The Origin and Evolution of Mutations in Acute Myeloid Leukemia John S. Welch, Timothy J. Ley, Daniel C. Link, Christopher A. Miller, David E. Larson, Daniel C. Koboldt, Lukas D. Wartman, Tamara L. Lamprecht, Fulu Liu, Jun Xia, Cyriac Kandoth, Robert S. Fulton, Michael D. McLellan, David J. Dooling, John W. Wallis, Ken Chen, Christopher C. Harris, Heather K. Schmidt, Joelle M. Kalicki-Veizer, Charles Lu, Qunyuan Zhang, Ling Lin, Michelle D. O’Laughlin, Joshua F. McMichael, Kim D. Delehaunty, Lucinda A. Fulton, Vincent J. Magrini, Sean D. McGrath, Ryan T. Demeter, Tammi L. Vickery, Jasreet Hundal, Lisa L. Cook, Gary W. Swift, Jerry P. Reed, Patricia A. Alldredge, Todd N. Wylie, Jason R. Walker, Mark A. Watson, Sharon E. Heath, William D. Shannon, Nobish Varghese, Rakesh Nagarajan, Jacqueline E. Payton, Jack D. Baty, Shashikant Kulkarni, Jeffery M. Klco, Michael H. Tomasson, Peter Westervelt, Matthew J. Walter, Timothy A. Graubert, John F. DiPersio, Li Ding, Elaine R. Mardis, Richard K. Wilson  Cell  Volume 150, Issue 2, Pages 264-278 (July 2012) DOI: 10.1016/j.cell.2012.06.023 Copyright © 2012 Elsevier Inc. Terms and Conditions

Cell 2012 150, 264-278DOI: (10.1016/j.cell.2012.06.023) Copyright © 2012 Elsevier Inc. Terms and Conditions

Figure 1 Coverage and Number of Validated SNVs by Tier per Genome (A) Gigabase pairs of sequence obtained for each genome assessed by whole-genome sequencing. (B) Mean number of reads obtained for each variant assessed per genome during validation. (C and D) Number of tier 1, 2, 3, and total variants validated in each sample. (E) Total number of validated SNVs by tier in M1 AML (red) versus M3 AML (blue) cases. (F) Total number of validated SNVs per genome in nonredundant regions (tier 1, 2, and 3) versus age of patient; M1 AML (red), M3 AML (blue). The R2 for the combined 24 cases is 0.5, p < 0.0001. See also Figures S1 and S2 and Tables S1–S4. Cell 2012 150, 264-278DOI: (10.1016/j.cell.2012.06.023) Copyright © 2012 Elsevier Inc. Terms and Conditions

Figure 2 Genomic Distribution of Somatic Variants (A) Percentage of validated SNVs in each tier versus the number of genomic megabases occupied by that tier; M1 (red), M3 (blue). Average R2 M1: 0.999 (range: 0.998–1.000); M3: 0.999 (range: 0.998–1.000). (B) Distribution of validated SNVs throughout the genome. Each point represents the genomic position of an SNV: x axis represents chromosomes in numerical order; y axis represents base pair position on each chromosome in megabase pairs; M1 AML (red), M3 AML (blue). The large discontinuous regions represent regions of the reference genome without defined sequence. (C) Average (circle), standard deviation (bars), and maximal (triangles) number of SNVs per megabase by chromosome in all 24 cases. (D) Frequency of SNVs per megabase of non-tier 4 genomic space compared with the expected Poisson distribution if SNVs were randomly distributed across non-tier 4 space. (E) Mutational spectrum of validated SNVs. Percentage of validated SNVs that occur in nonredundant regions of the genome (tiers 1, 2, and 3) with indicated nucleotide changes in M1 AML (red) and M3 AML (blue). (F) Mutational spectrum as defined by tier of each event (tier 1, genic; tier 2, highly conserved and/or with regulatory potential; tier 3, unique in the genome, and not in tier 1 or tier 2, see Mardis et.al. 2009 for details). Cell 2012 150, 264-278DOI: (10.1016/j.cell.2012.06.023) Copyright © 2012 Elsevier Inc. Terms and Conditions

