ARTICLE
Received 7 Sep 2015 | Accepted 13 Jul 2016 | Published 12 Sep 2016
DOI: 10.1038/ncomms12605
OPEN
Mutational signatures of ionizing radiation
in second malignancies
Sam Behjati1,2,*, Gunes Gundem1,*, David C. Wedge1,3, Nicola D. Roberts1, Patrick S. Tarpey1,
Susanna L. Cooke1, Peter Van Loo4,5, Ludmil B. Alexandrov1, Manasa Ramakrishna1, Helen Davies1,
Serena Nik-Zainal1, Claire Hardy1, Calli Latimer1, Keiran M. Raine1, Lucy Stebbings1, Andy Menzies1, David Jones1,
Rebecca Shepherd1, Adam P. Butler1, Jon W. Teague1, Mette Jorgensen6, Bhavisha Khatri7, Nischalan Pillay6,7,
Adam Shlien1,8, P. Andrew Futreal1,9, Christophe Badie10, ICGC Prostate Groupz, Ultan McDermott1,
G. Steven Bova11, Andrea L. Richardson12,13,w, Adrienne M. Flanagan6,7, Michael R. Stratton1
& Peter J. Campbell1,14
Ionizing radiation is a potent carcinogen, inducing cancer through DNA damage. The signatures
of mutations arising in human tissues following in vivo exposure to ionizing radiation have not
been documented. Here, we searched for signatures of ionizing radiation in 12 radiation-associated second malignancies of different tumour types. Two signatures of somatic mutation
characterize ionizing radiation exposure irrespective of tumour type. Compared with 319
radiation-naive tumours, radiation-associated tumours carry a median extra 201 deletions
genome-wide, sized 1–100 base pairs often with microhomology at the junction. Unlike deletions
of radiation-naive tumours, these show no variation in density across the genome or correlation
with sequence context, replication timing or chromatin structure. Furthermore, we observe a
significant increase in balanced inversions in radiation-associated tumours. Both small deletions
and inversions generate driver mutations. Thus, ionizing radiation generates distinctive mutational signatures that explain its carcinogenic potential.
1 Cancer Genome Project, Wellcome Trust Sanger Institute, Wellcome Trust Genome Campus, Hinxton, Cambridgeshire CB10 1SA UK. 2 Department of Paediatrics, University of
Cambridge, Cambridge CB2 0QQ, UK. 3 Oxford Big Data Institute and Oxford Centre for Cancer Gene Research, Wellcome Trust Centre for Human Genetics, Roosevelt Drive,
Oxford OX3 7BN, UK. 4 The Francis Crick Institute, London WC2A 3LY, UK. 5 Department of Human Genetics, University of Leuven, Leuven B-3000, Belgium. 6 University College
London Cancer Institute, Huntley Street, London WC1E 6BT, UK. 7 Histopathology, Royal National Orthopaedic Hospital NHS Trust, Stanmore, Middlesex HA7 4LP, UK.
8 Department of Paediatric Laboratory Medicine, The Hospital for Sick Children, Toronto, Ontario, Canada M5G 1X8. 9 Department of Genomic Medicine, MD Anderson Cancer
Center, University of Texas, Houston, Texas 77030, USA. 10 Cancer Mechanisms and Biomarkers Group, Radiation Effects Department, Centre for Radiation Chemical and
Environmental Hazards, Public Health England, Chilton, Didcot OX11 0RQ, UK. 11 Institute of Biosciences and Medical Technology, BioMediTech, University of Tampere and Fimlab
Laboratories, Tampere University Hospital, Tampere FI-33520, Finland. 12 Dana-Farber Cancer Institute, Boston, Massachusetts 02215-5450, USA. 13 Brigham and Women’s
Hospital, Harvard Medical School, Boston, Massachusetts 02115 USA. 14 Department of Haematology, University of Cambridge, Hills Road, Cambridge CB2 2XY, UK. * These
authors contributed equally to this work. w Present address: Sibley Memorial Hospital, Johns Hopkins Medicine, Washington, District Of Columbia 20016, USA. Correspondence
and requests for materials should be addressed to P.J.C. (email: pc8@sanger.ac.uk).
zA full list of consortium members appears at the end of the paper.
NATURE COMMUNICATIONS | 7:12605 | DOI: 10.1038/ncomms12605 | www.nature.com/naturecommunications
1
ARTICLE
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12605
E
xposure to ionizing radiation increases the risk of
subsequent cancer. This risk exhibits a strong dose–
response relationship, and there appear to be no safe limits
for radiation exposure1. This association was first noted by March
who observed an increased incidence of leukaemia amongst
radiologists2. A leading cause of radiation-induced cancers
appears to be exposure to medical radiation, either in the form
of radiotherapy for an unrelated malignancy3 or diagnostic
radiography4,5. These iatrogenic tumours arise as de novo
neoplasms in a field of therapeutic radiation after a latency
period that can span decades6, and are not recurrences of the
original cancer7.
Many, but not all, environmental carcinogens induce cancer by
increasing the rate of mutation in somatic cells. The physicochemical properties of a given carcinogen govern its interaction
with DNA, leading to recurrent ‘signatures’ or patterns of
mutations in the genome. These can be reconstructed either from
experimental model systems8,9 or from statistical analyses of
cancer genomes in exposed patients10–12. Ionizing radiation
directly damages DNA, and can generate lesions on single bases,
single-stranded nicks in the DNA backbone, clustered lesions at
several nearby sites and double-stranded DNA breaks13. In
experimental systems exposed to radiation, including the murine
germline and Arabidopsis thaliana cells, ionizing radiation can
cause all classes of mutations, with possible enrichment of
indels14–22. Targeted gene screens in radiation-induced sarcoma
have indicated an increased burden of deletions and substitutions
with frequent inactivation of TP53 and RB1 (refs 23–25). In
addition, a transcriptome profile that represents a state of chronic
oxidative stress has been proposed to be specific to radiationassociated sarcoma26.
