Genetically engineered mouse models of head and neck cancers

The head and neck region is one of the anatomic sites commonly afflicted by cancer, with ~1.5 million new diagnoses reported worldwide in 2020 alone. Remarkable progress has been made in understanding the underlying disease mechanisms, personalizing care based on each tumor’s individual molecular characteristics, and even therapeutically exploiting the inherent vulnerabilities of these neoplasms. In this regard, genetically engineered mouse models (GEMMs) have played an instrumental role. While progress in the development of GEMMs has been slower than in other major cancer types, several GEMMs are now available that recapitulate most of the heterogeneous characteristics of head and neck cancers such as the tumor microenvironment. Different approaches have been employed in GEMM development and implementation, though each can generally recapitulate only certain disease aspects. As a result, appropriate model selection is essential for addressing specific research questions. In this review, we present an overview of all currently available head and neck cancer GEMMs, encompassing models for head and neck squamous cell carcinoma, nasopharyngeal carcinoma, and salivary and thyroid gland carcinomas.


INTRODUCTION
Head and neck cancers (HNC) include squamous cell carcinomas (HNSCC) arising from oral, oropharyngeal, nasopharyngeal, hypopharyngeal, and laryngeal epithelia in addition to adenocarcinomas arising from the salivary gland carcinomas (SGC) and thyroid cancers (TC) glands. Despite significant differences in incidence across different countries [1], HNSCC is the most common malignancy of the head and neck [2] and the sixth most common cancer overall with~878,000 new diagnoses globally in 2020 [3]. Variations in global HNSCC burden reflect differences in exposure to known risk factors including smoking and tobacco use [4], alcohol consumption [5], betel nut chewing [6], oncogenic human papillomavirus (HPV) infection [7], and low socioeconomic status [8,9]. However, there has been a shift in the pattern of risk factors that contribute to HNSCC development in recent decades. Most notably, smoking and tobacco use have been declining [10] and, consequently, the number of HNSCC cases attributed to them. In stark contrast, HPV-associated HNSCC (HPV+ HNSCC) cases have been steadily increasing, with~80-90% of oropharyngeal squamous cell carcinomas (OPSCCs) being HPV+ [11,12]. HPV negative HNSCC (HPV-HNSCC) tumors frequently harbor alterations in TP53 and CDKN2A, although several other tumor suppressor genes are also commonly mutated (FAT1, NOTCH1, KMT2D, NSD1, and TGFBR2) [2,13]. Notably, RAS alterations are very rare in HPV− HNSCC despite carcinogen exposure, and PIK3CA is the only frequently mutated oncogene [13].
While the rates of mutation appear to be similar between HPV+ HNSCC and HPV− HNSCC [14], distinct mutational patterns exist between the two entities. Specifically, HPV− HNSCC tumors often present with PIK3CA, TP53, and CDKN2A alterations [13], while HPV+ HNSCC tumors have been shown to exhibit PIK3CA, TRAF3, CYLD, and E2F1 alterations [2,15]. Several recent advances in our understanding of the role that HPV infection plays in OPSCC tumorigenesis have changed the field dramatically [16]. HPV16 is the most prevalent strain among HPV+ HNSCC malignancies [17], accounting for~90% of all HPV+ HNSCC cases. Akin to cervical squamous cell carcinoma in women [18], E6 and E7 are the key oncogenes driving HPV+ oropharyngeal tumor development [19] by inactivating p53 and pRb, respectively [20], and their ability to subsequently induce proliferation and malignant transformation of epithelial cells [21,22]. Notably, the molecular pathways driving HPV+ HNSCC tumors are also associated with distinct clinicopathological characteristics such as age at diagnosis, immunologic profile, and microbiome status [23,24]. Thus, patients with HPV+ and HPV− HNSCCs require different treatment strategies in the management of disease [25,26]. Moreover, the development of HPV vaccines has raised the hopes for more effective prevention and management of HPV-associated neoplasms, including those originating within the head and neck epithelia [20]. However, until preventative vaccination is broadly adopted within the general population, the incidence of HPV+ OPSCCs is expected to continue to rise [27].
While the survival for HNSCC patients has improved over the past decades, it remains one of the most lethal malignancies worldwide, with 444,339 reported deaths in 2020 [2,3]. Early-stage HNSCC is usually managed surgically, and results are optimal in this group with long-term survival rates exceeding 80% in some cohorts [2,28,29]. In laryngeal carcinoma in particular, primary radiotherapy is also an option [2,29,30]. However, pathologic risk features (e.g., extranodal extension and perineural invasion) determine the need for adjuvant treatment (radiotherapy and/or cisplatin-based chemotherapy). In contrast, late-stage disease and recurrent/metastatic disease remain difficult to manage despite an expansion in available treatment options over the past two decades, and patient prognosis is generally dismal [2,29].
Targeted therapy [31] and immunotherapy [32,33] have proven effective for only a small subset of the patient population in prolonging survival. Efforts to unravel the molecular pathogenesis of HNSCC through The Cancer Genome Atlas Project (TCGA) and other large scale genomic and transcriptomic analyses have identified a significant number of potentially actionable targets (e.g., PI3K, NOTCH1, TRAF3), although there remains a dearth of effective therapies [19,34]. Thus, the current treatment landscape underscores the need for tools that can reliably recapitulate the molecular and cellular complexities of HNSCCs. These include unique genetic and transcriptional differences, as well as molecular and cellular heterogeneity between HNSCC subtypes, which all contribute to a wide diversity in treatment responses that need to be accounted for in order to enable the development and preclinical validation of new therapeutic options.
