Mating type in Saccharomyces cerevisiae is determined by two nonhomologous alleles, MATa and MATα. These sequences encode regulators of the two different haploid mating types and of the diploids formed by their conjugation. Analysis of the MATa1, MATα1, and MATα2 alleles provided one of the earliest models of cell-type specification by transcriptional activators and repressors. Remarkably, homothallic yeast cells can switch their mating type as often as every generation by a highly choreographed, site-specific homologous recombination event that replaces one MAT allele with different DNA sequences encoding the opposite MAT allele. This replacement process involves the participation of two intact but unexpressed copies of mating-type information at the heterochromatic loci, HMLα and HMRa, which are located at opposite ends of the same chromosome-encoding MAT. The study of MAT switching has yielded important insights into the control of cell lineage, the silencing of gene expression, the formation of heterochromatin, and the regulation of accessibility of the donor sequences. Real-time analysis of MAT switching has provided the most detailed description of the molecular events that occur during the homologous recombinational repair of a programmed double-strand chromosome break.
SACCHAROMYCES cerevisiae is a budding yeast that propagates vegetatively either as MATa or MATα haploids or as MATa/MATα diploids created by conjugation of opposite haploid types (Figure 1). Mating type is determined by two different alleles of the mating-type (MAT) locus. Like many other fungi, budding yeast has acquired the capacity to convert some cells in a colony from one haploid mating type to the other (Figure 1). This process is called homothallism. The subsequent mating of cells to the opposite mating type enables these homothallic organisms to self-diploidize. The diploid state provides yeast with a number of evolutionarily advantageous strategies unavailable to haploids, most notably the ability to undergo meiosis and spore formation under nutritionally limiting conditions. Mating-type gene switching in S. cerevisiae is a highly choreographed process that has taught us much about many aspects of gene regulation, chromosome structure, and homologous recombination.
The MAT locus lies in the middle of the right arm of chromosome III, ∼100 kb from both the centromere and the telomere. The two mating-type alleles, MATα and MATa, differ by ∼700 bp of sequences designated Yα and Ya, respectively (Figure 2). Yα and Ya contain the promoters and most of the open reading frames for proteins that regulate many aspects of the cell’s sexual activity (for reviews, see Klar 1987; Herskowitz 1988; Strathern 1988; Haber 1992, 1998, 2006, 2007). The MAT locus is divided into five regions (W, X, Y, Z1, and Z2) on the basis of sequences that are shared between MAT and the two cryptic copies of mating-type sequences located at HMLα and HMRa (Figure 1). HMLα and HMRa serve as donors during the recombinational process that allows a MATa cell to switch to MATα or vice versa.
Functions of the MAT Proteins
MATα encodes two proteins, MATα1 and MATα2. MATα1, in conjunction with a constitutively expressed protein, Mcm1, activates a set of α-specific genes (Klar 1987; Hagen et al. 1993; Bruhn and Sprague 1994), including those encoding the mating pheromone, α-factor, and Ste2, a trans-membrane receptor of the opposite mating pheromone, a-factor (Figure 3). As noted above, these mating pheromones trigger G1 arrest of the budding yeast cell cycle and facilitate conjugation, ensuring that the zygote will contain two unreplicated nuclei. MATα2 encodes a homeodomain helix-turn-helix protein that acts with Mcm1 to form a repressor that binds to a 31-bp symmetric site with Mcm1 in the center and Matα2 at the ends (Smith and Johnson 1992). Matα2-Mcm1 represses a-specific genes including those that produce a-factor (MFa1and MFa2) and the Ste3 transmembrane receptor protein that detects the presence of α-factor in the medium. Repression also requires the action of Tup1 and Ssn6 proteins (Keleher et al. 1989; Herschbach et al. 1994; Patterton and Simpson 1994; Smith and Johnson 2000).
When the bidirectional promoter controlling MATα1 and MATα2 is deleted (or if MAT is entirely deleted), haploid cells have an a-like mating behavior (e.g., they mate identically to MATa cells), because a-specific genes are constitutively expressed in the absence of Matα2 and α-specific genes are not transcribed in the absence of Matα1 (Strathern et al. 1981). But although MATa is not required for a-mating, the MATa1 gene is required, along with MATα2, to extinguish haploid-specific gene expression and to manifest diploid-specific attributes. A MATa/MATα cell is nonmating, whereas the diploid resulting from mating matΔ and MATα is α-mating. The nonmating phenotype of MATa/MATα diploids results from the action of a very stable corepressor of Mata1 and Matα2 proteins (Jensen et al. 1983; Goutte and Johnson 1988; Strathern 1988; Li et al. 1995; Johnson et al. 1998; Tan and Richmond 1998). This repressor turns off a set of haploid-specific genes and allows expression of diploid-specific genes.1 The a1-α2 repressor turns off transcription of MATα1, the activator of α-specific genes, but allows expression of MATα2, the repressor of a-specific genes; hence the diploid is nonmating.
The MATa locus actually has two divergently transcribed open reading frames, MATa1 and MATa2; but although MATa2 is remarkably well conserved evolutionarily, it has yet to be assigned a biological function (Tatchell et al. 1981). MATa2 and MATα2 share part of the same protein sequence as the transcript extends into the MAT-X region and the evolutionary preservation of MATα2 may account for some of the conservation of MATa2, despite its lack of apparent function.
Haploid-Specific and Diploid-Specific Genes Under MAT Control
There are a number of important mating-type–dependent differences. These distinctions are not simply a question of haploidy vs. diploidy: MATa/MATα diploids are notably different from diploids homozygous for either MATa or MATα. First of all, they are nonmating. Second, MATa/MATα cells can initiate meiosis and spore formation, whereas a-mating or α-mating diploids cannot. The ability to enter the meiosis and sporulation program is controlled by the repression of RME1 (repressor of meiosis 1) by the a1-α2 repressor. If RME1 is deleted, then a MATa/MATa or a matΔ/MATα diploid can undergo meiosis and produce viable spores.
Another key gene under mating-type control is NEJ1, which encodes a necessary component of nonhomologous end joining (NHEJ) (Frank-Vaillant and Marcand 2001; Kegel et al. 2001; Ooi and Boeke 2001; Valencia et al. 2001). Double-strand breaks (DSBs) in chromosomes can be repaired either by homologous recombination (HR) or by nonhomologous end-joining (NHEJ) (reviewed by Pâques and Haber 1999; Haber 2006). In haploids, both processes are efficient; for example a DSB at MAT created by the HO endonuclease is repaired ∼90% of the time by HR, using HML or HMR as the donor, but ∼10% of cells use NHEJ to religate the DSB ends, recreating the cleavage site.2 But if cells are arrested in the G1 phase of the cell cycle, by treating MATa cells with α-factor, NHEJ is the predominant pathway. These arrested cells fail to activate the critical early steps of HR because the Cdk1 protein kinase is inactive (Ira et al. 2004; Aylon and Kupiec 2005). Nevertheless, in MATa/MATα diploids, NHEJ is turned off by a1-α2 repression of the NEJ1 gene and by the partial repression of another NHEJ component, LIF1. The idea that NHEJ is repressed because diploids have two copies of each chromosome and therefore can always repair from a homolog is hard to accept if indeed HR is impaired in G1 cells.3 Perhaps NHEJ would be a dangerous pathway in meiosis in MATa/MATα cells, where there are ∼100 DSBs to generate meiotic crossovers.
Some other aspects of DNA repair are under mating-type control (Friis and Roman 1968; Heude and Fabre 1993; Nickoloff and Haber 2000; Valencia-Burton et al. 2006; Fung et al. 2009). MATa/MATα diploids are substantially more resistant to ionizing radiation than are MATa/MATa or MATα/MATα cells. In addition, spontaneous rate of recombination, for example between alleles of various biosynthetic genes, is higher in MATa/MATα cells. Both rad55Δ and rad57Δ mutants are partly defective in recombinational repair of DNA damage, being X-ray sensitive at lower temperatures, but this defect is suppressed in haploids in which both MATa and MATα are expressed (termed mating-type heterozygosity). rad55Δ is also sensitive to the topoisomerase inhibitor, camptothecin; this sensitivity is partially suppressed by mating-type heterozygosity and can be mimicked by knocking out NHEJ (Valencia-Burton et al. 2006). In a similar fashion, defective alleles of recombination proteins Rad52 (rad52-327) and Rad51 (rad51-K191R) can be suppressed by mating-type heterozygosity, apparently through repression of other genes including Pst2 or Rfs1, respectively. Why it would be advantageous for diploid cells to overcome some of the defects caused by the Rad mutants is not evident to this reviewer.
Another haploid/diploid distinction is found in the pattern of bud formation (Figure 1): haploids or diploids expressing only one MAT allele exhibit an axial pattern of budding that appears to be designed to facilitate efficient mating in homothallic cells (see below), while nonmating MATa/MATα diploids (or haploids expressing both mating-type genes) have a bipolar budding pattern (Chant 1996). Axial budding depends on the Axl1 protein that is expressed in haploids but not diploids. Axl1 localizes to the constriction at the mother–bud neck and, in conjunction with a number of other gene products (Bud3, Bud4, Bud5, and Bud10), establishes the emergence of a bud in this position (Chant and Pringle 1991; Lord et al. 2002).
Finally, MAT heterozygosity plays a key role in the switching of mating-type genes. Homothallic strains expressing the HO endonuclease gene, HO, are able to switch from MATa to MATα and vice versa, but once cells of opposite mating type conjugate to form a diploid, HO expression is repressed, again by the a1-α2 repressor.
The phenotypic switch from MATα to MATa is quite rapid—within a single cell division. For example if one places MATα cells in the presence of the mating pheromone, α-factor, they will grow because they are insensitive to the pheromone, but if one plates HO MATα spores, then by the time the cell has divided (and switched to MATa) the mother and new daughter cells are unable to progress beyond the G1 phase of the cell cycle, because they are now arrested by α-factor. Thus the cell must have the capacity to shift rapidly from expressing α-specific genes to expressing the a-specific program. Consequently the newly switched MATa cell should turn off expression of the Ste3 pheromone receptor for a-factor and should also stop producing α-factor; at the same time it must quickly be able to respond to α-factor by inserting Ste2 into the cell membrane, while at the same time start pumping out a-factor. Hence it is not surprising that Mata1, Matα1, and Matα2 transcription regulators are quite rapidly turned over, degraded by ubiquitin-mediated proteolysis by the proteasome (Laney and Hochstrasser 2003; Laney et al. 2006). In contrast, the a1-α2 corepressor is much more stable (Johnson et al. 1998).
Mating-Type Switching: a Model of Cell Lineage, Gene Silencing, and Programmed Genomic Rearrangement
S. cerevisiae has evolved an elaborate set of mechanisms to enable cells to switch their mating types. Learning how these processes work has provided some of the most fascinating observations in eukaryotic cell biology. MAT switching depends on four phenomena: (1) the presence of two unexpressed (silenced) copies of mating-type sequences that act as donors during MAT switching; (2) the programmed creation of a site-specific double-strand break at MAT that results in the replacement of Ya or Yα sequences; (3) a cell lineage pattern that ensures that only half of the cells in a population switch at any one time, to ensure that there will be cells of both mating types in close proximity; and (4) a remarkable mechanism that regulates the selective use of the two donors (donor preference). Each of these important mechanisms is reviewed below.
Silencing of HML and HMR
The presence of intact, but unexpressed copies of mating-type genes at HML and HMR implied that these two loci had to be maintained in an unusual, silent configuration. The study of the mechanism of silencing of these donors has occupied the attention of many labs and has provided some important insights into the way in which chromatin structure influences gene expression and recombination (see reviews by Laurenson and Rine 1992; Loo and Rine 1994; Sherman and Pillus 1997; Aström and Rine 1998; Rusche et al. 2003; Hickman et al. 2011). Our current understanding can be summarized as shown in Figure 4A. Both HML and HMR are surrounded by a pair of related but distinct silencer sequences, designated HML-E, HML-I, HMR-E, and HMR-I. These cis-acting elements interact, directly or indirectly, with several trans-acting factors to repress the transcription of these genes (Figure 4A). Among the trans-acting proteins that play critical roles in this process are four Silent Information Regulator (Sir) proteins, a set of silencer binding proteins, histone proteins, the multipurpose Rap1 protein, as well as several chromatin modifiers. Together, these gene products and cis-acting sequences create short regions (∼3 kb) of heterochromatin, in which the DNA sequences of HML and HMR are found as a highly ordered nucleosome structure (Nasmyth 1982; Weiss and Simpson 1998; Ravindra et al. 1999) (Figure 4B). These heterochromatic regions are transcriptionally silent for both PolII- and PolIII-transcribed genes (Brand et al. 1985; Schnell and Rine 1986) and resistant to cleavage by several endogenously expressed endonucleases, including the HO endonuclease (Connolly et al. 1988; Loo and Rine 1994).