We studied the genomes of 12 radiation-associated second
malignancies of four different tumour types: osteosarcoma;
spindle cell sarcoma; angiosarcoma; breast cancer. These were
secondary tumours that arose within a field of therapeutic
ionizing radiation and were not thought to be recurrences of the
original malignancy treated with radiation. We chose this
experimental design for several reasons: the tumours are classic
radiotherapy-induced cancers with high attributable risks for the
radiation exposure; the radiation exposure occurs over a short
time period relative to the evolution of the cancer; and the
mutational signatures of sporadic breast cancers and sarcomas
have been well documented10,27–29. It should be noted that in the
absence of biomarkers, a diagnosis of a tumour being radiationinduced cannot be definitively made (see Supplementary Note 1
for clinical details and further discussion).
We subjected these 12 tumours, along with normal tissues from
the same patients, to whole-genome sequencing and obtained
catalogues of somatic mutations. We compared our findings to
319 radiation-naive breast cancers and sarcomas processed by the
same sequencing and bioinformatics pipeline: 251 breast
tumours; 33 breast tumours with pathogenic BRCA1 or BRCA2
germline mutations; 35 osteosarcomas (see Methods for cohort
details). In addition, we validated our findings in a published
series of radiation-naı̈ve and radiation-exposed prostate tumours
from ten patients30.
The main aim of our analyses was to search for tumour-type
independent, overarching signatures of ionizing radiation. Overall
we identified two such signatures in radiation-associated second
malignancies, an excess of balanced inversions and of small
deletions.
Results
Tumour-type specific features. The 12 radiation-associated
tumours harboured 1,506–9,245 substitutions per genome
2
(median 4113), 135–943 indels per genome, (median 429) and
6–321 rearrangement break points per genome (median 74;
Supplementary Data 1–3). The observed driver mutations
followed the patterns expected for the tumour type consistently
(Supplementary Table 1). Angiosarcomas harboured PTPRB and
PLCG1 mutations. In spindle cell sarcoma and osteosarcoma
driver alteration of TP53 and CDKN2A were found. Canonical
PIK3CA mutations were seen in the radiation-associated breast
tumours. Similarly, many of the mutational signatures seen in
sporadic cancers were also present in radiation-associated
tumours, such as chromothripsis in sarcomas (Supplementary
Fig. 1). Against this backdrop of genomic diversity, we found
evidence for two mutational signatures in the radiation-associated
cancers that transcended tumour type: small deletions and
balanced inversions.
Enrichment of deletions in radiation-associated tumours.
Although the absolute burden of indels varied across the 12
radiation-associated tumours, in each tumour the indel burden
was high compared with that tumour’s substitution burden
(Fig. 1a). Compared with 319 radiation-naive tumours, the indel/
substitution ratio was significantly increased in radiationassociated tumours (P ¼ 0.0003, linear mixed effects model, see
Methods).
Deletions and insertions were not equally enriched. There was
a significant excess of deletions relative to insertions in radiationassociated second malignancies (Fig. 1b; Po2.2 10 16, linear
mixed effects model). This excess of deletions was also seen in
BRCA1 or BRCA2 germline-deficient breast tumours, as previously described27, but was not seen in radiation-naive sporadic
breast tumours or sarcomas. In each radiation-associated tumour,
the radiotherapy had been given over a relatively short time
period many years earlier. If the excess deletions we observed
were directly attributable to ionizing radiation, then the
enrichment should only be evident amongst the early, clonal
mutations and not in the late, subclonal mutations. We were able
to define subclones in three of the radiation-associated tumour
genomes. Strikingly, compared with subclonal mutations,
deletions were significantly increased compared with insertions
amongst clonal mutations in all three cases (Fig. 1c, Po0.00005;
Fisher’s exact test).
Distribution of deletions across the genome. Many mutational
signatures show uneven distribution across the genome, especially
those associated with carcinogens such as tobacco smoke and
ultraviolet light31, thought to arise due to higher order chromatin
organization and accessibility of the carcinogen and repair
proteins to DNA (ref. 32). Deletions found in radiation-naive
tumours showed considerable long-range variation in density
across the genome, as did insertions in both radiation-associated
and radiation-naive tumours, correlating with several genomic
features (Fig. 1d,e; Supplementary Data 4). In stark contrast,
deletions in radiation-induced cancers showed almost no
variability across the genome and minimal correlation with
genomic properties such as replication timing, sequence
complexity or GC content. We hypothesize that this is because
of the pervasive penetration of ionizing radiation through tissue,
meaning that its interaction with DNA is stochastic and
unaffected by higher order chromatin structure. Since small
deletions are the predominant read-out of this damage, they show
no association with the genomic features that influence other
mutational processes.