Figure 3 Clonality Assessment in NK M1 AML and t(15;17)-Positive M3 AML (A–D) Clusters of variants that occur with similar frequencies identify AML founding clones and subclones. Upper panels: probability density function using kernel density estimation of data in lower panels. The graphed line shows the relative probability of each variant allele frequency in the total sample. Lower panels: for each SNV, the frequency of the variant allele within the bone marrow sample is plotted versus the total number of sequencing reads covering the corresponding nucleotide position. Triangles indicate SNVs on the X chromosome. (E) The standard deviation of the variant frequency for each cluster in all 24 cases: M1 AML (red); M3 AML (blue). F. Number of clusters identified by kernel density estimation in M1 AML (red) and M3 AML (blue). See also Figures S3, S4, and S5. Cell 2012 150, 264-278DOI: (10.1016/j.cell.2012.06.023) Copyright © 2012 Elsevier Inc. Terms and Conditions

Figure 4 Background Mutations in Normal HSPCs Accumulate as a Function of Age (A) Number of validated SNVs per exome identified in each of 3 clones that were derived from individual hematopoietic stem/progenitor cells from 7 anonymous, healthy donors with normal blood counts. (B) Clustering of mutations by clone. The y axis represents the analyzed clones from each donor. The x axis represents the genes with mutations in each clone (red). Note that the two clones with ZFHX4 variants contained unique mutations (Q1660K versus P3497A). (C) Mutational spectrum in normal donor exomes (81 validated variants) versus 24 tier 1 AML genomes (343 validated variants). See also Figure S6 and Table S6. Cell 2012 150, 264-278DOI: (10.1016/j.cell.2012.06.023) Copyright © 2012 Elsevier Inc. Terms and Conditions

Figure 5 Recurrently Mutated Genes in M1 and M3 AML (A) 108 total cases of M1 and M3 AML were screened for somatic variants in the 284 genes identified by WGS and in 53 genes previously reported to be mutated in AML. Mutations with translational consequences that occur in at least three unique genomes are represented in red. FAB subtype indicated above graph. Overall survival is indicated by red (≤12 months), pink (12–24 months), and white (>24 months) bars above graph. (B) The total number of validated genic variants per genome identified by WGS in 12 NK M1 AML cases (red) and 12 M3 AML cases (blue), including PML-RARA. (C) The total number of recurrently mutated genes per genome identified in 12 NK M1 AML cases (red) and 12 M3 AML cases (blue), including PML-RARA. Only mutations with translational consequences were assessed. (D) Total number of recurrently mutated genes per genome identified in 65 M1 AML cases (red) and 43 M3 AML cases (blue), including PML-RARA. Only mutations with translational consequences were assessed. (E) Somatic variants in cohesin complex genes (SMC3, SMC1A, STAG2, and RAD21) identified in 199 cases of AML (red). Additional mutations with translational consequences in highly recurrent genes are also shown for each genome (red). See also Figures S7 and S8 and Table S7. Cell 2012 150, 264-278DOI: (10.1016/j.cell.2012.06.023) Copyright © 2012 Elsevier Inc. Terms and Conditions

Figure 6 Pathway Analysis of Key Mutations in the 24 Sequenced Genomes Mutations identified in Table 1 were individually curated in Pubmed for biological function. Functional categories represented by 2 or fewer mutated genes were excluded. Mutations in NPM1, DNA methylation, and cohesin genes occurred only in the M1 cases, suggesting that they may substitute for PML-RARA as initiating events, or do not productively cooperate with PML-RARA. Mutations in ubiquitin pathway genes occurred more often in M3 AML. Genes representing other functional categories were mutated in both M1 and M3 AML cases. Cell 2012 150, 264-278DOI: (10.1016/j.cell.2012.06.023) Copyright © 2012 Elsevier Inc. Terms and Conditions

