Rapid adaptive evolution of myo1Δ yeast cells
At the start of the evolution experiment (), we dissected 24 tetrads sporulated from a MYO1/myo1Δ heterozygous diploid strain to isolate fresh myo1Δ haploid spores. After three days of growth, the tetrads exhibited the expected 2:2 segregation for growth between wild type (MYO1) and mutant (myo1Δ) spores (). After seven more days of incubation, 15 out of the 48 myo1Δ spores (31%) gave rise to small colonies, whereas 100% of the MYO1 spores grew to large colonies. Each of the 15 myo1Δ survivor colonies was streaked onto a new plate for growth into isolated colonies (P2 in ). After three days, the three largest colonies from each plate were selected and streaked onto a new plate. Subsequently, the 45 myo1Δ strains were evolved separately by a repeated process of manually selecting the largest colony, which was then streaked onto a new plate for propagation (). At each passage, we compared size and number of viable colonies and cytokinesis efficiency (see for an example) with those of a wild type strain propagated in parallel. A strain was deemed to have evolved to a steady state when colony growth remained constant for at least four consecutive passages. This experiment ceased after the 20th passage (corresponding to ~400 generations for wild type) and resulted in 29 stably evolved strains (e-strains) with different evolutionary history and a variety of terminal phenotypes ().
Adaptive evolution and phenotypic characterization of myo1Δ e-strains
To obtain a quantitative phenotypic classification of the 29 myo1Δ
e-strains, we determined their cytokinesis index (CI), rates of biomass and cell number increase, cell viability and DNA content over a time course of 10 hours. This dataset was subjected to a Principal Components Analysis (PCA). Projection of the phenotypic data in the space defined by the first two principal components, describing >80% of the total variance in the dataset, discriminated all wild type controls (, diamonds) from all the e-strains by a higher cell mortality in the latter. PCA further separated the 29 analyzed e-strains into distinct phenotypic classes. A cluster of ten strains was characterized by higher cytokinesis indexes, biomass increase rates and cell division rates in comparison to the remaining e-strains and were thus referred to as the “10 fittest e-strains” throughout this study (, large dark circles, and Figure S1
). Even thought these strains divide and grow well under optimal conditions, competition experiments and growth assays under a variety of conditions showed that they were still less fit than wild-type cells (Figures S2
Loss of ability to assemble the contractile ring in myo1Δ e-strains
As a first step to characterize the evolved cytokinesis in the e-strains, we analyzed the localization of actin and Chs2, known partners of Myo1 during normal cytokinesis. Both proteins join Myo1 in the cytokinetic ring at the bud neck in late anaphase (Balasubramanian et al., 2004
). Compared to wild-type cells, the actin ring and Chs2-GFP ring were observed at drastically reduced frequencies in all ten fittest e-strains (Figure S3
). In those cells that did show Chs2 localization the intensity of the signal was often dim and time-lapse microscopy data further showed that these rings often faded without evidence of constriction (Figure S3
). Since these e-strains exhibited cytokinesis efficiencies comparable to wild type (Figure S1
), these data indicated that it was unlikely that the evolved cytokinesis in the myo1Δ
e-strains relied on formation of an actin or Chs2-containing cytokinetic ring.
We further tested if the e-strains retained the ability to assemble a myosin ring at the bud neck by re-introducing Myo1-GFP, carried on a centromeric plasmid under the control of its own promoter, into the wild type and the 10 fittest e-strains. Consistent with published observations, Myo1-GFP localizes at the bud neck in all budded wild-type cells, but most e-strains, with the exception of 12d-1 and 12d-2, exhibited drastically reduced ability to correctly localize Myo1 (), further suggesting that the e-strains were unlikely to perform cytokinesis through restoration of the previous cytokinetic apparatus.
