Integrative genomic and functional analysis of human oral squamous cell carcinoma cell lines reveals synergistic effects of FAT1 and CASP8 inactivation

Oral squamous cell carcinoma (OSCC) is genetically highly heterogeneous, which contributes to the challenges of treatment. To create an in vitro model that accurately reflects this heterogeneity, we generated a panel of HPV-negative OSCC cell lines. By whole exome sequencing of the lines and matched patient blood samples, we demonstrate that the mutational spectrum of the lines is representative of primary OSCC in The Cancer Genome Atlas. We show that loss of function mutations in FAT1 (an atypical cadherin) and CASP8 (Caspase 8) frequently occur in the same tumour. OSCC cells with inactivating FAT1 mutations exhibited reduced intercellular adhesion. Knockdown of FAT1 and CASP8 individually or in combination in OSCC cells led to increased cell migration and clonal growth, resistance to Staurosporine-induced apoptosis and, in some cases, increased terminal differentiation. The OSCC lines thus represent a valuable resource for elucidating the impact of different mutations on tumour behaviour.


Introduction
The global annual incidence of head and neck squamous cell carcinoma (HNSCC) is approximately 600,000, with over 300,000 deaths attributable to the disease each year [1]. HNSCC lesions frequently recur and metastasize to other locations, contributing to significant morbidity and mortality [2]. Despite improvements in surgical care and adjuvant therapy, the 5-year survival rate is only 60% in the United States [3].
Previous studies have used next-generation whole exome sequencing of primary tumours to reveal the mutational landscapes of human HNSCCs and suggest novel oncogenic drivers to target for potential therapies [4e6]. HNSCC is strikingly heterogeneous at both the genetic and the cellular level. No single genetic lesion predominates and patients often develop more than one primary tumour.
Oral cancer is a subtype of HNSCC, comprising tumours of the lip and oral cavity. Worldwide, over 250,000 new cases and 125,000 deaths from oral cancer occurred in 2008 [1]. Approximately 90% of oral cancers are squamous cell carcinomas (SCCs), arising in the multilayered epithelia that line the mouth and lips. Oral SCC (OSCC) frequently involves the tongue, and can also occur in the floor of the mouth, gingiva, cheek and palate. The two major risk factors for OSCC are smoking and alcohol consumption [7]. In contrast to HNSCC of the tonsils and oropharynx, OSCC tends not to be associated with HPV infection [8,9]. Despite advances in treatment, the survival rate for OSCC remains stubbornly low [3].
As in other solid tumours [10,11], the properties of the malignant epithelial cells in HNSCC elicit an immune infiltrate and activation of fibroblasts in the surrounding connective tissue stroma [1,3,12]. Integrative analysis of the genetic and cellular heterogeneity of HNSCC is beginning to reveal new biologically distinct subtypes of the disease [13], and greater understanding of their significance could inform future therapies. One way to achieve this is by studying the properties of cells isolated from HNSCC.
Cells can readily be cultured from human oral tumours and normal oral mucosa, and several well-annotated collections of HNSCC cells are available [14e16]. However, genetic changes in individual cell lines have not been compared with control DNA from the original donors. Thus it has been difficult to discover whether the lines truly reflect the genetic heterogeneity of OSCC or are dominated by mutations acquired during long-term culture. The value of such lines lies in characterising the biology of genes that are frequently mutated in OSCC, including NOTCH1, CASP8 (Caspase 8) and FAT1 (an atypical cadherin) [6], which are not conventional oncogenes/tumour suppressor genes and can potentially have pleiotropic effects on tumour properties. For example, the association between FAT1 mutation and overall survival in HNSCC differs according to the HPV status of the tumour [17], and loss of Caspase 8 not only has cell intrinsic effects [18] but can also trigger inflammation [19]. Furthermore, there is evidence for biological interactions between FAT1 and Caspase 8, with FAT1 acting as an antagonist of Caspase 8 in a synthetic lethal screen in cancer cell lines [20].
In this study, we set out to develop new OSCC lines, discover which mutations are tumour-acquired and determine whether they are indeed representative of mutational burden in primary tumours. We then used the lines to explore the impact of mutations in CASP8 and FAT1 on cell behaviour.

