Downloaded from https://academic.oup.com/clinchem/article/58/3/580/5620563 by guest on 17 September 2023
`
`Clinical Chemistry 58:3
`580–589 (2012)
`
`Molecular Diagnostics and Genetics
`
`COLD-PCR Enrichment of Rare Cancer Mutations prior to
`Targeted Amplicon Resequencing
`Coren A. Milbury,1 Mick Correll,2 John Quackenbush,2 Renee Rubio,2 and G. Mike Makrigiorgos1,3*
`
`BACKGROUND: Despite widespread interest
`in next-
`generation sequencing (NGS), the adoption of personal-
`ized clinical genomics and mutation profiling of cancer
`specimens is lagging, in part because of technical limita-
`tions. Tumors are genetically heterogeneous and often
`contain normal/stromal cells, features that lead to low-
`abundance somatic mutations that generate ambiguous
`results or reside below NGS detection limits, thus hinder-
`ing the clinical sensitivity/specificity standards of muta-
`tion calling. We applied COLD-PCR (coamplification at
`lower denaturation temperature PCR), a PCR methodol-
`ogy that selectively enriches variants, to improve the de-
`tection of unknown mutations before NGS-based ampli-
`con resequencing.
`
`METHODS: We used both COLD-PCR and conventional
`PCR (for comparison) to amplify serially diluted
`mutation-containing cell-line DNA diluted into wild-
`type DNA, as well as DNA from lung adenocarcinoma
`and colorectal cancer samples. After amplification of
`TP53 (tumor protein p53), KRAS (v-Ki-ras2 Kirsten rat
`sarcoma viral oncogene homolog), IDH1 [isocitrate de-
`hydrogenase 1 (NADP⫹), soluble], and EGFR (epidermal
`growth factor receptor) gene regions, PCR products were
`pooled for library preparation, bar-coded, and sequenced
`on the Illumina HiSeq 2000.
`
`RESULTS: In agreement with recent findings, sequencing
`errors by conventional targeted-amplicon approaches
`dictated a mutation-detection limit of approximately
`1%–2%. Conversely, COLD-PCR amplicons enriched
`mutations above the error-related noise, enabling reliable
`identification of mutation abundances of approximately
`0.04%. Sequencing depth was not a large factor in the
`identification of COLD-PCR–enriched mutations. For
`the clinical samples, several missense mutations were not
`called with conventional amplicons, yet they were clearly
`detectable with COLD-PCR amplicons. Tumor heteroge-
`neity for the TP53 gene was apparent.
`
`CONCLUSIONS: As cancer care shifts toward personalized in-
`tervention based on each patient’s unique genetic abnormal-
`ities and tumor genome, we anticipate that COLD-PCR
`combined with NGS will elucidate the role of mutations in
`tumor progression, enabling NGS-based analysis of diverse
`clinical specimens within clinical practice.
`© 2011 American Association for Clinical Chemistry
`
`Rapidly evolving sequencing technologies have em-
`powered enormous growth in the breadth and depth of
`cancer genome characterization. Second-generation
`massively parallel sequencing approaches are increas-
`ing the throughput and decreasing the costs of
`nucleotide-resolution oncogenomics, thereby making
`the characterization of entire transcriptomes, exomes,
`and genomes readily achievable (1, 2 ). Currently, the
`pace of acquisition of genomic data for cancer patients
`far outstrips the utility of that information for choosing
`specific therapeutic avenues for individualized patient
`care (1 ). Targeted amplicon resequencing is an alter-
`native that balances the amount of information ob-
`tained, affordability, and the ability to include muta-
`tion profiling of the most meaningful genes (3–5 ). The
`opportunity provided by clinical genomics is unique,
`because for a growing number of tumor types, clinical
`decision-making for patients with diagnosed cancers
`will increasingly be driven by the status of mutated can-
`cer genes (1 ). Whether these new approaches will affect
`routine clinical practice and the treatment of disease is
`no longer debatable, but how that will happen is a
`source of ongoing speculation and development (1 ).
`There are both conceptual and technical chal-
`lenges to assessing the wealth of information for indi-
`vidual tumors obtained with next-generation sequenc-
`ing (NGS)4 technologies (1 ). For example, the clinical
`importance of the minor alleles frequently encoun-
`tered in key cancer genes and often appearing at abun-
`
`1 Division of DNA Repair and Genome Stability, Department of Radiation Oncol-
`ogy; 2 Center for Cancer Computational Biology; and 3 Division of Medical
`Physics and Biophysics, Department of Radiation Oncology, Dana-Farber Cancer
`Institute, Harvard Medical School, Boston, MA.
`* Address correspondence to this author at: Dana-Farber/Brigham and Women’s
`Cancer Center, Brigham and Women’s Hospital, Level L2, Radiation Ther-
`apy, 75 Francis St., Boston, MA 02115. Fax 617-525-7123; e-mail
`
`mmakrigiorgos@lroc.harvard.edu.
`Received September 20, 2011; accepted November 17, 2011.
`Previously published online at DOI: 10.1373/clinchem.2011.176198
`4 Nonstandard abbreviations: NGS, next-generation sequencing; COLD-PCR, coam-
`plification at lower denaturation temperature PCR; TL, lung tumor; CT, colorectal
`tumor; Tc, critical denaturation temperature; Tm, amplicon melting temperature.
`
`580
`
`Personalis EX2020
`
`

