Development, Implementation and Assessment of Molecular Diagnostics by Next Generation Sequencing in Personalized Treatment of Cancer: Experience of a Public Reference Healthcare Hospital

The establishment of precision medicine in cancer patients requires the study of several biomarkers. Single-gene testing approaches are limited by sample availability and turnaround time. Next generation sequencing (NGS) provides an alternative for detecting genetic alterations in several genes with low sample requirements. Here we show the implementation to routine diagnostics of a NGS assay under International Organization for Standardization (UNE-EN ISO 15189:2013) accreditation. For this purpose, 106 non-small cell lung cancer (NSCLC) and 102 metastatic colorectal cancer (mCRC) specimens were selected for NGS analysis with Oncomine Solid Tumor (ThermoFisher). In NSCLC the most prevalently mutated gene was TP53 (49%), followed by KRAS (31%) and EGFR (13%); in mCRC, TP53 (50%), KRAS (48%) and PIK3CA (16%) were the most frequently mutated genes. Moreover, NGS identified actionable genetic alterations in 58% of NSCLC patients, and 49% of mCRC patients did not harbor primary resistance mechanisms to anti-EGFR treatment. Validation with conventional approaches showed an overall agreement >90%. Turnaround time and cost analysis revealed that NGS implementation is feasible in the public healthcare context. Therefore, NGS is a multiplexed molecular diagnostic tool able to overcome the limitations of current molecular diagnosis in advanced cancer, allowing an improved and economically sustainable molecular profiling.


Introduction
Cancer is a complex and heterogeneous disease with considerable variation in histological and biological features. Understanding the role of genetic alterations involved in cancer development has led to its reclassification into different molecular subtypes that reflect biological behavior and may lead to further effective therapeutic targets to achieve improved outcome [1,2]. For this purpose, single gene testing approaches are traditionally used to identify individual alterations, currently targeted with approved drugs. However, comprehensive molecular characterization of tumors is hampered by the limited amount of cytology samples and/or formalin-fixed paraffin-embedded (FFPE) tissue biopsies, and by the turnaround time to assess multiple targetable genes and high economic costs. Moreover, the routine methods do not board the mutational co-occurrences, so they do not detect other alterations which in many cases are responsible for disease progression.
The knowledge of a tumor's genetic profile is crucial to improve clinical-decision making in the patient management. Consequently, laboratories must integrate high-throughput sequencing technologies in routine molecular diagnostics [3]. These allow the simultaneous testing of multiple genetic alterations (point mutations, insertions, deletions, copy number variations and translocations) and quantify molecular subclones by procedures that provide accurate, reliable and cost-effective results. In this sense, next-generation sequencing (NGS) has overcome the cited challenges, posing an attractive alternative to traditional molecular diagnostic testing for cancer [4][5][6]. In fact, the College of American Pathologists has suggested the use of expanded panels in its latest guideline [7][8][9] and both the National Comprehensive Cancer Network (NCCN) and the European Society for Medical Oncology (ESMO) have proposed broader molecular profiling to identify rare driver mutations in non-small cell lung cancer (NSCLC) and metastatic colorectal cancer (mCRC) patients, for which effective drugs are already available or under development in clinical trials [10][11][12][13]. However, given the implications of NGS studies on the treatment of cancer patients, the establishment of an internal quality management system is strongly recommended. In this regard, the UNE-EN ISO 15189:2013 accreditation has been recognized as the international standard for quality management systems for all fields in clinical laboratories [14].
The purpose of the current study is to evaluate the integration of NGS technology in a routine clinical setting. We describe the mutational profile of two highly prevalent cancers (advanced NSCLC and mCRC) and analyze its diagnostic potential to characterize molecular heterogeneity and to increase the therapeutic opportunities with targeted therapies; we assess NGS technology at a technical and economical level; and we describe our experience in clinical practice of an NGS pipeline for cancer molecular diagnostics in the UNE-EN ISO 15189:2013 accreditation scope.

