MIG6 Mediates Adaptive and Acquired Resistance to ALK/ROS1 Fusion Kinase Inhibition through EGFR Bypass Signaling Open Access

Drug resistance is a major challenge for tyrosine kinase inhibitors (TKI) targeting oncogenic ALK and ROS1 fusions, which occur in 3% to 9% of patients with non–small cell lung cancer (NSCLC; refs. 1, 2). Crizotinib is the first FDA-approved ALK/ROS1 inhibitor with a superior benefit compared with chemotherapy (3, 4). Followed by crizotinib, next-generation ALK inhibitors (ceritinib, alectinib, brigatinib, and lorlatinib), and ROS1 inhibitor entrectinib were approved by FDA to improve selectivity, potency, and brain penetration (5–9). Despite this progress in drug development, complete responses are rare, and therapeutic resistance ultimately develops. Several mechanisms that contribute to ALK/ROS1 TKI resistance have been identified including ALK/ROS1 kinase mutations hindering inhibitor binding and ALK/ROS1-independent alternative pathway activation, which can be addressed by ALK/ROS1 TKIs targeting resistance mutations and drug combinations cotargeting bypass signaling, respectively (10–15). However, current therapeutic strategies overcoming drug resistance are often applied sequentially after disease progression, and subsequent resistance is inevitable.

An alternate approach to addressing resistance with sequential therapy is to target residual disease preemptively (16, 17). In this paradigm, despite an initial tumor remission under targeted therapy, there are cancer cells that quickly adapt to primary oncogene inhibition and establish residual disease, which can fuel eventual disease progression. Understanding the mechanisms that contribute to this adaptive resistance is critical to help develop strategies that target residual disease and potentially prevent or delay resistance emergence. Unfortunately, resistance mechanisms previously reported were mostly identified in acquired resistance models after progression on long-term ALK/ROS1 inhibition (10–15). Few studies have addressed adaptive resistance mechanisms immediately following exposure to oncogene inhibition. Because ALK/ROS1 kinase domain mutations (KDM) only occur in less than half of patients after progressing to TKIs, there is also a strong need to identify biomarkers to guide alternative effective treatments for resistant patients who do not develop KDMs (13, 14). Another hindrance to studying the resistance mechanism to ALK/ROS1 inhibitors is a relative lack of cell line models harboring those fusion kinase oncogenes. In the past decade, our lab has generated numerous patient-derived ALK and ROS1 cell lines, which represent a unique and valuable resource for studying the resistance mechanisms to ALK/ROS1 inhibitors (13).

Cellular adaption through the signaling rewiring under primary oncogene inhibition is considered as a major mechanism contributing to residual disease (16). To interrogate the rapid signaling adaption following ALK/ROS1 inhibition, we used an unbiased phosphoproteomics approach. This approach has recently been successful in identifying adaptive resistance mechanisms to KRAS inhibitors (18). Herein, we found the phosphorylation and mRNA of MIG6, a negative regulator of EGFR, were rapidly downregulated following ALK and ROS1 inhibition, potentiating EGFR activity to support tumor cell survival under immediate primary oncogene inhibition. This adaptive resistance can also transform into acquired resistance where MIG6 remains depleted. These data provide a strong rationale for combining EGFR and ALK/ROS1 therapies as the first-line therapy to improve patient outcomes. For patients who progress after ALK/ROS1 TKIs, this study also identified several MIG6-related biomarkers to potentially guide therapeutic intervention with EGFR inhibitors.

Sample preparation for p-Tyr–enriched phosphoproteomics, global (pSTY) phosphoproteomics, and following peptide identification, normalization, and quantification were performed as described previously (18). Log₂ ratio (fold change) and t test were calculated between DMSO and crizotinib-treated biological triplicates. Phosphosites were determined differentially expressed if the following criteria were met: not entirely reverse or nonhuman contaminants, |log₂ ratio| ≥ ∼0.585 (1.5-fold), and P < 0.05.

Proteins with phosphorylation sites significantly regulated by crizotinib in either p-Tyr enriched phosphoproteomics or pSTY phosphoproteomics were combined for Gene Ontology (GO) enrichment analysis using the R package clusterProfiler v3.18.0 (19). For EGFR interactome analysis, 27 significantly crizotinib-regulated phosphoproteins from the GO term “regulation of ERBB signaling pathway” enriched in CUTO28 p-Tyr phosphoproteomics data were analyzed by STRING (https://string-db.org/, v11.5) to plot the protein-protein interaction (PPI) network. The EGFR interactome was illustrated in Cytoscape v3.7.2 with the edge thickness mapped to the protein–protein interaction confidence score.

H3122 was obtained from Dr. Paul A. Bunn at the University of Colorado Anschutz Medical Campus (Aurora, CO). All the CUTO cell lines (the “CUTO” moniker refers to the “University of Colorado Thoracic Oncology”) were derived from pleural effusion or tumor biopsy from ALK or ROS1 fusion-positive patients following informed, written consent from the patient under an Institutional review board–reviewed protocol (Supplementary Table S1). The cell line derivation process was performed as previously described (13). The fusions in all derived cells were verified by Archer Fusionplex in the Colorado Molecular Correlates Laboratory (CMOCO). The mutations of all patient-derived cells were profiled by the Variantplex assay (a targeted NGS panel covering 53 genes) while CUTO63 line was additionally profiled by the GOAL assay (an expanded NGS panel covering 498 genes including ERRFI1) in CMOCO. Although cells were often derived from patients who were being treated with ALK/ROS1 TKIs, cell line propagation was performed in the absence of TKIs. All the cell lines were authenticated by short tandem repeat profiling at the Cell Technologies Shared Resources and the Barbara Davis Center of the University of Colorado Anschutz Medical Campus, and were routinely tested for Mycoplasma. H3122 and all patient-derived cells were cultured in RPMI1640 supplemented with 10% FBS. Crizotinib, alectinib, entrectinib, afatinib, lorlatinib, and trametinib were purchased from Selleck Chemicals. EGF was purchased from R&D Systems.

