Abstract
The DNA double-strand breaks that initiate homologous recombination during meiosis are subject to extensive 5′→3′ exonucleolytic processing. This resection is a central and conserved feature of recombination, yet its mechanism is poorly understood. Using a purpose-made deep-sequencing method, we mapped meiotic resection endpoints genome-wide at high spatial resolution in Saccharomyces cerevisiae. Generating full-length resection tracts requires Exo1 exonuclease activity and the DNA-damage responsive kinase Tel1, but not the helicase Sgs1. Tel1 is also required for efficient and timely initiation of resection. We find that distributions of resection endpoints at individual genomic loci display pronounced heterogeneity that reflects a tendency for nucleosomes to block Exo1 in vivo, yet modeling experiments indicate that Exo1 digests chromatin with high apparent processivity and at rates approaching those for naked DNA in vitro. This paradox points to nucleosome destabilization or eviction as a determining feature of the meiotic resection landscape.
Main Text
Meiotic recombination plays a pivotal role in sexual reproduction by promoting proper segregation of homologous chromosomes1. Recombination initiates with DNA double-strand breaks (DSBs) formed by the topoisomerase-like transesterase Spo11, which remains covalently attached to DSB 5′ ends (Fig. 1a). Endonucleolytic cleavage dependent on the conserved Mre11– Rad50–Xrs2 (MRX) complex in cooperation with Sae2 generates nicks on the Spo11-bound strands that serve as entry points for the modest 3′→5′ exonuclease activity of Mre11 and the much more robust 5′→3′ exonuclease activity of Exo1 (refs 2,3–5). The net result is the release of Spo11 bound to a short oligonucleotide (oligo) and, more importantly, generation of long 3′ single-stranded DNA (ssDNA) tails. These tails are substrates for the strand-exchange proteins Dmc1 and/or Rad51, which carry out homology search and invasion of a repair template in the homologous chromosome1. A similar nick-plus-exonuclease mechanism operates in vegetative cells6, except that Exo1 is partially redundant with the concerted action of the Sgs1–Top3–Rmi1 complex along with the Dna2 nuclease for extensive 5′→3′ resection7,8. In meiosis, available data clearly implicate Exo1 but suggest that Sgs1 is dispensable4,9.
Although meiotic resection was first demonstrated experimentally more than 25 years ago and has long been known to be a fundamental step in recombination10–12, it remains poorly understood. In fact, high resolution, quantitative data are only available for one side of a single artificial DSB hotspot4 and whether this information can be extrapolated to natural DSB hotspots genome-wide is unknown. To overcome this limitation, we developed methods to map resection endpoints genome-wide specifically and with high sensitivity and spatial resolution. These maps allowed us to answer longstanding questions about the genetic control of resection; the handoff from MRX–Sae2 to Exo1; locus-to-locus variation in fine-scale resection patterns; the kinetics of resection; and how resection machinery interacts with barriers imposed by local chromatin structure.
Mapping DSB resection endpoints
We digested the 3′ tails of resected DSBs with ssDNA-specific nucleases to generate sequencing libraries of ssDNA–dsDNA junctions that were then compared to previously derived DSB maps from Spo11-oligo sequencing13 (Fig. 1a,b and Extended Data Table 1). For each strain or time point, data were pooled from two highly reproducible biological replicates (Pearson’s r = 0.95–0.99) (Extended Data Fig. 1a,b).
Reads from resection endpoints should be meiosis-specific and Spo11-dependent and should flank DSB hotspots with defined polarity (Extended Data Fig. 1c). Indeed, S1 sequencing (S1Seq) reads were enriched with the expected polarity around hotspots, and this enrichment was absent from premeiotic (0 h) samples and from meiotic samples from the catalytically inactive spo11-Y135F mutant (Fig. 1b,c and Extended Data Fig. 1d,e,f). S1Seq signal around hotspots correlated well with Spo11-oligo counts (Fig. 1d), and reads spread further from hotspots in a time-dependent manner in dmc1Δ (Fig. 1b and Extended Data Fig. 1e,g,h), consistent with gradual hyper-resection14. We conclude that S1Seq is a sensitive and quantitative measure of DSB resection endpoints.
Resection endpoints were located from ~200 to ~2000 nt away from hotspot centers, with a mean of 822 nt and a positive skew (Fig. 1c,e). This pattern resembles Southern blot analysis of the HIS4LEU2 hotspot4 (Fig. 1e, range 350–1550, mean 800 nt), but greater sensitivity of S1Seq captured less abundant species at extremes of the distribution. Genome-average profiles were smooth and left–right symmetric (Fig. 1c), but individual hotspots were heterogeneous with strong peaks and valleys that differed between hotspots and between sides of the same hotspot (Fig. 1b and Extended Data Fig. 1e). This heterogeneity reflects previously undescribed effects of chromatin structure on resection termination (see below).
