Mutational signatures of ionizing radiation in second malignancies

ARTICLE Recei ed 7 Sep 2015 |Accep ed 13 Jul 2016 |Published 12 Sep 2016 Mu a ional signa u es o ionizing adia ion in second malignancies Sam Behja i1,2,*, Gunes Gundem1,*, Da id C. Wedge1,3, Nicola D. Robe s1, Pa ick S. Ta pey1, Susanna L. Cooke1, Pe e Van Loo4,5, Ludmil B. Alexand o 1, Manasa Ramak ishna1, Helen Da ies1, Se ena Nik-Zainal1, Clai e Ha dy1, Calli La ime 1, Kei an M. Raine1, Lucy S ebbings1, Andy Menzies1, Da id Jones1, Rebecca Shephe d1, Adam P. Bu le 1, Jon W. Teague1, Me e Jo gensen6, Bha isha Kha i7, Nischalan Pillay6,7, Adam Shlien1,8, P. And ew Fu eal1,9, Ch is ophe Badie10, ICGC P os a e G oupz, Ul an McDe mo 1, G. S e en Bo a11, And ea L. Richa dson12,13,w, Ad ienne M. Flanagan6,7, Michael R. S a on1 & Pe e J. Campbell1,14 Ionizing adia ion is a po en ca cinogen, inducing cance h ough DNA damage. The signa u es o mu a ions a ising in human issues ollowing in i o exposu e o ionizing adia ion ha e no been documen ed. He e, we sea ched o signa u es o ionizing adia ion in 12 adia ion-asso- cia ed second malignancies o di e en umou ypes. Two signa u es o soma ic mu a ion cha ac e ize ionizing adia ion exposu e i espec i e o umou ype. Compa ed wi h 319 adia ion-nai e umou s, adia ion-associa ed umou s ca y a median ex a 201 dele ions genome-wide, sized 1–100 base pai s o en wi h mic ohomology a he junc ion. Unlike dele ions o adia ion-nai e umou s, hese show no a ia ion in densi y ac oss he genome o co ela ion wi h sequence con ex , eplica ion iming o ch oma in s uc u e. Fu he mo e, we obse e a significan inc ease in balanced in e sions in adia ion-associa ed umou s. Bo h small dele ions and in e sions gene a e d i e mu a ions. Thus, ionizing adia ion gene a es dis inc i e mu a- ional signa u es ha explain i s ca cinogenic po en ial. DOI: 10.1038/ncomms12605 OPEN 1Cance Genome P ojec , Wellcome T us Sange Ins i u e, Wellcome T us Genome Campus, Hinx on, Camb idgeshi e CB10 1SA UK. 2Depa men o Paedia ics, Uni e si y o Camb idge, Camb idge CB2 0QQ, UK. 3Ox o d Big Da a Ins i u e and Ox o d Cen e o Cance Gene Resea ch, Wellcome T us Cen e o Human Gene ics, Roose el D i e, Ox o d OX3 7BN, UK. 4The F ancis C ick Ins i u e, London WC2A 3LY, UK. 5Depa men o Human Gene ics, Uni e si y o Leu en, Leu en B-3000, Belgium. 6Uni e si y College London Cance Ins i u e, Hun ley S ee , London WC1E 6BT, UK. 7His opa hology, Royal Na ional O hopaedic Hospi al NHS T us , S anmo e, Middlesex HA7 4LP, UK. 8Depa men o Paedia ic Labo a o y Medicine, The Hospi al o Sick Child en, To on o, On a io, Canada M5G 1X8. 9Depa men o Genomic Medicine, MD Ande son Cance Cen e , Uni e si y o Texas, Hous on, Texas 77030, USA. 10 Cance Mechanisms and Bioma ke s G oup, Radia ion E ec s Depa men , Cen e o Radia ion Chemical and En i onmen al Haza ds, Public Heal h England, Chil on, Didco OX11 0RQ, UK. 11 Ins i u e o Biosciences and Medical Technology, BioMediTech, Uni e si y o Tampe e and Fimlab Labo a o ies, Tampe e Uni e si y Hospi al, Tampe e FI-33520, Finland. 12 Dana-Fa be Cance Ins i u e, Bos on, Massachuse s 02215-5450, USA. 13 B igham and Women’s Hospi al, Ha a d Medical School, Bos on, Massachuse s 02115 USA. 14 Depa men o Haema ology, Uni e si y o Camb idge, Hills Road, Camb idge CB2 2XY, UK. * These au ho s con ibu ed equally o his wo k. wP esen add ess: Sibley Memo ial Hospi al, Johns Hopkins Medicine, Washing on, Dis ic O Columbia 20016, USA. Co espondence and eques s o ma e ials should be add essed o P.J.C. (email: pc8@sange .ac.uk). zA ull lis o conso ium membe s appea s a he end o he pape . NATURE COMMUNICATIONS | 7:12605 | DOI: 10.1038/ncomms12605 | www.na u e.com/na u ecommunica ions 1 Exposu e o ionizing adia ion inc eases he isk o subsequen cance . This isk exhibi s a s ong dose– esponse ela ionship, and he e appea o be no sa e limi s o adia ion exposu e1. This associa ion was fi s no ed by Ma ch who obse ed an inc eased incidence o leukaemia amongs adiologis s2. A leading cause o adia ion-induced cance s appea s o be exposu e o medical adia ion, ei he in he o m o adio he apy o an un ela ed malignancy3o diagnos ic adiog aphy4,5. These ia ogenic umou s a ise as de no o neoplasms in a field o he apeu ic adia ion a e a la ency pe iod ha can span decades6, and a e no ecu ences o he o iginal cance 7. Many, bu no all, en i onmen al ca cinogens induce cance by inc easing he a e o mu a ion in soma ic cells. The physico- chemical p ope ies o a gi en ca cinogen go e n i s in e ac ion wi h DNA, leading o ecu en ‘signa u es’ o pa e ns o mu a ions in he genome. These can be econs uc ed ei he om expe imen al model sys ems8,9 o om s a is ical analyses o cance genomes in exposed pa ien s10–12. Ionizing adia ion di ec ly damages DNA, and can gene a e lesions on single bases, single-s anded nicks in he DNA backbone, clus e ed lesions a se e al nea by si es and double-s anded DNA b eaks13.In expe imen al sys ems exposed o adia ion, including he mu ine ge mline and A abidopsis haliana cells, ionizing adia ion can cause all classes o mu a ions, wi h possible en ichmen o indels14–22. Ta ge ed gene sc eens in adia ion-induced sa coma ha e indica ed an inc eased bu den o dele ions and subs i u ions wi h equen inac i a ion o TP53 and RB1 ( e s 23–25). In addi ion, a ansc ip ome p ofile ha ep esen s a s a e o ch onic oxida i e s ess has been p oposed o be specific o adia ion- associa ed sa coma26. We s udied he genomes o 