RAS testing in metastatic colorectal cancer: advances in Europe

To assess RAS mutation status, analysis is usually performed on DNA extracted from formalin-fixed, paraffin-embedded (FFPE) tissue blocks [43]. As a result of chemical modification and DNA fragmentation, the amplicon size in polymerase chain reactions (PCR) is limited to about 150 base pairs, when sequencing is performed on FFPE-derived DNA [44]. Moreover, for NGS, the initial procedure for DNA isolation from FFPE tissue blocks remains the same as for conventional DNA sequencing [43] and is, therefore, subject to some of the same issues of DNA quantity and quality.

Most sequencing methods can screen a gene locus for mutations. Newer sequencing methods provide quantitative information and increase the number of loci that can be analyzed in parallel.

Sanger sequencing (also known as “terminated chain sequencing” or “dideoxy sequencing”) is considered to be a standard method for analyzing DNA sequences and thus also for RAS testing [43, 44]. Overall, it is the most widely used method of RAS testing in Europe [45], although its use varies from country to country. Major improvements to the original Sanger technique have resulted in reduced reagent volumes and consumable costs, as well as increased throughput [46], although it is still considered time consuming compared with some other methods [44]. The advantage of this method is its ability to detect all types of mutation (single-nucleotide substitutions, insertions, and deletions) in the amplified locus. If DNA is isolated from FFPE tissue, the size of the amplicons to be analyzed should not exceed 150 base pairs. To ensure reproducibility, bidirectional sequencing reactions are performed using forward and reverse primers [47]. Each run gives the nucleotide sequence of one amplified locus; up to 96 runs can be performed in parallel (usually 16 to 24). In addition, the LOD for Sanger sequencing is relatively modest compared with other techniques. Multiple studies have shown that Sanger sequencing requires at least 10–25 % of the neoplastic cells in the sample to contain KRAS mutations for reliable detection [48]. Several enrichment methods are available, however, to increase the concentration of mutant DNA, thus improving the overall sensitivity of the detection process (Table 2) [49‐53]. Another limitation of Sanger sequencing is the low level of precision in the variant fraction (i.e., the percentage of tested alleles that are mutant); the error rate (i.e., the percentage of incorrectly identified bases) of the Sanger sequencing reaction is estimated at 0.001 to >1 %, depending on the software that is used post processing [54].

Pyrosequencing is an alternative method to Sanger sequencing that can provide both quantitative information and a low technical LOD (5–10 %) [55], although the sequencing error rate may be higher (4–25 %) [56]. During pyrosequencing, the incorporation of each nucleotide is followed in real time by the emission of a specific light unit based on the amount of pyrophosphate, which is released in stoichiometric amounts during the elongation reaction and converted by an enzymatic cascade into adenosine triphosphate (ATP) molecules each time a nucleotide is incorporated. The ATP serves as the energy donor in a bioluminescence reaction resulting in emission of light that can be measured as a peak in the pyrogram output. Pyrosequencing allows the local separation of mutant and WT signals, although the presence of homopolymers can limit the method, as linearity is present only for some nucleotides. In this approach, each run gives the nucleotide sequence of one amplified locus, but the size of the sequence (less than 20 base pairs) is often shorter than those produced via Sanger sequencing. Up to 96 runs can be performed in parallel.

NGS platforms involve massively parallel sequencing of bulks of clonally amplified DNA molecules that are spatially separated and have helped to shift the focus of sequence analysis from a few loci to multiple genes and loci. With NGS, each run can give the nucleotide sequence for several loci and several tumor samples. Importantly, NGS methods can detect any variant from single-nucleotide changes to insertions or deletions, and even translocations, in a single multiplex PCR reaction. Depending on the depth of sequencing, the LOD of NGS is about 1–5 %. Results are provided as quantitative information. The most sensitive part of the NGS is the bioinformatic pipeline generating Binary Sequence Alignment Map/Sequence Alignment Map format (BAM/SAM) files. Specific statistical algorithms are used to score the quality of a single read by aligning the sequence to a reference genome. As a consequence, some readouts do not detect all of the deletions, insertions, or other mutations that are detectable with Sanger sequencing, as some mutations are found by the basic algorithms but subsequently excluded and, therefore, do not occur in the resulting variant caller files. Therefore, extensive validation of the entire procedure is essential, as is the use of internal control to assess the reproducibility of the runs.

