Hotspot-centric analysis of DSBs

Most S. pombe DSB hotspots are in large intergenic regions (IGRs), and most protein-coding gene promoters lack associated hotspots (Cromie et al. 2007); this pattern is similar to that in mice (Brick et al. 2012) but unlike that in S. cerevisiae (Pan et al. 2011). As a result, S. pombe has fewer, more widely spaced hotspots than S. cerevisiae. We previously defined 288 DSB hotspots as regions with ChIP-chip hybridization signal at or above the level expected for a DSB frequency of 0.3%, based on linear regression of ChIP-chip signal vs. direct DSB Southern blot assays at 25 hotspots (Cromie et al. 2007; Fowler et al. 2013). This value (0.3%) represents the approximate threshold for detection on the Southern blots (Cromie et al. 2007).

In this study, we defined 603 Rec12-oligo hotspots as sites with significantly higher read densities than surrounding regions (Fig. 2A; Supplemental Material; Supplemental Table S3). Rec12-oligo densities in these hotspots ranged from 173 to 5021 RPM/kb. The density in the weakest hotspot scored was thus 247-fold higher than the density in the rDNA, and 2.2-fold higher than the genome average. This list accounted for 98% of the 288 previously defined hotspots (Fowler et al. 2013), plus 322 additional hotspots uncovered because of the greater sensitivity of Rec12-oligo sequencing. Consistent with previous findings, most Rec12-oligo hotspots overlapped with IGRs (Fig. 2B). Importantly, significant spatial enrichment of ChIP-chip signal for Rec12 and, albeit weakly, for hotspot-determinant proteins Rec25, Rec27, and Mug20 was seen even at the weak Rec12-oligo hotspots that would have been below detection limits by other means (Supplemental Fig. S2A), confirming that these are bona fide sites of preferential DSB formation.

Open in a separate window

Figure 2.

Features of DSB hotspots and comparisons with S. cerevisiae. (A) An example of adjacent strong and weak hotspots. “Strong” hotspots refer to those with an inferred break frequency of ≥0.3% of DNA, i.e., the limit of detection in prior studies. Rec12-oligo frequencies (blue) were compared with Rec12 ChIP-chip signal (red). For comparison, there are eight strong and 13 weak hotspots in Figure 1B (top). Orange arrows indicate open reading frames of protein-coding genes. (B) Rec12 oligos arise primarily from intergenic DNA and hotspots, although about a quarter arise from “cold regions,” i.e., outside the 603 hotspots defined here. NDR and Inter-NDR indicate the fraction of oligos that arise within or outside nucleosome depleted regions, respectively. (C) Rec12-oligo counts in hotspots follow a smooth continuum. (Left) Hotspots were rank-ordered by oligo counts (603 hotspots from S. pombe; 3604 hotspots from S. cerevisiae) (Pan et al. 2011). (Middle) To account for the greater number of meiotic DSBs per cell in S. cerevisiae, the rank-ordered S. pombe hotspots were scaled by a factor of 58/160, the approximate ratio of the number of DSBs per cell in S. pombe and S. cerevisiae. (Right) Plot of the cumulative fraction of hotspot oligos among the ranked hotspots. (D) Hotter hotspots tend to be wider than colder hotspots; S. pombe hotspots tend to be much wider than in S. cerevisiae. (E) Hotspot boundaries correlate poorly if at all with IGR boundaries, positions of protein-coding genes (orange arrows), or promotors (white boxes indicate transcribed regions). Gray bands indicate IGRs. Top bars show the defined hotspot regions, blue from this study and red from microarray hybridizations (Fowler et al. 2013). See Supplemental Figure S2 for further analyses of Rec12-oligo hotspots.

When hotspots were ranked by oligo count, they followed a smooth continuum with no obvious discontinuity to serve as a dividing line between quantitative classes (Fig. 2C, left). In particular, we emphasize that there was no clear distinction between previously discovered hotspots and the additional weak ones discovered here (i.e., those with an inferred DSB frequency of <0.3% of DNA). This pattern reinforces the view that the cutoff is arbitrary between sites that are hotspots and those that are not (Pan et al. 2011). Interestingly, the distribution of breaks among hotspots differed in the two species. For example, hotspots in general were hotter as a fraction of total breaks than those in budding yeast, as judged by comparing the cumulative curves in Figure 2C, left. Looked at a different way, if we scale hotspot counts to reflect the greater number of total DSBs in S. cerevisiae than in S. pombe (~160 vs. ~58, respectively, see below), S. pombe hotspots also tend to be hotter in absolute terms (Fig. 2C, middle). Furthermore, the weakest two-thirds of hotspots accounted for 22% of hotspot-associated oligos in S. cerevisiae, but only 16.4% in S. pombe (Fig. 2C, right); and 20% of all hotspot-associated oligos were contained in the 87 hottest hotspots in S. cerevisiae, but in only the 16 hottest hotspots in S. pombe. Thus, while there are fewer locations that score as hotspots in fission yeast, a larger fraction of these are exceptionally hot and account for a disproportionate share of breakage.

The more precise spatial information afforded by Rec12-oligo mapping allowed us to explore hotspot features in greater detail than previously possible. For example, the width of Rec12-oligo hotspots correlated strongly with heat (i.e., read count) (Fig. 2D). This relationship was not linear, however, reflecting an additional correlation of hotspot width with oligo density (Supplemental Fig. S2B). Thus, the wider break sites are broken disproportionately more frequently. Compared with budding yeast, S. pombe hotspots displayed more highly variable widths (~50 bp to 7 kb; the widest S. cerevisiae hotspot is 2 kb) and were wider on average: The mean width (1.4 kb) was comparable to that of the widest hotspots in budding yeast (Fig. 2D).

