Juvenile myelomonocytic leukemia (JMML) is an intractable pediatric leukemia with poor prognosis1 whose molecular pathogenesis is poorly understood, except for somatic or germline mutations of RAS pathway genes, including PTPN11, NF1, NRAS, KRAS and CBL, in the majority of cases2, 3, 4. To obtain a complete registry of gene mutations in JMML, whole-exome sequencing was performed for paired tumor-normal DNA from 13 individuals with JMML (cases), which was followed by deep sequencing of 8 target genes in 92 tumor samples. JMML was characterized by a paucity of gene mutations (0.85 non-silent mutations per sample) with somatic or germline RAS pathway involvement in 82 cases (89%). The SETBP1 and JAK3 genes were among common targets for secondary mutations. Mutations in the latter were often subclonal and may be involved in the progression rather than the initiation of leukemia, and these mutations associated with poor clinical outcome. Our findings provide new insights into the pathogenesis and progression of JMML.
At a glance
JMML is a rare myelodysplastic/myeloproliferative neoplasm unique to childhood, characterized by excessive proliferation of myelomonocytic cells and hypersensitivity to granulocyte-macrophage colony-stimulating factor1. A cardinal genetic feature of JMML is frequent somatic and/or germline mutation of RAS pathway genes, such as NF1, NRAS, KRAS, PTPN11 and CBL, which are mutated in more than 70% of JMML cases in a mutually exclusive manner2, 3, 4. However, it is still open to question whether RAS pathway mutations are sufficient for the development of JMML or if secondary mutations have a role in the development and progression of this cancer. To address these issues and to better define the molecular pathogenesis of JMML, we performed whole-exome sequencing of paired tumor-normal DNA from 13 cases (Supplementary Table 1). We obtained mean coverage in exome sequencing of 137× for tumor samples and 143× for normal samples (Supplementary Fig. 1). A Monte-Carlo simulation indicated that the study detected 88% of the existing somatic mutations (Online Methods and Supplementary Fig. 2).
Sanger sequencing of 25 candidate non-silent somatic nucleotide alterations confirmed 1 nonsense and 10 missense mutations (Table 1 and Supplementary Fig. 3), with the low true positive rate consistent with the very low numbers of somatic mutations in JMML. Of the 11 somatic mutations, 6 involved known RAS pathway genes. In addition, non-overlapping RAS pathway mutations (6 somatic and 6 germline) were confirmed in 11 of the 13 discovery cases (86%; Table 1). For the remaining two cases that lacked documented RAS pathway mutations, we intensively searched for possible germline mutations that could be relevant to the development of JMML. In total, 179 and 167 candidate germline mutations were detected in subjects 77 and 92, respectively, but these mutations did not affect known RAS pathway genes or other cancer-related genes, including the ones registered in the pathway databases (Online Methods). A frameshift deletion in KMT2D (also known as MLL2; encoding p.Val1670fs) was found in subject 92, who had been diagnosed as having Noonan syndrome on the basis of typical features such as hypertelorism, webbed neck and congenital heart disease (Supplementary Fig. 3) but lacked the distinctive facial appearance of Kabuki syndrome, which was shown to be caused by germline KMT2D mutations5.
