R-loop-mediated genome instability in mRNA cleavage and polyadenylation mutants

Rad52 foci screening of essential CIN genes

Rad52p is essential for homologous recombination and organizes into repair centers in response to double-strand breaks (DSBs) and other recombination events (Lisby et al. 2001; Mortensen et al. 2009). Increases in Rad52p foci can therefore indicate a number of genome-destabilizing conditions, including increased DSBs, inefficient resolution of recombination intermediates, or hyperrecombination (Alvaro et al. 2007). To determine which CIN mutants cause increased or prolonged engagement of the homologous recombination machinery, we introduced a RAD52-YFP fusion under the control of its native promoter by synthetic genetic array (SGA) (Tong et al. 2004). The resultant strains expressing a mutant CIN gene and RAD52-YFP were screened visually for increased levels of Rad52 foci (see the Materials and Methods). We screened 360 alleles of 305 essential CIN genes, including 306 ts (temperature-sensitive) and 54 DAmP alleles (Stirling et al. 2011). Similar to previous studies, we retested mutants in which ≥15% of cells had Rad52 foci in the primary screen (Alvaro et al. 2007). Triplicate retesting produced a list of 46 alleles in 44 unique genes whose mutation elicits an increased level of Rad52 foci (Fig. 1A). Eighteen additional mutants had increased but variable levels of Rad52 foci that did not meet our threshold across replicates (Supplemental Table S1). The largest functional group of mutants with increased Rad52 foci affected DNA replication, consistent with the S-phase function of Rad52 and its role in repairing damage caused by collapsed replication forks (Lisby et al. 2001). Interestingly, mutations in multiple genes involved in the proteasome, Smc5/6 complex, early secretion, transcription, and mRNA processing all caused increased rates of Rad52 foci (Fig. 1). These data stress the value of screening essential gene mutant collections, since a systematic screen of nonessential gene deletions uncovered a nonoverlapping set of biological processes (Alvaro et al. 2007).

Published genetic interactions exist between rad52Δ and 12 of 44 genes identified in our screen (i.e., BioGrid/DRYGIN databases). The presence of Rad52 foci in a mutant may indicate the dependence of that mutant on RAD52 function (Alvaro et al. 2007). To further this hypothesis, we directly tested selected mutants from different biological pathways for genetic interactions with rad52Δ. For pol31-ts, sld7-ts, rpn5-1, orc6-ts, and rna15-58 there were clear growth defects in the absence of RAD52 (Fig. 1B). For ssl1^T242I and pcf11-2, a growth defect was apparent when a low level of hydroxyurea (HU) was present, whereas nse3-ts and sec12-4 did not show obvious growth defects in this assay (Fig. 1B).

Overall, our screen indicates that a large proportion of essential CIN genes (i.e., 14%; 44 of 306 genes) exhibit significant levels of mutant-induced Rad52 foci and thus that disruption of unexpected cellular pathways create a requirement for homology-directed DNA repair. Of particular interest was the identification of seven mCP genes in the Rad52 foci screen (Fig. 1A). The mCP machinery is essential for processing nascent RNAs to free polyadenylated species (Gross and Moore 2001). Interestingly, unlike the majority of mutants identified in the screen, which arrested as large-budded G2/M cells, most mCP mutants exhibited Rad52 foci at all cell cycle stages, including G1. By comparison with those Rad52 foci-forming mutants involved in DNA replication, the role of mCP genes in genome integrity is not well characterized, and we decided to characterize it further.

Fragile sites in mCP mutants differ in transcribed ORFs and near replication origins

Phosphorylation of H2A-Ser129 is a mark of DNA damage analogous to human γ-H2AX modification in response to DNA damage and is critical for DSB repair (Downs et al. 2000). Recent analysis of sites of H2A-S129 phosphorylation in the yeast genome revealed a set of fragile loci or “γ sites” using ChIP–chip (Szilard et al. 2010). It is known that h2a-S129A mutants accumulate Rad52 foci and that both Rad52 and H2A phosphorylation contribute to efficient DNA repair (Downs et al. 2000; Mortensen et al. 2009; Szilard et al. 2010). We reasoned that mapping differences in fragile sites between wild type and mCP mutants could suggest a mechanism for CIN in the mutants. We performed γ-H2A ChIP–chip on the two mCP alleles with the highest levels of Rad52 foci: pcf11-2 and clp1-ts. We applied double-T7-amplified anti-phospho-Ser129-H2A chromatin immunoprecipitates from log phase wild-type, mCP mutant, or h2a-S129A control cells to a high-density tiling microarray containing 3.2 million probes with an average 5-base-pair (bp) resolution and 20-bp overlap between probes (Schulze et al. 2009). Normalizing our data to the h2a-S129A control produced a profile of yeast γ sites (Fig. 2A; see also Supplemental Table S2 for correlations; Supplemental Table S3 for all ChIP–chip peak coordinates).

