To investigate the effect of fibroblasts on various non–small cell lung cancer (NSCLC) cells, we cocultured nucleus-labeled KRAS-mutant, EML4-ALK fusion–positive, and EGFR-mutant NSCLC cells with CAFs or normal lung–associated fibroblasts (NAFs). CAFs were selected from a panel of patient-derived lung CAF lines (see Materials and Methods) based on having sufficient remaining cell culture life before onset of senescence. As expected, CAFs and NAFs both caused resistance to the mitogen-activated protein kinase kinase inhibitors trametinib and AZD8330 in KRAS-mutant A549 cells, with these CAFs exhibiting somewhat stronger effects than NAFs (Fig. 1A). Fibroblast effects on EML4-ALK fusion cell lines treated with the anaplastic lymphoma kinase (ALK)/MET inhibitor crizotinib, on the other hand, were variable: Both CAFs and NAFs induced resistance in H3122 cells and had more limited effects in H2228 cells, whereas only CAFs mildly sensitized STE-1 cells (fig. S1A). Similarly, whereas these CAFs showed a tendency toward protecting EGFR-mutant PC9 cells from the multitargeted SRC/Abelson kinase (ABL) inhibitor dasatinib, the same CAFs, in contrast to NAFs, mildly enhanced the response to the epidermal growth factor receptor (EGFR) inhibitor gefitinib (Fig. 1B). Furthermore, we observed pronounced sensitization of gefitinib-resistant PC9GR cells, which express the EGFR^T790M gatekeeper mutant, by these specific CAFs, but not NAFs, when treating with the second- or third-generation EGFR inhibitors afatinib or osimertinib, respectively (Fig. 1C). Given the differential CAF effects on cell viability depending on cell line and drug treatment, we confirmed that all fibroblasts expressed the common fibroblast markers α–smooth muscle actin (α-SMA) and fibroblast activation protein–α (FAP-α), as well as the mesenchymal marker vimentin, but did not express the epithelial and lung cancer markers E-cadherin and cytokeratin, respectively (Fig. 1D and fig. S1B). We also observed the expected mesenchymal, spindle-like morphology of the Moffitt-generated lung CAF cells (fig. S1C). Stromal effects on cancer growth can be mediated by both secreted factors and contact-dependent mechanisms (cell-cell or cell-ECM). To determine the relevance of paracrine effects, we examined the impact of CAF- or NAF-conditioned medium (CM) on the response of PC9GR cells to EGFR inhibition (Fig. 1E). Sensitization was evoked by CAF CM, but not by NAF CM, suggesting that potentially a large proportion of the sensitizing effect is the result of secreted proteins. Furthermore, caspase-3 and poly(adenosine 5′-diphosphate–ribose) polymerase 1 (PARP1) cleavage indicated strong drug-induced apoptosis of PC9GR cells by osimertinib in the presence of CAF CM, but not NAF CM (Fig. 1F). In summary, these findings suggest the presence of secreted CAF-specific cancer-suppressive factors and that the same CAFs, in a context-dependent manner, can cause either drug resistance or sensitization.

Because drug sensitization was only observed with CAFs, we aimed to determine the differences between CAFs and NAFs in an unbiased way. Global gene expression analysis of five fibroblast (IMR-90, WI38, MRC5, CAF7, and CAF12) and three NSCLC (H3122, PC9, and PC9GR) cell lines reflected the general differences between NSCLC cells and fibroblasts (fig. S2A), between lung CAFs and NAFs (Fig. 2A and fig. S2B), and between PC9 parental and PC9GR cells (fig. S2C) (see also data file S1). As expected, epithelial cancer cells displayed higher expression than fibroblasts of genes that encode cytokeratin epithelial markers but lower expression of genes that encode the common fibroblast markers α-SMA, FAP-α, and vimentin, which were similarly highly expressed across all fibroblasts (fig. S2D). Despite the similarities among fibroblasts, there were also notable differences between MRC5 NAFs and the two CAF cell lines (Fig. 2A and data file S1). Within the 172 genes that encode secreted proteins, multiple IGF-binding proteins (IGFBPs), such as IGFBP5, IGFBP6, and IGFBP7, were found to be significantly increased in CAFs, whereas IGF2 was decreased (Fig. 2B). Several IGFBPs were also decreased in PC9GR versus PC9 cells, a feature that has been reported for IGFBP3 to promote drug-resistance in wild-type EGFR-driven A431 skin carcinoma cells (Fig. 2C and data file S1) (26).

Considering that CM caused similar effects as physical coculture, we next investigated the differences between CAFs and NAFs regarding secreted proteins. Examination of CAF7 and MRC5 CM using cytokine array revealed the expected increased CAF secretion of cytokines associated with resistance, such as HGF and IL-6, but did not provide obvious candidates for sensitizing factors (fig. S3A). Therefore, we used an unbiased mass spectrometry (MS)–based secretomics approach. To prevent background signals from serum proteins, we collected serum-free CM (SFCM) from CAFs and NAFs. Subsequent high-resolution liquid chromatography–tandem MS (LC-MS/MS) analysis revealed multiple differences between the two cell types (data file S2). For example, consistent with the cytokine array results, IL-6 was secreted more from CAFs than MRC5 NAFs, albeit only at overall low levels (Fig. 2D). Some of the most abundant differentially secreted proteins by CAFs versus MRC5 were IGFBP5 and IGFBP6 (Fig. 2, D and E, and fig. S3B), corroborating the transcriptome analysis. IGFBPs are known to regulate IGF-mediated activation of IGF1R signaling, a pathway linked to EGFR inhibitor resistance that was also significantly enriched among the most abundant secreted proteins (top 2.5% of identified proteins), along with well-known ECM pathways containing collagen, collagen-binding proteins [for example, fibronectin, secreted protein acidic and rich in cysteine (SPARC), and biglycan], and proteoglycans such as lumican and decorin (fig. S3C and data file S3). Western blot analysis validated higher expression of several IGFBPs in CAFs than in MRC5 or cancer cells (Fig. 2F and fig. S9). In addition, although generally low in abundance relative to the IGFBPs, the stimulatory IGF1R ligands IGF1 and IGF2 were significantly less expressed in the CAF than in the NAF secretome (Fig. 2, D and E, and fig. S3B). Together, the above work shows that CAFs express and secrete higher levels of several IGFBPs, while at the same time expressing less IGFs, suggesting an overall inhibitory effect on IGF1R signaling in NSCLC cells.

