Genetic variation across and within individuals.

Our understanding of the contribution of both somatic and germline mutations to human diseases has substantially progressed in recent years. This progress has been applied to several therapeutic treatments for cancer and to increasing prevention and treatment options for largely monogenic non-cancer diseases, such as sickle cell anaemia, for which gene therapies were recently approved by the FDA 146 . Continued progress, particularly towards precision prevention and treatment, will necessitate integrating high-depth multi-omics data, as well as information on social determinants of health, lifestyle factors, health status and the environment. Given the dynamic nature of somatic variants, longitudinal analyses will also yield a more nuanced understanding of mutation dynamics over time 108 , 147 . Additionally, diverse representation in human genetic studies examining both germline and somatic variants will be necessary for a comprehensive understanding of the mutational landscape and to ensure the research findings and their subsequent applications are equitable and beneficial across populations. Finally, studies of somatic variants in human tissues or cell types beyond blood are currently sparse, but they will yield important new observations, especially on patterns of somatic mosaicism in normal tissues and on the role of somatic mutations in human disease. Notably, the recently established NIH Somatic Mosaicism across Human Tissues (SMaHT) Consortium, which intends to comprehensively study all classes of somatic variants by short-read and long-read sequencing across human tissues, is a pivotal step in this direction 148 . Future directions for technical progress include leveraging methodological advancements from studies of germline variants towards a better understanding of somatic mosaicism. For example, studies of the causal relationship between somatic variants and diseases can extend beyond current mice experiments into human genetics by utilizing causal inference methods from epidemiological and statistical domains 149 . Additionally, in situ omics technologies remain underutilized outside of neurobiology 150 , 151 and could thus become invaluable tools. For example, omics technologies could help track the evolution of somatic variants within an individual and dissect the molecular consequences over time. Finally, comprehensive reference databases and catalogues need to be constructed to account for the accelerating pace of research on somatic mutations. Building a robust and accessible database would facilitate cross-disciplinary studies and accelerate the translation of somatic variation research into biological and clinical applications. As research expands to include more diverse and longitudinal human studies, as well as more single-cell multi-omics and causal inference methods, new insights will continue to emerge on the similarities, differences and interplay of germline and somatic variations. These insights can, in combination with large-scale collaborative efforts to facilitate translation, lead to deeper biological understanding, therapeutic innovations and clinical care applications.

Genetic variations between (that is, germline variants) and within (that is, somatic mutations) individuals underpin many phenotypic differences. Germline variants are inherited and templated in all cells of an individual ( Fig. 1a ), with evolutionary forces such as selection, recombination and drift shaping their frequency and distribution in a population over generations 1 . By contrast, somatic mutations accumulate in a mosaic manner within an individual from conception onward as a result of DNA damage or errors in DNA repair ( Fig. 1a ). Somatic mutations originate in a single cell and are propagated within an individual by DNA replication and cell division 2 – 6 ( Fig. 1b ). These fundamental differences between germline and somatic variants underlie their analogous modes of study either between individuals (germline) or between cells within an individual (somatic) ( Fig. 1c ). Both germline and somatic variants contribute to human disease. More specifically, germline variants underlie inherited genetic conditions, such as Huntington disease 7 , familial hypercholesterolaemia 8 and breast cancer predisposition syndromes 9 , 10 . Although most somatic mutations do not have a noticeable phenotypic effect, some can alter key cellular functions and potentially culminate in cancer 11 . Furthermore, somatic mutations in multiple normal tissues play profound roles in non-oncologic diseases (reviewed in ref. 12 ), such as somatic mutations that contribute to Alzheimer disease in neurons 13 . Moreover, somatic mutations that lead to clonal expansion in non-neoplastic blood cells are associated with a range of non-oncologic conditions, such as atherosclerosis and chronic liver disease 14 – 16 . Germline and somatic variants are inherently connected. All germline variants originate as de novo somatic mutations either in parental germ cells or very early in embryonic development. As such, mutations introduced and retained in germ cells, if passed on to the next generation, effectively