Tumor Analyses Reveal Squamous Transformation and Off-Target Alterations As Early Resistance Mechanisms to First-line Osimertinib in EGFR-Mutant Lung Cancer

Purpose: Patterns of resistance to first-line osimertinib are not well-established and have primarily been evaluated using plasma assays, which cannot detect histologic transformation and have differential sensitivity for copy number changes and chromosomal rearrangements. Experimental Design: To characterize mechanisms of resistance to osimertinib, patients with metastatic EGFR-mutant lung cancers who received osimertinib at Memorial Sloan Kettering Cancer Center and had next-generation sequencing performed on tumor tissue before osimertinib initiation and after progression were identified. Results: Among 62 patients who met eligibility criteria, histologic transformation, primarily squamous transformation, was identified in 15% of first-line osimertinib cases and 14% of later-line cases. Nineteen percent (5/27) of patients treated with first-line osimertinib had off-target genetic resistance (2 MET amplification, 1 KRAS mutation, 1 RET fusion, and 1 BRAF fusion) whereas 4% (1/27) had an acquired EGFR mutation (EGFR G724S). Patients with squamous transformation exhibited considerable genomic complexity; acquired PIK3CA mutation, chromosome 3q amplification, and FGF amplification were all seen. Patients with transformation had shorter time on osimertinib and shorter survival compared with patients with on-target resistance. Initial EGFR sensitizing mutation, time on osimertinib treatment, and line of therapy also influenced resistance mechanism that emerged. The compound mutation EGFR S768 + V769L and the mutation MET H1094Y were identified and validated as resistance mechanisms with potential treatment options. Conclusions: Histologic transformation and other off-target molecular alterations are frequent early emerging resistance mechanisms to osimertinib and are associated with poor clinical outcomes. See related commentary by Piotrowska and Hata, p. 2441


Introduction
The identification of EGFR T790M as the dominant mechanism of resistance to firstand second-generation EGFR-tyrosine kinase inhibitors (TKI) resulted in the development of osimertinib, a thirdgeneration EGFR-TKI (1,2). Osimertinib's initial approval was in the setting of progression on initial EGFR-TKI in patients with tumors harboring EGFR T790M (3). More recently, osimertinib exhibited superior progression-free survival (PFS) compared with erlotinib or gefitinib as initial treatment in patients with EGFRmutant non-small cell lung cancer, positioning osimertinib as the preferred first-line treatment where available (4). Nevertheless, acquired resistance to osimertinib invariably develops with a median PFS of 19 months (4).
Analysis of circulating tumor DNA (ctDNA) has been the predominant method for investigating resistance, but cannot detect histologic transformation and has differential sensitivity for copy number changes and chromosomal rearrangements compared with tissue analysis (5,6,11,12,(21)(22)(23)(24)(25). Furthermore, published studies have lacked paired tumor samples pre-and post-osimertinib, which makes determination of acquired alterations and putative resistance mechanisms challenging. Therefore, we sought to use paired tumor tissue to detect molecular and histologic mechanisms of resistance to osimertinib and identify potential associations with clinical outcomes.

Materials and Methods
In accordance with the Belmont report and following the Institutional Review Board (IRB)/Privacy Board at Memorial Sloan Kettering Cancer Center (MSKCC, New York, NY) for retrospective review of records and waiver of consent, we retrospectively identified all patients with EGFR-mutant metastatic lung cancers who received osimertinib and had pretreatment and postprogression tumor samples (acquired resistance defined by Jackman criteria; ref. 26), where targeted hybrid capture, next-generation sequencing (NGS) of tumor DNA had been performed. The primary NGS platform was MSK-IMPACT (27), but other NGS platforms, such as MSK-Ampliseq (Supplementary Materials and Methods) and Foundation Medicine NGS (28), were occasionally utilized. For MSK-IMPACT, patients were consented to MSKCC IRB protocol 12-245. Patients were divided into two cohorts: (i) patients who received osimertinib without prior EGFR-TKI exposure ("first-line" osimertinib), and (ii) patients who received osimertinib after prior EGFR-TKIs ("later-line" osimertinib).
Patient records were reviewed to extract demographic information, clinical outcomes, and molecular and histologic data. Time-totreatment discontinuation (TTD) was defined as time from start of EGFR-TKI to last administered dose prior to a treatment change (29). Overall survival (OS) was defined as date of osimertinib initiation to date of death or last follow-up as of May 1, 2019. Fisher exact and log-rank tests were used to identify associations between clinical, molecular, and histologic features, and Kaplan-Meier methodology was used for TTD and OS. In all cases of transformation, the original pathology samples were re-reviewed to confirm the absence of preexisting neuroendocrine, squamous/ adenosquamous, or small cell histology using IHC performed for p40, TTF1, RB1, and p53.
MSK-Fusion Solid, a custom RNAseq panel, was used to detect fusions in cases where no resistance mechanism was identified by NGS and sufficient tissue was available (30). Single nucleotide variants and copy number variants identified by MSK IMPACT were analyzed on the cBioPortal (31). Specific mutations were assessed for enrichment with the McNemar test using paired samples and Fisher exact test in unpaired analyses. Additional methods are provided in the Supplementary Materials and Methods.

