Journal: bioRxiv

Article Title: A genome-wide atlas of meiotic recombination intermediates reveals distinct modes of DNA repair that direct crossovers away from transcriptionally marked genes

doi: 10.64898/2026.03.26.714455

Figure Lengend Snippet: (A) SPO11 induces programmed DNA double-strand breaks (DSBs), which are resected to generate single-stranded DNA (ssDNA) coated by RPA, which interacts with BLM. DMC1 replaces RPA to mediate homology search and strand invasion, forming strand-exchange intermediates (D-loops). BLM binds these intermediates, which are ultimately resolved as crossovers (∼10%) or non-crossovers (∼90%). (B) (Top) BLM ChIP-SSDS signal reflects BLM association with resected ssDNA and strand-exchange intermediates and is impacted by their lifespans. (Bottom) Representative hotspots showing SPO11 oligos (measuring DSBs, purple) and BLM ChIP-SSDS signal (read counts per million) in WT mice, illustrating variability in BLM signal relative to break frequency. (C) Scatterplot of BLM signal versus SPO11-oligo counts in WT mice across 16,658 autosomal hotspots (each point represents one hotspot). Points are coloured according to inferred class assignment (high, blue; low, red). Hotspots separate into two distinct classes with systematically higher and lower BLM signal relative to break frequency. (D) Histogram of the strand-exchange–dependent component of BLM signal, inferred through comparison between WT and Dmc1 -/- mice that lack strand exchange (Supplementary Text). The distribution reveals two classes of hotspots with a high (blue) and low (red) strand-exchange component. Hotspots with sufficient signal to infer class membership (>50 ChIP-SSDS reads in Dmc1 -/- , n=1,902) included; bins coloured by the mean inferred class probability of hotspots in each bin. (E) Scatterplot as in (C) for BLM signal in Dmc1 -/- mice versus WT SPO11-oligo counts. This demonstrates that the strand-exchange component of BLM signal disappears in Dmc1 -/- , which eliminates the difference between hotspot classes. (F) Average SPO11-oligo density in matched sets of high (blue) and low (red) BLM signal hotspots (10 bp smoothing), showing similar DSB levels between classes. (G) Average DMC1 signal in matched hotspots in WT mice (100 bp smoothing), shown for the top (dashed) and bottom (solid) strands. They demonstrate comparable strand-invasion efficiency between the two classes. (H) Average BLM signal in matched hotspots in WT mice (100 bp smoothing), shown for the top (dashed) and bottom (solid) strands. The ∼2.3-fold difference between classes reflects differential persistence of strand-exchange intermediates.

Article Snippet: A functional knock-out of Dmc1 was confirmed in meiotic spreads by immunohistochemistry using a rabbit polyclonal anti-DMC1 antibody (Proteintech 13714-1-AP) to test absence of DMC1 foci ( ) as previously described .

Techniques: Comparison

Journal: bioRxiv

Article Title: A genome-wide atlas of meiotic recombination intermediates reveals distinct modes of DNA repair that direct crossovers away from transcriptionally marked genes

