University of Groningen Next generation sequencing guided molecular diagnostic tests in non-small-cell lung cancer Wei, Jiacong

(1)

Next generation sequencing guided molecular diagnostic tests in non-small-cell lung cancer

Wei, Jiacong

DOI:

10.33612/diss.101317239

IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below.

Document Version

Publisher's PDF, also known as Version of record

Publication date: 2019

Link to publication in University of Groningen/UMCG research database

Citation for published version (APA):

Wei, J. (2019). Next generation sequencing guided molecular diagnostic tests in non-small-cell lung cancer. Rijksuniversiteit Groningen. https://doi.org/10.33612/diss.101317239

Copyright

Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).

Take-down policy

If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.

Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.

(2)

3

Manuscript in preparation

Jiacong Wei

1

_,

4

_{; Pei Meng}

2

_,

4

_{; Miente Martijn}

Terpstra

1

_{; Anke Rijk}

2

_{; Menno Tamminga}

3

_,

Frank Scherpen

2

_{; Arja ter Elst}

2

_{; Mohamed ZMA}

1

_;

Lennart Johansson

1

_{; T. Jeroen N. Hiltermann}

3

_;

Harry J.M. Groen

3

_{; Klaas Kok}

1

_{; Anthonie J. van der}

Wekken

3

_{; Anke van den Berg}

2

_*

1 Department of Genetics, University of Groningen, University Medical Center Groningen, 9700RB, Netherlands

2 Department of Pathology and Medical Biology, University of Groningen, University Medical Center Groningen, 9700RB, Netherlands

3 Department of Pulmonary Diseases, University of Groningen, University Medical Center Groningen, 9700RB, Netherlands

4 Department of Pathology, Collaborative and Creative Centre, Shantou University Medical College, Shantou, 515041, China

* Correspondent author

Clinical value of EGFR gene

amplifica-tions detected using ampliconbased

targeted next generation sequencing

data in lung adenocarcinoma patients

(3)

Abstract Objective

The relevance of EGFR gene amplification in patients with EGFR‐mutated advanced non‐small cell lung cancer (NSCLC) has not been fully elucidated. Moreover, a limited number of studies have been published on methods to estimate gene amplifications based on routinely performed targeted next generation sequencing (NGS). In this study we aimed to determine whether PCR‐ based targeted NGS data on tumor tissue can be used to estimate presence of gene amplifications and explored the prognostic value of EGFR gene amplification in EGFR mutated NSCLC patients.

Material and methods

A total of 3,194 good quality targeted NGS data files were retrieved from 2014 to 2017. Among those, 1,729 NSCLC samples originated from 1,586 NSCLC patients of whom 134 had an EGFR mutation (8.2%) and clinical data being available for 66 of the patients. Raw sequencing data were re‐analyzed using a custom designed pipeline. The presence of gene amplifications was based on ratio of amplicon reads of a given gene relative to the reference amplicons in the sample or relative to normal control samples (cut‐off ≥3). Reference amplicons were selected based on low degree of variation amongst all tested samples. Amplifications were regarded significant when the z score was ≥3.5. Technical validation of the amplification calling procedure was done by FISH and MLPA. Cox regression analysis on overall survival was done with each of the amplification parameters using SPSS.

Results

No amplifications were detected for the ALK, KIT, NRAS, PDGFRA, GNAQ and MAP2K1 gene loci, whereas amplifications for BRAF, ESR1, GNA11, HRAS, KRAS, MET and PIK3CA were observed at a low frequency. Depending on the amplification analysis strategy (within sample or relative to normal controls), 19% and 13% of the EGFR mutated group had an EGFR amplification, respectively. In EGFR wild type patients, amplifications were detected in 5% and 4% of the patients using the two methods, respectively. The sensitivity and specificity of the NGS based estimation of amplifications for EGFR were 94% (14/15) and 83% (29/35) for both data analysis approaches. Patients with concurrent EGFR mutations and amplifications (estimated within sample) treated with EGFR‐TKI had a worse overall survival compared with those without concurrent EGFR amplifications (ratio, p=0.047 and z score, p=0.015). Cox regression analysis indicated a borderline significant interaction between ratio and z score (p=0.052).

Conclusion

Routinely obtained amplicon‐based NGS data can be used to identify gene amplifications. The presence of EGFR amplifications in EGFR mutant patients is predictive of a worse overall survival.

(4)

3

Abstract Objective

The relevance of EGFR gene amplification in patients with EGFR‐mutated advanced non‐small cell lung cancer (NSCLC) has not been fully elucidated. Moreover, a limited number of studies have been published on methods to estimate gene amplifications based on routinely performed targeted next generation sequencing (NGS). In this study we aimed to determine whether PCR‐ based targeted NGS data on tumor tissue can be used to estimate presence of gene amplifications and explored the prognostic value of EGFR gene amplification in EGFR mutated NSCLC patients.

Material and methods

A total of 3,194 good quality targeted NGS data files were retrieved from 2014 to 2017. Among those, 1,729 NSCLC samples originated from 1,586 NSCLC patients of whom 134 had an EGFR mutation (8.2%) and clinical data being available for 66 of the patients. Raw sequencing data were re‐analyzed using a custom designed pipeline. The presence of gene amplifications was based on ratio of amplicon reads of a given gene relative to the reference amplicons in the sample or relative to normal control samples (cut‐off ≥3). Reference amplicons were selected based on low degree of variation amongst all tested samples. Amplifications were regarded significant when the z score was ≥3.5. Technical validation of the amplification calling procedure was done by FISH and MLPA. Cox regression analysis on overall survival was done with each of the amplification parameters using SPSS.

Results

No amplifications were detected for the ALK, KIT, NRAS, PDGFRA, GNAQ and MAP2K1 gene loci, whereas amplifications for BRAF, ESR1, GNA11, HRAS, KRAS, MET and PIK3CA were observed at a low frequency. Depending on the amplification analysis strategy (within sample or relative to normal controls), 19% and 13% of the EGFR mutated group had an EGFR amplification, respectively. In EGFR wild type patients, amplifications were detected in 5% and 4% of the patients using the two methods, respectively. The sensitivity and specificity of the NGS based estimation of amplifications for EGFR were 94% (14/15) and 83% (29/35) for both data analysis approaches. Patients with concurrent EGFR mutations and amplifications (estimated within sample) treated with EGFR‐TKI had a worse overall survival compared with those without concurrent EGFR amplifications (ratio, p=0.047 and z score, p=0.015). Cox regression analysis indicated a borderline significant interaction between ratio and z score (p=0.052).

Conclusion

Routinely obtained amplicon‐based NGS data can be used to identify gene amplifications. The presence of EGFR amplifications in EGFR mutant patients is predictive of a worse overall survival.

Keywords: Lung adenocarcinoma, EGFR, survival, tyrosine kinase inhibitor

Background

In the past decade, targeted therapies have dramatically improved clinical management of non‐ small cell lung cancer (NSCLC), especially of lung adenocarcinoma patients [1]. The therapeutic choices are determined by the presence of activating mutations in the epidermal growth factor receptor (EGFR) and V‐RAF murine sarcoma viral oncogene homolog B1 (BRAF) genes, and by gene fusions targeting anaplastic lymphoma kinase (ALK), c‐ROS Oncogene 1 (ROS1) and rearranged during transfection (RET) [1] and the more recently identified fusions involving neurotrophic tyrosine receptor kinase (NTRK) [2].