Next, we evaluated the functional role of IGF1R pathway components that were differentially expressed between CAFs and NAFs regarding the sensitization effects seen with CAF CM. Because IGFBP5, IGFBP6, and IGFBP7 were expressed at higher levels in CAFs than in NAFs, we first treated PC9GR cells with osimertinib in the presence of recombinant human (rh) IGFBP5, IGFBP6, or IGFBP7. All three IGFBPs significantly enhanced the response of PC9GR cells to osimertinib (Fig. 3, A to C). In contrast, the addition of exogenous growth factors IGF1 or IGF2, which we had observed to be secreted from CAFs at lower levels than from NAFs and which would be expected to cause the opposite effects from IGFBPs, resulted in significantly increased cell viability in the presence of osimertinib (Fig. 3D). Last, we collected CM from CAFs in which IGFBP5 had been silenced by either a small interfering RNA (siRNA) pool or two individual siRNAs. Consistent with the results from treatment with recombinant IGFBP5, compared with nontargeting siRNA controls, all three CM preparations from CAFs in which IGFBP5 was knocked down caused a small but significant reduction in sensitization of PC9GR cells to osimertinib treatment (Fig. 3E). This was similarly the case upon IGFBP6 silencing (fig. S3D). These results suggest that the cumulative differences between multiple tumor-suppressive IGFBPs and tumor-promoting IGFs are critical for driving CAF CM–induced drug sensitization.

Given that gene expression and secretome analysis of CAFs versus NAFs identified differential expression of IGF1R signaling components, i.e., IGFBPs and IGF1/2, and that we were able to show that the relative expression of these components plays a role for CAF-induced drug sensitization, we next sought to identify the downstream signaling pathways in PC9GR cells that are affected by CAF and NAF CM. We first examined receptor tyrosine kinase (RTK) phosphorylation using phospho-RTK arrays because RTKs are direct effectors of paracrine growth factor signaling. As expected, pEGFR was strongly reduced by osimertinib regardless of CM presence (Fig. 4A and fig. S4, A and B). Consistent with the expected effects of IGFBPs, pIGF1R was decreased by CAF7 or CAF12 CM compared with MRC5 CM (Fig. 4A and fig. S4, A and C). To interrogate proteome-wide phosphotyrosine signaling in dimethyl sulfoxide (DMSO) or osimertinib-treated PC9GR cells grown in MRC5 or CAF7 CM, we undertook an unbiased tyrosine phosphoproteomics approach (data file S4). EGFR autophosphorylation sites including Tyr¹⁰⁹², Tyr¹¹⁷², and Tyr¹¹⁹⁷, as well as canonical downstream phosphosites (for example, SHC1 Tyr⁴²⁷, GAB1 Tyr⁶²⁷, and CBL Tyr⁷⁰⁰), were similarly decreased upon osimertinib treatment in both NAF and CAF CM (Fig. 4B and fig. S5, A and B), suggesting that the sensitization to osimertinib was not due to changes in direct EGFR signaling. Consistent with higher levels of IL-6 in the CAF7 secretome, signal transducer and activator of transcription 3 (STAT3) phosphorylation (Tyr⁷⁰⁵) was higher in cells treated with CAF7 CM than those treated with MRC5 CM, which, as a prosurvival signal, was unlikely to account for any drug sensitization effect (Fig. 4, B and C, and fig. S5C). However, although osimertinib treatment caused significantly increased phosphorylation of the IGF1R adaptor and substrate protein insulin receptor substrate 2 (IRS2) by both CM (Fig. 4B and fig. S5B), there was a tendency toward a lower degree of up-regulation by CAF7 CM compared with MRC5 CM (data file S4). IRS2 Tyr⁶³² was furthermore less phosphorylated with CAF7 CM under DMSO treatment (fig. S5C). Several of the identified IRS2 phosphotyrosine sites have been shown to be crucial for phosphatidylinositol 3-kinase/AKT signaling (27), implying a CAF7-dependent dampening of IGF1R signaling through AKT downstream of IRS2. Further comparison of the altered phosphosites revealed that extracellular signal–regulated kinase 1/2 (ERK1/2; Tyr^204/187) and focal adhesion kinase 1 (FAK1) (Tyr³⁹⁷) phosphorylation were also reduced by CAF7 CM compared with MRC5 CM upon osimertinib treatment (Fig. 4, C and D). Reduction of FAK (Tyr³⁹⁷) phosphorylation by CAF CM was also discernible by immunoblot (fig. S5D). This is in line with reports of IGFBP interaction with integrins, leading to subsequent dephosphorylation of FAK, which can signal through both ERK and AKT (28–30). In support of these findings, pathway analysis of phosphoproteins significantly altered by osimertinib with either CM showed enrichment of the ErbB, insulin, and focal adhesion pathways (fig. S5E and data file S5). Consistently, immunoblot analysis of early signaling effects showed that, although pEGFR was similarly decreased upon cell culture with all CM, pAKT and pERK1/2 were both further decreased in the presence of CAF CM compared with MRC5 CM (Fig. 4E and fig. S10). This concurs with a CAF-dependent decrease of both proliferation and survival signals, especially upon osimertinib treatment. Moreover, stimulation with IGF1 rescued AKT phosphorylation, but not ERK phosphorylation, supporting the idea that, although AKT is downstream of IGF1R, the CAF-induced change in ERK signaling is neither EGFR- nor IGF1R-driven, but possibly rather due to suppression of FAK signaling through integrin-IGFBP interactions. Collectively, these data suggest that compensatory survival signaling in PC9GR cells in response to osimertinib treatment is reduced by CAF CM through inhibition of IGF1R and FAK signaling.

