Deletion of p53 and Hyper-Activation of PIK3CA in Keratin-15+ Stem Cells Lead to the Development of Spontaneous Squamous Cell Carcinoma

Squamous cell carcinoma (SCC) is the second commonest type of skin cancer, and SCCs make up about 90% of head and neck cancers (HNSCCs). HNSCCs harbor two frequent molecular alterations, namely, gain-of-function alterations of phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit alpha (PIK3CA) and loss-of-function mutations of tumor protein p53 (TP53). However, it remains poorly understood whether HNSCCs harboring different genetic alterations exhibit differential immune tumor microenvironments (TME). It also remains unknown whether PIK3CA hyperactivation and TP53 deletion can lead to SCC development spontaneously. Here, we analyzed the Cancer Genome Atlas (TCGA) datasets of HNSCCs and found that patients with both PIK3CA and TP53 alterations exhibited worse survival, significantly lower CD8 tumor infiltrating lymphocytes (TILs) and higher M0 macrophages than other controls. To better model human tumorigenesis, we deleted TP53 and constitutively activated PIK3CA in mouse keratin-15-expressing stem cells, which leads to the spontaneous development of multilineage tumors including SCCs, termed Keratin-15-p53-PIK3CA (KPPA) tumors. KPPA tumors were heavily infiltrated with myeloid-derived suppressor cells (MDSCs), with a drastically increased ratio of polymorphonuclear-MDSC (PMN-MDSC) versus monocytic-MDSC (M-MDSC). CD8 TILs expressed more PD-1 and reduced their polyfunctionality. Overall, we established a genetic model to mimic human HNSCC pathogenesis, manifested with an immunosuppressive TME, which may help further elucidate immune evasion mechanisms and develop more effective immunotherapies for HNSCCs.


Introduction
Head and neck cancers (HNC) are a heterogeneous group of tumors arising from the mucosal surfaces of the upper aerodigestive tract [1]. Collectively, HNC is the sixth most prevalent cancer worldwide [1]. Some 90% of all HNCs are head and neck squamous cell carcinomas (HNSCCs) and HNSCCs are often associated with either carcinogens, such as alcohol and tobacco use, or oncogenic human papillomavirus (HPV) infection [2,3], thereby categorized as HPV(−) or HPV(+) HNSCCs. HNSCCs have been found to be diverse with a high rate of genetic heterogeneity, resulting in hyper-activation of oncogenes (e.g., PIK3CA and HRAS) and loss-of-function mutations in tumor suppressor genes (e.g., TP53, CASP8, and NOTCH1) [4,5]. Phosphoinositide 3-kinase (PI3K) is a frequently deregulated pathway in HNSCCs with a phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit alpha (PIK3CA) gene mutation rate of approximately 16% and gene amplification of more than 30% in tumors [6,7]. PI3Ks are activated by receptor tyrosine kinases (RTKs), such as epidermal growth factor receptor (EGFR), and consist of different classes of enzymes vital for differentiation, proliferation and cell survival [8]. Mammalian target of rapamycin (mTOR) complexes (mTORC1 and mTORC2) and protein kinase B (also known as AKT) are also involved in this pathway that can activate transcription and other signaling molecules of the PI3K pathway [9]. Monoclonal antibodies (mAbs) that inhibit EGFR have been used for both HPV(−) and HPV(+) subtypes of HNSCCs; however, they were found to have limited efficacy and elicited resistance [10].
Another highly mutated gene in HNSSCs is the tumor protein p53 (TP53) gene, with over 80% of HPV(−) HNSCCs harboring loss-of-function mutations in TP53; however, TP53 mutations occur much less frequently in HPV(+) HNSCCs (~3%) [4]. TP53 is a tumor-suppressor gene encoding a transcription factor that maintains DNA repair, cell cycle, senescence and apoptosis [11]. These attributes make p53 an important cell sensor for oncogene activation and DNA damage. It has been found that the degradation of p53 is associated with HPV E6 oncoproteins [3]. Although there have been several therapies that target p53 in hopes to restore p53 function, they have yet to be proven effective in clinical trials [12]. By and large, TP53 mutations are associated with poor HNSCC prognosis and overall survival with increased rate of recurrence and resistance to therapies. It remains poorly understood whether HNSCCs harboring different genetic alterations exhibit differential immune tumor microenvironment (TME). For instance, it is unknown whether HNSCCs with the double mutations in TP53 and PIK3CA have a more immunosuppressive TME.
