Host-Dependent Phenotypic Resistance to EGFR Tyrosine Kinase Inhibitors

Both intrinsic and acquired patient resistance severely limit efficacy of nearly all successful cancer therapies (1). Understanding mechanisms underlying cancer drug resistance may take lessons from the much older field dealing with resistance to antibacterial therapies. For example, the tolerance of biofilms to antimicrobials is fundamentally different from the tolerance displayed by bacteria grown in planktonic cultures (2). Furthermore, multiple tolerance mechanisms confer bacterial phenotypic resistance, which might predispose to genetic resistance. A similar interplay between phenotypic and genetic resistance, as well as the influence imposed by environmental conditions, might be relevant to cancer treatment. For example, it has been proposed that clinical drug resistance is due to simultaneous changes in expression of a large number of genes, which have a reversible (nonmutational) cumulative impact on drug sensitivity (3). Along this line, reversible epithelial–mesenchymal transition (EMT) and acquired resistance to a kinase inhibitor, sunitinib, have been observed in a patient with renal cell carcinoma (4).

Lung cancer is responsible for the majority of cancer-related deaths worldwide (5). Most cases of lung cancer are characterized as non–small cell lung cancer (NSCLC; ref. 6), and many express the EGFR (7). Somatic driver mutations in the EGFR gene are frequently detected in NSCLC (8). To overcome the deleterious effects of such mutations, three generations of tyrosine kinase inhibitors (TKI) have been developed. The majority of patients whose tumors harbor EGFR-activating mutations initially respond to treatment with TKIs, but drug resistance inevitably evolves (9, 10). Approximately 55% of acquired resistance to the first-generation drugs is linked to the intrinsic T790M mutation (11–13). However, other processes might be involved, such as c-MET amplification (14), AXL overexpression (15), and activation of the epigenetic program called EMT (16, 17). While resistance to other EGFR TKIs is, in general, well characterized, the mechanism of resistance to dacomitinib is less clear. Dacomitinib, a highly selective TKI, covalently binds with three receptors of the EGFR family (EGFR, HER2, and HER4). It was approved in 2018 as a first-line treatment for patients with NSCLC harboring EGFR mutations, but only a few studies addressed mechanisms of resistance. For example, it was shown that chronic exposure of engineered myeloid cells to dacomitinib induced the T790M mutation, whereas cotreatment with a mutagen resulted in additional mutations, such as C797S and G719A (18).

In analogy to drug-tolerant subpopulations of bacteria, which play important roles in recurrent infections (19), a small subpopulation of drug-tolerant persister cells (DTP) has been reported (20). These cells demonstrate reversible tolerance and they can be inhibited by an inhibitor specific to the insulin-like growth factor 1 receptor (IGF1R), or with chromatin-modifying agents. Likewise, resistance of breast cancer to endocrine therapy is preceded by genome-wide reprogramming of the chromatin landscape (21). However, how the reversible DTP states are replaced by permanent drug-resistant states is currently unclear. According to one model, cancer cells enter a state of reversible cell-cycle arrest, which permits acquisition of mutations (22). Yet, according to another model, drug-treated cells transiently increase their mutation rates (adaptive mutability) and acquire resistance (23). To better understand mechanisms of resistance, we established dacomitinib-resistant cells from PC9 lung cancer cells. Cell viability assays revealed that the dacomitinib-resistant cells (PC9DR) also resist other EGFR TKIs. Whole-exome sequencing (WES), along with RNA sequencing (RNA-seq), cytokine arrays, and reverse phase protein arrays (RPPA) uncovered that PC9DR cells acquired a mesenchymal phenotype, which comprised upregulation of AXL. However, resistance to dacomitinib was reversed when PC9DR cells were implanted in animals. Moreover, when examined ex vivo, tumor-derived cell lines exhibited EMT and renewed resistance to dacomitinib. Taken together, these observations uncover a hitherto unknown interim state of drug-tolerant cells and indicate that host-dependent mechanisms can overcome phenotypic resistance.

Drugs were obtained from Medchem Express or from Sigma, and antibodies were purchased from Cell Signaling Technology, unless otherwise indicated. PC9 and HCC2935 cells were from ATCC and 3T3 cells from JCRB (JCRB9014). Periodic tests for Mycoplasma and authentication were performed using commercially available kits.