With the increasing number of biomarkers, NGS allows screening of a large number of patients and markers with a fixed and limited amount of DNA. For NGS using Ion Torrent technology, as little as 1 ng input material is needed, whereas conventional sequencing technologies need 10 to 20 ng per run and marker [57]. As a first step, a DNA library is generated by amplification of the DNA fragments to be sequenced (the gene panel) in a single reaction (sometimes in up to five different tubes) or capture hybridization followed by the amplification of this pre-library either by emulsion PCR (emPCR; e.g., on spheres) or by bridge PCR on solid flow cells [46]. As capture hybridization needs much more input material compared with the panel-amplification approach, the latter is mostly used in the analysis of biomarkers. Ultimately, a single NGS run may comprise many millions or even billions of spatially distributed, clonally amplified amplicons [58]. Custom assays can be developed (AmpliSeq™ for Ion Torrent, GeneRead for Qiagen, TruSeq™ for Illumina) using amplicons designed with a starting size of 150 bp, optimal for FFPE samples. Finally, the quality of the sample can be extremely limited, and strategies have been proposed to identify samples with low probability of success (e.g., the Illumina FFPE QC Kit based on qPCR validation) before any NGS run. This highlights that not all samples are suitable for analysis on these platforms.

Historically, three principal NGS technologies have been commercially available [58]. The first of these, sequencing-by-ligation, no longer has any clinical application. The second is sequencing-by-synthesis [59], and the third is based on the pyrosequencing approach. As an alternative to pyrophosphate measurement, hydrogen ions/protons can be used as a hydrogen ion is released for each nucleotide incorporated. These can be measured as an induced currency peak on a negatively charged semiconductor, which is the basis of the detection unit or chip. Currently available NGS platforms are summarized in Table 3.[60‐62] Of these, the Illumina MiSeq and Life Technologies’ Ion Torrent Personal Genome Machine™ (PGM™) are the most commonly used in Europe (H. Van Krieken, personal communication), and NGS sequencing methods are expected to replace older methods completely in the coming years.

For the detection of single mutation events in a background of WT sequences, digital PCR techniques [68] were developed that can achieve LODs of <0.1 % for specific hotspots. If the coverage, and thus the depth, of NGS techniques are increased, they can also reach similar LODs [69]. Sensitivity can also be increased by using single-molecule techniques that discriminate between biological and technical duplicates, such as single-molecule molecular inversion probes [70].

The potential applications for hypersensitive methods include detection of mutations in circulating plasma DNA (the so-called liquid biopsy), monitoring of metastatic disease under therapy, detection of emergence of resistance to targeted agents, analysis of small or poor-quality DNA samples, and as a possible additional parameter alongside histopathology for the staging of tumors [71]. Such approaches have been applied in research and for free plasma DNA detection; there will be no direct application for molecular RAS testing in routine clinical practice until the clinical validity of these techniques can be determined. The preparation of such assays is generally easier than for NGS assays, but the scope of the targeted markers is greatly reduced, mostly focusing on a single gene locus or small number of loci.

Commercially available kits, instruments, and reagents that are intended by the manufacturer to be used for in vitro examination of patient specimens for the purpose of providing information are governed by European Directive 98/79/EC on in vitro medical devices [77]. Key objectives of this directive are to ensure that in vitro diagnostic medical devices provide a high level of health protection and attain the attributed performance level, and products that conform to this standard carry “CE” marking. The directive does not regulate products for research-use only nor does it cover non-commercialized patient diagnostic tools produced within health institution laboratories for use solely within that environment. It is important to note that the CE mark simply means that the product itself has consistent quality and does not provide any information about specificity and sensitivity in the lab where the kit is used.

The advantage of commercial CE tests is the validation process that they have undergone. Thus, after the in-house verification of a CE-kit, no additional testing is needed, which is in contrast to non-CE-kits where a change in the batch of materials used requires quality testing of the new material. A list of CE-approved RAS tests available in Europe (complete as of December 2014, to the best of our knowledge) is shown in Table 4 [78‐92]. Some techniques identify the somatic mutation directly, while others are screening methods that involve a two-step process for positive samples. In addition, other CE-approved kits are available for assessing KRAS exon 2 status (with or without exon 3 codon 61), and laboratories may wish to use these, supplemented with laboratory-developed tests to complete the RAS analysis.