S1Seq profiles affirmed dispensability of Sgs1 (Fig. 1b,f and Extended Data Fig. 1e). In contrast, the nuclease-defective mutation exo1-D173A (exo1-nd)15 caused S1seq reads to cluster closer to DSB hotspots, reducing resection lengths to <1100 nt (mean = 373 nt), comparable to HIS4LEU2 in exo1Δ (mean = 270 nt)4 (Fig. 1b,e,f and Extended Data Fig. 1e). We consider it likely that the distribution of resection endpoints in exo1-nd is also the distribution of the most distal Mre11-dependent nicks that form in wild type. If so, the difference between Spo11-oligo lengths and exo1-nd endpoints indicates that there are multiple Mre11-dependent nicks on the same strand, or that there is a single distant nick plus extensive 3′→5′ digestion averaging 335 nt (Extended Data Fig. 1i).
Tel1 promotes initiation and extension of resection
Absence of Tel1 (a DSB-responsive kinase orthologous to human ATM) decreased resection length at HIS4LEU2 for early DSBs16. This and other findings led to the proposal that Tel1 controls resection when DSB numbers are still low, whereas higher DSB numbers later allow Mec1 (ATR in humans) to substitute16. S1Seq data refined this model and revealed that Tel1 acts at multiple steps in resection.
S1Seq reads fell closer to DSB hotspots in the tel1Δ mutant at both early (2 h) and later (4 h) time points, indicating shorter resection, but with peaks in similar positions as in wild type (Fig. 2a,b). Although a subset of tracts at 4 h in tel1Δ matched the longest in wild type (Fig. 2b), DSBs remained hypo-resected in the mutant even at this later time (Fig. 2a,b), so Tel1 influences resection length throughout meiosis.
DSBs in wild type have appeared to be maximally resected as soon as they are detectable14,16. S1Seq profiles at individual hotspots seemed stable over time in wild type (Fig. 2a), especially compared to dmc1Δ (Fig. 1b), but genome-wide averages revealed slightly shorter resection tracts at 2 h than at 4–6 h (Fig. 2b,c,d and Extended Data Fig. 1j). Locus-to-locus variability and the greater sensitivity of S1Seq compared to Southern blotting are nonexclusive possibilities for why this small change was not seen before. Although shorter tracts may reflect partially resected DSBs, we favor the interpretation that later-forming DSBs tend to be resected further, perhaps via increasing Tel1 activity16.
Unexpectedly, tel1Δ cells had higher S1Seq signal within hotspots, accounting for a greater fraction of reads at 2 h than 4 h in tel1Δ, but also present in wild type at lower levels (Fig. 2a,b and Extended Data Fig. 1e). We hypothesized that this signal might reflect presence of unresected DSBs. Within-hotspot S1Seq signal displayed peaks overlapping strong Spo11-oligo clusters (Fig. 2a) and it correlated with hotspot strength in both tel1Δ and wild type (Extended Data Fig. 2a,b). Fine-scale patterns matched expectation for preferred Spo11 cleavage 3′ of C residues and for the 2-nt 5′ overhang of Spo11 primary cleavage products (Extended Data Fig. 2c,d). We conclude that unresected DSBs are present in wild type at levels too low to detect readily by Southern blotting; higher levels in tel1Δ indicate that Tel1 is required for normal resection initiation, possibly via Sae2 phosphorylation17. Consistent with this interpretation, a subset of the DSBs in tel1Δ mutants detected by Southern blotting are unresected and have Spo11 still covalently attached (M. Neale, personal communication).
Recombination intermediates
Along with resection signal, there was weaker S1Seq signal with the “wrong” polarity, e.g., top-strand reads mapping to the left of hotspots (Fig. 1b, and Extended Data Fig. 1d,e). This signal was not resection from neighboring hotspots, it correlated with hotspot strength, it was meiosis-specific and Spo11-dependent, and it was essentially absent in dmc1Δ (Fig. 1b,c, and 3a; Extended Data Fig. 1e,f,g,h, and 3a). We therefore conclude that this signal derives from S1-sensitive recombination intermediates (RIs), probably displacement (D) loops from strand exchange (Extended Data Fig. 3b,c). Throughout this study, resection profiles were corrected by subtracting an estimate for the small amount of RI signal with the “correct” polarity that presumably masquerades as resection signal (Extended Data Fig. 3b,c; see Methods).
We explored spatial and temporal patterns and genetic control of these RIs. At 4 h in wild type, RI reads had a positively skewed distribution with an estimated mean of 606 bp from hotspot centers, i.e., ~25% shorter than the mean resection length (Fig. 3b). The sgs1 mutant was indistinguishable from wild type (Extended Data Fig. 3d), but RI distances were shorter in tel1Δ consonant with the modestly shorter resection tracts (Fig. 3b), suggesting that resection length influences RI position. In exo1-nd the RI signal was even closer to hotspot centers (Fig. 1f, 3b, and Extended Data Fig. 3e). The similarity of RI distances (mean of 350 bp) to resection lengths in exo1-nd suggests that much of the resection tract is “used up” making these RIs, whereas RIs in wild type usually form within a DSB-proximal subregion of resection tracts.
