CXCR7 Reactivates ERK Signaling to Promote Resistance to EGFR Kinase Inhibitors in NSCLC

Non–small cell lung cancers (NSCLC) with acquired EGFR TKI resistance that present with histologic transformations including epithelial-to-mesenchymal transition (EMT) have been identified in recent clinical studies (1–4) and remain as a significant clinical problem. In vitro, NSCLC cells with an EMT phenotype are de novo resistant to EGFR TKIs despite the overexpression of mutant EGFR (5–11). EMT is a physiologic program required in development, and is implicated in tumor progression and metastasis (12). Cancer cells presenting with a canonical EMT phenotype are poorly differentiated and often refractory to chemotherapy (13, 14). For this subset of EGFR TKI–resistant tumors, there are no specific targeted therapies and patients are limited to the conventional chemotherapeutic agents, radiation, epigenetic modifiers, or the mixing of targeted agents (15–17).

We reported that acquired resistance to first-generation EGFR TKIs with an EMT phenotype is a TGFβ-mediated process in HCC4006 EGFR-mutant cells that can be blocked with combined inhibition of EGFR and the TGFβ receptor (18). However, the cotreatment failed to prevent acquired EGFR TKI resistance due to an increased emergence of the EGFR T790M allele compared with cells treated with TKI alone (18). Our finding underscores the difficulty in suppressing the EGFR TKI–acquired resistance in NSCLC cells lines harboring EGFR kinase domain mutations as intratumoral heterogeneity gives rise to divergent resistance mechanisms in response to treatment. Furthermore, the clinical availability of third-generation EGFR TKIs including osimertinib (AZD9291) that overcome the T790M mutation in patients with NSCLC increases the prevalence of resistance cases with histologic transformation, acquired KRAS mutation, gene fusions, or an EGFR C797S mutation (19).

To date, little is known about the oncogenic drivers in EGFR-mutant cells with acquired resistance with EMT. Understanding the mechanisms of resistance underlying EMT may help in developing treatment strategies for this subset of resistant NSCLC. Prior in vitro studies have identified that the receptor tyrosine kinase AXL is frequently overexpressed in EGFR TKI–resistant NSCLC cell lines with an EMT phenotype (10, 17). AXL inhibition has been shown to sensitize this population to antimitotic agents but not to EGFR TKIs (17). This result suggests that sensitizing the resistant cells with EMT could potentially be difficult, and that a more thorough understanding of the molecular mechanisms by which the inhibition of mutant EGFR in NSCLC cells promotes EMT is required. Consequently, we decided to explore therapeutic targets beyond traditional TKIs and TGFβR in this subset of resistant cells with a hope to sensitize the resistant tumor to EGFR TKIs. We postulated that identifying and inhibiting an EMT-selective therapeutic target would prevent or reverse EMT and resistance to TKIs in EGFR-mutant cells.

In this study, we have employed an unbiased approach to find a molecular target that could compensate for the loss of EGFR signaling in NSCLC cell lines with acquired resistance to EGFR TKIs with an EMT phenotype. We have utilized NSCLC patient samples and mouse models of acquired EGFR TKI resistance to test whether our approach using these cell lines is instructive. Our studies identify a previously unrealized molecule that can be targeted to treat or prevent the emergence of EGFR TKI–resistant cancers with an EMT phenotype.

HCC827, HCC4006, and NCI-H1975 NSCLC cells were obtained from the ATCC and maintained as specified. To generate cell lines resistant to EGFR TKIs, cells were exposed to increasing concentrations of EGFR TKIs over 6 months in a manner similar to described previously (18); however, resulting resistant cell lines are polyclonal and not clones. For EGFR TKI-resistant HCC827 cells, clones were used as described previously (18). All resistant cells are able to proliferate normally in the presence of 10 μmol/L EGFR TKIs. Upon confirming resistance, cells were cultured without drugs and their resistance to TKI was examined periodically. All the cell lines including the drug-resistant and engineered cells were tested for the presence of Mycoplasma and the cell authenticity using the STR assay. Results for the STR assay are listed in the Supplementary Materials and Methods.

Live cells were counted using Countess (Thermo Fisher Scientific) and an equal number of live cells were seeded in each assay to compare cell growth kinetics using the Cell Counting Kit-8 (CCK-8) colorimetric assay (Dojindo) as described previously (18). The results were analyzed and graphed using Prism (GraphPad Software). For the crystal violet cell viability assays, cells were seeded in plates and grown to 70% confluence followed by drug treatments for the indicated times. Supernatant was removed and replaced with fixing/staining solution. Fixing/staining solution was removed, and plates were washed in dH₂O, allowed to dry, and scanned for imaging.

Lysate preparation and Western blotting were performed as described previously (18). A list of antibodies is listed in the Supplementary Materials and Methods.

cDNAs were cloned into pLX304 lentiviral expression vectors (Addgene) or pBabe (Addgene). pLKO.1 shRNA lentiviral vector construct designed by the RNAi Consortium were used as previously described (18). Lentivirus coding for cDNA or shRNA were packaged in 293LTV cells (Cell Biolabs) and transduced to the target cells as described previously (18). cDNA/shRNA/siRNA/sgRNA source, sequences, and RNA depletion procedures are provided in the Supplementary Materials and Methods.

RNA from cell lines and fresh frozen tumor or normal lung tissues was extracted and purified using QIAshredder and RNeasy Plus (Qiagen) according to the manufacturer's instructions. DNA was extracted using the DNeasy Blood and Tissue Kit (Qiagen). Samples were quantified using NanoDrop ND-2000 Spectrophotometer followed by Qubit (Thermo Fisher Scientific).

