Clinical next-generation sequencing in patients with non–small cell lung cancer
We thank Savita Shrivastava, Nobish Varghese, and Emily Brophy for informatics support.
Abstract
BACKGROUND
A clinical assay was implemented to perform next-generation sequencing (NGS) of genes commonly mutated in multiple cancer types. This report describes the feasibility and diagnostic yield of this assay in 381 consecutive patients with non–small cell lung cancer (NSCLC).
METHODS
Clinical targeted sequencing of 23 genes was performed with DNA from formalin-fixed, paraffin-embedded (FFPE) tumor tissue. The assay used Agilent SureSelect hybrid capture followed by Illumina HiSeq 2000, MiSeq, or HiSeq 2500 sequencing in a College of American Pathologists–accredited, Clinical Laboratory Improvement Amendments–certified laboratory. Single-nucleotide variants and insertion/deletion events were reported. This assay was performed before methods were developed to detect rearrangements by NGS.
RESULTS
Two hundred nine of all requisitioned samples (55%) were successfully sequenced. The most common reason for not performing the sequencing was an insufficient quantity of tissue available in the blocks (29%). Excisional, endoscopic, and core biopsy specimens were sufficient for testing in 95%, 66%, and 40% of the cases, respectively. The median turnaround time (TAT) in the pathology laboratory was 21 days, and there was a trend of an improved TAT with more rapid sequencing platforms. Sequencing yielded a mean coverage of 1318×. Potentially actionable mutations (ie, predictive or prognostic) were identified in 46% of 209 samples and were most commonly found in KRAS (28%), epidermal growth factor receptor (14%), phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit alpha (4%), phosphatase and tensin homolog (1%), and BRAF (1%). Five percent of the samples had multiple actionable mutations. A targeted therapy was instituted on the basis of NGS in 11% of the sequenced patients or in 6% of all patients.
CONCLUSIONS
NGS-based diagnostics are feasible in NSCLC and provide clinically relevant information from readily available FFPE tissue. The sample type is associated with the probability of successful testing. Cancer 2015;121:631–639. © 2014 American Cancer Society.
INTRODUCTION
Cancer is a disease of the cellular genome. Large-scale genomic sequencing studies have clearly demonstrated the molecular heterogeneity and diversity of malignant neoplasms. The availability of targeted, rationally designed therapeutics and the increased availability of tumor genotyping have led to widespread interest in personalized medicine.
The empiric treatment of non–small cell lung cancer (NSCLC) with platinum-based cytotoxic chemotherapy has reached a plateau.1 The discovery of activating epidermal growth factor receptor (EGFR) mutations as predictors of a response to EGFR tyrosine kinase inhibitors revolutionized the landscape of NSCLC treatment.2, 3 More recently, targeted therapies directed at tumor cells harboring anaplastic lymphoma kinase (ALK) fusions and ROS1 fusions have produced impressive results.4, 5 In addition to EGFR, ALK, and ROS1, there are several other abnormalities that could potentially be treated with drugs already approved for other malignancies or investigational agents.6 Molecular analysis to support personalized therapeutics has historically been obtained with several methods.7-9 In current clinical practice, a single-gene testing approach is often used to identify variants (eg, EGFR mutation, KRAS mutation, or ALK break-apart testing) to guide treatment decisions. Serial testing takes time and depletes tumor tissue. In addition, the cost of single-gene methods scales linearly with the number of genes interrogated.
An alternative approach is to pursue whole genome or whole exome tumor sequencing to perform an unbiased search for actionable genetic variation. This approach has several of its own limitations, including a high cost, complexities associated with bioinformatics and reporting, and a low yield of actionable data per nucleotide sequenced. The challenge is to develop genomic assays that are truly clinically oriented and focused on detecting actionable sequence variation.
Targeted next-generation sequencing (NGS) of cancer-related genes permits unbiased variant detection of commonly altered genes on a single platform. Apart from realizing economies of scale in comparison with consecutive single-gene testing, targeted NGS also makes economical use of limited tissue specimens.
