Germ-line CEBPA mutations are highly penetrant, causing early-onset de novo AML associated with favorable survival outcomes.
Familial CEBPA-mutated AML displays a unique model of disease progression, with recurrence caused by novel, independent leukemic episodes.
In-depth molecular investigation of familial leukemia has been limited by the rarity of recognized cases. This study examines the genetic events initiating leukemia and details the clinical progression of disease across multiple families harboring germ-line CEBPA mutations. Clinical data were collected from 10 CEBPA-mutated families, representing 24 members with acute myeloid leukemia (AML). Whole-exome (WES) and deep sequencing were performed to genetically profile tumors and define patterns of clonal evolution. Germline CEBPA mutations clustered within the N-terminal and were highly penetrant, with AML presenting at a median age of 24.5 years (range, 1.75-46 years). In all diagnostic tumors tested (n = 18), double CEBPA mutations (CEBPAdm) were detected, with acquired (somatic) mutations preferentially targeting the C-terminal. Somatic CEBPA mutations were unstable throughout the disease course, with different mutations identified at recurrence. Deep sequencing of diagnostic and relapse paired samples confirmed that relapse-associated CEBPA mutations were absent at diagnosis, suggesting recurrence was triggered by novel, independent clones. Integrated WES and deep sequencing subsequently revealed an entirely new complement of mutations at relapse, verifying the presentation of a de novo leukemic episode. The cumulative incidence of relapse in familial AML was 56% at 10 years (n = 11), and 3 patients experienced ≥3 disease episodes over a period of 17 to 20 years. Durable responses to secondary therapies were observed, with prolonged median survival after relapse (8 years) and long-term overall survival (10-year overall survival, 67%). Our data reveal that familial CEBPA-mutated AML exhibits a unique model of disease progression, associated with favorable long-term outcomes.
The transcription factor CCAAT/enhancer binding protein-α (CEBPA) is encoded on chromosome 19q13.1 and has an essential role in mediating granulocytic differentiation and cellular growth arrest.1 Somatic CEBPA mutations occur in 10% to 15% of sporadic normal karyotype acute myeloid leukemia (NK-AML), with double mutations (CEBPAdm) now considered to represent a unique molecular entity associated with favorable clinical outcomes.2-4 CEBPAdm tumors typically harbor N-terminal frameshift mutations (enforcing translation of the shorter p30 isoform) combined with C-terminal in-frame insertions or deletions (disrupting the DNA binding or leucine zipper domains). These lesions are frequently accompanied by mutations in GATA2 or WT1, which appear to be enriched within CEBPAdm AML.5-9
We initially described the autosomal dominant transmission of AML in a family harboring a germline CEBPA mutation in 2004.10 Following subsequent reports of further CEBPA-mutated families, this unique clinical entity was established and has been recently reviewed with other leukemia predisposition syndromes, including germline RUNX1 and GATA2 mutations.11 Although familial leukemia is rare, its true prevalence is likely underestimated due to variations in disease phenotype and latency, compounded by a lack of available family history. Clinical awareness and recognition is not only essential for the appropriate investigation, counseling, and management of families but also provides the basis for further scientific and clinical investigation. This study represents a multicenter collaboration, whereby families with germline CEBPA mutations were collectively identified to characterize both the clinical course of disease and the molecular events underlying disease evolution.
Familial AML patients
This study was performed in accordance with the declaration of Helsinki; ethical approval was obtained locally and within collaborating centers, with informed consent received from all participants or their guardians. Clinical follow-up data were collected for 10 pedigrees (A-J), 9 of which were previously reported.2,10,12-17 Pedigree C represents a novel AML pedigree and is detailed below. Germline CEBPA mutations were verified in affected individuals from pedigrees A to E, using bidirectional sequencing as previously described.18 The sources of germline (constitutional) DNA included peripheral blood (PB) or bone marrow (BM) during remission (A.II.1, A.II.5, A.III.2, B.II.2, C.III.1, D.II.2, and E.I.1), saliva (C.III.1 and D.II.1), or skin biopsy (B.I.1).