It should be noted that silencing also occurs adjacent to yeast telomeres and many of the genes involved in HML/HMR gene silencing also play a role in telomeric silencing (see reviews by Laurenson and Rine 1992; Loo and Rine 1994; Sherman and Pillus 1997; Grunstein 1998; Lustig 1998; Stone and Pillus 1998; Gasser and Cockell 2001; Rusche et al. 2003; McConnell et al. 2006; Hickman et al. 2011). There is a hierarchy of silencing, with HMR and HMR being more strongly silenced than telomeres. In general, telomeric silencing is less robust; several mutations that strongly affect telomeric silencing (e.g., yku70Δ (Moretti et al. 1994; Wotton and Shore 1997; Vandre et al. 2008) have either no effect on HM loci or have an effect only with a partially debilitated HMR-E sequence.
Cis-acting silencer sequences
There appear to be some important differences in the silencing of HML and HMR. Analysis of HMR-E (essential) and HMR-I (important) sequences showed that HMR-E alone was sufficient for silencing of the a1 transcript at HMRa or of other PolII- or PolIII-transcribed genes inserted in place of the Ya region, whereas HMR-I can not silence completely without HMR-E (Abraham et al. 1984; Hicks et al. 1984; Brand et al. 1985). Mutations of HMR-E that weaken silencing can be “tightened” by the presence of HMR-I. In contrast, either HML-E or HML-I is each sufficient to silence HML or other genes placed nearby (Mahoney and Broach 1989). Silencing also appears to be enforced by the fact that both HMR and HML lie relatively near chromosome ends (telomeres) that also exhibit gene silencing. When HML-E or HMR-E silencer sequences are inserted at other chromosome locations, further from telomeres, their ability to silence various adjacent genes is less strong (Thompson-Stewart et al. 1994; Shei and Broach 1995; Maillet et al. 1996; Marcand et al. 1996). This may explain why a circular chromosomal fusion of MATa and HMLα (i.e., containing HML-I but lacking HML-E and lacking telomeres) is expressed (Strathern et al. 1979).
The distance over which E and I silencers can act to completely silence genes is not very great. Normally the distance between E and I is <3 kb. Silencing is weakened if that distance is increased. For example, if most of the mating-type gene sequences lying between HMR-E and HMR-I are deleted and replaced by a 2.2 kb LEU2 gene fragment, LEU2 is completely silent; however if the same LEU2 gene is simply inserted into the middle of the mating-type sequences, thus moving HMR-E and HMR-I further apart, there is sufficient LEU2 gene expression to allow leu2 cells to grow (B. Connolly and J. E. Haber, unpublished results). Similarly, Weinstock et al. (1990) discovered that tandem insertions of the Ty1 retrotransposon within HML unsilenced the locus, with the degree of expression correlated to the size of the array and thus the distance between E and I sites.
At telomeres, there are no specific silencer sequences, but the telomere-associated Rap1 protein interacts with both Sir3 and Sir4 (Moretti et al. 1994; Wotton and Shore 1997; Mishra and Shore 1999). Moreover telomere termini are also enriched in yKu70-yKu80, which also recruit Sir4 (Roy et al. 2004; Ribes-Zamora et al. 2007). Telomeric silencing can extend >10 kb (Strahl-Bolsinger et al. 1997), but strong silencing is confined to the first 1–2 kb (Rusche and Lynch 2009). Part of the way silencing is established seems to be by tethering sequences to the nuclear periphery, as exemplified by Sternglanz’ experiment of artificially tethering a partially crippled HMR locus to the nuclear membrane by use of a Gal4-fusion protein that contained an integral membrane protein that bound to a Gal4-binding domain as part of the HMR-E silencer (Roy et al. 2004). However silencing is not dependent on peripheral localization (Gartenberg et al. 2004). Overexpression of Sir3 has the surprising effect of causing telomere delocalization toward the center of the nucleus but silencing is stronger (Ruault et al. 2011).
A striking feature of all four silencer sequences is that each is capable of acting as an autonomously replicating sequence (ARS) on a plasmid, thus allowing it to replicate (Abraham et al. 1984; Feldman et al. 1984; Kimmerly et al. 1988). Yet when one examines these same sequences on the chromosome by two-dimensional gel electrophoresis, to detect structures characteristic of the presence of an origin of DNA replication, neither HML-E nor HML-I is active, while HMR-E does appear to act as a chromosomal origin, although only in a fraction of cell cycles (Dubey et al. 1991; Rivier and Rine 1992; Collins and Newlon 1994). Whether silencing depends on origin activity has been the subject of much debate; current evidence suggests that the binding of the ORC proteins to silencer regions is a key step in establishing silencing, but it is not necessary that replication be initiated at that site (Fox et al. 1993; Ehrenhofer-Murray et al. 1995; Fox et al. 1997; Li et al. 2001). Indeed mutations of Orc5 were isolated that were defective in silencing but not in replication (Fox et al. 1995). A notable experiment in this regard was the demonstration by Sternglanz’ lab that the ORC sequence could be replaced by a Gal4-binding domain to recruit Gal4-Sir1 in the apparent absence of ORC proteins to establish silencing (Chien et al. 1993; Fox et al. 1997).
Another interesting aspect of silencer sequences is that they confer centromere-like behavior on plasmids, allowing them to segregate properly in most cell divisions (Kimmerly et al. 1988). One explanation for this behavior is that silencer sequences may be anchored to some nuclear structure that is involved in chromosome partitioning. Evidence that silencers are anchored has been presented in studies of scaffold-attachment regions by Amati and Gasser (1988) and Towbin et al. (2009) and in elegant topological experiments by Ansari and Gartenberg (1997) showing that DNA bound by a lexA-Sir4 fusion protein is anchored within the nucleus. However, whereas deleting yKu partially releases telomeres from their peripheral association, it appears to allow a stronger association of HML with the nuclear envelope. Only when both yKu proteins and either Esc1 or Sir4 are deleted does one see a release of both telomeres and HML from the periphery (Bystricky et al. 2009; Miele et al. 2009).
The most detailed dissection of a silencer has been carried out with HMR-E. A combination of deletion analysis and protein-binding experiments has demonstrated that HMR-E contains three distinct subdomains (Figure 4A): ARS consensus sequence to which ORC proteins bind, a binding site for the ARS-binding factor (ABF1) protein, and a binding site for Rap1, which plays a complex role in both gene activation and in gene silencing (McNally and Rine 1991). Deletion of any one of these three regions still allows substantial silencing, but deletion of any two removes all repression of transcription. As noted above, the ORC-binding domain can be replaced by a lexA-binding domain to serve as a target to localize a lexA-Sir1 fusion protein, suggesting that one important role for ORC binding is in tethering another protein such as Sir1 to facilitate the establishment of silencing (Chien et al. 1993; Fox et al. 1997). The arrangement of DNA-binding sites varies at each silencer; for example HMR-I does not contain a Rap1-binding domain.
Control of the spreading of silencing by boundary elements
Silencing is quite restricted around HML and HMR, though it extends ∼1 kb beyond the E and I sequences. At HMR-I the spread of silencing is blocked by a tRNA gene (Donze et al. 1999; Dhillon et al. 2009), but even when the tRNA is deleted, silencing does not spread much further (Lynch and Rusche 2010). At HML-E there is also a nearby sequence but its precise role has not been clarified (Bi 2002). In the case of HML-I, it seems that the silencer itself somehow has directionality so that inverting this region promotes longer silencing outside the HML locus and weakens silencing inside HML (Bi et al. 1999). Further study has shown that directionality can be attributed to the asymmetric positioning of nucleosomes around the silencer (Zou et al. 2006). A clever and more general search for silencer boundary elements was designed by Laemmli in which ADE2 and URA3 genes were placed inside HMR but with a pair of Gal4-binding sites flanking ADE2 (Ishii et al. 2002). Initially both genes are silenced but the binding of a putative boundary element to both Gal4 sites should isolate ADE2, allowing its expression but leaving URA3 still silent. The screen found a number of nuclear pore complex proteins and exportins; so, here, tethering to the nuclear periphery seems to prevent an extension of silencing.
Trans-acting silencing proteins
The establishment and maintenance of silencing requires several proteins. The first trans-acting silencing gene was identified by Klar et al. (1979a) based on the fact that coexpression of both mating types in a haploid cell produced a nonmating phenotype. Eventually four SIR genes were identified (Haber and George 1979; Rine et al. 1979; Rine and Herskowitz 1987). A deletion of three of these—SIR2, SIR3, and SIR4—completely abolished silencing, while loss of SIR1 had a less extreme phenotype (see below). The keystone of these silencing proteins is the NAD+-dependent Sir2 histone deacetylase, which is responsible for deacetylating a number of lysines on the N-terminal tails of histones H3 and H4 (Imai et al. 2000). Sir3 exhibits homology with the origin binding protein Orc1 and has a nucleosome binding BAH (bromo adjacent homology) domain (Wang et al. 2004; Gallagher et al. 2009; Hickman and Rusche 2010). None of the Sir proteins binds directly to DNA, but they interact with each other, with Sir3 and Sir4 directly interacting and Sir4 binding to Sir2 (Moazed et al. 1997). Sir3 and Sir4 bind to deacetylated histone H3 and H4 tails (Johnson et al. 1990; Hecht et al. 1995, 1996; Grunstein 1997; Gasser and Cockell 2001). The acetylated form of histone H4K16 promotes Sir2-Sir4 binding (which might promote deacetylation), while the acetylated form has a reduced affinity for Sir3 (Oppikofer et al. 2011). These interactions set up the basis for a processive silencing mechanism in which Sir2 deacetylates the adjacent histone (Figure 4A), allowing Sir3 to bind, and so on, propagating an array of highly ordered (heterochromatic) nucleosomes (Grunstein 1997; Moazed 2011). Sir4 also interacts with yKu70 (Tsukamoto et al. 1996), which is important in telomere associations with the nuclear periphery (Taddei et al. 2004).
Sir proteins play a number of other roles
Sir2, but not Sir3 and Sir4, acts to reduce the expression of PolII-transcribed reporter genes embedded in rDNA and reduce recombination in rDNA (Gottlieb and Esposito 1989; Fritze et al. 1997; Smith and Boeke 1997; Smith et al. 1998). Overexpression of Sir3p or Sir4p also causes a significant increase in chromosome instability (Holmes et al. 1997), which might have something to do with its effects on telomeres and their position in the nucleus (Palladino et al. 1993). The striking discovery that the NAD+-regulated SIR2 HDAC affects yeast lifespan (Kennedy et al. 1995; Smeal et al. 1996; Gotta et al. 1997) was attributed to the formation and asymmetric segregation of extrachromosomal rDNA circles (Sinclair and Guarente 1997), which is unlikely to be a general mechanism of regulating longevity, but, remarkably Sirtuins (proteins homologous to Sir2) have proven to play key roles in longevity in metazoans and to affect many other aspects of cell metabolism, including autophagy (reviewed by Blander and Guarente 2004; Donmez and Guarente 2010; Herranz and Serrano 2010).
Our understanding of Sir2 has been enlivened by the discovery that there are four other genes exhibiting homology to sir2 (HST genes) (Brachmann et al. 1995; Derbyshire et al. 1996). Individual deletions of these HST genes do not affect silencing with an intact HMR, but hst3 and hst4 mutants partially derepress telomere silencing and overexpression of HST1 suppresses a sir2 mutation in HM gene silencing. Moreover hst3 and hst4 mutations are also radiation sensitive and show significant increases in chromosome instability (Brachmann et al. 1995). Recent studies have shown that these two histone deacetylases are required to remove Rtt109-mediated acetylation of histone H3-K56, an event that is normally strongly cell-cycle regulated. Without Hst3 and Hst4, cells fail to replicate properly and have greatly reduced viability. Removing the acetylation site or deleting Rtt109 suppresses the lethality of hst3Δ hst4Δ (Yang et al. 2008).