Overall, our wild-type ChIP–chip profiles confirmed the findings of Szilard et al. (2010); for example, identifying γ sites enriched in subtelomeric regions, replication origins, long terminal repeats, repressed ORFs, MAT, and near rDNA and centromeres (Fig. 2A; Supplemental Figs. S1, S2; Supplemental Material). Initial examination of the mutant data revealed similar γ-site peak profiles in replicates of clp1-ts, pcf11-2, and wild-type strains (Fig. 2A; Supplemental Table S2). For example, Figure 2B shows similar mutant and wild-type γ sites associated with a repressed gene, HXT10, and a replication origin in the indicated 50-kb segment of chromosome VI. To highlight regions of the mutant-specific γ-H2A signal, which could be linked to mutant-induced DNA damage, we normalized the mutant profiles to that of the wild type (Fig. 2C). This analysis revealed reproducible and widespread differences in γ sites adjacent to replication origins (e.g., ARS603.5 in Fig. 2C).

To quantify the observed connection between mCP mutant γ-site enhancement and autonomous replicating sequence (ARS)-linked sites, we scanned a 2-kb window on either side of all ARSs for enriched regions ≥500 bp in length. This analysis found significantly more sites than would be expected at random for both clp1-ts and pcf11-2 mutants (i.e., P = 0.99997) (Fig. 3A compares the observed values with the predicted number of peaks if ARS start coordinates were randomized). Sixty-four ARSs met these stringent criteria for three of four mutant replicates (i.e., the sites are found at least once in both clp1-ts and pcf11-2 and are found in two replicates of either clp1-ts or pcf11-2) (Fig. 3A; Supplemental Table S4). Very few ARSs were themselves enriched in the mCP-specific γ sites; instead, the peaks appeared in ORFs in close proximity to the ARS. The peaks of 60 of the 64 ARS-linked sites could be assigned to single genes. Sixty-five percent of these genes were oriented toward the ARS, similar to the genome-wide frequency of ORF–ARS collision (60%). Notably, ARS-linked γ-H2A-enriched ORFs oriented in such a colliding fashion had a higher average transcriptional frequency than the complete set of colliding ORF–ARS pairs, while this trend was not seen for noncolliding ORF–ARS pairs (Fig. 3A, right panel; Supplemental Table S4). These observations are consistent with a mechanism contingent on R-loop formation, since they implicate the collision of transcription and DNA replication as a source of DNA damage in mCP mutants.

In addition to the ARS-linked γ sites, there was mutant-specific γ-H2A enrichment at a set of 918 ORFs throughout the genome. When we examined the transcriptional frequency of these 918 ORFs, we found significantly fewer of the lowest transcription category (i.e., less than one mRNA per hour) compared with all ORFs (Fig. 3B; Holstege et al. 1998). This represents a dramatic shift from the wild-type γ sites, which strongly enriched repressed or weakly transcribed genes and underrepresented higher transcriptional categories (Fig. 3B; Szilard et al. 2010). If we map the relative position of γ-site differences within the 918 ORFs, there is a slight bias toward the 3′ end of the gene for both pcf11 and clp1-ts mutants compared with enriched ORFs in the global γ-site profile of wild-type cells (Supplemental Fig. S3). Moreover, examining the 55 genes whose 3′ untranslated region (UTR) contains a γ site in both clp1-ts and pcf11-2 profiles (i.e., three of four replicates covered by >50% of a 500-bp window downstream) reveals that the associated genes tend to have higher transcriptional frequencies (Supplemental Fig. S3). Together, these data suggest that at least some transcription occurs at ORFs specifically enriched for γ-H2A in mCP mutants and that damage may be biased to the site of mCP function at 3′ ends. One potential reason that the bias is not more pronounced is that various 3′ end processing factors, including the mCP, also affect transcription elongation and could be causing damage within genes as well as downstream (Tous et al. 2011).

