Utility of different massive parallel sequencing platforms for mutation profiling in clinical samples and identification of pitfalls using FFPE tissue.

In the growing field of personalised medicine, the analysis of numerous potential targets is becoming a challenge in terms of work load, tissue availability, as well as costs. The molecular analysis of non-small cell lung cancer (NSCLC) has shifted from the analysis of the epidermal growth factor receptor (EGFR) mutation status to the analysis of different gene regions, including resistance mutations or translocations. Massive parallel sequencing (MPS) allows rapid comprehensive mutation testing in routine molecular pathological diagnostics even on small formalin-fixed, paraffin‑embedded (FFPE) biopsies. In this study, we compared and evaluated currently used MPS platforms for their application in routine pathological diagnostics. We initiated a first round‑robin testing of 30 cases diagnosed with NSCLC and a known EGFR gene mutation status. In this study, three pathology institutes from Germany received FFPE tumour sections that had been individually processed. Fragment libraries were prepared by targeted multiplex PCR using institution‑specific gene panels. Sequencing was carried out using three MPS systems: MiSeq™, GS Junior and PGM Ion Torrent™. In two institutes, data analysis was performed with the platform-specific software and the Integrative Genomics Viewer. In one institute, data analysis was carried out using an in-house software system. Of 30 samples, 26 were analysed by all institutes. Concerning the EGFR mutation status, concordance was found in 26 out of 26 samples. The analysis of a few samples failed due to poor DNA quality in alternating institutes. We found 100% concordance when comparing the results of the EGFR mutation status. A total of 38 additional mutations were identified in the 26 samples. In two samples, minor variants were found which could not be confirmed by qPCR. Other characteristic variants were identified as fixation artefacts by reanalyzing the respective sample by Sanger sequencing. Overall, the results of this study demonstrated good concordance in the detection of mutations using different MPS platforms. The failure with samples can be traced back to different DNA extraction systems and DNA quality. Unknown or ambiguous variations (transitions) need verification with another method, such as qPCR or Sanger sequencing.

Introduction

In the growing field of personalised medicine, the increasing number of molecular targets for individualised therapies requires the analysis of numerous, potential genetic alterations, which is becoming a challenge in terms of workload, tissue availability, as well as costs (1). For non-small cell lung cancer (NSCLC), molecular analysis has shifted from the analysis of the epidermal growth factor receptor (EGFR) mutation status to the analysis of additional gene target regions, including resistance mutations and gene fusion events (2).

Taking these developments into account, massive parallel sequencing (MPS) has come into focus, as it allows rapid, comprehensive and cost-effective mutation testing for routine molecular pathological diagnostics, even on small formalin-fixed, paraffin-embedded (FFPE) biopsies (3–6). However, the implementation of MPS platforms into routine diagnostics raises questions about feasibility, sensitivity and specificity, as the results of mutation testing are the basis for therapeutic decision making (1,7). The ever-increasing pace of MPS adoption presents enormous challenges, in terms of data processing, storage, management and interpretation, as well as sequencing quality control, which impede the translation of research into clinical practice (8,9).

Additionally, the preanalytical steps are important to consider: the manual macrodissection of selected tumour areas has become a standard procedure in molecular pathology and is a powerful tool to reduce false negative results resulting from wild-type contamination (10). Selecting the right tumour area influences not only the result of the analysis, but also the allele frequency, the value of which is pivotal when reporting diagnostic findings (11). Automated DNA extraction systems are helpful in a routine laboratory with respect to expenditure of time, sample tracking and reproducible sample quality. In addition, an accurate and reliable DNA quantification system is necessary for good and constant MPS performance (12).

In the present study, we compared three different MPS platforms: PGM Ion Torrent™ from Life Technologies™, MiSeq™ from Illumina® and GS Junior from Roche. We used lung cancer samples, obtained from the clinical setting, with a known EGFR and KRAS mutation status. Samples included large tumour resections, as well as small fine needle biopsies. In our comparison, three different multiplex primer panels, tailored to the needs of the respective sequencing platforms were used in the participating institutes, mirroring the individual approaches that may be used for routine testing.

Materials and methods

Samples

A total of 30 tumour samples was collected from 2010 to 2013. All samples were lung adenocarcinomas and each institute contributed 10 samples. Tumours were diagnosed by experienced pathologists and the tumour content was determined by the visual inspection of hematoxylin and eosin (H&E)-stained corresponding sections. The mutation status of the samples was determined previously in routine molecular diagnostics in each institute using conventional methods.

DNA isolation

All tissue specimens were fixed in neutral-buffered formalin prior to paraffin embedding (FFPE samples). Tumour areas were marked by a pathologist on an H&E-stained slide and DNA was extracted from corresponding unstained 10-µm-thick slides by manual macrodissection. Following treatment with proteinase K, the DNA was isolated by either automated or manual extraction: BioRobot M48 (institute A), the QIAamp DNA FFPE Tissue kit (institute B), QIASymphony SP (institute C) (all from Qiagen, Hilden, Germany) or the Maxwell 16 Research system (institute C; Promega, Madison, WI, USA) following the manufacturer's instructions.

DNA quality and quantity

The quality and quantity of the isolated DNA samples were assessed by agarose gel electrophoresis and measured fluorimetrically using the Qubit® HS DNA assay (Life Technologies, Darmstadt, Germany) in institute A. The quantity of the isolated DNA was measured spectrophotometrically using the NanoDrop 2000c spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA) in institute B. In institute C, the DNA content was measured fluorimetrically using the Qubit HS DNA assay (Life Technologies) and using a qPCR-based method (RNaseP Detection system; Life Technologies).

