Previous studies of laboratory strains of budding yeast had shown that when gene copy number is altered experimentally, RNA levels generally scale accordingly. This is true when the copy number of individual genes or entire chromosomes is altered. In a recent study, Hose et al. (2015) reported that this tight correlation between gene copy number and RNA levels is not observed in recently isolated wild Saccharomyces cerevisiae variants. To understand the origins of this proposed difference in gene expression regulation between natural variants and laboratory strains of S. cerevisiae, we evaluated the karyotype and gene expression studies performed by Hose et al. on wild S. cerevisiae strains. In contrast to the results of Hose et al., our reexamination of their data revealed a tight correlation between gene copy number and gene expression. We conclude that widespread dosage compensation occurs neither in laboratory strains nor in natural variants of S. cerevisiae.https://doi.org/10.7554/eLife.10996.001
DNA inside cells is packaged into structures called chromosomes. Different species can have different numbers of chromosomes, but when any cell divides it must allocate the right number of chromosomes to each new cell. If this process goes wrong, cells end up with too many or too few chromosomes. The presence of extra copies of the genes on the additional chromosomes can cause the levels of the proteins encoded by those genes to rise abnormally, which can in turn lead to cell damage and disease.
Proteins are produced using the information in genes via a two-step process. First, the gene’s DNA is copied to create molecules of RNA, and these molecules are then translated into proteins. In many organisms, the presence of extra chromosomes in a cell is matched by a corresponding increase in the RNA molecules encoded by the extra genes. Some organisms, however, counteract this effect through a process called dosage compensation. This process inactivates single genes or whole chromosomes by various means, and ensures that normal levels of RNA are produced, even in the presence of extra genes.
In 2015, researchers from the University of Wisconsin-Madison reported that dosage compensation occurs in wild strains of budding yeast and effectively protects the yeast cells against the harmful effects of having extra chromosomes. However, these findings conflicted with earlier studies of laboratory strains of this yeast, which had reported that RNA levels increased along with gene number.
Torres, Springer and Amon have re-analysed the data published in 2015, and now challenge the findings of the previous study involving the wild yeast strains. The new re-analysis instead showed that, like in laboratory yeast strains, gene number still correlates closely with RNA levels in the wild yeast. This led Torres, Springer and Amon to conclude that, in contrast with the previous report, there is currently no evidence that dosage compensation occurs in wild strains of yeast.
So why do the results of these two studies disagree? Torres, Springer and Amon identified several issues concerning the original analysis made by the researchers from the University of Wisconsin-Madison. For example, some of the strains included in the 2015 study were unstable and were naturally losing the additional chromosomes that they’d acquired. Also, the thresholds set in the analysis to identify dosage compensated genes do not appear to have been stringent enough. Together, the new findings indicate that dosage compensation is a rare event in both wild and laboratory strains of yeast.https://doi.org/10.7554/eLife.10996.002
Losses or gains of whole chromosomes, a condition known as aneuploidy, have a profound impact on cell physiology. Gene expression studies in budding yeast, fission yeast, mammalian cells, and plants revealed that this is due to the fact that changes in gene copy number result in changes in gene expression (Chikashige et al., 2007; Huettel et al., 2008; Pavelka et al., 2010; Sheltzer et al., 2012; Stingele et al., 2012; Torres et al., 2010; 2007). For example, in haploid budding yeast strains harboring single chromosome gains, RNA levels of more than 90% of genes located on the extra chromosome reflect the increased gene copy number (Dephoure et al., 2014; Torres et al., 2007). Only few genes, such as histone and some ribosomal genes defy this trend (Dabeva and Warner, 1987; Gunjan and Verreault, 2003; Libuda and Winston, 2006; Moran et al., 1990; Sutton et al., 2001; Vilardell and Warner, 1997). Given that aneuploidy has such a profound impact on the cell’s transcriptome and proteome it is not surprising that aneuploidy affects virtually all aspects of cell physiology, generally having a negative impact on fitness (Hassold and Jacobs, 1984; Hodgkin, 2005; Huettel et al., 2008; Lindsley et al., 1972; Niwa et al., 2006; Stingele et al., 2012; Torres et al., 2007; Williams et al., 2008).
Aneuploidy not only affects gene expression through changes in gene copy number, the condition also causes transcriptional responses. For example, when chromosome gains or losses lead to a decrease in growth rate, a stereotypic slow-growth transcriptional response known as the environmental stress response (ESR) ensues (Gasch et al., 2000). The ESR is characterized by the down-regulation of growth-promoting genes and the up-regulation of stress response genes and has been reported to occur in response to aneuploidy in many organisms including laboratory yeast strains (Sheltzer et al., 2012).
Changes in gene copy number not only can lead to transcriptional responses but also can elicit dosage compensation, a gene regulatory mechanism that specifically compensates for alterations in gene copy number at the gene expression level. Dosage compensation is best understood in the context of sex chromosome-encoded genes (reviewed in Straub and Becker, 2007). For example in mammals, an RNA-mediated mechanism silences expression of one copy of the X chromosome in females thereby equalizing X chromosome-encoded gene expression between males and females (Lee and Bartolomei, 2013). In Caenorhabditis elegans, gene expression of the two X chromosomes is reduced by half in the hermaphrodite to match the expression of the single X chromosome in males (Meyer, 2010). Dosage compensation can also affect specific loci. The perhaps best known example is the histone locus in budding yeast (Osley and Hereford, 1981). When an extra copy of the HTA1 gene (histone H2A) is introduced into budding yeast, mRNA turnover increases resulting in normal HTA1 transcript levels (Moran et al., 1990; Osley and Hereford, 1981). It is important to note that dosage compensation and transcriptional responses to aneuploidy can have the same effect on a gene. Both can cause the down-regulation of a gene, but the mechanisms are distinct. Transcriptional responses to aneuploidy are elicited by an aneuploid genome affecting a biological pathway and are not restricted to the aneuploid chromosomes but impact expression of genes located throughout the genome. In contrast, dosage compensation specifically alters the expression of a gene whose copy number has been varied and its effects are thus restricted to the aneuploid chromosome.