Unlike HNSCCs, SGCs only account for <0.5% of all cancers and around 3-5% of all head and neck cancers [35]. They arise in any of the three major salivary glands (submandibular, sublingual, and parotid gland) or the minor glands, with mucoepidermoid carcinomas (MEC), adenoid cystic carcinomas (AdCC), and polymorphous adenocarcinomas being the most frequent histopathological subtypes [36,37]. Notably, SGCs display remarkable histologic heterogeneity that presents significant challenges in their diagnosis and management [38]. Surgical resection and adjuvant radiotherapy remain the primary treatment modality, however, there is a lack of systemic treatment options largely in the recurrent or metastatic disease setting [39]. The rarity of SGCs and their vast clinicopathological diversity impedes large-scale patient accrual to conduct prospective trials. Furthermore, the published randomized clinical trials have amalgamated different histological types and hence the performance of therapy can be limited. Consequently, there is an unmet need of subtype-specific approaches for the management of SGCs. Thus, in vivo models that can recapitulate salivary gland carcinogenesis are indispensable in identifying the cellular and molecular mechanisms to develop tumor subtype-specific treatments.
TC comprised 2.3% of all new cancer diagnoses in 2022 and accounted for 0.4% of all cancer deaths [40]. Specifically, in HNC, thyroid malignancies accounted for roughly 41% of all diagnoses in 2020 (n = 586,202 worldwide), but only 9% (n = 43,646) of head and neck cancer-associated deaths [41]. Fortunately, TC mortality remains low, with 5-year survival rates of >98%, although certain subtypes of TC are classified among the deadliest malignancies described in medicine. The thyroid gland is composed of several cell types, each with different histology. Follicular and epithelial cells are the major cell types that function to concentrate iodine and produce thyroid hormones. C-cells or para follicular cells are neuroendocrine cells that synthesize calcitonin. Molecular subtyping has aided in the classification of follicular cell-derived neoplasms including differentiated TC: papillary thyroid carcinomas (PTC) and follicular thyroid carcinomas (FTC), poorly differentiated thyroid carcinoma (PDTC), and undifferentiated anaplastic thyroid carcinoma (ATC). PTC is the most common subtype of TC accounting for about 80% of all TC cases, followed by FTC (about 15%), PDTC (>5%) and ATC (1-2%) [42].
Differentiated TC in general has excellent prognosis as opposed to the more aggressive PDTC and ATC that account for half of the TC-related deaths [43]. Medullary thyroid carcinoma (MTC) is a rare type of TC representing about 4% of all TC that is derived from C-cells [44]. About 20-25% of MTCs can arise due to an inherited syndrome, such as Multiple Endocrine Neoplasia (MEN) type 2A or type 2B or familial MTC [45,46]. Management of TC differs according to the type of cancer. Surgery, with or without radioiodine therapy, is the primary treatment option for differentiated TC. While MTC is usually treated with upfront surgery, systemic targeted therapy can be given in the locoregionally advanced stages, either in an adjuvant or neoadjuvant setting [47]. ATC is commonly unresectable and metastatic at the time of diagnosis, making systemic treatment, with or without external beam radiation therapy, the first-line therapy. Development of targeted therapies for progressive metastatic radioactive iodine refractory differentiated TC, metastatic MTC, and ATC patients resulted in improved clinical outcomes and survival [48][49][50]. Comprehensively evaluating such therapies for advanced and aggressive TC has been historically challenging in clinical trials due to the rarity of ATC and MTC. Having reliable and established in vivo models for TC can help expedite novel therapeutic development.
Given the challenges associated with treating patients with any of the above HNCs, the identification, selection, and/or development of in vivo models suitable for interrogating the unique pathobiology of these cancers is crucial. Compared to other animal models, the development of mouse models offers easier genetic engineering, higher rates of successful tumor formation, faster reproduction, and lower development and maintenance costs [51]. Genetically engineered mouse models (GEMMs) in particular, can recapitulate most of the heterogeneous characteristics of various tumors including the tumor microenvironment [52,53]. In head and neck cancer, progress in the development of GEMMs has been slower than in other major cancer types, such as melanoma [54], non-small cell lung carcinoma [55], and pancreatic adenocarcinoma [56]. This is due in part because HNCs are primarily characterized by genetic instability leading to chromosomal translocations or the frequent loss or gain of chromosomal regions involving tumor suppressors (e.g., CKND2A, TP53) but many available GEMMs employ oncogene overexpression instead (e.g., RAS) to model HNC carcinogenesis [2]. However, several GEMMs with or without concurrent carcinogen exposure are now available, enabling research questions regarding many aspects of carcinogenesis [57,58], treatment response and resistance [59][60][61], and relapse/recurrent disease [62].
