Impact of Homologous Recombination on Silent Chromatin in Saccharomyces cerevisiae

Abstract

One of the best-studied domains of heterochromatin is the silent mating-type locus HML in baker’s yeast. Sieverman and Rine report that DNA transactions...

Specialized chromatin domains repress transcription of genes within them and present a barrier to many DNA–protein interactions. Silent chromatin in the budding yeast Saccharomyces cerevisiae, akin to heterochromatin of metazoans and plants, inhibits transcription of PolII- and PolIII-transcribed genes, yet somehow grants access to proteins necessary for DNA transactions like replication and homologous recombination. In this study, we adapted a novel assay to detect even transient changes in the dynamics of transcriptional silencing at HML after it served as a template for homologous recombination. Homologous recombination specifically targeted to HML via double-strand-break formation at a homologous locus often led to transient loss of transcriptional silencing at HML. Interestingly, many cells could template homology-directed repair at HML without an obligate loss of silencing, even in recombination events with extensive gene conversion tracts. In a population of cells that experienced silencing loss following recombination, transcription persisted for 2–3 hr after all double-strand breaks were repaired. mRNA levels from cells that experienced recombination-induced silencing loss did not approach the amount of mRNA seen in cells lacking transcriptional silencing. Thus, silencing loss at HML after homologous recombination was short-lived and limited.

IN the budding yeast Saccharomyces cerevisiae, the genes at the cryptic mating type loci HML and HMR are silenced, allowing haploid cells to maintain their identities as either the a or α mating type. Yeast silent chromatin shares characteristics with heterochromatin of metazoans and plants, including hypoacetylation of histones (Braunstein et al. 1993; Suka et al. 2001), epigenetically inherited repression (Pillus and Rine 1989; Xu et al. 2006), and compact, higher-order chromatin structure (Bi and Broach 1997; Cheng et al. 1998; Weiss and Simpson 1998; Ravindra et al. 1999). The Silent Information Regulator (SIR) proteins Sir1–Sir4 establish silencing via recruitment to regulatory sites called silencers that flank both HML and HMR. Sir2, the founding member of the highly conserved sirtuin family of deacetylases, removes acetyl marks on histone H4 at position K16 and on histone H3 at positions K9 and K14 across the silenced domain (Imai et al. 2000; Landry et al. 2000; Smith et al. 2000). H4K16 deacetylation creates high-affinity binding sites for the Sir complex, comprising Sir2, Sir3, and Sir4, resulting in a chromatin domain that represses transcription of a variety of RNA PolII- and PolIII-transcribed genes [reviewed in Gartenberg and Smith (2016)].

Silent chromatin offers limited accessibility to many DNA-binding proteins (Gottschling 1992; Singh and Klar 1992; Loo and Rine 1994), yet must allow certain transactions like replication and homologous recombination to occur. In fact, homologous recombination within silent chromatin is a key aspect of yeast biology, as the process of mating-type switching depends upon it (Strathern et al. 1982; Kostriken et al. 1983). The mating type of haploid yeast is determined by the a or α allele present at the MAT locus. Unexpressed copies of the MATa and MATα genes reside within silent chromatin at HML and HMR. Mating-type switching initiates when the HO endonuclease creates a double-strand break at MAT, which is then repaired by homologous recombination templated from either HML or HMR. SIR proteins prevent access of HO to its recognition sequences at HML and HMR, ensuring that only MAT is available for cleavage. Recombination between MAT and one of the heterochromatic donor loci results in gene conversion of the sequences at MAT from either HML or HMR.

Mating-type switching in S. cerevisiae has provided a foundation for elucidating much of what is known about the repair of double-strand breaks by homologous recombination. Repair of the double-strand break at MAT begins with resection in the 5′ to 3′ direction on each side of the break (White and Haber 1990; Sun et al. 1991). The Rad51 protein binds the exposed single strands and coordinates with other DNA-repair proteins, including Rad54 and Rad52, to facilitate recognition of homologous sequences at either HML or HMR and carry out strand invasion of the donor locus. Repair of the double-strand break at MAT occurs through a type of recombination called synthesis-dependent strand annealing (SDSA), resulting in unidirectional transfer of genetic information from the donor locus, either HML or HMR, to MAT (Haber et al. 1980; Klar and Strathern 1984; Ira et al. 2006).

Considering the ability of silent chromatin to block DNA–protein interactions and the broad range of proteins needed to repair the cleaved MAT locus from the heterochromatic donors, one might expect an obligatory loss of silencing during mating-type switching to allow the recombination machinery access to the silenced template used for repair. To date, there has been no evidence that mating-type switching causes any loss of transcriptional silencing at either HML or HMR. However, traditional assays of silencing loss that rely on mating phenotypes have limited ability to reveal whether a donor locus becomes expressed as a result of a mating-type switch. In this study, we used a recently developed assay capable of detecting even transient disruptions of silencing (Dodson and Rine 2015) to investigate whether changes to silent chromatin dynamics at the HML locus result from its participation in homologous recombination.

Materials and Methods

All strains in this study (Table 1) were derived from strain JRY9628 (Dodson and Rine 2015), which is derived from W303 (R. Rothstein, Columbia University). Deletion of HMR was accomplished via one-step integration (Gueldener et al. 2002) of the Kluyveromyces lactis LEU2 gene using the hmr∆::K.lac.LEU2 forward/reverse primers and confirmed with sequencing. pGAL10:HO (Herskowitz and Jensen 1991) was integrated at the LEU2 locus with the leu2∆::pGAL10:HO forward/reverse primers.

