THE GENOMIC LANDSCAPE OF HYPODIPLOID ACUTE LYMPHOBLASTIC LEUKEMIA

Subtype specific genetic alterations in hypodiploid ALL

We identified several genes and pathways recurrently targeted by sequence mutations and submicroscopic deletions that were significantly associated with hypodiploid ALL subtype (Table 1 and Supplementary Tables 7–8 and 10–13). There were no significant differences in lesion frequency or transcriptional profiles between non-masked cases and cases that were masked or harbored a clone of at least 30% doubled hypodiploid cells. These cases were grouped together for subsequent analyses (Table 1, Fig. 2b and Supplementary Tables 13–15). Alterations targeting RTK- and Ras signaling were a hallmark of near haploid ALL (70.6% of cases; Table 1). Other recurrent alterations identified in near haploid ALL were deletion of the lymphoid transcription factor IKZF3 (AIOLOS; 13.2%) and deletions of a histone cluster at chromosome 6p22 (19.1%; Table 1).

In contrast, low hypodiploid ALL was highly enriched for alterations in TP53 (91.2%), RB1 (41.2%) and the IKAROS family gene IKZF2 (HELIOS; 52.9%; Table 1). Adult low hypodiploid ALL cases also had a remarkably high frequency of TP53 mutations (90.9%), indicating that mutations of TP53 are a hallmark of low hypodiploid ALL (Supplementary Table 16).

Mutations targeting RTK- and Ras signaling in near haploid ALL

More than two-thirds (70.6%) of near haploid ALL cases harbored genetic alterations known or predicted to result in activation of RTK- or Ras signaling, including deletion, amplification and/or sequence mutation of NF1, NRAS, KRAS, MAPK1, FLT3 and PTPN11. RTK- and Ras signaling alterations were less frequent in low hypodiploid (8.8%) and near diploid ALL (31.8%; P=2.47×10⁻⁵; Table 1, Fig. 3 and Supplementary Table 12).

Thirty of 68 near haploid cases (44.1%) had focal deletions or sequence mutations of NF1 (Fig. 3a and Supplementary Table 12). Due to aneuploidy the NF1 alterations were biallelic in 76.7% of near haploid cases. NF1 encodes Neurofibromin, a tumor suppressor, Ras-GTPase Activating Protein (GAP) and negative regulator of Ras signaling¹⁶. Alterations of NF1 result in neurofibromatosis^16–19 and juvenile myelomonocytic leukemia (JMML)²⁰. NF1 alterations also occur in T-lineage ALL, acute myeloid leukemia²¹ and non-hematopoietic malignancies such as glioblastoma.²² Furthermore, neurofibromatosis is associated with an increased risk of developing leukemia.^23,24

In contrast to other tumors, in which NF1 deletions usually involve the entire gene²⁵, the focal NF1 deletions in hypodiploid ALL were always intragenic, involving exons 15 through 35 in 17 of 21 cases(Fig. 3b–c, Supplementary Table 12 and Supplementary Fig. 5). This focal deletion has not previously been described, and results in deletion of the GAP domain. Sequencing of the genomic breakpoints suggests that these deletions result from aberrant recombinase-activating gene activity (Fig. 3c and Supplementary Fig. 5b), which has been implicated in the generation of other chromosomal rearrangements and deletions in ALL.^26,27

To further characterize the transcriptional consequences of the exon 15–35 NF1 deletion, we performed RT-PCR for NF1 in 17 hypodiploid ALL cases, and mRNA-seq for NALM-16²⁸, a near haploid cell line harboring the same deletion of NF1 exons 15–35. The deletion results in splicing of exon 14 to 36, causing a frame-shift in exon 36 (Fig. 3d and Supplementary Fig. 5). Immunoblotting of NALM-16 indicated that the deletion results in loss of NF1 expression (Supplementary Fig. 6).