doi: 10.64898/2026.03.26.714455

Figure Lengend Snippet: (A) Scatterplot of BLM ChIP-SSDS signal versus SPO11 oligo counts across 16,658 autosomal wild-type B6 hotspots (each point represents one hotspot), illustrating hotspot-to-hotspot variation in BLM signal relative to break frequency. (B) Scatterplot of BLM versus RPA ChIP-SSDS signal in wild-type B6 hotspots, showing close concordance between BLM and RPA, consistent with BLM interacting with RPA. (C) Scatterplot of BLM versus DMC1 ChIP-SSDS signal in wild-type B6 hotspots. Although binding of both proteins depends on resection, BLM exhibits substantial hotspot-to-hotspot variation relative to DMC1, consistent with association of BLM with strand-exchange intermediates and DMC1 with homology search intermediates. (D) (Left) Schematic illustrating that in Dmc1 -/- mice strand exchange does not occur; BLM ChIP-SSDS signal therefore reflects binding only to resected ssDNA intermediates. In wild-type (WT), BLM associates additionally with strand-exchange intermediates. (Right) Representative hotspots showing BLM ChIP-SSDS signal in WT and Dmc1 -/- mice. (E) As in (A) but for BLM in Dmc1 -/- relative to SPO11 oligo counts, exhibiting higher correlation relative to BLM in wild-type. (F) Scatterplot of BLM versus RPA ChIP-SSDS signal in Dmc1 -/- mice, showing close concordance. This high correlation is consistent with BLM and RPA binding the same resected-ssDNA substrate in the absence of strand exchange. (G) Scatterplot of BLM ChIP-SSDS signal in wild-type B6 versus Dmc1 -/- , with hotspots coloured by the inferred probability of assignment to the high (blue) or low (red) signal classes. The plot shows the same class separation evident in : hotspots with elevated and reduced WT-specific BLM signal form distinct groups. (H) Average BLM signal in matched hotspots in Dmc1 -/- (100 bp smoothing, matching for DSB frequency as in ) for the top (dashed) and bottom (solid) strands. No significant difference is observed between classes on average when strand-exchange intermediates are absent. (I) Scatterplot of DMC1 ChIP-SSDS signal versus SPO11 oligo counts, showing no separation between hotspot classes, consistent with comparable strand invasion efficiency. (J) END-seq in wild-type B6 hotspots shown for matched sets of fast- (red) and slow- (blue) resolving hotspots (matched for DSB frequency as in ; data from ) for the top (solid) and bottom (dashed) strands. END-seq detects ssDNA–dsDNA junctions generated during DSB processing and strand exchange. Its characteristic profile reflects two components: (i) a central component where strand-exchange structures accumulate, and (ii) a flanking signal, which arises from a combination of resection endpoints and ssDNA in incomplete strand-exchange intermediates located outside the D-loop on the DSB-initiating chromosome . In wild-type, slow-resolving hotspots show excess of both the central signal and the flanking signal, consistent with higher strand-exchange intermediate persistence. In strand-exchange mutants, the central component is absent and the flanking signal reflects resection-derived junctions alone. The similar END-seq profiles observed for fast- and slow-resolving hotspots in Hop2 -/- and Dmc1 -/- mice (panels K and L) provide further evidence that the divergence in wild type depends on strand-exchange intermediate persistence. (K) END-seq signal in Hop2 -/- mice, shown for the same matched hotspot sets as in (J). This mutant lacks strand exchange; no separation between fast- and slow-resolving hotspots is observed, demonstrating that the divergence in WT END-seq signal in (J) is due to persistence of strand-exchange intermediates. (L) END-seq signal in Dmc1 -/- mice, shown for the same matched hotspot sets as in (J). As for Hop2 -/- mice, no separation between fast- and slow-resolving hotspots is observed in the absence of strand exchange. This again demonstrates that the divergence in WT END-seq signal in (J) is due to persistence of strand-exchange intermediates. (M) RPA ChIP-SSDS signal in wild-type B6, shown for matched sets of fast- (red) and slow-resolving (blue) hotspots (matching as in ) for the top (dashed) and bottom (solid) strands. RPA, which binds resected ssDNA, reproduces the class separation seen with BLM in wild type, supporting differences in ssDNA persistence at strand-exchange intermediates. (N) RPA ChIP-SSDS signal in Dmc1 -/- mice, shown for the same matched hotspot sets as in (M). The difference disappears in Dmc1 -/- , consistent with absence of strand-exchange intermediates. (O) Scatterplot of RPA ChIP-SSDS signal versus SPO11-oligo counts in WT mice across 16,658 autosomal hotspots (each point represents one hotspot). Points are coloured according to inferred class assignment as in (G) (high, blue; low, red). The plot shows the same class separation evident in , consistent with strand-exchange–dependent ssDNA persistence. (P) As in (O), but comparing RPA in Dmc1 -/- to SPO11 oligos, confirming absence of strand-exchange–dependent separation. (Q) Principal Component Analysis (PCA) plot of key recombination measures across 16,658 autosomal hotspots. Points are coloured by inferred BLM cluster probability (high, blue; low, red). PCA separates hotspots according to strand-exchange–dependent behaviour, corroborating the clustering used in the main text.

Article Snippet: A functional knock-out of Dmc1 was confirmed in meiotic spreads by immunohistochemistry using a rabbit polyclonal anti-DMC1 antibody (Proteintech 13714-1-AP) to test absence of DMC1 foci ( ) as previously described .

