PIK3CA and p53 Mutations Promote 4NQO-Initated Head and Neck Tumor Progression and Metastasis in Mice

This article is featured in Highlights of This Issue, p. 799

Head and neck squamous cell carcinoma (HNSCC) represents the sixth most common cancer worldwide, leading to significant morbidity and mortality and resulting in an estimated 10,860 deaths in 2019 in the United States alone (1). Although a clinically detectable preneoplastic lesion frequently precedes development of frank squamous cell carcinoma (SCC), most patients with HNSCC are still diagnosed at advanced disease stages and often fail to respond to available therapies (2).

Understanding of the molecular mechanisms underlying HNSCC progression may afford opportunities to develop novel, targeted strategies for prevention and treatment. HNSCC is a heterogeneous disease involving deregulation of multiple pathways linked to cellular differentiation, cell-cycle control, apoptosis, angiogenesis, and metastasis (3). The PI3K/AKT/mTOR signaling pathway has emerged as one of the most frequently altered pathways in HNSCC (4–7) and multiple upstream and downstream components such as EGFR, PI3K, protein kinase B (AKT/PKB), phosphatase and tensin homolog (PTEN), and mTOR have been found to be highly dysregulated in HNSCC, making this pathway very attractive for molecularly targeted therapies (8, 9).

PIK3CA mutations have been demonstrated in HNSCC (6, 7, 10). One of the most common and direct mechanisms to activate the PI3K pathway is through gain-of-function (GOF) mutations in the PIK3CA gene (11–13). GOF mutations have been described and are associated with increased AKT activity and oncogenic transformation (14, 15). We and others have reported mutations of the PIK3CA gene in 2.6% to 20% of head and neck tumors including oral SCC (OSCC; refs. 5–7, 10, 16–19). PIK3CA mutations are reported to be particularly common in human papillomavirus (HPV)-positive oropharyngeal tumors (reaching 24%–28%; refs. 20, 21). The majority of PIK3CA mutations cluster within the helical (exon 9) and catalytic (exon 20) protein domains (22). Furthermore, three hotspot mutations have been identified: E542K (exon 9), E545K (exon 9), and H1047R (exon 20), which have been shown to increase PI3K oncogenic activity and confer transforming properties in vitro and in vivo (13–15).

To interrogate the role of oncogenic PIK3CA in transformation of upper aerodigestive track epithelium and/or to test the efficacy of therapeutics targeting the PI3K pathway in HNSCC, we developed novel genetically engineered mouse models (GEMM) carrying conditionally expressed mutant PIK3CA and/or p53 GOF alleles in the basal layer of the stratified squamous epithelium of the tongue. Using these GEMMs, we evaluated the impact of the PIK3CA H1047R mutation on the progression and metastasis of 4-nitroquinoline 1-oxide (4NQO)-induced tumors in the oral cavity. The H1047R mutation in PIK3CA is one of the most common hotspot mutation in HNSCC (9) and has been shown to cause activation of PIK3CA, attenuation of apoptosis, and facilitation of invasion (23). Here, we use H1047R mutation to mimic the activation effect of PIK3CA in mice (24). Inactivation of tumor suppressor p53 is one of the key events during malignant transformation into HNSCC. Furthermore, in addition to loss of p53 tumor suppressor function, some p53 mutations are associated with GOF activity that can enhance tumor progression, metastatic potential, or drug resistance. Thus, p53 mutations are associated with shorter survival time and increased resistance to radiotherapy and chemotherapy in patients with HNSCC (25, 26). Accordingly, we also explored the implications of an additional p53 mutation on tumorigenesis for two reasons. First, genetic alterations of the p53 gene are found in HNSCC at high frequency, with LOH of 17p and TP53 point mutations seen in 40% to 50% of cases of premalignant lesions and in HNSCC (27, 28). Second, p53 transcriptionally regulates PTEN, an antagonist of the PI3K pathway (29). We hypothesized that an additional p53 mutation could further alter susceptibility and synergistically induce carcinogenesis, which has been described in other tissues, such as the mammary gland (30).

Introduction of cisplatin has been a significant landmark in the treatment of HNSCC; however, there remains room for improvement in enhancing treatment response and patient outcomes. Numerous potential mechanisms for resistance to cisplatin have been proposed (31). Among these mechanisms, loss of p53 function and/or the activation of the PI3K/AKT pathway are known to inhibit the propagation of DNA damage signal to the apoptotic machinery induced by cisplatin (32). Thus, specific inhibition of a deregulated PI3K/AKT pathway could help in overcoming cisplatin resistance. In recent years, PI3K p110α-specific inhibitors have demonstrated effectiveness in cancer cell lines harboring PIK3CA mutations in large cell line screens (33). Therefore, we also took advantage of our model to generate SCC primary cell lines with different genetic backgrounds to test the efficacy of combination treatment of PIK3CA inhibitor BYL719 and cisplatin.

