Establishment and Genetic Landscape of Precancer Cell Model Systems from the Head and Neck Mucosal Lining

The 5-year survival rate of patients diagnosed with head and neck squamous cell carcinoma (HNSCC) has not much improved during the last decades, and remained around 50% to 60% (1, 2). Despite invasive treatment protocols, patients frequently develop locoregional recurrences and second primary tumors. Primary HNSCCs develop in precancerous mucosal changes coined “fields,” which can be centimeters in diameter and also cause approximately half of the local relapses (3). Field cancerization was first mentioned in 1953 by Slaughter and colleagues (4), who postulated the concept of cancer and relapse development from microscopic changes present in the mucosal lining of the head and neck. Some of these fields are also macroscopically visible as white (leukoplakia) or red (erythroplakia) lesions, but most are only recognized under the microscope as dysplastic mucosal epithelium in biopsies of these lesions or in the surgical margins of excised head and neck cancer specimen, and graded as mild, moderate or severe. Molecular research during the last decades revealed that a variety of tumor-associated genetic changes occur in these fields (1, 5, 6), an observation that formed the basis of the very first genetic progression model of head and neck cancer (7). In addition, the presence of these cancer-associated genetic alterations appeared to be the best predictor of malignant transformation of (erythro)leukoplakia as well as of the development of local relapses in the surgical margins of treated HNSCC patients (8–11).

Although treatment of these precancerous fields is attractive to prevent both primary tumors and local relapses, the options are limited. A number of chemopreventive approaches to eradicate these fields have been studied amongst which retinoids and cetuximab, but efficacy was limited and the burden on the patients high (3, 12–15). Visible lesions can be excised or removed by laser vaporization depending on their dimensions, but lesions tend to recur or patients develop tumors elsewhere in the upper aerodigestive tract limiting efficacy (16). For larger lesions and precancerous fields that can be only recognized under the microscope or by non-invasive genetic cytology (17), treatment options are lacking. This may change as several lines of evidence indicate that precancerous fields that cannot be recognized macroscopically may be visualized by autofluorescence (18). Furthermore, it was demonstrated in a preliminary study that fluorescence-guided surgery improved the local recurrence-free survival of patients, providing the proof of concept that effective treatment of precancerous fields will prevent cancer and relapses (19). However, since excision is not always possible and of limited efficacy, chemopreventive treatments are urgently awaited. Toxicity should be limited for preventive therapies, and targeted treatment approaches seem therefore most suited. To be able to select, test and validate such approaches cell culture models of patient-derived precancerous fields are required.

Previously, we generated a precancer cell culture designated as VU-preSCC-M3, by culturing a biopsy from the resection margin of a patient with a T4aN0 glottic larynx carcinoma. Genetic analysis of this culture revealed loss of heterozygosity (LOH) of chromosome arms 3p and 9p and a nonsense mutation of TP53 (20). Others have generated several immortal cell models from leukoplakia, erythroplakia and erythroleukoplakia lesions (21, 22) that we included in our studies. To establish a more representative collection of precancer cell culture models we cultured biopsies from the surgical margins of a large cohort of HNSCC patients. We genetically characterized a selection of the obtained cell cultures with low-coverage whole-genome sequencing (lcWGS) and target-enrichment mutation sequencing.

The study was approved by the Institutional Review Board of the VU University Medical Center (protocols 2008-71 and 2015-345). Tissue samples were collected and cultured after signed informed consent. Use of pseudonymized residual tissue samples from uvulo-palatopharyngeal reconstructions was carried out according to the guidelines of the Dutch Medical Scientific Societies (www.federa.org). TNM staging was performed based on the guidelines of the American Joint Committee on Cancer (23).

Primary fibroblasts were cultured in Dulbecco's Modified Eagle Medium (DMEM, Lonza, cat. no. BE12-709F), with 10% fetal bovine serum (FBS, Biological Industries, cat. no. 04-007-1A) and 2 mmol/L L-Glutamine (Lonza, cat. no. BE17-605E). Primary oropharyngeal keratinocytes were cultured in serum-free KGM medium (KGM-SFM, Gibco, cat. no. 17005042), supplemented with 0.1% BSA (Biovision, cat. no. 714100), 25 mg bovine pituitary extract (Gibco, cat. no. 17005042), 2.5 μg human recombinant EGF (Gibco, cat. no. 17005042), 250 μg Amphotericin B (Gibco, cat. no. 15290026) and 250 μg Gentamycin (Sigma-Aldrich, cat. no. G1272).

