Patient-Derived Three-Dimensional Culture Techniques to Model Tumor Heterogeneity in Head and Neck Cancer

Abstract

Head and neck squamous cell carcinoma (HNSCC) outcomes remain stagnant, in part due to a poor understanding of HNSCC biology. The importance of tumor heterogeneity as an independent predictor of outcomes and treatment failure in HNSCC has recently come to light. With this understanding, 3D culture systems, including patient derived organoids (PDO) and organotypic culture (OTC), that capture this heterogeneity may allow for modeling and manipulation of critical subpopulations, such as p-EMT, as well as interactions between cancer cells and immune and stromal cells in the microenvironment. Here, we review work that has been done using PDO and OTC models of HNSCC, which demonstrates that these 3D culture models capture in vivo tumor heterogeneity and can be used to model tumor biology and treatment response in a way that faithfully recapitulates in vivo characteristics. As such, in vitro 3D culture models represent an important bridge between 2D monolayer culture and in vivo models such as patient derived xenografts.

Patients with head and neck squamous cell carcinoma suffer from poor patient stratification tools

In 2020, the head and neck was the eighth most common site of new cancer cases (4.6%, 880,000) and cancer-related deaths (4.5%, 440,000) globally.¹ Ninety percent of head and neck cancers are squamous cell carcinomas (HNSCC),^2,3 which arise in the upper aerodigestive tract. The most common risk factors for HNSCC are tobacco and alcohol use, which pose dose-dependent and synergistic risks.^4,5 Betel or areca nut chewing is an important risk factor in South and Southeast Asia.⁶ Epstein-barr virus (EBV) and human papilloma virus (HPV) are infectious risk factors.^7,8

Although five-year survival of HNSCC has improved slightly over the past few decades, from 55% to 66%,⁹ this improvement is largely due to the rising incidence of HPV-related HNSCC. Patients with HPV-positive oropharyngeal SCC (OPSCC) experience significantly better survival than HPV-negative OPSCC; HPV-positive disease is increasingly being viewed as a distinct entity, now with its own staging system^10,11 Accordingly, while survival in the HPV-positive OPSCC population has increased, by contrast, survival in HPV-negative OPSCC has stagnated in spite of advances in staging, surgery, radiation, and systemic therapy.¹⁰

One contributing factor to the challenges in HPV-negative HNSCC management reflects the difficulties in accurately stratifying these patients for treatment. The mainstays of treatment in HNSCC remain surgery and radiation therapy, with or without chemotherapy,¹² with immunotherapy typically reserved for recurrent and metastatic disease, but under investigation for other indications as well.¹² Treatment planning relies heavily on clinical and pathologic features, such as depth of invasion, perineural invasion (PNI), or lymphovascular invasion (LVI). These are relatively imprecise tools that inform treatment decisions that may confer significant morbidity, such as whether to perform an elective neck dissection on a node-negative neck or whether to administer adjuvant radiation therapy.¹³

Other than HPV, which can help to limit overtreatment with tri-modality therapy, few biomarkers are in regular clinical use. Epithelial growth factor receptor (EGFR) is overexpressed in 80–90% of HNSCC, and although higher levels of EGFR expression are correlated with worse outcomes,¹⁴ EGFR-directed therapies such as cetuximab confer only a modest benefit that is not enhanced in subgroups defined by alterations in EGFR expression. Programmed death ligand 1 (PD-L1) is expressed in 50–60% of HNSCC,¹⁵ and while the likelihood of benefit from treatment with the anti-PD-1 antibody, pembrolizumab, is increased in patients with higher levels of PD-L1 in the tumor microenvironment, its predictive capacity is limited and the vast majority of patients do not durably benefit.¹²

Tumor heterogeneity predicts outcomes and treatment response

New therapeutic targets have been elusive, despite comprehensive genetic and expression profiling of HNSCC over the past decade. Microsatellite analysis first established a sequential “genetic progression” of driver genetic changes that transform dysplasia into invasive HNSCC, namely inactivation of tumor suppressor genes TP53, CDKN2A, and PTEN, and amplification of CCND1.¹⁶ Subsequent bulk sequencing studies have identified TP53, FAT1, CDKN2A, PI3KA, and NOTCH to be commonly mutated in HNSCC.^17,18 Genetic alteration profiles may differ by HPV status. While TP53 and CDKN2A alterations predominate in HPV-negative tumors, TP63, TRAF3, and E2F1 alterations are common in HPV-positive tumors.^17–19