We reasoned that quantitative differences between Spo11 oligos and S1Seq can reveal regional differences for lifespans of steps in recombination, because turnover of Spo11-oligo complexes is tied to meiotic prophase exit18 whereas lifespans of resection and RI signals from S1Seq reflect more directly the progress of the recombination reaction. Smaller chromosomes tend to incur more DSBs per kb because of negative feedback regulation of DSBs tied to engagement of homologous chromosomes13,18. It was proposed that smaller chromosomes take longer on average to engage homologous partners because multiple DSBs per chromosome are usually needed for successful engagement18,19. This hypothesis predicts that resected DSBs should tend to persist longer on smaller chromosomes. Consistent with this prediction, the ratio of S1Seq resection signal to Spo11-oligo counts at 4 h correlated negatively with chromosome size, i.e., the yield of resected DSBs (S1Seq) per DSB formed (Spo11 oligos) was higher on smaller chromosomes (Fig. 3c). This anticorrelation was not seen at 2 h (Fig. 3c), as expected because early DSBs have not had time to progress further in recombination regardless of chromosome size. The ratio of RI signal to Spo11-oligo counts was uncorrelated with chromosome size (Extended Data Fig. 3f), suggesting that RI lifespan is a function of local features rather than large-scale chromosome pairing kinetics.
The ratio of S1Seq resection reads to Spo11 oligos was lower for pericentromeric hotspots than for other sub-chromosomal domains (Fig. 3d). This apparently shorter lifespan could reflect delayed DSB formation and/or more rapid DSB repair. Consistent with either possibility, kinetochore components suppress DSB formation and promote sister chromatid recombination in pericentric regions20.
Modeling Exo1 mechanism and speed
We tested a model in which Exo1 enters DNA at an Mre11-generated nick and digests DNA until it dissociates from its substrate, which predicts that Exo1 run lengths should follow a geometric distribution defined by the average probability of dissociation at each nucleotide step. We therefore determined the geometric distribution that provided the best fit to wild-type resection lengths when combined with the distribution of presumptive Exo1 entry points (i.e., the exo1-nd endpoint distribution) (Fig. 4a). The geometric model fit the data poorly and could be ruled out, but an excellent fit arose when the Exo1 geometric distribution was shifted (Fig. 4a). This simple alternative can be conceptualized in terms of a high probability of Exo1 resecting for a minimum distance (best-fit estimate of 220 nt), after which resection termination follows a geometrically distributed process. This model, plus findings below, provides a framework for understanding the resection mechanism (Conclusions).
How fast is Exo1? In vegetative cells, unrepairable DSBs are continuously resected for tens of kb at ~4.4 kb/h (refs 8,21,22). Long-range resection by Exo1 only (i.e., in an sgs1 mutant) is even slower, at ~1 kb/h (ref 8). To test whether the vegetative rate can account for meiotic resection, we performed Monte Carlo simulations to generate populations of resected DNA molecules to compare with S1Seq patterns (Fig. 4b,c, Extended Data Fig. 4 and Supplemental Movies 1 and 2. Rates of 4 kb/h or even 8 kb/h were not fast enough: these speeds predicted that DSBs should be resected less far than was observed, particularly at early time points. In contrast, simulations matched observed tracts well when we used rates from 16 to 40 kb/h, the latter being the value from single-molecule studies for Exo1 resecting naked DNA23. Since rates ≤8 kb/h appear implausibly slow, we conclude that meiotic resection is much faster than long-range resection in vegetative cells. Because 16 kb/h or higher is plausible, it suggests that Exo1 processes meiotic DSBs in vivo nearly as quickly as it can degrade naked DNA in vitro. Hyperresection in dmc1Δ was much slower, with an estimated average speed of 0.19 kb/h (Fig. 4d). Together, these findings clarify the striking differences between the extreme rapidity but limited length of wild-type meiotic resection, the sluggish continuity of hyper-resection in the absence of Dmc1, and the moderate pace but largely unconstrained distance of long-range resection in vegetative cells.
Chromatin shapes the resection landscape
The strong peaks we observed in S1Seq resection profiles were unexpected. Verifying that this heterogeneity was not a sequencing artifact, S1 treatment converted resected DSB fragments at the GAT1 hotspot from the usual featureless smears into discrete banding patterns on Southern blots (Fig. 5a and Extended Data Fig. 5a). Prominent bands occupied similar positions in mutants with very different average resection lengths, so this banding pattern is a reproducible feature of the GAT1 locus itself. S1 generated no discrete banding at HIS4LEU2 (Extended Data Fig. 5b), consistent with a lack of preferred resection endpoints4.
The stereotyped positions of nucleosomes around natural hotspots allowed us to ask if chromatin structure contributes to this resection heterogeneity. Most yeast promoters have a nucleosome depleted region (NDR), where many DSBs form, flanked by positioned nucleosomes with the transcription start site (TSS) in the first (+1) nucleosome13,24 (Fig. 5b). When S1Seq data from wild type were averaged around +3 nucleosomes, the broad peak showed modest scalloping in register with average nucleosome occupancy (Fig. 5c). We speculated that locus-to-locus variation in nucleosome positions blurs the chromatin signature, so we compiled S1Seq averages for three groups of genes divided by linker width between +3 and +4 nucleosomes (Fig. 5d). An S1Seq peak overlapping the left edges of +4 nucleosomes moved progressively to the right with increasing linker width (arrows in Fig. 5d). The widest group also accumulated S1Seq reads within the linkers. Focusing on +2 and +3 nucleosomes revealed similar patterns for the shorter resection tracts in tel1Δ (Extended Data Fig. 5c,d), and +1 and +2 nucleosomes yielded similar patterns for Mre11-dependent clipping in exo1-nd (Fig. 5e, f). These findings establish a spatial correlation between nucleosome positions and preferred endpoints for both the Mre11 and Exo1 resection steps. For Exo1, the patterns are consistent with digestion proceeding for a short distance into a nucleosome, with increasing likelihood of resection termination as Exo1 approaches the nucleosome dyad possibly reflecting the greater stability of histone-DNA binding near the dyad25. RIs displayed a tendency for local peaks at nucleosome linkers (Extended Data Fig. 5e), indicating that RI positions are also influenced by chromatin structure, possibly that of the homologous recombination partner.