Prior studies have generated murine models that mimicked the alterations of PIK3CA or p53 in HNSCCs. Transgenic mice that overexpressed wild-type PIK3CA in head and neck epithelium were generated; however, PIK3CA overexpression alone was not sufficient to initiate HNSCC formation [6]. Nevertheless, these PIK3CA Tg mice were much more susceptible to 4-nitroquinoline 1 oxide (4NQO)-induced HNSCC carcinogenesis [6]. Conditional deletion of p53 in mouse epithelial cells with K14.CrePR1 led to SCC development in about half of mice after 20 months [13]. To establish a mouse model that more closely resembles the genetic alterations in HNSCCs and allows us to better investigate immune evasion mechanisms of HNSCCs, we generated a novel genetic model by deleting p53 and constitutively activating PIK3CA in mouse keratin 15-expressing (K15 + ) stem cells, which leads to the development of multilineage tumors including SCCs, termed Keratin-15-p53-PIK3CA (KPPA) tumors. We found that the TME of KPPA tumors appeared to be highly immunosuppressive. We suggest that the KPPA tumor model may help further elucidate immune evasion mechanisms in HNSCCs and develop more effective HNSCC immunotherapies.
We uploaded RNA-seq data of HNSCC patients onto CIBERSORT (see details in Methods), which estimated the relative proportions of 22 immune cell types (Supplemental Figure S2), with a more in-depth dissection shown in Supplemental Table S1. Both innate and adaptive immune cells varied in their expression levels depending on the genetic alterations in 4 groups ( Figure 1C). In particular, we found that the expression of CD8 T cell signature genes was significantly lower in PIK3CA Amp /TP53 Mutated group, compared with PIK3CA WT /TP53 WT and PIK3CA Amp /TP53 WT groups ( Figure 1D). In addition, PIK3CA Amp /TP53 Mutated group had significantly lower expression of activated NK cell-associated genes compared with PIK3CA WT /TP53 WT and PIK3CA WT /TP53 Mutated groups (Supplemental Table S1). PIK3CA Amp /TP53 Mutated group expressed significantly higher level of resting macrophage (M0) signature genes but lower level of activated macrophage (M1) genes than PIK3CA WT/ TP53 WT and PIK3CA Amp /TP53 WT groups ( Figure 1D). We conclude that HNSCCs with the genotype of PIK3CA Amp /TP53 Mutated appear to have a highly immunosuppressive TME.

A Genetic Mouse Model of PIK3CA Hyperactivation and p53 Deletion in K15 + Cells
To establish a mouse model that can mimic the two most frequently mutated genes in human HNSCCs, we crossed a K15.CrePR1 transgenic mouse model that expressed a RU486-inducible Cre recombinase in K15 + bulge epithelial stem cells with a floxed trp53 allele (p53 f/f ) and a knock-in allele of the constitutively active pik3ca gene in the ROSA26 locus (PIK3CA c/c ) to generate K15.CrePR1(+)p53 f/f PIK3CA c/c mice. RU486 application causes the activation of homozygous PIK3CA c/c knock-in allele and conditional deletion of trp53 in K15 + epithelial stem cells. Upon RU486 application, we observed that none of the mice in the Cre − cohort developed tumors, whereas, a majority of K15.CrePR1(+)p53 f/f PIK3CA c/c mice developed cutaneous tumors 2−3 months after induction of Cre-mediated recombination, termed KPPA tumors (Figure 2A, Supplemental Table S2). Individual mice developed a varying number of tumors (ranging from one to multiple tumors per mouse), which is likely due to the efficiency of Cre-mediated recombination.