We identified RTK/Ras pathway alterations in matched non-tumor DNA from two near haploid cases (Supplementary Fig. 7). A PTPN11 p.Gly503Arg substitution(case SJHYPO036) was heterozygous in remission DNA, and homozygous in the tumor due to aneuploidy. This substitution has been reported previously as a somatic and inherited variant in JMML and Noonan syndrome (NS), and is located in a mutational hotspot region in hematopoietic malignancies and NS.^29,30 Exchange of this amino acid results in constitutive activation of the phosphatase.²⁹ An NRAS p.Gly12Ser substitution was identified in remission DNA from case SJHYPO020, and verified by sequencing DNA from normal T-cells purified from bone marrow in this patient (Supplementary Fig. 7). Inherited NRAS mutations are rare in hematopoietic disorders³¹ and have not previously been described in ALL. The p.Gly12Ser may have a lower transforming potential than other p.Gly12 alterations and thus may be tolerated as inherited.³² These results suggest either an inherited predisposition to leukemia, or acquisition of the RTK/Ras pathway mutations as de novo germline mutations or in a primitive hematopoietic progenitor cell in these patients.

Recurrent alterations of PAG1 in near haploid ALL

PAG1, encoding Phosphoprotein associated with glycosphingolipid microdomains 1, also known as Csk Binding Protein (CBP), was altered in 10.3% of near haploid ALL, but rarely in the other hypodiploid ALL subtypes (Supplementary Table 1). Mutations in this gene have not previously been described in leukemia. The majority of PAG1 deletions was homozygous and involved the upstream region and the first exon, leading to a complete loss of PAG1 expression (Supplementary Fig. 8).

PAG1 is mainly known as an adapter for C-terminal Src kinase in T-cells, and thereby having a role within negative regulation of Src family kinases.^33–35 PAG1 has also been identified as a putative Ras signaling inhibitor³⁶, and to have a negative regulatory function in proximal B-cell receptor signaling.³⁷

TP53 mutations in low hypodiploid ALL

A striking finding was TP53 alterations in 91.2% of pediatric low hypodiploid ALL, in comparison to less than 5% of non-low hypodiploid B-ALL^38,39(P=1.65×10⁻²⁰; Table 1, Fig. 4 and Supplementary Table 18). Thirty (88.2%) cases had a sequence mutation, the majority of which were homozygous due to aneuploidy. RT-PCR identified an additional case that expressed an aberrantly spliced isoform lacking exons 2–6, including the translational start site, with no expression of the wild-type allele (data not shown). The TP53 mutations included missense, nonsense and insertion/deletion mutations and are in general predicted to be loss-of-function mutations (Fig. 4a and Supplementary Table 18).

Almost half (43.3%) of the TP53 mutations in pediatric low hypodiploid ALL were present in non-tumor hematopoietic cells (Supplementary Table 18), suggestive of an inherited origin or acquisition as de novo mutations in the germline or hematopoietic compartment (a comprehensive listing of putative deleterious sequence variants present in non-tumor cells is provided in the Supplementary Note and Supplementary Table 7). Moreover, many of the mutations have previously been reported as LFS associated mutations.⁴⁰ The most common LFS substitution p.Arg248Trp was present in three low hypodiploid cases, and prevents p53 from binding into the minor groove of its DNA binding site.⁴¹ Further, the p.Arg306* alteration present in three cases has previously been identified as both somatic and inherited.⁴⁰

In support of an inherited origin of the TP53 mutations, we studied a kindred in which a child with low hypodiploid ALL harbored a p53 p.Gly302fs alteration that was homozygous in leukemic cells, and heterozygous in a skin biopsy lacking tumor involvement (Fig. 4c–d and Supplementary Note). The proband’s father developed glioblastoma multiforme, a LFS associated malignancy, and the same homozygous mutation was identified in a tumor sample obtained at diagnosis (Fig. 4d–e).

TP53 mutations were also identified in 10 of 11 adult low hypodiploid cases (90.9%) but only in 8 of 106 (7.5%)non-low hypodiploid ALL cases (P=5.03×10⁻⁹; Fig. 4b and Supplementary Tables 2 and 16). In contrast to pediatric low hypodiploid ALL, all TP53 mutations observed in adult cases with available remission DNA were somatic. As previously described^42,43, TP53 mutations were also identified at relapse (5 of 18 adult ALL relapse cases(27.8%) and one of 5 pediatric hypodiploid ALL cases with available relapse material).

