Different mechanisms of adaptation have different timescales. Epigenetic changes are often rapid and reversible, while most genetic changes have nearly negligible rates of reversion (1). This poses a challenge for genetic adaptation to transient conditions such as drug treatment; mutations that confer drug resistance are often deleterious in the absence of drug, and the second-site suppressor mutations are required to restore fitness (2, 3). Preexisting tandem repeats (satellite DNA) undergo frequent expansion and contraction (4⇓–6). While repeats are rare inside of most coding sequences and functional elements, there is some evidence for conserved repetitive regions that undergo expansion and contraction to regulate protein functions or expression (6⇓–8). RNA interference– or Chromatin-based epigenetic states have been associated with transient drug resistance in fungi (9) and cancer cells (10, 11), and transient resistant states have been characterized by differences in organelle state, growth rate, and gene expression in budding yeast (12, 13). In bacteria and in fungi, copy-number gain and subsequent loss can result in reversible drug resistance (14⇓⇓⇓–18). However, all genetic systems developed so far for studying unstable genotypes rely on reporter genes and thus investigate only one genetic locus and only one type of genetic change.
Unbiased, next-generation sequencing-based approaches could give a more global view, allowing us to understand the rules that govern unstable genotypes at a genome-wide scale. However, genetic changes with high rates of reversion tend to remain subclonal (19⇓–21), and it is challenging to distinguish most types of low-frequency mutations from sequencing errors (22), especially in complex genomes with large amount of repetitive DNA or de novo duplicated genes. Thus, fast-growing organisms with relatively small and simple genomes are particularly well suited for determining whether transient mutations exist, for the genome-wide characterization of such mutations, and for identification of the underlying mechanisms.
Microhomology-Mediated Tandem Duplications in Specific Genes Caused Reversible Phenotypes in Schizosaccharomyces pombe.
To discover transient drug-resistance mechanisms in a eukaryote, we performed a genetic screen in the fission yeast Schizosaccharomyces pombe for spontaneous mutants that are reversibly resistant to rapamycin plus caffeine (caffeine is required for rapamycin to inhibit growth in S. pombe) (23) (Fig. 1A). We plated 107 cells from each of two independent wild-type strains to YE5S+rapamycin+caffeine plates and obtained 173 drug-resistant colonies, 14 (7%) of which exhibited reversible drug resistance following serial passage in no-drug media (Fig. 1 B and C). In contrast, resistance for deletion mutants such as gaf1Δ (24) is irreversible, suggesting the existence of a type of genetic or epigenetic alteration allowing for reversible drug resistance in the newly isolated strains (Fig. 1 B and C).
We used genetic linkage mapping and whole-genome sequencing to identify the molecular basis of reversible rapamycin+caffeine resistance. We identified two linkage groups (SI Appendix, Fig. S1A); we could not identify any common mutations in the first linkage group, suggesting an epigenetic or nonnuclear genetic mutation or an inheritable variation that remains to be detected. In contrast, all eight strains in the second linkage group contained tandem duplications in the gene ssp1, a Ca2+/calmodulin-dependent protein kinase (human ortholog: CAMKK1/2) which negatively regulates TORC1 signaling, the pathway inhibited by rapamycin, suggesting that mutations in ssp1 were causal for drug resistance (25).
The ssp1 linkage group contained three insertion alleles, all of which were tandem duplications of a short DNA segment (55/68/92 base pairs [bp] in length) and had 5 to 8 bp of identical sequence (microhomology pairs, MHPs) at each end (Fig. 1D and SI Appendix, Fig. S1B and Dataset 7). We postulate that these microhomology-mediated tandem duplications (MTDs) (26⇓–28) are important for de novo generation of reversible mutations.