Techniques: Binding Assay, Generated, Derivative Assay, Mutagenesis

Journal: bioRxiv

Article Title: A genome-wide atlas of meiotic recombination intermediates reveals distinct modes of DNA repair that direct crossovers away from transcriptionally marked genes

doi: 10.64898/2026.03.26.714455

Figure Lengend Snippet: (A) Scatterplot of BLM ChIP-SSDS signal in wild-type hybrid relative to Dmc1 -/- hybrid across 8,293 symmetric autosomal hotspots. Points are coloured by inferred probability of belonging to the slow-resolving (blue) or fast-resolving (red) cluster (as in ). Hotspots with insufficient information for classification are shown in grey. Separation of hotspots observed in B6 is also observed in hybrid symmetric hotspots. (B) As in (A) but restricted to PRDM9 CAST hotspots (n=4,740), demonstrating that class separation is maintained in these hotspots. (C) As in (A) but restricted to PRDM9 HUM hotspots, showing similar separation and confirming allele independence of the separation. (D-F) As (A-C) but for RPA ChIP-SSDS signal. RPA reproduces the same strand-exchange–dependent class separation observed with BLM, confirming that divergence is not specific to BLM occupancy. (G) Principal Component Analysis (PCA) of key recombination measures across 8,293 symmetric autosomal hybrid hotspots. Colour scale reflects the inferred probability of each hotspot belonging to the slow-resolving (blue) or fast (red) clusters. Principal components separate hotspots according to strand-exchange–dependent behaviour, corroborating the cluster structure. (H) Fast- and slow-resolving hotspots differ systematically in single-nucleotide polymorphism (SNP) density, which impacts non-crossover detection. To correct for this bias, non-crossover tract lengths were explicitly modelled for each set (STAR Methods). The inferred mean tract length distributions are shown. (I) Average HFM1 ChIP-SSDS signal in matched sets of hybrid hotspots from slow-resolving (blue) and fast-resolving (red) clusters (matching and smoothing as in ). Unlike fast-resolving hotspots, slow-resolving hotspots exhibit strongly elevated HFM1 association, consistent with stabilisation of strand-exchange intermediates.

Article Snippet: A functional knock-out of Dmc1 was confirmed in meiotic spreads by immunohistochemistry using a rabbit polyclonal anti-DMC1 antibody (Proteintech 13714-1-AP) to test absence of DMC1 foci ( ) as previously described .

Techniques:

Journal: bioRxiv

Article Title: A genome-wide atlas of meiotic recombination intermediates reveals distinct modes of DNA repair that direct crossovers away from transcriptionally marked genes

doi: 10.64898/2026.03.26.714455

Figure Lengend Snippet: (A) Scatterplot of BLM signal in wild-type hybrid relative to the Dmc1 -/- hybrid for symmetric hotspots. Each point represents one hotspot and is coloured by its inferred class assignment (slow-resolving, blue; fast-resolving, red; n=866). The separation of hotspots reflects differences in strand-exchange intermediate persistence similar to B6. Hotspots controlled by PRDM9 CAST (circles) and PRDM9 HUM (squares) are indicated. (B) Fast-resolving hotspots are strongly depleted for crossovers relative to slow-resolving hotspots, whereas non-crossover outcomes are comparatively enriched. At first glance it may seem paradoxical that sites with higher BLM signal (slow-resolving hotspots) produce more crossovers, given that BLM promotes non-crossover repair , . However, recall that the higher BLM signal reflects longer persistence of strand-exchange intermediates rather than increased BLM loading. Fold-differences in recombination outcomes normalised for estimated DNA break frequency (using RPA signal in Dmc1 -/- as proxy ) are shown. Crossovers identified by single-sperm sequencing (left), and crossovers and non-crossovers from a Prdm9 HUM pedigree (right) are shown (data from and ). Non-crossover counts were corrected for local polymorphism density . Error bars indicate 95% confidence intervals (bootstrap n=2000). (C) Average HFM1 ChIP-SSDS signal in matched sets of B6 hotspots from slow-resolving (blue) and fast-resolving (red) classes (matching as per ), indicating preferential stabilisation of strand-exchange intermediates in slow-resolving hotspots. (D) Scatterplot of RPA ChIP-SSDS signal in Mlh3 -/- hybrid versus Dmc1 -/- hybrid testes. Hotspot class separation is preserved in the absence of MLH3-dependent crossover designation, indicating that separation of breaks into fast and slow repair modes precedes crossover specification.

Article Snippet: A functional knock-out of Dmc1 was confirmed in meiotic spreads by immunohistochemistry using a rabbit polyclonal anti-DMC1 antibody (Proteintech 13714-1-AP) to test absence of DMC1 foci ( ) as previously described .

