To characterize the structure of derivative chromosomes produced by three-way or more complex translocations, we utilized WGS or mate-pair sequencing (MPS) for breakpoint junction analysis, and SNP microarray for copy number analysis. We extracted discordant reads from the WGS or MPS results and confirmed them by PCR and Sanger sequencing. We did not find any simple CCRs, which indicated an exchange of more than three segments in turn among the derivative chromosomes. A total of 193 breakpoints were identified among 14 cases with CCR. The number of DNA breaks was greater than predicted by the number of derivative chromosomes, ranging from 4 to 40 in each case (Table 1, Fig. 1A). The derivative chromosomes and breakpoints involved in the chromosomal rearrangements were illustrated using a circos plot (Fig. 2). We also generated a schematic subway plot showing breakpoints of the derivative chromosomes and their genome position (Fig. 3, Supplementary Fig. 1). All of the cases had one or more cluster breakpoint regions showing a close distance between each DNA break. The median size of the intervals between two DNA breaks was 370 kb and the quartile was 2 Mb (Fig. 1B). Cases 2, 3, 5, 8, and 13 showed a multiple cluster breaks region (Fig. 2). With regard to the transcribed genes, more than half of the breakpoints were located within protein coding regions (212 in 368 breakpoints, 57.3%) (Fig. 1C).

Characterization of three-way translocation breakpoints. A Number of breakpoints in each patient. B Histogram of breakpoint intervals. Fragment sizes were measured between DNA breaks. The vertical axis shows the number of interstitial fragments and the horizontal axis indicates their length. C Locations of the breakpoints and whether they were intragenic or intergenic. D Intra-chromosomal or inter-chromosomal rejoining of the DNA breaks

Circos visualization of genomic rearrangements. A circos plot with arcs is shown and depicts the breakpoint connections among derivative chromosomes. The colors used to denote derivative chromosome correspond to those used in the other figures

Putative structures of three-way translocations and CCRs. Illustration showing putative derivative chromosomal structures on the basis of the breakpoint junctions revealed by WGS of MPS results. Connected lines indicate breakpoint junctions. Genome positions of the breakpoint junctions were determined with reference to the human genome version GRCh37/hg19. Translucent lines indicate a deleted region

We next analyzed how the DNA breaks were reassembled to each other. Most of the shattered fragments were recovered and reassembled with either an inverted or non-inverted orientation almost without missing any segments (Fig. 3, Supplementary Fig. 1). Of note, case 3, which involved as many as 40 breakpoints, did not show any deleted region. Only a small number of copy number losses of larger than 50 kb were identified (case 5, 12, and 13) (Supplementary Fig. 2). Three cases (case 4, 11, and 12) carried simple interstitial deletions on the chromosome not related to the translocation detected microscopically. In contrast, we did not observe any copy number gain. Since the CCRs in our present series involved more than two breakpoints in one or more chromosomes, some breakpoints were fused to others on the same chromosome (intra-chromosomal rearrangement), and on a different chromosome (inter-chromosomal rearrangement) (Fig. 1D). Notably, most of the breakpoint clusters included both intra-chromosomal and inter-chromosomal rearrangements, whereas a subset of breakpoint clusters predominantly included inter-chromosomal rearrangements. For instance, most of the breaks within the breakpoint clusters were found to be fused via an inter-chromosomal rearrangement or to a far distant region on the same chromosome in case 5 (Fig. 2).

A total of 193 breakpoint junctions were validated by PCR and Sanger sequencing at a nucleotide resolution. Most of these junctions were rejoined by blunt ending or via microinsertion or microhomology, irrespective of whether the fusion was intra- or inter-chromosomal (Fig. 4). In some breakpoint junctions, small fragments of unknown origin had been inserted. Of note, the insertion observed in BP13 of case 2, BP37 of case 3, and BP11 of case 8 constituted tandem or inverted repeats with the adjacent sequence of the breakpoint, which was reminiscent of backward or serial slippages that can occur in replication-based mechanisms (Supplementary Fig. 3).