A strain of mice carrying a modified and conditionally expressed PIK3CA allele with a “hotspot” exon20 GOF mutation [Loxp-STOP-Loxp (LSL)-PIK3CA^H1047R or E20 thereafter] has been described previously (24). The mutation was introduced by gene targeting (knockin) into the endogenous locus, which can be transcriptionally activated after Cre-mediated excision of a “floxed” DNA segment that blocks its expression. Conditional expression of mutant PIK3CA^H1047R in the squamous epithelium of the upper aerodigestive tract was attained by crossing E20 mice to a K14-Cre transgenic line (01XF1, NCI Mouse Repository; ref. 34). LSL-PIK3CA^H1047R;LSL-p53^+/R172H;K14-Cre (E20) mice were also crossbred with heterozygous LSL-p53^+/R172H mutants (01XAF, NCI Mouse Repository; ref. 35) to yield LSL-PIK3CA^H1047R;LSL-p53^+/R172H;K14-Cre double mutants (E20/p53). These were then crossed with the ROSA26-YFP reporter mice (006148, the Jackson Laboratory; ref. 36). Genomic DNA was isolated from tail clips and genotyped by Transnetyx (Cordova) with real-time PCR using genomic DNA and primers specific to the Cre-recombinase, the 5′ LoxP site of PIK3CA and p53 mutants, as well as the wild-type (WT) sequence to confirm successful cross-breeding.

For the in vivo study involving subcutaneous implantation and therapeutic treatments, 5 to 6 weeks old nude mice (NU-Foxn1nu) were obtained from Charles River Laboratory. Half million of EPK-S20-658 cells were suspended in 1:1 mixture of medium/Matrigel and injected subcutaneously in both flanks of each mouse. Treatment started after tumors became palpable, approximately 2 weeks after injection. Mice were grouped (n = 5 per group) and treated with either saline, cisplatin (Sigma, 3 mg/kg, twice per week via intraperitoneal injection), or the combination of cisplatin and BYL719 (MedChemExpress, 30 mg/kg, once daily via gavage feeding) for a maximum of 6 weeks. Tumor volume, calculated as W² × L/2, was measured twice a week for the duration of the study.

All animal studies were carried out in compliance with Institutional Animal Care and Use Committee guidelines and with approved protocols at Columbia University Irving Medical Center (New York, NY).

To induce tumor formation, mice starting at approximately 2 months of age were treated for 8 weeks with 50 μg/mL of 4NQO (Sigma-Aldrich) in drinking water ad libitum. Drinking water was changed weekly. After 8 weeks, 4NQO administration was discontinued. Mice were examined biweekly for development of lesions in the oral cavity for 8 or 16 weeks, and were weighed and euthanized at 8 or 16 weeks.

After recording the total number of gross oral lesions at necropsy, oral tissues were harvested (tongue, lips, and submandibular lymph nodes), fixed in 10% formalin overnight, and embedded in paraffin. Tongue specimens were cut longitudinally to visualize both sides. Serial sections (4 μm) were stained with hematoxylin and eosin (H&E). Liver and lung specimens were similarly processed to monitor for distant metastases. Histologic grading was assessed using the criteria established by the World Health Organization (37) and the lesions were classified as: epithelial atypia limited to the basal layer, epithelial dysplasia (mild, moderate, and severe), carcinoma in situ (CIS), and SCC (well, moderate, and poorly differentiated). Histologic diagnoses were rendered by two board-certified oral pathologists blinded to all data.

Paraffin-embedded oral tissues were processed as described previously (38) and slides were incubated overnight at 4°C with primary antibodies diluted in antibody diluent buffer (Dako; Supplementary Table S1). The Dako LSAB-System-HRP Kit was used for signal amplification. Slides were counterstained with hematoxylin, dehydrated sequentially in ethanol, cleared with xylenes, and mounted with Cytoseal 60 (Thermo Fisher Scientific). For lymph node metastasis determination, lymph node sections were immunolabeled with cytokeratin-14 (CK-14) antibody or anti-YFP IHC. CK-14 and anti-YFP coimmunofluorescence was performed to confirm the cell of origin.

Freshly isolated oral tumors were minced with a sterile razor blade and digested with 0.25% trypsin for 15 minutes at 37°C and 5% CO₂, passed through an 18-gauge syringe 5 times, resuspended in DMEM with 20% FBS, and seeded in 60-mm tissue-culture plates. At 80% confluence, cells were trypsinized (0.05%) and passaged onto 10% FBS DMEM. Further experiments were performed on YFP-expressing cells, after cell sorting using a FACSAria Cell Sorter (Becton Dickinson) on cells that had already been stabilized after at least five passages.