become de novo germline variants. Additionally, germline variants, especially those in genes encoding DNA repair proteins, can influence somatic mutation rates and patterns 17 . This Review draws on insights gained from decades of DNA sequencing and more recent omics approaches ( Box 1 ) to delineate the fundamental features of germline and somatic variants. More specifically, we define how germline and somatic variants compare in their mutation rates, types and patterns, and we describe approaches to their detection and analytical considerations for genetic association studies. We also highlight instances of interplay between germline and somatic variants, including somatic reversal of germline variants and germline predisposition for somatic mutations. Finally, we underscore how the biological and clinical significance of germline and somatic variants will continue to be elucidated by diverse and longitudinal human studies, single-cell multi-omics and causal inference methods. Germline and somatic variants have many shared characteristics, given that germline variants originate as somatic mutations in germ cells. However, fundamental differences in their heritability, effect and prevalence can affect their respective mutational types, rates and patterns. Genetic variants can be divided into four main classes: substitutions, which are mainly comprised of single-nucleotide variants (SNVs); short (<50 bp) insertions and deletions (indels); structural variants, including large deletions, segmental duplications, inversions, translocations and transposable element insertions; and other large chromosomal abnormalities, including whole-chromosome losses and gains. Segmental duplications, large deletions and often whole-chromosome alterations are also referred to as copy number variations. These distinct classes are consequences of different mutagenic processes: SNVs and indels are a consequence of erroneous DNA damage repair or replication, whereas structural variants can result from errors introduced during DNA double-strand break repair, mitotic or meiotic recombination, chromosome lagging or chromosomal missegregation, the latter of which can also cause somatic aneuploidy. Notably, acquired or inherited deficiencies (through somatic or germline variants, respectively) in DNA repair pathway components or DNA polymerases can change the frequency and type of mutations 18 – 20 . Finally, the genome may be mutated by the insertion of mobile genetic elements, such as retrotransposons 21 . Different mutagenic processes produce distinct rates and patterns (known as mutational signatures) of somatic mutations. For example, mutational signatures specifically for SNVs typically refer to the distribution of base changes in specific trinucleotide contexts 22 , 23 , and similar signatures are defined for other classes of mutations, including indels 22 , chromosomal alterations 24 and structural variants 25 (see COSMIC database). Recent studies have shown that somatic mutation rates differ across tissues and age ranges in the same individual 26 – 36 ( Fig. 2a ). Notably, somatic mutation rates in different tissues are influenced by exposure to both cell-intrinsic and cell-extrinsic causes, such as smoking, ultraviolet light and chemical mutagens. Even when restricted to endogenous mutagenic sources, the mutation rate of SNVs greatly varies across cell types. For example, approximately 17 SNVs are acquired per year in neurons 30 and haematopoietic stem cells 32 , whereas 28 and 44 SNVs are acquired per year in endometrial stem cells 34 and colonic stem cells, respectively 31 . The pattern of somatic mutations, or mutational signature, also varies based on the source of mutations ( Fig. 2b ). Normal tissues vary in their mutational signatures, but all tissues have some trace of the linear, clock-like mutational signatures 27 referred to as single base substitution signature 1 (SBS1; C>T mutations in a CpG context) and SBS5 (a flat signature) in the COSMIC database 22 . In addition, many tissues also exhibit SBS18, which is characterized by C>A mutations and is a consequence of oxidative damage. Expectedly, some mutagens are restricted to certain organs; for example, SBS7, the consequence of ultraviolet light damage, is a common mutational signature in the skin, and it consists of C>T mutations. Normal tissues also show mutational patterns due to other exposures, such as smoking (SBS4, dominated by C>A mutations) 35 , chemotherapy (including SBS31, dominated by C>T mutations in CCC and CCT contexts; and SBS35, a collection of C>A, C>T and T>A mutations), and a genotoxic strain of Escherichia coli (SBS88; substitutions of T when preceded by A or T) 27 , 31 . The vast majority of somatic mutations will have no phenotypic effect on the cells that harbour them, especially those that affect non-coding regions of the genome or induce a synonymous change in genes. However, occasionally a cell will acquire a mutation that carries a selective advantage, such as increased proliferation or survival. These clonal expansions become more widespread with age (as mutations accumulate over time) and are enriched in rapidly dividing tissues in which clones are