Techniques: Sequencing

Journal: bioRxiv

Article Title: A genome-wide atlas of meiotic recombination intermediates reveals distinct modes of DNA repair that direct crossovers away from transcriptionally marked genes

doi: 10.64898/2026.03.26.714455

Figure Lengend Snippet: (A) Prediction accuracy of individual genomic features for classifying hotspots as fast- or slow-resolving. Logistic regression models (STAR methods) were trained on B6 hotspots (black) and tested on hybrid hotspots (orange). The dashed line indicates random classification (50%). See for a more detailed assessment. (B) Cells with fast-resolving hotspots are strongly enriched for overlap with genes. Scatterplot of BLM signal in wild-type B6 relative to Dmc1 -/- mice, with hotspots aggregated into cells of a 75x75 grid (each point represents one grid cell). Point size reflects the number of hotspots per cell; colour indicates the proportion of hotspots overlapping gene bodies. (C) Heatmap of single-cell RNA-seq levels across spermatogenesis in B6 mice. Each row represents a gene; columns represent pseudotime intervals across meiosis. RNA abundance in each time window was used to predict hotspot class; the zygotene window (green lines) is highlighted (Data from ). (D) Prediction accuracy of RNA level for classifying genic hotspots as fast- or slow-resolving across time windows in meiosis. Four distinct metrics (STAR Methods) consistently identify zygotene expression as the most informative stage. (E) As in (B), but for RNA level of genes overlapping hotspots. Grid cells are coloured by the mean log-scaled zygotene RNA level of genes harbouring the hotspots, illustrating concordance between zygotene expression and rapid strand-exchange resolution. (F) Relationship between strand-exchange–dependent BLM signal (WT/ Dmc1 -/- ratio) and zygotene RNA level for genic hotspots. The sigmoidal Hill-type fit (blue curve), indicates a threshold-like association between RNA abundance and intermediate lifespan.

Article Snippet: A functional knock-out of Dmc1 was confirmed in meiotic spreads by immunohistochemistry using a rabbit polyclonal anti-DMC1 antibody (Proteintech 13714-1-AP) to test absence of DMC1 foci ( ) as previously described .

Techniques: Single Cell, RNA Sequencing, Expressing

Journal: bioRxiv

Article Title: A genome-wide atlas of meiotic recombination intermediates reveals distinct modes of DNA repair that direct crossovers away from transcriptionally marked genes

doi: 10.64898/2026.03.26.714455

Figure Lengend Snippet: (A) Scatterplot of strand-exchange–dependent BLM signal (wild-type B6 relative to Dmc1 -/- ) across hotspots aggregated within cells of a 75 × 75 grid (as in ). Each point represents one grid cell; colour indicates average distance of hotspots in that cell from the non-centromeric telomere, and point size reflects hotspot density. No systematic gradient is observed. (B) As in (A) but coloured by replication timing (Data from ), showing no consistent relationship between replication timing and strand-exchange–dependent class separation. (C) As in (A) but coloured by GC-content within 1 kb of hotspot centres, indicating that GC-content does not explain strand-exchange–dependent class separation. (D) As in (A) but coloured by the proportion of bases within 1 kb of hotspot centres annotated as repetitive elements by RepeatMasker, demonstrating that repeat density does not account for the observed separation. (E) Prediction accuracy for hotspots in fast- and slow-resolving classes separately (as in ). (F) Receiver operator characteristic (ROC) curves for training (B6) and test (hybrid) datasets for selected predictors, demonstrating robust generalisation of Zygotene RNA level-based prediction across genetic backgrounds. (G) Representative fast- and slow-resolving hotspots on chromosome 1, showing SPO11-oligo and BLM ChIP-SSDS signal in WT and Dmc1 -/- mice. (H) Average RNA level of genes harbouring fast- and slow-resolving hotspots across pseudotime in spermatogenesis (top 500 most confidently classified genes per class). Germ-cell stages, namely, spermatogonia (SPG), Leptotene (L), Zygotene (Z), Pachytene (P), Diplotene (D), Meiotic divisions (MD) & Sperm production (SP), are annotated (Data from ).

Article Snippet: A functional knock-out of Dmc1 was confirmed in meiotic spreads by immunohistochemistry using a rabbit polyclonal anti-DMC1 antibody (Proteintech 13714-1-AP) to test absence of DMC1 foci ( ) as previously described .

Techniques:

Journal: bioRxiv

Article Title: A genome-wide atlas of meiotic recombination intermediates reveals distinct modes of DNA repair that direct crossovers away from transcriptionally marked genes

doi: 10.64898/2026.03.26.714455

Figure Lengend Snippet:

Article Snippet: A functional knock-out of Dmc1 was confirmed in meiotic spreads by immunohistochemistry using a rabbit polyclonal anti-DMC1 antibody (Proteintech 13714-1-AP) to test absence of DMC1 foci ( ) as previously described .

