EGFR amplification status for clinical trial inclusion: where do we draw the line?

See the article by French et al. in this issue, pp. 1263–1272.

Epidermal growth factor receptor (EGFR) is commonly altered in isocitrate dehydrogenase (IDH) wild type glioblastoma and is an attractive therapeutic target. While genomic alterations in EGFR are diverse, the most common include gene amplification, present in more than 40% of tumors, and the constitutively active internal deletion mutant EGFR variant III (EGFRvIII), present in approximately 20%.¹ EGFR is a known proto-oncogene in glioblastoma and detection of EGFR gene amplification in the setting of an IDH-wildtype diffuse astrocytic glioma is routinely used to differentiate World Health Organization grade IV from lower grade diffuse glioma.² Given the availability of EGFR inhibitors, EGFR is a potential therapeutic target in glioblastoma. However, thus far, clinical trials of EGFR inhibitors (single-agent and in combination) have not shown clinical effect despite the high frequency of EGFR alterations in glioblastoma. Reasons for the lack of efficacy may be due to various factors, including: insufficient drug concentration in the tumor; unselected patient populations; decreased drug concentrations due to interactions with anti-epileptic drugs; lack of predictive biomarkers; and/or escape mechanisms via compensatory pathways.^3,4 Currently, multiple trials are under development with inhibitors that target EGFR and vaccines that target EGFRvIII. The future success of these trials relies on preclinical molecular understanding of EGFR inhibitor resistance, combinatorial approaches to avoid escape mechanisms, and the ability to accurately define EGFR amplified patient populations.

Several clinical assays have been developed to assess EGFR status in tumors, including fluorescence in situ hybridization (FISH), chromogenic in situ hybridization (CISH), targeted next generation sequencing (NGS), real-time quantitative (RT-q)PCR, and whole genome array based comparative genomic hybridization.⁵ Although FISH is routinely used in many centers, for clinical trials there is neither consensus on the optimal assay to use nor well-defined guidelines on how different assays can be compared across centers. Lack of consensus for the most appropriate assays and corresponding cutpoints to determine EGFR status have added to the difficulty in correlating genomic alterations in EGFR with response to EGFR inhibitors.

Until now, few studies have directly compared different assays for EGFR. In this issue of Neuro-Oncology, French et al⁶ use data from a large clinical trial cohort to compare 3 different assays (FISH, RT-qPCR, and NGS) and investigate corresponding cutpoints. The Intellance 2 trial is a controlled, randomized phase II trial in EGFR-amplified recurrent glioblastoma patients to assess a survival benefit for those treated with depatux-M, a microtubule toxin targeted to EGFR amplified cells, in combination with temozolomide compared with depatux-M alone versus an alkylator only–based control arm. Assessment of EGFR amplification via FISH and expression via RT-qPCR were available for over 1000 patients assessed for eligibility. EGFR amplification for eligibility was defined as 15% amplified nuclei of the 50 examined per tumor. Of the 260 patients included in the trial, NGS was available for 226.

In this study, EGFR amplification as measured by FISH appears to have 2 distinct subpopulations, either very low (<15% amplified nuclei) or very high (>90% amplified nuclei) with just over 10% of the tumors falling in between. On the other hand, EGFR expression as measured by RT-qPCR appears to be a bimodel distribution with peaks relatively close to each other. To compare the 2 assays, receiver operating characteristic (ROC) plots were drawn for gene expression in determining EGFR amplification by FISH. For the ROC, an optimal threshold of EGFR amplification was defined as >77% amplified nuclei, and 2 cutpoints for expression were chosen based on: the best sum of sensitivity plus specificity as well as restricting the type I error to 0.05. Targeted NGS to determine EGFR copy number was available for patients deemed to be EGFR amplified and trial eligible (defined as >15% of amplified nuclei by FISH). In this study, copy number counts correlate with amplification. A subset, however, had low copy numbers (<4) and correspondingly low amplification, and lacked EGFRvIII expression. Thus, the authors postulate that a higher percentage of cells with EGFR amplification (eg, 50%) would be a more appropriate threshold for future eligibility criteria, compared with the 15% criteria for eligibility in the Intellance 2 trial.

Previous analyses using FISH have suggested a mosaic pattern of receptor tyrosine kinase amplification within glioblastoma.⁷ For EGFR, however, amplification is less heterogeneous,⁸ and French et al⁶ demonstrate that tumors can be considered EGFR amplified or not. In contrast, EGFRvIII is subclonal and demonstrates intratumoral heterogeneity. Furthermore, deep sequencing and single cell analyses have revealed additional subclonal alterations in EGFR that may be therapeutically relevant.⁸ In glioblastoma, EGFR gene amplifications are extrachromosomal.⁹ As these DNA elements are not equally segregated at mitosis, they can contribute to both intratumoral heterogeneity and EGFR inhibitor resistance. They may also contribute to the variable stability of these subclonal alterations at recurrence.¹⁰ Future studies will be needed to determine the role of subclonal alterations and extrachromosomal DNA in therapy response.

In conclusion, criteria to rigorously define amplification status of EGFR or EGFRvIII expression for clinical trial eligibility do not exist. French et al⁶ compare 3 assays for determination of EGFR amplification. Using FISH, they find 2 distinct populations of tumors exhibiting either a very low or a very high percentage of amplified nuclei with few tumors between. Using targeted NGS, they observed a correlation with amplification, but also identify a small group of tumors deemed amplified by a FISH cutpoint of >15% but with <4 gene copies by NGS. As the majority of these tumors contained relatively few EGFR-amplified cells by FISH, the authors suggest that a higher threshold for percent amplified nuclei is appropriate to exclude the truly low copy numbers. Due to a lack of sequencing data on the non-amplified tumors excluded from the trial, it is not possible to extrapolate an association between FISH and NGS in unamplified tumors and, thus, to define a threshold of copy number by NGS.

Using RT-qPCR, they suggest that expression is correlated with amplification by FISH and that it could act as a surrogate. As noted, expression cutpoint specification is complicated by its lack of distinctly separated populations and, additionally, by the significant overlap in expression between amplified and non-amplified tumors. Furthermore, reliance on a provisional FISH cutpoint in the comparison of expression and amplification, as well as the determination of suitable criteria for expression thresholding (sensitivity/specificity vs type I error) make it challenging to specify criteria for patient eligibility for future trials. Clinical outcome data and external test set(s) are needed to validate cutpoints for amplification and expression; however, at the current time neither are available. When faced with a lack of external validation, internal validation via resampling (ie, repeated training/test set splits of the data) is appropriate and can assist with fine-tuning cutpoint selection (eg, 15% vs 50% vs 77% amplified nuclei) as well as reduce bias in estimates of misclassification and area under the curve.¹¹

As the authors state, cutpoints to define EGFR amplification in this study may not reflect clinical efficacy of targeted agents. Furthermore, it may be advantageous to explore multiple cutpoints for the same assay to identify patients with varying likelihoods of response. As this study highlights, rigorous definition/validation of predictive biomarkers is critical for the application of precision medicine to glioblastoma.

Funding

This work was supported in part by the National Institutes of Health/National Cancer Institute [U01 CA229345 to A.M.M. and J.J.P.].