The index case in pedigree C was C.III.1, who presented in 2012 at 18 years of age. BM examination revealed 70% blasts, morphologically classified as French-American-British (FAB) subtype M4, with aberrant CD7 expression; cytogenetic analysis repeatedly failed. Her mother and maternal aunt had also developed AML at 32 and 3 years of age, respectively. A germline CEBPA mutation (c.218insC, p.H24AfsX84) was detected in tumor, remission, and salivary DNA from C.III.1. A further C-terminal CEBPA mutation (c.1047_1088dup, p.R300_K313dup) was detected in tumor DNA alone, and the nucleotide sequence of this insertion was subsequently confirmed with cloning and Sanger sequencing of individual colonies (Original TA kit; Invitrogen). A second maternal aunt, C.II.3, aged 41 years, was found to be a healthy carrier of the germline mutation (Figure 1).
Whole-exome sequencing of familial AML
Whole-exome profiling of somatic mutations was performed in AML samples from 4 pedigrees (A-D). Primary tumors were investigated in 7 patients (A.II.1, A.II.5, A.III.2, B.I.1, B.II.2, C.III.1, and D.II.2) with 2 episodes of disease recurrence (B.I.1). Matched DNA from the first complete remission (CR1) was used as a germline reference for primary leukemic episodes, whereas skin DNA was used as a germline reference for the 3 sequential tumors (T1-T3) in B.I.1. Capture libraries were constructed from 1.5 to 3 μg DNA using the Agilent SureSelect XT Human All Exon V5 kit. Enriched exome libraries were multiplexed and sequenced on the Illumina 2500 platform to generate 100-bp paired-end reads. Whole-exome sequencing (WES) data have been deposited in the European Genome-phenome Archive under accession no. EGAS00001000949.
Sequence alignment and variant detection
WES reads were first de-multiplexed and aligned to the reference genome hg19 using the Burrows-Wheeler Aligner version 0.6.2.19 Picard tools software (version 1.86) was used to generate BAM files and remove polymerase chain reaction duplicates, with local realignment around insertions or deletions (indels) and base quality score recalibration performed with the Genome Analysis Toolkit (GATK) v188.8.131.52 Somatic single nucleotide variants (SNVs) were called using the MuTect algorithm,21 and short indels were identified using the GATK Somatic Indel Detector module, with a minimum coverage of 8 reads at the identified site. The minimum average mapping and base quality scores for Indel supporting reads were 10 and 20, respectively, in both tumor and matched normal samples. Pindel22 was used to detect longer known indels, such as CEBPA C-terminal insertions (>10 bp). Only somatic variants with homozygous reference alleles in the matched normal samples were considered and annotated with the SNPnexus tool23 ; variants corresponding to dbSNP137 entries were also noted. Mutations were then manually reviewed to remove sequencing and misalignment errors, prior to validation with Sanger and deep sequencing. To ensure that preleukemic mutations (recently described in sporadic AML24-29 ) had not been excluded from our analytical pipeline, variant calling in candidate genes was performed for all tumor and remission samples using the reference genome, hg19, as detailed in the supplemental Data available on the Blood Web site. Further genetic profiling of tumors involved the detection of copy number alterations (CNAs) and loss of heterozygosity (LOH) from WES reads, with copy-neutral LOH (also known as acquired uniparental disomy [aUPD]) revealed by intersecting these data (supplemental Data).
Tumor subpopulation identification and clonality analysis
To analyze the clonal architecture and population dynamics in tumor samples, all somatic SNVs, indels, LOH variants, and CNAs were combined to estimate clonal expansions and the cellular frequency of each subpopulation (SP) using EXPANDS.30 Somatic LOH variants were identified according to the method described by Murtaza et al.31 All somatic variants were assigned to nested populations, with the largest SP forming the dominant clone and any SPs with lower cellular frequencies regarded as subclonal populations.
Integrated whole-exome and deep sequencing analysis of clonal evolution
Somatic CEBPA mutations were tested at diagnosis and relapse in 6 patients (B.I.1, B.II.2, E.I.1, E.II.1, F.IV.2, and G.II.2) using Sanger sequencing. Targeted deep sequencing was subsequently performed in tumor pairs from B.I.1 and E.I.1 to determine whether relapse-associated CEBPA mutations preexisted at diagnosis or represented a de novo event. To comprehensively characterize disease progression, we performed deep sequencing of 28 selected variants across 3 sequential tumors (T1-T3) in patient B.I.1, enabling us to decipher the pattern of clonal evolution over a disease course of 17 years. For detailed methods, see supplemental Data.