Many wide hotspots contained clusters of closely spaced break sites, as exemplified by mbs1 (Fig. 1D). This pattern reinforces the points that the genome is not simply divided into alternating domains that favor or disfavor break formation, and that hotspot definitions are inherently arbitrary. From ChIP-chip data, it appeared that most hotspots cover the majority or entirety of the IGRs where they reside (Cromie et al. 2007). At finer scale, however, Rec12 oligos generally arose from only a portion of each IGR (Fig. 2E; Supplemental Fig. S2C). On average, intergenic hotspots occupied only 57% of their IGRs, with hotter hotspots tending to encompass more, and in only 21% of IGRs with a hotspot did oligos extend over >90% of the IGR (Supplemental Fig. S2C). Prominent clusters of largely intergenic Rec12 oligos often extended into adjacent coding regions (Fig. 2E) and spatially correlated with some sites of noncoding RNA (ncRNA) generation (Rhind et al. 2011; Chen et al. 2012). For intergenic ncRNAs, Rec12 oligos were enriched within the ncRNA region but not within their flanking regions (Supplemental Fig. S2D). Thus, DSBs are not as constrained by coding regions as it previously appeared (Cromie et al. 2007).

Susceptibility of IGRs to DSB formation did not correlate with the presence of promoters or adjacent gene expression (data not shown). Within IGRs, breakage was rarely promoter associated: Only 19% of IGR oligos were within 200 bp of coding gene transcription start sites (TSSs), and this number dropped to 5% for IGRs wider than 2.5 kb. Separating IGRs by the orientation of their flanking genes showed that oligos were most dense between divergent genes and least dense between convergent (Supplemental Fig. S2E). This overall pattern is similar to S. cerevisiae IGRs, but S. pombe IGRs had overall lower oligo densities (because many DSBs arise outside of IGR-associated hotspots; see below), and S. pombe showed a much smaller quantitative distinction between promoter-containing (divergent and tandem) and promoter-lacking (convergent) IGRs. Thus, unlike in S. cerevisiae, promoters of protein-coding genes are only rarely prominent targets for DSB formation. At least superficially, this pattern is reminiscent of that observed in mice, where DSB hotspots, measured as RAD51- or DMC1-bound DNA, are generally intergenic but not promoter associated (Smagulova et al. 2011; Brick et al. 2012).

Abundant non-hotspot DSBs that were previously undetectable

A hotspot-centric view is biased toward the idea that hotspots produce all or the vast majority of DSBs and crossovers. However, 28% of all Rec12 oligos mapped outside of definable hotspots (Figs. 1B, 2A,B). This fraction is greater than that in S. cerevisiae (11.4%) despite similar minimum thresholds for calling hotspots (approximately twofold over genome average oligo density in both cases). While the mean density of these “cold-region” Rec12 oligos was low (22.6 RPM/kb), it was 32-fold higher than that within the rDNA (Supplemental Fig. S1G). Significantly, these oligos share a hallmark local base composition bias with their hotspot counterparts (discussed further below), indicating that cold-region and hotspot oligos are formed by similar or identical mechanisms. Similar findings were obtained from both sequencing platforms used here and from an independent study (data not shown; P Schlögelhofer, pers. comm.). Thus, we conclude that the overwhelming majority of cold-region oligos represent bona fide Rec12-generated breaks. The low, nearly uniform distribution of these breaks (Fig. 1B, bottom) would render them invisible to lower sensitivity methods such as ChIP-chip or Southern blot hybridizations.

Non-hotspot DSBs account for crossovers in “DSB-cold” regions

Meiotic crossing-over occurs frequently at great distances from DSB hotspots in S. pombe; these crossovers are Rec12 dependent, but prior studies were unable to detect enough DSBs to account for them (Young et al. 2002). The large number of Rec12 oligos arising from regions outside of hotspots strongly argues that this recombination originates from DSBs outside hotspots, and that such DSBs make up much of the total chromosome breakage and crossovers (see below). In support of this hypothesis, genetic intervals that had a substantial fraction of cold-region oligos and small intervals without hotspots displayed a linear relationship, over a >100-fold range, between the number of Rec12 oligos and the genetic distance (cM) (Fig. 3A; Supplemental Fig. S3A; Supplemental Table S4), consistent with the interpretation that each DSB in these intervals yields a crossover with roughly equal probability. As expected for the relatively uniform distribution of cold-region DSBs, intervals with few or no hotspot-associated oligos showed a linear relation between genetic distance and DNA length, but very short intervals (<5 kb) with the hotspots mbs1 and ade6-3049 were outliers (Supplemental Fig. S3B).

Open in a separate window

Figure 3.

DSBs in hotspots are much less likely to give rise to crossovers than are DSBs in cold regions. (A) Genetic map length is proportional to Rec12-oligo count, but regions containing strong hotspots behave differently from regions lacking hotspots. Rec12 oligos were summed across intervals with known genetic distances; replicate map lengths were averaged. Sources of map length data are in Supplemental Table S4. The Pearson correlation coefficient r is based on black points; red points are short intervals containing strong DSB hotspots. (B) Different ratios of crossing over to DSB frequency (cM/1000 oligos) at the mbs1 hotspot and an adjacent DSB “cold region.” Genetic distances are from Cromie et al. (2005). (C) DSBs from “cold regions” are more likely to yield a crossover than DSBs in “hot” regions, and the degree of bias in crossover fate correlates with the region’s heat. For each genetic interval in A, the fraction of Rec12 oligos that originated from hotspots was determined (see B for an example). Red points are the outliers in A. See Supplemental Figure S3 for other correlations and estimates of the contribution of different types of intervals to crossover numbers.

