Background
MET is the receptor tyrosine kinase (RTK) for scatter factor/hepatocyte growth factor (SF/HGF). Aberrant activation of the SF/HGF-MET axis is common in a wide range of human malignancies and promotes cell growth, invasiveness, metastatic potential, angiogenesis, and resistance to cell death following chemotherapy and radiotherapy [
1‐
7]. The established relevance of SF/HGF-MET in tumorigenesis and progression suggests that MET inhibition is a potential strategy in cancer therapy, a notion supported by a growing body of preclinical evidence. Moreover, several strategies to impair SF/HGF-MET aberrant activation are currently under investigation in numerous clinical trials [
7,
8], and cabozantinib (MET, RET, and VEGFR2 multikinase inhibitor) received FDA approval for the treatment of medullary thyroid carcinoma [
9].
Paralleling the preclinical and clinical developments in the field of RTK-targeted therapy, awareness on potential mechanisms of therapeutic resistance is an issue of major relevance. One such resistance mechanism is the presence of activating mutations affecting different members of signaling pathways downstream of RTKs, such as the MAPK and the PI3K pathways. The actual clinical relevance of this phenomenon was clearly outlined by Karapetis et al. [
10], who assessed the impact of activating mutations in exon 2 of
KRAS in cetuximab-treated patients with colorectal cancer. The authors found that 42.3% of the tumors harbored at least one
KRAS mutation. Cetuximab was beneficial in terms of overall survival only in patients with wild-type
KRAS. In order to optimize patient selection in this subgroup of patients, assessment of
KRAS mutational status is currently performed prior to cetuximab therapy [
10]. In a later study, Douillard et al. demonstrated that other
KRAS mutations (exon 3 and 4) equally had an impact on responses to panitumumab in patients with colorectal cancer [
11].
Mutations along the MAPK pathway are well-established sources of resistance to RTK inhibitors, particularly against EGFR targeted therapies [
12]. Importantly however, recent comprehensive genomic approaches have revealed a high prevalence of mutations in members of the PI3K pathway in various human cancer types, including head and neck cancer (HNC), breast cancer, or endometrial carcinoma [
13‐
15]. These mutations are most commonly encountered in the
PIK3CA-encoded p110α catalytic subunit of PI3K (PIK3CA) and tend to cluster at hotspots E542 and E545 within the helical domain, as well as H1047 in the kinase domain. Previous in vitro studies have shown that PIK3CA hotspot mutations lead to increased kinase activity and confer variable oncogenic features [
14,
16]. More specifically, Meyer et al. showed that both PIK3CA
E545K and PIK3CA
H1047R were tumorigenic in mouse mammary models [
17]. However, PIK3CA
E545K-induced tumors developed more slowly and were histologically less aggressive than PIK3CA
H1047R-driven tumors. In contrast to these findings in mammary tumors, Trejo et al. [
18] reported that presence of PIK3CA
H1047R alone was not sufficient to promote initiation in lung cancer models. Instead, co-presence of BRAF
V600E was required for initiation. In such context both mutations cooperated in tumor progression.
The PI3K-AKT-mTOR pathway is a critical regulator of cell metabolism, proliferation, survival, apoptosis and cell motility [
19]. PI3K activity can be countered by its negative regulator phosphatase and tensin homologue (PTEN), whose loss-of-function is common in cancer [
20]. Upon PI3K activation, AKT is recruited to the cell membrane and phosphorylated. In turn, phosphorylated AKT activates an extensive signaling network, resulting in cell growth by promoting ribosome biogenesis, protein translation, synthesis of lipids and nucleotides, and regulation of autophagy [
13,
21]. Moreover, AKT exerts a pro-survival activity by phosphorylating MDM2 and thus inhibiting p53-regulated cell-cycle arrest and apoptosis [
19,
22].
The implications of synchronous presence of aberrantly active MET and PIK3CA mutations in cancer has not been previously explored. Therefore, we hypothesized that in preclinical models of MET-driven tumors sensitive to MET inhibition, PIK3CA mutations may confer resistance to MET inhibition. Our data demonstrate that PI3K signaling remains active upon MET inhibition in tumors harboring PIK3CA mutations, rendering such tumors resistant to MET inhibition. In vivo tumor models harboring PIK3CAH1047R are equally more resistant to MET inhibition than their wild-type counterpart. In all cases, PIK3CA-induced resistance to MET inhibitors could be reverted by co-targeting PI3K.
