Analytical Validation of Ion AmpliSeq Cancer Panel 2.0Weihua Liu, Zhenyu Yan, Peng Fang, Agus Darwanto, Stephane Wong, Chad Galderisi, and Jin Li, MolecularMD Corp., Cambridge, MA and Portland, OR

Introduction Results continuedBli d d S d F h 37 bli d d FFPE l f i (l l GIST d i l b i d l ) 222 i ll d f TS2 2 f h h h

Materials and MethodsThe Ion AmpliSeq Cancer Panel 2.0 (Ion Torrent) allows for targeted sequencing of 189 regions ofinterest in 46 cancer‐related genes. We used well characterized cancer cell lines harboringclinically relevant variants as positive controls and HapMap samples NA12878 and NA19240 aswild type control samples. A total of 37 FFPE samples of a variety of cancer types were analyzedin a blinded fashion to evaluate the diagnostic sensitivity and specificity. With no gold standardavailable as a reference method to detect mutations with comparable sensitivity, we analyzed

Blinded Study: For the 37 blinded FFPE samples from various cancer types (lung, colon, GIST, endometrial, breast, ovarian and larynx), 222 raw variants called from TS2.2 were further run throughMolecularMD’s proprietary pipeline to allow for reporting 105 variants (Figure 3), 95 of which were cross confirmed by Illumina panel. Three mutations, which were evident when reviewed in IGVbut missed by Torrent Suite variant caller were confirmed as false negatives by Illumina cancer panel (Table 3) and Sanger sequencing (data not shown). The 10 variants identified by Ion, but notconfirmed by Illumina (Table 4), were further evaluated either by Sanger sequencing (for those with >10% frequency; data not shown) or a restriction endonuclease‐based fragment analysismethod described by Liu and colleagues and run on CaliperGx (ref. 3; Figure 4 contains representative data for the FGFR2 A389T variant, additional data not shown). Nine of the ten variants wereconfirmed as true positives. The variant in SMARCB1 remains to be confirmed, and is defined as false positive at this stage.

Genomic DNA (10ng) was amplified using Ion AmpliSeq™ Primer Pools (ref.1). The resultingamplicons were treated with FuPa Reagent to partially digest primer sequences and ligated toIon sequencing adapters. For sample multiplexing, libraries were barcoded using the IonXpress™ Barcode Adapters 1‐16. The ligated amplicons were purified by Ampure XP andquantified by qPCR using Ion Library Quantitation kit. Quantified libraries were qualified withCaliper GX (optional) and amplified on OneTouch system using OneTouch template preparationavailable as a reference method to detect mutations with comparable sensitivity, we analyzed

the same sets of samples with both the Ion AmpliSeq and the Illumina TruSeq Amplicon CancerPanel for comparison. These two cancer panels use distinct library preparation methods, as wellas different sequencing chemistries and analysis pipelines. They share largely overlappingregions of interest and have comparable sensitivity of minor allele detection. Variants detectedby both panels were considered as true positive variants. While most variants can be crosslid t d b t I d Ill i i t th t l d b l b t t th

Sample name Gene name Chromosome coordinate & cDNA change Amino acids Mut Freq (%) 

by Illumina

R09‐0060 KIT 4:55595550_c.2028delT p.Leu678X 22.8

R07‐0690 PDGFRA4:55152092‐55152103_c.25242535d lGACATCATGCAT

p.Ile843Asp846del(842 845DIMH> )

26.0

Table 3. False negatives in blind study of 37 FFPE samples

p p gCaliper GX (optional) and amplified on OneTouch system using OneTouch template preparationv2 kit and sequenced on PGM using Ion PGM 200 bp sequencing kit. Sequence alignment wasperformed by Torrent Suite 2.2 (ref. 2) and variants were called by TS 2.2 variant caller withmodified parameters to allow optimal sensitivity and specificity. MolecularMD proprietaryscripts were used for filtering recurring false positives, variant annotation and reporting. Inaddition to reporting hotspot variants, we report all variants at frequency >=2% with potentialimpact on protein

Figure 3. 3D projection of variants detected; genes and variants spread on x and y axes, number of variants on z axis 

validated between Ion and Illumina, variants that were only covered by one panel but not theother were confirmed by a third method.