Unequal recombination potential and crossover invariance

Further examination of quantitative relationships between Rec12 oligos and crossing over provided new insight into the phenomenon of crossover invariance, in which numerous intervals display a nearly uniform crossover density (i.e., nearly constant cM/kb) despite some intervals having a prominent DSB hotspot(s) and others not (Hyppa and Smith 2010). These investigators showed that an important component of crossover invariance is locus-dependent differences in partner choice for DSB repair. DSBs at the strong hotspots mbs1 and ade6-3049 are biased toward repair by formation of a Holliday junction with the sister chromatid and less frequently produce an interhomolog crossover; distinct genetic requirements for recombination in hotspots vs. cold regions suggest that DSBs in cold regions are more likely to be repaired with the homolog and often produce a crossover (Hyppa and Smith 2010).

In agreement with prior findings (Cromie et al. 2005; Hyppa and Smith 2010), crossover frequencies between markers closely flanking the mbs1 and ade6-3049 hotspots were much lower than in other intervals with similar numbers of oligos (Fig. 3A). We now find similar behavior at other DSB hotspots in the short ade3–ura3 (55-kb) and ura2–leu2 (13-kb) intervals, where most Rec12 oligos (88% and 92%, respectively) arise from sites that score as hotspots (Fig. 3A; Supplemental Fig. S3A). We infer that this is a general pattern: Unlike small intervals without hotspots (see above), small intervals with a strong DSB hotspot experience much less crossing over than predicted from Rec12-oligo frequency. We therefore estimated how much each DSB contributes to the genetic map (cM per oligo). In the ura1–mbs1L interval, which is directly adjacent to mbs1 but does not contain a hotspot, we observed 4.6 cM/1000 oligos, whereas the nearby mbs1L–mbs1R interval containing the mbs1 hotspot displayed only 1.1 cM/1000 oligos (Fig. 3B). Similar results were obtained from the newly implicated hotspots: The ade3–ura3 and ura2–leu2 intervals, which contain DSB hotspots, had 0.7 and 0.5 cM/1000 oligos, respectively.

We considered that partner choice might vary continuously across the genome, such that crossover invariance is stronger at more intense hotspots, i.e., sites more frequently broken tend to be repaired more frequently using the sister chromatid. This view fits with the notion that hotspots exist as a continuum with the largely DSB-cold genome. As predicted by this hypothesis, we found for a large collection of genetic intervals a strong anticorrelation between cM/oligo values and the fraction of Rec12 oligos that were hotspot derived (Fig. 3C). There was a four- to fivefold difference in cM/oligo values between the coldest and hottest intervals, which is similar to the ratio of intersister:interhomolog Holliday junctions at the strong DSB hotspots mbs1 and ade6-3049 (4:1 and 3:1, respectively) (Cromie et al. 2006; Hyppa and Smith 2010). This result implies that the degree of repair template bias is proportional to the frequency of hotspot DSBs in the region and suggests that this pattern is pervasive.

One way to rationalize these patterns is to hypothesize that the frequency of DSB repair with the sister vs. with the homolog depends on the fraction of DSBs in the interval that depend on the Rec25–Rec27–Mug20 complex. This complex binds loci to variable extents and, in hotspots, in proportion to DSB frequency (Fowler et al. 2013). Loci with a high population-average level of Rec25–Rec27–Mug20 bound also have a high frequency of DSBs (hotspots) and are repaired primarily with the sister (IS repair) (Cromie et al. 2006; Hyppa and Smith 2010). Loci rarely bound by Rec25–Rec27–Mug20 have less frequent DSBs (cold regions) and, we infer from genetic requirements (Hyppa and Smith 2010), are repaired primarily with the homolog (IH repair). In this view, since the population average amount of Rec25–Rec27–Mug20 bound is a continuous variable, so is the population average IS:IH ratio of partner choice for repair, a consequence of the hypothesized dual functions of Rec25–Rec27–Mug20.

We conclude that crossovers are often formed outside hotspots from low-level, widely distributed DSBs. Indeed, we estimate below that about half of all crossovers come from intervals outside the previously identified (i.e., stronger) hotspots. These previously undetectable DSBs are repaired preferentially with the homolog to give more crossovers per DSB than do hotspot DSBs, which are repaired preferentially with the sister chromatid (Cromie et al. 2006; Hyppa and Smith 2010). Importantly, these results show that crossover invariance is not limited to the two previously tested intervals but likely extends to the entire genome.

Number of DSBs and crossovers per meiotic cell: Low-level DSBs account for approximately half of all crossovers

The nearly linear relation between Rec12-oligo count and percent DNA broken at a range of hotspots (Supplemental Fig. S1H) allowed us to estimate that the average number of DSBs formed per cell is ~58 (95% CI = 48–68) (Supplemental Material). This is fewer than half the breaks in S. cerevisiae (~160 on average) (Pan et al. 2011) despite similarly sized genomes.