Five of the 11 somatic mutations were non–RAS pathway mutations, involving SETBP1 (3 p.Asp868Asn alterations), JAK3 (1 p.Arg657Gln alteration) and SH3BP1 (1 p.Ser277Leu alteration), which had not been reported in JMML cases. SETBP1 was originally isolated as a 170-kDa nuclear protein that interacts with SET, a small protein inhibitor of the putative tumor suppressors PP2A and NM23-H1 (ref. 6). Several lines of recent evidence suggest that SETBP1 has a role in leukemogenesis (Supplementary Fig. 4)7, 8, 9, 10, 11. SETBP1 participates in translocations that result in an aberrant fusion gene (NUP98-SETBP1) and overexpression of SETBP1 in T cell acute lymphoblastic leukemia (T-ALL) and acute myeloid leukemia (AML), respectively12, 13. SETBP1 is one of the downstream targets induced by the Evi-1 oncoprotein14 and, together with EVI1 and its homolog PRDM16 (also known as MEL1), was reported to be activated through retrovirus integration. SETBP1 is also known to augment the recovery of granulopoiesis after gene therapies for chronic granulomatous disease15. SETBP1 overexpression is found in more than 27% of adult AML cases and is associated with poor survival13. The discovery of recurrent hotspot mutations of SETBP1 provides unequivocal evidence for the leukemogenic role of deregulated SETBP1 function. Notably, the SETBP1 mutation encoding p.Asp868Asn was identical to one of the de novo mutations reported to be causative in Schinzel-Giedion syndrome (SGS; MIM 269150), which is a highly recognizable congenital disease characterized by severe mental retardation, distinctive facial features and multiple congenital malformations. Individuals with SGS with this mutation have a higher than normal prevalence of tumors, including of neuroepithelial neoplasia16, although development of myeloid malignancies has not been reported so far.
To further validate our findings, we screened the entire cohort of 92 JMML cases for gene mutations in the newly identified 3 genes together with known RAS pathway targets using deep sequencing17 (Supplementary Fig. 5).
RAS pathway mutations were found in 82 of 92 cases (89%) in a mutually exclusive manner, with PTPN11 mutations predominant, followed by NRAS, KRAS, CBL and NF1 mutations (Fig. 1a and Table 2). In accordance with previous reports, most of the CBL (8/14) and NF1 (4/9) mutations were biallelic (Fig. 1a,b and Supplementary Table 2)2, 3, 18, whereas the majority of mutations in PTPN11, NRAS and KRAS were heterozygous4. The individuals without RAS pathway mutations (n = 10) were vigorously investigated by whole-genome sequencing of tumor-normal paired samples (n = 2; Supplementary Fig. 6) or by whole-exome sequencing of only tumor samples (n = 8; Supplementary Fig. 7). As anticipated, we found no known RAS pathway mutations.
On the other hand, 18 mutations were found in SETBP1 (n = 7) or JAK3 (n = 11) in 16 cases (Fig. 1a,b, Table 2 and Supplementary Table 2), with these mutations more frequent in cases with mutated PTPN11 (and possibly NF1) than in cases with mutated NRAS, KRAS or CBL (Fig. 2a). Mutations in SH3BP1, encoding SH3 domain–binding protein 1, were not recurrent. All SETBP1 mutations were heterozygous and occurred within the portion of the gene encoding the SKI domain, with six identical to the de novo recurrent mutations reported in SGS and five identical to the mutation encoding the p.Asp868Asn alteration (Fig. 1b). RT-PCR analysis showed that the wild-type and mutant alleles of SETBP1 were equally expressed (Supplementary Fig. 8). Similarly, 8 of the 11 JAK3 mutations in 10 cases were the well-described activating mutation (encoding a p.Arg657Gln alteration) found in various hematological malignancies, including Down syndrome–associated acute megakaryoblastic leukemia19, 20, 21, 22, 23, ALL24, 25 and natural killer (NK)/T cell lymphoma26, and the remaining 3 were also within the portions of the gene encoding the pseudokinase or kinase domain, suggestive of gain of function.
Deep sequencing of the relevant mutant alleles enabled an accurate estimation of allele frequencies for individual mutations (Supplementary Fig. 9). SETBP1 and JAK3 mutations showed lower allele frequencies (but not with statistical significance for SETBP1) than did the corresponding RAS pathway mutations (Supplementary Fig. 10a), indicating that the former mutations represent secondary genetic hits that contributed to clonal evolution after the main tumor population was established (Supplementary Fig. 10b). Individuals with secondary mutations had shorter lengths of survival compared to those without mutations: 5-year overall survival (hazards ratio (HR) = 1.90, 95% CI = 0.87–4.19). In addition, none of the individuals with JMML who survived without hematopoietic stem cell transplantation (HSCT; n = 26) harbored any of the secondary mutations, and individuals with secondary mutations showed significantly inferior 5-year transplant-free survival (HR = 2.18, 95% CI = 1.18–4.02) (Fig. 2b–d and Table 2).