Performance Evaluation of Next Generation Sequencing
2.1.1. Next Generation Sequencing Quality Assessment NGS assay was able to detect the seven low frequency variants (between 1-3%) present in two reference materials used as positive controls. The variant allele frequency (VAF) detected was consistent to the data obtained by digital droplet PCR (ddPCR) assays except for EGFR p.Leu858Arg and KRAS p.Gly13Asp mutations in which NGS VAF was slightly higher (5% versus 3% and 4% versus 3%, respectively).
Quality control analysis revealed excellent performance of the NGS panels ( Figure S1). The median of total reads per sample was 347,362 with a median read depth of 3310 reads per amplicon. Uniformity was 97.1% on average and the average "on-target" reads per sample was 92.6%. Moreover, 96.6% of targeted bases showed ≥500 × read depth.
On the other hand, NGS showed an invalid test rate of 3.8% (8/208 FFPE specimens). Six NSCLC samples and two mCRC failed due to low sequencing quality metrics (total reads <100,000). These samples were subsequently excluded from the study.

Comparison of Next Generation Sequencing with Conventional Methods
For ROS1 rearrangements and NRAS mutations the overall agreement (OA) was 100% (Table 1). Regarding EGFR, one specimen showed the p.Thr790Met mutation by conventional methodology which was not reported by NGS (OA; 99.0%). Instead, NGS and Sanger Sequencing (SS) revealed a synonymous change in homozygosis in 787 codon p.(Gln787Gln). NGS technology allowed the detection of an ALK rearrangement not detected by immunohistochemistry (IHQ) or fluorescence in situ hybridization (FISH) (OA: 99.0%). In KRAS, seven specimens gave discordant results when comparing with Real Time (RT)-qPCR assay (OA: 94.7%). BRAF pVal600Glu mutation was detected by High Resolution Melting (HRM) in nine out of eleven NGS BRAF p.Val600Glu mutated samples (OA: 96.4%).

Turnaround Time (TAT) and Cost Comparison
In order to compare NGS with conventional methodologies under theoretical conditions, we calculated the turnaround time and cost for three mandatory testing genes in NSCLC (EGFR, ALK and ROS1) and mCRC (KRAS, NRAS and BRAF). Starting in both cases from FFPE tissue blocks we were able to prepare libraries, sequence eight NSCLC or ten mCRC samples, and analyze data in five working days. Conventional methodologies for molecular testing of EGFR, ALK and ROS1 resulted approximately in three days while testing KRAS, NRAS and BRAF in mCRC resulted approximately in four working days. However, if KRAS is positive, TAT is reduced to three days. TAT and cost comparison between NGS and conventional methods is shown in Table 2.

Clinical Laboratory Accreditation
This thorough validation consisting of reference material analysis, quality control metrics assessment, experimental validation with conventional methodologies and TAT and cost comparison allowed us to include the assay within the scope of the recently granted UNE-EN ISO 15189:2013 accreditation (Entidad Nacional de Acreditación, ENAC, Nº1302/LE2445). Moreover, this NGS assay was externally validated within the European Genetics Quality Network (EMQN) External Quality Assessment Scheme for Oncogene Panel Testing, obtaining satisfactory results in 2017 and 2018 editions [15].  In the NSCLC cohort, the most prevalently mutated gene was TP53 (49%) followed by KRAS (31%), EGFR (13%), BRAF (11%) and PIK3CA (7%). Rearrangements were found in ALK and ROS1 (5% and 1%, respectively). In the entire group, 9% of patients did not carry any somatic mutation; 56% harbored one somatic mutation and 35% two or more ( Figure 1, Table S1).