`

`Downloaded from https://academic.oup.com/clinchem/article/58/3/580/5620563 by guest on 17 September 2023
`
`COLD-PCR Improves Mutation Calling via NGS
`
`dances of ⬍10% in the tumor cell population appears
`to be case dependent. Low-abundance clones contain-
`ing TET25 (tet methylcytosine dioxygenase 2) muta-
`tions in chronic myelomonocytic leukemia patients
`confer no prognostic value (6 ). In contrast, detection
`of low-level KRAS (v-Ki-ras2 Kirsten rat sarcoma viral
`oncogene homolog) mutations in metastatic colorectal
`cancer enhances the prediction of resistance to treat-
`ment with an anti– epidermal growth factor receptor
`monoclonal antibody (7 ). Mutations in 1%–5% of
`cells in primary breast tumors can be found at preva-
`lent (clonal) status in the secondary metastasis, a find-
`ing consistent with the mutations having obtained
`“driver” status in the microenvironment of the meta-
`static site (8 ). Clearly, a fraction of the low-abundance
`genetic alterations constitutes precious clinical infor-
`mation that one should be able to capture.
`Before we can identify which low-abundance
`DNA variants revealed by NGS are clinically meaning-
`ful, the question of the confidence in the generated data
`must be addressed. Although NGS has been demon-
`strated to be reliable for DNA with high-prevalence
`tumor somatic mutations (5, 8, 9 ), the required depth
`of sequence interrogation remains problematic (10 ),
`and the detection of low-prevalence somatic mutations
`at levels below approximately 2%–5% in tumors with
`heterogeneity, in tumors with stromal contamination,
`or in bodily fluids, is fraught with false positives, irre-
`spective of the coverage (4, 11 ). Because physicians are
`not likely to make clinical decisions on the basis of
`DNA-mutation signals that are close to background
`levels, it is important to develop procedures that enable
`signals arising from low-abundance mutations to be
`separated from the background noise and thereby
`boost the confidence in NGS results.
`We evaluated the use of COLD-PCR (coamplifica-
`tion at lower denaturation temperature PCR), a newly
`developed methodology from our laboratory (12, 13 ),
`to enhance mutation detection via massively parallel
`sequencing with the Illumina HiSeq 2000 analyzer.
`This approach enabled genuine low-abundance muta-
`tions to be magnified by enrichment before NGS-based
`amplicon resequencing, thereby enabling a clear dis-
`tinction of mutations from the background sequencing
`noise. The combination of COLD-PCR with NGS im-
`proved the detection limit of targeted amplicon rese-
`
`5 Human genes: TET2, tet methylcytosine dioxygenase 2; KRAS, v-Ki-ras2 Kirsten
`rat sarcoma viral oncogene homolog; TP53, tumor suppressor protein 53; IDH1,
`isocitrate dehydrogenase subunit 1 (NADP⫹), soluble; EGFR, epidermal growth
`factor receptor; KIF1C, kinesin family member 1C; USP28, ubiquitin specific
`peptidase 28; KIT, v-kit Hardy–Zuckerman 4 feline sarcoma viral oncogene
`homolog.
`
`quencing of
`100-fold.
`
`low-abundance mutations by 50- to
`
`Materials and Methods
`
`DNA TEMPLATE AND SERIAL DILUTIONS OF MUTANTS
`Human cell line DNA was obtained from the ATCC
`and the Dana-Farber Cancer Institute (Boston, MA).
`Frozen tissue was obtained from clinical glioblastoma,
`lung, and colon tumor specimens according to the ap-
`proval from the Internal Review Board. Genomic DNA
`was isolated with the DNeasy™ Blood & Tissue Kit
`(Qiagen) according to manufacturer instructions.
`DNA quality and concentration were measured with a
`NanoDrop 1000 spectrophotometer (Thermo Scien-
`tific). Data for all evaluated DNA are presented in Ta-
`ble 1 in the Data Supplement that accompanies the
`online
`version of
`this
`article
`at http://www.
`clinchem.org/content/vol58/issue3.
`Genomic DNA from cell lines and 1 previously
`evaluated glioblastoma sample (14 ) was serially diluted
`into human male wild-type genomic DNA (Human
`Genomic DNA: Male; Promega) to generate the fol-
`lowing preamplification mutant DNA abundances:
`2%, 1%, 0.5%, 0.2%, 0.1%, 0.05%, 0.02%, and 0%. In
`each PCR reaction, 100 ng of genomic DNA was used
`to ensure an efficient representation of minor alleles.
`Clinical lung tumor (TL) and colorectal tumor
`(CT) samples containing naturally occurring medium-
`and low-level somatic mutations that had previously
`been documented by at least 1 independent method in
`addition to COLD-PCR (12, 15–17 ) were analyzed in
`parallel with their paired, putatively normal samples,
`which were obtained from tumor margins and proxi-
`mal regions during surgery. Naturally occurring muta-
`tion abundances in the clinical samples varied from
`⬍1% to heterozygous status.
`
`TARGET AMPLICON REGIONS
`Nine regions in 4 frequently mutated oncogenes were
`evaluated: exons 5–10 of the TP53 (tumor suppressor
`protein 53) gene, exon 2 of the KRAS gene, exon 4 of
`the
`IDH1 [isocitrate dehydrogenase
`subunit 1
`(NADP⫹), soluble] gene, and exon 20 of the EGFR
`(epidermal growth factor receptor) gene. The locations
`of the amplicons and their primers are presented in
`Table 2 in the online Data Supplement.
`
`AMPLIFICATION STRATEGIES
`COLD-PCR is a recently developed PCR-based ap-
`proach for enriching low-abundance DNA mutations
`and minor allele variants (12 ). COLD-PCR enriches
`unknown mutations at any position within the ampli-
`con through the use of a critical denaturation temper-
`ature (Tc) during the PCR. The Tc is lower than stan-
`
`Clinical Chemistry 58:3 (2012) 581
`
`Personalis EX2020
`
`

