Genomic Landscape of Primary Resistance to Osimertinib Among Hispanic Patients with EGFR-Mutant Non-Small Cell Lung Cancer (NSCLC): Results of an Observational Longitudinal Cohort Study

An observational longitudinal cohort study was conducted by the Latin American Consortium for Investigation of Lung Cancer (CLICaP) using the data deposited in a centralized database (CLICaP Real World Data/Evidence database). Patients included in the EGFRm NSCLC database were assessed from January 2018 to May 2022. All patients were residents of Bogota, DC, Colombia, and demographic variables such as age, sex, and smoking history were collected from the clinical record. Eligible patients were those with EGFRm analyzed by molecular diagnosis (N = 578) (Fig. 1). Among these patients, we defined cohort A as cases with intrinsic resistance to first-line osimertinib with measurable disease progression within the first 3 months of treatment (N = 16). In contrast, cohort B were patients with long-term responses to osimertinib treatment or progression-free survival (PFS) with osimertinib > 18 months. Finally, this control group was selected from a cohort of long survivors with genomic evaluation by next-generation sequencing (NGS) and complete clinical follow-up. Cases with similar clinical characteristics were sought without being able to match them, due to the unusual nature of patients with primary resistance.

At diagnosis, tumor tissue was acquired from all included patients to undergo molecular characterization, including PD-L1 expression by immunohistochemistry (IHC) and a standardized NGS test. A liquid biopsy was also obtained before starting osimertinib to evaluate Bcl-2 interacting mediator of cell death (BIM) polymorphisms. In the baseline tumor tissue, BIM and AXL mRNA expression levels were quantified using reverse transcription-polymerase chain reaction (RT-PCR). Cohort A and B patients received first-line osimertinib (80 mg orally once daily) until disease progression or unacceptable toxicity was documented. Patients were followed for 48 months and assessed for progression. After 8 weeks of treatment with osimertinib, a new liquid biopsy was performed to estimate the EGFRm status by droplet digital PCR (ddPCR). The NGS test in liquid biopsy was repeated when progression was found, and tissue collection was attempted when there was suspicion of histological transformation (SCLC). The study design and patient inclusion are summarized in Fig. 1.

Patients included in the study were assessed clinically every month. Radiological follow-ups were completed every 9–12 weeks. Clinical, demographic, and molecular variables (including all mutations and tumor mutational burden [TMB]) were stored in a centralized and deidentified database at the CLICaP/Foundation for Clinical and Applied Cancer Research—FICMAC (Bogotá, Colombia). All included patients provided signed informed consent. In addition, an Institutional Review Board and Privacy Board waiver was obtained to facilitate retrospective clinical-pathological and molecular data (Lung Cancer-FICMAC/CLICaP Platform—Registration No. 2012/014, Kayre, Bogotá, Colombia, and Research Ethics Committee of Universidad El Bosque—Registration No. PCI-2018-10171).

Overall response rate (ORR) was defined as the sum of complete responses (CRs) and partial responses (PRs), as assessed by an independent, blinded radiologist according to the Response Evaluation Criteria in Solid Tumors (RECIST) V1.1. For PFS, time to event was calculated from the date of treatment start until radiographic disease progression as assessed by a blinded radiologist according to RECIST V1.1 or death by any cause. Observations for patients who did not experience progression were censored at the patient-specific last follow-up. OS was estimated from diagnosis until death by any cause or loss to follow-up.

PD-L1 expression was determined by IHC using the Dako 22C3 pharmDx kit, considering more than 100 tumor cells in the slide section for accurate PD-L1 readings. PD-L1 testing was completed on biopsies taken at diagnosis. For survival analyses, patients were stratified according to PD-L1 status (i.e., ≥ 50%, 1–49%, and < 1% subgroups). Patients with unknown PD-L1-expression status were also included in this study.

The first blood sample was collected at the time of diagnosis before the start of treatment. Circulating free DNA (cfDNA) was quantified using the QX200 Droplet Digital PCR System (Bio-Rad Laboratories, Hercules, CA). For the EGFR multiplex assays, a final PCR mix volume of 20 μL was manually loaded into wells of a DG8 cartridge (Bio-Rad Laboratories) with 70 μL of Droplet Generation Oil for probes (Bio-Rad Laboratories). After droplet generation by the QX200 Droplet Generator (Bio-Rad Laboratories), 40 μL of the sample was transferred into a 96-well PCR plate and amplified with a C1000 Touch Thermal Cycler (Bio-Rad Laboratories), using the following thermal cycling conditions: 95 °C for 10 min, 40 cycles at 94 °C for 30 s, 55 °C for 1 min (2 °C/s), and 98 °C for 10 min, and a final cooling step to 12 °C (1 °C/s). The droplets were analyzed using the QX200 Droplet Reader (Bio-Rad Laboratories), and data were analyzed using QuantaSoft software version 1.7.4.0917 and QuantaSoft Analysis Pro software version 1.0.596 (Bio-Rad Laboratories). Thresholds were placed manually, and the fractions of positive and negative droplets were used to calculate the concentration and fractional abundance of target DNA sequences with their 95% Poisson-based confidence intervals (CI). EGFRm primers and probes were based on previous studies [27, 28].