Genomic DNA was extracted from SCC cell lines using the DNeasy Mini kit (Qiagen), following the manufacturer's protocols. To validate successful recombination of the mutant PIK3CA allele, genomic DNA from SCC cell lines was amplified by PCR with the primers PIK3CA-E20Fwd, 5′-CTGAGAAAAACAAGGGAGTTGGC-3′ and PIK3CA-E20Rev, 5′-CCACTTCTTGGCCCTGGTGAGAA-3′. Upon electrophoresis, a band of 280 bp corresponding to the mutant recombined allele and 190 bp corresponding to the WT allele were detected. Similarly, to test the recombination of the mutant p53^R172H allele, the following primers flanking the insertion site of the remaining loxP site were used: p53Rec-Fwd, 5′-AGCCTGCCTAGCTTCCTCAGG-3′; p53Rec-Rev, 5′-CTTGGAGACATAGCCACACTG-3′; yielding a 290 bp WT band and a 330-bp mutant p53 band.

Protein was extracted from SCC cell lines using RIPA Lysis buffer (Millipore). PhosSTOP and proteinase inhibitors cocktail (complete; Roche Diagnostic GmbH) were added according to the manufacturer's instructions. Proteins were resolved using NuPAGE 4%–12% Bis-Tris Gel (Invitrogen), transferred to polyvinylidene difluoride membrane (Bio-Rad), and probed with primary antibodies (Supplementary Table S1).

To assess drug efficacy, 2 × 10³ cells from each cell line were seeded in 96-well plates and treated with 0.05–10 μmol/L of BYL719 (Selleckchem) or 0–20 μmol/L of cisplatin to determine IC₅₀. After determining the IC₅₀ and optimal concentrations, all five cell lines were subjected to treatments of 1 μmol/L of BYL719 and escalating concentrations of cisplatin (0–20 μmol/L) either separately or in combination for 48 hours and then cell proliferation was analyzed with Cell Proliferation Kit (MTT; Roche Diagnostic GmbH) to determine the effect of combination therapy on PIK3CA and p53 mutants. Combination index was calculated using the method of constant ratio drug combination proposed by the Chou and Talalay (39) and the COMPUSYN software (www.combosyn.com), where combination index < 1, =1, and >1 indicate synergism, additive effect, and antagonism, respectively. These results were obtained using a 96-well spectrophotometer with KC Junior Software (Bio-Tek Instruments).

Cells were seeded in 6-well tissue culture plates (2 × 10⁴ per well), cultured for 48 hours under the conditions specified, washed with PBS, and fixed and stained using the Senescence β-Galactosidase Staining Kit (Cell Signaling Technology). Images were captured with a Nikon Labophot-2 Microscope using the Nikon Imaging Software NIS-Elements-F 2.20. Five random, nonoverlapping, 200 × images were collected from each cell line and condition. For each image, positive cells (blue color) and total cells were counted using ImageJ 1.48g software (NIH, Bethesda, MD). Error bars represent SEM.

All statistical analyses were performed using SPSS Software version 17.0 (SPSS Inc.). The Fisher exact test was used for comparison between different genotypes. Survival curves were calculated using Kaplan–Meier product-limit estimate. Deaths due to causes other than head and neck cancer (including mice sacrificed at 16 weeks) were treated as censored observations. Statistical significance in differences between survival times was evaluated by a log-rank test. P ≤ 0.05 was considered statistically significant.

RNA-seq was performed at Macrogen Lab. Conditions and replicates are given in Supplementary Table S2. Clones were subdivided into samples, which were cycled several times before harvesting. The TruSeq Stranded mRNA Sample Preparation kit was used to prepare RNA libraries. 40M. 150 bp Paired end sequencing was performed on an Illumina NovaSeq6000 S4 flowcell. Reads were aligned to the mm10 assembly of the mouse genome with Rsubread (40). Expression was quantified on a gene level with featureCounts (41). Reads and Fastq files were deposited in Gene Expression Omnibus (GSE143716). Differential expression was analyzed with weighted Limma-Voom (42), with a significance cutoff of the Benjamini–Hochberg FDR ≤ 0.05. Correlations between samples from the same clone were taken into account using the duplicateCorrelation method in Limma (42). The comparisons performed are given in Supplementary Table S3. Venn diagrams were drawn with BioVenn (43). GOslim and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment were analyzed, and depicted graphically, with WEBGESTALT (44).

To investigate the role of the PI3K pathway in the development of HNSCC in vivo, we generated E20 GEMM (LSL-PIK3CA^H1047R;K14-Cre) expressing constitutively active p110α-H1047R protein in the epithelium of the upper aerodigestive tract. No premalignant lesions beyond atypia were detected in E20 mice after 1 year of follow-up (data not shown). The same observations (the absence of premalignant abnormalities and tumor development) were made in p53 mice (LSL-p53^+/R172H;K14-Cre) and E20/p53 mice (LSL-PIK3CA^H1047R;LSL-p53^+/R172H;K14-Cre). Accordingly, we tested whether, after tumor initiation using the standard 4NQO model, tumor progression was affected by PIK3CA and/or p53 mutations. As such, E20, p53, E20/p53, and WT control mice were treated for 8 weeks with 4NQO, a synthetic compound that acts as a surrogate of tobacco exposure (45), added in drinking water (Supplementary Fig. S1).

