HNRNPU (GenBank: NM_031844.3) encodes the heterogeneous nuclear ribonucleoprotein U, a highly conserved RNA-binding protein involved in RNA splicing, chromatin organization, and nuclear architecture.^1,2 Haploinsufficiency of HNRNPU is associated with a rare neurodevelopmental disorder (developmental and epileptic encephalopathy 54; MIM: 617391) characterized by developmental delay, moderate to severe intellectual disability, epileptic encephalopathy, and muscular hypotonia.^3,4,5,6 Additionally, postzygotic mutations in HNRNPU have been proposed to cause autism spectrum disorder, indicating that both germline and mosaic variation in this gene can be pathogenic.⁷

In the present study, whole-genome sequencing of monozygotic twins with a clinical phenotype consistent with HNRNPU-related neurodevelopmental disorder revealed an unexpected genetic constellation: two closely spaced, mosaic single-nucleotide deletions on the same parental allele, mutually exclusive at the molecular level. Because such a configuration is highly improbable under the conventional assumption of independent mutational events, we systematically interrogated its origin. We assessed two competing models—independent dual mutagenesis versus a single mutational event resolved through divergent sister-chromatid repair—and screened large-scale mosaic and somatic mutation datasets (MosaicBase and COSMIC) for analogous configurations. These analyses lead us to propose clustered monoallelic mosaicism (cMoMa) as a previously unrecognized mosaic configuration, with implications for variant interpretation, recurrence risk estimation, and the broader framework of clustered mutational processes.

Here, we report monozygotic twin males born at 36 + 1 week of gestation after an uncomplicated pregnancy. Both individuals displayed highly similar clinical features. At birth, they exhibited postaxial hexadactyly of the feet, feeding difficulties, and early failure to thrive. Anthropometric measurements showed short stature and microcephaly. Development was characterized by mild global developmental delay, generalized muscular hypotonia, neurogenic hip dysplasia, and pes planovalgus. Both twins experienced febrile seizures. Behavioral features included reduced risk awareness and limited peer interaction.

Subsequent whole-genome sequencing (WGS) of leukocyte-derived DNA identified two closely spaced de novo 1-bp deletions in HNRNPU: c.1463del (p.Pro488Glnfs∗13) and c.1466del (p.Lys489Argfs∗12), each present in mosaic state (Figure 1). Variant-allele fractions (VAFs) ranged from approximately 20% to 39%. Sanger sequencing of DNA from leukocytes and buccal swabs independently confirmed both variants and their mosaic distribution (Figure S1; full details on sequencing methods are provided in supplemental text S1). Combined, the genetic results and clinical phenotype established the diagnosis of HNRNPU-related neurodevelopmental disorder.Figure 1

Overview of the genotype discovered in this study

(A) Schematic of HNRNPU pre-mRNA and the three allelic haplotypes. Two closely spaced 1-bp deletions (c.1463del and c.1466del) occur on the paternal allele, with ambiguous positioning of c.1466del. Long-read phasing confirmed both variants on the paternal haplotype and mutual exclusivity at the molecular level.

(B) Pedigree of the family studied. Checkered boxes resemble mosaic occurrences of the variants c.1463del and c.1466del.

(C) Illustration of the three different allelic combinations identified in blood leukocytes. Observed allele fractions based on whole-genome sequencing differed between twins, consistent with early postzygotic segregation.

Notably, the deletions were mutually exclusive at the molecular level: sequencing reads harbored either c.1463del or c.1466del but never both (Figures 1 and S2A). Long-read sequencing further demonstrated that both deletions are located on the paternal allele (Figures 1 and S2B). Together, these findings indicate that each twin harbors three cellular lineages distinguished by the sequence of the paternal allele: one carrying the c.1463del variant, one carrying the c.1466del variant, and one retaining the wild-type sequence, while the maternal allele remains wild-type in all cells. This configuration—two closely spaced de novo deletions on the same allele segregating into distinct mosaic lineages and never co-occurring on the same DNA molecule—may represent a previously unrecognized genetic constellation, which we term “clustered monoallelic mosaicism” (cMoMa).

Recognizing cMoMa as a rare genetic constellation, we systematically explored potential mechanisms for its origin. Two models could explain this pattern (Figure 2A).Figure 2

Mechanistic hypotheses for the origin of cMoMa

(A) Colored clones indicate lineage segregation through the blastocyst stage and into both twins.

(B) Option 2: a singular event. A single mutational event occurs before sister-chromatid formation. Asymmetric processing during repair creates two distinct 1-bp deletions that are subsequently segregated to opposite sister chromatids after DNA replication.

