Integrative genomic analysis reveals cancer-associated mutations at diagnosis of CML in patients with high-risk disease

Chronic myeloid leukemia (CML) represents the prototype of genetically based diagnosis and management, and tyrosine kinase inhibitors (TKIs) exemplify the success of molecularly targeted therapy, with average life expectancy approaching that of the general population.¹  However, therapy still fails in a proportion of patients. At 10 years of first-line imatinib, 47% of patients were alive and still receiving imatinib.²  Up to 26% of all imatinib discontinuation was for drug resistance.^2-4  More-potent inhibitors rescued response for a proportion of patients with resistant chronic-phase (CP) disease; however, the risk of transformation and death for imatinib-resistant patients was 34% to 43%.^5,6 

BCR-ABL1 kinase domain (KD) mutations remain the major known mechanism of acquired TKI resistance,^7-12  but they are rarely responsible for primary resistance. The events initiating BCR-ABL1–independent resistance remain poorly understood.¹³  However, studies have demonstrated that transformation to blast crisis (BC) is due to the accumulation of genetic abnormalities detectable as additional cytogenetic changes and mutations in individual genes identified by deep sequencing.^14-22  Furthermore, gene expression changes may distinguish CP from BC and predict response to TKI therapy.^23-27  The increasingly mainstream use of next-generation sequencing for studying cancer genomes is advancing our understanding of disease pathogenesis and prognosis.^28-32  In CML, this technology may reveal the most pathologically relevant additional genetic events to identify high-risk patients and direct trials to test new combinational treatment approaches.

Here, we used whole-exome (WES) and whole-transcriptome sequencing (RNA-Seq) to determine somatic point or small insertion/deletion (indel) variants, copy-number variation (CNV), and fusion genes that underlie disease transformation. Patients with outlier responses have been studied to identify biomarkers of drug response,^33-35  which demonstrates the potential of such cases to advance understanding of the genetic basis of treatment outcome. We therefore included diagnosis samples of patients at the extremes of TKI response: achievement of a major molecular response (MMR) vs BC transformation. Our study revealed clinically relevant variants in addition to BCR-ABL1 at CP diagnosis in patients with poor outcome and in all patients at BC. The types of variants were diverse and included gene fusions and CNVs. Many of the mutated genes have been identified in CML and other hematologic diseases.^15-22,29,36  However, some of the recurrent variants, mutated genes, fusions, and splice variants are novel in CML. We highlight the importance of incorporating multiple sequencing platforms to fully characterize the various mutation classes.

Sixty-five patients were studied, and the type of sequencing technique performed was dependent on the sample type available: DNA and/or RNA (Figure 1). DNA was available for WES for 38 of 65 patients. The somatic status of variants was assessed with a matching nonleukemic control sample in 35 of these 38 patients. Two sources of control were used: mesenchymal stromal cells or remission samples. RNA-Seq was possible for 59 of 65 patients.

At diagnosis, patients were selected from among >500 patients for sequencing in CP on the basis of their subsequent response to first-line TKIs and availability of appropriate samples. Nineteen patients had an optimal response, defined as durable MMR (median time to MMR, 3 months; range, 3-55 months). MMR was durable, with a median follow-up of 34 months (range, 18-174 months). Twenty-seven patients had a poor outcome, where 26 developed BC after a median of 6 months (range, 1-60 months), and for 1 patient, 4 TKIs had failed by 18 months (best response, complete hematologic response). These 27 patients were collectively termed poor-outcome patients. At BC, samples of 39 patients were sequenced, including 20 of the first-line TKI–treated patients who were also sequenced at diagnosis. The study was approved by the Human Research Ethics Committee of the Royal Adelaide Hospital, and patients provided written informed consent.