To examine whether inhibition of IGF1R signaling using small molecules phenocopied CAF sensitization effects, we treated PC9GR cells with two different IGF1R tyrosine kinase inhibitors (TKIs), namely, GSK1838705A and linsitinib. Similar to CAF CM (Fig. 1E), the combination with either IGF1R TKI enhanced osimertinib activity (Fig. 5A and fig. S6A). Next, we tested the combination of osimertinib, linsitinib, and defactinib, a specific FAK inhibitor, for activity in PC9GR cells. We found that both defactinib and linsitinib increased osimertinib sensitivity, but the combination of all three had the strongest effect (Fig. 5B). Considering that the Food and Drug Administration (FDA)–approved ALK inhibitor ceritinib also inhibits IGF1R and FAK1 in the low- and mid-nanomolar range, respectively (31, 32), we tested the combination of ceritinib and osimertinib on PC9GR cells. This combination closely resembled the linsitinib, defactinib, and osimertinib combination, showing an even stronger effect on cell viability than IGF1R inhibition alone (Fig. 5C). We hypothesized that ceritinib’s polypharmacology targeting both IGF1R and FAK may explain this additional effect and found reduced FAK and ERK phosphorylation upon treatment of PC9GR cells with ceritinib versus linsitinib, whereas both drugs inhibited phosphorylation of IGF1R and AKT similarly (Fig. 5D and figs. S6, B to E, and S11). Likewise, the combination of ceritinib with an alternate EGFR inhibitor, nazartinib, decreased PC9GR cell viability (fig. S6F), confirming that this effect is not limited to osimertinib. The ceritinib and osimertinib combination also decreased cell viability in multiple other EGFR-driven cell lines, suggesting broader applicability (fig. S6, G to I). Long-term clonogenic growth assays furthermore revealed strong synergy, leading to almost complete elimination of PC9GR cells (Fig. 5E). Together, these data show that IGF1R and FAK inhibitors, by targeting the mechanisms modulated by CAFs, can sensitize EGFR-driven cancer cells to EGFR TKIs.

Osimertinib has been recently approved for the treatment of naïve EGFR-mutant lung cancer. We therefore aimed to determine the efficacy of combined targeting of EGFR, IGF1R, and FAK, the latter two of which we observed to mimic CAF effects, in such a first-line setting. Parental PC9 cells were treated with the combination of EGFR TKIs and various IGFBPs. Although PC9 cells are inherently more sensitive to EGFR inhibition by osimertinib and gefitinib, significant sensitization was observed by the addition of IGFBPs for multiple combinations (Fig. 6A). Furthermore, combination of osimertinib or nazartinib with ceritinib was synergistic in PC9 cells and pushed the viability curves both to the left and down to baseline (Fig. 6, B and C), which suggested the elimination of persister cells that could give rise to bona fide resistant clones in the future. To test this, we set up long-term, high-plating density live-cell imaging assays with final crystal violet readouts at either 14 or 28 days (Fig. 6D and fig. S7, A to C). Although 1 μM ceritinib alone had no effect, PC9 cell confluence initially decreased as expected with both EGFR inhibitors alone. Consistent with our short-term viability data, however, the combination with ceritinib was significantly more pronounced. Moreover, the single-drug–treated cells began to recover and grow out at about 21 days, whereas the EGFR TKI combinations with ceritinib eliminated essentially all cells through 28 days. Furthermore, in PC9 three-dimensional (3D) spheroids, osimertinib strongly reduced spheroid size and viability, but the ceritinib combination was strongly synergistic in decreasing spheroid cell viability (∆Bliss = 0.14) (fig. S7D). Last, although the individual contributions of the various involved CAF proteins remain to be determined, the combination of ceritinib with osimertinib more effectively inhibited tumor growth as determined by tumor volume than single-drug treatment in vivo using PC9 mouse xenografts (Fig. 6E). Consistently, the size and weight of excised tumors were significantly reduced by the drug combination compared with osimertinib treatment alone (Fig. 6, F and G). Mouse body weight, as a measure of toxicity, stabilized at 2 weeks after an initial slight reduction, indicating general tolerability of this drug combination (fig. S7E). Together, this work shows that imitating tumor-suppressive CAF effects on survival pathways, such as IGF1R and FAK signaling, with small-molecule drugs can also enhance first-line TKI efficacy in EGFR-mutant NSCLC.