12 adia ion-associa ed second malignancies o ou di e en umou ypes: os eosa coma; spindle cell sa coma; angiosa coma; b eas cance . These we e seconda y umou s ha a ose wi hin a field o he apeu ic ionizing adia ion and we e no hough o be ecu ences o he o iginal malignancy ea ed wi h adia ion. We chose his expe imen al design o se e al easons: he umou s a e classic adio he apy-induced cance s wi h high a ibu able isks o he adia ion exposu e; he adia ion exposu e occu s o e a sho ime pe iod ela i e o he e olu ion o he cance ; and he mu a ional signa u es o spo adic b eas cance s and sa comas ha e been well documen ed10,27–29. I should be no ed ha in he absence o bioma ke s, a diagnosis o a umou being adia ion- induced canno be defini i ely made (see Supplemen a y No e 1 o clinical de ails and u he discussion). We subjec ed hese 12 umou s, along wi h no mal issues om he same pa ien s, o whole-genome sequencing and ob ained ca alogues o soma ic mu a ions. We compa ed ou findings o 319 adia ion-nai e b eas cance s and sa comas p ocessed by he same sequencing and bioin o ma ics pipeline: 251 b eas umou s; 33 b eas umou s wi h pa hogenic BRCA1 o BRCA2 ge mline mu a ions; 35 os eosa comas (see Me hods o coho de ails). In addi ion, we alida ed ou findings in a published se ies o adia ion-naı ¨ e and adia ion-exposed p os a e umou s om en pa ien s30. The main aim o ou analyses was o sea ch o umou - ype independen , o e a ching signa u es o ionizing adia ion. O e all we iden ified wo such signa u es in adia ion-associa ed second malignancies, an excess o balanced in e sions and o small dele ions. Resul s Tumou - ype specific ea u es. The 12 adia ion-associa ed umou s ha bou ed 1,506–9,245 subs i u ions pe genome (median 4113), 135–943 indels pe genome, (median 429) and 6–321 ea angemen b eak poin s pe genome (median 74; Supplemen a y Da a 1–3). The obse ed d i e mu a ions ollowed he pa e ns expec ed o he umou ype consis en ly (Supplemen a y Table 1). Angiosa comas ha bou ed PTPRB and PLCG1 mu a ions. In spindle cell sa coma and os eosa coma d i e al e a ion o TP53 and CDKN2A we e ound. Canonical PIK3CA mu a ions we e seen in he adia ion-associa ed b eas umou s. Simila ly, many o he mu a ional signa u es seen in spo adic cance s we e also p esen in adia ion-associa ed umou s, such as ch omo h ipsis in sa comas (Supplemen a y Fig. 1). Agains his backd op o genomic di e si y, we ound e idence o wo mu a ional signa u es in he adia ion-associa ed cance s ha anscended umou ype: small dele ions and balanced in e sions. En ichmen o dele ions in adia ion-associa ed umou s. Al hough he absolu e bu den o indels a ied ac oss he 12 adia ion-associa ed umou s, in each umou he indel bu den was high compa ed wi h ha umou ’s subs i u ion bu den (Fig. 1a). Compa ed wi h 319 adia ion-nai e umou s, he indel/ subs i u ion a io was significan ly inc eased in adia ion- associa ed umou s (P¼0.0003, linea mixed e ec s model, see Me hods). Dele ions and inse ions we e no equally en iched. The e was a significan excess o dele ions ela i e o inse ions in adia ion- associa ed second malignancies (Fig. 1b; Po2.2 1016, linea mixed e ec s model). This excess o dele ions was also seen in BRCA1 o BRCA2 ge mline-deficien b eas umou s, as p e- iously desc ibed27, bu was no seen in adia ion-nai e spo adic b eas umou s o sa comas. In each adia ion-associa ed umou , he adio he apy had been gi en o e a ela i ely sho ime pe iod many yea s ea lie . I he excess dele ions we obse ed we e di ec ly a ibu able o ionizing adia ion, hen he en ichmen should only be e iden amongs he ea ly, clonal mu a ions and no in he la e, subclonal mu a ions. We we e able o define subclones in h ee o he adia ion-associa ed umou genomes. S ikingly, compa ed wi h subclonal mu a ions, dele ions we e significan ly inc eased compa ed wi h inse ions amongs clonal mu a ions in all h ee cases (Fig. 1c, Po0.00005; Fishe ’s exac es ). Dis ibu ion o dele ions ac oss he genome. Many mu a ional signa u es show une en dis ibu ion ac oss he genome, especially hose associa ed wi h ca cinogens such as obacco smoke and ul a iole ligh 31, hough o a ise due o highe o de ch oma in o ganiza ion and accessibili y o he ca cinogen and epai p o eins o DNA ( e . 32). Dele ions ound in adia ion-nai e umou s showed conside able long- ange a ia ion in densi y ac oss he genome, as did inse ions in bo h adia ion-associa ed and adia ion-nai e umou s, co ela ing wi h se e al genomic ea u es (Fig. 1d,e; Supplemen a y Da a 4). In s a k con as , dele ions in adia ion-induced cance s showed almos no a iabili y ac oss he genome and minimal co ela ion wi h genomic p ope ies such as eplica ion iming, sequence complexi y o GC con en . We hypo hesize ha his is because o he pe asi e pene a ion o ionizing adia ion h ough issue, meaning ha i s in e ac ion wi h DNA is s ochas ic and una ec ed by highe o de ch oma in s uc u e. Since small dele ions a e he p edominan ead-ou o his damage, hey show no associa ion wi h he genomic ea u es ha influence o he mu a ional p ocesses. E idence o non-homologous end-joining causing dele ions.In wo aspec s, he dele ions o adia ion-associa ed cance s ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12605 2NATURE COMMUNICATIONS | 7:12605 | DOI: 10.1038/ncomms12605 | www.na u e.com/na u ecommunica ions esembled hose o BRCA1 o BRCA2 ge mline-deficien b eas umou s (Supplemen a y Fig. 2): en ichmen o dele ions 4 2–3 bp in leng h and significan ly highe