We wished to estimate the total number of crossovers per cell, but doing so from DSB numbers is complicated by crossover invariance, particularly variability in partner choice (see above). Motivated by the analysis above (Fig. 3C), we assumed that the likelihood of interhomolog (rather than intersister) repair is a linear function of the DSB site’s break frequency (Supplemental Fig. S3C). We then calculated the partner choice bias for each hotspot, ranging from 20% interhomolog repair at the hottest hotspot to 100% at the coldest. We further assumed that 75% of interhomolog repair events result in a crossover, the approximate frequency observed among gene convertants in both hotspots and cold regions (Grimm et al. 1994; Cromie et al. 2005). We thus estimated ~36 crossovers per meiosis, which agrees well with prior estimates of total genetic map length: 45 crossovers (Munz 1994) or 34 crossovers (Egel 2003). Importantly, our estimate was relatively robust to changes in the shape of the function used to model partner choice bias (linear, sigmoidal, exponential, and step-function variations ranged from ~27 to 38 crossovers per meiosis; Supplemental Fig. S3C). The concordance of our estimates with experimental values reinforces our conclusion that crossover invariance is based on differences in partner choice for DSB repair genome-wide.

We estimated the fraction of all crossovers arising from DSBs outside the strong hotspots that were previously detectable (i.e., those with more than ~0.3% breakage) by using the relation between hotspot intensity and frequency of repair with the homolog, as above. Using the linear relation (Supplemental Fig. S3C), we calculated that 46% of all crossovers arise from these DSB-poor regions, of which the majority (32% of total) came from DSBs that did not map to a scoreable hotspot at all. The sigmoidal, exponential, and step functions yielded similar estimates of 43%, 59%, and 44%, respectively. These results show that previously undetected DSBs account for nearly half of all crossovers and underscore the importance of sensitive assays when assessing genome-wide features. They also show that spatially dispersed meiotic DSBs are more important than previously suspected.

Relationship of DSBs to local chromatin structure

Chromatin is known or hypothesized to shape DSB distributions in many organisms (Lichten 2008), but precisely how it does so and the degree of evolutionary conservation are unknown. Until now, the only available high-resolution genome-wide data in meiosis for both chromatin structure and DSB distributions were from S. cerevisiae (Jiang and Pugh 2009; Pan et al. 2011; Zhang et al. 2011). In this yeast, most DSB hotspots correspond to places where the chromatin is hypersensitive to digestion with exogenous nucleases (DNase I or micrococcal nuclease [MNase]) (Ohta et al. 1994; Wu and Lichten 1994), and a large majority of Spo11 oligos map to clear NDRs, mostly in promoters (Pan et al. 2011). Thus, Spo11 takes advantage of constitutively accessible regions in chromatin and targets these preferentially for cleavage, further suggesting that Spo11 only rarely if ever cuts DNA wrapped into nucleosomes (Pan et al. 2011). Importantly, although the degree of nucleosome occupancy within or adjacent to an NDR is not a good predictor of hotspot heat, nucleosome occupancy is nevertheless an excellent predictor of local DSB spatial patterns (Pan et al. 2011). As a result, it is often easy to guess where DSB hotspots might be simply by looking at a nucleosome map, or conversely to guess where NDRs are by looking at a Spo11-oligo map (Fig. 4A, bottom). It has been argued that S. pombe hotspots are similarly lacking in nucleosomes and that DSBs are largely restricted to places that are nucleosome-depleted in most cells in a population (Hirota et al. 2007; de Castro et al. 2012; Yamada and Ohta 2013). However, prior genome-wide data had insufficient resolution to fully test this hypothesis. We therefore explored this question using recent maps of MNase-resistant mononucleosomal DNA generated from S. pombe meiotic chromatin (Soriano et al. 2013). As detailed below, relationships between population-average chromatin structure and DSB distributions differ substantially between fission and budding yeasts.

Open in a separate window

Figure 4.

Markedly different relationship between DSBs and chromatin structure in S. pombe and S. cerevisiae. (A) Rec12 oligos are not especially enriched in NDRs, unlike Spo11 oligos. High-resolution mononucleosome coverage maps (MNase-seq) from meiotic S. pombe (Soriano et al. 2013) and S. cerevisiae (Pan et al. 2011) cells were compared with their respective oligo maps for representative regions on each organism’s chromosome 1. Nucleosome maps were normalized by taking the number of times each base pair was sequenced and dividing it by the mean genome coverage (i.e., the mean genome coverage is 1). Horizontal blue and red bars indicate hotspots. Orange arrows mark protein-coding genes and white boxes indicate transcripts. (B) Fission yeast hotspots are much wider than budding yeast hotspots and not generally centered on TSSs or NDRs. Rec12- and Spo11-oligo hotspot midpoints were aligned and the mean oligo distributions were smoothed with a 201-bp Hann window. Relative positions of annotated meiotic NDRs (Soriano et al. 2013) and TSSs (Lantermann et al. 2010) in S. pombe are indicated by blue ticks at top; S. cerevisiae NDRs and TSSs (Jiang and Pugh 2009) are in red. (C) Unlike Spo11-oligo hotspots, Rec12-oligo hotspots tend not to be strongly MNase sensitive in meiotic chromatin. Hotspots were aligned as in B and the mean mononucleosome coverage was determined at each position.

