RET fusions as primary oncogenic drivers and secondary acquired resistance to EGFR tyrosine kinase inhibitors in patients with non-small-cell lung cancer

DNA extraction, library construction, and targeted NGS were performed as previously described in a Clinical Laboratory Improvement Amendments (CLIA)-certified and College of American Pathologists (CAP)-accredited clinical testing laboratory (Nanjing Geneseeq Technology Inc., Nanjing, China) [25, 26]. FFPE samples were de-paraffinized with xylene followed by genomic DNA extraction using QIAamp DNA FFPE Tissue Kit (Qiagen Cat. No. 56404) according to the manufacturer’s instructions. Genomic DNA from fresh tumor tissue was extracted using the DNeasy Blood and Tissue Kit (Qiagen Cat. No. 69504) according to standard protocols. Peripheral blood samples were centrifuged at 1800g for 10 min, followed by cell-free DNA (cfDNA) extraction and purification using QIAamp Circulating Nucleic Acid Kit (Qiagen Cat. No. 55114). Genomic DNA of white blood cells in sediments was extracted using the DNeasy Blood and Tissue Kit (Qiagen Cat. No. 69504) as normal control. Genomic DNA was qualified using Nanodrop2000 (Thermo Fisher Scientific, Waltham, MA), and cfDNA fragment distribution was analyzed on a Bioanalyzer 2100 using the High Sensitivity DNA Kit (Agilent Technologies, Santa Clara, CA, 5067-4626). DNA quantification was performed using the dsDNA HS assay kit on a Qubit 3.0 fluorometer (Life Technology, US). NGS libraries were prepared using the KAPA Hyper Prep kit (KAPA Biosystems) with an optimized manufacturer’s protocol for different sample types. Targeted capture enrichment was performed as previously described [27]. The target-enriched library was then sequenced on HiSeq4000 or HiSeq4000 NGS platforms (Illumina) according to the manufacturer’s instructions.

Sequencing data were processed as previously described [25]. In brief, the data was first demultiplexed and subjected to FASTQ file quality control using Trimmomatic [28]. Low-quality data (QC below 15) and N bases were removed. Raw reads were then mapped to the Human Genome (hg19) using Burrows-Wheeler Aligner (BWA-mem, v0.7.12; https://​github.​com/​lh3/​bwa/​tree/​master/​bwakit). Genome Analysis Toolkit (GETK 3.4.0; https://​software.​broadinstitute.​org/​gatk/​) was employed to perform local realignment, base quality score recalibration, and detect germline mutations. Picard was used to remove PCR duplicates. VarScan2 was applied to detect single-nucleotide variations (SNVs) and insertion/deletion mutations. The limit of detection (LOD) of tumor tissues and plasma samples under specific sequencing depths has been tested repetitively in Nanjing Geneseeq Technology Inc. to ensure the mutation calling results are consistent and have optimal performance for identifying genetic alterations. According to its internal specifications, sequencing of tissue and plasma ctDNA via targeted NGS can reach a sensitivity of 98% and a positive predictive value (PPV) of 95% [26, 29, 30]. SNVs were filtered out if the VAF was less than 1% for tumor tissue and 0.3% for plasma samples. Common SNVs were excluded if they were present in > 1% population in the 1000 Genomes Project or the Exome Aggregation Consortium (ExAC) 65,000 exomes database. The resulting mutation list was further filtered by an in-house list of recurrent artifacts based on a normal pool of whole blood samples. Parallel sequencing of matched white blood cells from each patient was performed to remove sequencing artifacts, germline variants, and clonal hematopoiesis. Genomic fusions were identified by FACTERA [31] with default parameters (≥ 2 reads). The fusion reads were manually reviewed and confirmed on Integrative Genomics Viewer (IGV). By definition, RET fusions were annotated as a protein fusion involving: (i) a 5′ non-RET partner gene and (ii) an intact 3′ RET kinase domain (NM_020975: exon 12–18). In this study, RET fusions were classified into canonical (single or compound KIF5B-RET and CCDC6-RET) and noncanonical RET. In this regard, a noncanonical RET fusion includes: (i) a rearrangement with a partner rather than KIF5B and CCDC6; (ii) A rearrangement with a novel partner gene; and (iii) a rearrangement with an intergenic region (IGR). Tumor mutational burden (TMB, mutation per Megabase) was determined based on the number of missense mutations in the targeted regions of the gene panel covering 0.85 Mb of coding genome, excluding known driver mutations as they are over-represented in the panel. Chromosome instability score (CIS) was defined as the proportion of the genome with aberrant (purity-adjusted segment-level copy number ≥ 3 or ≤ 1) segmented copy number [32]. For gene-level analyses in the baseline cohort, genetic alterations with a mutation frequency ≥ 2% were considered frequently mutated. All 139 genes in the PULMOCAN™ gene panel (Geneseeq Technology Inc.) were subjected to pathway enrichment analyses. Specifically, the genetic alterations identified by targeted panel sequencing were first categorized into ten signaling pathways associated with common hallmarks of cancer that control cell-cycle progression, apoptosis, cell proliferation and growth [33]. The proportion of baseline patients with specific RET fusion variants harboring mutation(s) in the relevant pathways was compared to reveal the differences of co-existing mutations in baseline RET+ patients.