EGFR belongs to the family of receptor tyrosine kinases (RTKs) that regulate multiple downstream signaling pathways involved in survival and proliferation [3,4]. Activating mutations in EGFR are observed in 10‐35% of lung adenocarcinoma patients, with deletions in exon 19 and the p.L858R mutation in exon 21 being the most common [5]. Patients with these activating mutations are treated with first or second generation tyrosine kinase inhibitors (TKIs) including erlotinib, gefitinib, afatinib, osimertinib [6]. Other TKI‐sensitive mutations are observed at amino acid positions 719, 768, 797 and 861 [7]. Patients with insertions and deletions in exon 20 do in general not respond to TKI treatment.

Several studies have shown amplifications of the EGFR gene region in lung adenocarcinoma using fluorescence in situ hybridization (FISH), chromogenic in situ hybridization (CISH) or multiplex ligation‐dependent probe amplification (MLPA) [8‐11]. These amplifications were observed more frequently in tumor samples with EGFR mutations (8% to 81%) as compared to tumor samples without EGFR mutations (1%‐29%). In one of these studies the presence of concurrent EGFR amplification in EGFR mutant patients was associated with improved response to TKI treatment as compared to patients without EGFR amplifications [8]. In several other studies, high mutant allele frequencies (MAF) for EGFR exon 19 deletion or L858R were associated with a better response to TKI treatment compared to those with low MAF [12‐14]. One study suggested that gain or amplification more frequently involved the mutated EGFR allele, but did not investigate the predictive value of concurrent EGFR amplifications [15]. NGS data are commonly used for the concomitant detection of copy number aberrations, and validated protocols are available for whole genome sequencing data and hybridization‐based targeted enrichment sequencing data sets [16]. The use of NGS data obtained by PCR‐based target enrichment is more challenging, and a consensus of “best practices” especially for aneuploidy tumor samples still needs to be reached. In a recent study, the ratio of the normalized read counts per amplicon and/or per gene compared to those in normal samples was used as an estimation of the copy number [17]. The z score was calculated to indicate the reliability of the calculated ratios. The use of normal control samples can potentially be hampered by some factors, e.g. the limited number of normal samples and variability in inter‐ assay and experimental conditions. For targeted NGS data with limited numbers of amplicons per gene a modified approach has been proposed, using median ratio values and a z score cutoff of 3.5 [18,19]. An alternative approach, using internal reference within the tested sample such as fixed number of stable amplicons, has not been explored.

(5)

In this study, we analyzed an amplicon based diagnostic IonTorrent hotspot panel dataset of patients with advanced NSCLC submitted over a period of 3 years for molecular diagnostic testing within our center. We used amplicon read depth to identify amplifications of EGFR and other genes by defining an internal reference amplicon set and by comparison to normal control samples. In addition, we evaluated whether the presence of an EGFR amplification in addition to the TKI sensitive EGFR mutations is a predictive marker for tumor response to EGFR‐TKI. Material and Methods

Patient samples

We retrieved data from 3,540 diagnostic samples that were subjected to NGS analysis in the period 2014 to 2017 (Figure 1). Three hundred forty‐six samples were excluded based on poor quality NGS data (i.e. median read counts per amplicon <50). For validation of amplification detection using read counts per amplicon, 42 NGS datasets were retrieved from samples with or without amplifications of ERBB2 (n=23) and MET (n=35) as established by FISH in the diagnostic setting from January 2018 until April 2019. The study protocol is consistent with the Research

Code of the University Medical Centre Groningen

(https://www.rug.nl/umcg/research/documents/research‐code‐info‐umcg‐nl.pdf) and national ethical and professional guidelines (“Code of conduct; Dutch federation of biomedical scientific societies”, htttp://www.federa.org).

DNA isolation, library preparation and sequencing

DNA was isolated from neoplastic cell rich areas of four to eight 5‐μm formalin fixed paraffin embedded (FFPE) tissue section slides, using Cobas kit (Roche Diagnostic Systems, Inc., Branchburg, USA) according to the instructions of the manufacturer. The DNA concentration was measured by Qubit™ dsDNA High Sensitivity assay using Qubit 2.0 Fluorometer (Invitrogen, Carlsbad, USA). 6 ng or more DNA was used as input for the amplicon‐based enrichment step. Resulting libraries were generated and processed for sequencing on the Ion PGM sequencing system (Life Technologies, San Francisco, USA).

Two custom designed AmpliSeq™ panels have been used in the period between 2014‐2017. The first custom designed AmpliSeq panel (Design 1) was used from September 2014 to September 2016, and consisted of 30 amplicons covering mutational hotspots of 11 genes (Supplementary Table 1). The second design (Design 2) implemented in September 2016 included two separate amplicon pools (Pool 1 and Pool 2) with 44 and 40 different amplicons respectively of 36 genes (Supplementary Table 1). Barcoded Ampliseq libraries were pooled and subjected to emulsion polymerase chain reaction (PCR) using the OneTouch2 (Life Technologies, San Francisco, USA). Sequencing was performed on the Ion Torrent PGM.

Copy number analysis

Read counts for each amplicon were generated using a Copy Number Variation Detection In Next‐generation sequencing gene panels (CoNVaDING) based pipeline [16]. Of note, only reads that covered at least 80% of the amplicon were regarded to originate from the involved amplicon. In Design 2, one amplicon was excluded, i.e. AMELY, a gender differentiating gene

(6)

3

In this study, we analyzed an amplicon based diagnostic IonTorrent hotspot panel dataset of patients with advanced NSCLC submitted over a period of 3 years for molecular diagnostic testing within our center. We used amplicon read depth to identify amplifications of EGFR and other genes by defining an internal reference amplicon set and by comparison to normal control samples. In addition, we evaluated whether the presence of an EGFR amplification in addition to the TKI sensitive EGFR mutations is a predictive marker for tumor response to EGFR‐TKI. Material and Methods

Patient samples

We retrieved data from 3,540 diagnostic samples that were subjected to NGS analysis in the period 2014 to 2017 (Figure 1). Three hundred forty‐six samples were excluded based on poor quality NGS data (i.e. median read counts per amplicon <50). For validation of amplification detection using read counts per amplicon, 42 NGS datasets were retrieved from samples with or without amplifications of ERBB2 (n=23) and MET (n=35) as established by FISH in the diagnostic setting from January 2018 until April 2019. The study protocol is consistent with the Research

Code of the University Medical Centre Groningen

(https://www.rug.nl/umcg/research/documents/research‐code‐info‐umcg‐nl.pdf) and national ethical and professional guidelines (“Code of conduct; Dutch federation of biomedical scientific societies”, htttp://www.federa.org).

