Somatic mutations in aging and disease

Abstract  

Time always leaves its mark, and our genome is no exception. Mutations in the genome of somatic cells were first hypothesized to be the cause of aging in the 1950s, shortly after the molecular structure of DNA had been described. Somatic mutation theories of aging are based on the fact that mutations in DNA as the ultimate template for all cellular functions are irreversible. However, it took until the 1990s to develop the methods to test if DNA mutations accumulate with age in different organs and tissues and estimate the severity of the problem. By now, numerous studies have documented the accumulation of somatic mutations with age in normal cells and tissues of mice, humans, and other animals, showing clock-like mutational signatures that provide information on the underlying causes of the mutations. In this review, we will first briefly discuss the recent advances in next-generation sequencing that now allow quantitative analysis of somatic mutations. Second, we will provide evidence that the mutation rate differs between cell types, with a focus on differences between germline and somatic mutation rate. Third, we will discuss somatic mutational signatures as measures of aging, environmental exposure, and activities of DNA repair processes. Fourth, we will explain the concept of clonally amplified somatic mutations, with a focus on clonal hematopoiesis. Fifth, we will briefly discuss somatic mutations in the transcriptome and in our other genome, i.e., the genome of mitochondria. We will end with a brief discussion of a possible causal contribution of somatic mutations to the aging process.

Introduction

As the carrier of hereditary information, which ultimately dictates the function of all cells and tissues, somatic nuclear genomes need to be preserved over the lifetime of an organism. Early in the 1950s, visionary physicists and biologists proposed that the rate of aging is determined by the rate of spontaneous mutation accumulation in the somatic genome [1, 2]. Despite the absence of any information about how mutations in DNA arise and how genomes function, these early hypotheses are still relevant. Indeed, in spite of the enormous improvements in our knowledge of the genome and major advances in studying its DNA sequences and the RNA and proteins derived from it, these original assumptions may still proof accurate and at least some of them are now being gradually confirmed by experimental studies made possible by the technological advances we will discuss below.

The early somatic mutation theories of aging were suffering from the scarcity of technology capable of measuring somatic mutations in tissues and organs of organisms. For example, Leo Szilard’s model of somatic mutation accumulation assumed that an “aging hit” of a chromosome would inactivate all genes carried by that chromosome [1]. Now we know that while mutations can indeed involve loss or gain of whole chromosomes (aneuploidy), virtually all of them involve only one base pair, i.e., base substitutions and one base pair insertions or deletions. Nevertheless, the Szilard paper was groundbreaking as the first to provide a quantitative model that explains the demographic trends of aging through the lens of genetic damage. Indeed, in light of the new findings, various aspects of Szilard’s model still hold. Most importantly, Szilard’s model takes into account the nonlinear pattern of human mortality through a mechanism of genetic and cellular redundancy which kept mortality low until the redundancy was exhausted, at which point mortality rapidly rose. He postulated that the aging hit rate is fairly stable among individuals within species but can be variable among species and that the surviving fraction of functional somatic cells determines the biological age of an individual [3, 4]. In other words, it was hypothesized that mutation accumulation could serve as a biomarker of biological age.

While not proposing a similarly elaborate model of aging as Szilard, Gioacchino Failla’s inclusion of cancer as a disease of old age, the risk of which increases exponentially with age [5, 6], was also novel [2]. Indeed, while Theodor Boveri had already proposed that chromosomal abnormalities promote cancer [7], it took until the 1970s until a generally acceptable model of cancer as a disease caused by mutations was put forward [8].

Here, we will briefly review the latest knowledge about the possible causal role of somatic mutations in aging and age-related diseases, how mutations arise and can be studied, how the somatic mutation rate differs from the germline mutation rate, how we can use mutational signatures as markers for aging and environmental exposure, and the evidence for and against a causal contribution of somatic mutations to aging.

How to measure somatic mutations in non-clonal, normal tissues

Each cell in a population carries a unique set of somatic mutations different from that in all other cells. This is due to the essentially random occurrence of mutations and the large size of the genome, i.e., 6 × 10⁹ for a diploid human genome. While there are certain preferences, for example, related to the source of the somatic mutation, e.g., C > A changes due to oxidative damage, and hotspot regions [9], somatic mutations overall occur randomly. When the same mutation is found in more than one cell in a tissue, this is usually a sign of clonal amplification, i.e., when a random mutation provides a survival advantage to the cell and is therefore propagated to all its daughter cells. Clonal amplification can be detected by bulk sequencing at high depth when the mutation is present in at least 5% of the cells comprising the DNA sample. It is strongly associated with age and occurs across most if not all tissues. Clonally amplified somatic mutations, including clonal hematopoiesis (CH) and their possible impact on aging, will be discussed below. Here, it suffices to mention that clonally amplified mutations can best be measured in small populations of cells, for example, from microbiopsies across a tissue, using high-depth sequencing [10, 11]. But even then the intrinsic error rate only allows to detect a variant allele frequency of about 5% at a sequencing depth of about 500 [12]. Because of sequencing errors, it is necessary to put in a threshold of a minimum number of reads containing the mutation to avoid false positives. For CH, targeted sequencing of panels of genes known to be frequently mutated, such as DNMT3A, TET2, ASXL1, TP53, JAK2, is often used.

