Somatic genetic variation shapes clonal selection of memory NK cells

Timo Rückert; Matthias Günther; Oliver Knight; M Mandry; Karl Junker; Simon A. Grassmann; Joseph Chai-Yuen Sun; Thomas Höfer; Chiara Romagnani

doi:10.64898/2026.09.17.752421

bioRxivSep 18, 2026ImmunologyPreprint

Somatic genetic variation shapes clonal selection of memory NK cells

Timo Rückert iD, Matthias Günther iD, Oliver Knight iD, M Mandry iD, Karl Junker, Simon A. Grassmann iD, Joseph Chai-Yuen Sun iD, Thomas Höfer iD, Chiara Romagnani iD

PDF DOI Code

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

Abstract

In adaptive immunity, antigen receptor specificity and affinity drive clonal selection, but what enables individual innate lymphocytes to gain a selective advantage without rearranged antigen receptors remains unknown. We traced somatic mutations in human natural killer (NK) cells, revealing extensive clonal dominance in cytomegalovirus-associated memory, with individual genetically diversified clones comprising up to 95% of the population, originating early in life and expanding over years. Their genetic evolution demonstrated positive selection of acquired coding mutations, linking somatic variation to NK cell clonal fitness. Among candidate drivers, we identified mutations in the chromatin regulator KMT2D, whose induced deletion in mature NK cells conferred a competitive advantage following infection in vivo. These findings establish somatic evolution as a fundamental mechanism shaping innate immune memory.

In-depth summary

Claude Sonnet 4.6 · xhigh effort · 8 Nov 2026 · from the full text of version 3 (4 Nov 2026) · CC BY 4.0

Whole-exome sequencing of memory NK cells from 81 CMV-seropositive donors reveals dominant somatic clones, early-life origin, and positive selection, with Kmt2d loss shifting mouse NK fitness.

Clonal selection is well described for T and B cells, where receptor affinity and precursor frequency matter, but it was unclear whether innate lymphocytes without rearranged receptors are selected. This work uses somatic mutations as lineage barcodes to quantify clonal dominance, timing and selection in human memory NK cells.

Key results

ShownMost analysed healthy CMV-seropositive donors carry a dominant memory NK-cell clone above 25% of the memory compartment, and these clones make up a sizable share of circulating lymphocytes.
49 of 76 individuals (64%) had a leading clone above 25% of the NKG2C+ memory compartment (largest 95%, median 39%). Clones reached up to 10% (median 0.9%) of lymphocytes (N=66). Mitochondrial clonotypes underestimated dominance. · Massive NK-cell clones dominate the memory…; Figure legends
ShownSingle-cell genotyping resolves dominant clones as branching families of related subclones, each with its own coherent receptor and chromatin profile.
Three donors had 2-5 clonal families each, comprising 2.3-63% of total NK cells, with up to five internal nodes supported by 11-16 variants. Leading-clone frequency agreed between single-cell and bulk sequencing. · Extensive branching in clonal phylogenies links…
SuggestedMost dominant clonal founder cells were estimated to arise early in life, mostly in the first decade, and clones grew slowly, with only a small share of divisions contributing to growth.
WGS in three donors (aged 28, 29, 49) put founder origins between 7 and 11 years. Nearly 70% of 49 WES clones arose within ten years, approximately 20% at 10 to 20. 60-100 SNVs per doubling vs about 1.5 per division gives one in 40-70 divisions self-renewing. · Memory NK cell clones emerge at a young age and…; Estimation of age-at-clonal-origin distribution
ShownUnder the assumed model parameters, neutral drift appears insufficient to explain the clonal architecture, and coding mutations show a significant positive-selection signal.
Neutral drift in a pool of 10^5 cells dividing once daily would need over a century. dN/dS 1.24 (95% CI 1.11-1.39) exome-wide and 1.98 (95% CI 1.09-3.61) in cancer genes; 1 in 5 (range 1 in 10 to 1 in 4) nonsynonymous mutations advantageous. · Somatic mutations in memory NK cells undergo…; Expected time to clonality under neutral drift
ShownLoss of Kmt2d, found as two somatic KMT2D variants in humans, changes mouse NK fitness in opposite directions depending on timing relative to infection.
Kmt2d-KO (N=9-10) and Het (N=5) NK cells gained advantage in steady-state chimeras but expanded less during MCMV (N=5-8). Inducible deletion at days 5-7 (iKO N=12-15; iHet N=5-10) gave an advantage from day 21, dominant by day 35. · Kmt2d loss affects NK-cell competitive fitness; Figure legends

Methods, data and reproducibility

Somatic variants from exome and genome sequencing of sorted NK subsets were combined with single-cell genotyping, SCIFER population-genetic inference, dN/dS selection tests, and mouse Kmt2d competition experiments.

Whole-exome sequencing of sorted NKG2C+ memory NK cells, paired conventional NK cells, and monocyte germline controls from healthy CMV-seropositive donors · human · n = 81 memory, 15 conventional
Targeted single-cell DNA mutation plus surface-protein profiling (Mission Bio Tapestri) of total NK cells, donors HC02, HC09, HC12 · human · n = 3 donors; 100 mutations per donor panel
Whole-genome sequencing of NK subsets (90x) with monocyte or buccal swab controls (30x), donors aged 28, 29, 49 · human · n = 3 donors
ASAP-seq with mtDNA recovery (new data) plus previously published scATAC-seq data · human · GSE197037
Mouse mixed bone marrow chimeras, adoptive transfers and tamoxifen-inducible Kmt2d deletion during MCMV infection · mouse · n = N=5-15 per group

Methods & tools: SCIFER (ABC-SMC, OligoCloneDynamics model), dndscv and genesetdnds, GATK Mutect2, Strelka2, Manta, DeepSomatic, Varlociraptor, SigProfilerExtractor, infSCITE, Seurat and Signac, mgatk, Mission Bio Tapestri pipeline and Mosaic, OptiType and kir-mapper, AlphaMissense, Ensembl VEP
Code: https://github.com/ollieeknight/imprint
Software & environment: Single Nextflow pipeline (imprint); fastp v1.3.5, bwa-mem3 v0.6.0, GATK v4.6.2.0, Seurat v5.4.0, Signac v1.16.0, SigProfilerExtractor v1.1.25, dndscv v0.0.1.0.
Reproducibility gaps: Data accessions and the custom-code link are placeholders until publication; raw sequencing data will be under controlled access via a Data Access Committee; the new SCIFER version is documented only in a manuscript in preparation.

Limitations (5)

With 81 donors, power to detect selection on individual genes was limited, and the aggregate dN/dS signal does not establish the contribution of any single variant. · stated by the authors
Mouse models cannot unequivocally reconstruct selection of KMT2D-mutated human clones, and the timing of human mutations relative to infection cannot be established retrospectively. · stated by the authors
SCIFER infers a single clone, so clones below 25% are often oligoclonal and were not treated as monoclonal. · stated by the authors
Single-cell, WGS, and clonal growth modelling rest on only three donors; clone age uses a naive T-cell mutation clock because NK-cell data were unavailable. · noted in this summary
Only two KMT2D variants were found, and the human variants were not functionally tested; mouse Kmt2d deletion is a surrogate for those specific mutations. · noted in this summary