Evidence of non-homologous end-joining causing deletions. In
two aspects, the deletions of radiation-associated cancers
NATURE COMMUNICATIONS | 7:12605 | DOI: 10.1038/ncomms12605 | www.nature.com/naturecommunications
ARTICLE
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12605
non-homologous end-joining are the pathways for repairing
radiation-induced DNA damage, rather than homologous
recombination. Possible explanations for this include that
damage occurs at phases of the cell cycle when homologous
recombination pathways are less active33 or because the damaged
resembled those of BRCA1 or BRCA2 germline-deficient breast
tumours (Supplementary Fig. 2): enrichment for deletions 4
2–3 bp in length and significantly higher rates of microhomology
at the breakpoint junction (P ¼ 2 10 16, Kolmogorov–Smirnov
test)27. This similarity suggests that microhomology mediated or
*
*
Primary Breast
Primary osteo- BRCA1/ Radiation
breast sarcoma BRCA2 tumours
a
b
10
Ratio genome wide deletions / insertions
a
30
75
PD
0.4
Ratio genome wide indel / substitution burden
*
Primary Breast
Primary osteo- BRCA1/ Radiation
breast sarcoma BRCA2 tumours
0.3
0.2
0.1
8
Radiation tumours
Angiosarcoma
Breast
Spindle cell sarcoma
Osteosarcoma
6
4
2
0
0
n
251
35
33
n
12
251
35
33
12
c
PD7191a
–5
P = 1.1 x 10
500
0
0
0
Mutation likelihood
l
lo
Su
Deletions
Insertions
Chr14
20
Insertions
na
na
l
C
lo
bc
Su
C
lo
na
na
l
l
na
na
lo
lo
bc
Su
d
Deletions
l
500
l
500
Complex
P = 4.1 x 10–5
lo
P = 9.3 x 10
PD7192a
bc
–10
C
Number of mutations
PD7189a
Chr14
40
60
80
100
20
40
60
80
100
Genomic position (MB)
Non-radiation associated mutations
Radiation-associated mutations
e
Insertions
Non-radiation associated indels
Radiation-associated indels
*
*
–1
0
Background
distribution
1
Z DNA
Triplex
Telomere
Simple repeat
Short tandem repeat
Sequence complexicity
Replication timing
MIR
LTR
LAD
L2
L2
Gene
GC content
G quadruplex
DNA repeat
Direct repeat
Cruciform
CpG islands
Chromatin
Centromere
ALU
Deletions
*
*
*
*
*
*
*
*
*
–1
0
Background
distribution
1
NATURE COMMUNICATIONS | 7:12605 | DOI: 10.1038/ncomms12605 | www.nature.com/naturecommunications
3
ARTICLE
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12605
DNA ends contain structures that interfere with homologous
recombination34.
Enrichment of balanced inversions. For structural variants, we
found enrichment of a rare type of rearrangement, balanced
inversions, among radiation-associated second malignancies,
irrespective of tumour type (Table 1; Fig. 2). While rearrangements with an inverted orientation are common in cancer
genomes, they are typically unbalanced, associated with copynumber changes and caused by processes such as breakagefusion-bridge cycles35, chromothripsis29 and chromoplexy36. We
found a significant enrichment of balanced inversions in
radiation-associated cancers: 52 in 11/12 tumours compared
with 66 balanced inversions in 43 of the 286 radiation-naive
tumours studied (P ¼ 2 10 16, generalized linear model). Of
note, complete inversions were also significantly enriched
amongst BRCA1 and BRCA2 germline-deficient breast tumours
(Table 1, P ¼ 2 10 16). Balanced inversions ranged in size from
a few hundred base pairs to nearly 100 megabases, and at
breakpoint junctions showed variability in microhomology and in
non-templated sequence inserted (Supplementary Data 5).
Validation of findings in prostate tumours. To validate our
observation that deletions and balanced inversions are genomic
imprints of ionizing radiation, we examined the genomes of
primary and/or metastatic prostate tumours from ten patients,
previously published30. Five patients had developed metastases
after irradiation of the primary lesion; four patients had never
received radiation treatment; and one patient, PD11331, received
radiotherapy to the primary lesion after metastases had already
formed. Consistent with the observations made in the 12
radiation-associated second malignancies, we found a significant
enrichment of deletions in prostate cancer lesions exposed
to radiotherapy compared with radiation-naive tumours’
(P ¼ 0.0002, generalized linear model, Fig. 3a). Strikingly, in
patient PD11331, the radiation-exposed primary tumour (sample
PD11331c), but not radiation-naive metastases, exhibited a
preponderance of deletions (Fig. 3b, P ¼ 10 15, Fisher’s exact
test). Similarly, balanced inversions were enriched amongst
radiation-exposed lesions (P ¼ 0.04, generalized linear model;
Supplementary Table 2). In patient PD11331, again it was the
radiation-exposed primary tumour, but not any of the metastases,
that harboured a balanced inversion.
Driver events generated by deletions and inversions. The
oncogenic potential of a mutational process derives from its
capacity to generate driver mutations. With their absence of
copy-number effects, functional consequences of balanced
inversions most commonly arise from genes broken at either end
of the inversion, notwithstanding the possibility of long-range
gene-enhancer disruption. In our data, 48/104 inversion break
points disrupted or fused genes (Supplementary Data 5), with one
forming a driver mutation through disruption of TP53. For the
mutational signature of small deletions that we observe, we estimate that the median excess of indels in the radiation-induced
cancers sequenced here is 201 indels per genome (linear mixed
effects model, s.d. 348 indels). Among these are a 14 base pair
deletion in CASP8 and a 4 base pair deletion in TP53, both
disrupting essential splice sites and thus generating driver events.
Discussion
Overall we identified two genomic imprints of ionizing radiation,
an excess of deletions and of an exceedingly rare type of
rearrangement, balanced inversions. The validity of our study
may be limited by the overall number of tumours we examined
and the small number of each tumour type. Yet it would seem
unlikely that the enrichment in radiation-associated tumours of
deletions and of balanced inversions occurred by chance. This
view is supported by our statistical analyses as well as the fact that
the signatures were tumour-type independent. Both signatures
were present across four different tumour types and could be
validated in a cohort of radiation-exposed prostate cancer lesions,
despite differences in the biological context of radiation-exposed
prostate tumours and radiation-associated second malignancies
(Supplementary Note 2). Particularly striking is patient PD11331
whose primary prostate lesion was irradiated after metastases had
formed. The primary lesion, but not the metastases, exhibited the
genomic features of ionizing radiation.
The relatively low number of mutations that we directly linked
to ionizing radiation may seem surprising for such a well-known
carcinogen. It is certainly considerably less than seen for cancers
associated with tobacco, sunlight or aristolochic acid exposure10.