The vast majority of somatic mutations are unique and do not clonally amplify. Their frequency of 10⁻⁷–10⁻⁶ is significantly lower than the sequencing error rate of next-generation sequencing (NGS), i.e., 10⁻⁴–10⁻² [13, 14]. This means that even when a cell contains thousands of somatic mutations they can still not be detected when sequencing bulk DNA. To address this technical challenge, researchers have developed multiple tools to detect extremely low-abundant somatic mutations in cells and tissues. In the 1990s, mouse models were developed harboring selectable transgenic reporter genes that could be excised from the mouse genome, recovered into E. coli, and analyzed for mutations in the reporter gene [15–17]. Such models provided the first systematic insight into somatic mutation accumulation in different tissues during aging. However, this approach cannot be extended to humans and the mutational target is limited to a bacterial reporter gene rather than the whole genome.

With the emergence of next-generation sequencing, efforts were made to sequence single cells. Sequencing the genome of individual cells is necessary to detect the vast bulk of somatic mutations, which have an extremely low abundance and are often unique for an individual cell. However, this requires methods to amplify the whole genome of single cells (SC-WGA), which first emerged in 2012 [18, 19]. These methods have been refined, including the use of error correction software [20, 21]. Alternatively, clones derived from single cells, which carry all the mutations generated in the original cell, can be obtained by in vitro culture and sequenced. The genome-wide mutations detected, then serve as surrogates for single cell analysis. However, growing such clones is labor intensive and essentially limited to stem cells. Alternatively, one can use tumors as a surrogate for clonally amplified single cells. Mutation burden in tumors is representative for the mutation burden of the cells from which the tumor originates. However, once neoplastically transformed, the subsequent tumor lineage will often develop additional mutations, which explains the observation that the somatic mutation burden in tumors range from less than one mutation per Mbp to more than 10, on average at least 3 ~ 4 times higher than in normal somatic tissues [22–24].

A third category of methods for detecting somatic mutations is single-molecule analysis. The first such method was so-called duplex sequencing, developed by the Loeb lab [25]. This approach relies on assigning unique molecular identifiers (UMIs), i.e., short sequences to uniquely tag each molecule in a sample library, to complementary strands of DNA fragments. In subsequent analysis following sequencing, these tags are employed to computationally merge the individual fragments’ strands into clusters or families. True mutations manifest on both strands, whereas sequencing artifacts are exclusive to one strand and can be disregarded. In this way, the burden of true mutations in a DNA sample from a tissue can be accurately determined [25, 26]. Further optimized versions of duplex sequencing, such as BotSeqS [27], NanoSeq [28], and SMM-seq [29], have been developed later.

Using these advanced technologies, listed in Table 1 and illustrated in Fig. 1, enormous progress has been made in unraveling the role of somatic, postzygotic mutations in aging and related diseases. Indeed, their widespread application has made it clear that somatic mutations occurring postzygotically during embryonic development, adulthood and aging, are the cause of not only cancer, but a wide variety of human diseases other than cancer [30, 31]. It is expected that with ongoing advances of accurate sequencing technology this list of diseases will expand further [32].

Table 1.

Methods to quantitatively analyze somatic mutations in normal cells and tissues

Category	Method	Key principles	Starting material	Main advantage	Limitation	References
SC-WGS (single-cell whole genome sequencing)	MDA	Isothermal amplification using Phi29 and strand-displacement mechanism	Single cells/nuclei	No cell type limitation, good coverage, high accuracy and DNA yield, long product	Expensive, labor-intensive, low throughput, artificial C > T	[21, 36–40]
	MALBAC	Quasilinear amplification	Single cells/nuclei	No cell type limitation, improved uniformity	Expensive, labor-intensive, low throughput	[19, 115, 116]
	LIANTI	Linear amplification through transposition and in vitro transcription	Single cells/nuclei	No cell type limitation, good coverage and uniformity	Expensive, labor-intensive, low throughput, complicated	[117]
	SCMDA	Reconfigured MDA, low temperature cell lysis and DNA denaturation	Single cells/nuclei	No cell type limitation, reduced cytosine-deamination artifacts, high accuracy, good coverage, high DNA yield, long product	Expensive, labor-intensive, low throughput	[9, 34, 35, 96, 106]
	PTA	Exonuclease-resistant terminators, primary template	Single cells/nuclei	> 95% coverage, good amplification uniformity, all MDA advantages	Very expensive, labor-intensive, low throughput	[38, 39, 118]
Duplex consensus sequencing	Duplex sequencing	Rule out artifacts by tagging and sequencing consensus DNA	Bulk DNA	Highly accurate, no single-cell isolation needed	Complicated operation, no SV information	[25, 26]
	BotSeqS	“Bottleneck” dilution of DNA templates before PCR enrichment	Bulk DNA	Simple procedure, minimal amount of sequencing	End-repair of SSB may generate artifacts, no SV information	[27]
	NanoSeq	ddBTPs for A-tailing to avoid errors from nick extension	Bulk DNA	Reduce errors caused by end repair, maximize duplex coverage	No SV information	[28]
	SMM-seq	RCA of UMI tagged DNA strand families	Bulk DNA	Higher efficiency	No SV information	[29]
Clone from single cell	In vitro single-cell clone	Colonies derived from single cells carry all the mutations generated in the original cell	Single viable cells	Very good coverage and uniformity	Limited in proliferate active cells, extensive cell culture	[33, 44, 61, 84, 119–122]
	LCM-seq	LCM, low-input sequencing	Microbiopsy (100 ~ 1000 cells)	Natural clones in normal tissues, no single cell isolation, no whole genome amplification	Cost, special instruments and skills required	[45, 78, 79, 93, 122–126]
	Tumor sequencing	Variant allele frequency	Bulk DNA	Public data set available	Highly mutated sample	[127–129]