For WES, we used the Roche SeqCap EZ Exome v3.0 kit for 100-bp paired-end sequencing. Somatic variants were called using published algorithms (supplemental Methods, available on the Blood Web site). CNV detection in normal and tumor samples was performed using a method developed in house (supplemental Methods). Structural variants and genomic fusions were additionally assessed using Manta.³⁷ 

RNA-Seq libraries were prepared using the Illumina TruSeq Stranded Total RNA protocol for 100-bp or 150-bp paired-end sequencing. The protocol proved capable of identifying sense and antisense spliced fusion transcripts as well as genomic breakpoints from intron-retaining precursor messenger RNA that are usually absent in standard poly(A)-enriched RNA-Seq. Gene fusions and other genomic rearrangements were analyzed using in-house tools based on output provided by the STAR RNA-Seq mapper³⁸  (supplemental Methods), and single-nucleotide variants (SNVs) and indels in the RNA-Seq data were detected using FreeBayes.³⁹  Differential gene expression analysis is described in the supplemental Methods. Sequencing was performed on Illumina HiSeq2500 and NextSeq 500 instruments.

The sequence data have been deposited at the European Genome-phenome Archive, which is hosted by the European Bioinformatics Institute, under accession #EGAS00001003071.

Variants were selected for validation (>3500) using independent methods (supplemental Methods and Results), including Sanger sequencing, reverse transcription polymerase chain reaction, and single-nucleotide polymorphism array. A comparison was performed between the somatic variant calls when either mesenchymal stromal cells or remission samples were used as the source of nonleukemic control, and the minimum detectable somatic variant allele frequency was established (supplemental Results).

Annotation of variants is described in the supplemental Methods. Indels in protein-coding regions and nonsynonymous and essential splice-site SNVs predicted to be deleterious using in silico functional prediction models were retained and classified according to clinical relevance, as were fusions and CNVs (supplemental Methods). Missense-coding variants designated as deleterious were predicted to be damaging by at least 3 of the 4 functional prediction algorithms, which indicates that they are likely to modify protein function. These variants also met minimum evolutionary conservation scores (supplemental Methods).

Classification systems for clinical relevance were adapted to enrich for potentially pathogenic variants.^40-42  Retained deleterious variants were classified as: 1) clinically relevant, 2) possibly relevant, or 3) unknown relevance (supplemental Methods). Briefly, clinically relevant variants were associated with hematologic malignancy and occurred in genes listed in the Catalogue of Somatic Mutations in Cancer Cancer Gene Census.⁴³  These genes have high-level evidence as drivers of hematologic malignancy and are hereafter termed cancer genes. Possibly relevant variants occurred in genes recurrently mutated in our cohort. Novel somatic fusions and recurrent CNVs were also considered possibly relevant.

Groups were compared using the Mann-Whitney rank sum test. Fisher’s exact test and the χ² test were used to compare frequencies.

Patient characteristics for the 46 patients diagnosed in CP and treated with first-line TKIs are listed in Table 1, including the best response to first-line TKI therapy. The type of sequencing performed for each patient is outlined in supplemental Table 1 and summarized in Figure 1.

At diagnosis, variants that met our criteria for clinically relevant or possibly relevant were detected in 23 (50%) of 46 patients (Figure 2; supplemental Table 1). A majority occurred in cancer genes and were detected in patients with a poor outcome: 19 (70%) of 27 patients compared with 4 (21%) of 19 patients with subsequent MMR (P = .002). All coding variants are listed in supplemental Tables 6 and 7. When available, RNA-Seq data established whether somatic variants were detected in the transcripts. The coding variants classified as relevant were expressed, whereas some of the others were not. These were likely to be passengers that do not play a role in treatment response.

Among the 27 poor-outcome patients, 26 developed BC. The median time to BC was earlier for those with relevant variants at diagnosis compared with those without: 5 months (range, 1-47 months; n = 18 patients) vs 15 months (range, 6-60 months; n = 8 patients; P = .018). The most frequently mutated gene at diagnosis was ASXL1 in 9 patients. Six of these developed BC. Patients with mutated ASXL1 at diagnosis had a significantly longer time to BC than patients with other mutated genes at diagnosis that did not include ASXL1: 21 months (range, 4-47 months) vs 4.5 months (range, 1-23 months; P = .037). Furthermore, 2 patients with ASXL1 variants (allele frequencies, 22% and 40%) achieved an MMR at 3 months of first-line imatinib. The variants were not detected at remission, suggesting they were present in a leukemic clone and were reduced or eradicated with effective imatinib treatment.