We analyzed the primary KPPA tumors by Western blotting to examine the expression of p53 and PIK3CA (a.k.a. p110α) proteins. Our data demonstrated that the primary KPPA tumors did not express p53 ( Figure 2B), and harbored the constitutively active PIK3CA allele that encoded a protein with a slightly higher molecular weight than the WT PIK3CA protein ( Figure 2B). Of note, we still detected the WT PIK3CA protein in all of the KPPA tumors ( Figure 2B), because the WT endogenous PIK3CA locus remained intact. Based on H&E histological assessment, KPPA tumors were characterized as either well-to-moderately differentiated SCC or pleomorphic carcinoma ( Figure 2C). Immunofluorescent staining of cytokeratin 5 (CK-5) and vimentin (Vim) confirmed these results ( Figure 2D). Overall, we conclude that deleting p53 and constitutively activating PIK3CA in mouse K15-expressing stem cells leads to the development of multilineage tumors including SCCs. The expression of CD8 T cell signature genes: PIK3CA Amp /TP53 Mutated group (3.77 ± 5.32) was significantly lower (p < 0.0001) than groups of PIK3CA WT /TP53 WT (8.97 ± 7.86) and PIK3CA Amp /TP53 WT (8.22 ± 8.12). The expression of M0 signature genes: PIK3CA Amp /TP53 Mutated group (21.07 ± 14.08) was significantly higher (p < 0.0001) than groups of PIK3CA WT /TP53 WT (

Characterization of the Immune TME in KPPA Tumors
To better understand how KPPA tumors evaded the host's immunity, we performed flow cytometry analysis to characterize different subsets of immune cells in KPPA TME. As controls, we

Characterization of the Immune TME in KPPA Tumors
To better understand how KPPA tumors evaded the host's immunity, we performed flow cytometry analysis to characterize different subsets of immune cells in KPPA TME. As controls, we analyzed the splenocytes collected from either wildtype (WT) B6 mice or mice harboring KPPA tumors that spontaneously arose upon RU486 induction. To test whether tumors harboring different oncogenic drivers exhibit differential immune profiles in the TME, we also transplanted a SCC line (A223) [14] derived from primary K15.Kras G12D .Smad4 −/− SCCs [15] into WT B6 recipient mice and analyzed the immune cells in these A223 tumors.
To differentiate hematopoietic cells from other cell lineages, we performed flow analysis on the single-cell suspension of the WT spleen control, KPPA tumor-bearing (TB) spleen control, A223 and KPPA tumors, and gated on the CD45 + population (a marker for hematopoietic cells). We found that the percentage of CD45 + cells was significantly less in both A223 and KPPA tumors compared to both splenocyte controls, while the percentage of CD45 + cells did not differ in A223 and KPPA tumors ( Figure 3A, Supplemental Figure S3A). Within the CD45 + population, we also identified the non-B cell/non-T cell (TCRβ − CD19 − ) population in these samples and found that both A223 and KPPA tumors harbored a significantly higher percentage of TCRβ − CD19 − population than the splenic controls ( Figure 3B, Supplemental Figure S3B). Further classification of the tumor-infiltrating immune cells showed that both SCC tumor models exhibited a much higher percentage of myeloid population (TCRβ − CD19 − CD11b + ) compared to splenic controls ( Figure 3C, Supplemental Figure S3C). We also examined the tumor infiltrating lymphocytes (TILs) including CD4 and CD8 T cells and found that both A223 and KPPA tumors contained a much lower percentage of CD4 T cell than splenic controls ( Figure 3C, Supplemental Figure S3C). While the percentage of CD8 T cells did not differ between KPPA tumors and the TB splenic control ( Figure 3C), it was significantly lower in KPPA tumors than in the WT splenic control (Supplemental Figure S3C). The difference in the percentage of cell type between WT and TB spleens may be due to metastases in the TB spleens, as indicated by the higher percentage of non-CD45 population (Supplemental Figure S4, TB-Spleen 54.8% vs. WT-spleen 2.08%). Taken together, we concluded that both SCC tumors (KPPA and A223) were heavily infiltrated by myeloid cell populations (CD11b + ) but not by CD4 or CD8 TILs.