MDA, multiple displacement amplification; MALBAC, multiple annealing and looping-based amplification cycles; LIANTI, linear amplification via transposon insertion; SCMDA, single-cell multiple displacement amplification; SHM, somatic hypermutation; PBBC, lung proximal bronchial basal cell; PTA, primary template-directed amplification; LCM, laser-capture microdissection; SSB, single strand DNA breaks; RCA, rolling circle–based linear amplification; SV, structural variation; SMM-seq, single-molecule mutation sequencing

Fig. 1.

Methods to detect somatic mutations. Three different categories of methods to detect somatic mutations are shown: single-cell whole genome sequencing (SC-WGS), duplex consensus sequencing, and bulk sequencing of single cell-derived clones. For SC-WGS, the four main genomic amplification techniques are SCMDA (single-cell multiple displacement amplification), MALBAC (multiple annealing and looping based amplification cycles), LIANTI (linear amplification via transposon insertion), and META-CS (multiplexed end-tagging amplification of complementary strands). For duplex consensus sequencing, representative technologies are BotSeqS (bottleneck sequencing system), NanoSeq (nanorate sequencing), and SMM-seq (single-molecule mutation sequencing). Bulk sequencing of single-cell derived clones include tumor sequencing, high-depth sequencing of tissues to detect natural clonally amplified mutations, and in vitro amplified single-cell clones. The outer circle shows the different tissues that can be targeted by these methods. The in vitro clonal amplification is limited to cells that can be cloned in culture. Single-cell (or single nuclei) and duplex consensus assays can be applied to all tissues

Application of the new methods in studying aging has by now yielded overwhelming evidence for an age-related accumulation of somatic mutations in human tissues, varying from lymphocytes [9, 33], liver hepatocytes [34] and bronchial lung cells [35], to neurons [36–39] and cardiomyocytes [40] (see also our recent review [41]). The focus in all these studies has been on base substitution mutations (single-nucleotide variants or SNVs) and small insertions and deletions (Indels). Indels also increase with age, but their frequency is about 10-fold lower than SNVs. This is most likely due to the potentially higher impact of these mutations; for example, in a gene coding region they could lead to frameshifts. Other potentially impactful types of mutations, such as copy number variation (CNV), retrotransposition (RT), and genome structural variation (SV), are more difficult to detect. They occur at even lower frequency than SNVs or Indels, i.e., one or less than one event per cell [42–44]. Especially SVs (large deletions, inversion, translocations, duplications) are difficult to detect and cannot be detected at all by using single-cell whole genome sequencing (because of artifacts associated with the amplification procedures) or single-molecule sequencing (because of a high background of self-ligation). The number of such variants reported varies from less than one to ~ 10 of such events per cellular genome. In all these cases, the largest events can be detected the easiest and it is possible that there are in fact hundreds of CNVs or SVs in the size range of ~ 1000 bp or less per cell. Recent progress in whole genome amplification and long-read sequencing will likely improve the situation substantially, which should lead to the generation of a complete index of somatic mutation burden for different cell types during aging.

Importantly, one aspect that has emerged thus far is the variation of mutation accumulation rate among tissues and cell types. In general, germ cells appear to accumulate mutations at a lower rate than somatic cells. Sperm contains less than 100 mutations per cell [28, 45] and the number may be even lower for oocytes, but this has as far as we know never been determined. This low number is in contrast to the thousands of base substitution mutations found to accumulate in human differentiated cells, as mentioned above. In the next section, we will discuss this germline-somatic contrast in some more detail.

Somatic versus germline mutations

As discussed above, for a long time, somatic mutation rate could not be quantified due to a lack of methods to accurately measure random mutations in normal cells and tissues. Before the emergence of methods to assess somatic mutations in human and animal tissues, it was generally assumed that somatic mutation rates would be similar to germline mutation rates. Germline mutation rate could be fairly accurately determined, simply based on phenotypic characteristics. J.B.S. Haldane estimated the human germline mutation rate based on the heritable mutation causing hemophilia. He calculated the rate to be roughly 10⁻⁵ per disease gene per generation [46]. Haldane’s calculations for the per gene human mutation rate compared well with John Drake’s estimate of a germline mutation frequency in humans of approximately 1 × 10⁻⁸ per base pair per generation [47]. Further refining these estimates, Kong et al. leveraged next-generation sequencing of parents and offspring to report an average de novo germline mutation rate of 1.20 × 10⁻⁸ per nucleotide per generation [48]. This translates to roughly 60 new SNVs per generation, the vast majority of which are derived from the father.

In a direct comparison of mutation rate in somatic cells (fibroblasts) with germline mutation rates in humans and mice, Milholland et al. showed that in both species the somatic mutation rate is almost two orders of magnitude higher than the germline mutation rate [49]. This is in keeping with the almost 10-fold increase in mutation burden in human somatic tissues as compared to germ cell tissue mentioned above [45]. Interestingly, humans tend to inherit more de novo germline mutations from their fathers than from their mothers [48], i.e., the predominant source of germline mutations is likely to reside in mature spermatozoa. Moreover, across diverse species, it has been consistently found that a paternal bias in germline mutation is common, regardless of variations in life history, physiology, and gametogenesis among these species [50].