Establishment and maintenance of silencing
Regulation of acetylation of the N-terminal tails of histones H3 and H4 is directly implicated in silencing, first by mutations that replace the four evolutionarily conserved lysine residues (Megee et al. 1990; Park and Szostak 1990; Thompson-Stewart et al. 1994; Fisher-Adams and Grunstein 1995; Hecht et al. 1995). More direct evidence came from the fractionation of chromatin4 in terms of the state of acetylation of lysine-16 of histone H4 (Braunstein et al. 1996), showing that HML and HMR are preferentially recovered in the hypoacetylated fraction. A similar analysis, taking advantage of affinity chromatography that detected an alteration in histone H3, had also shown a difference in chromatin structure between MAT and the two silent loci (Chen-Cleland et al. 1993).
The use of modified histones H3 and H4 has also revealed dramatic differences between HML and HMR silencing. A single H4-K16A mutation strongly unsilences HMLα but has little effect at HMRa, and even mutation of four lysines in the H4 tail only weakly affects silencing at HMR (Park and Szostak 1990). Some of this difference reflects chromosomal context; however, it seems that silencing at HMR is made substantially stronger by its chromosome location near the right end of chromosome III (although HMR is 23 kb from its telomere compared to 12 kb for HML on the left arm). If both HML and HMR are inserted ∼50 kb from a telomere on chromosome VI, both HML and HMR show equivalent silencing defects with a histone H4-K16N mutation (Thompson et al. 1994). Quite possibly there are “booster” sequences (other Rap1- or Abf1-binding sites or even ARS sequences) in the vicinity of HMR that make it so much more silent. A large deletion of the H4 N-terminal tail does unsilence HMR as well as HML (Park and Szostak 1990). In contrast, a tail deletion of histone H3 causes a loss of silencing at HMLα, but the same mutations have no effect on HMR silencing, whether HMRa or HMRα. For example, in a hmlΔ matΔ HMRα strain that is a-like in its mating phenotype, deleting histone H3’s N-terminal tail does not change the mating phenotype from a-like to α-mating behavior, as would be expected if HMRα were expressed even at a low level (J. A. Kim, Q. Wu, and J. E. Haber, unpublished results).
Silencing at HML also requires the protein transacetylases encoded by NAT1 and ARD1, whose target appears to be the N terminus of Sir3 (Wang et al. 2004). The Nat1/Ard1 complex plays many roles in addition to silencing, including a failure to arrest in G1 after starvation (Whiteway et al. 1987; Mullen et al. 1989). Deletion of Ard1 (and presumably Nat1) also fails to unsilence HMRa or HMRα, while allowing HML to be strongly expressed (Whiteway et al. 1987). Silencing is influenced by two other protein transacetylases encoded by SAS2 and SAS3 (Reifsnyder et al. 1996; Ehrenhofer-Murray et al. 1997). The pleiotropic deletion of the RPD3 deacetylase also influences silencing (De Rubertis et al. 1996; Rundlett et al. 1996).
At least some of the difference in silencing at HML and HMR comes from the fact that the two cassettes normally carry different Y sequences. As further discussed below, silencing of HMR requires passage through S phase (but not replication per se). However, “substantial silencing of HMLα” could be established without passage through S phase (Ren et al. 2010). These authors attributed this difference in promoter strength—with the α1-α2 promoter being weaker than a1-a2—because HMRα is partially silenced without S phase, while HMLa requires S phase passage.
One key question in silencing is whether the establishment of silencing differs from its subsequent maintenance. Although tethering Sir1 in place of ORC will indeed establish silencing (Chien et al. 1993; Fox et al. 1997), in a wild-type cell its role is more subtle. Cells carrying a sir1 deletion exhibit a striking epigenetic variegation in HML silencing (Pillus and Rine 1989). Some cells express HML and others do not, but each state is persistent through many cell divisions. A sir1 cell with HML in a silent state gives rise predominantly to silent HML for many generations, but occasionally a cell will arise where HML is not silenced, and this unsilenced state will also persist for many generations, until a cell reestablishes silencing, and so on. This epigenetic inheritance suggests that the establishment of silencing and its subsequent maintenance are separable, an idea that is strongly supported by other observations (reviewed by Stone and Pillus 1998). The epigenetic nature of silencing has been more directly visualized by Xu et al. (2006), who inserted URA3::GFP inside HML and URA3::CFP within HMR. Strikingly in a sir1Δ cell, the silent states of HML and HMR were independent of each other (that is, some cells expressed only CFP while others expressed only GFP and other expressed both, or neither). Moreover, there was a significant difference between the silencing of HML and HMR, as ∼60% of cells expressed HML::YFP, whereas ∼90% of all cells expressed HMR::CFP. By reducing the strength of URA3 expression, with a ppr1Δ mutation, they found that silencing was nearly normal in sir1Δ for HML::YFP but still HMR::CFP was expressed in most cells. This result is striking in part because, as discussed earlier, many other mutations more strongly unsilence HML than HMR. It is also curious that sas2Δ sir1Δ completely loses silencing, whereas sas2Δ suppresses sir1Δ’s defect at HMR (Xu et al. 1999).
The idea that tethering HML or HMR to the nuclear periphery enhances silencing has been substantiated by the finding that the silencing that does occur in a sir1Δ is dependent on the yKu70 and YKu80 proteins (Patterson and Fox 2008; Vandre et al. 2008). The Ku proteins, along with Esc1, redundantly tether telomeres and HM loci to the nuclear periphery (Taddei et al. 2004, 2005) and this sequestering may place the HM locus near a higher concentration of Sir2 and other silencing factors. However Bystricky et al. (2009) has reported that HML, but not HMR, becomes more tightly associated with the nuclear envelope, though this was not assessed in sir1Δ.
An early study by Miller and Nasmyth (1984) showed that raising a temperature-sensitive sir3 mutant to its restrictive temperature caused immediate loss of silencing, but returning cells to their permissive temperature did not restore silencing until cells had passed through the next S phase. However this experiment does not mean that DNA replication per se is required; silencing may only require an enzyme whose synthesis is confined to the S phase of the cell cycle (Holmes and Broach 1996; Bi and Broach 1997; Kirchmaier and Rine 2001; Li et al. 2001). But additional experiments argue that full silencing requires cells to progress through mitosis (Lau et al. 2002; Martins-Taylor et al. 2004, 2011).
Site-specific recombination has been used to “pop out” a DNA circle from HML either with or without a silencer sequence. When silent chromatin was popped out such that the excised circle lacked E and I silencers, silencing could not be maintained, even in G1- or G2-arrested cells (Cheng and Gartenberg 2000). Silencing was lost if cells passed though S phase. Surprisingly, the loss of silencing is apparently not directly caused by the partitioning of nucleosomes to newly synthesized DNA, as the popped-out circle does not contain an origin of replication. This result suggests that some events in maintaining silencing depend on progressing through the S phase part of the cell cycle but are not intimately involved in replication itself. Nevertheless, it is entirely possible that, in normal cells, when replication does occur, the maintenance of silencing is closely connected to chromatin assembly. Enomoto and Berman (1998) further showed that a deletion of the Cac1 subunit of chromatin assembly factor 1 (CAF-1) affects the maintenance of HML silencing but apparently not its establishment. It has been suggested that cac1Δ and asf1Δ have the global effect of reducing histone acetylation, which in turn might attract Sir2 away from the silent loci to many other loci (Rusche et al. 2003).
The list of genes affecting silencing continues to grow, although their contributions are relatively minor. Two mutations affecting the ubiquitylation of proteins affect silencing (Moazed and Johnson 1996; Huang et al. 1997). The case of Ubp3 protein is especially interesting, as it has been reported to be affinity purified in a complex with Sir4 (Moazed and Johnson 1996). Uls1—a member of the Swi2/Snf2 family of proteins implicated in chromatin remodeling—has been shown to tighten silencing when it is deleted and weaken silencing when a truncated gene is overexpressed (Zhang and Buchman 1997). Similar phenotypes have been found for still another antisilencing factor (ASF1) (Sharp et al. 2001, 2005). The absence of Asf1 or Rtt109 leads to a loss of histone H3-K56 acetylation and in some way this change alters the efficiency of Sir2-mediated silencing. Similarly, Dot1 methyltransferase, which modifies histone H3-K79, also plays some minor role in HM gene silencing (Takahashi et al. 2011). In addition temperature-sensitive mutations in two essential genes, NLP3 and YCL054, also perturb silencing by unknown means (Loo and Rine 1995). The CDC7 gene, encoding a protein kinase necessary to initiate chromosomal DNA replication, also plays a role in silencing (Axelrod and Rine 1991), though how this relates to the role of ORC proteins remains unknown. Finally, recent studies have shown that the MAP kinase pathway(s) responsible for mating pheromone, starvation, and heat-shock response all cause the hyperphosphorylation of Sir3 (Stone and Pillus 1996), though how phosphorylation affects normal silencing is not yet established.
Further analysis of silencing has suggested that the full establishment is more complex than had been imagined. It takes several generations for the complete silencing of HMR (Katan-Khaykovich and Struhl 2005; Kirchmaier and Rine 2006), although 90% of the transcriptional repression is accomplished in the first cell cycle. These results were supported by Xu et al. (2006), who used single-cell analysis of strains carrying GFP inserted inside HML, allowing them to use microscopic analysis of silencing in individual cells. Their work showed that the establishment of silencing after returning sir3-ts cells to their permissive temperature was stochastic, and that silencing of all cells took several generations.
Rine’s lab has also studied silencing at the microscopic level in single cells when Sir3 is turned on (Osborne et al. 2009). An α-mating sir3Δ strain carrying HMLα matΔ hmrΔ is mated to an a-like strain lacking all three mating loci, but SIR3. When two G1 cells conjugate, a zygote is formed that initially expresses MATα1 and MATα2 (from HMLα), but as this locus is silenced, the zygote should become inhibited in its growth by α-factor. This clever assay demonstrated that silencing was accomplished in 1–2 generations. In a later assay, Rine’s group used GFP embedded in HML in a similar assay to ask how rapidly GFP intensity decreased (the assay relies on GFP turning over rapidly) (Osborne et al. 2011). These assays revealed “unexpected complexity” in the contributions of a histone acetyltransferase (Sas2), two histone methytransferases (Dot1 and Set1), and one histone demethylase (Jhd2) to the dynamics of silencing and suggested that removal of methyl modifications at histone H3-K4 and -K79 were important steps in silent chromatin formation and that Jhd2 and Set1 had competing roles in the process.
Despite intensive work in this area, there are still many mysteries yet to solve. Moreover, two recent articles have called into question some previously published studies examining silencing at crippled HM silencers or at telomeres. First, it seems that in contrast to the most frequently used telomere silencing construct created at a truncation at the left arm of chromosome VII, most subtelomeric regions do not show strong position effect variegation and do not show much change in gene expression when Dot1 is absent (Takahashi et al. 2011). Moreover, the effects of some mutations such as alterations of PCNA, may not actually affect normal silencing. In many studies, URA3 has been used as a reporter gene, but it now appears that its inhibitor, 5-FOA, itself alters nucleotide pools and dNTP ratios and is sensitive to other perturbations of nucleotide metabolism (Rossmann et al. 2011). Alterations in dNTP levels appear to account at least in part for the previous finding that mutants of PCNA affected telomere silencing (Li et al. 2009).
Silencing in the absence of Sir proteins
Surprisingly a gain-of-function mutation in the SUM1 gene will establish silencing even in the absence of the SIR genes (Klar et al. 1985; Laurenson and Rine 1991; Chi and Shore 1996). A single-amino-acid substitution in SUM1 is sufficient to convert it from a localized transcriptional repressor to a broad repressor of gene expression. At HML and HMR, SUM1-1 silencing still depends on histone deacetylation, but now it turns out to be mediated not by Sir2 but by its homolog, Hst1 (Rusche and Rine 2001; Yu et al. 2006; Safi et al. 2008). Curiously, the S. cerevisiae SUM1-1 allele behaves like the wild-type allele of SUM1 in Kluyveromyces lactis, where it normally plays a role in HML and HMR silencing (Hickman and Rusche 2009).