Concurrent Molecular Alterations
Thirty-five NSCLC patients harbored co-occurring mutations. One patient carried an indel in EGFR exon 19 accompanied by KRAS codon 12 mutation. Three EGFR mutations and one exon 19 deletion appeared in concurrency with TP53 mutations. Two EGFR mutated patients also presented PIK3CA mutations; and one harbored a double EGFR codon 18 mutation in concurrency with CTNNB1 mutation. Among the 31 KRAS mutated samples, two were in concurrency with BRAF mutations; one carried also a mutation in ERBB2 and one sample harbored double KRAS mutations. One sample carrying a NRAS mutation also presented a TP53 frameshift mutation. PIK3CA mutations were found to be concurrent with EGFR, BRAF, TP53 mutations or ALK fusion. We identified two samples with three concurrent mutations. In one sample we identified mutations in PIK3CA, CTNNB1 and TP53. Another patient carried mutations in KRAS, STK11 and TP53. Moreover, one sample harbored four concurrent mutations in PIK3CA, BRAF, FBXW7 and TP53 (Figure 2, Table S1). Fifty mCRC patients harbored concurrent mutations. KRAS mutations were found to be concomitant with TP53 (n = 11), PIK3CA (n = 5), SMAD4 (n = 3) and FBXW7 (n = 1), CTNNB1 (n = 1) and AKT (n = 1). BRAF p.Val600Glu mutation was found in concurrency with TP53 (n = 3), PTEN (n = 1), PIK3CA (n = 1) and SMAD4 (n = 1). NRAS and PIK3CA were concomitant in one sample. Fifteen patients carried three concurrent mutations, KRAS-PIK3CA-TP53 being the most frequent combination (n = 5). Interestingly, one patient harbored concurrent mutations in KRAS, BRAF and TP53 and other carried mutations in NRAS, BRAF and TP53. One patient carried four concurrent mutations in KRAS, SMAD4, FBXW7 and TP53 ( Figure 3, Table S2).

Clinically Relevant Genetic Variants
NGS identified actionable genomic alterations in 58% of NSCLC patients (Figure 4). The most prevalent changes detected in EGFR were exon 21 mutations (6%) followed by exon 19 alterations (5%). Codon 12 was the most frequently mutated in KRAS (23%) followed by codon 13 (3%) and codon 61 (2%). In NRAS only codon 61 was found mutated (1%). Regarding BRAF, three out eleven detected mutations occurred on the hotspot Val600. In regard to PIK3CA, codon 542 mutations were the most frequent (2%) followed by mutations in codons 545 and 1047 (1% for both). Duplication in exon 20 (p.Ala771_Met774dup; 2%) and the hotspot mutation p.Arg784His (1%) were found in ERBB2. Four patients showed fusions between ALK and EML4, in all of them, the rearrangement involved exon 20 of ALK, in three patients with exon 6 of EML4 and in other with exon 13. One patient showed a fusion of ALK with an unknown partner. Finally, one patient presented a fusion between ROS1 (exon 35) and CD74 (exon 6).  Regarding targeted therapy in mCRC patients, 50% of patients harbored RAS mutations as a primary resistance mechanism to anti-EGFR therapies. Additionally, in the RAS wild type patients, NGS identified eight BRAF V600E mutated patients, five PIK3CA mutated patients and one patient harboring p.Lys57Asn in MAP2K1 gene ( Figure 5).