Missense TP53 mutations may result in elevated p53 levels.⁴⁴ Immunoblotting, immunohistochemistry- and flow cytometry analysis of primary hypodiploid ALL patient material and cells harvested from NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ mice xenografted with primary hypodiploid ALL cells demonstrated increased levels of p53 in cases harboring missense mutations, but not in cases lacking TP53 mutations (Fig. 4f–g, Supplementary Fig. 9 and data not shown). Expression of CDKN1A, encoding the cyclin-dependent kinase inhibitor p21, is directly regulated by p53.⁴⁵ The inactivity of one of the p53 mutants (p.Arg280Ser, xenograft SJHYPO120-X) was demonstrated by lack of increase in p21 levels upon treatment of the leukemia cells with etoposide (Supplementary Fig. 9a).

Frequent deletions of IKZF2 and IKZF3 in hypodiploid ALL

Alteration of IKZF1 is a hallmark of high-risk non-hypodiploid B-progenitor ALL^5,27,39, but alteration of other IKAROS family genes is rare. In contrast, IKZF1 alterations were infrequent in hypodiploid ALL, while IKZF2 deletion was highly recurrent in low hypodiploid ALL (18 of 34 cases, 52.9%; Fig. 5; Supplementary Table 17). IKZF2 deletions were also common in adult low hypodiploid ALL, with 4 of 11 cases (36.4%) harboring deletions, compared to 2 of 106 adult non-low hypodiploid ALL cases (1.9%; P=0.0005; Supplementary Table 19 and Supplementary Fig. 10).

Alterations of IKZF3 were common in near haploid ALL, with deletions and one frame-shift mutation in 9 of 68 cases (13.2%), but rare in non-near haploid ALL (P=0.061; Fig. 5c, Supplementary Table 17). Alterations of IKZF2 and IKZF3 were bi-allelic due to aneuploidy(Supplementary Table 12).

The IKAROS gene family encodes zinc finger transcription factors involved in lymphoid development and differentiation.⁴⁶ The founding member IKZF1 is essential for lymphopoiesis, and Ikzf1 inactivation leads to complete lack of B-lymphoid cells in mice.^47,48 Aiolos has a key role in B-cell differentiation and associates with Ikaros in an Ikaros-NuRD complex⁴⁹, while Helios has a role in regulatory T-cell development.^50–52 During B cell development, Ikzf2 expression is highest in common lymphoid progenitors and pre-pro-B cells, whereas Ikzf3 expression is highest in more mature lymphoid precursors (Supplementary Fig. 11 and refs ^53,54), suggesting that the alterations of Ikzf2 and Ikzf3 in low hypodiploid and near haploid ALL, respectively, occur in cells at different stages of lymphoid maturation. Accordingly, the frequency of antigen receptor rearrangements and expression of CD19 are lower in low hypodiploid ALL than non-low hypodiploid B-ALL (Supplementary Fig. 12a–c). Further, low hypodiploid cases have a similar transcriptional profile as immature murine lymphoid progenitors (Hardy stage B; Supplementary Fig. 12d). Together, these observations suggest that low hypodiploid ALL arises from transformation of a less mature lymphoid progenitor than near haploid ALL.

RB1 alterations in low hypodiploid ALL

Alterations of RB1, encoding Retinoblastoma 1, were present in 14 of 34 (41.2%) low hypodiploid cases compared to 6 of 68 (8.8%) near haploid cases and were absent in near diploid ALL (P=1.19×10⁻⁵; Table 1 and Supplementary Fig. 13a–b). Thirteen (92.9%)RB1-altered low hypodiploid cases had concomitant TP53 mutations, while only one (7.1%) had co-occurrence of RB1 and CDKN2A/CDKN2B alterations. A total of 61.8% of low hypodiploid ALL harbored either RB1 or CDKN2A/CDKN2B alteration, suggesting that these are complementary alterations disrupting the RB1 tumor suppressor pathway (Fig. 6a and Supplementary Fig. 14).

Similarly, 2 of 11 (18.9%) adult low hypodiploid ALL cases had RB1 deletions, compared to 10 of 106 (9.4%) non-low hypodiploid ALL cases (Supplementary Table 19 and Supplementary Fig. 13c).

High frequency of alterations targeting histones and histone modifiers

Focal deletions of a histone gene cluster at 6p22 were common in near haploid ALL (19.1%; Supplementary Fig. 15a). The minimal region of deletion involved HIST1H2BE, HIST1H4D, HIST1H3D, HIST1H2AD and HIST1H2BF. These deletions have been previously observed in non-hypodiploid ALL, albeit at a lower frequency (8.1% of B-ALL cases⁵).