`

`Downloaded from https://academic.oup.com/clinchem/article/58/3/580/5620563 by guest on 17 September 2023
`
`dard denaturation temperatures and preferentially
`denatures heteroduplexed molecules (those formed by
`hybridization of mutant and wild-type sequences) and
`amplicons possessing mutations that lower the ampli-
`con melting temperature (Tm), such as G:C⬎A:T or
`G:C⬎T:A. Minor-allele enrichment by COLD-PCR
`has been demonstrated in combination with several
`downstream approaches, such as Sanger sequencing,
`denaturing HPLC/Surveyor, MALDI-TOF, pyrose-
`quencing, real-time TaqMan, single-strand conforma-
`tion polymorphism, PCR-based mutation-specific re-
`striction enzyme digestion,
`and high-resolution
`melting analyses (12, 14, 15, 18 –21 ). The combination
`of COLD-PCR and NGS and the impact of the former
`on NGS have not previously been reported.
`The details of the thermocycling conditions and am-
`plification protocols we have used are presented in Table 3
`in the online Data Supplement. We evaluated 2 COLD-
`PCR platforms: fast COLD-PCR and ice-COLD-PCR.
`COLD-PCR amplifications used a Tc to preferentially de-
`nature and enrich allelic variants containing a lower Tm,
`as previously described (13–17, 22–26). PCR reactions
`were performed on SmartCycler II thermocyclers (Ce-
`pheid). The incidence of PCR errors was reduced by per-
`forming all reactions (25-␮L final volume) with Phu-
`sion™ polymerase (New England Biolabs), which
`possesses a very high replication fidelity. Conventional
`and COLD-PCR reactions were performed with 1⫻
`manufacturer-supplied HF Buffer (New England Bio-
`labs), 0.2 mmol/L of each deoxynucleoside triphosphate,
`0.3 ␮mol/L primers, 1.0⫻ LCGreen威 Plus⫹ dye (Idaho
`Technologies), and 0.02 U/␮L Phusion polymerase. Ice-
`COLD-PCR reactions were performed with the 1⫻
`manufacturer-supplied HF buffer, 0.2 mmol/L of each
`deoxynucleoside triphosphate, 0.9 ␮mol/L primers, 1.0⫻
`LCGreen Plus⫹ dye, 0.02 U/␮L Phusion polymerase, and
`25 nmol/L reference sequence oligonucleotide.
`
`SANGER SEQUENCING CONFIRMATION
`PCR products from the wild type and from samples
`with preamplification mutation abundances of 1% and
`0.2% were digested with exonuclease I (New England
`Biolabs) and shrimp alkaline phosphatase
`(Af-
`fymetrix). Products were processed for Sanger se-
`quencing at the Molecular Biology Core Facility of the
`Dana-Farber Cancer Institute (for primer sequences,
`see Table 2 in the online Data Supplement).
`
`AMPLICON LIBRARY PREPARATION FOR ILLUMINA NGS
`Amplicons were purified with the QIAquick™ PCR Puri-
`fication Kit (Qiagen) and quantified on a NanoDrop 1000
`spectrophotometer. Purified PCR products were pooled
`in equivalent concentrations across the serially diluted
`mutant DNA abundances. Final amplicon mixtures (ap-
`proximately 1–2 ␮g) were precipitated in ethanol and 0.3
`
`582 Clinical Chemistry 58:3 (2012)
`
`mol/L sodium acetate, washed in 700 mL/L ethanol,
`dried, and resuspended in 30 ␮L water.
`Library preparation for paired-end NGS on the
`Illumina HiSeq 2000 instrument was performed at the
`Center for Cancer Computational Biology at the Dana-
`Farber Cancer Institute. PCR products underwent end
`repair and A-tailing according to outlined protocols
`(Illumina). Products were purified with the Agencourt
`AMPure® XP Bead system (Beckman Coulter Genom-
`ics). Paired-end adaptors (TruSeq Sample Prep Kit;
`Illumina) were ligated to the products after the multi-
`plex paired-end protocols, as outlined. Ligation prod-
`ucts were purified via the AMPure system, and target
`bands (220 –270 bp) were collected and purified (Min-
`Elute Gel Extraction Kit; Qiagen). Phusion polymerase
`was used in the PCR to enrich the adapter-modified
`products. In each case, 12 bar codes were multiplexed
`per pool of amplicons. Library validation for quality
`and concentration was performed on the Bioanalyzer
`(Agilent) before immobilization on the flow cell and
`sequencing on the HiSeq 2000 instrument. Products
`were paired-end sequenced with 100 cycles and an in-
`dex read.
`
`DATA ANALYSIS
`Primary analysis, including base calling, read filtering,
`and demultiplexing were performed according to the
`standard Illumina processing pipeline (CASAVA
`1.7.1). Sequence read pairs were mapped indepen-
`dently to the human genome assembly GRh37/hg19
`(build 37.2, February 2009) with Bowtie (27 ). We al-
`lowed up to 3 mismatches across the entire length of
`the read and reported only reads that were uniquely
`aligned (⫺v3 ⫺ml). SAMtools were used to calculate
`read depth and nucleotide frequencies for each posi-
`tion of the amplicons (28 ), and custom PERL scripts
`were used to calculate the observed frequency of each
`nucleotide at each base position and to compile the
`results across samples into an integrated report.
`
`Results
`
`MUTATION DETECTION IN SERIALLY DILUTED
`
`MUTATION-CONTAINING DNA
`
`Sanger sequencing confirmation. Before library prepara-
`tion for NGS, amplicons generated from the 1%, 0.2%,
`and 0% (of wild type) preamplification mutation
`abundances were analyzed via Sanger sequencing (for
`estimates of the resulting mutation abundances, see
`Table 4 in the online Data Supplement). As anticipated,
`the samples with 1% mutation abundances amplified
`with conventional-PCR amplicons fell below the ana-
`lytical sensitivity of Sanger sequencing and could not
`be detected. After COLD-PCR enrichment, however,
`
`Personalis EX2020
`
`

