Spectra and characteristics of somatic mutations induced by ionizing radiation in hematopoietic stem cells

Significance

The increased neoplasm risk is one of the major late effects of ionizing radiation, and widely believed to be a consequence of somatic mutations caused by DNA damages. However, genome-wide characteristics of radiation mutagenesis or molecular mechanisms for subsequent neoplasm development have not been revealed. This study elucidated whole spectra and frequencies of somatic mutations in mouse hematopoieticstem cells (HSCs) with or without prior whole-body X-irradiation. We thus determined the radiation-sensitivity and -specificity of each type of somatic mutations, and identified mutational signatures of ionizing radiation. Further, we found that a large fraction of postirradiation HSCs originated from a single stem cell that survived irradiation. These findings contribute to mechanistic elucidation of and biomarker development for late radiation effects.

Abstract

Spectra and frequencies of spontaneous and X-ray-induced somatic mutations were revealed with mouse long-term hematopoietic stem cells (LT-HSCs) by whole-genome sequencing of clonal cell populations propagated in vitro from single isolated LT-HSCs. SNVs and small indels were the most common types of somatic mutations, and increased up to twofold to threefold by whole-body X-irradiation. Base substitution patterns in the SNVs suggested a role of reactive oxygen species in radiation mutagenesis, and signature analysis of single base substitutions (SBS) revealed a dose-dependent increase of SBS40. Most of spontaneous small deletions were shrinkage of tandem repeats, and X-irradiation specifically induced small deletions out of tandem repeats (non-repeat deletions). Presence of microhomology sequences in non-repeat deletions suggested involvement of microhomology mediated end-joining repair mechanisms as well as nonhomologous end-joining in radiation-induced DNA damages. We also identified multisite mutations and structural variants (SV), i.e., large indels, inversions, reciprocal translocations, and complex variants. The radiation-specificity of each mutation type was evaluated from the spontaneous mutation rate and the per-Gy mutation rate estimated by linear regression, and was highest with non-repeat deletions without microhomology, followed by those with microhomology, SV except retroelement insertions, and multisite mutations; these types were thus revealed as mutational signatures of ionizing radiation. Further analysis of somatic mutations in multiple LT-HSCs indicated that large fractions of postirradiation LT-HSCs originated from single LT-HSCs that survived the irradiation and then expanded in vivo to confer marked clonality to the entire hematopoietic system, with varying clonal expansion and dynamics depending on radiation dose and fractionation.

It has been established that ionizing radiation increases long-term risks of malignant neoplasms such as leukemia and solid cancers owing to epidemiological studies on victims of nuclear explosions or accidents, nuclear workers, medical radiation, and others (1, 2). More recently, radiation risks of nonmalignant diseases such as benign tumors and circulatory diseases have become evident (3, 4). Although it is widely believed that increased risks of neoplasms are due to increased somatic mutations by radiation exposure, precise pathophysiological mechanisms for those late adverse effects even for neoplasm development have not been revealed. Especially, the spectra or characteristics of somatic mutations directly introduced by ionizing radiation have not been elucidated comprehensively or quantitatively. Therefore, whether and how those radiation-induced somatic mutations contribute to the development of late adverse effects after varying latent periods is still a major enigma in radiation biology. In order to address these long-standing questions, we need to clarify the genomic landscape of somatic mutations directly introduced by radiation. Thereafter, we will be able to identify potential biological significance of those mutations, which may include functional or causal interactions with other genomic variations acquired later by aging or other environmental factors, or inherited variations, in the development of late adverse effects. Those radiation-induced mutations may also pathogenically interact with cellular or systemic effects elicited by radiation exposure such as persistent inflammation and immune dysfunction (5), or rather may play causal roles in such cellular or systemic effects.

However, even in the era of high-throughput sequencing, it is not a simple task to identify somatic mutations introduced by radiation exposure at the sequence level on a whole-genome scale. Because the frequencies of somatic mutations are usually lower than the error rates of Illumina sequencers by at least three orders of magnitude, it is not possible to identify somatic mutations just by sequencing bulk tissues that are normally mosaics of different somatic mutations (6). However, in recent years, several groups have conducted whole-genome sequencing (WGS) of clonal cell populations propagated in vitro from single cells to successfully identify somatic mutations in various tissues of humans and mice (7–9).

In this study, we elucidated whole spectra and frequencies of somatic mutations in mouse long-term hematopoietic stem cells (LT-HSCs) by WGS of clonal cell populations propagated in vitro from single LT-HSCs with or without prior whole-body X-irradiation. We thereby determined the dose–response of each type of mutations to X-irradiation and the radiation specificity, and identified mutational signatures of ionizing radiation. Further, through analysis of somatic mutations, we found that large fractions of postirradiation LT-HSCs originated from a single LT-HSC that expanded in vivo to confer marked clonality to the hematopoietic system, with varying clonal expansion and dynamics depending on radiation dose and fractionation. These findings should contribute to elucidation of molecular and cellular mechanisms for late adverse effects of ionizing radiation and the development of biomarkers for genomic insults by radiation exposure.

Results

WGS of LT-HSC-Derived Clones.

To unravel spectra, frequencies, and characteristics of somatic mutations induced by whole-body X-irradiation, we conducted WGS of clonal cell populations propagated in vitro from single mouse LT-HSCs isolated from bone marrow with or without prior irradiation. Male C57BL/6J mice were exposed to X-rays at the dose of 3.8 or 7.5 Gy at 8 wk of age, or at the total dose of 7.7 Gy in four equal fractions beginning at 5 wk and ending at 8 wk of age (four doses of 1.9 Gy per week) (Fig. 1A). At 16 wk of age (i.e., 8 wk after completion of irradiation), mice were euthanized, and LT-HSCs as defined by lineage⁻/c-Kit⁺/Sca-1⁺/CD34⁻/CD150⁺/CD48⁻ were isolated from bone marrow by FACS (SI Appendix, Fig. S1). Because in our initial set of experiments about 90% of mice exposed to single-dose 7.5-Gy radiation died within 3 wk, we decided to adopt fractionated irradiation at 7.7 Gy, which yielded a survival rate of 90% at 8 wk after exposure. The numbers of mice used for a final set of experiments are as follows: three mice without irradiation (0-Gy exposure), three mice exposed to 3.8 Gy, one mouse exposed to 7.5 Gy in a single dose, and two mice exposed to 7.7 Gy in four fractions (SI Appendix, Table S1). The isolated LT-HSCs were seeded in 96-well plates at one cell per well for clonal expansion, grown into colonies, replated, and then subjected to DNA extraction. There was no difference in the efficiency of colony formation from LT-HSCs among different radiation dosages or conditions. WGS of DNA from two independent clones and a matched tail from each mouse was conducted with the mean coverage depths of at least 33 for each sample (SI Appendix, Table S2). Sequences of matched-tail bulk DNA were used as a reference to call germline variants to be differentiated from somatic mutations.

Fig. 1.