Biweekly examination revealed that 1 week following the 8-week treatment the E20/p53 double mutant mice displayed small lesions presenting as flat leukoplakia and/or erythroplakia, exophytic papillary lesions, or raised/indurated masses suspicious for malignant transformation. In contrast, oral lesions were not evident in the E20, p53, and WT mice in this early time period. By 8 or 16 weeks after treatment, mutant (E20, p53, or E20/p53) and WT mice all had detectable oral cavity lesions including leuko- or erythroplakias, exophytic papillary masses, and fungating lesions highly suspicious for cancer (Fig. 1A). Macroscopic analyses at necropsy showed that at 8 weeks after treatment termination, the E20-mutant mice had developed significantly more lesions than the WT (P = 0.01) or the p53-mutant mice (P < 0.001), whereas the mice of the latter two genotypes did not show differences between them (Fig. 1B). On the other hand, the average number of gross oral lesions was practically the same in the E20 single mutants and the E20/p53 double mutants at the 8-week time point (3.2 ± 0.7 and 3.28 ± 0.2, respectively). Interestingly, at the 16-week time point, this number did not increase for E20 mutants (3.28 ± 0.6), whereas in the E20/p53 mice it was doubled in comparison with the number observed at 8 weeks (6.33 ± 0.8 vs. 3.28 ± 0.2; Fig. 1B) and was significantly higher than that detected in E20 mice (P < 0.05).

Weights of mice prior to euthanasia were also recorded. WT mice weighed significantly more than any of the mutant mice at 16 weeks' follow-up (E20, P = 0.0004; p53, P = 0.02; E20/p53, P = 0.002; Fig. 1C). Notably, the average weight corresponded nearly inversely with oral lesion count.

Overall, E20/p53 mice showed an average reduced latency (47.8 days), defined here as time to lesion formation after 4NQO treatment, when compared with the E20 (57.3 days), p53 (78.8 days), and WT (105.8 days) mice (Fig. 1D). Kaplan–Meier analysis, after monitoring of animals of the four genotypes over a period of 4 months following 4NQO exposure, indicated that all mutant mice exhibited decreased survival compared with the WT controls (Fig. 1E). At the end of the observation period, while 93% (14/15) of WT mice survived, the E20/p53 mice had the lowest percentage of survival (29%, 2/7), compared with 45% (5/11) of the E20 mice and 83% (10/12) of the p53 mice (Fig. 1E). The E20/p53 mice showed a decreased median survival (T₅₀ = 76 days) compared with the E20 mice (T₅₀ = 113 days), although the difference was not statistically significant (P = 0.3). However, both E20/p53 and E20 groups had significantly shorter survival than did the p53 and WT mice. There was neither statistically significant difference between sexes in survival nor lesion counts across the mutant genotypes.

Microscopic examination of tongue tissue revealed varying degree of neoplastic progression, ranging from atypia of epithelial cells confined to the basal layer, epithelial dysplasia (mild, moderate, and severe), CIS, to invasive SCC that faithfully recapitulates tumor progression in humans (Fig. 2A). At 8 weeks of follow-up, E20/p53 double mutants exhibited advanced tumor progression, in which almost half of all tongue lesions progressed to invasive SCC (4/9; 44.4%), compared with a minority in the E20 mutant (1/12; 8.3%), p53 mutant (1/10; 10%), and WT (1/15; 6.7%) groups (Fig. 2B). Notably, at 8 weeks of follow-up, the percentage of tongue lesions graded as CIS or SCC together was greater in E20/p53 mice than that of both single mutants (E20 and p53) combined. It is not possible to compare across the four groups at 16 weeks because as noted above, survival rates for the E20/p53 and E20 mice were significantly lower than those of p53 and WT mice (Fig. 1E), thus only the few mice that survived up to that time in the E20/p53 and E20 groups were examined and included here (Fig. 2C).