Dysfunctional TILs in KPPA Tumors Suggest an Immunosuppressive TME
To assess the expression level of immune checkpoint molecules, we performed flow cytometry analysis by comparing WT or TB splenic control, and CD8 TILs from A223 or KPPA tumors. We found that both WT and TB splenic CD8 T cells expressed a similar level of checkpoint molecules, lymphocyte-activation gene 3 (LAG-3), programmed cell death 1 (PD-1), and T cell immunoglobulin and mucin domain 3 (TIM-3) ( Figure 4A), thus, we compared the CD8 TILs to TB splenic control ( Figure 4B). CD8 T cells in all groups expressed a negligible level of TIM-3 ( Figure 4A,B). Only CD8 TILs in A223 tumors expressed a high level of LAG-3 [14], while CD8 T cells in other groups did not ( Figure 4A,B). Of note, we observed that CD8 TILs in KPPA tumors expressed a much higher level of PD-1 compared with the splenic CD8 T cells, while CD8 TILs in A223 tumors expressed the highest level of PD-1 ( Figure 4A,B).
Next, we measured the expression of PD-L1 in different types of cells ( Figure 4C,D). The percentage of CD11b + PD-L1 + population was significantly higher in A223 tumors than the TB spleens or KPPA tumors, while there was no statistically significant difference between the TB spleens and KPPA tumors ( Figure 4C). With further examination of the PD-L1 + population in CD45 − cells, our data showed that all of the CD45 − populations expressed a minimal level of PD-L1 (<0.1%), including CD45 − cells from the TB spleens, A223 tumors and KPPA tumors, which was significantly lower than that in CD45 + CD11b + population in the TB spleens ( Figure 4D).
IFN-γ and TNF-α are commonly examined cytokines for evaluating T cell effector functions, especially for the polyfunctionality of T cells, which means T cells can produce not only one cytokine but also additional different cytokines. Polyfunctional T cells are effector T cells that retain cytotoxic potential and may be more effective in tumor suppression [22,23]. In addition, the loss of double producers (IFNγ + TNFα + ) is often considered as a sign of CD8 T cell dysfunction [24]. To examine the functional changes in CD8 TILs of KPPA tumors, we performed intracellular cytokine staining by flow cytometry to detect the intracellular level of single IFN-γ + , single TNFα + or double IFN-γ + TNFα + production in CD8 T cells from different groups. As a negative control, unstimulated naïve CD8 T cells did not produce much cytokine (unstimulated) ( Figure 5A). As a positive control, we stimulated CD8 T cells from WT B6 mice with anti-CD3/anti-CD28 beads for 3 days, then cultured these cells in the presence of PMA/ionomycin/BFA for 4-6 h, and examined the IFN-γ and TNF-α level. As shown in Figure 5A,B, the anti-CD3/anti-CD28 stimulated CD8 T cells contained the highest level of double producers (IFNγ + TNFα + ), indicating a robust polyfunctionality and strong effector functions. Then, we compared the splenic CD8 T cells from tumor-bearing mice (CD8 T cells-TB spleen) and the CD8 TILs from KPPA tumors (CD8 TILs-KPPA). A vast majority of splenic CD8 T cells are naïve CD8 T cells and it has been shown that naïve CD8 T cells tended to produce a high level of TNF-α upon stimulation [25]. Consistently, we found that the percentage of single TNF-α + CD8 T cells was the highest in the splenic CD8 T cells from tumor-bearing mice (CD8 T cells-TB Spleen) ( Figure 5A,B).
In contrast, we found that CD8 TILs from KPPA tumors exhibited the lowest level of effector functions, evidenced by a lower percentage of double producer (IFNγ + TNFα + ) in CD8 TILs than not only anti-CD3/anti-CD28-stimulated but also splenic naïve CD8 T cells ( Figure 5A,B). We found that there was no statistically significant difference in the percentage of single IFN-γ + CD8 T cells in different groups ( Figure 5B). Taken together, these results are consistent with the notion that CD8 TILs in KPPA tumors were exhausted with impaired effector functions.

Discussion
Comprehensive genomic and epigenetic analyses and flow cytometry-based assay of HNSCC samples demonstrate the heterogeneity in HNSCC molecular signature and immune landscape [4]. Our study further examined this premise and grouped the HNSCC patient cohort based on their genetic alterations of TP53 and PIK3CA. We found that the patients with both TP53 and PIK3CA gene alterations have a significantly greater hazard ratio and worse OS in 5 years. Consistent with previous studies that used CD8 T cell as a prognostic biomarker [26], we found that PIK3CA Amp /TP53 Mutated HNSCC patients expressed significantly lower levels of CD8 T cell gene signature in their tumor biopsy than PIK3CA WT /TP53 WT and PIK3CA Amp /TP53 WT patients ( Figure 1D). Of note, we also found significantly higher expression levels of M0 subset and lower levels of M1 subset in PIK3CA Amp /TP53 Mutated group than PIK3CA WT /TP53 WT and PIK3CA Amp /TP53 WT ( Figure 1D). Overall, we suggest that identifying a correlation between specific genetic and molecular signatures and immune TME may help to predict the clinical outcomes and provide potential therapeutic targets.