A brief word about unsilencing
One measure of unsilencing is the susceptibility of HML or HMR to be cleaved by HO endonuclease. As noted before, these loci are normally protected from HO cleavage, except that HML will show a small amount of cleavage after several hours of overexpressing HO. When these loci are desilenced by raising a temperature-sensitive allele of sir3 to 37° or by adding nicotinamide to inhibit Sir2, HML is consistently cleaved 1 hr before HMR (A. Walther and J. E. Haber, unpublished results). If HMRα is inserted in place of HMLα, then it is cleaved ∼30 min after HMLα; so there seem to be both sequence and position effects on the persistence of silencing.
Cell Lineage and Cell Cycle Control of HO Gene Expression
MAT switching provided a powerful early model to study the determination of cell lineage.5 Only half of the cells in a colony are able to switch mating type in any one cell division (Figure 1). A germinating haploid spore grows, produces a bud, and divides without changing mating type. Then, in the next cell division cycle, the older mother cell and its next (second) daughter change mating type while the first daughter buds and divides without any change (Haber and George 1979). The axial budding pattern of haploids places two MATa cells immediately adjacent to two MATα cells and they readily conjugate, forming MATa/MATα diploids in which the HO endonuclease gene is turned off so that further mating-type switching is repressed. If cells are prevented from mating—for example, by micromanipulating cells apart before conjugation—one can establish that the lineage rules persist: any cell that has previously divided once is capable of switching MAT, while new daughter cells cannot (Haber and George 1979).
Nasmyth’s lab (Breeden and Nasmyth 1987; Nasmyth 1987a; Nasmyth et al. 1987) first demonstrated that the control of this lineage pattern depended on the asymmetric expression of the HO endonuclease gene, which is restricted to mother cells that have divided at least once. Control of HO expression depends on the Swi5 transcription factor, which is localized to the mother cell nucleus and not that of her daughter (Nasmyth 1987b). The absence of Swi5 expression in daughters is caused by the Ash1 repressor protein that exhibits a striking localization only in the daughter cell (Figure 1) (Bobola et al. 1996; Derbyshire et al. 1996; Sil and Herskowitz 1996). Ash1 acts upstream of Swi5 and may directly repress SWI5 transcription, thus restricting HO expression to the mother cell in the next G1 stage of the cell cycle. The localization of Ash1 occurs not by selective transport of the protein to the daughter but by localizing its mRNA prior to cell division (Long et al. 1997). mRNA localization apparently depends on the myosin-like protein Myo4 (Bobola et al. 1996), as well as actin (Takizawa et al. 1997). A complex of two other RNA-binding proteins, She2 and She3, act to effect cargo binding. Two other proteins have also been shown to be important for efficient ASH1 localization in yeast: Loc1p, a nuclear protein, and Khd1, which is thought to link translational repression to the localization process (Long et al. 2001; Irie et al. 2002; Hasegawa et al. 2008). Since Ash1 mRNA localization was first discovered, >20 other mRNAs have been shown to show similar localization in Saccharomyces (Shepard et al. 2003; Jambhekar et al. 2005).
Mother–daughter control is only one aspect of the regulation of HO expression. HO transcription is confined to a narrow window in the cell cycle, after the cell passes “start.” Start is the point at which α-factor arrests MATa cells (or a-factor arrests MATα cells) so that cells beginning conjugation should contain unreplicated nuclei that will fuse by karyogamy to create a diploid nucleus. Start is also the point at which the key cell division kinase Cdk1 is inactive. As cells pass start, Cdk1 in conjunction with its G1 cyclins (Cln1, Cln2, and Cln3) becomes active. Cdk1 activates two transcription factors, Swi4-Swi6, which together are called SBF, and which bind to the SCB DNA motif (Swi4,6-regulated cell cycle box). Swi6 also pairs with Mbf1 to form the MBF (Mlu1 cell cycle box) cell cycle regulatory factor that turns on genes prior to initiating new DNA synthesis. The HO upstream regulatory region is perhaps the largest of any yeast gene—on the order of 1.4 kb (Nasmyth 1993). The HO promoter contains 10 copies of SCB, which is bound by Swi4 and Swi6. There is also a binding site for the Mata1-Matα2 repressor that turns HO off in diploids.
The HO endonuclease protein is quite unstable and is rapidly degraded so that mother cells suffer a brief pulse of endonuclease activity. Were the HO protein to persist, it would be able to cut the switched locus and cause a second recombination event. Studies by Raveh (Kaplun et al. 2003, 2006) have shown that HO endonuclease is rapidly degraded by the ubiquitin-mediated SCF protein degradation complex. HO has a half-life of only 10 min.6 HO is apparently targeted for degradation by its phosphorylation by the Chk1 kinase of the DNA damage response; hence a kinase-dead mutation of the ATR homolog Mec1, as well as rad9Δ and chk1Δ stabilize HO-LacZ. However, degradation does not depend on the actual induction of a DSB and triggering the DNA damage checkpoint per se.
HO is a member of the LAGLIDADG family of site-specific endonucleases (reviewed by Haber and Wolfe 2005); it recognizes a degenerate 24-bp sequence that spans the MAT-Y/Z border (Nickoloff et al. 1986, 1990). A haploid yeast has three possible targets for HO: the MAT locus, HMLα, and HMRa, but only the MAT locus is accessible under normal conditions.7 So, combining all these controls, there is a single, programmed DSB inflicted on the MAT locus only in mother cells and prior to the initiation of DNA replication.
First Models of MAT Switching
Early studies of MAT switching recognized the existence of two additional key loci that were required for the replacement of MAT alleles: HML and HMR (Takahashi et al. 1958; Takano and Oshima 1967; Santa Maria and Vidal 1970). A remarkably insightful hypothesis by Oshima and Takano (1971) suggested that these loci were the seat of controlling elements that could transpose to MAT and activate opposite mating-type alleles. Coupled with the key experiments of Hawthorne (1963), these ideas led Herskowitz’ lab (Hicks et al. 1977; Haber and George 1979) to suggest a specific version of the transposition model known as the “cassette model” in which an unexpressed copy of Yα sequences was located at HML (HMLα) and unexpressed Ya sequences were found at HMRa. These sequences could be transposed to the MAT locus, where they would be expressed. In these early models, there was no suggestion that MAT switching involved homologous recombination; rather a site-specific duplicative transposition imagined. Subsequent studies (Nasmyth and Tatchell 1980; Strathern et al. 1980; Astell et al. 1981; Tatchell et al. 1981) confirmed that there were indeed two additional copies of mating-type information at HML and HMR. Most laboratory strains carry HMLα and HMRa, but natural variants exist that carry the opposite configuration: HMLa and HMRα (Takahashi et al. 1958; Naumov and Tolstorukov 1971; Tolstorukov and Naumov 1973). One early surprise in the molecular analysis of MAT, HML, and HMR was that the two donor cassettes did not carry simply the Ya and Yα donor sequences that could be “played” in the cassette player of the MAT locus, but were in fact intact, complete copies of mating-type genes carrying their own bidirectional promoters (Figure 2). But somehow these genes were not transcribed. The two unexpressed cassettes differ in the extent of homology they share with MAT. HMR, HML, and MAT all share two regions flanking the Y sequences, termed X and Z1. HML and MAT share additional sequences, termed W and Z2 (Figure 2).
We now know that during switching there is no change in either donor sequence; that is, MAT switching does not involve a reciprocal exchange of Ya and Yα sequences, but rather a copying of the sequences from either HMLα or HMRa and their insertion at MAT in place of the original MAT allele (Hicks et al. 1979). This asymmetric recombination event is termed a gene conversion. The idea that HML and HMR repeatedly served as donors during MAT switching provided an explanation for an early observation of Hawthorne (1963) that a mutant MATα cell could be replaced by MATa, which then switched to a wild-type MATα allele. Subsequent “healing” and “wounding” experiments were carried out in which mutations at MAT were corrected by recombination with the donor or in which a mutation at the donor was introduced into the MAT locus (Hicks and Strathern 1977; Klar et al. 1979b; Sprague et al. 1981). In some cases, the replacement of MAT information included not only the Y region but at least part of the flanking X and Z1 regions as well that were shared by MAT and its two donors (Sprague et al. 1981; McGill et al. 1989).
In the 35 years since the cassette model was articulated, the MAT switching system and other HO-induced DSBs have been the object of intense study, to learn both about gene silencing and about the multiple mechanisms of double-strand break repair by homologous recombination, nonhomologous-, and microhomology-mediated end-joining and new telomere addition (Rattray and Symington 1995; Pâques and Haber 1999; Aylon and Kupiec 2004; Krogh and Symington 2004; Daley et al. 2005; McEachern and Haber 2006; Sung and Klein 2006; Lydeard et al. 2007; Li and Heyer 2008; McVey and Lee 2008; San Filippo et al. 2008; Jain et al. 2009; Heyer et al. 2010; Schwartz and Heyer 2011). Here, we focus more specifically on MAT switching.
MAT Switching: a Model for Homologous Recombination
The conversion of one mating type to the other involves the replacement at the MAT locus of Ya or Yα by a gene conversion induced by a DSB by HO endonuclease (Strathern et al. 1982; Kostriken et al. 1983). The process is highly directional, in that the sequences at MAT are replaced by copying new sequences from either HMLα or HMRa, while the two donor loci remain unchanged by the transaction. Directional gene conversion reflects the fact that HO endonuclease cannot cleave its recognition sequence at either HML or HMR, as these sites are apparently occluded by nucleosomes in silenced DNA. Thus the MAT locus is cleaved and becomes the recipient in this gene conversion process. A very weak cleavage of HML has been observed when HO endonuclease is overexpressed (Connolly et al. 1988); moreover, rare “illegal” switches, where MAT is the donor and the silent locus is switched, have been observed when MAT itself cannot be cut because of a change in the HO recognition site (Haber et al. 1980a). In Sir− cells where HML or HMR is expressed, HO can readily cut these loci and they become recipients (Klar et al. 1981; Bressan et al. 2004).
Normally the HO gene is tightly regulated to be expressed only in haploid mother cells and only at the G1 stage of the cell cycle (Nasmyth 1987b); however, the creation of a galactose-inducible HO gene made it possible to express HO at all stages of the cell cycle and in all cells (Jensen and Herskowitz 1984). This made it possible to deliver a DSB to all cells simultaneously and to follow the appearance of intermediates and final products by physical analysis of DNA extracted at times after HO induction (Connolly et al. 1988; Raveh et al. 1989; White and Haber 1990). An example of Southern blot analysis of MATa switching to MATα is shown in Figure 5. Physical monitoring of recombination at MAT has yielded much of what we know about DSB-induced mitotic recombination (reviewed in Haber 1995, 2006; Pâques and Haber 1999; Krogh and Symington 2004; Hicks et al. 2011). Some related studies have been done by inserting small HO endonuclease recognition sites at other locations and from the induction of other site-specific endonucleases, most notably I-SceI (Rudin and Haber 1988; Nickoloff et al. 1989; Ray et al. 1989; Plessis et al. 1992; McGill et al. 1993; Liefshitz et al. 1995; Weng et al. 1996; Inbar and Kupiec 1999; Wilson 2002; Storici et al. 2003; Lydeard et al. 2007, 2010; Jain et al. 2009; Marrero and Symington 2010). Additional information about DSB repair has been gleaned from the analysis of DSB-induced recombination in meiotic cells (reviewed in Kleckner 1996; Borner et al. 2004; Keeney and Neale 2006; Longhese et al. 2009). By and large the results are sequence independent, though some interesting aspects particular to MAT switching are noted below.
The overall process of MATa switching to MATα is illustrated in Figure 6. Following HO cleavage of MATa, the ends are resected in a 5′ to 3′ direction, creating a 3′-ended ssDNA tail that assembles a filament of the Rad51 recombinase protein. This protein::DNA complex engages in a search for a homologous sequence (in this case HMLα) with which repair can be effected. Homology search culminates in strand exchange in which the ssDNA base pairs with the complementary sequence in the donor, creating a displacement loop, or D loop. The 3′ end of the invading strand is then used as a primer to initiate copying of one strand of the donor locus, and the newly copied strand is displaced until it can anneal with homologous sequences on the opposite end of the DSB. The 3′-ended nonhomologous tail is clipped off and the new 3′ end is used to prime a second strand of DNA synthesis, completing the replacement of MATa by MATα. Each of these steps is discussed in more detail below.