JMML is characterized by a paucity of gene mutations. The average number of mutations per sample (0.85; range of 0–4) was unexpectedly low compared to those reported in other human cancers (Supplementary Fig. 11); excluding common RAS pathway mutations, only 5 mutations were detected in 3 of the 13 discovery cases. This small number of mutations is only comparable to the figure reported for retinoblastoma (mean of 3.3 per case; range of 0–5) (ref. 27) and is in stark contrast to the abundance of gene mutations in chronic myelomonocytic leukemia (CMML) in adult cases, where the mean number of non-silent mutations was 12.4 per sample, of which 3.1 represented known driver changes (ref. 17 and K.Y., M.S., Y.S., D. Nowak, Y. Nagata et al., unpublished data), underscoring the distinct pathogenesis in these two neoplasms that show indistinguishable morphology. The impact of germline events is underscored by the fact that 6 of the 13 discovery cases harbored germline RAS pathway mutations and an additional case without known RAS pathway mutations showed constitutive abnormalities similar to Noonan syndrome. Despite the central role of RAS pathway mutations, a small subset of cases had no documented RAS pathway mutations, even after whole-exome analysis in the two RAS pathway mutation–negative cases, raising the possibility that the latter cases represent a genetically distinct myeloproliferative neoplasm in childhood.
Another key finding in the current study is the discovery of secondary mutations that involve SETBP1 and JAK3. Detected only in a subpopulation of leukemic cells, most of these mutations are thought to be involved in the progression rather than the establishment of JMML and were associated with poor clinical outcome. SETBP1 is a newly identified proto-oncogene, and identical mutations in this gene have recently been reported in 15–25% of adult cases with atypical chronic myeloid leukemia (CML)10, CMML and secondary AML28. Affecting one of three highly conserved amino acid positions, SETBP1 mutations have been shown to abolish the binding of an E3 ubiquitin ligase (β-TrCP1) to SETBP1, which prevents ubiquitination and subsequent degradation, leading to gain of function through the consequent increase in SETBP1 protein amounts10, 28. Although the precise leukemogenic mechanisms of SETBP1 mutations are still unclear, we have shown that mutant SETBP1 alleles confer self-renewal capability to myeloid progenitors in vitro, and SETBP1 mutations in adult leukemia were associated with increases in HOXA9 and HOXA10 expression28. Recurrent JAK3 mutations in JMML are also noteworthy. The JAK-STAT pathway is a key component of normal hematopoiesis29. As in other hematopoietic malignancies20, the p.Arg657Gln alteration represents the most frequent change in JMML. This alteration confers interleukin (IL)-3 independence to Ba/F3 cells and induces STAT5 phosphorylation20. Targeting the JAK-STAT pathway with a pan-JAK inhibitor such as CP-690550 (ref. 30) could be a promising therapeutic possibility for patients with JAK3-mutated JMML.
In conclusion, our whole-exome sequencing analysis identified the spectrum of gene mutations in JMML. Together with the high frequency of RAS pathway mutations, the paucity of non–RAS pathway mutations is a prominent feature of JMML. Mutations of SETBP1 and JAK3 were common recurrent secondary events presumed to be involved in tumor progression and were associated with poor clinical outcomes. Our findings provide an important clue to understanding the pathogenesis of JMML that may help in the development of novel diagnostics and therapeutics for this leukemia.
We studied 92 children (61 boys and 31 girls) with JMML, including 7 individuals with NS/MPD, who were diagnosed as having JMML in institutions throughout Japan. Written informed consent was obtained from subjects' parents before sample collection. This study was approved by the ethics committees of the Nagoya University Graduate School of Medicine and the University of Tokyo in accordance with the Declaration of Helsinki. Diagnosis with JMML was made on the basis of internationally accepted criteria1. Characteristics of the 92 JMML cases are summarized in Table 2. The median age at diagnosis was 16 months (range of 1–160 months). Karyotypic abnormalities were detected in 16 subjects, including in 8 with monosomy 7. Fifty-six of the 92 subjects (61%) received allogeneic HSCT.