Figure 7 Integrated Model for the Origin of Driver and Passenger Mutations during AML Evolution Hematopoietic stem/progenitor cells (HSPCs, shown in green) accumulate random, benign background mutations as a function of age. “X” represents these background mutations in the HSPCs, and may range from ∼100 to 1,000 events, depending on age. The initiating mutations are different for M1 AML cases and M3 AML cases. Because the initiating event provides an advantage for the affected cell, clonal expansion ensues, so that all of the preexisting mutations in that cell are “captured” by cloning. Each cooperating mutation gives the expanding clone an additional advantage; our data suggest that 1–5 events contribute to progression in most cases of AML. Each cooperating mutation is expected to capture all mutations that occurred between the initiating event and the progression event (designated as “Y” in the yellow cell). Although this number is unknown, the analysis of clonal progression of secondary AML (Walter et al., 2012) suggests that each cooperating mutation may capture dozens to hundreds of mutations. The “founding” AML clone is designated in red. Subclones arise from the founding AML clone by acquiring a small number of additional mutations that confer an advantage to that cell, along with any additional background mutations that may have occurred in the interim (represented as “Z”). In this study, the average number of mutations captured in subclones was 40 (range 6–110). Cell 2012 150, 264-278DOI: (10.1016/j.cell.2012.06.023) Copyright © 2012 Elsevier Inc. Terms and Conditions

Figure S1 Clinical and Molecular Characterization of 24 AML Patients, Related to Figure 1 (A) Overall survival by FAB of the 24 AML patients characterized with whole-genome sequencing. (B) Affymetrix U133 Plus 2 array profiling of the same 24 cases. (C) Structural changes assessed by Illumina SNP array in the same 24 cases. (D) Overall survival by FAB of the 108 AML patients characterized during the extension analysis. Cell 2012 150, 264-278DOI: (10.1016/j.cell.2012.06.023) Copyright © 2012 Elsevier Inc. Terms and Conditions

Figure S2 PML-RARA Breakpoints, Related to Figure 1 (A–C) Breakpoint positions within PML intron 3 (A), PML intron 5 (B), and RARA intron 2 (C). Blue triangles indicate breakpoint positions reported in this paper. Other triangles indicate previously reported breakpoint positions (orange: Hasan et al., 2008; green: Hasan et al., 2010; Mistry et al., 2005; purple: Reiter et al., 2003). (D and E) Genomic PML-RARA rearrangements associated with a large deletion in PML (D). AML10 and (E) AML15. Cell 2012 150, 264-278DOI: (10.1016/j.cell.2012.06.023) Copyright © 2012 Elsevier Inc. Terms and Conditions

Figure S3 Contamination of Leukemic Variants in Skin Control, Related to Figure 3 (A) White blood cell count (WBC) in M1 (red) and M3 (blue) cases subjected to whole-genome sequencing. (B) Contamination of skin sample by leukemic variants in M1 (red) and M3 (blue) cases. (C) Relationship of WBC and skin sample contamination by leukemic variants in M1 (red) and M3 (blue) cases. Cell 2012 150, 264-278DOI: (10.1016/j.cell.2012.06.023) Copyright © 2012 Elsevier Inc. Terms and Conditions

Figure S4 Accuracy of Variant Allele Frequency by Digital Reads in AML51, Related to Figure 3 Metaphase cytogenetics and SNP arrays identified del7, +8, and LOH on chromosomes 16 and 20. (A) Variant allele frequency of all validated SNVs in AML51. The x axis represents chromosome position, starting at chromosome 1 and continuing to the X and Y chromosomes. (B) SNP array analysis of AML51 demonstrating copy number abnormalities. The x axis in A and B are proportional. (C) Total read counts for each validated variant in A. Cell 2012 150, 264-278DOI: (10.1016/j.cell.2012.06.023) Copyright © 2012 Elsevier Inc. Terms and Conditions

Figure S5 Chromosome X Analysis and Calculation of Tumor Burden, Related to Figure 3 (A) Variant allele frequency for all variants on chromosome X in 24 AML cases. (B) Percentage of total variants, which are located on chromosome X in males versus female subjects. (C) Bone marrow tumor burden assessed by flow cytometry for M1 versus M3 AML cases. (D) Bone marrow tumor burden calculated by validated allele frequency on chromosome X (male patients) and in regions of LOH (males and females, when available). (E) Comparison of bone marrow tumor burden calculated by validated variant allele frequency and by clinical cytomorphologic blast count. (F) Comparison of bone marrow tumor burden calculated by validated variant allele frequency and by clinical flow cytometry. Cell 2012 150, 264-278DOI: (10.1016/j.cell.2012.06.023) Copyright © 2012 Elsevier Inc. Terms and Conditions