Clinical characteristics
Sixty-two patients were identified with acquired resistance to osimertinib and paired tumor tissue available for analyses.

Translational Relevance
Prior reports of mechanisms of resistance to osimertinib primarily focus on patients who received osimertinib after other EGFR-tyrosine kinase inhibitors and rely heavily upon circulating tumor DNA. We utilized paired tumor tissue with nextgeneration sequencing performed before osimertinib and after progression to analyze resistance mechanisms in a cohort of 62 patients with EGFR-mutant lung cancer treated with osimertinib, either as first-line or later-line treatment. We identified lineage plasticity and in particular, squamous histologic transformation, as unexpectedly frequent with first-line osimertinib and associated with considerable genomic complexity, highlighting the importance of tissue-based analyses to evaluated acquired resistance. We also detected a diverse array of off-target genomic resistance mechanisms that may be amenable to targeted therapy. Sensitizing EGFR mutation, time on osimertinib therapy, and line of therapy all may influence the resistance spectra identified. Finally, we validate two novel resistance alterations, EGFR S768 þ V769L and MET H1094Y and explore relevant potential treatment options. Concurrent genomic alterations seen with first-line osimertinib The molecular landscape of concurrent alterations identified before and after treatment with first-line osimertinib is depicted in Fig. 1A. The most frequent cooccurring pretreatment mutation was TP53 (70%, n ¼ 19). Common co-occurring amplifications/ deletions were EGFR amplifications (33%, n ¼ 9) and CDKN2A/B deletions (15%, n ¼ 4 each). There was no significant enrichment of specific alterations in either the pretreatment or posttreatment setting (Fig. 1B). Known mechanisms of acquired resistance to EGFR-TKIs were identified in 41% (n ¼ 11) of cases (Fig. 1C). In the first-line setting, EGFR G724S was the only on-target acquired mutation identified. Off-target resistance mechanisms included MET amplifications (Supplementary Table S1), MET H1094Y mutation, KRAS G12A mutation, TRIM24-BRAF fusion, and RUFY2-RET fusion.
Comparisons between first-line and later-line osimertinib Overall, EGFR-mediated acquired resistance was associated with a longer time on osimertinib and improved OS compared with other resistance mechanisms (median TTD 18.0 months; 95% confidence interval, 13.3-33.2; vs. 13.2 months, 95% confidence interval, 10.6-16.3; P ¼ 0.04; median OS not reached vs. 29 months 95% confidence interval, 24.6-not reached; P < 0.001). Notably, the proportion of offtarget resistance in the first-line setting was higher than the later-line cohort (P ¼ 0.01) suggesting that off-target resistance emerges earlier and/or treatment with first-line osimertinib may enrich for off-target resistance.

Associations with sensitizing EGFR mutation
Acquired alterations were analyzed by initial sensitizing EGFR mutation (exon 19 deletion vs. L858R mutation; Fig. 3A and B). EGFR C797S was more frequently seen with EGFR exon 19 deletion ( Fig. 3A and B; P ¼ 0.03). Posttreatment CDKN2A/B deletions and TERT amplifications were more commonly seen with EGFR L858R mutations (P ¼ 0.02; P ¼ 0.03, respectively). However, these associations were not significant when accounting for multiple comparisons.