To establish acquired resistance to dacomitinib in PC9 cells, we followed previously described protocols (24). In short, PC9 cells were seeded at approximately 70% confluence in RPMI1640 with 10% FBS. Dacomitinib was added at a starting concentration of 1 pmol/L, and cells were passaged once they reached confluence. Dacomitinib was increased once every 2 weeks in half-log intervals until a final concentration of 100 nmol/L was reached.

Cells were washed and resuspended in serum-free medium. Thereafter, the cells were added into transwell inserts with 8-μm pore polycarbonate filters precoated with invasion matrix (BD Biosciences). Following 18 hours of incubation, noninvaded cells on top of the membrane were removed with a cotton swab. Cells invaded into the bottom side of the membrane were fixed and stained. The number of invaded cells on the membrane was then determined using the ImageJ software.

Spheroids were generated by means of the hanging drop method. Medium (20 μL) containing cells (3 × 10³) was dropped in the cap of a 60-mm dish filled with saline. After 72 hours, the spheroids were treated with dacomitinib (100 nmol/L). On the third day, we captured images of spheroids under treatment.

Short hairpin RNAs (shRNA) targeting AXL (TRCN0000001039 and TRCN0000001040) and IGF1R (TRCN0000039675 and TRCN0000039677), or control shRNAs, were obtained from Sigma-Aldrich. Lentiviruses were packaged by cotransfecting HEK-293 cells with shRNAs vectors, psPAX2 (Addgene, #12260) and pMD2.G (Addgene #12259), along with the jetPEI reagent. PC9DR cells were infected and selected under puromycin (2 μg/mL).

All experiments involving animals were approved by the Weizmann Institute's Review Board and performed in accordance with the guidelines of the Institutional Animal Care and Use Committee. PC9 and PC9DR cells (3 × 10⁶ per mouse) were subcutaneously injected in the right flanks of 6-week-old female CD1 nude or NSG mice. Once tumors reached a volume of approximatively 500 mm³, mice were divided in different groups and orally treated daily with the indicated kinase inhibitors. Tumors were measured twice a week and body weight was measured once a week. Tumor volume was calculated by using the formula = 3.14 × (shortest diameter × longest diameter²)/6. Mice were euthanized when tumors reached 1,500 mm³.

PC9DR cells were cultured in the presence of 100 nmol/L dacomitinib, and 3 days before injecting them into the flanks of CD1 nude mice (3 × 10⁶ cells per mouse) dacomitinib was removed. Until day 21, all mice were kept without any treatment. On day 22, mice were divided into two groups: control (N = 4), and dacomitinib-treated mice (N = 6). Dacomitinib (1 mg/kg) was administered daily. Mice were treated for 7 days and on day 29, they were sacrificed. Tumor dissociation was conducted by means of enzymatic digestion in RPMI medium containing FBS (1%), DNase I (2 μg/mL; Sigma-Aldrich) and collagenase type II (1 mg/mL). Following incubation for 3 hours at 37°C, cell suspensions underwent vigorous pipetting (20–25 times) by using a 5-mL syringe. The enzymatic reaction was stopped by adding media containing 10% FBS. The cell suspension was then filtered using 40-μm cell strainers and cells were harvested by centrifugation, resuspended in media containing FBS and cultured in the absence of dacomitinib.

The RNA-seq and WES datasets generated in this study are available at Gene Expression Omnibus (accession number GSE168043; https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE168043) and at Sequence Read Archive (accession number PRJNA705746; https://www.ncbi.nlm.nih.gov/sra/PRJNA705746), respectively.

Results are presented as means ± SD or SEM. Experiments were analyzed using the software GraphPad Prism (version 7.0). Statistical analyses were performed using t test or one-way or two-way ANOVA with Tukey, Bonferroni, or Dunnet multiple comparison test (*, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001).

All correspondence and material requests should be addressed to Yosef Yarden ([email protected]).