Complementarily, we sought to assess the impact of dual MET/PI3K inhibition in models displaying ligand-induced MET activation and PI3K mutations. For this purpose, we selected cellular models of HNC endogenously harboring
PIK3CA mutations. In this particular disease,
PIK3CA are commonly encountered (over one-third of cases), in contrast with the more seldom MAPK or JAK/STAT mutations [
23]. Moreover, MET gene copy number gain or amplification is found in around 20% of HNCs, and protein overexpression in more than 80% of the cases [
4]. However, MET constitutive activation is rare in HNC, and receptor activation occurs mainly through oncogenic ligand-receptor loops [
24]. In HNC models, combination of MET and PI3K inhibitors was effective in abrogating invasive features regardless of PIK3CA mutational status, but had synergistic effects especially in PIK3CA-mutated cell lines.
Methods
Cell lines, plasmids and transfections
NIH3T3 cells stably expressing MET-activating mutation M1268T were provided by Dr. Laura Schmidt (NCI, Frederick, MD, USA) and maintained as previously described [
25].
NIH3T3 MET M1268T cells were transfected using standard reverse transfection protocols with Lipofectamine 2000 (Invitrogen), using the following plasmids: PIK3CA E545K and H1047R (Addgene plasmids #12524 and #12525 respectively, from Jean Zhao [
26]), as well as pBABE control vector (Addgene plasmid #14738,from Adrienne Cox [
27]). Stable transfectants were selected in the presence of puromycin (1.5 μg/mL).
HNC cell lines FaDu (PIK3CA wild-type) and Detroit-562 (PIK3CA H1047R) were obtained from the American Type Culture Collection (Manassas, VA, USA) and cultured in Minimum Essential Medium (MEM, Sigma) supplemented with FCS 10%, antibiotic-antimycotic, and non-essential aminoacids (NEAA 1% vol/vol; Sigma). SCC-61 (PIK3CA E542K) cells were provided by Prof. M. Pruschy (University of Zurich, Switzerland) and cultured in DMEM with supplements [
28,
29]. No cell line authentication was performed by the authors for this study.
Inhibitors and growth factor treatments
The MET tyrosine kinase inhibitor tepotinib (EMD1214063, MSC2156119J) was kindly provided by Merck KGaA (Darmstadt, Germany). PI3K was inhibited with pictilisib (GDC-0941; AbMole BioScience, Hong Kong). Both drugs were dissolved in DMSO and stored at−20 °C. DMSO concentrations were normalized in all experimental conditions.
Recombinant human SF/HGF was purchased from R&D Systems and prepared following the manufacturer’s instructions. Treatments were carried out as described in Results.
Immunoblotting and immunoprecipitation
Cells were lysed with Non-ident NP40 buffer as previously described [
25]. Xenograft tissues were lysed similarly, including a previous step of mechanical disruption. Total protein concentrations were determined with the Bio-Rad protein quantification reagent (Bio-Rad Laboratories, Inc.). Equal amounts of protein (30–50 μg) were resolved by SDS-PAGE on 7–12% gels under reducing conditions. Separated proteins were transferred onto PVDF membranes, blocked with 5% milk or BSA in TBS/T, and incubated overnight with the following primary antibodies: p-Y1234/Y1235 MET, p-Ser473 AKT, p-Thr202/Tyr204 ERK1/2, p-Ser235/236 S6, pan-AKT, and pan ERK (all from Cell Signaling Technology). Anti-β-actin antibody was obtained from Millipore Corporation. The anti-HA high affinity antibody (clone 3 F10) was purchased from Roche.
Membranes were incubated with appropriate secondary antibodies and signals were detected with the ECL kit (Amersham Pharmacia Biotech) or with infrared fluorescence on an Odyssey imager (Li-Cor biosciences).
Immunoprecipitation was performed using the Dynabeads® Protein G kit following manufacturer’s instructions (ThermoFisher Scientific, #10003D).
With the exception of samples from in vivo tumors and organotypic slices, immunoblots shown are representative of at least three independent experiments.
Cell viability/toxicity assays, determination of EC50 values and combinatorial effects
Cell viability was determined using a resazurin sodium salt reduction assay (Sigma). Briefly, after treatment for 48 h, cells were supplemented with medium containing 44 μM resazurin. Resazurin reduction was colorimetrically measured 1 and 6 h later (570/600 nm) with a Tecan Reader (Tecan Group Ltd.). Results were normalized to vehicle-treated controls and represent the mean of at least three independent experiments.