For cell viability assay, cells were seeded at 1,000 or 2,000 cells per well in 96-well plates 24 hours before treatments. After 72 hours of drug treatments, cell viability was assayed by MTS (CellTiter 96 Aqueous One Solution Cell Proliferation Assay, Promega). Dose-response curves were generated in GraphPad Prism software. For the clonogenic assay, cells were seeded 2,000 cells per well in 12-well plates and treated with drugs for 2 weeks with media changed every 3 or 4 days. Colonies were stained with crystal violet and the total colony area per well was quantified by MetaMorph software.

Mission nontarget shRNA control (#SHC002) and two MIG6 shRNAs, shMIG6#1 (TRC Clone ID TRCN0000291921) and shMIG6#2 (TRC Clone ID TRCN0000118128) were ordered from the Functional Genomics Facility at the University of Colorado Anschutz Medical Campus (Aurora, CO). WT MIG6 lentiviral plasmid (#RC206883L3) and the empty vector (#PS100092) were purchased from Origene. To generate ERRFI1 (MIG6) mutant lentiviral plasmid, a short sequence containing Y394FY395F mutations from a MIG6 Y394FY395F pBabe plasmid (a gift from Dr. Jeonghee Cho) was cut to substitute the corresponding sequence in the WT MIG6 lentiviral plasmid. Doxycycline-inducible WT and Y394FY395F MIG6 lentiviral plasmids were generated by the Functional Genomics Facility at the University of Colorado Anschutz Medical Campus. Lentivirus was produced by transfecting shRNA or protein-expressing plasmids along with pCMV-VSV-G and pCMVΔR8.2 into 293T cells using Mirus TransIT-293 reagent. Viral supernatants were collected 48 hours after transfection and applied to target cells for 24 hours. Transduced cells were selected by puromycin for one week (5 μg/mL for H3122 and H3122-CR cells, 2 μg/mL for CUTO28 cells, 1 μg/mL for CUTO37 and CUTO37-ER cells, and 0.5 μg/mL for CUTO63 cells). H3122-CR and CUTO37-ER cells were maintained free of TKI after transduction to avoid selection of clones expressing high levels of MIG6.

Cells were transfected for 48 hours with 25 nmol/L nontargeting control (#D-001810–10), or ERRFI1 siRNA (#L-017016–01) purchased from Horizon Discovery using DharmaFECT 1 transfection reagent (#T-2001).

Cell lysis and immunoblotting were performed as previously described (20). Antibodies used in this study were listed in Supplementary Table S2.

Cells were lysed in cold coimmunoprecipitation (co-IP) buffer (50 mmol/L Tris HCl pH 7.5, 100 mmol/L NaCl, 5 mmol/L EDTA, 1% Triton X-100) and 800 μg protein was incubated with 10 μg EGFR rabbit antibody (Cell Signaling Technology, #4267) in 500 mL lysis buffer at 4°C overnight. The protein–antibody complex was captured by PureProteome Protein A/G Mix Magnetic Beads (Millipore) for 1 hour at 4°C and washed with cold co-IP buffer. Buffer for lysing, incubation, and washing were all supplemented with Halt Protease and Phosphatase Inhibitor Cocktail (Thermo Fisher Scientific). Proteins were eluted from beads by boiling in Protein Sample Loading Buffer (LI-COR) supplemented with 100 mmol/L DTT for 5 minutes.

RNA was extracted using RNeasy Plus Kits (Qiagen) and reverse-transcribed to cDNA by SuperScript IV Reverse Transcriptase (Invitrogen). qPCR was performed on an Applied Biosystems QuantStudio3 using TaqMan Gene Expression Assays #4331182 and #4448490, respectively, to quantify ERRFI1 and 18S rRNA expression. ERRFI1 transcriptional changes under TKI treatments relative to control were normalized by 18S rRNA expression and analyzed using ΔΔC_t method.

RNA for H3122, CUTO37, and their resistant counterparts was extracted in biological triplicates using RNeasy Plus Kit (Qiagen) and submitted to Novogene for sequencing and analysis. Briefly, the Ultra II RNA Library Prep Kit for Illumina (NEB) was used for library construction. Sequencing was performed on NovaSeq 6000 (Illumina) and low-quality reads were trimmed. Trimmed reads were mapped to the reference genome hg19 using STAR (v2.6.1d) and quantified by FeatureCounts (v1.5.0-p3). RNA-seq raw counts were normalized and the following differential expression analyses were performed using DESeq2 R package v1.26.0 (21).

DNA for H3122, CUTO37, and their resistant counterparts was extracted using QuickDNA Plus Kit (Zymo), submitted to the Genomics and Microarray Core at the University of Colorado Anschutz Medical Campus for sequencing and analyzed by the Biostatistics and Bioinformatics Shared Resource. Briefly, the SureSelect Human All Exon V8 exome probes (Agilent) and SureSelectXT Reagent Kits (Agilent) were used to generate WES libraries. Paired-end sequencing reads of 150 bp were generated on NovaSeq 6000 (Illumina) at a target depth of 100X coverage per sample. Variants and associated biological significance were analyzed with the nf-core/sarek pipeline (v2.71), using default annotations for the human genome assembly GRCh38 and Mutect2, Strelka, and FreeBayes tools (22).