Microscopy characterization of cytokinesis in 10 fittest myo1Δ e-strains
The fittest e-strains display three major mechanisms of septum formation
To gain further insights into how cytokinesis is achieved in the absence of MYO1
, we carried out thin-section transmission electron microscopy (TEM) to examine septum morphologies of the 10 fittest e-strains. Wild-type cells formed a single, straight primary septum, which developed behind the cleavage furrow intruding symmetrically from opposite sides of the bud neck (). A secondary septum was then deposited along both sides of the primary septum. Four different septum morphologies were observed in dividing cells of myo1Δ
e-strains: I) a primary and secondary septum that appeared morphologically similar to those in wild type cells, with only subtle differences such as occasional unilateral formation of a wavy primary septum; II) a single but slightly aberrant primary septum, frequently characterized by cytoplasmic entrapment; III) multiple randomly oriented septa, some of which emanated even from regions distant from the bud neck; IV) thickening of the cell wall around the bud neck without the formation of a primary septum (). Though none of the e-strains homogeneously displayed any single septum morphology, quantification of TEM images showed that one of three morphologies (I, III or IV) was predominant in each e-strain (). Serial sectioning TEM confirmed that the different septum morphologies observed were not simply due to sectioning through different planes of the bud neck (Figure S4
We next used 3D time-lapse imaging to visualize plasma membrane dynamics during cytokinesis in e-strains representative of all septum morphologies. These e-strains expressed RAS2-GFP
, used as a plasma membrane marker (Whistler and Rine, 1997
), and GFP-TUB1
, to visualize the mitotic spindle and mark cell cycle progression. In wild-type cells, closure of the mother-bud junction is achieved through bilateral invagination of the plasma membrane at the bud neck 4–6 min after spindle breakdown ( and Movie S1
). Seven time-lapse movies made of 12d-1 strain all showed behaviors consistent septum morphology I, four of which formed monolateral and three bilateral invaginations that extended across the bud neck to complete cytokinesis within 4–8.5 min after spindle breakdown ( and Movie S2
). Similar membrane closure dynamics were also observed in 7a-2 (Movie S3
) and 24d-3 e-strains (Movie S4
). Five of 9 movies of 13c-2 (representing morphology III) revealed formation of multiple, unstable invaginations from both sides of (and also far away from) the bud neck ( and Movie S5
). Cytoplasm separation was accomplished more slowly (6–30 min) than in the wild type. Movies of e-strains 7a-1 and 23a-1 (representing morphology IV) showed that in the majority of cells recorded (3/5 for 7a-1 and 5/6 for 23a-1), there was no evidence of furrowing, but membrane closure was achieved by a slow (19–33 min) inwardly pushing of the plasma membrane ( and Movies S6
), consistent with the idea that cytokinesis was achieved through gradual thickening of the cell wall around the bud neck. These results revealed the presence of three distinct physical mechanisms of membrane closure among the 10 fittest e-strains, each consistent with one of the three major septum morphologies observed by TEM.
Different septum morphologies correlate with distinct patterns of gene expression
To identify molecular pathways underlying the evolved cytokinesis mechanisms, we performed a microarray-based mRNA profiling of the ten fittest e-strains (Supplemental Experimental Procedures
and Spreadsheet S1
). After filtering the data to remove non-changing genes, a statistical approach was applied to detect gene expression changes that significantly correlated with each of the four septum morphologies (). Gene expression changes detected by microarray analysis were validated by quantitative reverse transcription PCR (qRT-PCR) on 12 genes across all ten fittest e-strains (Figure S5
Transcriptome changes underlying the evolved phenotypes
FatiScan analysis (Al-Shahrour et al., 2006
) was applied to determine enrichment of Gene Ontology (GO) annotations among genes correlated with each of the septum morphologies ( and Figure S6
). While several GO Biological Process annotation terms that positively correlated with septum morphology I and II included ribosomal biogenesis, septum morphology IV was positively associated with cell wall biogenesis. Few biological processes were found to associate with septum morphology III possibly due to a small number of genes found to correlate with this morphology (data not shown). In addition, manual inspection of the microarray data revealed that both Hsp90 genes (HSP82
) were more up-regulated in the e-strains that preferentially divided through morphology I (). Specific correlation of above pathways with septum morphology I and IV was further corroborated by a second analysis applied to the gene expression data of strains 7a-1, 7a-2 and 7a-3, which descended from the same spore but showed divergent septum morphologies (Figure S7
Functional validation of pathways revealed by microarray analyses through genetic manipulation was challenging due to various genomic changes in the e-strains (see below) and their propensity to undergo adaptive evolution when perturbed. Chemical inhibitors were therefore used to acutely inhibit specific cellular pathways in the e-strains. We first used radicicol, an Hsp90 inhibitor (Roe et al., 1999
), to test the requirement for Hsp90 in the growth of the e-strains. Consistent with the correlation of Hsp90 expression with septum morphology I, the five e-strains that displayed predominantly this morphology were hypersensitive to radicicol compared to wild type and the other 5 e-strains favoring morphology III or IV (). Liquid-culture experiments found that radicicol inhibited cytokinesis in the 5 e-strains that predominantly showed septum morphology I (). This provides a functional validation of a specific involvement of Hsp90 in the evolved cytokinesis represented by septum morphology I. Hypersensitivity to cycloheximide, an inhibitor of ribosome function, was also observed among the same group of e-strains (), consistent with a higher requirement for ribosome function in those e-strains. Since cycloheximide strongly inhibited budding and cell cycle progression of these e-strains (data not shown), we were unable to test specifically its effect on cytokinesis. Taken together, above results suggest that distinct cytokinesis mechanisms arose from expression changes of genes defining specific cellular functions.