a es o mic ohomology a he b eakpoin junc ion (P¼21016, Kolmogo o –Smi no es )27. This simila i y sugges s ha mic ohomology media ed o non-homologous end-joining a e he pa hways o epai ing adia ion-induced DNA damage, a he han homologous ecombina ion. Possible explana ions o his include ha damage occu s a phases o he cell cycle when homologous ecombina ion pa hways a e less ac i e33 o because he damaged PD7189a PD7191a PD7192a 0 500 P = 9.3 x 10–10 Clonal Subclonal P = 1.1 x 10–5 0 500 P = 4.1 x 10–5 0 500 Numbe o mu a ions Os eosa coma B eas Spindle cell sa coma Angiosa coma c 251 35 33 12 P ima y b eas P ima y os eo- sa coma B eas BRCA1/ BRCA2 Radia ion umou s n 0 0.1 0.2 0.3 0.4 Ra io genome wide indel / subs i u ion bu den a * P ima y b eas P ima y os eo- sa coma B eas BRCA1/ BRCA2 Radia ion umou s PD7530a 251 35 33 12n 0 2 4 6 8 10 Ra io genome wide dele ions / inse ions b Dele ions Inse ions Complex 20 40 60 80 100 Mu a ion likelihood 20 40 60 80 100 Genomic posi ion (MB) Ch 14 Ch 14 Dele ions Inse ions Radia ion-associa ed mu a ionsNon- adia ion associa ed mu a ions d –1 0 1 0 1 * * * * * * * * * * * –1 e Z DNA T iplex Telome e Simple epea Sho andem epea Sequence complexici y Replica ion iming MIR LTR LAD L2 L2 Gene GC con en G quad uplex DNA epea Di ec epea C uci o m CpG islands Ch oma in Cen ome e ALU Non- adia ion associa ed indels Radia ion-associa ed indels Backg ound dis ibu ion Backg ound dis ibu ion Inse ions Dele ions Radia ion umou s ** Clonal Subclonal Clonal Subclonal NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12605 ARTICLE NATURE COMMUNICATIONS | 7:12605 | DOI: 10.1038/ncomms12605 | www.na u e.com/na u ecommunica ions 3 DNA ends con ain s uc u es ha in e e e wi h homologous ecombina ion34. En ichmen o balanced in e sions. Fo s uc u al a ian s, we ound en ichmen o a a e ype o ea angemen , balanced in e sions, among adia ion-associa ed second malignancies, i espec i e o umou ype (Table 1; Fig. 2). While ea ange- men s wi h an in e ed o ien a ion a e common in cance genomes, hey a e ypically unbalanced, associa ed wi h copy- numbe changes and caused by p ocesses such as b eakage- usion-b idge cycles35, ch omo h ipsis29 and ch omoplexy36.We ound a significan en ichmen o balanced in e sions in adia ion-associa ed cance s: 52 in 11/12 umou s compa ed wi h 66 balanced in e sions in 43 o he 286 adia ion-nai e umou s s udied (P¼21016, gene alized linea model). O no e, comple e in e sions we e also significan ly en iched amongs BRCA1 and BRCA2 ge mline-deficien b eas umou s (Table 1, P¼21016). Balanced in e sions anged in size om a ew hund ed base pai s o nea ly 100 megabases, and a b eakpoin junc ions showed a iabili y in mic ohomology and in non- empla ed sequence inse ed (Supplemen a y Da a 5). Valida ion o findings in p os a e umou s. To alida e ou obse a ion ha dele ions and balanced in e sions a e genomic imp in s o ionizing adia ion, we examined he genomes o p ima y and/o me as a ic p os a e umou s om en pa ien s, p e iously published30. Fi e pa ien s had de eloped me as ases a e i adia ion o he p ima y lesion; ou pa ien s had ne e ecei ed adia ion ea men ; and one pa ien , PD11331, ecei ed adio he apy o he p ima y lesion a e me as ases had al eady o med. Consis en wi h he obse a ions made in he 12 adia ion-associa ed second malignancies, we ound a significan en ichmen o dele ions in p os a e cance lesions exposed o adio he apy compa ed wi h adia ion-nai e umou s’ (P¼0.0002, gene alized linea model, Fig. 3a). S ikingly, in pa ien PD11331, he adia ion-exposed p ima y umou (sample PD11331c), bu no adia ion-nai e me as ases, exhibi ed a p eponde ance o dele ions (Fig. 3b, P¼1015, Fishe ’s exac es ). Simila ly, balanced in e sions we e en iched amongs adia ion-exposed lesions (P¼0.04, gene alized linea model; Supplemen a y Table 2). In pa ien PD11331, again i was he adia ion-exposed p ima y umou , bu no any o he me as ases, ha ha bou ed a balanced in e sion. D i e e en s gene a ed by dele ions and in e sions. The oncogenic po en ial o a mu a ional p ocess de i es om i s capaci y o gene a e d i e mu a ions. Wi h hei absence o copy-numbe e ec s, unc ional consequences o balanced in e sions mos commonly a ise om genes b oken a ei he end o he in e sion, no wi hs anding he possibili y o long- ange gene-enhance dis up ion. In ou da a, 48/104 in e sion b eak poin s dis up ed o used genes (Supplemen a y Da a 5), wi h one o ming a d i e mu a ion h ough dis up ion o TP53. Fo he mu a ional signa u e o small dele ions ha we obse e, we es i- ma e ha he median excess o indels in he adia ion-induced cance s sequenced he e is 201 indels pe genome (linea mixed e ec s model, s.d. 348 indels). Among hese a e a 14 base pai dele ion in CASP8 and a 4 base pai dele ion in TP53, bo h dis up ing essen ial splice si es and hus gene a ing d i e e en s. Discussion O e all we iden ified wo genomic imp in s o ionizing adia ion, an excess o dele ions and o an exceedingly a e ype o ea angemen , balanced in e sions. The alidi y o ou s udy may be limi ed by he o e all numbe o umou s we examined and he small numbe o each umou ype. Ye i would seem unlikely ha he en ichmen in adia ion-associa ed umou s o dele ions and o balanced in e sions occu ed by chance. This iew is suppo ed by ou s a is ical analyses as well as he ac ha he signa u es we e umou - ype independen . Bo h signa u es we e p esen ac oss ou di e en umou ypes and could be alida ed in a coho o adia ion-exposed