Of the available CE-approved kits, only five detect mutations in both the KRAS and NRAS genes: (1) CRC RAScan™ (SURVEYOR®/WAVE®), (2) CRC RASseq™ (both Transgenomics, Inc.), (3) Therascreen® KRAS and NRAS Pyro® kits (Qiagen), (4) OncoCarta (Agena Bioscience), and (5) cobas® KRAS mutation test (Roche) together with LightMix® kits (TIB MOLBIOL) for the cobas system.

CRC RAScan™ (SURVEYOR®/WAVE®) uses Transgenomics’ proprietary DNA mismatch-cutting enzyme SURVEYOR® nuclease, with cleavage products separated by denaturing high-pressure liquid chromatography using the WAVE® platform [77]. The nature of the mutation is then confirmed by Sanger sequencing. CRC RASseq™ uses Transgenomics’ proprietary primer sets for PCR amplification and Sanger sequencing of exons 2, 3, and 4 of the KRAS and NRAS genes [79]. This method was used to identify RAS mutation status in mCRC patients included in the PRIME study [14].

Therascreen® KRAS and NRAS Pyro® kits are based on pyrosequencing technology and consist of two assays: one for codons 12 and 13 and a second for codon 61 [80, 81]. Qiagen has recently launched a new RAS Extension Pyro Kit, covering codons 58/59, 117, and 146 of KRAS and NRAS, which has recently received CE-marked status, although the LOD of this extension kit is not reported [93].

The Agena Bioscience OncoCarta panel includes 12 KRAS and eight NRAS mutations [82]. Detection is based on the Agena MassARRAY® system, which uses matrix-assisted laser desorption/ionization–time of flight mass spectrometry [94].

The choice of CE-marked RAS mutation test by individual laboratories is largely dependent on the available equipment, experience, and costs of the test [95]. The rapid change in knowledge of biomarkers and loci implies that development of additional commercial kits is necessary, and the situation for RAS screening, with only five available kits covering all loci, highlights the relatively low reactivity of the market.

Many laboratory-developed methods for RAS testing are used for in vitro diagnostics, and still more methods exist for research-use only [42]; the latter are beyond the scope of this review. This is especially true for NGS, as commercially available kits are almost inaccessible, particularly those of CE-quality. CE-marked assays for NGS could be the next step to ensure the quality of reagents and primer pools. Initiatives to standardize a primer set, such as the community panel from Life Technologies and the Ion AmpliSeq™ Colon and Lung Cancer Research Panel v2, are underway, although such primer sets are currently for research-use only [57]. Most commonly, Sanger sequencing- and pyroseqencing-based laboratory-developed methods are used for RAS testing [43]. In some countries, such as France, allele-specific PCR is very popular (Dequeker et al., manuscript submitted). Here, mutant alleles are selectively amplified, making this technique highly sensitive for the detection of mutations in the RAS genes. There are many different variations of this general principle (Table 5) [96‐100], which is reflected in the many variations of allele-specific hybridization methods [43]. In general, laboratory-developed methods offer the possibility of more rapid adaptation to new knowledge of biomarkers and loci, although their use requires a high level of experience and quality management within the laboratory. Moreover, mostly as a result of the costs of CE-marked kits, laboratory-developed techniques are very suitable for testing the sensitivity (LOD) and specificity of RAS mutation detection, which is important in the maintenance of quality.

The sensitivity of mutation detection is influenced by multiple factors, mostly LOD and the minimum amount of required template. The LOD is dependent on the technique used and might therefore be considered as the technical sensitivity, which is defined by the ratio of the measurement signal (mutation) to the background signal (mutant signal in a WT sample) [43]. The sensitivities of available tests range from <1 to 10 % for CE-approved testing methods (Table 4). For laboratory methods, the most sensitive techniques are allele-specific PCR and hybridization assays (0.1–1 %) and gel electrophoresis (e.g., temporal temperature gradient or constant denaturant capillary electrophoresis) (1 %) [43]. Pyrosequencing (5 %), high-resolution melting curve analysis (~10 %), and Sanger sequencing (10–25 %) have a lower level of technical sensitivity [43, 101]. This highlights the benefit of histopathology as the basis of mutation detection: sensitivity can be controlled, as the total amount of tissue—reflecting the amount of template, and relative fraction of neoplastic cells within the tumor (measured as area or total nuclei)—reflecting the LOD, can be determined (semi) quantitatively. For example, a KRAS mutation in a small metastatic focus of neoplastic cells in a lymph node can escape detection when entire tissue sections of the lymph node are used for DNA isolation (i.e., normal cell DNA content is far greater than neoplastic cell DNA). After careful microdissection of the neoplastic cells, however, the KRAS mutation can unequivocally be identified. Thus, in clinical practice, the sensitivity of the entire process must be considered, not just the sensitivity of the technique being used.

