Tumor Genotyping – How Accurate Are You?

At a Glance

• Molecular diagnostics are making personalized medicine a reality, with companion diagnostics supporting progress towards the goal of a precise diagnosis and a tailored therapy
• Europe has seen the approval of an EGFR tyrosine kinase inhibitor to have a label indicating the use of cell-free DNA obtained from a blood sample; the first of its kind, but large variances in concordance rates between cfDNA and tumor tissue have been reported
• More is clearly needed to ensure accuracy of mutation testing; an incorrect outcome could be potentially life-threatening
• Inaccuracies and errors made by diagnostic labs using a wide range of methodologies can be reduced though, using reference materials, and annual participation in EQA should be seen as the norm

Personalized medicine, with the aid of molecular diagnostics, is providing the exciting possibility of cost-effective tailored therapies, based on an individual patient’s genetic code. This is particularly true in the case of cancer where a single nucleotide polymorphism (SNP) out of a three billion-base genome can be the difference between having, and not having, an actionable drug therapy. However, identifying this one-in-a-billion can be tricky; with the multiple steps of a diagnostic workflow (Figure 1), any variability that creeps into each step is further compounded downstream potentially leading to incorrect diagnoses. The need for consistent accuracy in order to provide a precise diagnosis and effective tailored therapy is therefore critical. So what progress is being made?

Companion diagnostic developments

Companion diagnostics are certainly making good headway towards achieving the ultimate goal. For example, the most recent collaboration between AstraZeneca and Qiagen provides the first companion diagnostic approach to guide the use of cell-free DNA (cfDNA) in the treatment of patients with advanced non-small cell lung cancer (NSCLC). The therapy, Iressa (gefitinib), is the first epidermal growth factor receptor (EGFR) tyrosine kinase inhibitor to have a European label indicating the use of cfDNA obtained from a blood sample.

However, the clinical feasibility of using cfDNA to detect EGFR mutations was assessed in a recent Phase III trial of a Japanese subset of patients (1). The trial found that the proportion of patients identified with mutant EGFR was lower when assessed in cfDNA (23.7 percent) compared with tumor tissue (61.5 percent). A high rate of false negatives (56.9 percent) was also observed. The large variance in concordance rates for mutation results between cfDNA and tumor tissue are shown in Figure 2.

Although companion diagnostic technologies undergo thorough regulatory review before being released to the market, there is still a need to maintain clinical vigilance, particularly where limitations are identified within a workflow approach, sampling method or limit of detection. As with any clinical protocol, sample handling will require clinical vigilance through sound quality assurance and control methodologies, including routine validation activities.

Outside of cfDNA, the need for accuracy is shown in External Quality Assessment (EQA) schemes; for example, the worldwide EQA proficiency scheme (2014) reports that of laboratories tested, only 72 percent correctly identified EGFR mutations in patient samples (2).

While substantial advances continue to be made, it’s clear that more is needed, and one technology that has seen an explosion in recent years is single-molecule sequencing (Figure 3). The new generation of these technologies (third-generation sequencing) is now emerging, with the potential for even higher throughput, longer reads and shorter time to result, which will lead eventually to a lower overall cost. However, as with any new technology, new challenges arise along with new workflow steps and therefore new sources of variability. Similarly, with all the data now being provided by next-generation sequencing (NGS) technologies in greater quantities, volume and speed, how is it actually being used?

How is Big Data being used?

According to Boehringer Ingelheim’s recent ‘Let’s Test Campaign’ (4) – not enough. The survey, conducted between December 2014 and January 2015, found that, although 81 percent of newly diagnosed NSCLC patients received testing for EGFR mutations, only 50 percent of oncologists reported their treatment decision was effected by a patient’s EGFR mutation subtype. It further found that they started one in four patients on first-line treatment before they had even received results on mutation status.

Cited reasons state lack of tumor histology and insufficient tumor samples. The lack of tissue samples has been a longstanding problem, particularly in hard-to-find lung cancers, hence the development of alternatives such as cfDNA tests. But lack of material for both clinical testing and validation and set up of diagnostic tests has always been an issue.

So what happens when therapies go wrong? Consider colorectal cancer as an example: EGFR targeting therapies have been developed for the treatment of patients with metastatic colorectal cancer to great effect. However, mutations within the KRAS gene are found in 30–40 percent of colorectal tumors (5) and people who have this particular mutation show a poor response to the popular therapies of cetuximab and panitumumab (6), with patients even experiencing worsening side-effects in some cases.

To put this into perspective; there are over 1.4 million people worldwide each year who are diagnosed with colorectal cancer (7). Combine this with the conservative number that 30 percent of these patients have a mutated KRAS gene, you can estimate that at a cost of $18,882 per treatment, it could potentially be costing payers over $8 billion worldwide per year because of incorrect tumor genotyping results in molecular diagnostics.