The high resistance to mutation induction is not confined to germ cells but can also be observed in other cell types. Indeed, there is evidence that stem cells have lower mutation frequencies than their differentiated counterparts. For instance, studies have demonstrated that the mutation rate in human induced pluripotent stem (iPS) and embryonic stem (ES) cell lines can be up to tenfold lower compared to somatic cells [51, 52]. In a recent study, we found similar results when sequencing organoids derived from liver stem cells [53]. The mutation frequencies were considerably lower in liver stem cell-derived organoids than in differentiated hepatocytes. These observations suggest that germ cells and pluripotent stem cells are better protected against genomic instability than their somatic counterparts. However, other factors are likely to be important too. For example, germ cells or stem cells may be more sensitive to apoptosis to prevent mutation induction, or the number of cell divisions could be lower, which would reduce the occurrence of replication errors, a major source of SNVs.

Additional research is needed to fully characterize somatic mutation burden in germ cells, stem cells, and somatic cells. Nevertheless, the concept of differential resistance to mutation induction depending on the required functional life span of a cell makes intuitive sense when somatic mutation accumulation would be harmful and is indeed a causal factor in aging. Germline cells, vital for the transmission of genetic information across generations, need to be preserved indefinitely. Similarly, stem cells, which continuously replenish functional tissues, require robust genomic stability. However, fully differentiated cells, after serving their specific functions for a limited duration, are usually replaced by new cells. Hence, evolutionarily, allocating the resources required for maintaining stringent genome maintenance mechanisms beyond the functional lifespan of a differentiated somatic cell, or far beyond an organism’s reproductive period, would serve no function [54]. Indeed, allocating resources to where there is no need would lead to fitness loss. This evolutionary strategy thus leads to varying levels of mutation resistance across different cell types and organisms, possibly through differences in DNA repair capacity.

Somatic mutation signatures in aging

Age reveals itself not only by an increased number of somatic mutations, but also by a characteristic pattern of mutations termed a signature. Somatic mutational signatures were first obtained from whole exome or whole genome sequenced tumors [23]. Since a cancer usually originates from one cell only, its genome sequence is representative for this, originally normal cell [55]. Hence, sequencing tumors is essentially a surrogate for single-cell analysis (like in vitro clonal amplification) and it has been demonstrated that tumors from older subjects carry significantly more mutations than tumors from younger subjects, providing additional evidence that somatic mutations accumulate with age [24, 56]. From the somatic mutations in human tumors, available in the Cancer Genome Atlas (TCGA), signatures have been extracted using an algorithm for nonnegative matrix factorization. Based on the different types of base substitution mutations, i.e., C > A, C > G, C > T, T > A, T > C, T < G, and their flanking 5′ and 3′ bases, a classification system has been developed comprising 96 classes [23, 57]. Similarly, a classification system has been developed for small insertions and deletions (Indels) based on the size of the Indel (one or more bases) and the identity of the base(s) lost or gained. This has resulted in the Catalogue Of Somatic Mutations In Cancer (COSMIC, https://cancer.sanger.ac.uk/cosmic/signatures) compendium of mutational signatures [23, 55]. COSMIC is now used as a standard for matching somatic mutation signatures discovered in many normal cells and tissues.

Somatic mutational signatures can provide information on specific mutational conditions in a tissue at a given age, such as DNA replication infidelity, genotoxic exposure, defective DNA repair, and DNA enzymatic editing. For example, specific mutational signatures have been found in lung cancers from smokers [58]. The signature most strongly associated with tobacco smoking is single-base substitution signature signature 4 (SBS4), which is also characterized by transcriptional strand bias for C > A mutations [59]. Transcription-coupled repair protects the transcriptional strand more efficiently against genotoxic insult than the non-transcribed strand, something that is also revealed by the somatic mutational signatures [60]. Somatic mutational signatures specific for exposure to many different environmental agents have been determined [61].

For the aging process in vivo, specific clock-like signatures, such as SBS1 and SBS5, SBS40, doublet-base substitution signature DBS2 and DBS4, as well as the small insertion-and-deletion signatures ID1, ID2, ID5, and ID8, have been distilled from the TCGA data [55]. SBS1 features C > T mutations at NpCpG trinucleotides resulting from spontaneous deamination of 5-methylcytosine, which leads to T-G mismatches. If not repaired before DNA replication at mitosis, these will be converted into C > T mutations [55, 57]. Generally, SBS1 is more common in fast-proliferating tissues, and cell proliferation rate may be the critical factor to influence the rate of mutations that dominate this signature. It is therefore possible to use the SBS1 rate as a clock to estimate the number of mitoses a cell has experienced during the lineage of cell divisions from the fertilized egg. By contrast, SBS5 accumulation rate does not correlate with cell division rate, and SBS5 is the dominant clock-like signature in postmitotic cells where cell division is absent. SBS40 and SBS5 are difficult to deconvolute separately and have been recommended to be combined and designated SBS5/40 [45, 59, 62]. SBS18 is predominantly characterized by C > A substitutions and possibly due to reactive oxygen species [62]. Recently, mutations attributed to SBS18 have been found to increase with age in endometrial epithelium [62], and in inflammatory bowel disease (IBD) SBS18 burden in colonic crypts is affected by age [63]. Choudhury et al. compared mutational signatures between cardiomyocytes, neurons, hepatocytes, and lymphocytes and found the SBS18-associated signature C to increase with age in all four cell types [40]. The etiology of DBS2 has been proposed to be exposure to tobacco smoke as well as other endogenous and/or exogenous mutagens like acetaldehyde [61]. Using clock-like mutational signature analysis, researchers are now able to explore and compare mutational processes in aging across tissues, as summarized in Fig. 2 and Table 2.