Techniques: CRISPR, Biomarker Discovery

Journal: bioRxiv

Article Title: A genome-wide atlas of meiotic recombination intermediates reveals distinct modes of DNA repair that direct crossovers away from transcriptionally marked genes

doi: 10.64898/2026.03.26.714455

Figure Lengend Snippet: (A) SPO11 induces programmed DNA double-strand breaks (DSBs), which are resected to generate single-stranded DNA (ssDNA) coated by RPA, which interacts with BLM. DMC1 replaces RPA to mediate homology search and strand invasion, forming strand-exchange intermediates (D-loops). BLM binds these intermediates, which are ultimately resolved as crossovers (∼10%) or non-crossovers (∼90%). (B) (Top) BLM ChIP-SSDS signal reflects BLM association with resected ssDNA and strand-exchange intermediates and is impacted by their lifespans. (Bottom) Representative hotspots showing SPO11 oligos (measuring DSBs, purple) and BLM ChIP-SSDS signal (read counts per million) in WT mice, illustrating variability in BLM signal relative to break frequency. (C) Scatterplot of BLM signal versus SPO11-oligo counts in WT mice across 16,658 autosomal hotspots (each point represents one hotspot). Points are coloured according to inferred class assignment (high, blue; low, red). Hotspots separate into two distinct classes with systematically higher and lower BLM signal relative to break frequency. (D) Histogram of the strand-exchange–dependent component of BLM signal, inferred through comparison between WT and Dmc1 -/- mice that lack strand exchange (Supplementary Text). The distribution reveals two classes of hotspots with a high (blue) and low (red) strand-exchange component. Hotspots with sufficient signal to infer class membership (>50 ChIP-SSDS reads in Dmc1 -/- , n=1,902) included; bins coloured by the mean inferred class probability of hotspots in each bin. (E) Scatterplot as in (C) for BLM signal in Dmc1 -/- mice versus WT SPO11-oligo counts. This demonstrates that the strand-exchange component of BLM signal disappears in Dmc1 -/- , which eliminates the difference between hotspot classes. (F) Average SPO11-oligo density in matched sets of high (blue) and low (red) BLM signal hotspots (10 bp smoothing), showing similar DSB levels between classes. (G) Average DMC1 signal in matched hotspots in WT mice (100 bp smoothing), shown for the top (dashed) and bottom (solid) strands. They demonstrate comparable strand-invasion efficiency between the two classes. (H) Average BLM signal in matched hotspots in WT mice (100 bp smoothing), shown for the top (dashed) and bottom (solid) strands. The ∼2.3-fold difference between classes reflects differential persistence of strand-exchange intermediates.

Article Snippet: A functional knock-out of Dmc1 was confirmed in meiotic spreads by immunohistochemistry using a rabbit polyclonal anti-DMC1 antibody (Proteintech 13714-1-AP) to test absence of DMC1 foci ( ) as previously described .

Techniques: Comparison

Journal: bioRxiv

Article Title: A genome-wide atlas of meiotic recombination intermediates reveals distinct modes of DNA repair that direct crossovers away from transcriptionally marked genes