unconstrained, such as the sheet-like epithelia of the skin 37 , oesophagus 38 and bladder 29 , as well as blood 32 , 39 . Clonal expansions in the blood-forming system are referred to as clonal haematopoiesis. By contrast, large clonal expansions are rare in glandular epithelium, such as the colon, because of physical constraints imposed by the tissue architecture, barring a history of damage and regrowth, such as in inflammatory bowel disease or the normal endometrial epithelium. In many normal tissues, clones with somatic mutations that are under positive selection and have been previously associated with cancer may lead to malignant transformation 34 , 37 , 38 . However, some somatic mutations can induce clonal expansion without leading to cancer, such as those underlying clonal haematopoiesis that is linked to many non-cancer diseases 15 , 39 , 40 . Somatic mutations can even protect against cancer; NOTCH1 -mutant clones in the oesophageal epithelium are known to outcompete pre-cancer clones 41 , 42 . Alternatively, recurrent somatic mutations may mitigate the effects of a disease. Such mitigation was recently demonstrated in chronic liver diseases, in which different clones recurrently and independently acquired somatic mutations that lead to escaping disease-related toxicity 43 . Somatic mutations that accumulate in germ cells, when passed on to the next generation, become de novo germline variants. Therefore, the mutational rates and patterns specifically within these germ cells greatly influence the generation of germline variants. The seminiferous tubules of the testes, which produce the spermatocytes, accrue approximately 2.7 SNVs per year, the lowest observed mutation rate among the examined tissues 27 . Notably, this mutation rate is estimated for the diploid genome of spermatogonial stem cells rather than the haploid spermatocytes. Therefore, spermatocytes should accumulate somatic mutations at half the rate of spermatogonial stem cells. Indeed, the mutation rate in seminiferous tubules reflects the estimated paternal age effect on de novo germline variants, which is approximately 1.4 per year of the father’s age, in contrast to 0.37 per year of the mother’s age 44 – 46 . Although these tissues mostly exhibit age-related mutational signatures (SBS1 and SBS5), mutagenic exposures, such as chemotherapy treatment, can increase the mutation rates in germ cells 47 . The effects of somatic variants in germ cells can greatly influence the rate of transmission to the next generation. Some somatic variants are termed ‘selfish’ because they can subvert normal germline processes to increase the likelihood of propagating to the next generation, which often leads to major developmental disorders 48 , 49 . For example, some mutations in the FGFR2 gene are under positive selection in the seminiferous tubules of the father and lead to large clonal expansions; however, when passed on to the next generation, these FGFR2 mutations cause Crouzon syndrome, a genetic disorder characterized by the premature fusion of certain skull bones 50 . Alternatively, somatic variants in germ cells that are ultimately fertilized can result in embryonic lethality and therefore never manifest as germline variants. For example, most germline aneuploidies are lethal, with the notable exception of trisomy 21 51 . In addition, many variants classically associated with cancer, such as BRAF V600E, have never been reported as germline variants, suggesting that they are lethal despite assumedly driving increased propagation in germ cells 52 . In summary, the mutation rates of germ cells and the effects of these variants on embryogenesis are major determinants of de novo germline variants. Future work is needed to compare somatic variants in germ cells with de novo germline variants to understand the contribution of embryonic lethality to rates and patterns of germline variants. Germline variants are mostly static throughout an individual’s lifetime, but the pattern of these variants is dynamic across populations and generations. Evolutionary forces, such as selection and genetic drift, can shape germline variant patterns, leading to distinct genetic signatures among different populations and species. These genetic signatures can offer insights into evolutionary processes, adaptation and the genetic basis of heritable diseases. An illustrative case is the positive selection observed on variants in the G6PD gene, which confer a protective advantage against malaria and are more prevalent in regions where this disease is endemic 53 . Understanding the intricate landscape of germline and somatic variation depends on their accurate identification. Not only have advancements in detection technologies improved the accuracy of variant identification, but they have also enhanced our ability to delve into the genetic underpinnings of disease at a more granular level. Given that germline variants are present in most cells within an individual, the variants most challenging to detect are those in regions difficult to resolve through whole-genome sequencing, such as repeat-rich regions. However, recent technological advancements, such as long-read sequencing technologies and novel