Comprehensive genomic characterization of adenocarcinoma and squamous cell carcinoma of the lung has been reported recently with high-quality (high tumor cellularity and limited tumor necrosis), fresh-frozen tumor specimens procured mainly from resected specimens from established cancer centers.10-12 However, the feasibility of NGS in the clinic, in an environment complying with the Clinical Laboratory Improvement Amendments (CLIA) of 1988, and with formalin-fixed, paraffin-embedded (FFPE) specimens has not been well described. The turnaround time (TAT), success rate, and clinical utility of NGS in routine clinical care have not been described to the best of our knowledge. We have implemented a clinical genomics workflow with which we perform targeted NGS of a set of cancer-related genes to detect actionable somatic mutations. We report here the spectrum of results, including laboratory parameters and sequence findings, obtained for the initial cohort of patients.
MATERIALS AND METHODS
Overview of Specimens and Test
Institutional review board approval, including a waiver of consent, was obtained to conduct this retrospective analysis. These data represent analyses of consecutive NSCLC samples submitted to Genomics and Pathology Services at Washington University13 between March 1, 2012 and October 23, 2013. Genomics and Pathology Services is a College of American Pathologists–accredited, CLIA-certified clinical laboratory within an academic pathology department. Tests were requested by medical oncologists. Target-capture sequencing on an Illumina platform was performed across all exons of 23 cancer-related genes: BRAF, CTNNB1, DNMT3A, EGFR, FLT3, IDH1, IDH2, JAK2, KIT, KRAS, MAP2K2, MAPK1, MET, NPM1, NRAS, PDGFRA, PIK3CA, PTEN, PTPN11, RET, RUNX1, TP53, and WT1. The test is described in further detail in the online supporting information.
Variant Annotation and Reporting
Identified sequence variants were initially classified into 1 of 5 levels by the bioinformatics pipeline according to an internally established classification scheme (Table 1 and Supporting Fig. 1 [see online supporting information]). This scheme is stratified by clinical actionability and relies on previously reported genetic data, including sequence variation databases such as the Single Nucleotide Polymorphism Database (version 132),14 the 1000 Genomes project,15 the Exome Variant Server of the National Heart, Lung, and Blood Institute,16 and the Catalogue of Somatic Mutations in Cancer somatic mutation database17 as well as a locally curated clinical-grade database of variants and corresponding interpretations. The classification of each variant was further reviewed by the clinical genomicist (pathologist) signing out each report. Variants were reported according to Human Genome Variation Society nomenclature.18 For the purposes of this publication, a list of all identified variants was extracted from the laboratory information system and manually curated (I.S.H. and C.M.L.) to resolve inconsistencies and remove duplicates. Although 2 different gene sets were tested, we report here only the 23 genes that were sequenced for all tumors.
| Variant Level | Significance |
|---|---|
| 1 | Potentially clinically actionable (predictive or prognostic) in patient's cancer type |
| 2 | Potentially clinically actionable (predictive or prognostic) in another cancer type |
| 3 | Previously reported as a somatic variant in cancer (COSMIC or other database) or reported as clinically significant in another disease |
| 4 | Novel variant of unknown significance; rare and private polymorphisms (<1% minor allele frequency) included in this group |
| 5 | Known polymorphism (≥1% prevalence in any population as recorded in dbSNP or EVS) |
- Abbreviations: COSMIC, Catalogue of Somatic Mutations in Cancer; dbSNP, Single Nucleotide Polymorphism Database; EVS, Exome Variant Server.
Workflow and turnaround time for clinical genomics samples. (A) Workflow labeled with definitions of key turnaround time metrics. (B) Histogram of the preanalytic time measured in calendar days. (C) Histogram of the turnaround time from the start of DNA extraction to report sign-out (laboratory time) measured in calendar days. Samples for which a variant was confirmed by Sanger sequencing of tumor DNA are shown as hatched bars, whereas samples without Sanger confirmation are shown as filled bars. Hatched and filled bars together add up to 100%. (D) Histogram of the laboratory time as a function of the sequencing platform.
TAT
A CoPathPlus laboratory information system (Cerner Corp., Kansas City, Mo) was used for specimen tracking. Cases were scanned to track their status as they progressed through the steps of the laboratory workflow. Time tracking data were extracted via a query. After DNA extraction, some cases were held for a number of days while authorization or predetermination from payers was awaited. These hold days were subtracted from the laboratory time if no other laboratory steps were in progress at that time.
Statistics
Descriptive statistics were calculated with Excel 2010 (Microsoft Corp, Redmond, Wash). Tests were performed in Prism 6 (GraphPad Software, San Diego, Calif).