Massive parallel sequencing

Illumina® MiSeq™ platform

MiSeq (Illumina, San Diego, CA, USA) was used in institute A. The custom-made lung cancer panel consisted of 102 amplicons for the detection of hotspot mutations in 14 lung cancer-related genes. A full list of the covered amplicons is provided in Table I. Isolated DNA (20 ng) was amplified with 2 customised Ion AmpliSeq™ Primer Pools for 15 sec at 99°C and 4 min at 60°C for 29 cycles, with an initial denaturating step at 99°C for 2 min. PCR products from the same patient were pooled following treatment with FuPa reagent. Following purification with Agencourt AMPure XP (Beckman Coulter, Brea, CA, USA), the PCR products were incubated with NEXTflex™ DNA Adenylation Mix (Bioo Scientific Corp., Austin, TX, USA). Adapters were supplied by NEXTflex™ DNA Barcodes (Bioo Scientific Corp.). After the bead-mediated size selection, NEXTflex™ PCR Master Mix (Bioo Scientific Corp.) was used for the final PCR amplification at 98°C for 15 sec and 60°C for 1 min for 10 cycles, with an initial denaturating step at 98°C for 2 min. Library products were quantified using a Qubit® 2.0 Fluorometer (Qubit® dsDNA HS kit; Life Technologies), diluted and pooled in equal amounts. A total of 6–8 pM was spiked with 5% PhiX DNA and sequenced using the MiSeq™ reagent kit V2 (300 cycles) (both from Illumina). Data were exported as FASTQ files.

Table I

Overview of the institute-specific gene panels.

