The Secretome Engages STAT3 to Favor a Cytokine-rich Microenvironment in Mediating Acquired Resistance to FGFR Inhibitors

Acquired resistance emerges as the major obstacle for targeted cancer therapeutics (1). In contrast to primary resistance, acquired resistance hinders the effectiveness of nearly all small-molecule inhibitors during prolonged treatment, regardless of their initial efficiency. Thus, understanding the mechanism of acquired resistance is important. In addition to well-accepted mechanisms, such as target mutation and pathway reactivation, the interplay between cancer cells and the surrounding stromal cells or the tumor microenvironment (TME) in mediating acquired resistance has attracted attention (2–4) and paved the way for strategies targeting the TME, including the modulation of specific cytokines, such as Hepatocyte growth factor and insulin-like growth factor (5–7). More recently, a short report showed that the cytokine secretome, rather than a specific cytokine, could also mediate acquired resistance to RAF inhibitors in melanoma cell lines (8). These emerging data indicated an important role of the secretome in the TME in acquired resistance.

FGFR is overactivated in a wide subset of tumors (9–11). Up to 20% of squamous non–small cell lung cancers (NSCLC), 10% of gastric cancers, 4% of triple-negative breast cancers, and 50%–60% of bladder cancers are connected to FGFR aberrations (12, 13). Therefore, agents targeting FGFR are under extensive investigation. Nevertheless, acquired resistance ultimately emerges with prolonged treatment with FGFR inhibitors (FGFRi), within 4–6 months for patients with intrahepatic cholangiocarcinoma and within 14 months for patients with gastric cancer (14, 15). Although classical mechanisms, such as FGFR mutation, feedback activation, and epithelial–mesenchymal transition, have been reported (15–22), no reports have discussed the contribution of the TME to acquired FGFRi resistance.

Here, we focused first on lung cancer because it is the most common cause of cancer-related death around the world in a 2017 cancer report (23). FGFR1 amplification is found in 20% of NSCLCs, and we thus chose one of the two available FGFR1-amplified lung cancer cell lines, NCI-H1581. Compared with DMS-114, the other FGFR1-amplified cell line, this cell line was not only highly sensitive to FGFRis but also had a rapid growth rate, which facilitated our research. We explored the role of the secretome in acquired resistance in NCI-H1581 cells and then expanded our findings to other FGFR-aberrant cancer cell lines.

NCI-H1581, SNU16, Huh7, MRC-9, IM-9, IMR-90, WI38, and U937 cells were obtained from ATCC. UMUC14 and MFM-223 cells were obtained from the European Collection of Cell Cultures. RT112 and OPM2 cells were obtained from Deutsche Sammlung von Mikroorganismen und Zellkulturen GmbH. EA.hy926 cells were obtained from the Academy of Military Medical Sciences of China, and THP-1 cells were obtained from the Shanghai Institute of Biochemistry and Cell Biology (Shanghai, P.R. China). All cell lines were maintained in appropriate medium, as suggested by the manufacturer, and authenticated via short tandem repeat analysis by Genesky Biopharma Technology (last tested in 2017).

All inhibitors used in the kinase library screening assay were purchased from Selleck Chemicals. PubChem IDs can be found in Supplementary Table S1. For in vivo studies, AZD4547 was obtained from Melone Pharmaceutical Co., Ltd. All these reagents were dissolved in DMSO for in vitro studies and in normal saline with 1% Tween-80 for in vivo studies. Phorbol 12-myristate 13-acetate (PMA) and lipopolysaccharide (LPS) were used for THP1 stimulation and were obtained from PeproTech Co., Ltd. The siRNA constructs are synthesized by Genepharma Co., Ltd, with the sequences listed in Supplementary Table S1.

To generate cells resistant to FGFRi, NCI-H1581 cells were exposed to AZD4547, BGJ398, or lucitanib; the drug concentrations increased in a stepwise manner from 30 nmol/L to 1 μmol/L when the cells resumed growth kinetics similar to those of the untreated parental cells. After approximately 6 months, resistant subpopulations, which were named NCI-H1581/AR, NCI-H1581/BR, and NCI-H1581/LR, respectively, were obtained.