Focusing first on hotspots, we found that Rec12 oligos were not strongly associated with NDRs. Figure 4A (top) illustrates a typical hotspot region: At this site, Rec12 oligos were not restricted to NDRs but instead arose frequently from positions whose nucleosome occupancies varied over a wide range. Thus, in stark contrast to S. cerevisiae, it would not be possible to guess the locations of DSB hotspots from the chromatin map or to guess where NDRs are from the Rec12-oligo map. When hotspots were superimposed, there was little if any tendency toward local clustering of either annotated meiotic NDRs (Soriano et al. 2013) or TSSs (Lantermann et al. 2010), again very different from S. cerevisiae (Fig. 4B). Indeed, 16% of IGRs with a hotspot lacked an NDR entirely. When averaged over all hotspots, nucleosome coverage tended to be modestly lower than in flanking regions (Fig. 4C), but this pattern was weak compared with that in S. cerevisiae, where hotspots tightly overlap on average with a deep and narrow NDR flanked by arrays of relatively well-positioned nucleosomes (Fig. 4C). Thus, the strong tendency for S. cerevisiae DSBs to prefer NDRs is not true for S. pombe DSBs. As these are population-average measures, it is not possible to distinguish whether Rec12 often cuts DNA bound to nucleosomes or only cuts in subsets of cells where nucleosomes are absent from these positions. Interestingly, S. pombe hotspots do resemble a small number of exceptionally wide (>1.5 kb) S. cerevisiae hotspots, which tend to have low overall occupancy of relatively disordered nucleosomes (Pan et al. 2011).

To further investigate chromatin-DSB connections, we examined Rec12-oligo distributions from the perspective of specific chromatin contexts (i.e., without regard for the arbitrary designation of hotspots). Aligning TSSs or meiotic NDRs revealed previously described stereotypical nucleosome patterning around these features, such as the tendency for nucleosome depletion in promoters and positioned arrays of nucleosomes in transcription units and upstream flanking regions (Fig. 5A,B; Lantermann et al. 2010; Soriano et al. 2013). Many of these patterns are similar to those in S. cerevisiae, but overall, chromatin structure appears more “irregular’” in fission yeast, i.e., lower amplitude of occupancy signal for positioned nucleosomes, implying more variability from location to location.

Open in a separate window

Figure 5.

DSBs have complex relationships with local chromatin features. (A) Spatial pattern of Rec12- and Spo11-oligo formation adjacent to TSSs. Annotated TSSs in S. pombe (top) and S. cerevisiae (bottom) were aligned and oriented with transcription to the right (arrow). The mean mononucleosome coverage (Pan et al. 2011; Soriano et al. 2013) around the TSSs was determined as in Figure 4C. To determine the average spatial relationship between DSBs and TSSs, oligo counts around individual TSSs were divided by the total oligos in the 2-kb region plotted, then these normalized profiles were averaged across all TSSs and smoothed with a 101-bp Hann window for clarity. Normalization prevents sites with many oligos from dominating the average profile. Note that this procedure reveals the average shape of the Rec12- and Spo11-oligo distributions, not the average intensity. The fraction of total Rec12 or Spo11 oligos found within 500 bp upstream of TSSs (dashed lines) is indicated. (B) Rec12-oligo profiles are strongly patterned around NDRs. The midpoints of NDRs in S. pombe (constitutive and meiosis-specific) (Soriano et al. 2013) and S. cerevisiae (Jiang and Pugh 2009) were aligned and the mean normalized oligo profiles were determined as in A. The fraction of total Rec12 or Spo11 oligos found within 500 bp of NDRs (dashed lines) is indicated. Additional data are in Supplemental Figure S4. (C) Heatmap of mononucleosome occupancy around S. pombe NDRs. The NDRs from B (top) were ranked by width and the degree of mononucleosome coverage was determined around each site. The coverage profiles were smoothed for clarity and plotted on a color scale (top). NDRs were binned based upon their width for subsequent analysis (right ruler; bin 1, ≤174 bp; bin 2, 175–199 bp; bin 3, 200–249 bp; bin 4, 250–349 bp; bin 5, ≥350 bp). (D) Rec12 oligos are biased toward NDR boundaries and the flanking nucleosomes, not NDR centers. The NDRs within each bin in C were aligned and mean normalized profiles were calculated as in A. The distance between the maxima in the Rec12-oligo peaks in each bin (dashed lines) is indicated at the top. The mean width of NDRs in bin 1 is 162 bp; bin 2, 187 bp; bin 3, 221 bp; bin 4, 287 bp; bin 5, 458 bp.

To examine the spatial distribution of DSBs around these sites, we divided the oligo count at each base pair by the total number of oligos in the 2-kb window centered on an individual NDR or TSS. These normalized maps were then averaged to obtain a single Rec12-oligo profile. This normalization prevents profiles being dominated by the subset of sites with the highest numbers of oligos nearby and allows us to examine the global tendencies of local spatial patterns separate from variation in DSB frequency from one region to another. Importantly, unlike in budding yeast where TSS- or NDR-proximal regions account for a large majority of all Spo11 oligos, these regions in S. pombe account for only a minority fraction of Rec12 oligos (Fig. 5A,B).

Rec12-oligo distributions were spatially correlated with patterns of nucleosome coverage at TSSs and NDRs, with broad zones of enrichment overlapping with the regions with lowest nucleosome coverage (Fig. 5A,B). However, oligos tended to be most highly enriched along the NDR-proximal sides of the nucleosomes flanking TSSs (Fig. 5A) and to an even greater extent at meiotic NDRs (Fig. 5B). Oligos also frequently spilled over into the regions covered by the flanking nucleosomes, such as the typically well-positioned first nucleosomes of transcriptional units. (The particularly high-level enrichment to the left of TSS-associated NDRs in Fig. 5A is likely due to averaging over many large IGRs that include very wide hotspots.) NDR centers do not exhibit biased nucleotide composition relative to the genome, and sequences from them can be mapped as well as or better than random S. pombe sequences (Supplemental Fig. S4); thus, the observed central depletion of Rec12 oligos is not a mapping artifact.