doi: 10.64898/2026.03.26.714455

Figure Lengend Snippet: (A) Scatterplot of BLM ChIP-SSDS signal versus SPO11 oligo counts across 16,658 autosomal wild-type B6 hotspots (each point represents one hotspot), illustrating hotspot-to-hotspot variation in BLM signal relative to break frequency. (B) Scatterplot of BLM versus RPA ChIP-SSDS signal in wild-type B6 hotspots, showing close concordance between BLM and RPA, consistent with BLM interacting with RPA. (C) Scatterplot of BLM versus DMC1 ChIP-SSDS signal in wild-type B6 hotspots. Although binding of both proteins depends on resection, BLM exhibits substantial hotspot-to-hotspot variation relative to DMC1, consistent with association of BLM with strand-exchange intermediates and DMC1 with homology search intermediates. (D) (Left) Schematic illustrating that in Dmc1 -/- mice strand exchange does not occur; BLM ChIP-SSDS signal therefore reflects binding only to resected ssDNA intermediates. In wild-type (WT), BLM associates additionally with strand-exchange intermediates. (Right) Representative hotspots showing BLM ChIP-SSDS signal in WT and Dmc1 -/- mice. (E) As in (A) but for BLM in Dmc1 -/- relative to SPO11 oligo counts, exhibiting higher correlation relative to BLM in wild-type. (F) Scatterplot of BLM versus RPA ChIP-SSDS signal in Dmc1 -/- mice, showing close concordance. This high correlation is consistent with BLM and RPA binding the same resected-ssDNA substrate in the absence of strand exchange. (G) Scatterplot of BLM ChIP-SSDS signal in wild-type B6 versus Dmc1 -/- , with hotspots coloured by the inferred probability of assignment to the high (blue) or low (red) signal classes. The plot shows the same class separation evident in : hotspots with elevated and reduced WT-specific BLM signal form distinct groups. (H) Average BLM signal in matched hotspots in Dmc1 -/- (100 bp smoothing, matching for DSB frequency as in ) for the top (dashed) and bottom (solid) strands. No significant difference is observed between classes on average when strand-exchange intermediates are absent. (I) Scatterplot of DMC1 ChIP-SSDS signal versus SPO11 oligo counts, showing no separation between hotspot classes, consistent with comparable strand invasion efficiency. (J) END-seq in wild-type B6 hotspots shown for matched sets of fast- (red) and slow- (blue) resolving hotspots (matched for DSB frequency as in ; data from ) for the top (solid) and bottom (dashed) strands. END-seq detects ssDNA–dsDNA junctions generated during DSB processing and strand exchange. Its characteristic profile reflects two components: (i) a central component where strand-exchange structures accumulate, and (ii) a flanking signal, which arises from a combination of resection endpoints and ssDNA in incomplete strand-exchange intermediates located outside the D-loop on the DSB-initiating chromosome . In wild-type, slow-resolving hotspots show excess of both the central signal and the flanking signal, consistent with higher strand-exchange intermediate persistence. In strand-exchange mutants, the central component is absent and the flanking signal reflects resection-derived junctions alone. The similar END-seq profiles observed for fast- and slow-resolving hotspots in Hop2 -/- and Dmc1 -/- mice (panels K and L) provide further evidence that the divergence in wild type depends on strand-exchange intermediate persistence. (K) END-seq signal in Hop2 -/- mice, shown for the same matched hotspot sets as in (J). This mutant lacks strand exchange; no separation between fast- and slow-resolving hotspots is observed, demonstrating that the divergence in WT END-seq signal in (J) is due to persistence of strand-exchange intermediates. (L) END-seq signal in Dmc1 -/- mice, shown for the same matched hotspot sets as in (J). As for Hop2 -/- mice, no separation between fast- and slow-resolving hotspots is observed in the absence of strand exchange. This again demonstrates that the divergence in WT END-seq signal in (J) is due to persistence of strand-exchange intermediates. (M) RPA ChIP-SSDS signal in wild-type B6, shown for matched sets of fast- (red) and slow-resolving (blue) hotspots (matching as in ) for the top (dashed) and bottom (solid) strands. RPA, which binds resected ssDNA, reproduces the class separation seen with BLM in wild type, supporting differences in ssDNA persistence at strand-exchange intermediates. (N) RPA ChIP-SSDS signal in Dmc1 -/- mice, shown for the same matched hotspot sets as in (M). The difference disappears in Dmc1 -/- , consistent with absence of strand-exchange intermediates. (O) Scatterplot of RPA ChIP-SSDS signal versus SPO11-oligo counts in WT mice across 16,658 autosomal hotspots (each point represents one hotspot). Points are coloured according to inferred class assignment as in (G) (high, blue; low, red). The plot shows the same class separation evident in , consistent with strand-exchange–dependent ssDNA persistence. (P) As in (O), but comparing RPA in Dmc1 -/- to SPO11 oligos, confirming absence of strand-exchange–dependent separation. (Q) Principal Component Analysis (PCA) plot of key recombination measures across 16,658 autosomal hotspots. Points are coloured by inferred BLM cluster probability (high, blue; low, red). PCA separates hotspots according to strand-exchange–dependent behaviour, corroborating the clustering used in the main text.

Article Snippet: A functional knock-out of Dmc1 was confirmed in meiotic spreads by immunohistochemistry using a rabbit polyclonal anti-DMC1 antibody (Proteintech 13714-1-AP) to test absence of DMC1 foci ( ) as previously described .