Total RNA was converted to cDNA using a High-Capacity RNA-to-cDNA Kit (Thermo Fisher Scientific). Assays were conducted using TaqMan Gene Expression Master Mix and Gene Expression Assays (Thermo Fisher Scientific) for the given targets with 50 ng of cDNA template. Human β-D-glucuronidase (GUSB) was used as an endogenous control. Samples were run in triplicate using a QuantStudio 6 Flex Real-Time PCR System and analyzed via comparative ΔΔC_t method. Probes used are listed in the Supplementary Materials and Methods.

Gene expression profiling methods are detailed in the Supplementary Materials and Methods.

Adherent cells were removed from plates using Accutase (Thermo Fisher Scientific) and pooled. Apoptosis was assessed using an Annexin V-FLUOS Staining Kit (Sigma) according to the manufacturer's instructions. For the detection of the cell surface expression of CXCR7 and CXCR4, pelleted cells were incubated with fluorochrome-conjugated antibodies (see Supplementary Materials and Methods), and/or DAPI. After washing, cells were resuspended in buffer and samples were run using either an Amnis ImagestreamX Imaging Flow Cytometer or a BD LSRFortessa. Appropriate unstained and single-stained CXCR7, CXCR4, or DAPI-only samples were used for controls and compensation. Imagestream data were analyzed using IDEAS software (Amnis/EMD Millipore), while BD LSRFortessa data was analyzed with FlowJo software.

Details for Luminex-based multibead assays are available in the Supplementary Materials and Methods. Concentration of cytokines and chemokines (pg/mL) were log₂ transformed and relative amount for each chemokine and cytokines was displayed using a heatmap.

Cells were grown on Nunc Lab-Tek II Chamber Slides (Thermo Fisher Scientific) to 70% confluence. Following treatments, the cells were fixed in 4% formaldehyde, permeabilized in 0.1% Triton X-100, and washed in PBS. The cells were incubated with Image-iT FX (Thermo Fisher Scientific) and blocked prior to addition of anti-CXCR7 (Genetex) followed by incubation with goat anti-rabbit Alexa Fluor 546–conjugated secondary antibody and Prolong Gold Antifade Reagent with DAPI (Thermo Fisher Scientific). Samples were imaged using an Olympus BX53 Scope with Hamamatsu ORCA-Flash 4.0 LT Digital Camera and analyzed with CellSens Software.

All animal treatment studies were reviewed and approved by the Institutional Animal Care and Use Committee at the Dana-Farber Cancer Institute (Boston, MA). Human EGFR^{Del E746_A750/T790M} (TD)-inducible bitransgenic mice were generated and have been previously characterized (20). After continuous exposure to doxycycline diets for more than 8 weeks, bitransgenic mice were subjected to MRI to document the lung tumor burden. After initial imaging, mice were treated either with vehicle (10% 1-methyl-2-pyrrolidinone: 90% PEG-300) alone or WZ4002, the mutant-selective EGFR TKI with activity against T790M, at 25 mg/kg gavage daily.

This study was carried out in accordance with the Declaration of Helsinki including all amendments, and participating hospitals required Institutional Research Ethics Board approval. All patients provided their written informed consent prior to being included in the study and before study procedures, sampling, and analyses. Following routine clinical practice, patients’ data were retrospectively obtained from patients’ medical records. All 29 EGFR-mutant NSCLC patients who progressed on EGFR TKI therapy underwent second biopsy of their tumor at Hospital Clinico Universitario and Hospital Universitario La Fe in Valencia, Spain and were included in the study cohort. Samples used for analyses had to have both pre- and posttreatment tumor specimens available. Seven samples were excluded from the analysis due to the lack of sufficient materials for the IHC. The medical records were reviewed retrospectively by board-certified physicians to obtain clinical information under an Institutional Review Board–approved protocol. All relevant ethical regulations were followed. The mutational status of EGFR (both primary and secondary mutations) was determined by the institutions as part of their diagnostic routine. Detailed procedures are available in Supplementary Materials and Methods.

In mice studies, the left lung of each mouse was dissected and snap-frozen for biochemical analysis after sacrifice. The right lung was inflated with buffered 10% formalin for 10 minutes and fixed in 10% formalin overnight at room temperature. The specimen was washed once in PBS, placed in 70% ethanol, and embedded in paraffin, from which 5-μm sections were generated. All biopsy specimens from patients with NSCLC were stained with hematoxylin and eosin, Nkx2.1, and CXCR7 and reviewed by the board-certified pathologists in the Department of Pathology at Hospital Quirónsalud (Valencia, Spain). Detailed procedures and the list of antibodies used are available in the Supplementary Materials and Methods.

Unless otherwise stated, comparisons of statistical significance were performed using ANOVA and Tukey multiple comparison test or t test where applicable. P < 0.05 was considered statistically significant.

Studies show that the EMT phenotype is associated with a subset of EGFR TKI–resistant tumors (2, 10, 21). We hypothesized that an EMT-specific signaling molecule promotes persistent activation of downstream signaling and blocks the apoptosis or impaired cellular proliferation that is normally mediated by EGFR inhibition.