Fisher’s exact tests were used to compare the categorical variables between groups. Kaplan–Meier curves were used to analyze the PFS of various patient groups, and the statistical difference was analyzed using the log-rank test. A two-sided P value of less than 0.05 was considered significant for all tests unless indicated otherwise (*P < 0.05, 0.01 < **P < 0.05, ***P < 0.001). All statistical analyses were performed using R (version 4.1.2).

After excluding patients with either poor NGS quality samples or incomplete RET kinase domain fusions, 451 patients with NSCLC harboring RET fusions were included in the study cohort (Fig. 1). In particular, 380 patients had baseline RET fusions in EGFR-TKI treatment-naïve tissue samples, whereas 71 patients who carried baseline EGFR mutations acquired RET fusions after resistance to EGFR-TKIs. According to the EGFR-TKI treatment regimens, the 71 acquired RET+ patients were subdivided into three groups: (i) patients treated with first-line (1L) 1st- or 2nd-generation (G) EGFR-TKIs (N = 13), (ii) patients treated with second-line (2L) 3^rd-G EGFR-TKIs (N = 51), and (iii) patients treated with 1L 3^rd-G EGFR-TKIs (N = 7).

In the baseline cohort, 58.9% of patients were under 60 years (58.9% vs. 41.1%, P < 0.001), and the proportion of females was significantly higher than males (55.3% vs. 44.7%, P = 0.038, Table 1). Of the 71 patients with acquired RET fusions, a higher proportion of younger (63.4% vs. 36.6%, P = 0.024) and female patients (64.8% vs. 35.2%, P = 0.013) was observed. In both baseline and acquired RET+ patient cohorts, ADC was the predominant histological subtype in NSCLC patients. All 71 patients harbored baseline EGFR mutations, including 47 with exon 19 deletions (19-Del, 66.2%), 23 with substitution mutation L858R (32.4%) and one G719C/S768I double-mutant patient (1.4%).

By targeted NGS, a total of 491 RET fusions were identified in 451 RET+ patients (Fig. 2a), of which KIF5B-RET (51.1%) was the most frequently observed RET rearrangement, followed by CCDC6-RET (23.4%). Notably, 39 patients (8.6%) carried more than one RET fusion, of whom 36 were baseline and 3 were acquired RET+ patients. Next, we tried to depict functional domains and breakpoints of canonical RET fusions. In particular, 78% of KIF5B-RET (195/251) fusions contained the kinesin motor domain and coiled coil domain from exons 1–15 of KIF5B and the kinase domain from exons 12–18 of RET (Fig. 2b). In comparison, 78% (90/115) of CCDC6-RET fusions contained a fused coiled coil domain from exon 1 of CCDC6 and the kinase domain from exons 12–18 of RET. These results were consistent with previous findings, showing that most genomic breakpoints are located within intron 11 of RET, while translocation events may also occasionally occur within intron 7 and 10, resulting in the inclusion of the RET transmembrane domain [34].

The next question we wanted to address was whether the percentage of specific RET fusion variants differs in baseline and acquired RET+ patients. We identified 417 and 74 RET rearrangements in these two cohorts, respectively (Fig. 2c, d). The most commonly observed fusion variants in baseline patients were KIF5B-RET (53.7%), followed by CCDC6-RET (22.5%, Fig. 2c). Surprisingly, KIF5B-RET accounted for only 8.1% of total translocation events in baseline EGFR-mutated NSCLC patients resistant to EGFR-TKIs, whereas CCDC6-RET was identified in 47.3% of translocations events, ranking as the predominant RET fusion variant in acquired RET+ patients (Fig. 2d). The identification of CCDC6-RET was the most common fusion variant, followed by NCOA4-RET, was consistent with previous findings [15]. Furthermore, these results implied that KIF5B-RET and non-KIF5B-RET fusions might have different functionalities in NSCLC. At the same time, noncanonical RET fusions, including those with hitherto unreported partner genes [10], were identified in 23.8% and 44.6% of baseline and secondary patient cohorts, respectively (Additional file 1: Table S1).