Fig. 2.

Clock-like mutational signatures. Clock-like mutational signatures that accumulate with age and have been identified in normal human tissues are shown, including SBS1, 5, 18 and ID1, 2, 5. SBS1 shows no transcript and strand bias and cell proliferation rate may be the critical factor to influence this signature, which is more common in fast-proliferating tissues. SBS5 is enriched in mutations at the transcribed DNA strand and its accumulation rate does not correlate with cell division rate. SBS5 can be attributed to endogenous DNA damage and repair. In postmitotic cells where cell division is absence, SBS5 is the dominant clock-like signature. SBS18 is generated by exposure to endogenous reactive oxygen species and, recently, SBS18 burden has been identified to be affected by donors’ age in endometrial epithelium, colonic crypts, cardiomyocytes, neurons, hepatocytes, and lymphocytes. SBS1 and SBS5 are generated through independent mutation processes, associated with genome replication and DNA damage repair, respectively. Their contributions to the mutation landscapes, therefore, vary with different tissues and cell types, and the ratio of SBS1/SBS5 may shift along tissue development. Compared with adults, the contribution of the SBS1 signature is higher in fetal and childhood tissues, for example, in children bronchial epithelium. During tissue development, a switch from SBS1 to SBS5 has been observed. (SBS: single substitution signatures; ID: small insertion-and-deletion signatures; if not indicated, tissues and cells in the figure are from adult)

Table 2.

Signatures of somatic mutations accumulating with age in different tissues

Mutation type	Name*	Features*	Clock-like	Proposed etiology	Transcribed strand bias	Tissue bias	Non-cancer tissue or cell type identified with high contribution
Single substitution	SBS1	C > T at NpCpG	Yes	Spontaneous deamination of 5-mC [55, 57]	No	Fast-proliferating cells/tissues	Fetal tissue [119], intestinal stem cell [107, 119], duodenum [78], colon/rectum [63, 78, 93], umbilical cord blood [119]
	SBS5/40	C > T and T > C	Yes	Endogenous DNA damage and repair, deamination of mC to T in CpG context [40, 130]	Yes	Postmitotic cells/tissues	Neurons [38], cardiomyocytes [40], hepatocytes [34, 125], adult HSPCs [44, 119]
	SBS18	C > A	No	Reactive oxygen species	Yes	–	Endometrial epithelium [62], colonic crypts [63], cardiomyocytes [40]
Doublet-base substitution	DBS2	CC > AA with smaller numbers of CC > AG and CC > AT	Yes	Exposure to tobacco smoking and acetaldehyde [59, 61]	Yes	–	–
	DBS4	GC > AA, TC > AA and TC > CA, with small numbers of TC > GA and GC > AT	–	Unknown, endogenously generated [59]	–	–	–
Small insertion-and-deletion	ID1	Predominant insertions of T at long (≥ 5) T repeats	No	Slippage during DNA replication, DNA mismatch repair deficiency	–	–	Colorectal epithelial cells [123]
	ID2	Deletion of T at long (≥ 5) T repeats	No	Same as ID1	–	–	Colorectal epithelial cells [123]
	ID5	–	No	Unknown, correlated with SBS40	–	–	Colorectal epithelial cells [123], neurons [39]
	ID8	≥ 5-bp deletions with no or 1 bp of microhomology at their boundaries	No	DSB repair by non-homologous end-joining, p.K743N mutation in TOP2A	–	–	Neurons [39]

mC, methylated cytosines; DSB, DNA double strand break; TOP2A, transcription-associated topoisomerase 2α; IGHV, immunoglobulin heavy variable genes

–: no relevant data yet available

^*: data from https://cancer.sanger.ac.uk/cosmic/signatures

Somatic mutational signatures have also been found associated with defects in DNA repair processes, most notably DNA mismatch repair (MMR) [64, 65]. These have been derived from mutations in tumors from patients with heritable defects in MMR, most notably colorectal cancer [66, 67]. For example, ID1 is featured by predominant insertions of thymine at long (≥ 5) thymine mononucleotide repeats, while ID2 is determined by the deletion of thymine on the same motif. The process to forming ID1 and ID2 is the slippage during DNA replication of the replicated DNA strand, which is a sign of defective MMR.

Hence, analysis of somatic mutations in aging can serve not only as an aging clock, but also reveals clues as to tissue-specific factors contributing to age-related cell functional loss and increased disease incidence. However, this area of research on aging is merely in its infancy. While somatic mutation data across most organs and tissues as a function of age have now been obtained from RNA sequence data generated by the Genotype-Tissue Expression (GTEx) project [68, 69], only clonally amplified mutations can be obtained from such bulk data, while single-cell RNA seq has a too low coverage to account for more than a tiny fraction of all mutations. Meanwhile, the vast bulk of all somatic mutations that are outside expressed sequences are missed, most notably mutations in gene regulatory regions. To obtain full advantage of mutational signature analysis, genome-wide data are needed and those can only be obtained using the advanced single-molecule or single-cell assays described above, which are still expensive. Finally, the most impactful type of mutations, i.e., genome structural variants, remains very difficult to analyze using single-cell or single-molecule assays [70].