By comparison, normalized Spo11-oligo distributions around S. cerevisiae TSSs and NDRs correlated much more tightly with the regions of MNase-sensitivity and appeared more strongly restricted by the flanking nucleosomes (Fig. 5A,B). Notably, Spo11 oligos within NDRs exhibited a modest central depletion akin to the strong depletion of Rec12 oligos in NDR middles, possibly indicating a shared spatial patterning that is much more pronounced in S. pombe. To better assess how the genome-wide average reflects specific sites, we rank-ordered NDRs by width and divided them into groups of similar-sized NDRs (Fig. 5C,D). Strikingly, differences in NDR widths between groups were matched by a concomitant shift in the spatial pattern of Rec12-oligo density, confirming that Rec12 oligos tend to be strongly enriched across and into the well-positioned nucleosomes flanking the NDRs, with a paucity in the NDR center itself. The group with the narrowest NDRs had widths comparable to those in budding yeast; this group displayed the same spatial bias as for wider NDRs (Fig. 5D), demonstrating that differences between the species are intrinsic to the DSB–chromatin relationship and not simply a consequence of different average NDR widths.

These findings illustrate themes in Rec12-oligo formation with regard to chromatin structure. First, loci largely depleted of nucleosomes are not especially favorable for break formation. Although IGRs containing hotspots often contain an NDR (de Castro et al. 2012), only a minority (~20%) of Rec12 oligos originate from within NDRs themselves (Fig. 2B) and hotspots can extend several kilobases from an NDR. Indeed, most NDR cores are actually depleted locally for oligos (Fig. 5B,D). Thus, while the transplacement of large DNA fragments containing an NDR can generate an ectopic DSB hotspot (de Castro et al. 2012), the spatial patterning observed here indicates a more complex relationship. Second, Rec12 oligos formed within NDRs tended to be adjacent to the NDR boundary and often arose from the NDR-flanking regions where well-positioned nucleosomes lie. This pattern agrees well with direct comparison of DSBs and MNase cleavage at two hotspots (mbs1 and cds1) previously analyzed by Southern blotting (Hirota et al. 2007). This striking spatial positioning could reflect Rec12 preferentially cleaving DNA near, or exiting, a nucleosome. However, these analyses rely on population averages, so we do not yet know whether a mechanistic relationship exists between Rec12 and histone proteins. Nucleosomes may be absent at the subset of sites or chromatids where the breaks are made, or Rec12 or one of its partner proteins may have a DNA-binding preference similar to that of nucleosomes.

Influence of transcription factor binding on DSBs

Several sequence-specific TFs influence hotspot activity and DSB formation, perhaps through effects on local chromatin organization and/or by direct interaction with the break-forming machinery (Kon et al. 1997; Petes 2001; Steiner and Smith 2005). Although it has been proposed that TFs play a dominant role in shaping and evolving DSB landscapes (Wahls and Davidson 2010; Steiner et al. 2011), it was not previously possible to test this hypothesis in S. pombe. Rec12 oligos allowed us to investigate the genome-wide relationship between TF binding and DSBs.

In S. pombe, DNA binding by the CREB-family TF Atf1-Pcr1 alters local chromatin structure and generates DSB hotspots in a context-dependent manner: Transplacement of the ade6-M26 allele, a binding site for Atf1-Pcr1, on 3- to 6-kb DNA fragments generates a hotspot in only a few chromosomal contexts (Ponticelli and Smith 1992; Virgin et al. 1995; Kon et al. 1997; Steiner et al. 2002; Hirota et al. 2008). We searched for the Atf1-Pcr1 core binding motif (5′-TGACGT) (Steiner and Smith 2005) and analyzed Rec12 oligos nearby; similar results were obtained with the more extended sequence 5′-ATGACGT (Schuchert et al. 1991; data not shown). For the 448 motifs (of 2141 total) with at least as many Rec12 oligos nearby as in the weakest hotspot (i.e., ≥140 RPM within ±1 kb), oligos were enriched on average for ~2 kb on either side and were depleted in the ~50-bp region centered on the motif (Fig. 6A). (Averages of locally normalized Rec12-oligo profiles are depicted in this and subsequent figures of sequence motifs to separate local spatial patterns from site-to-site variation in total oligo number.) This pattern was more pronounced if we considered only sites within IGRs (Supplemental Fig. S5A). Local depletion over the motif agrees with results of direct DSB detection by Southern blot hybridizations (Steiner et al. 2002) and resembles average patterns around binding sites of the S. cerevisiae TFs Abf1 and Reb1 (Pan et al. 2011).

Open in a separate window

Figure 6.

Atf1-Pcr1 and select short DNA sequence motifs are spatially correlated with DSBs. (A) Rec12-oligo distribution around Atf1-Pcr1 motifs. Sites with the motif 5′-TGACGT and ≥140 RPM total within 2 kb were oriented and aligned (n = 448). Mean normalized oligo profiles were determined as in Figure 5A and smoothed with a 51-bp Hann window for clarity (inset shows central 1 kb). (B) Atf1-Pcr1 associated oligo distributions are often asymmetric at individual sites. Oligo counts around each site in A were binned, clustered according to oligo distribution (cluster 1 loci have ≥50% more oligos to the left than to the right of the motif; cluster 2 loci have ≥50% more oligos to the right), and plotted as a heatmap. Clustering by k-means reveals a similar pattern (data not shown). (C) Transcription and Rec12 oligos predominate on opposite sides of asymmetric Atf1-Pcr1 motifs. Oligo counts in the clusters from B were averaged after normalization as in Figure 5A and compared with mean meiotic transcript abundance (red and cyan lines) (Chen et al. 2012). Oligos were smoothed with a 201-bp Hann window for clarity. (D) Rec12 oligos around simple sequence motifs. A PWM was constructed for 5′-ACACAC- and 5′-ACTGCT-like motifs (top logos), and genomic positions scoring ≥95% match to a PWM and having ≥50 oligos within the 500-bp window (n = 1021 and 1024, respectively) were identified. Averages of normalized Rec12-oligo distributions at these sites were determined as in Figure 5A and smoothed with a 51-bp Hann window (blue lines). The mean mononucleosome coverage around the aligned motifs during meiotic (solid gray lines) and mitotic (dashed gray lines) growth was then determined (Soriano et al. 2013). (E) Spatial pattern of Rec12 oligos around 5′-AAATTT motifs (n = 2200). Sites with the motif and ≥10 RPM total within 100 bp were aligned, and normalized Rec12-oligo profiles were determined as in Figure 5A. Oligo profiles were clustered by k-means (k = 10) and plotted as in B. The distance between adjacent oligo clusters is ~10 bp; see Supplemental Figure S6F for the composite mean profile. See Supplemental Figures S5 and S6 for further analyses of sequence motifs associated with Rec12 oligos.

