To appreciate why this matters, it helps to understand what EED actually does. The gene encodes a core subunit of the Polycomb Repressive Complex 2, or PRC2, one of the most important epigenetic machines in human cells. Alongside its partners EZH2, SUZ12, and the retinoblastoma binding proteins RBBP4 and RBBP7, EED helps silence genes by recognizing and propagating a chemical mark on histones, the protein spools around which DNA is wound. Specifically, the complex deposits and maintains tri-methylation of histone H3 at lysine 27, a modification that compacts chromatin and shuts down transcription. EED plays the crucial role of reading this mark and extending it along the chromosome, a mechanism essential for regulating developmental genes and preserving cellular identity. When one copy of EED is disabled, the fine-tuning of gene expression during growth and brain development goes awry, producing the overgrowth syndromes that have fascinated geneticists for the past decade.

The clinical family of PRC2-related overgrowth disorders is small but distinctive. Heterozygous germline variants in EZH2 cause Weaver syndrome, marked by tall stature, macrocephaly, advanced bone age, and characteristic facial features. Variants in EED produce Cohen–Gibson syndrome, while alterations in SUZ12 underlie Imagawa–Matsumoto syndrome. All three overlap substantially with one another and with Sotos syndrome, featuring prenatal and postnatal overgrowth, hypotonia, and intellectual disability of variable severity. The facial gestalt of EED-related cases can usually be distinguished from Weaver syndrome, particularly in older children, by features such as telecanthus, narrow palpebral fissures, ptosis, downslanted upper eyelids, and eversion of the lower lateral eyelid margins. Some individuals also present with congenital heart defects, genitourinary anomalies, structural brain abnormalities, and epilepsy, reflecting the variable expressivity of these epigenetic disorders.

The girl at the center of the new report was born at 36 weeks and three days of gestation by cesarean section, prompted by maternal edema and hypertension. From the very first moments of life, the signs of overgrowth were unmistakable: she weighed 4500 grams, more than three and three-quarter standard deviations above the mean, measured 55 centimeters in length, and had a head circumference of 38 centimeters. Postnatal hypotonia was noted immediately. Developmental milestones arrived late, with unsupported walking at around 24 months and first words at about 30 months. Today she attends a school for children with intellectual disabilities and receives a broad program of physiotherapy, occupational therapy, and speech therapy. At age 12, her height measured 173.5 centimeters, nearly three standard deviations above average, and facial examination revealed hypertelorism, narrow palpebral fissures, periorbital fullness, and large, posteriorly rotated ears, alongside genu valgum, thin calves, and slender fingers with short first and fifth metacarpals.

Notably, her bone age at 10 years and 4 months matched her chronological age, but projected adult height, calculated by the Bayley and Pinneau method, reached an estimated 192 centimeters, dramatically exceeding the mid-parental target height of 163 centimeters. Endocrinological workup was unremarkable, chromosomal analysis showed a normal 46,XX karyotype, and the family history across three generations was entirely uneventful, with no other affected individuals. Intriguingly, two artificial intelligence tools designed to recognize rare disease from facial photographs, GestaltMatcher and Face2Gene, both flagged Cohen–Gibson syndrome among their top suggestions, with Face2Gene assigning it to the ultra-rare category at 76 percent gestalt similarity. That computational nudge helped steer the diagnostic team toward the right corner of the genome.

Whole-exome sequencing of blood-derived DNA then delivered the decisive clue: a missense variant in exon 10 of EED, designated c.1091G>T, which substitutes leucine for the conserved tryptophan at position 364 of the protein. Crucially, the variant was present at a variant allele fraction of only about 28 percent, far below the 50 percent expected for a heterozygous germline change. Targeted deep sequencing confirmed a fraction of 31 percent in blood, 30 percent in saliva, and between 11 and 17 percent in two independently collected buccal mucosa samples. Neither parent carried the variant above the assay’s validated detection threshold of roughly 1 percent, and the trace reads observed in their samples were consistent with sequencing artifacts. The tissue distribution, spanning both mesoderm-derived blood and ectoderm-derived buccal mucosa, pointed to a postzygotic mutation arising early in embryogenesis, after fertilization but before the embryo’s cells had fully committed to their developmental fates.