DNA isolation, library preparation and sequencing

DNA was isolated from neoplastic cell rich areas of four to eight 5‐μm formalin fixed paraffin embedded (FFPE) tissue section slides, using Cobas kit (Roche Diagnostic Systems, Inc., Branchburg, USA) according to the instructions of the manufacturer. The DNA concentration was measured by Qubit™ dsDNA High Sensitivity assay using Qubit 2.0 Fluorometer (Invitrogen, Carlsbad, USA). 6 ng or more DNA was used as input for the amplicon‐based enrichment step. Resulting libraries were generated and processed for sequencing on the Ion PGM sequencing system (Life Technologies, San Francisco, USA).

Two custom designed AmpliSeq™ panels have been used in the period between 2014‐2017. The first custom designed AmpliSeq panel (Design 1) was used from September 2014 to September 2016, and consisted of 30 amplicons covering mutational hotspots of 11 genes (Supplementary Table 1). The second design (Design 2) implemented in September 2016 included two separate amplicon pools (Pool 1 and Pool 2) with 44 and 40 different amplicons respectively of 36 genes (Supplementary Table 1). Barcoded Ampliseq libraries were pooled and subjected to emulsion polymerase chain reaction (PCR) using the OneTouch2 (Life Technologies, San Francisco, USA). Sequencing was performed on the Ion Torrent PGM.

Copy number analysis

Read counts for each amplicon were generated using a Copy Number Variation Detection In Next‐generation sequencing gene panels (CoNVaDING) based pipeline [16]. Of note, only reads that covered at least 80% of the amplicon were regarded to originate from the involved amplicon. In Design 2, one amplicon was excluded, i.e. AMELY, a gender differentiating gene

located on the Y chromosome. Moreover, for Design 2, the amplicons from the two PCR pools were analyzed separately.

Variability in coverage per amplicon was standardized using amplicon coverage divided by the total read counts of all other amplicons in the library pool. Reference amplicons were selected per pool based on coefficient of variations (CV) of standardized read counts per amplicon over all samples with median read count >50. The 25% amplicons with the lowest CV values were selected as internal reference amplicons. Starting from the raw read counts per pool and per sample, we calculated the coverage of each amplicon relative to the internal reference amplicons. To identify outlier amplicons per gene we calculated correlation coefficients of each amplicon relative to the other amplicons of the same gene in the same amplification pool. Amplicons with an R‐square less than 0.5 to the other amplicons of the same gene were excluded from subsequent calculations.

For the lung cancer samples, we subsequently calculated the gene ratio per amplification pool using the formula: [median of the coverage per amplicon]/[median coverage of reference amplicons]. This ratio indicates the relative gene‐specific copy number state within a sample. For design 1 with one library pool, this is the final relative abundance, for design 2 we calculated the average value of the two pools as a measure for the relative gene‐specific copy number state. We calculated the modified z score as a measure for the significance of the identified copy number changes [18,19]. The within‐sample modified z score is calculated as [gene ratio ‐ median ratios of internal reference amplicons] / [Median absolute deviation (MAD) of ratios of internal reference amplicons].

For comparison with normal control samples, the gene specific ratio was calculated as above and divided by the median gene specific coverage in the normal control samples. The modified z score relative to normal was defined as { [gene ratio]sample ‐ [median gene ratio]normal controls }/{MAD of gene ratios in normal control samples}.

Genes represented by a single amplicon in the designs were excluded from the analyses as we considered single amplicon values unreliable for determining amplifications.

To summarize, we followed two approaches to define gene amplification, namely the within sample comparison and the relative to normal approaches. For both approaches, we regarded a ratio ≥3 and a z score ≥3.5 as proof of amplification according to a previous study [19].

Validation of the NGS-based amplification analysis

To validate our amplification calling approach, NGS data of patients with FISH results of ERBB2 or MET were retrieved and analyzed in the same way to call amplifications. FISH procedures were carried out using PathVysion HER2 DNA Probe Kit and Vysis MET SpectrumRed Probes (Abbott Molecular, Abbott Park, USA) following the manufacturer's recommendations. Stained tissue sections were scored by two independent observers by referring to American Society of Clinical Oncology guidelines for ERBB2 in breast cancer and for MET in lung cancer [20‐22]. In case of disconcordant results for ERBB2 or MET a third observer was consulted.

ERBB2 and CEP17 signals were counted in 50 non‐overlapping nuclei and ERBB2/CEP17 ratios were calculated. ERBB2 amplification was defined as positive when ERBB2/CEP17 ratio ≥2, or the ratio <2 with ≥6 ERBB2 signals per cell. An equivocal result was defined by ERBB2/CEP17

(7)

ratio <2 with between 4 and 6 ERBB2 signals per cell. No copy number changes were defined as ERBB2/CEP17 ratio <2 and<4 ERBB2 signals per cell.

For MET amplification 20 non‐overlapping nuclei were counted and ratios relative to the CEP7 signals were calculated. For MET/CEP7 ratios ≥15 in 10% of the cells or presence of MET clusters was defined as high amplification. MET/CEP7 ratios ≥2 with 5 to 14 MET signals/cell were defined as low amplification. High polysomy of MET was defined as ratio <2 with ≥4 MET signals/cell. Low polysomy of MET was defined as ratio <2 with 3 MET signals per cell. Cases with a ratio <2 and <3 MET signals per cell were regarded as normal MET copy number cases. To validate EGFR copy number changes, a multiplex ligation‐dependent probe amplification (MLPA) analysis was performed [23] using the SALSA MLPA P105 Glioma‐2 probe mix (MRC Holland, Amsterdam, the Netherlands), which is a validated assay in the molecular diagnostics group for glioblastoma.MLPA was performed in accordance with the manufacturer’s instruction on the same DNA sample as used for targeted NGS. For each run, we included three normal controls and one positive control sample with known EGFR amplification. DNA samples used for the analysis were retrieved from the molecular diagnostics archive, based on availability. We aimed to include all amplified cases and a similar number of non‐amplified cases based on our NGS‐based results. CNV analysis was performed using Coffalyser net software (MRC Holland, Amsterdam, the Netherlands).

Figure 1. Flow diagram of the study showing inclusion of patient samples for amplification calling and the number

of patient samples available for survival analysis. Survival analysis

(8)

3

ratio <2 with between 4 and 6 ERBB2 signals per cell. No copy number changes were defined as ERBB2/CEP17 ratio <2 and<4 ERBB2 signals per cell.

For MET amplification 20 non‐overlapping nuclei were counted and ratios relative to the CEP7 signals were calculated. For MET/CEP7 ratios ≥15 in 10% of the cells or presence of MET clusters was defined as high amplification. MET/CEP7 ratios ≥2 with 5 to 14 MET signals/cell were defined as low amplification. High polysomy of MET was defined as ratio <2 with ≥4 MET signals/cell. Low polysomy of MET was defined as ratio <2 with 3 MET signals per cell. Cases with a ratio <2 and <3 MET signals per cell were regarded as normal MET copy number cases. To validate EGFR copy number changes, a multiplex ligation‐dependent probe amplification (MLPA) analysis was performed [23] using the SALSA MLPA P105 Glioma‐2 probe mix (MRC Holland, Amsterdam, the Netherlands), which is a validated assay in the molecular diagnostics group for glioblastoma.MLPA was performed in accordance with the manufacturer’s instruction on the same DNA sample as used for targeted NGS. For each run, we included three normal controls and one positive control sample with known EGFR amplification. DNA samples used for the analysis were retrieved from the molecular diagnostics archive, based on availability. We aimed to include all amplified cases and a similar number of non‐amplified cases based on our NGS‐based results. CNV analysis was performed using Coffalyser net software (MRC Holland, Amsterdam, the Netherlands).