Genomic DNA was extracted using the QIAamp DNA Blood Mini kit and the QIAamp DNA Investigator kit (Qiagen) according to the manufacturer's instructions. The T Cell Activation/Expansion kit, human (Miltenyi Biotec) was used for the expansion of CD3+ T cells from subjects' peripheral blood or bone marrow mononuclear cells3.
Exome capture from paired tumor-reference DNA was performed using SureSelect Human All Exon V3 (Agilent Technologies), covering 50 Mb of coding exons, according to the manufacturer's protocol. Enriched exome fragments were subjected to massively parallel sequencing using the HiSeq 2000 platform (Illumina). Candidate somatic mutations were detected through our in-house pipeline (Genomon) as previously described17.
Detection of mutations from whole-exome sequencing data.
Detection of candidate somatic mutations was performed according to previously described algorithms with minor modifications17. Briefly, the number of reads containing single-nucleotide variations (SNVs) and indels in both tumor and reference samples was determined using SAMtools31, and the null hypothesis of equal allele frequencies in tumor and reference samples was tested using the two-tailed Fisher's exact test. A variant was adopted as a candidate somatic mutation if it had P < 0.01, if it was observed in bidirectional reads (in both plus and minus strands of the reference sequence) and if its allele frequency was less than 0.25 in the corresponding reference sample. For the detection of germline mutations in RAS pathway genes, SNVs and indels having allele frequencies of more than 0.25 (SNVs) and 0.10 (indels) were interrogated for 46 genes, which consisted of known JMML-related RAS pathway genes and genes registered in the pathway databases ('Ras signaling pathway' in BioCarta and 'signaling to RAS' in Reactome32). For variant calls in tumor samples for which the paired normal reference was not available, candidate variants in the RAS pathway were detected at an allele frequency of >0.10. Finally, the list of candidate somatic and/or germline mutations was generated by excluding synonymous SNVs and other variants registered in either dbSNP131 or an in-house SNP database constructed from 180 individual samples. All candidates were validated by Sanger sequencing as previously described.
Estimation of tumor content.
The tumor content of bone marrow specimens was estimated from the allele frequency of the somatic mutations identified by deep sequencing. For homozygous mutations, as indicated by an allele frequency of >0.75, the tumor content (Ftumor) was calculated from the observed frequency (Fobserved) of the mutation according to the following equation: Ftumor = 2 × Fobserved – 1. For heterozygous mutations, the tumor content was calculated by doubling the allele frequency.
Power analysis of whole-exome sequencing.
The power of detecting somatic mutations at each nucleotide position in whole-exome sequencing was estimated by Monte-Carlo simulation (n = 1,000) on the basis of the observed mean depth of coverage for each exon in germline and tumor samples and the observed tumor content for each sample, which were estimated using the allele frequencies of the observed mutations. For the samples with no observed somatic mutations, the average tumor content of the informative samples was employed. Simulations were performed across a total of 192,424 exons.
Copy number analysis in whole-exome sequencing data.
To detect copy number lesions at a single-exon level, the mean coverage of each exon normalized by the mean depth of coverage of the entire sample was compared with that of 12 unrelated normal DNA samples. Exons showing normalized coverage greater than 3 s.d. from the mean coverage of the reference samples were called as candidates for copy number alterations. All candidate exons of RAS pathway genes were visually inspected using the Integrative Genomics Viewer33 and were validated by Sanger sequencing of corresponding putative breakpoint-containing fragments.
Targeted deep sequencing.
Deep sequencing of the targeted genes was performed essentially as described in the 'deep sequencing of pooled target exons' section in ref. 17, except that target DNA was not pooled. Briefly, all exons of PTPN11, NF1, KRAS, NRAS, CBL, SETBP1, JAK3 and SH3BP1 were PCR amplified with Quick Taq HS DyeMix (TOYOBO) and the PrimeSTAR GXL DNA Polymerase kit (Takara Bio) using primers including the NotI restriction site (Supplementary Table 3). The PCR products from an individual sample were combined and purified with the QIAquick PCR Purification kit (Qiagen) for subsequent digestion with NotI (Fermentas). Digested PCR product was purified, concatenated with T4 DNA ligase (Takara Bio) and sonicated to generate fragments with an average size of 150 bp using Covaris. Fragments were processed for sequencing according to a modified Illumina paired-end library protocol, and sequences were read by a HiSeq 2000 instrument using a 100-bp paired-end read protocol.
