• © 2012 by American Society of Clinical Oncology

Concordance and Discordance in Tumor Genomic Profiling

Systematic tumor genomic profiling has the following two overarching objectives in clinical oncology: the matching of a cancer drug to a specific tumor genetic context and the identification of tumors for which use of a specific therapy would prove futile or harmful. The impact of KRAS-oncogene mutations on cetuxumab response in colorectal cancer (CRC) comprises an instructive example of the latter, whereas BRAF and PIK3CA mutations (among many others) are being used to assign patients with cancer to clinical trials of exciting new targeted agents. However, the extent to which genetic heterogeneity modifies both the spectrum of actionable genetic alterations during tumor evolution and their diagnostic evaluation before patient stratification remains a subject of debate.

In the article that accompanies this editorial, Vakiani et al¹ present a genomic profiling study of 736 frozen CRC specimens obtained from more than 600 patients. The authors used mass spectrometric genotyping to identify known activating mutations in KRAS, NRAS, BRAF, and PIK3CA oncogenes, which are commonly mutated in CRC. The mutational status of TP53, which is the most common CRC tumor-suppressor gene, was determined by using Sanger sequencing. A subset of tumors was analyzed by using array-based comparative genomic hybridization.

The spectrum of genomic aberrations identified was characteristic of CRC. RAS and PIK3CA mutation frequencies (43% and 12%, respectively) were virtually identical in primary and metastatic tumors. BRAF mutations were notably less frequent in metastatic specimens, which may have been related to the aggressive disease course that is typical of this subtype (eg, comparatively few patients with BRAF mutatations may have survived or been candidates for a rebiopsy). TP53 mutation frequencies increased steadily from 8% of adenomas to 53% of metastases.

The key findings were derived from the following two subgroups: a set of 84 patients for whom both primary and metastatic specimens were available and a second group of 31 metastatic pairs obtained from the same patient. An initial analysis of these subgroups uncovered the following pivotal observation of the study: the mutational concordance between matched pairs across the five cancer genes analyzed ranged from 89% to 95%. At one level, this result could be interpreted favorably because it provided reassurance against major technical variances in sample preparation or mutation detection. In contrast, a 5% to 10% discordance rate raised two worrisome possibilities that either the genetic testing platforms rendered inaccurate mutation calls in some instance, or patient-matched specimens occasionally harbored distinct mutation profiles. The approach of the authors to sort through the basis for discordant mutational spectra provided a generally instructive framework for clinicians and translational investigators who seek to apply mutational profiling more broadly in the cancer genomic era.

Clearly, a 5% to 10% discordance rate between assay results and the ground truth of the lethal tumor fraction could prove highly detrimental to treatment or clinical trials on the basis of genetic criteria. The failure of a diagnostic assay to detect KRAS or NRAS mutations present in CRC tumors (false-negative results) may lead to the futile use of cetuximab (ie, several weeks of ineffective and costly treatment and the potential for unwarranted adverse effects). Conversely, false-positive KRAS results could result in the failure to administer cetuximab-based therapy, which affords clinical benefit in some KRAS wild-type tumors. In other solid tumor contexts (eg, EGFR mutations and erlotinib treatment in non–small-cell lung cancer), mutational false positives might also lead to the erroneous use of anticancer agents.

Assortments of methodologic and biologic factors influence the likelihood of false-positive or false-negative tumor genomic profiling results (Fig 1). Certain technical aspects are germane to the genetic assay itself, such as the robustness and reproducibility of the technology used. Another procedural component relates to the quality of the tumor DNA obtained for analysis. In the study by Vakiani et al,¹ fresh or frozen tumor material was used for the primary analysis, which generally yields high-quality tumor genomic DNA. More often, archival specimens stored as formalin-fixed, paraffin-embedded (FFPE) tissue comprise the most accessible clinical option. However, formalin can fragment the genomic DNA and introduce chemical modifications during the fixation and storage process that lead to false-positive mutation calls. The latter effect poses a particular risk when polymerase chain reaction (PCR) followed by Sanger sequencing is applied to small quantities of degraded genomic DNA from FFPE material. Techniques that augment the quantity of genetic material (such as whole-genome amplification) can also produce false-positive mutation calls.

View larger version:

• In this window
• In a new window

• PowerPoint Slide for Teaching

Fig 1.