Whole-exome sequencing and RNA sequencing were performed at Beijing Genomics Institute (http://www.genomics.cn).

Cell proliferation or viability assays were conducted using CCK8 or SRB cells, according to our previously established methods (24). For CCK8, the cells were seeded in a 96-well plate on day 1. After 2–4 hours, inhibitors or siRNAs were added to the predesigned wells in a concentration gradient or at a specific concentration. The cells were seeded at a density of 8,000–1 × 10⁴ per well, depending on the cell type, or at a density of 20%–30% when the cells were ready for passage. The plates were then placed in a 37°C incubator containing 0.05% CO₂ for 3 days. On day 4, the 96-well plates were removed from the incubator, treated with 10 μL of the CCK8 reagent per well (for floating cells), and then moved back to the same incubator for 1–4 hours. During this process, the plates were analyzed using a microplate spectrophotometer until the OD (optical density) value of the control well reached 0.8–1.2. Cell viability was calculated as (OD_{test cells} − OD_{control cells})/OD_{control cells} × 100%. The inhibition rate equals 1 − cell viability.

The short hairpin RNAs (shRNA) used to stably knockdown STAT3 were purchased from Sigma-Aldrich (shRNA1, TRCN0000329888; and shRNA2, TRCN0000353630). The negative control shRNA was selected from the study performed by Liu and colleagues, (24). To generate cells with stable knockdown, the plasmids were transfected into 293FT packaging cells with Lipofectamine 2000 (Invitrogen).

The siRNAs used in our study were generated by GenePharma, Co. Ltd., and the assays were conducted using Lipofectamine RNAiMAX (Invitrogen) according to a standard transfection procedure. The complete sequences of these constructs are listed in Supplementary Table S1.

Western blotting was performed using our previously established methods (24). For the tyrosine kinase and receptor tyrosine kinase (RTK) array chips, samples of H1581 and H1581/AR cells were prepared according to the Human Phospho-Kinase Array Kit (catalog no. ARY003B, R&D Systems) and the Human Phospho-RTK Array Kit (catalog no. ARY001B, R&D Systems) according to the manufacturer's instructions.

To immunoprecipitate pFGFR, 4 × 10⁷ cells of both the parental and resistant cells were collected and lysed using RIPA lysis buffer (strong) with protease inhibitor cocktail tablets (Roche, catalog no. 04693132001) and phosphatase inhibitors (Roche, PhosStop, catalog no. 04906837001). The supernatants of the protein lysates were recovered by centrifugation at 12,000 r.c.f at 4°C and protein concentrations were normalized to a concentration of 2–4 mg/mL in a 1-mL volume using a BCA Assay Kit (Thermo Fisher Scientific). Then, the lysates were mixed with 30 μL of glutathione sepharose and 15–20 μL of an anti-FGFR antibody [Cell Signaling Technology, FGF Receptor 1 (D8E4) XP Rabbit mAb, catalog no. 9740] overnight for each lane. After elution using RIPA buffer, the sepharose samples were lysed with 4 × loading buffer, and the supernatants were centrifuged (12,000 r.c.f, 4°C) and analyzed by normal Western blotting procedures. For the detection of pFGFR, an antibody specific for phosphorylated-FGFR [phospho-FGF Receptor (Tyr653/654) (55H2) mouse mAb catalog no. 3476; Cell Signaling Technology] was used.

The cells were seeded and cultured for 1 day, and conditioned medium was collected. Then, each conditioned medium was mixed with the normal medium for each tested cancer cell line to conduct conditioned medium assays.

For the coculture assay, the indicated stromal cells were seeded into the bottom wells, and H1581 cells were cultured into the top wells in 6-well coculture plates. After incubating the cells for 1 day, the conditioned medium was collected for secretome detection. Then, inhibitors were added, and the cells were incubated for 3 more days before proliferation was analyzed. The interference of STAT3 in the coculture assay was conducted 1 day before the initiation of the coculture and cell viability assay. With respect to cell numbers, the H1581 cells were grown at 1.0–1.6 × 10⁵ cells per 2 mL per well, and MRC9, THP1, and other stromal cells were grown at 5 –8 × 10⁴ cells per 2 mL per well.