Clonally expanded somatic mutations in aging

As mentioned above, certain somatic mutations can clonally expand, either by genetic drift or due to a selective growth advantage. This allows their detection by bulk sequencing above a certain threshold. Clonal expansion was discovered in blood of elderly subjects, in which it is known as clonal hematopoiesis [71]. Because many clonally expanded mutations in blood were identified as cancer driver mutations, which could ultimately lead to uncontrolled proliferation and the formation of hematopoietic malignancies, CH was initially assumed to represent premalignant expansion of mutated hematopoietic stem cells. Indeed, CH has also been termed CHIP (clonal hematopoiesis of indeterminate potential), referring specifically to the presence of cancer-related genetic variants in the blood cells of individuals without apparent malignancy or any other recognized clonal disease. Certain CH-associated mutations, for example, mutations in TET2 and DNMT3A, can impact immune cell function and promote inflammatory responses [72, 73]. CH has also been correlated with other age-related diseases, most notably cardiovascular disease, as well as with overall mortality. Clonal expansion of mutations in cells of the hematopoietic lineage differs from that in solid tissue (below) where the expansion of a clone is constrained locally, while the migration of hematopoietic cells throughout the body may affect a wide range of age-related diseases. Interestingly, clonal expansion of somatic mutations may sometimes be beneficial. Indeed, recent evidence suggests that CH may protect against Alzheimer’s disease, with the mutations found in blood also detected in microglial cells in the brain [74]. Also in blood from Parkinson’s disease patients, the percentage of CH carriers was significantly lower than in age-matched control subjects [75].

By now it has been well documented that blood is not unique in showing clonally amplified mutations, which has now been demonstrated for most if not all human tissues [63, 76–81]. Some of these expanded mutations have also been linked to cancer and it has been argued that age-related changes in the tissue microenvironment, with a declining ability to clear dysfunctional cells, fuels aberrant clonal expansions, establishing a direct biological link between the aged phenotype and cancer risk [82]. An interesting example of a gene harboring a clonally amplified somatic mutation at the interface of cancer and aging is NOTCH1. NOTCH1 is the most frequently mutated gene in physiologically normal esophageal epithelium, with the prevalence of NOTCH1 mutations in normal esophagus (> 30%) several times higher than in esophageal cancers (< 20%) [80, 81, 83]. In contrast, mutations in the cancer-promoting gene TP53 are found in most cancers with a prevalence higher than 80%, but are present at a much lower frequency (30%) in normal esophageal samples [81]. Clones carrying NOTCH1 mutations expand progressively with age, emerge in childhood at multiple sites in esophageal epithelia, and increase in number and size over time. Eventually, these mutated clones replace almost the entire esophageal epithelium in extremely old individuals. The Notch pathway plays a crucial role in controlling cell fate during development and in tissue homeostasis. It can both promote and suppress carcinogenesis, but the mechanism is complex and environment-dependent. Abby et al. proposed that NOTCH1 mutations drive clonal expansion in normal esophageal epithelium and prevent tumor expansion; conversely, wild-type NOTCH1 favors tumor expansion. This hypothesis was confirmed in a mouse model, where NOTCH1 deletion increases clonal fitness, enable epithelial colonization, and reduce tumor growth [83]. Evidence for positive selection of NOTCH1 mutations has also been found in normal skin and bronchial epithelium, possibly preventing esophageal and other cancers [76, 84].

A substantial proportion of the expanded mutations were observed in various tissues and appeared unrelated to cancer, raising the question of a potential causal relationship to aging phenotypes. The clonal expansion of somatic mutations has the potential to influence the function of organs and tissues, affecting disease susceptibility and potentially serving both detrimental and beneficial roles in the aging process [85].

Age-related mutations beyond the nuclear genome

Age-related somatic mutation accumulation can also be found outside the nuclear genome, i.e., in the transcriptome and in the mitochondrial genome. When occurring in transcribed DNA, somatic mutations can be detected at the RNA level, e.g., from RNA-seq data sets [76]. However, mutant RNA can also be a result of RNA editing or transcription errors. RNA editing is the posttranscriptional modification of RNA at one or more positions, leading to transcripts that differ from the genome template. Such RNA sequence differences between mature transcripts and their encoding genome sequences represent a form of genetic recoding. RNA editing is different from other forms of RNA processing, such as 5′-capping, 3′-polyadenylation, or splicing, in the sense that it represents a true form of recoding, with the capacity to amplify genetic diversity and alter gene product function [86]. Age-related differential RNA editing, leading to differentially expressed genes, has been observed [87].

Transcription errors are very similar to random somatic mutations in the genome, in the sense that they are the result of infidelity, in this case infidelity of the transcription machinery. This is a relatively understudied field, but work from Vermulst and others has shown, relatively recently, that transcription errors are much more frequent than DNA mutations and since they are generated continuously, highly likely to lead to many mutated proteins over the lifespan of a cell [88]. Transcription errors can contribute to protein aggregation [89].