To test this hypothesis, we undertook an unbiased approach by inducing EMT in EGFR-mutated HCC827 cells by TGFβ1 stimulation (18), depleting E-cadherin by shRNA, or ectopically expressing C-terminus V5-tagged SNAIL (SNAIL-V5). The induction of EMT was confirmed by immunoblot with canonical EMT markers, demonstrating the depletion of an epithelial marker, E-cadherin (CDH1), and the increase of mesenchymal markers, N-cadherin (CDH2), and vimentin (Fig. 1A). Gene set enrichment analyses (GSEA) confirmed the induction of an EMT phenotype and predicted gefitinib resistance (Supplementary Fig. S1A). As we reported previously (18), the EMT induction was sufficient to promote gefitinib resistance (Fig. 1B; Supplementary Fig. S1B) while EGFR and ERBB3 expressions were decreased (Supplementary Fig. S1C). As oncogenic drivers used to bypass EGFR signaling are often overexpressed, we identified 68 genes significantly upregulated at the mRNA level (FDR < 0.01, fold change > 3) in all HCC827 cells with the EMT induction (Fig. 1C; Supplementary Table S1). Among the mRNAs, typical G protein-coupled receptor (GPCR), C-X-C chemokine receptor 7 (CXCR7) was significantly overexpressed. (Supplementary Table S1). Consistent with this result, TaqMan RT-PCR (Fig. 1D) and Western blot (Fig. 1E) confirmed the CXCR7 overexpression, which was TGFβ1 dependent as previously reported (22). ImageStream revealed that CXCR7 is expressed on the surface and in the cytosol of mesenchymal HCC827 cells (Supplementary Fig. S1D).

The results in NSCLC cell lines prompted us to analyze consecutive formalin-fixed paraffin-embedded (FFPE) tissue samples from 29 advanced stage EGFR-mutant lung adenocarcinoma patients who radiographically progressed on EGFR TKI treatment for the presence of a T790M secondary mutation and the overexpression of CXCR7. Overall, 12 of the 22 posttreatment specimens (54.5%) harbored a T790M mutation. Of those who had sufficient and evaluable FFPE tissues (Fig. 2A), board-certified pathologists scored IHC samples for the expression of CXCR7. Eleven of the 22 (50%) patients showed increased CXCR7 protein expression at the time of disease progression; of these 11 patients, 7 also harbored a T790M secondary mutation. However, we were unable to find correlations between CXCR7 expression and T790M status or progression-free survival (PFS). Interestingly, 8 of the 22 posttreatment specimens (36.4%) showed decreased CXCR7 expression at the time of disease progression.

We found significant correlation between high CXCR7 expression and poor overall survival in very few GEO datasets, most of them composed of resectable stages; GSE3141 (P = 0.0027, n = 111), GSE13213 (P = 0.016, n = 117), and GSE30219 (P = 0.0001, n = 307, Supplementary Fig. S2A) using PROGgeneV2 (23, 24). In contrast, no significant correlation was found in the The Cancer Genome Atlas lung cancer dataset. To extend these results to an unbiased and more homogeneous manner, we also assessed the impact of CXCR7 overexpression on PFS in our own cohort of patients with surgically resected stage I/II lung adenocarcinoma, comparing normal adjacent tissue versus tumor in each patient by TaqMan RT-PCR. All patients (n = 55) underwent surgery prior to any therapies and had thorough clinical annotations. The result indicates that high CXCR7 expression levels significantly (P = 0.0127) predict poorer PFS in patients with lung adenocarcinoma (Fig. 2B).

Genetically engineered mouse models (GEMM) of NSCLC harboring EGFR^L858R/T790M (TL) or EGFR^{Del E746_A750/T790M} (TD) mutation have been used extensively to develop therapeutics to overcome EGFR TKI resistance (18, 20, 25–29). WZ4002, an irreversible EGFR TKI, is effective in TL and TD GEMMs over a 2- to 6-week treatment; however, prolonged treatment resulted in the development of acquired resistance with an increase in tumor volume due to ERK reactivation (25, 26). To investigate whether the WZ4002-resistant tumors overexpress CXCR7, we treated TD tumors for 12 weeks with WZ4002 (Supplementary Fig. S2B). We took serial MRIs and measured tumor volumes (30) (Supplementary Fig. S2C and S2D). We used a threshold of 30% decrease in tumor volume to define partial response and a 20% increase in tumor size as a progressive disease based on our previous studies (30, 31). At the time of resistance (12 weeks), the expression of CXCR7 was increased compared with tumors from untreated or 2-week treatment mice (Supplementary Fig. S2E). While no change in epithelial marker, E-cadherin, mesenchymal markers, both vimentin and N-cadherin, increased in WZ4002-resistant tumors compared with untreated or 2-week–treated tumors (Supplementary Fig. S2F).

To investigate EGFR TKI–induced EMT, several EGFR-mutant cell lines were grown in increasing concentrations of EGFR TKIs (Fig. 3A; Supplementary Fig. S3A–S3C; refs. 18, 27). CXCR7 was overexpressed in EGFR TKI–resistant mesenchymal HCC4006Ge-R or HCC4006O-R cells (heretofore Ge-R and O-R, respectively) derived by exposing EGFR TKI-sensitive, EGFR-mutated, epithelial HCC4006 cells to increasing concentrations of gefitinib or osimertinib, respectively (Fig. 3A). CXCR7 was also overexpressed in a mesenchymal subset of HCC827 cells grown resistant to erlotinib (ERC4) and a mesenchymal subset of NCI-H1975 grown resistant to an irreversible EGFR TKI, CL-387,785 (NCI-H1975CL-R).

CXCR7 could homodimerize or heterodimerize with CXCR4 upon CXCL12 stimulation (32) to activate PI3K and MAPK pathways (33); however, the CXCR7 signaling mechanisms in EGFR-mutant NSCLC remain elusive (Fig. 3B). EGFR-mutant cell lines are instructive in delineating mechanisms of EGFR TKI–acquired resistance that have been confirmed to exist in clinical samples (9, 34–39). We sought to analyze the acquired EGFR TKI–resistant cells with an EMT phenotype for the presence of canonical CXCR7 signaling pathways.

