In vivo NSCLC acquired resistance to ceritinib is mediated by cell autonomous and non-cell-autonomous mechanisms.
To study the mechanisms of in vivo acquired resistance to ceritinib, we subcutaneously injected ALK-addicted NSCLC cells (H3122) in immunocompromised mice. When the tumors reached an approximate volume of 500 mm3, mice started a continuous treatment with ceritinib, at the highest tolerated dose, resulting in tumor shrinkage followed by disease stabilization. After this initial response, however, all the tumors started to regrow in the presence of the drug, displaying acquired resistance (RES-CER# tumors, Fig. 1A). Tumors grown in vehicle-treated animals represented matched controls (VEH# tumors, Fig. 1A). From resistant and control tumors we derived cells ex vivo, devoid of stroma cells contamination (for details see Methods and14). We then evaluated the sensitivity of the cells derived from RES-CER# tumor masses to increasing concentrations of ceritinib, and observed that 2 out of 5 cell lines (RES-CER#8 and RES-CER#15) were not resistant in vitro to the ALK TKI, showing an IC50 similar to control cells (Fig. 1B), thus suggesting that the mechanism of resistance in the two original tumor masses was likely non-cell-autonomous. In agreement with this hypothesis, RES-CER#8 and RES-CER#15 tumor cells did not display EMT features (Suppl. Figure 1A) nor any known mutation in the ALK kinase domain (data not shown).
CAF-increased GAL1 production sustains non-cell autonomous resistance to ceritinib.
To evaluate the role of the tumor microenvironment and, in particular, of cancer associated fibroblasts (CAFs) in the observed adaptive resistance to ceritinib, we derived ex vivo mouse CAF cultures from the RES-CER#15 tumor and from control tumors (for details see Methods and14). Unfortunately, we were not able to derive CAFs growing in culture form the RES-CER#8 tumor. H3122 parental cells were treated with increasing concentrations of ceritinib, in presence of RES-CER#15 CAF or VEH#11 CAF conditioned media. As shown in Fig. 2A, the supernatant of CAFs derived from the resistant tumor was able to confer resistance to ceritinib.
GAL1 activates ALK and its downstream pathway in spite of ceritinib treatment.
To identify which soluble factor(s) in the RES-CAF conditioned medium was (were) responsible for resistance, we performed a Real Time PCR screening for a panel of mouse growth factors that we and others found frequently involved in resistance to targeted agents: HGF, FGF2, IGF1, IGF2, TGFb, LGALS1. In CAFs derived from RES-CER#15, we identified a significant overexpression of the LGALS1 gene (encoding GAL1, Fig. 2B) and an increased secretion of GAL1 (Fig. 2C), compared to matched controls. Consistently, a significant increase in mouse LGALS1 gene expression was observed not only in the RES-CER#15 tumor mass (from which RES-CER#15 CAFs derived), but also in the RES-CER#8 tumor, suggesting the identity of the non-cell-autonomous mechanisms of resistance in the two independently generated models (Fig. 2D). It is known that, in CAFs, the NF-kB transcription factor can regulate the expression of genes mediating resistance.14 Indeed, we found that GAL1 overexpression in CAFs of resistant tumors was mediated by NF-kB, as treatment of CAFs with the NF-kB inhibitor IKK abrogated it (Supplementary Fig. 1B). We have recently demonstrated that GAL1, a soluble ligand of Neuropilin-1, known to regulate relevant cancer biological processes, is a driver of cell-autonomous, in vitro developed resistance to BRAF inhibitors in melanoma17. To evaluate if GAL1 was causatively involved in the adaptive resistance to ceritinib, parental (sensitive) H3122 cells were treated with ceritinib alone or in the presence of purified GAL1. As shown in Fig. 2E, we observed that GAL1 was sufficient to sustain drug resistance. Conversely, the ability of RES-CER#15 CAF medium to induce ceritinib resistance was lost in presence of GAL1 blockers (TDG or Lactose19, Fig. 2F, upper and lower graphs respectively), reinforcing the idea that GAL1 was the crucial mediator of non-cell-autonomous resistance to ceritinib.
Western blot analysis revealed that the presence of purified GAL1 preserved a partial ALK phosphorylation and activation of AKT/MAPK pathways in tumor cells, despite treatment with ceritinib (Fig. 3A). As in melanoma cells GAL1 drives resistance to BRAF inhibitors through EGFR activity17, we verified EGFR status in our model and found that EGFR phosphorylation, affected by ALK inhibition (likely for impaired transphosphorylation by ALK), was restored in the presence of GAL1 (Fig. 3B).