Histologic transformation
Histologic transformation was identified in 15% of cases (9 patients), 15% in the first-line setting and 14% in the later-line setting (Fig. 4A). We identified five cases of squamous cell transformation (Fig. 4B) Unknown  transformation. One patient acquired a PIK3CA E726K mutation; no preexisting or acquired SOX2 amplifications were identified, but 1 patient had low-level chromosome 3q gain and 1 had low level PIK3CA copy number gain in both the pretreatment and posttreatment samples (Fig. 4A). The patient with pleomorphic transformation acquired a very high level chromosome 3q amplification (49-fold) and FGF3/ FGF4/FGF19 amplification (15-fold; Fig. 4A). Consistent with prior literature, the small-cell lung cancer tumors (SCLC) all had preexisting alterations of RB1 and TP53 identified in their pretreatment samples (Fig. 4A).
Clinical course after treatment with osimertinib Patient outcomes post-osimertinib are summarized by resistance mechanism and line of therapy (Fig. 5). Five of the 9 patients with histologic transformation have died (4 within 10 months postosimertinib progression). The patients with small-cell transformation all received platinum with etoposide as a part of their subsequent therapy (Fig. 4C). Among the 5 patients with squamous transformation, varying treatment strategies were employed and outcomes were mixed with limited follow-up (Fig. 4C). In several instances, patients with acquired off-target alterations were treated with targeted therapies. For instance, 2 patients with tumors that acquired ALK fusions were treated with osimertinib and ALK TKIs (crizotinib, alectinib, and lorlatinib) with durable responses (Fig. 5C and D). The clonal evolution of identified alterations is illustrated over interval biopsy samples for 2 patients (Fig. 5C  and D).
Functional studies of MET H1094 and EGFR SV768IL, as novel resistance mechanisms to osimertinib We hypothesized that acquired alterations in MET H1094, identified in the first-line setting and EGFR S768I þ V769L (EGFR SV768IL) mutations found in the later-line setting could represent putative resistance mechanisms and explored these alterations in preclinical models.

MET H1094 mutations confer resistance to osimertinib
We expressed wild-type MET, MET H1094R, and MET H1094Y, and a kinase dead MET H1094Y (K1110A mutant) in the PC9, EGFR exon 19 deleted lung adenocarcinoma cell line. Mutations were introduced into MET using site-directed mutagenesis and mutations confirmed by DNA sequencing (Fig. 6A). Western blotting of wholecell extracts confirmed overexpression of MET and the corresponding mutants (Fig. 6B). The two MET mutants demonstrated similar levels of MET phosphorylation versus wild-type MET. The kinase dead MET H1094Y mutant showed a low level of MET phosphorylation. Sensitivity to growth inhibition by osimertinib was reduced in PC9 cells expressing either MET H1094 mutant compared with PC9 cells expressing an empty vector (EV) plasmid (PC9-EV) or PC9 cells overexpressing wild-type MET (PC9-MET; ref. Fig. 6C). The MET H1094R mutant induced greater resistance to osimertinib compared with the MET H1094Y mutant (959-vs. 136-fold increase in IC 50 value for growth inhibition, compared with PC9-EV cells). PC9 cells expressing the two MET H1094 mutants were slightly more sensitive to crizotinib than the PC9-EV or PC9-MET cells (Fig. 6D).

Combined inhibition of MET and EGFR overcomes MET H1094mediated osimertinib resistance
To determine whether combination crizotinib and osimertinib would be an effective therapeutic strategy for EGFR-mutant cancers expressing MET-H1094, we examined whether these two drugs act synergistically to inhibit growth. We used the method of Chou-Talalay (32), where cells are treated with drug combinations and then a parameter called the combination index (CI) is determined. CI < 1 indicates synergy, CI > 1 indicates antagonism between the two drugs, and CI ¼ 1 indicates an additive effect. Growth was inhibited in PC9-MET-H1094Y cells when the two drugs were combined, compared with either drug alone ( Fig. 6E and F). The CI values were <1 for all drug concentrations tested (Fig. 6G). We next examined the effect of the MET H1094Y mutation on osimertinib-induced caspase 3/7 activity as a measure of apoptosis. Treatment of PC9-EV cells with   CLINICAL CANCER RESEARCH osimertinib induced activation of caspase 3/7 as expected but did not stimulate caspase 3/7 activity in PC9-MET or PC9-MET-H1049Y cells (Fig. 6H). In contrast, treatment of the PC9 cell lines with crizotinib alone did not stimulate caspase 3/7 activity in any of the cell lines (Fig. 6H). However, a combination of crizotinib and osimertinib resulted in a significant increase in caspase 3/7 activity in PC9-MET and PC9-MET-H1094Y (Fig. 6G). Taken together, these results endorse MET H1094Y to be a novel mechanism of resistance to osimertinib that can be overcome by combined inhibition of MET and EGFR.
Compound EGFR S768IþV769L (EGFR SV768IL) mutations confer resistance to osimertinib To determine whether the observed mutations in EGFR exon 20 confer resistance to osimertinib, we generated cell lines with the S768I and V769L mutations (EGFR-SV768IL) by site-directed mutagenesis ( Supplementary Fig. S1A). Wild-type EGFR or EGFR-SV768IL was expressed in PC9 and HCC827 cell lines and expression confirmed by Western blotting (Supplementary Fig. S1B). EGFR-SV768IL was more heavily phosphorylated than wild-type EGFR in both cell lines, indicating higher level of activity ( Supplementary Fig. S1B) of the SV768IL variant. Similarly, expression of EGFR-SV768IL resulted in phosphorylation of ERK1/2 and AKT to a higher extent than that observed with expression of wild-type EGFR. Growth of PC9 (Supplementary Fig. S1C) and HCC827 ( Supplementary Fig. S1C) expressing EGFR-SV768IL was resistant to the inhibitory effects of osimertinib, compared with cells expressing either an empty vector (EV) or wild-type EGFR. The IC 50 value of osimertinib for PC9-EGFR-SV768IL cells was 166-fold higher than that for PC9-EV cells (Supplementary Fig. S1C). Similarly, the IC 50 value of osimertinib in HCC827-EGFR SV768IL cells was 244-fold higher than that of HCC827-EV cells (Supplementary Fig. S1C). We also examined the effect of EGFR-SV768IL on osimertinib-induced caspase 3/7 activity in PC9 cells. Whereas osimertinib caused a significant increase in caspase 3/7 activity in PC9-EV and PC9-EGFR cells, no increase was observed in PC9-EGFR-SV768IL cells (Supplementary Fig. S1D).