Variant calls in targeted deep sequencing.
Data processing and variant calling were performed with modifications to the protocol described in a previous publication17. Each read was aligned to the set of targeted sequences from PCR amplification, with BLAT34 instead of Burrows-Wheeler Aligner (BWA)35 used with the -fine option. Mapping information in the .psl format was converted to the .sam format with paired-read information. Of the successfully mapped reads, reads were excluded from further analysis if they mapped to multiple sites, mapped with more than four mismatched bases or had more than ten soft-clipped bases. Next, the Estimation_CRME script was run to eliminate strand-specific errors and exclude PCR-derived errors. A strand-specific mismatch ratio was calculated for each nucleotide variant for both strands using the bases from read cycles 11 to 50 on the next-generation sequencer. By excluding the top five cycles showing the highest mismatch rates, strand-specific mismatch rates were recalculated, and the smaller value between both strands was adopted as a nominal mismatch ratio for that variant. After excluding variants found in dbSNP131 or the in-house SNP database, non-silent variants having a mismatch ratio of greater than 0.05 were called as candidates, unless they had median values of the mismatch ratio at the relevant nucleotide positions in the 92 samples of greater than 0.01, as such variants were likely to be caused by systematic PCR problems. Finally, candidates with mismatch ratios of >0.15 were further validated by Sanger sequencing.
Annotation of the detected mutations.
Detected mutations were annotated using ANNOVAR36. The positions of the mutations were based on the following RefSeq transcript sequences: NM_002834.3 for PTPN11, NM_000267.3 for NF1, NM_002524.4 for NRAS, NM_004985.3 for KRAS, NM_005188.3 for CBL, NM_015559.2 for SETBP1 and NM_000215.3 for JAK3. The effect of the mutations on protein function was assessed by SIFT37, PolyPhen-2 (ref. 38) and MutationTaster39.
Paired tumor-reference DNA samples were sequenced with the HiSeq 2000 platform according to the manufacturer's instructions to obtain 30× read coverage for reference samples and 40× coverage for tumor samples. Obtained FASTQ sequences were aligned to the human reference genome (hg19) using BWA35 0.5.8 with default parameters. Alignment of pairs of sequences, at least one of which was not mapped or was considered to have possible mapping problems (with mapping quality of less than 40, insertions or deletions, soft-clipped sequence of more than 10% of the length of the original sequence, irregular paired-read orientation or mate distance of greater than 2,000 bp), was attempted with BLAT34 using default parameters, except for stepSize = 5 and repMatch = 2,253. Mapping statistics were calculated by counting the bases at each genomic position with SAMtools31. For variant calling, variant and reference bases with base quality of >30 were counted in both germline and tumor samples, and the Fisher's exact test was applied. Variants with P of <0.01 were called. Variants having allele frequency of >0.25 in the germline sample were excluded. Variants found in 12 unrelated germline samples with an allele frequency of >0.01 on average were also excluded owing to the high probability that they represented false positive calls. Copy number estimation was performed by calculating the averaged ratio of read depths in germline and tumor samples in 10,000-base bins. An allele-specific copy number plot was generated by measuring the allele frequency of the tumor sample at the positions in which more than 25% of the allele mismatch was observed in germline samples. For the detection of chromosomal structural variations, soft-clipped sequences that could be mapped to a unique genomic position were selected. Structural variation candidates that had more than four supporting read pairs in total and at least one read pair from each side of the breakpoint were called. Contig sequences were generated by assembling the reads within 200 bp of the breakpoint with CAP3 (ref. 40), and structural variations having the contig sequence that could be aligned to the alternate assembly of the hg19 genome with more than 93% identity were excluded as false positives. Structural variations with read depth of greater than 150 on at least one side of the breakpoint were considered to be mapped to a repeat element and were also excluded. For detection of viruses, unmapped sequences were aligned to the collection of all viral genomes in the RefSeq database using BLAT. A virus was considered to be detected if its genome was covered by mean read coverage of >1.