algorithms 54 , have improved overall germline variant detection ( Box 1 ). More specifically, long reads can cover repetitive genomic regions that are inaccessible to short reads 55 and deep learning models, such as DeepVariant, enhance variant detection by minimizing the dependence on arbitrary rules and filters 56 . Despite the remaining challenges of highly polymorphic and duplicate-rich regions of the genome that often complicate calling variants in clinically relevant genes (for example, PMS2 and SMN1 ), current germline variant calling achieves high accuracy, especially for SNVs (>99.9% in benchmark regions 57 ). The detection of somatic mutations requires additional considerations, because they are present in relatively few cells. Many insights into somatic mutations originally came from whole-genome sequencing of cancer tissue, which are large single-cell-derived clones. The genome sequence of the cancer cells is then compared to that of a large aggregate of normal cells (often blood) and any sequence differences are attributed to somatic mutations. However, somatic variants in normal tissues are not typically found in large clonal aggregates, so they tend to be rare and difficult to differentiate from errors introduced during library preparation and sequencing. This error rate depends on a variety of factors, including the chemistry and specific technology used for sequencing. Recent technological advancements have enabled researchers to surmount these obstacles and increase the detection sensitivity for somatic variants ( Table 1 ). The most straightforward method to improve the detection of somatic variants is to increase the sequencing depth. However, to avoid dramatically increasing sequencing costs, this approach is often paired with a ‘bait capture’ of genomic regions of interest, such as genes known to be under selection or often harbouring cancer-associated mutations. Bait capture relies on hybridization of oligonucleotide probes to enrich the DNA of the genomic regions of interest. Prior work using targeted sequencing experiments can reliably detect variants in ~0.1–1.0% of cells 37 , 38 , below which the mutation detection is constrained by the error rate of sequencing. Because of its sensitivity and depth, this approach excels at cell population-level inferences on the selection of mutations in specific genes, which is important to identify ‘driver’ mutations that confer a selective advantage to cells. However, given the limited genomic footprint, relatively few mutations are detected overall, which hampers the study of mutational burdens and signatures. Sequencing polyclonal populations of cells also precludes the precise reconstruction of phylogenetic trees, as it is impossible to prove that different somatic mutations co-occur in the same cells or occur in different cell populations. Of note, when the sample consists of a dominant clone with subclones, as is the case for cancers, a rudimentary phylogeny can be reconstructed through clustering mutations by their variant allele frequency into clones 58 . An alternative approach to increasing detection sensitivity of somatic variants is to sequence the DNA of a single cell to obtain mutational readouts of a single lineage. However, a single cell does not have enough DNA to for reliable whole-genome sequencing and somatic variant detection. Single-cell DNA can be amplified using one of three methods: first, directly isolating the DNA of a single cell and biochemically amplifying the DNA 59 ; second, isolating single cells and expanding into clones in vitro 32 , 60 ; third, isolating the clonal progeny of a single cell in vivo, often using laser capture microdissection 27 , 31 , 36 , 61 . However, each of these approaches has downsides. Amplifying the DNA of a single cell can introduce artefactual variants and can exclude genomic regions that are difficult to amplify 59 . Cloning cells in vitro is laborious and works better for stem cells than differentiated cell types. The microdissection approach relies on the presence of typically rare clonal units in a tissue, but it yields high-quality genomic readouts and retains spatial information on tissue architecture 27 , 31 , 36 , 61 . Obtaining single-cell-derived readouts is crucial for lineage tracing and reconstructing the mutation-based phylogenies of normal cells. Prior work has minimized the error rate of library preparation and sequencing by sequencing both the forward and reverse strands of a DNA duplex molecule 30 , 62 . A true somatic variant will be observed in both strands, whereas variants introduced during library preparation and sequencing will only appear in one of the two strands. This approach quadratically lowers the probability that a variant is observed erroneously and can lower the error rate to approximately 10 −8 per base from the usual 10 −4 per base on commonly used short read sequencing platforms. Notably, this increased sensitivity only applies to the detection of SNVs and indels. By achieving a lower error rate, duplex sequencing provides better estimations of average mutational burdens and signatures in cell populations. Finally, the bioinformatic approach to variant calling from aligned