The expression pattern of proteins, such as p16^ink4a, p53, E-cadherin, and known to be associated with the development of human HNSCC, was analyzed by IHC in the tongue tissues. The expression of phospho-S6 (p-S6, or S6 ribosomal protein) was also analyzed as a surrogate marker of PI3K pathway activation. The expression of p16 was primarily localized to the nuclei of surface epithelial cells in the normal tissue, but absent from epithelial cells in the basal layer (Fig. 3A; Supplementary Fig. S2 i). Strong p16 nuclear staining was also observed throughout the surface epithelium of premalignant lesions (atypia/dysplasia), but was progressively lost in CIS and SCC (Fig. 3A; Supplementary Fig. S2 ii and iii). Nuclear p16 expression was scored (weak, moderate, or strong) for the E20, p53, and E20/p53 mutants with matching tumor stages, and increased p16 expression at the premalignant stage and decreased p16 expression during the CIS and SCC stages were observed in all genotypes but without statistical significance (Fig. 3B). Low level of nuclear p53 expression was found in the basal layer of normal surface epithelium (Fig. 3A; Supplementary Fig. S2 iv). As with p16 expression, progressively dysregulated p53 nuclear staining was observed in the epithelium of premalignant lesions (Fig. 3A; Supplementary Fig. S2 v) and in the tumor islands of SCC (Fig. 3A; Supplementary Fig. S2 vi). E-Cadherin was strongly expressed in the normal epithelium, whereas it was weak and/or diffuse in premalignant and SCC lesions (Supplementary Fig. S2 vii–ix).

Strong p-S6 expression was evident in the normal epithelium of the E20/p53 (Supplementary Fig. S2 x), which was sustained throughout the progression from normal to SCC (Supplementary Fig. S2 xii). A similar pattern of p-S6 expression was detected in the epithelium of the E20 mutants, while absent or weak p-S6 expression was found in the p53-mutant epithelium (data not shown).

In HNSCC, distant metastasis leads to dismal prognosis, and risk factors associated with distant metastasis include regional node positivity and extranodal extension (46, 47). To evaluate the presence of lymph node metastasis, the YFP allele was introduced into our GEMMs to trace the dissemination of tumor cells. IHC against YFP and/or cytokeratin-14 was performed (Fig. 3C), verified by double immunofluorescence labeling (Fig. 3C, ii–v) of lymph node tissues, and further supported by the isolation of YFP⁺ cells from lymph nodes in tissue culture (Supplementary Fig. S3). Consistent with the survival data, the E20/p53 double mutant mice presented the highest rate of lymph node metastasis (57.1%), while the frequency was lower in the E20 and the p53 groups (33.3% and 11.1%, respectively). In regards to distant metastasis, 40% (n = 2/5) of the E20/p53 mice exhibited micrometastasis to the lungs (Fig. 3C, vi), whereas none of the E20 (n = 0/5) or p53 (n = 0/5) mice harbored detectable distant metastasis.

Further molecular analyses were carried out in primary tumor cell lines derived from the OSCC tumors developed in these GEMMs. The cell lines were first examined for recombination of PIK3CA^H1047R-and p53^+/R172H-mutant alleles, verifying that the tumors were indeed driven by engineered mutations and not merely sporadic products of the 4NQO treatment (Fig. 4A). As expected, PIK3CA^H1047R recombination was detected in the tumor cell lines derived from E20 and E20/p53 mutant but not from WT mice. Similarly, p53^+/R172H recombination was confirmed in the p53 and E20/p53 mutants and not in WT controls. Interestingly, one of the E20/p53 tumor cell lines harbored only the mutant allele at the p53 locus, indicating LOH of the WT allele, which was further confirmed by sequencing. Sporadic mutations of the p53 gene were found in some tumor cell lines without the engineered p53^+/R172H allele. RT-PCR sequencing was used to further confirm the expression of the mutants PIK3CA and p53 alleles (Supplementary Table S4). Western blot analysis showed elevated accumulation of p-Akt (both T308 and S473) in all cell lines derived from tumors with E20 mutation, confirming activation of the PI3K pathway. Cell lines derived from tumors with WT PIK3CA genotype exhibited comparatively minimal Akt activation (Fig. 4A).

RNA-seq was also performed to reveal genes and pathways that may be involved in SCC progression in our GEMMs in addition to the engineered mutations. Gene expression profiles were compared between SCC cell lines derived from mutant mice (p53, E20, and E20/p53) to the WT control. The number of differentially expressed genes in each comparison is given in a Venn diagram (Fig. 4B) and Supplementary Table S3. The Venn diagram, describing the overlap between genes differentially expressed between each of the three mutants and WT, depicts that 3,502 of these genes were shared in all three comparisons. These genes were then analyzed in terms of GOslim and KEGG enrichment with WEBGESTALT (Fig. 4C and D) and the results imply that PIK3CA and p53 mutations both individually and together drive SCC progression by changing cell proliferation and growth, the cell cycle, miRNA expression, apoptosis, and focal adhesion in these mice.