Chromosome	From (hg19)	To (hg19)	Gene name	Exon
Custom panel Heidelberg
chr1	27056234	27056365	ARID1A	2
chr1	27057662	27057775	ARID1A	3
chr1	27057875	27058001	ARID1A	3
chr1	27092899	27093023	ARID1A	10
chr1	27094337	27094460	ARID1A	11
chr1	27099336	27099464	ARID1A	14
chr1	27100275	27100411	ARID1A	17
chr1	27105906	27106030	ARID1A	20
chr1	27106449	27106570	ARID1A	20
chr1	27106750	27106883	ARID1A	20
chr1	115256484	115256587	NRAS	3
chr1	115258676	115258805	NRAS	2
chr1	150549826	150549952	MCL-1	3
chr1	150551531	150551670	MCL-1	1
chr2	178098765	178098890	NFE2L2	2
chr3	41266029	41266147	CTNNB1	3
chr3	41266893	41267010	CTNNB1	5
chr3	41275089	41275211	CTNNB1	9
chr3	178916892	178917000	PIK3CA	2
chr3	178921523	178921633	PIK3CA	5
chr3	178928050	178928160	PIK3CA	8
chr3	178936022	178936106	PIK3CA	10
chr3	178938830	178938960	PIK3CA	14
chr3	178952038	178952157	PIK3CA	21
chr3	181430178	181430283	SOX2	1
chr3	181430516	181430649	SOX2	1
chr4	1803550	1803636	FGFR3	7
chr4	1808277	1808409	FGFR3	16
chr4	55131108	55131222	PDGFRA	5
chr4	55139749	55139881	PDGFRA	10
chr4	55140692	55140818	PDGFRA	11
chr4	55141036	55141156	PDGFRA	12
chr4	55152001	55152128	PDGFRA	18
chr4	55156632	55156764	PDGFRA	22
chr4	55592107	55592203	KIT	9
chr4	55593595	55593684	KIT	11
chr4	153245407	153245522	FBXW7	11
chr4	153247237	153247369	FBXW7	10
chr4	153249405	153249530	FBXW7	9
chr5	1264501	1264634	TERT	11
chr5	1293392	1293528	TERT	2
chr6	66115100	66115214	EYS	7
chr6	66204680	66204810	EYS	5
chr7	55241602	55241732	EGFR	18
chr7	55242411	55242544	EGFR	19
chr7	55248974	55249100	EGFR	20
chr7	55259416	55259546	EGFR	21
chr7	92300724	92300853	CDK6	5
chr7	92403995	92404124	CDK6	3
chr7	116411944	116412066	MET	14
chr7	116417426	116417508	MET	16
chr7	140453110	140453232	BRAF	15
chr7	140481387	140481511	BRAF	11
chr8	38275705	38275835	FGFR1	10
chr8	38282107	38282241	FGFR1	7
chr8	128751156	128751293	MYC	2
chr8	128752956	128753086	MYC	3
chr9	5069993	5070100	JAK2	12
chr9	5073678	5073788	JAK2	14
chr9	5126715	5126797	JAK2	25
chr9	21970912	21971032	CDKNA2	2
chr9	21971086	21971218	CDKNA2	2
chr9	21974672	21974792	CDKNA2	1
chr9	139401722	139401834	NOTCH1	22
chr9	139404170	139404306	NOTCH1	18
chr9	139412260	139412400	NOTCH1	8
chr9	139413034	139413159	NOTCH1	6
chr10	89624207	89624322	PTEN	1
chr10	89685258	89685374	PTEN	3
chr10	89692864	89692987	PTEN	5
chr10	89711806	89711936	PTEN	6
chr10	89717622	89717747	PTEN	7
chr10	89720778	89720902	PTEN	8
chr10	123256020	123256129	FGFR2	13
chr10	123279495	123279622	FGFR2	7
chr11	533800	533929	HRAS	3
chr11	534220	534349	HRAS	2
chr11	69456096	69456216	CCND1	1
chr11	69458624	69458747	CCND1	3
chr11	119103162	119103275	CBL	2
chr11	119148912	119149006	CBL	8
chr11	119149215	119149290	CBL	9
chr12	25380249	25380348	KRAS	3
chr12	25398183	25398310	KRAS	2
chr12	69210596	69210679	MDM2	4
chr12	69233038	69233165	MDM2	11
chr13	48881433	48881526	RB1	2
chr13	48916793	48916902	RB1	3
chr13	48923124	48923208	RB1	6
chr13	48951050	48951160	RB1	13
chr13	48954320	48954437	RB1	16
chr13	48955427	48955539	RB1	17
chr13	49027105	49027191	RB1	18
chr13	49033834	49033935	RB1	20
chr13	49037844	49037955	RB1	21
chr13	49039144	49039221	RB1	22
chr13	49039304	49039410	RB1	23
chr14	36987081	36987213	NKX-2.1	2
chr14	36988227	36988351	NKX-2.1	1
chr14	105246470	105246589	AKT1	3
chr17	7573886	7574019	TP53	10
chr17	7576836	7576950	TP53	9
chr17	7577028	7577157	TP53	8
chr17	7577492	7577629	TP53	7
chr17	7578180	7578289	TP53	6
chr17	7578425	7578555	TP53	5
chr17	7579278	7579397	TP53	4
chr17	7579454	7579566	TP53	4
chr17	37880169	37880287	ERBB2	19
chr17	37880958	37881089	ERBB2	20
chr18	48581196	48581323	SMAD4	5
chr18	48584702	48584826	SMAD4	7
chr18	48591813	48591934	SMAD4	9
chr18	48604680	48604811	SMAD4	12
chr19	1206977	1207113	STK11	1
chr19	1218379	1218488	STK11	2
chr19	1220390	1220504	STK11	4
chr19	1220594	1220684	STK11	5
chr19	1221205	1221340	STK11	6
chr19	1223020	1223155	STK11	8
chr19	10599879	10600011	KEAP1	5
chr19	10600372	10600496	KEAP1	4
chr19	10602263	10602390	KEAP1	3
chr19	10602579	10602708	KEAP1	3
chr19	10602796	10602912	KEAP1	3
chr19	10610088	10610218	KEAP1	2
chr19	10610289	10610416	KEAP1	2
chr19	10610465	10610599	KEAP1	2
chr19	11094812	11094945	SMARCA4	2
chr19	11136088	11136220	SMARCA4	22
chr19	11138426	11138556	SMARCA4	23
chr19	11141448	11141561	SMARCA4	25
chr19	11144042	11144179	SMARCA4	26
chr19	30308024	30308156	CCNE1	5
chr19	30313134	30313262	CCNE1	10
chrX	47028755	47028888	RBM10	3
chrX	47034396	47034523	RBM10	5
chrX	63411268	63411399	FAM123B/AMER1	1
chrX	63412836	63412964	FAM123B/AMER1	1
Custom panel Cologne
chr1	115256352	115256453	NRAS	3
chr1	115256453	115256550	NRAS	3
chr1	115256550	115256672	NRAS	3
chr1	115258676	115258798	NRAS	2
chr1	162688829	162688951	DDR2	3
chr1	162722872	162722995	DDR2	4
chr1	162724359	162724466	DDR2	5
chr1	162724466	162724586	DDR2	5
chr1	162724586	162724687	DDR2	5
chr1	162724850	162724967	DDR2	6
chr1	162724967	162725094	DDR2	6
chr1	162725447	162725572	DDR2	7
chr1	162729566	162729694	DDR2	8
chr1	162729681	162729782	DDR2	8
chr1	162730973	162731107	DDR2	9
chr1	162731107	162731197	DDR2	9
chr1	162731197	162731276	DDR2	9
chr1	162735765	162735879	DDR2	10
chr1	162736904	162737029	DDR2	11
chr1	162737029	162737154	DDR2	11
chr1	162740090	162740201	DDR2	12
chr1	162740201	162740327	DDR2	12
chr1	162741756	162741887	DDR2	13
chr1	162741887	162742002	DDR2	13
chr1	162742002	162742088	DDR2	13
chr1	162743204	162743301	DDR2	14
chr1	162743301	162743421	DDR2	14
chr1	162745384	162745513	DDR2	15
chr1	162745513	162745634	DDR2	15
chr1	162745915	162746038	DDR2	16
chr1	162746038	162746162	DDR2	16
chr1	162748317	162748432	DDR2	17
chr1	162748432	162748519	DDR2	17
chr1	162749866	162749977	DDR2	18
chr1	162749977	162750066	DDR2	18
chr2	29432650	29432776	ALK	25
chr2	29436843	29436974	ALK	24
chr2	29443565	29443688	ALK	23
chr2	29443688	29443772	ALK	23
chr2	29445200	29445332	ALK	22
chr2	29445369	29445489	ALK	21
chr3	41266072	41266193	CTNNB1	3
chr3	178935940	178936023	PIK3CA	9
chr3	178936023	178936105	PIK3CA	9
chr3	178936092	178936180	PIK3CA	9
chr3	178951824	178951942	PIK3CA	20
chr3	178951942	178952063	PIK3CA	20
chr3	178952063	178952155	PIK3CA	20
chr7	55241596	55241679	EGFR	18
chr7	55241679	55241800	EGFR	18
chr7	55242411	55242539	EGFR	19
chr7	55248984	55249117	EGFR	20
chr7	55249117	55249200	EGFR	20
chr7	55259367	55259486	EGFR	21
chr7	55259484	55259567	EGFR	21
chr7	116411701	116411801	cMET	intron 13/14
chr7	116411801	116411909	cMET	14
chr7	116411894	116411998	cMET	intron 13/14
chr7	116411998	116412072	cMET	14
chr7	140453023	140453099	BRAF	15
chr7	140453099	140453224	BRAF	15
chr7	140481297	140481387	BRAF	11
chr7	140481387	140481511	BRAF	11
chr10	89624207	89624322	PTEN	1
chr10	89653745	89653817	PTEN	2
chr10	89653816	89653930	PTEN	2
chr10	89685258	89685374	PTEN	3
chr10	89690819	89690917	PTEN	4
chr10	89692713	89692819	PTEN	5
chr10	89692819	89692920	PTEN	5
chr10	89692920	89693032	PTEN	5
chr10	89711802	89711928	PTEN	6
chr10	89711917	89712018	PTEN	6
chr10	89717580	89717695	PTEN	7
chr10	89717694	89717792	PTEN	7
chr10	89720692	89720768	PTEN	8
chr10	89720769	89720842	PTEN	8
chr10	89724948	89725061	PTEN	9
chr10	89725058	89725147	PTEN	9
chr10	89725207	89725320	PTEN	9
chr12	25380167	25380240	KRAS	3
chr12	25380240	25380357	KRAS	3
chr12	25398183	25398304	KRAS	2
chr12	25398304	25398379	KRAS	2
chr14	105246406	105246502	AKT1	4
chr14	105246500	105246583	AKT1	4
chr15	66727356	66727487	MAP2K1	2
chr15	66727487	66727602	MAP2K1	2
chr17	7577017	7577142	TP53	8
chr17	7577140	7577233	TP53	8
chr17	7577392	7577509	TP53	7
chr17	7577508	7577611	TP53	7
chr17	7578141	7578234	TP53	6
chr17	7578234	7578362	TP53	6
chr17	7578310	7578425	TP53	5
chr17	7578425	7578555	TP53	5
chr17	7579278	7579385	TP53	4
chr17	7579385	7579502	TP53	4
chr17	7579502	7579590	TP53	4
chr17	37880155	37880283	HER2	19
chr17	37880960	37881074	HER2	20
chr17	37881074	37881206	HER2	20