Adaptive evolution of myo1Δ yeast correlates with and is enhanced by polyploidization
To identify the genetic changes that led to heritable phenotypic variation in the e-strains, we first used flow cytometry to determine the cellular DNA content of the 29 stably evolved myo1Δ
e-strains (Figure S8
). 20 of the 29 e-strains, including the 10 fittest e-strains, had increased their ploidy from the haploid state to a state ranging from diploid to tetraploid, and the rest remained haploid (). Cytokinesis failure in budding yeast normally results in multiple-budded, multi-nucleate cells, which could account for a ploidy increase. However, ~90% cells in each of the 10 fittest e-strain population were mononucleate (Figure S9
), thus these myo1Δ
e-strains had undergone mononucleate polyploidization (referred to hereafter as polyploidization). In contrast, 7 of 8 e-strains that were still defective in cytokinesis at the end of the evolution experiment were still haploid, showing that polyploidization was not an obligatory outcome of the cytokinesis defect. Instead, this analysis revealed a strong correlation between polyploidization and restoration of cytokinesis at the end of the evolution ( and Table S3
myo1Δ e-strains are characterized by polyploidy and aneuploidy
The above observation raised the possibility that polyploidization facilitates adaptive evolution of myo1Δ cells. To test this, a MYO1/MYO1/myo1Δ/myo1Δ tetraploid strain was used to generate fresh myo1Δ/myo1Δ diploid cells through sporulation. The adaptive evolution experiment was then repeated, comparing the rate of cytokinesis improvement between 7 myo1Δ/myo1Δ diploid spores and 22 myo1Δ haploid spores. Both haploid and diploid myo1Δ spores initially exhibited 100% cytokinesis defect (), but diploid strains showed significantly faster evolution during early passages (). Interestingly, the cytokinesis efficiency of myo1Δ haploid strains did reach that of diploid strains in later passages (), consistent with polyploidization in the haploids during earlier passages.
Whole chromosome aneuploidy occurred in all 10 fittest myo1Δ e-strains
Flow cytometry data indicated that the DNA content of some e-strains was between that of a pure diploid and a pure tetraploid, suggesting the possibility of aneuploidy (Figure S8
). Array-based Comparative Genomic Hybridization (aCGH) indicated that all 10 fittest e-strains, but not the wild type strain propagated in parallel, carried at least one chromosome in aneuploidy (). The frequency at which individual chromosomes were observed in aneuploidy was not evenly distributed: some chromosomes (e.g. II, III, XIII and XVI) were in aneuploidy across the majority of the e-strains, while others (i.e. V, VII and XIV) were always found in euploid number. The karyotype of the e-strains was stable during the time frame of this work, as shown by replicate aCGH on three independent cultures of strain 7a-1 and on two independent cultures of strain 7a-2 (data not shown).
The spontaneous rate of chromosome missegregation is related to chromosome size in both yeast (Murray et al., 1986
) and vertebrate cells (Spence et al., 2006
). However, aneuploidization frequencies in the e-strains were uncorrelated with either chromosome length or chromosomal gene content (Figure S10
). In contrast, shared aneuploid chromosomes can be found among e-strains that preferentially divided with the septum morphology (). To further determine the correlation between karyotype and evolved phenotypes, aCGH was carried out on four additional e-strains, each sharing an ancestral spore with one or more of the 10 fittest e-strains but resulting in less fit evolved phenotypes. In every case, the poorly evolved e-strains exhibited a drastically different aneuploidy pattern from their well evolved relatives (Figure S11
Besides the expected deletion of the MYO1
locus and the presence of the TRP1
marker used for generating the deletion, close inspection of the aCGH data across all the chromosomes indicated absence of sub-chromosomal amplifications or deletions in all ten fittest myo1Δ
e-strains (). The only local changes in hybridization intensity mapped exclusively to highly repetitive sequences, such as transposons and telomeric repeats (), for which aCGH data cannot be interpreted (Pinkel and Albertson, 2005
). Pulse-field gel electrophoresis revealed no gross chromosomal rearrangements or non-homologous end joining (Figure S12
Identification of point mutations in e-strains by Illumina whole-genome re-sequencing
Phenotypic adaptation during experimental evolution can often be attributed to point mutations in DNA sequences that result in changes in protein activities or expression levels (Elena and Lenski, 2003
). However, given the fact that all of the e-strains have become polyploid and aneuploid, it would be difficult to identify mutations in these strains by commonly used genetic mapping techniques. We therefore used the Illumina sequencing system to obtain full genomic sequences from five strains: the parental MYO1/myo1Δ
diploid, the wild type strain at passage 1 and at passage 20, and two e-strains, 7a-1 and 7a-2, which exhibited the same basal ploidy, similar but non-identical karyotypes, and different predominant septum morphologies.