Other recurrently mutated genes at CP diagnosis included cancer genes RUNX1, TP53, and IKZF1. IKZF1 variants occurred in 6 patients and comprised whole-gene (n = 1) or exon deletions (n = 3) and novel fusions (n = 2). Five of these 6 patients developed lymphoid BC (LBC) after 3 to 23 months of imatinib. A novel recurrently mutated gene was the histone methyltransferase SETD1B, in which somatic frameshift/nonsense variants occurred in 3 patients with subsequent BC. This gene, although categorized as a cancer gene in the Catalogue of Somatic Mutations in Cancer, is currently not associated with hematologic malignancy.

Thirty of the 46 patients sequenced at diagnosis had WES analysis performed and were evaluable for somatic CNV analysis. We used an algorithm developed in house, which demonstrated a high concordance (97.6%) with a gold-standard technique: Affymetrix CytoScan HD array (supplemental Results; 123 somatic CNVs were evaluable from the total cohort of patients). At diagnosis, CNVs were detected in 7 (23%) of 30 patients (supplemental Table 8). These included deletions associated with the Ph translocation in 4 of 7 patients. Six of the 7 patients with CNVs at diagnosis developed BC (n = 5) or had no response to 4 TKIs (n = 1).

Among 23 patients with poor outcome where the best response to first-line TKIs was known, 8 (30%) had primary resistance, defined as failure to achieve a major cytogenetic response. All of these patients had relevant variants at diagnosis (supplemental Table 1), compared with 6 of 9 with a major cytogenetic response (67%) and 2 of 6 with a complete cytogenetic response/MMR (33%; P = .027). The 1 patient who achieved but did not sustain an MMR transformed rapidly at 25 months of imatinib and acquired an MLL fusion.

The Sokal risk score was known for 37 of 46 CP patients. There was an association between Sokal score and relevant variants at diagnosis: high, 86% (6 of 7 patients); intermediate, 42% (5 of 12 patients); and low, 28% (5 of 18 patients; P = .03). We also tested for differentially expressed genes between good and poor responders at diagnosis to determine if an expression signature could be identified for use as a classifier for outcome. However, the expression analysis did not result in any genes differentially expressed at the level of P < .05 (Benjamini-Hochberg adjusted; supplemental Figure 18).

Thirty-nine patients had sequencing performed at BC, including 20 of the first-line TKI–treated patients who were sequenced at CP diagnosis described in “Patient samples and type of sequencing” (Figure 1). All 39 patients had 1 to 6 variants in cancer genes at BC (supplemental Table 1; Figure 3). Variants were acquired in genes that were also somatically mutated in other patients at CP diagnosis: IKZF1, RUNX1, ASXL1, BCORL1, and IDH1.

The most frequently mutated gene was ABL1, where KD mutations were detected in 19 (58%) of 33 TKI-treated patients. These mutations were more frequently detected at LBC (16 [84%] of 19 patients) compared with MBC (3 [21%] of 14 patients; supplemental Results). The average allele frequency of the KD mutations was 31% detected by WES (range, 13%-65%) and 47% by RNA-Seq (range, 10%-88%). All of these mutations were known imatinib-resistant mutations,^44,45  and all were confirmed to be present in the BCR-ABL1 transcript by Sanger sequencing.⁸  Only 2 of 19 patients had ABL1 KD mutations as the sole mutated cancer gene. Pairwise interactions between mutated genes were assessed and potential interactions identified (supplemental Results; supplemental Table 10). The number of mutants was low, but no patient with an MLL fusion had an ABL1 KD mutation. Conversely, ABL1 KD mutations were frequently observed in patients with IKZF1 variants.

Twenty patients had WES analysis performed at BC and were evaluable for somatic CNV analysis. CNVs were detected in 18 (90%) of 20 patients (supplemental Table 8; supplemental Figures 4 and 5). Recurrent CNVs involved large regions of chromosomes 8, 17, and 19 and the Ph chromosome [der(22)]. Patients with LBC had a high frequency of submicroscopic deletions of immunoglobulin and T-cell receptor loci, consistent with clonal populations of lymphoid progenitors. A novel recurrent ∼60-Kb deletion comprising exons 1 to 2 of HBS1L and an intergenic sequence between HBS1L and MYB (termed HBS1L-MYB locus deletion) was detected in 3 of 10 patients with LBC (supplemental Figure 6). Genes with recurrent submicroscopic deletions included RUNX1 (Figure 4) and IKZF1 (supplemental Figure 9), and atypical splice junctions were frequently detectable by RNA-Seq in patients with exon deletions (Figure 4; supplemental Figures 8 and 9).