GS Junior platform

GS Junior (Roche, Basel, Switzerland) was used in institute B. Genomic DNA (10–250 ng) was used for the amplification of EGFR exons 18–21 in a single multiplex reaction using the EGFR 18–21 MASTR assay and the 454 MID kit 1–8 (both from Multiplicom N.V., Niel, Belgium) according to the manufacturer's instructions. Libraries were purified, quantified, diluted to a final concentration of 1×106 molecules, multiplexed, clonally amplified by emulsion PCR and sequenced on the GS Junior (Roche) following the manufacturer's instructions. Amplicon libraries were sequenced in two runs on 454 GS Junior with 15 samples each.

PGM Ion Torrent platform

PGM Ion Torrent (Life Technologies) was used in institute C. For library preparation, the multiplex PCR-based Ion Torrent™ AmpliSeq™ technology (Life Technologies) with a custom-made lung cancer panel was used. The panel consisted of 139 primer pairs for the detection of hotspot mutations in 41 lung cancer-related genes. A full list of the covered amplicons is provided in Table I. Amplicon library preparation was performed with the Ion AmpliSeq™ Library kit v2.0 using approximately 10 ng of DNA as advised by the manufacturer. The PCR cycling conditions were as follows: initial denaturation: 99°C for 2 min, cycling: 21 cycles of 99°C, 15 sec and 60°C, 4 min. PCR products were partially digested using FuPa reagent as instructed, followed by the ligation of barcoded sequencing adapters (Ion Xpress Barcode Adapters 1–16 kit; Life Technologies). The final library was purified using Agencourt AMPure XP magnetic beads (Beckman Coulter) and quantified using qPCR (Ion Library Quantitation kit) on a StepOne qPCR machine (both from Life Technologies). The individual libraries were diluted to a final concentration of 100 pM and eight to ten libraries were pooled and processed to library amplification on Ion Spheres using an Ion PGM™ Template OT2 200 kit. Unenriched libraries were quality-controlled using Ion Sphere quality control measurement on a Qubit instrument. Following library enrichment (Ion OneTouch ES), the library was processed for sequencing using the Ion Torrent 200 bp sequencing v2 chemistry and the barcoded libraries were loaded onto a single 318 chip.

Data analysis

Illumina MiSeq platform

The FASTQ files were aligned against reference NCBI build 37 (hg19) and annotated using a modified version of a previously described method (13). The resulting BAM files were visualized using the Integrative Genomics Viewer (IGV; http://www.broadinstitute.org/igv/). Called variants were then imported into a FileMaker (FileMaker GmbH, Germany) database for further analysis, annotation and reporting. A 5% cut-off for variant calls was used and the results were only interpreted if the coverage was >100x.

