brief communications
`
`sequencing small genomic
`targets with high efficiency
`and extreme accuracy
`Michael W Schmitt1–3, Edward J Fox2, Marc J Prindle2,
`Kate S Reid-Bayliss2, Lawrence D True2,
`Jerald P Radich3 & Lawrence A Loeb2
`
`the detection of minority variants in mixed samples requires
`methods for enrichment and accurate sequencing of small
`genomic intervals. We describe an efficient approach based
`on sequential rounds of hybridization with biotinylated
`oligonucleotides that enables more than 1-million-fold
`enrichment of genomic regions of interest. in conjunction
`with error-correcting double-stranded molecular tags, our
`approach enables the quantification of mutations in individual
`dna molecules.
`
`Diseases such as cancer or viral infections do not manifest as a
`single population of cells but rather as a heterogeneous mixture of
`subclonal populations1. Although massively parallel sequencing
`has made it feasible to scan whole genomes for clonal nucleotide
`variations, this approach cannot readily delineate the heteroge-
`neity of mutations within a cell population. In order to detect
`rare, subclonal mutations, sequencing must be carried out to
`depths that can be prohibitively expensive, and at low frequencies
`it becomes difficult or impossible to distinguish sequencing-
`related errors from true variation. We overcome these challenges
`by coupling extensive purification of targeted sequences with
`highly accurate DNA sequencing.
`Targeted capture approaches2 sequence large genomic regions
`(typically hundreds of kilobases to several megabases); sequenc-
`ing at great depth is impractical for targets of this size owing to
`the large amount of sequencing capacity that would be required.
`These approaches do not scale to small targets (<50 kb) and typi-
`cally result in recovery of targeted DNA sequences of 5% or less.
`Small targets can be amplified by methods such as PCR or molec-
`ular inversion probes3; however, these methods are error prone
`and generate artifactual mutations that overwhelm the detection
`of subclonal variants4.
`Detection of subclonal and random mutations in a target gene
`also requires extremely accurate sequencing. High-throughput
`sequencing has a high error rate of 0.1–1%, averaging one artifactual
`
`mutation in every sequencing read. Thus, millions of sequencing
`errors occur in every sequenced genome5,6. These errors can be
`averaged to obtain a single consensus sequence for a population
`of cells; however, owing to this high error rate it is not feasible to
`reliably detect mutations present in fewer than 5% of cells. Molecular
`tagging of ssDNA before amplification7–9 can reduce the frequency
`of erroneously called variants—but only by approximately 20-fold,
`as it cannot correct errors that occur in the first round of amplifica-
`tion and are propagated to subsequent copies10.
`To overcome these limitations, we developed an alternative
`approach based on sequential rounds of capture with individual
`biotinylated DNA oligonucleotides in conjunction with duplex
`sequencing, which uses double-stranded, complementary molec-
`ular tags to separately label and amplify each of the two strands
`of individual duplex DNA molecules10. In duplex sequencing,
`mutations are scored only if they occur at the same position on
`both DNA strands, whereas amplification and sequencing errors,
`which appear in only one strand, are not scored.
`As a demonstration, we attempted to detect rare mutations in
`the ABL1 gene that confer resistance to imatinib (Gleevec) therapy
`of chronic myeloid leukemia11. We synthesized 5′-biotinylated
`DNA oligonucleotides corresponding to exons 4–7 of ABL1
`(Supplementary Table 1). Duplex sequencing adaptors containing
`complementary molecular tag sequences that identify each of the
`two strands of individual DNA molecules were ligated to sheared
`human genomic DNA (Online Methods). The product was then
`PCR amplified and hybridized to the pooled ABL1-targeting
`oligonucleotides, and hybridization was followed by recovery
`with streptavidin beads. Elution and sequencing revealed a 50,000-
`fold enrichment of the target; however, owing to the small size of
`the target, this enrichment resulted in only 2–5% of reads being
`on-target (Fig. 1a). The recovered DNA was then subjected to
`iterative rounds of PCR and capture. In two independent experi-
`ments, two rounds of capture resulted in >97% of reads mapping
`to the ABL1 gene. A third round of capture provided no further
`improvement (Fig. 1a).
`The double-capture approach resulted in extremely high depth
`and uniformity of coverage (Fig. 1b). Conventional capture
`yielded a maximum on-target depth of 25,000×. In contrast, with
`equivalent use of sequencing capacity, double capture gave up to
`1,000,000× depth, with average and minimum depths of 830,000×
`and 250,000×, respectively. The duplex tags were then used to
`collapse into consensus sequences the PCR duplicates for which
`the two strands of individual DNA molecules were perfectly com-
`plementary. This yielded an average of more than 1,000 unique
`DNA molecules sampled at every nucleotide position within the
`ABL1 target (Supplementary Fig. 1).
`
`1Department of Medicine, Divisions of Hematology and Medical Oncology, University of Washington, Seattle, Washington, USA. 2Department of Pathology, University
`of Washington, Seattle, Washington, USA. 3Clinical Research Division, Fred Hutchinson Cancer Research Center, Seattle, Washington, USA. Correspondence should be
`addressed to L.A.L. (laloeb@uw.edu).
`Received 19 NovembeR 2014; accepted 19 maRch 2015; published oNliNe 6 apRil 2015; doi:10.1038/Nmeth.3351
`
`nature methods  |  VOL.12  NO.5  |  MAY 2015  |  423
`
`© 2015 Nature America, Inc. All rights reserved.
`
`npg
`
`00001
`
`EX1068
`
`