RNA-sequencing (RNA-seq) data from two independent patient cohorts were used to determine EGFR to MIG6 ratios. The first patient cohort contains 5 ALK inhibitor (ALKi) naïve and 7 ALKi-resistant tumors (23). The processed RNA-seq data with gene expression normalized to RPKM values was obtained from Dr. Frederick H. Wilson. The second patient cohort contains 28 pre- and 14 post-ALKi–treated tumors. The processed RNA-seq gene data with gene expression normalized to TPM values was obtained from Caris Life Sciences. Patients in the post-ALKi group received at least two cycles of treatment within 6 months while the disease progression status upon the sample collection was unknown. All the tumor samples used for EGFR to MIG6 ratio analysis do not contain ALK kinase domain mutations.

MSK-IMPACT (24), MSK-MetTropism (25), and AACR GENIE v13.0-public (26) datasets were accessed through https://www.cbioportal.org/ and https://genie.cbioportal.org/ websites. ERRFI1 alteration status and structural variant profiling status from all patients in AACR GENIE v13.0-public cohort were downloaded and analyzed in R to determine (i) the frequency of patients with ERRFI1 deep-deletion and EBD truncation mutations (truncation mutations occurring after EBD were mutually examined and excluded) and (ii) frequency of structural variant profiling in the patients with ERRFI1 deletion and EBD truncation.

Statistical analysis for phosphoproteomics, RNA-seq, and whole-exome sequencing (WES) are described under corresponding method sections. Two-tailed unpaired Student or Welch t tests were performed for two-group comparisons and ANOVA with Tukey post hoc test was performed for multiple-group comparisons in GraphPad Prism 8. A P value less than 0.05 was considered statistically significant.

The authors declare that all data generated in this study are available within the article and the Supplementary Data. pY and pSTY phosphoproteomics data for H3122 and CUTO28 cells were deposited in ProteomeXchange (http://www.proteomexchange.org/) with the accession numbers PXD036771 and PXD036772. RNA-seq data for H3122 and H3122-CR cells (GSE239666), CUTO37 and CUTO37-ER cells (GSE214715), and a panel of 7 ROS1 cells (GSE239844) were deposited in GEO. WES data for H3122 and CUTO37 and their resistant counterparts (PRJNA1002775) was deposited in SRA.

To explore the early adaptive signaling reprogramming after ALK/ROS1 oncogene inhibition, we conducted quantitative phosphoproteomics on H3122 (EML4-ALK) and CUTO28 (TPM3-ROS1) cells following a 2-hour crizotinib treatment (Fig. 1A). We then integrated all significantly regulated phosphoproteins (|fold change| > 1.5, P < 0.05) identified from global (pSTY) and pY-enriched phosphoproteomics for downstream pathway analyses. We found 1252 S/T/Y phosphorylation sites and 798 phosphoproteins were significantly regulated under crizotinib treatment in H3122 cells, while 2,201 S/T/Y phosphorylation sites and 1,095 phosphoproteins were significantly regulated in CUTO28 cells (Supplementary Tables S3–S6). GO enrichment analysis for those regulated phosphoproteins reveals that the ERBB signaling pathway was highly enriched following crizotinib treatments (Fig. 1B). To interrogate the mechanism contributing to the ERBB signaling enrichment, we focused on the GO term “regulation of ERBB signaling pathway”, which identified 27 proteins with significantly altered phosphorylation after crizotinib treatment. To pinpoint the potential regulators for EGFR, we constructed a protein–protein interaction network for those 27 candidates and focused on the ones with predicted strong interactions with EGFR using the STRING database (Fig. 1C; Supplementary Table S7). In addition, we examined the phosphorylation change on each protein residue in relation to its known functional role in regulating EGFR, identifying ERRFI1 (ERBB Receptor Feedback Inhibitor 1) as the most promising regulator for EGFR in response to ALK/ROS1 inhibition. ERRFI1 encodes MIG6, a known EGFR protein inhibitor and tumor suppressor (27–31). Phosphorylation of Y394 and Y395 has been reported to be critical for MIG6 binding and inhibition on EGFR (30, 31). Notably, MIG6 Y394 phosphorylation in H3122 cells and both Y394 and Y395 phosphorylation in CUTO28 cells were decreased by crizotinib (Fig. 1D and E). The decreased phosphorylation was not caused by the changes in MIG6 protein abundance, which was sustained during the 2-hour crizotinib treatment (Fig. 1F and G). Overall, we showed a significant ERBB signaling enrichment associated with a rapid downregulation of MIG6 Y394/Y395 phosphorylation in response to short-term ALK/ROS1 inhibition.

Despite no changes in MIG6 protein abundance observed following the 2-hour crizotinib treatment, we evaluated MIG6 protein levels in an extended ALK/ROS1 inhibition for 48 hours. Strikingly, the MIG6 protein decrease started after 6-hour treatments of various ALK/ROS1 inhibitors in H3122 and CUTO28 cells and correlated with a moderate rebound in ERK phosphorylation (Fig. 2A and B). Given the enriched ERBB signaling pathway and downregulation of MIG6 phosphorylation and protein level following ALK/ROS1 inhibition, we asked whether upfront inhibition of EGFR signaling could enhance cellular response to ALK/ROS1 inhibitors. To potently inhibit wild-type EGFR, we selected an irreversible pan-ERBB family inhibitor, afatinib (32). Indeed, afatinib combination led to a more profound inhibition of cell viability and downstream ERK signaling in H3122 and CUTO28 cells compared with ALK/ROS1 inhibition alone (Fig. 2C–F). Furthermore, this combination eliminated residual colony formation of H3122 and CUTO28 cells in response to ALK/ROS1 TKI monotherapies (Fig. 2G and H).