Considering that targeting of IGF1R and FAK signaling enhanced osimertinib efficacy in gefitinib-resistant EGFR-mutant PC9GR lung cancer cells, which as their dominant mechanism of resistance feature the EGFR gatekeeper mutation, this suggested that multiple adaptive signaling and genetic mechanisms can cooperate to mediate overall drug resistance in these cells. To further explore this concept, we evaluated the effects of IGFBPs on EGFR-mutant cell lines, which have developed resistance to EGFR TKIs through other primary mechanisms than the EGFR gatekeeper. rhIGFBPs also increased osimertinib sensitivity of two osimertinib-resistant cell lines, namely, HCC827AZR, which has undergone EMT as shown by increased expression of N-cadherin and vimentin, and PC9AZR, which has an unknown mechanism of resistance (Fig. 7, A and B, and figs. S8A and S12). This suggests that IGF1R and/or FAK signaling is also contributing to some extent to resistance to third-generation EGFR TKIs. Similarly, A431GR epidermoid skin carcinoma cells that express high levels of EGFR and are reportedly resistant to gefitinib due to loss of IGFBP3 expression are sensitive to combined EGFR TKI and IGFBP treatment (fig. S8B) (26), but addition of HGF led to significant osimertinib resistance in these cells. The EGFR-mutant lung cancer cell line HCC827ER4 exhibits resistance to erlotinib due to MET amplification, the expression of which we confirmed by Western blot (fig. S8A) (33). These cells showed sensitivity to the combination of osimertinib with the ALK/MET inhibitor crizotinib (Fig. 7, C and D). In addition, treatment of HCC827ER4 cells with osimertinib and crizotinib in the presence of IGFBPs led to significant inhibition of viability over osimertinib when combined only with crizotinib (Fig. 7C). Consistently, treatment with either an IGF1R TKI (linsitinib or ceritinib), crizotinib, or combinations thereof led to reduced viability compared with osimertinib alone (Fig. 7D). Osimertinib in combination with crizotinib was more potent than osimertinib combined with either of the IGF1R inhibitors, which is in line with MET amplification being the main resistance mechanism in these cells. However, the triple combination was yet more potent than the dual MET/EGFR TKI combination, indicating that IGF1R signaling also contributes to drug resistance in these cells. Collectively, these data show that, even in the presence of dominant resistance mechanisms, such as cancer cell–intrinsic gatekeeper mutations and MET amplification, additional, less obvious or less strong mechanisms, some of which are modulated by CAFs, can contribute to overall drug resistance (Fig. 7E).

A549, PC9, PC9GR, H3122, H2228, STE1, H1975, A431, A431GR, HCC827ER4, and WI38 VA13 subline 2RA (WI38V; human lung fibroblast, SV40-transformed) cells were obtained from the Moffitt Lung Cancer Center of Excellence Cell Line Core. HCC827 (CRL-2868) cells were purchased from American Type Culture Collection (ATCC). PC9AZR were generated as described previously (51). The NAF cell lines MRC5 (CCL-171), WI38 (CCL-75), and IMR-90 (CCL-186), as well as the bone marrow fibroblast cell line HS-5 (CRL-11882), were purchased from ATCC. Frozen aliquots of lung CAFs (CAF7, CAF12, and CAF10), generated with patient consent as previously described (70), were thawed, expanded, and refrozen at low passage to have ample cells for completing the project. Cells were cultured in RPMI 1640 medium containing 10% fetal bovine serum (FBS; RPMI10) at 37°C and 5% CO₂. All cell lines tested negative for mycoplasma contamination and have been authenticated by short tandem repeat analysis. Nuclear mKate2-labeled cells were made according to the manufacturer’s directions with the following modifications: IncuCyte NucLight Red Lentivirus Reagent (Essen BioScience, no. 4476) was added to the cells at a multiplicity of infection of 1.5 in the presence of polybrene (4 μg/ml; Millipore, no. TR-1003-G). Virus was removed after 48 hours, and the cells were allowed to recover for 24 hours in fresh medium before the addition of puromycin (Invivogen, no. ant-pr) to select stably transduced cells.

To generate CM for viability assays, phosphoproteomics, and cytokine analysis, we plated fibroblasts in RPMI10 at 0.8 × 10⁶ cells in a 15-cm dish and allowed them to grow for 4 days (~60 to 70% confluence). The serum-containing RPMI was then collected, clarified by centrifugation (10 min at 1000g), aliquoted, and stored frozen at −80°C until needed.

AZD8330, trametinib, crizotinib, osimertinib, linsitinib, ceritinib, and defactinib were purchased from ChemieTek. Dasatinib and afatinib were from LC Labs. Gefitinib, GSK1838705A, and nazartinib were from SelleckChem. All drugs were dissolved in DMSO (10 mM), aliquoted, and stored at −20°C. Growth factors IGF1, IGF2, and HGF (PeproTech, nos. AF100-11, AF100-12, and 100-39H, respectively) were reconstituted at 100 μg/ml in sterile 0.1% bovine serum albumin (BSA)/phosphate-buffered saline (PBS), aliquoted, and stored at −80°C. rhIGFBP5, rhIGFBP6, and rhIGFBP7 (R&D Systems, nos. 875-B5, 876-B6, and 1334-B7, respectively) were reconstituted at 100 μg/ml in sterile PBS, aliquoted, and stored at −80°C.

To create resistant lines, we cultured drug-sensitive “parental” HCC827 cells with increasing concentrations of osimertinib, starting at the 30% inhibitory concentration. At 80 to 90% confluence, cells were trypsinized and divided into two tubes. Half were frozen, and the rest were reseeded into a new dish at a 20 to 30% higher dose of osimertinib. Fresh drug was added every 72 to 96 hours.

