henhaobio
Sign in

bioRxivGenomicsPreprint

Biobank-scale genotyping of Robertsonian translocations reveals hidden structural variation on the human acrocentric chromosomes

Arang RhieiD, Juhyun KimiD, Francisco Rodríguez-AlgarraiD, Steven J. SolariD, Sergey KoreniD, Dmitry AntipoviD, Caralynn M. WilczewskiiD, George L. Maxwell, Jennifer L. GertoniD, Justin E. Paschall, Tamara PotapovaiD, Tyra G. WolfsbergiD, Sumeeta Singh, Sandra del CastilloiD, Clesson TurneriD, Vardhman K. RakyaniD, Adam M. PhillippyiD

PDFDOICodeData
In-depth full-text digest available · Jump to digest ↓

Abstract

Balanced Robertsonian translocation (ROB) is the most common chromosomal rearrangement in humans, with an estimated occurrence of 1 in 800 in newborn studies. Carriers are at increased risk of cancer and often diagnosed at fertility clinics after facing recurrent miscarriages, infertility, or aneuploid offspring. Genotyping carriers with DNA sequencing has been challenging because of gaps and misrepresentation of the translocation fusion site in the human reference genome. Only recently, telomere-to-telomere (T2T) human genomes successfully revealed sequences of the acrocentric short arms, including the most common ROB fusion site. A ROB results in loss of two ribosomal DNA (rDNA) arrays and its adjacent distal sequences, including the highly conserved distal junction (DJ). Here, we present a new method to type ROB carriers directly from short and long sequencing reads by estimating DJ copy number. We demonstrate that our method successfully genotypes ROBs using a reference-free approach or alignments to either T2T-CHM13v2 or GRCh38. Applying this method to a cohort of healthy newborns and family members (n=4,172) as well as the UK Biobank (n=490,416), we find candidate ROBs at a frequency consistent with the previously reported incidence (0.11-0.12%). In addition to ROB carriers, we report the frequency of one DJ loss (9, 2.8-3.4%) or gain (11+, 8.4-9.3%) from the two cohorts and the 1000 Genomes Project (n=3,202), and characterize the underlying structural variation in near-T2T genome assemblies from the Human Pangenome Reference Consortium. Importantly, we provide the first screening method for ROB carriers using low-coverage sequencing data, enhancing its clinical applicability and enabling new studies of structural variation on the acrocentric chromosomes.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 3 (13 Oct 2026) · CC BY 4.0

Short-read genomes can flag Robertsonian translocation carriers by counting rDNA distal junction (DJ) copies (8 rather than 10); UK Biobank screening found 583 candidates, 1 in 841.

Robertsonian translocations (ROBs), carried by 1 in 800 people and tied to infertility, miscarriage and cancer risk, are mostly found by karyotyping, fluorescence in situ hybridization (FISH) or Hi-C on live cells, because reference gaps and an unclear breakpoint kept whole-genome sequencing from detecting them. Using telomere-to-telomere (T2T) assemblies of the acrocentric short arms, the authors offer what they call the first sequencing-based ROB diagnostic, usable on raw reads or GRCh38 alignments.

Key results

  1. ShownIn five control genomes, DJ copy number separated known ROB carriers from non-carriers. This held for estimates from masked T2T-CHM13 alignments, from GRCh38 alignments and from reference-free 31-mer counts. The three estimates also agreed in cohort samples.

    All three approaches gave ~10 DJs for CHM13 and HG002 and ~8 DJs for ROB cell lines GM03417, GM03786 and GM04890; T2T assemblies and FISH had confirmed the ROBs. In 92 Reverse Phenotyping Core (RPC) samples spanning every DJ copy bin, GRCh38 Fast-precise and k-mer estimates correlated strongly with CHM13-based estimates. · Detecting ROBs by the loss of rDNA distal…; One in 800 ROB candidates in newborn family…

  2. SuggestedCandidate ROB carriers (8 DJs) occurred near the expected 1 in 800 in a newborn-family cohort and in UK Biobank. One 1000 Genomes Project (1KGP) candidate had a karyotype showing a chromosome 13 and 21 fusion.

    RPC: 4 individuals had 8 DJs, 3 of them among 2,612 sequenced unrelated parents, giving 1 in 871-883 (0.11%) vs an expected 1 in 800 (0.125%). UK Biobank: 583 samples with 8 DJs, 1 in 841 (0.12%). 1KGP: 7 individuals with 8 DJs. One of them, HG01204, showed a 13 and 21 fusion on G-banding done for Human Pangenome Reference Consortium (HPRC) quality control. · One in 800 ROB candidates in newborn family…; DJ and PHR copy number distribution in UK…; Uncovering structural variation within the DJ of…

  3. ShownSingle DJ losses (3.1% of people) and gains (~8.9%, the more frequent change) recur across cohorts and can be inherited. Copy number of the ROB-associated pseudo-homolog region (PHR) is broadly polymorphic and not significantly associated with DJ count, so the authors judge it not a reliable ROB marker.

    Across RPC, UK Biobank and 1KGP, 2.8-3.4% of individuals had 9 DJs and 8.4-9.3% had 11+. Monozygotic twins had identical DJ and PHR counts. In RPC, parents' DJ counts were slightly higher than offspring's (Mann-Whitney p = 3.7 × 10^-6). PHR copies in RPC: 6 in 52.3%, 5 in 39.3%, 4 in 6.2%. PHR vs DJ for 8-10 DJ copies: Pearson r=0.01, p=0.39. · Abstract; One in 800 ROB candidates in newborn family…; DJ and PHR copy number distribution in UK…; False positive and false negative rate

  4. ShownNear-T2T HPRC Release 2 assemblies resolved the sequence behind DJ changes: each single loss had its own breakpoint, while most gains shared one partially duplicated block.