Model 1 (null hypothesis) assumes that two independent mutational events produced the two 1-bp deletions (Figure S3A). For the two variants to reside on the same allele yet remain mutually exclusive, the deletions must have arisen in distinct embryonic lineages. Since both twins carry both variants, the events must have preceded twinning and occurred within the first approximately eight embryonic cell divisions.^13,14 Quantitative modeling under permissive assumptions for early embryonic mutation rates indicated that the expected occurrence of two independent 1-bp deletions within a ±3 bp window during this period is <10⁻⁷ (Figure 3A and supplemental text S1).^15,16 Restricting the analysis to the second and third embryonic cleavages, which are more consistent with the observed VAFs, the probability decreases further to approximately 10⁻¹⁰. Although this calculation is necessarily approximate, as local mutation rates vary across the genome, these estimates indicate that the independent occurrence of two closely spaced deletions during early embryogenesis is unlikely.Figure 3

Mathematical calculations and previously identified cMoMa cases

(A) Probability calculation for model 1 (two independent de novo 1-bp deletions on the same paternal allele during early embryogenesis) using a Poisson near-pair model. Assuming 12 daughter cells, 28.8 SNVs per cell division, an indel/SNV ratio of 1:13.78, and a 1-bp deletion fraction of 33.8%, the expected probability of two independent deletions occurring within ±3 bp on the same allele is ≤2.3 × 10⁻¹⁰ (D2–D3 window) and <10⁻⁷ even under permissive assumptions, effectively rejecting model 1.

(B) Systematic screening of MosaicBase (4,647 postzygotic mosaic variant entries) for haplotype-resolved cMoMa-like cases. Filtering for ≥2 variants in the same gene within ≤20 bp on the same phased allele identified two cases; WAS (homopolymer tract) and ACVRL1 (GC-rich secondary structure context).

(C) Combined, this gives a total of four reports (five affected individuals) with the specific cMoMa genotype: one from this study (HNRNPU variants), two from MosaicBase (WAS and ACVRL1 variants), and one additional case that we were able to identify in the literature (AHDC1 variants).

Model 2 (alternative hypothesis) assumes that a single mutational event occurred during the earliest embryonic divisions and that subsequent error-prone repair gave rise to divergent sister-chromatid outcomes: a 1-bp deletion on one chromatid and a distinct, nearby 1-bp deletion on the other (Figures 2B and S3B). Following cell division, segregation of the repaired chromatids would produce daughter lineages carrying either variant 1 or variant 2. This mechanism naturally generates a mosaic pattern in which both variants reside on the same parental allele but remain mutually exclusive on individual sequencing reads. It therefore accounts for the molecular exclusivity of the variants, their shared paternal phase, and the mosaic VAFs observed in leukocytes and buccal swabs, including tissue-specific skew. Importantly, this model requires only a single mutational event rather than two independent deletions arising in extremely close proximity.

If cMoMas arise from a reproducible mutational mechanism, similar configurations should occur in other genomic datasets. We therefore explored large-scale mutation resources for comparable patterns, focusing on single-nucleotide deletions. In an exploratory analysis of the COSMIC dataset (229,087 tumor samples), we observed an increased prevalence of closely spaced single-nucleotide deletions compared with a random distribution (Figure S4), suggesting that clustered short deletions occur in large somatic mutation datasets. However, because COSMIC does not provide systematic haplotype-level resolution and cancer variant datasets are influenced by clonal dynamics and variant-calling constraints, these events cannot be assigned to cMoMa. To overcome this limitation, we screened MosaicBase for individuals carrying two distinct variants in the same gene located within 10 bp on the same allele (Figure 3B).¹⁰ Two published cases met these criteria, each involving haplotype-resolved, mutually exclusive single-base deletions consistent with the cMoMa definition (Table S1).^11,12 We acknowledge that, given the small number of haplotype-resolved cases currently available, chance occurrence cannot be formally excluded for the two MosaicBase examples in isolation. In addition, a recently published AHDC1/Xia-Gibbs syndrome case reported two adjacent mosaic deletions, each with allele fractions of approximately 30%–36%, which were mutually exclusive on sequencing reads and assigned to the same parental haplotype by long-read sequencing.¹⁷ Notably, all four cMoMa descriptions, including the twins described here, consist of pairs of adjacent 1-bp deletions in short repetitive or secondary-structure-prone sequence contexts (Figure 3C). In HNRNPU and WAS, the deletions occur in short homopolymeric tracts, whereas in ACVRL1, they occur within a GC-rich region capable of forming secondary DNA structures. Moreover, in all four instances, the dual-deletion pattern was present across multiple tissues and likely occurred early during development.

We present a case of twins with the newly defined cMoMa genotype, which, combined with previously reported configurations, suggests that cMoMa is not unique to the present case but may represent a recurrent mutational configuration arising through a shared underlying mechanism.