`

`Single capture
`
`Double capture
`
`100
`90
`80
`70
`60
`50
`40
`30
`20
`10
`
`b
`
`Targeted nucleotides (%)
`
`Flanking introns
`
`Targeted exons
`
`100,000
`Sequencing depth
`
`1,000,000
`
`0 1
`
`0,000
`
`1×
`3×
`2×
`capture
`capture
`capture
`Experiment 1
`
`3×
`2×
`1×
`capture
`capture
`capture
`Experiment 2
`
`100
`90
`80
`70
`60
`50
`40
`30
`20
`10
`0
`
`a
`
`On-target reads (%)
`
`brief communications
`
`figure 1 | High on-target recovery with sequential
`rounds of capture. (a) Human genomic DNA
`captured with biotinylated probes targeting ABL1
`exons 4–7 results in low on-target recovery after
`one round of capture, whereas two rounds result
`in >97% of reads mapping to the targeted gene.
`Experiment 1 was carried out with conventional
`blocking oligonucleotides mws60 and mws61;
`experiment 2 used chemically modified high-
`affinity blocking oligonucleotides mws58 and
`mws59 (supplementary table 1). (b) Percentage
`of targeted nucleotides covered at a given
`sequencing depth after single and double capture.
`Both samples were sequenced on an equivalent
`fraction of a HiSeq 2500 lane (5 × 106 paired-end
`reads, corresponding to 3% of a single lane).
`
`single-strand consensus sequences10. This analysis resulted
`in mutations at hundreds of positions in the ABL1 target
`(Supplementary Fig. 2), in contrast to the one true mutation
`that was found by duplex sequencing. The discrepancy indicates
`that >99% of mutations identified by the single-stranded tagging
`approach are artifacts.
`We next determined whether our approach could scale to
`multiple targets. We obtained biotinylated oligonucleotides
`corresponding to the coding exons of the five human replicative DNA
`polymerases13 (total target size, 19.4 kb) and applied the double-
`capture approach to DNA extracted from histologically normal
`human prostate and colon. More than 90% of reads mapped to
`the targeted genes, revealing mutation frequencies of 1 × 10−7
`to 4 × 10−7 (Supplementary Table 2). Among the mutations,
`six were in introns and two changed the coding sequence of DNA
`polymerase epsilon (Supplementary Table 3). The frequency of
`mutations is in accord with prior estimates14,15 of the spontane-
`ous mutation rate in human cells and thus could be the result
`of multiple rounds of cell division and endogenous mutagenic
`processes. Alternatively, these mutations could represent artifacts
`in our assay. However, the error frequency of duplex sequencing
`has been estimated to be <4 × 10−10, as complementary errors
`would need to occur in both strands to be scored10.
`Our approach allows for the study of small genomic regions,
`such as individual human exons or viral sequences present at
`low levels in human samples. Owing to the high level of enrich-
`ment, significant depth can be obtained with modest sequencing.
