A budding yeast hybrid showing growth heterosis
We generated hybrids by mating haploids of S. cerevisiae and S. paradoxus, two closely-related sensu-stricto species that express largely the same set of genes, with 90% and 80% sequence identity in coding or inter-genic regions, respectively [27]. The hybrids are limited in meiosis, producing less than 1% viable gametes [28], but can propagate vegetatively without signs of genetic instability or aneuploidy.
When provided complete media (SD), both diploid parents grow with a division time of approximately 90 minutes, typical of rapidly growing strains. Still, the hybrid grows approximately 20% faster than both diploid parents (one-way ANOVA, P < 10–7) (Fig. 1a and Additional file 1: Figure S1A). Growth heterosis was observed also in other conditions, including high temperature, high ethanol concentrations, and low Pi concentrations (Fig. 1a). We used live-cell microscopy to quantify the duration of the different cell cycle phases (Fig. 1b, Additional files 2, 3 and 4: Movies S1–S3). The two parents regulate their cell cycle differently; S. cerevisiae cells have short G2 phase, thereby generating small daughter cells that extend their G1 phase to retrieve their mother’s size. S. paradoxus cells, on the other hand, regulate their size by extending the G2 phase, followed by a short G1 [29]. We find that the hybrid G2 phase was as short as in S. cerevisiae, yet it did not extend its daughter’s G1, which was as short as in S. paradoxus (Fig. 1c and Additional file 1: Figure S1B).
Additional file 2: Movie S1. Growth of S. paradoxus. Note the pseudo hyphae-like growth. (M4V 2559 kb)
Additional file 3: Movie S2. Growth of S. cerevisiae. Note S. cerevisiae’s asymmetrical division. (M4V 766 kb)
Additional file 4: Movie S3. Growth of the hybrid. Note the synchronized growth in the hybrid due to loss of G1 and G2 delay in the daughter cells. (M4V 455 kb)
Sustained rapid growth entails a more efficient production of biomass. Biomass and energy production are regulated by the routing of carbon through central carbon metabolism. We previously noted that respiratory gene expression is higher in the hybrid relative to its parents, even when grown in glucose (Figure S13 in [18], Fig. 2a), which was surprising, as budding yeast exhibits glucose-mediated catabolite repression of the TCA cycle. We therefore asked whether glucose repression is reduced in the hybrid, enabling more efficient energy generation through respiration. Measuring oxygen usage along the growth curve confirmed that hybrids consumed oxygen at a high rate throughout the growth curve, even when glucose was abundant, in contrast to S. cerevisiae and S. paradoxus, where oxygen consumption was lower when glucose was present (Fig. 2b and Additional file 6: Figure S2A, B). Consistently, hybrid mitochondria were larger and contained more cristae compared to S. cerevisiae and S. paradoxus, as visualized by electron microscopy (Fig. 2c, d and Additional file 6: Figure S2C). Finally, heterosis was reduced upon the addition of the respiration blocker Antimycin A to the level of the best parent (Fig. 2e and Additional file 6: Figure S2D). Together, our results suggest that reduced glucose repression in the hybrid enables it to respire even in the presence of glucose.
Increased respiration may lead to oxidative stress that can cause strand breaks in DNA. Therefore, we assessed DNA damage in the hybrid by following Rad52-GFP, a protein that localizes to foci of DNA double-strand breaks [30]. Notably, the frequency of cells with Rad52-GFP foci increased by approximately two-fold in the hybrid compared to either parent (Fig. 3a). Further, we noted that, in the hybrid, a group of cytosolic chaperones were localized into punctate structures, a known marker for stress, such as DNA damage (Fig. 3b and Additional file 7: Table S2).
A genome-wide genetic screen detects hundreds of alleles contributing specifically to hybrid growth in an environment-dependent manner
Classical models of heterosis, including dominance, overdominance, or dosage models, attribute the hybrid’s superior performance to the action of specific alleles. We reasoned that, working with budding yeast, we could systematically screen for heterotic alleles, defined here as alleles that contribute to the hybrid’s growth but do not show a dosage effect in the parental background. This general definition includes alleles that function through dominance, dosage, overdominance, epistasis, or more complicated (e.g., cis and trans) effects.
Our screen is based on the availability of a deletion library, corresponding to all non-essential S. cerevisiae genes. Starting from this library, we generated a library of hemizygote hybrids that lack a specific S. cerevisiae allele but contain the corresponding S. paradoxus allele, and a control library containing hemizygote S. cerevisiae diploids (Fig. 4a). Two independent libraries were generated for each genetic background.
The deletion library was specifically designed to enable sensitive measurement of the growth rate of individual strains, while growing all strains in one pool [31, 32]. This approach has the advantage that all strains are exposed to the same environment, making a comparison between the individual strains more controlled. Specifically, each strain in the library is marked by a specific sequence barcode that is flanked by a common sequence, so that high-throughput sequencing can be used to quantify the relative abundance of each strain within the growing pool (Fig. 4b, c, see Methods and Ref [33]). Temporal changes in a strain’s abundance during pool growth indicate its growth rate relative to the pool’s average, i.e., slow growing strains will be gradually outcompeted, while the fast growing ones will become increasingly more abundant. Note that the measured growth rate of the pool will approximate the wild-type growth rate as the majority of mutants do not show a growth defect at any given condition [33].