Thirty-three patients had RNA-Seq performed at BC and were assessed for gene fusions. Fourteen patients (42%) had 18 known or novel fusions that were not associated with the Ph translocation (see “Novel class of variant associated with the Ph translocation”). MLL (KMT2A) fusions were the most frequent (n = 5), and 2 of these were cytogenetically cryptic. The other known fusions were CBFB-MYH11 and PAX5-ZCCHC7. Novel fusions in 6 patients involved cancer genes implicated in hematologic malignancy: RUNX1, IKZF1, MECOM, and MSI2 (supplemental Table 11).

In total, 14 (78%) of 18 fusions at BC were known leukemia-associated fusions (n = 7) or were fusions in which a fusion partner was a known leukemia-associated gene (n = 7). A number of the novel fusions were cytogenetically cryptic, including an intrachromosomal RUNX1-MX1 out-of-frame fusion. A submicroscopic 6.6-Mb inversion brought the RUNX1 gene into the same transcriptional orientation as MX1 (supplemental Figure 10). A similar inversion generated a novel intrachromosomal MBNL1-MECOM fusion. A complex series of events generated a cryptic MSI2-PRMT2 fusion, including a 3-way exchange of sequence from chromosomes 17, 19, and 21 (supplemental Figure 11). In total, 151 of the fusions that met our criteria for relevance were evaluated using reverse transcription polymerase chain reaction, and 150 (99%) were validated (supplemental Results).

Gene expression analysis among patients with specific mutated genes was confounded by multiple mutations in different cancer genes in most patients. These multiple variants included SNVs, indels, and fusions, plus CNVs in some patients. However, expression analysis between disease phases revealed differentially expressed genes (supplemental Results). Samples of TKI-treated patients at MBC (n = 11) or LBC (n = 18) were compared with the diagnosis samples of 43 patients who were subsequently treated with first-line TKIs. Supplemental Figure 18A shows a heat map for hierarchical clustering of 1156 differentially expressed genes with an adjusted P value <.001. Two clusters were dominated by the BC samples, whereas there was no discernable clustering of the diagnosis samples according to treatment outcome. One BC cluster only included patients with LBC (n = 15), and the predominant genes mutated were ABL1 and IKZF1 in 93% and 53% of patients, respectively (supplemental Figure 18B). Most patients in the other BC cluster had MBC (10 [63%] of 16 patients), and the predominant variants were fusions (not Ph-associated) in 56% of patients. All 5 patients with MLL fusions at BC clustered in this group, including 2 patients with LBC.

Twelve of 59 patients with RNA-Seq performed had gene rearrangement and novel fusions on the chromosome arms involved in the Ph translocation (supplemental Table 11; Figure 5). These are termed Ph-associated rearrangements. They were detected in 11 patients at diagnosis and in 1 patient tested only at BC. The somatic status of the rearrangements was confirmed in 5 of 5 patients with available nonleukemic control. No newly emerging Ph-associated rearrangements were detected at BC, suggesting they occurred with the initiating translocation.

The Ph-associated rearrangements were more frequently detected at diagnosis in patients with poor outcome: 9 (33%) of 27 first-line TKI–treated CP patients who developed BC (n = 8) or for whom 4 TKIs failed (n = 1) vs 2 (11%) of 19 with sustained MMR (P = .07). The events generated 15 novel fusion transcripts, plus 10 genomic fusions where no corresponding fusion transcript was detected. The fusions frequently involved inversions that brought genes into the same transcriptional orientation. In addition, 4 patients had intragenic inversions of ABL1 or BCR. Figure 5D shows a 93-Kb intragenic inversion of ABL1 intron 1. In total, sequence inversions were detected in 10 (83%) of 12 patients.