Effects of whole-body X-irradiation on small somatic mutations in LT-HSC-derived clones. (A) The diagram shows experimental procedures to identify somatic mutations by WGS of clonal cell populations derived from single mouse LT-HSCs isolated from bone marrow with or without prior whole-body X-irradiation. (B) The graphs show the numbers of small somatic mutations per genome, i.e., single-nucleotide variants (SNVs), small insertions, small deletions, repeat deletions, non-repeat deletions, and microhomology deletions, identified in LT-HSC-derived clones for each radiation dose. Each circle represents each clone, and two independent clones were obtained from each mouse, that is, six clones from three nonirradiated mice (0 Gy), six clones from three mice exposed to single-dose 3.8-Gy radiation, two clones from one mouse exposed to single-dose 7.5-Gy radiation, and four clones from two mice exposed to fractionated 7.7-Gy radiation (denoted by fr7.7). Horizontal lines adjacent to circles indicate mean values and SD for each radiation dose. Asterisks indicate P-values by binomial test in two-way comparisons with Holm correction for multiple comparisons within each graph: *P < 0.05, **P < 0.001, and ***P < 10⁻¹⁰.

Identification of Small Somatic Mutations.

Sequence reads were mapped to the mouse reference genome, and variants were called by GATK HaplotypeCaller. For accurate identification and comparison of somatic mutations across all the samples, we defined effective whole-genome coverage (EWC) regions (10, 11) based on read-depth and quality cutoffs. The EWC regions shared by all the samples (covering 76.4% of the whole diploid genome) (SI Appendix, Table S3) were analyzed for small somatic mutations including SNVs and indels. Somatic mutations were defined as genomic alterations that were identified specifically in LT-HSC-derived clones but not in any of the tail samples (SI Appendix, Fig. S2).

Overall, 1,505 SNVs, 694 small insertions, 1,134 small deletions, and 77 multisite mutations (see below) were identified as somatic mutations within the EWC regions. Of those, randomly selected 88 SNVs and 50 non-repeat deletions (see below) produced single amplicon bands by PCR, and were all successfully validated as genuine somatic mutations by PCR-amplicon sequencing (Dataset S1), which indicated no detectable false positivity in our calling of those somatic mutations. Estimated false-negative rates of our mutation calling for small somatic mutations were less than 7.5% as described below in more detail. The spontaneous somatic mutation rate for SNVs in LT-HSCs from nonirradiated mice was 55.0 ± 16.4 (average ± SD) per genome (10.9 ± 3.3 per 10⁹ nucleotides) (Fig. 1B, Table 1, and SI Appendix, Table S14), which was lower than those in other murine tissues by up to two orders of magnitude (6, 7). Spontaneous mutation rates for small insertions and deletions in nonirradiated mice were 33.8 ± 7.3 and 42.5 ± 11.5 per genome, respectively. The numbers of SNVs, small insertions, and deletions increased by X-irradiation in a dose-dependent manner up to 3.2, 2.1, and 3.0-fold, respectively.

Table 1.

Radiation sensitivity and specificity of somatic mutations.

	Mutation rate (per genome)
Mutation type	Spontaneous* (0 Gy) mean ± SD	Induced (per Gy) (95%CI†)	Radiation specificity index‡
SNV	55.0 ± 16.4	14.7 (13.0–16.5)	0.27
Small insertion	33.8 ± 7.3	4.70 (3.50–5.90)	0.14
Small deletion (total)	42.5 ± 11.5	10.5 (8.9–12.0)	0.25
Repeat deletion	42.1 ± 12.0	4.43 (3.12–5.74)	0.11
Non-repeat deletion	0.44 ± 0.62	6.31 (5.62–7.04)	14.5
Non-repeat deletion w/o microhomology	0.22 ± 0.49	4.63 (4.04–5.26)	21.2
Microhomology deletion	0.22 ± 0.49	1.65 (1.28–2.05)	7.6
Multisite mutation	1.09 ± 1.40	1.08 (0.66–1.50)	0.99
Structural variant (total)	0.83 ± 1.46	0.78 (0.52–1.05)	0.93
Large deletion	0.17 ± 0.37	0.40 (0.25–0.59)	2.4
Structural variant except retroelement insertion	0.17 ± 0.37	0.72 (0.51–0.96)	4.3

^*Spontaneous mutation rates were experimentally determined with nonirradiated mice.

^†95% CI denotes the credible interval of the regression for the per-genome mutation rate induced by 1-Gy radiation.

^‡Radiation specificity index was calculated by dividing the per-Gy mutation rate by the spontaneous mutation rate.

Base Substitution Signatures of SNVs.

Of the six possible base substitutions in SNVs, the most common spontaneous substitution in nonirradiated mice was the C > T transition (either at CpG or non-CpG sites), followed by the oppositely directed T > C transition (Fig. 2 A and B). By X-irradiation, C > A and T > A transversions and the C > T transition at non-CpG sites increased more significantly than others. These three base substitutions coincide with the somatic mutations previously identified by a shuttle vector assay as SNVs specifically induced in mice administered with a chemical agent to generate reactive oxygen species (ROS) (12), and are also known as the mutations introduced via several DNA base damages induced by ROS, including 8-oxo-guanine that causes the C > A transversion (13, 14). These notions collectively suggest a causal role of ROS in X-ray mutagenesis (Discussion). The C > T transition at CpG sites is often caused by deamination of 5-methylcytosine, and correlated with the cell division number (15), but the predominance of the C > T transition at non-CpG sites in irradiated mice suggests the role of ROS rather than cell division.

Fig. 2.

Base substitution signatures of SNVs. (A) The graph shows per-genome numbers of all six possible base substitutions, in which C > T transitions are further categorized into those at CpG or non-CpG sites, in SNVs identified in LT-HSC-derived clones for each radiation dose. The same sequencing data from the same clones as in Fig. 1 were analyzed. “fr7.7” denotes fractionated 7.7-Gy radiation. Bars represent mean and SD, and asterisks indicate P-values by binomial test in two-way comparisons with Holm correction for multiple comparisons within each substitution type: *P < 0.05, **P < 0.001, and ***P < 10⁻¹⁰. (B) The graph shows the proportions of the base substitutions categorized as in A. Error bars represent SD. (C) The graph shows the single base substitution (SBS) profiles across all 96 trinucleotide contexts comprising all six possible base substitutions with 5′ and 3′ flanking bases. (D) The graph shows relative contributions to the SBS profiles of previously established SBS signatures deduced in the pancancer analyses by the Catalogue of Somatic Mutations in human Cancer (COSMIC).

To gain further insights into mutational processes responsible for spontaneous and radiation-induced somatic mutations in LT-HSCs, we next determined relative frequencies of SBS across all 96 trinucleotide contexts comprising the six base substitutions with 5′ and 3′ flanking bases (Fig. 2C). The SBS profiles were compared with the established SBS signatures that have been previously deduced in the pancancer analyses by the COSMIC, and were deconvoluted to determine relative contributions of the established signatures (Fig. 2D) (16). The SBS signature of spontaneous SNVs in LT-HSCs from nonirradiated mice was attributable to SBS5 (75%) and SBS1 (25%) signatures, which are both known as clock-like signatures that correlate with the age of sample donors (17). Particularly, the underlying mechanism of the SBS1 signature is spontaneous deamination of 5-methylcytosines mostly at CpG sites, which likely reflects the cell division number. By X-irradiation, the SBS5 contribution was markedly diminished and replaced by SBS40 in a dose-dependent manner, demonstrating a strong link between radiation exposure and SBS40, which is known to contribute to mutations in multiple cancer types, but considered as a featureless flat signature with unknown etiology (16).

