Heterozygous inversion breakpoints suppress meiotic crossovers by altering recombination repair outcomes

Heterozygous chromosome inversions suppress meiotic crossover (CO) formation within an inversion, potentially because they lead to gross chromosome rearrangements that produce inviable gametes. Interestingly, COs are also severely reduced in regions nearby but outside of inversion breakpoints even though COs in these regions do not result in rearrangements. Our mechanistic understanding of why COs are suppressed outside of inversion breakpoints is limited by a lack of data on the frequency of noncrossover gene conversions (NCOGCs) in these regions. To address this critical gap, we mapped the location and frequency of rare CO and NCOGC events that occurred outside of the dl-49 chrX inversion in D. melanogaster. We created full-sibling wildtype and inversion stocks and recovered COs and NCOGCs in the syntenic regions of both stocks, allowing us to directly compare rates and distributions of recombination events. We show that COs outside of the proximal inversion breakpoint are distributed in a distance-dependent manner, with strongest suppression near the inversion breakpoint. We find that NCOGCs occur evenly throughout the chromosome and, importantly, are not suppressed near inversion breakpoints. We propose a model in which COs are suppressed by inversion breakpoints in a distance-dependent manner through mechanisms that influence DNA double-strand break repair outcome but not double-strand break formation. We suggest that subtle changes in the synaptonemal complex and chromosome pairing might lead to unstable interhomolog interactions during recombination that permits NCOGC formation but not CO formation.


Introduction
Chromosome inversions have far-ranging impacts on reproduction and speciation when paired with a non-inverted homolog. At the molecular level, heterozygous inversions disrupt fundamental aspects of meiosis by suppressing both the formation and recovery of meiotic crossovers (COs) within the inversion and in the regions nearby but outside the inversion breakpoints [1] (Fig 1B). At the population level, suppressing COs prevents genetic exchange between an inversion and its non-inverted homolog, dramatically reducing gene flow for that portion of the genome [2,3]. Within the inversion, recombination is unable to separate combinations of alleles, which can be beneficial for advantageous or adaptive alleles [4][5][6]. Alternatively, suppression of exchange can have negative outcomes such as harboring selfish genetic elements and meiotic drive systems (reviewed in [7]). Lastly, the ability of inversions to prevent gene flow locally in the genome is the basis of the chromosomal theory of speciation [8,9]. Given the far-ranging impacts of heterozygous inversions on reproduction and speciation, it is critical to understand how they disrupt meiosis at the molecular level. Despite a rich history of studying inversions, the mechanisms of how they suppress COs remain unknown. Here, we exploit the unique properties of inversion breakpoints to provide crucial insight into the mechanisms of CO suppression in Drosophila melanogaster.
Early work on heterozygous inversions focused on two possible explanations for how they might suppress COs. The first possibility is that when inversions are heterozygous with a noninverted chromosome, meiotic chromosome pairing and synapsis are defective, which prevents CO formation [10,11]. The second possibility is that COs can form, but because COs within the inversion will lead to chromosome rearrangements, the gametes containing such chromosomes will be inviable [1,11] (Fig 1B). This will make it appear as if COs are suppressed, when in reality, those chromosomes simply are not recovered in the offspring. This debate was mostly settled in 1936 [12,13] when Sturtevant demonstrated that COs do form within an inversion yet, paradoxically, there is no increased embryonic lethality as expected; he suggested a model where chromosomes resulting from COs within an inversion are aligned on the meiotic spindle such that they are preferentially placed in the polar bodies and eliminated (see [14] for a summary of this model). Despite the lack of direct cytological evidence for Sturtevant's model, the field has generally accepted the notion that heterozygous inversions suppress COs because the aneuploid offspring are not recovered [15][16][17].
Concomitant with the early observation that COs are suppressed within heterozygous inversions was the observation that they are also severely reduced in regions immediately outside the inversion breakpoints [11,12,18]. For example, on a chromosome carrying the dl-49 inversion in D. melanogaster, COs are reduced to approximately 30% of wildtype in the proximal interval and to about 4% of wildtype in the distal interval [11]. This phenomenon is not exclusive to D. melanogaster as it was also seen in D. pseudoobscura by Dobzhansky and Epling [4]. Additionally, in interspecies crosses between D. pseudoobscura and D. persimilis, crossover frequency ranges from 0.0% to 0.5% in regions up to approximately 2.5 Mb outside of the inversion breakpoints, depending on the inversion [19,20]. Since COs that occur outside of the inversion breakpoints will not lead to chromosome rearrangements, there must be a mechanistic explanation for why they do not form in these regions.
COs are made by repairing a DNA double-strand break (DSB) made during meiosis ( Fig 1A). However, only a few select DSBs will be repaired into COs, while the majority of DSBs will be repaired into noncrossovers (reviewed in [21]). If these noncrossovers are associated with gene conversions, then there are small tracts of unidirectional gene transfer from one homolog to the other. A major open question is if heterozygous inversions impact noncrossover gene conversions (NCOGCs) in a manner similar to crossovers.
Data are sparse on how heterozygous inversions affect NCOGCs because they are notoriously difficult to detect due to their small size, less than 1 kb in Drosophila species [22][23][24]. It is possible to select for NCOGCs using purine selection for intragenic recombination events at the rosy locus in D. melanogaster, but these events are so rare that it requires selecting against 500,000 to 1 million offspring in each experiment [23]. Previous work using this approach to study NCOGC frequencies in the interior of an inversion selected against 5,000,000 offspring to recover 66 NCOGCs; this work demonstrated that NCOGCs occur within a heterozygous inversion at the same frequency as in standard arrangement chromosomes [25]. Even attempts at comprehensively assessing the frequency and location of NCOGCs using genome-wide analyses are limited by small sample sizes. Sequencing-based analyses in D. pseudoobscura and D. persimilis showed that NCOGC frequencies inside inversions were at least as high as in 2) The DSB is resected into single-stranded ends, one of which invades the homologous chromosome and primes DNA synthesis. This forms a displacement loop (D-loop) and during synthesis dependent strand annealing (SDSA), this structure can be unwound by a helicase into a NCO. 3) If the second end of the DSB is captured by the D-loop, it also primes synthesis. 4) The second-end capture intermediate is ligated into a double Holliday Junction, which is cleaved by a meiosis-specific endonuclease to form mostly COs, although some NCOs can occasionally form. B) Predicted pairing arrangement between an inversion and a standard arrangement homologous chromosome. Single COs within the inversion breakpoints lead to acentric and dicentric chromosomes with deletions and duplications. If these COs occur, they are not recovered in the offspring (the so-called transmission distortion effect). COs are also suppressed in the collinear regions outside of the inversion breakpoints, even though COs in these regions would not lead to chromosome rearrangements. This suggests that COs are prevented from forming, as opposed to being suppressed due to transmission distortion. https://doi.org/10.1371/journal.pgen.1010702.g001