Discussion
NGS has emerged as a promising strategy to achieve precision medicine [16,17]. These approaches are able to identify multiple cancer genes simultaneously with low sample requirement, reducing sequencing costs and molecular diagnostics turn-around time. However, the integration a high-throughput technology into clinical routine practice of a public health system represents a major challenge. Laboratories performing clinically-relevant tests must improve their quality and competence [18] and for this purpose, accreditation and participation in External Quality Assessment (EQA) programs are strongly recommended [19].
In this study, we show that NGS technology is able to efficiently amplify and sequence multiple genes using only 10 ng of DNA or RNA obtained from FFPE samples. The quality metrics analysis revealed excellent read depth and coverage for all the targeted regions, allowing confident somatic variant detection [20]. These results technically validate the NGS assay and grant the identification of low VAF variants. The invalid test rate obtained (3.8%) is concordant with previously reported studies in FFPE samples [21,22].
In particular, these panels were able to confidently detect variants at <5% VAF, which can be relevant, in challenging, low tumor percentage FFPE samples. Therefore, and taking into account that current recommendations of NGS studies in FFPE samples propose not reporting variants with a VAF lower than 5% [23,24] (unless they have an important therapeutic or prognostic impact) we established this value as a threshold for reporting variants.
Moreover, we found an excellent correlation between NGS and single-gene conventional methods. The analysis of discordant results revealed that NGS is a more robust method compared with conventional approaches. Concerning EGFR mutations, one patient reported as positive with Cobas ® assay and negative by NGS also carried a homozygous and synonymous variant near codon 790 (p.Gln787Gln). However, since both NGS and SS did not detect this mutation, we hypothesize that the synonymous variant could affect primer or probe hybridization of the Cobas ® assay, resulting in a false positive detection of the p.Thr790Met mutation. In KRAS testing, we found seven discordant cases. Two samples resulted positive by RT-qPCR assay but were not detected by NGS. These samples were re-tested using a new lot of the AmoyDx assay, providing then concordant results with the NGS assay. Among the five negative samples for KRAS mutations by RT-qPCR, NGS reported mutations at low VAF in two cases (3.9% and 5.0%, respectively). In theory, these VAFs should be detected by the AmoyDx assay, which has a limit of detection (LOD) of 1-2%, established by using cell line DNA. However, we suspect this LOD could be higher when using highly degraded DNA obtained from FFPE samples. In the three remaining cases, NGS revealed KRAS mutations at high VAF (12%, 12% and 20%), which could be confirmed by SS. Moreover, these samples were re-tested by a technician in another institution, showing concordant results with the NGS assay. For BRAF p.Val600Glu mutation, NGS revealed two mutated samples with VAF of 4% and 5%, not detected by HRM (LOD = 10%). Regarding fusion transcripts, an OncoNetwork collaborative research study was able to detect EML4/ALK fusion up to 1% dilution [25]. In our cohort, we detected five fusions by NGS, of which one was not detected by IHQ nor FISH (53 nuclei counts). Our results are in agreement with Velizheva et al., who conclude that targeted NGS is a more robust and reliable method for fusion detection, especially in borderline cases, compared to single target assays such as FISH [26]. Taken together, NGS has proven to be a valid alternative to conventional molecular testing in terms of diagnostic accuracy [27].
TAT and economic costs are essential for NGS implementation in routine molecular diagnostics in a public healthcare hospital. Here, we found a great economic benefit in the employment of NGS technology versus conventional methodologies when it came to wild type KRAS mCRC patients. In KRAS mutated patients this benefit is not observed, however, a complete NGS test is achieved with a €30 difference per sample, therefore being an economically sustainable approach. In NSCLC, the extra cost associated with NGS studies (€51.5/patient) can be assumed based on the ability to identify actionable alterations with significant impact on patients' outcome.
The NGS approach described in this study requires a manual library and template preparation (emulsion PCR, enrichment and chip loading). Consequently, hands-on time is clearly higher than conventional studies in both NSCLC and mCRC samples. However, the development of new automatized devices for library and template preparation has drastically reduced hands-on time to approximately 1 hour, making NGS implementation in terms of technical staff much easier. Global time duration of NGS studies has also been higher than conventional approaches in NSCLC (5 versus 3 working days) and in mCRC (5 versus 3-4 working days depending on KRAS mutational status). This delay in molecular studies should not be an important limitation of NGS implementation because of its ability to identify clinically relevant alterations beyond the routinely tested genes. Moreover, the coexistence of both strategies may allow the choice of a faster conventional strategy when needed, especially in patients whose clinical situation requires a molecular result in a short period of time.
The implementation of this workflow in diagnostics routine was our first experience with NGS which allowed us to acquire a great expertise in amplicon-based NGS approaches. This has permitted us improve and optimized the process by implementing more complex gene panels (such as Oncomine Focus Assay, able to detect hotspot mutations in 35 genes, copy number variation in 19 genes and fusion transcripts of 23 driver genes) and automatizing the process (IonChef Instrument). Moreover, the development of new and faster sequencers (Ion S5 Instrument) is able to reduce global time duration to 4 days. However, it is important to acknowledge that NGS is economically sustainable when the appropriate number of samples is studied in the same experiment. In this sense, and according to the number of samples received for NGS studies, we are reporting NGS results under routine laboratory conditions in approximately 10-15 working days, as recommended [28][29][30].
The major advantage of the NGS approach is to provide information about potential therapeutic targets to improve clinical outcomes of patients with advanced cancer. Multiple biomarker testing has become a major challenge for molecular diagnostic laboratories because of the increasing number of approved targeted therapies and clinical trials. In this scenario NGS has been postulated as a technology with clinical applicability able to provide an exhaustive molecular profiling, deciphering tumoral heterogeneity that can in certain cases have a prognostic value and/or explain treatment resistance [31].
The mutation prevalence identified in our study for NSCLC and mCRC samples is concordant with previously published studies [32][33][34]. Exhaustive molecular profiling can provide relevant information for a patient's clinical management. In our study, 58% of NSCLC patients harbored a potential clinically-actionable alteration, confirming the applicability of these studies for candidate selection. Regarding mCRC, in RAS wild type patients (n = 50) NGS identified four patients with mutated PIK3CA. Response to anti-EGFR treatment in these patients is still controversial [35] although RAS-RAF and PIK3CA wild-type patients seem to have better responses [36]. Moreover, one patient harbored the p.Lys57Asn mutation in MAP2K1 that has been described as a primary resistance mechanism to anti-EGFR treatment [37]. Taken together NGS identified 49% of patients without resistance mechanisms to this targeted therapy.
Concurrent mutations have been detected in 35% of NSCLC patients and in 50% of mCRC patients revealing tumor biology complexity. Although there are no well-established molecular prognostic factors neither in NSCLC nor mCRC certain passenger mutations may be associated with an adverse prognosis. In this sense, TP53 [38] or STK11 in concomitancy with KRAS mutations [39] have been associated with a worse prognosis in NSCLC. In mCRC, TP53 [40] or SMAD4 [41] mutations have been related to a worse response to anti-EGFR therapy and FBXW7 [42] has recently been described as a strong worse prognostic factor.