Cell viability assays were conducted using the following techniques: For short-term assays, cells were seeded in RPMI10 in a 384-well microtiter plate (Corning, no. 3764) and treated after 24 hours. Drugs were diluted in RPMI10 and added to the cells at the indicated concentrations. After 72 hours of drug treatment, CellTiter-Glo (CTG) reagent was added according to the manufacturer’s instructions for the CTG Luminescent Cell Viability Assay (Promega, no. G7573), and the resulting luminescence was read on an M5 Spectramax plate reader (SoftMax Pro Software 6.2.1, Molecular Devices). Cell viability was determined relative to DMSO-treated cells. For cell viability assays in CM, the cells were plated in a 1:1 mixture of normal growth medium (RPMI10) and CM. Cell viability in coculture was determined using a IncuCyte ZOOM live-cell analysis system (software versions 2016A/B, Essen BioScience). Fluorescently nuclear-labeled NSCLC cells were plated with unlabeled fibroblasts in either a 1:1 or 1:2 ratio (500:500, 500:1000, or 1000:1000 cells), as indicated. For monoculture controls, labeled NSCLC cells were plated to match their cell number in coculture (500 or 1000 cells, respectively). After 72 hours of drug treatment, cell viability was measured as the red object count (red-labeled NSCLC nuclei) per well relative to the DMSO-treated monoculture wells. Long-term cell viability assays were quantified using crystal violet staining. Cells were plated in six-well dishes and treated after 24 hours with the indicated drug concentrations. For low-density plated cells, drug was added once, and then, the plates were incubated for 10 to 14 days. For high-density plated cells, drug was replenished in fresh RPMI10 every 3 to 4 days for the duration of the assay out to 28 days. At the end of the incubation, the cells were washed once with ice-cold PBS, fixed for 10 min on ice with ice-cold methanol, and then incubated while rocking for 30 min at room temperature (RT) in diluted crystal violet solution (Sigma-Aldrich, no. HT90132; 1:10 in PBS). After extensive rinsing with water to remove excess dye, the plates were dried overnight and then scanned on a flat-bed scanner. To quantify the stained cells, we added RT methanol to each well, rocked the plate for 30 to 60 min at RT, and then measured the extracted crystal violet by reading absorbance at 540 nm on an M5 Spectramax plate reader (Molecular Devices). For spheroid assays, PC9 cells were plated at 6400 cells per well in 96-well ultralow attachment plates (Sigma-Aldrich, no. CLS4520), centrifuged 200g for 10 min, and allowed to form spheroids over 72 hours. Cells were then drug-treated and incubated for an additional 72 hours before addition of CellTiter-Glo 3D Cell Viability Assay reagent (Promega, no. G9682) and processed as described above. Raw data were analyzed using GraphPad Prism 7. One unpaired t test per row was determined without (except for dose-response curves) correction for multiple comparisons, α = 0.05%, and assuming all rows are sampled from populations with the same scatter. Drug combination effects were evaluated using the Bliss method.

Cells were lysed using a 0.2% NP-40, 50 mM tris (pH 7.5), 5% glycerol, 1.5 mM MgCl₂, and 100 mM NaCl lysis buffer containing phosphatase (Sigma-Aldrich, no. P5726) and protease (Roche, no. 11873580001) inhibitors. Proteins were resolved on SDS–polyacrylamide gel electrophoresis gels, transferred to activated polyvinylidene difluoride membranes using the BioRad TransBlot Turbo System, and incubated with primary antibodies. Antibodies were from Abcam: α-SMA (ab32575, RRID:AB_722538; 1:1000); from BD Pharmingen: vimentin (550513, RRID:AB_393716; 1:1000); from Cell Signaling Technology: E-cadherin (3195, RRID:AB_2291471; 1:1000), p-EGFR Tyr¹⁰⁶⁸ (aka Tyr¹⁰⁹²) (2234, RRID:AB_331701; 1:1000), EGFR (4267, RRID:AB_2246311; 1:1000), p-AKT Ser⁴⁷³ (9271, RRID:AB_329825; 1:500), p-AKT Thr³⁰⁸ (13038, RRID:AB_2629447; 1:1000), AKT (9272, RRID:AB_329827; 1:1000), p-p44/42 MAPK (ERK1/2) Thr²⁰²/Tyr²⁰⁴ (4370, RRID:AB_2315112; 1:2000), p-IGF1R β Tyr¹¹³¹/InsR β Tyr¹¹⁴⁶ (3021, RRID:AB_331578; 1:500), IGF1R β (9750, RRID:AB_10950969, 1:1000), p-FAK Tyr³⁹⁷ (8556, RRID:AB_10891442; 1:1000), FAK (13009, RRID:AB_2798086; 1:1000), PARP1 (9542, RRID:AB_2160739; 1:1000), cleaved caspase-3 (9661, RRID:AB_2341188; 1:1000), and α-tubulin (2125, RRID:AB_2619646; 1:1000); from R&D Systems: IGFBP5 (AF875, RRID:AB_355678; 1:2000), IGFBP6 (AF876, RRID:AB_355679; 1:2000), and IGFBP7 (AF1334, RRID:AB_2264436; 1:400); from Sigma-Aldrich: β-actin (A5441, RRID:AB_476744; 1:15,000) and MAPK (ERK1/2) (M5670, RRID:AB_477216; 1:10,000); and from Thermo Fisher Scientific: pan-cytokeratin (MA5-12231, RRID:AB_10980711; 1:50). Secondary immunoglobulin G antibodies were horseradish peroxidase (HRP)–conjugated anti-goat (R&D Systems, HAF109, RRID:AB_357236), anti-mouse (GE Healthcare, NA931, RRID:AB_772210), or anti-rabbit (GE Healthcare, NA934, RRID:AB_772206). After the addition of HRP detection reagent [Clarity (Bio-Rad, no. 1705061), VisiGlo Select (VWR, no. 89424-018), or SuperSignal West Femto (Thermo Fisher Scientific, no. 34095)], the membranes were imaged on the Odyssey Fc LI-COR Dual-Mode Imaging System. Images were visualized and, where indicated, quantified using LI-COR’s Image Studio Lite (version 5.2) software.