Whole genome- and exome sequencing identified a range of mutations in genes encoding proteins involved in histone modifications (Fig. 6b, Supplementary Note and Supplementary Table 20). Twenty-six histone modifier gene mutations were identified in 16 near haploid cases (64% of the 25 next-generation sequenced near haploid cases), and 9 mutations in 9 low hypodiploid ALL samples (60% of 15 cases). The most common target of alteration was the transcriptional co-activator and histone acetyltransferase CREBBP⁵⁵, with 8 of 25 near haploid cases (32%) harboring a deletion or sequence mutation (Supplementary Fig. 15b). Mutations in CREBBP are uncommon at diagnosis in ALL but enriched at relapse⁶, and this is the first ALL subtype found to have a high frequency of CREBBP mutations at diagnosis. Two of the CREBBP substitutions(p.Arg1446Cys and p.Gln1500Pro) occur at residues in the histone acetyltransferase domain, important for the histone acetylase activity of CREBBP.^6,39 Further, a CREBBPp.Lys389_Met395>Lys alteration overlaps with a deletion in murine Crebbp that enhances Myc-driven lymphomagenesis.⁵⁶ Three missense mutations with predicted deleterious effect were identified in EZH2, the product of which catalyzes trimethylation of histone 3 lysine 27.⁵⁷ Two of these mutations appeared in the same relapse sample and were not present at diagnosis (Supplementary Tables 7 and 20).

Gene expression profiling distinguishes subgroups of hypodiploid ALL

The observation of distinct genomic alterations in near haploid and low hypodiploid ALL was recapitulated by gene expression profiling (Table 1 and Supplementary Table 1). Near haploid, low hypodiploid and near diploid cases formed discrete clusters in unsupervised hierarchical clustering and principal component analysis (PCA; Fig. 2b). Further, differential expression analysis between near haploid and low hypodiploid ALL resulted in more than 15,000 differentially expressed probesets (Supplementary Table 21). Similar PCA clustering was observed when restricting the analyses to probes on chromosomes showing identical patterns of aneuploidy between near haploid and low hypodiploid ALL (Supplementary Fig. 16). This indicates that aneuploidy is not responsible for the major differences observed, but that there is a global difference in expression profiles between the hypodiploid ALL subgroups. Furthermore, more than 600 gene sets showed subtype-specific enrichment on gene set enrichment analysis (Supplementary Table 22).

Activation of signaling pathways in hypodiploid ALL

The finding of a high frequency of Ras- and RTK signaling lesions suggested that activation of these pathways is important in the pathogenesis of near haploid ALL. We examined activation of Ras- and related signaling pathways in 8 near haploid and 6 low hypodiploid tumor xenografts. Flow cytometric analysis of pERK, pmTOR and pS6 and assessment of Ras-GTP levels demonstrated activated Ras- and PI3K signaling in tumors with canonical Ras pathway mutations, but also in low hypodiploid tumors lacking mutations in genes known to regulate Ras- and PI3K signaling (Fig. 7 and Supplementary Fig. 17). These findings suggest that alternative mutational mechanisms activate these pathways in low hypodiploid cases lacking Ras/PI3K pathway mutations. Deletions of IKZF2 in low hypodiploid ALL were exclusively observed in cases lacking Ras pathway mutations(Fig. 6a). To investigate the potential role of IKZF2 inactivation in regulation of Ras signaling, we performed shRNA-mediated knockdown of Ikzf2 and Ikzf3 in murine Ba/F3 and Arf^−/− pre-B cell lines (Supplementary Fig. 18), which resulted in an increase of both pERK and pS6, suggesting that HELIOS and AIOLOS, in addition to their roles within lymphoid development, may also modulate Ras- and PI3K signaling (Supplementary Fig. 19).