PLOS GENETICS
collinear regions, but the number of unique NCOGC events was only 32 [26]. Similar sequencing analyses in D. melanogaster show that NCOGCs within inversions occur at rates higher than on standard arrangement chromosomes [27]. While this work was able to analyze the location of 79 unique NCOGCs, it was done using multiply inverted balancer chromosomes, which may not be representative of how single or naturally occurring inversions behave.
The same sequencing-based analyses described above provide some limited insight into how NCOGCs behave outside of inversion breakpoints. In D. melanogaster, 20 of the 79 unique NCOGCs occurred between 0.5 and 1 Mb away from the inversion breakpoints and one NCOGC was detected only 14kb away [27]. In the D. pseudoobscura and D. persimilis analysis, there was one NCOGC 37 kb away from an inversion breakpoint [26]. The low-but non-zero-frequency of NCOGCs within 500 kb of inversion breakpoints suggests that these regions are at least competent to form DSBs, although it remains to be seen if DSBs form at the same frequency in these regions as in collinear portions of the genome. Critically, the datasets on NCOGCs outside inversion breakpoints are too small to draw robust mechanistic conclusions.
The emerging picture of recombination outside of inversion breakpoints is that CO frequency is severely reduced and that NCOGCs can occur. However, it remains unclear whether NCOGCs are reduced by similar levels as COs, preventing crucial insight into whether CO suppression is mediated by a reduction in DSBs or whether recombination is biased away from CO repair ( Fig 1A). To address this, we generated a high-resolution map of rare CO and NCOGC events outside of a single X chromosome inversion in D. melanogaster. Critically, we built full-sibling wildtype and inversion stocks and recovered recombination events in syntenic regions, enabling us to directly compare frequencies and distributions of recombination events. We find that COs are suppressed in a distance-dependent manner from the inversion breakpoint and that NCOGCs occur at wildtype frequencies outside of inversion breakpoints. Together, these data suggest that inversion breakpoints suppress COs by altering recombination outcomes as opposed to suppressing DSB formation.