To quantify FAP-α expression, adhered cells were detached in the presence of Accutase cell detachment solution (Innovative Cell Technologies, no. AT104), centrifuged (500g/5 min), rinsed, and resuspended with BD Pharmingen Stain Buffer (FBS) (no. 554656). FAP-α–positive cells were detected with FAP-α phycoerythrin-conjugated antibody (R&D Systems, no. FAB3715P) in the presence of 4′,6-diamidino-2-phenylindole as a viability dye. Cells were analyzed using a LSRII SORP (BD Biosciences) with DIVA software. The data were analyzed with FlowJo (v9.9.3/4.; BD Biosciences).

For RNA extraction, each cell line was grown in a 10-cm dish to 80 to 90% confluence. RNA was extracted from cell pellets using the Qiagen RNeasy Mini Kit (no. 74106) according to the manufacturer’s protocol, and RNA quality was assessed on the Agilent TapeStation RNA ScreenTape. One hundred nanograms of total RNA was amplified and labeled with biotin using the Ambion Message Amp Premier RNA Amplification Kit (Thermo Fisher Scientific) following the manufacturer’s protocol and as described elsewhere (71). Hybridization, staining, and scanning of the chips followed the procedure outlined in the Affymetrix technical manual, as previously described (72). The Human Genome U133 Plus 2.0 Arrays used contain over 54,000 probe sets representing over 47,000 transcripts. The array output files were visually inspected for hybridization artifacts and then analyzed using Affymetrix Expression Console v.1.4 with the MAS 5.0 algorithm; scaling probe sets had an average intensity of 500. Lysates for each cell line were collected from three sequential early passages of the respective cells.

CEL files were normalized against the median sample (MRC5_2) using iterative rank-order normalization (IRON). A total of 54,675 probes were reported across nine cell lines. Two-group comparisons used the following criteria for filtering to determine a “Score” for significance: Probe set must not be antisense to the annotated gene, maximum of the two averages must be >5, |log₂ ratio| ≥ ~0.585 [log₂(1.5-fold)], t test P value < 0.05 (two-tailed, two-sample unequal variance), and Hellinger distance > 0.25. Fibroblast versus NSCLC (CAF + NAF versus NSCLC) required additional filtering to avoid being led astray by large differences between CAFs and NAFs, because we were looking for genes that are differentially expressed in both CAF and NAF versus NSCLC. Therefore, the following filters were set: pass CAF + NAF versus NSCLC filter, pass CAF versus NSCLC filter, and sign of CAF versus NSCLC agrees with sign of NAF versus NSCLC. To select differential probes, we removed “unknown” probes as well as probes not passing the Score criteria for each respective comparison. Heatmaps were created using Heatmapper (73). Values were scaled across rows, and clustering was performed across rows and columns using complete linkage and the Pearson distance measuring method. Genes that were significantly increased or decreased between CAF7 and CAF12 versus MRC5 were selected for DAVID pathway analysis (74). The DAVID “UP_KEYWORDS – SECRETED” term was used to determine secreted proteins.

The RayBio C-Series Human Cytokine Antibody Array C5 (RayBiotech, no. AAH-CYT-5-4) was used to determine relative levels of secreted cytokines in fibroblast CM and was performed according to manufacturer’s instructions. Briefly, after blocking the membrane for 30 min at RT, we added 1 ml of undiluted CM to each respective membrane and allowed it to incubate for 5 hours at RT. The membranes were then washed before incubating with Biotinylated Antibody Cocktail overnight at 4°C. A second wash was performed, and then, the membranes were incubated with HRP-streptavidin for 2 hours at RT. After the final wash step, HRP-Detection reagent [either provided by the kit or a more sensitive HRP-reagent, such as VisiGlo Select (VWR) or SuperSignal West Femto (Thermo Fisher Scientific)] was added, and the membranes were imaged on the Odyssey Fc LI-COR Dual-Mode Imaging System. Images were quantified using LI-COR’s Image Studio Lite (version 5.2) software. The cytokine array was performed in duplicate.

To generate SFCM for secretomics analysis, we plated fibroblasts in RPMI10 at 0.8 × 10⁶ cells in a 15-cm dish and allowed them to grow for 4 days (~60 to 70% confluence). The serum-containing RPMI10 was aspirated, and the plates were rinsed with PBS and then incubated with serum-free RPMI (SF-RPMI) for 20 min at 37°C and 5% CO₂ to allow serum-derived proteins to detach from the cells and plate surface. After aspiration and a further quick rinse with SF-RPMI, 18 ml of SF-RPMI was added to the plate, and the cells were incubated at 37°C and 5% CO₂ for 48 hours. The SFCM was then collected, clarified by centrifugation (10 min, 1000g) followed by filtration (0.45 μm; GE Healthcare Puradisc 25 AS, no. 6780-2504), aliquoted, and stored at −80°C until needed. Samples were prepared as biological triplicates; 3 fibroblast types × 3 replicates = 9 samples.