The molecular properties of the variant strengthen the case for its pathogenicity. The altered tryptophan sits within the WD40 repeat domain of EED and forms part of the aromatic cage that binds trimethylated H3K27, the very interaction that allosterically activates PRC2 and allows the complex to propagate its repressive marks. A single nucleotide change at this residue is extraordinarily rare: population databases record it exactly once among nearly 1.6 million alleles, and it appears in neither ClinVar nor the Human Gene Mutation Database. A battery of computational predictors, including AlphaMissense, CADD, PolyPhen, REVEL, PhyloP, PrimateAI-3D, and SIFT, uniformly predicted a deleterious effect. Under American College of Medical Genetics and Genomics guidelines, the variant was classified as likely pathogenic on the strength of its confirmed de novo status, its extreme rarity, the computational evidence, and the highly specific match between the patient’s phenotype and known EED-related disease.

Perhaps the most elegant piece of evidence came not from DNA sequencing at all but from the epigenome. Genome-wide DNA methylation profiling on the Illumina EPIC BeadChip platform, analyzed with the EpiSign METRIC software, revealed a methylation pattern that matched the established episignature of PRC2 complex disorders, which encompasses both EED-associated syndrome and EZH2-associated Weaver syndrome. In hierarchical clustering heatmaps, multidimensional scaling plots, and t-distributed stochastic neighbor embedding visualizations, the patient’s sample clustered tightly with the PRC2 reference cohort and separated cleanly from controls. In other words, the mutation is not merely present; it is functionally active, leaving a measurable genome-wide fingerprint of dysregulated gene silencing. This convergence of sequencing, structural biology, and epigenetic profiling illustrates how modern molecular diagnostics can triangulate a diagnosis from multiple independent lines of evidence.

The discovery also reshapes thinking about mosaicism in overgrowth syndromes more broadly. Postzygotic mosaicism is a well-established mechanism in disorders driven by somatic activating variants in genes such as PIK3CA and AKT1, which cause conditions like CLOVES syndrome and Proteus syndrome. But for PRC2-related overgrowth, mosaicism had never been observed, and variant databases list only germline alterations. The new case shows that intermediate-level mosaicism, with variant allele fractions between 11 and 31 percent across tissues, is sufficient to produce a multisystemic phenotype of moderate severity, including cardiac valvular insufficiency, musculoskeletal anomalies, and increased patellar reflexes. It also carries a practical lesson: had the diagnostic workup relied on blood alone, or on a single tissue, the variant might have been missed or its significance underestimated. Multi-tissue sequencing, the authors argue, should be considered whenever clinical suspicion remains high despite negative or ambiguous initial testing.

Limitations remain, as the researchers themselves acknowledge. This is a single-case observation, and no functional assays were performed to directly demonstrate how the variant perturbs PRC2 activity in cells. Whether the degree of mosaicism correlates with phenotypic severity is still an open question, and systematic studies in larger cohorts will be needed to establish standardized approaches for assessing the clinical relevance of mosaic PRC2 variants. Nevertheless, the message for clinicians is clear: mosaicism deserves a routine place in the differential diagnosis of PRC2-related syndromes, particularly in patients with unexplained or atypical overgrowth and developmental delay. For the families of children who have endured years of inconclusive genetic testing, the recognition that a postzygotic mutation can masquerade as, or produce, Cohen–Gibson syndrome offers a new and potentially decisive avenue toward the answer they have been searching for.