As RET fusion variants distributed differentially in baseline and acquired RET+ patients, we investigated whether RET fusion types could be associated with the patient's clinical manifestations. Interestingly, females were more likely associated with KIF5B-RET (P = 0.006) and noncanonical RET fusions (P = 0.015) than CCDC6-RET in baseline patients (Additional file 2: Fig. S1a). However, neither the patient’s age nor cancer stage at diagnosis was directly associated with RET fusion types in baseline or secondary patients (Additional file 2: Fig. S1b-e).

RET fusions are oncogenic drivers that usually occur mutually exclusive to other driver genes in NSCLC patients. Research to date has not yet addressed why RET-rearranged patients with different fusion partners responded differently to treatments. We hypothesized that this could result from different mutational profiles among baseline RET+ patients. Accordingly, we analyzed the NGS data of 380 baseline RET+ patients and profiled their genomic landscape. Genetic alterations, such as TP53, MDM2, CDKN2A/B, ATM, and RB1, were frequently detected in baseline RET+ patients (Fig. 3a). Notably, although 94.5% (360/380) of our baseline RET+ patients contained only RET fusions as the primary oncogenic driver of NSCLC, 21 patients (5.5%) harbored additional driver gene aberrations, including EGFR (11/21), KRAS (6/21), BRAF (2/21), ERBB2 (1/21) and ROS1 (1/21) (Additional file 3: Fig. S2a). However, there was no significant correlation between the patient number and the patient’s clinical characteristics (Additional file 3: Fig. S2b).

Our next objective was to investigate whether specific genomic alterations could be associated with baseline patients harboring specific RET fusion variants. Hence, we performed both gene- and pathway-level enrichment analyses using targeted NGS results of tumor tissues collected from 380 baseline RET+ patients. EGFR mutations were more commonly found in patients with CCDC6-RET (5.0% vs. 0.4%, P = 0.014) and noncanonical RET fusions (7.1% vs. 0.4%, P < 0.001) compared to KIF5B-RET fusions (Fig. 3b). Meanwhile, co-existing mutations in PTEN were more frequently associated with noncanonical RET fusions than KIF5B-RET fusions (7.1% vs. 0.4%, P = 0.042). In addition, CDKN2A mutations were more frequently co-existed with CCDC6-RET than KIF5B-RET (11.3% vs. 2.4%, P = 0.003), while SMAD4 mutations (5.3% vs. 0.0%, P = 0.044) and MYC amplification (6.9% vs. 0.0%, P = 0.009) were more frequently found in patients with KIF5B-RET fusions than those with CCDC6-RET fusions. By performing pathway-level analyses, we noticed that noncanonical RET fusions were more likely to be associated with mutations in the PI3K and RAS/RTK pathways (Fig. 3c). In great contrast, the aberrant MYC pathway more frequently co-occurred with KIF5B-RET than CCDC6-RET (7.8% vs. 0.0%, P < 0.001) in baseline RET-rearranged patients with NSCLC. None of the other oncogenic pathways examined showed a significant difference among patients harboring KIF5B-RET or CCDC6-RET fusions.

Lastly, we examined tumor mutational burden (TMB) and chromosomal instability in baseline RET+ patients, where we found no significant difference in TMB, chromosomal instability score (CIS), or arm-level copy number variations (CNVs) among different types of RET fusions (Additional file 4: Fig. S3). From the above analysis, we depicted mutational profiles of RET+ patients and identified concomitant mutations that may contribute to the differential treatment responses in RET-rearranged patients.

