Genomic Profiling Identifies Outcome-Relevant Mechanisms of Innate and Acquired Resistance to Third-Generation Epidermal Growth Factor Receptor Tyrosine Kinase Inhibitor Therapy in Lung Cancer

PURPOSE Third-generation epidermal growth factor receptor (EGFR) tyrosine kinase inhibitors (TKIs) are effective in acquired resistance (AR) to early-generation EGFR TKIs in EGFR-mutant lung cancer. However, efficacy is marked by interindividual heterogeneity. We present the molecular profiles of pretreatment and post-treatment samples from patients treated with third-generation EGFR TKIs and their impact on treatment outcomes. METHODS Using the databases of two lung cancer networks and two lung cancer centers, we molecularly characterized 124 patients with EGFR p.T790M-positive AR to early-generation EGFR TKIs. In 56 patients, correlative analyses of third-generation EGFR TKI treatment outcomes and molecular characteristics were feasible. In addition, matched post-treatment biopsy samples were collected for 29 patients with progression to third-generation EGFR TKIs. RESULTS Co-occurring genetic aberrations were found in 74.4% of EGFR p.T790-positive samples (n = 124). Mutations in TP53 were the most frequent aberrations detected (44.5%; n = 53) and had no significant impact on third-generation EGFR TKI treatment. Mesenchymal-epithelial transition factor (MET) amplifications were found in 5% of samples (n = 6) and reduced efficacy of third-generation EGFR TKIs significantly (eg, median progression-free survival, 1.0 months; 95% CI, 0.37 to 1.72 v 8.2 months; 95% CI, 1.69 to 14.77 months; P ≤ .001). Genetic changes in the 29 samples with AR to third-generation EGFR TKIs were found in EGFR (eg, p.T790M loss, acquisition of p.C797S or p.G724S) or in other genes (eg, MET amplification, KRAS mutations). CONCLUSION Additional genetic aberrations are frequent in EGFR-mutant lung cancer and may mediate innate and AR to third-generation EGFR TKIs. MET amplification was strongly associated with primary treatment failure and was a common mechanism of AR to third-generation EGFR TKIs. Thus, combining EGFR inhibitors with TKIs targeting common mechanisms of resistance may delay AR.


Study Design, Patient Selection, and Tumor Tissue Collection
To determine the frequency of co-occurring genetic aberrations in samples of EGFR p.T790M-mediated resistance to early-generation EGFR TKIs, we systematically searched the databases of the Network Genomic Medicine, the NOWEL network, the Department of Thoracic Oncology of the Netherlands Cancer Institute, and the Institute of Oncology at the Vall d'Hebron University Hospital for patients with non-small-cell lung cancer (NSCLC) who fulfilled the following selection criteria (cohort A; patients a1 to a68/b1 to b56;  In a subset of patients from cohort B, a rebiopsy was performed at disease progression for identification of mechanisms of AR. These patients were grouped in cohort C (Fig 1).
In all patients, tumor tissue was collected in growing lesions by aspiration biopsy, core needle biopsy, or excisional biopsy (Data Supplement). All patients consented to the procedures according to local and Good Clinical Practice standards. Procedures were approved by the local ethics committees or review boards.
We identified three patients with EGFR p.G724S mutations (see Results). A more detailed description will be published elsewhere. 29

Efficacy Assessments
Patients treated within the osimertinib CUP or in clinical routine received scans as clinically indicated and per local practice. In patients treated within clinical trials, scans were performed according to the protocols. Scans were evaluated according to RECIST 17 Amplification of ERBB2 was positive if the ERBB2/ CEP17 ratio was greater than or equal to 2.0 or the average ERBB2 GCN per cell was greater than or equal to 6.0. In the post-treatment samples (cohort C) of b41 to b56, MET and ERBB2 status was assessed by fluorescence in situ hybridization or chromogen in situ hybridization only if CNVs were detected by MPS.
Small-cell lung cancer transformation was assessed using microscopy by experienced pathologists. Transformation was defined by the occurrence of small-cell lung cancer histology.

Statistical Analyses
RR was defined as the percentage of complete remissions and PR as best response. Progression-free survival (PFS) indicated the time from treatment start until PD or death. Overall survival (OS) was defined as the time from first diagnosis until death. Time-to-event end points were analyzed using the Kaplan-Meier estimator. Qualitative variables were summarized by count and percentage; quantitative variables were summarized by mean, median, and range. Differences in time-to-event distribution were evaluated by the log-rank test, and statistical association between any two categorical variables was assessed by Fisher's exact test; 95% CIs for proportions were calculated using the Clopper-Pearson (binominal) formula. P values less than or equal to .05 were considered statistically significant. The frequencies of the genetic changes were calculated on the basis of the number of patients screened for each aberration. Calculations were performed in Excel