Repeat and Non-repeat Small Indels.

We subdivided small indel mutations into two categories; one is “repeat insertion/deletion” that expands or shrinks a tandem repeat sequence comprising two or more repeat units, and the other is “non-repeat insertion/deletion” that occurs out of tandem repeats. As for small deletions, the mutation rate of repeat deletions that occurred spontaneously in nonirradiated mice was 42.1 ± 12.0 per genome (Fig. 1B) with a size distribution from one to eight bases, of which one-base deletions comprised 75% (Fig. 3A, blue bar and SI Appendix, Table S4). The repeat deletions were increased by X-irradiation with a wider size distribution up to 12 bases and in a dose-dependent manner up to 1.9-fold. In contrast, non-repeat deletions were hardly detectable in nonirradiated mice (mutation rate 0.44 ± 0.62 per genome) (Fig. 1B), and were specifically induced by irradiation, with a much wider size distribution from one to 34 bases (Fig. 3A, red bar and SI Appendix, Table S4), in a dose-dependent manner up to 35.0 ± 7.0 per genome with fractionated 7.7-Gy radiation. When the size of small deletions were compared between mice with different radiation doses or conditions, those in irradiated mice were significantly larger than those in nonirradiated mice, primarily because the size distribution of non-repeat deletions were wider than that of repeat deletions (SI Appendix, Table S4). We then examined spectra of small deletions across the deletion size and the number of tandem repeat units in the reference genome. Most (99%) of spontaneous small deletions in nonirradiated mice were repeat deletions in tandem repeats (including homopolymers) of 4 to 21 units (Fig. 3B, blue bar); the majority (70%) of them were one-base deletions in A/T homopolymers of 6 to 18 bases (SI Appendix, Fig. S3A, blue bar). By X-irradiation, non-repeat deletions (Fig. 3B, red bar), and repeat deletions in tandem repeats of five units or less, most notably one-base deletions in A/T or G/C homopolymers of five bases or less were specifically induced (SI Appendix, Fig. S3A, blue bar), and thus revealed as mutational signatures of radiation.

Fig. 3.

Size distributions and repeat number profiles of small indels. (A) The graph shows size distributions of small insertions and deletions for each radiation dose. Blue bars indicate per-genome numbers of repeat insertions/deletions of each size, whereas red bars indicate numbers of non-repeat insertions/deletions. The same sequencing data from the same clones as in Fig. 1 were analyzed. “fr7.7 Gy” denotes fractionated 7.7-Gy radiation. (B) The graph shows spectra of small deletions across the deletion size and the number of tandem repeat units in the reference genome. Blue bars indicate per-genome numbers of repeat deletions of each size and of each repeat unit number, whereas red bars indicate numbers of non-repeat deletions that occurred out of tandem repeats.

To gain insights into mutational processes induced by radiation exposure, we further subcategorized some of the non-repeat deletions as “microhomology deletions” based on the following criteria; the deletion size is three bases or larger, and the deleted sequence has a homologous sequence of two bases or more at one end, which is identical with a flanking sequence adjacent to the other end of the deletion. Such microhomology deletions are typically generated by microhomology mediated end-joining (MMEJ) repair mechanisms dependent on polymerase theta, whereas deletions without microhomology are mostly generated by nonhomologous end-joining (NHEJ) mechanisms (18, 19). Microhomology deletions were specifically induced by X-irradiation, comprising 23 to 35% of non-repeat deletions in average for each radiation dose, with the deletion size up to 34 bases and the microhomology length of two to five bases, in a dose-dependent manner up to 13.7 ± 0.7 per genome with single-dose 7.5-Gy radiation (Fig. 1B). We then examined frequency spectra of non-repeat deletions with or without microhomology across the deletion size and the microhomology length (SI Appendix, Fig. S3B). The proportion of microhomology deletions among non-repeat deletions, or the spectral profile was not related to the radiation dose, suggesting the invariable involvement of both NHEJ and MMEJ repair mechanisms for DNA double-strand breaks induced by X-irradiation at different doses.

Spontaneous small insertions in nonirradiated mice were all repeat insertions with the size distribution from one to six bases, of which one-base insertion was the most frequent (Fig. 3A, blue bar and SI Appendix, Table S5). By X-irradiation, repeat insertions were increased in a dose-dependent manner with similar size distributions. In contrast, non-repeat insertions were observed exclusively in irradiated mice albeit in a small proportion (only 1.7% of small insertions) with the size up to 19 bases with no apparent dose-dependency in the frequency or insertion size (Fig. 3A, red bar and SI Appendix, Table S5). There was no statistically significant difference in the size of small insertions between mice with different radiation doses. We then examined spectra of small insertion frequencies across the insertion size and the number of repeat units in the reference genome (SI Appendix, Fig. S3C). Spontaneous small insertions in nonirradiated mice were found in tandem repeats of 5 to 23 units; about one-third of them were one-base insertions in A/T homopolymers of 5 to 16 bases. By X-irradiation, non-repeat insertions and repeat insertions in tandem repeats of 2 to 4 units were specifically induced, and thus revealed as mutational signatures of radiation.

Multisite Mutations.

It has been reported that germline multisite mutations, i.e., clusters of base substitutions or small indels in close proximity, can be a distinctive marker for parental exposure to ionizing radiation (20–22). We thus counted multisite mutations, defined here as two or more alterations (base substitutions or small indels) occurring within 100 bases in the reference genome on the same allele, distinct from isolated SNVs or indels. Overall, 77 multisite mutations were identified within the EWC regions and manually confirmed by Integrative Genomics Viewer (IGV) (SI Appendix, Fig. S4) (see SI Appendix, Supporting Text for more details). The spontaneous rate for multisite mutations in LT-HSCs from nonirradiated mice was 1.09 ± 1.40 per genome (Fig. 4A). The number of multisite mutations increased significantly by X-irradiation up to 7.8-fold with 3.8-Gy irradiation, but not linearly correlated with the dose. The sequence length spanning multisite mutations in the reference genome ranged from one to 78 bases (SI Appendix, Fig. S5A), and was not related to the radiation dose. Of the seven types of multisite mutations (SI Appendix, Supporting Text), doublet base substitution (DBS, 39%), SBS plus deletion (SBS+Del, 29%), and two nonconsecutive SBS (2SBS, 23%) were more common and radiation-inducible than others (SI Appendix, Fig. S5 B and C). Collectively, 99% of multisite mutations involved one or more base substitutions, whereas the deletion and insertion were involved only in 30% and 5.2% of them, respectively (SI Appendix, Fig. S5B). Total numbers of base substitutions and deletions within multisite mutations increased significantly by irradiation, but insertions did not (SI Appendix, Fig. S6A). Of the six possible base substitutions, the dominant species involved in multisite mutations were C > A and T > A transversions and the C > T transition at non-CpG sites, which are the same three substitutions that were more significantly induced among isolated SNVs by X-irradiation than others, suggesting the major role of ROS also in the generation of multisite mutations, not only in radiation-induced SNVs (12–14). The C > T transition at CpG sites was totally absent (Fig. 4B and SI Appendix, Fig. S6B) in multisite mutations, making a sharp contrast to its substantial contribution to isolated SNVs (Fig. 2B), suggesting noninvolvement of cytosine methylation, deamination, or cell division number in multisite mutations (15). The size of deletions within multisite mutations ranged from one to 17 bases (SI Appendix, Fig. S5D), whereas that of insertions ranged from one to 16 bases; those sizes were not related to the radiation dose.