Yeast strains

To create the SNP-INC at HML::cre, site-directed mutagenesis was performed on the HML::cre sequence from JRY9628 using the HML-INC forward/reverse primers. To construct the pseudo-MAT locus, two rounds of site-directed mutagenesis were performed on the HML::cre sequence from JRY9628 using the SNP-R forward/reverse primers and the SNP-L forward/reverse primers. The pseudo-MAT locus was inserted between the YHL045W and YHL044W open reading frames on the left arm of ChrVIII, replacing the sequences between base-pairs 13,192 and 13,237, using the ChrVIII::pseudo-MAT forward/reverse primers. For strain JRY10818, the HML::cre sequence with SNP-INC was inserted at the same location on ChrVIII with the ChrVIII::pseudo-MAT forward/reverse primers.

To remove the endogenous HO recognition sequence from HML::cre in strain JRY10819, Gibson assembly (Gibson et al. 2009) was performed with two overlapping fragments of HML::cre from JRY9628 using primers ∆HO-5′ and ∆HO-3′ in a way that deleted base pairs 356–475 of the α1 sequence, leaving α1 with base pairs 1–355;476–528 intact.

To insert the HO recognition sequence into the K. lactis URA3 sequence, a three-step Gibson assembly was performed that inserted base pairs 377–465 of the α1 sequence between base pairs 200 and 201 of the K. lactis URA3 sequence using primers K.lac.URA3-HO-5′ and K.lac.URA3-HO-3′. The K.lacURA3-HO construct was then inserted at the same position on the left arm of ChrVIII as pseudo-MAT in JRY10817 with the ChrVIII::K.lac.URA3-HO forward/reverse primers.

To insert the K. lactis URA3 sequence containing the HO recognition sequence with SNP-INC and SNP-R onto ChrIX, a three-step Gibson assembly was performed that inserted base pairs 377–465 of the α1 sequence containing SNP-INC and SNP-R between base pairs 200 and 201 of the K. lactis URA3 sequence using primers K.lac.URA3-HO-5′ and K.lac.URA3-HO-3′. This construct was then inserted on the right arm of ChrIX in between base pairs 428,440 and 428,441 with the ChrIX::K.lac.URA3-HO-INC-SNPR forward/reverse primers.

To insert the swi2∆10R allele into the SWI2 locus, a sequence to create a guide RNA to target cleavage of SWI2 by Cas9 was first cloned into the BsmBI sites of pYTK050 (Lee et al. 2015) using the SWI2-sgRNA forward/reverse primers. The full guide RNA sequence was excised from this plasmid and inserted into the BsaI sites of pWCD2257 [described in Figure S5B of Lee et al. (2015)], which contains the Cas9 sequence on a yeast CEN/ARS plasmid, thus creating pJR3417. pJR3417 was then transformed into JRY10817 along with a portion of swi2∆10R from CP1413 (Manning and Peterson 2014) that was PCR-amplified using the swi2∆10R forward/reverse primers to create strain JRY10820. Successful replacement of SWI2 by swi2∆10R was confirmed with sequencing.

All genomic positions described are from the S288C reference genome version R64-2-1, which can be accessed at http://downloads.yeastgenome.org/sequence/S288C_reference/genome_releases/.

Galactose induction of double-strand breaks

Strains were grown at 30° in liquid Complete Supplement Mixture (CSM)-Trp (Sunrise Science Products, San Diego, CA) with raffinose (2% w/v) as a carbon source and containing G418 to select for red cells, then diluted into liquid CSM-Trp medium with raffinose and allowed to grow to an A600 of 0.2–0.4. Cultures were then split and received either galactose or raffinose to a final concentration of 2%. For experiments with a glucose recovery, cultures were centrifuged at 4000 rpm for 15 min, then resuspended in CSM-Trp medium containing glucose (2% w/v) and allowed to grow. For each time point, a sample of culture was diluted in CSM-Trp with glucose and plated onto CSM-Trp glucose plates.

Colony imaging and silencing-loss analysis

Colonies were grown for 5–7 days on CSM-Trp glucose plates and then imaged with a Zeiss Axio Zoom.V16 microscope equipped with ZEN software (Zeiss, Jena, Germany), a Zeiss AxioCam MRm camera, and a PlanApo Z 0.5× objective. Sectoring patterns were scored manually. Colonies that were at least one-quarter green were counted as “early silencing-loss” events. Red colonies with wild-type levels of sectoring were scored as colonies that “maintained silencing.” Colonies that were petite or with morphologies suggestive of reciprocal crossover leading to a dicentric chromosome were omitted from analysis.

DNA blots

DNA hybridization blots were performed as previously described, with only minor changes (Southern 2006). Probes were radiolabeled by random priming with ³²P-αdCTP using either the Amersham Rediprime II Random Primer Labeling System (GE Healthcare) or the Amersham Megaprime DNA Labeling System (GE Healthcare). Membranes were exposed 16–72 hr with a Storage Phosphor Screen (GE Healthcare) and imaged with a Typhoon Trio (GE Healthcare). To determine the ratio of pseudo-MAT molecules cut, the intensity of each pseudo-MAT band was quantified using the Gel Analysis function of ImageJ software (National Institutes of Health, Bethesda, MD). Ratios were calculated by dividing the sum of intensities of the two cut pseudo-MAT bands by the sum of intensities of the cut and uncut pseudo-MAT bands.

For blots probing the pseudo-MAT and HML::cre loci (Figure 1C, Figure 5C, and Figure 6B), DNA was digested with XbaI and PacI. The probe comprised a 505 bp sequence centered around the HO recognition sequence that was amplified from the HML::cre sequence of JRY10817 using the HO-α1 probe forward/reverse primers.