Subject of Research: Mosaic EED variant causing Cohen–Gibson syndrome, a PRC2-related overgrowth and neurodevelopmental disorder

Article Title: Mosaicism for Cohen–Gibson Syndrome in a Female Individual With Overgrowth and Neurodevelopmental Delay

Article References: Song, F., Michels, M., Betz, B., Vorhoff, W., Neuhann, T., Senninger, L., & Wieczorek, D. (2026). Mosaicism for Cohen–Gibson Syndrome in a Female Individual With Overgrowth and Neurodevelopmental Delay. Molecular Genetics & Genomic Medicine, 14(10), Article e70286. https://doi.org/10.1002/mgg3.70286

Image Credits: AI Generated

DOI: 10.1002/mgg3.70286

Keywords: Cohen-Gibson syndrome, EED, PRC2, mosaicism, overgrowth syndrome, Weaver syndrome, epigenetics, DNA methylation, episignature, whole-exome sequencing, neurodevelopmental delay, rare disease

The significance of that result becomes clear when it is set against the track record of double-strand break editing in human embryos. Previous work had shown that Cas9-induced breaks are genotoxic in this setting, producing frequent aneuploidy, a condition in which cells end up with the wrong number of chromosomes, as well as large deletions that scramble the very region the edit was meant to correct. The reason lies in how embryos handle broken DNA. Early human embryos rely on relatively inefficient and error-prone repair pathways, so a deliberate double-strand break can cascade into chromosome loss, mosaicism in which different cells carry different genomes, and developmental arrest. By contrast, the lesions created by base editors, single-strand nicks and mismatched base pairs, appear to be handled far more gracefully by the embryo’s repair machinery, and the new study provides the most direct evidence yet that this is the case.

The base editor at the heart of the experiment, known as ABE8e-V106W, belongs to the adenine base editor family. These enzymes fuse a Cas9 variant that nicks only one DNA strand to an engineered deaminase enzyme that chemically converts adenine into inosine, a nucleotide that the cell’s replication and repair machinery reads as guanine. The result is a targeted A-to-G transition, or, on the opposite strand, a T-to-C change. The V106W mutation in the deaminase domain is a precision refinement designed to constrain the enzyme’s activity window and reduce collateral edits at nearby positions. In the study, the team targeted the editor at the PCSK9 locus, a gene whose loss-of-function variants are known to lower LDL cholesterol dramatically and protect against cardiovascular disease, and at the HBG locus, a fetal hemoglobin gene relevant to blood disorders.

Delivery was a critical design choice. The researchers introduced the editor as a purified protein together with its guide RNA at fertilization, rather than as messenger RNA or DNA that would force the embryo to manufacture the enzyme itself. Protein delivery has a built-in advantage: the enzyme is present immediately but is also degraded quickly, limiting the window during which off-target editing can occur. With this approach, editing was achieved at all PCSK9 alleles, meaning both the maternal and paternal copies of the gene were converted in the edited embryos. The team then derived homozygous edited stem cell lines from the resulting blastocysts, confirming that the edit had been installed uniformly and could be propagated through cell division. Notably, sequencing detected no insertions or deletions at the target site, the class of collateral damage that most commonly accompanies nuclease-based editing.

Yet the study is emphatically not a green light for editing human reproduction. Although the overall picture was far cleaner than with Cas9 breaks, the researchers documented rare on-target chromosome breakage and other chromosomal abnormalities among the treated embryos. These events were infrequent, but in the context of reproductive medicine even a low rate of chromosomal damage is a serious concern, because an embryo carrying such an abnormality could, in principle, develop into a pregnancy with severe consequences. The findings underscore a recurring theme in genome editing: eliminating one failure mode does not eliminate risk, it merely shifts the risk profile, and every new editing modality must be evaluated on its own terms in the most sensitive biological context available.

Two further problems emerged that complicate the picture. First, editing at bystander sites, positions within the enzyme’s activity window near the intended target, and at off-target sites elsewhere in the genome was mosaic, meaning that only a subset of cells in a given embryo carried the unintended edits. Mosaicism is particularly troublesome because it makes it difficult to verify the final genetic makeup of an embryo from a single biopsy, and it means that some cell lineages could carry edits that others do not. Second, and more dramatically, when the team delivered the editor as mRNA instead of protein, the embryos frequently arrested in development. The culprit was guide-independent deaminase activity: the deaminase portion of the editor, once produced in abundance inside the embryo, chemically modified adenines in RNA and DNA without needing the guide RNA to direct it, and this untargeted activity proved toxic to early development.