Figure 1. Flow diagram of the study showing inclusion of patient samples for amplification calling and the number

of patient samples available for survival analysis. Survival analysis

Clinical data were independently retrieved from the patient files. Data retrieved included gender, stage of disease, pathology, TKI, tumor response and survival. Patients who were selected for this study signed an informed consent for bio‐banking to use their tissue and patient characteristics.

Presence of gene amplifications in patients with and without EGFR mutations were tested for prediction of overall survival. The predictive value of NGS‐determined amplification was studied in the patient cohort treated with first line TKI treatment using ratio and z score of both approaches as independent variables.

Overall survival was defined as the time between the date of starting first TKI treatment and the date of death or censored for loss of follow up. Cox regression on overall survival was separately performed with each ratio and z score, as well as with interaction between ratio and z score from the two approaches. Kaplan‐Meier plots were generated to show the time to event distributions. All survival analyses were performed using SPSS 23.

Results

Clinical data on EGFR mutant patients

NGS data were retrieved from 3,540 tumor samples of which 3,194 had a median coverage per amplicon of >50. These included 1,729 NSCLC samples from 1,586 patients, 786 colorectal tumor samples from 748 patients, 435 melanoma samples from 413 patients, 211 samples of other malignancies from 189 patients and 33 normal samples.

From the diagnostic reports of the 1,586 NSCLC patients, we identified 134 (8.4%) patients with an EGFR mutation (Table 1 and Table 2). Exon 19 deletions (E19 DELs) were most common with a frequency of 2.5% followed by the L858R mutation with a frequency of 2.1%. Other common mutations included G719A/C/S with a frequency of 1.1%, exon 19 INDELs with a frequency of 0.5%, S768I with a frequency of 0.44%, E709A/K and L861Q both with a frequency of 0.25%. The TKI unresponsive exon 20 INDEL mutations (E20 INDELs) were observed in 14 patients. The T790M mutations occurred in combination with E19 DEL (13 samples), L858R (11 samples), E19 INDEL (2 samples), and S768I (1 sample). In one sample with L858R a secondary EGFR C797S mutation that leads to TKI resistance was found. Out of the 134 patients, we managed to retrieve detailed clinical information of 66 patients. EGFR amplification analysis was not performed as part of the routine diagnostic tests. We defined the EGFR amplification patterns, by reanalysis of the raw NGS data.

Defining reference amplicons and removing outlier amplicons

After normalization of the read counts obtained by reanalysis of the BAM files, we calculated coefficients of variation for all amplicons per amplification pool and selected the 25% amplicons with the lowest coefficient of variation across all samples as reference amplicons. For design 1 with a total of 2,444samples with median read count per amplicon >50 this resulted in eleven reference amplicons (Table 3). For design 2 with a total of 1,096 samples with a median read

(9)

count per amplicon >50, we selected 11 and 10 reference amplicons for pool 1 and pool 2, respectively.

To identify possible outlier amplicons for each pool we calculated correlation coefficients for read counts of all amplicons per gene and per pool. For design 1 all but one amplicon, i.e. PIK3CA, showed an R square above 0.5 (Supplementary Figure 1). The genes from design 1 included in the amplification analysis were ALK, BRAF, EGFR, ERBB2, KIT, PDGFRA, NRAS and KRAS. For design 2, we excluded 10 amplicons from pool 1 and 8 from pool 2 based on an R square below 0.5 with the other amplicons of the same gene. The genes included for the amplification analysis were ALK, BRAF, EGFR, ERBB2, ESR1, HRAS, KIT, MAP2K1, MET, NRAS, GNA11, GNAQ, PIK3CA and PDGFRA.

Table 1. Characteristics EGFR mutation positive NSCLC patients with clinical data

Characteristics n=66 (%) Age upon TKI treatment

Median 67.5 Range 40‐85 Gender Male 23 (35%) Female 43 (65%) Smoking status Never Smoker 26 (39%) Smoker 40 (61%) Histologic type Adenocarcinoma 66

Treatment of EGFR TKI

1st_line _{49 (74%)}

2nd_line _{30 (45%)}

3rd_line _{10 (15%)}

Gene ratios and z scores for amplification

EGFR gene amplifications based on a ratio of ≥3 and a z score of ≥3.5 were identified in 96 (6.1%) samples by the internal comparison approach and in 70 samples (4.4%) by the normal comparison approach. The overlap between both methods was limited with 65 of the samples being positive for both approaches (Figure 2). Amplifications of ERBB2 were observed at higher frequencies (n=120, 7.6%) by internal comparison approach and at slightly lower frequencies (n=55, 3.5%) by the normal comparison approach as compared to EGFR. Gene amplifications were observed at lower frequencies for BRAF (0.4%), ESR1 (0.4%), GNA11 (1.1%), HRAS (1.1%), KRAS (0.6%), MET (0.1%) and PIK3CA (0.2%) using the internal comparison approach and BRAF (0.7%), ESR1 (0.2%), GNA11 (0.9%), HRAS (1.3%), KRAS (0.6%), MET (0.3%) and PIK3CA (0.2%)

(10)

3

count per amplicon >50, we selected 11 and 10 reference amplicons for pool 1 and pool 2, respectively.

To identify possible outlier amplicons for each pool we calculated correlation coefficients for read counts of all amplicons per gene and per pool. For design 1 all but one amplicon, i.e. PIK3CA, showed an R square above 0.5 (Supplementary Figure 1). The genes from design 1 included in the amplification analysis were ALK, BRAF, EGFR, ERBB2, KIT, PDGFRA, NRAS and KRAS. For design 2, we excluded 10 amplicons from pool 1 and 8 from pool 2 based on an R square below 0.5 with the other amplicons of the same gene. The genes included for the amplification analysis were ALK, BRAF, EGFR, ERBB2, ESR1, HRAS, KIT, MAP2K1, MET, NRAS, GNA11, GNAQ, PIK3CA and PDGFRA.