Unfortunately, the sensitivity of mutation detection—which needs to be related to the clinical effect—in the reference studies establishing the validity of RAS as a biomarker is unknown. This is because only the technical sensitivity of detection can be derived from the method used and the minimum required neoplastic cell fraction required as input, not the actual values for individual samples. Thus, the clinical significance of high versus lower sensitivity, indicating the number of neoplastic clones with or without a RAS mutation, is currently unclear. Achievement of higher sensitivity must, therefore, be weighed against the possibility of false positives or the identification of mutations present only in a small minority of the neoplastic cells, reflecting a small tumor clone population with RAS mutations, which is of unknown clinical relevance. Additionally, sensitivity depends on the experience of the laboratory, as results for the LOD of Sanger sequencing were found to be similar to allele-specific methods during blind testing [102]. Furthermore, a recent publication suggests that tumors containing <1 % mutant KRAS DNA allele-fraction in microdissected samples might still benefit from treatment with anti-EGFR agents [103]. Overall, it is important to take into account histopathological information from the tissue samples used for mutation detection in future clinical studies, to better define cutoffs for sensitivity and clinical response, and to improve the benefit to patients of targeted therapies.

Turnaround times for testing remain a key challenge for pathologists and oncologists, and it is important that a strategy is in place to optimize the process, whether this is through stratified reflex testing based on patient risk factors or through logistical management of new or archived tumor samples. In general, European EQA schemes recommend that the process of obtaining a RAS test result should take no more than 10 working days [104], and it is important that pathology and oncology centers work together to address barriers that prevent prompt turnaround of test requests. NGS processes generally take more than 2 days to complete, which is the main bottleneck when choosing new technology in comparison to conventional methods.

Reflex sampling involves the immediate sampling and testing of mCRC at the time of diagnostic biopsy. The major benefit of reflex testing is that information on RAS status is available as soon as EGFR-targeted therapies are considered, so no time is lost in starting treatment. Reflex testing in which all colorectal cancers are tested to anticipate the risk of metastasis is also possible, although this remains an area of debate. Underlying RAS mutations appear to occur early in tumor development, and therefore, testing for RAS mutations is potentially informative, although the prognostic significance of such information is not known. While delays in obtaining RAS test results can affect treatment strategies, the implementation of reflex testing is likely to be associated with a high degree of unnecessary testing. Whatever the timing of testing, it is important that interpretation of results must take into account the patient’s clinical history.

The function of the bioinformatics pipeline with regard to RAS analysis is to convert the raw sequencing data into clinically useful information through the use of appropriate software algorithms [105]. Integration of NGS technologies into the clinical setting has the potential to verify suggestive relationships and remove ambiguity more quickly than current approaches. Validation of the output from such algorithms can be achieved through the use of reference sequences, control samples containing specific RAS variants, and by comparing the output with that from alternative sequencing platforms. Guidelines issued by the National Collaborative Study of Dutch Genome Diagnostic Laboratories state that data analysis and quality checks can be divided into five main steps: base calling and quality assessment of raw data, mapping of reads against a given reference sequence, enrichment analysis, coverage analysis, and variant calling and annotation [106]. The guidelines further state that minimal quality criteria must be established in each laboratory to determine the overall quality of platform performance, although the specific criteria usually depend on the sequencing platform being used.