Experimental evolution studies suggest that selective pressures cause changes in karyotype such as chromosome gains or losses (Dunham et al., 2002; Gresham et al., 2008). However, such aneuploidies are usually transient evolutionary intermediates that, given time, are replaced with more optimal solutions (Yona et al., 2012). A key question that arises from these studies is how prevalent whole chromosome gains and losses are in wild yeast strains and how aneuploidies affect cell physiology. Hose et al. (2015) addressed these questions. They isolated 47 wild yeast strains to identify 12 (26%) that harbored whole chromosome gains and/or losses. The detailed analysis of six of these strains led them to the conclusion that aneuploidies are prevalent, stable and well-tolerated in wild yeast strains. Based on gene expression analyses, they further concluded that tolerance to aneuploidy is caused by dosage compensation mechanisms that buffer gene amplifications thereby protecting cells against the adverse effects of aneuploidy. They reported that gene-dosage compensation functions at >30% of amplified genes.
To understand why dosage compensation mechanisms are rare in laboratory strains of budding yeast, but highly prevalent in wild isolates, we reevaluated the karyotype and gene expression studies performed by Hose et al. (2015). This reexamination revealed that gene copy number and expression are tightly correlated in wild S. cerevisiae strains. We conclude that dosage compensation is a rare occurrence in both, laboratory and natural variants of S. cerevisiae.
Hose et al. (2015) isolated 47 wild yeast variants and determined their karyotypes by inferring the copy number from genome sequencing data using depth of coverage. This analysis showed that 12 of these 47 strains harbor whole chromosome aneuploidies. DNA and RNA sequencing data for 6 of these 12 aneuploid strains were deposited in the NCBI Sequence Read Archive (SRA) under accession SRP047341 and NCBI Gene Omnibus under accession GSE61532 referenced in Hose et al. (2015). Three of these strains harbored one or two single chromosome gains in a diploid background. Strain K9 is a diploid strain carrying an extra copy of chromosomes IX and X (Figure 1A,G), YPS1009 is diploid with an extra copy of chromosome XII (Figure 1B,G), and diploid strain NCYC110 carries two extra copies of chromosome VIII (Figure 1C,G). In addition, Hose et al. (2015) analyzed three strains with high levels of aneuploidy. These strains were strains YJM428, Y2189 and K1 (Figure 1D–G).
We examined the karyotypes and gene expression of these strains and found the aneuploid strains K9, YPS1009 and NCYC110 with low levels of aneuploidy to harbor relatively stable karyotypes (Figure 1A–C). As discussed in more detail below, RNA levels also generally correlated well with DNA levels, with aneuploid chromosomes overall showing a corresponding increase in gene expression (Figure 1A–C). It is, however, noteworthy that strain K9 which harbors extra copies of chromosome IX and X in the Hose et al. (2015) study was previously reported to be trisomic for chromosome IX only, indicating that this strain exhibits some karyotypic instability (Kvitek et al., 2008).
In contrast to the relatively stable strains K9, YPS1009, and NCYC110, a different picture emerged from our analysis of strains YJM428, Y2189, and K1 that harbor complex karyotypes. Based on the presence of non-integer DNA copy number states, we conclude that the described aneuploidies are only present in subpopulations of cells (Figure 1D–F). The comparison between RNA and DNA levels further revealed significant inconsistencies between the two data sets indicating that some strains had changed their karyotypes between the two analyses (e.g. DNA and RNA copy numbers are very different in strains Y2189 and K1; Figure 1E,F). This discrepancy is problematic as Hose et al. (2015) used the standard deviations (SD) of the DNA measurements to establish cutoffs in their RNA data set to identify dosage compensated genes (discussed in detail below).
We also analyzed the karyotypes of the other six aneuploid variants UC5, WE372, T73, Y3, Y6, and CBS7960 that were not characterized in detail by Hose and coworkers (both Figure 1 and Supplementary file 1 in Hose et al., 2015; log2 ratios of normalized DNA copy numbers were provided by A. Gasch). We found that strains T73, which is tetrasomic for chromosome VIII (analyzed below; Figure 4A), and WE372, which is trisomic for chromosome I to harbor stable karyotypes (Figure 2A). However, DNA copy numbers in strains UC5, Y3, Y6, and CBS7960 exhibited non-integer DNA copy number states indicating that the strains are heterogeneous (Figure 2B–E).
Analysis of the karyotypes of the other 35 wild strains (both Figure 1 and Supplementary file 1 in Hose et al. (2015)) revealed that more than half of the strains harbored karyotype profiles consistent with heterogeneity. Importantly, strains K10, YJM308, and Y2209 utilized as the euploid reference in the gene expression analysis of the aneuploid wild strains YJM428, Y2189, K1, and K9 (Figure 2 in Hose et al., 2015) appeared to harbor heterogeneous karyotypes (Figure 2F–H). In particular, strain YJM308 harbors an amplification of chromosome XV and has lost part of chromosome III (Figure 2G). We conclude that only 10.6% (5 out of 47) of the strains analyzed by Hose et al. (2015) harbor relatively stable aneuploidies that are confined to 1 – 2 chromosomes.
As all strains studied by Hose et al. (2015) were derived from single colonies, our finding of significant karyotype heterogeneity indicates that a large fraction of wild yeast strains grown under standard laboratory conditions are unstable. The observed instability and heterogeneity of many wild S. cerevisiae strains makes it likely that the aneuploidies in these wild isolates are a consequence of culturing the natural variants under laboratory conditions to which they may be ill-adapted to, instead of these strains being naturally aneuploid. Caution is therefore warranted when analyzing growth rates, gene expression patterns and phenotypes of such wild yeast strains under laboratory growth conditions.
In our previous studies, we found RNA and DNA levels to be well-correlated in haploid laboratory W303 strains carrying additional chromosomes (Dephoure et al., 2014; Torres et al., 2007). Hose et al. (2015) reported that this coordination between DNA and RNA levels was not evident in wild budding yeast isolates. Their conclusion was based on three analyses. In the first analysis, they characterized six wild aneuploid isolates; in the second, they studied three euploid-aneuploid strain pairs; and in the third analysis, they investigated two sets of strains each comprised of a series of strains with increasing aneuploidies of one particular chromosome. To begin to understand the mechanisms that could have led to the loss of dosage compensation mechanisms in laboratory strains, we reanalyzed the data generated by Hose et al. (2015) using the methods we previously employed to examine the effects of aneuploidy on gene expression in laboratory strains.