Table 1. Characteristics EGFR mutation positive NSCLC patients with clinical data

Characteristics n=66 (%) Age upon TKI treatment

Median 67.5 Range 40‐85 Gender Male 23 (35%) Female 43 (65%) Smoking status Never Smoker 26 (39%) Smoker 40 (61%) Histologic type Adenocarcinoma 66

Treatment of EGFR TKI

1st_line _{49 (74%)}

2nd_line _{30 (45%)}

3rd_line _{10 (15%)}

Gene ratios and z scores for amplification

EGFR gene amplifications based on a ratio of ≥3 and a z score of ≥3.5 were identified in 96 (6.1%) samples by the internal comparison approach and in 70 samples (4.4%) by the normal comparison approach. The overlap between both methods was limited with 65 of the samples being positive for both approaches (Figure 2). Amplifications of ERBB2 were observed at higher frequencies (n=120, 7.6%) by internal comparison approach and at slightly lower frequencies (n=55, 3.5%) by the normal comparison approach as compared to EGFR. Gene amplifications were observed at lower frequencies for BRAF (0.4%), ESR1 (0.4%), GNA11 (1.1%), HRAS (1.1%), KRAS (0.6%), MET (0.1%) and PIK3CA (0.2%) using the internal comparison approach and BRAF (0.7%), ESR1 (0.2%), GNA11 (0.9%), HRAS (1.3%), KRAS (0.6%), MET (0.3%) and PIK3CA (0.2%)

using normal comparison approach. No amplifications were detected for KIT, NRAS, PDGFRA, GNAQ and MAP2K1 with either approach. For ALK, six amplification positive cases were found with the normal control sample approach, but none of them was observed with the internal reference amplicon approach. The overlap between the two methods was limited for ALK and MET, whereas it was more than 40% for BRAF, ERBB2, ESR1, HRAS, GNA11, KRAS, and PIK3CA (Supplementary Figure 2). This might be related to the combination of more amplicons per gene and the exclusion of a higher proportion of outlier amplicons in design 2.

Table 2. EGFR mutations reported in the diagnostic setting in all NSCLC patients.

EGFR Mutations No. Sample / Patients Frequency in cohort based on 1586 patients No. of patients with first‐line EGFR‐ TKI E19 DEL 52 / 40 2.52% 21 L858R 50 / 34 2.14% 13 G719A/C/S 26 / 18 1.13% 2 E19 INDEL 9 / 8 0.50% 3 E20 INDELb _{19 / 14} _0.88% ₄ S768I 9 / 7 0.44% 3 E709A/K 6 / 4 0.25% 0 L861Q 5 / 4 0.25% 0 L747Pb _{3 / 2} _<0.25% ₀ T488Na _{1 / 1} _<0.25% ₀ V689L 1 / 1 <0.25% 0 K714Na _{1 / 1} _<0.25% ₀ T751_S752dela _{1 / 1} _<0.25% ₀ A755V 1 / 1 <0.25% 0 D761N 2 / 1 <0.25% 1 R776Ga _{2 / 1} _<0.25% ₀ R776H 1 / 1 <0.25% 1 R776L 1 / 1 <0.25% 1 T790Sa 1 / 1 <0.25% 0 C797Sb _{2 / 1} _<0.25% ₀ G810Va _{1 / 1} _<0.25% ₀ D830Y 1 / 1 <0.25% 0 V834L 1 / 1 <0.25% 0 R836La _{1 / 1} _<0.25% ₀ A840Tb _{2 / 2} _<0.25% ₀ P848Lb _{1 / 1} _<0.25% ₀ L861R 1 / 1 <0.25% 0

a_{response to EGFR‐TKI is unknown}

b _{unresponsive or resistant to EGFR‐TKI}

Validation of NGS-based EGFR amplifications by MLPA

For the validation of EGFR amplifications, we analyzed 50 samples by MLPA, 15 were scored as amplification positive, 35 as non‐amplified using a cutoff value of 3 (Table 3). Using the internal comparison approach, 14 out of 15 samples were correctly scored as EGFR amplified resulting in

(11)

a positive predictive value of 94%. Using the normal comparison approach, 14 out of 15 samples were correctly scored positive, with 13 samples overlapping with the internal comparison approach. For the negative samples, 29 out of 35 (83%) were negative using either of the two approaches. The 6 cases that were negative by MLPA were scored positive by both NGS based approaches.

Figure 2. NGS‐analysis based EGFR amplification using two different strategies. a) Ratios and z scores of NSCLC

patients as calculated by the internal comparison approach. b) Venn diagram showing the overlap in samples with EGFR amplifications between the two approaches. c) Ratios and z scores of NSCLC patients as calculated by the normal sample approach. The lines in the graphs of panel a and c represent the cut‐off values used for ratio (>3) and z score (>3.5). Green dots indicate normal samples, black dots indicate EGFR wild type samples, and red dots indicate EGFR mutant samples.

Validation of NGS-based ERBB2 and MET amplifications with FISH

We retrieved NGS data of 42 samples that were analyzed by FISH for ERBB2 (n=23) and/or MET (n=35) in the diagnostic setting (Table 3). Of the samples analyzed for ERBB2 by FISH, 3 were scored as positive, 15 as negative and the others were scored as equivocal or polysomy (n=5). Using the internal comparison approach, two of the three FISH positive cases were confirmed and all 15 negative cases and 5 equivocal or polysomy cases were correctly defined as non‐ amplified (negative predictive value of 100%). Using the normal comparison approach, none of the three ERBB2 FISH positive cases were scored as positive, whereas all FISH negative cases were correctly scored amplification‐negative based on NGS relative to normal comparison (n=20).Of the 35 samples analyzed for MET by FISH, 3 were scored as positive, 23 as negative, and 9 with polysomy. Using the internal comparison approach, one out of three FISH positive cases was correctly scored as positive by NGS, 20 out of 23 were correctly scored as negative and all 11 polysomy cases were correctly scored as negative (negative predictive of 91%). Using the normal comparison approach, all three cases were scored positive, and 19 out of 23 FISH negatives were correctly scored as negative and 8 out of 11 cases with polysomy were correctly scored as negative.

(12)

3

a positive predictive value of 94%. Using the normal comparison approach, 14 out of 15 samples were correctly scored positive, with 13 samples overlapping with the internal comparison approach. For the negative samples, 29 out of 35 (83%) were negative using either of the two approaches. The 6 cases that were negative by MLPA were scored positive by both NGS based approaches.

Figure 2. NGS‐analysis based EGFR amplification using two different strategies. a) Ratios and z scores of NSCLC

patients as calculated by the internal comparison approach. b) Venn diagram showing the overlap in samples with EGFR amplifications between the two approaches. c) Ratios and z scores of NSCLC patients as calculated by the normal sample approach. The lines in the graphs of panel a and c represent the cut‐off values used for ratio (>3) and z score (>3.5). Green dots indicate normal samples, black dots indicate EGFR wild type samples, and red dots indicate EGFR mutant samples.

Validation of NGS-based ERBB2 and MET amplifications with FISH

We retrieved NGS data of 42 samples that were analyzed by FISH for ERBB2 (n=23) and/or MET (n=35) in the diagnostic setting (Table 3). Of the samples analyzed for ERBB2 by FISH, 3 were scored as positive, 15 as negative and the others were scored as equivocal or polysomy (n=5). Using the internal comparison approach, two of the three FISH positive cases were confirmed and all 15 negative cases and 5 equivocal or polysomy cases were correctly defined as non‐ amplified (negative predictive value of 100%). Using the normal comparison approach, none of the three ERBB2 FISH positive cases were scored as positive, whereas all FISH negative cases were correctly scored amplification‐negative based on NGS relative to normal comparison (n=20).Of the 35 samples analyzed for MET by FISH, 3 were scored as positive, 23 as negative, and 9 with polysomy. Using the internal comparison approach, one out of three FISH positive cases was correctly scored as positive by NGS, 20 out of 23 were correctly scored as negative and all 11 polysomy cases were correctly scored as negative (negative predictive of 91%). Using the normal comparison approach, all three cases were scored positive, and 19 out of 23 FISH negatives were correctly scored as negative and 8 out of 11 cases with polysomy were correctly scored as negative.