Our current study addresses a previously recognized limitation in the field, which is the inadequate preclinical models mimicking human HNSCCs characterized with specific genetic alterations, thereby hindering studies to further elucidate the mechanistic link between the immune TME and oncogenic drivers in SCCs and to develop new immunotherapies. In this regard, a recent study reported the establishment of a syngeneic mouse HNSCC model induced by 4-NQO that resembles the human tobacco-related HNSCC mutanome, including mutations in TP53 and FAT3 but not in PIK3CA [27]. To our knowledge, we first showed that genetic mutations of both PIK3CA hyper-activation and TP53 deletion in K15 + stem cells resulted in spontaneous development of KPPA tumors including SCCs. Consistent with findings in TCGA HNSCC patients of PIK3CA Amp /TP53 Mutated cohort, we found KPPA tumors harbored a low level of CD8 TILs that expressed a higher level of PD-1 and exhibited reduced polyfunctionality, suggesting that these CD8 TILs were chronically activated and experienced exhaustion. However, the CD8 TILs in KPPA tumors did not express TIM-3 or LAG-3, an observation different from our previous studies of a different SCC model in which the CD8 TILs coexpressed a high level of PD-1 and LAG-3 [14]. These data suggest that the immune phenotypes of CD8 TILs vary and may be influenced by tumor cells with different genetic alterations or differentiation status. Although anti-PD-1/PD-L1 have been approved by FDA for treating HNSCCs, the overall response rate is still below 20% [28]. Various clinical studies of treating HNSCC patients with anti-PD-1/PD-L1 in combination with targeted therapy, radiation and chemotherapy are ongoing [28]. By further developing syngeneic transplanted mouse models of KPPA tumors, we may be able to provide a platform for better understanding the immune TME and developing combinatorial treatment and predictive markers for clinical outcomes of immunotherapies.
Notably, KPPA tumors were heavily infiltrated with MDSCs that exhibited a drastic increase in the ratio of PMN-MDSC versus M-MDSC. VEGF and TGF-β have been reported to recruit MDSCs into the TME and affect their differentiation [29]; however, it remains to be addressed how PMN-MDSCs were preferentially increased in KPPA tumors. Treating SCC with anti-CD11b monoclonal antibodies has been shown to prevent the recruitment of myeloid cells into tumors, which attenuates tumor growth, and enhances antitumor response to radiation [30]. PD-L1 expression on myeloid cells can induce T cell exhaustion and reduce the efficacy of T cell-associated immunotherapy in solid tumor [28]. Thus, anti-PD-L1 in combination with myeloid-targeted therapy may likely lead to better responses to T cell-mediated immunotherapy. In this regard, by generating new KPPA SCC lines, it will allow us to better investigate the mechanisms and responses to combinational targeted therapy for HNSCCs and beyond.

Analysis of Patient Samples Obtained by TCGA
The Cancer Genome Atlas (TCGA) RNA-seq data and clinical data of HNSCC cohorts were obtained from the cBioPortal (https://cbioportal.org). The TCGA datasets provided comprehensive genomic sequencing and signatures which allowed the identification of patients with TP53 mutations and PIK3CA copy number changes. Within cBioPortal and under the category of head and neck cancers, we downloaded data from two cohorts of HNSCCs (TCGA, Firehose Legacy, n = 528 samples; and TCGA, PanCancer Atlas, n = 489 samples). We used the second HNSCC dataset (TCGA, PanCancer Atlas, n = 489 samples) for analysis in Figure 1A. These two data sets were merged utilizing patient IDs (n = 528) and used for analysis in Figure 1B-D; however, we only had 489 analyzable records due to 39 having missing group information (n = 489). Patients were divided into four different groups: (1) amplification and gain of PIK3CA copy number (PIK3CA Amp ) and truncation and missense of TP53 gene (TP53 Mutated ) (n = 294); (2) PIK3CA Amp and wildtype TP53 gene (TP53 WT ) (n = 85); (3) no amplification and gain of PIK3CA copy number (PIK3CA WT ) and TP53 Mutated (n = 56); and (4) PIK3CA WT and TP53 WT (n = 54). Only patients with available survival data were included for survival analysis.