Clinical profiling of an extended familial AML cohort
Clinical demographic and survival data were collected across 10 pedigrees including 24 members with AML. The median duration of follow-up in surviving patients was 7.5 years (range, 2-46 years); 4 patients died within a year of diagnosis, and 1 (F.III.2) was lost to follow-up. Overall, 96% of patients (n = 23) were treated with induction chemotherapy comprising an anthracycline and/or cytarabine regimen. Patient A.I.1 presented in 1963 and achieved a 16-month remission following prednisolone and mercaptopurine therapy. He experienced 2 episodes of disease recurrence: the first was treated with prednisolone and methotrexate and the second with cyclophosphamide, resulting in a long-term remission of 45 years.10 Consolidation strategies are summarized in supplemental Table 1 and comprised chemotherapy alone (n = 11), autologous stem cell transplantation (n = 5), or allogeneic stem cell transplantation (n = 3). Due to treatment-related complications, 2 patients did not receive consolidation therapy after remission (A.I.1 and I.I.2), whereas J.I.2 and C.II.1 died due to hemorrhage and chemotherapy-resistant disease, respectively.
Clinical end point definitions follow revised International Working Group Criteria,32 except that count recovery was not required for complete remission. Survival percentages were calculated using Kaplan-Meier methodology and cumulative incidence of relapse using death as a competing risk. Univariate analyses were performed using the log-rank test. Clinical outcomes in familial AML were compared with sporadic AML in younger adults, harboring presumed somatic single (n = 48, CEBPAsm) and double (n = 59, CEBPAdm) CEBPA mutations. All sporadic cases were treated within the UK Medical Research Council AML 10 and 12 trials, as previously described.3 (supplemental Table 1).
Characteristics of disease presentation
Familial AML presented early, at a median age of 24.5 years (range, 1.75-46 years; Figure 1). Three asymptomatic carriers were identified (C.II.3, E.II.2, and I.III.3) aged 41, 24, and 19 years, respectively; they continue to receive annual follow-up with normal PB counts and BM morphology. All episodes of familial AML presented de novo and were predominantly characterized by an NK or FAB M1 or M2 morphology and aberrant expression of the T-cell antigen CD7. Germline CEBPA mutations were universally located in the N-terminal, each causing a frameshift preceding the internal p30 start codon. Somatic C-terminal CEBPA mutations were detected in all diagnostic tumors tested (n = 18), with the majority representing in-frame insertions or deletions within the DNA-binding and leucine zipper domains (Figure 2A; supplemental Table 2).
Profiling of somatic mutations within familial AML
WES of 9 familial AML samples yielded a mean exonic coverage of 89×, with 94% of bases covered >10-fold (supplemental Table 3). A total of 300 tumor-specific mutations were detected within 252 genes, including 160 nonsynonymous/splice site (protein-altering), 57 synonymous, and 83 untranslated region variants (supplemental Figure 1 and supplemental Tables 4-6). We verified 107 of 118 variants with Sanger and deep sequencing, achieving concordance of 83% for insertions or deletions (indels) and 93% for SNVs, as detailed in supplemental Table 7.
We then focused our analysis on somatic nonsynonymous mutations and compared these with recurrently mutated genes (n ≥ 2 nonsynonymous mutations) identified from 200 sporadic AML exomes, sequenced by the Cancer Genome Atlas Research Network (TCGARN).33 C-terminal CEBPA mutations were acquired in all 9 exome-sequenced tumors, with further mutations observed in GATA2 (n = 5), WT1 (n = 3), EZH2 (n = 2), SMC3 (n = 1), TET2 (n = 1), NRAS (n = 1), DDX41 (n = 1), and CSF3R (n = 1). These genes were also found to be mutated in sporadic CEBPA-mutated tumors, with the exception of EZH2 and DDX41 (Figure 2B).
In pedigrees A and B, we performed WES on 2 or more affected members, enabling comparison of somatic leukemic variants. Three members of pedigree A acquired 1 or more GATA2 mutations, all of which localized to the first zinc finger domain (A.II.1, p.L321F; A.II.5, p.L321F and p.R330Q; A.III.1, p.N317I). Furthermore, mother (B.I.1) and son (B.II.2) in pedigree B both acquired aUPD of chromosome 11p, with an underlying WT1 mutation detected in the mother’s leukemic DNA (WT1 p.D223-S233dup; supplemental Figure 2). This apparent intrafamilial convergence of tumors suggests that genetic lesions are acquired in a nonrandom manner that may be predetermined by features within the host genotype.