Alignment against reference NCBI build 37 (hg19) and variant calling was carried out using AVA software (Roche). Thresholds for variant calling were set to a minimum allele frequency of 5% with a coverage of at least 100x. All variants were visually inspected using the AVA software (Roche). Annotation of variants was done according to the HGVS nomenclature.

Raw data processing, sequence generation and alignment to the reference hg19 genome were conducted using the Torrent Suite software (version 4.0; Life Technologies). Variants were identified using the variant caller plug-in package. For hotspot mutations, a minimum allele frequency of 3% was set and for novel mutations, at least a 5% allele frequency was set as the cut-off level (with coverage >100x). Annotation of variants was performed with the CLC genomics workbench (version 6.5) followed by the visual inspection of putative mutations using the IGV browser.

Results

DNA concentration

DNA extraction from the 30 NSCLC samples was carried out with three different DNA extraction systems and the DNA concentration was measured using individual methods as described above. Table II summarises the resulting DNA concentrations. While the DNA concentration ranges measured with the Qubit 2.0 fluorometer in institutes A and C were comparable, the values measured using the NanoDrop 2000c spectrophotometer in institute B were generally higher due to the different principles of measurement. We observed a 1.4- to 856-fold and a 3.9- to 156-fold difference in the concentrations of institute B compared with the concentration values in institutes A and C, respectively with average differences of 133- and 30-fold. Particulary in samples with concentrations below 10 ng/µl, the measurements showed high deviations (Table II). Although only minimal amounts of DNA were measured in some samples from institutes A and C, the maximum volume possible was used for the massive parallel analysis for comparative purposes.

Table II

DNA concentration.

Sample no.	Institute A (ng/µl)	Institute B (ng/µl)	Institute C (ng/µl)
1	31	362.9	20.8
2	2.9	7.84	0.85
3	3.32	109.16	7.81
4	0.1	4.03	0.41
5	12.8	186.49	11.7
6	7.5	14.92	1.15
7	16.6	374.76	4.55
8	2.44	24.24	1.48
9	26.6	504.4	44.8
10	0.1	3.61	<0.5
11	10.3	26.1	3.42
12	5.7	266.83	2.36
13	8.06	11.58	2.94
14	4.56	21.62	1.18
15	2.06	28.72	4.94
16	2.7	15.64	1.25
17	1.29	25.68	3.58
18	3.78	17.32	1.99
19	0.1	19.24	4.3
20	8.8	204.08	1.3
21	0.1	1.3	0.1
22	18.4	470.92	12.2
23	0.83	204.61	6.08
24	0.97	52.5	5.34
25	0.16	103.01	2.85
26	0.3	103.01	8.52
27	0.24	60.02	1.31
28	0.1	85.57	1.06
29	0.1	47.74	0.7
30	0.1	56.77	4.2

DNA extraction from 30 non-small cell lung cancer (NSCLC) samples was carried out with three different DNA extraction systems from Qiagen: BioRobot M48, QIA Symphony SP as well as manual extraction. After the extraction, concentration was measured with the Qubit 2.0 fluorometer in institutes A and C, or with the NanoDrop 2000c spectrophotometer in institute B.

Platform comparison summary

The median amplicon sizes for all platforms ranged from 125–345 bp, allowing the amplification of target sequences from degraded DNA obtained from FFPE material (Table III). The number of analysed amplicons ranged from 4 up to 137. Depending on the platform used, the number of samples analysed in one single run varied from 8 up to 48. The maximum number of median reads per sample was approximately 500.000 on the PGM followed by approximately 350.000 on the MiSeq and 5007 reads on the GS Junior. In general, the read coverage for each amplicon was considered to be sufficient for each sample with median values of between 1290 and 7409.

Table III

Sequencing statistics.

	MiSeq™	PGM Ion Torrent™	GS Junior
No. of Amplicons	102	137	4
Median amplicon size	150 bp	125 bp	345 bp
Samples/run	48	8–10	15
Median reads/sample	~350.000	~ 500.000	5007
Median coverage/amplicon	7409x	2500x	1290x

Overview of the different massive parallel sequencing (MPS) platforms. bp, base pairs.

Influence of macrodissection

Manual macrodissection of marked regions on unstained sections was performed to enrich for tumour cells in the extraction. Depending on the strictness of separating tumour cells from normal cells, the resulting allele frequencies for mutant vs. wild-type alleles can vary. This is of particular importance when analysing samples with low tumour cell content or when allele frequencies are expected to be low. Depending on the size of the marked area, the proportion of tumour and normal cells and therewith the allele frequencies could differ in the same sample. This is exemplified in Fig. 1; the area used for DNA extraction was larger in institute B than in institute A. Thus, the corresponding allele frequencies for the EGFR mutation of this sample were determined to be 14 and 54%, respectively.

Figure 1

Macrodissection. Tumor cells on H&E-stained slides were marked by experienced pathologists. Manual macrodissection of marked regions in (A) resulted in an AF of 14% whereas manual macrodissection of tissue in (B) resulted in 54% AF. H&E, hematoxylin and eosin; AF, allele frequency.

Detection of EGFR mutations

Concerning the expected EGFR mutation status, we found concordance in 26 out of 26 samples (Table IV). In all samples, the EGFR mutation status was correctly identified by all participants using a 5% threshold for allele frequencies and at least a coverage rate of 100 (Table IV). The EGFR mutation status of our sample cohort was comprised of 12 single point mutations, 9 complex exon 19 deletions/insertions and 11 wild-type samples. In three cases, two EGFR mutations were present (Table IV, nos. 1, 20 and 21).