Overlap between EGFR amplification and mutation status

Using the pre‐defined cut‐off values for calling amplifications 26 (19%) of the 134 EGFR mutant patients had an EGFR amplification according to the internal comparison approach and 18 (13%) according to the normal sample comparison approach. All 18 cases called as being amplified with the normal sample approach, were also scored positive by the internal comparison approach. In the 1440 EGFR wild type patients, EGFR amplifications were called in 70 (5%) and 52 (4%) of the patients by the internal reference amplicon comparison and normal sample comparison approach, respectively (Figure 2 and Table 4). A consistent positive result for both approaches was observed in 47 cases.

Table 3. Validation of NGS‐based amplifications by FISH and MLPA

Internal Comparison Normal Comparison

NGS + NGS ‐ NGS + NGS ‐ ERBB2 FISH + 2 1 0 3 FISH ‐ 0 20 0 20 MET FISH + 1 2 3 0 FISH ‐ 3 29 4 28 EGFR MLPA + 14 1 14 1 MLPA ‐ 6 29 6 29

Table 4. Cross table for NGS‐based EGFR amplification detection by two different approaches.

Internal comparison approach

amplification No amplification

Mutant 26 (19%) 108 (81%)

Wild type 70 (5%) 1382 (95%)

Normal comparison approach

amplification No amplification

Mutant 18 (13%) 116 (87%)

Wild type 52 (4%) 1400 (96%)

Comparison NGS EGFR amplification with clinical outcome

For 66 patients with EGFR mutations clinical information could be retrieved. Amongst them, 49 were treated with 1st_{line TKI drugs (Table 1, Figure 1). For this cohort the median PFS and OS} was 7.6 months (95% CI 5.6‐9.6) and 32 months (95% CI 22.1‐41.9), respectively. The ratio and z score by internal comparison approach were both significant (p<0.05) with a hazard ratio (HR) of 1.134 and 1.009 respectively (Table 5). The normal comparison approach showed a trend for the ratio, but no significant result for the z score. The interaction was borderline significant for both approaches.

(13)

The Kaplan Meier plots for the internal comparison approach showed worse OS for the fourth interquartile group with the highest z score relative to the lower three quartiles. For the ratio the two highest quartiles both show worse OS as compared to the lowest two groups (figure 4).

Table 5. Cox regression analysis on EGFR amplification for overall survival in 1st_{line TKI treated EGFR mutant NSCLC}

patients.

Approach Factor Significance Hazard Ratio 95% Confidence Interval Lower Upper Internal Comparison ratio 0.047 1.134 1.001 1.285 z score 0.015 1.009 1.002 1.015 Interaction* 0.052 0.997 0.995 1.000 Normal Comparison ratio 0.057 1.016 1.000 1.033 z score 0.174 1.002 0.999 1.004 interaction 0.058 1.000 1.000 1.000

* indicates interaction between ratio and z score.

Figure 3. Kaplan Meier plots of EGFR mutant NSCLC patients treated with 1st_{line EGFR‐TKIs.}_{a) OS for ratio shown in}

interquartile groups. b) OS for z score shown in interquartile groups. First interquartile group represents the lowest

quartile of the ratio or the z score, respectively. Discussion

In this study, we evaluated two approaches for calling gene amplifications by analyzing read counts of the amplicons obtained via an amplicon‐based targeted NGS approach. The NGS‐ based amplification calling strategies were validated by routine diagnostic tests, i.e. FISH for ERBB2 and MET, and MLPA for EGFR. We were able to demonstrate feasibility of our approach, with good validation results especially for calling EGFR amplifications. Subsequently, we looked into the clinical value of our amplification calling by evaluating the predictive value of ratio and z score on survival of a cohort of EGFR mutated patients who were treated with EGFR‐TKI. The

(14)

3

The Kaplan Meier plots for the internal comparison approach showed worse OS for the fourth interquartile group with the highest z score relative to the lower three quartiles. For the ratio the two highest quartiles both show worse OS as compared to the lowest two groups (figure 4).

Table 5. Cox regression analysis on EGFR amplification for overall survival in 1st_{line TKI treated EGFR mutant NSCLC}

patients.

Approach Factor Significance Hazard Ratio 95% Confidence Interval Lower Upper Internal Comparison ratio 0.047 1.134 1.001 1.285 z score 0.015 1.009 1.002 1.015 Interaction* 0.052 0.997 0.995 1.000 Normal Comparison ratio 0.057 1.016 1.000 1.033 z score 0.174 1.002 0.999 1.004 interaction 0.058 1.000 1.000 1.000

* indicates interaction between ratio and z score.

Figure 3. Kaplan Meier plots of EGFR mutant NSCLC patients treated with 1st_{line EGFR‐TKIs.}_{a) OS for ratio shown in}

interquartile groups. b) OS for z score shown in interquartile groups. First interquartile group represents the lowest

quartile of the ratio or the z score, respectively. Discussion

In this study, we evaluated two approaches for calling gene amplifications by analyzing read counts of the amplicons obtained via an amplicon‐based targeted NGS approach. The NGS‐ based amplification calling strategies were validated by routine diagnostic tests, i.e. FISH for ERBB2 and MET, and MLPA for EGFR. We were able to demonstrate feasibility of our approach, with good validation results especially for calling EGFR amplifications. Subsequently, we looked into the clinical value of our amplification calling by evaluating the predictive value of ratio and z score on survival of a cohort of EGFR mutated patients who were treated with EGFR‐TKI. The

presence of EGFR amplification on top of EGFR mutations was associated with shorter OS as compared to patients without EGFR amplification.

In this study, we specifically focused on the use of diagnostic amplicon‐based NGS data for calling EGFR amplifications. The use of diagnostic data for amplification calling enables broad implementation, as targeted NGS data are available for most advanced stage NSCLC patients. Several studies have analyzed the presence of EGFR amplifications in EGFR mutant and/or EGFR wild type NSCLC patients using FISH, CISH Southern blot, or MLPA [9‐11,24‐28]. In three studies with a Caucasian population cohort, the incidence of EGFR amplification was 6%, 7.9% and even 29.6% [9,10,26]. In two studies from an Asian cohort, it was 3.5% and 32% [11,25]. In EGFR wild type cases, amplification percentages were 2%, 3%, and 22% in the three Caucasian cohorts, and 1% and 29% in the two Asian cohorts. In EGFR mutant cases, amplification percentages were usually higher, ranging in different studies from 8% to 81%. Of note, these studies applied different techniques (e.g., FISH, MLPA, Southern Blot, Chromogenic in situ hybridization and dual in situ hybridization) and used different cut‐off values to define amplifications. The high percentages were reported in studies that were less strict for setting amplification cut‐off values, e.g. ≥3 or ≥4 EGFR copies / cell [10,11,24,27,28]. One study with relatively strict criteria for amplification estimation showed EGFR amplifications in 19% of EGFR mutant patients. Their criterion was >5 EGFR copies / signals per cell as a cut‐off value [26]. We used a ratio cut‐off of 3, which is in the same range as the stricter study [26].