For all five strains we obtained high genome coverage, high sequencing depth and quality, within or above the expected performance of this technology (Tables S4 and S5
). Within the non-repetitive genome, 820 single nucleotide polymorphisms (SNPs) were called as homozygous mutations in the parental diploid strain, compared to the published yeast genome sequence (Spreadsheet S2
). 803 (~98%) of these SNPs were also identified in 3 of the 4 remaining strains and were therefore likely to be true mutations present in the parental genetic background. Of the 33 SNPs predicted in the parental diploid strain as heterozygous, 23 (~70%) were also detected in either the wild type haploid strains or the 7a e-strains or both (Spreadsheet S2
). Six of these predicted heterozygous mutations were validated by Sanger sequencing (data not shown), further confirming the ability of this technology to identify heterozygous mutations.
Nine and twelve mutations were predicted to be unique to 7a-1 and 7a-2, respectively. These SNPs, if confirmed, could contribute to phenotypic evolution; however, SNPs predicted by a high-throughput sequencing technology to be unique to a single strain are more prone to bear false positives compared to SNPs consistently identified across several strains. Indeed, Sanger sequencing showed that all 9 mutations predicted to be unique to e-strain 7a-1 were false positives, whereas only a single mutation was confirmed in e-strain 7a-2. This mutation mapped to the coding sequence of YGR130C
, encoding a protein of unknown function, and causes a lysine to asparagine change that was predicted by the SIFT software (Ng and Henikoff, 2001
) not to affect protein function (Spreadsheet S2
). Taken together, within the limitation of available technology, current evidence has not indicated point mutations to be the primary genetic changes accounting for the divergent cytokinesis mechanisms observed in myo1Δ
Aneuploidy affects gene expression at and beyond the chromosomal average level
Above findings left us with the unexpected possibility that aneuploidy, observed in all e-strains, was the primary genetic change that could account for the heritable phenotypic variation. To test this, we first asked whether aneuploidy could account for the global transcriptome changes that occurred in the myo1Δ
e-strains. Consistent with a global effect of aneuploidy on transcriptome, the two pairs of e-strains with identical karyotypes (i.e. 7a-2/7a-3 and 23a-1/23b-1) displayed the highest overall transcriptome similarity: the microarray data of these pairs were in fact as similar to each other as any two biological replicates of a same strain (Figure S13
). Previous studies showed that genes encoded on extranumerary chromosomes tend to be more highly expressed than genes on euploid chromosomes (Hughes et al., 2000
; Torres et al., 2007
). Above observation was confirmed by directly comparing our aCGH data with our microarray gene expression data in the myo1Δ
e-strains (). In addition, there was a direct proportionality between chromosome average gene expression change and the chromosome copy number stoichiometry (): i) gaining two extra copies of a given chromosome produced a chromosomal average expression change roughly twice as high as that caused by gaining one extra copy; and ii) gaining an extra chromosome in a triploid background had a smaller effect on the chromosomal average gene expression level than gaining the same chromosome in a diploid. These data demonstrate that chromosome stoichiometry quantitatively determined the chromosomal average level of gene expression.
Effects of aneuploidy on gene expression
Besides the above effect, careful inspection of gene-specific expression changes in each e-strain revealed the presence of genes whose expression change deviated considerably from the chromosome average change (). Whereas most genes indeed changed their expression within one standard deviation from the chromosome average change (referred to as inlier genes), a small number of genes changed their expression level more than three standard deviations away from the chromosome average change (referred to as outlier genes). This small number was nonetheless significantly higher than the number expected by chance (Figure S14
Since a subset of the outlier genes encompassed some of the genes identified to be correlated with specific cytokinesis phenotypes (Spreadsheet S1
), we investigated if and how aneuploidy might be able to cause such gene expression changes. Two possibilities were considered (). First, karyotype changes could directly give rise to outlier gene expression changes, if some genes were intrinsically more sensitive or more resistant to gene copy number changes. If this was correct, outlier genes should be preferentially located on aneuploid chromosomes; however, they were found to be evenly distributed across the whole genome (). The second possibility was that outlier gene expression changes might be an indirect consequence of karyotype changes, mediated by inlier gene expression changes. In support of this hypothesis, e-strains with similar inlier gene expression also displayed similar outlier gene expression and vice versa
, indicating that outlier gene expression changes correlated with inlier gene expression changes ().