We also examined MIG6 protein level under ALK/ROS1 inhibition across additional patient-derived ALK and ROS1 lines generated in our lab. CUTO29.1 and CUTO29.2 are EML4-ALK–driven cells derived from the same patient at different points in a treatment course. CUTO29.1 was derived during brigatinib treatment; the patient had previously progressed on crizotinib and alectinib. CUTO29.2 was derived later after this patient switched to lorlatinib treatment. CUTO29.1 harbors a crizotinib-resistant mutation ALK p.C1156Y, but is sensitive to lorlatinib (Supplementary Fig. S1A). CUTO29.2 harbors ALK p. C1156Y, I1171T, and L1198F mutations, which induces lorlatinib resistance, but resensitizes cells to crizotinib (Supplementary Fig. S1B; ref. 33). Interestingly, MIG6 protein was decreased by lorlatinib in lorlatinib-sensitive CUTO29.1 cells but not in lorlatinib-resistant CUTO29.2 cells (Supplementary Fig. S1C), suggesting the reduction of MIG6 protein level is a direct result of effective ALK inhibition. Indeed, MIG6 could be decreased by crizotinib in CUTO29.2 cells, in line with the mutation-mediated resensitization to crizotinib (Supplementary Fig. S1D).

The CUTO52 cell line harbors EML4-ALK and the ALK p.L1196M and D1203N resistance mutations, derived from a patient following treatments with crizotinib, alectinib, and brigatinib. CUTO52 is resistant to crizotinib and the MIG6 protein level was not reduced by crizotinib (Supplementary Fig. S1E and S1F). CUTO23 harbors a CD74-ROS1 fusion and is sensitive to crizotinib and demonstrated a reduction of MIG6 following crizotinib treatment (Supplementary Fig. S1G and S1H). Collectively, these data suggest the MIG6 protein reduction correlates with the cellular sensitivity to ALK/ROS1 inhibitors across multiple models, implying a role of MIG6 in mediating adaptive resistance. Consistently, the afatinib combination further suppresses cell viability when its partner fusion kinase inhibitor effectively inhibits ALK/ROS1 fusions (Fig. 2C and D; Supplementary Fig. S1A, S1B, and S1G; Supplementary Table S8).

Because MIG6 downregulation correlates with cellular sensitivity to ALK and ROS1 inhibition, we speculated that MIG6 expression is regulated by the common downstream signaling driven by the fusion kinases. MAPK pathway is a critical downstream pathway shared by ALK and ROS1 kinases and has also been reported to regulate MIG6 mRNA expression (28). We found ALK/ROS1 inhibition reduced MIG6 transcripts (Supplementary Fig. S2A and S2B). Treatment with trametinib, a MEK1/2 inhibitor, also suppressed ERRFI1 mRNA expression in ALK and ROS1 cells (Supplementary Fig. S2C and S2D) and recapitulated the MIG6 protein decrease observed under ALK/ROS1 inhibition (Supplementary Fig. S2E). These results suggest MIG6 is transcriptionally regulated by ALK and ROS1 fusions through the MAPK pathway.

To functionally validate the role of MIG6 in modulating responses to ALK/ROS1 inhibitors, we depleted MIG6 by shRNA knockdown in H3122 and CUTO28 cells. We found MIG6 knockdown induced resistance to various ALK/ROS1 inhibitors, whereas the addition of the pan-ERBB inhibitor afatinib restored ALK/ROS1 inhibition sensitivity in the absence of MIG6 expression (Fig. 3A and B). Of note, H3122 and CUTO28 remained insensitive to afatinib single-agent treatment under MIG6 knockdown, suggesting that ALK/ROS1 inhibition is necessary to trigger the survival signaling rewiring to EGFR (Fig. 3C and D). Consistent with the responses in cell viability, ERK signaling was also rescued by MIG6 knockdown under ALK/ROS1 inhibition while this rescue could be blocked by the addition of afatinib (Fig. 3E and F). To interrogate whether MIG6 knockdown mediated resistance through EGFR specifically or through all ERBB family members, we tested an EGFR-selective inhibitor, gefitinib. Similar to afatinib, we found gefitinib combination was able to completely abrogate ALK/ROS1 TKI resistance induced by MIG6 knockdown (Supplementary Fig. S3A–S3F). These data suggest MIG6 modulates responses to ALK/ROS1 inhibition primarily through EGFR in H3122 and CUTO28 cells. Previous studies have shown SHC1, a critical signaling adaptor shared by EGFR, ALK, and presumably, ROS1, competes with MIG6 for the same EGFR substrate–binding cleft and is phosphorylated at Y239 by EGFR after binding (30, 34–36). We therefore hypothesized that MIG6 depletion increases SHC1 binding to EGFR to maintain cell survival following ALK/ROS1 fusion kinase inhibition. Indeed, the SHC1 Y239/Y240 phosphorylation was reduced by crizotinib and this reduction could be partially rescued by MIG6 knockdown. However, when cotreated with afatinib or gefitinib, SHC1 Y239/Y240 phosphorylation rescue by MIG6 knockdown was blocked (Fig. 3E and F; Supplementary Fig. S3E and S3F). We observed a similar SHC1 and ERK phosphorylation rescue on cells that were transiently depleted of MIG6 by siRNA knockdown and treated with crizotinib (Supplementary Fig. S4). Finally, co-immunoprecipitation of EGFR showed MIG6 knockdown increases SHC1 binding to EGFR (Fig. 3G). Collectively, these results suggest MIG6 orchestrates the signaling switch from ALK/ROS1 primary oncogene to EGFR by modulating SHC1 adaptor binding to EGFR.