In our study, both the positive and negative predictive value of the EGFR amplification validation was high, indicating feasibility of the approach. Six out of 35 samples were positive of EGFR amplification by both NGS approaches but negative by MLPA. For part of the samples, we did observe gain by MLPA, including one sample with a cutoff value close to the cutoff value. The negative predictive value for ERBB2 and MET amplification was high, yet we cannot draw a firm conclusion on the positive predictive value due to limited number of positive samples. For EGFR, NGS‐based amplification positive and negative results are highly reliable, although its clinical value needs to be further established in larger patient cohorts. The poorer performance for ERBB2 and MET in comparison to EGFR might also lie in the fact that amplification scoring by FISH is generally done on a limited number of tumor cells, whereas the NGS based approach gives a result based on the bulk of the tumour mass. Validation of EGFR by MLPA was done on the same DNA sample as used for NGS, with both approaches being based on a bulk analysis. In our study we showed an effect of EGFR gene amplification on OS. Patients with both mutation and amplification had a worse survival as compared to patients with EGFR mutation without amplification. Among the above‐mentioned studies, only two compared survival of EGFR amplification positive and negative groups in EGFR mutant NSCLC patients treated with EGFR‐TKIs. In contrast to our findings, both studies reported EGFR amplification as a favorable factor for survival compared with those without amplification. In the cohort with Asian patients, the EGFR amplification group (n=41) had a median PFS of 16 months compared with median PFS of 9 months in the group without EGFR amplifications (n=45), p<0.01 [24]. In a Latino cohort, the median PFS of the EGFR amplification group (n=22) was 28.5 months versus 11.0 months in the no amplification group (n=50) (p=0.002). The OS was 37.8 months versus 27.1 months (p=0.009) [28]. Besides presence of EGFR amplification, another potential predictive biomarker is the MAF of the EGFR mutation. A high MAF of EGFR L858R and/or EGFR E19Del predicted

(15)

better PFS (26.3 m vs. 12.3 m, p=0.0155) in a cohort of 77 patients [12]. In two other studies, a high MAF of the EGFR L858R mutation was shown to be a favorable factor for PFS (9.5 m vs. 3.0 m, p=0.0027 and 12.3 m vs. 2 m, p<0.001) [13,14]. A high MAF might be associated with an increase in copy number, but this was not further investigated. Cut‐off values for MAF varied among the studies and no correction was done for the tumor cell percentages in the tissue samples. Despite these differences in methodology, we do not have a clear explanation for the differences in survival with our study. In the published studies the authors proposed a higher dependency of the tumor cells to EGFR in amplification positive cases as a possible explanation for the improved survival. An alternative explanation for our finding might be that a poor response in amplification positive cases is caused by the higher EGFR expression in cases with amplifications, which might require higher TKI‐doses to effectively inhibit EGFR [29,30]. Alternatively it might be hypothesized that cases with amplifications have a higher chance of developing resistant mutations in one of the EGFR mutant copies.

Limitations of our study lie in setting cut‐off values during the process of data analysis. For example, a median coverage per sample of 50 was chosen as cut‐off for sample quality. When selecting internal reference amplicons, we included the most stable 25% amplicons based on CV values. When evaluating the correlation of each amplicon with the other amplicons in the same gene, an R square cut‐off value of 0.5 was taken. A deficiency of our limited panel is that we cannot differentiate amplification from polysomy, due to the limited number of genes per chromosome in the panel. The number of EGFR mutation‐positive patients with available clinical data for the first line TKI treatment was limited and needs to be expanded to draw firm conclusions. We have shown that EGFR amplification can be determined using PCR‐target enrichment‐based NGS data. Overall read count normalization by the internal comparison approach was more reliable than the comparison approach using normal samples. This might be related to batch wise analysis of the normal samples, whereas tumor samples were retrieved over a period of three years.

One of the advantages of our study is that the patient cohort is derived from a relatively stable population over time from the northern Netherlands. According to the literature, the frequency of EGFR mutations in northern Netherlands is 10% (38 out of 368) in lung cancer patients [31]. In the cohort of our study, the EGFR mutation frequency is 8.4% (134 out of 1586).

In conclusion, we recommend to always validate NGS based amplification positive cases of ERBB2 and MET by FISH. For EGFR, the NGS‐based amplification calling seems to be robust and also have clinical value as shown in the OS analysis. Amplicon‐based targeted NGS data can be used to estimate presence of EGFR amplifications and can be used as predictive biomarker in EGFR mutant patients treated with EGFR‐TKIs.

Funding

This work was supported by KWF grant (RUG2015‐8044) and the University Medical Centre Groningen.

Conflicts of Interest

The authors declare that they have no competing interests. Acknowledgement

(16)

3

better PFS (26.3 m vs. 12.3 m, p=0.0155) in a cohort of 77 patients [12]. In two other studies, a high MAF of the EGFR L858R mutation was shown to be a favorable factor for PFS (9.5 m vs. 3.0 m, p=0.0027 and 12.3 m vs. 2 m, p<0.001) [13,14]. A high MAF might be associated with an increase in copy number, but this was not further investigated. Cut‐off values for MAF varied among the studies and no correction was done for the tumor cell percentages in the tissue samples. Despite these differences in methodology, we do not have a clear explanation for the differences in survival with our study. In the published studies the authors proposed a higher dependency of the tumor cells to EGFR in amplification positive cases as a possible explanation for the improved survival. An alternative explanation for our finding might be that a poor response in amplification positive cases is caused by the higher EGFR expression in cases with amplifications, which might require higher TKI‐doses to effectively inhibit EGFR [29,30]. Alternatively it might be hypothesized that cases with amplifications have a higher chance of developing resistant mutations in one of the EGFR mutant copies.

Limitations of our study lie in setting cut‐off values during the process of data analysis. For example, a median coverage per sample of 50 was chosen as cut‐off for sample quality. When selecting internal reference amplicons, we included the most stable 25% amplicons based on CV values. When evaluating the correlation of each amplicon with the other amplicons in the same gene, an R square cut‐off value of 0.5 was taken. A deficiency of our limited panel is that we cannot differentiate amplification from polysomy, due to the limited number of genes per chromosome in the panel. The number of EGFR mutation‐positive patients with available clinical data for the first line TKI treatment was limited and needs to be expanded to draw firm conclusions. We have shown that EGFR amplification can be determined using PCR‐target enrichment‐based NGS data. Overall read count normalization by the internal comparison approach was more reliable than the comparison approach using normal samples. This might be related to batch wise analysis of the normal samples, whereas tumor samples were retrieved over a period of three years.