These findings position cMoMa as a mosaic counterpart to other clustered mutational phenomena: in cancer genomes, kataegis describes localized hypermutation characterized by clusters of single-nucleotide substitutions; in the germline, multinucleotide mutations capture clusters of closely spaced mutations that occur more often than expected by chance.^16,18 Engineered double-strand breaks induced by Cas9 reproducibly yield heterogeneous, sequence-dependent indel spectra at cut sites, often including 1-bp deletions.¹⁹ Whereas other clustered mutational phenomena originate from episodes involving multiple events, cMoMa is more precisely defined by a single event. It is further characterized by lineage segregation: rather than co-occurrence within the same molecule: divergent outcomes are partitioned across sister chromatids and subsequently across descendant cell lineages, yielding mutually exclusive, monoallelic variants within one individual.

We acknowledge that, given the small number of haplotype-resolved cases currently available, chance occurrence cannot be formally excluded. Further work will be needed to confirm whether our descriptive findings translate into a prevalence of cMoMa that exceeds what would be expected by chance. Future work should aim to characterize the underlying mechanism and determine the true prevalence of cMoMa. Long-read, haplotype-resolved sequencing of trio datasets could enable unbiased detection beyond clinically apparent loci, while single-cell approaches could resolve lineage architecture and developmental timing. Controlled systems such as Cas9-induced breaks at homopolymeric or secondary-structure-prone contexts could directly test whether asymmetric repair reproducibly generates divergent sister-chromatid outcomes.

Discovery of this mechanism might have immediate diagnostic implications. First, it illustrates that mosaicism at a locus does not necessarily reflect expansion of a single allele; instead, multiple distinct alleles may arise from a single event. Second, accurate recognition requires sequencing data with sufficient depth and phasing resolution to detect mutual exclusivity at the molecular level, determine parental origin, and distinguish true clustered variants from complex or recurrent artifacts. Third, for genetic counseling, recurrence risk equals the population baseline. Like other forms of mosaicism, cMoMas arise postzygotically and are therefore incompatible with parental germline mosaicism or inherited transmission. Therefore, in a canonical cMoMa case, future children of the respective parents do not have an increased risk of recurrence. This point is particularly counterintuitive in twins: while shared disease in monozygotic twins is usually assumed to reflect inheritance, here both twins are affected by a postzygotic event. Therefore, caution should be exercised to prevent misclassification as germline transmission, as this was the case in a previous cMoMa description.¹¹ Conversely, since analogous events occurring prezygotically in the germline would be observed as conventional heterozygous variants, dual mosaic cases may represent one of the few observable manifestations of this mechanism. A second source of underascertainment is genomic context: currently recognized cases derive largely from clinical sequencing, which preferentially detects coding or disease-relevant variants. cMoMa-like events in intronic, intergenic, or otherwise clinically neutral regions would therefore usually remain undetected, suggesting that the true spectrum of this mechanism may be broader than the reported cases indicate.

In summary, we define a previously under-recognized form of mosaicism likely resulting from a single mutational event that generates distinct pathogenic lineages through divergent sister-chromatid repair. We term this configuration “clustered monoallelic mosaicism” (cMoMa), defined by mutually exclusive, same-allele variants that never co-occur on a molecule. By illustrating that a single mutational event may yield multiple lineage-segregated variants, cMoMa challenges the conventional assumption that one mutation produces one variant and expands the current framework for interpreting mosaic and clustered mutational processes.

All data supporting the findings of this study are available within the article and its supplemental information. Raw sequencing data (short-read WGS, long-read sequencing, and Sanger traces) are available from the corresponding author upon reasonable request, in accordance with institutional ethics approval and patient consent.

The authors thank Samantha Sarli and Tim Schubert for their intellectual insight and help in revising the manuscript.

J.B., conceptualization, data curation, formal analysis, investigation, methodology, project administration, visualization, writing – original draft, and writing – review and editing; J.G.M., conceptualization, data curation, formal analysis, investigation, methodology, project administration, visualization, writing – original draft, and writing – review and editing; J.D., conceptualization, data curation, formal analysis, investigation, methodology, project administration, resources, visualization, writing – original draft, and writing – review and editing; S.B., resources and writing – review and editing; L.T.K., formal analysis, investigation, supervision, and writing – review and editing; M.B., investigation and writing – original draft; K.G., investigation and writing – original draft; A.K.-S., formal analysis; T.H.-L., formal analysis; L.K.S., visualization, writing – original draft, and writing – review and editing; K.H., resources, supervision, and writing – review and editing; M.H., writing – review and editing; C.P.S., resources, supervision, and writing – review and editing.

The authors declare no competing interests.

During the preparation of this work, the authors used ChatGPT (OpenAI) and Claude (Anthropic) in order to refine the language and improve clarity. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article.