To determine if the resection correlation reflects causality, we examined mutants lacking transcription factors Bas1 or Ino4, in which specific hotspots experience altered chromatin structure in adjacent transcription units while retaining promoter-associated DSBs26. In ino4Δ the nucleosome array in CHO2 shifts toward the promoter as revealed by micrococcal nuclease (MNase) digestion of chromatin (Fig. 5g). In bas1Δ the normally variable nucleosome positioning in SHM2 (manifesting as a broad, shallow nucleosome ladder in wild type) becomes more regular (Fig. 5h). Concordantly in both cases, S1Seq reads shifted closer to the promoters, whereas neither chromatin structure nor resection in the adjacent PTI1 or REX2 genes were much affected (Fig. 5g,h). A control locus where chromatin structure was unchanged in bas1Δ showed unaltered resection profiles as well (Extended Data Fig. 5f). Parenthetically, these findings demonstrate that S1Seq heterogeneity cannot be a consequence solely of biases from DNA sequence. More importantly, we can conclude that chromatin structure directly affects Exo1 stopping points, and probably shapes Mre11-dependent clipping as well. To our knowledge, this is the first direct evidence that nucleosome positions determine resection termination positions in vivo.
Conclusions
Given the typical chromatin structure around yeast DSB hotspots, Mre11-dependent incision often occurs within +1 or +2 nucleosomes and Exo1 then traverses several nucleosomes’ worth of DNA (Fig. 5b). This creates an apparent paradox: the feeble activity of Exo1 on nucleosomal substrates in vitro27 and the pronounced tendency for resection to stop near nucleosome boundaries in vivo indicate that nucleosomes are a strong block to Exo1. Yet Exo1 resects for several hundred nt extremely quickly and with high apparent processivity in vivo, as if nucleosomes are initially little or no barrier at all.
A solution to this paradox is for nucleosomes to be destabilized or removed from DNA prior to digestion by Exo1, with a major constraint on resection length simply being how many nucleosomes are removed from Exo1’s path (or are not present to start with) (Extended Data Fig. 5g). Additional factors contributing to resection control may include inhibitors of Exo1, spatially regulated Exo1 activity, and/or iterative Exo1 loading (Extended Data Fig. 5h). Nucleosome eviction at DSBs has been proposed for vegetative yeast and somatic mammalian cells, but experimental support is indirect and conflicting and has generally been unable to distinguish whether observed chromatin changes are prerequisite or consequence of resection28–30 (see Supplementary Discussion). Our parameterization of resection speed, apparent processivity, and preferred endpoint positions provides a strong new line of support for this model. Chromatin remodeling enzymes implicated in vegetative resection31 are good candidates for nucleosome destabilization in meiotic resection. Interestingly, DSB resection in mouse meiosis extends a similar average distance as in yeast and is inferred to pass through multiple nucleosomes (J. Lange, S.Y., M. Jasin, and S.K., submitted). Conservation of the scale over which DSB-provoked chromatin remodeling occurs could explain this similarity between species.
If nucleosomes are indeed evicted, we can infer that this occurs after DSB formation because Spo11 rarely cuts within nucleosome positions13. Furthermore, the strong chromatin signature in the exo1-nd mutant implies that the full extent of eviction must occur after positions of Mre11-dependent incision are established. We also note that detection of apparently unresected DSBs in wild type may indicate that Mre11 incision is a rate-limiting step, as in Schizosaccharomyces pombe32. Tel1 emerges as a prime candidate for promoting Mre11-depended incision given the greater number of unresected DSBs in its absence. Tel1 also regulates net resection distance without changing the positions of preferred stopping points. We speculate that Tel1 controls the efficiency or distance over which nucleosome destabilization occurs (but does not affect translational positions of nucleosomes), which could be via effects on chromatin remodelers, histone modifications, or both.
This study captures the first comprehensive yet finely detailed portrait of the meiotic resection landscape and uncovers components that govern its shape and dynamics. S1Seq gives a selective, sensitive and quantitative measure of DSB resection tracts and should be readily applicable to map resection or DSBs in other settings and organisms.
Methods Summary
Yeast strains are of the SK1 background (Extended Data Table 2). Synchronized meiotic cultures were prepared according to standard methods. Resection overhang removal took place in agarose plugs with a sequential digestion with S1 nuclease and T4 DNA polymerase, followed by ligation of the first adaptor in plugs. After plug extraction, size selection and shearing, the biotin-containing DNA fragments were affinity purified with streptavidin beads and a second adaptor was ligated. The library was then amplified with low-cycle PCR using primers specific for the adaptors, and submitted for deep sequencing.
Full Methods and any associated references are available in Supplementary Information.