Hose et al. (2015) compared mRNA levels with DNA copy number of amplified genes across six aneuploid wild yeast strains called K9, YPS1009, NCYC110, YJM428, Y2189, and K1 and concluded that 38% (838 of 2,204) of amplified genes showed lower expression than predicted by their gene copy number (light blue points in Figure 4A in Hose et al., 2015). We reevaluated their findings. Because of karyotype heterogeneity in strains YJM428, Y2189, and K1 (Figure 1D–F), we did not reanalyze these strains except to determine the false discovery rate discussed in detail below.
Strains YPS1009 and K9 are trisomic for chromosomes XII and IX+X, respectively, while NCYC110 harbors a tetrasomy of chromosome VIII. Our analysis revealed that the expression of all genes on the aneuploid chromosomes increased proportionally with gene copy number (Figure 1A–C, 3A,B). As predicted by a null model with no compensation, we found that the log2 ratios of expression values of genes encoded by the triplicated chromosomes of these strains to fit a normal distribution with a mean value very close to the predicted log2 ratio of 0.58 (mean log2 ratio = 0.55, R2 = 0.99, Figure 3A, middle panel, Figure 3B) for the trisomic strains and a log2 ratio of 1 (mean log2 ratio = 0.95, R2 = 0.97, Figure 3A, right panel) for the tetrasomic chromosome. No skewness in the distributions - more compensating or exacerbating - was noted as would be expected if a large fraction of the genes encoded on the aneuploid chromosome were dosage compensated (skewness = 0.02 (3n) and 0.07 (4n); Figure 3A). The distribution of log2 ratios of expression values of genes encoded by euploid chromosomes also fit a normal distribution with the predicted log2 ratio of 0 (mean log2 ratio = 0.00, R2 = 0.99, Figure 3A left panel). These data are very much in line with what is observed in aneuploid laboratory strains. RNA quantification of two disomic W303 strains (disomes V and XVI) showed that the log2 ratios of expression values of genes encoded by the duplicated chromosomes fit a normal distribution with a mean value very close to the predicted log2 ratio of 1 (mean log2 ratio = 1.03, R2 = 0.98, Figure 3C).
To determine how many genes were potentially subject to dosage compensation, we used 2 SD from the means of the log2 ratios of each amplified chromosome and found that between 0% (0 gene in NCYC110) and 3% (19 genes in K9) of amplified genes showed values lower than expected (Table 1). Importantly, a similar number of genes was found to exhibit higher than expected expression (between 1% in YPS1009 and 2% in NCYC110, Table 1). Using the same cutoff on the euploid chromosomes, we found between 0.1% (7 genes in NCYC110) and 3% (153 genes in K1) genes with values lower than expected. The nature of the distributions of gene expression patterns (normal distribution with expected means) and these analyses are inconsistent with high levels of dosage compensation occurring in wild yeast strains. Instead, they indicate that gene expression proportionally increases with copy number without signs of dosage compensation in wild yeast strains. The fact that the euploid chromosomes encode the same proportion of up and downregulated genes as the aneuploid chromosomes further indicates that any effects on gene expression seen in these strains are likely to be the consequence of measurement noise or a transcriptional response elicited by the aneuploid state rather than dosage compensation.
To further characterize dosage compensation in wild variants Hose et al. (2015) generated a panel of isogenic euploid and aneuploid strain pairs. They isolated a disomic strain for chromosome VIII (YPS163-chr VIII-2n) of the euploid strain YPS163, and euploid versions of strain T73, which is tetrasomic for chromosome VIII (T73-chrVIII-4n) and of strain YJM428, which is tetrasomic for chromosome XVI (YJM428-chrXVI-4n). They then determined DNA copy number state and gene expression levels in these strains and concluded that between 11 and 36% of genes were expressed at lower than expected levels, that is, they were dosage compensated.
We compared the average chromosome copy number in the three aneuploid strains with the average RNA copy number in these strains and found that RNA levels proportionally increased with DNA copy number (Figure 4A,B). The aneuploidies in the three strains represent duplications. We were, therefore, able to combine the duplicated values of the DNA and RNA copy of all the three strains. The 941 duplicated genes showed a mean log2 ratio of 1.02 (SD = 0.29, R2 = 0.99) for DNA copy number and a nearly identical mean log2 ratio (mean = 0.97; SD = 0.36, R2 = 0.99) for RNA copy number (Figure 4C). Furthermore, the distribution of expression values fit a normal distribution and was indistinguishable from the distribution of the gene expression values of genes encoded by the euploid chromosomes. The standard deviations of the RNA distributions were similar for euploid and aneuploid chromosomes (Figure 4C bottom graphs) and each distribution showed skewness of 0.01 and 0.02, respectively. These observations indicate that the variance of the euploid genes is the same as that of the aneuploid genes. If dosage compensation were to occur, variance and skewness, would be different between genes encoded by euploid and aneuploid chromosomes. Lastly, using 2 SD as cutoff to find potential dosage compensated genes, we identified a small number of outliers. Importantly, the number of up and downregulated outliers was similar (Figure 5). We conclude that RNA levels correlate well with DNA copy number in aneuploid strains YPS163, T73, and YJM428.
The third set of strains that Hose et al. (2015) analyzed was comprised of two series of yeast strains that carry increasing numbers of a specific chromosome. Starting with strain YPS1009, which carries three copies of chromosome XII (YPS1090-chrXII-3n), Hose et al. (2015) derived a euploid strain (YPS1009-chrXII-2n) and a strain that is tetrasomic for chromosome XII (YPS1009-chr XII-4n; Figure 6A,B). Using strain NCYC110, which carries four copies of chromosome VIII (NCYC110-chrVIII-4n), they isolated a strain trisomic for chromosome VIII (NCYC110-chrVIII-3n) and a diploid strain (NCYC110-chrVIII-2n; Figure 6A,B). They then determined DNA copy number state and gene expression levels in these strain series and concluded that 11% of genes encoded on chromosome VIII and 29% of genes encoded on chromosome XII were dosage compensated.