Four- to 6-week-old nude or SCID mice were obtained from the Shanghai Laboratory Animal Center of the Chinese Academy of Sciences (Shanghai, China). Cancer cells (1 × 10⁷) were suspended in 200 μL of PBS and inoculated subcutaneously into the right flanks of the mice. When the volume of NCI-H1581, H1581/AR-NC, sh1-STAT3-AR, or sh2-STAT3-AR xenograft tumors reached approximately 700–800 mm³, they were excised and cut into approximately 1.5-mm³ segments, which were further implanted subcutaneously in mice via a Trocar needle. In evaluating the resistance potential of NCI-H1581/AR cells in vivo, the tumor reached 150–350 mm³ when the mice were randomly divided into control and treated groups. For THP1/H1581 coculture assays in vivo, PMA was dissolved in DMSO to obtain a concentration of 200 μg/mL and added to THP1 cells at a dilution of 1:1,000 to generate a final concentration of 200 ng/mL, and the cells were cultured for 48 hours. The cells were digested using trypsin and mixed with H1581 cells at a 1:1 ratio, with 1 × 10⁷ cells each. The mixture was then injected subcutaneously into the right flanks of the mice. The individual relative tumor volume (RTV) was calculated as follows: RTV = V_t/V₀, where V_t is the volume on each day, and V₀ is the volume at the beginning of the treatment. Tumor growth inhibition (TGI) was measured using the following formula: TGI (%) = [1−(RTV in treated group)/(RTV in control group)] × 100%. All efficacy studies were conducted as described in our previous reports (24). The experiments adhered to the Institutional Animal Care and Use Committee guidelines and the animal welfare policies of Shanghai Institute of Materia Medica (Shanghai, China). Animal procedures were approved by the Institutional Animal Care and Use Committee at Shanghai Institute of Materia Medica (approval nos. 2014-03-DJ-13, 2015-04-DJ-17, 2016-04-DJ-21, and 2018-05-DJ-37, Shanghai, China).

The differences between experimental groups in in vitro and in vivo studies were compared using an unpaired two-tailed Student t test. P < 0.05 was considered statistically significant. Bioinformatics analyses were mostly performed using OmicsBean (http://www.omicsbean.com:88/), an online bioinformatics tool, and by the companies that performed the next-generation sequencing (NGS) and cytokine array assays.

The IC₅₀ values of FGFRi against sensitive cells were approximately 60 nmol/L; however, the IC₅₀ values against resistant cell lines, which were designated H1581/AR and H1581/BR, were approximately 11–17 μmol/L (Fig. 1A). These results indicated that H1581/AR and H1581/BR are resistant to FGFRi. Given that FGFRi induce cell-cycle arrest to elicit a drug response (24), we examined the cell-cycle distribution upon FGFR inhibition to further confirm the loss of FGFR dependence in resistant cells. As expected, in contrast to its effect on parental cells (Fig. 1B, left), AZD4547 was unable to induce G₁–S cell-cycle arrest in the resistant cells (Fig. 1B, right).

We next probed the activation status of FGFR and its downstream effectors. As expected, FGFRi strongly abrogated the activation of FGFR1 and FRS2α, indicating effective target inhibition (Fig. 1C). With respect to FGFR key downstream effectors, FGFRi significantly suppressed ERK phosphorylation only in FGFRi-sensitive H1581 cells, but not in FGFRi-resistant H1581/AR or H1581/BR cells. In addition, we also noticed a higher basal level of the FGFR protein in cells with acquired resistance, and the mRNA level and DNA copy number of FGFR were slightly higher (2-fold) in both types of resistant cells than in the parent cells (Supplementary Fig. S1A); however, knockdown of FGFR using a specific siRNA failed to restore the sensitivity of H1581/AR cells to AZD4547 (Supplementary Fig. S1B), suggesting that increased FGFR levels did not contribute to FGFRi resistance. In addition, although both resistant cell lines exhibited a much higher ERK phosphorylation level than the sensitive H1581 cell line, inhibitors that suppressed ERK phosphorylation combined with AZD4547 still failed to inhibit the proliferation of H1581/AR cells, which also excluded the possibility that ERK activation was responsible for acquired resistance (Supplementary Fig. S1C).