These average Rec12-oligo profiles seemed to show that DSBs tend to be symmetrically arrayed around this TF’s binding sites, and indeed such a pattern was seen at the Atf1-Pcr1-dependent ade6-3049 hotspot (Supplemental Fig. S5B). However, closer examination showed this was often not the case. Instead, about half of motif-containing sites had higher oligo frequencies to one side or the other (Fig. 6B). This asymmetry correlated strongly with nearby transcription such that lowest oligo density occurred in transcribed regions (Fig. 6C), but did not correlate with local chromatin structure (measured by MNase resistance; data not shown). Atf1-Pcr1 binding at some locations may stimulate local DSB formation and transcription in opposite directions, and/or transcription may inhibit DSB formation in the transcribed region. The remaining motif-containing sites had more symmetric oligo distributions and no bias in transcript abundance (data not shown).

The spatial correlation of oligos with Atf1-Pcr1 motifs is tempered by the fact that most motifs were not associated with high Rec12-oligo density: Only 19% were above the genome average, a nonstringent threshold. Even motifs in large IGRs (which generally contain a hotspot) had oligo densities above the large-IGR mean only 27% of the time. Although 28% of all hotspots contained a motif, only 12% of motifs were in hotspots (~6% of the genome is in hotspots). Thus, the motif alone is a poor predictor of break location or frequency.

Because DNA sequence often does not predict TF binding in vivo, including for Atf1-Pcr1 (Eshaghi et al. 2010), we examined experimentally defined Atf1-Pcr1 binding sites. Atf1-Pcr1 responds to oxidative stress in addition to pre-meiotic nitrogen starvation (Shiozaki and Russell 1996). Since binding in meiosis has not been reported, we analyzed the 250 sites bound by Atf1 during oxidative stress as a proxy (Eshaghi et al. 2010). These sites—but not those bound by Atf1 only prior to oxidative stress—displayed Rec12-oligo profiles similar to that around the Atf1-Pcr1 motif, i.e., central depletion flanked by zones of enrichment (Supplemental Fig. S5C). We found that 49% of the bound sites overlapped a hotspot, but only a minority (19%) of hotspots was bound by Atf1. Note that the number of detectably bound sites is about 10 times fewer than the number of motifs.

Taken together, our findings imply that Atf1-Pcr1 can stimulate DSB formation over a region of ~4 kb at select sites, with further spatial patterning conferred by other features of the region. In certain contexts, such as the 2.6-kb-wide ade6-3049 hotspot, breaks are induced to both sides. Importantly, most Atf1-Pcr1 motifs are not strong DSB sites, and the vast majority genome-wide are not bound by Atf1-Pcr1 in vivo and/or do not give rise to appreciable DSB activity. These findings fail to support the hypothesis that the Atf1-Pcr1 motif alone, or even the more stringent criterion of Atf1 binding, is sufficient to target high levels of DSB formation nearby (Wahls and Davidson 2010).

Other DSB-associated motifs

Two additional TF-associated motifs were more recently demonstrated to activate DSB hotspots at particular DNA consensus sequences in S. pombe: oligo-C hotspots depend on Rst2, and CCAAT hotspots depend on the Php2–Php3–Php5 complex (Steiner et al. 2011). We therefore assessed Rec12-oligo patterns near positions of their core motifs, 5′-CCCCGC (n = 456) and 5′-CCAATCA (n = 1408). Only 20% of oligo-C and 9% of CCAAT motifs fell within hotspots (Supplemental Fig. S5D); these fractions are higher than expected by chance but represent a small minority of all motif occurrences. Conversely, only 10% and 16% of hotspots contained the respective motif. Thus, as for Atf1-Pcr1 motifs, their presence was a poor hotspot predictor. However, when we aligned these motif-containing sites, Rec12 oligos were spatially correlated: oligos tended to be locally depleted directly at the motif with higher Rec12-oligo frequency in flanking regions (Supplemental Fig. S5E,F). As at Atf1-Pcr1 motif sites (Fig. 6C), Rec12-oligo frequency was anticorrelated spatially with meiotic transcript abundance (data not shown). These findings suggest that, although these TFs probably contribute fairly little to the overall hotspot map, the subset of sites they activate are also subject to spatial patterning conferred by the TF.