RESULTS
Feasibility of Sequencing in the Clinic With FFPE Samples
A total of 381 FFPE samples from patients with newly diagnosed or recurrent NSCLC were submitted for clinical NGS during the study period. Patients had newly diagnosed metastatic lung cancer (67%), persistent/progressive disease (20%), or recurrent/metastatic disease after treatment with curative intent (8%; Supporting Table 3 [see online supporting information]). For the remainder of the cases, the exact clinical scenario was undetermined. Samples were obtained by biopsy or resection from a variety of sources, including lungs, lymph nodes, or an extranodal distant site of metastasis. In 172 of 381 samples (45%), NGS could not be performed for a variety of reasons (Table 2), including insurance denial or an inability to procure paraffin blocks (15%), insufficient tissue material at intake according to validated metrics in the laboratory (neoplastic cellularity and viability; 21%), and inadequate DNA after extraction (8%). Finally, for a small number of samples (2%), the DNA met quality and quantity cutoffs, but postsequencing analysis indicated that the data did not meet our validated laboratory standards for library complexity.19
| Cases accessioned, n | 381 |
| NGS testing completed, n/N (%) | 209/381 (55) |
| NGS testing not completed, n/N (% of accessioned cases) | 172/381 (45) |
| Study cancelleda | 56/381 (15) |
| Insufficient tissue at intake (ie, at pathologist review) | 78/381 (21) |
| Insufficient DNA quality or quantity after extraction | 32/381 (8) |
| Failed library preparation | 6/381 (2) |
| Overall rate of NGS success according to specimen type, n/N (% of accessioned cases)b | |
| Core biopsy | 40/100 (40) |
| Endoscopic biopsy | 47/71 (66) |
| Excision | 99/104 (95) |
| Fluid | 1/3 (33) |
| FNA | 11/33 (33) |
| Incisional biopsy | 11/14 (79) |
- Abbreviations: FNA, fine-needle aspiration; NGS, next-generation sequencing.
- a In the majority of cases for which the study was cancelled by the ordering clinician, the reason was a denial of insurance coverage or an inability of the laboratory to locate tissue blocks. These are recorded together because, in both cases, the test was cancelled for nonlaboratory reasons.
- b Cancelled cases have been excluded.
Impact of Specimen Type on Feasibility
We examined the possibility that certain specimen types were more suitable than others for NGS testing. For primary (lung) tumor tissue, excision specimens led to the completion of NGS testing in 97% of samples (Table 3), whereas endoscopic biopsies yielded results for only 55%, and core biopsies were sufficient for 31%. The probability of successful test completion for primary tumors was significantly different across specimen types (P < .0001 by the chi-square test).
| Failed Cases | Successful Cases | Total Cases | |||||
|---|---|---|---|---|---|---|---|
| Insufficient Tissue at Intake, % | Insufficient DNA Yield, % | Library Preparation Failure, % | Tests Completed Successfully, % | DNA Yield, Mean (Range), µg | Mean Unique Coverage | ||
| I. Test performed on primary tumor tissue | |||||||
| Core biopsy | 45 | 24 | — | 31 | 0.69 (0.12-2.44) | 1063 | 49 |
| Endoscopic biopsya | 34 | 6 | 4 | 55 | 2.13 (0.10-9.37) | 1358 | 47 |
| Resectionb | 3 | — | — | 97 | 8.44 (0.24-39.33) | 1306 | 58 |
| FNA | 44 | 19 | 6 | 31 | 0.78 (0.14-1.40) | 970 | 16 |
| II. Test performed on a tissue metastasis | |||||||
| Core biopsy | 34 | 20 | 2 | 43 | 1.30 (0.11-4.68) | 1246 | 44 |
| Endoscopic biopsya | 13 | — | — | 87 | 2.80 (0.25-7.6) | 1100 | 8 |
| Resectionb | 5 | 5 | 5 | 85 | 11.58 (0.49-54.90) | 1414 | 19 |
| Fluid | 67 | — | — | 33 | 1.28 | 961 | 3 |
| FNA | 50 | 50 | — | — | — | — | 2 |
| Incisional biopsy | — | 14 | 7 | 79 | 6.85 (1.44-20.79) | 1446 | 14 |
| III. Test performed on a lymph node metastasis | |||||||
| Core biopsy | 14 | — | — | 86 | 1.08 (0.31-2.17) | 1515 | 7 |
| Endoscopic biopsya | 13 | — | — | 87 | 2.82 (0.23-8.67) | 1544 | 15 |
| Excisional biopsy | — | — | — | 100 | 10.39 (0.10-38.61) | 1420 | 27 |
| FNA | 53 | 7 | — | 40 | 0.83 (0.29-1.36) | 1266 | 15 |
- Abbreviation: FNA, fine-needle aspiration.