Figure S6 Analysis of Cell-Specific Variants in Hematopoietic Stem/Progenitor Cells from Normal, Healthy Donors, Related to Figure 4 (A) Age-dependent clonogenic capacity of flow sorted CD34+CD38− lineage-peripheral blood cells and the number of cells used for exome sequencing for each clone. B. Exome coverage of each peripheral blood control sample, and sequenced clone. (C and D) Validation of two selected variants that create endonuclease recognition sites. To determine whether we could detect clone-specific mutations in the starting peripheral blood samples, genomic DNA samples from the peripheral blood cells from the volunteers in their 20 s (C) or 30 s (D) were used as input to PCR amplify regions. The amplicon in (C) was designed to identify a mutation that causes L124V in the PSPC1 gene, which creates an MslI site. The amplicon in (D) was designed to identify a G1220D mutation in the C12orf51 gene, which creates a BamHI site. The amplicons were ligated into pCR2.1, transformed into E. coli, and plated on 10 cm agar plates containing ampicillin, where colonies were grown overnight, counted, and pooled for DNA preparation. The numbers of colonies evaluated in each pool is indicated above each lane. The HSPC clone known to contain the test mutation was used as a positive control for each experiment (“positive clone”). Plasmid DNA was prepared from the pooled colonies and digested with EcoRI and MslI (C) or BamHI (D). Digested DNA was separated on 7% polyacrylamide gels, transferred to nitrocellulose, and analyzed by Southern blotting using a probe that consisted of the full length PCR product for the target amplicon. Expected fragment lengths are indicated to the left of the autoradiographs. The time of exposure is noted for each blot. Note that the positive HSPC clone for each blot contains strong subbands corresponding to the presence of the mutation in nearly all cells in the sample. The pool containing 411 plasmids is positive for the mutation in (C) and the pool with 177 plasmids is positive in (D). (E) The positive pools from C and D were subjected to deep digital sequencing, and readcounts were assembled for the WT and mutant alleles for each. The total frequency of the mutation in the entire pool of tested clones suggests that less than 1 starting cell in 1,000 in the peripheral blood samples contained each of the mutations detected in the clones. Cell 2012 150, 264-278DOI: (10.1016/j.cell.2012.06.023) Copyright © 2012 Elsevier Inc. Terms and Conditions

Figure S7 Pair-Wise Analysis of Co-Occurring Mutations, Related to Figure 5 (A) Pair-wise analysis of co-occurrence of mutations in FLT3, NPM1, DNMT3A, and IDH1 in 108 M1 and M3 cases. p value calculated by pair-wise hypergeometric distribution comparison. (B) Pair-wise analysis of co-occurrence of mutations in FLT3, NPM1, DNMT3A, and IDH1 in 200 AML cases. Only genes that significantly co-occur with one another are included. Cell 2012 150, 264-278DOI: (10.1016/j.cell.2012.06.023) Copyright © 2012 Elsevier Inc. Terms and Conditions

Figure S8 Co-Occurrence of DNMT3A, NPM1, IDH, and FLT3 within the Founding Clone or Subclones, Related to Figure 5 The variant allele frequency and read counts for each validated SNV in the eight cases with co-occurring DNMT3A, NPM1, IDH1, and FLT3 mutations (A-H). The variant allele frequency for each of these mutations is indicated with large circles: DNMT3A, red; NPM1, magenta; IDH1, blue; and FLT3, green. Small black circles indicate the variant allele frequency of all other validated variants for each case. Note that in AML21, chromosome 13 (including FLT3) has undergone UPD. Note also that the variant allele frequency of indels is not as accurate as SNVs. This is relevant in interpreting AML8, AML21 and AML40, which contain NPM1 and FLT3 indels with variant allele frequencies outside the major clones. Cell 2012 150, 264-278DOI: (10.1016/j.cell.2012.06.023) Copyright © 2012 Elsevier Inc. Terms and Conditions