TAM family receptors in conjunction with MAPK signalling are involved in acquired resistance to PI3Kα inhibition in head and neck squamous cell carcinoma

Cell lines were obtained from the sources listed (Additional Table 1). We previously used short tandem repeat profiling (The Center for Applied Genetics; Toronto) to confirm cell line identities [20]. 93-VU-147T cells were cultured in DMEM/F12, with 10% fetal bovine serum (FBS; GIBCO), penicillin (100 IU/mL; Invitrogen) and streptomycin (100 μg/mL; Invitrogen). Cal33 cells were cultured in DMEM, with 10% heat-inactivated FBS, 1x non-essential amino acids (Wisent), penicillin (100 IU/mL) and streptomycin (100 μg/mL). Resistant cell lines were obtained after chronic exposure to increasing concentrations of alpelisib for 3–4 months [22]. All cells were maintained in a 37 °C humidified atmosphere at 5% CO₂. The inhibitors alpelisib and BI-D1870 were purchased from Selleckchem. Compounds were dissolved in DMSO for in vitro experiments.

Mice were handled in accordance with the AUP 1542 approved by the University Health Network Animal Care Committee and in accordance with the CCAC regulations. Xenografts were established and handled as described previously [21]. Details are provided as Supplemental Methods & Materials.

Once tumor volumes reached 80-120 mm³ mice were randomized to either daily (5x/week) alpelisib (Novartis; 50 mg/kg) by oral gavage or a vehicle control (corn oil) [12, 21]. Individual tumor volumes were calculated using the formula: [length x (width)²] × 0.52. Where possible, STR profiling was used to confirm matching identities of primary tumors, xenograft tumors, patient blood and PDX-derived cell lines where available (Additional Table 2). Tumors were classified as HPV-positive using immunohistochemistry (IHC) for p16.

Cells were seeded in 96-well plates at 2400 cells/well and cultured overnight. Drugs were then added over 10-point ranges (0-40 μM). Viability was determined 72 h later using the PrestoBlue® Reagent (Thermo Fisher Scientific) on a Synergy™ H4 Hybrid Reader (BioTek) with 560 nm excitation and 590 nm emission wavelengths. For each dose, viability values were normalized to no-drug controls and average viability for each dose was calculated. To determine the half-maximal inhibitory concentration (IC₅₀) values, normalized relative fluorescence values of drug-treated replicates were calculated as a percentage of the mean RFU of the control replicates and then drug doses were transformed to a logarithmic scale. IC₅₀ values were subsequently calculated by non-linear regression. Values are plotted as mean + standard deviation (SD) using Prism® 7 GraphPad Software.

Parental and resistant cell lines were counted and seeded at 500 cells per well into 24-well dishes. Cells were allowed to adhere for 48 h and then were treated with media containing 5 μM alpelisib. For the next 7–14 days, cells were monitored and media replaced every 3 days until visible colonies were formed. Colonies were rinsed with 1x PBS, fixed with cold 100% methanol (MeOH) and stained with 0.5% crystal violet in 25% MeOH/1x PBS. The colonies were then gently washed with water and air-dried. Visible colonies were counted.

Cells were prepared for reverse phase protein arrays (RPPAs) as follows: 10 cm plates were washed twice with cold 1x PBS. Cold lysis buffer (containing: 1% Triton X-100, 50 mM HEPES pH 7.4, 150 mM NaCl, 1.5 mM MgCl2, 1 mM EGTA, 100 mM NaF, 10 mM Na pyrophosphate, 1 mM Na3VO4, 10% glycerol and 1% freshly-added protease and phosphatase inhibitors) was added to the plates which were then incubated 20mins on ice with occasional shaking. Lysed cells were centrifuged at 14000 rpm for 10mins at 4 °C. Protein concentration determined by Bradford Assay. Lysates were combined with sample buffer (40% glycerol, 8% SDS, 0.25 M Tris-HCl pH 6.8 and 1/10 volume β-mercaptoethanol –added just before use) at 3 parts lysate:1 part sample buffer. Samples were boiled for 5 mins and stored at − 80 °C.