We reported that HCC4006 cells consist of a heterogeneous population that can be grown to develop resistance to gefitinib with an EMT phenotype (Ge-R) or with a T790M secondary mutation if concurrently grown in 1 μmol/L TGFβ receptor inhibitor, SB431542 (HCC4006GeSB-R, heretofore GeSB-R; ref. 18). First, we used gene expression profiling followed by GSEA (40, 41). The GSEA revealed significant enrichment of gefitinib-resistant, mesenchymal, and TGFβ gene signatures in Ge-R cells but not in GeSB-R cells (Supplementary Fig. S3D). Comparative marker selection analysis identified a marked increase in the expression of CXCR7 and its ligand CXCL12 (Fig. 3C). In contrast, β-arrestin and CXCR4 expression did not meet the criteria for a significant increase in mesenchymal Ge-R cells (FDR<0.05, FC>2).

Chemokine or cytokine binding to GPCRs is a prerequisite for receptor activation and it is known that CXCL12 binds to CXCR7 with increased affinity compared to CXCR4, and CXCR7 binds CXCL12 with higher affinity than CXCL11 (33, 42–44). To identify the chemokines and cytokines in the supernatants from gefitinib-resistant HCC4006 cells, we used a Luminex-based multiplex assay (Fig. 3D). CXCL12, CCL7, and CCL8 were uniquely found in the supernatants from mesenchymal Ge-R but not in epithelial HCC4006 and the T790M-enriched GeSB-R. Of note, little CXCL11 was detected in the supernatant from the mesenchymal Ge-R cells.

As the CXCR7 activation by ligand occurs on the cell surface, we quantified surface expression of CXCR7 and CXCR4 in HCC4006, Ge-R and GeSB-R cells with or without the depletion of CXCR7 using validated shRNA (Supplementary Fig. S3E) using fluorescently labeled anti-CXCR7 or anti-CXCR4 antibodies in FACS analyses (Fig. 3E). CXCR7 was markedly overexpressed on the surface of mesenchymal Ge-R cells and lentiviral shRNA significantly depleted the expression. Modest surface CXCR7 expression was found in HCC4006 or GeSB-R thus CXCR7 shRNA did not substantially change the surface expression. Of note, the depletion of CXCR7 promoted the decrease in CXCR4 expression at protein (Fig. 3E; Supplementary Fig. S3F) and mRNA (Supplementary Fig. S3G) levels.

Because relative quantity between CXCR7 and CXCR4 cannot be assessed by the cell surface staining using antibodies with dissimilar epitope reactivity, the quantity of CXCR7 in Ge-R cells was calculated using TaqMan RT-PCR, with expression being significantly (P < 0.0001) higher than CXCR4 when normalized with endogenous control GUSB (Fig. 3F).

In summary, the mesenchymal Ge-R overexpresses CXCR7 at a considerably higher level than CXCR4, suggesting that the CXCR7 homodimer might be functionally more relevant in TKI-resistant cells. The overexpression of CXCL12 that binds tightly to CXCR7 homodimers is also unique in the mesenchymal Ge-R cells.

Unlike CXCR4 that predominantly transduces signals through the conventional heterotrimeric G-protein, Gαi, CXCR7 is an atypical GPCR that transduces signals through β-arrestin upon activation by CXCL12 (Fig. 4A; refs. 32, 33, 43). While CXCR7 homodimers are likely more prevalent than CXCR4-CXCR7 heterodimers in Ge-R cells (Fig. 3F), it is unclear whether the CXCR7 homodimers compensate for the loss of the EGFR signal. Lentiviral shRNAs against nontarget, CXCR4, or CXCR7 were transduced to Ge-R cells and the decrease in the surface expression of CXCR4 or CXCR7 was confirmed (Supplementary Fig. S4A). Ge-R cells constitutively overexpressing the shRNAs were incubated with DMSO or osimertinib (Fig. 4B). As expected, CXCR4 depletion did not attenuate the phosphorylation of AKT (pAKT^S437) and ERK (pERK^T202/Y204), surrogate markers for PI3K and MAPK pathways, respectively. Interestingly, the phosphorylation of ERK but not AKT was markedly decreased by the depletion of CXCR7 following 500 nmol/L osimertinib exposure (Fig. 4B; Supplementary Fig. S4B). Similarly, combined CXCR7 depletion and EGFR inhibition in mesenchymal HCC827 cells ectopically expressing SNAIL-V5 markedly reduced ERK phosphorylation and the expression of mesenchymal markers, vimentin and N-cadherin (Supplementary Fig. S4C, lane 4).

To investigate whether CXCL12 overexpressed in Ge-R cells (Fig. 3D) activates CXCR7 establishing an autocrine loop, Ge-R cells constitutively overexpressing lentiviral shRNA against nontarget control or CXCR7 were stimulated with CXCL12 (SDF1α) in the presence of gefitinib. In the control Ge-R cells that overexpress CXCR7 (ΔNT), gefitinib treatment reduced but maintained, phosphorylation of ERK (Fig. 4C, lane 2). In the presence of SDF1α, phosphorylation of ERK increased upon gefitinib treatment in the Ge-R control cells (Fig. 4C, lane 3). In contrast, ERK phosphorylation in Ge-R cells lacking CXCR7 was not maintained to the same level nor increased upon SDF1α addition (Fig. 4C, lanes 5 and 6).