We next examined the potential for targeting Ras-, PI3K- and mTOR signaling in hypodiploid ALL. The NALM-16 near haploid cell line and a panel of near haploid and low hypodiploid xenograft cells were treated with the MEK inhibitors PD0325901 and MEK162, the PI3K inhibitor GDC-0941 and the dual PI3K/mTOR inhibitor BEZ235. Notably, MEK inhibition did not affect proliferation of the samples tested, but the PI3K- and dual PI3K/mTOR inhibitors substantially inhibited proliferation of all tumors examined, and efficiently attenuated intrinsic PI3K signaling in NALM-16 (Fig. 7d–f and Supplementary Table 23) suggesting that inhibition of the PI3K pathway may serve as a novel treatment alternative in both near haploid and low hypodiploid ALL.

Patients and samples

One hundred and twenty-four pediatric patients with hypodiploid ALL treated on St Jude Children’s Research Hospital (SJCRH; N=29) and Children’s Oncology Group (COG; N=95) protocols were studied. Relapse material was available for 5 cases, and 89 cases had a matched remission sample (Supplementary Table 1). This cohort was compared with an adult ALL cohort of 117 cases, including 11 low hypodiploid ALL cases (Supplementary Table 2).

Informed consent was obtained in accordance with the Declaration of Helsinki. Study approval was obtained from the SJCRH and COG institutional review boards (IRB), and the IRBs of the Royal Adelaide Hospital.

Mononuclear cells were purified from bone marrow or peripheral blood by density gradient centrifugation, and cryopreserved in liquid nitrogen. Diagnostic samples were characterized as hypodiploid ALL samples by karyotyping and DNA index measurement.

DNA was directly extracted from tumor samples with more than 70% blasts and remission samples using either phenol/chloroform or a DNA Blood Mini Kit (Qiagen, Valencia, CA), and DNA integrity was assessed by agarose gel electrophoresis. Tumor samples with less than 70% blasts and sufficient material were flow sorted prior to DNA extraction using FACSVantage SE flow cytometers and phycoerythrin (PE) labeled CD19 (BD Biosciences, San Jose, CA) and fluorescein isothiocyanate (FITC) labeled CD45 (Life Technologies, Carlsbad, CA) antibodies. Tumor samples from cases harboring TP53 mutations lacking a matched remission sample were flow sorted using CD45-FITC and allophycocyanin (APC) labeled CD7 (Life Technologies) antibodies to get access to normal T-cells to test the somatic status of the mutations. DNA was whole genome amplified using Qiagen Repli-G mini kit (Qiagen). DNA concentrations were measured using a NanoDrop ND-1000 spectrophotometer or fluorometrically by the Quant-iT BR kit (Life Technologies) using a Synergy HT (BioTek). RNA was extracted from tumor samples using TRIzol (Life Technologies).

Cell lines

Cell lines used were the near haploid cell line NALM-16 (DSMZ #ACC-680), and the murine cell line Ba/F3 and primary Arf^−/− pre-B cells (harvested from Arf^−/− C57BL/6 mouse bone marrow; e.g. Ref.⁶⁹). Ba/F3 and Arf^−/− pre-B cells were cultured in the presence of 10nM IL-3 and IL-7 (PeproTech), respectively. shRNAs targeting Ikzf2 and Ikzf3 are listed in Supplementary Table 24.

Next generation sequencing

Twenty cases underwent whole genome sequencing (WGS) of tumor and matched non-tumor DNA, and 20 cases whole exome sequencing (WES, Agilent SureSelect Human All Exon 50MB). DNA and RNA library preparations and paired-end sequencing was performed using either Illumina GAIIx or HiSeq 2000 analyzers as previously described.^15,70,71

Analysis of sequence and structural variants was performed as previously described.¹⁵ Validation was performed on whole genome amplified tumor and matched normal DNA by PCR and either Roche 454 FLX sequencing or Sanger sequencing.

For the mRNA-seq data from the cell line NALM-16, putative variants were compared to a database of variants identified in 254 non-tumor samples sequenced as part of the SJCRH – Washington University Pediatric Cancer Genome Project (PCGP⁷²), and variants present in dbSNP⁷³ (excluding known cancer variants in OMIM and local databases). The remaining variants were also compared with exome variants identified by NHLBI Exome Sequencing Project (Exome Variant Server, NHLBI Exome Sequencing Project, Seattle, WA (http://evs.gs.washington.edu/EVS/)). Data are presented in Supplementary Table 25.