SFCM was lyophilized and then redissolved in urea buffer [aqueous 8 M urea, 20 mM Hepes (pH 8), 1 mM sodium orthovanadate, 2.5 mM sodium pyrophosphate and 1 mM β-glycerophosphate]. A Bradford assay was carried out to determine the protein concentration. The proteins were reduced with 4.5 mM dithiothreitol and alkylated with 10 mM iodoacetamide. Trypsin digestion was carried out at RT overnight, and tryptic peptides were then acidified with 1% trifluoroacetic acid (TFA) and desalted with C18 Sep-Pak cartridges according to the manufacturer’s procedure. After lyophilization, the peptides were redissolved in 400 μl of aqueous 20 mM ammonium formate (pH 10), which was used as peptide fractionation solvent A. To build a comprehensive peptide library, a portion from each sample was pooled and fractionated using a high-pH reversed-phase separation on an XBridge 4.6-mm–inner diameter (ID) × 100-mm-long column packed with BEH C18 resin (3.5-μm particle size, 130 Å pore size; Waters). The peptides were eluted as follows: 5% B [5 mM ammonium formate and 90% acetonitrile (pH 10)] for 10 min, 5 to 15% B in 5 min, 15 to 40% B in 47 min, 40 to 100% B in 5 min, and 100% B held for 10 min, followed by reequilibration at 1% B. The flow rate was 0.6 ml/min, and 12 concatenated fractions were collected. All final peptides (individual samples and the fractioned pool samples) were dried via SpeedVac and redissolved in 20 μl of aqueous 2% acetonitrile with 0.1% formic acid spiked with Pierce peptide retention time calibration mixture standards (20 fmol total). For LC-MS/MS analysis, 5 μl was injected once for each pooled fraction (×12 runs) or in technical duplicates for the individual samples (9 samples × 2 runs = 18 runs).

A nanoflow ultrahigh performance liquid chromatograph [rapid separation LC (RSLC); Dionex, Sunnyvale, CA] interfaced with an electrospray bench top orbitrap mass spectrometer (Q-Exactive Plus; Thermo Fisher Scientific, San Jose, CA) was used for MS/MS peptide sequencing. The sample was first loaded onto a precolumn (100 μm ID × 2 cm packed with C18 reversed-phase resin, 5-μm particle size, and 100 Å pore size) and washed for 8 min with aqueous 2% acetonitrile and 0.04% trifluoroacetic acid. The trapped peptides were eluted onto the analytical column (C18, 75 μm ID × 25 cm length, 2-μm particle size, and 100 Å pore size; Dionex). The 90-min gradient was programmed as follows: 95% solvent A (aqueous 2% acetonitrile + 0.1% formic acid) for 8 min, solvent B (90% acetonitrile + 0.1% formic acid) from 5 to 38.5% in 60 min, then solvent B from 50 to 90% B in 7 min and held at 90% for 5 min, followed by solvent B from 90 to 5% in 1 min and reequilibration for 10 min. The flow rate on the analytical column was 300 nl/min. Sixteen tandem mass spectra were collected in a data-dependent manner after each survey scan. MS/MS scans were performed using a 15-s exclusion for previously sampled peptide peaks.

MaxQuant (75) with match between runs selected was used to quantify the intensities. After IRON (76), the resulting log₂ iBAQ (intensity-based absolute quantification) values were converted back to their respective intensities (77, 78). Missing values, having been converted to 1, were then imputed to the minimum whole-number intensity in each respective sample. From the total 2632 proteins identified, three types of entries were deleted: proteins detected only in the pool, nonhuman (for example, Bos taurus) proteins, and unnamed and/or contaminating proteins, resulting in a final 2330 proteins (data file S1). Each iBAQ intensity was then divided by the sum of the iBAQ intensities for that run (×18 runs), resulting in a relative iBAQ (riBAQ) for each protein in each run. Technical replicates (2× MS injections) were averaged, and the remaining values (×9) were subjected to the t test (two-tailed, two-sample equal variance) comparing the 6× riBAQ values for CAF7 and CAF12 to the 3× riBAQ values for MRC5 to determine P values for significant differences between proteins in CAF versus NAF SFCM. The biological replicates (6× or 3×) were then averaged into two single final values (riBAQave), which, as a measure of relative abundance for each protein in either the CAF or NAF SFCM samples, was used to determine protein ratios (CAF/NAF) between samples. Although the relative ratio for some lesser expressed proteins may be larger, in absolute terms, selecting highly abundant proteins with a reasonable relative ratio accounts for more active molecules, which is likely to have a higher impact on cell viability. Therefore, we selected the top 2.5% (based on riBAQ values) of all identified proteins (2330 proteins total × 2.5% = 58 proteins). These proteins were subsequently subjected to DAVID pathway analysis (74). Because we observed not just the absence of resistance but actual sensitization, we focused on pathways and proteins with anticancer activity.

CAF12 cells were transfected by reverse transfection using Lipofectamine RNAiMAX Transfection Reagent (Thermo Fisher Scientific, no. 13778150) according to the manufacturer’s instructions. In brief, 20 nM final concentration siRNA [Dharmacon, IGFBP5 ON-TARGETplus SMARTpool (no. L-010897-00-0005), IGFBP6 ON-TARGETplus SMARTpool (L-006625-00-0005), individual IGFBP5 siRNAs (si-IGFBP5-8 and si-IGFBP5-10, no. LQ-010897-00-0002), or the control nontargeting pool (si-NT, no. D-001810-10-20)] was incubated for 20 min in a six-well dish containing 5 μl of Lipofectamine RNAiMAX in 500 μl of Opti-MEM (Gibco, no. 31985062) medium. Subsequently, 1.5 × 10⁵ CAF12 cells were added in 2 ml of RPMI10, bringing the final volume to 2.5 ml. After 24 hours at 37°C and 5% CO₂, the medium was exchanged for 3 ml fresh RPMI10. The resulting CM after siRNA-mediated gene silencing (siCM) was harvested at 96 hours and processed as described above. The cells were also collected to confirm knockdown efficiency by Western blot analysis.