BYL-719 (Alpelisib) is a selective inhibitor of PIK3CA and is being evaluated in numerous clinical trials for patients with HNSCC (9). To examine whether BYL719 could enhance the antiproliferative effects of cisplatin on the SCC primary cell lines, we first determined the IC₅₀ of BYL-719 and cisplatin in each cell line, then treated these cell lines with a combination of 1 μmol/L of BYL719 and cisplatin in escalating concentrations (1.25–20 μmol/L; Fig. 5). For these experiments, we selected two E20/p53 double mutant cell lines, one of them exclusively carrying the p53^+/R172H-mutant allele and the other carrying an E168Stop mutation (p53^−/R172H). We also selected two E20-mutant cell lines without p53 mutations. For validation purposes, we also included the human HNSCC cell line Detroit 562, which carries both PIK3CA^H1047R and p53^R175H mutations (17), to the latter corresponding to the R172H in mouse. The IC₅₀ for cisplatin was 11.7 μmol/L in Detroit 562 cell line, which is in agreement with a previous report (48). The E20/p53 double mutant cell lines had an IC₅₀ for cisplatin close to that of Detroit 562 and higher than that of E20 single mutant cell lines (9.5 and 10.9 μmol/L vs. 7.4 μmol/L; Fig. 5F). Cell viability was significantly decreased by combination treatment of BYL719 and 2.5 μmol/L cisplatin in all cell lines except the E20/p53 cell lines not expressing the p53 WT allele (S26-330 and Detroit 562). The reduction in cell viability by the combination treatment was more pronounced in the two E20 single mutant cell lines (53% and 44% inhibition, respectively) than in the E20/p53 double mutant cell lines (28% and 31% inhibition, respectively) including Detroit 562 (31% inhibition; Fig. 5). The combination index calculation and the leftward shift in the dose–response curve (reduced concentration for IC₅₀) resulting from combination treatment demonstrate that there could be a benefit from using BYL719 in combination with cisplatin in HNSCC carrying a PI3K mutation (Fig. 5F). Although the presence of TP53 mutations predicts a worse response to cisplatin, our results suggest that combination with a PI3K inhibitor could sensitize SCC cells to cisplatin regardless of TP53 status.

To assess the determinants of sensitivity to p110α inhibition, we analyzed phosphorylation status of Akt as a proximal marker of PI3K inhibition and p-S6 as a distal component of the PI3K pathway after treatment with cisplatin, BYL719, or both. There was a marked inhibition of p-Akt 48 hours after treatment with BYL719 but not with cisplatin on all cell lines, indicating successful inhibition of PI3K (Fig. 5G). These levels were comparable with those found in cells treated with the combination of BYL719 and cisplatin, indicating that p-Akt levels only reflected PI3K inhibition by BYL719 but not by cisplatin. Persistent mTORC1 activity has been correlated with intrinsic resistance to BYL719 (49). Both E20/p53 double mutant cell lines presented significant sustained activation of p-S6 (s235/236) with BYL719 and/or combination treatment. Only the E20-mutant cell lines treated with combination treatment showed mTOR inhibition, as indicated by almost complete suppression of p-S6 expression. These findings correlate with a higher cisplatin IC₅₀ shift in the E20 cell lines (3.8–2 fold change) versus E20/p53 (1.7–1.3 fold change) with the BYL719 treatment, which suggests that the contribution of TP53 mutations to BYL719 resistance may be due to persistent mTORC1 activity.

p16^ink4a is a negative regulator of the CDK/Rb pathway (50) and an effector of oncogene-induced senescence (OIS) in premalignant tumors. It is therefore inferred that premalignant tumors that have undergone p16 or p53 loss are able to overcome OIS and progress to malignancy (51). Western blot analysis showed that both E20/p53 double mutant cell lines preserved p16 expression, while the E20 single mutant cell lines either lost (S26–117 E20) or had low (S26-234 E20) p16 expression (Fig. 5G). This observation suggests that tumor progression in E20 single mutant mice requires a second event involving inactivation of a tumor suppressor gene like p53 or p16 to overcome OIS to develop SCC (Supplementary Table S4).

To evaluate the senescence response after combination treatment with BYL719 and cisplatin, we analyzed senescence-associated β-galactosidase activities in our cell lines (Fig. 6A and B). We found that the combination treatment enhanced the senescence in all the cells except in the cell line with complete p16 loss (S16-117 E20). Interestingly, both E20/p53 double mutant cell lines presented an increased senescence response similar to that of the E20 single mutant with low p16 expression (Fig. 6A), although we did not observe elevated p16 expression induced by the combination treatment (Fig. 5G).

To investigate whether the efficacy of combination therapy observed in vitro can be translated to in vivo feasibility, we subcutaneously implanted an E20/p53 SCC cell line (s24-658) that has demonstrated enhanced sensitivity to the combination treatment of BYL719 and cisplatin over single therapy in vitro (Supplementary Fig. S4). Statistical significant inhibition of tumor growth was observed when comparing the combination treatment with the untreated control (P < 0.01) or cisplatin alone (P < 0.05; Fig. 6C).

In this study, we have shown that a GOF mutation in the exon 20 of the PIK3CA (H1047R) in GEMMs can promote 4NQO-induced tumors of the oral mucosa that faithfully simulate the complex nature of human HNSCC tumors. Moreover, we found that E20/p53 double mutants exhibited a more aggressive phenotype (characterized by increased incidence of HNSCC, metastasis, and shortened survival) than did the mice with a single mutant allele or WT control after the same 4NQO treatment.