If DNA damage in pcf11-2 and clp1-ts mutants occurs at transcribed ORFs, it is plausible that reducing transcription could mitigate the increase in Rad52 foci. To test this hypothesis, we treated cells with 6-azauracil (6AU), an inhibitor of transcription elongation, and scored the presence of Rad52 foci. Wild-type cells showed an increase in Rad52 foci when treated with 6AU (Fig. 3C). Remarkably, both clp1-ts and pcf11-2 strains showed a significant decrease in Rad52 foci when treated with 6AU (Fig. 3C). This effect was not seen in the strong hits srm1-ts or cdc24-11, which both retained high levels of Rad52 foci in 6AU (Fig. 3C). We conclude from this experiment that at least some of the DNA damage in mCP mutants is occurring within transcribed regions.

CIN in mCP mutants occurs via R loops

Our data show that mutants in the mCP pathway accumulate Rad52 foci, and our ChIP–chip data suggest a connection between DNA replication and the role of mCP in transcription. It is known that transcription can act as a replication fork barrier (Deshpande and Newlon 1996; Takeuchi et al. 2003) and that mutations in mCP components can cause transcription defects (Birse et al. 1998; Prado and Aguilera 2005). To investigate the connection of these phenotypes to CIN, we first confirmed reported mCP phenotypes in three standard CIN assays, investigating primarily chromosome loss (chromosome transmission fidelity [Ctf]), loss of heterozygosity (BiMater [BiM]), and rearrangement/gene conversion (a-like faker [ALF]). Mutants in CFT2, CLP1, FIP1, PCF11, and RNA15 had a robust Ctf phenotype and a weak ALF phenotype; RNA15, CLP1, and PCF11 mutants also had a detectable BiM phenotype (Supplemental Table S5). Each mCP mutant was hypersensitive to the genotoxic agents HU, cisplatin, bleomycin, and UV (Fig. 4A; data not shown). These agents may synergize with the DNA damage that occurs in mCP mutants; for example, failed transcription termination in mCP mutants could act as a block to replication forks, thereby sensitizing cells to replication inhibitors like HU or cisplatin (Birse et al. 1998; Prado and Aguilera 2005). Indeed, HU and certain mCP mutations both increase recombination rates; thus, the effects may be additive (Luna et al. 2005).

Our data and the literature support the notion that transcriptional defects underlie genome instability in mCP mutants. One common way by which this occurs is through formation of transcription-coupled RNA:DNA hybrids called R loops. The identification of pcf11-2 and clp1-ts γ-site differences in ARS-proximal genes, combined with the suppression of Rad52 foci by 6AU, makes this an especially attractive model (Fig. 3). Consistent with this explanation of our CIN data, we found that known R-loop-forming mutations in the helicase SEN1 and the THO components MFT1 and THP2 lead to strong CTF and ALF phenotypes (Supplemental Table S5). To test the R-loop hypothesis for mCP mutants, we expressed a recombinant RNaseH enzyme that specifically degrades RNA:DNA hybrids. Analysis of Ctf in pcf11, clp1, fip1, and rna15 mutant strains showed a remarkable suppression by RNaseH, which should reduce R loops, supporting a direct role for R loops in genome instability (Fig. 4B). Suppression of Ctf by RNaseH was not seen in Ctf control strains representing other biological pathways (i.e., rpf1-1 or scc2-4 in Fig. 4B; quantified in Supplemental Fig. S4). Thus, RNaseH suppression of Ctf is highly specific to mCP mutants.

S. cerevisiae encodes at least two RNaseH enzymes encoded by RNH1 and RNH201. Deletion of both RNaseH genes, which increases the frequency of R loops (see below), caused synthetic growth defects at a semipermissive temperature (32°C) in CFT2, CLP1, PCF11, or RNA15 mutants (Fig. 4C). While we did not identify a growth defect associated with RNaseH deletion in fip1-ts, the triple-mutant strains did show enhanced sensitivity to low concentrations of bleomycin (Supplemental Fig. S5). These data indicate that mCP mutants require RNaseH activity under certain conditions to promote viability, consistent with a detrimental accumulation of RNA:DNA hybrids.