The association of mutation grouping with survival was evaluated with Cox regression. The hazard ratios and associated p-value are presented with Kaplan Meier curves. Pairwise comparisons utilizing the log-rank test were made between each group. Censoring occurred when a patient was denoted as alive at the end of the observed time period. p-values are reported based on a null hypothesis of no effect against a two-sided alternative. Analyses were performed using SAS 9.4 (SAS Institute; Cary, NC, USA).
The downloaded RNA-seq data from HNSCC tumor biopsies were uploaded onto CIBERSORT (https://cibersort.stanford.edu/index.php). CIBERSORT input the RNA-seq data into a matrix of reference gene expression signatures (LM22), which contains 547 genes that distinguish phenotypes of 22 human hematopoietic cell [31,32]. This output was used to estimate the relative proportions of the immune cell type composition of a tumor biopsy. Cell types were summarized by group using the mean, standard deviation, median, and range for continuous variables (Supplemental Table S1). The differences between groups were explored with the omnibus Kruskal−Wallis test due to violations of the normality assumption. Cell types with at least 90% of the subjects in a group having no expression were not tested due to suspect validity of the omnibus test. Pairwise comparisons between groups using Dunn's Test were performed on cell types that had achieved statistical significance with the omnibus test at the 0.05 level. p-values are reported based on a null hypothesis of no effect against a two-sided alternative. Analyses were performed using SAS 9.4 (SAS Institute; Cary, NC, USA).

Mouse Models
Mice were bred to contain the following alleles: a K15 promoter-driven Cre recombinase (K15.CrePR1) [33], a R26Stop FL P110* conditional allele that carries a loxP-flanked Neo-STOP cassette preceding a constitutively active PIK3CA allele (encoding p110α protein) targeted to the Gt(ROSA)26Sor locus [34], and a conditional TP53 gene knockout [35]. Induction of the recombinase activity was achieved by applying 100 µL of RU486 (0.2 µg/µL in 70% ethanol) orally or to the shaved dorsal flank skin of 5 to 12 week-old mice for 5 consecutive days. Afterwards, mice were examined weekly for tumor development (Supplemental Table S2). When tumor size reached 2 cm in any dimension or other humane end points were met, mice were euthanized in accordance with institutional guidelines. Mice were maintained under specific pathogen-free conditions in the vivarium facility of University of Colorado Anschutz Medical Campus (AMC). Animal work was approved by the Institutional Animal Care and Use Committee (IACUC, 00037, 21 Aplril 2020) of University of Colorado AMC (Aurora, CO, USA). PBS, and incubated on the slides overnight at 4 • C. Slides were washed with PBS three times, and 1:400 secondary antibodies, goat-anti-chicken conjugated to Alexa Flour 594 (red) and goat-anti-mouse conjugated to Alexa Flour 488 (green) (ThermoFisher Scientific, Cat# A-11042, Catalog # A-11029, respectively) in 0.05% TBST were added and incubated for 60 min at RT. Finally, slides were observed and evaluated using an Olympus IX83 microscope.

Statistical Analysis of Murine Samples
Statistical analysis was performed using either two-way ANOVA with Tukey's multiple-comparison test correction or Kruskal−Wallis test with Dunn's multiple-comparison test correction. GraphPad Prism 8.4.3 software (GraphPad Software, La Jolla, CA, USA) was employed, with significance determined at p < 0.05.  to Z.C., X.W. was supported by an AAI Careers in Immunology Fellowship. R.A.W. is supported by a NIH F31 fellowship (F31DE027854). S.M.Y.C. is supported by a NIH T32 fellowship (T32 AI007405) and a NCI T32 fellowship (T32 CA174648). A.G.N. and D.G. were partially supported by NCI P30CA046934.