A potentially major target for somatic mutagenesis is the mitochondrion [90, 91]. Mitochondria, the membrane-bound organelles located in the cytoplasm that house the respiratory complex, have their own genome. Mitochondrial DNA (mtDNA) is separated from nuclear DNA, has its own DNA replication and RNA transcription systems, and lacks some major DNA repair systems that are active in the nuclei of most cell types. The somatic mutation rate of mtDNA is about 100-fold higher than that of nuclear DNA, possibly because it is closer to the main source of oxidative DNA damage, i.e., the electron transport chain, and its deficiency in active DNA repair pathways. However, since each cell has many mitochondria, and each mitochondrion multiple genome copies, mutations in the mitochondrial genome have less of an impact than when they occur in the nuclear genome. Because the mitochondrial genome is so small, i.e., 16.5 kb, analysis of low abundant mutations is easier than in the nuclear genome and several methods have emerged all showing an increase of mtDNA mutations with age, in mice as well as in humans, and in multiple tissues [92]. In the earlier described study on age-related mutation accumulation in colonic crypts from different mammalian species, it was found that mitochondria harbor heavy burdens of somatic mutations, and mtDNA mutation burden is inversely correlated with species’ lifespan [93].

Mitochondrial DNA has mutational spectra distinctive from nuclear genomic DNA, which can be attributed to the independent DNA replication system in mitochondria and the higher oxidative burden caused by the respiratory complex. In the above mentioned paper on mitochondrial DNA mutations in intestinal crypts across mammals, the mutational spectra showed a preponderance of C > T and A > G substitutions, which were suggested to result from mtDNA replication errors rather than DNA damage [93]. An increase of A to G transitions with age in the heavy strand of mitochondria across multiple tissues was found by Mikhailova et al. [94] in mice and humans. Further, A > G transitions on the heavy strand were found to be more frequent in mammals with long generation times. A to G transition is thus considered as an age-associated mutational signature specific for mtDNA of long-lived mammals [94]. More recently, the above described duplex sequencing method has been used to analyze multiple tissues of aging mice, finding significant tissue-specific increases during aging across all tissues examined [95]. Interestingly, consistent with results of Cagan et al. [93] and Mikhailova et al. [94], the most common mutations were G > A/C > T substitutions, indicative of replication errors and/or cytidine deamination. G > T/C > A substitutions, indicative of oxidative damage, were less frequent and did not increase with age in any tissue. This lack of accumulation of oxidative damage-linked mutations with age confirms earlier observations that somatic mutations in mtDNA mostly result from mtDNA replication errors rather than errors in the repair of DNA damage and suggest that oxidative lesions are efficiently removed from mtDNA genomes, possibly by base excision repair pathways that are highly active in mitochondrial genomes [95].

Functional impact of somatic mutations

Failla already predicted that in short-lived animals the somatic mutation rate per unit time would be higher than the rate in long-lived ones [2]. However, due to the lack of methods to quantitatively analyze somatic mutations, it is only very recently that such a reverse correlation between somatic mutation rate and species lifespan has been demonstrated, first by studying mutagen-induced mutation burden in cells from rodents greatly differing in maximum life span [96], and shortly thereafter across a set of different mammals, not limited to rodents, by comparing the age-related accumulation of mutations in colonic crypts [93]. Both data sets indicate an inverse correlation between somatic mutation rate and maximum species life span. However, such an inverse correlation could be secondary to other variables and be unrelated to the aging process itself. The demonstration that somatic mutations accumulate with age to much higher numbers than previously thought possible is also not conclusive. Indeed, it is possible that somatic mutation rates have been selected to be low enough to keep cancer at bay until well after the reproductive period, but still too low to cause any other age-related pathogenic phenotype [97].

Thus far, somatic mutations have only been functionally implicated in disorders of proliferative homeostasis, including hyperplasias, benign neoplasms, and malignant neoplasms, the incidence of which increase with age in mammalian species, including humans [98]. An increasing number of these diseases, but more recently also diseases other than cancer, have been found associated with clonally amplified somatic mutations in multiple tissues, including skin, esophagus, brain, and liver [31, 32]. It is reasonable to consider the age-related accumulation of somatic mutations as discussed above as a major risk factor of clonal amplification and therefore a risk factor for age-related disease. However, in addition to clonally amplified somatic mutations, it has been suggested that an increased burden of random somatic mutations, possibly with a major role for epimutations as well, can have detrimental consequences for the cell by increasing transcriptional heterogeneity among cells; in turn this would adversely affect gene regulatory pathways [85, 99, 100]. However, it will be necessary to demonstrate such a functional impact of increased somatic mutation burden in aged human tissues. Thus far, the only evidence for it is the observed inverse correlation between increased somatic mutation burden and proliferative potential of human muscle satellite cells from aged human subjects [101].

Evidence that somatic mutation accumulation could lead to aging comes from the analysis of somatic cells of patients with progeroid diseases (Table 3). Progeroid diseases are rare genetic disorders showing features of aging, such as hair loss, loss of subcutaneous fat, and age-related diseases such as cardiovascular diseases and osteoporosis. Most of these diseases are caused by defects in DNA repair systems, including Werner’s syndrome (WS) and Cockayne syndrome (CS). WS is caused by a genetic defect in a RecQ helicase, which is involved in DNA repair. Lymphocytes of WS patients show increased chromosomal instability and genome structural variation [102–104], but have not yet been analyzed carefully using the advanced mutation analysis method discussed above. Interestingly, the heterozygous BRCA1 mutations that predispose so strongly to breast and ovarian cancer have also been shown to cause accelerated ovarian aging and systemic aging-related pathophysiology [105]. Indeed, somatic mutation frequency is elevated in non-cancer carriers of BRCA1/2 mutations [106].