Samples were submitted to MD Anderson’s Functional Proteomics RPPA Core Facility. Briefly, lysates were serially diluted and arrayed onto nitrocellulose-coated glass slides. Samples were probed with 307 antibodies and visualized by DAB colorimetric reaction. Slides were then scanned and spot densities quantified by Array-Pro Analyzer. All data points were normalized for protein loading and transformed to a linear value. Resistant replicates were then normalized to the mean of their respective parental replicates. Values were then log2-transformed and we restricted our analysis to the top 50% of differentially-expressed proteins for each cell line. Unsupervised hierarchical clustering was performed using the average agglomeration method and Euclidean distance measurements. Clustering was performed in R using the ComplexHeatmap package (version 2.1.1).

Cell lysates were prepared for immunoblotting as described previously [23]. A list of primary antibodies used is provided in Additional Table 3. Membranes were visualized following exposure to enhanced chemiluminescence reagent (Luminata™ Crescendo or Luminata™ Forte, Western HRP Substrate; Millipore) on a Bio-Rad ChemiDoc™MP Imaging System.

ImageJ was used to select and determine the background-subtracted density of the bands in all immunoblots. Values were then normalized to their corresponding α-tubulin band. All values are presented below the associated band.

TMAs were constructed for two of the xenograft models. In brief, the FFPE block for each tumor was sectioned and stained with hematoxylin & eosin (H & E) to confirm the presence of human tumor. Guided by these sections, a Manual Tissue Arrayer (MTA-1; Beecher Instruments Inc.) was used to punch out 3–4 cylindrical cores of 0.6 mm diameter from each sample. Cores were arrayed into recipient paraffin blocks. Eleven control tissues (tonsil, stomach, prostate, pancreas, lung, kidney, skin, thyroid, spleen, adipose, liver) were also included on each block. Cores were sealed into recipient blocks by heating at 40 °C for ~40mins. Blocks were sectioned into 1.5 μM sections and affixed to glass slides. Every ninth slide was stained with H & E to provide a reference. Additional details are available in the MTA-1 Instruction Manual (www.​beecherinstrumen​ts.​com). IHC staining was completed in collaboration with the Department of Pathology & Laboratory Medicine and the Molecular Pathology Core Facility (Western University). Tissues were examined using an Aperio ScanScope® slide scanner and staining quantification was performed using the Fiji plugin for ImageJ.

Knockdown of AXL and TYRO3 was performed using specific pooled siRNAs purchased from Dharmacon (Cat No’s. L-003104-00-0005 and L-003183-00-0005, respectively), as described previously [23]. Scrambled control siRNA (siCT) (Thermo Fisher Scientific; Cat No. 4390843) was also used. Knockdowns were confirmed by immunoblotting.

For drug testing, cells were seeded into 96-well dishes at 2400cells/well. Alpelisib was added the next day at 5 μM and cells were incubated for 72 h. Cell viability was then determined indirectly using the PrestoBlue® Reagent (Thermo Fisher Scientific) on a Synergy™ H4 Hybrid Reader (BioTek) with 560 nm excitation and 590 nm emission wavelengths. For each condition, alpelisib-treated cells were compared with normalized untreated cells to determine the relative effect of RNAi-mediated knockdown.

Using cells dissociated from first-passage xenograft tumors, we attempted to establish cell lines from the patient tumors that were used to generate the PDX models (Additional Fig. 1a). Specifically, the generation of cell lines was attempted from tumor tissues that were never treated with alpelisib nor the vehicle agent. A cell line (called PDX-C Cell Line) was successfully established from one model, PDX-C (Additional Fig. 1b). STR profiling, immunoblotting and flow cytometry for cell surface expression of EpCAM (CD326) were all completed as described previously, validating the line as a human epithelial line from the same patient as the PDX-C tumor (Additional Fig. 1c, Additional Table 2).

All analyses were performed with Prism® 7 GraphPad Software. Experimental groups were compared with controls using Student’s unpaired, two-tailed t-tests. Multiple groups were compared across a single condition using one-way ANOVA. P < 0.05 was used to define significant differences from the null hypothesis.