RESULTS
Clinical and Molecular Characteristics of Patients With p.T790M-Positive AR to Early-Generation EGFR TKI Therapy (cohort A) and Impact on Outcome of Third-Generation EGFR TKI Treatment (cohort B) The molecular characteristics of cohort A (n = 124) and the impact on OS are illustrated in the Data Supplement. A total of 56 patients (45%) from cohort A fulfilled the selection criteria for cohort B and showed the clinical characteristics outlined in the Data Supplement. Patients received third-generation EGFR TKI treatment with osimertinib (n = 37; 66.1%), nazartinib (n = 11; 19.6%), and rociletinib (n = 8; 14.3%).
The RR in the overall population was 61% (95% CI, 46.8% to 73.5%), and median PFS was 8.0 months (95% CI, 6.9 to 9.1 months; Table 1). Efficacy of osimertinib and nazartinib treatment was not significantly different. One PR was confirmed while the patient was taking rociletinib, and RR was 12.5% (95% CI, 0.3% to 52.7%). Median PFS with rociletinib was 3.7 months (95% CI, 0.0 to 7.9 months). Initial tumor stage, gender, smoking status, and the number of prior EGFR TKIs had no significant impact on treatment outcomes (Table 1). A map of molecular aberrations found in patients from cohort B is displayed in Figure 2 Fig 3).

Mechanisms of AR to Third-Generation EGFR TKI Therapy (cohort C)
In total, 44 patients (79%) in cohort B had disease progression, and tumor samples were available from 29 patients (52%; cohort C; Figs 1 and 2). The results of the molecular analyses were matched with pretreatment samples and one earlier sample, if possible, to distinguish between passenger and acquired aberrations. The calculation of the frequency of changes in a gene compared with the pretreatment sample was performed in matched samples only (Fig 2; Data Supplement)   percentage of samples in which we detected acquired changes in the molecular pattern was 89% (n = 23). Loss of EGFR p.T790M was by far the most common molecular change (n = 13 of 29; 45%). Isolated loss of p.T790M without any other genetic change was detected in four samples (n = 4 of 26; 15%). However, we found small-cell lung cancer transformation in one sample (4%), which showed loss of p.T790M. Acquisition of high-level MET amplification was detected in seven samples (n = 7 of 25; 29%), and the mean MET copy number increased significantly between pretreatment and post-treatment biopsies (GCN mean, 2.8 v 6.3; two-tailed, pairwise t test P = .02; Data Supplement). The third most common genetic changes in cohort C were acquisition of EGFR p.C797S (n = 3 of 29; 10%), of which two were in cis and one in trans position, and loss of p.T790M with acquisition of p.G724S (n = 3 of 28; 11%). Amplification of ERBB2 was observed in two samples (7%) and occurred together with MET amplification. Both patients were MET and ERBB2 WT in pretreatment samples and had a long PFS of 15.1 and 19.7 months, respectively. Common KRAS mutations were detected in two samples (7%)-KRAS p.G12S and p.G12C. The KRAS p.G12C mutation involved the change of two consecutive nucleotides c.33_34delinsCT on the same allele, with an allelic fraction of 2.7%. Both patients are illustrated in Figure 4. Acquired mutations in BRAF (p.V600E), TP53 (p.E180*), and PTEN (p.S229*) were detected in one sample each (4%  Progression-Free Survival (%) Overall Survival After PD (%)