While CXCR7 could heterodimerize with CXCR4 or other GPCRs, or even with EGFR to activate PI3K and MAPK pathways (32, 33, 45), CXCR7 itself is an atypical GPCR that associates with β-arrestin 2 upon activation by either CXCL11 or CXCL12 (Fig. 4A; refs. 33, 43, 46). To delineate whether CXCR7 regulates PI3K and MAPK pathways, we tested whether the depletion of β-arrestins 1/2 would inhibit either signaling pathway. While β-arrestin 2 is known to preferentially associate with CXCR7, β-arrestin 1 also binds CXCR7 upon CXCL12 stimulation (33). Therefore, we transfected HCC4006 parental cells, Ge-R cells, or O-R cells with siRNA for a scrambled sequence or the combination of siRNAs against β-arrestins 1 and 2 (Fig. 4D). Osimertinib alone was sufficient to dephosphorylate AKT and ERK in parental HCC4006 with or without depleting β-arrestins 1/2; however, osimertinib failed to dephosphorylate AKT and ERK in Ge-R and O-R cells if β-arrestins 1/2 are expressed (Fig. 4D, Ge-R and O-R, lane 2). In contrast, osimertinib-mediated EGFR inhibition dephosphorylated ERK but not much AKT in the Ge-R and O-R cells with depleted β-arrestins 1/2, even though the siRNAs only decreased β-arrestins 1/2 protein levels by roughly half (Fig. 4D, Ge-R and O-R lane 4). In IHC samples from the GEMM model of WZ4002 resistance, β-arrestins 1/2 were ubiquitously expressed in the course of treatment (Supplementary Fig. S4D).

Interestingly, a long-term depletion of CXCR7 in the mesenchymal Ge-R cells diminished the expression of canonical mesenchymal markers, vimentin, and CD44 but not N-cadherin, while increasing the expression of an epithelial marker, E-cadherin (Fig. 4E, lane 4). While EGFR inhibition with gefitinib failed to suppress ERK phosphorylation in Ge-R cells transduced with lentivirus coding for shRNA targeting nontarget (ΔNT) sequence (Supplementary Fig. S4E, lane 6), CXCR7 depletion in Ge-R cells, followed by gefitinib treatment resulted in profound dephosphorylation of EGFR and ERK (Supplementary Fig. S4E, lane 8). Furthermore, the CXCR7 depletion resulted in a marked increase in apoptosis following exposure to osimertinib 48 hours (Fig. 4F; Supplementary Fig. S4F).

In summary, EGFR inhibition combined with CXCR7 depletion in the mesenchymal EGFR-mutant cells attenuated β-arrestins 1/2-mediated ERK phosphorylation in the HCC4006Ge-R and O-R cells. Exposure of HCC4006Ge-R cells to CXCR7 ligand, CXCL12 (SDF1α), resulted in increased ERK phosphorylation.

We next investigated whether ectopic expression of CXCR7 in EGFR-mutant cells promotes phosphorylation of ERK and EGFR TKI resistance. Lentivirus coding for V5-tagged CXCR7 (CXCR7-V5) or LacZ-V5 were transduced to HCC4006 cells and surface expression of the receptor was confirmed (Fig. 5A; Supplementary Fig. S5A) and by immunoblot with anti-V5 antibody (Fig. 5B). The transduced cells were treated with 1 μmol/L osimertinib for the days indicated. The appearance of heavier bands that represent posttranslationally modified CXCR7-V5 suggests but not proves internalization and activation of the CXCR7-V5 (Fig. 5B, lanes 2, 4, and 6). The phosphorylation of ERK was maintained in the osimertinib-treated cells ectopically overexpressing CXCR7-V5 at days 7 and 14 (Fig. 5B, lanes 4 and 6). Interestingly, increased expression of mesenchymal markers, N-cadherin, CD44, and vimentin was found in the cells ectopically expressing CXCR7-V5 chronically exposed to osimertinib but not in the control cells expressing LacZ-V5 (Fig. 5C). Nonetheless, the increased ERK activity and the induction of mesenchymal markers in the cells ectopically expressing CXCR7-V5 coincided with more surviving cells under 7 and 14 days osimertinib treatment (Fig. 5D and E).

In the WZ4002-resistant EGFR TD GEMM tumors with increased CXCR7 expression (Supplementary Fig. S2E), both phospho-AKT (pAKT^S437) and phospho-ERK (pERK^T202/Y204) are significantly increased (Supplementary Fig. S5B). Because the depletion of CXCR7 or β-arrestins 1/2 suppressed MAPK signaling in our cell line models (Fig. 4), we hypothesized that CXCR7-activated MAPK signaling is necessary for the survival of mesenchymal Ge-R cells. To test the hypothesis, we first challenged the HCC4006 and Ge-R cells with EGFR, MEK, or PI3-K inhibitors for 72 hours (Fig. 5F). The treatment of HCC4006 with EGFR TKI induced cleaved PARP while the same treatment in Ge-R cells did not. The treatment of HCC4006 cells with MEK inhibitor promoted PARP cleavage; however, the resulting apoptosis was less than that induced by EGFR TKI (Fig. 5G; Supplementary Fig. S5C). Of note, the inhibition of PI3K alone was not sufficient to induce cleaved-PARP or cell death in both HCC4006 and Ge-R cells. Cell viability assays confirmed these results (Supplementary Fig. S5D). Exposure of mesenchymal Ge-R or O-R cells to MEK inhibitor was the only treatment that completely eliminated the cells in the long-term cell viability assay (Supplementary Fig. S5E). Similarly, HCC827 cells ectopically expressing SNAIL-V5 with CXCR7 overexpression (Fig. 1A) remained sensitive to MEK inhibitor (Supplementary Fig. S5F).

