty elements, except those located on chrII, which are known to be mediated by another mechanism (Brewer, et al. 2015). In contrast, copy number variants in the hybrid were rarely facilitated by repetitive elements, perhaps in part because S. uvarum has no full length Ty elements.
Frequently in nutrient limited evolution experiments, copy number variants involve amplification of the nutrient specific transporter, and indeed, we also observed amplification of these transporters in many of the clones. In sulfate limitation, the S. cerevisiae allele of the high affinity sulfate transporter SUL1 is amplified in 7/7 hybrid clones and 4/4 S. cerevisiae clones ( Figure 1, Tables 1-2, Supplemental Figures 1, 3). Interestingly, SUL2 is the preferred sulfate transporter in S. uvarum (Sanchez, et al. 2016) and was not observed to be amplified in the evolved hybrids (Supplemental Figure 2, Table 2). In glucose limitation, previous S. cerevisiae evolution experiments found consistent amplification of the high affinity glucose transporter genes HXT6/7 (Brown, et al. 1998;Dunham, et al. 2002;Gresham, et al. 2008;Kao and Sherlock 2008;Kvitek and Sherlock 2011). In our experiments, the S. uvarum alleles of the HXT6/7 transporters are amplified in 3/3 hybrid clones and both S. uvarum clones, but are not amplified in evolved S. cerevisiae clones, suggesting that the S. uvarum HXT6/7 alleles confer a greater fitness advantage compared to S. cerevisiae (Figure 1, Tables 1-2, Supplemental Figures 1-3).
Finally, in phosphate limitation, the S. cerevisiae copy of the high affinity phosphate transporter PHO84 is amplified, while the S. uvarum allele is lost in 3/6 hybrid clones in an event known as loss of heterozygosity (Figures 1-2 Figure 3). Intriguingly, the evolved S. cerevisiae also display loss of heterozygosity and accompanied amplification favoring the allele derived from strain GRF167 over the S288C allele in 4/4 clones (Figure 2, Table 2). All hybrid clones carry the "preferred" GRF167 S. cerevisiae allele, as this was the allele used to create the de novo hybrid.

Loss of heterozygosity is a common event in heterozygous evolving populations
Selection on heterozygosity, as a loss of heterozygosity event could represent, is an underappreciated source of adaptation in microbial experimental evolution, as typical experiments evolve a haploid or homozygous diploid strain asexually and as a result, have little variation to select upon. Loss of heterozygosity is observed in natural and industrial hybrids (Albertin and Marullo 2012;Louis, et al. 2012a;Pryszcz, et al. 2014a;Wolfe 2015), but here we document its occurrence in both intra-and interspecific hybrids in the laboratory as a result of short term evolution (also see (Burke, et al. 2014;Dunn, et al. 2013)). Loss of heterozygosity is observed across all nutrient conditions, with twelve independent loss of heterozygosity events detected in S. cerevisiae, and nine independent events documented in the hybrids (Figures 1-2,   Tables 1-2, Supplemental Figures 1, 3). It thus appears that this type of mutational event is both common, and can occur over short evolutionary timescales.
The loss of heterozygosity event can result in copy-neutral (where one allele is lost and the other allele is amplified) or non-neutral chromosomal segments (where one allele is lost, rendering the strain hemizygous at that locus), and can favor the retention of either allele. In S. cerevisiae, there is a bias in resolution where loss of heterozygosity events favor retaining the GRF167 allele over the S288C allele (Table 2, Supplemental Figure 4). One unique case in clone Sc4 has a small ~5 kb loss of heterozygosity event on chrXV favoring GRF167, which switches to favoring S288C for the rest of the chromosome. The retention of S. cerevisiae is slightly more common in the hybrids (5/9 events, Table 1 (Nakao, et al. 2009). The size of the event ranges from approximately 25 kb to the whole chromosome level in the hybrids, and from 5kb to 540kb in S. cerevisiae. Where loss of heterozygosity is accompanied by an amplification event, the loss of heterozygosity event always occurs first; unlike many CNV events, almost all loss of heterozygosity events do not appear to be mediated by existing repetitive sequence such as a transposable element in the hybrid or S. cerevisiae, and are most likely a product of break induced replication or mitotic gene conversion (Hoang, et al. 2010). The exceptions are in hybrid clones Ph4, Ph5, and Sh1, where there is a non-copy neutral loss of S. cerevisiae mediated by a Ty element or Long Terminal Repeat (LTR), and S. cerevisiae clones Sc1 and Sc4, where there is a 6.5 kb deletion of the S288C allele flanked by two Ty elements.
Loss of heterozygosity events in hybrids could signify several ongoing processes in hybrid genome evolution: loss of heterozygosity regions may represent (1) loci with incompatibilities; (2) selection on existing variation; or (3) genetic drift eroding genomic segments. While our sample size is modest, failing to see repeated loss of heterozygosity events across nutrient conditions disfavors the hypothesis that loss of heterozygosity is resolving some sort of hybrid incompatibility. Furthermore, loss of heterozygosity events observed in evolved S. cerevisiae suggest that this mutation type is not unique to interspecific hybrids. Instead, repeated events within a particular condition, such as the repeated loss of heterozygosity at PHO84 in phosphate limitation or the 6.5kb segment on chrXIII in sulfate limitation, suggest that these events are beneficial, and are indeed selection on one allele over the other.