Overall, genetic alterations in HNSCC are diverse, and the driver genetic alterations are loss of function mutations in tumor suppressor genes, which are difficult to target therapeutically. Additionally, mechanisms beyond driver mutations are increasingly appreciated as important aspects of tumorigenesis and tumor progression, including intratumoral heterogeneity (ITH), interaction with tumor microenvironment (TME), and non-mutational epigenetics.²⁰

Studied across multiple cancer types, ITH has been proposed to contribute to poor outcomes, as specific cellular subpopulations within a tumor may possess unfavorable characteristics, such as propensity for metastasis or treatment resistance.^21–25 Accordingly, Mroz et al. developed a universal metric called the mutant allele tumor heterogeneity (MATH) score to quantify genetic heterogeneity in a tumor based on whole exome sequencing data.²⁶ They demonstrated that higher MATH scores, and thus greater ITH, are indeed independently predictive of worse overall survival in HNSCC.^27,28

Puram et al. subsequently used single cell RNA-sequencing (scRNA-seq) to define this cellular heterogeneity in HNSCC tumors. They uncovered a novel partial epithelial-to-mesenchymal transition (p-EMT) program that was coherently expressed across tumors. EMT is a cellular transition that involves downregulation of anchoring proteins, a key feature of epithelial cells, and a resultant increase in cellular mobility. This process is normally active during embryogenesis and is proposed to be coopted by cancer cells to facilitate invasion and metastasis.^29–33 The p-EMT program defined by Puram et al. was characterized by expression of a unique set of mesenchymal markers, without loss of epithelial markers or expression of canonical EMT transcription factors and was specifically expressed by invasive cells at the tumor-stromal interface.^34–38 With this localization, Puram et al. further hypothesized that this program may be driven by signaling between cancer-associated fibroblasts (CAF) and cancer cells, representing a non-mutational epigenetic source of ITH. ³⁴

Two-dimensional cell line models are limited in their ability to capture heterogeneity

Given the localization of p-EMT and its association with invasiveness, ^34,35 modeling and manipulation of the p-EMT subpopulation is a critical step in understanding the biology of HNSCC. However, despite the availability of more than 300 well characterized cell lines reflecting HPV-positive and HPV-negative HNSCC,^39,40 traditional two-dimensional (2D) cancer cell lines grown in monolayer culture have been limited in their ability to model p-EMT⁴¹ and capture general cellular heterogeneity.⁴² Underscoring this limitation, Puram et al. profiled five commonly used HNSCC cell lines by scRNA-seq and found only one with a subpopulation that partially recapitulated the in vivo p-EMT program. As the in vivo program was present in the majority of tumors, this discrepancy confirmed that cell lines are limited in their ability to capture such heterogeneity.⁴¹

More broadly, a subsequent single cell transcriptomic analysis of cellular heterogeneity in immortalized cell lines across cancer types revealed that only a small minority of profiled cell lines demonstrated discrete expression subpopulations. Further, most of these had only one detected subpopulation,⁴² rather than the multiple populations seen in most tumors in vivo.⁴¹ Although an EMT program was a common aspect of variability across cell lines,⁴² the inconsistency of cell lines to capture this program underscores the need for in vitro models that better capture and model in vivo heterogeneity in HNSCC. Underlying this inability to capture heterogeneity is likely the loss of key features of in vivo tumor biology, including a necessary loss of tissue architecture and intercellular interactions and alterations in gene expression following plating in monolayer, relative to the tumors from which they are derived.³⁹

Early three-dimensional techniques

While the limitations of 2D cell lines have recently been highlighted by investigations of cellular heterogeneity, it has long been evident that monolayer culture may be inadequate for modeling in vivo behavior. Cell lines require significant time for development and maturation from fresh tumor tissue and have relatively low rates of successful immortalization.³⁹ In fact, success rates of deriving HNSCC cell lines range from just 11 to 33%.⁴³ As a result, three-dimensional (3D) culture techniques have been developed in an attempt to preserve multicellular structures and tissue architecture. Histocultures and multicellular spheroids were among the earliest and most widely used of these techniques.⁴⁴