To confirm that the mCP mutants accumulate R loops, we transiently grew cells at the nonpermissive temperature of 37°C and harvested them for chromosome spreads. We performed immunofluorescent detection of RNA:DNA hybrids using the S9.6 monoclonal antibody in chromosome spreads of wild type; mCP mutants; known or predicted R-loop-forming mutants, including rnh1Δrnh201Δ double mutants; and a panel of Rad52 foci-forming strains from other biological pathways (Fig. 4D). For each allele of the mCP machinery tested, we observed an increased number of chromosome spreads with RNA:DNA hybrid staining. The rnh1Δrnh201Δ double mutant also showed high levels of RNA:DNA hybrids, confirming that the role of endogenous RNaseH is actively suppressing R-loop formation (Fig. 4D). Indeed, the rnh1Δrnh201Δ strain exhibits a strong CIN phenotype (Supplemental Table S5). As predicted, mutants in the THO subunits MFT1 and THP2 and in SEN1 caused RNA:DNA hybrids (Fig. 4D; Gomez-Gonzalez et al. 2009; Mischo et al. 2011). Deletion of the transcription termination factor RTT103 also increased RNA:DNA hybrids (Fig. 4D). Among the panel of Rad52 foci-forming control strains, only srm1-ts showed an increase in RNA:DNA hybrids (Fig. 4D). SRM1 plays a role in nucleocytoplasmic transport of mRNAs, which could account for the accumulation of RNA:DNA hybrids in srm1-ts mutants. As an antibody specificity control, we show that pretreatment of chromosome spreads with recombinant RNaseH in vitro significantly reduces the R-loop signal (Supplemental Fig. S6). Together, these data suggest that mCP mutants are similar to THO and SEN1 mutants in that the mechanism of genome instability requires the accumulation of RNA:DNA hybrids.

Comparative analysis of mCP mutants with known R-loop-forming mutants

Known R-loop-forming mutants (e.g., THO complex and SEN1) work in parallel with mCP in the continuum of transcription termination and mRNA processing and export. Since mCP mutants have defects in transcription elongation and termination and mRNA processing and export, we wanted to address their similarity to known R-loop-forming mutants to probe the mechanism by which mCP mutant-induced R loops cause genome instability (Brodsky and Silver 2000; Luna et al. 2005; Tous et al. 2011). Mutants in THO and the SEN1 helicase exhibit CIN phenotypes and RNA:DNA hybrid formation in our cytological assay (Fig. 4D; Supplemental Table S5). Moreover, representative alleles in mCP (i.e., fip1-ts, clp1-ts, rna15-58, and alleles of pcf11) exhibit synthetic growth defects or lethality with deletion of the THO subunit THP2 or the termination factor RTT103 or with the sen1-1 ts allele, as assessed by tetrads and growth curve analysis (Supplemental Fig. S7). This analysis confirms that the cellular defects caused by our mCP mutants are buffered by the action of THO and SEN1 and vice versa.

Another phenotype common to R-loop-forming mutations is hyperrecombination, and a weak phenotype has been reported for certain components of mCP (Chavez et al. 2000; Luna et al. 2005; Mischo et al. 2011). To clarify these observations, we subjected the mCP mutants to a well-characterized direct repeat recombination assay (Prado et al. 1997). At 30°C, the mCP mutants clp1-ts, rna15-58, and pcf11-ts10 exhibited significant hyperrecombination (Supplemental Fig. S7). In this assay, mutations in SEN1 or the THO complex lead to strong hyperrecombination phenotypes; however, the introduction of a transcription terminator after the first homology segment blocks hyperrecombination in THO mutants but stimulates recombination in the sen1-1 allele (Mischo et al. 2011; Supplemental Fig. S7). Thus, while THO and SEN1 mutants both induce R loops, this reporter system mechanistically differentiates RNA packaging defects (i.e., THO mutants) from SEN1 helicase defects. Unlike sen1-1, hyperrecombination in mCP mutants is not increased by the transcription terminator, but rather is generally reduced (Supplemental Fig. S7). Thus, mCP mutants behave like THO mutants, consistent with the known interdependence of THO and mCP function in mRNA processing and export (Luna et al. 2005).

Truncation fusion of yeast FIP1 causes genome instability

Our data suggest that the mCP machinery prevents genome instability by suppressing the formation of RNA:DNA hybrids in a manner at least partially similar to the THO complex. The human ortholog of the mCP gene FIP1, FIP1L1, comprises the N-terminal moiety of a fusion protein with PDGFRα, which plays a causative role in 10%–20% of hypereosinophilic syndrome/eosinophilic leukemia (Cools et al. 2003; Gotlib and Cools 2008). These oncogenic fusion proteins truncate the C terminus of FIP1L1 and liberate the tyrosine kinase domain of PDGFRα, which becomes dysregulated. While the PDGFRα moiety is responsible for the proliferative phenotype, the impact of the truncation on FIP1L1 function is not clear (Stover et al. 2006). However, the fusion is under the control of the FIP1L1 promoter, and in vitro colony-forming assays indicate that FIP1L1 does have an effect on the fitness of transformed cells (Buitenhuis et al. 2007; Gotlib and Cools 2008). Thus, the role of FIP1L1 may be important to the etiology of this leukemia.