GEMMs enable researchers to study the biological effects of oncogene overexpression or tumor-suppressor gene inactivation on downstream signaling and target genes in a defined genetic background [63]. Thus, tumors that develop in GEMMs are more likely to be histologically and genetically accurate representations of human cancer. Constitutive knockout transgenic mice were developed by depleting or silencing the gene of interest in germline cells leading to permanent inactivation of the target gene in every cell of the organism. However, this is inconsistent with the HNSCC mutational landscape, and the timing of gene loss is critical in carcinogenesis. To overcome this limitation, conditional and conditionally inducible models were developed that enable tissue-and time-specific regulation of the gene of interest by external stimulation by chemicals or virus [64]. Exposure to the carcinogen 4-NQO has been integrated with certain models for development of spontaneous tumors. DNA damage by chemicals occurs randomly, thus the stochastic nature of 4-NQO induced mutations makes these GEMMs less genetically defined [65]. It is therefore important to note that the approach employed in GEMM establishment and implementation, recapitulates only certain aspects of the disease course with accuracy. Moreover, models are usually developed for specific tumor sites within the head and neck region (e.g., oral cavity). Thus, appropriate model selection is essential for addressing specific research questions. At the same time, as more and more information on the molecular background of HNCs comes to light, new models which accurately represent the underlying mechanisms of head and neck carcinogenesis are necessary to improve our understanding of the disease and develop improved precision medicine therapies. In this review, we present a critical overview of the existing head and neck cancer GEMMs and discuss strengths and limitations of each model with respect to the research questions under investigation and the different anatomic sites where these cancers can develop.
Overexpression of Cyclin D1 (CCND1) has been reported as an early event in oral and esophageal SCC [82,83]. An Epstein-Barr virus (EBV) lytic promoter ED-L2 resides within the 3′ UTR of the EBV latent membrane protein-1 gene and extensive characterization of its promoter activity revealed tropism for human aerodigestive epithelial cells including the tongue, esophagus, and forestomach. Thus, the ED-L2 promoter-mediated expression of CCND1 combined with Trp53 knockout leads to the development of SCC within the buccal mucosa, tongue, and esophagus of C57BL/6 mice [84]. However, buccal SCC develops far less frequently than tongue SCC. Notably, this model was also able to produce nodal metastasis, with approximately a quarter of mice displaying positive lymph nodes. A major caveat of this model is that while CCND1 is reported to be overexpressed in HNSCC [85,86], it remains unclear if it is a main driver of carcinogenesis at these sites [34]. Thus, while its combination with Trp53 loss in mice is able to produce squamous carcinomas within the oral cavity, the molecular landscape of these tumors in humans differs substantially from those commonly associated with HPV-HNSCC.
GEMMs have successfully used combinations of oncogene overexpression and carcinogen exposure to promote malignant tumor development. Specifically, Kalish et al. [78] created a conditional and doxycycline-inducible C57BL/6 model where Bmi1 is overexpressed in basal epithelial cells and that forms tongue SCC upon concurrent 4-Nitroquinoline 1-oxide (4NQO) exposure. While the conditional nature of this model limits transgene impact to squamous epithelia, administration of doxycycline to the drinking water leads to transgene activation not only in the mucosal epithelia of the oral cavity but also in the cutaneous epithelia of the skin. Perhaps more importantly, BMI1 overexpression is only reported in <4% of HNSCC patients in the TCGA and is therefore not considered a primary driver of HNSCC oncogenesis [34]. This is further highlighted by the inability of Bmi1 to form malignant tumors without concurrent 4NQO exposure. Importantly, some alterations induced by 4NQO exposure in mice such as Hras mutations are mutually exclusive with BMI1 overexpression in human HNSCCs, further limiting the utility of this GEMM. PIK3CA mutations and/or overepression are commonly identified in >40% of all HNSCCs in the TCGA and this pathway is associated with cell survival and malignant transformation. Du et al. developed a conditionally inducible GEMM where PIK3CA overexpression is regulated by the anti-progesterone and anti-glucocorticosteroid agent RU486 (mifepristone). While in this model PI3K pathway deregulation alone is insufficient for tumor development [81], concurrent RU486-induced PIK3CA overexpression and 4NQO administration results in over 40% of tumors exhibiting increased invasion and metastasis. This model of concurrent carcinogen exposure and PIK3CA overepression closely recapitulates drivers of human HPV-HNSCC, where tobacco exposure is a common carcinogen and of PI3K overexpression and/or hyperactivation of its downstream target PDK1 promote a more aggressive biological behavior [87].