Despite their rarity, it is clear from previous studies that RTK fusions, such as RET rearrangements, are actionable resistance mechanisms to EGFR-TKIs. We aim to increase awareness of this emerging paradigm by comprehensively profiling baseline EGFR-mutated NSCLC patients who acquired RET fusions after developing resistance to EGFR-TKIs. In acquired RET+ patients, second-site EGFR mutations, such as T790M (50.7%), C797S/G (16.9%) and L718V/Q (5.6%), were among the top frequently mutated gene alterations (Fig. 4a). It was also apparent that cell cycle control pathway gene alterations, such as TP53, RB1, CDKN2A, CDKN2B, CDKN1B, and CDK6, co-existed with acquired RET fusions. In addition, ALK and NTRK1 fusions that belong to the RTK family were identified to co-occur with RET fusion in two patients. These results implied that RTK fusions and other genetic variants could potentially serve as resistance mechanisms to anti-EGFR treatment in NSCLC.

To characterize the AR mechanism through RET fusions in EGFR-mutated NSCLC patients and their survival outcomes, we compared the PFS among patients treated with different EGFR-TKI regimens. However, no significant difference was observed (Additional file 5: Fig. S4a). Alternatively, as 2L 3rd-G EGFR-TKI treatment has been employed as a standard practice for EGFR-mutated NSCLC patients with progressive disease on first-line targeted therapy, we performed integrated survival analyses using group 2 patients. Notably, we excluded one patient with the rare EGFR double-mutants and one with SCC histology, leaving 49 patients with only EGFR 19-Del or L858R mutations in the refined cohort (Additional file 5: Fig. S4b). We first performed a univariate analysis on various demographic and mutational features of refined cohort 2 patients. Interestingly, bypass pathway genetic alterations (KRAS and PIK3CA activation mutations, ERBB2 and MET amplification, ALK and NTRK fusions), RB1 and TP53 co-mutation, and ERBB2 copy-number gain had significant associations with the prognosis of acquired RET+ patients who underwent 2L 3^rd-G EGFR-TKI treatment (Additional file 1: Table S2). We then performed multivariate analysis using these three features, and we discovered that they were no longer statistically significant, with bypass mutations and RB1and TP53 co-mutations being close to significance. This result suggested that there might be some interactions or correlations among these prognostic features. Nonetheless, the PFS of patients harboring bypass pathway gene alterations was significantly shorter than wild-type patients (median: 6.7 vs. 12.0 months, P = 0.020, Fig. 4b). In addition, the PFS of patients with TP53 and RB1 co-mutations was significantly shorter than wild-type counterparts (median: 5.5 vs. 10.0 months, P = 0.02, Fig. 4c). ERBB2 gene amplifications that were present in 5.6% of acquired RET+ patients were also associated with poor prognosis in NSCLC patients (median: 5.6 vs. 10.0 months, P = 0.041, Fig. 4d). Notably, although the majority of acquired RET fusions were detected by circulating nucleic acid method using plasma ctDNA, 4 out of 49 patients in the refined cohort contributed only FFPE samples, raising a concern of missing rare fusions compared to tumor-based comprehensive genomic profiling. We, thereby, excluded patients with only tissue samples and re-performed the analysis in Fig. 4 using only ctDNA data (Additional file 6: Fig. S5a). Overall, our results about the co-occurring genetic alterations were consistent when using either all samples (tissue and ctDNA) or only ctDNA (Additional file 6: Fig. S5b-d), suggesting that our conclusions are not likely to be affected by different sample types.

Collectively, comprehensive molecular profiling of acquired RET + patients showed that co-existing genomic alterations, such as TP53 and RB1 co-mutations and ERBB2 amplification, might be associated with a poor prognosis in EGFR-mutated NSCLC patients upon drug resistance.

Previous case reports have linked RET fusions, such as CCDC6-RET, TRIM24-RET, NCOA4-RET and ERC1-RET, to osimertinib resistance in EGFR-mutated NSCLC patients [35‐39]. Here, we demonstrated a higher incidence rate of RET fusions in patients who developed AR to 3rd-G EGFR-TKIs from a population perspective. Interestingly, the incidence of RET fusions was highest in patients who underwent 2L 3rd-G EGFR-TKI treatment (1.5%, 51/3330), followed by patients treated with 1L 3rd-G EGFR-TKIs (1.2%, 7/589, Fig. 4e). These numbers were about ten-fold higher than those in patients treated with first- or second-generation EGFR-TKIs at the front line (0.15%, 13/8732). Overall, we demonstrated that RET fusions were more likely associated with EGFR-mutant NSCLC patients who received therapeutic interventions targeting EGFR with third-generation EGFR-TKIs. The complexity of non-RET secondary mutations in acquired RET+ patients might contribute to the differential treatment responses to follow-up therapies.