- Cases for which the study was cancelled by the ordering clinician have been omitted. The data in each row are expressed as percentages of the total cases in that row except for the percentage of completed cases with low-input DNA, which is expressed as a fraction of successfully sequenced cases. Blank cells represent zero occurrences.
- a Includes small biopsy specimens obtained by bronchoscopy, mediastinoscopy, and thoracoscopy.
- b Includes wedge biopsy specimens obtained by video-assisted thoracic surgery.
Specimens were also obtained from various other anatomic sites, including nodal metastases (mediastinal, axillary, supraclavicular, and distant) and distant metastatic sites (including pleura, diaphragm, liver, brain, and skin). Among nodal metastasis specimens, excisional biopsy samples (including nodes obtained by mediastinoscopy) were successfully sequenced 100% of the time, whereas other specimen types had differing success rates (Table 3; P = .0012). Among tissue metastases, the most frequently successful specimen types were endoscopic biopsies, resections, and incisional biopsies (Table 3), with significant differences between types (P < .0001).
Total and Laboratory TAT
Clinically relevant targeted sequencing requires results to be reported in a clinically useful timeframe. We tracked TAT by retrieving specimen status data from the laboratory information system. Figure 1A shows key steps in the progress of a generic sequencing case, culminating in reporting to the patient's electronic medical record. The median preanalytic time, including delays related to obtaining tissue from outside institutions, was 7 days (range, 1-63 days; Fig. 1B). The median TAT in the molecular pathology laboratory, from the start of DNA extraction to sign-out, was 21 days (range, 9-51 days; Fig. 1C). Although some cases required orthogonal (Sanger) confirmation of certain variants, this was not a major determinant of TAT (Fig. 1C). The platform used for sequencing was a much more meaningful determinant of TAT (Fig. 1D), with major decrements in TAT realized as successive generations of Illumina sequencers were implemented (with medians of 33 days on HiSeq 2000, 22 days on MiSeq, and 18 days on HiSeq 2500).
Sequencing Quality Metrics
For samples that were successfully sequenced, DNA was extracted from 1 or more 1-mm cores of paraffin-embedded tissue. DNA extraction performed according to our standard workflow yielded a mean of 5.7 µg DNA per tumor, which was well in excess of the 1 µg called for by our standard protocol and the 150-ng cutoff established for low-input sequencing. As expected, the DNA yield varied according to the specimen type, with excisional specimens giving greater quantities of DNA than minimally invasive biopsies (Table 3).
Downstream library preparation and sequencing were initiated with 1 µg of DNA or the amount available if that was 0.15 to 1 µg, as described in the Materials and Methods section. The HiSeq 2000 instrument yielded a mean of 5.57 × 106 unique on-target reads (standard deviation, 1.8 × 106), the HiSeq 2500 yielded 5.38 × 106 (standard deviation, 1.4 × 106), whereas the MiSeq yielded 2.3 × 106 (standard deviation, 8.0 × 105). The mean unique depth of coverage across the capture region was 1318× (1838× for HiSeq 2000, 1144× for HiSeq 2500, and 1141× for MiSeq). Quality control metrics were established by the laboratory,20 and samples failing these metrics were excluded (Table 3).
Overview of Identified Nonsynonymous Variants
Sequencing and analysis were completed for 209 specimens encompassing various histologic types of NSCLC (Supporting Table 3 [see online supporting information]). Across the entire series, a total of 1164 nonsynonymous variants were called within the coding regions of the 23 sequenced genes, and they included 744 known polymorphisms and 420 nonpolymorphic variants. There were a mean of 2.01 nonpolymorphic, nonsynonymous variants per specimen. The nonpolymorphic, nonsynonymous variants included 381 substitutions, 27 deletions, 6 insertions, and 4 combined insertion-deletion events (Supporting Fig. 2A [see online supporting information]). Ninety percent of the samples had at least 1 variant that was not a polymorphism (levels 1-4). After the exclusion of polymorphisms, the most frequently mutated genes were TP53 (138 variants in 120 samples), KRAS (60 variants in 59 samples), and EGFR (46 variants in 40 samples; Supporting Fig. 2B [see online supporting information]). Twenty-one percent of the 420 total nonpolymorphic, nonsynonymous variants were classified as level 1, and 4% were classified as level 2; together, these are the levels used to denote potentially actionable variants as judged by the clinical genomicist (Supporting Fig. 2C [see online supporting information]).