Fig. 4.

Multisite mutations and structural variants. (A) The graphs show per-genome numbers of multisite mutations, total structural variants (SV), and SV except retroelement insertions for each radiation dose. Each circle represents each LT-HSC-derived clone. The same sequencing data from the same clones as in Fig. 1 were analyzed. fr7.7 denotes fractionated 7.7-Gy radiation. Horizontal lines adjacent to circles indicate mean values and SD for each radiation dose. Asterisks indicate P-values by binomial test in two-way comparisons with Holm correction for multiple comparisons within each graph: *P < 0.05, and **P < 0.001. (B) The graph shows the proportions of the six base substitutions within the multisite mutations with C > T transitions subcategorized into those at CpG or non-CpG sites. (C) The graph shows average per-genome numbers of different SV types, i.e., large deletion, large insertion, inversion, reciprocal translocation, and complex variant.

Structural Variants.

We identified SV defined as alterations spanning more than 50 bases in the reference genome, which comprised large deletions and insertions, inversions, reciprocal translocations, and complex variants, by local de novo assembly of split reads using three software tools, SvABA (23), Manta (24), and RUFUS (25). The complex variants were defined as SV involving three or more breakpoints. All of the SV were manually confirmed by IGV. Overall, we identified 69 SV in the entire genome (not restricted to the EWC regions) of the LT-HSC-derived clones, which were not detected in matched tails; 67% of them were identified by SvABA, 88% were by Manta, 83% were by RUFUS (SI Appendix, Table S6), and collectively 52% were identified by all three tools. As described in the following section in more detail, 40 of the 69 SV were shared by the two independent LT-HSC-derived clones from the same mice exposed to 7.5- or 7.7-Gy radiation, and thus the total number of unique SV was 49. Of the 49 unique variants, 20 (41%) were large deletions, 13 (27%) were large insertions, and 8 (16%) were inversions, whereas the numbers of reciprocal translocations and complex variants were both four (8%) (SI Appendix, Tables S7–S11 and Figs. S7–S11). Those SV were examined for authenticity by Sanger sequencing of PCR products spanning breakpoint junctions. Overall, 43 (88%) of the 49 unique variants were verified by at least one Sanger-validated breakpoint junction (SI Appendix, Figs. S12 and S13 and Dataset S1).

The size of the 20 unique large deletions ranged from 52 bases to 5.0 Mbp with the median of 86 bases (SI Appendix, Table S7 and Fig. S7), and the size was unrelated to the radiation dose. Of the 13 unique large insertions, 11 (85%) were insertions of retroelements, which comprised ten LINE1 and one SINE insertions, while the other two were a 61-base tandem duplication and a 227-base insertion of a gamma-satellite sequence (SI Appendix, Table S8 and Fig. S8). Of the 11 retroelement insertions, nine LINE1 and one SINE insertions had both poly(A) tracts and target-site duplications (TSD), which are typical features of endonuclease mediated insertion (26), with undetermined insertion sizes, whereas the other was a 217-base partial LINE1 insertion. The size of the eight unique inversions ranged from 681 bases to 15.9 Mbp with the median of 2.1 Mbp, and the size seemed unrelated to the radiation dose (SI Appendix, Table S9 and Fig. S9). Of the four unique complex variants, two were interstitial translocations where deleted fragments were inserted into other chromosomes, and the other two were inversions associated with a reciprocal translocation or an insertion of a minor satellite sequence (SI Appendix, Table S11 and Fig. S11). Of the 38 unique SV except retroelement insertions that are generally mediated by RNA intermediates, 20 (53%) had pairs of 2- to 4-base microhomology sequences adjacent to breakpoints in the reference sequences, suggesting involvement of both NHEJ and MMEJ repair mechanisms for DNA double-strand breaks induced by X-irradiation in the generation of those SV, similarly to the non-repeat deletions.

The overall spontaneous mutation rate for SV in nonirradiated mice was 0.83 ± 1.46 per genome (Fig. 4 A and C and SI Appendix, Table S12). Those spontaneous mutations were four LINE1 insertions and one large deletion, identified in six independent LT-HSC-derived clones. By X-irradiation, a statistically significant and dose-dependent increase in SV was observed, leading to a 9.0-fold increase with fractionated 7.7-Gy radiation. Among those mutations, large deletions increased up to 24-fold with statistical significance (SI Appendix, Fig. S14), whereas large insertions increased only up to 2.3-fold without statistical significance. In contrast to these unbalanced mutations (indels), balanced mutations (inversions and reciprocal translocations) and complex variants were identified exclusively in irradiated mice, but did not show statistically significant association with the radiation dose. Because the retroelement insertions were identified even in nonirradiated mice and not significantly induced by X-irradiation, they did not seem characteristic of radiation-induced somatic mutations, which is consistent with the current notion that retroelement insertion is mediated by RNA intermediates, rather than by DNA repair processes. Thus we examined mutation rates for SV except retroelement insertions; the rate was 0.17 ± 0.37 per genome in nonirradiated mice, and was induced by X-irradiation in a statistically significant and dose-dependent manner up to 36-fold with fractionated 7.7-Gy radiation (Fig. 4A).

Commonality of Somatic Mutations among Multiple LT-HSCs.

All the somatic mutations identified in LT-HSC-derived clones from nonirradiated or 3.8-Gy irradiated mice were unique and specific to each clone, and did not show any commonality. However, to our surprise, the majority of somatic mutations identified in mice exposed to single-dose 7.5-Gy or fractionated 7.7-Gy radiation were shared by the two independent clones, yet substantial portions of mutations were unique to each clone. The proportions of mutations shared by the two independent clones varied depending on the mutation type. In the bar graphs in Fig. 5A, red portions represent shared mutations that fully satisfied the criteria for small somatic mutations (see Materials and Methods) in both of the two clones with the minimum variant allele fractions (VAFs) of 30% in autosomes or 60% in sex chromosomes (defied here as tier-1 shared mutations), whereas yellow portions represent shared mutations (defined as tier-2) that satisfied the criteria with only one of the two clones and showed VAFs of 15 to 30% in autosomes or 30 to 60% in sex chromosomes with the other clone (thus not called as somatic mutations on their own). In the mice exposed to 7.5 or 7.7-Gy radiation, 59 to 84% (combining tier-1 and -2 shared mutations) of SNVs, small insertions, and repeat deletions were shared by the two independent clones, whereas 99 to100% of non-repeat deletions and multisite mutations were shared by the two clones. The proportions of tier-2 shared mutations in the total shared mutations (combining tier-1 and -2) should approximate the false-negative ratios of our somatic mutation calling by WGS, which were 1.2% for SNVs, 7.5% for short insertions, 3.9% for repeat deletions, 7.5% for non-repeat deletions, and 3.0% for multisite mutations.

Fig. 5.