To test the in vivo efficacy, mice bearing H1581 and H1581/AR tumors were treated with AZD4547 at an optimal dose of 12.5 mg/kg, as confirmed in our previous studies and other reports (24, 25). In contrast to the high efficacy observed in the parental H1581 model (TGI = 83.63% at end point), mice bearing H1581/AR cells responded poorly to AZD4547 (Fig. 1D, TGI = 31.57% at end point), further confirming that the resistance potential was sustained in vivo.

To exclude the possibility that a specific chemotype contributes to acquired resistance, we tested the ability of resistant cells to exhibit cross resistance to a wide range of FGFRi. The tested inhibitors included both specific and nonspecific inhibitors with different chemotypes. Our results showed that both resistant cell lines demonstrated cross resistance to another well-accepted FGFRi, LY2874455, as well as to two multitarget FGFRis, lucitanib and ponatinib (Fig. 1E).

To systematically identify possible determinants of acquired resistance to FGFR inhibition, we first examined whether potential RTK pathway feedback activation drives acquired resistance. Using a RTK array (Supplementary Fig. S1D), we found that c-Met was obviously upregulated among the RTKs, which was further validated using independent immunoblotting assays (Supplementary Fig. S1E). However, the suppression of c-Met using both inhibitors (Supplementary Fig. S1F) and siRNAs (Supplementary Fig. S1G) failed to reverse the observed resistance to FGFRi, suggesting that c-Met activation may not be the dominant contributor to acquired resistance. This result is different from the positive c-Met inhibition result from a previous report that used the “spindle-like” H1581 cells (26), which differed from the standard phenotype suggested by ATCC: “cells grow mostly as floating aggregates of round cell clusters with some attached cells”.

We next examined the molecular differences between the parental and resistant cell lines using NGS, and we used NCI-H1581/AR as the representative cell line. Our data showed that although the two cell lines shared many somatic genetic aberrations, none of the point mutations, gene amplifications, or gene fusions identified were closely associated with common mediators that contribute to resistance (mutation, amplification, and gene fusion data in Supplementary Table S2). However, using RNA sequencing, we found differential expression of secreted proteins (Supplementary Table S3; RNAseq cytokine raw data), including ILs, growth factors and their cognate growth factor receptors, TNF, and transforming growth factors. These data strongly suggest that acquired resistance to FGFRi may arise from the reprogramming of various cytokines.

We next examined the secretome of resistant cells. We analyzed conditioned medium derived from H1581/AR (AR-CM) and H1581/BR (BR-CM) cells, as well as the H1581 parental cell line (P-CM), using a human cytokine chip array that detects 507 human secreted proteins, including cytokines, chemokines, and growth factors. We found that 202 (40%) and 379 (76%) proteins were upregulated in AR-CM and BR-CM, respectively (Fig. 2A). In addition, 178 secreted proteins were commonly upregulated between the two types of conditioned medium compared with P-CM. Notably, the fold change of the cytokine protein level between AR-CM or BR-CM and P-CM varied from 1.5- to 3-fold and 1.5- to 6-fold, respectively (Fig. 2B). Moreover, a uniform distribution of secreted proteins was upregulated, indicating that the entire conditioned medium may be involved in resistance in the resistant cells.