Drosophila husbandry and stocks used
Flies were maintained at 25˚C on standard cornmeal media. The X chromosome inversion dl- 49, v Of , f 1 was ordered from the Bloomington Drosophila Stock Center (stock #779). Previous sequencing of the FM7 balancer chromosome showed the dl-49 inversion breakpoints occur at nucleotide position 4,897,260 and between nucleotides 13,426,854 and 13,427,212 [28]. We confirmed that these same inversion breakpoints are in the single dl-49 inversion stock using PCR (S1 Fig) and Nanopore sequencing (all Nanopore sequencing data can be found at https://doi.org/10.5061/dryad.rxwdbrvdf). We isogenized chrX and chr2 of the dl-49 stock by first making a dl-49, v Of , f 1 ; Pin/CyO stock. This stock was then crossed to a fully isogenized Oregon-RM stock (courtesy of R. Scott Hawley), and a single male of the genotype dl-49, v Of , f 1 ; +/CyO was crossed back to dl-49, v Of , f 1 ; Pin/CyO females. The male and female offspring of this cross that were dl-49, v Of , f 1 ; +/CyO were crossed together to establish a stock that was dl-49, v Of , f 1 ; +/+, with chr2 in this stock descending from Oregon-RM. The y 1 cv 1 IP3K2 wy-74i f 1 stock was generated by isolating a recombinant between cv 1 and IP3K2 wy-74i from y 1 cv 1 v 1 f 1 / sc 1 ec 1 IP3K2 wy-74i f 1 females (Bloomington Drosophila Stock Center #1274). This new y 1 cv 1 IP3K2 wy-74i f 1 stock was partially isogenized by crossing the single male recombinant to FM7w females [29]. We refer to IP3K2 wy-74i as wy in the main text for clarity. The X and 2 nd chromosomes of the y 1 w 1118 stock were partially isogenized by crossing a single male to FM7w; Pin/ CyO females. All stocks are available upon request.

Generating full-sibling dl-49 and wildtype stocks
To control for differences in genetic background that might influence recombination rates [30], we generated full-sibling dl-49 and Oregon-RM stocks. A fully annotated crossing scheme can be found in S2 Fig. Briefly, in generation 1, males from the previously isogenized dl- 49, v Of , f 1 ; +/+ stock were crossed to previously isogenized Oregon-RM females (stock courtesy of R. Scott Hawley). In generation 2, heterozygous +/ dl-49, v Of , f 1 females were crossed to sibling Oregon-RM males. In generation 3, a single male that contained the dl-49, v Of but had lost f 1 was crossed to homozygous Oregon-RM sisters. The goal of crossing off f 1 was to ensure at least one recombination event near the dl-49 inversion breakpoint, which would increase the amount of shared genetic material between dl-49 and the wildtype full-sibling stock. In generations four through nine, the dl-49, v Of chromosome was maintained in females in a heterozygous state to allow recombination between it and the wildtype Oregon-RM X chromosome. Additionally, in all nine generations, chr2 and chr3 were freely recombining. At the tenth generation, homozygous dl-49, v Of and wildtype Oregon-RM full-sibling stocks were established. It's important to note that the dl-49 chromosome, chrX and chr2 are fully isogenized, while chr3 has shared genetic background but is not isogenized.

Recovering recombinants
Crossovers were recovered in two separate crosses, with slightly different experimental set ups (Fig 2). In the first experimental set up, we generated females that were dl-49, v Of , f 1 / y 1 w 1118 . These females were crossed to males of the reference genome stock y; cn bw sp/CyO. We collected individual male offspring that came from a CO between y and f in the maternal genome.
After completing our first experimental set up, we decided to control for genetic background by generating full sibling stocks as described above. To collect COs in the new full sibling stocks, we generated females that were dl-49, v Of / y 1 cv 1 IP3K2 wy-74i f 1 (Fig 2). These females were crossed to males of the reference genome stock y; cn bw sp/CyO. We again collected male offspring that came from a CO between y and f in the maternal genome. An unforeseen complication arose in this cross because we were unable to detect the difference between nondisjunction and a CO until the CO offspring were sequenced. 42% of the sequenced offspring were nondisjunction events (as determined by the lack of a Y chromosome in the sequencing data). This means approximately 78 of the 185 CO offspring were instead the result of nondisjunction. With a total sample size of 9174 flies, the nondisjunction rate is 0.009%, well within the wildtype rates of nondisjunction. While these nondisjunction events prevent our ability to accurately calculate overall CO frequency in the interval, it does not affect our analysis of CO distributions within the interval.
To recover crossovers in the syntenic regions of chrX in the full-sibling wildtype Oregon-RM stock, we generated females that were +/ y 1 cv 1 IP3K2 wy-74i f 1 (Fig 2). These females were crossed to males of the reference genome stock y; cn bw sp/CyO. We collected male offspring that resulted from crossovers between y 1 and cv 1 or between IP3K2 wy-74i and f.
We were not able to directly select for noncrossover gene conversion events in any of the above crosses. To recover offspring that potentially resulted from NCOGC events, we sequenced 100 randomly chosen male offspring that maintained the parental genotype in the dl-49 and Oregon-RM full sibling crosses.