To model the putative effects of FIP1 truncation fusions, we fused a commonly used globular domain (GFP) to two orthologous breakpoints in the yeast FIP1 gene. The fusions Fip1^1-213-GFP and Fip1^1-279-GFP correspond to human exons 7 and 9, respectively, in the most closely related human FIP1L1 isoform, isoform 4 (Fig. 5A, left panel). Tetrad dissection of heterozygous diploids revealed that each truncation fusion affected fitness; Fip1^1-213-GFP did not support viability, whereas Fip1^1-279-GFP produced small colonies (Fig. 5A, right panel). A control in which GFP was fused to the C terminus of full-length FIP1 did not appear to affect fitness. Since the Fip1^1-279-GFP construct was viable, we focused our subsequent characterization on this allele. Similar to the fip1-ts allele, we found that Fip1^1-279-GFP caused a Ctf phenotype not seen in the FIP1-GFP control (Figs. 4B, 5B). In addition, the truncation fusion protein increased formation of Rad52 foci and RNA:DNA hybrids, although in this case, the full-length FIP1-GFP also increased focus and hybrid formation above wild-type levels (Fig. 5C,D). This indicates that perturbations of FIP1 function, even those that do not grossly affect fitness, can induce R-loop-mediated DNA damage and supports the idea that human FIP1L1, in the context of a PDGFRα fusion, may be only partly functional. Our data also predict that more severe truncations would further inactivate FIP1L1.

Human FIP1L1 may be required for genome stability

Our examination of CIN phenotypes in yeast is ultimately directed at predicting human genome integrity pathways (Yuen et al. 2007; Barber et al. 2008; Stirling et al. 2011). Therefore, an important confirmation of our data is a cross-species validation of the CIN phenotype. Partial reduction of yeast FIP1 activity by a ts allele or truncation fusion leads to Rad52 foci and CIN in yeast cells (Figs. 1, 5; Supplemental Table S5). Thus, we predicted that reduction of human FIP1L1 function would also lead to genome instability in human cells.

We transiently reduced the protein levels of FIP1L1 in the near-diploid colorectal cell line HCT116 by transfection with pooled or, as a specificity control, individual siRNAs (Fig. 6A). We directly assessed chromosome integrity in cells with reduced FIP1L1 by analysis of mitotic metaphase chromosome spreads. When cells were treated with FIP1L1 siRNA, we reproducibly observed a significant increase in the appearance of chromosome breaks and chromosome fragments as compared with cells treated with a GAPDH control siRNA (P < 0.05) (Fig. 6A,B). The aberrant chromosomal structures resemble the damage seen in chromosome spreads from bleomycin-treated cells (data not shown). Consistently, high-content analysis of viability suggested that FIP1L1 siRNA treatment was toxic to cells (data not shown). In yeast, the CIN phenotype of fip1-ts is related to DNA damage (Figs. 1, 4). Therefore, reduction of FIP1L1 should also lead to DNA damage. In support of this, we found that FIP1L1 siRNA treatment increases the number of 53BP1-mCherry foci (Fig. 6C,D). 53BP1 foci mark sites of DNA damage and H2AX phosphorylation and have recently been shown to increase in human cells depleted for THO (Dominguez-Sanchez et al. 2011). These data support the hypothesis that, at least for FIP1L1, the essential roles of yeast mCP genes in suppressing DNA damage and genome instability are conserved to human cells.

Rad52 foci screening of essential CIN genes

Focused secondary screens are a powerful way to define specific functions within primary functional genomic data (e.g., Doheny et al. 1993; Michaelis et al. 1997; Carroll et al. 2009; Stirling et al. 2011). By direct screening of more than one-quarter of essential genes, we linked 44 essential CIN mutants to increased Rad52 foci formation, including seven mCP factors with an unappreciated role in genome integrity (Fig. 1). Rad52 foci indicate organization of the homologous recombination machinery into a repair center in response to DNA damage (Lisby et al. 2001). Our genetic data support the idea that increased Rad52 foci often indicate a requirement for Rad52 function (Fig. 1B; Alvaro et al. 2007).