Commonality of somatic mutations among multiple LT-HSCs. (A) The graphs show the commonality of various somatic mutations between the two independent LT-HSC-derived clones from each mouse. In each bar, the blue portion represents the average per-genome number of mutations unique to each clone (i.e., not shared by the two clones), and the red portion represents the average number of shared mutations that fully satisfied the criteria for somatic mutations in both of the two clones (defied as tier-1 shared mutations), whereas the yellow portion represents the average number of mutations that satisfied the criteria in only one of the two clones and showed VAFs of 15 to 30% in autosomes or 30 to 60% in sex chromosomes in the other clone (defied as tier-2 shared mutations). Error bars represent SD. The same sequencing data from the same clones as in Fig. 1 were analyzed. fr7.7 denotes fractionated 7.7-Gy radiation. (B) The grid-like diagrams show results of PCR-amplicon sequencing as to whether or not randomly selected SNVs or non-repeat deletions identified by WGS in either of the two independent LT-HSC-derived clones were shared by other clones from the same mouse. The clone numbers are shown at the top of the diagrams where the two clones subjected to WGS are indicated as 1W and 2W. On the left side are the numbers for primer pairs used to amplify the SNVs or deletions by PCR (Dataset S1). For each clone, SNVs or deletions detected by the primers are indicated with different colors; SNVs and deletions unique to one of the two WGS clones are indicated by orange and green, respectively, while those shared by the two clones are indicated by red and blue, respectively. The graphs on the right side show the proportions of cells with the SNVs or deletions in bone marrow (BM) and matched tails based on their VAFs determined by PCR-amplicon sequencing. LT-HSC-derived clones from the same mice as in Fig. 1 (denoted by 0C, 3.8C, 7.5A, and fr7.7A in SI Appendix, Table S1) were examined. (C) The graph shows the commonality of SV between the two independent LT-HSC-derived clones from each mouse. In each bar, the blue portion represents the average per-genome number of mutations unique to each clone, and the red portion represents the average number of shared variants.

The significant commonality of the somatic mutations between the two independent LT-HSCs from a mouse exposed to 7.5- or 7.7-Gy radiation indicated that the two LT-HSCs originated from a single common LT-HSC that survived the irradiation. The commonality in all three pairs of LT-HSC-derived clones from three different mice suggested that viable LT-HSCs in those mice were severely depleted by the 7.5- or 7.7-Gy irradiation, but not by the 3.8-Gy irradiation. Based on the presumption that the two LT-HSCs originated from a single LT-HSC, we assume that most of the somatic mutations unique to each LT-HSC from mice exposed to 7.5 or 7.7-Gy radiation occurred after the radiation exposure, whereas those shared by the two independent clones were introduced directly by irradiation, or occurred spontaneously sometime before the exposure such as during developmental or hematopoietic processes. These notions are consistent with the finding that the non-repeat deletions and multisite mutations, which are barely detectable in nonirradiated mice, were almost all shared by the paired clones, indicating that such mutations hardly occurred after the radiation exposure, supporting our contention that these mutation types are highly specific to radiation.

To interrogate our presumption that the LT-HSCs from mice exposed to 7.5- or 7.7-Gy radiation originated from a single LT-HSC which survived the irradiation, we examined whether or not the SNVs and non-repeat deletions identified by WGS were shared by other LT-HSCs isolated from the same mice by PCR-amplicon sequencing for randomly selected mutations (Fig. 5B and SI Appendix, Fig. S15). In nonirradiated or 3.8-Gy irradiated mice, the majority of SNVs identified by WGS existed exclusively in the clones in which they were identified. However, three out of 14 SNVs (21%) identified in a nonirradiated mouse, and two out of eight SNVs (25%) identified in a 3.8-Gy irradiated mouse also existed in clones other than the paired clones subjected to WGS. These data indicated that some of the somatic mutations were shared by multiple LT-HSCs even in nonirradiated mice, and we call those mutations “shared mutations.” We then deduced the proportions of cells sharing the SNVs based on their VAFs in bone marrow, spleens, and matched tails measured by PCR-amplicon sequencing. The VAFs of those shared autosomal SNVs in bone marrow were 2.3 to 7.5%, and in matched tails were 0.1 to 9.6%, and thus the deduced cellular proportions were 4.6 to 15% in bone marrow and 0.2 to 19% in tails (Dataset S1). In contrast, the VAFs of all other SNVs were below 1.1% in bone marrow and below 0.5% in tails. Those shared SNVs probably occurred during normal development or hematopoiesis, thus presenting shared clonal origins of some LT-HSCs even in nonirradiated mice (SI Appendix, Fig. S16). In contrast to SNVs, all of the non-repeat deletions identified in nonirradiated or 3.8-Gy irradiated mice existed exclusively in the clones in which they were identified, suggesting that non-repeat deletions hardly occur spontaneously during normal development or hematopoiesis in contrast to SNVs.

In mice exposed to single-dose 7.5-Gy or fractionated 7.7-Gy radiation (Fig. 5B and SI Appendix, Fig. S15), the majority of SNVs identified by WGS in only one of the paired clones existed exclusively in the clones in which they were identified except for five SNVs out of 23 (22%), which is consistent with the assumption that most of the SNVs unique to one of the two clones occurred after the irradiation (SI Appendix, Fig. S16). In contrast, all the SNVs and non-repeat deletions shared by the paired clones from the same mice existed in large fractions of the clones examined. These data are consistent with the notion that LT-HSCs were severely depleted by the irradiation, and that the majority of isolated LT-HSCs originated from a single LT-HSC that survived the irradiation. In the mice exposed to fractionated 7.7-Gy radiation, cellular proportions of the shared SNVs or non-repeat deletions were 6.1 to 30% in bone marrow and 1.2 to 17% in spleens (Dataset S1). These cellular proportions were relatively consistent with the proportions of the LT-HSCs having shared mutations, which is 38 to 71% (5 or 8 clones out of 13 in one mouse, and 12 clones out of 17 in the other). However, in the mouse exposed to single-dose 7.5-Gy radiation, cellular proportions of shared mutations were only 1.0 to 2.4% in bone marrow and 1.2 to 3.4% in spleen, although almost 100% of the LT-HSCs had those shared SNVs (13 or 14 clones out of 14). The reason for the lower cellular proportions of shared mutations in bone marrow and spleen in the mouse exposed to single-dose 7.5-Gy irradiation at 8 wk of age may be the greater persistence of downstream progenitors such as multipotent progenitors (MPPs) that originated from preirradiation LT-HSCs that were lost by the irradiation and thus unrelated to the isolated LT-HSCs. In contrast, fractionated 7.7-Gy radiation beginning at 5 wk of age to span 3 wk might have reduced such downstream progenitors derived from preirradiated LT-HSCs more severely, thereby resulting in the higher cellular proportions of shared mutations. Such interpretations could be consistent with the in vivo dynamics and residence times of mouse hematopoietic stem cells (HSCs) and MPPs (27).

Similarly to the small mutations, all the SV identified in the nonirradiated or 3.8-Gy irradiated mice were unique to each clone, but most (91 to 100%) of SV identified in the mice exposed to single-dose 7.5- or fractionated 7.7-Gy radiation were shared by the two independent clones (Fig. 5C) suggesting that those SV were mostly introduced directly by the irradiation, not after the irradiation.