sequence data must be considered. Many different variant callers, including GATK MuTect2 (refs. 63 , 64 ), VarScan2 (ref. 65 ) and CaVEMan 66 , are used to detect somatic variants. To distinguish between germline and somatic variants, the sample may be compared with another normal tissue sample, and any common variants are assumed to be from the germline 34 . Alternatively, unmatched calling approaches avoid the risk of filtering out shared variants that arose during very early embryogenesis, in which case the variant allele frequency across tissues can be used to distinguish between germline and somatic variants 2 ( Fig. 1b ). Furthermore, some variant callers, such as MoChA 40 and DeepMosaic 67 , are specifically designed for mosaic variant calling by explicitly modelling or training on data of mosaic mutations. Generally, variant callers filter out recurrent artefacts by utilizing a panel of unmatched normal samples, ideally subjected to the exact same library preparation and sequencing protocol as the sample of interest 63 , 64 . Genetic association analyses play a crucial role in uncovering the relationships between genetic variation and phenotypic traits, both in the context of germline and somatic variants. Germline analyses typically involve large-scale population studies to identify variants associated with specific traits or diseases. Conversely, many somatic mutation studies are focused on mutations within individual cells or tissues, often utilizing single-cell sequencing to discern their association with phenotypic changes. Multiple advancements have drastically increased the breadth and depth of association analyses of germline variants ( Fig. 3 ). First, studies have transitioned from single-population investigations to well-powered studies of mega-biobanks or multiple diverse cohorts due in part to sequencing cost reductions 68 ( Fig. 3a ). Second, high-throughput sequencing technologies and bioinformatics tools 69 , 70 have facilitated a shift from focusing solely on common coding variants to exploring the complex polygenic architecture across diverse populations, as well as rare coding, non-coding and structural variants 71 – 78 ( Fig. 3b ). Third, the functional effects of germline variants with phenotypic associations are now explored through the integration of techniques such as single-cell sequencing, CRISPR, Hi-C, assay for transposase-accessible chromatin with sequencing (ATAC-seq), chromatin immunoprecipitation followed by sequencing (ChIP–seq) and spatial molecular profiling 79 – 86 ( Fig. 3c and Box 1 ). These innovative integrations have helped dissect regulatory and disease pathways acting at cell-specific or tissue-specific levels 87 – 90 . Fourth, further integration of these new types of data with enhanced statistical and machine learning methods has enabled researchers to identify disease causal variants and target genes 91 – 93 , infer cell developmental trajectories 94 and predict gene expression accurately 95 . The recently introduced STAAR (variant-set test for association using annotation information) series exemplifies these advancements by developing novel statistical methods for aggregated rare variant association tests. These methods incorporate multiple functional annotations and are scalable to large whole-genome sequencing datasets, as demonstrated using multi-ancestry whole-genome sequencing data from the TOPMed program 75 , 78 , 96 As germline variant research evolves, conventional analytical considerations, such as statistical power, remain pertinent. Strategies to enhance power include increasing sample size and diversity of ancestries, utilizing more accurate models (such as those that account for gene–gene or gene–environment interactions), and aggregating genomic information (such as polygenic risk scores) 97 – 101 . Innovations accounting for linkage disequilibrium, population structure using principal component analysis, and relatedness using linear mixed models have more recently focused on scalability, given ongoing development and expansion of large-scale biobanks. These analytical considerations have been extensively addressed in the literature 102 – 104 . Association analyses of somatic variants differ from those of germline variants in several ways ( Fig. 4 ). For somatic mutation studies, the sample collection required to identify mutations across tissues is inherently invasive, and thus tissue-comprehensive studies are often limited in sample size. To date, datasets with a wider range of tissues have typically been procured from a limited number of deceased donors 89 . For example, recent efforts to analyse somatic mosaicism across developmental stages have used tissue samples from deceased paediatric donors 105 . Therefore, given that many tissue types from living research participants are inaccessible, studies of population-based somatic mosaicism at scales comparable to germline mutation studies have focused mainly on blood DNA in adults. Given the acquired and dynamic nature of somatic mutations, association analyses for somatic mutations need to also account for potential exposures that may jointly or separately influence the acquisition, fitness and clinical