`

`Downloaded from https://academic.oup.com/clinchem/article/58/3/580/5620563 by guest on 17 September 2023
`
`COLD-PCR Improves Mutation Calling via NGS
`
`mutations were pronounced and clearly evident in the
`samples with 1% preamplification abundances of each
`amplified region. The 1% preamplification abundance
`demonstrated an overall mean enrichment of approx-
`imately 57-fold (SD, 11-fold), whereas the 0.2% pre-
`amplification abundance demonstrated a mean en-
`richment of approximately 146-fold (SD, 72-fold). The
`mutation enrichment for the 0.2% preamplification
`mutation abundance varies among the amplicons and
`potentially illustrates the lower limits of detection via
`COLD-PCR and Sanger sequencing analysis.
`
`Targeted amplicon resequencing: serial dilutions. Ampli-
`con regions produced by both conventional and
`COLD-PCR were sequenced (paired-end) on the Illu-
`mina HiSeq 2000 and aligned to the reference genome
`(GRh37/hg19). Frequency calls were generated for
`each nucleotide aligned within the amplicon locations.
`Frequency calls for all discordant nucleotides (relative
`to the wild-type sequence) were plotted for each nucle-
`otide position of each amplicon. We therefore devel-
`oped “variant-and-noise plots” to display both the fre-
`quencies of mutation calls and the background signals.
`For clarity, the wild-type sequence calls were not plot-
`ted but were reflected by the presentation of sequence
`depth relative to the nucleotide frequencies of the vari-
`ant alleles. The overall sequence depth is presented in
`each plot and is defined by the right-hand y axis and
`denoted by the blue plotted line. Fig. 1 presents repre-
`sentative variant-and-noise plots for the serial-dilution
`study of TP53 exon 10. After conventional PCR, the
`exon 10 mutation had a limit of detection of about 2%,
`in view of the maximum sequencing “noise.” In con-
`trast, the enriched mutation after COLD-PCR was ev-
`ident down to a preamplification abundance of 0.02%,
`despite the noise. The noise was defined by the maxi-
`mum observed aberrant calls, such that an error would
`not be mistaken for a mutation or variant. In another
`example, variant-and-noise plots were presented after
`ice-COLD-PCR (see Fig. 1 in the online Data Supple-
`ment). Ice-COLD-PCR involves a more elaborate PCR
`cycling protocol than fast COLD-PCR but can enrich
`for all possible mutations. Genomic DNA from 2 cell
`lines with TP53 exon 8 mutations (HCC1008, Tm-
`equivalent mutation G⬎C; PFSK-1, Tm-increasing
`mutation T⬎G) were mixed, and both mutations were
`simultaneously evaluated in serial dilutions with wild-
`type DNA. The mutation enrichment enabled by ice-
`COLD-PCR allowed reliable detection down to a pre-
`amplification abundance of 0.2%. Table 5 in the online
`Data Supplement summarizes the details of the results
`for the serially diluted mutated DNA. The observed
`nucleotide frequency at the mutation position being
`evaluated is presented for each of the preamplification
`mutation abundances. Table 5 in the online Data Sup-
`
`plement also presents the maximum observed back-
`ground noise along the amplicon sequence (deter-
`mined by the maximum observed frequency of an
`aberrant nucleotide call), the mean noise (the mean
`observed frequency of aberrant calls within the ampli-
`con), and the sequence depth (the number of aligned
`paired-end sequence reads). Serially diluted mutation
`abundances of ⬍2% were not possible to discern in
`most cases with the conventional-PCR amplicons. Se-
`quencing COLD-PCR amplicons, however, demon-
`strated a median detection limit of a 0.04% mutation
`abundance, with an overall range of 0.02%– 0.2%. Ac-
`cordingly, the ability to detect the mutations was im-
`proved by a mean of 50-fold after COLD-PCR.
`The observed frequencies of the mutated nucleotide
`after amplification by conventional PCR and COLD-PCR
`were plotted and compared. Representative plots of mu-
`tation nucleotide frequency are presented in Fig. 2 in the
`online Data Supplement. Observed nucleotide frequen-
`cies after conventional PCR were consistent with the pre-
`pared serial dilutions, whereas observed mutation fre-
`quencies after COLD-PCR reflected the achieved
`enrichment. In some cases, such as with TP53 exons 7 and
`10, preamplification mutation abundances of ⬍1% were
`enriched to ⬎50% (see Table 4 in the online Data Supple-
`ment), which represents mutation enrichments by
`COLD-PCR of up to 300-fold.
`The influence of amplification method and
`sequence-interrogation depth on noise was also as-
`sessed. From the wild-type replicates, we evaluated the
`mean background noise of each amplicon and plotted
`it against sequence depth (see Fig. 3 in the online Data
`Supplement). We calculated a mean noise estimate
`across all amplicons of 0.08% (SD, 0.03%) for the
`conventional-PCR amplicons and 0.15% (SD, 0.06%)
`for the COLD-PCR amplicons. We observed no de-
`crease in noise with increasing depth of interrogation,
`indicating that the observed noise was due not to sam-
`pling error but rather to sequencing errors or upstream
`preparation (polymerase misincorporations). Indeed,
`higher noise was associated with COLD-PCR amplifi-
`cation compared with conventional PCR, possibly ow-
`ing to polymerase errors generated and enriched
`throughout COLD-PCR amplification.
`A comparison of the data in Figs. 2 and 3 in the online
`Data Supplement indicates that genuine mutations are
`enriched by COLD-PCR much more than polymerase er-
`rors and that their enriched abundances overcome the
`sequencing errors. Accordingly, the overall signal-to-
`noise ratio increased sharply after COLD-PCR.
`
`MUTATION DETECTION VIA TARGETED AMPLICON
`
`RESEQUENCING: CLINICAL SAMPLES
`Lung adenocarcinoma and colorectal cancer clinical
`samples with previously demonstrated (12, 15–17 )
`
`Clinical Chemistry 58:3 (2012) 583
`
`Personalis EX2020
`
`