For half maximal effective concentration (EC50) value determination, resazurin assays were carried out after 72 h of treatment with increasing doses of tepotinib and pictilisib (0 to 10 μM). EC50 values were determined using the GraphPad software (version 5.03).
The combinatorial effects between MET and PI3K inhibition were estimated by Bliss independence analyses as previously described [
30]. Excess over Bliss values equal to 0 indicate additive effects, >0 activity greater-than-additive, and <0 combination less-than-additive.
Cell death and viability were assessed with the Live/Dead Assay Kit (Molecular Probes). For this assay, cells were treated for 3 days and stained with green-fluorescent calcein-AM (viable cells) and red-fluorescent ethidium homodimer-1 (dead cells). Images from four independent experiments were captured under a fluorescent microscope (Leica DC 300 F) and number of live and dead cells quantified using ImageJ (imagej.nih.gov/ij/).
Caspase-3 enzymatic activity was determined via a fluorogenic assay based on the caspase-3 specific substrate Ac-DEVD-AMC (Calbiochem). The substrate was added to cell lysates after 72 h of treatment and fluorescence was measured at excitation and emission wavelengths of 380 nm and 460 nm respectively, with an Infinite 200 plate reader (Tecan Group Ltd.). Caspase-3 activity was normalized to protein content and results shown are representative of at least three independent experiments.
Cells were plated the day before treatment at densities ranging from 500 to 2500 cells/well in 6-well plates and then treated with the indicated drug concentrations. Colonies were allowed to form for 5–7 days and subsequently fixed and stained with 2% crystal violet dissolved in 1:3 methanol and 2:3 acetic acid (v:v). Colonies (>50 cells) were scored and quantified using the “Analyze Particles” plug-in of ImageJ (imagej.nih.gov/ij/). Experiments were performed three times.
Wound-healing assay
Cell migration was assessed with the Oris
TM
Cell Migration Assembly Kit (AMS Biotechnology). Cells were seeded in 96-well plates containing cell stoppers (which create a 2 mm central circle) and left to attach overnight. Thereupon, cell stoppers were removed and cells were treated as indicated. Pictures were captured at baseline and after 48 h using a Leica DC 300 F microscope. Images obtained from at least three independent experiments were quantified with the ImageJ software (imagej.nih.gov/ij/). Results are presented as number of invading cells for NIH3T3 cells (scattering invasion pattern) or as percentage of wound-closure relative to baseline for HNC cell lines (pushing-front invasion pattern).
In vivo tumor growth delay experiments and generation of organotypic tissue cultures
All animal experiments were conducted in strict compliance with Swiss Federal guidelines. Vector and PIK3CA
H1047R-transfected NIH3T3 MET
M1268T cells were injected subcutaneously in both flanks of 7 to 9 weeks-old female nude mice (RjOrl:NMRI-
Foxn1
nu
/
Foxn1
nu
, Janvier Labs, Le Genest-Saint-Isle, France). Tumor growth was estimated by regular caliper measurements using the following formula:
$$ V = \left( L\ x\ {W}^2\right)/ 2 $$
where V = tumoral volume, L = larger dimension and W = shorter dimension. Upon tumors reaching an average of 150–200 mm3, animals were randomly allocated to one of four treatment groups (four animals per group): vehicle (Solutol HS 15, BASF ChemTrade GmbH), tepotinib 50 mg/kg alone, pictilisib 50 mg/kg alone, and combination of both drugs. Drugs were administered via oral gavage and daily measurements were recorded. Animals were euthanized 4 h after the last treatment, either at the scheduled endpoint or prematurely when interruption criteria were met (tumor volume superior to 1 cm3 or persistent tumor bleeding). For determination of in vivo proliferation, 5-bromo-2′-deoxyuridine (BrdU; 2 mg dissolved in 0.9% NaCl) was injected intra-peritoneally 4 h before euthanasia.
As a complementary assessment method, we generated organotypic tissue cultures as previously described [
31] from freshly removed fragments of vehicle-treated tumors in each group.