Our study demonstrated that a hotspot mutation in the PIK3CA gene, expressed from its endogenous locus, was sufficient to increase susceptibility to chemical carcinogenesis in the oral epithelia of mice. We observed that PIK3CA^H1047R mutation significantly increased the number of visible lesions compared with p53 GOF alone at 8 weeks follow-up. At 16 weeks follow-up, E20/p53 double mutant mice presented significantly more grossly visible lesions than did the E20 or p53 single mutants, pointing to a synergy between p53^R172H and PIK3CA^H1047R in tumor development (Fig. 1B). At the histologic level, PIK3CA^H1047R mutation was not associated with more advanced histologic grade than that of p53^R172H. The p53 and E20 single mutants exhibited similar numbers of advanced tumors at 8 weeks (CIS and SCC). However, consistent with the gross anatomic findings, the E20 single mutants displayed increased dysplasia compared with the p53 mutant and WT cohorts (Fig. 2B). The E20 single mutant mice also had shorter average tumor latency than did the p53 mutant and WT mice (Fig. 1D). Moreover, synergy between PIK3CA^H1027R and p53^R172H at gross examination of the E20/p53 double mutants (Fig. 1B) was also confirmed at histologic analyses, where the presence of both mutations resulted in greater percentage of mice with advanced tumors compared with single mutant mice (Fig. 2B). It is not possible to compare across the four groups at 16 weeks because only <50% of the E20/p53 and E20 mice survived by that timepoint (Fig. 1E), thus the histologic analyses of the surviving E20/p53 and E20 mice included in Fig. 2C do not represent the overall pathogenesis of those two genotypes at 16 weeks.

Our latency and survival data also support a synergistic effect between PIK3CA^H1047R and p53^R172H mutations (Fig. 1D and E). The E20/p53 mice demonstrated the worst survival (Fig. 1E), lowest body weight (Fig. 1C), and higher histologic grade of all groups (Fig. 2). The more aggressive nature in E20/p53 double mutant mice was also characterized by increased rates of lymph node and distant metastases when compared with the single mutants. The E20/p53-mutant mice presented a higher rate of lymph node metastasis (57%) than did the E20 and the p53 groups (33% and 11%, respectively). Furthermore, 40% (n = 2/5) of the E20/p53 mice exhibited micrometastasis to the lungs (Fig. 3C, vi), whereas none of the E20 (n = 0/5) or p53 (n = 0/5) mice harbored detectable distant metastasis. Some of the mice also developed esophageal lesions which might have impacted their oral intake and weight loss; however, only mice with confirmed gross and histologic oral lesions were included in this study. Overall, our E20/p53 double mutant mouse model closely mimics the histology and etiology of human HNSCC.

Molecular analysis of the primary SCC cell lines demonstrated the expected activation of the PI3K pathway and the malignant transformation of the tumors in our mice were significantly driven by the engineered PIK3CA^H1047R and/or p53^R172H mutations (Fig. 4A). Genotypic analysis of the p53 gene revealed that in one of the E20/p53 cell lines, a spontaneous loss of the WT p53 allele had occurred (Supplementary Table S4). This was subsequently confirmed through sequencing exon 4 of the p53 gene. In the absence of p53 mutation, spontaneous or engineered, p16 expression was suppressed in tumor cell lines derived from E20 single mutant mice (Fig. 5G; Supplementary Table S4), indicating that a second event such as p53 LOH or p16 inactivation was necessary for progression to invasive cancer in our GEMMs. RNA-seq analyses also revealed that in addition to the engineered genetic mutations, expression levels of genes associated with DNA replication and cell-cycle pathways were highly dysregulated (Fig. 4D), which is consistent with the progressive inactivation of p16 and p53 observed in these GEMMs (Fig. 3A and B). Our finding of spontaneous p53 LOH/mutation and suppressed p16 expression underscores the notion that these models accurately simulate the natural progression of HNSCC in humans at the genetic level (52). Presumably, then, molecular changes brought about by the 4NQO treatment (53) can fruitfully collaborate with the GOF alterations in the PI3K and p53 pathways toward the stepwise development of full-fledged malignancy.