HO endonuclease cleaves a degenerate recognition site of 24 bp in vitro (Nickoloff et al. 1986), although sites of 117 bp down to 33 bp are generally used when the HO recognition site is inserted at other locations. A site with only 21 bp results in inefficient single-strand nicking that, by replication, can be converted to a DSB (Cortes-Ledesma and Aguilera 2006). Single-base-pair MAT-inc (inconvertible) or MAT-stk (stuck) substitutions in the recognition site abolish or greatly reduce switching8 (Weiffenbach and Haber 1981; Ray et al. 1991). HO cutting generates 4 bp, 3′-overhanging ends, both of which are accessible to exonucleases in vitro (Kostriken et al. 1983). In vivo, however, the DSB is processed almost exclusively by several 5′ to 3′ exonucleases to create long 3′-ended tails (White and Haber 1990). As discussed more fully below, the 3′ end is remarkably resistant to exonucleolytic removal. It is possible that there are no 3′ to 5′ exonucleases that act on a 3′-overhanging end or that the end is protected in vivo by the binding of RPA or Rad51 or other proteins.
MAT switching, induced by a galactose-regulated HO endonuclease, is a surprisingly slow process, requiring 1 hr to complete, independent of the time during the cell cycle (Connolly et al. 1988; Raveh et al. 1989; White and Haber 1990). It is possible that normal MAT switching may be more rapid, when HO is expressed in G1, only in mother cells, and in a coordinated fashion with other genes; this may be inferred from the low level of steady-state HO-cleaved MAT DNA in cells that can continually switch (i.e., MATa cells with HMLa and HMRa cassettes) (Strathern et al. 1982). However, additional experiments using HO whose expression is restricted to the G1 phase of the cell cycle (Nasmyth 1987b) shows similar slow kinetics, as does expressing a conditional allele of HO under normal cell cycle control (M. Yamaguchi, M. Gartenberg, and J. E. Haber, unpublished results).
5′ to 3′ resection of the DSB ends
5′ to 3′ resection proceeds rather slowly, at a rate of ∼4 kb/hr—about 1 nt/sec—(Fishman-Lobell et al. 1992; Zhu et al. 2008) and in a strain where there is no repair of the DSB, resection will continue at roughly this rate for more than 24 hr!9 Physical analysis of the rate of 5′ to 3′ degradation initially implicated the trio of interacting proteins, Rad50, Mre11, and Xrs2 (MRX complex), in this process (Sugawara and Haber 1992; Ivanov et al. 1994; Tsubouchi and Ogawa 1998). MRX somehow associates with Sae2, though a direct interaction has not been demonstrated, and together these proteins appear to do the initial 5′ to 3′ resection at HO-induced DSB ends. Mre11 has 3′ to 5′ exonuclease activity and both Mre11 and Sae2 have endonuclease activity (Bressan et al. 1998; Lee et al. 2002; Nicolette et al. 2010); however, mutation of the nuclease activity of Mre11 has very little effect on resection (Bressan et al. 1998; Moreau et al. 1999; Lee et al. 2002). Mre11’s nuclease activity is required in mitotic cells to cleave hairpin ends (Lobachev et al. 2004; Yu et al. 2004) and in meiosis to remove the Spo11 protein from DSB ends (but in both of these cases Sae2 is also necessary). Deleting Sae2 does significantly retard resection (Clerici et al. 2005). While MRX-Sae2 appears to get resection started, extensive resection depends on two competing pathways of resection, one comprising the 5′ to 3′ exonuclease Exo1 and the other consisting of the Sgs1-Top3-Rmi1 (STR) helicase complex coupled to the nuclease function of Dna2 (which itself has nuclease activity not relevant for this process) (Huertas et al. 2008; Mimitou and Symington 2008; Zhu et al. 2008; Niu et al. 2010). Exo1 activity is increased when the Ku proteins are ablated (Mimitou and Symington 2010; Shim et al. 2010).
When both Sgs1 and Exo1 are deleted, there is very limited resection, which appears to depend on MRX and Sae2. This result suggests that MRX acts first and hands off resection to Exo1 or the STR-Dna2 complex; however, this idea is contradicted by the long-observed fact that deletion of MRX proteins or Sae2 only reduces resection to about half in cycling cells. However in G2-arrested cells deletion of Rad50 eliminates nearly all resection (Diede and Gottschling 2001); possibly an alternative resection activity is absent in G2-arrested cells. Moreover in G1 cells prior to the activation of Cdk1 at the start point of the cell cycle, there is nearly no resection (Aylon et al. 2004; Ira et al. 2004). Inhibition of Cdk1 at other points in the cell cycle also blocks resection (Aylon et al. 2004; Ira et al. 2004). Taken together, it seems that the modest inhibition of resection in cycling cells by deleting MRX or Sae2 argues that MRX-Sae2 does not act as the obligate gatekeeper of resection and that in S phase, MRX-Sae2 may have a very minimal role.
Resection of course must plow through chromatin and it is not yet clear how these complexes accomplish the necessary chromatin remodeling. Deleting the Arp8 subunit of the Ino80 complex has—in some hands—a modest effect on resection. But a much more profound inhibition of resection is seen when the Swi2/Snf2 homolog Fun30 is deleted (G. Ira, personal communication; B. Llorente, personal communication; V. Eapen, N. Sugawara, M. Tsabar, and J. E. Haber, unpublished results). Fun30 is an ATPase that has been shown to displace a positioned nucleosome in vitro (Neves-Costa et al. 2009; Awad et al. 2010).
Recruitment of Rad51 recombinase and the search for homology
Once long 3′ tails have been generated, they can associate with the Rad51 recombination protein that facilitates a search for homologous regions, to initiate recombination (Figure 6). Chromatin immunoprecipitation (ChIP) experiments have shown that once ssDNA is generated, it is first bound by the ssDNA-binding protein complex, RPA, which is then displaced by Rad51 (Wang and Haber 2004). The loading of Rad51 depends completely on the Rad52 protein (Sugawara et al. 2003; Wolner et al. 2003). In the absence of the Rad55 and Rad57 proteins, which are known as Rad51 paralogs, Rad51 filament assembly is slow and apparently incomplete, and MAT switching fails to occur. In other DSB-mediated repair events where the donor is not silenced, recombination also fails in the absence of Rad55 or Rad57 (Sugawara et al. 2003; Wolner et al. 2003).
ChIP, using an anti-Rad51 antibody, allows one to visualize the kinetics of Rad51 loading onto ssDNA (Figure 7). The same approach permits visualization of the synapsis between the MAT DSB and the donor, as Rad51 will be associated with both the invading MAT strand and the HMLα duplex DNA (Figure 7). This step takes ∼15 min after appearance of Rad51 assembly at the DSB (Sugawara et al. 2003; Wolner et al. 2003; Hicks et al. 2011). The time to achieve pairing between MAT and HML has also been seen microscopically by examining GFP-tagged LacO and TetR arrays situated close to HML and MAT, respectively (Bressan et al. 2004; Houston and Broach 2006). It should be noted that the time to pair with donor sequences located interchromosomally is significantly longer than what occurs between HML and MAT. The relatively rapid encounter between these two loci is undoubtedly aided by the cis-acting recombination enhancer (RE), located ∼17 kb centromere proximal to HML, which will be discussed in detail below.
It is striking that the amount of homology shared by MAT and its donors is quite small, especially on the Z side, which seems to initiate copying of the donor. MAT and HMR only share 230 bp, while MAT and HML share 327 bp. In contrast, there is much more extensive homology on the W/X side, beyond the Y nonhomologous sequences. But the efficiency of repair is largely dictated by the smaller Z side. We will discuss the mechanism of donor preference in detail below, but suffice it to say that one can set up an experiment in which MAT will normally switch with HMR as a partner and HML is the “wrong donor.” By artificially increasing the size of the homology on the Z side of HML from 327 to 650 to 1800 bp, one can significantly increase its use as a donor in competition with HMR (Coïc et al. 2011). This is a surprising result because there is always 1400 bp of homology in the W/X region; it suggests that the engagement of the donor is done by the side of the DSB that can initiate new DNA synthesis. The results suggest that although the preferred donor will be encountered in a population of cells 90% of the time, this choice is not irrevocable. On the basis of a model in which an encounter between the DSB end and a donor of normal size is only likely to culminate in a switching event, but that the success of the encounter will depend on the size of the homologous region, it appears that encounters between the Rad51 filament and the preferred donor will happen on average four times before some irreversible step will lead to the completion of recombination (Coïc et al. 2011). If the wrong HML donor carries a donor that shares much more homology with MAT on the right side, then whenever the DSB end encounters HML it will have a higher probability to complete repair.
Once resection exposes both the X and Z homologous regions flanking the DSB, they can each synapse with their homologous sites in HML (Hicks et al. 2011), but until the nonhomologous Y region is removed from the left end of the DSB, it does not seem to participate in the next key step in gene conversion, the initiation of new DNA synthesis, primed by a 3′ end of the strand-invaded DNA. The 3′ end of the invading strand in HML-Z1 acts as a primer to initiate new DNA synthesis, copying the Y region of the donor (Figure 7). This step can be seen by a PCR reaction using two primers—one complementary to sequences distal to MAT and one in the Yα region of the donor (White and Haber 1990). PCR amplification can take place only after the invasion of the 3′-ended single strand into the Z region of the donor locus and the copying of at least 50 nt, primed from the 3′ end, thus creating a recombination intermediate that covers both primers. This step occurs ∼15–20 min after synapsis between MAT and HML is observed by ChIP and 15–30 min prior to the completion of gene conversion, as monitored by both Southern blots and a second PCR assay, detecting the time when the donor Y sequences are joined to the proximal side of MAT (White and Haber 1990; Hicks et al. 2011).
The region that is replaced during MAT switching is substantially longer than the Y region itself. McGill et al. (1989) used artificial restriction sites inserted at different places in the X and Z regions to show that replacement of the Y segment often extends well into the flanking homologous regions. They further showed that there was no reciprocal transfer of markers from MAT to the donor. This observation was supported by studies of the mismatch repair of a single-bp mutation only 8 bp from the 3′ of the HO cut, in the Z region (Ray et al. 1991). In the absence of mismatch repair, this mutation was most often retained during switching (thus confirming physical studies showing that there was almost no 3′ to 5′ removal of the 3′-ended tail). Usually only one of the two daughter cells carried the mutation. This type of postswitching segregation is analogous to postmeiotic segregation observed among meiotic segregants when the DNA inherited into one spore is heteroduplex (mutant/wild type) in the absence of mismatch correction. A kinetic analysis (Haber et al. 1993) further demonstrated that, in repair-proficient cells, mismatch correction occurred very rapidly (as quickly as the PCR-amplified intermediate could be detected), suggesting that correction occurred soon after the strand invaded the donor locus. Moreover the heteroduplex DNA was corrected in a highly biased way, such that mutant sequence in the invading Z DNA was corrected to the genotype of the donor. This observation is probably the most direct in vivo demonstration of the idea that mismatch repair will preferentially correct a mismatch adjacent to a nick (in this case, the 3′ end of an invading strand) (Porter et al. 1993; Leung et al. 1997).
Recently, it has been possible to examine strand invasion at the level of chromatin, using a PCR-based nucleosome protection assay after treating chromatin with micrococcal nuclease (Hicks et al. 2011). As noted before, HML has very highly positioned nucleosomes, whereas there is little order to nucleosomes at MAT. At the time of synapsis, there is a transient reduction in protection of several nucleosomes at the site of strand invasion. By delaying the initiation of new DNA synthesis, it was possible to see a longer period of change in nucleosome organization. This study produced two interesting results. First, the region of nucleosome protection extended several nucleosomes beyond the point where strand invasion occurred. (Recall that MAT and HML share only 327 bp on the right side of the break where strand invasion is seen.) This observation suggests that the D loop is extended—perhaps by helicases—prior to the initiation of new DNA synthesis. It has not been determined yet whether the extended and apparently open region would bind RPA.
The second important finding was that nucleosome rearrangement at HML did not occur in a rad54Δ mutant. Rad54 is a Swi2/Snf2 homolog that has been shown to engage in chromatin remodeling in vitro (Jaskelioff et al. 2003). The surprising finding here is that some sort of strand invasion occurs in the absence of Rad54, as seen by ChIP for Rad51 associating with HML-adjacent sequences (Sugawara et al. 2003); but this association apparently is distinct from the full chromatin rearrangement necessary to complete DSB repair. Without Rad54, there is no primer extension and new DNA synthesis. One possible explanation for this result is that the kind of association of MAT and HML strands mediated by Rad51 is a paranemic joint in which the invading strand does not intertwine with the donor duplex, whereas with Rad54 a plectonemic, interwound structure is formed with the displaced strand in an extended D loop (see Figure 5).