Illumina whole genome sequencing
We extracted genomic DNA from frozen single flies using the Qiagen DNeasy Blood and Tissue kit (Qiagen #69506) with some modifications. Single flies were homogenized using a pestle (Kimble #749520-0090) in 180 μl of ATL buffer, after which 5 μl of RNase A was added (10 mg/ml, Thermo Scientific #EN0531). We followed all other standard manufacturer protocols, except we eluted the DNA in either 40 or 50 μl of 10 mM Tris (pH 8.0) to avoid incorporating EDTA into the final DNA.
To prepare libraries for sequencing, we used the Illumina DNA Prep kit using the standard manufacturer protocols (cat no #20018705, previously known as the Nextera DNA Flex Library Prep). Libraries were amplified using 6 or 7 rounds of PCR depending on the exact concentration of the genomic DNA. We used the IDT for Illumina UD Indexes, sets A-D (#20027213, #20027214, #20027215, #200272136). The final concentration of the libraries was determined using the ThermoFisher Qubit 1X dsDNA High Sensitivity Kit (#Q33230) and a Qubit 4 Fluorometer. The 260/280 and 230/260 ratios were quantified on a Denovix DS-11 Spectrophotometer.  [47]. B) Cross schemes for all three experimental set ups. COs were identified by scoring for new combinations of parental alleles. In cross 2, cv and wy were not scored because the Cy phenotype interfered with scoring for wy; for consistency, we also did not score cv and used only y and f. https://doi.org/10.1371/journal.pgen.1010702.g002

PLOS GENETICS
Libraries from experimental set up 1 were sent to Nationwide Children's Hospital Genomics Services Laboratory and sequenced on an Illumina NovaSeq 6000 in 2 lanes with S1 2x150 chemistry. Libraries from experimental set up 2 were sent to the University of Minnesota Genomics Center and sequenced on an Illumina NovaSeq 6000 in 2 lanes with S4 2x150 chemistry. All data are available at https://www.ncbi.nlm.nih.gov/bioproject/PRJNA944242/.

Nanopore sequencing
DNA for Nanopore sequencing was prepared from 20 females of the dl-49, v Of stock generated in this study using the method in [31]. Libraries for sequencing were prepared using the ligation sequencing kit (SQK-LSK110) starting with 2 μg of DNA and following the manufacturers' instructions, except that the ligation time was extended to 30 minutes. Prepared libraries were quantified with a Qubit Fluorometer (ThermoFisher) and 600 ng of prepared library was loaded onto a R9.4.1 PromethION flow cell that had been previously run for 48 hours and run for 24 hours. FASTQ files were generated using Guppy version 6.1.1. (Oxford Nanopore Technologies) and aligned to Drosophila melanogaster genome version dm6 using minimap2 [32]. Sniffles2 was used to identify structural variants [33]. Data are deposited at 10.5061/dryad. rxwdbrvdf.

Analysis to identify COs and NCOGCs
Illumina short read data was aligned to the Drosophila melanogaster genome (dm6) using minimap2 [32,34]. Single nucleotide variants (SNVs) were identified using samtools [35]. COs and NCOGCs were identified as previously described [24,27]. Briefly, for all parental stocks, SNVs on the X chromosome that were unique to one of the two parental lines were identified. For each offspring, the parental origin of each SNV was determined and any SNV that represented a switch from one parent to the other was flagged as a possible CO or NCOGC event.

Statistical methods and modeling
When analyzing CO frequencies in dl-49 heterozygotes, we limited our analyses to COs that occurred between the proximal breakpoint and f. COs that occurred proximal to f were recovered in the nonrecombinant offspring and had not been selected for, thus we excluded them from our analyses (Fig 3). Because we only sequenced a subset of the total COs recovered, we did not measure CO frequencies using centiMorgans (cMs), the traditional measurement of CO frequency. Instead, we normalized CO frequencies within each interval by the total number of COs that were sequenced for each genotype. This provided a more accurate representation of the distribution of COs within our samples.
When analyzing CO frequencies in Oregon-RM, we selected for COs that occurred between y and cv or IP3K2 wy-74i and f. However, 21 of the offspring we sequenced had two COs on the same chromosome, one of which occurred between cv and IP3K2 wy-74i . These COs are shown in Fig 3, but were not included in any analyses.
All Fisher's exact tests were performed using QuickCalcs at https://www.graphpad.com/ quickcalcs/. All other statistical tests were performed in GraphPad Prism version 9.4.
To calculate confidence intervals of CO frequencies, the exact 95% confidence intervals of binomial proportions were calculated using the method of Clopper and Pearson [36] implemented at https://statpages.info/confint.html.
The number of NCOGCs that could be recovered from 100 nonrecombinant offspring were predicted using the methods in [24].

Immunofluorescence imaging
2-3 day old mated females were put on yeast paste overnight. Ovaries were dissected in PBS and fixed for 20 minutes in 1000 μl of solution containing 2% paraformaldehyde (Ted Pella When selecting for COs between y and cv or wy and f, we recovered chromosomes that had two COs, one of which was between cv and wy. These are shown in the figure, but the COs were not included in any analyses. CO counts were normalized to sample size by dividing the number of COs per interval by the total number of COs. 95% confidence intervals were calculated and can be found in S1 File. In Oregon-RM, CO distribution between the dl-49 breakpoint and f are not correlated with distance (Spearman's rank correlation, r = 0.18, p = 0.35). In dl-49 heterozygotes, CO distribution between the dl-49 breakpoint and f are correlated with distance (Spearman's rank correlation, r = 0.47, p = 0.01). https://doi.org/10.1371/journal.pgen.1010702.g003