Samples were lysed in a buffer containing EDTA, SDS, and proteinase K. Then RNA was extracted with phenol-chloroform-isoamyl alcohol, followed by precipitation with isopropanol in the presence of glycogen and sodium acetate. RNA was homogenized in water and treated with DNAse I to avoid DNA contamination. Complementary DNA (cDNA) was synthesized using M-MLV (Moloney Murine Leukemia Virus Reverse Transcriptase) retro-transcriptase enzyme. cDNA was added to Taqman Universal Master Mix in a 12.5-μL reaction with specific primers and probes for each gene. The primer and probe sets were designed using the LightCycler^® Probe Design Software 2.0 (Roche Diagnostics Corporation, Indianapolis, Indiana) according to their Ref Seq (National Center for Biotechnology Information). Quantification of gene expression was performed using the LightCycler^® 480 Real-Time PCR System (Roche Diagnostics Corporation, Indianapolis, Indiana) and calculated according to the comparative Ct method. The final results were determined as follows: 2-(ΔCt sample-ΔCt calibrator), where ΔCt values of the calibrator and sample are determined by subtracting the Ct value of the target gene from the value of the endogenous gene (β-actin). Commercial RNA controls were used as calibrators (Stratagene, La Jolla, CA, USA). In quantitative experiments, a sample was considered non-viable when the standard deviation (SD) of the Ct values was ≥ 0.30 in two independent analyses. To dichotomize the mRNA expression levels, AXL 1.42 (high AXL [AXL-H] ≥ 1.42) and BIM 0.98 (high BIM [BIM-H] ≥ 0.98) were used as cut-off points; the cut-off points are based on the values described in the existing evidence [29, 30].

cDNA synthesis prior to library preparation for the RNA panel was carried out using SuperScript™ VILO™ cDNA Synthesis Kit (Thermo Fisher Scientific, 2018, 11754050). Library preparation was carried out using the Oncomine™ Comprehensive Assay Plus (Thermo Fisher Scientific, A49667) and RNA Oncomine™ Fusion's assay (Thermo Fisher Scientific) kits following the instructions of the manufacturer and using a total of 10 ng input DNA and/or RNA per sample (minimum 0.83 ng/μL sample DNA concentration). A maximum of seven DNA samples were prepared per run (six samples if both DNA and RNA analyses were required) on an Ion 540™ Chip (Thermo Fisher Scientific, A27765) using the Ion S5™ sequencing system (see the electronic supplementary material, Supplementary Data 1 and 2). The DNA panel can identify hotspot mutations and copy number variations (CNVs), full coding DNA sequences (CDSs), and RNA fusions (Supplementary Table 1 shows the genes and alterations included in the analysis). Template preparation was performed on the Ion Chef System (Thermo Fisher Scientific) using the Ion PGM Hi-Q Chef Kit and/or the Ion One Touch™ 2 System using the Ion PGM Template OT2 200 Kit. Sequencing was performed using the Ion PGM Hi-Q Sequencing Kit on the Ion Torrent Personal Genome Machine. TMB was measured by running the Oncomine Tumor Mutation Load Assay (Thermo Fisher Scientific, A37909), which enables accurate measurements of somatic mutations without needing a matched normal sample. This assay covers a large genomic footprint of 1.7 Mb, encompassing 1.2 Mb of exonic sequence, to enable accurate mutation counts for samples with a range of TMBs. For the analysis, TMB–high (TMB-H) was defined as ≥ 10 mutations/megabase [mut/Mb]) using the same cut-off as in the KEYNOTE-158 study (NCT02628067) [31].

Ion Torrent Suite™ Browser version 5.0 and Ion Reporter™ version 5.0 were used to analyze the NGS data. The Torrent Suite™ Browser was used to perform initial quality control, including chip loading density, median read length, and the number of mapped reads. The Coverage Analysis plugin was applied to all data to assess amplicon coverage for regions of interest. Variants were identified by the Ion Reporter filter chain 5% Oncomine™ Variants (5.0). A cut-off of 500X coverage was applied to all analyses. All identified variants were checked for correct nomenclature using Alamut Visual v.2.7.1 (Interactive Biosoftware). Any discrepancies in variant identification between Ion Reporter and Alamut were validated manually using the Integrative Genomics Viewer and NextGENe^® v2.4.2 (SoftGenetics^®) [32, 33].