(A) High-level view of sequencing results showing the distribution of cases according to the most significant variant that was detected. (B) Distribution of detected actionable mutations (level 1 or 2) according to the gene in which they fall. (C) Heat map describing somatic mutations identified in each case. Each column represents 1 sample; each row represents 1 gene. Potentially actionable mutations (classified as level 1 or 2) are shown in red. Nonactionable somatic mutations (classified as level 3 or 4) are shown in blue. Only genes that were sequenced for all cases and in which a presumed somatic mutation was detected in 1 or more samples are shown. The column on the right (n) indicates the number of samples with an actionable mutation. COSMIC indicates Catalogue of Somatic Mutations in Cancer; VUS, variant of uncertain significance.
Yield of Actionable Variants
A major objective of NGS in the clinic is to identify samples with potentially actionable variants (defined in Table 1). Forty-two percent of the samples able to be sequenced had a level 1 variant, that is, a variant recognized by the clinical genomicist as being either prognostically significant or predictive of a therapeutic response in the patient's tumor type. Seven percent of the samples had a level 2 variant, that is, a variant considered predictive or prognostic in another tumor type. Because targeted therapies are often chosen on the basis of molecular profiles rather than tumor histology, it may be most relevant to consider level 1 and 2 variants together. By this measure, at least 1 potentially actionable variant (level 1 or 2) was identified in 46% of the samples that were able to be sequenced (Fig. 2A).
Actionable mutations were most commonly seen in KRAS (28% of samples sequenced), EGFR (14%), PIK3CA (4%), PTEN (1%), and BRAF (1%; Fig. 2B,C). Within KRAS, 98% of actionable mutations were at codon 12 or 13; an insertion-deletion at codons 19 and 20 was also annotated as actionable (Table 4). In EGFR, 10% of actionable mutations were in exon 18, 36% were in exon 19, 13% were in exon 20, and 43% were in exon 21.
| Potentially Actionable Mutations Identified, n (%) | ||||||
|---|---|---|---|---|---|---|
| Adenocarcinoma (n = 147) | Large Cell Neuroendocrine (n = 4) | Poorly Differentiated (n = 9) | Sarcomatoid (n = 6) | Squamous Cell (n = 36) | All histologies (n = 209) | |
| KRAS | 51 (35) | 1 (25) | 2 (22) | 2 (33) | 3 (8) | 59 (28) |
| Codon 12 | 48 | 1 | 2 | 2 | 3 | 56 |
| Codon 13 | 2 | 2 | ||||
| L19_T20delinsFS | 1 | 1 | ||||
| EGFR | 27 (18) | 1 (11) | 2 (6) | 30 (14) | ||
| Exon 18 | ||||||
| E709_T710delinsD | 2 | 2 | ||||
| G719S | 1 | 1 | ||||
| Exon 19 | ||||||
| E746_A750del | 4 | 1 | 1 | 6 | ||
| E746_S752delinsV | 2 | 2 | ||||
| E746_T751delinsI | 1 | 1 | ||||
| E746V | 1 | 1 | ||||
| L747_A750delinsP | 1 | 1 | ||||
| Exon 20 | ||||||
| D770_N771insN | 1 | 1 | ||||
| D770_N771insNP | 1 | 1 | ||||
| D770_P772dup | 1 | 1 | ||||
| T790M | 1 | 1 | ||||
| Exon 21 | ||||||
| L833V | 1 | 1 | ||||
| L858R | 11 | 11 | ||||
| L861Q | 1 | 1 | ||||
| PIK3CA | 6 (4) | 1 (11) | 1 (17) | 1 (3) | 9 (4) | |
| E542K | 1 | 1 | 2 | |||
| E545K | 3 | 1 | 4 | |||
| H1047L | 1 | 1 | ||||
| H1047R | 1 | 1 | 2 | |||
| BRAF | 2 (1) | 2 (1) | ||||
| D594G | 1 | 1 | ||||
| V600E | 1 | 1 | ||||
| PTEN | 1 (1) | 1 (3) | 2 (1) | |||
| F81Y | 1 | |||||
| R233* | 1 | 1 | ||||
| V290I | 1 | 1 | ||||
| JAK2 | 1 (1) | 1 (0) | ||||
| V617F | 1 | 1 | ||||
| KIT | 1 (1) | 1 (0) | ||||
| V526I | 1 | 1 | ||||
| NRAS | 1 (3) | 1 (0) | ||||
| Q61H | 1 | 1 | ||||
- Abbreviations: EGFR, epidermal growth factor receptor; JAK2, Janus kinase 2; PIK3CA, phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit alpha; PTEN, phosphatase and tensin homolog.