Frequencies and Dose–Response Profiles of Somatic Mutations.

The average overall spontaneous somatic mutation rate for all the mutation types comprising SNVs, small indels, multisite mutations, and SV in LT-HSCs from nonirradiated mice was 133.3 ± 21.7 per genome (26.4 ± 4.3 per 10⁹ nucleotides), in which the number of multisite mutations represents the number of clusters involving multiple alterations (Fig. 6A). The number of overall somatic mutations was increased by irradiation in a dose-dependent manner up to 2.9-fold. Proportions of SNVs and total small deletions among the overall mutations did not significantly change by irradiation, but that of small insertions slightly decreased. Among the small deletions, the proportion of non-repeat deletions including microhomology deletions significantly increased by irradiation, but that of repeat deletions decreased in contrast. Proportions of multisite mutations and SV also significantly increased by irradiation. These data indicate that the spectra and characteristics of somatic mutations induced by X-irradiation are distinct from those occurring spontaneously.

Fig. 6.

Frequencies and dose–response profiles of somatic mutations. (A) The graphs show the per-genome numbers (Left) and proportions (Right) of all types of somatic mutations, i.e., SNV, small insertion, repeat deletion, non-repeat deletion without microhomology, microhomology deletion, multisite mutation, and SV, identified in LT-HSC-derived clones for each radiation dose. Error bars represent SD. The same sequencing data from the same clones as in Fig. 1 were analyzed. fr7.7 denotes fractionated 7.7-Gy radiation. (B) The graphs show linear regression of dose–response relationship for each type of somatic mutations. Each circle represents each LT-HSC-derived clone, and red circles indicate clones from the mice exposed to fractionated 7.7-Gy radiation. Dashed lines indicate the 95% credible interval of the regression. “R” and “P” represent Pearson’s correlation coefficient and its associated two-tailed P-value, respectively. (C) The graph shows the average mutation rates of SNVs for each radiation dose across four equal quartiles of the mouse genome from early to late replicating regions based on the reference data of mouse ES cells (28). Error bars represent SD. Asterisks indicate P-values by binomial test in comparison between one quartile and the other three combined, with Holm correction for multiple comparisons: *P < 0.05, and **P < 0.01.

To further examine characteristics and radiation-sensitivity of each type of mutations, we determined mutation rates per 1-Gy dose of radiation for each mutation type by linear regression (Fig. 6B, Table 1, and SI Appendix, Tables S13 and S14). SV including or excluding retroelement insertions were analyzed separately, because most of the retroelement insertions were associated with typical features of endonuclease mediated insertion, and did not show dose-dependent increase. As a quantitative measure of radiation specificity of each mutation type, we devised “radiation specificity index” that is calculated by dividing the mutation rate per 1-Gy dose by the spontaneous mutation rate determined with nonirradiated mice. The radiation specificity index was highest with non-repeat deletions, especially those without microhomology, followed by microhomology deletions, SV except retroelement insertions, and then multisite mutations, indicating that these are mutational signatures of ionizing radiation. On the contrary, the index was lowest with repeat deletions, followed by small insertions and then SNVs.

It has been reported that the SNV rate is higher in late replicating regions in human and mouse stem cells of other tissues (7, 8). In order to gain insights into spontaneous and X-ray-induced mutational processes in LT-HSCs, we compared SNV rates across four equal quartiles of the mouse genome from early to late replicating regions using the reference data of mouse ES cells (28). The spontaneous SNV rate in LT-HSCs from nonirradiated mice was significantly higher in the late region (fourth quartile), consistent with the studies with other tissue stem cells (7, 8), and was significantly lower in the second quartile (Fig. 6C). The plausible mechanism for the variable mutation rates across the replicating regions is the differential activities of DNA mismatch repair (MMR) and nucleotide excision repair (NER), which are both lower in the late S-phase (29, 30). The similar bias in the SNV rate was also observed in LT-HSCs from X-irradiated mice with varying statistical significance, suggesting that SNV rates are determined by the balance between the amounts of DNA damages or replication errors and the differential activities of MMR and NER during the S-phase. Small indels also showed the similar trend of higher mutation rates in late replicating regions in LT-HSCs from nonirradiated or X-irradiated mice with less statistical significance compared to SNVs (SI Appendix, Fig. S17), suggesting that DNA damages responsible for small indels may be less amenable to MMR or NER than damages causing SNVs.

Discussion

In this study, we elucidated types and frequencies of spontaneous somatic mutations in mouse LT-HSCs, and impacts of whole-body X-irradiation on those mutations. In nonirradiated 16-wk-old mice, average spontaneous mutation rates of SNVs, small deletions, and insertions were 55.0, 42.5, and 33.8 per genome, respectively, and they were increased in a dose-dependent manner up to twofold to threefold by 3.8 to 7.7-Gy X-irradiation. While base substitution patterns in spontaneous SNVs were dominated by base transitions, X-irradiation significantly increased C > A and T > A transversions and the C > T transition at non-CpG sites, which may be introduced by ROS via oxidative DNA base damages. The most common oxidative base damage is 8-oxo-guanine that can mispair with adenine, leading to the C > A (i.e., G:C-to-T:A) transversion (13). The T > A (A:T-to-T:A or vice versa) transversion can be introduced either through another oxidatively damaged purine, 2-oxo-adenine, that can mispair with adenine, or through oxidatively damaged thymine such as 5-formyluracil that can mispair with thymine. The C > T (C:G-to-T:A) transition can be introduced through oxidatively damaged cytosine such as uracil glycol and 5-hydroxypyrimidines that can mispair with adenine (14). Moreover, the C > A, T > A, and non-CpG-site C > T substitutions coincide with the somatic mutations identified by the gpt mutation assay as highly induced SNVs in mice administered with potassium bromate, a ROS-inducing agent (12). These notions collectively suggest that these three base substitutions are characteristic of ROS-induced mutations, and suggest a causal role of ROS in X-ray mutagenesis. While SBS signatures of spontaneous SNVs were attributable to SBS5 (75%) and SBS1 (25%) signatures, both known as clock-like signatures (17), X-irradiation markedly diminished SBS5 and reciprocally increased SBS40, demonstrating a strong link between SBS40 and radiation.

We subdivided small deletions into two types; one is the repeat deletion, i.e., shrinkage of tandem repeats, which accounted for most of spontaneous deletions, and the other is the non-repeat deletion that occurred out of tandem repeats, and was specifically induced by X-irradiation. Over a quarter of non-repeat deletions were associated with two- to five-base microhomology sequences, and thus subcategorized into microhomology deletions, which suggested involvement of not only NHEJ but also MMEJ repair mechanisms for radiation-induced DNA damages (18, 19). We further examined multisite mutations, having two or more alterations (base substitutions or small indels) in close proximity, and SV comprising large indels, inversions, reciprocal translocations, and complex variants identified by local de novo assembly of split reads. The radiation-sensitivity of each mutation type was evaluated by the mutation rate per 1-Gy dose estimated by linear regression, and then the radiation-specificity of each mutation type was determined by dividing the per-Gy mutation rate by the spontaneous rate. The radiation-specificity was highest with non-repeat deletions, particularly those without microhomology, followed by microhomology deletions, SV except retroelement insertions, and then multisite mutations. These types of somatic mutations were thus revealed as mutational signatures of ionizing radiation.