One of the advantages of our study is that the patient cohort is derived from a relatively stable population over time from the northern Netherlands. According to the literature, the frequency of EGFR mutations in northern Netherlands is 10% (38 out of 368) in lung cancer patients [31]. In the cohort of our study, the EGFR mutation frequency is 8.4% (134 out of 1586).

In conclusion, we recommend to always validate NGS based amplification positive cases of ERBB2 and MET by FISH. For EGFR, the NGS‐based amplification calling seems to be robust and also have clinical value as shown in the OS analysis. Amplicon‐based targeted NGS data can be used to estimate presence of EGFR amplifications and can be used as predictive biomarker in EGFR mutant patients treated with EGFR‐TKIs.

Funding

This work was supported by KWF grant (RUG2015‐8044) and the University Medical Centre Groningen.

Conflicts of Interest

The authors declare that they have no competing interests. Acknowledgement

This work was supported by the University Medical Centre Groningen. We would like to acknowledge the UMCG molecular diagnostic team in the Pathology Department for technical assistance with the experimental work. We thank the UG Center for Information Technology and their sponsors BBMRI‐NL & TarGet for storage and compute infrastructure. We would like to thank the Exome Aggregation Consortium and the groups that provided exome variant data for comparison. A full list of contributing groups can be found at http://exac.broadinstitute.org/about.

(17)

References

1. Hirsch, F.R.; Suda, K.; Wiens, J.; Bunn, P.A. New and emerging targeted treatments in advanced non‐small‐ cell lung cancer. The Lancet 2016, 388, 1012‐1024.

2. Hsiao, S.J.; Zehir, A.; Sireci, A.N.; Aisner, D.L. Detection of tumor ntrk gene fusions to identify patients who may benefit from tyrosine kinase (trk) inhibitor therapy. The Journal of Molecular Diagnostics 2019, 21, 553‐571. 3. ALovly, C.; Horn, L.; Pao, W. Molecular profiling of lung cancer. My cancer genome https://www.Mycancergenome.Org/content/disease/lung‐cancer/ (updated march 16). My Cancer Genome https://www.mycancergenome.org/content/disease/lung‐cancer/ (Updated March 16) 2018.

4. Yarden, Y. The egfr family and its ligands in human cancer: Signalling mechanisms and therapeutic opportunities. Eur. J. Cancer 2001, 37, S3‐S8.

5. Ladanyi, M.; Pao, W. Lung adenocarcinoma: Guiding egfr‐targeted therapy and beyond. Modern Pathol 2008, 21, S16.

6. Ettinger, D.S.; Wood, D.E.; Aisner, D.L.; Akerley, W.; Bauman, J.; Chirieac, L.R.; D'Amico, T.A.; DeCamp, M.M.; Dilling, T.J.; Dobelbower, M., et al. Non–small cell lung cancer, version 5.2017, nccn clinical practice guidelines in oncology. 2017, 15, 504.

7. Marchetti, A.; Martella, C.; Felicioni, L.; Barassi, F.; Salvatore, S.; Chella, A.; Camplese, P.P.; Iarussi, T.; Mucilli, F.; Mezzetti, A., et al. Egfr mutations in non–small‐cell lung cancer: Analysis of a large series of cases and development of a rapid and sensitive method for diagnostic screening with potential implications on pharmacologic treatment. Journal of Clinical Oncology 2005, 23, 857‐865.

8. Shan, L.; Wang, Z.; Guo, L.; Sun, H.; Qiu, T.; Ling, Y.; Li, W.; Li, L.; Liu, X.; Zheng, B. Concurrence of egfr amplification and sensitizing mutations indicate a better survival benefit from egfr‐tki therapy in lung adenocarcinoma patients. Lung cancer 2015, 89, 337‐342.

9. Lewandowska, M.A.; Czubak, K.; Klonowska, K.; Jozwicki, W.; Kowalewski, J.; Kozlowski, P. The use of a two‐tiered testing strategy for the simultaneous detection of small egfr mutations and egfr amplification in lung cancer. PloS one 2015, 10, e0117983.

10. Miller, V.A.; Riely, G.J.; Zakowski, M.F.; Li, A.R.; Patel, J.D.; Heelan, R.T.; Kris, M.G.; Sandler, A.B.; Carbone, D.P.; Tsao, A. Molecular characteristics of bronchioloalveolar carcinoma and adenocarcinoma, bronchioloalveolar carcinoma subtype, predict response to erlotinib. Journal of Clinical Oncology 2008, 26, 1472‐1478.

11. Morinaga, R.; Okamoto, I.; Fujita, Y.; Arao, T.; Sekijima, M.; Nishio, K.; Ito, H.; Fukuoka, M.; Kadota, J.i.; Nakagawa, K. Association of epidermal growth factor receptor (egfr) gene mutations with egfr amplification in advanced non‐small cell lung cancer. Cancer science 2008, 99, 2455‐2460.

12. Hung, M.‐S.; Lung, J.‐H.; Lin, Y.‐C.; Fang, Y.‐H.; Hsieh, M.‐J.; Tsai, Y.‐H. The content of mutant egfr DNA correlates with response to egfr‐tkis in lung adenocarcinoma patients with common egfr mutations. Medicine 2016, 95, e3991.

13. Ono, A.; Kenmotsu, H.; Watanabe, M.; Serizawa, M.; Mori, K.; Imai, H.; Taira, T.; Naito, T.; Murakami, H.; Nakajima, T., et al. Mutant allele frequency predicts the efficacy of egfr‐tkis in lung adenocarcinoma harboring the l858r mutation. Annals of Oncology 2014, 25, 1948‐1953.

14. Li, X.; Cai, W.; Yang, G.; Su, C.; Ren, S.; Zhao, C.; Hu, R.; Chen, X.; Gao, G.; Guo, Z., et al. Comprehensive analysis of egfr‐mutant abundance and its effect on efficacy of egfr tkis in advanced nsclc with egfr mutations. Journal of Thoracic Oncology 2017, 12, 1388‐1397.

15. Li, A.R.; Chitale, D.; Riely, G.J.; Pao, W.; Miller, V.A.; Zakowski, M.F.; Rusch, V.; Kris, M.G.; Ladanyi, M. Egfr mutations in lung adenocarcinomas: Clinical testing experience and relationship to egfr gene copy number and immunohistochemical expression. The Journal of molecular diagnostics : JMD 2008, 10, 242‐248.

16. Johansson, L.F.; van Dijk, F.; de Boer, E.N.; van Dijk‐Bos, K.K.; Jongbloed, J.D.H.; van der Hout, A.H.; Westers, H.; Sinke, R.J.; Swertz, M.A.; Sijmons, R.H., et al. Convading: Single exon variation detection in targeted ngs data. Human Mutation 2016, 37, 457‐464.

17. Eijkelenboom, A.; Tops, B.B.J.; van den Berg, A.; van den Brule, A.J.C.; Dinjens, W.N.M.; Dubbink, H.J.; Ter Elst, A.; Geurts‐Giele, W.R.R.; Groenen, P.; Groenendijk, F.H., et al. Recommendations for the clinical interpretation and reporting of copy number gains using gene panel ngs analysis in routine diagnostics. Virchows Arch 2019, 474, 673‐680.

(18)