Derivation of OSCC lines
Anonymized biopsies of OSCC or normal oral mucosa were collected with appropriate ethical approval (UK National Research Ethics Service (08/H0306/30). Cells were isolated and cultured on a feeder layer of J2 3T3 cells in complete FAD medium as described previously [16].

Whole exome sequencing
Genomic DNA was extracted from OSCC lines (passage 2e4) and patientmatched blood. Whole exome sequencing was performed by Beijing Genomics Institute (Hong Kong). Raw image files were processed by Illumina base calling Software 1.7 or base calling with default parameters, and the sequences of each individual were generated as 90 bp paired-end reads. High-quality reads were aligned against the NCBI human reference genome (hg19) using Burrows-Wheeler Aligner (v0.5.9) with default parameters. Picard (v1.54) was employed to mark duplicates and was followed by Genome Analysis Toolkit (v1.0.6076, GATK Indel-Realigner) to improve alignment accuracy. Putative somatic single nucleotide variations (SNVs) were predicted by VarScan2.25 with the parameters as "mincoverage 5 -min-coverage-normal 5 -min-coverage-tumour 5 -min-var-freq 0.1 -min-freq-for-hom 0.75 -min-avg-qual 0 dsomatic-p-value 0.15". In order to obtain high confidence somatic SNVs, an in-house pipeline was applied. Somatic InDels were predicted by GATK SomaticInDelDetector with default parameters. A pipeline was developed to obtain high confidence somatic InDels; normal and tumour bam were reused to perform local realignment and germline indels were filtered for high confidence indels, with normal coverage and tumour coverage no less than 5. High confidence somatic single nucleotide variants and InDels were annotated using ANNOVAR. Functional impacts of missense mutations were predicted using SIFT, PolyPhen2, PhyloP, MutationTaster and LRT annotations.

Prediction of driver genes and pathways
The Oncodrive-fm method was applied, as previously published, to identify significantly mutant genes and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways [21]. Pathway enrichment analysis was also performed to identify additional significantly mutated KEGG pathways.

KEGG pathway analysis and clustering
Whole exome sequencing data from The Cancer Gene Atlas (TCGA) HNSCC collection [6] were accessed from cBioPortal.org. KEGG pathway analysis was performed; elog 2 (P) values were calculated and presented in heat maps created using GENE-E software (Broad Institute). OncoPrint, mutation diagrams and tests for mutual exclusivity of genomic alterations were downloaded from cBioPortal.

Quantitative real-time RT-PCR
RNA was harvested using the RNeasy Mini Kit and QIAshredder columns according to the manufacturer's instructions (Qiagen). cDNA was synthesized using SuperScript™ III First-Strand Synthesis SuperMix (Invitrogen). qPCR was performed using TaqMan Fast Universal PCR Master Mix without AmpErase UNG (Life Technologies) on the CFX96 Touch Real-Time PCR Detection System (Bio-Rad) with the following primers: CASP8 (Hs01018151_m1), FAT1 (Hs00170627_m1) and GAPDH (Hs02758991_g1) as the internal control (Life Technologies). Relative quantification was performed using the 1/DCt or DDCt methods.

Transient transfection
SMARTpool ON-TARGETplus CASP8 and FAT1 (GE Healthcare) and Silencer ® Negative Control No. 1 (Thermo Fisher) siRNAs were used. Prior to transfection, wells were coated with 20 mg/mL Rat Tail Collagen I (Corning) and incubated for 1 h at 37 C. Cells were transfected using either Lipofectamine RNAiMax or Lipofectamine 2000 with Opti-MEM medium (Thermo Fisher). The viability of transfected cells was confirmed by Trypan Blue exclusion prior to performing cell behaviour assays.