Author contributions
EPM developed S1Seq and performed experiments. SY developed in silico modeling of resection length and speed. EPM and SK designed the study, analyzed data and wrote the paper with contributions from SY.
Methods
Yeast Strains and culture methods
All strains used in this study were SK1 background47 and the genotypes are listed in Extended Data Table 2. The sgsl-ΔC795 allele was provided by N. Hunter and the exo1-Dl73A allele by R. Liskay15. The dmcl and tell deletions were made by replacing the coding sequence with the hygromycin B phosphotransferase gene (hphMX4). Gene disruption was verified by Southern blot. The basl and ino4 deletion alleles were made as described26. Diploid cells were streaked from a frozen stock onto YPD plates and allowed to grow at 30°C for two days. A single colony was inoculated into 20 ml liquid YPD medium and grown for >24 hr at 30°C. The saturated YPD culture was used to inoculate the appropriate volume of YPA medium (1% yeast extract, 2% Bacto peptone, 2% potassium acetate, 0.001% antifoam 204 (Sigma)) to OD600 0.2 and grown in 2.8 l baffled Fernbach flasks at 250 rpm at 30°C for 14 hr. Cells at indicated time points (75 ml for 0 and 2 h, 66 ml for 4 h, 60 ml for 6 h and 54 ml for 8 h) were harvested, washed with 50 mM EDTA pH 8.0 and stored at -80°C until plug preparation.
Preparation of DNA and sequencing libraries for S1Seq
Genomic DNA was extracted in low-melting point (LMP) agarose to yield intact high molecular weight DNA. Each cell pellet was resuspended in 150 μl 50 mM EDTA pH 8.0 and allowed to thaw at 40°C. 498 μl LMP agarose (2% in 50 mM EDTA pH 8.0) was mixed with 102 μl Solution 1 (SCE (1 M sorbitol, 0.1 M sodium citrate, 60 mM EDTA pH 7.0) plus 5% β-mercaptoethanol and 1 mg/ml zymolyase 100T) and the mix was prewarmed at 40°C before adding to the cells. The resuspended cells were pipetted into plug molds and chilled at 4°C for 30 min. Once solidified, the agarose plugs (three at a time) were expressed into 3 ml of Solution 2 (0.45 M EDTA pH 8.0, 0.01 M Tris-HCl pH 7.5, 7.5 % β-mercaptoethanol, 10 μg/ml RNase A) and incubated at 37°C for 1 h. Solution 2 was then replaced with Solution 3 (0.25 M EDTA pH 8.0, 0.01 M Tris-HCl pH 7.5, 1% SDS, 1 mg/ml proteinase K) and plugs were incubated at 50°C overnight. Plugs were finally washed three times with 3 ml 50 mM EDTA pH 8 and stored in plug storage solution (0.05 M EDTA pH 8.0, 50% glycerol) at -20°C until all plugs were ready to be processed for library preparation.
For efficient in-plug overhang removal and adaptor ligation, a sequential treatment with S1 nuclease and T4 DNA polymerase was required before ligation. Even though S1 nuclease treatment was sufficient to remove the overhang tails, the subsequent ligation step was not efficient, unless a clean-up reaction with T4 DNA polymerase was performed (E.P.M., unpublished observations). Ten plugs of each sample to be analyzed were equilibrated in 500 μl 1 × S1 buffer (50 mM sodium acetate pH 4.5, 0.28 M NaCl, 4.5 mM ZnSO4) per plug in 2 ml conical tubes for 30 min and this was repeated three times. Fresh 1 × S1 buffer (500 ul) containing 9 U of S1 nuclease (Promega) was added to each tube and the plugs were incubated on ice for 15 min to allow the enzyme diffuse into the plugs, followed by 20 min incubation at 37°C. S1 nuclease was inactivated by addition of EDTA pH 8.0 to a final concentration 10 mM and incubation on ice for 15 min. Plugs were rinsed with 1 × TE and further equilibrated in 500 μl T4 polymerase buffer (1 × T4 DNA ligase buffer (NEB), 1 × BSA, 100 μM dNTPs), four times for 30 min each. Fresh T4 polymerase buffer (500 μl) containing 30 U of T4 DNA polymerase (NEB) was added to each tube and the plugs were incubated on ice for 15 min, followed by 30 min incubation at 12°C. T4 polymerase was inactivated by addition of EDTA pH 8.0 to a final concentration of 10 mM and incubation on ice for 15 min. At this point all buffer was aspirated, the plugs were rinsed with 1 × TE and incubated at 75°C for 20 min to fully inactivate T4 polymerase. Plugs were allowed to gradually cool and solidify, before equilibrating in 250 μl 1× T4 ligase buffer (NEB), four times for 15 min each. After the last equilibration step, 200 μl of the buffer were removed leaving each plug immersed in 50 μl 1 × T4 DNA ligase buffer, to which 1 μl of 50 μM adaptor and 1 μl of 2000 U/μl T4 DNA ligase were added. Ligation took place at 4°C for ≥18 h. The adaptor was prepared the day of the experiment by mixing oligos P5-top (5BiosG/ACACTCTTTCCCTACACGACGCTCTTCCGATCT, biotin attached to the 5′-end of the oligo) and P5-bottom (5Phos/AGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT/3InvdT, 5′-end phosphorylated with inverted dT incorporated at the 3′-end to block ligation) at equimolar concentration, boiling for 5 min and cooling down at room temperature for at least 1 h.