A previous attempt to resolve mechanisms of resistance to dacomitinib employed a murine myeloid cell line, the pro–B-cell line Ba/F3 (18). As an alternative, we made use of the PC9 NSCLC cell line, which is frequently utilized as a model system because these cells naturally express the most abundant class of EGFR mutations, exon 19 deletions. While the majority of PC9 cells were killed within 9 days of exposure to dacomitinib, small fractions of viable cells DTPs (20) survived treatment with increasing doses of the drug (Fig. 1A). Potentially, surviving cells might seed long-term resistant clones. Hence, we characterized the dacomitinib-tolerant persisters, especially in terms of their predicted sensitivity to a histone deacetylase (HDAC) inhibitor and an inhibitor of IGF1R, linsitinib (20). Cell viability assays confirmed that DTPs established under dacomitinib (either 100 nmol/L or 1 μmol/L) gained tolerance to the TKI and concurrently acquired sensitivity to both trichostatin A (TSA), an inhibitor of HDACs, and linsitinib (Fig. 1B). Additional assays revealed that PC9DTPs acquired resistance to all three EGFR inhibitors we tested (i.e., erlotinib, afatinib, and osimertinib), but they remained sensitive to chemotherapeutic drugs (Fig. 1C).

Notably, long-term resistance to EGFR inhibitors frequently entails emergence of a secondary mutation, T790M (11). Alternatively, resistance might be due to amplification of c-MET (14), phenotypic alterations (16) or activation of bypass routes involving IGF1R (25) or AXL (15). Probing extracts of dacomitinib DTPs unveieled strong inhibition of EGFR autophosphorylation, along with a partly inhibited downstream pathway, ERK (Fig. 1D). These observations implied that an ERK-independent pathway compensated for the extiguished EGFRs. Along this line, we detected upregulation of both AXL and IGF1R, as well as their phosphorylated forms (Fig. 1D). Notably, phosphorylation of c-MET was inhibited rather than enhanced in dacomitinib DTPs, and according to recent reports inhibitors of AXL can suppress emergence of DTPs (26, 27). Next, we replicated the experiments with two additional NSCLC lines, HCC2935 and H3255 cells, which, respectively, harbor an EGFR exon 19 deletion and the L858R mutation. The results obtained with DTPs established from HCC2935 cells are presented in Supplementary Fig. S1. In similarity to PC9DTPs, the new DTPs acquired resistance to several EGFR inhibitors (Supplementary Fig. S1A and S1B), downregulated MET and upregulated AXL, IGF1R, and vimentin (Supplementary Fig. S1C). Collectively, the dacomitinib-tolerant persisters we derived from several NSCLC lines shared functional features with the previously described gefitinib DTPs (20), and they depend on histone acetylation and two survival receptors, IGF1R and AXL.

To identify molecular mechanisms conferring resistance to dacomitinib, we followed a previously established protocol (24). Dacomitinib was incubated with PC9 cells at a starting concentration of 1 pmol/L, which was increased in half-log intervals up to 100 nmol/L, approximately 4.5 months later. Drug concentrations were increased once every other week, and both medium and drug were repeatedly replenished (Fig. 2A). Once established, we used cell viability assays to confirm the phenotype of the dacomitinib-resistant cells (PC9DR; Fig. 2B). In addition, we performed a DNA synthesis assay (Supplementary Fig. S2A) and an alternative cell viability assay (Supplementary Fig. S2B). Both tests demonstrated that PC9DR cells acquired faster rates of cell proliferation and metabolism. Next, we used immunoblotting, which revealed that dacomitinib completely blocked phosphorylation of EGFR in both PC9 and PC9DR cells, but nevertheless both ERK and AKT retained their activities (Fig. 2C). In similarity to dacomitinib DTPs, analyses of PC9DR cells detected upregulation of AXL and IGF1R.

The ability of dacomitinib to block EGFR autophosphorylation in PC9DR cells predicted absence of EGFR-activating secondary mutations. To validate this prediction, we used PCR to amplify and later sequence exons 19, 20, and 21, which harbor the most frequent sites of EGFR-activating mutations. This analysis detected no new genetic aberrations (Supplementary Fig. S2C). To detect mutations in other genes and exclude EGFR-activating mutations affecting other exons, we applied whole-exome DNA sequencing. Genomic DNA was isolated from PC9 and PC9DR cells and analyzed by DNA link (https://www.dnalink.com/english/service/exome_sequencing.html). While no new EGFR mutations, other than the original exon 19 deletion, were identified, we detected several mutations that were not shared by PC9DR and PC9 cells. Supplementary Table S1 lists all differences, including mutations in ALK and RAF1. Notably, multiple ALK fusion partners and distinct mutations may act as drivers of NSCLC (28), while mutations in BRAF, a family member of RAF1, are found in 2% to 4% of all NSCLC.