Patients and Samples
The study included a series of 106 advanced NSCLC (stages III-IV) and 102 mCRC (stage IV) patients diagnosed in the Department of Medical Oncology at the University Hospital La Fe (Valencia, Spain) from 2015 to 2017. The epidemiological, clinical and pathological features of these patients are summarized in Table 3. All patients showed their agreement by signing the informed consent elaborated in accordance with the recommendations of the Declaration of Human Rights, the Conference of Helsinki [43]

DNA and RNA Preparation
Genomic DNA was isolated from three 5 µm thick FFPE sections using Deparaffinization Solution and the GeneRead DNA FFPE Kit (Qiagen, Hilden, Germany). RNA was extracted from three 15 µm thick FFPE sections employing the RecoverAllTM Total Nucleic Acid Isolation Kit (ThermoFisher Scientific, Waltham, MA, USA). DNA and RNA concentration was assessed using Qubit 3.0 fluorometer with DNA HS or RNA HS Assay Kit (ThermoFisher Scientific).

Molecular Analysis by Next Generation Sequencing
Molecular analysis was performed at the Molecular Biology Unit (University Hospital la Fe) using Conformité Européenne-In vitro diagnostic (CE-IVD) approved kits and workflows.

Ion Torrent Library Preparation
For DNA libraries preparation, multiplex PCR was performed on 10 ng of DNA. After primer digestion and barcode ligation, library fragments were purified with Agencourt ® AMPure ® XP (Beckman Coulter, Brea, CA, USA). Finally, quantification and dilution (100 pM) of the amplified libraries was performed using the Ion Library Equalizer Kit (ThermoFisher Scientific) as described by the manufacturer.
RNA libraries preparation included a previous cDNA synthesis step from 10 ng of RNA using the SuperScript kit VILO cDNA synthesis kit (ThermoFisher Scientific). In this case, a multiplex PCR amplification of cDNA was performed. Library quantification was carried out by qPCR, inferring the concentration from a standard curve generated with Ion Library Quantification Kit (ThermoFisher Scientific). RNA libraries were diluted to a concentration of 100 pM.
In NSCLC, DNA and RNA libraries from eight patients were combined in a 4:1 proportion, generating the library pool. In mCRC samples, 10 DNA libraries were combined in equal proportion.

Clonal Amplification and DNA Sequencing
The library pool was clonally amplified in an emulsion PCR reaction using Ion Sphere Particles (ISPs) in the One Touch 2 Instrument. Subsequently, template-positive ISPs were enriched using the Ion One Touch ES with the Ion PGM Hi-Q OT2 kit following manufacturer´s protocol. Enriched template-positive ISPs were subjected to sequencing on the Ion Torrent Personal Genome Machine (PGM) on a 318v2 Ion Chip using Ion PGM Sequencing Hi-Q kit (all kits from ThermoFisher Scientific).