Interestingly, fusion partner genes were located both upstream and downstream of BCR and ABL1 (Figure 5), and some fusions were associated with deletions on chromosomes 9 and 22. These were detected in 4 of 7 patients with CNV analysis performed. An insertion of NUP214 exon 23 occurred in 1 patient between BCR exon 19 and ABL1 exon 2, which generated an in-frame BCR-NUP214-ABL1 transcript (supplemental Results). NUP214 is normally located downstream of ABL1. The additional exon could potentially alter the conformation of a generated protein and modify sensitivity to imatinib. The observed gene fusions in some patients and their transcriptional orientation provide evidence of a complex series of events, including chromosome fragmentation, sequence inversion, and imperfect reassembly generating random fusion (supplemental Figures 14-16).

Various patterns of mutation acquisition with disease stage were evident. First, 5 patients diagnosed in accelerated phase/BC had 1 to 2 cancer gene variants at diagnosis: CBFB-MYH11, RUNX1, GATA2, BCOR, BCORL1, and an MSI2 fusion.

Second, 18 CP patients had 1 to 4 cancer gene variants at diagnosis, including recurrently mutated genes TP53, RUNX1, IKZF1, and ASXL1. Fifteen of these 18 developed BC (n = 14; median, 6 months; range, 3-47 months) or had no response to 4 TKIs (n = 1). Eleven of 12 patients who were tested during therapy gained 1 to 5 additional variants in cancer genes (Figure 6). The other 3 of 18 patients achieved and maintained MMR: 2 had somatic ASXL1 variants, and 1 had a novel somatic IKZF1-TCF4 fusion at diagnosis.

The third pattern involved 14 patients tested at CP diagnosis who did not have variants in cancer genes but developed BC (median, 14 months; range, 1-64 months). Eleven of these 14 were tested at BC, and all gained 1 to 4 cancer gene variants. Four patients had samples available to track the time point of variant acquisition by Sanger sequencing (Figure 6).

Of 16 patients with ABL1 KD mutations at BC who were sequenced at prior time points, 10 (62%) had cancer gene variants that predated the ABL1 mutations. Clonal competition between variants was evident over time, where the allele frequency decreased after diagnosis for some variants and increased for others (Figure 6).

One patient had DNMT3A and TET2 variants detected in a remission sample (Figure 6N), and the pattern of clonal selection is consistent with their presence in a BCR-ABL1⁻ clone, which is detailed in the supplemental Results.

Our study has demonstrated that somatic variants in cancer genes in addition to the BCR-ABL1 fusion are present at diagnosis in a significant subset of patients, particularly in those patients who, despite TKI therapy, experience early BC transformation. The variants included gene fusions, splice anomalies, and CNVs, and this diversity of mutation type would not have been detected by DNA sequencing alone. This underscores the importance of multiple modes of variant detection to reveal relevant genomic events in addition to the BCR-ABL1 fusion.

Among the genes mutated at diagnosis were RUNX1, ASXL1, and TP53, which confer a poor prognosis for patients with acute myeloid leukemia.⁴⁶  Their role in prognosis and treatment outcome for patients with CML is yet to be established. RUNX1 is a frequent and aggressive driver of hematologic malignancies^15,46-48  and has been described in multiple CML studies, but it is not always associated with poor outcome.^16-22  In our cohort, mutated RUNX1 was rarely detected as the sole mutated cancer gene. We speculate that in optimal responders to TKIs, certain leukemic clones with mutated genes at diagnosis may be eradicated by TKIs. This may be dependent on a number of factors, including cooccurring mutations, the variant allele frequency, and the varying degree of innate TKI resistance conferred by particular variants. In our cohort, patients with an ASXL1 variant at diagnosis had a longer median time to BC than other patients. Furthermore, 2 of 9 patients with an ASXL1 variant at diagnosis achieved an MMR. One of these variants occurred in exon 11. A vast majority of ASXL1 variants are exon 12 frameshift/nonsense and may be gain-of-function mutations,⁴⁹  whereas the functional significance of variants in other regions are unknown. The significance of mutated genes at diagnosis and their association with outcome warrant further investigation.