Colony formation
48 h after transfection, cells were seeded at clonal density in 6-well plates on a layer of feeders. 10e14 days after seeding, feeders were detached and keratinocytes were fixed in 10% formalin and stained with a solution of 1% Rhodamine B and 1% Nile blue. Well images were acquired on the Gel Doc™ XRþ (Bio-Rad). Percent well area covered by cells was determined using the ColonyArea plugin [22] for ImageJ.

Cell migration assay
24 h after transfection, cells were seeded on collagen-coated well plates in the Oris™ Cell Migration Assay (Platypus Technologies) and cultured in complete FAD medium. 24 h after plating, cell seeding stoppers were removed. Phasecontrast images were acquired for 16e24 h using an IncuCyte ® ZOOM (Essen BioScience). Area closure was quantified in ImageJ, following the manufacturer's instructions.

Induction of apoptosis
24 h after transfection, cells were seeded on collagen-coated 96-well plates in complete FAD medium. 24 h later cells were incubated with varying concentrations of Staurosporine (1285; Tocris) in complete FAD medium for 4 h at 37 C.

Statistical analyses
Data were analyzed using an unpaired Student's t-test or one-way ANOVA with Tukey's Multiple Comparison Post Test on GraphPad Prism 5.0b, and a P value of <0.05 was considered significant, unless otherwise noted.

Whole exome sequencing of OSCC lines
We derived multiple low passage polyclonal cell lines from primary oral squamous cell carcinoma biopsies by culture on a 3T3 J2 feeder layer in order to minimise selection for rapidly dividing cells [16]. Whole exome sequencing was performed on 16 lines, together with patient-matched blood. We achieved 37and 43-fold mean sequence coverage of targeted exonic regions, with 73 and 77% of loci covered at !20-fold from tumour and matched blood samples, respectively ( Supplementary Fig. S1). Mutation rates varied from 2.50 to 44.7 mutations/megabase (mean 16.9 ± 13.5), with 80e1431 somatic mutations per sample (mean 539 ± 432) ( Fig. 1A; Supplementary Table S1). A total of 8629 single nucleotide variations across 2611 genes were found, of which 5839 (68%) were synonymous, 2621 (30%)  Table S2). Ninety-five insertions/deletions (indels) were found across 83 genes, of which 36 (43%) and 27 (32.5%) were non-frameshift and frameshift deletions, respectively. Thirteen (16%) and 19 (23%) were nonframeshift and frameshift insertions, respectively. The ratio of nucleotide transitions to transversions ranged from 1.17 to 3.00 (mean 2.24 ± 0.436) (Fig. 1B). The frequency of C:G to A:T transversions and C:G to T:A transitions varied inversely with mutation rate, while the frequency of T:A to C:G transitions increased with mutation rate.
Ten of the 16 cell lines were derived from cancers of the tongue, four from the alveolus, one from the buccal mucosa and one from the hard palate ( Fig. 1C; Supplementary Table S1). No associations were detected between self-described alcohol consumption, tobacco history or age, or between mutation rate or differentiation status of the primary tumour. However, welldifferentiated tumours tended to have higher mutation rates than moderately differentiated tumours. All tumours were negative for common genotypes of human papilloma virus (HPV), determined by the absence of viral DNA integration in the sequenced exons.
To identify putative driver genes and pathways that, when mutated, contribute to OSCC tumourigenesis, the Oncogene-fm method was applied [21]. We considered significant only those mutated genes or Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways with Q values < 0.10. Only two genes were found to be significantly mutated e the canonical tumour suppressor TP53 (p-value 1.49 Â 10 À9 ; q-value 3.72 Â 10 À7 ) and CSMD2 (CUB and sushi domain-containing protein 2; p-value 0.00024; q-value 0.03013). While CSMD2 has a tumour suppressive role in several cancers [23,24], CSMD2 mutations in HNSCC have not been described previously.
Twenty KEGG pathways were significantly mutated across the 16 OSCC lines (Supplementary Table S3). Unsurprisingly, ten were pathways for specific cancers and three for other human diseases profiled in KEGG. Three cellular community pathways (Adherens Junction, Focal Adhesion and Tight Junction) and two signalling pathways (Wnt and B Cell Receptor Signalling Pathways) were also significantly mutated. Pathway enrichment analysis was performed in parallel, and seven KEGG pathways were significantly mutated with a corrected P value < 0.10 (Supplementary  Table S4). Common pathways identified by both methods were Focal Adhesion and Wnt Signalling. The contribution of upregulated Wnt/b-catenin signalling to cell survival and invasion in HNSCC has been previously discussed [15,25,26]. Recent work suggests that transcriptional regulation of focal adhesion genes plays a crucial role in OSCC pathogenesis and that focal adhesion kinase (FAK) is a key mediator of extracellular matrix interactions and cellular signalling, contributing to HNSCC onset and progression [27,28].
In 2015, The Cancer Gene Atlas (TCGA) Network published genomic profiles for 279 HNSCCs [6]. Of the 11 genes identified in that study as significantly mutated and eight trending towards significance, nine and five, respectively, were identified as harbouring somatic mutations across 13 of our 16 cell lines (Fig. 1D). These include inactivating NOTCH1 mutations (found in 10e15% of HNSCC) [4e6], which cluster to the EGF-like binding domain and domains necessary for transactivation of target genes. Tumourassociated NOTCH1 mutations were identified in three of our OSCC lines via whole exome sequencing (Fig. 1D). In addition, Sanger sequencing of the entire genomic region of NOTCH1-3 was performed on all OSCC lines, including lines for which matched blood was not available. This identified several additional novel bona fide somatic (nonsynonymous, nonsense and indel) mutations ( Supplementary Fig. 2). 44% of the cell lines contained mutations in NOTCH1, many of the mutations clustering in the EGF-like domain. Mutations in NOTCH2 and 3 were less common. The presence of mutations in the PEST domain of NOTCH3 could potentially result in increased receptor activity due to decreased ubiquitin-mediated degradation.