`For example, a 1-kb target can be sequenced to 100,000-fold depth
`with 4 × 105 paired-end 125-nt reads, and thus hundreds of samples
`can be sequenced simultaneously on a single lane of an Illumina
`HiSeq 2500. The approach is therefore highly scalable and cost
`effective for sequencing small targets. Duplex sequencing on larger
`targets, such as whole exomes, is also possible in principle with a
`greater use of sequencing capacity. For example, under optimized
`
`figure 2 | Removal of sequencing artifacts by duplex sequencing.
`(a) Exons in ABL1 spanning the active site of the enzyme were enriched
`by the double-capture protocol and sequenced conventionally on an
`Illumina HiSeq 2500. Despite extremely stringent quality filtering
`(minimum Phred score = 50) and removal of end-repair artifacts by 5-nt
`trimming from read ends, true mutations cannot be discerned among the
`thousands of sequencing errors that persist. (b) Duplex sequencing of the
`same sample reveals a single point mutation in ABL1 that confers imatinib
`resistance. The mutation was verified by reverse-transcription PCR and
`Sanger sequencing.
`
`We used our protocol to sequence the ABL1 gene from an
`individual with chronic myeloid leukemia who relapsed after
`treatment with the targeted therapy imatinib. Conventional high-
`throughput sequencing was unable to resolve any mutations in
`the sample (Fig. 2a). Even stringent quality filtering (requiring a
`minimum Phred quality score of 50) was unable to remove back-
`ground errors, as many sequencing errors occur during PCR
`amplification and thus cannot be removed by quality filtering10.
`In contrast, duplex sequencing revealed a single mutation with a
`mutant fraction of 1% (Fig. 2b). This mutation, E279K, is known
`to confer imatinib resistance11.
`Alternative methods to detect subclonal mutations have been
`described that result in multiple copies of ssDNA linked together
`by concatemerization12 or tagged with a molecular identifier
`sequence8,9. These approaches are inherently more error prone
`than duplex sequencing because they use information from only
`one of the two DNA strands and thus have less capability for error
`correction. To compare our double-stranded tagging approach
`to these methods, we reanalyzed our data using information
`from only one of the two tagged strands, which we refer to as
`
`500
`
`1,000
`
`1,500
`2,000
`Genome position
`
`2,500
`
`3,000
`
`3,500
`
`500
`
`1,000
`
`1,500
`2,000
`Genome position
`
`2,500
`
`3,000
`
`3,500
`
`2.5a
`
`2.0
`
`1.5
`
`1.0
`
`0.5
`
`0
`
`1
`
`2.5
`
`2.0
`
`1.5
`
`1.0
`
`0.5
`
`0
`
`1
`
`Mutant fraction (%)
`
`b
`
`Mutant fraction (%)
`
`424  |  VOL.12  NO.5  |  MAY 2015  |  nature methods
`
`© 2015 Nature America, Inc. All rights reserved.
`
`npg
`
`00002
`
`