`

`Downloaded from https://academic.oup.com/clinchem/article/58/3/580/5620563 by guest on 17 September 2023
`
`Fig. 1. Variant/noise frequency plots (antisense strand orientation) for TP53 exon 10 mutant serial dilutions, as
`amplified by conventional PCR (A) or COLD-PCR (B).
`The UACC-893 mutation (c.1024C⬎T) cannot be called in conventional-PCR amplicons at mutation abundances of ⱕ1%
`because of background noise. In contrast, after COLD-PCR the preamplification mutational abundance of 0.02% is enriched to
`approximately 14% and is detectable. Sequence read depth (right y axis) is presented as a blue line. WT, wild type.
`
`584 Clinical Chemistry 58:3 (2012)
`
`Personalis EX2020
`
`

`

`COLD-PCR Improves Mutation Calling via NGS
`
`Downloaded from https://academic.oup.com/clinchem/article/58/3/580/5620563 by guest on 17 September 2023
`
`Fig. 2. Mutation analysis in tumor sample TL8 (TP53, c.853G>T) and matched normal sample NL8.
`(A), Antisense Sanger sequences of conventional-PCR or COLD-PCR amplicons, including an estimate of abundance after
`enrichment. (B), In NGS variant-and-noise frequency plots (antisense), the TL8 mutation at approximately 1% cannot be called
`in conventional amplicons because of the background noise; however, COLD-PCR has appreciably enriched the mutation
`(approximately 29%) above the noise.
`
`low-abundance mutations were selected for the present
`study. Putatively normal samples, which were sampled
`from the tumor margin and proximal regions, were
`examined in parallel. We used the Illumina HiSeq 2000
`to generate variant-and-noise plots for these samples.
`For some samples, naturally occurring low-abundance
`mutations and heterogeneity were identified in COLD-
`PCR amplicons, although they were not detectable by
`conventional PCR– based NGS analysis. For example,
`lung adenocarcinoma sample TL8 contained a mis-
`sense mutation of ⬍1% abundance that could not be
`called when conventional-PCR amplicons were used
`(Fig. 2). Conversely, this mutation was enriched by
`COLD-PCR to almost 30% abundance and was easily
`detectable via both Illumina and Sanger sequencing
`(Fig. 2). Similarly, colorectal cancer sample CT20 con-
`tained 3 mutations within TP53: 2 low-level mutations
`in exons 8 and 9 and a heterozygous mutation in exon
`5 (Fig. 3). Whereas the heterozygous mutation in exon
`5 was clearly evident in conventional-PCR amplicons,
`the 2 low-level mutations were borderline detectable
`(approximately 3% abundance in exon 9) to nonde-
`tectable (exon 8). COLD-PCR amplification, however,
`enriched the 3% exon 9 mutation to nearly 76% and
`enriched the exon 8 mutation to ⬎50% (Fig. 3). Wide-
`
`spread intratumoral heterogeneity was evident in sam-
`ple CT20. Furthermore, analysis of the matched, puta-
`tively normal sample after enrichment by COLD-PCR
`revealed the TP53 exon 9 mutation. The final enriched
`abundance was just 14%, which was below the limit of
`detection via Sanger sequencing and therefore not pre-
`viously detected. Given the simultaneous collection of
`the matched samples, it is possible that the matched
`normal sample was collected from a tumor margin that
`incidentally harbored the same mutations found in the
`tumor, albeit at very low abundance.
`improved
`the
`Further
`demonstration
`of
`mutation-detection limit with COLD-PCR is pre-
`sented for clinical samples TL6, CT2, TL121, TL22, and
`TL119 (Table 1; see Figs. 4 – 8, respectively, in the on-
`line Data Supplement). Sequencing revealed a com-
`mon (29, 30 ) yet previously undetected mutation in
`sample TL22 (TP53 exon 5, c.527G⬎T) (Table 1; see
`Fig. 7 in the online Data Supplement) in addition to the
`2 previously documented mutations. Because 2 rounds
`of COLD-PCR enriched this mutation to approxi-
`mately 10%, it remained below the detection limit of
`COLD-PCR/Sanger sequencing, and thus remained
`undetected in our previous analyses (15 ). The im-
`proved detection capability enabled by the Illumina se-
`
`Clinical Chemistry 58:3 (2012) 585
`
`Personalis EX2020
`
`