We found the gene expression distribution of genes located on euploid and aneuploid chromosomes to fit normal distributions without any skewness (Figure 6C,D). The two trisomic strains YPS1009-chrXII-3n and NCYC110-chrVIII-3n together harbored 776 triplicated genes. Their averaged log2 ratio of DNA copy number was 0.57 (SD = 0.15, R2 = 0.99) and 0.60 (SD = 0.53, R2 = 0.97) for RNA copy number (Figure 6C). A similar coordination between DNA and RNA copy number was observed in the tetrasomic strains. The mean log2 ratio of DNA copy number of genes located on the tetrasomic chromosome was 0.99 (SD = 0.16, R2 = 0.99), the mean mRNA expression of genes located on the tetrasomic chromosome was 0.99 (SD = 0.62, R2 = 0.96; Figure 6D). Importantly, the distributions of DNA and mRNA copy number were similar for genes located on euploid, trisomic and tetrasomic chromosomes, with similar SDs and no evidence of skewness (skewness varied between 0.00 and 0.03).
In summary, we were not been able to detect dosage compensation in the strains described in Hose et al. (2015). RNA levels of genes encoded by the aneuploid chromosomes are normally distributed with the expected or close to expected mean. No difference was observed between the number of down-regulated genes located on aneuploid and euploid chromosomes. Furthermore, no skewness was observed for any of the distributions. Figure 7 shows how distributions exhibit negative values of skewness when dosage compensation occurs. Importantly, in a previous study, we were able to detect attenuation in the expression of certain genes in aneuploid yeast strains using the method employed here. In Dephoure et al. (2014), we examined the proteomes of haploid disomic laboratory yeast strains and found that production of ribosomal proteins encoded on disomic chromosomes is significantly attenuated causing the distributions to exhibit negative skewness (Dephoure et al., 2014).
We identified two problems in their data analysis. The first regards data normalization. The ratios are off by a factor of log2 = 0.1–0.2 (normalized data utilized in Hose et al. (2015) were kindly provided by A. Gasch). Most normalization protocols do not take into account that aneuploid strains harbor a different number of gene copies compared to euploid strains. When this is not manually corrected, data are shifted by a factor that depends on the degree of aneuploidy and results in incorrect values as shown in Figure 8A. The degree by which the data used for analysis by Hose et al. (2015) deviate from the correctly normalized expression values is of the same magnitude as some of the cutoffs used to define dosage compensated genes (detailed next).
The second problem with the data analysis concerns cutoffs used to define dosage compensated genes. To establish cutoffs for designating whether a gene is dosage compensated or not Hose et al. (2015) used the SD of the DNA measurements, which ranged between 0.1 and 0.45 (Table 1 column 11, data kindly provided by A. Gasch) as cutoffs for the RNA measurements (Figure 4 in Hose et al., 2015). Genes whose expression deviated by the DNA SD value from the expected RNA expression level were considered dosage compensated. This is not the correct cutoff tool because the DNA copy number measurements are less variable than mRNA measurements. As seen in Figure 8B, transcript levels can vary by several orders of magnitude depending on the expression levels of a particular gene. Therefore, the distributions of relative RNA changes will show bigger SDs than gene copy number distributions. Indeed, the RNA measurements conducted by Hose et al. (2015) show SDs between 0.53 and 1.01 (Table 1, column 6). Employing the SD derived from the DNA measurements, which are fairly lower (Table 1, column 8), will therefore not identify genes that are dosage compensated in a statistically significant manner (see false discovery rate discussion below). This is of particular importance as genes identified in Figure 4 of Hose et al. (2015) as dosage compensated were included in a group of 245 dosage compensated genes used to establish GO term enrichments among dosage compensated genes.
To determine how Hose et al. (2015) identified 838 of 2204 genes as dosage compensated we re-evaluated their analysis. Figure 4A in Hose et al. (2015) displays an unusual behavior. The null model shown by the diagonal of equal RNA and DNA in this figure did not bisect the blue (compensated) and magenta (exacerbated) points. Instead, the vast majority of points below this line were considered compensated while the vast majority of points above this line were considered not exacerbated. This suggests that there could be a high number of false positives amongst the 838 genes determined to be dosage compensated.
To address this possibility, we used two methods to determine the false discovery rate. First, we scrambled the data by randomly permuting the RNA/DNA ratio between genes. We did this independently for each replicate. This preserves the RNA/DNA ratios but unlinks the values from their replicate measurements and genes. Then, we used the same effective significance cutoffs used by Hose et al. (2015) to determine the number of dosage compensated genes (see Materials and methods). As this is a randomized dataset, genes identified by this method are noise and can be used to determine the number of genes the analysis method would find just by chance. Based on 10,000 randomizations, we determined that on average, 779 genes would have passed the threshold method used by Hose et al. (2015) by chance. This yields a false discovery rate (FDR) of 92.9%. This high false discovery rate was also seen at much lower cutoffs. The FDR was between 93 and 100% at cutoffs from 0.1 STD to 2 STDs.
Second, we calculated the average SD for each RNA sequencing measurement. As the DNA measurements for each strain were not reported independently, we calculated the average chromosome-wide DNA error from all the sequencing data that were deposited and used the lowest of these as an estimate for all analyses. We combined these errors together (square root of summed squares of the two composite noises) to give a measurement noise distribution for the experiment. We then randomly sampled from a normal distribution where the SD for this normal distribution was randomly sampled from the measurement noise distribution. Using this method, we found that on average, 754 genes would have passed the effective threshold used by Hose et al. (2015). This corresponds to a false discovery rate of 89%; this value is likely a small underestimation of FDR given our method for estimation of DNA error. We conclude that both methods that we applied to determine false discovery rate strongly suggest that only a handful, at most ~70 genes or <3%, are actually dosage compensated. These results are completely in line with previous findings from laboratory strains (Springer et al., 2010; Dephoure et al., 2014; Torres et al., 2007).