For blots probing the URA3-HO sequences (Figure 3B), DNA was digested with PacI and PvuII. The probe comprised a 501 bp sequence surrounding the HO recognition sequence that was amplified from the K. lactis URA3-HO sequence of JRY10819 using the K.lac.URA3-HO probe forward/reverse primers.

RNA preparation and quantitative RT-PCR

RNA was extracted from cells using the hot-acid-phenol method (Kebaara et al. 2012). 10 μg of RNA was digested with DNase I (Invitrogen, Carlsbad, CA) then purified with phenol-chloroform extraction followed by precipitation with 100% ethanol and 0.3 M sodium acetate (pH 5.2). Complementary DNA was synthesized from 2 μg of DNAse-treated RNA with the SuperScript III First-Strand Synthesis System for RT-PCR (Invitrogen) and oligo(dT) primers. Quantitative PCR of CDNA was executed to detect the 3′ end of the cre transcript specific to HML::cre with the cre-3′-RTPCR forward/reverse primers using the Thermo Scientific DyNAmo HS SYBR Green qPCR Kit (Fisher Scientific, Chicago, IL) and a Mx3000P machine (Stratagene, acquired by Agilent Technologies, Santa Clara, CA). Expression levels were normalized first to ACT1 mRNA, which was measured using the ACT1-RTPCR forward/reverse primers, and then to cre mRNA levels from strain JRY10822. Samples were analyzed in technical triplicate.

Calculating gene conversion frequencies

For colony genotyping, colonies were resuspended in 200 μl TE buffer and 100 μl of glass beads. 200 μl of 25:24:1 phenol:chloroform:isoamyl alcohol was added, and tubes were subjected to two rounds of 20-sec bead-beating using a FastPrep-24 (MP Biomedicals, Santa Ana, CA). Tubes were centrifuged for 10 min at 15,000 rpm, and 1 μl of the aqueous phase was removed for PCR. PCR amplification of the pseudo-MAT locus on ChrVIII was carried out using the pseudo-MAT-amplify forward/reverse primers and Phusion High-Fidelity DNA polymerase (New England Biolabs, Ipswich, MA). A PCR cleanup reaction was performed using Agencourt AMPure XP reagent and the Beckman NX automated liquid handler by the University of California at Berkeley DNA Sequencing Facility, and PCR products were sequenced with primers pseudo-MAT-seq-A and pseudo-MAT-seq-B. Sequencing traces were analyzed using SnapGene software (from GSL Biotech; available at snapgene.com). For each of the three differentiating SNPs, sequencing traces that showed peaks of both the original pseudo-MAT sequence and the sequence at HML::cre were scored as partial gene conversions and those that showed only the HML::cre sequence were scored as full gene conversions. Multiple negative control samples in the sequencing process with no template DNA failed to produce PCR amplicons for sequencing, indicating that the presence of both alleles within a single colony reflected heterogeneity of genotype within the colony rather than cross-contamination of sequencing samples.

Measuring half-life of cre mRNA

Strain JRY10823, bearing the plasmid pJR2538 with pGAL:cre (Table 2), was grown to saturation in CSM-His + galactose medium, then diluted to an OD of 0.25 in the same medium and grown to an OD of 0.5. At this point, a preinduction sample was removed. The culture was centrifuged for 10 min at 4000 rpm and resuspended in CSM-His with glucose to turn off cre expression. A time 0 sample was immediately taken, and samples were taken every 10 min for the first 60 min and then every 15 min for the next 60 min. RNA extraction and quantitative RT-PCR were performed as described above.

Plasmids used in this study

Data availability

All strains and plasmids are available upon request. Supplemental Material, Table S1 in File S1 contains the oligonucleotide sequences used for strain construction.

Results

Design of a pseudo-MAT locus to test stability of gene silencing at a heterochromatic donor locus

To ask whether silencing stability is affected by homologous recombination within the silenced domain, we created a strain that allowed temporal control of homology-directed repair of a double-strand break at a custom-designed locus using HML as a donor. To monitor silencing stability, we employed an assay previously developed by our lab called the Cre-Reported Altered States of Heterochromatin, or CRASH, assay (Dodson and Rine 2015). In this assay, the cre recombinase gene replaces the native α2 sequence at HML. Loss of silencing at HML::cre leads to the production of Cre protein, which then acts on a reporter cassette with loxP sites positioned around RFP and GFP genes, such that Cre-mediated recombination creates a permanent and heritable switch in fluorescence from red to green and a loss in resistance to the drug G418 (Figure 1A).

During mating-type switching, a double-strand break at MAT results in transposition of the sequence from the HML donor locus to the MAT locus, where it is then expressed. Hence, transposition of cre from HML to MAT would cause all cells to switch from RFP expression to GFP expression, rendering the CRASH assay ineffective. To induce a double-strand break that would be repaired by recombination targeted to HML::cre yet circumvent this problem, we constructed a pseudo-MAT locus on the left arm of ChrVIII (Figure 1B). This pseudo-MAT locus consisted of a 2.4 kb region of HML::cre, omitting the silencers as well as the 3′ end of the cre open reading frame, which encodes the protein’s active site required for Cre function. The lack of these sequences and flanking homology guaranteed that repair of the double-strand break at the pseudo-MAT locus could not restore a functional cre gene at that locus. The endogenous recognition sequence of the HO endonuclease found within the α1 open reading frame served as a target for double-strand break induction at the pseudo-MAT locus. The HO gene under control of the GAL10 promoter allowed temporal regulation of double-strand break induction by addition of galactose to the growth medium. Under physiological conditions of HO expression, Sir proteins prevent access of HO to its recognition sequence at HML, but HO overexpression can lead to low levels of HML cleavage (Connolly et al. 1988; K. Sieverman 2017). To ensure that HO caused double-strand break formation only at the pseudo-MAT locus and not at HML::cre itself, we mutated the α1 sequence at HML::cre to contain an SNP that eliminates HO’s ability to cleave it, historically known as MATα-inc (Nickoloff et al. 1990) but called SNP-INC here. To further distinguish the homologous sequences at HML::cre and pseudo-MAT, we added two SNPs to the pseudo-MAT locus on either side of the site of HO cleavage: one 721 bp away from the site of HO cleavage, within the cre sequence, called SNP-L, and one 16 bp to the right of the HO recognition sequence, called SNP-R. We also deleted the MAT and HMR loci in this strain to prevent cleavage by HO at its recognition sequence within them.