To examine resistance of PC9DR cells to other EGFR TKIs, we tested the effects of erlotinib, afatinib, and osimertinib. In similarity to the respective DTPs, PC9DR cells showed resistance to all three EGFR TKIs (Supplementary Fig. S3). In contrast, cell viability assays that used doxorubicin and paclitaxel showed that PC9DR cells acquired no chemoresistance. In conclusion, similar to the respective DTPs, the established PC9DR cells remained sensitive to chemotherapy but acquired resistance to all four EGFR inhibitors we tested. Interestingly, the mechanism of pan-TKI resistance bypassed EGFR and made no use of secondary EGFR mutations. Presumably, the mechanism of evasion utilizes mutations in other genes, or it epigenetically engages RTKs previously implicated in survival of TKI-treated cancer cells.

To fully resolve the transcriptional landscape of PC9DR cells, we conducted RNA-seq analysis. The results reflected upregulation of a large group of genes associated with TGFβ signals, EGFR pathway, and a mesenchymal phenotype (Fig. 3A and B; see Supplementary Table S2). For example, genes encoding fibronectin, vimentin, AXL (along with its ligand, GAS6), and SNAI2 (Slug) were highly active in PC9DR cells, while epithelial phenotype genes, such as a subset of the keratin family, along with OVOL1 transcripts, were downregulated (Fig. 3C and D). As expected, analysis of PC9DR extracts confirmed upregulation of vimentin, snail and AXL, and downregulation of OVOL1 (Fig. 3E). Likewise, an ELISA specific to GAS6-detected upregulation in PC9DR cells (Fig. 3F).

Next, we utilized two high-throughput platforms, RPPAs (Supplementary Fig. S4A and S4B) and cytokine arrays (Supplementary Fig. S5A and S5B). Prior to RPPA, cells were treated with dacomitinib for increasing time intervals. Spotted cell lysates were probed using pre-calibrated antibodies (Supplementary Table S3). Evidently, the phenotype of PC9DR cells extended beyond EMT to survival pathways and the cell cycle. Furthermore, we observed consistent time-dependent upregulation of several RTKs, including not only AXL, c-MET, and IGF1R, but also ERBB4 (Supplementary Fig. S4A). Western blots further revealed that unlike AXL, MERTK, and TYRO3, its family members, displayed only minor differences (Supplementary Fig. S4B). In addition to RTKs, PC9DR cells upregulated two ligands of EGFR, amphiregulin, and TGFα. To identify additional components of the secretome, we subjected media conditioned by PC9 and PC9DR cells to cytokine array analysis (Supplementary Fig. S5A). Interestingly, we observed increased secretion by PC9DR cells of a metalloproteinase, MMP9 (>10-fold), along with elevated secretion of complement factor D and macrophage migration inhibitory factor, and downregulation of resistin (an adipokine, >10-fold), ST2 (>7-fold), and IGFBP2 (>4-fold; Supplementary Fig. S5B). The latter was confirmed by the RPPA results. In addition to IGFBP2, the array detected downregulation of two RTK ligands, FGF19 and PDGF-AA.

Because previous studies linked EMT to both stemness (29) and resistance to EGFR inhibitors (16, 30, 31), we assayed aldehyde dehydrogenase (ALDH), a marker of embryonic and cancer stem cells. The results indicated that PC9DR cells are characterized by relatively high ALDH activity (Supplementary Fig. S6), consistent with cancer stem or progenitor states. Taken together, high-throughput analyses of dacomitinib-resistant cells uncovered a complex evasive response that concurrently controls secretion of growth factors and proteases, as well as upregulates several RTKs, while launching the interlinked stem- and EMT-like programs.