Instability of somatic CEBPA mutations at disease recurrence
Sanger sequencing of CEBPA at diagnosis and recurrence revealed discordant somatic mutations in 83% of tumor pairs (n = 5), suggesting the presentation of new leukemic episodes (Table 1). Novel mutations were predominantly acquired within the C-terminal (B.I.1, E.I.1, and F.IV.2), although 1 patient (B.II.2) acquired a somatic N-terminal mutation, and another reverted to a wild-type sequence at relapse (E.II.1). To verify that these mutations were unrelated to preexisting leukemic clones, we performed targeted deep sequencing of somatic CEBPA mutations at diagnosis and relapse in patients B.I.1 and E.I.1. Relapse-associated mutations were absent from preceding diagnostic tumors at sequencing depths exceeding 16 000× (B.I.1) and 143 000× (E.I.1), signifying that each mutation had developed de novo within an independent clonal expansion. Furthermore, somatic CEBPA mutations from the diagnostic tumor were not detected at relapse with sequencing depths exceeding 11 000× (B.I.1) and 212 000× (E.I.1), suggesting the primary episode had been cured. In a third patient, B.II.2, who presented with recurrence 15 months after diagnosis, we investigated the relapse status of 10 variants identified from WES of the primary tumor (supplemental Table 6). Notably, all 10 variants (7 clonal and 3 subclonal) were absent at disease recurrence, with detection of a novel somatic CEBPA mutation, further indicating a new clonal expansion. These data highlight the tumor specificity of somatic CEBPA mutations and signify that disease recurrence may be initiated by new leukemic episodes that are independent of the preceding tumor.
|Patient .||CEBPA somatic mutation-diagnosis .||CEBPA somatic mutation-relapse .||Time to relapse, years .|
|Amino acid .||Mutation .||Amino acid .||Mutation .|
|E.II.1||p.K313dup||c.1087_1089dup||Wild type||Wild type||6.6|
|Patient .||CEBPA somatic mutation-diagnosis .||CEBPA somatic mutation-relapse .||Time to relapse, years .|
|Amino acid .||Mutation .||Amino acid .||Mutation .|
|E.II.1||p.K313dup||c.1087_1089dup||Wild type||Wild type||6.6|
Molecular profiling reveals 2 distinct models of AML recurrence
Integrated analysis of WES and deep sequencing data across 3 consecutive tumor episodes in B.I.1 identified 2 distinct models of AML recurrence. B.I.1 first presented at 23 years (T1), achieving a 14-year remission following conventional chemotherapy. She developed AML recurrence at 37 years (T2) and underwent autologous stem cell transplantation (harvested during CR1), culminating in a third presentation at 40 years of age (T3). WES data revealed a high degree of genetic semblance between T1 and T3, with both tumors sharing clonal loss of chromosome 7q (∼50-Mb deletion, encompassing EZH2) and 12 identical mutations, including 10 coding variants (Figure 3A; supplemental Figure 2; supplemental Table 6). Subclonal mutations from T1 were absent in T3, most likely reflecting eradication by chemotherapy or clonal regression. In contrast, T2 was molecularly and structurally distinct from T1 and T3, with no shared somatic mutations or LOH. This tumor was characterized by 11p aUPD with 16 novel nonsynonymous mutations (including CEBPA p.K313del and WT1 p.D223-S233dup), consistent with the emergence of a novel leukemia.
To further characterize the ancestral relationship between T1, T2, and T3, we used deep sequencing to trace a selection of 28 variants across each disease episode (supplemental Figure 3A). Sequencing depths exceeded 70 000× for all samples, providing a variant detection sensitivity of 0.02% (15-20 supporting reads for 70 000-100 000× depth), with variant allele frequency (VAF) measurements supporting WES data (supplemental Figure 3B). All T1 and T3 somatic variants were absent from T2, confirming that the latter represented an independent leukemic episode (Figure 3B). WES data suggested that T1 and T3 had evolved from an identical ancestral clone, but deep sequencing more accurately defined the T3 founding population as a minor divergent subclone of T1, which harbored the TET2 p.R571X mutation in an estimated 3% of cells (supplemental Figure 3C). This nascent subpopulation of T1 appeared to survive chemotherapy and markedly expanded 3 years after autologous transplantation, to reemerge as the T3 founding clone (Figure 3C). T3 was characterized by chemotherapy resistance, indicating that this tumor was biologically distinct from T1 and T2. Although convergent evolution was observed between T1 and T3, with each acquiring different EZH2 mutations, the gain of novel subclonal GATA2 (p.L321P) and CSF3R (p.T618I) mutations led to further clonal heterogeneity in the final disease episode.