Total RNA was extracted using the RNeasy Mini kit (Qiagen) and was reverse transcribed with the ThermoScript RT-PCR system (Life Technologies). Target sequences were PCR amplified with the PrimeSTAR GXL DNA Polymerase kit using the primers listed in Supplementary Table 3 and were sequenced.
For comparison of the frequency of mutations or other clinical features between disease groups, categorical variables were analyzed using the Fisher's exact test, and continuous variables were tested using the Mann-Whitney U test. Overall survival and transplantation-free survival were estimated by the Kaplan-Meier method. Hazard ratios for survival with 95% CIs were estimated according to the Cox proportional hazards model, and difference in survival was tested by log-rank test. STATA version 12.0 (StataCorp) was used for all statistical calculations.
We deposited whole-genome and whole-exome sequence data in the European Genome-phenome Archive under accession EGAS00001000521.
- Differentiating juvenile myelomonocytic leukemia from infectious disease. Blood 91, 365–367 (1998). et al.
- Mutations in CBL occur frequently in juvenile myelomonocytic leukemia. Blood 114, 1859–1863 (2009). et al.
- Mutations of an E3 ubiquitin ligase c-Cbl but not TET2 mutations are pathogenic in juvenile myelomonocytic leukemia. Blood 115, 1969–1975 (2010). et al.
- Genetic typing of CBL, ASXL1, RUNX1, TET2 and JAK2 in juvenile myelomonocytic leukaemia reveals a genetic profile distinct from chronic myelomonocytic leukaemia. Br. J. Haematol. 151, 460–468 (2010). et al.
- Exome sequencing identifies MLL2 mutations as a cause of Kabuki syndrome. Nat. Genet. 42, 790–793 (2010). et al.
- Identification and characterization of SEB, a novel protein that binds to the acute undifferentiated leukemia–associated protein SET. Eur. J. Biochem. 268, 1340–1351 (2001). et al.
- SETBP1 mutations in 658 patients with myelodysplastic syndromes, chronic myelomonocytic leukemia and secondary acute myeloid leukemias. Leukemia 27, 401–403 (2013). et al.
- SETBP1 mutations in 415 patients with primary myelofibrosis or chronic myelomonocytic leukemia: independent prognostic impact in CMML. Leukemia published online; doi:10.1038/leu.2013.97 (5 April 2013). et al.
- SETBP1 mutations occur in 9% of MDS/MPN and in 4% of MPN cases and are strongly associated with atypical CML, monosomy 7, isochromosome i(17)(q10), ASXL1 and CBL mutations. Leukemia published online; doi:10.1038/leu.2013.133 (30 April 2013). et al.
- Recurrent SETBP1 mutations in atypical chronic myeloid leukemia. Nat. Genet. 45, 18–24 (2013). et al.
- SETBP1 mutation analysis in 944 patients with MDS and AML. Leukemia published online; doi:10.1038/leu.2013.145 (7 May 2013). et al.
- Fusion of NUP98 and the SET binding protein 1 (SETBP1) gene in a paediatric acute T cell lymphoblastic leukaemia with t(11;18)(p15;q12). Br. J. Haematol. 136, 294–296 (2007). et al.
- SETBP1 overexpression is a novel leukemogenic mechanism that predicts adverse outcome in elderly patients with acute myeloid leukemia. Blood 115, 615–625 (2010). et al.
- Evi-1 is a critical regulator for hematopoietic stem cells and transformed leukemic cells. Cell Stem Cell 3, 207–220 (2008). et al.
- Correction of X-linked chronic granulomatous disease by gene therapy, augmented by insertional activation of MDS1-EVI1, PRDM16 or SETBP1. Nat. Med. 12, 401–409 (2006). et al.
- De novo mutations of SETBP1 cause Schinzel-Giedion syndrome. Nat. Genet. 42, 483–485 (2010). et al.