Histocultures are generated from slices or fragments of tumor that have only been mechanically divided, without enzymatic dissociation, to maintain cancer cells within their original microenvironment.⁴⁴ Tissue pieces are maintained at the air liquid interface using culture inserts.⁴⁴ A number of studies have demonstrated the ability of histoculture treatment response assays to recapitulate the in vivo response of HNSCC tumors to a variety of standard of care therapies.^45–49 While the biggest advantage of histocultures is maintenance of the original tumor microenvironment, enabling modeling of therapeutic strategies that target the microenvironment, these cultures suffer from rapid deterioration and dissolution of the tissue architecture, with viability decreasing significantly after seven days, declining fractions of non-cancer cells over that period, inability to passage or perform genetic modulation, and low throughput.^44,49

Multicellular spheroids utilize a different approach, with spontaneous formation of cell aggregates from single cell suspensions using a variety of methods such as ultra-low attachment plates and hanging-drop cultures.^50,51 Several studies have utilized HNSCC spheroid cultures to model biological process, including EMT,^52,53 and differential treatment response,^51,54,55 including in co-culture with chimeric antigen receptor (CAR)-T cells.⁵⁶ Spheroid cultures may be generated with reagents and techniques similar to monolayer culture, except for ultra-low attachment or hanging drop plates,⁵⁴ increasing accessibility to those familiar with standard culture techniques. However, multicellular spheroids demonstrate limited fidelity to the tumors from which they are derived⁵⁷ and, like histocultures, they can be maintained only for a short period of time, with an average of 10–15 days.⁴⁴

Three-dimensional patient derived organoids offer a novel platform to study heterogeneity

Adult stem cell-derived 3D organoids address many of the limitations of early 3D culture methods. These cultures are mini-organs grown within a basement membrane gel that induces epithelial differentiation. HNSCC patient derived organoids (PDO) can be generated from a variety of tissue sampling methods, including excess tissue from surgical resection, surgical or core needle biopsies, or even fresh frozen tissues, though success rates are lower with the latter as a source.⁵⁸ Tissue biopsies are dissected to remove non-epithelial tissue (e.g., sub-epithelial fat, muscle, necrotic debris), minced into small fragments, and then physically and enzymatically dissociated to a single cell suspensions,⁵⁸ and 1,000–20,000 cells are typically seeded per well of a 24-well plate. Cells are suspended within a basement membrane gel and treated with a cocktail of growth factors to promote a regenerative response of the epithelial stem cells (Figure 1).⁵⁸ PDO are then passaged every 10–14 days and can be maintained in culture for > 10 passages.^58,59

Figure 1. Generation of PDO.

(A) Schematic showing workflow for generation of PDO. Freshly tumor biopsies are dissociated to single cells, suspended in Matrigel, and plated. After 7–11 days, single cell-derived organoids are formed and may be utilized for further experiments or characterization. (B) 20X hematoxylin and eosin (H&E) image of a representative HNSCC PDO showing a multicellular structure with layers of squamous epithelial cells. Scale bar represents 100 μm.

Multiple HNSCC PDO protocols are described.^58–60 Published protocols are largely similar but vary on a few key parameters. The first is the basement membrane gel used (e.g. basement membrane extract (BME)⁵⁸ or Matrigel⁵⁹) and the percent of gel in which organoids are suspended, particularly for drug treatment assays.^58,59 The second is the medium used for culture. While some have described the use of growth factors such as CHIR99021 (GSK3β inhibitor/Wnt activator), FGF2, FGF10, prostaglandin E, and forskolin,⁵⁸ our group has had comparable success rates of organoid generation without these factors,⁵⁹ suggesting they may not be critical for HNSCC PDO formation. Inhibitors of transforming growth factor (TGF)-β (e.g., A83-01) have also been used to prevent epithelial differentiation,⁵⁸ but in our experience these can also be eliminated without loss of efficiency.⁵⁹ With these protocols, reported success rates of HNSCC PDO generation are 60–70%.^43,61 Tanaka et al. used an alternate cancer tissue originated spheroid (CTOS) method⁶² with embryonic stem cell medium, though their rates of success were lower at 30%.⁶⁰

HNSCC PDO capture heterogeneity.