Sequence microhomology at the breakpoint junctions. The vertical axis denotes the number of breakpoint junctions. The horizontal axis indicates the length of the microhomology. Minus number indicates microinsertion. The presence of both microhomology and microinsertion suggests an MMEJ pathway as a mechanism of DNA repair

Four cases in our present study series were found to be of de novo origin, while in the remainder cases, we could not obtain parental samples. We determined a parental origin of these four de novo cases using the genotypes for the derivative and normal chromosomes near to the breakpoint junctions. All four cases were found to be of paternal origin since all of the single nucleotide variants near the breakpoints on all of the relevant chromosomes showed a paternal allele type (Fig. 5, Supplementary Fig. 4).

Parental origin of de novo three-way translocations. The genotypes of the proband, father, and mother are shown from top to bottom. The data indicate that all of the derivative chromosomes in de novo cases were of paternal origin. A Case 13. B Case 12

To further analyze when and how complex rearrangements develop during male gametogenesis, we examined the correlation between the breakpoint distribution and the peak sequence reads detected by Chip-seq, ATAC-seq, and MNase-seq for tissues at various developmental stages of male gametogenesis and early embryogenesis. We first analyzed our four samples with a confirmed de novo paternal CCR origin using a Poisson regression model, but found no significant correlation between the breakpoint distribution and the peak sequence reads. However, when we used all 193 breakpoints among our 14 samples for these analyses, we observed a significantly high enrichment of the breakpoints on the chromatin accessible regions obtained from ATAC-seq of mature sperm using univariate analysis for all set window sizes (window size: 1 kb, RR, 7.8; 95% CI 3.01–20.27; P < 0.001; 5 kb, RR, 73.5; 95% CI 15.94–338.44; P < 0.001; 10 kb, RR, 251.2; 95% CI 26.10–2417.32; P < 0.001; Fig. 6A). In contrast, no significant enrichment was observed in other assays, indicating a chromatin status of mature sperm (ChIP-seq for H3K4me3, ChIP-seq for H3K27me3, Histone-MNase-seq and MNase-seq).

Mature sperm-specific DNA breaks in constitutional CCRs. A Correlations between breakpoint locations and the peaks of the sequence reads in chromatin accessibility assay for mature sperm. The data for ATAC-seq, ChIP-seq (H3K4me3, H3K27me3), Histone-MNase-seq and MNase-seq are shown. B Correlations between breakpoint locations and the peaks of the sequence reads in ATAC-seq for various tissues at various developmental stages of male gametogenesis and early embryogenesis. The data for mature sperm (Liu et al. 2019), zygote, four-cell stage embryo, morula, and trophectoderm are indicated in the left column, while the data for mature sperm (Jung et al. 2019), two-cell stage embryo, eight-cell stage embryo, inner cell mass, and whole testis are indicated in the right column. C Correlations between breakpoint locations and the peaks of the sequence reads in ATAC-seq for various somatic tissues. The data for adipose tissue, bone marrow, fibroblast, and lung are indicated in the left column, while the data for blood, colon, kidney, and prostate are indicated in the right column. Correlations were analyzed by univariate analysis using the Poisson regression model. The dependent variable and the presence of DNA breaks as the objective variable in each window were used to calculate the risk ratio and 95% confidence intervals (95% CIs). All of the database we used are shown in Supplementary Table 1 with the reference

To assess whether the enrichment of breakpoints on a chromatin accessible region on ATAC-seq is specific for mature sperm, we performed similar analysis by Poisson regression for other developmental stages of male gametogenesis and early embryogenesis using samples from whole testes including spermatogonia and spermatocytes, zygotes, and early embryos. Notably, a significant correlation between the breakpoint distribution and the peak sequence reads was obtained only in mature sperm (Fig. 6B). We also examined these correlations using oocytes or other somatic tissues, but no enrichment of the breakpoints on chromatin accessible regions was observed on ATAC-seq (Fig. 6C). The oocytes could not be properly analyzed because of the small number of peak sequence reads in the data set.