Table IV

EGFR mutation status.

Case	Expected result	A	B	C	Tumor cell content	A	B	C	A AF%	B AF%	C AF%
1	p.G719A	√	√	√	50	13936	4917	3001	20	15	24
1	p.V834L	√	√	√	50	9112	4917	4829	17	18	22
2	p.L838R	√	√	√	80	1430	10143	5885	17	17	17
3	p.E746_A750del	√	√	√	60	10584	3379	9216	79	45	44
4	p.E746_A750del	√	√	√	10	1102	512	5116	23	18	22
5	wt	√	√	√	90	wt	wt	wt
6	wt	√	√	√	70	wt	wt	wt
7	wt	√	√	√	60	wt	wt	wt
8	wt	√	√	√	30	wt	wt	wt
9	wt	√	√	√	30	wt	wt	wt
10	–	n.a.	n.a.	n.a.	80	n.a.	n.a.	n.a.
11	p.E746_A750del	√	√	√	60	9562	5020	1947	67	60	49
12	p.L858R	√ + p.T790M	√	√	50	29429/34779	8291	2820	28/1.03	21	12
13	p.E746_A750del	√	√	√	40	9936	11132	2820	31	29	25
14	p.L858R	√	√	√	30	35355	6911	5693	36	33	13
15	p.L858R	√	√	√	50	14143	1381	3407	31	41	31
16	p.E746_A750del	√	√	√	70	11546	1472	1975	34	51	33
17	p.L858R	n.a.	n.a.	√	70	n.a.	n.a.	3336	n.a.	n.a.	20
18	p.E746_A750del	√	√	√	n.d.	4179	406	1521	54	14	10
19	p.L747_A751delinsP	√	n.a.	√	70	7445	n.a.	1585	75	n.a.	54
20	p.L747_P753delinsS	√	√	√	80	8010	5816	4221	75	59	49
20	p.A755D	√	√	√	80	7297	5816	4221	74	59	59
21	p.E709A	√	√	√	80	716	3273	4662	23	24	22
21	p.G719S	√	√	√	80	2102	3273	4640	9	24	20
22	p.E746_A750del	√	√	√	50	8391	33615	1968	62	56	50
23	p.L858R	√ + p.T790M	√	√	30	20413/10246	11389	1994	27/1.42	20	18
24	p.L858R	√	√	√	30	9794	16509	1714	34	26	30
25	wt	n.a.	√	√	60	wt	wt	wt
26	wt	√	√	√	60	wt	wt	wt
27	wt	√	√	√	70	wt	wt	wt
28	wt	√	√	√	60	wt	wt	wt
29	wt	√	√	√	70	wt	wt	wt
30	wt	√	√	√	n.d.	wt	wt	wt

Concerning the epidermal growth factor receptor (EGFR) mutation status, we found concordance in 26/26 samples. The mutation status was analysed previously with conventional methods. Institute A found two resistance mutations in samples 12 and 23. AF%, allele frequency; hook, concordant EGFR result; n.a., not analysable, n.d., not determined; wt, wild-type.

In only one case (no. 10), parallel sequencing was unsuccessful due to either failed PCR amplification or insufficient coverage. This case, which could not be analysed by conventional methods previously, was included intentionally to test the limits of parallel sequencing. In three cases (nos. 17, 19 and 25) with limited tumour material, parallel sequencing failed depending on the DNA extraction method. Institute A, using the BioRobot M48, did not get any sequencing results for samples 17 and 25, which was due to high salt concentrations that inhibited the multiplex PCR. Samples 17 and 19 could not be analysed by institute B due to the high degradation of samples and failed amplification.

In 2 out of the 30 samples, minor p.T790M clones of the EGFR gene were detected (nos. 12 and 23) by institute A. The underlying mutation was found with 1.03 and 1.42% allele frequency with a coverage of 34779 and 10246, respectively and balanced forward and reverse reads (Fig. 3). A qPCR system (therascreen® EGFR RGQ PCR kit; Qiagen) with a detection limit of 1% allele frequency was used for the verification of originally extracted DNA samples (BioRobot M48; Qiagen), newly extracted DNA samples (Maxwell 16 Research system; Promega) from both samples as well as the corresponding DNA samples from institutes B and C. The minor variants could not be confirmed in any of the DNA samples. Thus, the EGFR p.T790M found in the first analysis most likely constitutes a fixation artefact.

Figure 3

Minor variants. Minor variants could be detected in two out of 30 samples in institute A (nos. 12 and 23). The resistance mutation p.T790M in EGFR was found with 1.03 and 1.42% AF with a coverage of 34779 and 10246. AF, allele frequency; cov, coverage.

Additional mutations and fixation artefacts

Besides the EGFR mutations, additional variants were identified by institutes A and C using more comprehensive primer sets (Table V). Concordance was found in 15 additional variants, whereas 16 variants could not be confirmed due to the missing inclusion of the respective primers in the individual panels. Seven samples (nos. 1, 4, 8, 13, 20, 24 and 30) showed no additional mutations, which was confirmed by both institutes.

Table V

Additional variations.