Loss of heterozygosity is driven by selection on one allele
To test the hypothesis that loss of heterozygosity events provide a selective advantage, we used allele replacement, in which the allele of one species/strain is replaced with the allele of the other species/strain in an otherwise isogenic background. We tested this hypothesis using the most commonly seen loss of heterozygosity event, loss of heterozygosity at PHO84. While the region extends from 25-234kb in length in the hybrids and 40-85 kb in S. cerevisiae, PHO84 was a prime candidate driving this event. PHO84 is one of only 10 genes encompassed in the region extending from the telomere to the breakpoint of the shortest loss of heterozygosity event, and is included in every other loss of heterozygosity event on chromosome XIII ( Figure   1). It is responsible for sensing low phosphate, and previous work identified a point mutation in PHO84 (an Alanine to Valine substitution at the 5' end of the gene), which increased fitness by 23% in phosphate limited conditions (Sunshine, et al. 2015). Finally, prior work with other nutrient transporters has shown amplification of nutrient transporters to be a key event in adapting to nutrient limited conditions. We thus selected a region of approximately 2.5kb encompassing the PHO84 ORF, its promoter, and 3'UTR (Cherry, et al. 2012;Nagalakshmi, et al. 2008;Yassour, et al. 2009). We created allele replacement strains using the two alleles of S. cerevisiae in a S. cerevisiae diploid background; the two alleles are 99.1% identical in this region and each strain is identical to the ancestral strain used in our evolution experiments except at the PHO84 locus. The S. cerevisiae ancestor carries one copy of GRF167 ("preferred") and one copy of S288C ("un-preferred"), so named due to which allele was retained and amplified in the evolved clones. To measure any resultant changes in fitness, we competed each strain individually against a fluorescent ancestral strain and measured which strain overtook the culture. Two copies of the un-preferred allele decreased fitness by -5.31 (+/-1.86), while two copies of the preferred allele increased fitness by 9.93 (+/-0.27). This displays an overall difference in fitness of 15.24 between the unpreferred and preferred alleles. By comparing the fitness of these allele replacement strains to the evolved clones (Table 2), the allele replacement does not fully recapitulate the fitness gain observed in the evolved clone. One explanation is that the additional mutations present in the evolved strains also contribute to their total fitness. Another explanation could be the increased copy number of the PHO84 region that we see in these evolved clones. To further explore this fitness difference, we cloned the GRF167 allele onto a low copy number plasmid and transformed the allele replacement strain carrying two preferred S. cerevisiae alleles to simulate increased copy number of PHO84, and saw a further fitness increase of 1.76. This supports the conclusion that relative fitness gains in the evolved clone are largely due to the loss of the S288C allele, and selection and amplification of the GRF167 allele, with little additional benefit from further amplification. It could also be the case that co-amplification of other genes in the segment is required to attain the full benefit, as previously observed by the contribution of BSD2 to the SUL1 amplicon (Sunshine, et al. 2015;. Unfortunately, we were unable to obtain a successful strain carrying the preferred S. cerevisiae allele in an S. uvarum background, so we were unable to test the fitness effect of carrying two preferred S. cerevisiae alleles in the hybrid background (which typically carries one preferred S. cerevisiae allele and one S. uvarum allele). However, we were able to generate a S. uvarum strain carrying the unpreferred allele and use this to create a hybrid. Carrying one preferred allele and one un-preferred S. cerevisiae allele has an increased fitness of 4.35 compared to two S. uvarum alleles in a hybrid background. Furthermore, using the GRF167 PHO84 plasmid, we found that the hybrid has an increased fitness of 16.47 (+/-1.13) when an extra preferred allele is added. Together, these results support the conclusion that the S. cerevisiae GRF167 allele is preferred over the S288C one, and that S. cerevisiae alleles are preferred over the S. uvarum allele in the hybrid, and hence, that the loss of heterozygosity events seen in both intra-and interspecific hybrids are the product of selection.