- Mutations are listed as amino acid changes in the Human Genome Variation Society protein nomenclature. At the gene level, the percentage of samples with a variant is also stated in parentheses. Histologies for which no potentially actionable variant was detected in any sample have been omitted.
One rationale for using NGS instead of single-gene testing is the possibility that actionable variants are present in more than 1 gene or pathway, and this renders a multigene approach more efficient. In addition, clinical characteristics are not sufficiently reliable to predict the order in which genes can be selected for sequential evaluation. In our series of 209 samples, 10 samples had 2 actionable variants, and 1 sample had 3 (Supporting Table 4). In 7 samples, these actionable variants were present in more than 1 gene. These are samples in which a single-gene approach might have missed therapeutically relevant information.
Another measure of the clinical significance of NGS testing is the institution of matched therapy. Patients received a mean of 1.25 lines of therapy (standard deviation, 1.20 lines) after NGS testing. This therapy was specifically matched to the results of NGS testing in 22 of the patients sequenced (11%) or in 6% of the patients for whom testing was initially requested (Supporting Table 5 [see online supporting information]). Patients receiving matched therapy included 53% of those with sensitizing EGFR mutations (Supporting Table 5 [see online supporting information]).
One hundred eighty-six of the 209 samples sequenced in this series underwent concurrent fluorescence in situ hybridization (FISH) testing for rearrangements or copy number variants in EGFR, MET, ROS, ALK, and/or ERBB2 (data not shown). In most of the remaining 23 samples, there was insufficient tissue to permit both NGS and FISH. Another 8 patients were placed on matched therapy as a result of concurrent FISH testing (data not shown).
DISCUSSION
We report here summary statistics for the initial 381 NSCLC samples submitted for clinical NGS at our institution. The assay consists of targeted sequencing of a panel of cancer-related genes in which somatic mutations have been shown to be prognostically significant or to predict a response to a therapy. The test focuses on nonsynonymous somatic mutations because validated predictive and prognostic variants at present fall into this category. In the test described herein, rearrangements were not detected by NGS, although this capability has now entered the clinical space in our laboratory and others.21, 22
TAT is a key aspect of clinical feasibility (Fig. 1) but has not previously been well documented for a clinical NGS cancer test. Several factors affecting TAT were noted. Preanalytic delays were related to insurance authorization and logistics required to obtain paraffin blocks and slides from outside laboratories. DNA was normally extracted and held while insurance authorization was awaited to minimize the subsequent processing time. These realistic delays are minimized in reports from laboratories that require upfront submission of paraffin blocks or do not seek third-party reimbursement. Diverse factors also affected in-laboratory TAT. Some samples required multiple attempts at library preparation; samples were sequenced on 3 platforms with differing time requirements (11 days for HiSeq 2000, 27 hours for HiSeq 2500, and 24 hours for MiSeq); some samples required Sanger sequencing for orthogonal confirmation of certain NGS-detected variants (particularly insertion/deletions); some samples showed variants that required laborious interpretation; and there existed some amount of interpathologist variation in sign-out practices. TAT improved as successive platforms were implemented, but our experience indicates that as the sequencing time falls, overall TAT comes to be dominated by human factors such as variant interpretation and report sign-out. The time required to interpret whole-gene sequence data is necessarily longer than that needed for hotspot genotyping because of the wider range of variations detected.