All three MTDs resulted in frameshifts and inactivation of ssp1. A similar level of drug resistance was found in the ssp1Δ, and replacement of the MTD alleles by transformation with wild-type ssp1 restored sensitivity (Fig. 1E). Sanger sequencing showed that all 16 randomly selected drug-sensitive revertants of the MTD alleles had the wild-type ssp1 sequence. Finally, ssp1Δ and ssp1MTD strains are temperature sensitive. Spontaneous drug-sensitive nonts revertants were frequently recovered for all the ssp1MTD alleles at a frequency of roughly 1/10,000 cells but not for the ssp1 deletion (Fig. 1F). The frequency of revertants is thus 100× higher than the forward MTD frequency (8/107), and MTDs in ssp1 are causal for reversible temperature sensitivity and drug resistance.
Supporting the notion that MTDs may not be specific to rapamycin/caffeine treatment and/or the target gene ssp1, in an unrelated genetic screen for suppressors of the slow growth defect of cnp1-H100M, a point mutation in the centromere-specific histone gene, we identified MTDs in the transcription repressor genes yox1 and lsk1 (SI Appendix, Figs. S1B and S2 and Dataset 7). These MTDs increase fitness in the cnp1-H100M background, and therefore, unlike ssp1MTDs, revertants do not increase in abundance in the mutant background. However, in the ssp1wt background, these MTDs are deleterious, and revertants accumulate (SI Appendix, Figs. S1 and S2). Thus, MTDs are not gene specific and likely occur throughout the genome.
10,000× Whole-Genome Sequencing Identified Thousands of Subclonal MTDs within a Clonal Population.
Based on the scale of the initial genetic screen and assuming drug resistance is not induced by rapamycin, the frequency of cells with any protein-inactivating MTD in ssp1 in an exponentially growing nonselected wild-type population is estimated ∼8 × 10−5. This result suggests that a clonal, presumed “isogenic” population contains a wide variety of subclonal MTDs at multiple loci throughout the genome. The frequency of any single MTD will depend on the rate of MTD formation, the rate of reversion, and the fitness cost of the MTD (19⇓–21). The fitness defect imposed by the MTD can be due to altered gene expression or protein function or from the fitness cost of ∼0.025% per kb of additional DNA (29, 30).
To identify the cis-encoded determinants of MTD frequency, we developed a computational pipeline for detecting subclonal MTDs in high-coverage Illumina sequencing data (see Materials and Methods for details). This method first identifies all MHPs in a DNA segment or genome and generates “signatures” for sequences that would be created by each possible MTD. It then identifies sequencing reads that match these signatures and thus provides experimental support for the existence of a particular MTD within the population (Fig. 2A). This method is capable of identifying subclonal MTDs present at very low frequencies in the population.
To determine whether subclonal MTDs captured by sequencing represent the true genetic variation or are technical artifacts (31), we performed two orthogonal tests. In the first, we tested whether MTDs are specific to genomic DNA or also exist in chemically synthesized DNA. We performed 105x − 106x coverage sequencing of ssp1 DNA fragments PCR-amplified from genomic DNA from a cloned copy of the gene in a plasmid in Escherichia coli or chemically synthesized 150-nt and 500-nt fragments of the gene as well as direct sequencing of chemically synthesized short DNA fragment and plasmid-borne fragment without PCR amplification. We observed far more MTDs in the pombe genomic DNA than in the chemically synthesized or plasmid borne controls (Fig. 2B and SI Appendix, Fig. S3), suggesting that MTDs are largely not caused by PCR or an artifact of Illumina sequencing. It is unclear why the plasmid-borne copy of ssp1 has fewer MTDs than the chemically synthesized DNA, but it raises the possibility that MTDs may be more frequent in eukaryotes (Fig. 2B and SI Appendix, Figs. S6–S8). We detect MTDs in the E. coli genome 1/20th as often as in S. pombe and 1/60th as often as in Saccharomyces cerevisiae (SI Appendix, Fig. S9).