_2535delGACATCATGCAT 2‐845DIMH>‐)R12‐0320 PIK3CA 3:178936092_c.1634A>G p.Glu545Gly(545E>G) 36.0

Table 4. Mutations in need of confirmation by a third method

Sample name Gene name Chromosome coordinate & cDNA change Amino acids Mut Freq (%) 

by Ion Torrent

Figure 4. Confirmation of FGFR2 A389T mutation

WT digested MT digested

K iti (6 ll li i ) FFPE l 8% ll li i C i ith Ill i lt

impact on protein.

Validation Flowchart

R08_0312 TP53 17:7578208_c.641A>G p.His214Arg(214H>R) 3.5R08_0312 TP53 17:7577121_c.817C>T p.Arg273Cys(273R>C) 2.8R07‐0017 FGFR2 10:123274753_c.1165G>A p.Ala389Thr(389A>T) 4.6R07‐0017 NRAS 1:115258723_c.59C>T p.Thr20Ile(20T>I) 4.8R07‐0017 SMARCB1 22:24176339_c.1103G>A p.Arg368His(368R>H) 2.5R07‐0017 PIK3CA 3:178916860_c.247T>G p.Phe83Val(83F>V) 40.5R07‐0017 PIK3CA 3:178916876 c 263G>A p Arg88Gln(88R>Q) 37 9

Mutant peak

Known positives (6 cancer cell line mix)+

Known negatives (HapMap samples)

Serial dilution

Analytical Sensitivity

FFPE sampleswith known mutations

(n=4)

Robustness in FFPE Samples

8% cell line mixand

FFPE samples (n=10)

Intra‐ and Inter‐assay

HorizonDx samples(n=4)

Mut Freq Accuracy

Blinded StudyClinical FFPE Samples

(n=37, assorted tumors)

MolecularMDdata analysis pipeline

Comparison with Illumina results

Confirmation by a third method

Diagnostic Sensitivity  3

4

5

6

7

8

9

10

R07 0017 PIK3CA 3:178916876_c.263G>A p.Arg88Gln(88R>Q) 37.9R07‐0017 KDR 4:55962485_c.2639G>A p.Arg880Gln(880R>Q) 3.1R07‐0017 MET 7:116340225_c.1087T>C p.Cys363Arg(363C>R) 3.7MMD_A2012 PIK3CA 3:178921549_c.1031T>G p.Val344Gly(344V>G) 53

Reproducibility  andMut Freq Precision

and Specificity

Analytical Sensitivity: Equimolar mixture of gDNA from six cell lines (G361, H1650, H1395,Results

Analytical Accuracy: AmpliSeq cancer panel reported mutation frequencies very close to those Mutation Spectrum:We compared the mutation spectrum from Uniformity of Coverage: For 72 distinct libraries analyzed, the average and standard deviation of relative coverage (coverage of a

0

1

2

ABL1

ATM

BRAF

CTNNB1

FBXW

7FG

FR2

FGFR3

HRA

SJAK3

KDR

KIT

KRAS

MET

NRA

SPIK3

CAPTEN RB1

RET

SMAD

4SM

ARCB

1SM

OSTK1

1TP53

Catalog# Sample Description Cell line AmpliSeq Mut FreqChromosome

di t DNA h G A i id M i f (%) R d D h

Table 2. Mutation frequency determination for HorizonDx samplesFigure 6. Relative coverage of 189 ROIs in Ion AmpliSeq Cancer Panel for 72 librariesTable 1. Analytical sensitivity defined by serial dilution of cancer cell line DNAs                             

H1975, DU145 and MDAMB231) was diluted into NA12878 gDNA with final concentrations of16%, 8% and 4% cancer cell line DNA (Table 1). The LOD for single base substitutions and indelswas determined to be 2‐5%.

expected for samples provided by HorizonDx (Table 2), with the exception of the EGFR T790Msample which had two overlapping amplicons that diluted the true mutation frequency.