To retrieve the DNA from the agarose plugs, the Epicentre GELase Enzyme Digestion protocol was used, followed by phenol extraction and ethanol precipitation. During this process plugs of DNA from a single culture and time point were pooled. The purified DNA was then subjected to size selection by electrophoresis through a 1% LMP 1× TAE agarose gel at 80 V/cm for 2 h to separate the high-molecular-weight genomic DNA from the excess unligated adaptor. The genomic DNA was excised and extracted from the gel using the Epicentre GELase enzyme protocol, phenol extraction and ethanol precipitation; then the DNA was dissolved in 100 μl 1 × TE. To ensure complete removal of unligated adaptor, each sample was further purified by passing three times through Chroma Spin-1000 (Clontech) columns. The eluates were then subjected to shearing according to the Covaris instrument protocol to DNA fragment sizes ranging between 200-500 bp.
Following shearing, fragments containing the biotinylated adaptor were enriched by affinity purification with streptavidin. For each sample, 50 μl of streptavidin M-280 beads (Roche) were used, prewashed twice with 1 × TE and twice with 1 × B&W (binding and washing buffer, 10 mM Tris-HCl pH 7.5, 1 mM EDTA, 2 M NaCl), according to the manufacturer’s instructions. Binding was carried out at 20°C for 30 min, followed by two washes with 500μl 1× B&W buffer and two washes with 500 μl 10 mM Tris-HCl pH 7.5. The remaining steps were performed with the DNA fragments still bound to the streptavidin beads. Because sheared ends are not always readily ligatable, an end-repair step was included according to Epicentre’s End-it DNA Repair kit protocol. The beads for each sample were incubated in 100 μl of reaction mix according to manufacturer’s instructions and incubated at 20°C for 45 min. Following washes with 500 μl 1× B&W buffer (twice) and 500 μl 10 mM Tris-HCl pH 7.5 (twice), the beads were incubated in 100 μl of 1 × T4 DNA ligase buffer containing 400 U T4 ligase and 1 μM P7 adaptor. The adaptor was prepared by annealing oligos P7 top (5Phos/GATCGGAAGAGCACACGTCTGAACTCCAGTCAC/3InvdT) and P7 bottom (GTGACTGGAGTTCAGACGTGTGCTCTTCCGATC) at equimolar concentration, boiling for 5 min and allowing to cool down at room temperature for at least one hour. Ligation of the P7 adaptor was allowed at 20°C for 4 h and was followed by washes with 500 μl 1 × TE and 500 μl 10 mM Tris-HCl pH 7.5 (three times each). Beads of each sample were resuspended in 20 μl 10 mM Tris-HCl pH 7.5.
Each sample of beads was split into two PCR reactions in a total of 50 μl containing 10 μl of the resuspended beads, 1 × Phusion HF buffer, 0.2 mM dNTP, 1 U Phusion HF polymerase (Thermo Scientific), 0.4 μM P5 universal primer (AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATC T) and 0.4 μM P7 indexed primer (CAAGCAGAAGACGGCATACGAGATCGTGA TGTGACTGGAGTTCAGACGTGTG, shown here with Illumina TruSeq index 1 underlined). PCR was initiated by a denaturation step at 98°C for 30 sec, followed by 16 cycles of amplification (98°C for 10 sec, 65°C for 20 sec, and 72°C for 20 sec). PCR products were pooled, and the DNA was precipitated by adding ammonium acetate to 2.5 M and 2.5 volumes of ice-cold 100% ethanol. Following an overnight incubation at −20°C, the samples were spun at 16000 g for 30 min, washed with ice-cold 70% ethanol, air-dried and dissolved in 20 μl 10 mM Tris, pH 7.5. PCRs from each sample were pooled and separated on a 5% non-denaturing polyacrylamide gel in 1 × TBE. The gel piece between 200–600 bp was excised, crushed, and eluted in 300 μl 10 mM Tris pH 8.0 at 37°C overnight. The elution mixture was spun through a SPIN-X column, and DNA was precipitated with ammonium acetate and ethanol as above. The DNA pellet was dissolved in 20 μl 10 mM Tris pH 7.5 and sequenced on the Illumina HiSeq platform (50 bp single-end) in the Integrated Genomics Operation at Memorial Sloan Kettering Cancer Center.
While we were optimizing this protocol, a similar method designed to map DSBs at high spatial resolution, BLESS, was described48. Another method, called EndSeq, has also been recently described49. A key difference between S1Seq and both BLESS and EndSeq is our use of S1 nuclease, which we found necessary to efficiently remove long stretches of ssDNA. In addition, BLESS uses a formaldehyde fixation step which we do not use, and which we speculate might cause unwanted damage to the DNA samples.