In a second approach to identify dosage compensated genes, Hose et al. (2015) defined genes to be dosage compensated when the RNA levels did not increase with DNA copy number in their YPS1009 (2N, 2N+1 chromosome XII, 2N+2 chromosomes XII) and NCYC110 (2N, 2N+1 chromosome VIII, 2N+2 chromosomes VIII) ploidy series. For this, they developed a mixture of linear regression (MLR) model to classify genes based on the slope and intercept of the RNA-gene copy relationships. When RNA levels did not increase proportionately as DNA copy increased as evidenced by slopes lower than 1 in the MLR model, a gene was classified as dosage compensated and categorizes as Class 3a in Table 1 in Hose et al. (2015). Thirty genes on chromosome VIII and 142 genes on chromosome XII were identified as dosage compensated through this method. This method of identifying dosage compensated genes is problematic in several ways. First, because there are only three data points per analysis, a single deviating data point can have a significant impact on the slope. For example, a gene with values of log2 ratio = 0.3, 0.6 and 0.8 representing, two, three, and four copies, respectively, will perfectly fit a straight line with the slope of 0.5 and hence would be classified as dosage compensated according to the criteria in Hose et al. (2015) even though none of the three data points significantly deviates from the mean value given a SD of 0.3 or higher (Figure 8C, Table 2). In fact, the majority (103 of 172) of class 3a genes (Table 1 and Supplemental File 3 in Hose et al. (2015)) fit the MLR model with slopes of 0.5 or higher indicating that their gene expression increases with copy number.
Because of these considerations, we reanalyzed the dosage compensation in chromosomes VIII and XII by two methods. In the first, we calculated the mean and standard deviations for each of the biological replicates in the NCYC110 and YPS1009 strain series and found that only two genes on aneuploid chromosome VIII and seven genes on aneuploid chromosome XII show log2 ratios 1 SD lower than the mean in three biological replicates and were reproducibly lower when present in 3 or 4 copies. Not a single gene passed the cutoff of 2 SD below the mean. We conclude that for the majority of genes only one of the two data points supports the conclusion that a gene is expressed at lower than the expected value, calling into question that the genes identified by this approach are indeed dosage compensated.
In a second approach, we defined the false discovery rate (not determined by Hose et al., 2015) to determine whether the genes identified as dosage compensated were statistically distinguishable from noise. Using the same subset of genes that Hose et al. (2015) examined, we calculated a slope based on the nine RNA measurements and matching DNA measurements (three replicates of three strains) for both YPS1009 and NCYC110. Using the genes identified as dosage compensated by Hose et al. (2015), we determined the effective cutoff of their MLR method (see Materials and methods). We then randomly permuted the positions of the RNA and DNA data and recalculated the slopes for each gene. From this analysis we determined the false discovery rate was within error of 100%. We conclude that there is no significant dosage compensation in these aneuploid series.
Our analyses indicate that a large fraction of wild S. cerevisiae strains are unstable and heterogeneous when grown under laboratory conditions. This result suggests that at least some wild S. cerevisiae strains may not be naturally aneuploid but could become aneuploid due to an adaptive response to laboratory growth conditions. Reevaluation of the DNA and RNA copy number data generated by Hose et al. (2015) further indicates that dosage compensation is rare in both wild and laboratory strain of S. cerevisiae. Both types of strains lack mechanisms that allow them to attenuate gene expression in response to gene copy number alterations. We conclude that wild variants of S. cerevisiae do not have mechanisms in place that protect them from changes in gene copy number. Their regulation of gene expression is thus the same as that of laboratory strains of budding yeast.
We consider any chromosome whose copy number was significantly different from an integral value to be heterogeneous. To determine which chromosomes were significantly different than the nearest integer value, we used a one sample t test using the copy number of each gene on the chromosome as the input which compares a distribution of values to an expected value and then corrected for multiple hypothesis testing. In strain YJM428, the expected value for chromosome III is 2 and the expected value for chromosome XV is 2. In strain Y2189, the expected value for chromosome I is 4, and for chromosomes IX and X is 3. In strain K1, the expected value for chromosomes I and VI is 2. In strains Y3 and Y6, the expected value for chromosome VIII is 3. In strain UC5, the expected value for chromosome I is 2. In strain CBS7960, the expected value for chromosomes I and III is 1. In strain WE372, the expected value for chromosome I is 3.
To avoid any discrepancies in data processing, Hose and coworkers kindly provided all the relative log2 ratios of the relative DNA copy number for all 47 wild strains and for the different panel of isogenic strains. In addition, Hose and coworkers kindly provided all gene expression data utilized in their manuscript. In addition, genome sequences for 16 distinct karyotypes (eight aneuploid and eight euploids) could be obtained from the NIH Sequence Read Archive (SRA) under accession SRP047341. Gene expression data could also be obtained from NIH GEO under accession GSE61532.
Log2 ratios provided by Hose et al. (2015) were normalized by centering the euploid chromosome ratios to 0. This was accomplished by calculating the mean of the log2 ratios of non-duplicated genes and subtracting this factor from all data points. Chromosome copy numbers were calculated by taking the average copy number of all genes within each chromosome. For diploids copy number equals 2*2^(log2 ratio euploid vs. aneuploid), for haploids copy number equals 1*2^(log2 ratio euploid vs. aneuploid).
Gene expression data of each aneuploid strain were compared to their reference genome as described in Hose et al. (2015). Log2 ratios of aneuploid/euploid genes were normalized to the euploid chromosomes as described above for DNA. RNA copy numbers per chromosome were calculated by averaging gene copy number of the genes within each chromosome.
To analyze the distributions of euploid or amplified genes, DNA and RNA log2 ratios of aneuploid/euploid we first calculated the distribution of the log2 ratios binned by a value of 0.2. The frequency distributions were plotted and the data were fit to normal distribution utilizing PRISM software. Means, medians, SD, skewness and R2 of the fits are reported in each figure. Gene expression data for disomes V and XVI (Figure 3C) were previously published in Dephoure et al. (2014). Gene expression profiles were visualized with Treeview.
First, we needed to determine an effective cutoff to classify a gene as dosage compensated. We calculated the RNA/DNA ratio or slope of RNA versus DNA for all genes and binned the data (0.1 width bins in log space). For each bin, we then determined the percent of genes in that bin that were classified as dosage compensated by Hose et al. (2015). Second, we randomized the data. We took the processed data (RNA/DNA) or raw data (RNA and DNA measurements, for slope analysis) and randomized the position of this information in the dataset. This decouples all the replicate measurements. Hose et al. (2015) supplied us with the RNA and DNA values for each gene and for each strain that they used to assess dosage compensation. Starting with this table as our input for randomization, we then calculated the RNA/DNA ratio for every replicate of every gene. We then permuted each column of the table (the replicates) independently and then calculated the average dosage compensation per 'gene' by averaging across the replicates; this is identical to how Hose et al. (2015) calculated dosage compensation from the unpermuted table.