The top-ranked cytokines in both AR-CM and BR-CM included IL family (IL21, IL1R) and platelet-derived growth factors (PDGF-BB; Supplementary Fig. S2A). Note that FGF was higher in both resistant cells, suggesting that autocrine loop is formed. Given the availability of cytokines, we choose IL6 and PDGF-BB, the representative top-ranked or classic cytokines, to test their potential to elicit acquired resistance. Unexpectedly, adding IL6 or PDGF either separately or in combination showed no ability to elicit resistance in the NCI-H1581 parent cell lines (Supplementary Fig. S2B). Thus, we next asked whether the secretome as an entity could functionally drive acquired resistance. We mixed conditioned medium from resistant cell lines with different proportions of normal medium to test their ability to induce acquired resistance. An increased percentage of AR-CM gradually abrogated the inhibitory effect of 1 μmol/L AZD4547 on H1581 cells in a dose-dependent manner (Fig. 2C). Consistent with this finding, AR-CM–induced resistant cells also demonstrated cross resistance to both selective FGFRi and the multitarget FGFRi lucitanib, regardless of the percentage of conditioned medium used (Fig. 2D). These effects were also confirmed using conditioned medium from BR cells and lucitanib-resistant H1581/LR cells (Supplementary Fig. S3A–S3D).

We next attempted to ascertain whether secretome-induced acquired resistance is a common event in cancer cells with different aberrantly activated FGFR members and other oncogenic kinases. FGFR1OP-driven KG1 cells, FGFR2-amplified KATOIII, SNU-16, MFM223, and SUM52PE cells, FGFR3-mutated OPM2 and UMUC14 cells, FGFR3-amplified RT112 cells, FGFR4-driven Huh7 cells, MET-driven MKN45 cells, and ALK-driven H3122 and H2228 cells were used. Our data showed that mixtures containing 50%–100% AR-CM could significantly abrogate the sensitivity of SNU16, MFM223, OPM2, UMUC14, RT112, and Huh7 cells to FGFRi (Fig. 2E), but had no influence on other tested cell lines (Supplementary Fig. S3E), suggesting that secretome-induced acquired resistance has certain commonalities in different types of cells.

We next sought to understand the mechanism by which the secretome induces acquired resistance. Pathway enrichment analysis of the upregulated cytokines shared between AR-CM and BR-CM revealed several pathways that were strongly correlated with the secretome (Fig. 3A). The top-ranked pathway according to significance level was the JAK–STAT pathway. Consistent with this result, a kinase chip array also indicated strong activation of STAT3 (Fig. 3B, top), which was further confirmed by independent immunoblotting analysis (Fig. 3B, bottom).

Next, to determine the contribution of STAT3, we used siRNA to knockdown STAT3 expression. Compared with the weak effects of siSTAT3 on the NCI-H1581 parent cell line (Fig. 3C), we found that STAT3 knockdown significantly inhibited the growth of H1581/AR cells combined with 1 μmol/L AZD4547 (Fig. 3D). Moreover, other STAT family members (STAT1, STAT2, STAT5, and STAT6) are activated in the resistant cells, as indicated by immunoblotting (Supplementary Fig. S4A). However, using a cut-off value of 40% of cell viability for efficient inhibition as we observed in STAT3 depletion, the knockdown of these proteins failed to induce the efficient inhibition in combination with AZD4547 (Supplementary Fig. S4B), highlighting the dominant role of STAT3 in acquired resistance. In addition, various kinase inhibitors targeting other kinases that were enriched in the pathway analysis failed to reverse the resistance of the cells to FGFRi, demonstrating that other kinases were not responsible for the resistant phenotype (Supplementary Fig. S4C).

To further verify our findings, we used shRNA to stably knockdown STAT3 and found that 1 μmol/L AZD4547 strongly suppressed the viability of shSTAT3-H1581/AR cells by approximately 60%–70% (Supplementary Fig. S4D). In addition, 1 μmol/L AZD4547 induced G₁–S cell-cycle arrest in shSTAT3-H1581/AR cells (Supplementary Fig. S4E). Furthermore, we extended our study to H1581/BR cells. Similarly, enhanced STAT3 activation was observed (Supplementary Fig. S4F). STAT3 knockdown restored sensitivity to BGJ398 in H1581/BR cells (Supplementary Fig. S4G), strongly suggesting that STAT3 signaling is the core pathway responsible for acquired resistance.