Physical analysis of resected and blunt-ended DSBs
For direct detection of DSB fragment migration before and after ssDNA removal at GAT1, the genomic DNA isolated in plugs was subjected to restriction endonuclease digestion, gel electrophoresis and Southern blot analysis. The plugs were treated with S1 endonuclease as described above but instead of T4 DNA polymerase and T4 DNA ligase treatment they were digested with PstI-HF (NEB) in plugs equilibrated in the appropriate restriction enzyme buffer as described13. Digested DNA was separated on 0.8% agarose gels in 0.5× TBE with electrophoresis at 80 V/cm for 16 h and re-circularization of the buffer, then detected by Southern blot hybridization using a 32P-labeled DNA fragment adjacent to one of the PstI sites as a probe. The primer sequences for amplification of the probe were 5′-CGCGCTTCACATAATGCTTCTGG and 5′-TTCAGATTCAACCAATCCAGGCTC.
Bioinformatics analysis
Mapping of the reads onto the S288c reference genome (SacCer2) was performed using the SHRiMP mapper50 (gmapper-ls) with arguments:
-E -U -n 1 -Q --sam-unaligned --strata -o 10001 -N 20
Before mapping, adaptor sequences were removed using fastx_clipper (http://hannonlab.cshl.edu/fastx_toolkit/) and a custom script. Code used for read processing and mapping is available online at https://github.com/soccin/S1Seq. After mapping, the reads were separated into unique and multiple-mapping sets, but only uniquely mapping reads were analyzed in this study.
Analyses were performed using the R package (RStudio version 0.99.879, R version 3.1.2). Before any analysis, maps were curated by masking (set to NA) 500 bp at the ends of each chromosome, because telomeres represent naturally occurring resected DNA ends and therefore give a high frequency of reads. We further masked regions that gave a high frequency of meiotic DSB-independent reads, defined as positions with more than 10 reads in the high-depth biological replicates and/or more than 5 reads in the lower depth biological replicates from the 0-h maps from wild type, spo11-Yl35F, dmc1Δ, and exo1-nd plus the 4-h map from spo11-Yl35F. A table of mask coordinates is provided at https://github.com/soccin/S1Seq. Moreover, reads mapping to mitochondrial DNA or the 2 μ plasmid were excluded. Each map was then normalized to reads per million remaining mapped reads, then biological replicates were averaged. The Gene Expression Omnibus (GEO) (http://www.ncbi.nlm.nih.gov/geo) accession number for the raw and processed sequence reads is pending.
All analyses were performed using a recent hotspot list compiled from a combination of multiple independent wild-type Spo11-oligo maps51. For the S1Seq resection and recombination lifespan analyses, the left arm of chromosome III was censored to avoid inconsistencies because some strains had the HIS4LEU2 and leu2:hisG artificial hotspots on this chromosome arm. Published maps of nucleosome occupancy13 and histone H3 midpoints52 were used. We defined subtelomeric hotspots (n=60) as those residing in the 20-kb telomere-proximal region; pericentromeric hotspots (n=82) are those residing in the 20-kb zone surrounding centromeres. For all modeling analyses, resection length histograms, resection length calculations, resection endpoint distributions centered on nucleosome midpoints, and analyses of RIs, we used the subset of hotspots for which no other hotspot was located within 3 kb, and for which hotspot width was less than 400 bp (n=405). S1Seq reads of polarity opposite to expectation for resection endpoints were subtracted to correct for RI signal. Adjustment for RIs had only a small effect on calculated values. For example, wild-type mean resection length was 779 nt without RI subtraction and 822 nt with subtraction, and the shift required for best fit in the shifted-geometric modeling experiment was 260 nt without RI subtraction versus 220 nt with subtraction.
Modeling Exo1 resection run lengths in wild type
We built a model based on the assumption that Exo1 starts resecting at Mre11-dependent nicks or gaps, whose distribution is measured by the resection maps from the exo1-nd mutant. We also assumed that a single Exo1 molecule resects DNA from 5′ to 3′ until it dissociates from its substrate without chance to rebind. If there is an equal probability of dissociation at each nucleotide step, Exo1 resection tract lengths will follow a geometric distribution determined by that stepwise dissociation probability. Note that application of this model to S1Seq data yields an estimate of the average apparent probability of resection termination at each step. We do not exclude the possibility of there being considerable microscopic variation in this probability at different locations in the genome or between individual resection events in the population.
After subtraction of RI signal from resection endpoint signal (see above), resection profiles for loner hotspots were averaged and binned (10 bp) to make the analysis computationally tractable. Moreover, the signal close to hotspot midpoints was excluded from the analysis by setting values of positions <200 bp for wild type and <20 bp for exo1-nd to zero. Finally, an estimated background was removed by subtracting from all values the value of signal 2 kb away from hotspot midpoints.
Let Mbe the observed exo1-nd mutant resection endpoint distribution in 10-bp bins: M = {m0, m1,…, m200}. Similarly, let W be the binned wild-type resection endpoint distribution: W = {w0, w1,…, w200}. Thus, mi and wi are empirical estimates of how many resection endpoints are located between (10i – 5) bp to (10i + 4) bp away from hotspot midpoints. We only considered signals located within 2 kb of hotspot midpoints (i.e., max i = 200). Let R be the distribution model for the Exo1 resection run lengths, R = {r0, rl,…, r200}, where rk is the probability that Exo1 stops resecting 10k nucleotides away from the Exo1 entry sites. We empirically chose 2 kb (i.e., max k = 200) as the resection length limit based on the average wild-type resection signal around hotspots.