As a second test, we hypothesized that most MTDs in essential genes should be deleterious and recessive. We therefore analyzed raw sequencing data from 209 S. cerevisiae haploid and diploid mutation accumulation (MA) lines (32) and identified all MTDs that occur in only one MA line. In haploids, MTDs were depleted in both essential genes and nonessential genes whose deletion causes a fitness defect (Fig. 2C). In addition, the MTDs that did occur were more likely to maintain the correct reading frame; the single MTD in an essential gene in a haploid was subclonal, maintained the correct reading frame, and was just 112 bp from the 3′ end of CCT7, a 1,652-bp gene. In contrast, there was no such reading-frame bias in diploids, nonessential genes, or intergenic regions (Fig. 2C). Therefore, rare subclonal MTDs identified by ultra-deep sequencing are likely real biological events mostly not experimental artifacts.
To assess the prevalence of MTDs and to identify the sequence-based rules that determine the probability of formation of each tandem duplication, we grew a single diploid fission yeast cell up to ∼108 cells (25 generations) and performed whole-genome sequencing to an average coverage of 10,000× the diploidy relaxed selection, allowing mutations to accumulate throughout the S. pombe genome.
With 10,000× genome sequencing, we identified 5,968 (0.02%) MHPs in which one or more sequencing reads supported an MTD. We observed zero MTDs in most genes, likely due to under-sampling (SI Appendix, Fig. S4). However, 20 genes contained more than 10 different MTDs in a single “clonal” population (Fig. 3A). To understand this heterogeneity across the genome, we used a logistic regression machine-learning model to predict the probability of duplication at each MHP. MH length, guanine and cytosine (GC) content, inter-MH distance, measured nucleosome occupancy, transcription level, and a local clustering on the scale of 100 nt were able to predict which MHPs give rise to duplications with an area under the curve score of 0.9 with 10-fold cross validation (Fig. 3 B and C and SI Appendix, Fig. S5 and Dataset 5). We note that the peak at 150-nt inter-MH spacing is independent of read length, was not found in E. coli or in mitochondrial DNA, and varies between haploid and diploid (SI Appendix, Figs. S5–S8). This analysis revealed that properties of MHPs significantly affect the likelihood of MTD formation; for example, and consistent with previous work in E. coli (34), long GC-rich MHP is 1,000× more likely to generate a tandem duplication than a short AT-rich one.
While MHPs are spread roughly uniformly throughout the genome (Fig. 3D, red), we observed both hot spots, in which MH-mediated generation of tandem duplications are common, and cold spots, in which they are rare (Fig. 3E). Local differences in MHP density can only explain some of the hotspots, while our logistic regression model explains the vast majority, suggesting that hotspots with frequent formation of tandem duplications are mostly determined by the local DNA sequence features in addition to microhomologies. The consequence is that duplications are more than 10× more likely to occur in some genes than others, and this variation is correctly predicted by our model (Fig. 3F). We detected no MTDs in ura4, which has a score of 52, placing it in the bottom third of genes (SI Appendix, Fig. S10 and Dataset 4) and providing a possible explanation why MTDs have not been noticed in 5-FOA-based screens of mutations in ura4 (35). Our results also emphasize that high-coverage sequencing is necessary to identify sufficient numbers of MTDs; 1 billion reads would be required to identify half of the 25 million possible MTDs in the S. pombe genome (SI Appendix, Fig. S4).
We identified three different subclonal MTDs in the SAGA complex histone acetyltransferase catalytic subunit gcn5, placing gcn5 in the top 5% of genes for both observed and predicted MTDs, suggesting that MTDs in gcn5 should be found frequently in a genetic screen. Indeed, examination of 16 previously identified (36) suppressors of htb1G52D identified MTDs in gcn5, as well as in ubp8, in which we also observed an MTD in our high-coverage sequencing data (SI Appendix, Fig. S1B). These results suggest that MTDs arise in most genes at a high-enough frequency within populations in order to be the raw material on which natural selection acts.
Replication Slippage Modulates the Rate of MTD Reversion at ssp1.