OSCC lines are representative of primary tumour heterogeneity
To discover whether the mutational spectrum of OSCC is distinct from that of other HNSCCs, we determined the site of origin of each of 267 tumours in TCGA and then performed unsupervised clustering of the samples according to the significance of mutated KEGG pathways (Fig. 1E). This indicated that the spectrum of KEGG pathways mutated in OSCC was similar to that of HNSCC arising in other sites.
To determine whether the KEGG pathways that were mutated in the OSCC lines were similar to those in primary tumours, we performed unsupervised clustering of TCGA OSCC samples ranked according to the frequency of mutated KEGG pathways in the cell lines. As shown in Fig. 1F, the pathways that were mutated in the lines were indeed representative of those in primary tumours.

Mutations in CASP8 and FAT1
To evaluate the impact of mutations and expression of putative driver genes established by genomic studies on OSCC cell behaviour, we conducted a systematic literature review of genes implicated in homoeostasis or disease in stratified squamous epithelia through regulation of adhesion, invasion or motility. We focused on genes that were members of multiple mutated nontissue specific KEGG pathways and/or were most significantly mutated in our study and TCGA. We prioritized those genes harbouring mutations in more than 8% of HNSCCs and at least two of our cell lines. Seven genes were identified e PLEC, CASP8, SYNE2, FAT1, LRP1B, COL11A1 and HUWE1 (Fig. 2A). We also queried TCGA for pair associations of mutations across these genes in HNSCC (Fig. 2B). CASP8 and FAT1 (P value ¼ 0.0003), and SYNE2 and LRP1B (P value ¼ 0.015) were identified as frequently co-mutated in HNSCC.
Of the 279 patients with HNSCC reported by TCGA in 2015 [6], 172 patients had tumours of the oral cavity and the other 107 had tumours of the oropharynx (n ¼ 33), hypopharynx (n ¼ 2) and larynx (n ¼ 72). Considering copy number alterations and truncating, missense and in-frame mutations, 23% of tumours of the oral cavity harboured mutations in FAT1 alone, 5.8% in CASP8 alone and 9.3% in both FAT1 and CASP8, compared with 22%, 2.8% and 1.9%, respectively, in tumours of other areas of the head and neck. The increased frequency of CASP8 and FAT1 co-mutations in OSCC relative to other HNSCC, recent evidence suggesting functions of CASP8 that extend beyond programmed cell death [29], and uncertainty surrounding the role of FAT1 in oncogenesis [30] led us to focus on these two genes.
Whole exome sequencing revealed eight putative inactivating somatic mutations in FAT1 across four of the 16 OSCC lines, and three in CASP8 across three lines (Fig. 2C). Inactivating mutations were defined as frameshift indels, and splice-site, nonsense and missense mutations predicted to be deleterious to protein function by ANNOVAR. Two lines simultaneously harboured mutations in both FAT1 and CASP8. Similar to the gene span of inactivating mutations in TCGA, mutations in FAT1 and CASP8 did not localize to any particular protein domain.
OSCC cells and normal oral keratinocytes. However, there was a trend for lower FAT1 mRNA levels in FAT1 mutant lines compared to wild type (P value < 0.05). In addition, FAT1 mRNA levels were higher in normal oral keratinocytes than OSCC lines (Fig. 2D). These observations were consistent with primary tumour data from TCGA, since CASP8 mRNA levels in tumours with mutant CASP8 did not differ from those with wild type CASP8 at a given copy number, whereas there was a trend to lower FAT1 mRNA levels in tumours bearing FAT1 mutations (Fig. 2E).