Previous studies employing WGS to interrogate mutational signatures of ionizing radiation in somatic cells have been limited to a couple of studies on radiation-associated malignancies in humans. Behjati et al. reported that second malignancies after radiotherapy have higher numbers of small deletions and balanced inversions, compared to radiation-naive tumors (31). More recently, the genome analysis of papillary thyroid carcinomas after exposure to radioactive iodine from the Chernobyl accident demonstrated a dose-dependent increase in the number of small and large deletions, and simple balanced SV, but did not identify any unique radiation-related biomarker (32). Those studies investigated somatic mutations in human cancers that acquired clonality by malignant transformation, and thus they could include not only mutations directly introduced by radiation but also mutations acquired during and after malignant transformation. Thus, somatic mutations identified in radiation-associated cancers could be affected by genetic or pathophysiological characteristics of those cancers, which are often linked with genomic instability leading to or resulting from malignant transformation. Despite such possible biases associated with cancer genomes, their findings are partly consistent with and thus supporting our findings that the numbers of small and large deletions and inversions increased in a dose-dependent manner. However, in contrast to their findings, our data indicated that the dose–response of SNVs is greater than that of small deletions, which may be ascribed to either difference between normal and transformed cells, or species difference.

There are a couple of previous mouse studies of radiation-induced germline mutations by whole-genome analysis. Adewoye et al. reported that de novo mutation rates of small indels and multisite mutations, identified by WGS, and copy number variants (CNVs) (i.e., large deletions and duplications), identified by microarray comparative genome hybridization, were significantly increased in the offspring of male mice exposed to 3-Gy X-rays, whereas the SNV rate was not significantly affected (20). In their study, while the spontaneous de novo mutation rates in the control mice were 1.17 small indels, 0.167 multisite mutations, and 0.0108 CNVs per genome, the de novo mutation rates in the offspring of exposed males were 2.83 small indels, 1.50 multisite mutations, and 0.0828 CNVs (2.4-, 9.0-, and 7.7-fold increases, respectively). The significant increases of de novo germline small indels and multisite mutations were also observed in the offspring of female mice as well as males, which were exposed to 4-Gy γ-rays (22). In contrast to these germline mutation studies, our study of somatic mutations found that SNVs and various SV are also significantly induced by ionizing radiation, not only those identified by the germline studies (i.e., small indels, multisite mutations, large deletions, and duplications), and identified non-repeat deletions and SV except retroelement insertions as the mutation types of the highest radiation-specificity, exceeding even multisite mutations.

Although somatic mutations identified in LT-HSC-derived clones from nonirradiated or 3.8-Gy irradiated mice were mostly unique to each clone, the majority of somatic mutations in 7.5- or 7.7-Gy irradiated mice were shared by large fractions of LT-HSCs, and by up to 30% of bone marrow cells and 17% of spleen cells. We assume that those postirradiation LT-HSCs with shared somatic mutations originated from a single common LT-HSC which survived the irradiation that severely depleted LT-HSCs, and then expanded in vivo to confer marked clonality to the entire hematopoietic system. Such dominance of a clonal cell population deriving from a single LT-HSC might be a consequence of an in vivo selection due to selective advantages of the LT-HSC over other HSCs for survival or propagation. A possible underlying mechanism for the putative in vivo selection may be that the somatic mutations introduced by radiation conferred selective advantages to the particular LT-HSC and its progenies, or on the contrary the particular LT-HSC had only neutral or less deleterious mutations compared to other LT-HSCs, leading to the phenomenon like “survival of the fittest.” However, several modeling studies have shown that if the size of an HSC pool is constrained, a large fraction of HSCs will eventually derive from a single HSC clone as a result of neutral drift without any selective advantages (33, 34). The simulation study further indicated that clonal expansion by neutral drift would occur more rapidly when the size of an HSC pool is smaller, or when the frequency of self-renewal cell divisions of HSCs to produce two daughter HSCs is higher. Certainly, by higher dose irradiation, the HSC pool size should have drastically diminished, and then expanded rapidly during the recovery of hematopoiesis possibly by increasing the self-renewal symmetrical cell divisions of HSCs. Thus the dominance of a clonal cell population deriving from a single HSC might be a consequence of neutral drift accelerated by radiation exposure, rather than selective advantages of certain LT-HSCs due to particular somatic mutations. Actually, we did not find any protein-altering mutations in mouse orthologs of the 92 genes implicated in myeloid neoplasms in humans (35).

Of the two mice exposed to fractionated 7.7-Gy radiation, maximal proportions of LT-HSCs sharing somatic mutations were 62% (8 out of 13 clones) in one mouse (fr7.7A) and 71% (12 out of 17 clones) in the other (fr7.7B) (Fig. 5B and SI Appendix, Fig. S15). In contrast, 100% (14 out of 14 clones) of LT-HSCs from the mouse exposed to single-dose 7.5-Gy radiation shared mutations. This difference in the fraction of LT-HSCs with shared mutations may be due to a greater reduction of a LT-HSC pool size by the single high-dose irradiation, and a higher demand for LT-HSCs to undergo self-renewal symmetric cell divisions for rapid hematopoietic recovery as compared to the fractionated irradiation. The consequential loss of clonal diversity of LT-HSCs by the single-dose irradiation might have at least in part contributed to the much higher lethality compared to the fractionated irradiation. The loss of clonal diversity of LT-HSCs might be also causally related to the lower cellular proportions of shared mutations in bone marrow and spleen, possibly due to impaired differentiation or proliferation of LT-HSCs or their progenies.

One (fr7.7A) of the two mice exposed to fractionated 7.7-Gy radiation presented two distinct groups of LT-HSC-derived clones with shared mutations (Fig. 5B and SI Appendix, Fig. S16); one group (clones 1W, 2W, and 3 to 5) had No. 1 to 3 of the shared SNVs, but the other group (clones 6 to 8) did not, indicating that the former represented a subclone diverged during or after radiation exposure. Furthermore, the mutations shared by those clones could be subdivided into three groups by the proportion of cells with those mutations in bone marrow: one group of 6.1 to 7.0% (No. 1 to 3 of the shared SNVs), another of 21 to 25%, and the other of 28 to 30%. Such a gradation in cellular proportions also suggests the presence of hierarchical subclones in bone marrow, which might be generated by multiple exposures of this mouse to fractionated irradiation. These data indicate that comprehensive sequencing analysis of somatic mutations in progenitor-derived clones is an effective approach to study in vivo clonal dynamics of the progenitors and their progenies not only in steady-state hematopoiesis but also in perturbed hematopoiesis induced by ionizing radiation or other agents, and in disease conditions.

In conclusion, we elucidated whole spectra and frequencies of somatic mutations in mouse LT-HSCs by WGS, and impacts of whole-body X-irradiation on those mutations. We thereby determined the sensitivity and specificity of each type of mutations to X-irradiation, and identified mutational signatures of ionizing radiation. These findings should advance our understanding of underlying molecular and cellular mechanisms for chronic radiation toxicities, particularly neoplasm development, and enhance our efforts to identify biomarkers for radiation injuries.