It has not been demonstrated whether ectopic activation of the MAPK pathway is sufficient for survival. On the basis of our results that CXCR7 drives the MAPK pathway but not PI3K pathway (Figs. 4B and 5B), we wanted to investigate whether the CXCR7-activated MAPK pathway is sufficient to confer resistance in EGFR-mutant NSCLC. Constitutively active MEK, MEKDD, was ectopically expressed in EGFR TKI–sensitive HCC827. Because HCC827 cells have a propensity to develop EGFR TKI–resistant MET amplification (39), we performed apoptosis assays with or without the MET inhibitor crizotinib. In HCC827 cells constitutively overexpressing vector control, osimertinib treatment, regardless of the use of crizotinib, resulted in dephosphorylation of ERK and induction of Bim and cleaved PARP (Fig. 5H, lane 4). In contrast, osimertinib treatment of HCC827 cells constitutively overexpressing MEKDD did not diminish phosphorylation of ERK and induced little Bim and cleaved PARP (Fig. 5H, lane 8). This result was also confirmed with Annexin V apoptosis assays (Fig. 5I; Supplementary Fig. S5G).

In summary, ectopic overexpression of CXCR7-V5 in EGFR-mutant cells promoted survival under the inhibition of EGFR by activating the MAPK–ERK signaling that is directly linked to the survival of the EGFR TKI–resistant cells overexpressing CXCR7.

EGFR TKI-resistant cells with a secondary T790M EGFR mutation can both pre-exist and/or emerge from drug-tolerant cells called persisters (47). We and others reported that exposure of EGFR-mutant NSCLC cells (epithelial) to an EGFR TKI results in the emergence of persisters in the short term (16, 18, 47–49). Previous studies have shown that ERK is reactivated in the persisters and EGFR-MEK dual inhibition significantly delays their emergence (25, 26). We have reproduced ERK reactivation in HCC4006 and HCC827 cells (Fig. 6A) and realized that the ERK reactivation occurs in parallel with the increase in CXCR7 expression increase in the persister cells (Fig. 6B). The overexpression of shRNA targeting CXCR7 in HCC4006 and HCC827 cells, while incomplete, significantly impaired the induction of CXCR7 following 24 hr (HCC4006) or 72 hr (HCC827) osimertinib treatment (Fig. 6C). On the basis of our results that CXCR7 activated the MAPK pathway (Fig. 4) in long-term resistance, we hypothesized that CXCR7 reactivates the MAPK pathway in persister EGFR-mutant NSCLC cells.

We then treated EGFR TKI–sensitive HCC4006 cells with or without impaired CXCR7 expression with osimertinib to monitor ERK phosphorylation (Fig. 6D, top). While the reactivation of ERK was evident in the control cells with CXCR7 expression, it was significantly attenuated in the cells with CXCR7 depletion. We have also developed CXCR7-depleted mass culture of HCC4006 cell lines using CRISPR/Cas9-mediated biallelic disruption of the CXCR7 locus to obtain identical results (Fig. 6E; Supplementary Fig. S6A). Similar results were obtained in HCC827 cells (Supplementary Fig. S6B), PC9 cells (Supplementary Fig. S6C), NCI-H1975 cells (Supplementary Fig. S6D). In HCC4006 cells ectopically expressing CXCR7-V5, the ERK reactivation in drug tolerant cells was more profound than that in the control HCC4006 cells expressing LacZ-V5 (Fig. 6D, bottom). Similarly, HCC827 cells ectopically overexpressing CXCR7-V5 (Supplementary Fig. S6E) resulted in the increased maintenance of phosphorylated ERK upon EGFR inhibition even after 24 hours of osimertinib/crizotinib combination treatment.

Motivated by these findings, we sought to investigate whether CXCR7 depletion potentiates efficient cell death in EGFR TKI–treated NSCLC cells. HCC4006 cells constitutively overexpressing shRNA against nontarget or CXCR7 were treated with osimertinib and apoptotic markers were assessed (Fig. 6F). Annexin V assay demonstrated significantly (P < 0.01) increased apoptosis by osimertinib treatment in HCC4006 cells (Fig. 6G, left; Supplementary Fig. S6F) or HCC827 cells (Fig. 6G, right; Supplementary Fig. S6G) where normal EGFR TKI–induced CXCR7 production is attenuated by shRNA depletion.

In summary, our results demonstrate that CXCR7 expression is an early event following EGFR inhibition and depletion of CXCR7 hampers the efficient reactivation of ERK, resulting in increased apoptosis. The expression of CXCR7 following EGFR inhibition was a ubiquitous event among the EGFR-mutant NSCLC cells that we tested. Ectopic expression of CXCR7 in HCC4006 cells potentiated the reactivation of ERK following initial EGFR inhibition.