Having established that local cis-encoded features determine the frequency with which tandem duplications arise from MHPs, we next sought to identify the trans-genes that affect the MTD process. ssp1MTD alleles fail to grow at 36 °C, and their reversion back to wild type suppresses the temperature sensitivity, providing a way to measure the effects of mutations on reversion frequency. We screened a panel of 364 strains with mutations in DNA replication, repair, recombination, or chromatin organization genes for mutants that affect the rate of ssp1MTD reversion back to wild type (Dataset 6) and found three mutants that significantly increased and eight that significantly decreased the frequency of ssp1WT revertants (Fig. 4 A–C).
Replication fork collapse is a major source of double-stranded breaks (DSBs), and the ensuing homologous recombination (HR)–related restarting process is error prone and is known to generate microhomology-flanked insertions and deletions via replication slippage, a process in which, when replication resumes for a stalled or collapsed fork, the unwound nascent strand may anneal with a homologous segment on the template, either at the vicinity (37, 38) or at a distance (34) of the paused site, with ensuing replication on noncontinuous template. Inactivation of Rad50, Rad52, or Ctp1 results in decreased replication slippage and decreased MTD reversion (37, 39) (Fig. 4 A–C). Deletions of mhf1 and mhf2, two subunits of the FANCM–MHF complex, which is involved in the stabilization and remodeling of blocked replication forks, also decreased the frequency of MTD revertants. It is therefore likely that replication slippage during HR-mediated fork recovery causes reversion of MTDs in these mutants and could be one contributing factor in wild-type background.
Replication stresses activate a checkpoint that promotes DNA repair and recovery of stalled or collapsed replication forks and delays entry into mitosis (40, 41). The inactivation of replication checkpoint kinase cds1 or its regulator mrc1 may thus result in a failure to restore the replication fork, causing increased genome instability and MTD reversion. The replication checkpoint would thus be required for the stability of MTDs. Consistently, we found that deletion of the DNA damage checkpoint kinase cds1 or its regulator mrc1 increased the frequency of ssp1WT revertants. Deletion of the single-stranded DNA binding A (RPA) subunit ssb3 (RPA3/RFA3) or the multifunctional 5′-flap endonuclease rad2 also increased the frequency of revertants (Fig. 4C).
Many genes identified in the screen are multifunctional and play roles in both replication and repair. We therefore performed quantitative epistasis analysis to determine the relation between six of the identified genes and the Mediator of the Replication Checkpoint, mrc1, which interacts with and stabilizes Pol2 at stalled replication forks. In addition to the checkpoint activator cds1, deletion of rad2 had no effect in an mrc1Δ background, suggesting that all three of these genes act in the same pathway (Fig. 4D). In contrast, deletion of ssb3 increased the frequency of revertants in both wild-type and mrc1Δ backgrounds, and deletion of pds5 or rik1 decreased the frequency of revertants in both wild-type and mrc1Δ backgrounds, though not to the extent expected for genetic independence, suggesting partial epistasis. In contrast, the effects of rad50 deletion were completely independent of mrc1 (Fig. 4D).
While the observed numbers of MTDs in ultra-deep sequencing experiments are a function of both duplication and reversion rates and all of the above genes may play a role in both processes, the above results suggested that due to increased reversion rates, the number and frequency of MTDs would be reduced in cds1Δ and rad2Δ strains. To test this idea, we performed 106x coverage sequencing of the hotspot gene SPCC1235.01. We observe MTDs at fewer MHPs and an overall decrease in the number of MTDs in both mutants (Fig. 4 E and F).
Half of Insertions and Tandem Duplications in Natural Isolates Are MH Mediated.
It was baffling that MTDs are prevalent within populations and that the first theoretical proposal for microhomology-mediated processes in the generation of tandem duplications is 20 y old (5); yet, relatively little is known about the forward process and even less about the reversion, suggesting that these events are not often encountered or identified as such. To better understand the dynamics of MTDs within a population, we used a simple model of neutral mutations within a growing population that takes into account both forward and reverse mutation rates and began with 100% of individuals as wild type (see Materials and Methods). The mutant frequency always increases and over short timescales (Fig. 5 A, Left), increasing the reverse rate from being equal to the forward mutation rate (gray) to being 10,000 times higher (yellow) has little effect.