Oropharynx
Numerous GEMMs intended to model HPV+ HNSCC have been developed by employing the HPV oncogenic proteins E6 and/or E7 [67][68][69][70][71][72][73]84]. While initially developed as an HPV16 model for cervical SCC, these animals were also used to model HPV+ HNSCC upon identification of an etiologic role for HPV in oropharyngeal HNSCCs. A keratin 14 promoter (K14) was used to express the viral  Oropharyngeal SCCs were often carcinomas in situ although invasive cancer with regional lymph node metastasis does occur [79] oncogenes in FVB/N mice, where the independent expression of both E6 and E7 in bi-transgenic mice (i.e., K14-E6; K14-E7) demonstrates greater transforming potential than each oncogene alone. However, the constitutive expression of these oncogenes only results in a hyperkeratotic and hyperplastic skin phenotype, with wrinkled and hairless skin. Also, embryonal transgene activation induces developmental defects and immune tolerance. An improved HPV+ HNSCC GEMM based on combination of inducible expression of HPV E6/E7 and oncogenic Kras G12D (i.e., LSL-Kras G12D ) in FVB/NJ mice was developed by Zhong et al. [73].  [58]. Notably, this GEMM uses a PIK3CA H1047R transgenic allele but this mutation is not common in HPV+ HNSCCs. Unconventional GEMMs have also been developed that utilize the Sleeping Beauty transposon mutagenesis system (SB100) which provides an opportunity to screen for cooperating oncogenes/tumor suppressors. For example, the Luc-HPVE6/E7-NRAS G12V -SB100 model is a constitutive HPV16 E6/E7 model that employs luciferase to track cells in the oral cavity and draining lymph nodes that get transfected with the SB100 plasmid system [66]. This model uses submucosal delivery via injection, thus avoiding a more time-consuming breeding approach. However, the model depends on mutant NRAS to form tumors. As The Cancer Genome Atlas (TCGA) and others have reported, RAS is infrequently involved in HNSCC [2,34], thus limiting the relevance of this model of human HNSCC. Acknowledging this limitation, Lin et al. replaced NRAS G12V with constitutively active AKT, a PI3K downstream target but tumor formation is only reported among immune depleted mice [66]. Of note, NRAS mutant tumors only develop after depletion of CD3 + T-cells with a sarcoma-like histology consisting of spindle-shaped tumor cells instead of squamous cells. Development of lymph node metastases and the non-invasive visualization offered by the luciferase reporter are important features of this model.
While the aforementioned models have employed HPV oncogenes, these GEMMs have only been shown to develop oral cavity tumors and high-risk HPV infections have been associated almost exclusively with oropharyngeal SCC (OPSCC) in humans. In stark contrast, Carper et al. reported a conditional and inducible knock-in model using C57BL/6J mice, that offers balanced postnatal E6 and E7 expression with the appearance of spontaneous oropharyngeal tumors (Table 1). This was achieved by establishing a protocol for local administration of tamoxifen to the submucosa of the murine equivalent of a human oropharyngeal region to achieve mosaic transgene expression [79]. While most tumors were early-stage carcinomas, frank malignancy is observed when the mice are exposed to chemical carcinogens (e.g., 4NQO) or by combining E6/E7 overexpression with expression of mutant PIK3CA, specifically PIK3CA E545k . Consequently, this is the first model to use a knock-in system to accurately recapitulate many of the unique molecular features of HPV-associated tumors of the oropharynx [79]. Additionally, robust leukocyte infiltration was observed in response to exogenous viral antigen exposure in premalignant lesions and the tumors exhibited histopathological and molecular features observed in human HPV(+) OPSCC [79]. Thus, this model forms the full spectrum of premalignant lesions (mild to severe dysplasia) and OPSCC marked by consistent lymphocytic infiltration, making it ideal for studies of the tumor immune microenvironment and host-tumor interactions.

Nasopharynx
Nasopharyngeal carcinomas (NPCs) have been stratified into three subtypes: keratinizing SCC, non-keratinizing SCC, and undifferentiated or poorly differentiated carcinoma. The etiology of nonkeratinizing NPC is associated with EBV infection [88]. QingLing et al. established L2/LMP1 B95 -8 /EGFP transgenic mice using the EBV ED-L2 promoter to drive expression of the EBV Latent membrane protein 1 (LMP1) in aerodigestive epithelial cells. LMP1 expression inhibits Wilms' tumor gene on the X chromosome expression, leading to increased expression of β-catenin in these mice. Squamous epithelial hyperplasia and atypia was observed in the nasopharynx and oropharynx of founder (F0) and first generation (F1) animals at 50% penetrance; however, progression to neoplasms was not observed [89]. Therefore, the development of GEMMs to model human NPCs is highly warranted.

Current limitations of HNSCC GEMMs
Despite technological improvements in surgery [90] and radiation [91,92] and the rapid development of immunotherapies and targeted therapies in recent decades [2,29,31], clinical outcomes of HNSCC patients remain mostly unchanged, especially in HPV− cases where tumor mutational burden is higher [2,3]. On the one hand, tumors in the oral cavity, oropharynx, and larynx have different clinical features and biological characteristics. On the other hand, HPV+ and HPV− HNSCC are now considered distinct disease entities with different genetic landscapes, biologic behaviors, treatments, and prognoses [21]. This is also reflected in the most recent 8th Edition of the TNM classification system of the American Joint Commission on Cancer [93]. Therefore, personalized treatment based on HPV status, tumor histology, and the specific genetic alterations of each tumor will improve the survival and life quality of HNSCC patients.
While the intrinsic complexity of HNSCC tumor cells is the key determinant driving tumorigenesis, the role of the tumor microenvironment and tumor immune microenvironment is more and more acknowledged as a critical parameter of tumor progression [94,95], treatment response [96] and prognosis [97]. In GEMMs, normal cells undergo the multi-step process of malignant transformation and thus allow for the development of a stromal response that can recapitulate the complex tumor microenvironment and immune microenvironment. This enables GEMMs to accurately recapitulate tumor growth, treatment responses, and tumor-stroma and tumor-immune system interactions, features often lacking in xenograft or syngeneic transplant models [53]. However, while currently available GEMMs provide invaluable information on head and neck tumorigenesis and play critical roles in identifying and testing novel therapeutic approaches, most HNSCC GEMMs suffer from low tumor incidence. This largely hampers the reproducibility of preclinical results in clinical follow-up studies. Improving current models to generate high-efficiency, versatile HNSCC GEMMs is essential to understand the molecular mechanisms and develop new therapies. To achieve this, there are several obstacles that must be overcome.