DISCUSSION
In summary, we sought to understand forces underlying genome stabilization and evolution in interspecific and intraspecific hybrids as they adapt to novel environments. We evolved and sequenced clones from 16 hybrid populations and 16 parental populations to reveal a variety of mutational events conferring adaptation to three nutrient limited conditions.
Of particular note, we find loss of heterozygosity in both evolved intraspecific and interspecific hybrid clones in all nutrient environments, potentially signifying areas where selection has acted on preexisting variation present in the ancestral clone. We used an allele replacement strategy to test this hypothesis for a commonly repeated loss of heterozygosity event and show that selection is indeed driving the homogenization of the genome at this locus. Though other studies in natural, industrial, and lab-evolved isolates have observed loss of heterozygosity, we present the first empirical test of the causal evolutionary forces influencing these events. This work is particularly informative for understanding past hybridization events and subsequent genome resolution in hybrids in natural and artificial systems.

The predictability of evolution
We now have many examples of predictable evolution in natural systems (Conte, et al. 2012;Elmer and Meyer 2011;Losos, et al. 1998;Martin and Orgogozo 2013;Rundle, et al. 2000;Wessinger and Rausher 2014), and in laboratory experimental evolution, in which there often appears to be a limited number of high fitness pathways that strains follow when adapting to a particular condition (Burke, et al. 2010;Ferea, et al. 1999;Gresham, et al. 2008;Kawecki, et al. 2012;Kvitek and Sherlock 2013;Lang and Desai 2014;Salverda, et al. 2011;Woods, et al. 2006). For example, it is well established that amplifications of nutrient transporters are drivers of adaptation in evolution in nutrient limited conditions. Previous work in our group has particularly focused on the amplification of the high affinity sulfate transporter gene SUL1 in sulfate limited conditions, which occurs in almost every sulfate limited evolution experiment and confers a fitness advantage of as much as 40% compared to its unevolved ancestor. The amplification of phosphate transporters has been markedly less common, and thus drivers of adaptation in this condition have been less clear. Gresham et al. (2008) identified a whole chromosome amplification of chrXIII in one population. In a follow up study, Sunshine et al. (2015), found whole or partial amplification of chrXIII in 3/8 populations.
A genome wide screen for segmental amplifications found a slight increase in fitness for a small telomeric segment of chromosome XIII, and a point mutation in PHO84 was observed to increase fitness by 23%. However, screens by  showed that although PHO84 is In our work, we observed amplification of the S. cerevisiae GRF167 allele of PHO84 in 4/4 S. cerevisiae clones from 4 populations and 3/6 hybrid clones from 6 populations. This amplification was always preceded by the loss of the S288C allele in S. cerevisiae clones or occurred in conjunction with the loss of the S. uvarum allele in hybrids. Furthermore, there is a 15% fitness difference between carrying two copies of the S288C allele of PHO84 compared to carrying two copies of the GRF167 allele of PHO84. It thus appears that amplification of PHO84 has been less predictable as the S288C allele does not confer a fitness advantage unless mutated.
We note that the preferred GRF167 allele of PHO84 does not carry this particular polymorphism. Together, these results imply that strain background can constrain adaptive pathways.
The infusion of variation created by hybridization provides new templates for selection to act upon, which can be more important than either point mutations or copy number variants alone. Our work shows that outcrossing need not be common to have long-lasting effects on adaptation. This implication is particularly relevant in yeast where outcrossing may occur quite rarely followed by thousands of asexual generations (Greig and Leu 2009;Liti 2015;Ruderfer, et al. 2006).