Aside from the mCP mutants that were the focus of follow-up studies, the Rad52 foci screen identified mutants with diverse functions. The highest levels of foci were seen in srm1-ts and cdc24-11 alleles (Fig. 1). SRM1 strongly interacts with nucleosomes, and mutants cause gross chromosomal defects, although we also found evidence that srm1-ts may cause RNA:DNA hybrids, potentially due to its role in RNA export (Fig. 4; Koerber et al. 2009). CDC24 encodes a guanine–nucleotide exchange factor for the small GTPase Cdc42, and while the mechanism of cdc24-11 induced Rad52 foci is unknown, a recent high-content screen identified foci of the DNA damage marker Ddc2-GFP in alleles of CDC24 (Li et al. 2011). Mutations affecting DNA replication were the class that most frequently exhibited enhanced Rad52 foci and usually arrested as large-budded G2/M cells (Fig. 1). DNA replication mutants were not significantly identified in the Rad52 foci screen of nonessential genes because so many of the components are essential. The same is true of essential transcription/mRNA processing, and proteasome subunits that were not identified in the literature (Alvaro et al. 2007). Conversely, Alvaro et al. (2007) identified DNA repair, mitotic checkpoint, and other genes not found in our survey of essential CIN mutants. Screening of both essential and nonessential gene mutants therefore provided complementary, nonoverlapping data sets.

Profiling fragile site changes in Rad52 foci-forming mRNA processing mutants

By comparing fragile sites in wild-type and mCP mutant yeast, we formulated a putative mechanism involving transcription-coupled R loops, which we confirmed by direct tests (Figs. 2–4). The robustness of the wild-type γ-site profiles to perturbations by different mutants was somewhat surprising but similar to findings in strains lacking RRM3, a helicase that facilitates replisome processivity (Szilard et al. 2010). Despite the similarities in wild-type and mutant γ sites, we were able to extract meaningful differences within transcribed ORFs adjacent to ARSs and with a slight bias toward the 3′ end of the ORF (Figs. 2, 3; Supplemental Fig. S3). These observations are consistent with the function of mCP in transcription termination and 3′ end processing of pre-mRNAs and suggest that DNA damage may occur near the site of mCP function. Given the role of mCP in transcription, we were surprised that we did not recover the most highly transcribed genes; indeed, this category was underrepresented (Fig. 3B). It is possible that highly transcribed genes may have mechanisms in place to protect the genome from the deleterious effects of high transcription. This idea has been supported by work in mammalian cells showing that highly expressed genes orient codirectionally with replication origins (Huvet et al. 2007; Tuduri et al. 2009). Based on the success of our approach, we predict that mapping changes in fragile sites (e.g., with ChIP–chip of γ-H2A or another DNA repair protein) in other CIN mutant backgrounds could be an informative method to suggest basic mechanisms.

Mechanism of genome instability in mCP mutants

R loops underlie genome instability in a growing number of transcription and RNA processing mutants, including SPT2, TOP1, SEN1, the THO complex, and ASF/SF2 (Li and Manley 2005; Sikdar et al. 2008; Gomez-Gonzalez et al. 2009; Tuduri et al. 2009; El Hage et al. 2010; Mischo et al. 2011; Wahba et al. 2011). R loops have been linked to various genome instability phenotypes, including hyperrecombination, replication defects, DNA damage, and chromosome loss. Our Rad52 screening and ChIP–chip studies suggested a mechanism for genome instability in mCP mutants that involves DNA damage induced by the formation of transcription-coupled R loops (Figs. 1–3). The fact that Rad52-foci are suppressed by reducing transcription with 6AU (Fig. 3D) shows that recombination is linked to transcription in the mCP mutants. We confirm the R-loop mechanism by rescuing Ctf with RNaseH, immunofluorescent detection of RNA:DNA hybrids, and demonstrating genetic interactions of mCP mutants with endogenous RNaseH mutants and known R-loop-forming THO/SEN1 mutants (Fig. 4; Supplemental Fig. S7).

mCP mutants have reported defects in transcription termination, elongation, and mRNA export (Birse et al. 1998; Brodsky and Silver 2000; Tous et al. 2011). More recently, mRNA 3′ end processing has been shown to directly influence transcription reinitiation (Mapendano et al. 2010). Therefore, several potential molecular mechanisms could explain the R-loop formation underlying genome instability in mCP mutants. Importantly, by comparing hyperrecombination phenotypes with mutants in THO and SEN1, we found that mCP mutants behave similarly to THO mutants (Supplemental Fig. S7). This is consistent with the established strong interdependence of THO and mRNA 3′ end processing factors (Luna et al. 2005). It is known that THO occupancy at gene 3′ ends is decreased by ∼50% when mCP genes are mutated, and mCP occupancy at gene 3′ ends is significantly increased in THO mutants (Luna et al. 2005, Rougemaille et al. 2008). Thus, mCP acts upstream of THO in the biogenesis of mRNA export complexes, but THO is required to disassemble mCP machinery at 3′ ends. Therefore, our data best support a mechanism where mCP mutants facilitate RNA:DNA hybrid formation by reducing THO recruitment and/or increasing the time that transcripts are present at gene 3′ ends due to inefficient termination.