Genetic Clustering of AR Mechanisms to Third-Generation EGFR TKIs and Impact on Third-Generation EGFR TKI Efficacy (cohort C)
Occurrence of multiple mechanisms of AR followed a distinctive pattern (Fig 5A). Changes in EGFR, such as loss of p.T790M and acquisition of p.C797S, were mutually exclusive. Except for one patient, CNV in MET and/or ERBB2 did not occur together with p.C797S or loss of p.T790M. In the samples with new BRAF and TP53 mutations, as well as in one of the patients with KRAS-mutant disease, p.T790M was lost. ERBB2 amplifications were all found in samples that also harbored amplifications of MET.
We therefore clustered the patients in four groups: (I) changes outside of EGFR only, (II) changes in EGFR and outside of EGFR, (III) changes in EGFR only, and (IV) no changes found (Fig 5A). Seven patients (24%) belonged to cluster I, and 11 belonged to cluster III (38%). Five patients (17%) had changes in and off the target at the same time (cluster II). No changes were found in six patients (21%; cluster IV). In patients treated with osimertinib, a larger fraction belonged to cluster III than cluster I or II (n = 10; 47. 6% for III v n = 5; 23.8% for I and II). In patients treated with rociletinib, this trend was inversed (changes in EGFR, n = 0; 0% v no changes found, n = 4; 100%). Of the four patients treated with nazartinib, two (50%) displayed changes outside of EGFR. In one patient (25%), changes in EGFR were found. No changes were found in another patient (25%). The statistical significance for a cross table stratified by cluster and type of EGFR TKI was P = .002 (Fisher's exact test). Differences in PFS by cluster were not statistically significant (Fig 5B). Similarly, OS after PD was also not significantly different between the clusters (Fig 5C).
Nine patients (31%) received a treatment trying to match the targets identified in the molecular analysis. Median duration of treatment was 1.8 months (95% CI, 0.3 to 3.3 months) for targeted approaches versus 2.6 months (95% CI, 0.0 to 5.2 months) for chemotherapy (n = 4; P = .891; Data Supplement).
Our first analysis revealed a high genomic heterogeneity in patients with p.T790M-positive resistance to earlygeneration EGFR inhibitors. Some of these aberrations, for example, amplifications of MET, are known to cause AR to any EGFR TKI. [17][18][19] The role of others, such as TP53, PTEN, PIK3CA, and CTNNB1, however, is still not well characterized.
We therefore sought to determine the effect of these aberrations on third-generation EGFR TKI treatment outcomes. Overall efficacy and OS were similar in patients treated with osimertinib and nazartinib and in concordance with the data reported so far. However, patients treated with rociletinib had a worse outcome than reported previously, which may be caused by the low patient number. Several groups have reported on an association of TP53 mutations and shorter OS in patients with EGFR-mutant NSCLC. However, most of these reports were not statistically significant, and similarly, OS, RR, and PFS were only numerically reduced in patients with TP53 mutations in our study. [32][33][34][35][36][37][38] Patient numbers with aberrations in PTEN, PIK3CA, and ERBB2 were low, and the differences in treatment efficacy were not statistically significant. However, preclinical models and reports on small patient series suggest a negative impact of these aberrations on EGFR TKI therapy. 7,17,19,39,40 In contrast, survival and treatment efficacy were dramatically impaired in patients with MET-amplified tumors, putting MET in the front line of potential mechanisms of IR.
To define mechanisms of AR to third-generation EGFR TKIs, we analyzed post-treatment biopsies of 29 patients (cohort C) and found that loss of p.T790M was by far the most frequent genetic change. However, only a small fraction of patients had an isolated loss of p.T790M. It is likely that other genetic changes that we did not detect with our analysis may contribute to AR in these patients with a loss of p.T790M and no other genetic change. 23 The acquisition of p.C797S was detected in three patients, and several studies have confirmed the resistance-mediating effect of this substitution to osimertinib treatment. 23,41 In addition, we found the secondary EGFR mutation p.G724S in three samples. In contrast to p.C797S, p.G724S was also in part detected in the samples collected at progression to early-generation EGFR TKIs. 29,42 However, after failure of third-generation EGFR TKI treatment, p.G724S was always co-occurring with loss of p.T790M, suggesting the treatment-induced selection of this mutation. Acquisition of MET amplification was the second most frequent event associated with AR to third-generation EGFR inhibition, and similar frequencies have been described in the literature. 19,23 The high prevalence of MET amplification in IR and AR points out the crucial role of MET in EGFR inhibitor resistance. Interestingly, amplifications of MET and ERBB2 occurred together in two patients. It is unclear whether this reflects the existence of two independent tumor clones or whether both aberrations are acquired in the same clone and how they influence therapy outcome. We also found acquired mutations in KRAS in two patients and a BRAF p.V600E mutation in one patient. Activation of the MEK/ extracellular regulated kinase pathway through KRAS mutations as an escape mechanism and efficacy of the combined EGFR and MEK inhibition was reported previously. 17,26,27 Thus, taken together, treatment of EGFRmutant NSCLC with TKIs targeting EGFR as well as MET and MEK may delay the development of AR and prevent IR in selected patients.
By clustering the genetic findings at AR into four groupsmechanism of resistance off target (I), on target (III), or in both (II), and no changes detected (IV)-we found a distinct molecular pattern depending on the EGFR TKI applied. Changes in EGFR were almost exclusively found in patients treated with osimertinib. In contrast, no patient treated with rociletinib displayed changes in EGFR, and other studies have confirmed the absence of secondary EGFR mutations in patients with progression while taking rociletinib. 19,43 It is conceivable that this effect may be caused by a lower selection pressure of rociletinib on cells with on-target aberrations. We also found a statistically significant association between cluster and ORR, because patients in cluster IV had a markedly reduced ORR to thirdgeneration EGFR treatment. However, differences in PFS or OS after PD were not significant.
In summary, our study first shows that molecular heterogeneity of p.T790M-mutant lung cancer with AR to earlygeneration EGFR TKIs influences efficacy of thirdgeneration inhibitors. Our observations also show the need to integrate information on co-occurring alterations in the design of clinical trials, aiming at a more precise identification of patients who benefit from combined targeted treatment. Because osimertinib has been approved for first-line treatment of EGFR-mutant NSCLC in many countries, our analysis may be of relevance to a decreasing subgroup. But mechanisms of resistance to first-line osimertinib have not been well characterized, and it is conceivable that recurrent mechanisms of resistance to EGFR inhibition such as MET amplification, MET activation, and EGFR p.C797S may also play a major role in this setting.