This probably reflects the fact that although the attributable risk
of such cancers is high, the absolute risk is relatively low. For
example, 490% of angiosarcomas occurring after radiotherapy
for primary breast cancer are attributable to radiation, but only
one in a thousand women receiving such radiotherapy will
develop angiosarcomas37, with a latency of many years. This
suggests that although ionizing radiation clearly pushes bystander
cells in the radiotherapy field towards cancer, the absolute burden
of radiation-induced mutations per cell would not be high and
additional driver mutations would be required.
Figure 1 | Indels in radiation-associated tumours. (a) Indel/substitution ratio. Shown is the indel/substitution ratio for each tumour. The ratio was
significantly increased in radiation-associated second malignancies (0.0003, linear mixed effects model). Each dot represents a tumour. Different colours
represent different tumour types (see legend, top right). Boxplots: vertical line – median; whiskers – minimum and maximum without outliers. (b) Deletion/
insertion ratio. Shown is the deletion/insertion ratio of every tumour. Deletions were significantly (*) enriched in radiation-associated second malignancies
(Po2.2 10 16, linear mixed effects model) and in breast tumours with germline BRCA1 or BRCA2 deficiency. Symbols, boxplots as per a. (c) Clonal versus
subclonal indels in radiation-associated second malignancies. Shown are the absolute clonal (early) and subclonal (late) indel burdens of each tumour, by
indel type. Amongst clonal indels, deletions were significantly enriched. P-values refer to the comparison of proportion of deletions/other indels in clonal
versus subclonal indels (Fisher’s exact test). (d) Indel likelihood across the genome. Shown is the probability of deletion or insertion to occur (vertical axis)
across different regions of the genome (horizontal axis). The probability was modelled on the basis of associations between indels and genomic properties
(see Methods). Chromosome 14 is shown as a representative chromosome. Radiation-associated indels were compared to indels of 35 non-radiationassociated osteosarcomas. Radiation-associated deletions, but not insertions, followed a more uniform distribution across the genome than in radiationnaive samples. (e) Distribution of indels in relation to genomic features. Comparison of the mutation density of radiation versus non-radiation indels in
relation to genomic features. X axis: ratio of mutation density of non-radiation-associated indels or radiation-associated indels over background density. Y
axis: genomic feature. The distribution of insertions in both radiation-associated and radiation-naı̈ve tumours correlated with several genomic features, with
few significant differences (asterisk) between the two. In contrast, the distribution of deletions in radiation-induced cancers, but not in radiation-naive
tumours, showed little variability and resembled the background distribution more closely. Thus, significant differences (asterisk) were seen in the deletion
density in relation to genomic features comparing radiation-associated and radiation-naı̈ve tumours. P-values are detailed in Supplementary Data 4.
4
NATURE COMMUNICATIONS | 7:12605 | DOI: 10.1038/ncomms12605 | www.nature.com/naturecommunications
ARTICLE
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12605
Table 1 | Survey of balanced inversions in different tumour types.
Tumours with at least one balanced
inversion
39
4
19
11
Primary breast tumours
Primary osteosarcoma
BRCA1/2-deficient breast tumours
Radiation-associated second
malignancies
Number of tumours
screened
251
35
33
12
Overall number of balanced
inversions
59
7
46
52
Rearrangement catalogues of tumours were searched informatically for the presence of balanced inversions. The basic principle of the search was to find pairs of head-to-head and tail-to-tail inversions in
which the breakpoint coordinates overlap at both ends. Compared with all 286 primary tumours, balanced inversions were significantly enriched in radiation-associated second malignancies
(P ¼ 2 10 16, generalized linear model) and also in BRCA1 or BRCA1 deficient breast tumours (P ¼ 2 10 16, generalized linear model). Further, compared to BRCA1 or BRCA1 deficient breast tumours
balanced inversions were significantly enriched in radiation-associated second malignancies (P ¼ 0.0006, generalized linear model).
a
Deletions
Tandem duplication
of
re Tot
ar al
ra nu
ge m
m be
en r
ts
Fraction of
rearrangements
PD7188a
PD7530a
PD7192a
108
1.0
0.8
0.6
0.4
0.2
0.0
49
PD9972a
180
51
PD13489a
1.0
0.8
0.6
0.4
0.2
0.0
1.0
0.8
0.6
0.4
0.2
0.0
1.0
0.8
0.6
0.4
0.2
0.0
71
PD9056a
1.0
0.8
0.6
0.4
0.2
0.0
PD8618a
156
6
PD7190a
1.0
0.8
0.6
0.4
0.2
0.0
177
321
PD8622a
PD8623a
1.0
0.8
0.6
0.4
0.2
0.0
1.0
0.8
0.6
0.4
0.2
0.0
1.0
0.8
0.6
0.4
0.2
0.0
76
Translocation
PD7191a
PD7189a
1.0
0.8
0.6
0.4
0.2
0.0
1.0
0.8
0.6
0.4
0.2
0.0
1.0
0.8
0.6
0.4
0.2
0.0
Unbalanced inversion
Balanced inversion
59
Angiosarcoma
Spindle cell sarcoma
Osteosarcoma
Breast tumours
31
b
PD7188a (angiosarcoma)
Chr 9p
5′
27,829,511
28,098,340
27,829,512
28,098,341
3′
Breakpoints
Allele A
Allele B
Germline DNA
sequence:
Wildtype reads
at breakpoint
coordinates
SNP
Mutant
allele B
Split reads
4 base pair deletion
27,829,511
28,098,340
+ +
27,829,512
Head Tail
28,098,345
Tumour DNA
sequence:
Mutant reads
(split reads)
– –
Tail to tail
inversion
Tu
m
N ou
o r
Tu rma
m l
N ou
or r
m
al
Head to head
inversion
Figure 2 | Balanced inversions in radiation-associated tumours. (a) Overview of rearrangements. Tumours exhibited tumour-type specific features.