The pseudo-MAT locus was inserted 13 kbp from the end of the left arm of ChrVIII. We chose this location to encourage efficient recombination between HML::cre and pseudo-MAT based upon two considerations: recombination efficiency correlates with contact frequency, as measured by high-throughput chromosome conformation capture (Hi-C) (Lee et al. 2016), and with similarity in chromosomal arm length (Agmon et al. 2013). Mating-type switching involves only nonreciprocal recombination. To allow us to focus exclusively on the outcomes of nonreciprocal events, the orientation of HML::cre and pseudo-MAT was such that reciprocal recombination via crossover would create a dicentric chromosomal fusion. Such events can be recognized by the slow growth and abnormal colony morphology resulting from various resolutions of the dicentric chromosome. Focusing our analysis on colonies with typical morphology allowed us to evaluate only nonreciprocal recombination events that would not lead to cre expression from the pseudo-MAT locus. Upon induction of HO in this strain (JRY10817), double-strand breaks were evident within 30 min, with breaks persisting but reduced after 4 hr of HO expression (Figure 1C).

Homology-directed repair decreased the stability of silencing at the donor locus

Induction of HO and subsequent plating on solid medium lacking galactose yielded colonies in which at least one-quarter of cells had switched from RFP to GFP expression, indicating a silencing-loss event early in colony growth. Moreover, there was a clear trend of increasing silencing-loss events with increasing duration of HO induction (Figure 2A). A subculture incubated in noninducing (raffinose) medium yielded red colonies, indicating that without exposure to galactose, the stability of silencing at HML::cre was unaffected.

To determine whether the loss in silencing stability upon HO induction resulted from the presence of HO itself rather than from the resulting double-strand break, we created a strain identical to the assay strain but also harboring SNP-INC at pseudo-MAT (JRY10818), thus lacking any recognition sequence for HO to cleave. In this strain, induction of HO resulted in no discernable increased likelihood of silencing loss (Figure 2B). Hence, destabilization of silencing at HML::cre depended on double-strand-break formation at the pseudo-MAT locus, rather than growth in galactose medium or some yet unrecognized ability of HO to influence silencing loss.

To investigate whether recombination involving HML::cre itself was required for the increase in silencing-loss events or whether repair of a double-strand break anywhere in the genome would cause silencing instability, we induced HO cleavage at a locus whose template for homology-directed repair was not HML::cre. Specifically, an HO recognition sequence was engineered into a URA3 gene at the same position on ChrVIII as pseudo-MAT, with an uncleavable version of the same URA3 construct inserted on ChrIX (Figure 3A). As expected, HO induction in this strain (JRY10819) resulted in double-strand-break formation on ChrVIII with cleavage levels similar to those seen at pseudo-MAT (Figure 3B). Even in colonies grown from cells with the longest induction times, there was no increase in silencing loss at HML::cre (Figure 3C). Thus, the silencing instability induced at HML::cre after cleavage of ChrVIII required a double-strand break in a sequence homologous to HML::cre.

Gene conversion of SNPs at pseudo-MAT was diagnostic of recombination with HML::cre

In mating-type switching, homologous recombination via SDSA results in gene conversion of the double-strand break recipient sequence to that of its donor locus. We analyzed the DNA sequence at pseudo-MAT for evidence of gene conversion of the three SNPs differentiating it from HML::cre in colonies grown on solid medium lacking galactose after HO induction. After a 1-hr HO induction, 42% of the resultant colonies showed evidence of gene conversion at the pseudo-MAT locus of at least one of the three SNPs (Figure 4A). The fraction of the population that experienced gene conversion exceeded the fraction that lost silencing, as discussed below.

The SNP-INC was most likely to gene convert: ∼41% of colonies showed appearance of the SNP-INC sequence in the pseudo-MAT locus after 1 hr of HO induction. SNP-L and SNP-R were less likely to gene convert: ∼24.6% of colonies showed gene conversion at SNP-L and 26.2% at SNP-R after 1 hr of induction. Many colonies showed only partial gene conversion of any given SNP, meaning that both the original pseudo-MAT and the HML::cre sequences were detected at the pseudo-MAT locus within the DNA from a single colony. We discuss this genotypic heterogeneity further below.

Recombination could occur without silencing loss

Inducing recombination between pseudo-MAT and HML::cre led to a population of cells that experienced a greater likelihood of silencing loss, yet many cells gave rise to red colonies that had maintained silencing after 4 hr of HO expression. The DNA blot (Figure 1C) suggested that not all recipient loci had been cleaved at any time point, but that analysis could not distinguish between molecules that had been cut and repaired and molecules that had never received a double-strand break. Additionally, 34% of colonies showed no evidence of gene conversion after 4 hr of HO induction (data not shown). Therefore, it was possible that colonies that maintained silencing arose from cells that did not experience a double-strand break and subsequent recombination between pseudo-MAT and HML::cre. Alternatively, some cells may have managed to execute homology-directed repair from HML::cre with no effect on silencing stability.