While the GPCR family represents the largest family of membrane receptors, their therapeutic values have not been extensively explored in NSCLC. The upregulation of triheteromeric GPCR, adenosine A2a receptor, in gefitinib-resistant NSCLC cell models has been reported (50); however, its signaling mechanisms and therapeutic potential remain unknown. Our analyses of the EMT-promoted EGFR-mutant NSCLC cell line models revealed that GPCR, CXCR7, is ubiquitously overexpressed in acquired EGFR TKI–resistant cell line models with an EMT phenotype (Figs. 1E and 3A). Coincidentally, 50% of specimens from EGFR mutation-positive patients with NSCLC who progressed on EGFR TKI treatment exhibit increased CXCR7 expression (Fig. 2A). Furthermore, CXCR7 expression correlates with poor progression-free survival in treatment-naïve, early-stage NSCLC patients regardless of their EGFR or any other oncogene (i.e., KRAS) statuses, suggesting CXCR7 activity is important in tumor progression and relapse (Fig. 2B; Supplementary Fig. S2A). However, the correlation between CXCR7 expression and T790M status or PFS was not evident in our EGFR-mutant patient cohort (Fig. 2A). This discrepancy is due to the differences in disease stages and sample size of the specimens used for the IHC studies in Fig. 2A compared with survival analyses displayed in Fig. 2B and Supplementary Fig. S2A. Nonetheless, the overexpression of CXCR7 was also evident in the tumors from transgenic murine models with EGFR TKI–acquired resistance (Supplementary Fig. S2) supporting its implications in the acquired resistance.

These clinical samples indicate CXCR7 overexpression occurs with or without T790M secondary mutations in tumors from patients with NSCLC who progressed on EGFR TKI treatment samples (Fig. 2A). From the limited amount of patient specimens, it was unclear whether the CXCR7 overexpression and T790M emerge in the same subpopulation or in distinct subsets. However, our EGFR TKI–acquired resistance study using HCC4006 cells revealed that the cells with T790M or a mesenchymal phenotype exist as different populations and the treatments dictate the emergence of dominant resistance mechanisms (18). Thus, the types of drug treatment or environmental cues from the tumor microenvironment might impact the dominant EGFR TKI resistance mechanisms. Further studies are necessary in a larger cohort of patient samples to correlate basal CXCR7 expression with the TKI response, or to understand why tumors with high expression of CXCR7 prior to EGFR TKI therapy showed a decrease in expression following the treatment.

CXCR7 is an atypical GPCR and the CXCR7 homodimer associates with β-arrestin2 instead of the conventional heterotrimeric G-protein, Gαi, upon activation by either CXCL11 or CXCL12 (43). Accumulating evidence suggests that CXCR7 plays pivotal roles in metastasis (22, 51, 52), angiogenesis (53, 54), proliferation of cancer cells (33, 55, 56), and de novo drug resistance (57, 58). CXCR7 overexpression has recently been linked to acquired enzalutamide resistant prostate cancer (59) by activating MAPK signaling pathway in a ligand-independent fashion (60). While CXCR7 is linked to TGFβ1-induced EMT in lung cancer cell line models (61), its roles in acquired drug resistance, and in particular, its therapeutic value in lung cancer has not been fully elucidated.

Our results demonstrate that the CXCL12–CXCR7 axis is necessary for the activation of the MAPK signaling (Fig. 4D and E; Supplementary Fig. S4C) and ectopic expression of CXCR7 is sufficient to increase the MAPK signaling when the CXCR7-overexpressing cells were exposed to osimertinib (Fig. 5A and B). The ectopic CXCR7 overexpression also increased the expression of mesenchymal markers (Fig. 5C) and promoted resistance to osimertinib (Fig. 5D and E). Ge-R and O-R cells with CXCR7 overexpression (Fig. 3A) are exquisitely sensitive to MEK inhibitors (Fig. 5F and G) and aberrant activation of MEK in otherwise EGFR TKI–sensitive HCC4006 is sufficient for conferring resistance to EGFR TKI (Fig. 5H and I). These results cumulatively indicate CXCR7 activates the MAPK signaling to promote EMT and EGFR TKI resistance. CXCR7 could heterodimerize with other GPCRs, or even bind to EGFR (32, 45) to activate MAPK pathways (33); however, the activation of the MAPK signaling by CXCR7 occurs under the EGFR inhibition with TKIs. Accordingly, this resistance mechanism resembles the EGFR TKI acquired resistance with MET amplification (39, 62), in which activation of MET sustains downstream signaling despite inhibition of oncogenic mutant EGFR.

Several questions remain unanswered regarding the regulation of EMT by CXCR7 (Fig. 5C). While CXCR4 is known to promote EMT (63–65), the depletion of CXCR7 alone was sufficient to reverse the EMT phenotype in our model (Fig. 4E). While CXCR4 expression was significantly lower than CXCR7 (Fig. 3F), the depletion of CXCR7 itself attenuated the expression of CXCR4 (Fig. 3E, right). It is likely that the decreased expression of CXCR4 is due to the mesenchymal-to-epithelial transition following CXCR7 depletion; however, additional studies are required to dissect the CXCR7–CXCR4 interaction in this model. In addition, further study is required to investigate whether CXCR7-driven MAPK signaling exclusively activates EMT promoting transcription factors or other pathways are involved in the activation. We have shown that exposure of CXCR7-overexpressing Ge-R cells to CXCL12 increases ERK phosphorylation (Fig. 4C) and that these cells also express several other chemokines and cytokines known to activate CXCR7 (Fig. 3D). Additional experiments are required to elucidate whether specific chemokines preferentially promote EMT by activating CXCR7.