After confirming that STAT3 activation is required for acquired resistance, we next wanted to determine whether the ability of the secretome to induce resistance occurs via STAT3 activation. As resistance emerged, STAT3 phosphorylation also progressively increased in H1581 cells, peaking at a concentration of 67% or higher for both AR-CM and BR-CM (Fig. 3E). Consistent with this result, STAT3 phosphorylation was also enhanced in FGFR-dependent cell lines that demonstrated acquired resistance induced by AR-CM (Supplementary Fig. S4H) but not in cell lines that failed to demonstrate acquired resistance, including KG1, SUM52PE, MKN45, H3122, and H2228 cells (Supplementary Fig. S4I and S4J), which suggested that secretome-induced resistance is dependent on STAT3 activation. To further validate the dominant role of STAT3 in secretome-elicited resistance, we knocked down STAT3 in parental H1581 cells. STAT3 knockdown almost completely abrogated the FGFRi resistance induced by AR-CM (Fig. 3F). Collectively, our findings revealed that the secretome induces acquired resistance in a STAT3-dependent manner.

Finally, to confirm our findings in vivo, we performed xenograft studies using two independent shSTAT3-AR cell lines. In contrast to the mild inhibitory effects of AZD4547 on NCI-H1581/AR NC xenograft (TGI = 23.92% at end point), AZD4547 treatment significantly reduced the volume of both shSTAT3-AR xenograft at the end point, with TGI values reaching 79.24% and 60.21% at end point, respectively (Fig. 3G). The potency of the drug in shSTAT3-AR was comparable with that of AZD4547 in H1581 xenograft (Fig. 1D, TGI = 83.63%). Collectively, these results demonstrate that the secretome engages and depends on STAT3 to induce acquired resistance.

We next focused on identifying the molecules that link the secretome with STAT3. Surprisingly, we found that canonical JAK inhibitors, that were effective to hit the JAK targets, failed to reverse acquired resistance (Supplementary Fig. S5A), which could result from the fact that phosphorylation of STAT3 was not consistently inhibited, especially the Ser727 sites. In addition to JAK, all the other agents targeting single kinases failed to reverse resistance. These treatments included agents targeting the canonical IL6–STAT3 pathway (using siRNAs targeting gp130, the subunit of the type I cytokine receptor within the IL6 receptor family), the LIFR- and OSM-activated STAT3 pathway (using antibodies targeting LIFR and OSM), and noncanonical intracellular kinase–mediated STAT3 activation (inhibitors of Abl, Src and ALK; Supplementary Fig. S5A). However, Hsp90 inhibitors could significantly suppress the proliferation of resistant cells. These results suggest that multiple activators that act together to activate STAT3 are involved in FGFRi resistance, and the inhibition of either mediator alone will not work as well. Indeed, classical receptors that dominate STAT3 activation, including JAK families, the common cytokine receptor domain GP130 and RTKs, were highly phosphorylated (Supplementary Fig. S5D), and Hsp90 inhibitors could reduce the activation of representative tested mediators, as well as phosphorylation at the tyrosine 705 (Y705) and serine 727 (S727) sites in STAT3 (Fig. 3H). Moreover, Hsp90 inhibitors also significantly suppressed the secretion of cytokines (Fig. 3I, left), which could be attributed to STAT3 and many other secretion pathways, as revealed by pathway enrichment analysis (Fig. 3I, right).

In conclusion, cognate receptors and mediators upstream of STAT3 are activated and function as an ensemble to mediate signaling between the secretome and STAT3.

Resistance is known to emerge gradually over time. We next examined the status of STAT3 activation, the secretion of cytokines and their ability to induce acquired resistance, as well as STAT3 activation during dynamic adaptation to FGFRi. We selected cell lines with intermediate acquired resistance (Fig. 4A). As expected, the cells exhibited gradually declining sensitivity to AZD4547 and BGJ398 during the development of acquired resistance, and the IC₅₀ value increased more than 6-fold in H1581/BR cells with intermediate resistance, with values of 80 nmol/L and over 7 μmol/L, respectively (Fig. 4B). Accordingly, we observed an increase in STAT3 phosphorylation in both categories of intermediate resistant cell lines (Fig. 4C), demonstrating a clear positive association between increased resistance to FGFRi and gradually increased STAT3 activation. Moreover, we observed significantly upregulated cytokine levels in cells that could tolerate 500 nmol/L AR and 300 nmol/L BR compared with those that tolerated 80 nmol/L AR and 100 nmol/L BR, respectively (Fig. 4D, left). Notably, these upregulated cytokines could also be enriched in a series of pathways involving STAT3 (Fig. 4D, right). More importantly, the secretome of intermediate resistant cells could significantly induce acquired resistance and STAT3 activation (Fig. 4E, left). In addition, the phosphorylation of STAT3 in H1581 cells was also gradually enhanced by conditioned medium from cells that showed intermediate resistance to AZD4547 (Fig. 4F, left). Similar results were obtained using conditioned medium from cells that showed intermediate resistance to BGJ398 (Fig. 4E and F, right). In conclusion, our data confirmed that acquired resistance to FGFRi is exaggerated by the time-dependent augmentation of cytokine secretion and STAT3 activation.