`

`Downloaded from https://academic.oup.com/clinchem/article/58/3/580/5620563 by guest on 17 September 2023
`
`Fig. 3. Mutational analysis of 3 TP53 exons for sample CT20.
`Sanger sequencing (A) and conventional-PCR or COLD-PCR NGS variant-and-noise frequency plots (B). A heterozygous mutation
`(exon 5, c.527G⬎T) is detectable in both amplicons. The mutation in exon 8 (c.818G⬎A) occurring at ⬍1% and the mutation
`in exon 9 (c.925C⬎T) occurring at approximately 3% are not clear in conventional-PCR amplicons; however, COLD-PCR enriched
`the exon 8 and exon 9 mutations to approximately 53% and 76%, respectively.
`
`quencing platform allowed identification of this vari-
`ant. In the majority of the reads (99.6%), the mutated
`alleles appeared in different allelic strands, indicating
`the presence of high heterogeneity. As in sample CT20,
`the presence of the same mutations in the putatively
`normal matched sample (NL22), albeit at a lower fre-
`quency, as in the corresponding tumor sample poten-
`tially indicates the effect of the tumor margin within
`the tissue sample. For each sample we evaluated, this
`combination of COLD-PCR enrichment with Illumina
`HiSeq 2000 sequencing allowed the low-abundance
`mutations to be detected with confidence.
`
`Discussion
`
`Interest in the clinical application of NGS to individual
`cancer samples for personalized treatment/guidance,
`prognosis, and therapy follow-up is burgeoning
`(1, 31, 32 ). Clinical tumor samples, however, often
`
`586 Clinical Chemistry 58:3 (2012)
`
`come in forms that challenge the technical limits of
`NGS-based molecular diagnostics. Such samples may
`be: genetically heterogeneous tumors with subclones of
`widely differing clinical impact; infiltrating, diffuse-
`type tumor samples; suboptimally microdissected tu-
`mor samples; DNA from circulating nucleic acids, cir-
`culating cells, sputum, or other bodily fluids; and
`samples of tumor margins. The excessive concentra-
`tions of wild-type cells and DNA in such specimens
`hinder the reliable identification of low-level tumor
`mutations, which can have profound clinical implica-
`tions on disease progression, the development of me-
`tastasis, the choice of treatment, or early-detection
`strategies (1 ).
`In agreement with previous reports of studies that
`used amplicon-based NGS (4, 6, 33–35 ), we found that
`the current limit of mutation detection for NGS-based
`sequencing analysis was limited to a mutation abun-
`dance of approximately 1%–2%, primarily owing to
`
`Personalis EX2020
`
`