We tested strain performance under five growth conditions (YPD, YPD + Sorbitol 1 M, YPD at 37 °C, YPD + 8% ethanol, and YPD + 6% ethanol; Fig. 4d, Additional files 8 and 9: Table S3 and Table S4). In each condition, the pools were maintained in log-phase throughout the experiment by back-dilution (see Methods) to ensure constant conditions and limit possible effects of nutrient depletion or secretion. In total, 865 alleles reduced growth in at least one condition or genetic background (Z Score < –1.5 in both biological replicates, Additional file 10: Figure S3A). These genes were classified into functional groups based on databases and literature (Additional files 10 and 11: Figure S3B and Table S5). Inferred growth rates were highly correlated between two replicates corresponding to two independently pooled libraries.
The effects that the different alleles had on growth differed between the conditions (Fig. 4e). For example, stress conditions, such as heat and high ethanol concentration, increased dosage sensitivity to genes involved in peroxisome function, cell wall formation/breakdown, or protein and lipid modifications (Additional file 10: Figure S3C). Most notably, while previously annotated haploinsufficient genes in S. cerevisiae [34], such as ribosome components, were reproducibly identified as dosage sensitive in rich media, these strains showed no effect in growth conditions in which cells were growing more slowly (Fig. 4f and Additional file 10: Figure S3D). Thus, depending on the growth conditions, loss of one allele could range from deleterious to even being beneficial.
Surprisingly, perhaps, the set of genes sensitive to hemizygosity in the hybrid greatly differed from that of its S. cerevisiae parent, even within the same growth condition (Fig. 4e, f). Thus, hundreds of S. cerevisiae alleles contribute to hybrid growth but not to S. cerevisiae growth. These alleles are not confined to a single pathway, but are associated in multiple cellular processes (Fig. 4f and Additional file 10: Figure S3B, C).
Heterotic alleles consistent with hybrid phenotypes
We observed a correspondence between allele-specific sensitivity and the hybrid phenotypes noted above. First, the hybrid shows increased sensitivity to alleles involved in the G1/S transition (Fig. 5a) and to alleles whose deletion increases cell size, irrespectively of their functional association (hypergeometric P value: Hybrid < 10–10, S. cerevisiae: 0.65, Fig. 5b). This increased sensitivity may be explained by the shorter duration of this cell cycle phase in the hybrid compared to S. cerevisiae, i.e., as the G1/S transition is a major checkpoint where nutritional status and cell-size are monitored and its duration is correlated to birth size [35, 36], the hybrid’s shorter G1/S duration may limit the checkpoint capacity to correct for size perturbations that would require shortening its duration beyond the functional limit.
Also connected to cell cycle progression, the hybrid showed an increased sensitivity to DNA repair genes (Fig. 5c). This, together with the phenotypic results of increased presence of DNA damage markers in the hybrid (Fig. 3a, b), may suggest suboptimal performance of mechanisms that maintain genome integrity.
Finally, consistent with the reduced glucose repression seen in the hybrid, the hybrid was sensitive to the deletion of S. cerevisiae alleles coding for genes that are localized to mitochondria. This sensitivity exceeds that of its S. cerevisiae parent (P value < 10–2), and was particularly pronounced during growth on ethanol, but observed also on glucose (Fig. 5d).
The hybrid escapes a programmed cell cycle slow-down under severe ethanol stress
The largest difference in the pattern of allelic sensitivity between the hybrid and S. cerevisiae was observed under conditions of high ethanol stress. Under this condition, a reproducible minority of hemizygote strains overtook the population in the two S. cerevisiae replicates (Fig. 6a, b and Additional files 12 and 13: Figure S4A, B, Table S6). In contrast, the hybrid showed the typical pattern of allele sensitivity as seen in other conditions. This differential pattern of effects was also reflected in the overall growth of the pools (cf. Figs. 1a and 4d); while the hemizygote hybrids maintained steady growth throughout the experiment, similarly to all other conditions tested, growth of the hemizygote S. cerevisiae pool was initially rapid, then slowed down, and became rapid again after approximately 10 generations. The rapid growth of the hybrid in high ethanol is especially striking considering that the S. paradoxus parent fails to grow in this condition (Fig. 1a). Note that, as during most of the growth phase cells were maintained at low density, the possible metabolism of ethanol by the cells had no effect on ethanol concentration.
The hybrid therefore maintains a stable growth in high ethanol concentration, while the S. cerevisiae diploids slowdown their growth after some period. Notably, this slowdown can be overcome by decreased expression of individual genes. This unique dosage response suggests that the growth slow-down is an active and adapted strategy and not a passive reaction to unavoidable toxicity. In support of that, strains that are maladapted in rich conditions were enriched amongst the surviving hemizygote diploids (Figs. 4f and 6c), which may also explain the selection of euploid S. cerevisiae x S. uvarum hybrids upon ethanol stress [37].
To try and identify the basis of this increased ethanol resistance in the hybrid, we examined the pattern of allelic sensitivity. The hybrid showed an increased dependency on retrograde signaling (Additional file 12: Figure S4C). This response is triggered by damaged mitochondria to induce nuclear-encoded protecting mechanisms [38]. Induction of this pathway within the hybrid could render the hybrid more stress resistant. In support of that, over-activating the retrograde pathway increased growth under ethanol stress for both the hybrid and its S. cerevisiae diploid parent, although accounting for only a fraction of hybrid growth vigor (Fig. 6d and Additional file 12: Figure S4D).