Balanced inversions (black bars) were found in every tumour, except PD7530a. (b) Example of a balanced inversion in PD7188a. A 0.9 Mb inversion. The
inversion was validated by PCR across the breakpoint (gel image) and by split reads. Note that the split reads carried a heterozygous SNP at the head end.
NATURE COMMUNICATIONS | 7:12605 | DOI: 10.1038/ncomms12605 | www.nature.com/naturecommunications
5
ARTICLE
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12605
a
P = 0.0002
Radiation-exposed prostate tumours
Radiation-naive prostate tumours
Number of mutations
2,000
1,000
Deletions
5
2
PD
11
33
41
13
PD
33
11
PD
PD
11
33
4
2
0
11
PD
33
12
Complex
33
7
3
PD
PD
11
33
32
11
PD
PD
11
32
8
9
0
Insertions
b
PD11331
P = 10–15
Number of
mutations
500
In
d
t el
pr o ir s ex
im ra c
ar dia lus
y
le ted ive
si
on
Al
l
in oth
de e
ls r
0
Figure 3 | Indels in prostate tumours. (a) Indels in radiation-naive versus radiation-exposed prostate tumours. Shown is the indel burden, by indel
subtype, found in radiation-naive and in radiation-exposed prostate tumours. In radiation-exposed tumours radiotherapy had been administered to the
primary tumour before formation of metastases. Deletions were significantly enriched in radiation-exposed tumours (P ¼ 0.0002, generalized linear
model). Note that radiation-associated tumours with confounding BRCA1 or BRCA2 deficiency were excluded from the statistical analysis (cases PD13412
and PD11335). (b) Indels in tumours from a patient whose primary lesion was treated with ionizing radiation after formation of metastases. Shown are
indels that were found exclusively in the primary lesion and indels found in all other lesions. Deletions were significantly enriched amongst indels exclusive
to the primary lesion. Comparison by Fisher’s exact test, of the ratio of deletions over other indels.
Methods
Patient samples. Informed consent was obtained from all subjects and ethical
approval obtained from Cambridgeshire 2 Research Ethics Service (reference
09/H0308/165). Collection and use of patient samples were approved by the
appropriate institutional review board of each Institution.
Whole-genome sequencing. DNA was extracted from 12 radiation-associated
tumours and subjected to whole-genome sequencing, along with normal tissue
derived from the same individuals. All tumour samples had been freshly frozen and
were reviewed by reference pathologists. DNA extraction and preparation followed
standard methods as previously described38. Reads were aligned to the reference
human genome (NCBI37) by using BWA on default settings39. Reads which were
unmapped or PCR-derived duplicates were excluded from the analysis. The average
coverage of tumours was at least 40 and of normal DNA 30 , as per standard
set by the International Cancer Genome Consortium.
Variant detection. The CaVEMan (cancer variants through expectation maximization) algorithm was used to call single-nucleotide substitutions (github.com/
cancerit/CaVEMan). To call insertions and deletions, we used split-read mapping
implemented as a modification of the Pindel algorithm38. To call rearrangements
we applied the BRASS (breakpoint via assembly) algorithm, which identifies
rearrangements by grouping discordant read pairs that point to the same
breakpoint event (github.com/cancerit/BRASS). Post-processing filters were
applied to the output to improve specificity. Copy-number data were derived from
whole-genome reads using the ASCAT (version 2.2) algorithm40. Mutations were
annotated to Ensembl version 58.
Variant validation. The precision of indels and substitutions presented here was
assessed by manual inspection of 100 randomly selected substitutions and was
found to be at least of 90% across the 12 radiation-associated tumours. This
precision of coding indels and substitutions was confirmed by re-sequencing
through whole-exome sequencing. Structural rearrangements were validated by
defining exact break points through local reassembly, as implemented in BRASS.
6
Only rearrangements that could be validated have been included in this report
(listed individually in Supplementary Data 3).
Screen/validation of balanced inversions. Rearrangement catalogues were
screened for the presence of balanced inversions by means of a bespoke PERL
script. Pairs of rearrangement calls were sought that were inversions in opposite
directions with overlapping ranges of upper and lower break points. The search was
directly performed on output from the Brass algorithm with the following postprocessing filters: read count supporting the break point of greater than five reads
and size of inversion greater than 2,500 base pairs unless the read count supporting
the break point was greater than ten reads in which case no size threshold was
applied. This post-processing strategy removes inversion artefacts, which are small
and generally have a read count supporting the break point of less than five reads,
without excluding small, high confidence inversions (defined as break points
supported by at least ten reads). The precision of balanced inversion calls yielded
by this search were assessed in the 12 radiation-associated tumours. In all but one
balanced inversion, both rearrangements defining the inversion could be validated
by algorithmic local reassembly or manual split-read mapping. In addition, a
proportion of balanced inversions in 20/52 was subjected to PCR across the
breakpoint in stock DNA from tumour and normal tissue. These inversions were
all confirmed to be genuine and somatic (Supplementary Fig. 3).
Germline variants. Germline point mutations in TP53, BRCA1 and BRCA2 were
searched for in catalogues of germline indels and substitutions, as determined by
the point mutation variant calling algorithms employed here. Putative mutations
were compared against publicly available catalogues of pathogenic germline
mutations in these genes (www.iarc.fr).
Extraction of substitution signatures. Substitution signatures were extracted by
using non-negative matrix factorization, as previously described10.