With the introduction of third generation EGFR TKIs that overcome the T790M mutation in patients with NSCLC (50%), the prevalence of resistance cases with pathologic transformation including EMT may rise. This underscores the need for new biomakers and improved therapeutic strategies for this subset of patients. Here, we demonstrated that cotargeting EGFR and CXCR7 may be an effective combination strategy against EGFR TKI–resistant cells with a mesenchymal phenotype. The depletion of CXCR7 not only resensitized the TKI resistant EGFR-mutant NSCLC with a mesenchymal phenotype to EGFR TKI (Fig. 4F), but also promoted partial mesenchymal-to-epithelial transition (Fig. 4E). These results support the unique concept that the combination of osimertinib and CXCR7 inhibitor might be particularly effective in patients with NSCLC who developed resistance to first-line EGFR TKIs, as a significant number of patients who received the treatment developed resistant tumors with both T790M and CXCR7 overexpression (Fig. 2). CXCR7 and EGFR dual inhibition can also be applied for the prevention of drug resistance, which offers broad clinical applicability, as our results demonstrate that the EGFR inhibition combined with CXCR7 depletion diminishes ERK reactivation (Fig. 6). We have previously reported that a surviving population of TKI-naïve EGFR-mutant NSCLC cells treated with EGFR TKIs, so-called persister cells, secrete TGFβ upon EGFR TKI treatment priming the surviving cells for EMT (18). Surprisingly, the overexpression of CXCR7 following EGFR TKI treatment was observed in persisting populations from 4 different EGFR-mutant NSCLC cell lines (Fig. 6). The depletion of CXCR7 in the persister cells suppressed ERK reactivation following EGFR inhibition and the EGFR–CXCR7 dual inhibition promoted a significant increase in cell death. Taken together, our study gives credence to the future development of potent CXCR7 inhibitors and provides guidance as to where CXCR7 inhibition should be evaluated, specifically either as initial therapy in EGFR TKI–näive patients or in those who have developed EGFR TKI resistance with a mesenchymal phenotype. The proposed concurrent inhibition of EGFR and CXCR7 may offer significant advantages over cotargeting EGFR and MEK, as CXCR7 overexpression seems to be specific to a subset of NSCLC cells compared with the more ubiquitously expressed MEK allowing for more selective targeting while attenuating toxicity.

J.A. Fernández-Coronado reports receiving a commercial research grant from Patológika Laboratorio. J. Alcácer reports receiving a commercial research grant from Patológika Laboratorio. J.A. Borgia reports receiving a commercial research grant from Biodesix and Epigenomics. P.A. Janne reports receiving a commercial research grant from AstraZeneca, Boehringer Ingeleim, Daiichi Sankyo, Astellas, Takeda Oncology, Puma, and Eli Lilly; has ownership interest (including stock, patents, etc.) in Gatekeeper Pharmaceuticals, LOXO Oncology; is a consultant/advisory board member for AstraZeneca, Boehringer Ingelheim, Voronoi, Biocartis, Takeda, Daiichi Sankyo, Novartis, Eli Lilly, Mirati Therapeutics, Roche/Genentech, Pfizer, Merrimack Pharmaceuticals, Chugai Pharmaceuticals, Ariad, Ignyta, Loxo Oncology, and SFJ Pharmaceuticals; and has provided expert testimony for post-marketing royalties from DFCI-owned intellectual property on EGFR mutations licensed to Lab Corp. K.-K. Wong reports receiving a commercial research grant from Janssen Pharmaceuticals, Pfizer, Novartis, Mirati, BMS, AZ, Takeda, and Merck and has ownership interest (including stock, patents, etc.) in G1 Therapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J.H. Becker, J.M. Galbis-Caravajal, M.P. Kahle, O. Juan, J. Carretero, T. Shimamura

Development of methodology: J.H. Becker, I. Pulido, M.L. Rodríguez, A. Lafuente-Sanchis, A. Cremades, J.M. Galbis-Caravajal, K.-K. Wong, T. Shimamura

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.H. Becker, Y. Gao, M. Soucheray, E. Kikuchi, M.L. Rodríguez, A. Lafuente-Sanchis, M. Aupí, J.A. Fernández-Coronado, P. Martín-Martorell, A. Cremades, J. Alcácer, C.L. Christensen, P. Simms, A. Hess, H. Asahina, M.P. Kahle, A. Lahoz, A. Insa, O. Juan, P.A. Jänne, K.-K. Wong, J. Carretero, T. Shimamura

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.H. Becker, Y. Gao, M. Soucheray, M.L. Rodríguez, A. Lafuente-Sanchis, M. Aupí, J.A. Fernández-Coronado, J.M. Galbis-Caravajal, J. Alcácer, P. Simms, F. Al-Shahrour, J.A. Borgia, O. Juan, J. Carretero, T. Shimamura

Writing, review, and/or revision of the manuscript: J.H. Becker, Y. Gao, M. Soucheray, E. Kikuchi, M.L. Rodríguez, P. Martín-Martorell, A. Cremades, H. Asahina, M.P. Kahle, J.A. Borgia, A. Insa, O. Juan, P.A. Jänne, J. Carretero, T. Shimamura

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.H. Becker, R. Gandhi, O. Juan, T. Shimamura

Study supervision: J.H. Becker, O. Juan, K.-K. Wong, J. Carretero, T. Shimamura

We would like to acknowledge the INCLIVA Biobank (PT13/0010/0004) integrated in the Valencian Biobanking Network and the Spanish National Biobanks Network for its collaboration. This work is supported by American Cancer Society Illinois Division Basic Science Grant #254563 (to T. Shimamura), the American Cancer Society Research Scholar Grant 126638-RSG-14-229-01-TBG (to T. Shimamura), MINECO (SAF2013-49056 and SAF2017-85352 to J. Carretero), and Domingo Martinez Foundation (to J. Carretero), NIH grants CA140594, CA122794, CA163896, CA166480 (to K.-K. Wong), and CA114465, CA135257, and CA154303 (to P.A. Jänne).

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.