Given the observed MIG6 regulation by the MAPK pathway (Supplementary Fig. S2), we asked whether MIG6 expression is also downregulated and thus contributes to adaptive resistance in other oncogene-driven NSCLC when treated with corresponding targeted therapies. We focused on KRAS mutant–driven NSCLC because KRAS is the most frequently mutated oncogene in lung adenocarcinoma and MAPK is a well-characterized effector pathway of KRAS (37). H358 line harbors a KRAS^G12C mutation and is sensitive to sotorasib, a recently approved KRAS^G12C inhibitor (38). It has been reported that the ERBB family mediates adaptive resistance in H358 cells under KRAS^G12C or MEK inhibition (18, 39). First, we confirmed afatinib combination enhanced its sensitivity to sotorasib (Supplementary Fig. S5A). Next we observed that sotorasib decreased MIG6 protein levels in H358 cells over 48 hours, with a concomitant rebound in pERK signaling (Supplementary Fig. S5B). MIG6 knockdown markedly rescued the cell viability and pERK inhibition by sotorasib, whereas combining afatinib restored cellular sensitivity to sotorasib (Supplementary Fig. S5C and S5D). These data suggest that MIG6 could contribute to EGFR-mediated adaptive resistance in other oncogene MAPK–driven cancers.

Because we found MIG6 mediates adaptive resistance following rapid ALK/ROS1 inhibition, we asked whether MIG6 downregulation could also contribute to acquired resistance following long-term ALK/ROS1 inhibition. We generated three biological replicates of crizotinib-resistant ALK lines (H3122-CR1, -2, and -3) by continuously treating H3122 cells with 500 nmol/L crizotinib for approximately 3 months. CUTO37, a patient-derived line harboring the CD74-ROS1 fusion, was exposed to escalating doses of entrectinib for approximately 4 months and ultimately maintained in 500 nmol/L entrectinib to generate an entrectinib-resistant ROS1 line, CUTO37-ER. No ALK/ROS1 kinase domain mutations were identified in H3122-CR and CUTO37-ER cells by WES (Supplementary Table S9).

We found a substantial decrease of MIG6 and a modest increase of EGFR protein levels across all H3122-CR lines compared with the parental line (Fig. 4A). Transcriptionally, RNA-seq revealed MIG6 mRNA was decreased by 57% while EGFR mRNA was increased by 34% without expression changes of other ERBB family members in crizotinib-resistant cells (Supplementary Table S10). Afatinib was able to partially resensitize those resistant cells to crizotinib treatment (Fig. 4B and C). Only crizotinib and afatinib combination, but neither crizotinib nor afatinib alone, was able to suppress pERK in H3122 CR1–3 cells, indicating a dual dependency of ALK and EGFR in H3122 crizotinib-resistant cells (Fig. 4D).

Interestingly, CUTO37-ER cells demonstrated a complete signaling switch from ROS1 to EGFR for survival (Fig. 4E and F). Immunoblots showed the MIG6 protein was depleted in CUTO37-ER cells accompanied by a robust upregulation of EGFR expression and phosphorylation (Fig. 4G). RNA-seq revealed MIG6 mRNA was decreased by approximately 89% while EGFR mRNA was increased by approximately 265% without expression changes in other ERBB family members in entrectinib-resistant cells (Supplementary Table S11). ROS1 phosphorylation was eliminated in CUTO37-ER cells. Afatinib alone was able to suppress pERK in CUTO37-ER cells (Fig. 4G). WES did not reveal any ERBB gene family activating mutations (Supplementary Table S9). FISH also confirmed no EGFR or ERBB2 amplification in CUTO37-ER cells (Supplementary Fig. S6).

We also determined the MIG6 protein level in other fusion kinase inhibitor–resistant cells previously generated in our lab that displayed EGFR-mediated resistance. HCC78 cells, driven by SLC34A2–ROS1 fusion, and LC-2/ad cells, driven by CCDC6-RET fusion, were continuously exposed to their oncogene-targeted TKIs for months to derive HCC78-TR (TAE684-resistant) and PR2 (ponatinib-resistant) cells, respectively (15, 40). EGFR bypass signaling was reported as the acquired resistance mechanism for those cells. We found MIG6 expression was attenuated in HCC78-TR and PR2 cells compared with their parental counterparts (Supplementary Fig. S7), supporting the correlation between MIG6 downregulation and EGFR-mediated acquired resistance to fusion kinase inhibitors.

To functionally validate the role of MIG6 in acquired resistance, we investigated whether MIG6 reconstitution would resensitize resistant cells to ALK/ROS1 inhibition. Furthermore, because a decrease of phosphorylation of Y394/Y395 of MIG6 was observed under crizotinib treatment (Fig. 1D and E), we interrogated the biological relevance of phosphorylation of these two residues in regulating the ALK/ROS1 fusion–driven cell survival under primary oncogene inhibition. To test this, the tyrosine residues positioned at MIG6 394 and 395, previously shown to be critical for MIG6 binding and inhibition on EGFR, were mutated to phenylalanine, which mimics the structure of tyrosine but lacks the hydroxyl to be phosphorylated, resulting in an EGFR binding–deficient MIG6 mutant (30, 31). MIG6 overexpression was able to resensitize H3122 CR1 cells to crizotinib treatment by suppressing SHC1 and ERK activation (Fig. 5A and B). However, this ability to overcome crizotinib resistance by MIG6 was diminished by the introduction of the Y394F/Y395F mutation, suggesting the phosphorylation of MIG6 is important to suppress EGFR-mediated acquired resistance to crizotinib.