outcome of a somatic variant ( Fig. 4a ). For example, associations between somatic mutations and diseases may be confounded by factors such as smoking 35 or exhibit potential bi-directional causal relations (that is, each of two traits may be causal to the other at the same time) 14 , 106 , 107 . Given that somatic variants occur only in a subset of cells, association analyses of somatic variants often require tissue-specific and single-cell analyses ( Fig. 4b ). These analyses can identify rare subclones, track clonal evolution and investigate cellular phenotypes associated with specific variants 32 , 60 , 108 . Furthermore, recent advancements have surmounted a limitation of conventional single-cell technologies, so that cells are no longer destroyed in the process of sequencing. More specifically, DNA can be sequenced while simultaneously measuring other ‘omics’ phenotypes at the single-cell level, enriching our insights into the mechanisms of somatic variants 109 . Additionally, algorithmic innovations can now directly detect somatic variants in single-cell RNA sequencing (RNA-seq) and ATAC-seq reads without the need for matched DNA sequencing data, which allows repurposing many previously collected single-cell data sets that went through RNA-seq and ATAC-seq for somatic mutation calling 110 . These developments enable the capture and integration of multiple data modalities to inform how somatic mutations affect cellular function and regulation 111 , 112 . In addition, different driver genes in somatic mutations have diverse biological and clinical consequences and are typically better considered separately rather than as one. For example, biologically, two commonly mutated genes in clonal haematopoiesis, DNMT3A and TET2 , are involved in methylation and demethylation, respectively 113 , 114 . Clinically, TET2 mutations cause atherosclerosis in both animal experiments and human studies 14 , 107 whereas the role of DNMT3A mutations in atherosclerosis is less clear 115 ( Fig. 4c ). Genetic variation across and within individuals is a dynamic mosaic of germline and somatic variants that can influence one another to shape health and disease trajectories ( Fig. 5 ). Their interplay has important clinical implications through contributions to the pathogenesis of various diseases, but it also has the potential to provide more effective, personalized therapeutic strategies. Somatic variants can partially or fully reverse the pathogenic effects of inherited germline variants, a phenomenon known as somatic genetic rescue (SGR; Fig. 5a ). SGR has now been reported in over 30 different haematopoietic disorders and several other diseases, such as breast cancer, which is caused by autosomal recessive, autosomal dominant and X-linked mutations 116 – 118 . The genetic mechanisms underlying SGR are diverse, including site-specific mutations that revert the original germline variant to the ‘wild type’ (that is, its common form found in the general population that is typically associated with the absence of the disease phenotype), second-site mutations that compensate for the germline defect, copy-neutral loss of heterozygosity and chromosomal deletions or rearrangements. Clinically, SGR can lead to milder disease phenotypes and delayed diagnoses, but it can also have neutral or negative effects 117 . A classic example with therapeutic relevance is germline variants in BRCA1 or BRCA2 that are related to cancer. Individuals with such variants might be initially responsive to poly (ADP-ribose) polymerase (PARP) inhibitors due to DNA repair defects, but they then become resistant to these inhibitors following additional mutations that restore some DNA repair function 119 , 120 . Looking ahead, CRISPR–Cas9 gene editing has been proposed as a method to controllably induce SGR for the targeted treatment of inherited monogenic disorders, such as Duchenne muscular dystrophy and cystic fibrosis 121 . In summary, understanding the genetic mechanisms and clinical effects of SGR has important diagnostic and therapeutic value. Prior work has identified two mechanisms by which germline variants can influence the risks of developing somatic mutations. First, some germline variants can increase the baseline rate of somatic mutations through, for example, inherited defects in DNA repair pathways. This mechanism is responsible for Bloom syndrome and Fanconi anaemia pathways, which are diseases associated with genomic instability and a higher likelihood of somatic mutation 122 , 123 . Second, individuals might develop malignancies following somatic mutations because of certain germline variants that predispose an individual to the clonal expansion of these cells. The occurrence of this phenomenon is evidenced by the increased risk of cancer development among first-degree relatives of cancer patients 9 , 10 , 124 . Another example of this phenomenon is that families with long telomere syndrome from POT1 mutations also have increased familial risk of clonal haematopoiesis 125 , 126 . Further efforts to identify such germline variants have been enabled in recent years by larger biobanks and patient cohorts. However, most population-level studies have focused solely on