PLOS GENETICS
cat. no. 18505), 0.5% Nonidet P-40 (Sigma cat. no. I8896), 200 μl PBS and 600 μl heptane. Ovaries were then washed three times for ten minutes each in PBS with 0.1% Tween-20 (PBST), blocked for one hour at room temperature in PBS with 1% BSA (MP Biomedicals cat. no. 152401) and incubated with primary antibody diluted in PBST overnight at 4 degrees. Ovaries were then washed three times in PBST and incubated in secondary antibody diluted in PBST for 4 hours at room temperature. DAPI was added for the last 10 minutes at a concentration of 1 μl/ml. Ovaries were washed again three times for 15 minutes each in PBST. All wash steps and antibody incubations were done while nutating. Ovaries were mounted in ProLong Glass (Invitrogen cat. no. P36980) and allowed to cure for the manufacturer's suggested time.
Ovaries were imaged on a Leica Stellaris 5 confocal microscope using an HC PL APO 63x/ 1.4 NA Oil objective. Images were acquired using the Lighting module with an Airy pinhole size of 0.75 AU and the standard default settings dictated by the pinhole size. All images were deconvolved using the Leica Lightning internal software with default settings.
The number of phosphorylated H2AV foci was determined with manual counting in the Leica LASX software. The background signal was subtracted, C(3)G positive cells were identified in each region of the germarium, and individual foci were manually counted through the z-stack. Only nuclei with clearly discernable foci were selected. 2 nuclei from regions 2a and 2b and the single nucleus from region 3 were analyzed in each of 5 germaria from Oregon-RM and dl-49 heterozygotes. The average number of foci per region, standard error of the mean, and a two-way ANOVA with Tukey's posthoc analysis were calculated using GraphPad Prism version 9.4. Raw data are found in S2 File.

Identification of COs in wildtype and dl-49
To determine the mechanism of how heterozygous inversions suppress COs in the regions outside of the breakpoints, we built fine-scale genetic maps of rare CO and NCOGC events within approximately 4 Mb of the breakpoints. We used the X chromosome inversion dl-49 because it is completely euchromatic [28] and visible recessive markers exist at convenient locations to facilitate identification of COs (Fig 2). We performed this experiment in two different genetic backgrounds. First, we generated females that were heterozygous for dl- 49, v, f and a y w marker chromosome, then recovered COs that occurred between y and f (Fig 2). The map length between y and f was 1.76 cM in this genetic background, confirming that the dl-49 inversion does suppress CO formation (Table 1). Second, in order to have a directly comparable wildtype dataset from the same genetic background, we created congenic dl-49 and Oregon-RM stocks by out-crossing the dl-49 inversion into the Oregon-RM background for nine generations and creating full-sibling stocks at the tenth generation (Methods and S2 Fig). We generated females that were heterozygous for dl-49, v and a y cv wy f marker chromosome, then recovered COs between y and f (Fig 2). The map length between y and f was 2.01 cM in this genetic background; however, we were unable to discriminate between a CO and nondisjunction in this cross, so this is an overestimate of map length (Methods and Table 1). Lastly, we generated females using the full-sibling Oregon-RM stock that were heterozygous for the y cv wy f marker chromosome and recovered COs between y and cv or wy and f in order to limit our analyses to the syntenic regions of the inversion breakpoints (Fig 2). The map length was 9.5 cM for y-cv and 10.4 for wy-f (Table 1). These map lengths are statistically different (Fisher's exact test, p < 0.0001) confirming that dl-49 suppresses crossovers to approximately 10% of wildtype levels in the y-f interval.
We next sequenced 145 chromosomes that had COs between y and f in dl-49 heterozygotes and 96 chromosomes that had COs between y and cv or wy and f in Oregon-RM (S1 File). To determine the exact location of the CO, we bioinformatically identified where the genotype switched from the dl-49 or Oregon-RM genotype to the marker chromosome genotype (Fig 2). The dl-49 and y cv wy f stocks had a distribution of approximately 1 SNV every 367 bp on chrX and the Oregon-RM and y cv wy f stocks had a distribution of approximately 1 SNV every 346 bp on chrX (see S3 Fig for the SNV frequency across the entire X chromosome). These SNV densities allowed us to pinpoint the location of the CO with very high resolution. We analyzed CO frequencies by normalizing for sample size and binning CO counts into 150 kb non-overlapping windows (Figs 3 and S5). COs from both dl-49 crosses were combined into one dataset for further analysis because the CO distributions were not statistically different (S4 Fig, Kolmogorov-Smirnov test, p = 0.61).