    Of 6 assembled samples with 9 DJs, 3 resolved the loss junction, and their breakpoints did not overlap. In one, the rDNA array start was inverted (supported by ultra-long ONT reads). In two families, FISH found the same loss in a parent. Of 8 samples with the extra-DJ junction assembled, 6 shared a duplicated block that copied only the first 34 kb of the DJ arm. · Uncovering structural variation within the DJ of…

  5. SuggestedDJCounter works on shallow data and can call a single sample against a reference distribution, using a Gaussian mixture model (GMM) with fixed component means.

    On HG002 subsampled to 5-50×, Fast-precise and Fast-refine were stable above 5× but read +0.7 copies high. K-mer estimates came close to 10 copies above 10×. A GMM trained on 2,402 1KGP samples and tested on 800 gave every test sample a posterior above 0.5 and most above 0.9, including ROB samples. · Computing resources and prediction model; GMM fitting and classification

Methods, data and reproducibility

DJ copy number is computed from short reads in two ways. One is target-to-autosomal depth on a masked T2T-CHM13 or on GRCh38 (four GRCh38 modes, mapped onto the CHM13 scale by linear regression). The other is reference-free: the median count of 52,227 DJ-specific 31-mers relative to the 2-copy k-mer peak. 8 DJs flags a ROB candidate. DJ losses and gains were then inspected in HPRC assemblies and by FISH.

  • Control Illumina WGS: CHM13 and HG002 (10 DJs) plus ROB cell lines GM03417, GM03786, GM04890 (8 DJs) · human · n = CHM13, HG002 and 3 ROB cell lines · SRA SRX1009644 (CHM13); dbGaP phs003920.v1.p1 (ROB lines)
  • Reverse Phenotyping Core (RPC) cohort of healthy newborns and family members, Illumina WGS re-aligned from hg19 · human · n = n=4,172 (1,325 families, 4,187 samples)
  • UK Biobank WGS, CRAM files aligned to GRCh38 · human · n = n=490,444 (abstract: 490,416)
  • 1000 Genomes Project 30x Illumina NovaSeq 6000 WGS, BAM files aligned to GRCh38 · human (1000 Genomes Project) · n = 3,202 (2,504 unrelated)
  • HPRC Release 2 near-T2T assemblies plus Verkko v2.2.1 assemblies of the same genomes; acrocentric haplotypes released · human (Human Pangenome Reference Consortium) · n = 226 long-read sequenced samples; 59 trio and 159 Hi-C Verkko assemblies · 10.5281/zenodo.18895340
  • Simons Genome Diversity Project copy number estimates used to check DJ copy stability · human (Simons Genome Diversity Project)
Methods & tools
DJCounter, Meryl, Merqury, GenomeScope 2.0, BWA-MEM, samtools, bedtools, Minimap2, Picard MarkDuplicates, Verkko, Mashmap, FISH
Software & environment
Versions stated: BWA-MEM v0.7.17, samtools v1.20-v1.23, bedtools v2.31.1, Meryl v1.4.1, Minimap2 v2.30, Picard v3.3.0, Verkko v2.2.1, statsmodels v0.14.5, scipy v1.13.1; UK Biobank jobs ran via swiss-army-knife v4.5.0 on the UKB Research Analysis Platform.
Assay conditions & compute
On DNAnexus, samtools depth was too costly for all UK Biobank samples (1 hour vs 2 minutes with samtools idxstats). The k-mer method ran with 120 GB of memory. CPU time and peak memory per mode on HG002 at 5-50× were taken from SLURM accounting.
Reproducibility gaps
No accession or access route is printed for the RPC cohort sequencing data. UK Biobank CRAMs were reached through the UKB Research Analysis Platform. ROB control reads are in dbGaP (phs003920.v1.p1).
Limitations (6)
  • None of the four RPC individuals with 8 DJs could be recontacted or agreed to take part, and UK Biobank recontact required extensive review, so no RPC or UK Biobank candidate was confirmed by karyotype. · stated by the authors
  • Structural follow-up used 1KGP lymphoblastoid cell lines, whose DJ region may carry cell-line-specific variants. · stated by the authors
  • Copy number alone cannot tell a ROB from two DJ-less chromosomes, or catch a ROB plus a duplicated DJ. Rough estimates put false positives at up to one third of 8-DJ calls and false negatives near 7.2%, so karyotype confirmation is advised. · stated by the authors
  • Fast mode counts duplicate, secondary and supplementary reads, which causes batch effects tied to how BAMs were made and can make a strict 8-copy cutoff hard to apply. Cohort-level calling needs a large cohort to estimate the distribution. · stated by the authors
  • Positive ground truth is three known ROB cell lines plus one karyotyped 1KGP sample; at cohort scale, support rests mainly on the match to the expected 1 in 800 frequency. · noted in this summary
  • UK Biobank was typed only with the pilot Fast mode (unfiltered read counts, with a DJ target later extended), so its estimates are not on the same footing as the final DJCounter modes. · noted in this summary

Read this if you work with short-read WGS cohorts or clinical genomes and want to screen for Robertsonian translocations or study copy-number variation on the acrocentric short arms.

Citations
1

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISLima '25Guarracino '23Mostovoy '24Jammihal '26Čechová '25Chapman '24
61% semantic matchGenomics
Leonardo Gomes de Lima, Andrea Guarracino, Sergey Koren+17 more·Nature·Sep 24, 2025· 34 citations