While the molecular mechanism of R-loop formation and propagation is complex even for better-characterized mutants in THO and SEN1, we can also make a statement globally about the impact of R-loop-induced DNA damage in mCP mutants. Our ChIP–chip data show that mCP mutant-induced DNA damage occurs within transcribed genes and that the damaged 3′ UTRs tend to be associated with higher transcriptional frequencies (Fig. 3; Supplemental Fig. S3). More importantly, transcribed genes colliding with adjacent replication origins are enriched in DNA damage in mCP mutants, supporting a model in which collision with DNA replication is a major source of DNA damage in mCP mutants (Fig. 3; Supplemental Table S4). This is consistent with literature showing that collision of R loops with the replisome is commonly associated with genome instability (Prado and Aguilera 2005; Sikdar et al. 2008; Tuduri et al. 2009; Gan et al. 2011). In contrast, the fact that mCP mutants can increase Rad52 foci in G1 cells suggests that DNA replication is not strictly required to elicit foci, although it may play a major role (Fig. 1). It may be that mCP mutants cause lesions requiring Rad52 in more than one way, and it will be interesting in future studies to determine the relative contributions of transcription and replication to genome stability in these mutants. One confounding factor for the interpretation of these mechanisms is that we are studying a set of partial loss-of-function mutations in essential genes whose protein products form a multifunctional and dynamic protein complex. Therefore, despite their phenotypic overlap, there are certain to be gene and allele-specific effects within the mCP mutants described here that modulate the phenotypes observed.

Conservation of a CIN phenotype in the human cancer gene FIP1L1

Our data suggest that truncation fusion of human FIP1L1 in eosinophilic leukemia may cause a loss of FIP1L1 function based on homology with yeast (Fig. 5). By demonstrating that reduced FIP1L1 levels lead to increased DNA damage foci and chromosome breaks in mitosis (Fig. 6), we suggest a role for FIP1L1 in genome integrity that is conserved from yeast to humans. These data raise the possibility that genome stability in cancers with FIP1L1 truncation fusion proteins may be impacted by the partial loss of FIP1L1 function. FIP1L1 is dispensable for activation of the PDGFRα kinase activity, and therefore it has been proposed that some aspect of its promoter or the presence of an adjacent fragile site is responsible for the frequent observation of FIP1L1-PDGFRα gene fusions (Stover et al. 2006). Recent confirmation of FIP1L1-RARA gene fusions in acute promyelocytic leukemia shows that oncogenic translocation with FIP1L1 is not limited or specific to PDGFRα (Menezes et al. 2011). In addition, FIP1L1-PDGFRα fusions have been found in patients presenting with acute myeloid leukemia, which may have progressed from eosinophilic leukemia, and T-cell non-Hodgkins lymphoma, both of which are sometimes associated with aneuploidy (Metzgeroth et al. 2007). Translocation may impact the function of another mCP component, human CLP1, which is associated with an oncogenic translocation in mixed-lineage leukemia (MLL) (Tanabe et al. 1996; Cools et al. 2003). We speculate that one outcome of interrupting FIP1L1 or CLP1 could be to destabilize the genome of these leukemias and, when left untreated, in combination with the proliferative phenotype, facilitate further evolution of the cancer genome. It is worth noting that heterozygous mutation of other CIN genes has been shown to increase tumor incidence in mouse models (e.g., Bub1, Aurora kinase, and CENP-E) (for review, see Schvartzman et al. 2010).

A genome-wide compilation of yeast CIN genes identified dozens of transcription and RNA processing-related CIN mutants whose R-loop status is unknown (Stirling et al. 2011). We predict that other CIN mutants in these pathways also induce R-loop formation. Interestingly, a high-throughput study of γ-H2AX DNA damage foci in mammalian cells identified many RNA processing genes, suggesting that this pathway is generally important for genome integrity (Paulsen et al. 2009). Moreover, these investigators found that overexpression of RNaseH could suppress γ-H2AX foci in some instances, supporting the R-loop model for diverse RNA processing pathways (Paulsen et al. 2009). The prevalence of R-loop-mediated CIN in cancer is unclear, although it is tempting to speculate that it could occur in cancers where the orthologs of known R-loop-forming mutants are disrupted (e.g., TOP1 or FIP1) (Cools et al. 2003; Iwase et al. 2003).