Three groups, including ours, have reported on head and neck PDO.^43,60,61,63 HNSCC PDO are robustly distinguishable from organoids derived from normal oral mucosa, with a coherent set of differentially expressed genes across PDO.⁴³ Moreover, PDO models capture the inter- and intra-patient heterogeneity of the normal oral epithelium or primary tumors from which they are derived. At the genetic level, Driehuis et al. demonstrated that HNSCC PDO recapitulate single nucleotide variants and small insertions or deletions seen in the tumors from which they were derived. Broadly, they also found that the mutations commonly seen in HNSCC, including p53, were recapitulated in their cohort of PDO.⁴³ In addition, they found that treatment with Nutlin-3, which selects for p53 mutant cells, identified p53 mutant PDO at a rate similar to the rate observed in vivo p53 mutations, confirming that PDO recapitulate in vivo genetic changes.⁴³ At the expression level, Driehuis et al. found that normal oral mucosal organoids show patterns of p63, Ki-67, and KRT13 expression consistent with the epithelial differentiation of the oral mucosa,⁴³ while HNSCC PDO show inter-patient variability in epidermal growth factor receptor (EGFR) expression that corresponded to the primary tumor tissue and to response to EGFR-targeted photodynamic therapy.⁶³ Others demonstrated that PDO recapitulate the immunohistochemical staining patterns and rates of marker positivity of the primary tumors from which they were derived, including staining for epithelial marker pancytokeratin, mesenchymal marker vimentin, and stem cell markers CD44 and ALDH1A1.^60,61 To our knowledge, investigations into transcriptomic heterogeneity at the single cell level have been limited to mouse oral cancer organoids,⁶⁴ with no studies in PDO, but the presence of this heterogeneity in murine organoids supports the notion that PDO may capture it, as well.

HNSCC PDO predict response to therapy.

In accordance with capturing cellular genetic and expression heterogeneity, HNSCC PDO predict in vivo response to a variety of standard and targeted therapies, including chemotherapy, radiation therapy, and photodynamic therapy, emphasizing the notion that PDO may be viewed as ‘avatars’ to help predict therapeutic response. Tanaka et al. tested HPV- and HPV+ HNSCC organoids with cisplatin and docetaxel and found organoid responses to these drugs were consistent with responses in mouse xenograft tumors derived from the same organoid lines.⁶⁰ They further demonstrated that sensitivities of organoid lines differed from those of 2D cell lines derived from the same tumors,⁶⁰ supporting the notion that organoids may be better models of in vivo tumor biology. In larger cohorts, HNSCC PDO demonstrated differential responses to 5-FU,⁶¹ cisplatin, carboplatin, cetuximab, ionizing radiation,⁴³ and EGFR-targeted photodynamic therapy,⁶³ with PDO response to ionizing radiation corresponding closely with clinical patient response to adjuvant radiotherapy, and chemotherapy acting as a radiosensitizer,⁴³ as it is thought to do clinically. Additionally, PDOs demonstrated differential responses to non-standard biologic therapies, including PIK3CA inhibitor alpelisib, BRAF inhibitor vemurafenib, PARP inhibitor niraparib, mTOR inhibitor everolimus, and fibroblast growth factor receptor (FGFR) inhibitor AZD4547, with response to alpelisib and vemurafenib assessed in the context of PIK3CA and BRAF mutations, respectively.⁴³ Taken together, these findings suggest that PDO may be used as an important correlative biomarker platform within clinical trials investigating novel therapies or therapeutic strategies in HNSCC.⁶⁵

Limitations of PDO.

Despite the transformative potential of PDO as predictors of in vivo tumor behavior and treatment response, a major limitation of this model system is its restriction to the epithelial component of tumors. Although early passages of PDO lines may retain some immune or stromal cells, PDO quickly lose these components.⁶⁶ The lack of stromal and immune elements, in the absence of reconstitution, creates an inability to model interactions with the tumor microenvironment (TME).⁶⁶ As the critical role of the TME is better delineated, the need for adjunctive techniques to model these interactions, including 3D organotypic culture and humanized xenograft models, becomes more critically apparent. Finally, techniques and reagents involved in culturing PDO, particularly the use of basement membrane gel, may increase cost and involve a steeper learning curve than other 3D techniques.