This case series is significant for several reasons. This is only the second report, to the best of our knowledge, to document the feasibility, TAT, and impact of NGS in the clinical practice of lung cancer. This report complements that of the Lung Cancer Mutation Consortium,23 which studied a larger number of patients but used a more targeted multiplatform sequencing approach. Other previous reports have not focused on feasibility,24, 25 have not involved an NGS approach,23-25 have occurred outside the CLIA setting,26-28 and/or have used banked or highly controlled specimens.25, 29 Furthermore, the present series addresses the significance of varying specimen sites and types, and it is larger than all but 2 of the aforementioned studies.23, 24
This report underlines the importance of central pathologist review of submitted tissue at the time of intake. As many as 29% of the samples were flagged as insufficient before library preparation or sequencing (Table 2); without this step, many of these samples would have been processed, and this would incurred additional expenses and resulted in either no data or data with little clinical meaning. We believe that this report presents the true success rate of NGS in the clinical setting with archived material obtained from diagnostic procedures. Notably, some laboratories require a review of NGS samples to be performed by the submitting pathologist. These laboratories could experience unrealistically elevated success rates because the smallest samples, which are most likely to fail, are excluded before accessioning.25
At least 1 potentially actionable variant was identified in 46% of all samples submitted for sequencing (Fig. 2). In addition to detecting well-known actionable variants, assays based on complete gene sequencing (as opposed to targeted testing for specific single-nucleotide variants) have the potential to reveal noncanonical variants that may potentially be actionable. Examples detected in this series include BRAF G596R and D594G and NRAS Q61H (Table 4).
In the assay described herein, only single-nucleotide variants, insertions, and deletions were called by NGS, yet structural variants and copy number variants involving EGFR, ALK, ROS, and ERBB2, among others, are clinically important in NSCLC. It is now possible to identify translocations by target-capture NGS,21, 22 and the current iteration of the assay does report rearrangements in ALK detected by NGS. Detection of copy number variation is more difficult in target-capture NGS data, but approaches are under investigation.22, 30 The lower yield of actionable variants in our report (22% of squamous cell carcinomas) versus the Cancer Genome Atlas findings (69%) is largely due to the fact that the latter report incorporated copy number variants.10 The spectrum of mutations reported here is specific to the assay as currently structured and reflects the mix of tumor subtypes that have been submitted to date.
The test has been available for too short a time to allow for comprehensive clinical follow-up, but in the future, it will be important to confirm that treatment decisions have been made on the basis of NGS results. The lack of appropriately matched clinical trials and difficulties associated with targeting alterations such as activating KRAS mutations (Table 4) contribute to the low impact of NGS in the clinic at the present time. The Lung Cancer Mutation Consortium23 reported placing a greater percentage of patients on matched therapy (eg, 83% of those with activating EGFR mutations vs 53% in our series), but that study was explicitly structured to direct patients to a slate of readily available clinical trials, whereas patients in our series were receiving routine clinical care, albeit at an academic tertiary care center. (Supporting Table 5 [see online supporting information]).
In conclusion, NGS-based diagnostics can provide clinically relevant information from readily available FFPE tissue, but both the success rate of sequencing and the utility of the results may be lower in real-world clinical practice versus an experimental setting. Potentially actionable variants were identified in 46% of the samples for which sequencing was completed, and action was taken for 11%. The impact of NGS-based diagnostics is likely to be greater with the increasing availability of novel molecularly targeted agents as well as improved informatics allowing calling of more variant types. The current results, derived from a consecutive set of samples studied at a major tertiary care center, indicate the promise of clinical genomics technologies and provide guidance concerning the results that can be expected in clinical practice. A method must be found to estimate the benefits as well as the costs of NGS-based tumor sequencing. There is also a clear need to develop innovative clinical trials and programs to make sure that genotype-matched drugs are readily available to patients with cancer.
FUNDING SUPPORT
No specific funding was disclosed.
CONFLICT OF INTEREST DISCLOSURES
Shashikant Kulkarni reports personal fees from the National Institute of General Medical Sciences (Coriell), Swift Biosciences, and Bina Technologies outside the submitted work. Rakesh Nagarajan is the chief biomedical informatics officer at PierianDx and reports personal fees and other benefits from PierianDx outside the submitted work; the Clinical Genomicist Workstation that was used to manage and analyze patient data in this work is now exclusively licensed by PierianDx from Washington University for further development and sublicensing. Karen Seibert is a cofounder of PierianDx, a company that licenses NGS software. Maria Baggstrom has been a site principal investigator of clinical trials for Novartis, Merck, Wyeth, ImClone, Boehringer Ingelheim, Merrimack, Eli Lilly, Bristol Myers-Squibb, Endocyte, Astex, and Onyx outside the submitted work. John D. Pfeifer is a cofounder of PierianDx, a company that licenses NGS software. Ramaswamy Govindan reports personal fees from Boehringer Ingelheim, GlaxoSmithKline, Pfizer, Merck, Bayer, Covidien, Bristol Myers-Squibb, Genentech, and Mallinckrodt Pharmaceuticals outside the submitted work.