Assignment of multiply mapping sequences

The multiply mapping sequences from the SOLiD data set were enriched for very short read lengths (Fig. 1C), indicating that this population contained many oligos that mapped to multiple locations because they were too short rather than because they arose from bona fide repetitive DNA elements. A map incorporating these reads was generated as follows: All potential map sites were considered for each read and the number of uniquely mapping oligos within ±250 bp of each site was determined; for each ambiguous read, a fractional value was assigned to each of its map positions proportional to the number of unique oligos in the vicinity. For example, a read that mapped to two sites, one with 90 unique reads nearby and the other with 10, would be assigned to these sites as 0.9 and 0.1 oligos, respectively; see Supplemental Figure S1E for an actual example. This map, in which we have imputed each read’s originating locus, was added to the unique data to generate a combined map. The three maps (unique, multiple-mappers, and combined) were analyzed as described in the text. Sequences that mapped to two or more positions from the 454 data set were not analyzed further.

Comparison with recombination frequencies

To further validate our map and to explore the relationship between DSB formation and recombination, we used previously published recombinant frequencies (Supplemental Table S4). Most of the genetic map lengths (cM) were calculated from random spore analyses using standard auxotrophic markers, while recombination in the ade3–ura3 interval was determined by tetrad analysis and in the ade6L–ade6R by a physical measure of crossover DNA. Genetic distances were calculated from random spore analysis as cM = −50 ln(1 – 2R), where R is the fraction of recombinant spore colonies among total spore colonies analyzed; the distance for tetrad analysis was cM = 100 (T + 6NPD)/2 (PD + NPD + T), where T, NPD, and PD are the fractions of tetratype, nonparental ditype, and parental ditype asci among total asci analyzed. Crossover DNA at the ade6–3049 hotspot was measured using diploids with heterozygous restriction sites flanking the hotspot and Southern blotting to probe for the appearance of recombinant DNA fragments (Hyppa and Smith 2010). Since only one of the two recombinant DNA molecules (R2) was assayed, genetic distance was calculated as cM = 2 × (R2/total DNA), where R2 is the mean of the accumulated recombinant DNA at 6 and 7 h in a meiotic induction (R1 could arise by partial restriction digestion).

Chromatin structure analysis

Transcription start sites (TSS) in S. pombe were from previously published studies (Lantermann et al. 2010). Only TSSs with annotated start and end positions were used. Positions previously defined as depleted of nucleosomes (NDR) during meiosis in S. pombe (constitutive and meiosis-specific NDRs) were compiled from Soriano et al. (2013). The spatial distribution of Rec12 oligos around TSSs and NDRs was determined as in the analysis of sequence motifs described above. Comparisons with Spo11 oligos in S. cerevisiae used previously defined TSSs and NDRs (Jiang and Pugh 2009), and the curated S288C Spo11-oligo map and “N2” 4-h nucleosome map from Pan et al. (2011).

To allow direct comparisons between nucleosomes and Rec12 and Spo11 oligos, we used the MNase-seq data from Soriano et al. (2013) to derive mononucleosome coverage maps. Our primary interest was to determine where Rec12 oligos arise relative to DNA that resides in nucleosomes, so we wished to use nucleosome coverage maps (as in Pan et al. 2011) rather than the maps of nucleosome midpoints used by Soriano et al. (2013). We therefore mapped their raw sequence reads (GEO accession GSE41772) to the same S. pombe reference genome used above in mapping Rec12 oligos. For meiotic mononucleosome coverage we used the data set from 3 h after a synchronous pat1-114 meiotic induction (accession number GSM1024001); our analyses of mitotic coverage used data derived from asynchronous cultures grown in rich media (GSM1024003).

Since in the preparation of these libraries DNA was size selected to enrich for mononucleosome-sized fragments, sequence tags should map to the boundaries of nucleosomes on opposite strands. Since the tag positions are a function of nucleosome position and the degree of MNase digestion, we first analyzed the pairwise distances between sequence tags on the plus and minus DNA strands to estimate the degree of MNase digestion. Under ideal conditions the most frequent distance should be ~147 bp, the length of DNA typically protected from MNase by the nucleosome. However, we found that the most frequent tag distance was only ~127 bp (data not shown). We assume this is due to ~10 bp being “nibbled” by MNase from each end of nucleosomal DNA, so we generated coverage maps by computationally extending the 5′ end of each sequence read 10 bp to its 5′ side and 137 bp to its 3′ side (147 bp total), representing the inferred nucleosome footprint, and then calculating the number of occurrences of each base in the genome. These coverage maps agreed well with coverage maps generated by computationally extending the nucleosome midpoint maps of Soriano et al. (2013) by 73 bp in each direction (data not shown). Since the maps of Soriano and colleagues have been subjected to extensive smoothing, we elected to use our remapped version of their data instead. Note that using coverage maps instead of midpoint maps results in more rounded peaks and narrower internucleosomal linker regions in profiles of individual sites (e.g., Fig. 4A) or averages over many sites (e.g., Fig. 5A). Importantly, however, this method of analysis makes the S. pombe data directly comparable to the S. cerevisiae data. The maps were normalized by dividing the coverage at each position by the average coverage genome-wide (the genome average ≡ 1).

We are especially grateful to Peter Schlögelhofer (University of Vienna) and Michael Eickbush and Sarah Zanders (FHCRC) for sharing unpublished data. We thank Agnes Viale (MSKCC Genomics Core Laboratory) for sequencing; Nick Socci (MSKCC Bioinformatics Core Facility) for assistance with sequencing data processing and mapping; Sam Tischfield (MSKCC) for assistance with data analysis; and Sarah Zanders, Naina Phadnis, Sue Amundsen (FHCRC), and Shintaro Yamada (MSKCC) for helpful comments on the manuscript. This work was supported in part by NIH grants GM031693 and GM032194 (G.R.S.) and GM058673 (S.K.).