Materials and Methods

Animals and X-Irradiation.

Male C57BL/6J mice were purchased from Japan SLC, Inc, and exposed to X-rays with an animal irradiator (Faxitron CP-160) at a dose rate of 0.963 Gy/min: the total dose of 3.756 Gy (3.9-min exposure, abbreviated as 3.8 Gy) or 7.511 Gy (7.8 min, abbreviated as 7.5 Gy) at 8 wk of age, or the total dose of 7.704 Gy (abbreviated as 7.7 Gy) in four equal fractions of 1.926 Gy (2.0 min, abbreviated as 1.9 Gy) per week starting at 5 wk and ending at 8 wk of age (Fig. 1A). At 16 wk of age, X-irradiated mice (8 wk after irradiation) or nonirradiated mice were euthanized for isolation of LT-HSCs. All animal experiments were approved by the Experimental Animal Care Committee of the Radiation Effects Research Foundation (Approval No. RP P3-19), and performed in accordance with the institutional Guidelines for Animal Experiments.

Clonal Culture of LT-HSCs.

Each single LT-HSC in each well was cultured with the MethoCult GF M3434 (Stemcell) medium for 7 to 9 d for clonal growth in a colony. Cells grown in isolated colonies were then replated in 35-mm dishes with the same medium for further clonal growth. Cells were then collected for DNA isolation 5 to 6 d after the replating.

DNA Isolation and WGS.

Genomic DNA was extracted from clonal cell populations derived from single LT-HSCs and matched tails using the smart DNA prep (m) (Analytik Jena) or Monarch Genomic DNA Purification Kit (New England Biolabs). Libraries were prepared with DNA samples without PCR amplification using the TruSeq DNA PCR-Free Library Prep Kit (Illumina), and sequenced on a NovaSeq 6000 System (Illumina) to obtain paired-end 150-bp reads. The average total coverage depth was 36.7 reads for LT-HSC-derived clones, and 40.6 for tails (SI Appendix, Table S2).

Mapping and Variant Calling.

Sequence reads were mapped to the mouse reference genome (UCSC mm10) using BWA-MEM v.0.7.17 with the “−M” option compatible with Picard v2.18.26 (broadinstitute.github.io/picard) used to remove PCR duplicates. To minimize false variant calls, we used only high-mapping-quality reads defined as MQ60 reads that met the following conditions: 1) properly mapped according to the aligner, 2) having a minimum mapping quality of 60 by SAMtools-1.9 (samtools view -q 60 -f 0 × 2 -F 0 × 500), and 3) mapped to the reference without clipping. Sequence alterations from the mouse reference genome were called as genomic variants using GATK v4.1.0.0 HaplotypeCaller with a minimum base quality of 20.

Definition of EWC Regions.

For accurate identification and comparison of somatic mutations across all the samples, we defined EWC regions (10, 11) that satisfied the following criteria: 1) the MQ60 read depths were 50% to 300% of the peak coverage for each chromosome; 2) the depth ratio of MQ60 reads to all mapped reads was at least 80% at each site, and 3) the minimum base quality was 20 by SAMtools. The EWC regions were first defined for each sample, and then the regions shared by all the samples were used for mutation analyses except for SV analysis. The MQ60 read depths in the shared EWC regions were 13 to 132 reads for autosomes, and 5 to 60 for sex chromosomes. The EWC regions covered 76.4% of the mouse whole diploid genome (SI Appendix, Table S3).

Identification of Small Somatic Mutations.

Small somatic mutations in LT-HSC-derived clones were determined from variants called by GATK HaplotypeCaller within the EWC regions by subtracting germline variants identified in bulk DNA from matched tails. Somatic mutations in autosomes were called by filtering with following inclusion criteria: 1) a VAF less than 0.05 in all the tail DNA samples, 2) a VAF of at least 0.3 in a clone, and 3) a total allele count of at least 10 in all the clones and matched tails. For sex chromosomes, the following criteria were applied: 1) a VAF less than 0.1 in all tails, 2) a VAF of at least 0.6 in a clone, and the same 3) as for autosomes. Using these filters, we identified SNVs and small indels as candidate somatic mutations, which were then confirmed by visual inspection with IGV. The somatic mutation rate, represented as a mutation number per whole mouse genome, was calculated from the number of mutations within the EWC regions, except for SV.

Analysis of Structural Variants.

Structural variants were identified from paired-end 150-bp reads by genome-wide local de novo assembly of split reads using software tools, SvABA v1.1.3 (23), Manta v1.6.0 (24), and RUFUS (25). Structural variants in SvABA were called by following inclusion criteria: 1) an allele depth (AD) less than two in each tail DNA sample, 2) an AD of more than five in a clone, and 3) an AD of 0 in all clones from other mice. Structural variants in Manta were called by following inclusion criteria: 1) a split reads for alternative allele (SRA) less than two in each tail DNA sample, 2) an SRA of four or more in a clone, and 3) an SRA of 0 in all clones from other mice. Structural variants in RUFUS were called by following criteria: 1) SVTYPE INFO in a clone, or 2) the other INFO with ALT_LEN of 50 or more in a clone, or 3) the other INFO with REF_LEN of 50 or more in a clone. All structural variants identified in LT-HSC-derived clones but not in matched tails were manually validated by IGV. The analysis was not limited to the EWC regions.

Statistical Methods.

We performed the exact binomial test to examine whether mutation rates were equivalent between mice with different X-ray doses. In this test, we assumed that the total number of mutations in each dose group was distributed according to Poisson distribution and tested if their mean numbers per clone were equivalent. We also tested Pearson’s correlation between mutation rates and radiation doses, and performed Wilcoxon rank sum test to examine if sizes of short indels were equivalent between mice with different X-ray doses. We calculated two-tailed P-values for the above tests, and P-values <0.05 with Holm correction, when necessary, were considered significant. We performed the above tests using R version 4.0.3 (R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/). We also estimated mutation rates per 1-Gy dose by linear regression, assuming that the number of mutations was distributed according to Poisson distribution. We assumed the following linear model because the data showed linear dose–response:

$$y_i \sim \mathrm{P}\mathrm{o}\mathrm{i}\mathrm{s}\mathrm{s}\mathrm{o}\mathrm{n} \left({\beta }_0+{\beta }_1{d}_i\right),$$

where $y_i$ is the number of mutations for ith clone, ${\beta }_0\ge 0$ is the background mutation rate, ${\beta }_1$ is the induced mutation rate per 1-Gy dose, and $d_i$ is the radiation dose for ith clone. To avoid instability when ${\beta }_0$ was in neighbor of 0, we estimated ${\beta }_0$ and ${\beta }_1$ using Bayesian analyses fit with stan version 2.27 (Stan Development Team, http://mc-stan.org/). We also assumed uniform prior distributions on $[0, \infty ]$ and $\left(-\infty , \infty \right)$ for ${\beta }_0$ and ${\beta }_1$ , respectively.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Data, Materials, and Software Availability

All sequence data are deposited and available in the DDBJ Sequence Read Archive (DRA; https://www.ddbj.nig.ac.jp/dra/index.html) with accession no. DRR412397 to DRR412423 (36). All other study data are included in the main text and SI Appendix.