Factors that affect the accuracy of tumor genomic profiling. As the demand for tumor genetic information has increased in the clinical oncology arena, the need for robust, high-throughput approaches to genomic profiling has also grown. Toward this end, multiple factors may affect both the accuracy of the genetic test itself and the ability of such tests to reflect the salient driver genetic events within a given tumor. Some factors are germane to the tumor itself (eg, genetic heterogeneity or stromal admixture) or the means by which it was preserved (frozen or formalin-fixed, paraffin-embedded tissue). Other factors relate to the particular genetic platform and its intrinsic false-positive and false-negative rates. Finally, the extent to which validation tests are included to replicate or otherwise confirm the initial testing result may affect the overall accuracy of tumor genomic profiling.

Similarly, false-negative results may occur when heavily degraded FFPE DNA is used in small quantities. Known as the allele dropout phenomenon, these mistakes arise when the mutant allele fails to be sampled during PCR amplification. The likelihood of allele dropout increases when the mutation is heterogeneous or heavily contaminated by stromal or inflammatory cells; however, the effect may often be mitigated by repeat tissue sampling and genetic analysis. Indeed, a focused reassessment of the initial discordant results (by using independent FFPE specimens) reduced the rate of discordance from 5% to 10% to less than 3% in the study of Vakiani et al.¹

Results of genomic profiling may also be profoundly influenced by the cellular and genetic heterogeneity of the tumor. As noted, low tumor purity (which results from the admixture with stromal, vascular, or inflammatory cells) decreases assay sensitivity, even when high-quality DNA from fresh or frozen material is used. The assay sensitivity of mass spectrometric genotyping diminishes when the mutant allelic fraction is less than 20%² (which can result from polyploidy as well as stromal admixture), although alternative approaches such as allele-specific PCR³ or deep sequencing (eg, pyrosequencing or massively parallel sequencing)^4,5 may retain robust sensitivity for specific loci at allelic fractions ≤ 5%. The importance of genomic heterogeneity across distinct tumor foci, and even within the same tumor site, has received increasing attention in recent years. Although some baseline heterogeneity is inevitable among individual tumor cells (and is governed by the base mutation rage operant within tumor cells), occasional reports have suggested that actionable genetic alterations may exhibit unexpectedly high local and regional heterogeneity in some tumor types. Extensive genetic heterogeneity would pose a considerable challenge from the standpoint of personalized cancer therapy by mandating a rebiopsy of a metastatic site (even if primary tumor material was available) and also by reducing the chance that any single metastatic focus would be representative of the salient driver genetic events.

Fortunately, a massive intratumoral genetic heterogeneity does not seem to account for the etiology of the remaining genetic discordances in the CRC cohort of Vakiani et al.¹ Through careful inspection of associated clinical information, the authors arrived at two parsimonious explanations. The first mechanism (operant in two patients), involved the presence of a second primary tumor. In one case, the synchronous tumor was available for analysis; this allowed for the confirmation of the genetic origin of the metastatic clone. A second mechanism postulated that the discordance in TP53 mutations was linked to treatment effects, such as the use of neoadjuvant chemoradiotherapy before resection of a primary rectal tumor. However, several aforementioned technical factors also contributed to the TP53-related discordances (eg, the use of deep sequencing to identify two TP53 mutations that were missed by mass spectrometric genotyping and the absence of a tumor in one source of primary tissue).

Having accounted for each mutational discrepancy, Vakiani et al¹ concluded that the analysis of primary CRC specimens (instead of metastases) is generally sufficient for the identification of mutations in several common cancer genes. This notion was buttressed by the concomitant analysis of comparative genomic hybridization data. The conclusion of the authors seems sound for founder cancer gene mutations that are highly prevalent in CRC; however, it remains unclear whether this premise will hold firm as the scope of genomic profiling expands in the future. For example, mutations that inactivate tumor-suppressor genes such as PTEN (which may phenocopy many PIK3CA mutations) and NF1 (which may lead to dysregulated RAS signaling) are also recurrently mutated in CRC and other tumor types. The therapeutic significance of such events remains to be determined; however, some tumors may acquire mutations in these genes later in their evolution. In such cases, an analysis of the primary tumor may not always suffice for genetically informed treatment decisions.

Finally, it is instructive to revisit the initial 5% to 10% discordance rate linked to mass spectrometric genotyping and Sanger sequencing. The authors were alerted to these discordances by analyzing matched specimens; however, the typical clinical case will comprise a single (and often limited) tumor specimen. How can physicians be sure that any given tumor genomic profile has not been confounded by methodologic or tumor biologic constraints? The incorporation of redundant and/or inline validation assays for “mission-critical” genetic loci may constitute one possible solution.⁶ Alternatively, the advent of massively parallel sequencing may impel the design and implementation of more-powerful diagnostic platforms.⁷ Many additional studies such as the study described by Vakiani et al¹ will be needed to vet the interface between technology performance and tumor biology as cancer medicine becomes increasingly genetics driven.