EMT is a reversible epigenetic process whereby epithelial cells acquire mesenchymal features, including enhanced motility (32). In line with the proteomic and transcriptomic analyses, migration assays confirmed that PC9DR cells acquired a highly migratory phenotype (Fig. 4A). In a similar way, we found that these cells gained a 4-fold stronger capacity to cross an extracellular matrix barrier (Fig. 4B). In addition, PC9DR cells displayed enhanced clonogenic capacity (Fig. 4C). Along this line, we compared the ability of the two cell lines to rapidly spread and adhere to fibronectin, a property shared by mesenchymal stem cells (33). The results confirmed more rapid and extensive adhesion of PC9DR cells to fibronectin (Fig. 4D). Next, we performed actin immunofluorescence analysis (Fig. 4E) and three-dimensional (3D) spheroid assays (Fig. 4F), which evaluated the ability to form cellular assemblies with specific architecture (34). As shown, PC9DR cells exhibited an elongated morphology and more cortical actin filaments. Furthermore, treatment of PC9 cells with dacomitinib reduced spheroid size, but unlike the parental cells, PC9DR cells displayed reduced capacity to form spheroids (Fig. 4F). Notably, dacomitinib exerted no effect on the relatively loose structures formed by PC9DR cells. In conclusion, these results confirmed acquisition of an invasive phenotype by PC9DR cells, congruent with their EMT hallmarks.

Our observations proposed that PC9DR cells adopted compensatory epigenetic programs able to bypass EGFR by means of upregulating several alternative receptors (e.g., AXL, IGF1R, c-MET, and ERBB4), which support survival and instigate EMT. In line with this model, PC9DR cells were more sensitive than PC9 cells to relatively low concentrations of TSA (Fig. 5A). Next, we separately examined the consequences of inhibiting individual receptors. Focusing on IGF1R, we noted that PC9DR cells remained partly sensitive to linsitinib (Fig. 5B). Likewise, viability assays focusing on c-MET and AXL and utilizing specific inhibitors, capmatinib and TP-0903, respectively, unveiled reliance of PC9DR cells on AXL, rather than c-MET (Fig. 5C). In addition, whereas PC9 cells were completely inhibited by a combination of compounds inhibiting AXL, c-MET, and EGFR, the effect on PC9DR cells was much smaller. Hence, in comparison with PC9 cells, the robust growth of PC9DR cells might be driven by a wider spectrum of signaling routes. Two additional lines of evidence highlighted AXL's contribution: (i) TP-0903 partially inhibited migration of PC9DR cells, but this AXL-specific TKI did not affect migration of PC9 cells (Fig. 5D), and (ii) overexpression of AXL using an expression plasmid reduced the sensitivity of PC9 cells to dacomitinib (Fig. 5E).

Next, we depleted AXL and IGF1R from PC9DR cells by means of shRNA-mediated knockdown (Supplementary Fig. S7A). As expected, shIGF1R clones increased sensitivity of PC9DR cells to dacomitinib, depending on knockdown efficacy, such that the more potent clone displayed similar sensitivity to that displayed by the parental PC9 cells (Supplementary Fig. S7B). A similar, albeit weaker effect, was displayed by cells stably expressing short hairpin specific to AXL (shAXL). Consistently, depleting either AXL or IGF1R enhanced the effect of dacomitinib on pAKT (Supplementary Fig. S7C) and inhibited migration of PC9DR cells, but analysis of EMT markers unveiled complex relations between receptor knockdown and EMT markers (Supplementary Fig. S8A and S8B). In conclusion, resistance to dacomitinib appears to be driven by epigenetic enhancement of several bypass routes, including the AXL pathway, a well-characterized driver of EMT and resistance to EGFR inhibitors (35).