Prior to exploring resistance to PI3K inhibition, we first validated the efficacy of alpelisib in HNSCC cell lines. Alpelisib treatment reduced signalling through both the PI3K and MAPK pathways in Cal33 (PIK3CA mutant) cells and 93-VU-147T (PIK3CA amplified) cells (Fig. 1a), as indicated by reduced levels of phosphorylated (p)-Akt (Thr308), pERK1/2 (Thr202/Tyr204) and pP90RSK (Ser380) [21]. Based on previous studies, we hypothesized that the mechanism of action for alpelisib in HNSCC would involve cell cycle arrest [7]. Following 24 h of treatment with alpelisib, we observed a significant reduction in the proportion of proliferating (S-phase) cells (Fig. 1b) [24, 25]. To determine whether alpelisib was also able to induce cell death through apoptosis, we examined PARP cleavage (Fig. 1c). Following alpelisib treatment, cleaved PARP was readily detectable. Of note, we have also previously characterized the sensitivity of 28 HNSCC cell lines to PI3K inhibition and found both wildtype and PIK3CA-altered cell lines to be sensitive to PI3K inhibition [21].

To identify pathways associated with acquired resistance to alpelisib, we exposed two genomically-distinct HNSCC cell lines (Fig. 2a) to increasing concentrations of alpelisib over a 3–4 month period (schematic shown in Additional Fig. 2), ultimately yielding cell lines significantly more resistant to alpelisib than their parental counterparts (Fig. 2b-c) [21]. To verify the durability of the resistant cell lines, parental and resistant cells were challenged to grow as single cell colonies in the presence of alpelisib. Whereas alpelisib treatment led to a significant reduction in the number of colonies formed by both parental cell lines, it was much less effective in resistant cell lines (Fig. 2d). Although the alpelisib-resistant 93-VU-147T cells exhibited a modest reduction in colony formation following alpelisib treatment, the difference was much less than that of the parental line. Thus, cell lines treated for a prolonged period with alpelisib exhibit increased tolerance for this PI3K inhibitor. In addition, we examined the relative abundance of colony sizes (small, medium, large) in parental versus resistant cell lines as this reflects cell lines’ proliferative status (Additional Fig. 3a-b). Finally, to compare the ability of alpelisib to induce cell death through apoptosis, we examined PARP cleavage following alpelisib treatment in parental and alpelisib-resistant cell lines (Fig. 2e). We found that following alpelisib treatment, cleaved PARP was readily detectable in parental cells, but was less abundant in alpelisib-resistant cells, suggesting that alpelisib treatment is less effective at inducing cell death via apoptosis in cells that have acquired resistance to alpelisib.

To determine whether alpelisib continued to block signalling through the PI3K pathway in drug-resistant cells, we used immunoblotting to examine Akt phosphorylation following alpelisib treatment in parental and resistant cell lines. In parental and drug-resistant cell lines alike, alpelisib treatment suppressed Akt Thr308 and Ser473 phosphorylation, indicating that resistance was acquired via a compensatory mechanism and not loss of direct inhibition of PI3K activity (Fig. 3a).

To examine a broad array of signalling pathways that could mediate resistance to PI3K inhibition in HNSCC cells, we performed RPPAs on lysates from Cal33 and 93-VU-147T parental cells and resistant cell lines (Fig. 3b). In both resistant cell lines, RPPAs suggested that expression of the membrane-bound RTK AXL was elevated relative to parental cells. As mentioned, AXL has been previously shown to mediate resistance to various anti-cancer agents, including EGFR, HER2 and PI3K-targeted therapies [12, 26‐28]. AXL is part of a three-member RTK sub-family known as the ‘TAM family’ of receptors (TYRO3, AXL, MER-TK) [29‐32]. Interestingly, expression of TYRO3 was also found to be elevated in both Cal33 and 93-VU-147T alpelisib-resistant cells. To our knowledge, TYRO3 has never been implicated in PI3K-inhibitor resistance, nor in HNSCC as an effector of therapy response. In general, much less is known about TYRO3, including the role it plays in cancer development and progression [31‐34].