High-risk germline variant analysis

Novel, non-silent germline coding variants were identified as previously described.¹⁵ To further identify likely high-risk variants, we selected variants that fit one of the following categories: 1, match to any validated somatic mutation; 2, predicted truncating variants in known cancer genes (using a combined list of CancerGeneCensus genes, downloaded on 2011-03-22, and 120 frequently mutated genes in childhood ALL³⁹); 3, overlap with sites (either chromosome coordinates or amino acid) where confirmed somatic variants have been reported by COSMIC (Catalogue Of Somatic Mutations In Cancer). Variants initially discovered by somatic pipeline but validated as germline were also included in the search. To further filter out common polymorphisms caused by deficiency in current human assembly, we compared the selected germline variants with common germline variants identified by NHLBI Exome Sequencing Project, and an internal database of germline variants collected from PCGP. We further removed variants present in germline but absent in tumor due to aneuploidy.

Single nucleotide polymorphism microarrays

Samples were genotyped using Affymetrix SNP 6.0 microarrays according to the manufacturer’s instructions. CEL files were generated using GeneChip Command Console Software. SNP calls were generated using Genotyping Console (Affymetrix) and the Birdseed v2 algorithm with default parameters with at least 50 arrays in each analysis.

Array normalization and copy number inference were performed according to a previously published workflow.^3,27 The importance of correct normalization is clearly exemplified by the highly aneuploid hypodiploid ALL samples (Supplementary Fig. 20). Normalized data were viewed in dChip⁷⁴ and regions with abnormal copy number identified computationally by circular binary segmentation (CBS)⁷⁵ and analyzed as previously described.^3,27

Recurrent focal deletions were validated by genomic PCR and/or RT-PCR, followed by Sanger sequencing. Oligonucleotide primers for PCR amplification were designed using Primer 3 (Ref. ⁷⁶). Primers used for mapping the NF1 exon 15–35 deletion are present in Supplementary Table 26.

Targeted gene resequencing

Gene resequencing was performed for all 124 diagnosis samples, 2 relapse cases (SJHYPO056-R and SJHYPO117-R) and the hypodiploid cell line NALM-16 for the genes listed in Supplementary Table 9. Primer sequences are present in Supplementary Table 26. Sequencing was performed by PCR of whole genome amplified DNA, followed by sequencing using 3730xl instruments (Applied Biosystems) as previously described.⁶ Single nucleotide variations were detected by SNPdetector⁷⁷ and PolyScan⁷⁸, and validated by sequencing of both tumor and, where available, matched non-tumor samples. Variations identified in unpaired tumor samples were compared to online databases and to a local database of sequence variation.

Gene expression profiling

Gene expression profiling was performed using Affymetrix GeneChip HT HG-U133+ PM microarrays for human specimens (samples listed in Supplementary Table 1) and MG-430 2.0 microarrays for Hardy Fractions from murine cells (details listed in Supplementary Table 27), according to the manufacturer’s instructions. Arrays with outlier relative log expression metric were excluded from the data depository and downstream analysis. Statistical analyses, principal component analysis and unsupervised hierarchical clustering were performed using R 2.13.0 (Ref. ⁶⁰), Bioconductor 2.6 (Ref. ⁶¹) and Spotfire Decision Site 9.1.1 (Tibco). For HT HG-U133+ arrays, all samples were normalized by the RMA algorithm. Probesets not passing the background signal (twice of the average signal on the control probes with different GC content) across all samples were excluded for differential expression analysis, where limma⁶² with estimation of false discovery rate (FDR)⁶³ was performed between experimental subgroups. Probesets below a FDR cutoff of 0.05 were deemed differentially expressed. Weighted Pair-Group Method with Arithmetic mean clustering with Euclidean distance was performed using the 10,000 probesets with the highest standard deviation across samples. For MG-430 2.0 arrays, gene expression signals were scaled to a median value of 500 using the Affymetrix MAS 5.0 algorithm. Probesets with absent calls for all samples were excluded, and probeset signals were variance stabilized by adding the number 32 and subsequently log2-transformed. Top 50, 100, 250, and 500 gene signatures for each Hardy Fraction were identified using limma, and Gene Set Enrichment Analysis (GSEA)⁷⁹ were used to examine the enrichment of the Hardy Fraction genesets for each hypodiploid ALL subgroup. Gene sets with less than 10 or greater 500 genes were excluded, and significantly enriched gene sets after 1000 permutations at a FDR of <0.25 are reported.