E6/E7 sequences from HPV are widely used to generate transgenic mice prone to HNSCC tumorigenesis, aiming to recapitulate human HPV+ HNSCC. However, the random integration into the mouse genome, often as concatemers and/or at multiple loci, may lead to genetic instability or even phenotypic instability during breeding, as has been observed in studies where E6 transgenic mice were crossbred with E7 transgenic mice [73]. In comparison, the conditional knock-in of E6/E7 cDNA expression cassettes into the Rosa26 locus induced stable and balanced E6 and E7 expression [79]. Even so, E6 and E7 expression could only induce hyperplasia in most cases. Malignant transformation is rarely achieved without additional driver mutations or exogenous carcinogens, indicating that E6 and E7 oncoproteins are incapable to initiate and promote HNSCC tumorigenesis without further triggers. In this regard, a model that would include the full-length HPV genome could potentially more accurately recapitulate the pathogenesis of HPV+ HNSCC and induce tumors without the need for concurrent exposure to carcinogens (e.g., 4NQO).
The TCGA dataset includes 504 HNSCC samples with integrated mutation and copy number alteration (CNA) data. Notably, TP53 is mutant in 72% of samples, CDKN2A is altered in 54% of samples (22% mutation and 32% deletion), and PIK3CA is altered in 39% of samples (18% mutation and 21% amplification). Moreover, alterations of these 3 genes cooccur, indicating they may all be essential for HPV− HNSCC tumorigenesis. GEMMs harboring alterations of Tp53, Cdkn2a, and Pik3ca are available, K14-Cre ERT2 ; R26-LSL-Pik3ca H1047R ; Trp53 fl/fl ; Cdkn2a fl/fl mice could be generated to study the combinatorial effects of these three drivers. Considering that developing this strain requires several generations of breeding and thus is very time and labor consuming, CRISPR knockouts could be an alternative approach. Specifically, traditional embryonic stem cell (ESC) mediated targeting strategies employ homologous DNA recombination to perform genetic recombination followed by injection of these modified ESCs into blastocysts to generate chimeric embryos within pseudopregnant mice [98]. This process is highly technical and time-consuming since these chimeric animals must then be mated with wild-type mice to screen for passage of the genetic information to subsequent generations which can take 10 months to over 1 year. In contrast, CRSIPR/Cas9 methodologies employ a guide RNA (gRNA) mediated targeting strategy and a homologydirected repair mechanism to repair the damaged doublestranded DNA which can facilitate either knockin or deletion of genetic material. Specifically, Cas9 transgenic animals [99] can be crossed with an existing HNC GEMM and these compound mutant mice can then be used to study several knockout mutations following delivery of gRNA(s) targeting the gene of interest. Importantly, the desired genetic manipulation is achieved within the first generation of offspring reducing breeding schema and time needed to generate models to less than 4 months. Using this CRIPSR-based approach, other potential driver genes including CCND1 (25% amplification), SOX2 (16% amplification), NOTCH1 (18% mutation and 4% deletion), and FAT1 (23% mutation and 8% deletion) can be examined in less time. Interestingly, PIK3CA and SOX2 are frequently co-amplified since they are both localized on chromosome 3q26, but whether overexpression of PIK3CA and SOX2 have synergistic effects in HNSCC is unknown. To determine the roles of these frequently amplified genes in vivo, CRISPR activation (CRISPRa) could be an effective approach.
The availability of the Cre-LoxP (for instance, K14-Cre ERT2 ) and Tet-on systems enabled the development of HNSCC GEMMs where tumor development is induced in a spatiotemporal manner. However, the exact site and timing for HNSCC initiation could be further optimized. First, most HNSCC models activate Cre 4-8 weeks after birth, which may be too early given that HNSCC is a disease predominantly affecting the elderly. Second, to increase the effectiveness and specificity of Cre activation, intraperitoneal injection of Tamoxifen could be replaced by topical administration of 4-Hydroxytamoxifen (4OHT), including brushing in the oral cavity or dripping in the pharynx, or by targeted submucosal delivery of Tamoxifen as performed by Carper et al. [79]. Furthermore, adenoviral or lentiviral delivery of Cre to the oral cavity or pharynx could be used to initiate genetic recombination of driver alleles in squamous cells, which would decrease genotype complexity of experimental animals and thus shorten the breeding process.