An unexpected finding was complex rearrangements and novel fusions associated with the Ph translocation. Chromosome 9 and 22 sequences were possibly fragmented during the Ph translocation in some patients. Some fragments were deleted and others inverted. Remaining fragments were reassembled imperfectly and generated random fusions, while maintaining the leukemia-initiating BCR-ABL1 fusion. Limitations in previous technology may have precluded their discovery. These occurred more frequently in patients with a poor outcome vs optimal responders and may constitute an important additional genomic risk factor. Derivative 9 deletions, which mainly spanned the translocation breakpoint, were associated with poorer outcome in the pre-TKI era.^50-52  However, some of our deletions were noncontiguous and at locations not detectable by older techniques.^50-54  Interestingly, a novel NUP214-XKR3 9;22 fusion has been reported in 2 independent studies in the BCR-ABL1⁺ cell line K562 that would require a similar series of events for formation, including sequence inversion.^55,56  The role of the Ph-associated fusions in treatment outcome is unknown. We have revealed another layer of molecular heterogeneity, leading us to speculate that there may be a subcategory of patients in whom Ph-associated gene fusions/inversions modify outcome with TKI therapy.

Our study revealed a high rate of other gene fusions (42% of patients at BC), and most were either fusions known to be associated with hematologic malignancy or fusions in which 1 of the fusion partners was a known cancer gene.^46,57-60  However, it will be important to functionally characterize the novel fusions to establish whether they are transforming and whether they confer TKI resistance or reduced sensitivity, which is planned in future studies.

Many of the mutated genes may be targets for emerging therapy. IDH1/2 inhibitors are in clinical trial for the treatment of acute myeloid leukemia.^61,62 ASXL1 mutants were hypersensitive to recently developed bromodomain inhibitors.⁴⁹  We detected an XPO1 E571K mutation in an isolated patient at LBC. This variant may play a role in B-cell hematologic malignancies,⁶³  and XPO1 is implicated in ABL1-independent resistance.^64,65  XPO1 inhibitors have demonstrated antileukemic activity.^65,66 MLL rearrangements may be targeted by inhibitors.^67,68  These were the most frequent fusions at BC, and some were cytogenetically cryptic, which emphasizes the advantage of using additional genomic tools to identify potentially targetable mutations.

The mutational spectrum in cancer genes was diverse, and subcohorts of patients with specific mutations were small in number. For this reason, we generally could not find specific gene expression signatures correlated with mutations in a statistically meaningful fashion. General differences were observed between the transcriptomes of diagnosis and BC samples, and in the latter, differences were seen between MBC and LBC samples. These differences in expression were accompanied by differences in mutational patterns.

Our study is among the most comprehensive genomic analyses of patients with CML. It included matched diagnosis/BC samples in many patients, coupled with nonleukemic controls, to determine the somatic status of variants. BC is a rare event for TKI-treated patients, and whether these findings extend to a broader population of patients will require assessment of a large cohort of unselected and consecutively treated patients. Future genomic analyses may reveal prognostic risk groups, and combining genomic data with clinical parameters could improve prognostication and disease classification, as it has in other hematologic malignancies.^46,69 

Although we cannot generalize our findings to current clinical practice, there are already hints that we can generate clinically useful data. There was marked heterogeneity of mutated genes in individual patients, with up to 5 mutated cancer genes at BC. Furthermore, by tracking variants in prior samples, clonal competition among mutants was evident. Additional studies will add to our understanding of treatment response and clarify the clinical usefulness of genomic analysis. These analyses are becoming increasingly affordable and in the future may be routine at cancer diagnosis. It is important to prepare for this era by gaining a deeper understanding of the somatic genome and its association with outcome in CML. Furthermore, collection of a matched nonleukemic sample at diagnosis, such as hair follicles or nail clippings,⁷⁰  will aid the future interpretation of data.

Although generic imatinib has become the standard of care for patients with CML in many countries because of its cost effectiveness and safety profile, it is likely that selected patients would benefit from a therapeutic approach incorporating more potent kinase inhibition or combinational therapy targeting additional pathways. If very high-risk cases could be identified with confidence, upfront allogeneic transplantation may even be considered. A detailed understanding of the prognostic implications of specific genetic variants will be critical to the achievement of a risk-adapted approach to CML therapy.