Generation of tissue arrays, BrdU incorporation in vivo, and immunofluorescence
Tumor xenograft tissue arrays of 5 mm were generated using the EZ-TMA Manual Tissue Microarray Kit 4 (IHC World). Sections of 5 μm were then cut, deparaffinized and rehydrated. Antigen retrieval was performed by heating slides for 10 min in a microwave oven (600–700 W) in Tris-EGTA buffer (pH 9.0). Sections were incubated overnight at 4 °C with specific primary antibodies: BrdU (1:300), pS6 (1:800), and p-MET (1:150). BrdU was purchased from Abcam and the remaining antibodies from Cell Signaling Technology.
Slides were washed and incubated with secondary antibodies from Life Technologies (goat anti-rabbit 488 and goat anti-rat 555), 1:500, for 1 h at room temperature. Slides were scanned using the Pannoramic Midi digital slide scanner and analyzed by CellQuant software (3DHISTECH Ltd).
Data analysis
Statistical analysis and graphic representation of data was performed with GraphPad (version 5.03). Data are presented as relative averages ± SD or fold-induction ± SD, as indicated. Statistical comparisons were performed using the Mann-Whitney U-test unless otherwise reported. Tumor sizes were compared by one-way ANOVA. Statistical significance was set at p < 0.05. Definitions for p-values are as follows throughout the manuscript: *p = 0.01–0.05, **p = 0.001–0.01, ***p < 0.001. Absence of labels implies non-significant differences.
Discussion
The developments in the field of molecular targeted therapy have led to an unparalleled range of opportunities to therapeutically target a considerable number of oncogenes or oncogenic pathways in a variety of human cancers. In spite of the attractive rationale behind molecular targeted therapy and promising preclinical data, clinical experience has clearly shown that single target inhibition is rarely effective. As a case in point, single use of EGFR inhibitors does not lead to significant tumor remission or prolonged stabilization rates in unselected patients with colorectal cancer, non-small cell lung cancer, breast cancer, or HNC among other entities [
29,
36‐
39].
Emerging and established evidence currently shows that a major element hindering effective development and clinical implementation of anti-RTK targeted approaches is the presence of genomic alterations, mainly mutations, in cellular signal transduction pathways [
40]. The clearest example, as mentioned above, is the predictive value of pre-therapeutic screening of KRAS mutations in patients with colorectal cancer prior to cetuximab therapy, an approach allowing individually-tailored therapeutic strategies [
10,
11].
In addition, it has been clearly established that aberrant activation of members in the MAPK or the PI3K pathways result in resistance to EGFR inhibitors. For instance, Wang et al. [
41] demonstrated that both PIK3CA and KRAS mutations resulted in resistance to cetuximab when ectopically expressed in initially-sensitive cells. Co-targeting mTOR with rapamycin or everolimus led to successful sensitivity restoration. Along the same lines, Young et al. [
29] found that cell lines harboring PIK3CA mutations were generally more resistant to EGFR inhibitors such as erlotinib or gefitinib. Co-targeting EGFR and PI3K resulted in enhanced tumor growth control in vivo than either inhibitor alone. Moreover, with respect to MET signaling, we recently demonstrated that HRAS and KRAS mutations confer differential degrees of resistance to MET tyrosine kinase inhibitors in MET-expressing preclinical in vitro and in vivo models [
35].
This ensemble of observations further stresses the importance of pre-therapeutic molecular stratification. Relevant to this point, recent comprehensive genomic approaches have revealed variable prevalence rates of mutations affecting PI3K and MAPK pathways in different cancers. These results have major implications for implementing rational combinatorial approaches in specific cancer types. The Cancer Genome Atlas (TCGA) datasets available at the cBio Portal for Cancer Genomics (
http://www.cbioportal.org/public-portal/cross_cancer.do; [
42,
43]) reveal KRAS mutation rates of over 90% in pancreatic cancer versus PIK3CA mutation rates below 10%. In contrast, entities such as breast cancer or HNC display a predominance of PIK3CA mutations (around 40%, for an overall mutation rate of KRAS, NRAS and HRAS inferior to 5% in both cases). An intermediate group with prevalent mutations on both pathways would include endometrial carcinoma (PIK3CA 57%, KRAS 21%), colorectal adenocarcinoma (PIK3CA 31%, KRAS 51%), or gastric adenocarcinoma (PIK3CA 24%, KRAS 16%).