The initial RAS sequencing report should include all variants that were detected against a stated reference sequence, regardless of their clinical significance, although this information should be used to annotate each variant as far as possible (including whether or not clinical validation has been reported [105]. The report should also include information on the test used, including any important limitations, along with base quality scores and confirmatory testing. Notably, a study of genome sequencing across 30 international groups showed that, while there was generally good consistency across bioinformatics analysis and interpretation, there was a lack of concordance with regard to interpretation, report content, and patient consent procedures [107]. Suggested reasons for differences between groups included limited quality control measures; different read aligners and variant calling pipelines; variations in performance of analysis tools, software, and filters; errors in programming or manual analysis; and poor-quality data [105, 107].

Given the large variety of techniques and the importance of laboratories’ experience, it is essential to test and validate centers’ performance with standardized EQA schemes. Use of CE-marked kits or fully validated methods involves proof that quality is maintained after the implementation of the marker in the laboratory. Nevertheless, EQA validation should be done periodically to obtain assurance about the laboratory’s quality.

Guideline recommendations for a European EQA program to ensure accuracy and proficiency in KRAS mutation testing were published in 2008 [42]. A number of EQA programs are now available (Table 6), [42, 108‐114] such as the European program established by the European Society of Pathology (ESP) for testing biomarker mutations in colorectal cancer [42] or the German Quality Initiative in Pathology (QuIP) (Jung et al., manuscript in preparation). The European program aims to ensure optimal accuracy and proficiency in biomarker testing in colorectal cancer across all participating countries [115] and is organized in collaboration with a European working group and the Biomedical Quality Assurance Research Unit of the University of Leuven [115]. The program provides recommendations and overviews of laboratory methods, standardized operating procedures, and accreditation criteria relevant for RAS mutation testing [42, 115, 116].

The Organization for Economic Co-operation and Development (OECD) guidelines for Quality Assurance in Molecular Genetic Testing address processes relating to genetic testing for variations in germline DNA sequences that are also applicable to somatic DNA testing. The guidelines cover general principles and best practices, EQA systems, proficiency testing (monitoring the quality of laboratory performance), quality of result reporting, and education and training standards for laboratory personnel [117].

In March 2012, medical oncologists, pathologists, geneticists, molecular biologists, EQA providers and representatives from pharmaceutical industries developed a guideline to harmonize the standards applied by EQA schemes in molecular pathology. The guideline comprises recommendations on the organization of an EQA scheme, defining the criteria for reference laboratories, requirements for EQA test samples, and the number of samples that are needed for an EQA scheme [118]. All EQA schemes should be developed by an expert group with a representing EQA provider, which is also responsible for all organizational aspects and operation of the scheme according to International Organization for Standardization (ISO) 17043 standards. For each scheme, a team coordinator is responsible for the selection, distribution and receipt of cases, and the analysis and reporting of results. Importantly, the samples used in an EQA scheme should reflect the diagnostic and clinical reality as closely as possible, and the turnaround time, as defined by the EQA provider, should reflect the common clinical situation. For a reliable evaluation, at least 10 samples per laboratory need to be analyzed per year and can be provided in a single batch, or in multiple smaller batches. The limit for poor performance should be defined as a score 18 out of 20, with no false-positive or false-negative results. Finally, EQA providers should be encouraged to make the general reports available in the public domain. The ESP Colon EQA scheme is organized in accordance with these recommendations and with the OECD best practice guidelines [42]. OECD Principles of Good Laboratory Practice state that individual testing facilities should have a documented QA program to ensure that studies performed are in compliance with principles of good laboratory practice [119]. A number of authors have stated that participation in internal QA schemes, as well as external schemes, should be mandatory for diagnostic pathology laboratories [95, 120]. Furthermore, the validation or verification of testing methods is a formal requirement for the accreditation of laboratories according to the major international standard applicable to medical testing laboratories: ISO 15189 [121, 122].

The need for QA schemes was highlighted by recent research carried out under the European QA program [45], in which 27 % of participating laboratories genotyped at least 10 % of samples incorrectly, although <5 % of distributed specimens overall were genotyped incorrectly. Errors included false negatives, false positives, and incorrectly genotyped mutations. In addition, 20 % of laboratories reported a technical error for at least one sample. It is encouraging, however, that the majority of laboratories made no mistakes. There are now sufficient quality-assured laboratories to allow RAS testing across Europe, and even a small number of EQA rounds can improve quality to a high level [123].

In the future, as new techniques become routinely available, existing EQA programs will need to evolve to address the issues arising from the use of NGS and other increasingly sensitive techniques.