In addition to the time dependence of this process, it can never be neglected that cancer cells exist within the surrounding microenvironment, which may also promote the evolution of acquired resistance through secreted cytokines. Indeed, recent reports have frequently mentioned that a specific TME could support the expansion of drug-resistant clones when targeted therapy is utilized (5–7). Thus, to explore the contribution of secretome–STAT3-mediated acquired FGFRi resistance in the context of the microenvironment, we built a coculture model that made the interaction between cancer cells and stromal cells possible.

We first examined which types of these human stromal cells could interact with sensitive H1581 cells and induce acquired resistance. The main stromal cells, macrophages, fibroblasts, endothelial cells, and lymphocytes, were included. We chose a 60% survival rate as the cutoff. We found that the interaction between THP-1 macrophages or MRC-9 fibroblasts and H1581 cells could indeed significantly induce acquired resistance (Fig. 5A). Moreover, this interaction could activate the phosphorylation of pathways involved in cytokine regulation in THP1 cells, indicating the activation of THP1 cells (Fig. 5B). Consistent with this finding, we found that a large number of cytokines were upregulated in a H1581 cell coculture model with MRC9 or THP1 cells (Fig. 5C, left), with JAK–STAT signaling ranking at the top in the pathway enrichment analysis (Fig. 5C, right), which was highly consistent with the H1581AR results. Furthermore, we tested our findings in an in vivo coculture xenograft. The interaction of H1581 cells with cocultured THP1 cells resulted in a loss of sensitivity to AZD4547 treatment (Fig. 5D, TGI = 31.56%), as compared with that in the noncoculture model (Fig. 1D, TGI = 71.40%). Similar trend was observed in MRC9/H1581 coculture xenograft (Supplementary Fig. S6A). More importantly, both coculture models showed significant upregulation of the secretome compared with the parental H1581 model, which was also enriched in the top-ranked STAT3 signaling pathway (Fig. 5E). These results revealed that interactions between macrophages or fibroblasts and FGFR-amplified lung cancer cells could simultaneously and significantly activate stromal cells and enhance cytokine secretion as well as induce acquired resistance.

We next investigated whether the interaction between fibroblasts/macrophages and FGFR-related cancer cells could enhance the extent of acquired resistance. Cell lines with intermediate resistance to AR and BR, as mentioned above, were cocultured with MRC-9 and THP1 cells, respectively. THP-1 and MRC9 cells were seeded in 6-well plates, and after cell adhesion, H1581 cells were grown in the top chambers of coculture wells. The stromal and tumor cells of each pair (THP1/H1581 or MRC9/H1581) could exchange soluble factors through the coculture membranes. Our data clearly showed that coculture with THP-1 and MRC9 cells significantly increased the resistance potential of a series of H1581 cell lines with intermediate acquired resistance. The cell viability increased from 25% to 80% in the THP1 coculture model with AR-80 nmol/L, while a smaller increase was observed with AR-500 nmol/L (Fig. 5F, left), indicating that the maximum effect had been obtained. Similar results were observed during coculture with MRC9 cells (Fig. 5F, right). Consistent with these results, compared with intermediate resistant cells alone, THP-1 coculture substantially enhanced the secretion of cytokines, as illustrated in Fig. 5G. In addition, this interaction could also enhance the activation of THP1 cells, as demonstrated by the phosphorylation of STAT1, STAT3, and JUN (Fig. 5H) in THP1 cells, and simultaneously enhance the activation status of STAT3 in intermediate resistant cells (Fig. 5I). Using H1581/BR intermediate resistant cells, we also revealed similar findings (Supplementary Fig. S6B and S6C). To reverse acquired resistance, we found that Hsp90 inhibitors and pan-HDAC inhibitors (LBH589 and TSA) could restore the sensitivity of resistant cells (Fig. 6A), suppress STAT3 phosphorylation (Fig. 6B), and suppress the secretome (Fig. 6C). Collectively, our data confirmed that fibroblasts and macrophages could exaggerate the effects of the tumor secretome to induce STAT3 activation and highlighted the involvement of the TME in mediating and enhancing resistance to FGFRi.