These figures are globally rendered more complex when considering different subgroups of patients with different subtypes of a same cancer entity. With respect with breast cancer for instance, KRAS mutation rate in metastatic breast cancer was estimated at approximately 12% [
44]. Along the same lines, KRAS mutations were detected in only 2% of luminal A versus 17.4% in luminal B breast cancer [
45], suggesting accumulation of KRAS mutation in more aggressive forms of disease.
Another important aspect is that different tumor entities have variable degrees of dependence on signaling inputs from given RTKs. For this reason, in the present study we sought to assess differences between models featuring MET constitutive activation and models with ligand-induced activation. Our findings indicate that ectopically-expressed PIK3CA hotspot mutations confer resistance to MET inhibition in initially-sensitive MET-driven models, both in vitro and in vivo. In ligand-dependent HNC cellular models, dual MET and PI3K blockade was superior to single inhibition on downstream signaling and wound-healing ability. Importantly however, stronger synergism between MET and PI3K inhibition was observed only in HNC cells harboring PIK3CA mutations. Importantly, the effect of MET inhibition was not only restored but also enhanced by co-targeting PI3K in both sets of models.
The differences seen between MET-driven models and non-MET-driven models endogenously harboring PIK3CA mutations are most relevant also in terms of pre-therapeutic stratification. It is increasingly evident that human cancers and their derived preclinical models display differential degrees of dependence on signaling by specific RTKs as well as other kinases. Therefore, while some malignancies quite uniquely rely on one given RTK, the same is not true for some other cancer types. For instance, with respect to MET, McDermott et al. [
33] assessed responses to the MET inhibitor PHA665752 in a panel of 500 cancer cell lines. These authors identified a subset of highly-responsive cell lines, predominantly of gastric cancer and NSCLC origin, that displayed
MET-gene amplification and were extremely sensitive to MET inhibition. Further in the same direction, small cohorts of patients who experience drastic and prolonged disease remission after MET inhibition indicate that at least subgroups of gastric cancers are indeed strongly dependent on MET signaling [
32,
46]. Such reliance on a given oncogene has been termed
oncogene addiction [
47]. Classically, oncogene addiction has been associated with underlying gene amplification, but other types of aberrations such as activating mutations may also be the cause of RTK-driven cell growth, transformation, and responses to molecular targeted therapy [
5]. Most illustrative of this point is the existence of EGFR mutations which determine both dependence on EGFR signaling and responses to EGFR inhibitors in patients with lung cancer [
48].
In contrast, entities like HNC do not feature oncogene addiction for any given RTK, and tend to rely more on ligand-receptor oncogenic loops, as well as on RTK cooperation and redundant signaling [
4,
24,
49].
Our study also underlines significant differences in the way H1047R and E542K/E545K influence responses to MET inhibition. As mentioned above, while it has been previously shown that H1047R has a superior oncogenic potential than E545K in transgenic mammary cancer models [
17], the potential of different PIK3CA mutations in conferring resistance to upstream inhibition has not been previously explored. Our current results indicate that PIK3CA
H1047R is a stronger effector of resistance to MET inhibitors, but further translational studies should address this issue given the variable prevalence of helical or catalytic domain mutations seen in different human malignancies (
http://www.cbioportal.org/public-portal/cross_cancer.do, [
42,
43]).
Specifically regarding the potential relevance of MET/PI3K combinations, in the context of a retrospective review of over 2000 patients, Maryam et al. [
50] reported frequent MET-PI3K co-mutations in colorectal adenocarcinoma, thyroid carcinoma, breast adenocarcinoma, and HNC. To our knowledge, only one study previously focused on combinations of ARQ 197 (MET inhibitor) and NVP-BEZ235 (dual PI3K/mTOR inhibitor) in malignant pleural mesothelioma, suggesting a benefit of this combinatorial approach [
51]. This study however did not assess the impact of PIK3CA mutations on MET inhibition.
Our current study provides evidence for the first time supporting the notion of co-targeting MET and PI3K as a potential therapeutic strategy in common clinical scenarios, featuring both constitutive and ligand-induced MET activation along with PIK3CA mutations.
While the combinatorial strategy we evaluate here could be therapeutically exploited, it is crucial that upcoming clinical trials consider the characteristics described above, namely mutational status of downstream signaling pathways and activation status of selected RTKs, in order to identify the subgroups of patients most likely to benefit from treatment.
Acknowledgements
The outstanding technical support of Dr. Aurélie Quintin, Mr. Bruno Streit and Mrs. Melanie Leuener-Tombolini is gratefully acknowledged.