Distribution of COs in wildtype and dl-49
Of the 145 COs in dl-49 heterozygotes that we sequenced, only nine occurred between the telomere and the distal inversion breakpoint (Fig 3). We wondered if a cryptic inversion polymorphism could be suppressing COs in this region. We performed whole-genome long-read sequencing of dl-49 on the Nanopore platform followed by de novo assembly and did not identify any additional structural variants (data at [39]). Previous studies of the dl-49 chromosome performed in different decades in different labs show that COs are reduced to nearly 0% in this region, so this effect is not unique to the current experiment [11,12,18]. Since there were only nine COs on the distal end, we restricted our analyses to COs on the proximal end.
Historical work studying CO frequencies outside of inversion breakpoints used traditional recessive marker scoring; these markers are usually widely spaced and will typically divide an entire chromosome into five or six genetic intervals [11,12,18]. Modern molecular work using sequencing and SNVs has suggested that inversion breakpoints suppress COs up to 2-3 Mb away from the breakpoint in Drosophila species [19,20,27,40], raising the possibility that heterozygous inversion breakpoints may have stronger effects at shorter distances. Therefore, we asked if CO suppression within large genetic intervals is due to a uniform decrease in CO frequency while the overall distribution is the same, or if there is stronger suppression of COs nearest to the inversion breakpoint. We asked if CO frequency is correlated with distance from the proximal inversion breakpoint, and after normalizing for sample size, we found that in Oregon-RM, CO frequencies between position 13.4 Mb (the position of the dl-49 breakpoint) and f are not correlated with distance (Spearman's rank correlation, r = 0.18, p = 0.35). However, in dl-49 heterozygotes, CO frequencies between the proximal inversion breakpoint and f are correlated with distance, with the lowest rates of COs near the inversion breakpoint (Spearman's rank correlation, r = 0.47, p = 0.01). This suggests that within large genetic intervals, inversion breakpoints influence not only the CO frequency but the distribution as well.
There was a notable absence of COs in dl-49 heterozygotes within approximately 500 kb of the proximal inversion breakpoint, with the closest being 470 kb outside the inversion breakpoint (Fig 3 and S1 File). In Oregon-RM, there were 5 COs in this same interval (Fig 3 and S1  File). The difference in CO frequency in this 500 kb window is statistically different (Fisher's exact test, p = 0.0085). This is consistent with a model where COs are completely suppressed within approximately 500 kb of the inversion breakpoint, although we cannot rule out that CO frequencies are so low in this region that we simply did not recover them given our sample size.

Distribution of NCOGCs in wildtype and dl-49
It remains unclear how NCOGCs are affected by inversion breakpoints because their small size makes them difficult to detect and the existing data sets are small. While we could not select for NCOGCs in our system, we reasoned that by sequencing nonrecombinant offspring, we would recover NCOGCs that occurred close to the inversion breakpoints. More specifically, based on the number of expected NCOGCs per meiosis and the average SNV density, we predicted that if we sequenced 100 nonrecombinant offspring, we would recover 50 NCOGCs on the X chromosome (see methods for details). We performed whole-genome Illumina sequencing on 100 randomly selected nonrecombinant male offspring from dl-49 heterozygotes and Oregon-RM females to identify NCOGCs. These nonrecombinant offspring were selected from the same crosses used to select for the COs described above.
We recovered a total of 51 unique NCOGCs from dl-49 heterozygotes and 23 unique NCOGCs from Oregon-RM across the entire X chromosome (Fig 4 and S1 File). This difference is not due to overall SNV density as it is comparable between the two crosses (S3 Fig). While this difference in frequency is statistically significant (Fisher's exact test, p < 0.0001), it is likely that the increase in NCOGC frequency in dl-49 heterozygotes is at least partially due to transmission distortion as described by Sturtevant and Beadle. As described earlier, since chromosomes with a single CO within the inversion are never recovered in the offspring, only chromosomes with NCOGCs are transmitted and will make up a larger percentage of the offspring, making it appear as if NCOGCs form at a higher frequency. Indeed, Sturtevant and Beadle (1936) used compound X chromosomes to show that single COs occur within dl-49 heterozygotes at approximately 12% but that there is no decrease in embryo viability, making this type of transmission distortion likely specifically within this genotype. Given these considerations, it is not appropriate to conclude that NCOGCs on an inversion chromosome form at higher frequencies, just that they are transmitted at higher frequencies.
Only three of the 51 NCOGCs in dl-49 heterozygotes occurred between the telomere and the distal dl-49 inversion breakpoint, whereas four of the 23 NCOGCs in Oregon-RM occurred in this region (Fig 4 and S1 File). We re-examined the local SNV density in this region and found only 1 SNV every 81,540 bp (S3 Fig). NCOGC length is approximately 400 bp in D. melanogaster [24], thus we did not have enough SNV density to detect NCOGCs in this area and excluded it from further analysis.
In the dl-49 heterozygotes, 30 of the NCOGCs occurred between the proximal breakpoint and the centromere, and two of these 30 NCOGCs were within 500 kb of the inversion breakpoint (Fig 4 and S1 File). In the same region of Oregon RM, there were 11 NCOGCs, 2 of which were within the same 500 kb interval (Fig 4 and S1 File). These proportions are not statistically different (Fisher's exact test, p = 0.29), suggesting that NCOGCs occur in this 500 kb region at the same statistical frequency in Oregon-RM and dl-49 heterozygotes. Our ability to directly compare NCOGC frequencies between genotypes is enabled by our experimental framework, however, the small number of events that occurred in this region (2 vs. 2) limits our statistical power. Thus, an alternative interpretation of our data is that NCOGCs can occur