`

`brief communications
`
`conditions, the full exome from 100 individual cells would require
`approximately 2 × 1011 nt of sequence capacity, which is within the
`output range of currently available sequencers.
`Our ABL1 results indicate that it is possible to assay for the
`presence of preexisting subclones encoding resistance to targeted
`cancer therapies, which would be expected to clonally expand in
`the presence of corresponding inhibitors. Armed with this knowl-
`edge, physicians could treat patients with drugs chosen for their
`lack of any detectable resistance. Targeted, high-accuracy capture
`has additional applications in a wide range of fields, including
`the detection of tumor-specific circulating DNA as a biomarker
`for cancer treatment16, the detection of minimal residual disease
`in hematologic malignancies17, confirming candidate subclonal
`mutations that are found by conventional sequencing, analysis of
`mutational processes in cancer18 and testing for low-level resist-
`ance mutations in viral populations. Moreover, as the extreme
`accuracy of the approach results in a theoretical need of only 1×
`coverage of a genome to obtain an accurate sequence, we antici-
`pate applications in settings where sample availability is extremely
`limited, such as paleogenomics, forensics and the study of circu-
`lating tumor cells.
`
`methods
`Methods and any associated references are available in the online
`version of the paper.
`
`Accession codes. NCBI BioProject: PRJNA275267.
`
`Note: Any Supplementary Information and Source Data files are available in the
`online version of the paper.
`
`acknoWledgments
`Research reported in this publication was supported by the US National
`Institutes of Health under award numbers NCI P01-CA77852, R01-CA160674 and
`R33-CA181771 to L.A.L. and NCI U10-CA180861, P01-CA018029, R01-CA175008
`
`and R01-CA175215 to J.P.R. We thank T. Walsh and M. Lee for assistance with
`DNA sequencing.
`
`author contributions
`M.W.S., E.J.F., M.J.P., K.S.R.-B., L.D.T., J.P.R. and L.A.L. contributed to
`experimental design. M.W.S., E.J.F. and M.J.P. performed the experiments in the
`paper and analyzed data. E.J.F., L.D.T. and J.P.R. contributed patient samples.
`M.W.S. and L.A.L. wrote the manuscript.
`
`comPeting financial interests
`The authors declare competing financial interests: details are available in the
`online version of the paper.
`
`reprints and permissions information is available online at http://www.nature.
`com/reprints/index.html.
`
`1. Schmitt, M.W., Prindle, M.J. & Loeb, L.A. Ann. NY Acad. Sci. 1267,
`110–116 (2012).
`2. Mamanova, L. et al. Nat. Methods 7, 111–118 (2010).
`3. Hardenbol, P. et al. Nat. Biotechnol. 21, 673–678 (2003).
`4. Kanagawa, T. J. Biosci. Bioeng. 96, 317–323 (2003).
`5. Fox, E.J., Reid-Bayliss, K.S., Emond, M.J. & Loeb, L.A. Next Gener. Seq.
`Appl. 1, 106–109 (2014).
`6. Glenn, T.C. Mol. Ecol. Resour. 11, 759–769 (2011).
`7. Jabara, C.B., Jones, C.D., Roach, J., Anderson, J.A. & Swanstrom, R.
`Proc. Natl. Acad. Sci. USA 108, 20166–20171 (2011).
`8. Kinde, I., Wu, J., Papadopoulos, N., Kinzler, K.W. & Vogelstein, B.
`Proc. Natl. Acad. Sci. USA 108, 9530–9535 (2011).
`9. Hiatt, J.B., Pritchard, C.C., Salipante, S.J., O’Roak, B.J. & Shendure, J.
`Genome Res. 23, 843–854 (2013).
`10. Schmitt, M.W. et al. Proc. Natl. Acad. Sci. USA 109, 14508–14513 (2012).
`11. Soverini, S. et al. Leuk. Res. 38, 10–20 (2014).
`12. Lou, D.I. et al. Proc. Natl. Acad. Sci. USA 110, 19872–19877 (2013).
`13. Sweasy, J.B., Lauper, J.M. & Eckert, K.A. Radiat. Res. 166, 693–714
`(2006).
`14. Albertini, R.J., Nicklas, J.A., O’Neill, J.P. & Robison, S.H. Annu. Rev.
`Genet. 24, 305–326 (1990).
`15. Kunkel, T.A. J. Biol. Chem. 279, 16895–16898 (2004).
`16. Esposito, A. et al. Cancer Treat. Rev. 40, 648–655 (2014).
`17. Buckley, S.A., Appelbaum, F.R. & Walter, R.B. Bone Marrow Transplant. 48,
`630–641 (2013).
`18. Alexandrov, L.B. et al. Nature 500, 415–421 (2013).
`
`nature methods  |  VOL.12  NO.5  |  MAY 2015  |  425
`
`© 2015 Nature America, Inc. All rights reserved.
`
`npg
`
`00003
`
`