Base Calling, Variant Annotation and Prediction Tools Analysis
Raw data processing and alignment to the hg19 human reference genome was performed with Torrent Suite v5.6. Aligned sequences (Binary Alignment Map (BAM) files) were automatically transferred to the Ion Reporter Software (v5.6) to perform variant calling/annotation by using commercial workflows. Intronic variants and synonymous changes were filtered out. Variants with low total read depth (<500 total) and/or low variant read depth (<20 reads) were excluded. Additionally, Variants were visually examined using the Integrative Genomics Viewer (IGV) software (v.2.4). Subsequently, sequence variation databases such as Catalogue of Somatic Mutations in Cancer (COSMIC) [44], VarSome [45], The 1000 Genomes Project [46] and Single Nucleotide Polymorphism Database (dbSNPs) [47] were used to assess the pathogenicity of the detected variants. In variants with unknown significance, prediction tools like Provean [48], Sorting intolerant from tolerant (SIFT) [49] and PolyPhen-2 [50] were used in order to predict the effect of the amino acid substitution on the protein structure and function.

Experimental Verification
Verification instead of a full validation analysis was performed according to our national accreditation body (ENAC; Entidad Nacional de Acreditación), since the OST-DNA and OST-RNA kits are CE-IVD approved. The performance of NGS testing was extensively evaluated on different aspects. Firstly, we used well-characterized reference material to assess the presence or absence of somatic variants (point mutation and small insertions/deletions) and their allele frequencies. Secondly, we considered the pre-analytical conditions and assessed the quality of NGS analysis on determining FFPE samples as start material, allowing us to establish the sequencing quality metrics. Thirdly, diagnostic sensitivity and specificity were determined experimentally by comparing it with conventional methods for routinely tested alterations; additionally, clinical reporting was adapted according to international diagnostic standards and professional guidelines. Finally, to ensure a consistent high standard of performance, it was essential to establish an EQA program to monitor the quality of NGS testing in clinical practice and to propose corrective actions when needed.

Next Generation Sequencing Metrics
The number of reads, mean depth, "on-target" reads and uniformity were the parameters used as quality control check points for further sample analysis. A total number of reads higher than 100,000 together with "on-target" and uniformity values >80% were required for each DNA library and 20,000 total reads for each RNA library.

Assessment of the Diagnostic Sensitivity and Specificity of the NGS Assay
Detected missense mutations in EGFR, NRAS, KRAS and BRAF genes, as well as ALK and ROS1 genes rearrangements were tested by conventional methods. EGFR mutations were validated by Cobas ® EGFR Mutation Test v2 (CE-IVD) (Roche Diagnostics, Basel, Switzerland); NRAS and KRAS mutations were confirmed by Real Time (RT)-qPCR using AmoyDx ® KRAS Mutation Detection Kit and AmoyDx ® NRAS Mutation Detection Kit (AmoyDx, Xiamen, China); BRAF mutations were validated by High Resolution Melting (HRM) as previously described [51]. ALK and ROS1 rearrangements were studied by immunohistochemistry (IHQ) employing VENTANA ALK (D5F3) CDx Assay (Roche Diagnostics, Basel, Switzerland) and IHQ ROS1 Clon D4D6 (Cell Signaling Technology, Danvers, MA, USA), respectively. Positive IHQ assays were confirmed by fluorescence in situ hybridization (FISH) employing Vysis ALK Break Apart FISH Probe Kit (Abbott Laboratories, Chicago, Illinois, USA) and Vysis 6q22 ROS1 Break Apart FISH Probe Kit (Abbott Laboratories), respectively.

External Quality Assessment (EQA) Program
To ensure a consistent high standard of performance, this assay was externally validated by the participation in the EMQN External Quality Assessment Scheme for Oncogene Panel Testing (2017/2018).

Statistics
Quantitative variables were summarized by their mean and standard deviation, and categorical variables by absolute frequencies.

Conclusions
NGS is a technology able to assess multiple genetic biomarkers that has demonstrated a great concordance with conventional single target assays, providing an exhaustive molecular profiling of clinically relevant alterations at reasonable costs and turnaround times. The implementation of NGS in the diagnostic routine under the scope of UNE-EN ISO 15189:2013 accreditation has provided relevant information for patients' clinical management, improving the molecular diagnostic in our center.