p os a e cance lesions, despi e di e ences in he biological con ex o adia ion-exposed p os a e umou s and adia ion-associa ed second malignancies (Supplemen a y No e 2). Pa icula ly s iking is pa ien PD11331 whose p ima y p os a e lesion was i adia ed a e me as ases had o med. The p ima y lesion, bu no he me as ases, exhibi ed he genomic ea u es o ionizing adia ion. The ela i ely low numbe o mu a ions ha we di ec ly linked o ionizing adia ion may seem su p ising o such a well-known ca cinogen. I is ce ainly conside ably less han seen o cance s associa ed wi h obacco, sunligh o a is olochic acid exposu e10. This p obably eflec s he ac ha al hough he a ibu able isk o such cance s is high, he absolu e isk is ela i ely low. Fo example, 490% o angiosa comas occu ing a e adio he apy o p ima y b eas cance a e a ibu able o adia ion, bu only one in a housand women ecei ing such adio he apy will de elop angiosa comas37, wi h a la ency o many yea s. This sugges s ha al hough ionizing adia ion clea ly pushes bys ande cells in he adio he apy field owa ds cance , he absolu e bu den o adia ion-induced mu a ions pe cell would no be high and addi ional d i e mu a ions would be equi ed. Figu e 1 | Indels in adia ion-associa ed umou s. (a) Indel/subs i u ion a io. Shown is he indel/subs i u ion a io o each umou . The a io was significan ly inc eased in adia ion-associa ed second malignancies (0.0003, linea mixed e ec s model). Each do ep esen s a umou . Di e en colou s ep esen di e en umou ypes (see legend, op igh ). Boxplo s: e ical line – median; whiske s – minimum and maximum wi hou ou lie s. (b) Dele ion/ inse ion a io. Shown is he dele ion/inse ion a io o e e y umou . Dele ions we e significan ly (*) en iched in adia ion-associa ed second malignancies (Po2.2 1016, linea mixed e ec s model) and in b eas umou s wi h ge mline BRCA1 o BRCA2 deficiency. Symbols, boxplo s as pe a.(c) Clonal e sus subclonal indels in adia ion-associa ed second malignancies. Shown a e he absolu e clonal (ea ly) and subclonal (la e) indel bu dens o each umou , by indel ype. Amongs clonal indels, dele ions we e significan ly en iched. P- alues e e o he compa ison o p opo ion o dele ions/o he indels in clonal e sus subclonal indels (Fishe ’s exac es ). (d) Indel likelihood ac oss he genome. Shown is he p obabili y o dele ion o inse ion o occu ( e ical axis) ac oss di e en egions o he genome (ho izon al axis). The p obabili y was modelled on he basis o associa ions be ween indels and genomic p ope ies (see Me hods). Ch omosome 14 is shown as a ep esen a i e ch omosome. Radia ion-associa ed indels we e compa ed o indels o 35 non- adia ion- associa ed os eosa comas. Radia ion-associa ed dele ions, bu no inse ions, ollowed a mo e uni o m dis ibu ion ac oss he genome han in adia ion- nai e samples. (e) Dis ibu ion o indels in ela ion o genomic ea u es. Compa ison o he mu a ion densi y o adia ion e sus non- adia ion indels in ela ion o genomic ea u es. Xaxis: a io o mu a ion densi y o non- adia ion-associa ed indels o adia ion-associa ed indels o e backg ound densi y. Y axis: genomic ea u e. The dis ibu ion o inse ions in bo h adia ion-associa ed and adia ion-naı¨ e umou s co ela ed wi h se e al genomic ea u es, wi h ew significan di e ences (as e isk) be ween he wo. In con as , he dis ibu ion o dele ions in adia ion-induced cance s, bu no in adia ion-nai e umou s, showed li le a iabili y and esembled he backg ound dis ibu ion mo e closely. Thus, significan di e ences (as e isk) we e seen in he dele ion densi y in ela ion o genomic ea u es compa ing adia ion-associa ed and adia ion-naı¨ e umou s. P- alues a e de ailed in Supplemen a y Da a 4. ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12605 4NATURE COMMUNICATIONS | 7:12605 | DOI: 10.1038/ncomms12605 | www.na u e.com/na u ecommunica ions Table 1 | Su ey o balanced in e sions in di e en umou ypes. Tumou s wi h a leas one balanced in e sion Numbe o umou s sc eened O e all numbe o balanced in e sions P ima y b eas umou s 39 251 59 P ima y os eosa coma 4 35 7 BRCA1/2-deficien b eas umou s 19 33 46 Radia ion-associa ed second malignancies 11 12 52 Rea angemen ca alogues o umou s we e sea ched in o ma ically o he p esence o balanced in e sions. The basic p inciple o he sea ch was o find pai s o head- o-head and ail- o- ail in e sions in which he b eakpoin coo dina es o e lap a bo h ends. Compa ed wi h all 286 p ima y umou s, balanced in e sions we e significan ly en iched in adia ion-associa ed second malignancies (P¼21016, gene alized linea model) and also in BRCA1 o BRCA1 deficien b eas umou s (P¼21016, gene alized linea model). Fu he , compa ed o BRCA1 o BRCA1 deficien b eas umou s balanced in e sions we e significan ly en iched in adia ion-associa ed second malignancies (P¼0.0006, gene alized linea model). 