To distinguish these two possibilities, we analyzed gene conversion frequencies from colonies that were either red or at least one-quarter green to ask whether gene conversion was evident in both populations. After a 1-hr HO induction, ∼84% of colonies that lost silencing early in colony growth exhibited gene conversion at pseudo-MAT (Figure 4B), demonstrating a strong correlation between recombination with HML::cre and silencing instability. Interestingly, ∼30% of red colonies, which maintained silencing through multiple rounds of cell division, also showed evidence of gene conversion (Figure 4C). Therefore, homologous recombination between pseudo-MAT and HML::cre did not lead to an obligatory loss of silencing.

Conceivably, silencing might be maintained only in cells with short conversion tracts, while those with extensive gene conversion may lose silencing stability. However, gene conversion into the cre sequence at SNP-L, 721 bp from the HO cleavage site was possible without subsequent transcriptional activation of the cre gene at HML, as evidenced by the ∼13% of colonies that maintained silencing yet experienced gene conversion at SNP-L (Figure 4C).

Kinetics and extent of silencing loss

To investigate how quickly silencing loss at HML::cre appeared after HO induction and how long it persisted after homology-directed repair, we monitored mRNA levels of the cre transcript specific to HML::cre before, during, and after a 1-hr HO induction. We calculated the half-life of cre mRNA to be ∼3 min after glucose shutoff of a cre gene being driven by the pGAL1 promoter (Figure 5A). Thus, measurements of cre mRNA by quantitative RT-PCR closely reflected the real-time transcriptional output of cells. Prior to double-strand break induction, cre mRNA levels were ∼9000-fold lower than those in a sir3∆ strain lacking transcriptional silencing at HML::cre (Figure 5B). After a 1-hr induction, cre mRNA levels increased ∼10-fold. Cre expression peaked at nearly 100-fold higher than pre-double-strand break induction 2 hr after quenching HO expression, at which time all double-strand breaks at pseudo-MAT appeared to be repaired (Figure 5D). Five hours after quenching HO, cre mRNA levels had returned to ∼10-fold higher levels than before double-strand break induction (Figure 5B), suggesting that many cells that had lost silencing had restored it by this time and were no longer producing cre transcripts. In strain JRY10818 harboring the SNP-INC at pseudo-MAT, expressing HO did not cause an increase in cre transcript levels, indicating that the increase in cre transcripts was specific to cultures experiencing the double-strand break at the pseudo-MAT locus (Figure 5C).

To estimate the extent of silencing loss resulting from homologous recombination, we compared the number of colonies that experienced silencing loss before, during, and after a 1-hr HO induction and compared these values to the cre mRNA levels measured by quantitative RT-PCR. Prior to HO induction, ∼2% of cells gave rise to colonies with an early silencing-loss event (Figure 5E). After a 1-hr HO induction, ∼23% of resultant colonies experienced an early silencing-loss event. The percentage of colonies with silencing loss remained similar after a 2-hr quench of HO expression with glucose, suggesting that no additional cells lost silencing during that time. The highest levels of cre expression were seen at the 2-hr quench time, at which point they were ∼1% that of a sir3∆ population (Figure 5B). Assuming that the population of cells that experienced silencing loss reestablished silencing with similar kinetics, we inferred that 23% of the population expressed cre at a maximum of 1% of full derepression, as measured from mRNA from the entire population. Therefore, the cells that lost silencing after homologous recombination within HML::cre expressed cre at much lower levels per cell than did fully derepressed cells.

A special allele of SWI2 influenced homologous recombination frequency and likelihood of silencing loss after recombination

Previous studies indicate that the presence of Sir3, one of the structural components of silent chromatin, prevents an early step of homologous recombination (Sinha et al. 2009). The nucleosome remodeling activity of the SWI/SNF complex is required in vitro for successful synapsis between Rad51-coated filaments and Sir3-bound nucleosomal donors. SWI2 encodes the ATPase of the SWI/SNF complex and interacts directly with Sir3  in vitro (Manning and Peterson 2014). An allele of SWI2, swi2∆10R, lacks sequences necessary for Swi2’s interaction with Sir3 and eliminates the ability of the SWI/SNF complex to facilitate pairing between Rad51-coated filaments and Sir3-coated nucleosomal donors in vitro (Manning and Peterson 2014). Although swi2∆10R is not reported to inhibit recombination between MAT and HMR in vivo (Manning and Peterson 2014), we hypothesized that the stability of silenced chromatin during recombination may be affected in this mutant.

We repeated the HO induction time course in a swi2∆10R mutant (JRY10820) that was otherwise isogenic with our assay strain. The fraction of colonies with an early silencing-loss event was not significantly different after HO induction in the swi2∆10R mutant relative to the SWI2 parent strain (P value > 0.05 for 1-hr induction and 2-hr induction when analyzed with an unpaired t-test) (Figure 6A), suggesting no increased likelihood of silencing loss in the swi2∆10R mutant after double-strand break formation at pseudo-MAT. Furthermore, there was no obvious difference in the kinetics of break induction and repair in the swi2∆10R mutant relative to a wild-type culture (Figure 6B  vs.  Figure 1C and Figure 5D).

The overall recombination frequency between pseudo-MAT and HML::cre was slightly reduced in the swi2∆10R mutant after 1-hr of HO induction. 44.8% of wild-type colonies showed evidence of a gene conversion event, either full or partial, for at least one of the three SNPs, (Figure 4A), while 33.2% of swi2∆10R mutant colonies experienced a gene conversion event anywhere (Figure 6C) (P-value = 0.0036, unpaired t-test). Thus, the overall frequency of gene conversion events was somewhat greater in strains with a wild-type SWI2 allele. Gene conversion frequencies were reduced in swi2∆10R colonies that both maintained silencing (P-value = 0.0026, unpaired t-test) and experienced early silencing loss (P-value = 0.0095, unpaired t-test) (Figure 6, D and E), although the reduction was more pronounced in the population that maintained silencing. Hence, the mutant swi2∆10R allele slightly reduced the ability of the cleaved pseudo-MAT locus to participate in homologous recombination, and successful recombination events occurred more frequently in colonies that experienced an early silencing-loss event.