Three-dimensional organotypic culture models the tumor-stromal interface

3D organotypic culture, also known as raft culture, is a form of tissue engineering that aims to recapitulate the tumor-stromal interface. Originally established in the study of skin biology,^67,68 organotypic culture (OTC) has been applied to the upper aerodigestive tract over the past three decades.^69,70 Initially, an acellular type I collagen matrix is placed on the bottom of an insert within a plate, followed by a layer of fibroblasts embedded within type I collagen and Matrigel. Over five days, there is fibroblast-mediated constriction of the collagen matrix. Epithelial cells are then seeded on top of this matrix, and on day 9, the volume of media is reduced to create an air-liquid interface, which triggers epithelial stratification and differentiation. Finally on day 13, the resulting culture is harvested for analysis (Figure 2).⁷¹ Specific cell populations may be dissected by laser capture microdissection for selective analysis, or the entire epithelium may be peeled away from the underlying matrix.⁷¹ Variations on this technique have been reported, including the use of small, undissociated tumor fragments to seed the epithelial component of OTC systems, rather than a pure population of epithelial cells, thus enabling the possibility of including patient-derived immune cells in the OTC system.⁷²

Figure 2. Organotypic 3D culture.

(A) Workflow for 3D OTC. Acellular type I collagen is placed on the bottom of an insert within a plate, followed by a layer of fibroblasts embedded within type I collagen and Matrigel. By day 5, matrix contraction is observed, and epithelial cells are then seeded on top of this matrix. By days 9–13, epithelial cells proliferate and stratify, and the resulting culture is brought to the air-liquid interface. At approximately two weeks, the culture is harvested for analysis. (B) 20X H&E image of 3D OTC, showing invasive squamous cell carcinoma tumor fronts modeled at the interface of the epithelium and the underlying stroma. Scale bar represents 100 μm.

Early HNSCC OTC models.

A handful of studies have utilized OTC models of HNSCC, with the majority using immortalized HNSCC 2D cell lines and normal human dermal fibroblasts. Early studies provided a proof of concept that OTC recapitulate the histology of in vivo HNSCC and expression changes relative to normal epithelium.^73,74 OTC has since been used for mechanistic studies and assessment of treatment response. Eicher et al. modeled the efficacy of topically applied adenoviral gene therapy;⁶⁹ Lee et al. modeled the effect of zoledronic acid treatment;⁷⁵ and Nisa et al. investigated the MET receptor tyrosine kinase as a potential therapeutic target.⁷⁶ Srivastava et al. modeled HPV-related HNSCC using E6/E7-expressing human foreskin keratinocytes and Rb-depleted human foreskin fibroblasts in OTC to model the relationship between invasiveness and EMT.⁷⁷ They modeled the impact of p63 expression and SNAI2 knockdown on EMT and invasiveness,⁷⁷ highlighting the power of this technique to study processes involved in tumor invasion of the underlying stroma. Finally, Gemenetzidis et al. used OTC to model early oncogenesis, showing that induction of FOXM1 in primary human keratinocytes induced epithelial hyperplasia.⁷⁰

OTC model patterns of invasion.

Engelmann et al. described the largest cohort of patient-derived HNSCC OTC models, deriving models from 13 patients, both HPV-positive and HPV-negative, across multiple subsites.⁷² They demonstrated that OTC displayed similar histological and morphological characteristics to primary tumors at 14 days and that models displayed variable patterns of invasiveness and proliferation, with the one patient with clinically aggressive disease having a highly invasive pattern.⁷² Their models also captured some CD45+ immune cells until day 21, though the fraction of these cells declined over time, as well as α-SMA+ CAF-like cells derived from normal dermal fibroblasts.⁷² Finally, they found that fractionated irradiation of five OTC in the cohort had variable impact on proliferative index and markers of apoptosis; however, the ability to correlate clinically was limited by the small sample size.⁷²

Limitations of OTC.