Applications to other hybrids and cancer
The observation that loss of heterozygosity occurs in hybrid genomes is increasingly documented (Borneman, et al. 2014;Louis, et al. 2012b;Marcet-Houben and Gabaldon 2015;Pryszcz, et al. 2014b;Soltis, et al. 2014), although the reason(s) for this type of mutation has been unresolved. As most examples stem from allopolyploid events that occurred millions of years ago, understanding why loss of heterozygosity is important in hybrid genome evolution is difficult. Cancer cells are also known to experience loss of heterozygosity, sometimes involved in the inactivation of tumor suppressor genes, leaving only one copy of the gene that may be mutated or silenced (Lapunzina and Monk 2011;Thiagalingam, et al. 2001;Tuna, et al. 2009).
Data support the conclusion that loss of heterozygosity events are selected for during tumor development, as many loss of heterozygosity events involve specific chromosomal segments (Thiagalingam, et al. 2001), although the underlying molecular and genetic reasons for selection is an open debate (Ryland, et al. 2015).
Here, we experimentally demonstrate that loss of heterozygosity can occur in homoploid hybrids, as well as intraspecific hybrids. We provide an example in which homogenization of the genome is non-random, but instead driven by selection on one allele. We furthermore discover examples where one species allele is preferred over the other without loss of heterozygosity, such as the repeated amplification of the S. uvarum high affinity glucose transporters HXT6/7. Amplification of one species allele with or without loss of heterozygosity may be due to hybrid incompatibility within a particular protein complex, or other epistatic interactions (Piatkowska, et al. 2013). Together, our results show that the heterozygosity supplied by hybridization is an important contributor to adaptive routes explored by populations as they adapt to novel conditions. While we cannot generalize our results from the PHO84 locus across the many other loss of heterozygosity events discovered in our hybrids and S. cerevisiae, in the future we can use similar methodology to explore whether positive selection always drives loss of heterozygosity or whether other explanations such as incompatibility resolution contribute as well. Future experiments might also utilize a high throughput method to explore segmental loss of heterozygosity in hybrids at a genome wide scale, similar to ongoing experiments at the gene level (Lancaster S, Dunham MJ, unpublished data). While our sample size is modest, this is a 16 novel and necessary step in understanding forces underlying hybrid genome stabilization and highlighting an underappreciated mechanism of hybrid adaptation.

Conclusions
The mutation events we observe in our experimentally evolved hybrids are in many ways quite representative of mutations observed in ancient hybrid genomes, suggesting that hybrid genome stabilization and adaptation can occur quite rapidly (within several hundred generations). Furthermore, our results illustrate that the infusion of variation introduced by hybridization at both the intra-and inter-species level can increase fitness by providing choices of alleles for selection to act upon, even when sexual reproduction is rare. This may be particularly important for leveraging existing variation for agricultural and industrial processes, and as climate change potentially increases natural hybridization (Hoffmann and Sgro 2011;Kelly, et al. 2010;Muhlfeld, et al. 2014).

Strains
A list of strains used in this study is included in Supplemental Table 1. All interspecific hybrids were created by crossing a ura3 LYS2 haploid parent to a URA3 lys2 haploid parent of the other mating type, plating on media lacking both uracil and lysine, and selecting for prototrophs.

Evolution Experiments
Continuous cultures were established using media and conditions previously described (Gresham, et al. 2008;Sanchez, et al. 2016). Detailed protocols and media recipes are available at http://dunham.gs.washington.edu/protocols.shtml. Samples were taken daily and measured for optical density at 600 nm and cell count; microscopy was performed to check for contamination; and archival glycerol stocks were made daily. An experiment was ended when contamination, growth in tubing, or clumping appeared (number of generations at the endpoint for each population shown in Tables 1, 2). Samples from each endpoint population were colony-purified to yield two clones for further study.

Array Comparative Genomic Hybridization (aCGH)
Populations from the endpoint of each evolution were analyzed for copy number changes using aCGH following the protocol used in Sanchez et al. (2016). Microarray data will be made available upon publication in the GEO database and the Princeton University Microarray Database.

Sequencing
DNA was extracted from overnight cultures using the Hoffman-Winston protocol (Hoffman and Winston 1987), and cleaned using the Clean & Concentrator kit (Zymo Research). Nextera libraries were prepared following the Nextera library kit protocol and sequenced using paired end 150 base pairs on the illumina NextSeq 500 machine (sequencing coverage in Supplemental Table 2). The reference genomes used were: S. cerevisiae v3 (Engel, et al. 2014), S. uvarum (Scannell, et al. 2011), and a hybrid reference genome created by concatenating the two genomes. Sequence was aligned to the appropriate reference genome using bwa v0.6.2 ) and mutations were called using GATK (McKenna, et al. 2010) and samtools 0.1.19 ). Mutations in evolved clones were filtered in comparison to the ancestor to obtain de novo mutations. All mutations were first visually inspected using Integrative Genomics Viewer (Robinson, et al. 2011). Subsequently, point mutations in the hybrids were confirmed with Sanger sequencing (Supplemental Table 3). Copy number variants were visualized using DNAcopy for S. cerevisiae and S. uvarum (Seshan and Olshen 2016). Loss of heterozygosity events were called based on sequencing coverage in the hybrids, and by identifying homozygous variant calls in S. cerevisiae. All breakpoints were called by visual inspection of sequencing reads and are thus approximate.