Cytological analysis of DSB formation and repair
To confirm that our results are not caused by changes in global DSB formation and repair in dl-49 heterozygotes, we performed immunostaining of phosphorylated H2AV as a marker for DSBs (Fig 5 and S2 File). In wildtype germaria, we found an average of 25.2 (SEM = 1.5) foci per nucleus in region 2A, 24 (SEM = 2.27) foci per nucleus in region 2B, and zero foci per nucleus in region 3. In dl-49 heterozygotes, we found an average of 26.2 (SEM = 2.2) foci per nucleus in region 2A, 14.9 (SEM = 4.4) foci per nucleus in region 2B, and 0.6 (SEM = 0.6) foci per nucleus in region 3. A two-way ANOVA with Tukey's posthoc analysis shows that the number of foci is not different between genotypes (region 2A, adjusted p-value = .98; region 2B, adjusted p-value = .09; region 3, adjusted p-value = 0.99), showing that global DSB formation and repair is not disrupted in dl-49 heterozygotes. This is consistent with previous reports showing that global DSB number and repair is not altered in balancer heterozygotes [27,41,42].

Pre-meiotic NCOGCs
There were 5 instances of NCOGCs that occurred in identical locations in different individuals from the dl-49 heterozygotes (S1 File). We did not recover any similar instances in Oregon-RM. To confirm that these were not Illumina sequencing errors or mis-alignments, we confirmed a subset of them with Sanger sequencing (S6 Fig). Since DSBs do not form in hotspots in Drosophila species (44), it is most likely that these NCOGCs occurred pre-meiotically in the germline and were propagated to more than one gamete. This idea is supported by the observation that when meiotic recombination is completely eliminated by mutating the genes required for DSB formation (mei-W68 and mei-P22), there are residual "jackpot" COs; that is, there is a low frequency of COs that occur in the same genetic interval from the same set of parents [43]. Additionally, there are other reports of pre-meiotic jackpot events occurring in genotypes with at least one heterozygous inversion [23,26]. We are unable to conclude if and how heterozygous inversions increase the frequency of pre-meiotic homologous recombination in the germline because we lack a comprehensive analysis; however, this remains an open question for future work.

Discussion
It has been known for almost 100 years that heterozygous inversions suppress COs in the regions immediately outside of the breakpoints [1,11,12], however the underlying mechanism was unknown. Based on the data described here, we propose the following model. The genomic regions near inversion breakpoints are able to form DSBs, potentially at wildtype frequencies. Within approximately 500 kb of the inversion breakpoint, DSBs are preferentially, if not exclusively, repaired into NCOGCs. After 500 kb, DSBs can be repaired as COs, but the likelihood of CO directed repair is low near the breakpoint and increases as distance from the inversion breakpoint increases.
Importantly, our data argue that CO suppression by inversion breakpoints is not caused by a decrease in DSB formation. The frequency and location of the NCOGCs clearly show that DSBs can occur in genomic regions very close to the breakpoints. We and others have previously shown that the interchromosomal effect-the genome-wide increase in CO frequency caused by a heterozygous inversion-is also not mediated by DSB number [27,44]; rather, the number of DSBs stays the same and the increased COs form at the expense of NCOGCs [27].
Together, these data suggest that, in D. melanogaster, plasticity in local and global CO frequencies due to heterozygous inversions is generally mediated by altering recombination repair outcome as opposed to changes in DSB number.

PLOS GENETICS
We also show that COs are suppressed in a distance-dependent manner, where they are suppressed strongly near the inversion breakpoint and increase in frequency farther away. There is remarkable consistency between our direct measurements of CO frequencies and those obtained in other contexts. Indirect measurements of CO frequencies at the distal end of TM3, a multiply inverted third chromosome balancer, suggest that COs can occur as close as 2 Mb to an inversion breakpoint [35]. Direct measurements of CO frequencies during the interchromosomal effect identified COs as close as 1.7 Mb to an inversion breakpoint [27]. Direct measurements in interspecies crosses of D. pseudoobscura and D. persimilis show that COs are suppressed up to 2-3 Mb from the breakpoint [19,20]. Interestingly, in Arabidopsis thaliana, COs are only suppressed approximately 10 kb outside of inversion breakpoints, with some occurring as close as 1 kb [45]; this intriguing difference could be explained by the different relationship between DSB formation and chromosome synapsis in Arabidopsis and Drosophila species [40,46].
Previous data left uncertainty about rates of NCOGCs near inversion breakpoints. For example, our previous work showed that NCOGCs can occur within 500 kb of the breakpoints on multiply inverted balancer chromosomes [27]. However, without a wildtype dataset to compare to, it was unclear whether NCOGCs in these regions were occurring at wildtype or reduced frequencies. Similarly, previous work in D. pseudoobscura found that of 32 unique NCOGCs, only one was within 500 kb, although the sample size in that study may have been simply too small to detect NCOGCs near inversion boundaries. In the current work, we were able to directly compare NCOGC frequencies in congenic stocks and show that NCOGC frequencies in the regions outside inversion breakpoints are statistically the same in the syntenic regions in wildtype (Fig 4).