NATURE COMMUNICATIONS | 7:12605 | DOI: 10.1038/ncomms12605 | www.nature.com/naturecommunications
ARTICLE
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12605
Subclonality analyses. Subclonal tumour cell populations that exist within
tumours can be screened for by searching for non-heterozygous mutations in
mutation catalogues, as previously described, using a Dirichlet process41. This was
applied to substitution and indel catalogues of the 12 radiation-associated whole
genomes. However, with indels there is a concern that mutant read frequencies
may be underestimated for larger indels, as reads containing larger indels may be
less amenable to mapping. To overcome this bias, the indel mutant read
frequencies were corrected by extracting unmapped reads (split reads) from
sequencing reads. The Dirichlet process was then applied to indel catalogues with,
and without, correction. The results in terms of number of subclonal peaks were
indistinguishable whether corrected or uncorrected indel catalogues were analysed.
Three of the twelve genomes screened for subclones contained subclones which
corresponded to subclones defined by substitutions in these tumours. Thus, the
indel-defined subclones were considered genuine. Indels were subdivided into
clonal (peak of mutation copy number B1) or subclonal (peak of mutation copy
number o1).
Non-radiation tumours. A total of 319 tumours, 284 breast cancers and 35
osteosarcomas were included for comparison in analyses. These were spontaneous
(primary), non-radiation-associated tumours. These tumours were sequenced to
B40 or more, along with normal tissue DNA from the same patients.
These tumours were prepared, sequenced, analysed by the same pipeline as
the 12 radiation-associated tumours, including use of the same algorithms.
The osteosarcoma cases were a series of paediatric and adult tumours (sequencing
data published in the European Genome-phenome Archive, accession
EGAD00001000147). The breast tumours were comprised of oestrogen receptor
positive and negative tumours42. For the purposes of this analysis they were
subdivided into spontaneous cases (n ¼ 251) and those associated with pathogenic
germline BRCA1 or BRCA2 mutation (n ¼ 33). No control primary angiosarcoma
and spindle cell sarcomas were available for inclusion in our analyses.
Association of mutation density with genomic features. The genomic
properties listed in Supplementary Table 3 were calculated at every variant
position, and, for comparison, at 100,000 random positions sampled uniformly
from the callable regions of hg19. Only chromosomes 1–22 and X were considered.
To test for differences in the genomic properties of variants in radiation-induced
versus non-radiation-induced tumours, we used a two-proportion z-test for the
binary variables, a t-test for the other quantitative variables (large sample size
justifies central limit theorem), and a w2-test for the categorical chromatin variable.
A Benjamini-Yekutieli correction was applied to the raw P-values to account for
multiple testing in the presence of likely correlation between these properties.
Genomic properties are considered significantly different between radiation and
non-radiation samples if the adjusted q-value is o0.01 and there is at least a 5%
difference in magnitude between the two group means.
Other statistical analyses. To assess whether radiation-associated tumours
harbour significantly more indels relative to substitutions and more deletions
relative to insertions, a mixed linear effects model was implemented using the
R package lme4. After incorporating as fixed effects type of mutation (substitution,
deletion and insertion) and group of tumour, interactions between tumour group
and type of mutation were assessed. For comparison of indel size distribution
underlying the clustering in Supplementary Fig. 2a, the statistic of the
Kolmogorov–Smirnov test was used (command in R: ks.test(x,y)$statistic). Unless
indicated, R was used for calculations.
Data availability. Sequencing data have been deposited at the European GenomePhenome Archive (EGA, http://www.ebi.ac.uk/ega/), which is hosted by the
European Bioinformatics Institute; accession numbers EGAS00001000138;
EGAS00001000147; EGAS00001000195.
References
1. Leuraud, K. et al. Ionising radiation and risk of death from leukaemia and
lymphoma in radiation-monitored workers (INWORKS): an international
cohort study. Lancet Haematol. 2, e276–e281 (2015).
2. March, H. C. Leukemia in radiologists. Radiology 43, 275–278 (1944).
3. Travis, L. B. et al. Second malignant neoplasms and cardiovascular disease
following radiotherapy. J. Natl Cancer Inst. 104, 357–370 (2012).
4. Preston, D. L. et al. Radiation effects on breast cancer risk: a pooled analysis of
eight cohorts. Radiat. Res. 158, 220–235 (2002).
5. Ronckers, C. M., Doody, M. M., Lonstein, J. E., Stovall, M. & Land, C. E.
Multiple diagnostic X-rays for spine deformities and risk of breast cancer.
Cancer Epidemiol. Biomarkers Prev. 17, 605–613 (2008).
6. Cahan, W. G. et al. Sarcoma arising in irradiated bone; report of 11 cases.
Cancer 1, 3–29 (1948).
7. Allan, J. M. & Travis, L. B. Mechanisms of therapy-related carcinogenesis. Nat.
Rev. Cancer 5, 943–955 (2005).
8. Meier, B. et al. C. elegans whole-genome sequencing reveals mutational
signatures related to carcinogens and DNA repair deficiency. Genome Res. 24,
1624–1636 (2014).
9. Flibotte, S. et al. Whole-genome profiling of mutagenesis in Caenorhabditis
elegans. Genetics 185, 431–441 (2010).
10. Alexandrov, L. B. et al. Signatures of mutational processes in human cancer.
Nature 500, 415–421 (2013).
11. Pleasance, E. D. et al. A small-cell lung cancer genome with complex signatures
of tobacco exposure. Nature 463, 184–190 (2010).
12. Poon, S. L. et al. Genome-wide mutational signatures of aristolochic acid and its
application as a screening tool. Sci. Transl. Med. 5, 197ra101 (2013).
13. Ravanat, J. L. et al. Radiation-mediated formation of complex damage to DNA:
a chemical aspect overview. Br. J. Radiol. 87, 20130715 (2014).
14. Adewoye, A. B., Lindsay, S. J., Dubrova, Y. E. & Hurles, M. E. The genome-wide
effects of ionizing radiation on mutation induction in the mammalian germline.