`

`online methods
`Human subjects approval. Use of human samples was approved
`by the Institutional Review Board at the University of Washington.
`Informed consent was obtained from patients who contributed
`samples.
`
`DNA isolation. Genomic DNA was extracted from peripheral
`blood mononuclear cells or tissue by high-salt extraction using
`the Agilent DNA extraction kit #200600.
`
`Ligation of duplex sequencing adaptors. Duplex sequencing
`was initially described with use of A-tailed adaptors10,19; we have
`since found that T-tailed adaptors result in improved ligation
`efficiency, and we have published a detailed protocol for their syn-
`thesis and use20. In brief, DNA was sheared, end repaired, A tailed
`and then ligated to T-tailed duplex sequencing adaptors using a
`20× molar excess of adaptors relative to A-tailed DNA molecules.
`Following reaction cleanup with 1.0 volumes of Ampure XP beads
`(Agencourt), the adaptor-ligated DNA was PCR amplified for
`five cycles with the KAPA Biosystems hot-start high-fidelity kit
`using primers mws13 and mws20 (Supplementary Table 1).
`240 ng of input DNA were used in each 100-µl PCR reaction, with
`2–8 PCR reactions performed per sample. Owing to the small
`amount of on-target DNA present in the starting sample, multiple
`PCR reactions are needed to amplify sufficient on-target DNA for
`capture. Each PCR reaction results in sequence data representing
`approximately 500 independent genomes; the number of PCR
`reactions performed can be adjusted depending on the sequenc-
`ing coverage desired. The products from all reactions were pooled
`and purified with 1.2 volumes of Ampure XP beads, with a final
`elution volume of 50 µl.
`
`Targeted capture. One-third of the total amount of adaptor-
`ligated DNA generated by PCR was combined with 5 µg of Cot-I
`DNA (Invitrogen) and 1 nmol each of blocking oligonucleotides
`mws60 and mws61. The mixture was completely lyophilized and
`then resuspended in 2.5 µl water, 7.5 µl NimbleGen 2× hybridi-
`zation buffer and 3 µl NimbleGen hybridization component A.
`The mixture was heated to 95 °C for 10 min, the tempera-
`ture was adjusted to 65 °C and 3 pmol of pooled biotinylated
`oligonucleotides were added (Integrated DNA Technologies).
`
`After 4 h, M-270 streptavidin beads (Life Technologies) were
`added and washes were performed according to the IDT xGen
`lockdown probe protocol version 2.0. We found that the standard
`quantity of streptavidin beads (the IDT protocol calls for 100 µl
`of beads per 50-µl PCR reaction) can result in PCR inhibition,
`so the quantity of beads was decreased to 75 µl per reaction, and
`the PCR reaction volume increased to 100 µl. The product was
`PCR amplified for 16 cycles with primers mws13 and mws20
`and purified with 1.2 volumes of Ampure XP beads. The puri-
`fied DNA was combined with 2.5 µg Cot-I DNA and 500 pmol
`each of oligonucleotides mws60 and mws61, and a second round
`of capture21 was performed with 1.5 pmol of pooled biotinylated
`oligonucleotides. A final PCR reaction was carried out for 8–10
`cycles with primers mws13 and mws21, which contain a fixed
`index sequence for multiplexing. After cleanup with 1.2 volumes
`of Ampure XP beads, the product was sequenced on an Illumina
`HiSeq 2500.
`
`Data processing. Processing of duplex sequencing data was per-
`formed essentially as previously described20. Mutations identi-
`fied by duplex sequencing were individually inspected in the
`Integrated Genome Viewer22 to verify that they were not affected
`by alignment errors. Software for duplex sequencing is available at
`https://github.com/loeblab/Duplex-Sequencing/. Data from this
`paper have been uploaded to the Sequence Read Archive under
`BioProject ID: PRJNA275267.
`
`Reverse-transcription PCR of the ABL1 gene. Total RNA was
`extracted from peripheral blood using Trizol reagent (Invitrogen).
`An initial RT-PCR step with nested PCR was used to amplify
`exons 4–9 (codons 199–507) of the ABL1 kinase domain, and
`bidirectional Sanger sequencing of the PCR product was per-
`formed, as previously described23.
`
`19. Kennedy, S.R., Salk, J.J., Schmitt, M.W. & Loeb, L.A. PLoS Genet. 9,
`e1003794 (2013).
`20. Kennedy, S.R. et al. Nat. Protoc. 9, 2586–2606 (2014).
`21. Burgess, D. et al. SeqCap EZ Library: Technical Note http://www.
`nimblegen.com/products/lit/06870406001_NG_SeqCapEZ_DoubleCaptureSR_
`20Aug2012.pdf (Roche NimbleGen, (2012)).
`22. Robinson, J.T. et al. Nat. Biotechnol. 29, 24–26 (2011).
`23. Egan, D.N., Beppu, L. & Radich, J.P. Biol. Blood Marrow Transplant. 21,
`184–189 (2014).
`
`nature methods
`
`doi:10.1038/nmeth.3351
`
`© 2015 Nature America, Inc. All rights reserved.
`
`npg
`
`00004
`
`