Current mouse models indicate that HPV infection alone may not be capable of initiating HNSCC and additional exogenous stimuli are essential for malignant transformation [58,78,81,87]. 4NQO, a quinolone derivative to induce DNA lesions, has been used in some HNSCC models as a carcinogen to enhance HPVinduced tumorigenesis (Table 1). However, the use of 4NQO in HPV+ models has some shortcomings. First, 4NQO is a substitute for tobacco exposure, which is not considered a major risk factor for the majority of HPV+ HNSCC where incidence rates appear to be increasing primarily in male non-smokers. Second, transcriptional or epigenetic deregulation rather than genetic mutation alone may drive HPV+ HNSCC carcinogenesis. Thus, a comprehensive screening approach using siRNA or cDNA libraries may help to better understand HPV+ HNSCC initiation. The roles and mechanisms for many other natural carcinogens closely related to HPV− HNSCC, including PAH and nitrosamine in cigarettes, tannins in areca nut, as well as alcohol, are yet to be determined in vivo. These carcinogenic substances could be used to recapitulate the exposures of HPV− HNSCC to develop models that provide more accurate information on various questions related to HNSCC prevention and treatment. Third, most of the models failed to recapitulate the molecular and histological features of human HNSCC (Table 1), with the notable exception of Bmi-1 conditionally inducible transgenic mice, exposed to 4NQO [78].

SALIVARY GLAND CARCINOMA
Despite the remarkable heterogeneity of SGC, there have been several attempts to recapitulate these tumors in vivo. Conventionally, the mouse mammary tumor virus long terminal repeat promoter was used to develop GEMMs of breast cancer. Since this promoter directs gene expression to secretory glands, extramammary gland tumors were reported in the salivary glands of these mice [100]. Consequently, new murine models of salivary gland tumorigenesis driven by the MMTV promoter were characterized in detail (Table 2). Brodie et al. developed a MMTVdriven c-neu transgenic mouse strain that exhibits increased rates of parotid tumor onset on a Trp53-deficient background [101]. Moreover, Zboray et al. employed MMTV-rtTA animals crossed to TetO-Akt3 transgenic mice to direct Akt3 overexpression to the salivary glands. These bi-transgenic mice (MMTV-tTA; TetO-Akt3) exhibit tumors upon doxycycline administration that were characterized as AdCC [102]. Notably, incidence of AdCC in the salivary glands is frequently associated with recurrent translocation and oncongenic fusion of v-myb avian myeloblastosis viral oncogene homolog (MYB) and nuclear factor I/B (NFIB) and/or overexpression of the MYB or MYBL1 transcripts [103][104][105]. Hence, to examine AdCC, a MYB-NFIB fusion construct was conditionally expressed in the salivary gland by crossing to MMTV-Cre mice. There is no robust expression of MYB-NFIB transcripts in salivary gland tissues nor is malignancy observed [106]. The introduction of a conditional Trp53 knockout allele also fails to yield salivary tumors in these animals even though MYB-NFIB; MMTV-Cre; p53 +/fl mice present with tumors in the upper mammary glands [106]. A MYB-NFIB; MMTV-Cre; Cdkn2a +/− GEMM also does not develop salivary tumors and the predominant phenotype of this model is B-cell leukemia [107]. Even though MYB-NFIB fusions have been reported to be a key event in the development of AdCC in humans, these studies failed to recapitulate its role in inducing SGCs in mice. More robust attempts to target MYB-NFIB to salivary gland-specific cell types and/or addition of concurrent mutations in genes such as NOTCH1, CDKN2B, NRAS, HRAS, ALK, c-KIT, TSC, PIK3CA, PTEN, or NF1 might be required to promote AdCC initiation and progression [108].
Although numerous MMTV promoter-dependent models of SGCs have been developed, the appearance of mammary gland tumors has been the major limitation of these GEMMs. Therefore, more selective approaches that enable targeting of salivary gland cell types are highly warranted. Using an RU486-inducible keratin 5-driven Cre allele (K5-CrePR), the tumor suppressors Pten and Smad4 are specifically deleted in K5-positive basal epithelial cells in the salivary glands [109]. Deletion of either Pten or Smad4 results in spontaneous pleomorphic adenomas whereas deletion of both genes leads to the development of several subtypes of malignant SGCs with salivary AdCC being the most frequent. In addition, salivary duct carcinomas and salivary adenosquamous cell carcinomas are observed in the double knockout mice [109]. A K14-Cre-dependent β-catenin gain-of-function (β-cat GOF ) and Bmpr1a loss-of-function (Bmpr1a LOF ) double mutant model was developed using the β-catenin lox(ex3) and Bmpr1A flox alleles. While no tumors develop in the single mutant mice, double mutant mice develop aggressive squamous cell carcinomas (SCCs). These tumors form exclusively in the submandibular salivary glands requiring euthanasia after 75-90 days, and no tumors are observed in other K14 expressing tissues [110].
