Clonal haematopoiesis without identified genetic drivers: insights from analyses of 407,512 individuals

Clonal haematopoiesis (CH) becomes ubiquitous as humans age. The role of somatic driver mutations in its development has been studied widely, but little is known about CH without identified genetic drivers, also known as "CH with unknown drivers" (CH-UD). A fundamental unresolved question is whether CH-UD is driven by undiscovered somatic genetic drivers or by other cell-heritable traits. Here, to investigate this, we develop a new machine learning classifier to improve CH-UD detection from whole-genome sequencing data. After excluding 77,885 individuals with previously documented driver CH or mosaic chromosomal alterations (mCA), we applied our classifier to 407,512 UK Biobank participants and identified 26,963 (6.6%) with CH-UD. A genome-wide association study (GWAS) of common germline variants identified 31 polymorphic loci associated with predisposition to CH-UD. Of these, 25 were associated with other forms of CH at genome-wide significance. Linkage Disequilibrium Score Regression analyses revealed an unexpectedly high genetic correlation (rg=0.794) between CH-UD and non-DNMT3A driver CH, indicative of a remarkable overlap between the genetic aetiologies of the two phenomena. Analysis of 2,941 plasma protein measurements in 47,757 individuals revealed that TCL1A was the most significantly elevated plasma protein in CH-UD, mirroring the finding that the TCL1A locus was in the top two most significant associations of CH-UD GWAS and TET2-CH and ASXL1-CH GWAS, the two most common forms of non-DNMT3A-CH. Furthermore, TCL1A plasma levels rose steadily with age even in those without detectable CH, particularly among carriers of the common TCL1A risk variant (rs2887399-G), potentially via stochastic promoter demethylation as described in TET2-CH and ASXL1-CH. Phenome-wide association analysis of 13,225 binary and 1,682 quantitative traits revealed that, similarly to non-DNMT3A-CH, CH-UD was significantly associated with several malignant (haematological and solid organ) and non-malignant (including cardiovascular and renal) diseases. Our findings reveal striking genetic and phenotypic similarities between CH-UD and non-DNMT3A driver CH, including a strong dependence on TCL1A, a protein recently found to inhibit DNA methylation. Collectively, these observations propose that CH-UD develops through selection acting on ageing-associated epigenetic changes that mirror those of non-DNMT3A-CH, but without the need for somatic genetic drivers.