Clinical outcomes in familial AML
Survival probabilities were calculated at 10 years, revealing significant differences between CEBPA-mutated subgroups (Figure 4A-B). Overall survival (OS) in familial AML was superior to sporadic CEBPAsm (P = .003 for all ages and P = .03 for ≤45 years), with no significant difference observed between familial and sporadic CEBPAdm patients (P = .4 for all ages, P = .6 for ≤45 years). The CR1 rate of familial AML was 91% (n = 21/23), although disease recurrence was frequent, and often a late event, as shown in the clinical timeline (Figure 1). The cumulative incidence of relapse in familial AML was 56% at 10 years, with the first recurrence presenting at a median of 27 months (patients with relapse, n = 11; range, 4-167 months). The cumulative incidence of relapse rates in sporadic CEBPAsm and CEBPAdm were 62% and 44%, respectively. Relapse occurred earlier in sporadic AML (median time to relapse CEBPAsm, 11.8 months; CEBPAdm, 11.2 months), but this did not differ significantly from familial AML (P = .07, familial vs sporadic AML).
Recurrence of familial AML appeared to retain sensitivity to chemotherapy; the median survival after relapse was 8 years (familial) vs 16 months (CEBPAdm) vs 3.5 months (CEBPAsm) (P = .01; P = .006 for familial vs sporadic CEBPAsm and P = .03 for familial vs all sporadic; Figure 4C). This ongoing sensitivity to treatment was further illustrated by the clinical course of several patients including A.I.1 and F.IV.2, who achieved a CR3 of 45 and 17 years, respectively (Figure 1). Of 7 familial patients who died following treatment of disease recurrence, 71% (n = 5) were in remission, including 3 in CR3 or CR4. These patients had treatment (cardiotoxicity, G.II.2) or transplant-related complications (graft vs host disease [E.II.1], posttransplant lymphoma [F.V.1], and infection [F.IV.2]), with 1 developing esophageal cancer several decades later (A.I.1). Such observations of late chemotherapy-sensitive recurrence support the hypothesis of novel, independent clones initiating new leukemic episodes. More importantly for clinicians, they reveal important biological and prognostic insights into the clinical disease course of familial AML.
This multicenter study documents our collective experience of managing this rare subgroup of patients, providing extended clinical follow-up and molecular profiling of familial AML at diagnosis and relapse. Germline CEBPA mutations appear highly penetrant, although the identification of 3 unrelated healthy adult carriers suggests the possibility of inter- and intrafamilial variations in mutation penetrance. The majority of AML cases presented at an early age, and, in contrast to other leukemia predisposition syndromes, germline CEBPA mutations were associated with the occurrence of de novo disease, without a preceding dysplastic or cytopenic phase. Both the molecular and clinical profiles of familial tumors resembled those of sporadic CEBPAdm AML.
For the first time, we describe the molecular events underlying disease progression and evolution in familial AML. In contrast to multiple reports34-36 describing the stability of CEBPA mutations in sporadic AML, the opposite was observed in familial AML, with the majority of cases demonstrating variation of somatic CEBPA mutations at disease recurrence. Deep sequencing confirmed that these mutations were not derived from preexisting subclones but had developed de novo, suggesting the occurrence of novel, independent leukemic episodes arising from highly penetrant germline mutations.
This unique model of disease recurrence was distinguished from conventional patterns of AML relapse37 by in-depth molecular profiling of consecutive leukemic episodes within a single patient. We identified the expansion of a novel leukemic clone initiating the first episode of disease recurrence, whereas the second evolved from a subclone of the primary tumor (as observed in sporadic AML). Although the latter 2 episodes represented independent clonal expansions, their respective WT1 and TET2 mutations both targeted a common pathway of DNA hydroxymethylation.38
Residual leukemic variants from the primary episode were not detected by deep sequencing, suggesting that leukemia-initiating cells may have been reintroduced following autologous transplantation. Because nascent primary leukemic populations cannot be fully excluded, the expansion of an independent leukemic episode may also be explained by novel clones outcompeting preexisting clones. Further analysis of leukemia predisposition syndromes will elucidate whether this phenomenon is unique to germline CEBPA mutations or whether it universally applies to all familial leukemia subtypes. It seems likely that the variable penetrance of RUNX1 and GATA2 mutations (40-80%) may favor conventional models of relapse, with expansion of preexisting clones.