Over longer timescales, high reversion rates cause the mutant frequency to plateau and remain subclonal (Fig. 5 A, Right), reducing the fraction of neutral MTDs within a population. However, despite the high reversion rate, both drift and selection enable fixation of MTDs within a population. To identify fixed microhomology-mediated insertions, we searched the genome sequences of 57 wild S. pombe isolates (42) and found that 50% of insertions larger than 10 bp involve microhomology repeats (Fig. 5 B and C). Among these were 158 microhomology-mediated insertions that did not contain an obvious duplication and 113 MTDs with an MTD.
To test whether the propensity of MTD formation within the laboratory strain is predictive of extant sequence variation observed in natural isolates, we tested whether the MTD score for each gene predicts the likelihood of microhomology-mediated insertions in that gene. We found that genes with microhomology-mediated insertions in natural isolates tend to have higher predicted MTD scores and more experimentally observed MTDs (Fig. 5D), suggesting that the local features that affect MTD formation in the laboratory also shape evolution in nature.
MHPs with Longer MH Sequences Are More Likely to Generate MTDs that Maintain the Correct Reading Frame.
We found that MHPs with longer MH sequences are more likely to form MTDs. If the high propensity to generate MTDs has shaped the S. pombe genome, any signature of selection should be stronger at MHPs with longer MH sequences and should also be stronger in essential genes versus nonessential genes. We therefore divided the 25 million MHPs with an MH length of 4 to 25 nt and an inter-MH distance of 3 to 500 nt into those fully contained within in intergenic regions or fully contained within essential or nonessential genes and split them by MH sequence length. Specifically in coding sequences, MHPs at which an MTD would not disrupt the reading frame are more common than expected by chance, and this enrichment is higher in essential genes and at longer MH sequences (Fig. 6). At the same time, MHPs within genes are more common that expected by chance (SI Appendix, Fig. S12). Therefore, natural selection has acted to decrease the number of MHPs that would create potentially deleterious MTDs, and this selection is weaker for MHPs that would create an in-frame MTD.
Taken together, our results demonstrate that MTDs occur frequently and broadly throughout the genome within a clonal population. These findings indicate that high levels of subclonal genetic divergence are prevalent but are under detected using conventional sequencing approaches that tend to disfavor the detection of low-abundance subclonal variants. As many MTDs create large insertions, they are more likely to be deleterious. Nonetheless, MTDs provide plasticity to the genome and its functionality, for example, by allowing cells to become drug resistant, while allowing the resistant cell lineage to revert back to wild type and regain high fitness once the drug is removed. Selection can act on this genetic diversity for its reversibility or by using the tandem duplications as the initial step for the generation of higher copy number repeats, which are evolutionarily fixed in extant genomes and traditionally regarded as a major source of genome divergence. While previous work has shown that preexisting repeats undergo rapidly reversible changes, the sequence-encoded rules regulating the birth and death of such sequences were less studied (34). This work reveals that numerous sites throughout the genome have the potential of evolving into such repetitive elements. Furthermore, MH sequence length–dependent depletion of frame-shifting MHPs in essential genes shows that natural selection has shaped the genome to avoid MHPs that would frequently generate deleterious MTDs. Finally, much in the same way as repetitive DNA may have been positively selected for as a regulatory element to maintain reversible genetic diversity (7, 8), the large number of MHPs that would result in in-frame MTDs raises the possibility that some genes may maintain MHPs to generation functional genetic diversity, creating a dynamic protein-coding genome.