As an ultimate assessment of the de novo acquired ability of PC9DR cells to withstand treatment with dacomitinib, we implanted cells in the flank of immunocompromised mice. When tumors became palpable, we started daily treatments with the drug. Unexpectedly, tumor volumes were rapidly reduced, in similarity to the regression displayed by dacomitinib-treated PC9 tumors (Fig. 6A). Treating pre-established tumors with an AXL-specific inhibitor, TP-0903, only weakly influenced tumor growth and, likewise, the combination of TP-0903 and dacomitinib was nearly as effective as dacomitinib alone. These in vivo observations implied that the mutant form of EGFR regained, while AXL lost driver activities. To explore the reversible phenotype of PC9DR cells, we analyzed tumor extracts (two mice per treatment). Unlike TP-0903, dacomitinib-treated animals bearing either PC9DR or PC9 tumors displayed strongly decreased phosphorylation signals corresponding to EGFR, HER2, AKT, and ERK (Fig. 6B). In addition, AXL showed high expression levels in the control tumors but this, along with c-MET levels, decreased rather than increased, after treatment with dacomitinib. To try and simulate tumors treated with dacomitinib, we used the hanging drop method to generate 3D spheroids. Whole extracts of spheroids, along with extracts from adherent PC9 and PC9DR cells [two dimensional (2D)], were resolved using immunoblotting (Fig. 6C). The results indicated that the overall expression levels of AXL decreased when cells were grown in 3D formats, and the ability of dacomitinib to elevate AXL and vimentin was nullified. In summary, unlike 2D cultures, spheroids and tumors treated with dacomitinib showed no upregulation of AXL and this might contribute to the observed reversal to a drug sensitive state when PC9DR cells were implanted in animals.

Presumably, the in vitro applied procedures we used to establish dacomitinib-resistant cells caused emergence of epigenetic or metabolic rewiring, which are inhibitable in vivo by host factor(s). To test this model, we established ex vivo tumor cell lines and examined their sensitivity to dacomitinib. To this end, we firstly implanted PC9DR cells in 10 untreated CD1 nude mice and once tumors reached approximately 1,000 mm³, mice were randomized to two groups: (i) a “holiday group” (control) was maintained for 29 days prior to surgery, and (ii) a “treatment group,” which received dacomitinib on a daily basis, from day 22 through day 29 (Fig. 7A). After confirming tumor regressions in the treatment group, both groups underwent surgery on the same day and 10 cell lines were established, four control lines (C1–C4) and six additional lines (D1–D6) were derived from dacomitinib-treated mice. Cell viability assays revealed that all ex vivo lines were resistant to dacomitinib (100 nmol/L), unlike the parental PC9 cells (Fig. 7B). This observation lent support to the aforementioned model assuming stable rewiring that can be reversibly inhibited by soluble factors or physical parameters of the host. Next, we performed immunoblotting analyses of all clones, along with PC9, PC9DR, and a murine fibroblast cell line (3T3). The latter line was used to verify absence of contaminating murine fibroblasts, which can be detected by means of an antibody to smooth muscle actin (α-SMA; Fig. 7C). Interestingly, the newly established lines displayed a rather uniform expression pattern, which included the original characteristics of PC9DR cells, such as relatively high abundance of c-MET, AXL, and vimentin, and relatively low abundance of OVOL1.

To confirm retention of additional PC9DR characteristics by the ex vivo lines, as opposed to the parental PC9 cells, we examined sensitivity to TKIs (Supplementary Fig. S9A) and rates of cell migration/invasion (Supplementary Fig. S9B). As predicted, all newly established lines we examined, like PC9DR, displayed resistance to erlotinib, afatinib, and osimertinib, but they remained sensitive to paclitaxel. In contrast, PC9 cells displayed sensitivity to all TKIs. Consistently, all four tumor-derived lines displayed enhanced migration and invasion. In conclusion, ex vivo derivation of cell lines from dacomitinib-sensitive PC9DR tumors further supported the working model: host factors likely negate the ability of PC9DR cells to survive treatment with dacomitinib. In the absence of the putative factors, the rewired PC9DR cells regain resistance to EGFR inhibitors and display in vitro the original motile phenotype.

Clinical approvals of anti-NSCLC drugs targeting angiogenesis and immune checkpoints exemplify the critical roles played by the tumor microenvironment in the pathophysiology of lung cancer (36). To test involvement of immune cells, we used nude mice, which lack T cells but have natural killer (NK) cells, and NSG mice, which have no T, B, and NK cells. Despite these differences, in both strains we observed similar regressions of preestablished PC9DR tumors following treatment with dacomitinib (Supplementary Fig. S10). These observations weakened the possibility that immune cells are involved in the renewed sensitivity of PC9DR cells to dacomitinib. Notably, hypoxia contributes to resistance to drugs (37). For example, hypoxic tumor microenvironments promote innate resistance to kinase inhibitors (38). Hypoxia-inducible factor 1-alpha (HIF1a) controls both angiogenesis and metabolic reprogramming. Assuming that tumor hypoxia involves secretion of HIF-induced tumor factors able to modify drug resistance, we maintained PC9DR and PC9 cells under hypoxic or normoxic conditions and determined cell viability (Supplementary Fig. S11A). Although immunoblotting confirmed hypoxia-induced induction of HIF1a and activation of ERK and AKT in PC9DR cells (Supplementary Fig. S11B), the results of cell viability assays indicated that the response to dacomitinib was unaltered by the state of environmental oxidation.