108 PD7188a 0.0 0.2 0.4 0.6 0.8 1.0 71 PD7192a 0.0 0.2 0.4 0.6 0.8 1.0 49 PD7530a 0.0 0.2 0.4 0.6 0.8 1.0 51 PD7189a 0.0 0.2 0.4 0.6 0.8 1.0 177 PD7191a 0.0 0.2 0.4 0.6 0.8 1.0 321 PD7190a 0.0 0.2 0.4 0.6 0.8 1.0 180 PD9056a 0.0 0.2 0.4 0.6 0.8 1.0 76 PD9972a 0.0 0.2 0.4 0.6 0.8 1.0 156 PD13489a 0.0 0.2 0.4 0.6 0.8 1.0 6 PD8618a 0.0 0.2 0.4 0.6 0.8 1.0 59 PD8622a 0.0 0.2 0.4 0.6 0.8 1.0 31 PD8623a 0.0 0.2 0.4 0.6 0.8 1.0 Os eosa coma B eas umou s Spindle cell sa comaAngiosa coma Dele ions Tandem duplica ion Unbalanced in e sion Balanced in e sion T ansloca ion To al numbe o ea agemen s F ac ion o ea angemen s a b ++ Head o head in e sion Tail o ail in e sion PD7188a (angiosa coma) Ch 9p 28,098,34128,098,34027,829,511 Head Tail Tumou No mal Tumou No mal Ge mline DNA sequence: Wild ype eads a b eakpoin coo dina es Allele A Tumou DNA sequence: Mu an eads (spli eads) 27,829,512 28,098,34528,098,34027,829,511 3 ′ 5 ′ SNP Allele B Mu an allele B Spli eads B eakpoin s 27,829,512 4 base pai dele ion –– Figu e 2 | Balanced in e sions in adia ion-associa ed umou s. (a) O e iew o ea angemen s. Tumou s exhibi ed umou - ype specific ea u es. Balanced in e sions (black ba s) we e ound in e e y umou , excep PD7530a. (b) Example o a balanced in e sion in PD7188a. A 0.9 Mb in e sion. The in e sion was alida ed by PCR ac oss he b eakpoin (gel image) and by spli eads. No e ha he spli eads ca ied a he e ozygous SNP a he head end. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12605 ARTICLE NATURE COMMUNICATIONS | 7:12605 | DOI: 10.1038/ncomms12605 | www.na u e.com/na u ecommunica ions 5 Me hods Pa ien samples.In o med consen was ob ained om all subjec s and e hical app o al ob ained om Camb idgeshi e 2 Resea ch E hics Se ice ( e e ence 09/H0308/165). Collec ion and use o pa ien samples we e app o ed by he app op ia e ins i u ional e iew boa d o each Ins i u ion. Whole-genome sequencing.DNA was ex ac ed om 12 adia ion-associa ed umou s and subjec ed o whole-genome sequencing, along wi h no mal issue de i ed om he same indi iduals. All umou samples had been eshly ozen and we e e iewed by e e ence pa hologis s. DNA ex ac ion and p epa a ion ollowed s anda d me hods as p e iously desc ibed38. Reads we e aligned o he e e ence human genome (NCBI37) by using BWA on de aul se ings39. Reads which we e unmapped o PCR-de i ed duplica es we e excluded om he analysis. The a e age co e age o umou s was a leas 40 and o no mal DNA 30 , as pe s anda d se by he In e na ional Cance Genome Conso ium. Va ian de ec ion.The CaVEMan (cance a ian s h ough expec a ion max- imiza ion) algo i hm was used o call single-nucleo ide subs i u ions (gi hub.com/ cance i /CaVEMan). To call inse ions and dele ions, we used spli - ead mapping implemen ed as a modifica ion o he Pindel algo i hm38. To call ea angemen s we applied he BRASS (b eakpoin ia assembly) algo i hm, which iden ifies ea angemen s by g ouping disco dan ead pai s ha poin o he same b eakpoin e en (gi hub.com/cance i /BRASS). Pos -p ocessing fil e s we e applied o he ou pu o imp o e specifici y. Copy-numbe da a we e de i ed om whole-genome eads using he ASCAT ( e sion 2.2) algo i hm40. Mu a ions we e anno a ed o Ensembl e sion 58. Va ian alida ion.The p ecision o indels and subs i u ions p esen ed he e was assessed by manual inspec ion o 100 andomly selec ed subs i u ions and was ound o be a leas o 90% ac oss he 12 adia ion-associa ed umou s. This p ecision o coding indels and subs i u ions was confi med by e-sequencing h ough whole-exome sequencing. S uc u al ea angemen s we e alida ed by defining exac b eak poin s h ough local eassembly, as implemen ed in BRASS. Only ea angemen s ha could be alida ed ha e been included in his epo (lis ed indi idually in Supplemen a y Da a 3). Sc een/ alida ion o balanced in e sions.Rea angemen ca alogues we e sc eened o he p esence o balanced in e sions by means o a bespoke PERL sc ip . Pai s o ea angemen calls we e sough ha we e in e sions in opposi e di ec ions wi h o e lapping anges o uppe and lowe b eak poin s. The sea ch was di ec ly pe o med on ou pu om he B ass algo i hm wi h he ollowing pos - p ocessing fil e s: ead coun suppo ing he b eak poin o g ea e han fi e eads and size o in e sion g ea e han 2,500 base pai s unless he ead coun suppo ing he b eak poin was g ea e han en eads in which case no size h eshold was applied. This pos -p ocessing s a egy emo es in e sion a e ac s, which a e small and gene ally ha e a ead coun suppo ing he b eak poin o less han fi e eads, wi hou excluding small, high confidence in e sions (defined as b eak poin s suppo ed by a leas en eads). The p ecision o balanced in e sion calls yielded by his sea ch we e assessed in he 12 adia ion-associa ed umou s. In all bu one balanced in e sion, bo h ea angemen s defining he in e sion could be alida ed by algo i hmic local eassembly o manual spli - ead mapping. In addi ion, a p opo ion o balanced in e sions in 20/52 was subjec ed o PCR ac oss he b eakpoin in s ock DNA om umou and no mal issue. These in e sions we e all confi med o be genuine and soma ic (Supplemen a y Fig. 3). Ge mline a ian s.Ge mline poin mu a ions in TP53, BRCA1 and BRCA2 we e sea ched o in ca alogues o ge mline indels and subs i u ions, as de e mined by he poin mu a ion a ian calling algo i hms employed he e. Pu a i e mu a ions we e compa ed agains publicly a ailable ca alogues o pa hogenic ge mline mu a ions in hese genes (www.ia c. ). Ex ac ion o subs i u ion signa u es.Subs i u ion signa u es we e ex ac ed by using non-nega i e ma ix ac o iza ion, as p e iously desc ibed10. a Radia ion-nai e p os a e umou s Radia ion-exposed p os a e umou s Numbe o mu a ions b Indels exclusi e o i adia ed p ima y lesion All o he indels 0 500 PD11331 Numbe o mu a ions P = 10–15 P = 0.0002 Dele ions Inse ionsComplex PD11328 PD11329 PD11333 PD12337 PD11330 PD11332 PD11334 PD13412 PD11335 0 1,000 2,000 Figu e 3 | Indels in p os a e umou s. (a) Indels in adia ion-nai e e sus adia ion-exposed p os a e umou s. Shown is he indel bu den, by indel sub ype, ound in adia ion-nai e and in adia ion-exposed p os a e umou s. In adia ion-exposed umou s adio he apy had been adminis e ed o he p ima y umou be o e o ma ion o me as ases. Dele ions we e significan ly en iched in adia ion-exposed umou s (P¼0.0002, gene alized linea model). No e ha adia ion-associa ed umou s wi h con ounding BRCA1 o BRCA2 deficiency we e excluded om he s a is ical analysis (cases PD13412 and PD11335). (b) Indels in umou s om a pa ien whose p ima y lesion was ea ed wi h ionizing adia ion a e o ma ion o me as ases. Shown a e indels ha we e ound exclusi ely in he p ima y lesion and indels ound in all o he lesions. Dele ions we e significan ly en iched amongs indels exclusi e o he p ima y lesion. Compa ison by Fishe ’s exac es , o he a io o dele ions o e o he indels. ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12605 6NATURE COMMUNICATIONS | 7:12605 | DOI: 10.1038/ncomms12605 | www.na u e.com/na u ecommunica ions Subclonali y analyses.Subclonal umou cell popula ions ha exis wi hin umou s can be sc eened o by sea ching o non-he e ozygous mu a ions in mu a ion ca alogues, as p e iously desc ibed, using a Di ichle p ocess41. This was applied o subs i u ion and indel ca alogues o he 12 adia ion-associa ed whole genomes. Howe e , wi h indels he e is a conce n ha mu an ead equencies may be unde es ima ed o la ge indels, as eads con aining la ge indels may be less amenable o mapping. To o e come his bias, he indel mu an ead equencies we e co ec ed by ex ac ing unmapped eads (spli eads) om sequencing eads. The Di ichle p ocess was hen applied o indel ca alogues wi h, and wi hou , co ec ion. The esul s in e ms o numbe o subclonal peaks we e indis inguishable whe he co ec ed o unco ec ed indel ca alogues we e analysed. Th ee o he wel e genomes sc eened o subclones con ained subclones which co esponded o subclones defined by subs i u ions in hese umou s. Thus, he indel-defined subclones we e conside ed genuine. Indels we e subdi ided in o clonal (peak o mu a ion copy numbe B1) o subclonal (peak o mu a ion copy numbe o1). Non- adia ion umou s.A o al o 319 umou s, 284 b eas cance s and 35 os eosa comas we e included o compa ison in analyses. These we e spon aneous (p ima y), non- adia ion-associa ed umou s. These umou s we e sequenced o B40 o mo e, along wi h no mal issue DNA om he same pa ien s. These umou s we e p epa ed, sequenced, analysed by he same pipeline as he 12 adia ion-associa ed umou s, including use o he same algo i hms. The os eosa coma cases we e a se ies o paedia ic and adul umou s (sequencing da a published in he Eu opean Genome-phenome A chi e, accession EGAD00001000147). The b eas umou s we e comp ised o oes ogen ecep o posi i e and nega i e umou s42. Fo he pu poses o his analysis hey we e subdi ided in o spon aneous cases (n¼251) and hose associa ed wi h pa hogenic ge mline BRCA1 o BRCA2 mu a ion (n¼33). No con ol p ima y angiosa coma and spindle cell sa comas we e a ailable o inclusion in ou analyses. Associa ion o mu a ion densi y wi h genomic ea u es.The genomic p ope ies lis ed in Supplemen a y Table 3 we e calcula ed a e e y a ian posi ion, and, o compa ison, a 100,000 andom posi ions sampled uni o mly om he callable egions o hg19. Only ch omosomes 1–22 and X we e conside ed. To es o di e ences in he genomic p ope ies o a ian s in adia ion-induced e sus non- adia ion-induced umou s, we used a wo-p opo ion z- es o he bina y a iables, a - es o he o he quan i a i e a iables (la ge sample size jus ifies cen al limi heo em), and a w2- es o he ca ego ical ch oma in a iable. A Benjamini-Yeku ieli co ec ion was applied o he aw P- alues o accoun o mul iple es ing in he p esence o likely co ela ion be ween hese p ope ies. Genomic p ope ies a e conside ed significan ly di e en be ween adia ion and non- adia ion samples i he adjus ed q- alue is o0.01 and he e is a leas a 5% di e ence in magni ude be ween he wo g oup means. O he s a is ical analyses.To assess whe he adia ion-associa ed umou s ha bou significan ly mo e indels ela i e o subs i u ions and mo e dele ions ela i e o inse ions, a mixed linea e ec s model was implemen ed using he R package lme4. A e inco po a ing as fixed e ec s ype o mu a ion (subs i u ion, dele ion and inse ion) and g oup o umou , in e ac ions be ween umou g oup and ype o mu a ion we e assessed. Fo compa ison o indel size dis ibu ion unde lying he clus e ing in Supplemen a y Fig. 2a, he s a is ic o he Kolmogo o –Smi no es was used (command in R: ks. es (x,y)$s a is ic). Unless indica ed, R was used o calcula ions. Da a a ailabili y.Sequencing da a ha e been deposi ed a he Eu opean Genome- Phenome A chi e (EGA, h p://www.ebi.ac.uk/ega/), which is hos ed by he Eu opean Bioin o ma ics Ins i u e; accession numbe s EGAS00001000138; EGAS00001000147; EGAS00001000195. Re e ences 1. Leu aud, K. e al. Ionising adia ion and isk o dea h om leukaemia and lymphoma in adia ion-moni o ed wo ke s (INWORKS): an in e na ional coho s udy. Lance Haema ol. 2, e276–e281 (2015). 2. Ma ch, H. C. Leukemia in adiologis s. Radiology 43, 275–278 (1944). 3. T a is, L. B. e al. Second malignan neoplasms and ca dio ascula disease ollowing adio he apy. J. Na l Cance Ins . 104, 357–370 (2012). 4. P es on, D. L. e al. Radia ion e ec s on b eas cance isk: a pooled analysis o eigh coho s. Radia . Res. 158, 220–235 (2002). 5. Roncke s, C. M., Doody, M. M., Lons ein, J. E., S o all, M. & Land, C. E. Mul iple diagnos ic X- ays o spine de o mi ies and isk o b eas cance . Cance Epidemiol. Bioma ke s P e . 17, 605–613 (2008). 