`

`Downloaded from https://academic.oup.com/clinchem/article/58/3/580/5620563 by guest on 17 September 2023
`
`COLD-PCR Improves Mutation Calling via NGS
`
`Table 1. Analysis of clinical samples with Illumina amplicon resequencing, after conventional PCR and after
`COLD-PCR amplification.a
`
`Observed mutational
`abundance in tumor
`
`Observed mutational abundance
`in putative normal tissue
`
`Sample Gene
`
`Exon
`
`Protein change
`
`Mutation
`
`Conventional PCR
`
`COLD-PCR
`
`Conventional PCR
`
`COLD-PCR
`
`CT2
`CT2
`CT20
`CT20
`CT20
`TL6
`TL8
`TL22
`TL22
`TL22
`TL64
`TL71
`TL96
`TL119
`TL119
`TL121
`TL121
`TL135
`
`TP53
`TP53
`TP53
`TP53
`TP53
`TP53
`TP53
`TP53
`TP53
`TP53
`TP53
`KRAS
`TP53
`KRAS
`TP53
`TP53
`TP53
`TP53
`
`5
`7
`5
`8
`9
`8
`8
`5
`5
`5
`8
`2
`7
`2
`7
`8
`7
`6
`
`p.Arg175Ser
`p.Asn247Ile
`p.Cys176Phe
`p.Arg273His
`p.Pro309Ser
`p.Cys277Phe
`p.Glu285X
`p.Val157Phe
`p.Arg158Leu
`p.Cys176Phe
`p.Arg273His
`p.Gly12Cys
`p.Arg249Ser
`p.Gly12Phe
`p.Gly244Cys
`p.Arg273His
`p.Gly245Ser
`p.Val216Met
`
`c.523C⬎A
`c.739A⬎T
`c.527G⬎T
`c.818G⬎A
`c.925C⬎T
`c.830G⬎T
`c.853G⬎T
`c.469G⬎T
`c.473G⬎T
`c.527G⬎T
`c.818G⬎A
`c.34G⬎T
`c.747G⬎T
`c.34_35GG⬎TT
`c.730G⬎T
`c.818G⬎A
`c.733G⬎A
`c.646G⬎A
`
`Not detectedb
`50%
`50%
`1%b
`3%
`2%b
`1%b
`15%
`5%b
`3%b,c
`13%
`17%
`10%
`1b
`21%
`Not detectedb
`6%
`29%
`
`28%
`62%
`87%
`53%
`76%
`22%
`29%
`54%
`25%
`10%
`69%
`84%
`49%
`67%
`61%
`18%
`58%
`75%
`
`Not detected
`Not detected
`Not detected
`Not detected
`Not detectedb
`Not detected
`Not detected
`Not detectedb
`Not detectedb
`Not detected
`Not detected
`Not detected
`Not detected
`Not detected
`Not detected
`Not detected
`Not detected
`Not detected
`
`Not detected
`Not detected
`Not detected
`Not detected
`14%
`Not detected
`Not detected
`6%
`17%
`Not detected
`Not detected
`Not detected
`Not detected
`Not detected
`Not detected
`Not detected
`Not detected
`Not detected
`
`a Observed mutational abundances are presented.
`b Indicates when a mutation could not be detected or reliably scored in the conventional PCR yet could be clearly identified in the COLD-PCR.
`c A previously undocumented mutation was detected in TL22 because of the increased sensitivity of the COLD-PCR/NGS protocol.
`
`the error-related background noise. Such errors are
`typically caused by the library-preparation steps, the
`sequencing reaction, and the processing of sequence
`calls. This error-related noise can vary across sequence