Cancer-associated fibroblasts can either promote or inhibit carcinomas (37). Hence, we assumed that mouse fibroblasts can inhibit resistance to dacomitinib by means of either soluble factors or secreted vesicles. Hence, we cocultured lung cancer cells, using transwells, with 3T3 mouse fibroblasts, and performed a series of assays 6 days later. Cell viability assays were unable to detect differences between monocultures of NSCLC cells and cocultures comprising murine fibroblasts (Supplementary Fig. S11C). Likewise, when using flow cytometry and determining the fractions of cells undergoing apoptosis, we detected no effects of the cocultured fibroblasts (Supplementary Fig. S11D). Next, we used immunoblotting to resolve potential effects of fibroblasts on activation of EGFR and downstream effectors. Immunoblotting confirmed that dacomitinib inhibited pEGFR in PC9 and PC9DR cells growing either in monocultures or in cocultures, and AXL was highly expressed in drug resisters. Similarly, ERK and AKT were inhibited by dacomitinib in PC9 cells, but this TKI exerted weaker effects on the TKI-resistant cells, independent of the presence of fibroblasts (Supplementary Fig. S11E). Altogether, our assays were unable to support a model attributing to fibroblasts a functional role in overcoming resistance to dacomitinib.

In summary, because the dacomitinib-resistant cells we established in vitro reverted to a drug-sensitive state when implanted in animals, but they regained resistance when returned to culture, we assume that specific molecule(s) or physical conditions exclusively existing in vivo reversibly nullified drug resistance. Although we were unable to identify the putative host-originated factor, we assume that no de novo mutations were involved in either the gain or the loss of resistance to dacomitinib. As far as we are aware, no previous report has described a similar interim reversible state. According to our data, generation of the reversible state entails epigenetic rewiring of gene expression programs, particularly events regulating EMT, including AXL, IGF1R. Below we discuss the emerging relations between the reversible resister state and two other cellular states: the precursors, drug-tolerant persisters, and the irreversibly acquired resister state.

Mechanisms conferring resistance to TKIs might be divided into two classes: mechanisms involving emergence of new mutations and nonmutational modes of resistance (1, 39). For example, analyses of tumor biopsies from patients with drug-resistant NSCLC carrying EGFR mutations identified cancers expressing mutant forms of the PIK3CA gene (40), BRAF (40, 41), and MAPK1 (42). The nonmutational mechanisms of resistance are less understood. For example, five cases of transition to small cell lung cancer, as well as two EMT cases were identified in a survey of 37 patients with NSCLC who acquired resistance to EGFR inhibitors (40). In similarity to other TKIs, resistance to dacomitinib may involve both mutational and nonmutational mechanisms. Chronic dacomitinib treatment of murine myeloid cells ectopically expressing Del19 EGFR induced emergence of the T790M mutation (18). In contrast, our PCR and WES analyses, along with the reversible nature of PC9DR's resistance, weaken the possibility that tolerance was due to new mutations.

Several lines of evidence support the possibility that the nonmutational mechanism relevant to PC9DR cells borrowed functional features from EMT. Simultaneous upregulation of several RTKs, including AXL, accompanied the EMT phenotype. AXL has previously been linked to EMT and resistance to EGFR inhibitors (15, 43). It can activate EGFR and c-MET, as well as translocate EGFR to the nucleus (44) and facilitate ERK and PI3K signals (45). In our study, upregulation of AXL associated with PC9DR cells and with increased levels of GAS6. Thus, our results raise the possibility that AXL, its ligand, GAS6, and perhaps also the cleaved form of AXL (sAXL), might herald emergence of resistance to kinase inhibitors, hence serve as biomarkers. Moreover, cotargeting AXL and EGFR might offer a “roadmap” to overcoming resistance to EGFR inhibitors. Notably, previous studies proposed that sAXL might serve as a biomarker of response to kinase inhibitors (46) or showed that AXL confers intrinsic resistance to osimertinib (26).