A mouse strain bearing a chemically induced recessive point mutation in the Gon4-like (Gon4l) gene was generated in which B lymphopoiesis is blocked at an early stage without hampering other hematopoietic cell development [111]. Mice homozygous for this mutation were named Justy (just T-cells) and in an independent study were shown to spontaneously develop SGCs with myoepithelial and basaloid differentiation. This model suggests an association between SGC development, downregulation of Gon4l mRNA transcripts and protein levels and/or the consequent depletion of B-cells, which can be further explored to understand cellular and molecular mechanisms [112]. The submandibular gland secretory protein b (Smgb)-SV40 T antigen (Tag) transgenic mouse model is another powerful tool for delineating molecular mechanisms underlying salivary gland tumorigenesis [113]. Using the neonatal Smgb promoter, a Tag transgene, which functionally inactivates p53 and Rb, is expressed in neonatal submandibular gland pro-acinar cells and terminal tubule cells and in the intercalated ducts of the adult gland. Tag expression in the intercalated ducts triggers progressive hyperplasia, dysplasia, and adenocarcinoma [114]. The importance of identifying oncogenic fusion genes for the development of treatment modalities is emphasized in a transgenic mouse model where Cre-regulated conditional expression of cyclic AMP-regulated transcriptional coactivator 1 and mastermind-like 2 fusion gene (CRTC1-MAML2) results in the formation of SGC resembling human MEC in histological and molecular features [115]. Even though this model is valuable to characterize MEC, it lacks the ability to target the expression of CRTC1-MAML2 to specific ductal cell populations since the Cre recombinase is driven by the MMTV promoter. A model displaying SGCs with remarkable morphological similarity to human acinic cell carcinoma was generated by conditional inactivation of the Apc and Pten tumor suppressor genes using MMTV-Cre, thereby constitutively activating the Wnt and PI3K/AKT/mTOR signaling pathways. Treatment with Rapamycin leads to the regression of the tumors in these mice, suggesting a role for mTOR in salivary gland acinar cell carcinoma development [116].
THYROID CARCINOMA TC is the second most common HNC after HNSCC, and there is currently a broad array of treatment options that depend on TC type and individual clinicopathological characteristics [117]. A multitude of molecular and genetic analyses have revealed that thyroid tumorigenesis and progression involve multiple genetic alterations including point mutations of the BRAF or RAS genes and fusions of RET, NTRK, ALK PAX8, and PPARγ, ultimately leading to the activation of the MAPK and PI3K/AKT signaling pathways [118][119][120]. This understanding of TC genomics has supported the development of several TC GEMMs intended to model these genetic mutations and mimic the human condition, making these strains suitable for the preclinical evaluation of new therapeutic modalities for TC (Table 3). However, it is crucial to select an appropriate time-point during the disease process, at which the model should be tested for therapies against a specific histopathological subtype since de-differentiation has been observed in these models.
The TRβ PV/PV transgenic mouse model was developed via a knock-in of a dominant negative PV mutation into the thyroid hormone receptor TRβ gene locus [121]. TRβ PV/PV thyroid tissues develop sequential pathological progression from extensive papillary hyperplasia to anaplasia and, ultimately, metastasis to distant organs. The disruption of the thyroid-pituitary axis in these mice results in extremely high levels of circulating thyroid stimulating hormone (TSH) which can play a pivotal role in stimulating the proliferation of thyroid cells, thereby participating in the initiation of tumorigenesis [122]. However, TC tumors in these animals are not driven by clinically relevant oncogene or tumor suppressor mutations and do not recapitulate a specific histologic subtype of human TC [121].
Genomic studies in follicular cell-derived TC revealed mutually exclusive driver mutations in BRAF or RAS. BRAF activating mutations (e.g., BRAF V600E ) are the most common mutation in differentiated TC, seen in over 60% of sporadic PTCs [119,123]. Several mutant BRAF-driven GEMMs of TC have been generated. Transgenic mice were generated that targeted expression of BRAF V600E to thyroid cells by using the bovine thyroglobulin (Tg) promoter. Tg-BRAF V600E mice develop PTC that progresses to PDTC [17]. Chakravarty et al., generated Tg-rtTA/tetO-BRAF V600E mice that express BRAF V600E in thyroid follicular cells in a doxycycline-inducible manner which gives rise to PTCs with short latency [124]. Moreover, conditional transgenic mice were Relatively low penetrance of tumor formation and a long latency period. [112] [127]. However, no distant metastasis are observed, suggesting that mutant BRAF alone lacks sufficient oncogenic potential and additional driver mutations are required for tumor progression. Combination of BRAF V600E or RAS-driven cancers with TERT promoter mutation or Trp53 mutation resulted in clinically more aggressive TC [128][129][130]. Moreover, next generation sequencing showed higher prevalence of these mutations in 60% of the analyzed PDTC patient samples as compared to only 9% of PTCs from TCGA and was also comparable to the 73% prevalence in ATC tumor samples [130]. Most of the undifferentiated TC arising from well differentiated precursors do so by acquiring additional mutations and gene alterations. GEMMs with multiple genetic abnormalities develop more aggressive types of TC. TPOCreER; Braf CA/+ ; Trp53 Δex2-10/Δex2-10 mice harbor combined BRAF mutation and loss of p53 in thyrocytes by virtue of the thyroid peroxidase (TPO) promoter, and ultimately progress to ATC [131]. Similarly, Thyro::CreER; Braf CA/+ ; Pik3ca H1047R/+ conditional transgenic mice harboring the H1047R activating hotspot mutation in PIK3CA combined with BRAF mutation also develop tumors that progress to ATC [132].