The acquisition of shared genetic lesions among family members suggests intrafamilial convergence of tumors. Despite limited studies to date, this phenomenon does not appear to be restricted to germline CEBPA mutations. An analogous observation was reported within a GATA2-mutated pedigree, where 2 cousins developed myelodysplasia (MDS)/AML, both acquiring identical somatic ASXL1 mutations (c.1934dupG, p.G646WfsX12) and monosomy 7.39 Enrichment of somatic ASXL1 mutations was later reported in 30% of germline GATA2 -mutated patients, coinciding with progression to a proliferative MDS phenotype.40 More recently, Yoshimi et al41 reported the inter- and intrafamilial acquisition of somatic CDC25C mutations in families with germline RUNX1 mutations. These observations suggest it is possible that germline mutations may influence the acquisition or selection of specific cooperating mutations and that the susceptibility to mutation acquisition may be governed by inherited factors, often shared within families.
From a clinical perspective, our data reinforce the need for long-term surveillance of families with germline CEBPA mutations and, crucially, highlight the need for reassessment of somatic mutations at relapse. Although recurrent episodes arising from novel leukemic clones may retain chemotherapy sensitivity, allogeneic hematopoietic stem cell transplantation remains an important consideration for the prevention of future disease episodes. Comprehensive genetic counseling and screening is essential prior to the selection of related stem cell donors, with the possibility of identifying asymptomatic mutation carriers often weighing significant anxiety on families.
The accurate identification of familial leukemia requires clinical vigilance and appropriate investigation. Earlier reports suggest that 5% to 10% of all CEBPA-mutated leukemias harbor germline mutations.2,13 Evaluation of family history is of paramount importance, although, in practice, this information may not be readily available. As germline CEBPA mutations are unique among leukemia predisposition syndromes, presenting with de novo AML in the absence of prodromal cytopenias, it is more likely that they may escape clinical recognition. We therefore advocate a thorough investigation of family history and germline DNA assessment of patients <50 years, presenting with CEBPAdm AML. Importantly, the distinction of pathogenic CEBPA mutations has been facilitated by the creation of extensive databases from large sequencing consortia, minimizing the potential for confusion with common germline polymorphisms.42
By demonstrating a unique model of disease evolution, we reveal important differences underlying the biology of sporadic and familial leukemias. With improved clinical awareness and ongoing research efforts, we hope to provide a foundation for evidence-based genetic counseling and tailored management of this unique patient population.
The online version of this article contains a data supplement.
There is an Inside Blood Commentary on this article in this issue.
The publication costs of this article were defrayed in part by page charge payment. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.
The authors thank the patients and their families for donating specimens for research in this study and G. Clark at the London Research Institute for the automated DNA sequencing work.
This study was predominantly funded by Cancer Research UK (CR-UK) through a Clinical Research Fellowship awarded to K.T. and Leukaemia Lymphoma Research Programme award 14032 (to J.F. and I.D.). Additional funding was awarded from Children with Cancer (J.F.), Grants-in Aid for Scientific Research from Japanese Ministry of Education, Culture, Sport, Science and Technology JSPS KAKENHI grant 26461431 (to N.A.), German Research Foundation for the Cluster of Excellence (REBIRTH) (from Regenerative Biology to Reconstructive Therapy; to B.S.), Hungarian Scientific Research Fund (Országos Tudományos Kutatási Alapprogramok) grant PD-108805 (to C.B.), Kay Kendall Intermediate Fellowship (to F.W.V.D.), and the NovoNordisk Foundation (to B.P.).
Contribution: K.T., J.W., C.B., R.H., and J.F. wrote the paper; K.T., A.R., A.S., F.W.V.D., J.T., N.A., N.U., M.D., J.J., P.A., X.T., V.M., M. Smith, J.C., B.P., C.N., I.D., B.S., C.O., J.G., C.P., and J.F. designed research and contributed vital materials; K.T., C.L., P.G., E.U., E.W., S.I., A.-K.F., and F.B.L. performed research; and J.W., R.H., K.T., C.B., F.W.V.D., V.M., X.T., K.D., L.B., B.M., T.P., M. Stelljes, S.A., J.O., R.G., C.O., and C.C. performed data analysis.
Correspondence: Kiran Tawana, Centre for Haemato-Oncology, Barts Cancer Institute, Queen Mary, University of London, John Vane Science Centre, Charterhouse Square, London EC1M 6BQ, United Kingdom; e-mail: email@example.com.