All analyses were conducted on IBM SPSS Statistics software version 25.0 (SPSS Inc. Chicago, IL, USA). Descriptive analyses were used to provide an overview of the study population characteristics. Categorical variables were assessed via the Chi-square test or, whenever appropriate, Fisher’s Exact test. OS and PFS were reported as Kaplan–Meier survival curves. Univariable regression analysis models were generated to assess potential confounders. Mann–Whitney U test with a two-sided P value was established to determine statistically significant outcomes. There were no adjustments made for multiple comparisons, and in all cases, the significance level was 0.05.

The clinical information derived from 32 Hispanic patients with NSCLC and EGFRm is summarized in Table 1. Demographic characteristics indicate that all patients were diagnosed at stage IV, and most patients had an adenocarcinoma histology type (96.9%), mainly solid or lepidic patterns (56.3%). There is a higher proportion of female patients in both cohorts (60%), and at least 65% had a negative smoking history. Most patients had pleuropulmonary compromise (87.5%) and good functional status (Eastern Cooperative Oncology Group, ECOG performance status 0–1, 90.7%). Samples were obtained from the lung or pleural space in 90.6%. Most patients had a liquid biopsy at progression (96.9%), and in two cases, tumor tissue was obtained by repeated biopsy. Both cases confirm the suspicion of differentiation to SCLC. Overall, the distribution of EGFR-sensitizing mutations was comparable in the population, including ex19dels and L858R mutations. However, patients in cohort A had significantly more frequent alterations in exon 21 than in cohort B in exon 19 (P = 0.0001). Seven patients had secondary EGFRm (21.9%), which occurred more frequently in cohort B (P = 0.32). A baseline T790M mutation was identified in five cases, four in cohort A (P = 0.16), with a mean allele frequency of 39 (SD ± 20). Thirteen patients (87.5%) in cohort A had more than three co-mutations at the time of diagnosis, a significantly higher finding than for cohort B (P = 0.0001). Almost all baseline TP53 commutations (82.4%) and those involving the RAF pathway were identified among patients in cohort A (P = 0.001 and P = 0.009). Besides EGFR and TP53 genes, the main commutations identified at baseline included RB1, AT-rich interactive domain-containing protein 1A [ARID1A], and BRAF. Fig. 2 shows the primary baseline genomic alterations identified in both cohorts, and Supplementary Table 2 details the variants detected at diagnosis (see the electronic supplementary material).

The baseline TMB was obtained in 27 patients, among whom 22 (68.8%) had a low TMB (TMB-L). It is important to note that cohort A included five patients (31.3%) with TMB-H, while in cohort A, all patients with available data (11/16; 68.8%) had TMB-L (P = 0.054). Based on the mRNA levels stratification, there was a statistically significant difference when AXL and BIM levels were compared among both cohorts (81.3%, N = 13, cohort A patients had AXL-H; 93.8%, N = 15, had BIM-H; 75%, N = 12, cohort B patients had low AXL [AXL-L]; 85.7%, N = 14, had low BIM [BIM-L]). As shown in Supplementary Fig. 1, both differences in mRNA levels between cohorts A and B were statistically significant. Monitoring of response to osimertinib using ddPCR multiplex assays performed at week 8 was positive in 18 patients (56.3%); 93.8% and 18.8% of patients in cohorts A and B had cfDNA detectable by ddPCR, and this difference was statistically significant (P = 0.0001; Supplementary Fig. 2). In terms of PD-L1, most patients had < 1% (N = 3; 18.8%) and 1–49% (N = 11; 68.8%) in cohort A. Similarly, in cohort B, most patients had < 1% (N = 6; 37.5%) and 1–49% (N = 10; 62.5%). Only two patients had PD-L1 ≥ 50% in the study population, both from cohort A.