Discussion
We identified an array of acquired resistance mechanisms to osimertinib using paired pre-and posttreatment tissue samples. In our cohort of first-line patients with limited follow-up and shorter time

Resistance to Osimertinib by Paired Tumor Analyses
AACRJournals.org Clin Cancer Res; 26(11) June 1, 2020 on osimertinib, EGFR-mediated resistance was uncommon, whereas off-target resistance, including histologic transformation, was seen frequently. Notably, there appears to be a time-dependent pattern of resistance with off-target resistance emerging earlier resulting in less durable responses to osimertinib. This mirrors our earlier finding that on-target resistance mutations (i.e., EGFR T790M) are associated with more indolent disease and arise as after a longer time on EGFR-TKI and with better postprogression survival (33). Development of offtarget resistance after a short period may result from preexisting subclones that emerge quickly on treatment. To ascertain whether resistance mechanisms to firstand later-line osimertinib truly differ will require lengthier follow-up in patients on first-line osimertinib. EGFR C797S, the most common EGFR mutation acquired on laterline osimertinib, was not identified in our first-line cohort of patients. The frequency of EGFR C797S in the first-line FLAURA study was 8% (15); both first-line cohorts (ours and FLAURA) report lower frequencies of EGFR C797S compared with later-line osimertinib cohorts (15%-32%; Supplementary Fig. S2; refs. 6, 8, 10, 34, 35) again suggesting first-line and later-line osimertinib may have different resistance spectra. Later-line osimertinib is utilized only in tumors with acquired EGFR T790M; these tumors have demonstrated continued dependence on EGFR signaling and may be predisposed to acquire tertiary EGFR mutations (i.e., EGFR C797S) compared with EGFR-mutant tumors at large resulting in the disparate frequencies of EGFR-acquired mutations in the first-line and later-line setting. MET amplification was also identified at a lower rate (7%) than most reports in the later-line setting (10%-26%; Supplementary  Fig. S2; refs. 6,8,10,34,35) and is on the lower-end of first-line reports (5%-15%; Supplementary Fig. S2; refs. 15,36). Previous studies lacking pretreatment tissue or plasma may overestimate acquired MET amplifications, which can be seen concurrently with EGFR prior to treatment (6,35,37,38). In addition, plasma-based platforms typically have lower sensitivity to assess copy number changes (21)(22)(23)(24)(25).
Tissue analysis is critical to characterizing resistance mechanisms. Histologic transformation, which cannot be detected via plasma testing, was a frequent mechanism of resistance in our study. Rates of transformation and other off-target resistance mechanisms may be higher with osimertinib compared with earlier generation TKIs due to better on-target inhibition. Prior to this report, squamous cell transformation was identified infrequently (5,8,(16)(17)(18)(19)(20). This phenomenon is surely under recognized because of the increasing reliance on ctDNA for identification of resistance mechanisms. Recognition of histologic transformation is imperative as it has prognostic and therapeutic implications. Patients with squamous cell transformation in our cohort had short postprogression survival (Figs. 4 and 5). Similar to small-cell transformation (1,2,39)  TP53 mutations identified in small-cell lung cancers, exhibit considerable genomic complexity. Understanding the etiology of squamous cell transformation will require comprehensive investigation, made more challenging by the fact that de novo squamous cell lung cancers do not have an overarching genomic signature. Further study will include understanding the gene expression subtype of the transformed cases and assessing nongenomic processes that may play a role in histologic transdifferentiation such as transcription factor networks and the epigenome. Recent data suggest that the initial sensitizing EGFR mutation may bias the resistance mechanisms that emerge. To date, EGFR G724S has only been identified with EGFR exon 19 deletions, and structural and in vitro models support EGFR G724S as conferring resistance only when concurrent with an EGFR exon 19 deletion (40). We similarly demonstrate that EGFR C797S is preferentially coupled with EGFR exon 19 deletions. We also confirm the previous findings that EGFR C797S was only seen in tumors that retained EGFR T790M suggesting continued EGFR dependence in these tumors. Typically, off-target or unknown resistance mechanisms are seen in the absence of EGFR T790M suggesting a loss of EGFR dependence in these tumors.