`

`Supplementary Figure 1
`
`Number of unique duplex DNA molecules sampled at each targeted position in the ABL1 gene after single and double capture.
`
`Nature Methods: doi:10.1038/nmeth.3351
`
`00005
`
`

`

`Supplementary Figure 2
`
` 
`
`Mutations in the ABL1 gene identified by single-strand consensus sequencing (SSCS).
`
`Nature Methods: doi:10.1038/nmeth.3351
`
`00006
`
`

`

`Table  S1:  Oligonucleotide  Sequences
`
`Sequence
`
`Oligonucleotide  name
`mws13
`mws20
`mws21a
`mws58b
`mws59b
`mws60
`mws61
`mws62-­‐abl-­‐probe1
`mws63-­‐abl-­‐probe2
`mws64-­‐abl-­‐probe3
`mws65-­‐abl-­‐probe4
`mws66-­‐abl-­‐probe5
`mws67-­‐abl-­‐probe6
`mws68-­‐abl-­‐probe7
`mws69-­‐abl-­‐probe8
`
`AATGATACGGCGACCACCGAG
`GTGACTGGAGTTCAGACGTGTGC
`CAAGCAGAAGACGGCATACGAGAT XXXXXXXX GTGACTGGAGTTCAGACGTGTGC
`AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCTIIIIIIIIIIIITGACT
`GTCAIIIIIIIIIIIIAGATCGGAAGAGCACACGTCTGAACTCCAGTCAC
`AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCTIIIIIIIIIIIITGACT
`GTCAIIIIIIIIIIIIAGATCGGAAGAGCACACGTCTGAACTCCAGTCAC
`/5'Biotin/CTCTACGTCTCCTCCGAGAGCCGCTTCAACACCCTGGCCGAGTTGGTTCATCATCATTCAACGGTGGCCGACGGGCTCATCACCACGCTC
`/5'Biotin/CATTATCCAGCCCCAAAGCGCAACAAGCCCACTGTCTATGGTGTGTCCCCCAACTACGACAAGTGGGAGATGGAACGCACGGACATCACC
`/5'Biotin/ATGAAGCACAAGCTGGGCGGGGGCCAGTACGGGGAGGTGTACGAGGGCGTGTGGAAGAAATACAGCCTGACGGTGGCCGTGAAGACCTTG
`/5'Biotin/GAGGACACCATGGAGGTGGAAGAGTTCTTGAAAGAAGCTGCAGTCATGAAAGAGATCAAACACCCTAACCTGGTGCAGCTCCTTG
`/5'Biotin/GGGTCTGCACCCGGGAGCCCCCGTTCTATATCATCACTGAGTTCATGACCTACGGGAACCTCCTGGACTACCTGAGGGAGTGCAACCGGC
`/5'Biotin/AGGAGGTGAACGCCGTGGTGCTGCTGTACATGGCCACTCAGATCTCGTCAGCCATGGAGTACCTGGAGAAGAAAAACTTCATCCACAG
`/5'Biotin/ATCTTGCTGCCCGAAACTGCCTGGTAGGGGAGAACCACTTGGTGAAGGTAGCTGATTTTGGCCTGAGCAGGTTGATGACAGGGGACACCT
`/5'Biotin/ACACAGCCCATGCTGGAGCCAAGTTCCCCATCAAATGGACTGCACCCGAGAGCCTGGCCTACAACAAGTTCTCCATCAAGTCCGACGTCT