As a result, since 2008, the use of EGFR-targeting antibodies in metastatic colorectal cancer has been restricted to patients with wild-type KRAS tumors by the European Medicines Agency, based on data showing a lack of efficacy and potential harm in patients with mutant KRAS tumors (Figure 5). To add complexity, NRAS has also been presented to be involved in the prognosis of inefficient treatment at ASCO (2013) (8), but that is another story. In any case, the variability between laboratories and methods means that some patients still receive medication when they do not need it, and more importantly, others do not receive potentially life-saving treatment when they do.

Aiming for accuracy

There are ways to increase and ensure the accuracy of a laboratories’ tumor genotyping, including the use of reference material, EQAs and ISO standards. Simon Patton, Director of the European Molecular Quality Network (EMQN), believes that EQA proficiency testing schemes may be the answer. His organization is responsible for coordinating many EQA schemes including the most recent EGFR EQA scheme (2), which included three rounds. “EMQN has been organizing EQA schemes for rare single gene disorders for eighteen years. Because of this experience, we were approached by a number of clinical oncologists working in Europe to provide EQA for lung cancer testing,” he says.

“We had evidence from a pilot scheme that the quality of lung cancer testing and reporting of the results to clinicians was in need of improvement. This area of diagnostics has evolved very fast, and it’s been driven by pharma’s need to get their drugs into the clinical setting. This need has mainly been met by different diagnostic laboratories, predominantly genetics and pathology, which have been encouraged to set up testing for tumor markers, and the manufacturers have responded by developing new diagnostic kits and end-to-end diagnostic solutions. However working with compromised FFPE samples is challenging and EQA schemes are needed to ensure that the quality of testing delivers the right result, for the right patient at the right time,” Patton adds.

The EQA scheme

A steering group of five individuals was formed who planned, designed and assessed the results of the pilot EQA scheme involved in NSCLC testing. It was coordinated and administered by the EMQN and three rounds were organized within a period of 18 months. The first was restricted to a maximum of 30 laboratories to establish proof-of-principle and validate the materials. A subsequent second round was organized with no restriction on participation. Laboratories that failed the second round were provided with another set of samples in a restricted third round. The steering group evaluated the results according to a predefined scoring system, which assigned two points to correct genotype and zero points to false-positive or -negative results (Figure 4).

Once the data were analyzed, false-negative results were found to account for 85 percent of all the genotype errors made in the scheme, which could be a result of the low sensitivity of the method used for mutational analysis. For example, the expected minimum level of sensitivity is 15 percent for Sanger sequencing, and 5.43 percent for the p.(G719S) mutation as defined in version 1 of the Qiagen Therascreen kit packaging insert. Genotyping EGFR G719S in particular showed a 35.6 percent error.

PCR/sequencing was the most common method used in the scheme for scanning to detect point mutations. The major disadvantage of sequencing though is that it is not very sensitive (9), especially in samples with low tumor cell content. Real-time allele-specific tests are much more sensitive and specific, but only test for a subset of common mutations.

Following the study, Patton commented, “There is still considerable room for improvement in the quality of genotyping of tumor genes and the diagnostic error rate [an incorrect genotype that leads to a misdiagnosis] remains stubbornly high at 3.65 percent (as measured by the EQA). Errors are made by laboratories using a broad range of methodologies (see Figure 5), but we do have evidence that poor validation and/or verification of new tests contributes significantly to this problem. This is especially true when implementing an NGS strategy, or using a ‘black box’ commercial diagnostic solution.”

Not all doom and gloom

Although the inaccuracies and wide range of methodologies are evident in diagnostics, Patton does highlight some of the positives that have come from the EQA scheme: “We are seeing a significant improvement in clinical reporting with far less ‘over interpretation’  of the genotyping results with respect to treatment decision-making compared with previous EQA schemes. However, there still remains a tendency of participants to overstate the significance of the test result. EMQN has been pushing for standardization of reporting of sequence variants within the testing community by promoting best practice and the use of the Human Genome Variation Society (HGVS) mutation nomenclature guidelines. Both of these activities play an important role in improving the quality of the test result.”

When asked about his overall recommendations and future plans for the scheme, Patton felt that although the improvement of the quality of testing is happening, there’s still more to do: “Annual participation in EQA should be seen as the norm for all laboratories offering a diagnostic test if they are serious about ensuring that they offer a high quality testing service.”

When applied correctly, personalized medicine can help identify not only patients who are most likely to benefit from a particular therapeutic product, but also those likely to be at increased risk of serious side-effects as a result of treatment. Furthermore, accurate diagnostics can also monitor a response to treatment with a particular therapeutic product, to achieve improved safety. In order to ensure the accuracy and achieve confidence of diagnostic testing/tumor genotyping, a myriad of options are available of which sustained evaluation and validation through reference materials, such as the EQA, are essential.