Crossover suppression at the distal end of dl-49
We observed a striking paucity of COs on the distal end of dl-49. COs are generally reduced in sub-telomeric regions and peri-centromeric regions [24]. dl-49 is only 4.89 Mb from the telomere but is approximately 10 Mb away from the peri-centromeric heterochromatin. We suggest that the combined effects of the telomere and the inversion breakpoint severely restrict CO formation on the distal end, but the proximal breakpoint is too far away from the centromere to be affected. Consistent with this idea is the behavior of the ClB inversion, a larger X chromosome inversion with breakpoints at cytological position 4A5 (approximately position 4.1 Mb) and 17A6 (approximately position 18.35 Mb) [34]. Similar to dl-49, COs between y and the distal breakpoint are very low (0.77 cM), but unlike dl-49, COs are severely suppressed between the proximal breakpoint and the centromere [47]. The proximal breakpoint is only 5 Mb from the pericentromeric chromatin, consistent with the idea that in ClB, the centromere effect combined with the inversion breakpoint severely limits CO formation.
Alternatively, the lack of COs on the distal end of dl-49 could be explained by a pairing defect. The 4.89 Mb between the distal breakpoint and the telomere may not be sufficient for chromatin to maintain stable pairing. Lastly, DSB frequency might be reduced in this region for unknown reasons. Resolving the cause for this interesting observation would require the recovery of all NCOGCs in this region, but unfortunately the poor SNV density prevents us from knowing the true NCOGC frequency.

Chromosome pairing and synaptonemal complex formation in structural variant heterozygotes
Because the regions near heterozygous inversion breakpoints are clearly competent to form DSBs, the next obvious question is why do COs form at such low levels in these intervals? The synaptonemal complex (SC)-a tripartite proteinaceous structure that forms between homologous chromosomes during meiosis-is thought to regulate CO formation [48]. Both the chromosome axes and the SC are likely under increased physical stress near inversion breakpoints due to formation of inversion loops (Fig 1).
Surprisingly, there are no overt chromosome pairing or SC defects in inversion heterozygotes. Analyses of the SC protein C(3)G show that SC does form along the length of a heterozygous FM7 balancer. Additionally, the heterozygous FM7 balancer is paired with its noninverted homolog in 70-80% of meiotic nuclei, depending on the location of the FISH probe [41]. In translocation heterozygotes, which also suppress COs, the SC is built along the length of the chromosome, although it appears to be missing at the translocation breakpoint in 10-20% of nuclei [49]. Translocation heterozygotes are similarly paired in 80-90% of meiotic nuclei [49]. While the severity of the CO suppression in structural variant heterozygotes does not seem to correlate well with the mild pairing and SC defects, it is possible that there are defects in SC formation that cannot be detected with simple immunostaining and would require super-resolution imaging.
We are left with trying to understand how heterozygous inversion breakpoints suppress CO formation even though DSBs are made, NCOGCs form, chromosomes are mostly paired, and the SC is at least superficially intact. One possible explanation is that increased physical stress at the inversion breakpoints prevents DSBs from being repaired into COs very locally; as the distance from the breakpoint increases, the stress dissipates, and normal SC structure allows DSBs to be repaired into COs. Indeed, Sturtevant proposed that defects in chromosome synapsis could be a reason COs were suppressed by inversions when he first observed this in 1926 [1]. Others have proposed that inversion breakpoints might act as CO interference signals [44], which also dissipate with distance [50], or that COs require long uninterrupted blocks of synapsis in order to occur [48]. Our data are certainly consistent with these ideas and cannot discriminate between them.
It is possible that there is hyperlocal asynapsis at inversion breakpoints that could not be detected with previous imaging approaches; however, if this were the case, we would not have been able to detect NCOGCs. Clearly, regions near the breakpoints are able to interact with the homolog at least long enough to form NCOGCs. It is possible that unstable interhomolog interactions in these regions (due to inversion loops) prevent recombination intermediates from being repaired into COs and that the NCOGCs are representative of short-lived interhomolog interactions. Given this, we favor a model where heterozygous inversion breakpoints destabilize interhomolog interactions locally, such that chromosome interactions are stable enough to facilitate DSB formation and recombination in general, but not CO formation. Future experiments detailing how inversion breakpoints impact recombination mechanics at the molecular level and super-resolution imaging analysis of the SC should provide insight into this issue.  Fig. CO frequencies in dl-49 heterozygotes from cross 1 and cross 2 (Fig 2). These distributions are not significantly different from each other and were combined into one dataset