`a.  XXXXXX  indicates  a  fixed  8-­‐nucleotide  barcode  sequence  for  multiplexing
`b.  mws58  and  mws59  are  identical  in  sequence  to  mws60  and  mws61.  These  oligonucleotides  contain  proprietary  chemically  modified  nucleotides  synthesized  by  Integrated
`DNA  technologies,  which  are  reported  to  enhance  the  strangth  of  binding  of  the  blocking  oligonucleotide  and  thus  improve  on-­‐target  capture  performance.
`
`Nature Methods: doi:10.1038/nmeth.3351
`
`00007
`
`

`

`Table  S2:  Sequence  data  from  human  samples
`
`Sequencer  lane  fraction Paired-­‐end  reads  obtaineda Duplex  nucleotidesb Duplex  mutations Mutation  frequency
`Sample
`3%
`4.02E+06
`3.29E+06
`1
`3.04E-­‐07
`CML  relapse
`25%
`3.75E+07
`3.30E+07
`4
`1.21E-­‐07
`Normal  prostate
`55%
`8.30E+07
`6.53E+07
`4
`6.12E-­‐08
`Normal  colon
`a.  101  nucleotide  paired-­‐end  reads  were  obtained  on  an  Illumina  Hiseq  2500
`b.  Final  number  of  unique  nucleotides,  after  collapsing  duplicate  reads  arising  from  each  of  the  two  DNA  strands  into  consensus  reads
`
`Nature Methods: doi:10.1038/nmeth.3351
`
`00008
`
`

`

`Table  S3:  Sub-­‐clonal  mutations  identified  in  human  samples
`
`Gene Chromosome Nucleotide  position Reference  nucleotide Mutant  nucleotide Mutation Mutant  fraction
`Sample
`ABL1
`9
`133747528
`G
`A
`E279K
`1.11%
`CML  relapse
`Normal  prostate POLA
`X
`24745824
`G
`C
`intron
`0.45%
`POLE
`12
`133218421
`G
`C
`D1730E
`0.16%
`POLE
`12
`133233870
`G
`C
`intron
`0.11%
`POLE
`12
`133256498
`C
`T
`intron
`0.11%
`POLD1
`19
`50906529
`G
`A
`intron
`0.47%
`POLD1
`19
`50905074
`G
`A
`R119H
`0.13%
`POLE
`12
`133252947
`T
`C
`intron
`0.10%
`POLE
`12
`133256904
`C
`T
`intron
`0.09%
`
`Normal  colon
`
`Nature Methods: doi:10.1038/nmeth.3351
`
`00009
`
`