Techniques: Binding Assay, Generated, Derivative Assay, Mutagenesis

Journal: bioRxiv

Article Title: A genome-wide atlas of meiotic recombination intermediates reveals distinct modes of DNA repair that direct crossovers away from transcriptionally marked genes

doi: 10.64898/2026.03.26.714455

Figure Lengend Snippet: (A) Scatterplot of BLM ChIP-SSDS signal in wild-type hybrid relative to Dmc1 -/- hybrid across 8,293 symmetric autosomal hotspots. Points are coloured by inferred probability of belonging to the slow-resolving (blue) or fast-resolving (red) cluster (as in ). Hotspots with insufficient information for classification are shown in grey. Separation of hotspots observed in B6 is also observed in hybrid symmetric hotspots. (B) As in (A) but restricted to PRDM9 CAST hotspots (n=4,740), demonstrating that class separation is maintained in these hotspots. (C) As in (A) but restricted to PRDM9 HUM hotspots, showing similar separation and confirming allele independence of the separation. (D-F) As (A-C) but for RPA ChIP-SSDS signal. RPA reproduces the same strand-exchange–dependent class separation observed with BLM, confirming that divergence is not specific to BLM occupancy. (G) Principal Component Analysis (PCA) of key recombination measures across 8,293 symmetric autosomal hybrid hotspots. Colour scale reflects the inferred probability of each hotspot belonging to the slow-resolving (blue) or fast (red) clusters. Principal components separate hotspots according to strand-exchange–dependent behaviour, corroborating the cluster structure. (H) Fast- and slow-resolving hotspots differ systematically in single-nucleotide polymorphism (SNP) density, which impacts non-crossover detection. To correct for this bias, non-crossover tract lengths were explicitly modelled for each set (STAR Methods). The inferred mean tract length distributions are shown. (I) Average HFM1 ChIP-SSDS signal in matched sets of hybrid hotspots from slow-resolving (blue) and fast-resolving (red) clusters (matching and smoothing as in ). Unlike fast-resolving hotspots, slow-resolving hotspots exhibit strongly elevated HFM1 association, consistent with stabilisation of strand-exchange intermediates.

Article Snippet: A functional knock-out of Dmc1 was confirmed in meiotic spreads by immunohistochemistry using a rabbit polyclonal anti-DMC1 antibody (Proteintech 13714-1-AP) to test absence of DMC1 foci ( ) as previously described .

Techniques:

Journal: bioRxiv

Article Title: A genome-wide atlas of meiotic recombination intermediates reveals distinct modes of DNA repair that direct crossovers away from transcriptionally marked genes

doi: 10.64898/2026.03.26.714455

Figure Lengend Snippet: (A) Scatterplot of BLM signal in wild-type hybrid relative to the Dmc1 -/- hybrid for symmetric hotspots. Each point represents one hotspot and is coloured by its inferred class assignment (slow-resolving, blue; fast-resolving, red; n=866). The separation of hotspots reflects differences in strand-exchange intermediate persistence similar to B6. Hotspots controlled by PRDM9 CAST (circles) and PRDM9 HUM (squares) are indicated. (B) Fast-resolving hotspots are strongly depleted for crossovers relative to slow-resolving hotspots, whereas non-crossover outcomes are comparatively enriched. At first glance it may seem paradoxical that sites with higher BLM signal (slow-resolving hotspots) produce more crossovers, given that BLM promotes non-crossover repair , . However, recall that the higher BLM signal reflects longer persistence of strand-exchange intermediates rather than increased BLM loading. Fold-differences in recombination outcomes normalised for estimated DNA break frequency (using RPA signal in Dmc1 -/- as proxy ) are shown. Crossovers identified by single-sperm sequencing (left), and crossovers and non-crossovers from a Prdm9 HUM pedigree (right) are shown (data from and ). Non-crossover counts were corrected for local polymorphism density . Error bars indicate 95% confidence intervals (bootstrap n=2000). (C) Average HFM1 ChIP-SSDS signal in matched sets of B6 hotspots from slow-resolving (blue) and fast-resolving (red) classes (matching as per ), indicating preferential stabilisation of strand-exchange intermediates in slow-resolving hotspots. (D) Scatterplot of RPA ChIP-SSDS signal in Mlh3 -/- hybrid versus Dmc1 -/- hybrid testes. Hotspot class separation is preserved in the absence of MLH3-dependent crossover designation, indicating that separation of breaks into fast and slow repair modes precedes crossover specification.

Article Snippet: A functional knock-out of Dmc1 was confirmed in meiotic spreads by immunohistochemistry using a rabbit polyclonal anti-DMC1 antibody (Proteintech 13714-1-AP) to test absence of DMC1 foci ( ) as previously described .