If a subset of genes on a chromosome are compensated, as reported by Hose et al. (2015), their average RNA/DNA ratios should appear as outliers on a distribution of RNA/DNA for a whole chromosome. Randomization of the RNA/DNA values before averaging will eliminate most of these outliers, as the outlying values will be most often average with non-outlying values; hence one should observe fewer genes that have large deviations from the mean. To assess this, we took all genes that Hose et al. (2015) had reported as dosage compensated. We took the distribution of RNA/DNA for these compensated genes and called this the observed or reported compensated distribution. The existence of true compensators would lead to significantly more genes in the compensated distribution than in the randomized distribution for a given dosage compensation range. This was not the case. Instead, the distributions were indistinguishable suggesting that the vast majority of genes reported as dosage compensated by Hose et al. (2015) is noise.
Before calculating the false positive rate, one minor correction was needed. As the cutoff for calling a gene dosage compensated in Hose et al. (2015) did not take into account all the measurement noise we had to determine the effective cutoff used by Hose et al. (2015).
While the RNA values were reported for each of the replicates, the DNA value was only reported as the mean of all measurements. This meant that calculating a SD based on the RNA/DNA ratios reported by Hose et al. (2015) would underestimate the true error of the measurement of dosage compensation and hence would give an artificially low false discovery rate. We turned to the sequencing data deposited with the paper, but the DNA data was only deposited for a subset of the strains. We therefore calculated the per gene DNA copy number error from the strains from which the replicates were deposited. From this, we found that the average standard deviation in DNA copy number was approximately 10%. For each dosage compensation value, we therefore randomly sampled from a normal distribution with a SD of 10% and modified the dosage compensation value by this percentage (square root of squared sum of errors).
To determine the false discovery rate of this compensated distribution we compared the distribution of dosage compensated genes to the distribution of dosage compensation data from a distribution obtained by randomly sampling from a normal distribution with errors that came from a table of measurement errors. If we did not include the DNA error in this table of measurement estimates, the FDR rates dropped by about 10%. Thus, the vast majority of dosage compensated genes are most likely false positives irrespective of whether a correction was included or not.
Gene expression and distribution of Swi6 in partial aneuploids of the fission yeast schizosaccharomyces pombeCell Structure and Function 32:149–161.https://doi.org/10.1247/csf.07036
The yeast ribosomal protein L32 and its geneThe Journal of Biological Chemistry 262:16055–16059.
Quantitative proteomic analysis reveals posttranslational responses to aneuploidy in yeasteLife, 3, 10.7554/eLife.03023.
Characteristic genome rearrangements in experimental evolution of saccharomyces cerevisiaeProceedings of the National Academy of Sciences of the United States of America 99:16144–16149.https://doi.org/10.1073/pnas.242624799
Genomic expression programs in the response of yeast cells to environmental changesMolecular Biology of the Cell 11:4241–4257.https://doi.org/10.1091/mbc.11.12.4241
Trisomy in manAnnual Review of Genetics 18:69–97.https://doi.org/10.1146/annurev.ge.18.120184.000441
Dosage compensation can buffer copy-number variation in wild yeasteLife, 4, 10.7554/eLife.05462.
Segmental aneuploidy and the genetic gross structure of the drosophila genomeGenetics 71:157–184.
A yeast H2A-H2B promoter can be regulated by changes in histone gene copy numberGenes & Development 4:752–763.https://doi.org/10.1101/gad.4.5.752
A general lack of compensation for gene dosage in yeastMolecular Systems Biology 6:368.https://doi.org/10.1038/msb.2010.19
Global analysis of genome, transcriptome and proteome reveals the response to aneuploidy in human cellsMolecular Systems Biology, 8, 10.1038/msb.2012.40.
Dosage compensation: the beginning and end of generalizationNature Reviews Genetics 8:47–57.https://doi.org/10.1038/nrg2013
Yeast ASF1 protein is required for cell cycle regulation of histone gene transcriptionGenetics 158:587–596.
Ribosomal protein L32 of saccharomyces cerevisiae influences both the splicing of its own transcript and the processing of rRNAMolecular and Cellular Biology 17:1959–1965.https://doi.org/10.1128/MCB.17.4.1959
Chromosomal duplication is a transient evolutionary solution to stressProceedings of the National Academy of Sciences of the United States of America 109:21010–21015.https://doi.org/10.1073/pnas.1211150109
Duncan T OdomReviewing Editor; University of Cambridge / Cancer Research UK, United Kingdom
In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.
Thank you for submitting your work entitled "No evidence of widespread dosage compensation in wild S. cerevisiae strains" for consideration by eLife. Your article has been reviewed by four peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Randy Schekman as the Senior Editor.
The reviewers have discussed the reviews with one another and the Reviewing editor has drafted this decision to help you prepare a revised submission.
All the reviewers, as well as the Reviewing Editor, were entirely convinced that a revised and carefully re-written version of your manuscript should be publicly released in eLife. Major concerns centered around the overall tone being too aggressive, many analyses possibly being too stringent, and an unbalanced consideration of all possible explanations, particularly given how nascent our understanding as a field is of dosage compensation. It was also discussed that the authors need to describe the methods used in a more detailed manner.
All four reviewers brought up a highly compelling set of additional analysis and interpretation concerns to be addressed, and are therefore appended to this letter. A revision suitable for acceptance will not require any additional experimental data; however, to publish will require new analyses, corrected figures, a balanced Discussion, and substantial reworking of the text and argument structure.
1) Methods: Every analysis should be described in a detailed Methods section, with subsections clearly annotated and referenced in the main text by occurrence in the figures and in the Results section.