Case	Gene	Nucleotide change	AA change	AF A (%)	AF C (%)
1	–	–	–	–	–
2	TP53	c.469G>T	p.V157F	80	79
3	TP53	c.637C>T	p.R213*	79	34
4	–	–	–	–	–
5	NKX2.1	c.515A>C	p.Q172P	n.i.	23
	RB1	c.2267delA	p.Y756fs	n.i.	91
	TP53	c.733G>T	p.G245C	87	91
6	TP53	c.641A>G	p.H214R	33	23
7	TP53	c.830G>T	p.C277F	23	44
8	–	–	–	–	–
9	KRAS	c.35G>A	p.G12D	2	5
10				n.a.	n.a.
11	TP53	c.1073C>T	p.P295S	1	5
	JAK3	c.2164G>A	p.V722I	n.i.	37
12	TP53	c.610G>T	p.E204*	7	25
13	–	–	–	–	–
14	ATM	c.2572T>C	p.F858L	n.i.	66
15	TP53	c.913A>T	p.K305	26	20
	KIT	c.1621A>C	p.M541L	n.i.	57
16	SMO	c.979G>A	p.A327T	n.i.	45
17	–	–	–	n.a.	–
18	TP53	c.530C>G	p.P177R	26	8
19	TP53	c.725G>A	p.C242Y	81	34
	TP53	c.555C>G	p.S185R	73	n.i.
	KIT	c.1621A>C	p.M541L	n.i.	78
	PIK3CA	c.1633G>A	p.E545K	44	4
20	–	–	–	–	–
21	PIK3CA	c.1624G>A	p.E542K	18	17
22	CTNNB1	c.98C>G	p.S33C	33	31
23	NOTCH1	c.3604C>T	p.P1202S	n.i.	5
	RBM10	c.79delG	p.G27fs	n.i.	17
24	–	–	–	–	–
25	SMARCA4	c.3634G>A	p.E1212K	n.i./n.a.	5
	KRAS	c.35G>A	p.G12D	n.a.	10
26	KRAS	c.35G>A	p.G12D	26	29
27	KEAP1	c.1426G>T	p.G476W	n.i.	45
	MAP2K1	c.171G>T	p.K57N	45	n.i.
28	CDK6	c.584G>T	p.S195I	n.i.	13
	CDKN2A	c.253C>T	p.Q85	n.i.	6
29	HRAS	c.59C>T	p.T20I	n.i.	5
	BRAF	c.1406G>A	p.G469E	FA	–
	NRAS	c.178G>A	p.G60R	FA	–
	PIK3CA	c.1633G>A	p.E545K	FA	–
30	–	–	–	–	–

Besides the epidermal growth factor receptor (EGFR) mutations, additional mutations could be identified with the extended primer sets used in institutes A and C. Concordance was found in 15 additional variations whereas 16 variants could not be confirmed by the other institute due to missing primer panel inclusion. Fixation artefacts were observed in sample 29. AA, amino acid; AF, allele frequency; FA, fixation artefact; n.a., not analysable; n.i., not included in primer panel; -, no variant found.

Concordant results were found in the genes CTNNB1 (no. 22), PIK3CA (nos. 19 and 21) and most frequently in TP53 (nos. 2, 3, 5, 6, 7, 11, 12, 15, 18 and 19). In two samples (nos. 9 and 26), a recurrent KRAS p.G12D mutation was identified. Notably, in sample 9 this KRAS mutation with a low allele frequency of 2.36 and 5%, respectively, was identified by both institutes, thereby confirming the true nature of this mutation (Table V).

Divergent results were discovered in sample no. 29. The average number of reported variants for each sample was 172 for all allele frequencies and 23 for allele frequencies above 5% in institute A. Sample no. 29 showed a markedly higher number of variants (157) following bioinformatic analysis institute A. The sample from institute A had a very low DNA concentration (Table II) and the variants were predominantly G>A or T>C substitutions. The results included besides other variants different hotspot mutations such as BRAF c.1406G>A, p.G469E [allele frequency (AF), 51%; coverage (cov), 6813], PIK3CA c.1633G>A, p.E545K (AF, 18%; cov, 6190) and NRAS c.178G>A, p.G60R (AF, 39%; cov, 2187) (Table V and Fig. For verification, the respective regions were reanalysed with Sanger sequencing as previously described (14). The mutations could not be confirmed and were categorized as fixation artefacts.

Discussion

In routine pathological diagnostics mostly FFPE material is available for molecular characterisation. With decreasing sample sizes and increasing numbers of molecular analyses, a targeted sequencing approach using MPS systems seems to be required. Since it is well known that DNA extracted from FFPE is degraded, with a maximum size of about 350 bp (15), approaches such as whole genome, transcriptome or exome sequencing are, besides being labour-intensive and expensive, not suitable for routine diagnostics. Targeted sequencing with the focus on hotspot regions is suitable for analysing FFPE material, in a cost-effective and technically feasible way. Comparing the benchtop systems available for parallel sequencing, they show all method-specific advantages and disadvantages. The 454 GS Junior has a low throughput, but generates at the same time long runs (16,17). The Ion Torrent PGM™ is a cost-saving and fast system, but has a limited accuracy in homopolymeric regions, which also applies to the 454 GS Junior (1,16). The MiSeq has a very high throughput and low error rates, but the runtime is long (17) and it needs a higher number of samples per run to be cost efficient.

In this study, in comparing 30 lung cancer samples with three different MPS platforms, we observed good concordance in the detection of mutations using different DNA extraction methods, quantification systems and individually designed primer panels. All institutes analysed 26 out of 26 samples accurately concerning the EGFR status.