We found that WT and mutant MIG6 were expressed at a similar level in H3122 CR1 cells, and WT MIG6 alone without TKI treatment did not decrease ERK phosphorylation in H3122-CR1 cells (Fig. 5B), consistent with our finding showing single-agent afatinib treatment did not impact downstream signaling of H3122-CR1 cells (Fig. 4D). However, CUTO37-ER colony formation could be suppressed by WT MIG6 overexpression alone without entrectinib treatment, but not by the mutant MIG6 (Fig. 5C and D), in line with its complete signaling switch from ROS1 to EGFR (Fig. 4E and F). Despite a much lower expression of WT MIG6 compared with mutant MIG6 in CUTO37-ER cells, presumably due to the cytotoxic effect caused by WT MIG6 expression, WT MIG6, but not mutant MIG6, could further inhibit SHC1 and ERK activation in CUTO37-ER cells when combined with entrectinib treatment (Fig. 5E). Together, these data confirm MIG6 contribution to EGFR-mediated acquired resistance and demonstrate the phosphorylation of MIG6 Y394/Y395 is critical for MIG6-inhibitory effect on EGFR and hence the responsiveness to ALK/ROS1 inhibition.

Given the pivotal role of MIG6 in mediating resistance to ALK/ROS1 TKIs in patient-derived cell lines, we sought to identify MIG6-related biomarkers in patient samples to predict ALK/ROS1 TKI resistance and guide alternative effective treatments. A previous study has shown EGFR activity and EGFR TKI sensitivity could be more accurately predicted by MIG6 to EGFR ratio than either MIG6 or EGFR expression alone in a panel of cancer cell lines (41). We first calculated EGFR to MIG6 expression ratios from the RNA-seq data of 11 ALK/ROS1 cell lines without ALK/ROS1 kinase mutations identified (7 ROS1 lines, 1 ALK line, 1 ALK-resistant line, and 2 ROS1-resistant lines) and correlated the ratios with their responsiveness to crizotinib. We found a positive correlation suggesting a higher EGFR to MIG6 ratio is associated with resistance to crizotinib in those lines (Fig. 6A; Supplementary Table S12). We then evaluated the clinical implication of EGFR to MIG6 ratios using ALK⁺ NSCLC patient samples. In a small cohort with 7 treatment-naïve (TN) ALK patients and 5 unpaired ALK-TKI–resistant patients (23), we found a trend of elevated ratios of EGFR to MIG6 expression in resistant tumors versus untreated tumors (Fig. 6B). In another independent cohort with 28 pre-ALK TKI patients and 14 unpaired patients with post-ALK TKI, we found 4 patients with post-TKI have remarkable increased EGFR to MIG6 ratios (Fig. 6C). Collectively, these data demonstrate EGFR to MIG6 ratio increases after ALK TKI treatment, at least in a subset of patients, and correlates with ALK/ROS1 TKI resistance.

In addition, we asked whether ERRFI1 mutations that alter MIG6 expression and/or function could contribute to ALK/ROS1 TKI resistance. Targeted NGS identified a MIG6 Q241* nonsense mutation (variant allele frequency ∼100%) in the CUTO63 cell line, which harbors SLC34A2-ROS1 (S13:R32, S13:R34) fusion and was derived from a patient progressing on crizotinib treatment. This Q241* mutation results in a complete deletion of the EGFR-binding domain (EBD) on MIG6 (Fig. 6D). Indeed, the full-length MIG6 (∼55 kDa) was not detected in CUTO63 by an antibody recognizing MIG6 residues 156–280, but a truncated MIG6 protein with the predicted size (∼26 kDa) could be detected by an antibody raised against MIG6 residues 111–220 before the truncation site (Fig. 6E).

When compared with other ROS1 fusion–harboring cells, no significant overexpression and overactivation of the ERBB family were observed in CUTO63 (Fig.6E). In addition, no ERBB family activating mutations were identified in this cell line by the targeted NGS. However, the cell viability assay showed CUTO63 was remarkably sensitive to a variety of EGFR inhibitors and resistant to ROS1 inhibitors (Fig. 6F), indicating a complete signaling switch from ROS1-fusion to EGFR. The differential sensitivity between afatinib and gefitinib correlates with their potencies in inhibiting wild-type EGFR (32). Consistently, pSHC1 and pERK were not altered by ROS1 inhibition, but were suppressed by afatinib in a dose-dependent manner (Fig. 6G).

To functionally validate whether the loss of full-length MIG6 mediates the survival and EGFR dependency in CUTO63, we reconstituted the expression of the full-length WT MIG6 or the EGFR binding–deficient Y394FY395F MIG6 mutant by a doxycycline-inducible gene expression system. Indeed, 24-hour doxycycline treatment induced a similar protein expression level of WT and mutant MIG6, and only WT MIG6, but not the mutant MIG6, significantly suppressed pSHC1 and pERK in CUTO63 (Fig. 6H). This result confirms the full-length MIG6 loss in CUTO63 contributes to the EGFR-mediated resistance to ROS1 inhibitors.