`regions because of such factors as priming efficiency,
`sequence composition, and reagent or DNA quality. As
`our evidence shows, some inherent variability among
`the regions analyzed occurs in both conventional-PCR
`and COLD-PCR amplicons. Only routine application
`and evaluation of this approach will define the typically
`observed noise and detection sensitivities in a clinical
`setting; however, if NGS is to be widely adopted in the
`clinical pathology arena, it is imperative to ensure the
`reliable detection of the low-abundance genetic vari-
`ants that have profound clinical significance. For ex-
`ample, KIF1C (kinesin family member 1C) and USP28
`(ubiquitin specific peptidase 28) mutations, which are
`clonally expanded in metastasis, preexist at levels of
`ⱕ1% in primary breast tumors (8 ), and clinical resis-
`tance– causing KIT (v-kit Hardy–Zuckerman 4 feline
`sarcoma viral oncogene homolog) (gastrointestinal
`stromal tumor) and EGFR (lung adenocarcinoma)
`
`mutations can be present in tumors at levels ⬍⬍1%
`(36, 37 ). The assessment of crucial mutations down to
`a mutation abundance of 0.01% can be important for
`accurately assessing biomarker mutations throughout
`disease progression (33, 38 ).
`Importantly, increasing sequencing depth, which
`unavoidably affects NGS throughput, is not required for
`obtaining reliable results when COLD-PCR is used. En-
`riching low-abundance mutations via COLD-PCR en-
`ables NGS mutation-detection limits as low as 0.02% with
`just 28 aligned reads (Fig. 1). Despite widespread percep-
`tions to the contrary (5), increasing the number of reads
`generally does not improve the signal-to-noise ratio or
`lower the detection limit for either COLD-PCR or con-
`ventional PCR (see Fig. 3 in the online Data Supplement).
`The enrichment obtained via COLD-PCR varies
`and depends on the specific mutation and the sequence
`context; thus, quantification of the original mutation
`frequency on the basis of only a COLD-PCR result is
`difficult. The data presented in Fig. 2 in the online Data
`Supplement does suggest, however, that once a muta-
`tion above the noise level has been identified via
`
`Clinical Chemistry 58:3 (2012) 587
`
`Personalis EX2020
`
`

`

`Downloaded from https://academic.oup.com/clinchem/article/58/3/580/5620563 by guest on 17 September 2023
`
`COLD-PCR, the frequency of the same mutation as
`identified via conventional PCR is a good approxima-
`tion of the