Phenotypic effects of loss or inactivation of CASP8 and FAT1
To characterise the effect of FAT1 mutations on OSCC lines we first observed the morphology of cells grown on feeders. As shown in Fig. 3A, cells with FAT1 mutations tended to form less cohesive colonies than cells that were wild type for FAT1. However, FAT1 knockdown in SJG-33 cells, which are wild type for FAT1, was not sufficient to reduce cellecell adhesion ( Fig. 3A and B). We also observed that the level of FAT1 mRNA increased with increased cell density in cells with wild type but not mutant FAT1 (Fig. 3B). This did not correlate with the level of FAT1 mRNA in cells at low density and was more marked in OSCC lines than in control oral keratinocytes.
To determine the impact of FAT1 expression on the growth of OSCC cells in culture, we performed siRNA-mediated knockdown in cells with wild type FAT1 ( Supplementary Fig. 3A) and then compared clonal growth with cells expressing scrambled control siRNA ( Fig. 3C and D; Supplementary Fig. 3B). We also examined cells in which we knocked down CASP8, alone or in combination with FAT1 knockdown. We observed that knockdown of each gene, alone or in combination, stimulated clonal growth of OSCC in culture but had no effect on the growth of primary oral keratinocytes.
Given the association between FAT1 mutation and cellecell adhesion (Fig. 3A) and between CASP8 and integrin-mediated cell migration [29], we next examined the effect of knockdown on cell migration. We monitored the ability of cells to migrate into a circular area exposed in the centre of the culture dish during a 24 h period. In the case of mutant cells, there were frequent individual cells that had migrated into the area independently of the migrating epithelial front (Fig. 3E). Quantitation of the assays showed that knockdown of FAT1 or CASP8 alone and in combination increased cell migration in OSCC but not in primary keratinocytes (Fig. 3F). In some cases the effect of CASP8 knockdown was greater than that of FAT1 (e.g. SJG-8, SJG-32, which harbour FAT1 mutations), whereas in others knockdown of CASP8 and FAT1 individually or in combination was equally effective (SJG-6, SJG-17).
The impact of CASP8 and FAT1 knockdown on clonal growth could not be accounted for by an increase in proliferation, and in several cases Ki67 labelling was, rather, reduced (Fig. 4A). Knockdown of CASP8 and FAT1 did not have a consistent effect on the proportion of cells that underwent terminal differentiation, as judged by the percentage of cells expressing Transglutaminase 1: there was no effect of double knockdown on SJG-17 cells; differentiation was stimulated in SJG-33 cells; and decreased in SJG-41 cells (Fig. 4B). Knockdown of CASP8 and FAT1 did, however, increase resistance to Staurosporine-induced apoptosis ( Supplementary Fig. 4), as judged by Caspase-3 immunostaining (Fig. 4C).
We conclude that loss of function mutations in CASP8 and FAT1 are likely to confer a growth advantage on OSCC by reducing cellecell adhesion, promoting cell migration and conferring resistance to apoptosis. Furthermore, as expected, knockdown tended to have more pronounced effects in cells with wild type CASP8 and FAT1.