Intriguingly, in the absence of 4NQO, no premalignant lesions beyond cellular atypism limited to the basal layer were detected in the E20/p53 double mutants after 1 year of follow-up (data not shown). Our observations are in agreement with previous studies in GEMMs of mammary, ovarian, and intestinal carcinogenesis (54–57) showing that mutated PIK3CA expressed from the endogenous locus at presumably physiologic levels is a weak oncogene unable to initiate tumor formation by itself, as indicated by very long latency periods, and requiring additional events to contribute to tumorigenesis. While overexpression of PIK3CA^H1047R was sufficient to induce the development of lung tumors in some transgenic mouse lines at various latency (58), overexpression of WT PIK3CA in head and neck epithelium was only sufficient to induce epithelial hyperplasia and 4NQO treatment was necessary for progression to SCC (59). In contrast, Moral and colleagues reported spontaneous development of oral lesions progressing to dysplasia in a mouse model with constitutively active Akt (myristoylated-Akt) and malignant conversion of the lesions by subsequent ablation of the Trp53 gene (60). Comparison of this evidence with our and others' results, suggests that PIK3CA is not as potent in driving spontaneous tumor formation as a constitutively active Akt kinase. The activation of the PI3K pathway in our model was demonstrated by increased phosphorylation of both the T308 and S473 Akt sites in the PIK3CA^H1047R-mutant cell lines (Fig. 4, iii), but as stated previously, a secondary mutational event such as p53 mutation or p16 inactivation was necessary for tumor progression to SCC and metastasis as in humans. However, unlike in human HNSCC where numerous stepwise genetic alterations are required for the development of invasive cancer, inactivation of p16 is not seem to be necessary in addition to the engineered PIK3CA and p53 mutations for SCC development in E20/p53 mice. Therefore, while p16 inactivation via sporadic genomic loss or methylation is a frequent event in human HNSCC, retention of p16 expression was detected in some OSCC cell lines derived from our GEMMs, particularly those derived from E20/p53 mice (Fig. 5G; Supplementary Table S4).

Importantly, all OSCC primary cell lines derived from the E20 and E20/p53 GEMMs exhibited enhanced sensitivity to BYL719 and cisplatin combination treatment in comparison with cisplatin alone in vitro and in vivo, regardless of p53 and/or p16 status (Figs. 5A–F and 6C). The inhibition of AKT and S6 phosphorylation (Fig. 5G) and increased senescence (Fig. 6A and B) were likely contributing factors to this enhanced drug sensitivity. We did not observe inhibition of S6 phosphorylation with BYL719 alone at the 48 hour timepoint likely due to the fact that inhibition of the PI3K pathway by single agents tend to be short-lived, with p-S6 levels returning to almost normal by 24 hours. Therefore, combination treatment might be needed to demonstrate clinical efficacy (61). While no senescence was induced in the S26-117 cell line because it had lost p16 expression, the S26-117 cells displayed the highest sensitivity to the combination therapy in all the cell lines examined; therefore, other mechanisms may also be involved. Nevertheless, given the prevalence of mutations in p53 and the PI3K pathway in HNSCC in conjunction with loss of p16 genetically or epigenetically, the evidence that BYL719 can overcome cisplatin resistance in PIK3CA-mutant cancer cells offers new hope to patients with HNSCC in the coming era of precision medicine. Currently, there are multiple targeted therapies against the PI3K pathway under investigation (9). These therapies may allow for leveraging a patient's tumor vulnerability to enhance sensitivity to a standard chemotherapy to potentially improve survival.

In future investigations, HPV status may be an important consideration. HPV is a known risk factor in certain types of HNSCC (62), with cancers attributed to HPV in 18.5%, 3%, and 1.5% of oropharynx, oral cavity, and larynx cancer, respectively, in a large study of 3,680 samples (63). In contrast to HPV-negative HNSCC, HPV-positive HNSCC does not frequently have TP53 alterations due to inhibition of p53 by viral proteins. PIK3CA alteration, on the other hand, has been found in 34% of HPV-negative HNSCC and 56% of HPV-positive HNSCC (64). Thus, while HPV mutation status did not play a role in this study, it would be worthwhile to investigate the role of PIK3CA mutations and response to PI3K inhibitor therapy in the context of HPV-positive and HPV-negative HNSCC models.

No potential conflicts of interest were disclosed.

Conception and design: D. García-Carracedo, G.H. Su

Development of methodology: D. García-Carracedo, W. Qiu, K. Saeki, G.H. Su

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D. García-Carracedo, Y. Cai, K. Saeki, A. Lee, Y. Li, E.M. Goldberg, E.E. Stratikopoulos, R. Parsons, C. Lu, A. Efstratiadis

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D. García-Carracedo, Y. Cai, K. Saeki, R.A. Friedman, A. Lee, Y. Li, A. Efstratiadis, A.J. Yoon, G.H. Su

Writing, review, and/or revision of the manuscript: D. García-Carracedo, Y. Cai, R.A. Friedman, A. Lee, C. Lu, A. Efstratiadis, A.J. Yoon, G.H. Su

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D. García-Carracedo, A. Lee, G.H. Su

Study supervision: D. García-Carracedo, G.H. Su

Other (histopathology review): E.M. Philipone

This work was supported by NCI R56CA109525 and NCI R01CA109525. Y. Cai was partially supported by the American Medical Association Seed Grant Research Program. We also would like to acknowledge Xinjing Xu for her technical support.

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.