Yeast strains and plasmids

Yeast strains were grown in YPD or synthetic complete medium lacking the appropriate amino acid where nutritional selection was required. Serial dilution assays and growth curve analysis were performed as described (McLellan et al. 2009; Stirling et al. 2011). Supplemental Table S6 contains a list of yeast strains and plasmids.

Rad52-YFP foci screening and CIN assays

Rad52-YFP fusions were made by direct transformation and introduced into the relevant strains by SGA (Tong et al. 2004). Log phase cultures expressing Rad52-YFP were shifted for 3.5 h from 25°C to 37°C and then prepared for microscopy as described (Carroll et al. 2009). Both DIC and fluorescence images were collected with Metamorph (Molecular Devices) and analyzed using ImageJ (http://rsbweb.nih.gov/ij/index.html). In Figure 3, cells growing at 25°C were pretreated with 100 μg/mL 6AU for 20 min before shifting for 3.5 h to 37°C and were imaged as for untreated cells. For each putative hit, the experiment was triplicated and ≥100 cells were counted per replicate. Budding indices were derived from these images. CIN assays were performed exactly as described (Spencer et al. 1990; Yuen et al. 2007).

ChIP–chip

ChIP–chip experiments were performed as described (Schulze et al. 2009) using a ChIP-grade anti-phospho-Ser129 H2A (ab15083, Abcam), except ts alleles were inactivated by shifting for 3 h to 37°C before cross-linking. Data analysis and plotting of enriched features were performed using custom R, MatLab, and python scripts essentially as described (Schulze et al. 2009; Takahashi et al. 2011). All profiles were generated in duplicate, except the h2a-S129A control. The enrichment threshold was 2.5-fold for all analyses. Features were considered enriched if 25% of the feature length (base pair) met the enrichment threshold when normalized to h2a-s129A and if 10% of the feature met the enrichment criteria for mutants normalized to wild type.

ChIP–chip statistical analysis

To determine whether the data recovered more of a particular feature than would be predicted at random, a Monte Carlo simulation on the relevant data set was run using randomly generated start positions for a particular set of features (e.g., ARSs), while maintaining the total number of features within a chromosome (Schulze et al. 2009). Five-hundred simulations were run for each feature to generate a mean and standard deviation. These values were compared with the observed score using the cumulative normal distribution, which calculated the probability of seeing a lower score if a value was selected at random (http://www.stattrek.com). Transcriptional frequency categories and Rad52 foci counts were compared with controls using a Student's t-test (http://www.graphpad.com). Overall transcriptional frequencies for smaller samples were compared using the Mann-Whitney test. Significance levels are indicated in the text or figure legends.

Yeast chromosome spreads

Chromosome spreads were done essentially as described, except that cells were shifted for 4–6 h to 37°C before lysis (Hartman et al. 2000; Wahba et al. 2011). The S9.6 antibody (Hu et al. 2006) was used at 1 μg/mL in blocking buffer (1× PBS with 5% BSA, 0.2% skim milk powder) and detected with Cy3-conjugated goat anti-mouse antibody (Jackson Laboratories, #115-165-003). For RNaseH treatments, each spread was incubated with 2 U of RNaseH (Roche) in blocking buffer for 15 min prior to antibody treatment.

Mammalian tissue culture

The HCT116 cell line was cultured under standard conditions as previously described (Barber et al. 2008). siRNAs were obtained from Dharmacon and applied according to the manufacturer's instructions. Mitotic chromosome spreads were performed essentially as described (Barber et al. 2008). Briefly, actively growing cells were treated with 0.1 μg/mL colcemid for at least 90 min before washing, trypsinization, and addition of hypotonic (0.075 M KCl) for 5 min at room temperature. Cells were fixed in 3:1 methanol:glacial acetic acid, centrifuged (800 rpm for 5 min) and fixed again before mounting on clean slides, DAPI staining, and visualizing. 53BP1 foci were detected in paraformaldehyde-fixed HCT116 cells constitutively expressing 53BP1-mCherry. Foci were scored in Metamorph from three sets of images derived from two independent siRNA transfections.