somatic variants in the blood system, clonal haematopoiesis, due to the difficulty and cost of collecting non-blood tissue samples 17 , 127 . These population-level studies revealed heterogeneous germline genetic basis across different types of clonal haematopoiesis. For instance, genome-wide association studies on clonal haematopoiesis of indeterminate potential have identified over 20 loci near genes involved in haematopoietic stem cell self-renewal, proliferation, telomere maintenance and DNA damage response pathways 128 – 130 ( Fig. 5b ). By contrast, germline variants influencing mosaic loss of the X chromosome within blood cells, another type of clonal haematopoiesis, are primarily linked to genes with established roles in chromosomal missegregation, cancer predisposition and autoimmune diseases 131 . In addition, even within one type of clonal haematopoiesis, somatic mutations at different driver genes can have different germline genetic underpinnings. A relevant example is a germline locus on TCL1A where alleles associated with an increased risk of developing DNMT3A -mutant clonal haematopoiesis of indeterminate potential are also associated with a decreased risk of developing TET2- mutant clonal haematopoiesis of indeterminate potential 129 , 132 . Germline variants, especially those regulating inflammatory pathways, can modify disease risks or treatment effects associated with somatic mutations. For example, mice with Tet2 -mutant clonal haematopoiesis develop larger atherosclerotic burdens than mice under normal conditions, and prior work demonstrated marked blunting of Tet2 -mutant clonal haematopoiesis’s atherogenic effect upon chemically abrogating interleukin (IL)-1B secretion 107 . This finding led subsequent work to test and confirm in humans that IL-6 pathway inhibition (a downstream event of IL-1B secretion inhibition), proxied by an IL6R -disruptive coding mutation, substantially modifies the clonal haematopoiesis-associated cardiovascular disease risk 133 . Another study extended these findings to another driver mutation of clonal haematopoiesis, Jak2 V617F . The authors demonstrated that atherogenic mice with Jak2 V617F clonal haematopoiesis had increased atherosclerosis, which was then reduced in the presence of Aim2 deficiency (induced through Aim2 knockout bone marrow transplantation) 134 . More recently, additional human genetics findings have validated the JAK2 – AIM2 interaction in humans and shown that germline genetically determined expression levels of several other genes can selectively modify the associations between specific driver genes of clonal haematopoiesis and cardiovascular disease risk 135 ( Fig. 5c ). Many cancers, spanning from haematological malignancies to solid tumours, owe their pathogenesis to the complex interplay between germline and somatic variants. Although cancer typically results from the accumulation of somatic variants, germline variants can predispose individuals to developing cancer, both directly, such as the case of BRCA1 and BRCA2 genes increasing the risk of breast cancer 10 , 136 , and indirectly, by increasing the risk of developing somatic mutations, as discussed above. A recent example includes germline variants in TP53 , which can cause Li–Fraumeni syndrome, a rare genetic disorder that predisposes individuals to multiple cancers 137 . A recent study across 14 cancers identified a highly polygenic architecture, involving germline variants at thousands of loci, and suggested that polygenic risk prediction has potential for patient stratification 138 . Numerous non-cancer diseases, ranging from developmental diseases in early life to geriatric conditions, are shaped by the combination of somatic and germline variants. For instance, endometriosis, a condition exhibiting nearly 50% heritability based on family studies 139 , is also associated with somatic mutations in ARID1A , PIK3CA and KRAS 140 ( Fig. 5d ). Additionally, germline variants in STAT3 have been found in rheumatoid arthritis cases, and somatic variants in STAT3 are common in T cells from patients with Felty’s syndrome, a complication of rheumatoid arthritis 141 . Furthermore, rare genetic disorders such as autoimmune lymphoproliferative syndrome are associated with both germline and somatic mutations in FAS 142 . The observation that clonal haematopoiesis is linked to atherosclerosis 14 , 107 , 133 subsequently led to clonal haematopoiesis being associated with many other non-cancer, heritable conditions, such as chronic liver diseases and neurodegenerative disorders 15 , 40 , 143 , 144 . Finally, a recent study found that recurrent non-missense somatic mutations in blood cells are individually not oncogenic, but they are associated with blood cell traits, such as altered monocyte counts comparable to those of Mendelian variants in RASGRP1 and ELANE , that cause severe congenital neutropenia 145 . These mutations are not readily explained by other clonal phenomena and seem to have a germline genetic basis related to adaptive immune function, pro-inflammatory cytokine production and lymphoid lineage, highlighting the complex interplay between germline and somatic variation patterns and disease risk 57 .