That observation carries a broader lesson for the base-editing field. Much of the enthusiasm for adenine base editors has come from work in cell lines and animal models, where high expression of the editor is generally a virtue, driving editing efficiency upward. In a human zygote, however, the calculus inverts. A transient burst of enzyme activity, tightly constrained in time and directed by the guide RNA, is tolerable; a sustained flood of deaminase that wanders off its intended target is not. The contrast between the protein-delivered and mRNA-delivered experiments in this study provides an unusually clean demonstration that the delivery format, not just the editor itself, determines whether editing is compatible with normal embryonic development.

The choice of PCSK9 as the target gene is also worth unpacking. Loss-of-function mutations in PCSK9 are among the most celebrated protective variants in human genetics: people born with them have low LDL cholesterol and sharply reduced rates of heart disease, and this biology has already inspired a generation of cholesterol-lowering drugs, including PCSK9-targeting antibodies and, more recently, CRISPR-based therapies in adults. Editing PCSK9 in an embryo would, in theory, confer lifelong protection from birth. The HBG target, meanwhile, speaks to the reactivation of fetal hemoglobin as a strategy for sickle cell disease and beta-thalassemia. By demonstrating efficient editing at both loci, the study connects embryo-level genome engineering to two of the most clinically mature gene-editing applications in medicine today.

The authors are explicit in their conclusion that undesirable consequences for the genome and development currently preclude clinical use of this technology in reproduction. That caution matters because the regulatory and ethical landscape surrounding human embryo editing remains fraught, particularly after the 2018 case in which embryos were edited with Cas9 and led to the birth of children in China, an experiment widely condemned for its safety failures and ethical lapses. The new work does not change the ethical calculus, but it does change the technical one. It shows that the genotoxicity that made nuclease editing in embryos so dangerous is not an inherent property of all genome editing, but a consequence of how double-strand breaks are repaired in early development. In that sense, the study reframes the debate: the question is no longer only whether embryo editing should be done, but which molecular strategies, if any, could ever make it safe enough to consider.

What comes next will likely focus on engineering away the remaining hazards. Reducing guide-independent deaminase activity, narrowing the editing window to eliminate bystander edits, improving the specificity of guide RNAs to suppress off-target activity, and refining delivery to minimize chromosomal damage are all active areas of enzyme engineering, and the detailed datasets accompanying this study, including genome-wide off-target maps and single-embryo sequencing analyses, provide a benchmark against which future editor versions can be measured. The derivation of homozygous edited stem cell lines also offers a renewable experimental platform for studying the long-term consequences of precise edits in a human genomic context. For now, the study stands as a technical landmark and a sober warning at once: base editing can rewrite the human embryonic genome with remarkable efficiency and without the catastrophic breaks of earlier methods, but the genome of a human embryo remains an unforgiving place to work, and no current technology is ready to touch it in the clinic.

Subject of Research: Adenine base editing of the PCSK9 gene in human embryos and its effects on genome integrity and development

Article Title: Highly efficient base editing at PCSK9 and normal human embryo development

Article References: Jerabek, S., Jung, C., Kappy, M., Zhao, Q., Sung, J., Wang, N., Kim, E., Kim, J., Kulmann, M. I. R., King, M. B., McAndrew, M. J., Li, M., Bhatele, S., Isado, M., Jang, H.-S., Dolezal, M., Prosser, R., Xu, S., Hwang, G.-H., … Egli, D. (2026). Highly efficient base editing at PCSK9 and normal human embryo development. Nature. https://doi.org/10.1038/s41586-026-11118-x

Image Credits: AI Generated

DOI: 10.1038/s41586-026-11118-x

Keywords: base editing, human embryos, PCSK9, ABE8e, genome editing, aneuploidy, mosaicism, blastocyst, embryonic stem cells, chromosomal abnormalities, gene therapy, Columbia University