Nat. Commun. 6, 6684 (2015).
15. Belfield, E. J. et al. Genome-wide analysis of mutations in mutant lineages
selected following fast-neutron irradiation mutagenesis of Arabidopsis thaliana.
Genome Res. 22, 1306–1315 (2012).
16. St Charles, J. et al. High-resolution genome-wide analysis of irradiated (UV and
gamma-rays) diploid yeast cells reveals a high frequency of genomic loss of
heterozygosity (LOH) events. Genetics 190, 1267–1284 (2012).
17. Ray, F. A. et al. Directional genomic hybridization: inversions as a potential
biodosimeter for retrospective radiation exposure. Radiat. Environ. Biophys. 53,
255–263 (2014).
18. Mitchell, C. R. et al. Stable intrachromosomal biomarkers of past exposure to
densely ionizing radiation in several chromosomes of exposed individuals.
Radiat. Res. 162, 257–263 (2004).
19. Natarajan, A. T., Boei, J. J., Vermeulen, S. & Balajee, A. S. Frequencies of X-ray
induced pericentric inversions and centric rings in human blood lymphocytes
detected by FISH using chromosome arm specific DNA libraries. Mutat. Res.
372, 1–7 (1996).
20. Saberi, A., Salari, E. & Latifi, S. M. Cytogenetic analysis in lymphocytes from
radiation workers exposed to low level of ionizing radiation in radiotherapy,
CT-scan and angiocardiography units. Mutat. Res. 750, 92–95 (2013).
21. Hande, M. P. et al. Past exposure to densely ionizing radiation leaves a unique
permanent signature in the genome. Am. J. Hum. Genet. 72, 1162–1170 (2003).
22. Agrawala, P. K., Adhikari, J. S. & Chaudhury, N. K. Lymphocyte chromosomal
aberration assay in radiation biodosimetry. J. Pharm. Bioallied Sci. 2, 197–201
(2010).
23. Gonin-Laurent, N. et al. Specific TP53 mutation pattern in radiation-induced
sarcomas. Carcinogenesis 27, 1266–1272 (2006).
24. Gonin-Laurent, N. et al. RB1 and TP53 pathways in radiation-induced
sarcomas. Oncogene 26, 6106–6112 (2007).
25. Nakanishi, H. et al. Mutation of the p53 gene in postradiation sarcoma. Lab.
Invest. 78, 727–733 (1998).
26. Hadj-Hamou, N. S. et al. A transcriptome signature distinguished sporadic
from postradiotherapy radiation-induced sarcomas. Carcinogenesis 32, 929–934
(2011).
27. Nik-Zainal, S. et al. Mutational processes molding the genomes of 21 breast
cancers. Cell 149, 979–993 (2012).
28. Nik-Zainal, S. et al. The life history of 21 breast cancers. Cell 149, 994–1007
(2012).
29. Stephens, P. J. et al. Massive genomic rearrangement acquired in a single
catastrophic event during cancer development. Cell 144, 27–40 (2011).
30. Gundem, G. et al. The evolutionary history of lethal metastatic prostate cancer.
Nature 520, 353–357 (2015).
31. Schuster-Bockler, B. & Lehner, B. Chromatin organization is a major
influence on regional mutation rates in human cancer cells. Nature 488,
504–507 (2012).
32. Supek, F. & Lehner, B. Differential DNA mismatch repair underlies mutation
rate variation across the human genome. Nature 521, 81–84 (2015).
33. Mehta, A. & Haber, J. E. Sources of DNA double-strand breaks and models of
recombinational DNA repair. Cold Spring Harb. Perspect. Biol. 6, a016428
(2014).
34. Weinfeld, M. & Soderlind, K. J. 32P-postlabeling detection of radiation-induced
DNA damage: identification and estimation of thymine glycols and
phosphoglycolate termini. Biochemistry 30, 1091–1097 (1991).
35. Campbell, P. J. et al. The patterns and dynamics of genomic instability in
metastatic pancreatic cancer. Nature 467, 1109–1113 (2010).
36. Baca, S. C. et al. Punctuated evolution of prostate cancer genomes. Cell 153,
666–677 (2013).
37. Abbott, R. & Palmieri, C. Angiosarcoma of the breast following surgery and
radiotherapy for breast cancer. Nat. Clin. Pract. Oncol. 5, 727–736 (2008).
38. Behjati, S. et al. Recurrent PTPRB and PLCG1 mutations in angiosarcoma. Nat.
Genet. 46, 376–379 (2014).
39. Li, H. & Durbin, R. Fast and accurate long-read alignment with BurrowsWheeler transform. Bioinformatics 26, 589–595 (2010).
NATURE COMMUNICATIONS | 7:12605 | DOI: 10.1038/ncomms12605 | www.nature.com/naturecommunications
7
ARTICLE
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12605
40. Van Loo, P. et al. Allele-specific copy number analysis of tumors. Proc. Natl
Acad. Sci. USA 107, 16910–16915 (2010).
41. Bolli, N. et al. Heterogeneity of genomic evolution and mutational profiles in
multiple myeloma. Nat. Commun. 5, 2997 (2014).
42. Nik-Zainal, S. et al. Landscape of somatic mutations in 560 breast cancer
whole-genome sequences. Nature 534, 47–54 (2016).
K.M.R., L.S., A.M., D.J., and R.S. contributed informatics tools. A.B. and J.W.T.
co-ordinated informatics analyses. B.K. co-ordinated sample curation. P.A.F., A.S.,
C.B. and U.M. contributed to discussion. M.J., N.P., R.T., M.F.A., G.S.B., A.R. and A.M.F.
curated samples, clinical data, and/or provided clinical expertise. A.M.F., M.R.S. and
P.J.C. directed the research. S.B. and P.J.C. wrote the manuscript, with contributions
from A.M.F., D.C.W., G.G. and G.S.B.