The synthesis-dependent strand-annealing (SDSA) mechanism that seems to be used predominantly differs from normal DNA replication in that the newly synthesized strand is displaced from the template and is subsequently used by the second end as the template for completing gene conversion. This model predicts that all the newly copied DNA should be found at the recipient locus while the donor remains unaltered. Indeed this is the case, as measured by a heavy 15N/14N density-shift experiment analogous to that used by Meselson and Stahl to show that bacterial replication was semiconservative. Here the outcome is “conservative” (Ira et al. 2006).
For the second strand’s end to act as a primer, the nonhomologous Ya sequences have to be removed. One strand is of course removed by 5′ to 3′ resection, but the second one is clipped off—apparently only after the elongated first strand anneals in MAT-X region—by the Rad1-Rad10 endonuclease (Fishman-Lobell and Haber 1992; Lyndaker et al. 2008). In other HO-induced events (and probably in MAT switching) removal of the nonhomologous tail also requires a number of other proteins, including Msh2-Msh3, Saw1, and Slx4 (Colaiacovo et al. 1999; Li et al. 2008). This “clippase” acts apparently only on branched, annealed structure with a 3′-ended tail, which is also produced during single-strand annealing. This step also requires the action of either of the two DNA damage-responsive protein kinases, Mec1 or Tel1 to phosphorylate Slx4 (Toh et al. 2010).
Copying the donor sequences
Physical analysis has also made it possible to analyze conditional lethal mutants to ask which DNA replication enzymes are required for MAT switching. In contrast to another HO-induced repair event, break-induced replication,10 only a fraction of the proteins necessary for origin-dependent DNA replication are also required for MAT switching (Lydeard et al. 2010). In this process, which appears to involve the elongation of one strand and then the elongation of the second strand, DNA polymerase α is not required, while both DNA polymerases δ and ε appear to act either sequentially or redundantly. The PCNA clamp is required, but the GINS-Cdc45-Mcm helicase complex is dispensable. A mutation of the largest subunit of RPA, rfa1-t11 (L45E) is also required at or after strand invasion, as Rad51 can load and engage in synapsis with HML but there is no new DNA synthesis. In contrast, DNA synthesis during MAT switching does not need most of the loading factors required at an origin for normal replication, including the ORC proteins, Cdt1 and Cdc6. The exception is that the Dpb11-Sld2-Sld3 proteins are required. These proteins have been shown to be part of a preloading complex at origins (Muramatsu et al. 2010), but how they would work when DNA copying is not dependent on an ARS or on other early-functioning proteins is unknown.
The fact that DNA synthesis during gene conversion does not use all the processivity factors employed in normal replication may explain why gene conversion is much more susceptible to mutation of the replicated sequences. Taking advantage of several features of MAT switching, it was possible to select for mutations that arose during gene conversion (Hicks et al. 2010). First, the normal a1 ORF of HMRa was replaced by the K. lactis URA3 ORF, also removing part of the HO cleavage site (Figure 8). Because HMR is silenced, the cells remain Ura−. With HML deleted, when MATα is induced to switch with GAL::HO, the Ya::Kl-URA3 sequences are gene converted into MAT, allowing the ORF to be expressed. Thus nearly all cells become Ura+. However, if a mutation arose during switching, then the cells would be Ura− and thus resistant to selection by 5-FOA, which kills Ura+ cells. Such mutations arose at a rate of ∼1 × 10−5, ∼1000 times higher than the spontaneous rate of mutation for a MAT locus carrying the same Ya::Kl-URA3. That these were de novo mutations could be demonstrated by showing that these cells still carried a wild-type Kl-URA3 allele at HMR. After inhibiting Sir2, with the addition of nicotinamide to the medium, the cells became Ura+, evidence that the Ya::Kl-URA3 at HMR had not been altered. The majority of mutations were base-pair substitutions but ∼40% represented template jumps, as if the DNA polymerase was less processive than would be found during normal replication. There were three types of mutations: −1 frameshifts in homonucleotide runs, complex mutations explained by the use of quasipalindromes, and remarkable jumps from the Kl-URA3 sequences to the 73% homeologous S. cerevisiae ura3-52 sequences located on a different chromosome. In the last case, there had to be a second jump, back to Kl-URA3 sequences, to complete the “switch.” Even more surprising is that all three types of mutations were eliminated in a strain with a proofreading defect in DNA polymerase δ. We surmised, on the basis of some in vitro studies of a similar proofreading mutation (Stith et al. 2008), that the proofreading-defective mutant enzyme is in fact less prone to dissociate from the template. This result argues strongly that DNA Polδ is a major player in MAT switching. However, there was also evidence that DNA polymerase ε was active, since a proofreading-defective mutant of Polε resulted in the appearance of +1 frameshifts. The appearance of these mutations in gene conversion was apparently independent of the mismatch repair system and insensitive to the error-prone DNA polymerase Polζ or another translesion DNA polymerase, Polη.
Completion of switching
One of the other striking aspects of MAT switching is that it is very rarely accompanied by crossing over. Such exchanges produce lethal outcomes: a MAT-HML centromeric deficiency chromosome that would lack essential genes distal to MAT or a large internal and equally lethal MAT-HMR deletion (Haber et al. 1980b; Klar and Strathern 1984). Crossovers are not expected when the SDSA mechanism is used, because there is no stable single or double Holliday junction that would be cleaved to produce crossovers. On the basis of ectopic recombination studies in which HO induces a gene conversion between MATa on chromosome V and an uncuttable MATa-inc allele on chromosome III (and where the normal donors are deleted), it seems that crossovers are prevented predominantly by the action of two helicases, Sgs1 (with its partners Top3 and Rmi1) and Mph1 (Ira et al. 2003; Prakash et al. 2009). The Sgs1 complex acts as a dissolvase to remove double Holliday junctions that would otherwise become crossovers. Mph1 appears to ensure that the SDSA pathway is used rather than the alternative double-strand break repair mechanism that is much more prevalent in meiosis.
Finally, it is worth noting that MAT switching represents a case of gap repair rather than strictly break repair; that is, the regions of homology located by the two DSB ends are separated on the donor template by ∼700 bp. In a related study, using HO-cleaved LEU2 sequences, we found that break repair and gap repair, when the gap was larger than ∼2 kb, are surprisingly different repair processes (Jain et al. 2009). Break repair occurs with relatively rapid kinetics, whereas there is a delay of hours before new DNA synthesis is initiated when there is a long gap. This delay is quite similar to that seen when only one end of the DSB is homologous to a template and repair can only occur by assembling a complete replication fork, resulting in break-induced replication (BIR). Consequently long gap repair and BIR depend on Pol32, a nonessential subunit of DNA Polδ, while break repair (and MAT switching) is Pol32 independent. Apparently the two ends of the DSB need to be in contact with each other at the time of strand invasion; they must pair close enough to each other, and in the proper orientation, to permit some signal to be propagated. We termed the assessment of the nature of the strand invasion—the difference between break repair and gap repair—as a manifestation of a recombination execution checkpoint. MAT switching appears to have a small enough gap to be treated as break repair.
On top of the inherent directionality of switching, namely, that HML or HMR are not cut by HO and therefore only donate sequences to MAT, there is an elaborate mechanism that gives yeast the ability to choose between its two donors. It makes sense that MATa should seek out and recombine with HMLα rather than HMRa, so that the recombinational repair of the DSB will lead to a switch to the opposite mating type. Donor selection is however not dictated by the Ya or Yα content of the donors: a strain with reversed silent information (HMLa MATa HMRα) still chooses HML ∼85–90% of the time (Klar et al. 1982; Weiler and Broach 1992). Weiler and Broach showed that replacing the entire HML region including its silencers with a cloned HMR locus did not change donor preference, so it is the location of the donor, not the sequence differences between HML and HMR, that direct donor selection. There must therefore be one or more cis-acting sequences, outside of the donors themselves that activate or repress one or both donors, depending on mating type.
MATα’s choice of HMR over HML occurs independently of the MATα1 gene, but is strongly dependent on MATα2, the gene that acts as a repressor of a-specific genes (Hicks et al. 1977; Tanaka et al. 1984; Weiler et al. 1995; Wu et al. 1996; Szeto and Broach 1997; Szeto et al. 1997). MATa donor preference does not depend on a functional MATa1 gene (Wu and Haber 1995). These observations might suggest that MATa cells activate HML for switching through one or more a-specific gene products that are turned off in MATα cells, while α-specific proteins might activate HMR. However, as we will see, HMR appears to be used as the default locus and all the active regulation is in making HML more accessible (Figure 9). Thus a MATa cell deleted for HML can easily use HMR, but 10–20% of MATα cells die when their only choice of a donor is HML (Wu and Haber 1996; Wu et al. 1996, 1997). The failure of many MATα cells to use the “wrong” donor occurs despite the fact that cells experiencing an unrepaired DSB become arrested at a G2/M checkpoint (Sandell and Zakian 1993); this checkpoint should theoretically have allowed cells time to locate a donor and repair the DSB by gene conversion, but as DSB ends are resected, alternative, lethal recombination events—for example, between several Ty retrotransposons located 30 kb proximal to MAT and Tys on other chromosomes—may lead to the death of cells that cannot easily repair the DSB (VanHulle et al. 2007).
Identification of a recombination enhancer
To locate a cis-acting element (now called the recombination enhancer) on the left arm of chromosome III, Wu and Haber (1996) deleted HML at its normal location, 12 kb from the telomere, and inserted either HMLα or HMRα at a site 41 kb from the left end of the chromosome. At this position the donor was still strongly preferred in MATa cells. By a series of truncation deletions they identified a region that was required to activate the donor at 41 kb. A further series of internal deletions was created to pinpoint the key cis-acting element. A 2.5-kb deletion located 17 kb proximal to HML completely reversed donor preference, so that a MATa cell now used HML only 10% of the time instead of 90% (Figure 9). Deletion of this sequence also abolished MATa donor preference for donors located 41, 62, and 92 kb from the left end. In the absence of this sequence, donors at these positions were not used 50:50 with HMR; rather they were found only 10% of the time. Deletion of this region had no effect on MATα cells, which continued to use HMR most of the time.
Further mapping of RE was accomplished by inserting subfragments of the smallest deletion back into the chromosome. This led initially to the identification of a 700-bp RE that restored MATa donor preference almost to wild-type levels. Subsequent analysis has narrowed down the most important sequences to ∼250 bp, although full activity resides in a region of ∼400 bp. This further refinement was accomplished by showing that a syntenic region in S. carlsbergensis and in S. bayanus (but not in more distant species such as S. servazzii) contained an active RE that would substitute for the S. cerevisiae RE (Wu et al. 1998; Sun et al. 2002). By comparing the divergent sequences of these REs, we defined five well-conserved subdomains, named A–E (Figure 9B). Of these, domain B appears to be unimportant, because it can be deleted without significant effect on MATa or MATα donor preference. In a minimum enhancer of 250 bp lacking region E, deletions of subdomains A, C, and D all abolish MATa donor preference, causing cells to use HMR 90% of the time. Subdomains C and D can be inverted relative to A and B and still function properly. It was startling to discover that subdomain A was completely missing from the 700-bp RE defined by Wu and Haber (1996). Clearly, other sequences in the larger region must carry out redundant functions that subdomain performs in the minimum RE.
Subdomains D and E are intriguing because of their unusual sequence. There are 10 or 15 perfect repeats of TTT(G/A). Truncations of this region, leaving only 8 repeats, reduces donor preference by ∼50% and further truncations have no RE activity (Wu et al. 1998). However, further analysis has revealed that what domains A, D, and E have in common is that they each contain one or more binding sites for the Fkh1 transcription regulator (Wu et al. 1998; Sun et al. 2002). As we will see below, this protein plays a central role in the activity of RE. A key finding was that one could replace the entire RE with multimers of only region A, or similarly with multimers of D or E, and retain RE activity. Most strikingly, four copies of the 21-bp region A was sufficient to raise HML usage from ∼10% in the absence of any RE to 65%. A 2-bp site-directed mutation in the A sequence abolished this activity. Region A, but not the 2-bp mutant form, binds Fhk1. The activity of the 4xA construct is abolished in a fkh1Δ mutant.