different cancer types. For the five endometrial cancer samples,in addition to most frequent mutations reported in endometrialcancer, e.g. PTEN, PIK3CA, CTNNB1, TP53, FBXW7, and FGFR2(ref. 4), we also observed variations in NRAS, MET, STK11, SMOand other genes (Figure 5). One sample, R07‐0017, showedsignificant heterogeneity with minor clones containing NRAS

ROI/mean coverage of the library) for the 189 ROIs was determined (Figure 6). The data showed the coverage for the ROIs wasgenerally uniform. For each ROI, the percentage of libraries with relative coverage <0.2X showed inverse correlation with relativecoverage, indicating uniformity is higher in the high coverage regions relative to low coverage regions.

13 5

Figure 5. Variants detected in endometrial cancer

HD 103‐003‐02‐005 3.5% BRAF V600E SW48 3.6%

HD125 6.5% EGFR T790M RKO 3.65

HD129 6% EGFR L858R RKO 5.7%

HD132 4.5% EGFR L861Q RKO 4.8%

coordinate_cDNA change Gene name Amino_acids

16% 8% 4% 16% 8% 4%

11:108204681_c.7996A>G ATM p.Thr2666Ala(2666T>A) 23.8 10 5 1535 1412 1412

12:25398281_c.38G>A KRAS p.Gly13Asp(13G>D) 7.6 2031

13:49037903_c.2143A>T RB1 p.Lys715X(715K>*) 18 5.7 278 439

17:7577099_c.839G>A TP53 p.Arg280Lys(280R>K) 20.3 9 3.8 6291 5426 6524

17:7577118_c.820G>T TP53 p.Val274Phe(274V>F) 13.8 5.2 2.9 3781 3238 3693

17:7577120_c.818G>A TP53 p.Arg273His(273R>H) 16 7.5 2.8 5612 5060 6127

17:7578181_c.668C>T TP53 p.Pro223Leu(223P>L) 4.8 4230

Mutation frequency(%) Read Depth

A389T C363R N375S F83V G34V T41I H168R T170M R342X M541L P38S R173H Q298X A328T R368H R393Q R579X R880Q T20I V722I W537X

CTNNB1

FBXW7

R88Q T319X

significant heterogeneity with minor clones containing NRASand FGFR2 and other mutations with frequencies <15%, which isbelow the Sanger detection limit.

0.5

0.6

0.7

0.8

0.9

1

2

2.5

3

3.5

Mean relative coverage of libraries Percentage of libraries with relative avg_cov <0.2 (total 72ibraries) Linear (Mean relative coverage of libraries)

19:1220438-1220442_c.531_535delCAAGC STK11 p.Lys178GlyfsX86 4.4 500

4:55593464_c.1621A>C KIT p.Met541Leu(541M>L) 20 8 3.4 8487 4504 4629

4:55972974_c.1416A>T KDR p.Gln472His(472Q>H) 11.5 4.1 2583 2359

5:112175363_c.4072G>A APC p.Ala1358Thr(1358A>T) 6.5 3.6 5482 3994

7:140453136_c.1799T>A BRAF p.Val600Glu(600V>E) 18.9 10.3 4.2 7324 5466 6142

7:140481402_c.1406G>C BRAF p.Gly469Ala(469G>A) 14 7.7 3.4 5274 4250 4333

7:140481417_c.1391G>T BRAF p.Gly464Val(464G>V) 12.8 5.9 6402 5123

7:55242465

FGFR2

Jak3

KDR

Kit

Met

Nras

PIK3CA

PTEN

RB1

Smo

STK11

Determination of Phasing of Compound Mutations: The 260 flows (~100 bp) of sequencingused here allows for identifying the phasing status of compound mutations within the range of100 bp. For example, we observed PIK3CA cis and trans compound mutations in two samples(Figure 2). Figure 2A shows an IGV view of phasing of co‐existing mutations E542K (37.1%, A)and E545K (2 7% A) in the PIK3CA gene as trans mutations in a colon cancer sample while