In addition to the lower overall frequency of gene conversion events, the swi2∆10R allele slightly increased the likelihood of silencing loss at HML::cre after homologous recombination. Wild-type colonies and swi2∆10R colonies that did not show evidence of gene conversion experienced early silencing-loss events at frequencies that were not significantly different (P-value > 0.05, unpaired t-test) (Figure 7A  vs.  Figure 7C). In colonies that experienced a gene conversion event, full or partial, for at least one of the three SNPs, those that were swi2∆10R mutants were more likely to lose silencing than their wild-type counterparts (P-value = 0.0174, unpaired t-test) (Figure 7B  vs.  Figure 7D). Thus, homologous recombination led to silencing loss more frequently at HML::cre in the presence of the swi2∆10R mutant allele.

Investigating causes of genotypic heterogeneity within colonies

As mentioned earlier, we observed some partial gene conversion events at the pseudo-MAT locus, whereby both the original pseudo-MAT and HML::cre sequences were detected within a single colony after HO induction. This genetic heterogeneity suggested that not all cells within the colony experienced the same molecular outcome at the pseudo-MAT locus after double-strand break induction. Double-strand breaks can be repaired independently of homologous recombination by nonhomologous end joining (NHEJ), and breaks made by HO often lack a sequence scar when repaired by NHEJ (Kramer et al. 1994; Moore and Haber 1996; Haber 2012). We hypothesized that the genotypic heterogeneity observed within colonies might be a result of postreplication breaks in which one chromatid of ChrVIII underwent homology-directed repair at pseudo-MAT while the other was nonhomologously end joined, leading to two different repair outcomes and thus two genotypes within the colony. To test this, we deleted the DNA ligase IV gene, DNL4, which is required for NHEJ (Wilson et al. 1997), in the parental strain used for all of the experiments above. In this mutant (JRY10821), we did not observe a difference in likelihood of silencing loss upon HO induction after a 1-hr HO induction (P-value > 0.05, unpaired t-test, data not shown). The overall gene conversion frequencies after a 1-hr HO induction were not significantly different between wild type and dnl4∆ (P-value > 0.05 for gene conversion anywhere, unpaired t-test) (Figure 8A). Furthermore, the frequency of partial gene conversions was not significantly different in the dnl4∆ mutant (P-value > 0.05 for partial gene conversion anywhere, unpaired t-test) (Figure 8A). Thus, repair through NHEJ was not a major contribution to removal of double-strand breaks and did not account for the genotypic heterogeneity within colonies.

Yeast cells sometimes fail to separate fully after completing a round of cell division. To determine whether the genotypic heterogeneity within colonies might have resulted from incomplete separation of single cells at the time of plating, leading to colonies with more than one original founder cell, we used a micromanipulator to separate single cells after 1 hr of HO induction onto solid medium, ensuring that the resultant colonies were progeny of one cell. Sequencing results from these colonies showed that the appearance of partial gene conversion events persisted, yet were reduced relative to colonies plating by standard glass-bead–spreading (P-value = 0.0174 for partial gene conversion events anywhere, unpaired t-test) (Figure 8B). Thus, although incomplete separation after cell division contributed to the genetic heterogeneity seen within colonies, differences in sequence at the pseudo-MAT locus existed between founder cells and their progeny.

To further investigate the genetic heterogeneity observed within colonies, we analyzed the sequence at pseudo-MAT after a 1-hr HO induction in colonies from pedigrees of mother cells and their first two daughter cells. We collected sequencing data from 27 complete pedigrees, 25 of which showed no evidence of gene conversion in colonies arising from the mother or her first two daughters.

In the first complete pedigree with a gene conversion event, all three resultant colonies maintained silencing. The mother cell showed no gene conversion at SNP-L and full gene conversion at both SNP-INC and SNP-R. The first daughter showed no gene conversion at SNP-L or SNP-R, but full gene conversion at SNP-INC. The second daughter’s genotype matched that of the mother. Hence, the first cell division gave rise to one cell that did not experience gene conversion at SNP-R and one that did.

In the second complete pedigree with a gene conversion event, the mother and her first two daughters showed full gene conversion at all three SNPs. Interestingly, the mother cell gave rise to a colony that maintained silencing, the first daughter gave rise to a colony that was one-quarter green, and the second daughter gave rise to a colony that was all green. Thus, a single gene conversion event that occurred in the mother cell prior to the first division affected the likelihood of silencing loss in later cell divisions.

We observed one incomplete pedigree comprising a mother cell and only her first daughter cell, in which both resultant colonies were all green. The colony arising from the mother cell showed full gene conversion at all three SNPs, while the colony from the first daughter showed full gene conversion at SNP-L and SNP-INC but partial gene conversion at SNP-R. Hence, genotypic heterogeneity at pseudo-MAT could manifest after the first cell division.

Discussion

In this study, we developed an assay to test the consequences of homologous recombination within silent chromatin on the stability of gene silencing. We used the HO endonuclease gene under control of the GAL10 promoter to temporally control double-strand-break induction at pseudo-MAT, a locus that shared homology with HML::cre. We found that double-strand break repair templated from HML::cre increased the frequency of silencing loss events, although recombination was possible without disruptions to silencing even in repair events with extensive gene conversion tracts. Recombination-induced expression from HML::cre was rapid and lasted on the order of hours before silencing was reestablished. Double-strand-break induction and silencing-loss outcomes were not slightly affected in a swi2∆10R mutant background. Founder cells and their progeny did not always share the same sequence at the pseudo-MAT locus after gene conversion, leading to genetic heterogeneity within some colonies.