Let θ be the unknown parameter of a geometric distribution for R and let L(θ) = {l0, l2,…, l200} be the likelihood function of θ. We wished to estimate θ from the combination of the exo1-nd mutant and wild-type data. The wild-type resection endpoint frequency at a given distance from hotspot midpoints is the sum of the resection tracts that start anywhere closer to hotspot midpoints and that end at that given distance. Thus L can be described as a series of linear equations combining M, W, and R:
We optimized θ to maximize the log likelihood using the optimize function of R (θ = 0.00200). As an alternative, we considered a shifted geometric distribution R′ for Exo1 run lengths, which can be described as: where s is the number of 10-nucleotide bins by which to shift the geometric distribution. We tested shifts of 0–30 bins (i.e., 0–300 nucleotides) and optimized θ and s to maximize the log likelihood (θ = 0.00440, s = 22). The Exo1 run length distributions predicted by the geometric and shifted geometric models with these best-fit parameters are presented in the middle panel of Fig. 4a.
Simulating Exo1 resection speed
To evaluate possible rates for resection by Exo1, we performed Monte Carlo simulations to generate populations of resected DNA molecules as a function of time in meiosis, and compared them to observed resection distributions. Details are provided below, but an overview is as follows. In the simulations, a total of 1 million DSBs were allowed to form over the course of meiosis with timing similar to published time course data33. This procedure accommodates the fact that DSBs form in a semi-synchronous manner over several hours in meiosis. These DSBs were then assumed to be clipped instantaneously by MRX plus Sae2, with positions sampled from the empirical distribution of resection endpoints in the exo1-nd mutant. The assumption of instantaneous clipping biases the model in favor of slower Exo1 speeds, i.e., by having the clipping step take no time, Exo1 thereby is allowed more time to resect to completion. We further sampled to establish Exo1 run lengths using the best-fit shifted-geometric model. Given these three sampled values for each DSB (time of formation, position of Exo1 entry point, and distance traveled by Exo1), the spatial distribution of resection endpoints can be calculated arithmetically at any given time in meiosis as a function of resection speed. By comparing the simulated resection profiles to observed resection profiles, we can evaluate the plausibility of a given Exo1 speed, i.e., whether that speed is fast enough to recapitulate empirical resection lengths.
To establish a model for DSB timing in meiosis, we started with published analysis of DSB formation at the HIS4LEU2 hotspot (Time Course 36)33, in which DSBs formed from 1.5 h to 8 h after transfer to sporulation medium. From these data, a normal distribution fit well to the time profile of DSB formation, but we considered that a model for global DSB timing would be likely to display a positively skewed DSB time distribution given temporal differences between different parts of the genome53 and given cell-to-cell variability in meiotic prophase length. Therefore, we used a shifted log normal distribution as a DSB time distribution (Extended Data Fig. 4a). However, we reached the same overall conclusions about plausibility of resection speeds if we used a normal distribution for the DSB timing profile instead (data not shown).
To define the distribution of Exo1 entry points, we used the exo1-nd resection endpoints from 2 h, with the rationale that a significant number of DSBs have already formed at this point, but there has not been sufficient time for DSBs to disappear via recombination. To define Exo1 endpoints, we used the shifted geometric run-length model fitted as described above, except using the data at 2 h in meiosis (θ = 0.00382, s = 16). Note that, because total resection length in wild type is slightly shorter at 2 h than at 4 h (Fig. 2d), choosing to use input data from the 2 h time points also biases our simulations in favor of slower Exo1 speeds.
To carry out the simulations, we generated a random sample of 1 million DSB formation times (in min after meiosis induction) using the rlnorm function in R with log mean = 4.5, log standard deviation = 0.5 and shift = 1.5 h, taking into account values of time points ≥1.5 h and ≤8 h. A random sample of 1 million Exo1 start points was generated by sampling with replacement from 10-bp bins between 0 and 2 kb from hotspot midpoints with sampling weights matching the empirical distribution of resection endpoints from the 2-h exo1-nd map (Extended Data Fig. 4b). A random sample of 1 million Exo1 run lengths (also in 10-bp bins) was generated from the shifted geometric model with parameters listed above.
From these randomly generated values, the distribution of resection endpoints can be calculated directly at any time in meiosis (t) for any given Exo1 speed (v). If τi, μi and εi are the sampled values for formation time, Mre11-dependent clipping position, and Exo1 run length, respectively, for the ith simulated DSB, then the resection endpoint distance di as a function of time in meiosis can be described as:
Results of these simulations are shown in Supplemental Movies 1 and 2 for resection speeds of 4 kb/h and 40 kb/h, respectively. A copy of the R code for carrying out the simulations is provided at https://github.com/soccin/S1Seq.
Acknowledgments
We thank A. Viale and the MSKCC Integrated Genomics Operation for sequencing and N. Socci and the MSKCC Bioinformatics Core for assistance mapping S1Seq reads. Support for core facilities was provided by the NIH/NCI Cancer Center Support Grant P30 CA008748. We thank M. Neale for sharing unpublished information and N. Hunter and R. Liskay for strains. This work was supported by NIH grants R01 GM058673 and R35 GM118092 (to S.K.). E.P.M. was supported in part by a Fellowship from the Helen Hay Whitney Foundation and S.Y. was supported in part by a Kuro Murase MD-JMSA Scholarship.