Upstream of BRAF, mutations in RAS genes have been associated with 30-45% of FTCs, 30-45% of follicular variant PTC, 20-40% of PDTC, and 10-20% of ATCs [133]. The GEMM TPO-Cre; LSL-Kras G12D ; Pten fl/fl conditionally expressing mutant Kras G12D and deleting the Pten gene were generated and rapid development of FTCs is observed [134]. This model was modified to generate mice with combined deletion of Pten and Trp53, TPO-Cre; Pten fl/fl ; Trp53 fl/fl , resulting in development of ATC [135]. Furthermore, although BRAF and RAS driver mutations are mutually exclusive in TC development, acquired resistance to BRAF inhibitors may be facilitated by KRAS or NRAS activating mutations both in the clinical and in vitro settings [136,137]. TCs in this setting progress to PDTC/ATC. Utilizing GEMMs, such as the TPOCreER; Braf CA/+ ; Trp53 Δex2-10/Δex2-10 mice and analyzing them in the presence of long-term targeted therapy may replicate the RAS-driven formation of resistant disease observed in patients.
Chromosomal translocations leading to the formation of fusion oncogenes play a role in the pathogenesis of many follicular cellderived TCs and have been reported in over 15% of both PTC and FTC [119]. PTCs, particularly radiation-associated PTC, harbor the rearranged during transfection (RET) proto-oncogene and the resultant oncogenic fusion proteins are termed RET/PTCs [138]. Based on this, Jhiang et al. developed the first GEMM for TC in which the highly active bovine thyroglobulin (Tg) promoter was employed to drive transgene expression of RET/PTC1. The thyroid tumors developed in theses mice exhibited nuclear features and local invasion similar to the nuclear grooves, vesicular nuclei, and pseudo inclusions characteristic of human PTC [139]. Similarly, Powell et al. generated mice by expressing another member of the RET proto-oncogene family, RET/PTC3, exclusively in mouse thyroid [140]. Crossing Tg-RET/PTC3 mice with Trp53 −/− mice led to an increase in tumor burden and aggressiveness [141]. However, even after introduction of the additional Trp53 mutation, there is no evidence of local or distant metastasis. Furthermore, elevated levels of TSH are reported which is a major drawback of these models because chronic TSH stimulation is not involved in TC development in humans. The expression of the PAX8/PPARγ gene fusion is seen in~35% of FTCs [142,143]. To study the role of PAX8/PPARγ fusion protein (PPFP), a transgenic mouse model was generated that combined CAG promoter-driven Cre-dependent expression of PPFP with conditional homozygous deletion of Pten in the thyroid. Tg-Cre; CAG-LSL-PPFP; Pten fl/fl mice develop thyroid carcinomas with distant metastasis [144]. The STRN-ALK fusions are commonly detected in PTC, PDTC and ATC, therefore transgenic Tg-STRN-ALK mice based on this chromosomal rearrangement were established. These mice exhibit thyroglobulin promoter driven thyroid-specific expression of STRN-ALK that leads to the development of PDTCs [145].
RET mutations are crucial for MTC pathogenesis in humans as over 75% of MTC are driven by RET proto-oncogene mutation, 25% of which are inherited through the germline [146]. Several distinct RET-activating point mutations within the endogenously expressed RET gene induce C-cell transformation, producing MTC [147][148][149]. Spontaneous activating mutations of RET were also identified in 50% of MTC bearing mice [150]. It has been attempted to establish transgenic mouse models for MEN 2-associated MTC by introducing a specific point mutation in the RET proto-oncogene [33,34]. Thus, the development of preclinical models representing sporadic MTCs is highly desired. Furthermore, it has been demonstrated that the tumor penetrance can be modulated by the genetic background of the RET transgenic mice [151]. RET-induced tumors also develop in transgenic mice with loss of p18 (and p27) serving as an additional oncogenic hit required for MTC tumorigenesis [152]. Harvey et al. were the pioneers in generating GEMMs of endocrine tumor types with heterozygous deletion of Rb and Trp53 (Rb +/− ; p53 +/− ). However, in addition to MTC, these mice develop pancreatic islet cell carcinomas, pituitary adenomas, lymphomas and sarcomas [153]. The neuroendocrine specific calcitonin generelated peptide (CGRP) promoter was used to express v-Ha-ras oncogene in the thyroid C-cells in mice of the C57BL/6 x SJL strain. Tumors develop with high penetrance and secrete calcitonin [154]. Another model with NRAS mutation in Rb knock-out mice was established which develop metastatic MTCs. However, RAS mutations are rare in clinical cases and hence this model is more representative of the human MTC phenotype but not the genotype [155].

CONCLUSIONS
A plethora of GEMMs recapitulating both squamous and glandular HNCs are now available for investigating the molecular pathways involved in tumor development and progression. Numerous efforts have improved and enriched the repertoire of HNC GEMMs to exhibit the various histologic subtype features of this diverse group of cancers, thereby enabling the selection of a suitable model for evaluating personalized cancer therapies. However, genetic alterations are engineered and present in all cells of a certain type, and GEMMs therefore may not completely represent the tumor heterogeneity and complexity found in human HNCs. It is also difficult to recapitulate metastatic progression, since animals tend to become moribund due to large burden of the primary tumor and must be euthanized often at early stages of disease. Furthermore, generating GEMMs remains highly time and labor intensive, even with the advent of CRISPR-based methodologies, and tumor development is extremely variable due to differences in penetrance and latency between models. Despite these limitations, GEMMs serve as a robust tool for understanding basic tumor biology and for preclinical testing of novel therapeutic modalities. Thus, continued development and refinement of GEMMs that reproducibly display key features of the human malignancies are required to enable investigators to evaluate the local and systemic responses to next generation treatment.