2) Soften tone: As one reviewer eloquently said: "The tone throughout the article is rather strident and bordering on confrontational. It would be wise for the sake of rational open discussion to soften some of the more forceful statements." The overly confident or aggressive text must be adjusted before acceptance. In some sections of the manuscript, the authors’ interpretations were simplified or inappropriate, and exceptionally strict analysis cut offs may bias Torres' interpretations. Additional tonal comments found in the reviews below should not be neglected. Importantly, the title and section headers will have to be toned down as well.
3) Restructure: A careful introduction to what is meant by 'dosage compensation' must be laid out clearly in the Introduction. Sections should start with a paragraph laying out the reasoning that leads to why each analysis was performed (see for instance Reviewer 3 point 5). See also the Reviewing Editor’s comments below.
4) Aneuploidy versus instability: Many reviewers noted that Torres' equating aneuploidy and instability was not appropriate; this must be corrected throughout.
The Torres et al. manuscript re-examines the Hose et al. (eLife 2015) data to evaluate dosage compensation in wild yeast strains. At the heart of the Torres manuscript is whether the data of Hose et al. truly support dosage compensation. Their analysis is thorough, careful, and convincing, but the methods need to be better described.
1) They argue that the presence of non-integer DNA copy number states implies the strains are highly heterogeneous and unstable. But this is methodologically poorly described. The methods argue that chromosome copy number is calculated by taking the average number of all genes within each chromosome. So one assumes that non-integer is an average substantially different from an integer, but what variance is acceptable? The copy number plots (Figures 1, 2, 4) show considerable variability from point to point, depending on strain; for example Figure 1 NCYC110 is tightly distributed around 2 DNA copies whereas YPS1009 seems more variable. Is this variability taken into account? Is it influenced by depth? Means are more sensitive to outliers than the median – would you see a similar result using the median, or (in the case of numerous aneuploid chromosomes) is the median too skewed upward?
Furthermore, it is unclear that they can infer instability rather than simply heterogeneity, the exception being the one obvious case (K1) where the DNA and RNA copy numbers vary. That said, a devil's advocate argument is that K1 shows differences between DNA and RNA assays (specifically on chromosome VI) because of "dosage compensation". While highly unlikely (seeing how there are other aneuploid chromosomes that are not compensated), it seems an important point given the final overall findings of the paper.
2) The methodologies for assessing false positives are poorly described. Both the randomization and the error/noise distribution approach are only vaguely described. When the ratios were permuted relative to the gene list, what was then done to assess dosage compensation – they say "the method of Hose et al. (see Methods)” but in the Methods it is not described. (The false discovery rate analysis is described but that is used later in the paper). What does "not reported independently" (in this same section) mean? How does using the "lowest average chromosome-wide DNA error" bias your result? How exactly were errors combined?
3) The heart of the paper is the issue of how to assess dosage compensation. They miss an opportunity in the Discussion to discuss both the key points of "dosage compensation" as a concept (distinct from what they refer to as "transcriptional response to aneuploidy”). Likewise there are issues when assessing aneuploidy numbers – the constant need for a frame of reference and this issue of expectation (i.e. RNA copy numbers will vary relative to DNA simply because of expression levels being variable). This would help the reader to understand why skew is expected in the distribution if dosage compensation is present.
Amon and coworkers re-analyze data previously obtained by Hose et al. (2015). This re-evaluation identifies several flaws in the original analysis and yields a completely opposite conclusion; namely that there is no sign of widespread dosage compensation in (aneuploid/polyploid) feral S. cerevisiae strains. As far as I can see, the re-analysis is technically sound and I especially commend the authors for the permutation test applied to the RNA/DNA ratio for each gene.
I therefore recommend publication of this paper in eLife, even though I do have a few suggestions (below).
1) Most importantly, I would suggest giving A. Gash and her team the opportunity to co-publish their response to this new paper together with the publication of the new paper. I think it would be interesting to know what the original authors think about this re-evaluation, and it also seems courteous to offer this possibility to Dr. Gash. I also feel that it is important to give the opportunity to have the response published together (at the same time) with this new paper.
2) It would be interesting to expand the short Discussion section to further highlight that non-laboratory S. cerevisiae strains harbour natural copy number variants. The sentence "The fact that strains YJM428, Y2189, K1, UC5, Y3, Y6 and CBS7960 are unstable also means that these strains are less fit than euploid strains" might benefit from a more elaborate discussion to help the reader understand the rationale behind this argument, and to discuss literature showing that several experimental evolution experiments have identified transient aneuploidies as a common but potentially suboptimal solution to overcome harsh conditions.
3) The authors use non-integer changes in DNA read depth as evidence that a given strain is unstable. I see the logic behind this reasoning, but since this result directly contradicts previously published results, and since it is of great importance to the broad community working with (industrial and feral) S. cerevisiae yeasts, one has to be sure that there are no (cryptic) technical reasons for the non-integer changes, even if this scenario appears unlikely. If the strains are indeed so unstable, one would expect that analyses (whole-genome-sequencing) of different single colonies yield (very) different outcomes. Is this the case? Another, more elegant approach would be to investigate CNVs in single cells in a population, but this might be technically more challenging…
This manuscript is a report that is a rebuttal directed at a prior eLife paper by Gasch and her group (Hose et al, 2015). The paper by Gasch reports dosage compensation at the transcriptional level in aneuploid wild yeast. The main focus of the reviewed manuscript is to demonstrate that the observed dosage compensation reflects shortcomings of the analysis. Science should foster open disputation of controversial topics. As the field of aneuploidy is rather new and evolving, it is essential to address the disagreements early on and thoroughly. Torres et al. specifically mention the following issues: the used strains may be unstable, the normalization is suboptimal there is no test for significance of the number of dosage compensated genes and its correction for multiple testing the standard deviation of the DNA sequencing data is used as a cut off for determining dosage compensated genes on RNA sequencing data and the three-point analysis to identify dosage compensated genes is too limited and therefore prone to noise-related errors. Since the analysis by Hose et al. has some flaws, this referee feels that the manuscript by Torres et al. should be made public. However, there are several important points that should be addressed.
1) First, the authors make the point that the wild aneuploid yeast might be chromosomally unstable. Although this is probably true, it is not possible to conclude based on the presented data – the authors say as an example that the RNA and DNA levels are very different for K1 strain in figure 1A in Hose et al.; however, this reviewer can see only DNA levels in 1A. The fact that there are non-integer DNA copy numbers means strictly speaking only that the strain is heterogeneous.