Silencing loss often, but not always accompanied homologous recombination

After induction of HO, many cells gave rise to colonies that had lost silencing in at least one cell by the four-cell stage of growth, resulting in colonies that were at least one-quarter green. This increase in silencing loss events was specific to strains in which a double-strand break was repaired from HML::cre, as homologous recombination not involving HML::cre did not change the likelihood that it would lose silencing.

In approximately one-third of colonies that did not lose silencing after 1 hr of HO induction, gene conversion occurred at the pseudo-MAT locus. Thus, homologous recombination targeted to HML::cre led to distinct outcomes where cells either did or did not lose silencing after homology-directed repair.

Why homologous recombination led to silencing loss in some cells but not others remains an interesting puzzle. It was unlikely that the extent of gene conversion tract determined whether silencing was lost, as gene conversion of the distant SNP-L occurred in ∼13% of colonies that did not lose silencing. In mating-type switching, homology between the double-strand break at MAT and the donor locus occurs only on one side of the double-strand break, ensuring that repair events proceed via a type of homologous recombination known as SDSA (Haber et al. 1980; Klar and Strathern 1984; Ira et al. 2006). However, multiple mechanisms of homologous recombination can repair a double-strand break [reviewed in Pâques and Haber (1997)]. Perhaps in our assay strain, recombination events between pseudo-MAT and HML::cre could be repaired by different mechanisms of homologous recombination, some of which led to silencing loss and some of which did not. Mating-type switching is restricted to the end of G1 (Strathern and Herskowitz 1979), whereas our experiments were conducted in cycling cells. Conceivably, the timing of either the double-strand break or the recombination event itself could influence whether or not the HML::cre donor locus lost silencing. For example, the sirtuin deacetylase Hst3 is a cell cycle–regulated protein important for maintaining silencing stability at HML (Dodson and Rine 2015) and is degraded after S-phase but before anaphase (Delgoshaie et al. 2014). Finally, it is also possible that the likelihood of silencing loss after recombination reflected stochasticity in the access of the transcription machinery to HML::cre.

Silencing loss was transient and limited

We monitored the kinetics and levels of transcripts from HML::cre before, during, and after HO induction by quantitative RT-PCR. Transcription at HML::cre was highest after completion of homologous recombination, perhaps because the presence of repair proteins prevented access of the transcription machinery to the donor locus until recombination was complete. In the cells that lost silencing, transcription at HML::cre resulting from homologous recombination likely did not approach the full level of derepression. It is unclear whether the green cells contributing to the colonies that were at least one-quarter green but not all green had lost silencing at the time of plating, or whether those silencing-loss events occurred postplating. It is likely that both of these scenarios are possible, either due to incomplete separation of red and green cells or due to a delay in silencing loss after repair. In the latter scenario, the all-green colonies alone would contribute to the pool of mRNA seen by quantitative RT-PCR within the first few hours. Even so, the increase in all-green colonies seen after HO induction to 5–10% of the population would have given rise to only 1% of cre mRNA amounts expected with full derepression, further supporting the conclusion that silencing loss did not approach full derepression.

Cre mRNA levels had decreased substantially 5 hr after quenching HO expression, but were still ninefold higher than before the induction. A longer time course would presumably show transcriptional silencing returning to its preinduction levels. Conceivably, some cells within the population may have undergone reciprocal recombination, giving rise to a functional cre gene that might not be fully silenced at the pseudo-MAT locus on the resulting dicentric chromosome. Although colonies resulting from such events were excluded from our plating-based analyses, the samples collected for quantitative RT-PCR analysis could include a small percentage cells expressing cre from such a restored pseudo-MAT locus.

Interactions between Swi2 and Sir3 influenced silencing stability during homologous recombination

Silent chromatin prevents access of many DNA binding proteins to the sequences within it, and previous studies established a role for ATP-dependent nucleosome remodelers in successful recombination between the MAT locus and its heterochromatic donors (Chai et al. 2005; Sinha et al. 2009; Tsukuda et al. 2009). In vitro, the SWI/SNF complex is necessary to allow pairing between Rad51-coated filaments and Sir3-coated nucleosomes in an assay that simulates recombination with silent chromatin (Sinha et al. 2009). Physical interactions between Sir3 and the ATPase subunit of the SWI/SNF complex, Swi2 (Manning and Peterson 2014), are necessary for successful in vitro synapsis between Rad51 filaments and Sir3-coated nucleosomal donors in that assay. An allele of SWI2, swi2∆10R, which abolishes the interaction between Sir3 and SWI/SNF, prevents in vitro pairing of Rad51 filaments with Sir3-bound nucleosomes.

In our assay, the swi2∆10R mutant slightly reduced the recombination frequency between pseudo-MAT and HML::cre, and recombination events were slightly more likely to lead to an early silencing-loss event. This suggests a minor role for the interaction between Swi2 and Sir3 in both promoting homologous recombination within HML::cre and preventing silencing loss after recombination. Because the effects seen in the swi2∆10R mutant were not severe, perhaps Sir3 is successfully removed from nucleosomes at HML in vivo by another nucleosome remodeling complex. Alternatively, Sir3 removal from nucleosomes may not be critical for homologous recombination between the pseudo-MAT locus and HML::cre, similar to the in vivo findings (Manning and Peterson 2014).