A possible explanation for this observation is that altered expression of transcription factors (TFs) due to their gene location on aneuploid chromosomes could cause expression changes of target genes not necessarily encoded on aneuploid chromosomes. To test this, we took advantage of a recently published yeast transcriptional regulatory network, reconstructed based on functional identification of direct and indirect targets of 268 yeast TFs (Hu et al., 2007
). Although the genome coverage of this network is incomplete, 2570 (~40%) of yeast genes are represented as potential targets of the analyzed TFs (P
≤ 0.001). Moreover, both the TFs and their targets are evenly distributed across all chromosomes (Figure S15
). Based on this data, we found that 43–78% of the outlier genes found in the myo1Δ
e-strains were downstream targets of TFs encoded on aneuploid chromosomes (, black bars). In contrast, only 8–31% of non-outlier genes had predicted TFs on aneuploid chromosomes (, white bars). This indicates that a specific pattern of aneuploidy might be sufficient to predict gene expression changes both at and beyond chromosome average levels. Interestingly, a specific pattern of aneuploidy was not required to cause a specific set of outlier genes. In fact, for a given set of outlier genes in an e-strain, their upstream regulators were not biased toward those carried on aneuploid chromosomes in that particular e-strain (). This implies that similar patterns of outlier gene expression may be achieved through different patterns of aneuploidy, supporting the observation that e-strains with different karyotypes can converge to a similar cytokinesis phenotype.
Phenocopying myo1Δ e-strains by mimicking gene-dosage increase due to aneuploidy
To experimentally test the hypothesis that aneuploidy can contribute to phenotypic evolution, we extended the transcription network analysis to specific target genes implicated in achieving septum morphology I or IV. First, we analyzed how aneuploidy could contribute to expression changes in the Hsp90 genes, which correlated with septum morphology I. HSC82 and HSP82 are located on chromosome XIII and XVI, respectively. Either or both of these chromosomes were gained in all five e-strains that favored morphology I (), suggesting that aneuploidy could directly account for increased expression of Hsp90 in these strains. Note that HSC82 and HSP82 share ~97% sequence identity and it is therefore challenging to design microarray probes that would clearly distinguish the two isoforms without any cross-hybridization. This might explain why strain 24d-3 also showed increased expression of HSC82 even though chromosome XIII was in euploid number.
On the other hand, septum morphology IV was associated with up-regulation of genes involved in cell wall biogenesis (), consistent with the slow closure of the mother-bud junction through cell wall thickening (). We noticed that in e-strains 23a-1 and 23b-1, which show the highest preference for morphology IV (), chromosome XVI was the only aneuploid chromosome. Moreover, chromosome XVI gain was the only aneuploidy shared among all e-strains preferentially dividing by morphology IV (). E-strains 23a-1 and 23b-1 shared 58 outlier genes, 11 of which are known targets of RLM1
, a TF encoded on chromosome XVI and implicated in cell wall remodeling. Most of these 11 RLM1
targets encode either cell wall proteins or proteins involved in cell wall biogenesis. Microarray and qRT-PCR analyses showed that 9 of these 11 genes were up-regulated in 23a-1 and 23b-1 up to 16 fold compared to wild type (). In addition, a signaling molecule that positively regulates Rlm1, the MAP kinase kinase Mkk2 (Levin, 2005
), is also encoded on chromosome XVI. These observations provided an opportunity to test if modulating copy numbers of specific genes on chromosome XVI, such as RLM1
, could bring about the evolved phenotype displayed by e-strains 23a-1 and 23b-1.
Effects of increased copy numbers of RLM1 and MKK2 in myo1Δ cells
We introduced two extra copies of RLM1 and of MKK2 into the MYO1/myo1Δ heterozygous diploid strain at the RLM1 and MKK2 locus, respectively. After sporulation, tetrads were dissected and spores were genotyped retrospectively after the colonies grew up. The genotype of most dead spores could be inferred assuming Mendelian segregation. First, we observed that gaining both genes, but not either gene alone, significantly improved the initial spore viability (). Assessment at passage 1 confirmed that gaining both genes led to significantly improved cytokinesis compared to control myo1Δ spores (). The morphology of cytokinesis in the resulting strains was subsequently analyzed by TEM (). We found that whereas among myo1Δ spores <50% of the cells exhibited septum morphology IV, this morphology was the predominant phenotype (~75%) among myo1Δ cells carrying extra copies of RLM1 and MKK2 (). These observations indicated that gaining extra copies of RLM1 and MKK2 not only improved the initial survival and cytokinesis but also biased the phenotype toward morphology IV.