Why haven’t MTDs been identified more frequently in genetic screens and MA assays? There are several possible reasons. Mutation callers are ineffective in detecting long insertions from short reads: on simulated data, both Mutect2 and HaplotypeCaller often fail to detect tandem duplications longer than 85 bp. Also, MTDs are often identified as insertions but not specifically as MTDs (SI Appendix, Figs. S1B and S11), suggesting the need for computational tools for identifying MTDs. URA4 and URA3 have relatively few MHPs (SI Appendix, Fig. S10). In many 5-FOA–based mutation spectra papers, only substitutions were analyzed in detail, as indels do not occur with high-enough frequency to generate good statistics (43). Due to the high rate of reversion, it is likely that tandem duplications in URA4 would yield URA+ colonies when restruck onto -URA plates; this high reversion rate may lead to MTD-containing colonies being discarded in many different types of genetic screens. In our reanalysis of S. cerevisiae MA lines, almost all de novo MTDs were subclonal (SI Appendix, Fig. S11B). With new computational tools for identifying MTDs plus third-generation sequencing platforms with improved ability to detect long indels, it is likely that MTDs will be implicated in more phenotypes.
Unbiased genome-scale approaches have been very informative for the mechanisms that generate point mutations in both wild-type and mutant cells (44). It is clear that multiple molecular mechanisms can give rise to tandem duplications in microhomology-dependent and independent manners, and the mechanisms may differ in mutants and between species. In plant mitochondria genomes, longer MH sequences are associated with longer tandem duplications, suggesting that microhomology is involved in the generation of tandem duplications, likely via microhomology-mediated repairing of DSBs (45, 46) or slippage strand replication (47). In contrast, tandem duplications in the rice nuclear genome tend to have no or shorter microhomology sequences, suggesting that in the rice nuclear genome, tandem duplications likely form via patch-mediated DSB creation followed by NHEJ (48). In E. coli, the lagging-strand processing activity of Pol I is required for stress-induced, MH-mediated amplification of 7- to 32-kb segments (34). However, simple models cannot account for all features observed across studies, and it is clear that multiple mechanisms play a role (49⇓–51).
All current methods of measuring microhomology-mediated duplications and deletions impose artificial length scales, whereas genetic screens require the entire gene or a specific region be duplicated. Thus, while microhomologies are associated with deletions on the 500-bp to 1-kb range in E. coli (52) and with unstable amplifications of 7 to 37 kb (16) and selection for increased gene expression often enriches for ∼12-bp MH-mediated amplification of ∼10kb (53), these length scales are determined by the locations of MH sequences in the particular genomic region required to be duplicated in the genetic screen and the size of the region required to be duplicated or amplified.
Genome-wide sequencing-based approaches are less biased but still not bias free. We limit the length-scale to 500 bp and set lower and upper bounds of 3 bp and 25 bp for the MH sequences. While the MTD frequency relative to the number of 1-bp or 2-bp MH sequence pairs in the genome is likely to be low, the number of 2-bp and 1-bp MH sequences is high. Preferential flanking of tandem duplications by 2-bp and even 1-bp sequence identities have been reported (27, 28, 45, 54), suggesting that subclonal MTDs generated by short MH sequences may be common. While it is computationally intractable to apply the directed “signatures” method presented here to search MH pairs separated by more than 500 bp, this method could be extended to >500 bp if the search is limited to longer MH sequences, which are rare but drive high rates of MTD formation (28, 34). Similarly, the method could be extended to broken microhomologies (55) but with an even higher computational cost. Computational approaches using third-generation (Nanopore and PacBio) sequencing have the potential to provide a truly unbiased measure of duplication and deletion frequencies as well as answer questions about how often amplifications are extrachromosomal versus intrachromosomal (17, 56) and tandem versus inverted amplifications (57). Verification of these algorithms will take some work, as Nanopore and Pac-Bio sometimes give different results when sequencing tandem repeats (58).
Because different molecular mechanisms have different sequence requirements, replication strand biases, and length scales (34, 50), genome-wide unbiased methods are necessary to understand the relative contribution of each mechanism to MTD formation and collapse. As they can, in theory, measure events across any distance, from single base pair to interchromosomal, and at an unlimited number of different loci, with variation in chromatin contexts, transcription, and other genome-architecture features, ultra-deep sequencing is likely the best way to quantitatively understand the various biological mechanisms that contribute to the dynamic genome.
Read more here: Source link