6. Cahan, W. G. e al. Sa coma a ising in i adia ed bone; epo o 11 cases. Cance 1, 3–29 (1948). 7. Allan, J. M. & T a is, L. B. Mechanisms o he apy- ela ed ca cinogenesis. Na . Re . Cance 5, 943–955 (2005). 8. Meie , B. e al. C. elegans whole-genome sequencing e eals mu a ional signa u es ela ed o ca cinogens and DNA epai deficiency. Genome Res. 24, 1624–1636 (2014). 9. Flibo e, S. e al. Whole-genome p ofiling o mu agenesis in Caeno habdi is elegans.Gene ics 185, 431–441 (2010). 10. Alexand o , L. B. e al. Signa u es o mu a ional p ocesses in human cance . Na u e 500, 415–421 (2013). 11. Pleasance, E. D. e al. A small-cell lung cance genome wi h complex signa u es o obacco exposu e. Na u e 463, 184–190 (2010). 12. Poon, S. L. e al. Genome-wide mu a ional signa u es o a is olochic acid and i s applica ion as a sc eening ool. Sci. T ansl. Med. 5, 197 a101 (2013). 13. Ra ana , J. L. e al. Radia ion-media ed o ma ion o complex damage o DNA: a chemical aspec o e iew. B . J. Radiol. 87, 20130715 (2014). 14. Adewoye, A. B., Lindsay, S. J., Dub o a, Y. E. & Hu les, M. E. The genome-wide e ec s o ionizing adia ion on mu a ion induc ion in he mammalian ge mline. Na . Commun. 6, 6684 (2015). 15. Belfield, E. J. e al. Genome-wide analysis o mu a ions in mu an lineages selec ed ollowing as -neu on i adia ion mu agenesis o A abidopsis haliana. Genome Res. 22, 1306–1315 (2012). 16. S Cha les, J. e al. High- esolu ion genome-wide analysis o i adia ed (UV and gamma- ays) diploid yeas cells e eals a high equency o genomic loss o he e ozygosi y (LOH) e en s. Gene ics 190, 1267–1284 (2012). 17. Ray, F. A. e al. Di ec ional genomic hyb idiza ion: in e sions as a po en ial biodosime e o e ospec i e adia ion exposu e. Radia . En i on. Biophys. 53, 255–263 (2014). 18. Mi chell, C. R. e al. S able in ach omosomal bioma ke s o pas exposu e o densely ionizing adia ion in se e al ch omosomes o exposed indi iduals. Radia . Res. 162, 257–263 (2004). 19. Na a ajan, A. T., Boei, J. J., Ve meulen, S. & Balajee, A. S. F equencies o X- ay induced pe icen ic in e sions and cen ic ings in human blood lymphocy es de ec ed by FISH using ch omosome a m specific DNA lib a ies. Mu a . Res. 372, 1–7 (1996). 20. Sabe i, A., Sala i, E. & La ifi, S. M. Cy ogene ic analysis in lymphocy es om adia ion wo ke s exposed o low le el o ionizing adia ion in adio he apy, CT-scan and angioca diog aphy uni s. Mu a . Res. 750, 92–95 (2013). 21. Hande, M. P. e al. Pas exposu e o densely ionizing adia ion lea es a unique pe manen signa u e in he genome. Am. J. Hum. Gene . 72, 1162–1170 (2003). 22. Ag awala, P. K., Adhika i, J. S. & Chaudhu y, N. K. Lymphocy e ch omosomal abe a ion assay in adia ion biodosime y. J. Pha m. Bioallied Sci. 2, 197–201 (2010). 23. Gonin-Lau en , N. e al. Specific TP53 mu a ion pa e n in adia ion-induced sa comas. Ca cinogenesis 27, 1266–1272 (2006). 24. Gonin-Lau en , N. e al. RB1 and TP53 pa hways in adia ion-induced sa comas. Oncogene 26, 6106–6112 (2007). 25. Nakanishi, H. e al. Mu a ion o he p53 gene in pos adia ion sa coma. Lab. In es . 78, 727–733 (1998). 26. Hadj-Hamou, N. S. e al. A ansc ip ome signa u e dis inguished spo adic om pos adio he apy adia ion-induced sa comas. Ca cinogenesis 32, 929–934 (2011). 27. Nik-Zainal, S. e al. Mu a ional p ocesses molding he genomes o 21 b eas cance s. Cell 149, 979–993 (2012). 28. Nik-Zainal, S. e al. The li e his o y o 21 b eas cance s. Cell 149, 994–1007 (2012). 29. S ephens, P. J. e al. Massi e genomic ea angemen acqui ed in a single ca as ophic e en du ing cance de elopmen . Cell 144, 27–40 (2011). 30. Gundem, G. e al. The e olu iona y his o y o le hal me as a ic p os a e cance . Na u e 520, 353–357 (2015). 31. Schus e -Bockle , B. & Lehne , B. Ch oma in o ganiza ion is a majo influence on egional mu a ion a es in human cance cells. Na u e 488, 504–507 (2012). 32. Supek, F. & Lehne , B. Di e en ial DNA misma ch epai unde lies mu a ion a e a ia ion ac oss he human genome. Na u e 521, 81–84 (2015). 33. Meh a, A. & Habe , J. E. Sou ces o DNA double-s and b eaks and models o ecombina ional DNA epai . Cold Sp ing Ha b. Pe spec . Biol. 6, a016428 (2014). 34. Wein eld, M. & Sode lind, K. J. 32P-pos labeling de ec ion o adia ion-induced DNA damage: iden ifica ion and es ima ion o hymine glycols and phosphoglycola e e mini. Biochemis y 30, 1091–1097 (1991). 35. Campbell, P. J. e al. The pa e ns and dynamics o genomic ins abili y in me as a ic panc ea ic cance . Na u e 467, 1109–1113 (2010). 36. Baca, S. C. e al. Punc ua ed e olu ion o p os a e cance genomes. Cell 153, 666–677 (2013). 37. Abbo , R. & Palmie i, C. Angiosa coma o he b eas ollowing su ge y and adio he apy o b eas cance . Na . Clin. P ac . Oncol. 5, 727–736 (2008). 38. Behja i, S. e al. Recu en PTPRB and PLCG1 mu a ions in angiosa coma. Na . Gene . 46, 376–379 (2014). 39. Li, H. & Du bin, R. Fas and accu a e long- ead alignmen wi h Bu ows- Wheele ans o m. Bioin o ma ics 26, 589–595 (2010). NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12605 ARTICLE NATURE COMMUNICATIONS | 7:12605 | DOI: 10.1038/ncomms12605 | www.na u e.com/na u ecommunica ions 7 40. Van Loo, P. e al. Allele-specific copy numbe analysis o umo s. P oc. Na l Acad. Sci. USA 107, 16910–16915 (2010). 41. Bolli, N. e al. He e ogenei y o genomic e olu ion and mu a ional p ofiles in mul iple myeloma. Na . Commun. 5, 2997 (2014). 