Role of an Matα2-Mcm1 operator in turning RE off in MATα cells
RE lies in perhaps the largest “empty” region of the yeast genome, ∼2.5 kb with no open reading frames or regulatory sequences associated with the flanking genes, KAR4 and SPB1. Although RE regulates the usage of HML, it does not do this through any global change in apparent chromosome condensation or transcriptional activity, as measured by the level of expression of genes along the left arm of chromosome III. However, there are dramatic chromatin changes in the 2.5-kp intergenic region harboring RE. In MATa, the region is “open” and binds a number of proteins, whereas in MATα cells, the region is covered with highly positioned nucleosomes that cover the entire RE and extend between, but not into, two flanking open reading frames (Weiss and Simpson 1997; Wu et al. 1998). This change is mediated by the Matα2-Mcm1 repressor complex that represses transcription at all a-specific genes (Smith and Johnson 1994; Tan and Richmond 1998).
The 90-bp domain C harbors a conserved 31-bp consensus Matα2-Mcm1 binding site. This same repressor binding turns off the RE in MATα cells, in conjunction with the corepressor Tup1p (Szeto and Broach 1997). Mutations of the Matα2-binding sites is sufficient to alter donor preference in MATα, so that HML usage is increased from ∼10% to >50% (Szeto and Broach 1997; Wu et al. 1998). These results also suggest that any a-specific gene products are unlikely to play an essential role in activating RE, since these genes should still be repressed by Matα2-Mcm1 in the MATα cell. However, the difference between HML use in MATa (80%) and MATα (55%) in this mutant RE could be attributed to a-specific genes.
With the exception of the RE, all Matα2-Mcm1 binding sites are located just upstream of a-specific genes. The RE contains no open reading frame, but indeed there are two “sterile” transcripts of the RE region that are transcribed in MATa, but not MATα (Szeto et al. 1997). Despite the increasing evidence of the role of noncoding RNAs in regulating chromatin structure in organisms that also have RNAi silencing, it is unlikely that the sequence of the RNA transcript is important for RE activity, as truncations of RE that remove all of this transcribed region have full activity (Wu and Haber 1996; Wu et al. 1998; Li et al. 2012). Indeed, as we will see below, the entire RE can be replaced by completely foreign, LexA-binding domains and RE activity can be mimicked by the binding of a LexA-Fkh1 protein.
Activation of RE in MATa cells depends on Mcm1 protein
The Mcm1 protein is not only a corepressor; it can also act as a coactivator of transcription for both a-specific (Elble and Tye 1991) and α-specific genes (Hagen et al. 1993; Bruhn and Sprague 1994). In the case of α-specific genes, Mcm1 acts as a heterodimer with Matα1p (Figure 3); no possible coactivator with Mcm1 in MATa cells has been identified. A 2-bp mutation that eliminates Mcm1 binding in the Matα2-Mcm1 operator sequence abolishes MATa donor preference (Wu et al. 1998); HML is used only 10–20% of the time, even in the case where the 2-bp mutation is introduced into an otherwise unmodified chromosome III. Thus, although the minimum RE is ∼250 bp long and apparently contains some regions of redundant function, the elimination of Mcm1p binding is sufficient to completely inactivate RE. A single-amino-acid-substitution mutation in MCM1 (mcm1-R89A) also reduces Mcm1 binding and had a similar reduction in HML usage (Wu et al. 1998).
The analysis of Mcm1 binding has led to another surprising finding concerning the role of Matα2 and Mcm1 protein binding in the establishment of repression. The 2-bp operator mutation that prevents Mcm1 binding also causes a dramatic change in the chromatin structure of the RE. Even in MATa cells, where there is no Matα2 protein, the mutant RE has an array of highly positioned nucleosomes that is very similar to what is seen in normal MATα cells (Wu et al. 1998). Apparently other sequences within RE can organize a phased nucleosome structure in the absence of Matα2-Mcm1 binding, although the repressor proteins seem to more precisely position and “lock in” the repressing chromatin structure. The idea that the sequences surrounding the Mcm1-binding site play important roles in both activating and inactivating RE finds support in the otherwise paradoxical observations that a deletion of nearly the entire 31-bp Matα2-Mcm1 operator has a less profound effect on reducing HML usage in MATa cells than does the simple 2-bp deletion of the Mcm1p-binding site (Szeto and Broach 1997; Wu et al. 1998). The adjacent sequences may be important in determining how “cold” the left arm of the chromosome is in the absence of RE activation.
Clearly Mcm1 binding is critical in the activation of the normal RE sequence, but it is also evident that since multimers of domain A, D, or E (which lack Mcm1-binding sites) are sufficient to promote preferential use of HML; Mcm1’s importance is in regulating other chromatin features of the normal RE.
The RE affects other recombination events
The regulation of HML is not specific to these particular donor sequences nor to HO-mediated recombination. If HML is replaced with an allele of the leu2 gene and a second leu2 allele was placed elsewhere on chromosome III—or even on another chromosome—the rate of Leu+ spontaneous recombination was ≥10 times higher in MATa cells than in MATα (Wu et al. 1996). This difference is lost when the RE is deleted. There is no significant mating-type–dependent difference in mRNA levels for the leu2 gene inserted in place of HML despite a dramatic recombination difference. That RE should stimulate recombination between sequences unrelated to MAT, and even between chromosomes, ruled out the idea that there were specific pairing sequences that would bring MAT and a left-arm donor together. There was no significant mating-type–dependent difference when a similar experiment was done with one leu2 allele in place of HMR, thus supporting the conclusion that donor preference was effected through changes in the left arm of the chromosome, with HMR being a more passive participant.
RE acts over a long distance and is portable
When HML was deleted and either HML or HMR was inserted at other chromosome locations along chromosome III, the donor could be activated at several sites along the entire left arm of chromosome III in MATa cells, though the efficiency decreased as the donor was moved further from the telomere (Wu and Haber 1995, 1996). Conversely, RE itself can be moved to sites closer to the centromere and still stimulate the use of HML (Coïc et al. 2006a). RE can also stimulate HML usage when MAT is moved to a different chromosome (Wu et al. 1997). Finally, we have established a competition assay in which the LEU2 gene is placed near RE and used as a donor to repair an HO-induced DSB in a leu2 sequence on chromosome V. A second donor, carrying leu2-K, is located ∼100 kb centromere proximal on chromosome V. When RE is active, the use of the adjacent interchromosomal LEU2 sequence is ∼50% of all repair events, but when RE is deleted, the use of the interchromosomal LEU2 donor falls to ∼15%.
Evidence that RE is portable has come from two additional experiments. First, if a copy of RE is placed near HMR in a MATa strain that also has its normal RE, then the usage of HMR increases from ∼10–50%, suggesting that RE can activate a nearby HMR in its normal location (Coïc et al. 2006a). A second approach was to remove MAT, HML, and HMR from chromosome III and to insert them in roughly the same configuration on the larger chromosome V. Here the use of the more distant HML, on the opposite side of the centromere, was ∼40%, but when RE was inserted near HML, its use increased to >90% (Coïc et al. 2006a). So RE can work in a similar fashion in an entirely different chromosome context.
It should be emphasized that the “coldness” of the left arm in MATα cells does not depend on the presence of RE; in RE-deleted strains both MATa and MATα cells use HML only 10% of the time (Wu and Haber 1996). Whether there are sequences that intrinsically restrict the motion of the left arm without RE remains a question to be addressed.
RE binds Fkh1 and Swi4/Swi6
As indicated above, we found that several domains of RE should bind one or more copies of Fkh1. Actual binding was confirmed by ChIP, using a functional epitope-tagged Fkh1-HA construct. As expected, Fkh1 bound only in MATa and not in MATα, where the Matα2-Mcm1 repressor precludes such binding (Coïc et al. 2006b). Deletion of Fkh1 markedly reduces HML donor preference in MATa without affecting HMR’s use in MATα; however, the reduction is only from ∼85 to ∼35%, suggesting that there must be other proteins involved in the action of the complete RE (Coïc et al. 2006b). When a 4xA construct was used in place of the whole RE, fkh1Δ dropped HML usage to the same level as deleting RE (Coïc et al. 2006b).
Domain C is much larger than the Matα2-Mcm1 operator region. Further inspection revealed that it contains a SCB that binds the cell cycle regulators Swi4-Swi6, known together as SBF. When the SCB was mutated or if Swi6 was deleted, donor preference dropped, again to ∼35% (Coïc et al. 2006b). That both Swi4-Swi6 and Fkh1 play nonoverlapping roles in RE activity was shown by the fact that a deletion of SCB, coupled with fkh1Δ further reduced HML usage to ∼15%. This still leaves a small amount of HML usage greater than the 10% found in REΔ apparently contributed by several proteins.
Other donor preference mutations:
Two other trans-acting factors have been shown to play less decisive roles in donor preference. Screening directly for donor preference mutations has not been very productive, largely because mutations that affect HO expression and cis-acting mutations that reduce HO cleavage tend to interfere with the evaluation of donor choice scored at the colony level. Only chl1Δ has emerged in this way. A deletion of the CHL1 gene reduces donor preference in MATa switching from 80 to 60%; but it has no effect on MATα (Weiler et al. 1995). CHL1 is not a donor preference-specific gene; it was identified >20 years ago because chl1Δ causes a large increase in both the loss and gain of chromosomes and is therefore most likely a nondisjunction mutation (Liras et al. 1978; Gerring et al. 1990). Subsequent studies have shown that CHL1 encodes a nuclear protein with presumed but undemonstrated helicase activity that is implicated in the establishment and maintenance of sister-chromatid cohesion (Petronczki et al. 2004). It is possible that if sister-chromatid cohesion is reduced, some cells in which MATa is repaired after DNA replication might allow the two sisters to search independently for a donor.
The other factors implicated in donor preference are the yKu70 and yKu80 proteins. The Ku complex plays many roles in chromosome architecture and in DSB repair. They are required for the predominant mechanism of nonhomologous end joining that can rejoin the 4-bp 3′-overhanging DSB ends created by HO. They are also critical in associating telomeres with the nuclear periphery and in ensuring the full activity of telomerase. When Ku proteins are deleted, telomeric regions are delocalized from the periphery. But, paradoxically, it seems that HML is more frequently associated with the nuclear periphery in a yku70Δ mutant. This might explain why, when either Ku protein is deleted, there is an ∼10% reduction in HML usage in MATa but no discernible effect on MATα.
It seems that Swi4-Swi6, Yku70-yKu80, and Chl1 may all act in a common pathway, as double mutants among this set continue to use HML ∼30% of the time, whereas fkh1Δ yku80Δ and fkh1Δ chl1Δ both resemble fkh1Δ SCBΔ (Coïc et al. 2006b). There is some additional contribution of these genes, however, because fkh1Δ SCBΔ chl1Δ yku80Δ reaches to the same low level as REΔ.
How does RE work?
RE does not alter the silencing of HML. One way that HML might be used more efficiently in MATa would be if HML were much more accessible in MATa, but this does not seem to be the case. First, there is no obvious difference in the positioning of nucleosomes over HML in MATa vs. MATα strains (Weiss and Simpson 1998). Second, whereas HML is fully accessible to HO endonuclease in a sir3Δ mutant that abolishes silencing, there is no evident change in the trace of HO cleavage in a vs. α strains when GAL::HO is overexpressed. Finally, if HML and its adjacent silencer sequences are replaced by a similar-sized segment of MATα-inc (which HO cannot cut), it is not a better donor than the normally silenced HMLα locus in a RE-deleted MATa strain (Coïc et al. 2011).
One attractive idea to explain RE’s role is that it changes the localization or the higher-order folding of the entire left arm of chromosome III to make it more flexible in locating and pairing with the recipient site in MATa cells. In this view, the chromosome arm would be sequestered or immobilized (perhaps by being bound to the nuclear envelope) in such a way that HML is unavailable in MATα or REΔ MATa cells, even though the chromatin structure at HML itself was unchanged. Several approaches suggest that there are differences in chromosome arrangement in the two mating types, but these changes, prior to creation of a DSB, do not seem to explain donor preference. First, chromosome segments near MAT, HML, and HMR can be fluorescently tagged by binding LacI-GFP or TetR-GFP (or some other color) to LacO or TetO arrays. In one such study of fixed cells, there was a small difference in MATa vs. MATα (Bressan et al. 2004), but in MATa, HML is not closer than HMR. A more detailed study (I. Lassadi and K. Bystricky, personal communication) suggests that in MATa cells, HML is being drawn closer to the centromere. Both fluorescent and chromosome conformation capture (3C) techniques have suggested that HML and HMR are relatively closer together than either is to MAT (Miele et al. 2009), but again recent sequencing-based chromosome conformation capture (termed 5C) finds HML being more strongly associated with the centromere (J. Dekker, personal communication).