PC9GR cells were plated in 1:1 RPMI10:CM overnight and then treated with DMSO or 500 nM osimertinib for 3 hours before harvesting. To collect the cells, we rinsed the cells twice with ice-cold PBS, then scraped them off the plate and collected them in ice-cold PBS. Cells were centrifuged at 500g for 10 min, the PBS was aspirated, and the cell pellet was immediately lysed according to the manufacturer’s instructions for the Proteome Profiler Human Phospho-RTK Array Kit (R&D Systems, ARY001B). A Bradford assay was carried out to determine the protein concentration. Continuing with the provided protocol, the membranes were blocked for 1 hour before addition of 250 μg of protein diluted to 1.5 ml in array buffer, followed by an overnight incubation at 4°C. The membranes were then washed before incubating with phospho-tyrosine HRP antibody for 2 hours at RT. A second wash was performed, and then, HRP reagent—either that provided by the kit or a more sensitive HRP-reagent, such as VisiGlo Select (VWR) or SuperSignal West Femto (Thermo Fisher Scientific)—was added, and the membranes were imaged on the Odyssey Fc LI-COR Dual-Mode Imaging System. Images were quantified using LI-COR’s Image Studio Lite (version 5.2) software. Data were normalized using the following equation: X(Ny) = X(y)* P1/P(y), where P1 is the mean signal density of positive control spots on the reference (MRC5) array, P(y) is the mean signal density of positive control spots on array “y” (CAF7 or CAF12), and X(Ny) is the normalized signal intensity for X(y) [spot “X” on array “y” (CAF7 or CAF12)]. The Phospho-RTK Array was performed in duplicate. Data were analyzed using GraphPad Prism 7.

For each condition, PC9GR cells were plated in 10 15-cm dishes at 7 × 10⁶ cells per dish in a 1:1 mixture of RPMI10 and (serum-containing) CM from CAF or NAF cells and allowed to adhere for 24 hours before a 3-hour treatment with 500 nM osimertinib or DMSO, as control. To collect the cells, we decanted the medium, and the cells were rinsed twice with ice-cold PBS containing sodium vanadate, then scraped off the plate and collected in ice-cold PBS containing sodium vanadate. Cells were centrifuged at 500g for 10 min, the PBS was aspirated, and the cell pellet was shock-frozen in liquid nitrogen and stored at −80°C until further processing. Samples were prepared as biological triplicates. Cells were lysed in denaturing lysis buffer containing 8 M urea, 20 mM Hepes (pH 8), 1 mM sodium orthovanadate, 2.5 mM sodium pyrophosphate, and 1 mM β-glycerophosphate. A Bradford assay was carried out to determine the protein concentration, and samples were denatured, digested, and desalted as described above. After lyophilization, the dried peptide pellet was redissolved in immunoaffinity purification buffer containing 50 mM Mops (pH 7.2), 10 mM sodium phosphate, and 50 mM sodium chloride. Phosphotyrosine-containing peptides were immunoprecipitated with immobilized anti-phosphotyrosine antibody p-Tyr-1000 (Cell Signaling Technology). After 2-hour incubation, the antibody beads were washed twice with immunoaffinity purification buffer, followed by three washes with H₂O. The phosphotyrosine peptides were eluted twice with aqueous 0.15% TFA, and the volume was reduced to 20 μl via vacuum centrifugation before LC-MS/MS as described above. Samples were injected as technical duplicates, resulting in 6 values (3 biological × 2 technical) per sample (24 total injections).

MaxQuant was used to quantify the intensities (75). A total of 907 Tyr-phosphorylated peptides were identified with a posterior error probability (PEP) score less than 0.05. After log₂ conversion of the peptide intensities, the technical replicates were averaged, and peptides with 0 or 1 value only per treatment triplicate across all treatments were deleted, resulting in 847 peptides corresponding to 484 proteins. The samples were subjected to the t test (two-tailed, two-sample equal variance) comparing the four conditions (three biological replicates each) to determine P values for significant differences between them (for example, CAF CM + DMSO versus CAF CM + osimertinib, or CAF CM + DMSO versus NAF CM + DMSO). Biological triplicates were then averaged, and the log₂ ratio was determined between the samples. Data were analyzed using GraphPad Prism 7 and DAVID pathway analysis (74). Phosphopeptides were manually confirmed by extracted ion chromatograms.

PC9 cells were subcutaneously injected into 4- to 6-week-old NOD-scid IL2Rg^null (NSG) recipient mice, produced from in-house breeding colony, with breeders purchased from the Jackson Laboratory. Each animal received two contralateral injections containing 10⁶ tumor cells suspended in 100 μl of 1:1 mix of RPMI/Cultrex basement membrane extract type 3 (R&D Systems). Three weeks after injections, animals were randomized into treatment and control groups and subjected to daily oral gavage with osimertinib (2 mg/kg; ChemieTek, no. CT-A9291), ceritinib (25 mg/kg; ChemieTek, no. CT-LDK378), a combination of the two drugs, or vehicle control (0.5% methyl cellulose/0.5% Tween 80). Tumor diameters, measured by electronic calipers, and animal weights were measured weekly. For tumor volume calculations, spherical shape of tumors was assumed. After 4 weeks of treatment, animals were euthanized, and tumors were weighted. The results were reproduced in two independent experiments, in both males and females. Xenograft studies were performed in accordance with the guidelines of the Institutional Animal Care and Use Committee of the H. Lee Moffitt Cancer Center. Animals were maintained under Association for the Assessment and Accreditation of Laboratory Animal Care (AAALAC)-accredited specific pathogen–free housing vivarium and care and veterinary supervision following the standard guidelines for temperature and humidity, with a 12-hour light/12-hour dark cycle.