Table 3.

Somatic mutation burden in normal cells from patients with progeroid diseases

Disease	Tissue	Method	Number of SNVs per cell*		References
			In disease	Normal control
Cockayne syndrome	Prefrontal cortex	MDA	1564	1038	[37]
Xeroderma pigmentosum	Prefrontal cortex	MDA	2577	1038	[37]
Alzheimer’s disease	Prefrontal cortex	MDA	2286	1747	[38]
Alzheimer’s disease	Hippocampus	MDA	1917	1667	[38]
Inflammatory bowel disease	Colonic crypt	LCM-seq	2864	2684	[63]
BRCA1/2 mutation carriers	Mammary epithelia	SCMDA	1902 (BRCA1/2-mut)	1506 (BRCA1/2-wt)	[106]

BRCA1/2, breast cancer 1 (BRCA1) and breast cancer 2 (BRCA2) genes

^*Median values from x–y different cells from x–y patients or controls

While there is evidence for a causal role of somatic mutations in aging, it is important to briefly discuss two types of observations that seem to conflict with a causal role of somatic mutations in aging. First, genetic defects in DNA replication has been associated with increased somatic mutation rates, without accelerating aging [107–109]. This suggests that somatic mutations are not causal to symptoms of aging other than cancer. However, humans or mice harboring these genetic defects usually die early from cancer, and to detect symptoms of premature aging requires exhaustive phenotypic characterization of the subjects involved, which has not been done [110].

Second, it has been argued that since live and superficially healthy animals can be cloned by somatic cell nuclear transfer (SCNT) using nuclei from adult, increased somatic mutation burden could never be a major cause of aging. Indeed, successful serial recloning has been achieved in the mouse [111]. However, this reasoning overlooks the fact that somatic cloning is inefficient, with high abortion and fetal mortality rates commonly observed [112]. Indeed, the low efficiency of the process itself, usually far below 10% per reconstructed oocyte, could point to the possibility that only nuclei with the lowest possible somatic mutation burden result in live clones, which is supported by the fact that stem cells or fetal donor cells are by far the most likely to achieve success [113]. Selection of cells with the least amounts of somatic mutations from among early cell lineages already harboring low mutation loads could explain even how serial recloning could be achieved [111].

Overall, therefore, while there is no conclusive evidence yet for a causal role of an elevated somatic mutation burden in aging, clonally amplified somatic mutations have now been demonstrated as causes of a large number of age-related diseases, including cancers and diseases other than cancer. More research is needed to demonstrate causality of increased somatic mutation burden per se.

Summary and future prospects

Since the somatic mutation theory of aging was first proposed, enormous progress has been made in developing the methods needed to test if somatic mutations really do accumulate with age. This has now resulted in a solid body of data showing that base substitution mutations accumulate with age in their thousands, with small insertions and deletions about 10-fold less. We also know that somatic mutation rate correlates with functional life span of a cell, with a much higher rate in differentiated somatic cells than in stem cells, and the lowest rate in germ cells. A relationship to function is also suggested by observations that cells from longer-lived mammalian species generally have a lower somatic mutation rate than cells from shorter-lived species. In spite of all the progress in this field, two major challenges remain.

First, from a technical point of view, improvements still need to be made in the quantitative analysis of genome structural variation. Neither at the single-cell nor single-molecule level successful approaches have been developed to gain the comprehensive picture of all types of somatic mutations that is needed to reliably judge whether or not somatic mutations can have a functional impact on the aging process. This is critically important in assessing the functional relevance of somatic mutations in aging. A reliable picture of the burden of all types of somatic mutations in different cell types as a function of age will allow us to study their functional impact.

Second, model systems need to be developed allowing to study how different somatic mutation loads affect specific cellular functions known to decline with age.

In conclusion, while critically important methodological advances now enable us to study somatic mutations and genome mosaicism in some detail, insight into the causal role of somatic mutations in aging is still lacking. Nevertheless, a new milestone in this rapidly growing field is already on the horizon. Will it be possible to intervene and slow down or even abrogate somatic mutation accumulation with age that leads to increased risk of cancer and possibly other aging-related health problems? This question has been addressed in some detail elsewhere [114], but the general answer is that there are now multiple approaches to postpone and even reverse somatic mutations that have accumulated with age. We expect that some of these new approaches will come to fruition over the next decades.

Author contribution

J.V. and P.R. conceived of topic area. J.V. and P.R. wrote the initial draft of the manuscript. J.Z. drew the figures. J.V. reviewed and edited the manuscript; all authors commented on previous versions of the manuscript. All authors approved the manuscript for publication. We acknowledge the Bioinformatics Core in the Center for Single-Cell Omics (CSCOmics), Shanghai Jiao Tong University School of Medicine, for providing assistance with interpreting the bioinformatics and computational genomics aspects of this review.

Funding

This study was financially supported by the National Natural Science Foundation of China grant 82172461 to J.V.; NIH grants AG017242, AG047200, AG038072, ES029519, HL145560, and AG056278 to J.V.; DOD grant BC180689P1 to J.V.; and the Glenn Foundation for Medical Research to J.V.

Declarations

Competing interests

J.V. is co-founder of SingulOmics Corp. and MutagenTech Inc. The remaining authors declare no competing interests.