Acknowledgements
Additional information
This work was supported by funding from the Wellcome Trust (grant reference 077012/
Z/05/Z), Skeletal Cancer Action Trust, Rosetrees Trust UK, Bone Cancer Research Trust,
the RNOH NHS Trust, the National Institute for Health Research Health Protection
Research Unit in Chemical and Radiation Hazards and Threats at Newcastle University
in partnership with Public Health England. The views expressed are those of the
author(s) and not necessarily those of the NHS, the NIHR, the Department of Health or
Public Health England. Tissue was obtained from the RNOH Musculoskeletal Research
Programme and Biobank, co-ordinated by Mrs Deidre Brooking and Mrs Ru Grinnell,
Biobank staff, RNOH. Support was provided to AMF by the National Institute for Health
Research, UCLH Biomedical Research Centre, and the CRUK UCL Experimental Cancer
Centre. S.N.Z. and S.B. are personally funded through Wellcome Trust Intermediate
Clinical Research Fellowships, P.J.C. through a Wellcome Trust Senior Clinical Research
Fellowship. We are grateful to the patients for participating in this research and to the
clinicians and support staff involved in their care.
Author contributions
S.B. and G.G. performed analyses of sequence data. D.C.W. and N.D.R. performed
statistical analyses. P.S.T., M.R., H.D., S.N-Z contributed data and to data analysis. S.L.C.
contributed to rearrangements analyses. P.V.L performed copy-number analysis. L.B.A.
analysed substitution signatures. C.H. and C.L. performed technical investigations.
Supplementary Information accompanies this paper at http://www.nature.com/
naturecommunications
Competing financial interests: The authors declare no competing financial interests.
Reprints and permission information is available online at http://npg.nature.com/
reprintsandpermissions/
How to cite this article: Behjati, S. et al. Mutational signatures of ionizing radiation in
second malignancies. Nat. Commun. 7:12605 doi: 10.1038/ncomms12605 (2016).
This work is licensed under a Creative Commons Attribution 4.0
International License. The images or other third party material in this
article are included in the article’s Creative Commons license, unless indicated otherwise
in the credit line; if the material is not included under the Creative Commons license,
users will need to obtain permission from the license holder to reproduce the material.
To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/
r The Author(s) 2016
ICGC Prostate Group
Colin S. Cooper15,16, Rosalind A. Eeles16, Douglas Easton17, Christopher Foster18, David E. Neal19,20,
Daniel S. Brewer15,16,21, Freddie Hamdy22, Yong-Jie Lu23, Andrew G. Lynch24, Charlie E. Massi19, Anthony Ng25,
Hayley C. Whitaker19, Yongwei Yu26, Hongwei Zhang26, Elizabeth Bancroft16, Dan Berney23,
Niedzica Camacho16, Cathy Corbishley27, Tokhir Dadaev16, Nening Dennis16, Tim Dudderidge28,
Sandra Edwards16, Cyril Fisher28, Jilur Ghori15, Vincent J. Gnanapragasam30, Christopher Greenman29,
Steve Hawkins19, Steven Hazell28, Will Howat19, Katalin Karaszi22, Jonathan Kay19, Zsofia Kote-Jarai16,
Barbara Kremeyer1, Pardeep Kumar28, Adam Lambert22, Daniel Leongamornlert16, Naomi Livni28,
Hayley Luxton19, Lucy Matthews16, Erik Mayer16, Susan Merson16, David Nicol28, Christopher Ogden28,
Sarah O’Meara1, Gill Pelvender31, Nimish C. Shah30, Simon Tavare32, Sarah Thomas16, Alan Thompson28,
Claire Verrill31, Anne Warren19 & Jorge Zamora1
15Norwich Medical School and Department of Biological Sciences, University of East Anglia, Norwich NR4 7TJ, UK; 16Division of Genetics and Epidemiology,
The Institute Of Cancer Research, London SW7 3RP, UK; 17Centre for Cancer Genetic Epidemiology, Department of Oncology, University of Cambridge,
Cambridge CB1 8RN, UK; 18University of Liverpool and HCA Pathology Laboratories, London WC1E 6JA, UK; 19Urological Research Laboratory, Cancer
Research UK Cambridge Institute, Cambridge CB2 0RE, UK; 20Department of Surgical Oncology, University of Cambridge, Addenbrooke’s Hospital,
Cambridge CB2 0QQ, UK; 21The Genome Analysis Centre, Norwich NR4 7UH, UK; 22The University of Oxford, Oxford OX1 2JD, UK; 23Department of
Molecular Oncology, Barts Cancer Institute, Queen Mary University of London, John Vane Science Centre, London EC1M 6BQ, UK.; 24Statistics and
Computational Biology Laboratory, Cancer Research UK Cambridge Institute, Cambridge CB2 0RE, UK; 25The Chinese University of Hong Kong, Hong Kong,
China; 26Second Military Medical University, Shanghai 200433, China; 27St George’s Hospital, London SW17 0QT, UK; 28Royal Marsden NHS Foundation
Trust, London SW3 6JJ, UK; 29School of Computing Sciences, University of East Anglia, Norwich NR4 7TJ, UK; 30Cambridge University Hospitals NHS
Foundation Trust, Cambridge CB2 0QQ, UK; 31Oxford University Hospitals NHS Trust, John Radcliffe Hospital, Oxford OX3 9DU, UK; 32Statistics and
Computational Biology Laboratory, Cancer Research UK Cambridge Institute, Cambridge CB2 0RE, UK.
8
NATURE COMMUNICATIONS | 7:12605 | DOI: 10.1038/ncomms12605 | www.nature.com/naturecommunications