To identify additional clinical cases with MIG6 alterations that could potentially contribute to acquired resistance to fusion kinase inhibitors, we explored the clinical sequencing databases in cBioPortal. We focused on the ERRFI1 genetic alterations that include ERRFI1 copy-number deletions, and MIG6 EBD truncation mutations. By combining the MSK IMPACT and the MSK MetTropism studies (24, 25), we were able to identify a ROS1 fusion–positive patient with a concurrent ERRFI1 deep deletion (Fig. 6I). The first biopsy of this patient prior to targeted therapies revealed the SDC4-ROS1 fusion; after about 1-year of treatment using crizotinib and lorlatinib, the disease progressed and the second biopsy identified no ROS1 kinase mutation but an intragenic deletion of ERRFI1, which completely deletes MIG6 exons 3 and 4, including the EBD, accompanied by a loss of heterozygosity (LOH) at ERRFI1 (Fig. 6J). Although various generations of ROS1 inhibitors were attempted, this patient did not respond and passed away within a year. This clinical case suggests ERRFI1 loss is associated with resistance to ROS1 inhibition.

To broaden our exploration and determine the frequency of ERRFI1 deletion and EBD truncation mutations in clinical samples in pan-cancer, we selected the AACR GENIE v13.0-public cohort (26), which contains 131,996 samples from 115,014 patients across different cancer types. We identified 275 patients with either ERRFI1 deletion or EBD truncation mutations, suggesting a frequency of approximately 0.24% of those potential deleterious ERRFI1 alterations in pan-cancer. Unfortunately, only approximately 30% of patients profiled for those ERRFI1 alterations were coprofiled in structural variants, thus presenting an obstacle to precisely determining the frequency of ERRFI1 alterations cooccurring with gene fusions (Supplementary Fig. S8A). However, given the MAPK regulation on MIG6 transcripts and the potential MIG6 role in mediating therapeutic responses to other oncogenes (Supplementary Fig. S5), we focused on concurrent copy-number variation and somatic mutations of a panel of gene drivers (Supplementary Fig. S8A) and summarized those cases in Supplementary Table S13. Interestingly, 16% of patients (44 of 275 patients) with MIG6-damaging alterations also have EGFR amplification and/or activating mutations. Among the 8 patients with NSCLC with concurrent MIG6 alterations and KRAS mutations, 4 of them harbor KRAS^G12C mutations which are targetable by currently approved inhibitors. We also identified a patient with NSCLC harboring a BRAF^V600E with paired pre- and posttargeted therapy biopsies. After two years of debrafenib/trametinib (BRAFi/MEKi) treatment, this patient progressed and developed two distinct ERRFI1 frame-shift mutations that completely truncated the EBD (Supplementary Fig. S8B). Together, these findings provide further support for the clinical relevance of MIG6 genetic alterations in pan-cancer.

MIG6 has been well documented as a negative regulator of EGFR and a tumor suppressor in several cancers (27–31), but little is known about MIG6 role in ALK/ROS1 fusion kinase–driven NSCLC. This study reveals that MIG6 can mediate both early, adaptive resistance as well as acquired resistance following ALK/ROS1 inhibition (Supplementary Fig. S9). The rapid downregulation of MIG6, as an adaptive resistance mechanism, will enable cancer cells to tolerate oncogene inhibition and create a survival niche, from which acquired resistance can later arise. A previous study by our lab demonstrated superior in vivo efficacy of crizotinib and gefitnib combination compared with crizotinib alone in suppressing tumor growth of H3122 xenografts (42). The EGFR-mediated adaptive resistance proposed here is reminiscent of the EGFR feedback activation observed in BRAF^V600E-driving colon cancer following BRAF inhibition (43). Although mediated by a distinct mechanism, cetuximab, an anti-EGFR antibody, combined with encorafenib, a BRAF inhibitor, demonstrated improved clinical outcomes compared to BRAF or BRAF/MEK inhibition in metastatic patients with colorectal cancer with a BRAF^V600E (44). Therefore, implementing EGFR inhibitors with ALK/ROS1 TKIs early in the treatment paradigm may suppress or eliminate residual disease to delay disease progression.

Given the observed MIG6 regulation by the MAPK pathway, the resistance mechanism presented here may have broader implications for other oncogene-driven cancers. In EGFR mutant NSCLC, MIG6 depletion may not offer a significant survival advantage if cancer cells are already treated with EGFR inhibitors. However, for non-EGFR oncogene-driven cancers, MIG6 decrease may activate EGFR as a bypass signaling-mediated resistance mechanism. MIG6 was reported to be MEK-regulated and inhibit EGF-induced migration and invasion of NRAS mutant melanoma cells (45). Recently, ACK1, which contains a region homologous to the MIG6 EBD, was shown to be downregulated in vemurafenib-resistant BRAF-mutant melanoma, contributing to EGFR-mediated BRAFi resistance (46). We have demonstrated that the MIG6 protein level was reduced by sotorasib, a KRAS^G12C inhibitor, and MIG6 knockdown rescued cell viability and ERK inhibition by sotorasib in the H358 cell line bearing KRAS^G12C mutation (Supplementary Fig. S5). We hypothesize MIG6 could also contribute to EGFR-mediated adaptive resistance following KRAS^G12C TKI treatment, especially for the epithelial-like KRAS^G12C cells which have higher basal ERBB family expression with a tendency to depend on ERBB signaling for survival under KRAS inhibition (39).