y = ‐0.0073x + 1.7011R² = 0.9768

0

0.1

0.2

0.3

0.4

0

0.5

1

1.5

Figure 2. Identification of phasing of compound mutations

C l i

7:55242465-55242479_c.2235_2249delGGAATTAAGAGAAGC EGFR

p.Glu746_Ala750del(745-750KELREA>K) 12.9 5.6 3.8 498 496 496

7:55249071_c.2369C>T EGFR p.Thr790Met(790T>M) 6.3 3.6 7074 2951

7:55259515_c.2573T>G EGFR p.Leu858Arg(858L>R) 9.4 4.7 3.1 8744 6513 6809

9:21971153_c.205G>T CDKN2A p.Glu69X(69E>*) 19.7 4.4 870 1600

Mut Freq<15%

TP53

R080604R080914R090677R120292R070017

Analytical Reproducibility: Among 10 samples with known mutations (cell line DNA and FFPE),60 of 63 expected mutations were reproducibly detected in duplicate and triplicate runs Of

and E545K (2.7%, A) in the PIK3CA gene as trans mutations in a colon cancer sample, whileFigure 2B shows an IGV view of phasing of co‐existing mutations F83V (40.5%, G) and R88Q(37.9%, A) in the PIK3CA gene as cismutations in an endometrial cancer sample.

R f

ROI

Conclusions• Our validation demonstrates that the Ion AmpliSeq Cancer Panel combined with

MolecularMD’s proprietary analysis pipeline provides a robust and accurate test to profileclinically relevant gene mutations with analytical sensitivity of 2‐5% minor allele frequency.

• In addition to reporting hotspot mutations, we also report other variants that would impact

60 of 63 expected mutations were reproducibly detected in duplicate and triplicate runs. Ofthe three non‐reproducible variants, two were low‐level (frequencies close to 2%), and onewas a high‐frequency variant in a homopolymer region. The precision of variant frequencydetection was determined using 30 mutations in 6 samples with triplicates or duplicates(Figure 1). In general, the CV is between 0‐20%, showing inverse correlation with variantfrequency.

A References1. Life Technologies; Ion AmpliSeq™ Library Kit 2.0 manual2. Life Technologies; Torrent Suite 2.2 manual3. Liu WH, Kaur M, Makrigiorgos GM. Detection of hotspot mutations and polymorphisms

using an enhanced PCR‐RFLP approach. Hum. Mutat. 2003; 21(5):535‐41protein (all except synonymous and intronic mutations). The diagnostic sensitivity andspecificity calculated based on a blinded study of 37 FFPE samples is 97% and 99%, respectively.

• In this study, we developed a unique cross‐validation strategy using two different NGSplatforms. We also established a rapid and accurate restriction endonuclease based fragmentanalysis method for confirming variants at frequencies below 10%.O lid ti t d 102 i t ti i 23 ith 7 i d l 7 d 88

Figure 1. Assay precision for detection of mutation frequency

Variant Frequency (%)

g pp4. Bamford S, Dawson E, Forbes S, Clements J, Pettett R, Dogan A, Flanagan A, Teague J,

Futreal PA, Stratton MR, Wooster R. The COSMIC (Catalogue of Somatic Mutations inCancer) database and website. Br J Cancer. 2004: 91;355‐8

• Our validation study covers 102 unique mutations in 23 genes, with 7 indels, 7 nonsense and 88missense mutations.

• Ion AmpliSeq paired with the MolecularMD proprietary pipeline has excellent sensitivity forsingle base substitutions with only one mutation (PIK3CA E545G) missed due to thepseudogene interference. This can be resolved by manual review of Bam files in IGV for thisgenomic locus in the future.

BFor Further InformationPlease contact info@molecularmd.com or visit www.molecularmd.com.

genomic locus in the future.• One weakness of this cancer panel in its current version is the inability to detect single base

indels in homopolymer regions. However, it tolerates other indels fairly well; indels rangingfrom 1‐21 bp were successfully detected in this study.

• Ion AmpliSeq is a cancer panel well‐suited for small clinical samples with only 10ng DNA input.There was a zero failure rate for 41 FFPE samples that we tested.

10

• We are currently exploring methods to improve the variant calling for indels in homopolymerregions by using TS 3.0 and other pipelines.

0