Fitness assays
The pairwise competition experiments were performed in 20 mL chemostats (Miller and Dunham 2013). Each competitor strain was cultured individually until steady state was reached, and then was mixed 50:50 with a GFP-tagged ancestor. Each competition was conducted in two biological replicates for approximately 15 generations after mixing. Samples were collected and analyzed twice daily. The proportion of GFP+ cells in the population was detected using a BD Accuri C6 flow cytometer (BD Biosciences). The data were plotted with ln[(dark cells/GFP+ cells)] vs. generations. The relative fitness coefficient was determined from the slope of the linear region.

Strain construction
Allele replacements for the PHO84 locus were done following the protocol of the Caudy lab with further modifications described here. The native locus was replaced with Kluyveromyces lactis URA3. The pho84∆::URA3 strain was grown overnight in 5 mL of C-URA media, then inoculated in a flask of 100 mL YPD and grown to an OD of 0.6-0.8. Cells were washed then aliquoted. 275 µl of transformation mix (35 µl 1M Lithium Acetate, 240 µl of 50% 3500 PEG), 10 µl of Salmon sperm, and approximately 3 µg of PCR product were added to the cell pellet. It was incubated at 37°C (S. uvarum) or 42°C (S. cerevisiae) for 45 minutes, then plated to YPD. It was replica plated to 5FOA the following day and colonies were tested for the gain of the appropriate species allele. The GRF167 allele was cloned into the pIL37 plasmid using Gibson assembly (Gibson, et al. 2009). Correct assembly was verified by Sanger sequencing. All primers used can be found in Supplemental Table 3. and an amplification of S. cerevisiae chrXIII with corresponding deletion of S. uvarum chrXIII (copy neutral loss of heterozygosity). Sh4 has an amplification of S. cerevisiae SUL1 and a whole chromosome amplification of S. uvarum chrVIII, (note, there is a reciprocal translocation between chrVIII and chrXV). Note that Sh4 is plotted on a different scale. For specific coordinates of copy number variants, see Table 1.   Point mutations, copy number variants (CNV), and loss of heterozygosity events (LOH) are recorded for each evolved parental clone. Clones are identified by nutrient (G: glucoselimitation, P: phosphate-limitation, and S: sulfate-limitation), by species ("c" denotes S. cerevisiae, "u" denotes S. uvarum), and the number indicates its derivation from independent populations. Note that mutations in the S. uvarum genome use S. uvarum chromosomes and coordinates. All breakpoints were called by visual inspection of sequencing reads and are thus approximate.

Fig S1: Whole genome copy number variation in S. cerevisiae evolved clones
Sequencing coverage for each evolved clone was normalized using the ancestor, and copy number variants were inferred by changes from a copy number of two using DNAcopy for each nutrient condition A. glucose-limitation, B. phosphate-limitation, and C. sulfate-limitation.
Chromosomes are plotted in alternating grey and red. The average copy number is plotted as a black line.

Fig S2: Whole genome copy number variation in S. uvarum evolved clones
Sequencing coverage for each evolved clone was normalized using the ancestor, and copy number variants were inferred by changes from a copy number of two using DNAcopy for each nutrient condition A. glucose-limitation, B. phosphate-limitation, and C. sulfate-limitation.
Chromosomes are plotted in alternating grey and blue. The average copy number is plotted as a black line. Note that plots use S. uvarum chromosomes and coordinates. Allele frequency as determined from sequencing read depth data is plotted across chromosomes for each observed loss of heterozygosity event in evolved S. cerevisiae clones in A.
glucose-limited conditions, B. phosphate-limited conditions, and C. sulfate-limited conditions.
The ancestral sequence is plotted at the top of each figure in grey, followed by each sequenced clone from that condition.

SUPPLEMENTARY TABLES
Supplemental