Previous mechanistic insights into acquired resistance have largely focused on the contribution of intracellular kinases; however, recently, the involvement of the TME in mediating acquired resistance has drawn much attention (27, 28). Most reports identified single factors in the secretome of cancer cells that dominate acquired resistance (5, 6, 27–30). In contrast, we revealed that the cancer secretome drives FGFRi resistance, which could be further enhanced through interactions with fibroblasts and macrophages in NCI-H1581 cells. The cancer secretome as an entity activates multiple mediators that subsequently activate STAT3 to elicit resistance. A similar mechanism was also observed in B-Raf inhibitor–resistant melanoma cells (8, 31). Although one report showed that NRAS and c-Met account for resistance in NCI-H1581 cells and treated these changes as a primary resistance mechanism, we did not observe similar results here. In addition, in contrast to our gradient of inhibitor concentrations, this report used long exposure times to obtain resistant clones, which we believed still reflected acquired rather than primary resistance responses (32).

Our findings reveal potential translational insights and therapeutic value. First, patients with NSCLC and breast cancer harboring the FGFR amplification do not respond well to FGFRi (33–35), and we propose that this may be due to the abundance of cancer-associated fibroblasts and macrophages. Second, we provided mechanism-based strategies to overcome FGFRi resistance. Although we found that targeting STAT3 could be an optimal strategy, no STAT-inhibiting agents have been utilized in clinical applications (36). We then utilized inhibitors targeting a wide range of either preclinical or on-market inhibitors (Supplementary Figs. S4C and S5A). Both Hsp90 inhibitors and pan-HDAC inhibitors (LBH589 and TSA) could reverse acquired resistance (Figs. 3H and I; 6A). However, as with many reports, the toxicity of these inhibitors resulting from their wide effects on intracellular kinases led to the inability to test our results in vivo, and whether Hsp90 and HDAC inhibitors act solely through the modulation of STAT3 still needs to be determined.

In summary, our findings will motivate future studies of STAT3-related secretome as potential markers for acquired resistance to FGFRi and, together with the work of others, shed light on further clinical strategies to optimize Hsp90i and pan-HDACi to overcome acquired resistance to FGFRi.

No potential conflicts of interest were disclosed.

Conception and design: J. Ai, J. Ding, M. Geng

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): X. Wang, H. Liu, X. Peng, H. Chen, Y. Chen, Y. Su

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): X. Wang, J. Ai, Y. Su, A. Shen, X. Huang, M. Geng

Writing, review, and/or revision of the manuscript: X. Wang, J. Ai, M. Geng

Study supervision: J. Ai, M. Geng

We are thankful to Dr. Jing Zhang who assisted us in dealing with the coculture model of in vivo assays. This work was supported by the National Natural Science Foundation of China (grant numbers 81773762, to J. Ai; and 91629104, to X. Huang), the “Personalized Medicines–Molecular Signature-based Drug Discovery and Development” Strategic Priority Research Program of the Chinese Academy of Sciences (grant numbers XDA12020000, to M. Geng; XDA12020100, to J. Ding; and XDA12020103, to J. Ai), the National Science & Technology Major Project “Key New Drug Creation and Manufacturing Program” of China (grant number 2018ZX09711002-011-013, to J. Ai), and the Science and Technology Commission of Shanghai Municipality (grant number 17431902900, to J. Ai).

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.