- Frequent pathway mutations of splicing machinery in myelodysplasia. Nature 478, 64–69 (2011). et al.
- Genome-wide single-nucleotide polymorphism analysis in juvenile myelomonocytic leukemia identifies uniparental disomy surrounding the NF1 locus in cases associated with neurofibromatosis but not in cases with mutant RAS or PTPN11. Oncogene 26, 5816–5821 (2007). et al.
- Activating alleles of JAK3 in acute megakaryoblastic leukemia. Cancer Cell 10, 65–75 (2006). et al.
- Functional analysis of JAK3 mutations in transient myeloproliferative disorder and acute megakaryoblastic leukaemia accompanying Down syndrome. Br. J. Haematol. 141, 681–688 (2008). et al.
- Loss-of-function JAK3 mutations in TMD and AMKL of Down syndrome. Br. J. Haematol. 137, 337–341 (2007). et al.
- Analysis of JAK3, JAK2, and C-MPL mutations in transient myeloproliferative disorder and myeloid leukemia of Down syndrome blasts in children with Down syndrome. Blood 110, 1077–1079 (2007). et al.
- JAK3 mutations occur in acute megakaryoblastic leukemia both in Down syndrome children and non–Down syndrome adults. Leukemia 21, 574–576 (2007). , , &
- FERM domain mutations induce gain of function in JAK3 in adult T-cell leukemia/lymphoma. Blood 118, 3911–3921 (2011). et al.
- The genetic basis of early T-cell precursor acute lymphoblastic leukaemia. Nature 481, 157–163 (2012). et al.
- Janus kinase 3–activating mutations identified in natural killer/T-cell Lymphoma. Cancer Discov. 2, 591–597 (2012). et al.
- A novel retinoblastoma therapy from genomic and epigenetic analyses. Nature 481, 329–334 (2012). et al.
- Somatic SETBP1 mutations in myeloid malignancies. Nat. Genet. published online; doi:10.1038/ng.2696 (7 July 2013). et al.
- Drosophila haematopoiesis. Cell. Microbiol. 9, 1117–1126 (2007). &
- Prevention of organ allograft rejection by a specific Janus kinase 3 inhibitor. Science 302, 875–878 (2003). et al.
- The Sequence Alignment/Map format and SAMtools. Bioinformatics 25, 2078–2079 (2009). et al.
- Reactome knowledgebase of human biological pathways and processes. Nucleic Acids Res. 37, D619–D622 (2009). et al.
- Integrative Genomics Viewer (IGV): high-performance genomics data visualization and exploration. Brief. Bioinform. 14, 178–192 (2013). , &
- BLAT—the BLAST-like alignment tool. Genome Res. 12, 656–664 (2002).
- Fast and accurate short read alignment with Burrows-Wheeler transform. Bioinformatics 25, 1754–1760 (2009). &
- ANNOVAR: functional annotation of genetic variants from high-throughput sequencing data. Nucleic Acids Res. 38, e164 (2010). , &
- Predicting the effects of coding non-synonymous variants on protein function using the SIFT algorithm. Nat. Protoc. 4, 1073–1081 (2009). , &
- A method and server for predicting damaging missense mutations. Nat. Methods 7, 248–249 (2010). et al.
- MutationTaster evaluates disease-causing potential of sequence alterations. Nat. Methods 7, 575–576 (2010). , , &
- CAP3: A DNA sequence assembly program. Genome Res. 9, 868–877 (1999). &
We thank the subjects and their parents for participating in this study. This work was supported by the Research on Measures for Intractable Diseases Project from the Ministry of Health, Labor and Welfare, by Grants-in-Aid from the Ministry of Health, Labor and Welfare of Japan and KAKENHI (23249052, 22134006 and 21790907), by the Project for the Development of Innovative Research on Cancer Therapeutics (P-DIRECT) and by the Japan Society for the Promotion of Science through the Funding Program for World-Leading Innovative R&D on Science and Technology.
- Supplementary Text and Figures (1.9 MB)
Supplementary Figures 1–11 and Supplementary Tables 1–3