Double-strand-break induction kinetics

Kinetic studies of HO-mediated cleavage of the endogenous MAT locus, where it is possible to distinguish the uncut MATa and repaired MATα, report that full cleavage is achieved after ∼30 min of HO induction (Hicks et al. 2011), yet at no time did we see complete cleavage of the pseudo-MAT locus, and the percentage of cleaved pseudo-MAT molecules never exceeded 40% throughout the time course. In our study, DNA blot analysis could not distinguish molecules that had been cleaved and repaired from those that were never cleaved due to the sequence identity between the initial and repaired pseudo-MAT products. Thus, it was not clear whether double-strand break induction at our pseudo-MAT locus was slower than at the endogenous MAT locus, or whether cleavage kinetics were similar to those previously reported. It was also possible that the cleaved pseudo-MAT molecules seen at later time points reflect molecules that had been repaired by recombination but without gene conversion, leaving the HO recognition sequence intact for further rounds of cleavage. We would expect that after an adequate time of HO expression, all pseudo-MAT molecules would show gene conversion to the uncleavable SNP-INC sequence found at the donor HML::cre locus.

Genetic and phenotypic heterogeneity within colonies

Silencing loss and recombination events were assayed in colonies grown after HO induction, representing a temporal difference in the occurrence of these events and their detection. Interestingly, many colonies displayed genotypic and phenotypic heterogeneity regarding the repaired sequence at pseudo-MAT and the silencing loss from HML::cre.

Approximately half of gene conversion events at pseudo-MAT after a 1-hr HO induction were partial gene conversion events, in which both the original pseudo-MAT and the donor HML::cre sequences were detected in a single colony. Some of this genetic heterogeneity was likely due to the propensity of cells to remain attached after rounds of cell division, as micromanipulating single cells onto solid medium after HO induction to ensure a single founder cell reduced the frequency of colonies with partial gene conversion events. However, some of the separated cells still gave rise to colonies with partial gene conversion events, thus requiring other explanations. One possible explanation could be that recombination in the founder cell led to a mismatch at pseudo-MAT that was not resolved prior to replication, causing both alleles to be propagated in the colony. Two pedigrees that we observed support this model. In one of our complete pedigrees, the mother cell showed full gene conversion at SNP-R while the daughter cell showed no gene conversion at SNP-R, suggesting that repair leading to a mismatch happened prior to replication, and each cell inherited a chromosome with the opposite genotype. In our incomplete pedigree, the appearance of a partial gene conversion only in the daughter but not the mother suggests that the double-strand break and repair happened postreplication, and the chromosome inherited by the daughter contained a mismatch. Additionally, the partial gene conversion events may have resulted from differences in the way a founder cell experienced HO-mediated double-strand breaks or subsequent repair from its progeny. Perhaps the HO protein was asymmetrically distributed to either the mother or daughter cell, which could result in one cell experiencing a double-strand break but not the other. In this scenario, the genetic heterogeneity within a colony would result from different events in the first cells that form the colony rather than a single event that leads to two molecular outcomes. Although we did not observe an outcome in our pedigree analysis that supports this model, we cannot rule out that possibility.

As with the genetic heterogeneity seen within a colony, micromanipulating single cells onto solid medium after HO induction revealed some colonies that were at least one-quarter green but not completely green, reflecting silencing-loss events that occurred after the first cell division (Figure S1 in File S1). Our pedigree analysis showing gene conversion prior to the first cell division but silencing loss in the second and third cell divisions suggests that it is possible that homologous recombination with pseudo-MAT affected chromatin at the HML::cre locus in a way that did not manifest silencing loss immediately, resulting in a delay in cre transcription by one or two cell divisions. Perhaps the chromatin at HML::cre after homologous recombination was stable enough to immediately maintain silencing in some cells, but changes to epigenetic marks or another factor that influences silencing stability were not restored to their pre-recombination status. Despite these nuances, there was a clear causal link between double-strand-break repair template from HML::cre and subsequent silencing loss.

Implications beyond S. cerevisiae

In this study, homologous recombination targeted to HML led to an increased likelihood of transcriptional silencing loss in many, but not all, cells. The silencing loss that did occur at HML after homologous recombination was transient and limited, which likely resulted from the efficiency and rapid kinetics with which yeast cells can establish silencing de novo when all necessary components are available (Osborne et al. 2009). In organisms or genetic backgrounds that cannot rapidly reestablish transcriptional silencing within heterochromatic loci, homologous recombination might alter chromatin structures in a way that has long-lasting effects on the transcriptional output of such loci. Conceivably, other structures of chromatin may also be vulnerable to disruption by homologous recombination. This possibility may become evident with the increase in frequency of Cas9-mediated genome editing studies.

Acknowledgments

We thank Ryan Janke and Martin Kupiec for invaluable insights into the design of multiple experiments in this study. We thank Anne Dodson for thoughtful input and strain contribution, including construction of strain JRY10822. We thank Craig Peterson for providing the swi2∆10R construct, and the Deuber laboratory at University of California at Berkeley for the plasmid enabling Cas9-mediated strain construction. We thank Oskar Hallatschek for use of the Zeiss Axio Zoom.V16, Carlos Bustamante for use of the Typhoon Trio imager, and the University of California at Berkeley DNA Sequencing Facility for preparation of samples for sequencing. We thank Doug Koshland, Elçin Ünal, Dana Carroll, and Kathleen Ryan for helpful discussion. Finally, we thank members of our laboratory, Aisha Ellahi, Melanie Kaiser, Gavin Schlissel, Davis Goodnight, and Daniel Saxton, for thoughtful discussion and technical guidance. This work was supported by a grant from the National Institutes of Health (GM31105), and by a National Institutes of Health National Research Service Award Trainee appointment on grant number T32 HM 007232.

Footnotes

Literature Cited