Immunohistochemical analyses

Immunohistochemistry of human and mouse tissues were performed using routine techniques, hematoxylin and eosin staining and appropriate primary antibodies.

Generation of xenografted mice

The mouse studies were approved by the SJCRH Institutional Animal Care and Use Committee. Primary leukemic cells were transplanted by tail vein injection into sub-lethally irradiated (250 cGy) immunodeficient NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ mice (2–3 mice per tumor; The Jackson Laboratory). Engraftment was analyzed by bi-weekly retro-orbital bleeds followed by flow cytometry analysis of human CD45-FITC, CD19-PE and CD7-APC, and mouse CD45-PE-Cy7-stained peripheral blood cells. Mice were killed upon signs of disease, and tumors were characterized histologically, immunophenotypically (immunophenotypic markers as above) and by genomic profiling by Affymetrix SNP6.0 microarrays for a subset of samples (Supplementary Note and Supplementary Figs. 21–22).

Fluorescence activated cell sorting analysis

Cell lines were starved in serum free media for 5 hours prior to stimulation with PMA (50 nM; 15min) or treatment with BEZ235 or GDC-0941 (0.3uM and 0.7uM, respectively (IC₅₀ values for NALM-16); 1 hour). Cell lines and splenocytes and bone marrow cells from xenografted mice were fixed with 2% paraformaldehyde and permeabilized with 90–95% methanol prior to staining with antibodies (Supplementary Table 28), and analysis on LSR II or FACS Calibur (Becton Dickinson). Data were collected using DIVA or CellQuest software (Becton Dickinson) and analyzed using FLOWJO (Tree Star) and Cytobank.

Immunoblotting and RAS-GTP assay

Cells were lyzed and protein concentrations determined using the BCA Protein Assay Kit (Pierce). Samples were separated on Criterion polyacrylamide (BioRad) or NuPAGE (Life Technologies) gels and transferred to Immobilon-P (Millipore) or PVDF (Life Technologies) membrane. After staining with the respective antibodies (Supplementary Table 28), target proteins were visualized using ECL (Amersham Biosciences), ECL plus (GE Healthcare) or the Femto Chemiluminescent Kit (Thermo Fisher). Ras-GTP levels were measured using a Raf-Ras binding domain (RBD) bead pulldown assay that has been described previously.^80,81

Ex vivo drug studies

Cell viability and proliferation of bone marrow cells obtained from xenografted mice was determined using AO/PI staining (Nexcelom Bioscience LLC, Lawrence, MA) and WST-1 reagent (Roche, Indianapolis, IN) according to manufacturer’s instructions. Cells were treated with BEZ235, GDC-0941, MEK162, or PD0325901 at increasing concentrations as indicated in Supplementary Table 23 for 72 hours. The concentration inhibiting cell proliferation by 50% compared with control cells (IC₅₀) was determined using GraphPad Prism 5.0 (GraphPad Software, La Jolla, CA).

Outcome analysis

Event-free survival (EFS) was defined as the time from diagnosis until the date of relapse or until the last follow-up date for all event-free survivors. Death was treated as a competing risk. Associations between genetic variables and EFS were estimated by the methods of Kaplan and Meier. Standard errors were calculated by the methods of Peto et al.⁸² The Mantel–Haenszel test was used to compare EFS estimates for patients with and without lesion at each locus.⁸³ The proportional hazards model of Fine and Gray was used to adjust for age, presentation leukocyte count, aneuploidy subgroup and levels of minimal residual disease (MRD).⁸⁴ Analyses were performed using R (www.r-project.org)⁸⁵, SAS v9.1.2 (SAS Institute) and SPLUS 7.0 (Insightful Corp.). To evaluate associations between genetic alterations and MRD, and hypodiploid ALL subgroup and MRD, MRD data were converted into an ordinal variable (<0.01% (negative) and ≥0.01% (positive)) and association analyses performed using the Chi-Square test (FREQ procedure, SAS) with estimation of false discovery rate (MULTTEST, SAS). Significantly associated variables were adjusted for age, presentation leukocyte count and genetic subtype using logistic regression. Data are presented in Supplementary Note and Supplementary Tables 29–34.