We confirmed elevated protein expression of AXL and TYRO3 in alpelisib-resistant Cal33 and 93-VU-147T cells, relative to parental cells by immunoblotting (Fig. 3c). We also examined expression of the third member of the TAM RTK family, MER-TK, which was not included in the RPPA. MER-TK was weakly detectable in both HNSCC cell lines, in contrast to its high expression in HEK293T cells (long exposure blots shown in Additional Fig. 4a-b). As MER-TK was only weakly detected in both cell lines and no difference in expression was apparent between parental and drug-resistant cells, we did not examine it further. Based on our findings, we hypothesized that overexpression of TYRO3 and/or AXL receptors could play a role in mediating resistance to PI3Kα inhibition.

In parallel with our cell line models, we generated 5 PDX models from HNSCC patient tumors (clinical characteristics outlined in Additional Table 4). Of the 5 models, one (PDX-B) was both HPV-positive and contained a non-synonymous PIK3CA mutation (E545K) [21]. The remaining models were all PIK3CA wildtype. Several of the PDX models contained mutations frequently seen in HNSCC, including TP53 mutations, EGFR amplifications and CDKN2A alterations [4, 21]. Histological comparison of PDXs and their corresponding primary tumors (where available) revealed a high degree of similarity in their cellular morphology (Additional Fig. 5). In all 5 PDX models, treatment with alpelisib significantly suppressed tumor growth for the first 20–35 days, relative to the vehicle agent (Fig. 4a, boxed regions). However, beyond this initial response period, alpelisib-treated tumors began to resume growth or exhibit an increased rate of growth over time (Fig. 4a). Thus, PDX models behave similarly to cell lines in that they also spontaneously develop resistance to PI3Kα inhibition by alpelisib over time.

Ki67 staining was used to examine the proliferative activity of vehicle-treated tumors, alpelisib-sensitive tumors (not treated to resistance) and alpelisib-resistant tumors. Most vehicle-treated and alpelisib-resistant tumors exhibited strong positive Ki67 staining, whereas alpelisib-sensitive tumors showed weaker staining (representative sections shown in Additional Fig. 6). We next analyzed the expression of AXL and TYRO3 in PDX models using IHC. While no difference in AXL expression was detected for either model examined (Additional Fig. 7), TYRO3 expression following prolonged treatment with alpelisib was elevated in both PDX models (Fig. 4b).

Since expression of AXL and TYRO3 were both elevated in alpelisib-resistant models, we proceeded to determine the relative expression of both receptors at the cell surface in parental and resistant cells. Flow cytometric analysis demonstrated a significant increase in both AXL and TYRO3 surface levels in alpelisib-resistant Cal33 and 93-VU-147T cells, compared to parental cells (Fig. 5a).

To assess whether the upregulation of TYRO3 and/or AXL expression plays a causative role in mediating resistance to PI3Kα inhibition, we used siRNAs to silence each receptor in alpelisib-resistant cells. Knockdown of either TYRO3 or AXL re-sensitized resistant cells to alpelisib treatment, to a level almost comparable to their parental (baseline) sensitivity (Fig. 5b). In 93-VU-147T cells, combined TYRO3 and AXL knockdown sensitized resistant cells to an even greater extent than their parental sensitivity, highlighting the role both receptors play in modulating response to alpelisib treatment.

Since the expression of activated MAPK pathway members ERK1/2 and P90RSK was reduced by PI3K inhibition in parental cells (Fig. 1a) and the MAPK signalling pathway is a downstream target associated with TAM RTKs, we proceeded to examine the activation status of several MAPK pathway members in our resistant cells (Fig. 6a) [33]. Beginning upstream, we examined expression of the scaffold protein GAB2 that mediates signalling from the adaptor protein Grb2 on intracellular RTK domains to RAS. GAB2 expression was elevated in alpelisib-resistant 93-VU-147T cells (as suggested in the RPPA), although no difference was apparent in Cal33 cells (Fig. 6b, block 1). The next differentially-expressed pathway member was phosphorylated MEK1 (Ser298) which was apparent in alpelisib-resistant Cal33 cells (Fig. 6b, block 2). Moving down the MAPK pathway, elevated activating phosphorylation of ERK1/2 and P90RSK was detected in all alpelisib-resistant cell lines (Fig. 6b, block 3). Collectively, these observations reveal an induction of MAPK pathway activation upon prolonged treatment with alpelisib.