The observed in vivo acquisition of drug sensitivity by PC9DR cells seems relevant to a previously reported clinical phenomenon: patients who previously received EGFR TKI but developed resistance and then switched to chemotherapy, unexpectedly derived survival benefit from renewed TKI treatments (47–49). Similarly, xenografts established from a patient with renal carcinoma who initially had a response to sunitinib but eventually progressed, regained sensitivity to the drug (4). What mechanisms may reverse TKI resistance? We speculate that inhibitors of EMT might underlie reversibility. It is relevant that PC9DR cells acquired both drug resistance and a mesenchymal phenotype while under dacomitinib, and they lost both features when implanted in animals. Similarly, it has been reported that two gefitinib-resistant NSCLC cell lines, which exhibited EMT, regained sensitivity to gefitinib and lost EMT after long-term culture (50). Hence, exit from EMT might explain the reversible, host-dependent phenotype of PC9DR cells. Interestingly, these cells share features with the previously characterized drug-tolerant expanded persisters (20). Taken together, the isolation of PC9DR cells unveiled a novel interim state between DTPs and cells stably resistant to TKIs (see model in Fig. 7D). According to our model, PC9-DTPs undergo growth arrest while PC9DR cells adopt EMT markers. However, in the absence of secondary EGFR mutations, PC9DR cells cannot transform to a permanent TKI-resistant state, which is irreversibly rewired as a result of drug-induced adaptive mutability (23). Importantly, however, we still need direct demonstration that DTPs actually evolve in patients and they give rise to full resisters. Likewise, it remains unclear whether DTPs can reliably reflect clinical resistance, hence permit development of effective resistance-preventing therapies.

In analogy to the proposed triphasic process conferring irreversible TKI resistance, bacterial cultures, especially biofilms, often display either phenotypic or genetic resistance to antibiotic agents (51, 52). Multiple mechanisms confer tolerance of biofilms to antibiotics (phenotypic resistance), and this causes both infection persistence and predisposition to resistance (genetic resistance; ref. 2). The phenotypic resistance is often controlled by either the environment, including aerobic and planktonic growth conditions, or by slowly dividing bacteria showing diminished susceptibility to antibiotics (persisters). In conclusion, both cancer cells and bacterial populations likely develop dynamic survival strategies permitting individual cells to reversibly assume drug-tolerant states. The latter protect from stressful conditions and predispose to mutation-based permanent resistance. Predictably, understanding the interplay between phenotypic and genetic resistance, as well as resolving the influence of the tumor microenvironment, will permit optimal clinical applications of kinase inhibitors.

Y. Yarden reports grants from the Miriam and Sheldon G. Adelson Medical Research Foundation (AMRF), European Research Council (ERC), Israel Cancer Research Fund (ICRF), and Israel Science Foundation (ISF) during the conduct of the study. No disclosures were reported by the other authors.

Y. Haga: Conceptualization, data curation, validation, investigation, visualization, writing–original draft, writing–review and editing. I. Marrocco: Data curation, validation, investigation, visualization, writing–review and editing. A. Noronha: Data curation. M.L. Uribe: Data curation. N.B. Nataraj: Data curation. A. Sekar: Data curation. D. Drago-Garcia: Data curation, formal analysis. S. Borgoni: Data curation. M. Lindzen: Data curation. S. Giri: Data curation. S. Wiemann: Writing–review and editing. Y. Tsutsumi: Writing–review and editing. Y. Yarden: Conceptualization, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.

The authors thank Gilgi Friedlander and Michael Gershovis for WES analyses. This work was performed in the Marvin Tanner Laboratory for Research on Cancer. Y. Yarden is the incumbent of the Harold and Zelda Goldenberg Professorial Chair in Molecular Cell Biology. This study was supported by Grants-in-Aid for Scientific Research from the Japan Society for the Promotion of Science (no. 18J21507), the Israel Science Foundation (ISF), the Israel Cancer Research Fund (ICRF), the European Research Council (ERC), and the Miriam and Sheldon G. Adelson Medical Research Foundation (AMRF).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.