We and others have described acquired chromosomal rearrangements (ALK, RET, BRAF, ERBB2, and MET exon 14) as resistance mechanisms to EGFR-TKIs (5-7, 10-15, 41-43). We again identified BRAF, ALK, and RET fusions in this series. The relatively high frequency of these otherwise extremely rare oncogene fusions in the setting of acquired resistance to osimertinib requires further exploration. Eight percent of all RET fusions and 50% of all BRAF fusions identified in lung cancers at MSKCC by MSK-IMPACT over the time period of this study were found in patients with EGFR-mutant lung cancer and acquired resistance to osimertinib (31,44). This high frequency of acquired fusions supports a predisposition for genomic rearrangements driven by the selective pressure of osimertinib.
We also identified and validated the mutation MET H1094Y and the compound mutation EGFR S768 þ V769L as resistance mechanisms with potential associated treatments. Prior work demonstrated increased catalytic activity and cognate autophosphorylation of MET H1094Y as compared with the wild-type MET kinase domain, confirming MET H1094Y to be an oncogenic and transformative mutation. From a structural perspective, this mutation resides in close proximity to the ATP-binding site, and based on molecular modeling studies, these mutations may activate MET kinase by destabilizing the inhibitory conformation of the activation loop (45,46). Prior case reports of de novo EGFR S768I and V769L compound mutations have been published with mixed responses to firstand secondgeneration EGFR TKIs (47)(48)(49). It is not clear whether treating a patient who acquires on-target resistance to osimertinib will respond to early generation TKIs and the appropriate trials are underway (NCT03755102; ref. 9).
Over half of our first-line cohort had unknown mechanisms of resistance. In these cases, resistance may be due to epigenetic modifications, changes in protein expression, or novel genomic alterations. Further analyses will need to integrate epigenetic, RNA, and protein expression analyses to uncover the yet undetermined mechanisms of resistance to osimertinib. In addition, clonal evolution and tumor heterogeneity also play a fundamental role in resistance to targeted therapies and should be considered in future analyses. As osimertinib has only recently been integrated as firstline treatment, our first-line cohort was biased toward resistance mechanisms that emerge earlier on treatment and makes directed comparison with the later-line cohort challenging, but provides the unique perspective of identifying early emerging mechanisms of resistance. Another limitation of our study is that histologic transformation could also represent outgrowth of a preexisting clone of tumor cells that were not previously identified. However, multiple sections throughout each sample of pathologic tissue were reexamined to confirm no evidence of preexisting squamous cell or small cell histology. This will be an overarching limitation for all future studies of lineage plasticity in this patient subset since most metastatic patients only have small core-needle biopsies done. Also, molecular data from single-lesion biopsies may not reflect the entirety of genetic alterations due to tumoral heterogeneity. Finally, although our sample size was modest, this is the largest analysis to date of osimertinib resistance utilizing paired tumor tissue.
In conclusion, our study establishes that mechanisms of resistance to osimertinib are diverse, with sensitizing EGFR mutation, time on osimertinib therapy, and line of therapy all influencing the resistance spectra identified. Off-target resistance arises early on first-line osimertinib after a shorter duration of osimertinib treatment. Histologic transformation appears common with first-line osimertinib and highlights the continued importance of tissue-based assays to evaluate acquired resistance. With resistance mechanisms dependent on original sensitizing EGFR mutation, further assessment of how pretreatment alterations forecast resistance will be important as the field amends first-line treatments to delay or prevent resistance. Identifying and overcoming these resistance mechanisms will require a multifaceted approach utilizing both plasma and tissue molecular and histopathologic analyses.