Of course for switching to take place, HML and MAT or HMR and MAT must come into contact. Although HML and HMR have been shown to preferentially reside near the nuclear periphery, it appears that this tethering does not prevent the donors from engaging MAT away from the periphery, since Bystricky has observed among cells undergoing switching that MAT remains in the center of the nucleus (Bystricky et al. 2009).11
RE can be mimicked by tethering Fkh1 to the LexA operators
As noted before, RE can be replaced by as few as four copies of the A subdomain of the RE. The strong use of HML in using 4xA is eliminated by deleting Fkh1. To explore how Fkh1 might act, we took the approach of replacing RE with four copies of the lexA operator and expressing lexA-Fkh1 (J. Lin, E. Coïc, and J. E. Haber, unpublished results). This construct significantly increased the use of HML in comparison to expressing LexA or LexA-Swi4. Subsequently, LexA-Fkh1 was further truncated so that it contained only the first 120 amino acids, comprising the forkhead-associated (FHA) domain. This construct proved to have even stronger HML usage, whereas other fusions carrying the transcription regulatory domain near the C terminus had no activity (Figure 10). FHA domains have been shown to bind phosphothreonines. Taken together these data suggested that RE might act by binding to phosphothreonines that would only be created or exposed after induction of a DSB. There are a number of proteins that exhibit DSB-induced post-translational modification of both series and threonines, including histones, proteins bound near the DSB end such as Sae2 and RPA, and which might serve as the target for RE’s bound FHA domains. Supporting this hypothesis, we showed that mutation of conserved histidine and arginine residues, which are found in all FHA domains, abolished LexA-FHA activity (Figure 10).
If Fkh1’s FHA domains bound to RE are responsible for moving HML close to MAT, then it should be possible to show that FHA associates with MAT only after a DSB is induced; indeed ChIP experiments have shown this to be the case, even in strains where HML itself is deleted (J. Lin, E. Coïc, and J. E. Haber, unpublished results). Conversely, a DNA damage-dependent chromatin modification that spreads around the MAT locus, the phosphorylation of S129 of histone H2A (called γ-H2AX) also spreads to surround the normal RE region (K. Lee and J. E. Haber, unpublished observation). However this Mec1/Tel1-dependent modification is only seen in MATa cells when RE is able to bind Fkh1.
Neither γ-H2AX nor the DSB-dependent phosphorylation of histone H4-S1 is responsible for donor preference. What remains to be determined is the phosphothreonine-binding partner of the Fkh1-FHA domain and the identity of the damage-dependent protein kinase that is employed. Neither Mec1 nor Tel1 appear to play a role, but strong donor preference is dependent on casein kinase II (J. Lin, E. Coïc, and J. E. Haber, unpublished results).
So, at present we are coming close to understanding the mechanism that underlies donor preference. A simple model to explain donor preference is presented in Figure 11. The cluster of Fkh1 proteins bound at RE in MATa (but not MATα) comes into contact with DSB-induced, casein kinase II-dependent phosphorylated threonines in proteins that bind near the DSB ends. This association effectively tethers the nearby HML locus close to MAT and facilitates its use in MAT switching. It is still not clear whether there are any other constraints preventing HML use in MATα cells or whether there are any facilitating sequences that aid HMR usage. Of course this is hardly the entire story since RE also binds Swi4/Swi6 in domain C, in a cell-cycle–dependent fashion. The activity of the entire RE is likely to be substantially more elaborate.
Mating-type switching and donor preference in other yeasts
Although other distantly related yeasts switch mating type and have silent donor cassettes and even donor preference, these processes are surprisingly different. I once quipped that MAT switching in Schizosaccharomyces pombe is the same as that in S. cerevisiae except in every detail.
Mating-type gene switching in fission yeast, S. pombe (Klar 1993; Thon and Klar 1993) also has an expressed mating-type locus, mat1 locus, which can carry either nonhomologous P or M alleles. Switching from one allele to the other uses one of two silent donors (mat2P and mat3M), which are close together and located only a short distance away from mat1 on the same chromosome arm. mat1, mat2, and mat3 share regions of homology flanking the P or M region, but they are unrelated to those in budding yeast. The two donors are silenced by heterochromatin established in small part by Sir2 but mostly by a silencing system that is absent in yeast, using a homolog of the Su(var)9 histone methyltransferase, Clr4, and the HP-1 relative, Swi6 (Nonaka et al. 2002). And there is no HO-like enzyme. Instead, a persistent single-strand nick is created at mat1, which is converted to a DSB when cells enter S phase (Arcangioli 1998; Kaykov and Arcangioli 2004). Thus only one of the two daughter cells can switch, but the repair is directed to mat2 or mat3 rather than to repair from the intact sister. Repair therefore takes place in the context of normal replication and the repair is conservative in that both the strands at the mat1 locus are newly copied (Arcangioli and De Lahondes 2000). Mutations that alter donor preference also alter silencing, which is not the case for S. cerevisiae (Thon et al. 1994; Thon and Friis 1997). Donor preference involves dramatic changes in chromatin modification and structure (Jia et al. 2004). Thus, although the two systems seem to share some common features, they seem to be the consequence of convergent evolution.
The switching system of the more closely related K. lactis is substantially more similar to S. cerevisiae, in having sequences that are recognizably similar to MATa1, MATα1, and MATα2, but the shared flanking sequences are not closely related to S. cerevisisae’s MAT-W, X, or Z1/Z2. HML carries not only α1 and α2 but also a novel gene, α3, while HMR has a1 and a2; both HML and HMR are silenced by a Sir2-dependent mechanism (Sjostrand et al. 2002). Switching is dependent on the Mts1 protein that is the homolog of the S. cerevisiae repressor RME1; but here it is required to activate switching and is turned off in MATa/MATα cells by an a1-α2 repressor. But the big surprise is that there is no functional HO gene, although there is evidence of an eroded, ancient HO gene. In its place—at least for MATα to MATa switching—is the α3 gene, which proves to be a transposable element that can excise from the DNA as a circle and somehow catalyze switching, dependent on Mts1 (Barsoum et al. 2010). MATa to MATα switching proceeds without the α3 gene, but it proceeds by the formation of a DSB with hairpin intermediates, reminiscent of transposon excision. Like the repair of hairpin intermediates in S. cerevisiae generated by excision of the plant Ds transposon (Yu et al. 2004), the hairpin ends require Mre11 to be cleaved for further steps in repair (Barsoum et al. 2010). MATa switching seems to be under the control of a different transposable element (S. Aström, personal communication).
In reviewing what we have learned about mating-type genes and the recombination process leading to MAT switching, one tends to focus on recent experiments that have dramatically illuminated the subject. Inevitably some of the pioneering work that laid the foundation tends to become obscured. In some cases, gene names have been changed to reflect a more comprehensive understanding of their function (e.g., MAR1 is now SIR2, HMa is now HMLα, HDF1 is now YKU70). We owe particular debts of gratitude to: Donald Hawthorne (1963), who published little but inspired many with his recounting the first pedigree analysis of MAT switching and the creation of fusions of MAT and HMR (before HMR had been defined), which provided essential keys to the model proposed by Hicks, Strathern, and Herskowitz (1977); Isamu Takano and Yasuji Oshima, whose early studies of MAT switching in the early 1970s included the formulation of their seminal transposable “controlling element” model (Takano and Oshima 1970; Oshima and Takano 1971); Vivian MacKay and Thomas Manney, whose sterile mutations, including those in MATα1 and MATα2 provided not only insights about signal transduction but also most of the genetic reagents that were used to demonstrate the way MAT alleles were replaced during homothallic switching (Mackay and Manney 1974a,b); Robert Mortimer, whose lab identified most of the key RAD genes needed for recombination (Rodarte-Ramon and Mortimer 1972; Game and Mortimer 1974) and who, along with Seymour Fogel, established most of the basic rules about gene conversion (Fogel et al. 1979). Michael Resnick also was the first to propose DSB repair mechanisms that fundamentally resemble SDSA and the double Holliday junction mechanisms that we continue to invoke (Resnick 1976).
I am grateful for the helpful comments of Laura Rusche, Scott Holmes, and members of my lab. The part of this work that derives from my own lab has been carried out for more than 30 years by an exceptional contingent of graduate students, postdoctoral fellows, technicians, and Brandeis University undergraduates and has been generously supported, primarily by National Institutes of Health (NIH) grant GM20056, but also by NIH grants GM61766 and GM76020, as well as previous support from the National Science Foundation and the Department of Energy.
↵1 Although haploid-specific genes are turned off in MATa/MATα diploids, they can also be turned off in haploids if both MATa and MATα are coexpressed. This can occur in a strain disomic for chromosome III, which carries the MAT locus, or if a plasmid expressing the opposite MAT locus is transformed into a haploid strain. The repression of haploid-specific genes can also occur in a haploid if the two silent mating-type donor sequences HMLα and HMRa are expressed, for example in sir2Δ mutants that fail to maintain their silencing. In contrast, diploid-specific genes are not turned off simply because the cell has two copies of every chromosome. A MATa/MATa or a matΔ/MATα diploid behaves like an a- or α-mating haploid.
↵2 If HO is expressed only for a short time and the HO protein is then rapidly degraded, then nearly all of the religation events anneal the 4-bp overhanging 3′ ends and regenerate the HO cutting site. If HO is expressed continuously, then perfect religation is futile and most cells die. About 1/500 cells survives by imperfect end joining, destroying the cleavage site. The great majority of these imperfect NHEJ events result from misalignments of the 5′ AACA 3′ and 3′ TGTT 5′ complementary ends either to delete 3 bp or to insert CA or ACA (Kramer et al. 1994; Moore and Haber 1996).
↵3 Cells arrested at the “start” point of the cell cycle, e.g., by α-factor–mediated arrest, are unable to initiate HR; however, this is not the case for cells that have progressed past start but are still in G1, that is, prior to the initiation of S phase. Repair of a DSB by gene conversion does not depend on Cdc7-Dbf4, which are required to initiate DNA replication (Ira et al. 2004). Hence cells arrested by inactivating Cdc7 are competent for HR as well as NHEJ.
↵4 This experiment is probably the first chromatin immunoprecipitation experiment.
↵5 The subsequent development of a detailed description of cell lineage in Caenorhabditis elegans has overshadowed the importance of the analysis of cell fate in budding yeast.
↵6 The rapid degradation of HO, even after galactose induction, contrasts with the stability of the I-SceI protein, which is also used to study induced DSB repair.
↵8 One of the annoying aspects of the Saccharomyces Genome Database is that—because the original sequence was performed on a MATα strain—MATa does not exist! However the sequence of HMRa of course carries the same Ya region as MATa, so one can find the sequence. However, it happens that HMRa in the reference strain S288c and its derivatives (e.g., the oft-used BY4741 and BY4742) carries a “stuck” mutation at position Z11, so that this sequence is very poorly cleaved if used as a cleavage site. This is also the case when this HMRa sequence is switched to replace MATα. Worse, it turns out that, whereas MATa-stk is poorly cleaved, MATα-stk is not cut at all (Ray et al. 1991).
↵9 Most likely the exonuclease(s) are removed by a more rapidly moving repair DNA polymerase, which is needed to fill in ssDNA regions when the repair is completed. Genetic experiments suggest that the translesion DNA polymerase Polζ acts at this step (Holbeck and Strathern 1997). But there must be alternative polymerases because deleting REV3 does not impair the completion of repair.
↵10 When only one end of the DSB shares homology with a donor sequence, gene conversion is not possible. Instead, the strand invasion leads to the assembly of a complete leading- and lagging-strand replicative fork, which can process down the entire chromosome arm to the telomere, in some cases copying at least 100 kb (reviewed by McEachern and Haber 2006).
↵11 A number of studies have shown that when a DSB cannot be repaired, it becomes associated with the nuclear periphery (Gartenberg 2009). But this pathological state is only seen after several hours and does not seem to reflect what happens during a successful DSB repair event.
Communicating editor: J. Thorner
- Received September 10, 2011.
- Accepted November 4, 2011.
- Copyright © 2012 by the Genetics Society of America