42. Nik-Zainal, S. e al. Landscape o soma ic mu a ions in 560 b eas cance whole-genome sequences. Na u e 534, 47–54 (2016). Acknowledgemen s This wo k was suppo ed by unding om he Wellcome T us (g an e e ence 077012/ Z/05/Z), Skele al Cance Ac ion T us , Rose ees T us UK, Bone Cance Resea ch T us , he RNOH NHS T us , he Na ional Ins i u e o Heal h Resea ch Heal h P o ec ion Resea ch Uni in Chemical and Radia ion Haza ds and Th ea s a Newcas le Uni e si y in pa ne ship wi h Public Heal h England. The iews exp essed a e hose o he au ho (s) and no necessa ily hose o he NHS, he NIHR, he Depa men o Heal h o Public Heal h England. Tissue was ob ained om he RNOH Musculoskele al Resea ch P og amme and Biobank, co-o dina ed by M s Deid e B ooking and M s Ru G innell, Biobank s a , RNOH. Suppo was p o ided o AMF by he Na ional Ins i u e o Heal h Resea ch, UCLH Biomedical Resea ch Cen e, and he CRUK UCL Expe imen al Cance Cen e. S.N.Z. and S.B. a e pe sonally unded h ough Wellcome T us In e media e Clinical Resea ch Fellowships, P.J.C. h ough a Wellcome T us Senio Clinical Resea ch Fellowship. We a e g a e ul o he pa ien s o pa icipa ing in his esea ch and o he clinicians and suppo s a in ol ed in hei ca e. Au ho con ibu ions S.B. and G.G. pe o med analyses o sequence da a. D.C.W. and N.D.R. pe o med s a is ical analyses. P.S.T., M.R., H.D., S.N-Z con ibu ed da a and o da a analysis. S.L.C. con ibu ed o ea angemen s analyses. P.V.L pe o med copy-numbe analysis. L.B.A. analysed subs i u ion signa u es. C.H. and C.L. pe o med echnical in es iga ions. K.M.R., L.S., A.M., D.J., and R.S. con ibu ed in o ma ics ools. A.B. and J.W.T. co-o dina ed in o ma ics analyses. B.K. co-o dina ed sample cu a ion. P.A.F., A.S., C.B. and U.M. con ibu ed o discussion. M.J., N.P., R.T., M.F.A., G.S.B., A.R. and A.M.F. cu a ed samples, clinical da a, and/o p o ided clinical expe ise. A.M.F., M.R.S. and P.J.C. di ec ed he esea ch. S.B. and P.J.C. w o e he manusc ip , wi h con ibu ions om A.M.F., D.C.W., G.G. and G.S.B. Addi ional in o ma ion Supplemen a y In o ma ion accompanies his pape a h p://www.na u e.com/ na u ecommunica ions Compe ing financial in e es s: The au ho s decla e no compe ing financial in e es s. Rep in s and pe mission in o ma ion is a ailable online a h p://npg.na u e.com/ ep in sandpe missions/ How o ci e his a icle: Behja i, S. e al. Mu a ional signa u es o ionizing adia ion in second malignancies. Na . Commun. 7:12605 doi: 10.1038/ncomms12605 (2016). This wo k is licensed unde a C ea i e Commons A ibu ion 4.0 In e na ional License. The images o o he hi d pa y ma e ial in his a icle a e included in he a icle’s C ea i e Commons license, unless indica ed o he wise in he c edi line; i he ma e ial is no included unde he C ea i e Commons license, use s will need o ob ain pe mission om he license holde o ep oduce he ma e ial. To iew a copy o his license, isi h p://c ea i ecommons.o g/licenses/by/4.0/ The Au ho (s) 2016 ICGC P os a e G oup Colin S. Coope 15,16, Rosalind A. Eeles16, Douglas Eas on17, Ch is ophe Fos e 18, Da id E. Neal19,20, Daniel S. B ewe 15,16,21, F eddie Hamdy22, Yong-Jie Lu23, And ew G. Lynch24, Cha lie E. Massi19, An hony Ng25, Hayley C. Whi ake 19, Yongwei Yu26, Hongwei Zhang26, Elizabe h Banc o 16, Dan Be ney23, Niedzica Camacho16, Ca hy Co bishley27, Tokhi Dadae 16, Nening Dennis16, Tim Dudde idge28, Sand a Edwa ds16, Cy il Fishe 28, Jilu Gho i15, Vincen J. Gnanap agasam30, Ch is ophe G eenman29, S e e Hawkins19, S e en Hazell28, Will Howa 19, Ka alin Ka aszi22, Jona han Kay19, Zsofia Ko e-Ja ai16, Ba ba a K emeye 1, Pa deep Kuma 28, Adam Lambe 22, Daniel Leongamo nle 16, Naomi Li ni28, Hayley Lux on19, Lucy Ma hews16, E ik Maye 16, Susan Me son16, Da id Nicol28, Ch is ophe Ogden28, Sa ah O’Mea a1, Gill Pel ende 31, Nimish C. Shah30, Simon Ta a e32, Sa ah Thomas16, Alan Thompson28, Clai e Ve ill31, Anne Wa en19 & Jo ge Zamo a1 15No wich Medical School and Depa men o Biological Sciences, Uni e si y o Eas Anglia, No wich NR4 7TJ, UK; 16Di ision o Gene ics and Epidemiology, The Ins i u e O Cance Resea ch, London SW7 3RP, UK; 17Cen e o Cance Gene ic Epidemiology, Depa men o Oncology, Uni e si y o Camb idge, Camb idge CB1 8RN, UK; 18Uni e si y o Li e pool and HCA Pa hology Labo a o ies, London WC1E 6JA, UK; 19U ological Resea ch Labo a o y, Cance Resea ch UK Camb idge Ins i u e, Camb idge CB2 0RE, UK; 20Depa men o Su gical Oncology, Uni e si y o Camb idge, Addenb ooke’s Hospi al, Camb idge CB2 0QQ, UK; 21The Genome Analysis Cen e, No wich NR4 7UH, UK; 22The Uni e si y o Ox o d, Ox o d OX1 2JD, UK; 23Depa men o Molecula Oncology, Ba s Cance Ins i u e, Queen Ma y Uni e si y o London, John Vane Science Cen e, London EC1M 6BQ, UK.; 24S a is ics and Compu a ional Biology Labo a o y, Cance Resea ch UK Camb idge Ins i u e, Camb idge CB2 0RE, UK; 25The Chinese Uni e si y o Hong Kong, Hong Kong, China; 26Second Mili a y Medical Uni e si y, Shanghai 200433, China; 27S Geo ge’s Hospi al, London SW17 0QT, UK; 28Royal Ma sden NHS Founda ion T us , London SW3 6JJ, UK; 29School o Compu ing Sciences, Uni e si y o Eas Anglia, No wich NR4 7TJ, UK; 30Camb idge Uni e si y Hospi als NHS Founda ion T us , Camb idge CB2 0QQ, UK; 31Ox o d Uni e si y Hospi als NHS T us , John Radcli e Hospi al, Ox o d OX3 9DU, UK; 32S a is ics and Compu a ional Biology Labo a o y, Cance Resea ch UK Camb idge Ins i u e, Camb idge CB2 0RE, UK. ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12605 8NATURE COMMUNICATIONS | 7:12605 | DOI: 10.1038/ncomms12605 | www.na u e.com/na u ecommunica ions