Given the upregulation of pP90RSK (Ser380) observed in the resistant cell lines and its described involvement in HNSCC oncogenesis, we proceeded to examine its expression in our PDX models using IHC [35, 36]. We observed a non-significant trend towards elevated pP90RSK (Ser380) expression in alpelisib-resistant PDX-E tissues and a significant increase in expression in PDX-C tissues (Fig. 6c-d).

To evaluate the effect of MAPK signalling on resistant cells’ responsiveness to alpelisib, we targeted the downstream MAPK pathway member P90RSK using the small molecule inhibitor BI-D1870, alone and in combination with alpelisib [37]. In both alpelisib-resistant cell lines, BI-D1870 treatment resulted in a significant reduction in cell viability (Fig. 7a). When BI-D1870 and alpelisib treatments were combined, greater decreases in cell viability were observed.

We next evaluated the relation between expression of TYRO3 and AXL, and MAPK pathway activation. Following knockdown of TYRO3 and AXL in alpelisib-resistant cells, we used immunoblotting to detect phosphorylated (active form) members of the MAPK pathway, including ERK1/2, P90RSK and S6 (Fig. 7b). In Cal33 cells, silencing of either TYRO3 or AXL was associated with reduced phosphorylation of P90RSK (Ser380) and S6 (Ser235/6), while only TYRO3 silencing reduced ERK1/2 (Thr202/Tyr204) phosphorylation. In 93-VU-147T cells, TYRO3 knockdown reduced phosphorylation of ERK1/2 and P90RSK, while AXL knockdown did not have an apparent effect on MAPK pathway activation.

To determine whether expression of TYRO3 and/or AXL was associated with PI3K inhibitor sensitivity at baseline (without prolonged drug exposure), we examined the protein expression of both receptors in a panel of 25 HNSCC cell lines (Additional Table 1) [20]. We previously characterized the sensitivity of all 25 lines to alpelisib and ordered the cell lines accordingly [21]. While expression of both proteins varied between cell lines, we did not observe a trend in the expression of either TYRO3 or AXL that appeared to correlate with sensitivity to alpelisib (Additional Fig. 8). Given an absence of a correlation between baseline TAM RTK expression and response to PI3K inhibition, it appears the involvement of TYRO3 and AXL in drug response is contingent on either the activity of the receptors or a relative upregulation of receptors during the course of treatment that subsequently affects drug response.

Finally, in parallel with our in vitro and in vivo models of acquired resistance to PI3K inhibition, we treated our PDX-derived cell line (PDX-C cell line) with alpelisib to generate an alpelisib-resistant cell line. This model system provided the unique opportunity to use an early-passage tumor-derived cell line to validate the data from both our in vitro studies using established HNSCC cell lines and our in vivo studies using PDX models. The alpelisib-resistant PDX-C-derived cell line had a ~ 3-fold increase in IC50 (2.3 μM versus 6.3 μM), relative to its parental counterpart (Fig. 7c). Immunoblotting revealed elevated expression of AXL in the alpelisib-resistant cell line, while TYRO3 expression appeared stable (Fig. 7d). We proceeded to evaluate the relative expression of AXL and TYRO3 at the cell surface using flow cytometry. We observed a significant increase in AXL surface levels in alpelisib-resistant PDX-C cells, compared to in parental cells (Fig. 7e). Downstream, elevated phosphorylation of MAPK pathway members ERK1/2 and P90RSK was also detected in alpelisib-resistant PDX-C cells, confirming our previous in vitro and in vivo PDX findings (Fig. 7f).