The median time between the first sample and the start of therapy was 15 days, and the median time between the start of therapy and the first follow-up evaluation was 60 days (± 4 days). The ORRs for cohorts A and B were 6.3% and 100.0% (P = 0.0001), respectively. Only one of the patients with primary resistance to osimertinib had a short-lived PR, with subsequent multifocal progression, including the brain. On the contrary, in cohort B, 31.3% achieved a CR and 68.8% a PR. In both cohorts, the leading site of progression was the lung (≈60%), and cohort A patients had twice as many brain metastases, generally in the first 6 months after diagnosis (P = 0.67). PFS with osimertinib was 3.1 months (95% CI 0.47–7.0) for cohort A and NR for cohort B (P = 0.0001) (Fig. 3A). PFS was significantly higher in patients with ex19del compared with those with L858R mutations (24.5 months, 95% CI 18.2–NR, vs. 7.6 months, 95% CI 4.8–21.1; P = 0.001) (Supplementary Fig. 3; see the electronic supplementary material) and in those without brain metastases at diagnosis compared with patients with baseline brain metastases (24.5 months, 95% CI 18.5–30.5, vs. 7.2 months, 95% CI 1.1–13.2; P = 0.004) (Supplementary Fig. 4). The baseline T790M mutation also negatively impacted PFS (Supplementary Fig. 5); similarly, the presence of mutations in TP53 (P = 0.001), a higher number of commutations at diagnosis (P = 0.0001; Supplementary Fig. 6), and TMB-H (P = 0.017; Supplementary Fig. 7) were associated with worse PFS. In contrast, the PD-L1 expression level did not significantly affect PFS (P = 0.36). Given the biological differences between cohort A and B, high mRNA levels of AXL (P = 0.0001) and low BIM (P = 0.001) also affected PFS, as did the presence of cfDNA in liquid biopsies obtained 8 weeks after starting treatment with osimertinib (P = 0.001).

After a median follow-up of 25.1 months (95% CI 11.0–43.0), the OS for cohorts A and B was 20.1 months (95% CI 10.4–29.2) and NR (P = 0.0001, Fig. 3B). OS was higher for patients with ex19del (P = 0.002; Supplementary Fig. 3B), those without baseline brain metastases (P = 0.013; Supplementary Fig. 4B), those with two or fewer commutations (P = 0001; Supplementary Fig. 6B), and those with TMB-L (P = 0.0001; Supplementary Fig. 7B).

PD-L1 expression was significantly associated with OS, favoring the group of patients with a negative immunophenotype (PD-L1 < 1% 34.1 months, 95% CI 30.6–37.6, vs. PD-L1 1–49% 31.6 months, 95% CI 13.7–24.5, vs. PD-L1 ≥ 50% 19.1 months, 95% CI 13.7–24.5; P = 0.029). Low mRNA expression levels for AXL (P = 0.001) and high mRNA expression levels for BIM (P = 0.0001) and the absence of detectable cfDNA at week 8 of treatment (P = 0.001) were also significantly associated with OS (Supplementary Fig. 8A–C). Contrary to the findings for PFS, the presence of baseline T790M did not significantly impact OS (P = 0.06; Supplementary Fig. 5B). Univariate analysis showed that the variables that most negatively influenced PFS and OS were the presence of brain metastases (P = 0.002), evidence of a more significant number of commutations (P = 0.001), TMB-H (P = 0.0001), and the high levels of mRNA of AXL (P = 0.01) and low BIM (P = 0.02).

Figure 4 summarizes the main alterations in the liquid biopsy after progression to osimertinib (31 patients had a liquid biopsy, and one patient had pleural effusion cell block analysis). Sixty-four genomic alterations were found at the time of progression, 45 in cohort A (70%) and 19 in cohort B (30%). In cohort A, six patients (37.5%) lost the EGFR-sensitizing mutation, a condition that did not occur in any patient in cohort B (P = 0.01). In addition, in cohort A, off-target alterations were dominant, mainly in TP53 (13/35.5%), the RAS pathway (13/35.5%), PI3K (4/8.8%), and RB1 (2/4.4%). In contrast, most alterations in cohort B were due to on-target (6/31.2%) or included MET amplification (3/15.8%), the RAS pathway (2/10.5%), GNAS (2/ 10.5%), or RB1 (2/10.5%). After the second line, two patients had transformation to small cell carcinoma (6.3%), confirmed by biopsy, with mutations in TP53 and RB1. Post-osimertinib mean TMB was 5.0 (SD ± 3.6), with no significant differences between the two cohorts. Supplementary Fig. 9 shows a representative case of primary resistance to osimertinib, including genomic variations throughout the history of the disease (see the electronic supplementary material).

Thirty-one patients received second-line treatment (95.6%), mainly with the IMPOWER-150 study schedule (75%) or with the carboplatin/pemetrexed/bevacizumab combination (12.5%). The ORR for the second line was 68.8% (23 cases achieved PR), with no statistically significant difference between the two cohorts. The median PFS obtained with the second line was 8.1 months (95% CI 3.2–12.1) and 11.2 months (95% CI 9.8–12.5; P = 0.22) for cohorts A and B, respectively. The only variable that affected PFS for the second line was PD-L1 expression (PD-L1 < 1% 7.8 months, 95% CI 6.4–8.7, vs. PD-L1 1–49% 9.2 months, 95% CI 8.7–11.2, vs. PD-L1 ≥ 50% 11.6 months, 95% CI 7.3–13.7; P = 0.021).