Discussion
While many investigators have generated cell lines from human HNSCC and used them to study the biology of this tumour type [14e16], our study is the first to derive cell lines for which matching normal patient DNA is available. Furthermore, we demonstrate that the spectrum of tumour-acquired mutations is representative of that found in tumour biopsies, for example in the case of NOTCH1-3 [6,15]. These lines will be of considerable value for studying tumour cell behaviour in vitro and in xenograft assays.
As proof of the utility of the lines, we studied the impact of loss of functional CASP8 and FAT1, two genetic changes that, we found, frequently occur together in individual OSCC. We found that in the lines, as in primary tumours, inactivating mutations in CASP8 did not result in a reduction in CASP8 expression, whereas FAT1 mutation did correlate with reduced mRNA levels. While FAT1 mutant lines showed reduced intercellular adhesion, FAT1 knockdown in a wild type line was not sufficient to cause a significant change in morphology. It will therefore be of interest to discover whether the effect of FAT1 loss is dependent on other OSCC mutations (Fig. 1D). For example, mutations in PTEN will affect the actin cytoskeleton [31] and AJUBA mutations are likely to impact E-cadherin mediated cell adhesion [32]. Furthermore, although we have not explored the underlying mechanism by which FAT1 mRNA levels increase in some cell lines with increasing cell density but not in others (Fig. 3B), this has the potential to affect cell behaviour via an impact on HIPPO signalling [30].
Knockdown of each gene in cells that were wild type for CASP8 and FAT1 confirmed a clonal growth advantage, consistent with the observation that CASP8 mutant cells are more aggressive in tongue xenografts [15]. This could be attributed, at least in part, to increased cell migration and reduced apoptosis, rather than to an increase in the proportion of actively cycling cells. The consequences of stimulation of terminal differentiation, observed in some cells, remain to be explored but could potentially lead to increased release of cytokines such as IL-6 that render the microenvironment permissive for tumour growth [33] and altered inflammatory signalling in CASP8 mutant cells [19]. Cells that show a more pronounced phenotype on combined knockdown of CASP8 and FAT1 will be useful for exploring the cancer relevance of the observed antagonistic interactions between FAT1 and Caspase 8 [20].
In conclusion, OSCC is highly heterogeneous at both the cellular and genetic levels. The panel of cell lines that we have characterised will be useful for identifying the signalling pathways that are affected by particular combinations of mutations and for discovering how they act both at the level of the cells that bear the mutations and at the level of the surrounding stroma and immune infiltrate. In the long term, this type of approach has the potential to suggest new combination therapies to improve the outcome for patients with OSCC.
Sunadome and Simon Broad for their helpful input. The authors have no competing interests to declare.