Independently of the downstream methods used, the crucial step in mutation analyses from tumour material is macrodissection and therewith the selection of the right areas. A tumour burden of 40% is recommended for Sanger sequencing (18). As MPS is more sensitive than Sanger sequencing, the amount of tumour cells required may be lower (19,20). Samples with low tumour cell content are at risk of being reported as false-negative. In contrast to our results (21) found no correlation between H&E-based morphologic assessment of tumour burden and the actual mutant allele frequency. In our cohort, the absolute allele frequencies for certain variants showed differences between the three laboratories, depending mainly on the selection of the macrodissected area. Restricted marking of tumour cells increases the detection thresholds, which may be critical for variants with low allele frequencies. Unfortunately at the same time there is an enhanced risk of 'mispicking' during the manual dissecting process. The important role of manual macrodissection is also emphasized by Ausch et al because the combination of the content of tumour cells and the allele frequency leads to the diagnostic study (22). We recommend a careful pathologic review of each individual case because the minimum percentage of tumour cells for doubtless results has not yet been defined (23). From our results, we suggest a tumour cell burden of at least 10%, which can also be reached in small biopsies.

Through the development of minimally invasive techniques biopsy sizes are decreasing. This is in contrast to the ever increasing demands of immunohistochemistry stainings and molecular analyses. Minimally invasive biopsies often deliver insufficient amounts of tissue material for subsequent analyses. We included one extra small tissue sample (no. 10) on purpose, which was originally difficult to analyse by conventional methods, to explore how the different MPS systems would cope with such a sample. None of the institutes were able to extract sufficient DNA for a reliable molecular analysis using next-generation sequencing (NGS) technologies.

In institute A, two further samples could not be analysed due to the high salt concentrations in BioRobot M48 extracts (12). The multiplex PCR for the library generation was inhibited and samples failed completely. Institute B could not analyse two samples as well due to strong DNA degradation. This can be attributed to the manual extraction method chosen byin institute B as it has been reported that automated nucleic acid extraction ensures a standardisation of sample processing and decreases time and variability in the clinical laboratory (24,25). Additionally, it is well known that manual extraction delivers less DNA than automated extraction (26). In this study, a comparison of the total DNA amounts is not possible due to the different systems used for measuring of DNA concentration. In institute C, using the automated QIASymphony SP system, only one sample failed. This extraction system was previously shown to generate DNA extracts with higher quality and concentration [Heydt et al (12)].

In FFPE material, non-reproducible sequence artefacts caused by DNA deamination induced by the sample fixation are frequently detected by all sequence analysis methods. The characteristic nucleotide transitions G>A and T>C had been found by several groups (27–29). Sequence artefacts arising from FFPE DNA are especially problematic when only limited amounts of template DNA are used for PCR amplification [Wong et al (29)]. In one of our samples, we detected mutations in hotspot regions with the typical C>T and G>A exchange which could not be validated by Sanger sequencing although they had sufficient allele frequency and coverage in MPS (Fig. 2).

Figure 2

Fixation artefacts. In our cohort, sample 29 showed a high number of variants after the bioinformatic analysis in institute A. Hotspot mutations in BRAF, NRAS and PIK3CA were selected for validation by Sanger sequencing. The mutations could not be confirmed and were therefore assessed to be fixation artefacts. AF, allele frequency; cov, coverage.

Since the fixation artefacts are amplified during all PCR-based methods and appear as false-positive variants, it is advisable to reduce the DNA amplification steps during mutational analyses. Hybrid selection methods like Nanostring® or SureSelect (Agilent Technologies) work without a preamplification step. Also, an approach from Udar et al where the two DNA strands were processed individually minimises fixation artefacts (30). Two independent libraries were combined and sequenced on the MiSeq (Illumina) instrument. Variant frequencies were calculated using information from both strands and are narrowed down.

Notably, the KRAS mutation (c.35G>A, p.G12D) in sample nine, which could also be attributed to a fixation artefact, was identified by two institutes with allele frequencies of 2.36 and 5% confirming the true nature of this mutation (Table V). Most of the artefacts appear once but not in duplicates so one solution to detect C>T (and G>A) sequence artefacts when using FFPE-DNA is to prepare analysis in duplicates. Verification of such low allele frequencies with an alternative method is a challenge, because most methods (Sanger sequencing, high resolution melting) have a higher detection limit than MPS.

The majority of patients with lung cancer receiving EGFR-tyrosine kinase inhibitor (TKI) therapy acquire resistance after a median of 10–16 months (31). Intense study in these NSCLCs has identified two major mechanisms of developing resistance to first generation TKIs: secondary resistance mutations within the same gene and 'oncogene kinase switch' systems with an overlap into another pathway (32). Also, new sensitive detection methods like MPS have identified a proportion of TKI-naive tumours that carry the secondary resistance mutation p.T790M in the EGFR gene; these resistant clones may be selected after exposure to TKI inhibitors (32–35). In institute A, two samples (nos. 12 and 23) with minor clones for the EGFR resistance mutation p.T790M were found (Table IV). Due to the low allele frequency, validation with Sanger sequencing seemed to be impossible. We therefore used a qPCR approach with a detection limit of 1%. Neither the DNA extracts from institutes B and C, nor the newly prepared or the primary DNA extracts from institute A, showed the resistance mutation (data not shown). Therefore, for the analysis of DNA from FFPE tissues, a general detection limit of 5% seems to balance sensitivity vs. reproducibility.