2) In the second paragraph of "Evaluation of the analysis methods employed by Hose et al.", the authors state: "Most normalization protocols do not take into account that aneuploid strains harbor a different total number of genes than euploid strains". This statement by Torres et al. is incorrect. If something, they contain more copies of genes. Also, Torres et al. mention that the data has to be manually corrected, however, there is no description how they envision to do this nor is it obvious whether they performed it on the data from Hose et al. or not. Here, a description should be added at least in the Methods description. Next, Torres et al. state that "the data used for analysis by Hose et al. (2015) deviate from the actual expression values..."; here the "actual expression values" should be those that Torres et al. calculated. In fact, there is no way to know the "actual expression values"; the approach is to just try to normalize the data and hope that it reflects the reality.
3) To straighten their point, the authors should show at least one volcano plot of the 838 compensated genes (identified by Hose et al.) and their calculated FDRs.
4) It would be useful for the reader if the authors would add the 2SD RNA and 2SD DNA to Table 1 for comparison of the variance and further substantiate the incorrect use of the 2SD DNA as a cutoff in Hose et al.
5) For the understanding of the discussed issues it might be useful to restructure the manuscript in a way that each of the three paragraphs showing that there is no dosage compensation includes the corresponding critical points of the analysis by Hose et al. For each of the three subchapters first the criticism should be raised, followed by their re-analysis and their own analysis. It feels much more logical to explain first the criticism and then show the new results than to list first everything that is wrong and only then show the reasons.
This manuscript is a very thorough reanalysis of a recent paper on aneuploidy and wild strains recently published in eLife (Hose et al.). This paper reaches quite different conclusions than Hose et al. and instead shows a lack of evidence for what Hose et al. call dosage compensation. I found the Torres et al. analysis pretty compelling, and I was rather surprised that the original paper did not include similar analyses. Some of the identified mistakes include errors in normalization, apparent copy number variants in the strains used as euploid controls, and discrepancies between RNA and DNA for some strains. Overall I was convinced that there are significant issues with the Hose paper that were masked by the data presentation and analysis methods used.
I was particularly convinced by the demonstration that the distribution of gene expression of genes on euploid and amplified chromosomes was identical. One could argue whether the distributions of DNA and RNA should be compared (even given the vastly different dynamic ranges), or whether the exact cutoffs employed are too stringent or not stringent enough, but any model of "dosage compensation" seems like it should have genes behaving differently when they are at elevated dosage. This does not appear to be the case.
That said, I did find some of the reanalysis a little too demanding: buffering could still be present but below the 2 SD thresholds used throughout. That's one difficulty of working with single increment dosage effects, particularly in diploids: the fold changes expected are far less than the cutoffs generally employed for expression analysis. While in bulk it's clear that the average expression change scales with copy number, individual genes may not. The MLR analysis from the dosage series strains seems like the best way to detect these potentially subtle effects, though the false positive analysis was fairly convincing that even this analysis fell short in Hose et al.
To make the paper even more compelling, I would like to see some positive controls. That is, what would genes look like that truly do have some buffering against dosage? Can they ever be detected using the methods of either this paper or the Hose et al. one? The Introduction mentions ribosomal genes and histones, for example, so it would be interesting to see how they behave. It could be that detecting such subtle effects is currently beyond the abilities of current RNA measurement technologies.
In addition, the main argument of the Hose paper seems to be a difference between lab strains and natural isolates. A more direct comparison between the data from different strains in this paper would help compare these directly and demonstrate whether they are in fact showing identical patterns.
Reviewing Editor's comments:
The following editorial revisions are single examples of the type that must be made to the manuscript's tone, listed in order of occurrence in the manuscript. Re-review will focus on a detailed list of revisions, if edits are still needed.
New title: “No current evidence for widespread dosage compensation in yeast”.
“mostly lead to an according change in gene expression.” This is a nonsensical sentence, likely a typo.
“gene expression is not observed in wild…” to "gene expression can be violated in wild…".
“dosage compensation occurs neither in laboratory strains nor natural…” to "dosage compensation remains unproven in laboratory strains and natural variants."
All these should be re-worded to report results, not conclusions. For instance:
"Many wild yeast strains have unstable karyotypes." As the reviewers indicated, this is just one of the potential interpretations of the CNV data.
"No dosage compensation in…" (all sections). These are extrapolative and interpretive statements that should be toned down considerably.
Section titled "No dosage compensation in wild isolated YJM428…" should probably be split into two sections, based on the fact that the six yeast lines listed are explored as falling under two different explanations.https://doi.org/10.7554/eLife.10996.017
- Angelika Amon
- Eduardo M Torres
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
We are grateful to Audrey Gasch for providing data and analysis methods. This work was supported by the Richard and Susan Smith Family Foundation and the Searle Scholars Program to ET, and by the National Institute of Health (GM056800) to AA. AA is also an investigator of the Howard Hughes Medical Institute.
- Duncan T Odom, Reviewing Editor, University of Cambridge / Cancer Research UK, United Kingdom
- Received: August 19, 2015
- Accepted: December 29, 2015
- Version of Record published: March 7, 2016 (version 1)
© 2016, Torres et al.
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.
Downloads (link to download the article as PDF)
Download citations (links to download the citations from this article in formats compatible with various reference manager tools)
Open citations (links to open the citations from this article in various online reference manager services)
In our prior work by Hose et al., we performed a genome-sequencing survey and reported that aneuploidy was frequently observed in wild strains of S. cerevisiae. We also profiled transcriptome abundance in naturally aneuploid isolates compared to isogenic euploid controls and found that 10–30% of amplified genes, depending on the strain and affected chromosome, show lower-than-expected expression compared to gene copy number. In Hose et al., we argued that this gene group is enriched for genes subject to one or more modes of dosage compensation, where mRNA abundance is decreased in response to higher dosage of that gene. A recent manuscript by Torres et al. refutes our prior work. Here, we provide a response to Torres et al., along with additional analysis and controls to support our original conclusions. We maintain that aneuploidy is well tolerated in the wild strains of S. cerevisiae that we studied and that the group of genes enriched for those subject to dosage compensation show unique evolutionary signatures.