Techniques: Sequencing

Journal: bioRxiv

Article Title: A genome-wide atlas of meiotic recombination intermediates reveals distinct modes of DNA repair that direct crossovers away from transcriptionally marked genes

doi: 10.64898/2026.03.26.714455

Figure Lengend Snippet: (A) Prediction accuracy of individual genomic features for classifying hotspots as fast- or slow-resolving. Logistic regression models (STAR methods) were trained on B6 hotspots (black) and tested on hybrid hotspots (orange). The dashed line indicates random classification (50%). See for a more detailed assessment. (B) Cells with fast-resolving hotspots are strongly enriched for overlap with genes. Scatterplot of BLM signal in wild-type B6 relative to Dmc1 -/- mice, with hotspots aggregated into cells of a 75x75 grid (each point represents one grid cell). Point size reflects the number of hotspots per cell; colour indicates the proportion of hotspots overlapping gene bodies. (C) Heatmap of single-cell RNA-seq levels across spermatogenesis in B6 mice. Each row represents a gene; columns represent pseudotime intervals across meiosis. RNA abundance in each time window was used to predict hotspot class; the zygotene window (green lines) is highlighted (Data from ). (D) Prediction accuracy of RNA level for classifying genic hotspots as fast- or slow-resolving across time windows in meiosis. Four distinct metrics (STAR Methods) consistently identify zygotene expression as the most informative stage. (E) As in (B), but for RNA level of genes overlapping hotspots. Grid cells are coloured by the mean log-scaled zygotene RNA level of genes harbouring the hotspots, illustrating concordance between zygotene expression and rapid strand-exchange resolution. (F) Relationship between strand-exchange–dependent BLM signal (WT/ Dmc1 -/- ratio) and zygotene RNA level for genic hotspots. The sigmoidal Hill-type fit (blue curve), indicates a threshold-like association between RNA abundance and intermediate lifespan.

Article Snippet: A functional knock-out of Dmc1 was confirmed in meiotic spreads by immunohistochemistry using a rabbit polyclonal anti-DMC1 antibody (Proteintech 13714-1-AP) to test absence of DMC1 foci ( ) as previously described .

Techniques: Single Cell, RNA Sequencing, Expressing

Journal: bioRxiv

Article Title: A genome-wide atlas of meiotic recombination intermediates reveals distinct modes of DNA repair that direct crossovers away from transcriptionally marked genes

doi: 10.64898/2026.03.26.714455

Figure Lengend Snippet: (A) Scatterplot of strand-exchange–dependent BLM signal (wild-type B6 relative to Dmc1 -/- ) across hotspots aggregated within cells of a 75 × 75 grid (as in ). Each point represents one grid cell; colour indicates average distance of hotspots in that cell from the non-centromeric telomere, and point size reflects hotspot density. No systematic gradient is observed. (B) As in (A) but coloured by replication timing (Data from ), showing no consistent relationship between replication timing and strand-exchange–dependent class separation. (C) As in (A) but coloured by GC-content within 1 kb of hotspot centres, indicating that GC-content does not explain strand-exchange–dependent class separation. (D) As in (A) but coloured by the proportion of bases within 1 kb of hotspot centres annotated as repetitive elements by RepeatMasker, demonstrating that repeat density does not account for the observed separation. (E) Prediction accuracy for hotspots in fast- and slow-resolving classes separately (as in ). (F) Receiver operator characteristic (ROC) curves for training (B6) and test (hybrid) datasets for selected predictors, demonstrating robust generalisation of Zygotene RNA level-based prediction across genetic backgrounds. (G) Representative fast- and slow-resolving hotspots on chromosome 1, showing SPO11-oligo and BLM ChIP-SSDS signal in WT and Dmc1 -/- mice. (H) Average RNA level of genes harbouring fast- and slow-resolving hotspots across pseudotime in spermatogenesis (top 500 most confidently classified genes per class). Germ-cell stages, namely, spermatogonia (SPG), Leptotene (L), Zygotene (Z), Pachytene (P), Diplotene (D), Meiotic divisions (MD) & Sperm production (SP), are annotated (Data from ).

Article Snippet: A functional knock-out of Dmc1 was confirmed in meiotic spreads by immunohistochemistry using a rabbit polyclonal anti-DMC1 antibody (Proteintech 13714-1-AP) to test absence of DMC1 foci ( ) as previously described .

Techniques:

Journal: bioRxiv

Article Title: A genome-wide atlas of meiotic recombination intermediates reveals distinct modes of DNA repair that direct crossovers away from transcriptionally marked genes

doi: 10.64898/2026.03.26.714455

Figure Lengend Snippet:

Article Snippet: A functional knock-out of Dmc1 was confirmed in meiotic spreads by immunohistochemistry using a rabbit polyclonal anti-DMC1 antibody (Proteintech 13714-1-AP) to test absence of DMC1 foci ( ) as previously described .

Techniques: CRISPR, Biomarker Discovery