The major limitations of OTC remain the availability of stromal fibroblasts that mimic in vivo CAFs and the need to culture cells in 2D prior to use in OTC. When considering tumor heterogeneity and p-EMT driven by interactions with CAFs, the former limitation is especially important given our group’s previous data showing that in vitro CAFs lost expression of key ligands that may mediate interactions with cancer cells, including TGF-β3.⁴¹ The latter limitation is circumvented by the modified protocol described by Engelmann et al., but this approach limits the ability to perform mechanistic analyses that require manipulation of the epithelial compartment prior to plating.⁷²

Future directions

Although it has been established that PDO and OTC models of HNSCC capture tumor heterogeneity, the use of these models to study cancer cell subpopulations has been limited. In organoid models of esophageal SCC, our group has previously studied EMT, developing mechanistic insights into its regulation⁷⁸ and phenotypic insights into its impact on invasiveness.⁷⁹ In this vein, performing OTC with CAFs and epithelial cancer cells derived from the same patient is a critical next step in the developing a model that more faithfully recapitulates the in vivo tumor microenvironment and the intercellular interactions that mediate programs like p-EMT.

The contribution of epigenetic mechanisms to HNSCC heterogeneity remains largely unclear. Application of recently developed single-cell epigenome technologies that profile chromatin accessibility,⁸⁰ histone/DNA modifications,^81,82 and 3D genome interactions⁸³ in PDO and OTC models of HNSCC could facilitate the identification of epigenetically distinct cancer cell subpopulations. Furthermore, genetic and/or pharmacologic perturbation of chromatin regulators will provide key insights into the causal relationships between epigenetic, transcriptomic and phenotypic heterogeneity.

In addition, deriving 3D culture models that include an immune component is important in light of the addition of immunotherapy as a novel therapeutic modality over the past decade. Co-cultures of PDO with immune cells have been reported but these models are not well established.⁸⁴ Neal et al. generated PDO from non-enzymatically processed fragments of tumor and reported that the resulting organoids contained CD3+ tumor infiltrating lymphocytes, as well as natural killer (NK) cells and macrophages, though these cells decreased over the course of one month in culture.⁸⁵ They utilized these PDO to model immune checkpoint blockade, providing a potentially promising avenue to better understand which patients may respond to immunotherapy.⁸⁵ In HNSCC, where durable response to immunotherapy is approximately 15% and side effects are non-trivial,⁸⁶ this stratification may be critical.

Beyond simply capturing tumor heterogeneity, HNSCC PDO and OTC models may be utilized for more mechanistic analyses via genetic modification. Driehuis et al. demonstrated that these models are permissive to HPV16 infection,⁴³ thus facilitating in vitro generation of PDO that recapitulated HPV-related HNSCC. CRISPR-based genome editing has also been applied to PDO outside of the head and neck,^87,88 suggesting mechanistic analyses previously performed in cell lines may now be translated to PDO and OTC models. This strategy may prove critical in HNSCC, with the current lack of targeted therapies in widespread use.

Finally, the use of PDO as ‘avatars’ to predict treatment response is in its infancy. In head and neck cancer, questions regarding response to therapy often arise in the setting of choosing between surgical and non-surgical therapies for early stage laryngeal HNSCC patients and oropharyngeal HNSCC who are HPV-positive smokers or HPV-negative, and the decision to administer immunotherapy. As markers to guide these decisions are severely lacking, the ability to test in vitro responses to therapy may provide a novel strategy to help stratify patients. Furthermore, PDOs may provide a platform to facilitate investigation into the utility of novel therapeutic strategies, including CAR-T cell therapies⁸⁹ and oncolytic virotherapy,^90,91 with previous studies in other tumor types suggesting the utility of the PDO system in such studies.

Conclusions

PDO and OTC represent novel approaches study HNSCC biology and treatment response. These 3D culture systems may better capture HNSCC tumor heterogeneity, including critical subpopulations such as p-EMT, and interactions between cancer cells and immune and stromal cells in the microenvironment than 2D cell lines grown in monolayer culture. With an increasing understanding of the importance of heterogeneity in outcomes and treatment failure, these 3D culture models are an important next step in understanding HNSCC biology in a way that may facilitate the development of novel therapeutic targets. Moreover, the use of PDO as ‘avatars’ to predict treatment response in a personalized way is a promising strategy to improve patient stratification, an issue that underlies the continued poor prognosis of HPV-negative HNSCC.