The authors thank the Korea Leukemia Bank for providing biospecimens and data (NRF-2013M3A9B8031236), the SA Cancer Research Biobank, the Australasian Leukaemia and Lymphoma Group, the Centre for Cancer Biology (an alliance between SA Pathology and the University of South Australia), Rosalie Kenyan for assistance with the SNP array work, John Toubia for valuable discussion, Monique Sinton for assistance with variant validation, and Sarah Moore, Jeff Suttle, and Mario Nicola for cytogenetic analysis and interpretation.

This work was supported by the National Health and Medical Research Council of Australia (APP1027531 and APP1104425 [S.B.], APP1135949 [T.P.H.], APP1023059 [H.S.S.], and APP1146253 [D.T.Y.]), the Ray and Shirl Norman Cancer Research Trust, the Royal Adelaide Hospital Research Foundation, and a grant from The Leukemia and Lymphoma Society. T.P.H. and H.S.S. are supported by Cancer Council SA’s Beat Cancer Project on behalf of its donors and the State Government of South Australia through the Department of Health. The Australian Cancer Research Foundation (ACRF) Cancer Genomics Facility was established with funding from Therapeutic Innovation Australia and the ACRF. I.W. was supported by a travel grant from Bloodwise United Kingdom. W.T.P. was supported by a postdoctoral fellowship from the Leukaemia Foundation of Australia/Cure Cancer Australia.

Contribution: S.B. designed the research, analyzed and interpreted the data, and wrote the manuscript; P.W. and A.W.S. performed the bioinformatic analyses, developed in-house bioinformatic pipelines, analyzed and interpreted the data, and wrote the manuscript; D.T.Y. designed the research, interpreted the data, provided clinical insight, and contributed to manuscript preparation; T.P.H. interpreted the data, provided clinical insight, and contributed significantly to manuscript preparation; W.T.P. and J.G. designed the research, performed experiments, and reviewed the manuscript; A.P., D.T., C.W., N.H.S., J.E.M., N.N., D.S., Z.D., H.A., J.G., J.B., and I.W. performed experiments and reviewed the manuscript; C.D., S.-H.K., S.-Y.C., and S.-H.P. provided samples and data and reviewed the manuscript; J.F. developed bioinformatic pipelines and reviewed the manuscript; H.S.S., A.B., C.H., and D.L.W. provided scientific insight and performed critical review of the manuscript; and D.M.R., A.S.M.Y., N.S., M.C.M., and D.-W.K. provided clinical insight and performed critical review of the manuscript.

Conflict-of-interest disclosure: S.B. received research funding and honoraria from and served on advisory committees for Novartis and Bristol-Myers Squibb and received consultancy fees and honoraria from Qiagen and Cepheid; D.T.Y. received research funding and honoraria from and served on advisory committees for Novartis and Bristol-Myers Squibb and has received research funding from Ariad; N.S. received travel sponsorship and honoraria from Novartis and Bristol-Myers Squibb; M.C.M. received support from Novartis, Bristol-Myers Squibb, and Ariad; A.B. received honoraria from Novartis; D.L.W. received research support from Novartis and Bristol-Myers Squibb and honoraria from Bristol-Myers Squibb; A.S.M.Y. received research funding from Novartis, Bristol-Myers Squibb, and Celgene and honoraria from Novartis and Bristol-Myers Squibb; D.-W.K. received research funding and honoraria from Novartis, Bristol-Myers Squibb, and Ariad; D.M.R. received research funding from Novartis and Celgene and honoraria from Novartis and Bristol-Myers Squibb; H.S.S. received honoraria from Celgene; and T.P.H. received research funding and honoraria from and served on advisory committees for Novartis, Bristol-Myers Squibb, Celgene, and Ariad. The remaining authors declare no competing financial interests.

Correspondence: Susan Branford, Department of Genetics and Molecular Pathology, SA Pathology, PO Box 14 Rundle Mall, Adelaide, SA, 5000, Australia; e-mail: susan.branford@sa.gov.au.