To evaluate the mechanistic relevance of these findings, we treated parental H3122 cells with ceritinib in presence of purified GAL1, in absence or presence of the EGFR tyrosine kinase inhibitor erlotinib; we observed that EGFR inhibition was sufficient to restore sensitivity to ceritinib, bypassing GAL1-induced resistance (Fig. 3C). Accordingly, erlotinib blunted the ability of GAL1 to activate the MAPK pathway in tumor cells treated with ceritinib (Fig. 3D). Importantly, GAL1 was able to mediate ceritinib resistance in an EGFR-dependent manner also in the STE-1 NSCLC cell line, bearing the same EML4-ALK fusion gene carried by the H3122 model (Supplementary Fig. 2).
To explore the possible therapeutic potential of these findings, we moved to in vivo experiments. When RES-CER#15 cells, sensitive to ceritinib treatment in vitro, were re-injected in mice, they gave rise again to tumors able to grow in presence of ceritinib (while, as expected, re-injected VEH- cells gave rise to drug-sensitive masses (Fig. 3E), indicating once more that, in this model, tumor microenvironment is required for resistance onset. However, if mice injected with RES-CER#15 were co-treated with ceritinib and erlotinib, the tumor masses did not grow, demonstrating that EGFR blockade was sufficient to efficiently overcome non-cell-autonomous resistance to ceritinib (Fig. 3F). Interestingly, when we re-injected control (VEH-) cells and subjected the mice to a prolonged treatment, the combo was also able to significantly postpone the onset of resistance compared to ceritinib monotherapy (Fig. 3G).
CAF-mediated adaptive resistance can evolve into cell-autonomous resistance
Two resistant tumors derived from RES-CER#15 cells re-injection and re-treatment with ceritinib were explanted at the end of the experiment; ex-vivo cells were successfully obtained (RES-CER#15/#09 and RES-CER#15/#17). As control, VEH#11/#04 cells were derived ex-vivo from one untreated tumor obtained from VEH#11 cells re-injection. When tested in vitro for their acute (72h) response to ceritinib, RES-CER#15/#09 and RES-CER#15/#17 showed again in vitro sensitivity to the ALK TKI, with an IC50 similar to control cells and to the two originally-derived cell lines, RES-CER#08 and RES-CER#15(Fig. 4A), suggesting that, also in this case, the mechanism of resistance was likely non-cell-autonomous. CAFs derived from RES-CER#15/#09 and RES-CER#15/#17 tumors overexpressed GAL1 at levels comparable with the original RES-CER#15 CAFs (Fig. 4B). However, when all the ex-vivo derived cells were subjected to a prolonged (25 days) in vitro treatment with ceritinib, only RES-CER#15/#09 and RES-CER#15/#17 (but not VEH-derived cells) were able to acquire cell-autonomous resistance to the ALK TKI in this short time-frame (Fig. 4C), demonstrating that, under the continuous pressure of the drug, adaptive resistance can more likely evolve into cell-autonomous resistance. The cells derived from RES-CER#15/#09 and RES-CER#15/#17 and subjected to 25 days of treatment (named ‘RES-CER#15/#09_25days CER and RES-CER#15/#17_25days CER) were passaged in vitro and tested for ceritinib response in the ‘canonical’ acute drug treatment (72 hours). As shown in Fig. 4D, these cells showed resistance to ceritinib, thus definitively demonstrating that they had acquired cell-autonomous resistance to the drug. Consistently, in these cell-autonomous resistant cells, in the absence of CAF medium, MAPK signaling was constitutively more active than in the corresponding non-cell autonomous models of resistance, and MAPK activation was no longer modulated by ceritinib treatment (Fig. 4E). Real-time PCR analysis revealed a copy number gain of the translocated ALK gene in cell-autonomous resistant cells (approx. 6 ALK gene copies, detected only using a probe located within exon 29, but not using a probe for exon 1, which is lost in the EML4-ALK fusion). This result indicates that CAF-mediated adaptive resistance can evolve in genetically sustained resistance (Fig. 4F).
Stromal GAL1 is overexpressed in patients progressed upon ALK TKIs
To assess the clinical relevance of these findings, we evaluated GAL1 expression in three paired FFPE NSCLC samples, obtained at diagnosis (‘BASAL’) and upon onset of resistance to ALK TKIs (‘RESISTANT’). The tumor samples were analyzed through in situ hybridization for human GAL1 mRNA levels, followed by pan-cytokeratin IHC to differentiate cancer/stroma signal. In 3 out of 3 patients, we found a > 3-fold increase in the GAL1 signal in the tumor stroma upon resistance acquisition (Fig. 5A,B), indicating that GAL1 overexpression could be a frequent event in resistance onset to ALK TKIs. Interestingly, for one of the relapsed tumors, no genetic mechanism of resistance has been identified through Next Generation Sequencing analysis. The molecular analysis of the tumor biopsies revealed that the other two relapsed tumors had acquired, besides higher levels of stromal GAL1, an ALK mutation already described in conferring resistance to ALK TKIs8 (Fig. 5C). These data suggest that, also in patients, the non-cell-autonomous, adaptive resistance mediated by stromal GAL1 might be an intermediate step versus the acquisition of genetically sustained resistance.