We also identified two potential biomarkers that may allow identification of tumors that have MIG6-mediated EGFR activation. We demonstrated that MIG6 damaging mutations/deletions are associated with EGFR bypass signaling and these are readily detectable on existing comprehensive NGS assays. Furthermore, a high EGFR to MIG6 expression ratio was associated with decreased ALK/ROS1 TKI sensitivity in vitro and found to be elevated in some patient tumor samples following ALK inhibition. We showed the frequency of ERRFI1 deep deletion and EBD truncation mutations is approximately 0.24% in a pan-cancer analysis, however, our analysis may underreport the true frequency of ERRFI1 functional loss as we did not include ERRFI1 gene fusion, splice mutations, and missense mutations that might also diminish MIG6 binding and inhibition on EGFR. Because of the lack of treatment information records, we could also not further evaluate the frequency of MIG6 alterations under oncogene inhibition in this cohort. Interestingly, several MIG6 EBD truncation mutations and a R247K missense mutation were recently identified in ctDNA profiling in patients with ALK-TKI naïve neuroblastoma and the pathologic functions of those MIG6 alterations were validated in neuroblastoma cells. Low MIG6 expression was also shown to be associated with a poor prognosis in several neuroblastoma patient cohorts (47). Although we focused on MIG6 alterations in the resistance setting to existing targeted therapies to primary oncogenes, we did not exclude the possibility that MIG6 alterations alone could be oncogenic in absence of other driver alterations.

Another notable finding in this study is that we discovered a novel mechanism that regulates EGFR signaling activity without greatly impacting EGFR phosphorylation itself in ALK/ROS1 fusion–positive cells. Consistent with the structural evidence showing MIG6 and SHC1 bind to the same substrate-binding cleft of EGFR (30, 34), we demonstrated that MIG6 competes with SHC1 for EGFR binding in ALK/ROS1-driven cancer cells and suggest this competition could mediate the signaling switch between ALK/ROS1 fusion and EGFR. Because SHC1 forms a signaling complex with GRB2 (34), this finding might explain previously reported GRB2 adaptor switching between EGFR and ALK/ROS1 fusion kinases following ALK and ROS1 TKI treatment (42). In addition, the position equivalent to EGFR K879, the key residue mediating the SHC1 and MIG6 binding to EGFR (30, 34), is conserved in ERBB2 and ERBB4, suggesting MIG6 may also block SHC1 binding to the substrate binding clefts of ERBB2 and ERBB4 to be phosphorylated.

The observation of a rapid decrease of MIG6 Y394/Y395 phosphorylation under fusion kinase inhibition leads to the question of how MIG6 is phosphorylated in ALK/ROS1-driven cell lines. Interestingly, a prior biochemical study found ACK1, which contains a homologous region to the MIG6 EBD, physically interacted with ALK via GRB2 but not with kinase-inactive ALK (48). We speculate MIG6 could be phosphorylated by ALK/ROS1 fusion kinases directly or by their downstream effectors indirectly. Further study is required to understand the mechanism contributing to MIG6 phosphorylation in ALK/ROS1-driven cancer cells.

In summary, our data provide a strong mechanistic basis for evaluating the combination of ALK/ROS1 and EGFR inhibitors in the first-line setting to minimize residual disease and improve the therapeutic responses in ALK/ROS1 fusion–positive patients. This study also suggests using MIG6 alterations and/or high EGFR to MIG6 expression ratios as biomarkers to stratify patients who are likely resistant to ALK/ROS1 TKI monotherapy but may respond to EGFR TKIs or antibodies. Finally, this work shows that this MIG6-mediated resistance mechanism may apply more broadly in pan-cancer, given both MAPK-mediated regulation and MIG6 alterations in many oncogene-driven cancers.

N. Chen reports a patent for PCT/US2022/045903 pending. A.T. Le reports a license for CUTO cell lines with royalties paid. A. Elliott reports personal fees from Caris Life Sciences during the conduct of the study. T. Danhorn reports grants from NIH/NCI during the conduct of the study. G.J. Riely reports grants from Rain, grants from Novartis, grants from Takeda, grants from Roche, grants from Mirati, and grants from Pfizer outside the submitted work. E.B. Haura reports grants, personal fees, and non-financial support from Revolution Medicines and personal fees and non-financial support from Janssen Pharmaceuticals outside the submitted work. R.C. Doebele reports personal fees and other support from Rain Oncology, other support from Casma Therapeutics, other support from Takeda, other support from Loxo, other support from ThermoFisher, other support from Foundation Medicine, other support from Voronoi, other support from Black Diamond Therapeutics, and other support from Histocyte outside the submitted work; in addition, R.C. Doebele has a patent for PCT/US2022/045903 pending. No disclosures were reported by the other authors.

N. Chen: Conceptualization, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. L.C. Tyler: Resources, validation, investigation, writing–review and editing. A.T. Le: Resources, methodology, writing–review and editing. E.A. Welsh: Data curation, software, formal analysis, writing–review and editing. B. Fang: Data curation, investigation. A. Elliott: Resources, data curation. K.D. Davies: Formal analysis, investigation, writing–review and editing. T. Danhorn: Data curation, software, formal analysis, writing–review and editing. G.J. Riely: Resources, funding acquisition, project administration. M. Ladanyi: Resources, funding acquisition, project administration. E.B. Haura: Resources, supervision, funding acquisition, project administration, writing–review and editing. R.C. Doebele: Conceptualization, resources, supervision, funding acquisition, project administration, writing–review and editing.

This study was supported by the NIH/NCI R01 CA193935 grant (to R.C. Doebele), the University of Colorado Lung SPORE P50 CA058187 grant (to R.C. Doebele), and the Addario Lung Cancer Medical Institute (ALCMI) grant (to R.C. Doebele). The works from Bioinformatics and Biostatistics Shared Resource and Pathology Shared Resource were supported by NIH/NCI P30 CA046934 grant. We sincerely thank the ROS1ders patient advocacy group (https://www.theros1ders.org) for their support of this research and the donation of tissue samples to generate patient-derived ROS1 cell lines. We thank Dr. Frederick Wilson for providing the RNA-seq data for the ALK patient cohort. We thank Dr. Cecilia Caino for the critical reading of the manuscript.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.