The cell cultures of the mucosal biopsies were generated as described previously (20), with some small adaptations. In short, surgically removed tumor specimen from mainly the oral cavity and larynx were directly transported to the pathology department. Biopsies of normal appearing mucosa were taken by a pathologist from the tissue surrounding the tumor. When possible, multiple biopsies of different locations were collected, close to the tumor but ensuring that no tumor was biopsied (distance from the tumor edge was 5-10 mm). The tissue was kept at 4°C on storage medium, consisting of KGM-SFM supplemented with 250 μg/mL Amphotericin B, 50 mg/mL Gentamycin, 50 U/ml Penicillin and 50 μg/mL Streptomycin (Lonza, cat. no. DE17602E) and 1% FBS, for 24 to 72 hours to eradicate infectious agents. After this period, the biopsies were transferred to a 1x Dispase II (Roche, cat. no. 4942078001) solution in PBS (Lonza, cat. no. BE17-512F) for 24 hours at 4°C. Next, Dispase II was activated at 37°C for 15 minutes, which enables the separation of the upper mucosal epithelium layer from the underlying tissue. Both separated tissue samples were dissociated in Trypsin (Lonza, cat. no. BE17-161E) for 10 minutes at 37°C. Dissociated cells were rinsed once in serum-free KGM-SFM for the mucosal epithelium layer, or DMEM with 10% FBS for the connective tissue. Subsequently, the cells were cultured in either KGM-SFM supplemented as described above, or in DMEM with 10% FBS, 2 mmol/L L-Glutamine, 50 U/ml Penicillin and 50 μg/ml Streptomycin, respectively. VU-preSCC-M3 (20) was also cultured in KGM-SFM supplemented as described.

In addition, we obtained five immortalized precancer cell cultures from the University of Sheffield, which originated from oral leukoplakia or erythroleukoplakia lesions with different grades of dysplasia (Supplementary Table S1). These cultures; D4, D19, D20, D34 and D35 were generated on 3T3 feeder layers (21, 22), but can be cultured without feeder layer as well. D34 had to be cultured in KGM-SFM supplemented as described above and the other four cell lines were cultured in DMEM supplemented with 10% FBS, 2 mmol/L L-Glutamine, 1 × 10⁻⁷ M Insulin (Sigma, cat. no. 91077C), 2.5 × 10⁻⁴ M Hydrocortisone (Sigma, cat. no. H0888) and 2.5 μg/L EGF (Invitrogen, cat. no. PHG6045). All cells were cultured in a humidified atmosphere at 37°C under 5% CO₂ and were tested for the presence of mycoplasma as soon as possible (Mycoalert, Lonza, cat. no. LT07-318).

From the surgical specimens described above, a biopsy from the tumor was collected as well and brought in culture as described previously (24). These tumor cells were cultured in DMEM with 10% FBS, 2 mmol/L L-Glutamine, 50 U/ml Penicillin and 50 μg/ml Streptomycin. Once established, cell lines were tested for the presence of mycoplasma as soon as possible. Other established HNSCC tumor cell lines were cultured in DMEM, complemented with 5% FBS and 2 mmol/L L-Glutamine. VU-SCC-040 (formerly 92VU040), VU-SCC-096 (formerly 93VU096), VU-SCC-120 (formerly 93VU120), VU-SCC-147 (formerly 93VU147) and VU-SCC-OE, were reported by Hermsen and colleagues (24). VU-SCC-1131 (formerly VU1131) and VU-SCC-1365 (formerly VU1365) were described in van Zeeburg and colleagues (25), and VU-SCC-1604 in Stoepker and colleagues (26). UM-SCC-22A, UM-SCC-22B, UM-SCC-11B, UM-SCC-06, UM-SCC-11B, UM-SCC-38 and UM-SCC-47 were obtained from Prof. T. Carey (University of Michigan, Ann Arbor, MI; ref. 27). FaDu was obtained from the American Type Culture Collection. All tumor cells were cultured in a humidified atmosphere at 37°C under 5% CO_2. The established cell lines were mycoplasma checked regularly and authenticated by visual inspection (cell dimensions and growth characteristics) and by genetic profiling on indication. Genetic patterns of earliest passages of the UM-SCC- cell lines have been made publicly available for authentication (28), or were analyzed internally for the VU-SCC- cell lines. All cell lines have not been passaged for more than 6 months in our study after thawing.

From the start of culture, population doublings (PDs) were registered by counting cells every passage. The initial number of cells that started to proliferate, 24 hours after the keratinocytes were harvested, had to be estimated. PDs were calculated with the formula; 2^N = (Cf/Ci), where N is the number of PDs, Cf is the number of cells harvested and Ci is the total number of cells at start. On the basis of previous literature and our own experience with culturing normal keratinocytes, a PD up to 20 is considered as a normal life span (29, 30). Thus, every culture with a PD of 20 or above was assumed to have an extended life span.

Genomic DNA (gDNA) was isolated with the NucleoSpin Tissue kit (Macherey Nagel, cat. no. 740952) with an overnight proteinase K incubation step and yields were analyzed on a Qubit 2.0 (Invitrogen). The DNA was then prepared for sequencing with the TruSeq Nano kit (Illumina, cat. no. 20015965). In short, 150 ng of DNA was used as input and was sheared using a Covaris S2. Fragments were end-repaired and 3′ adenylated, after which adapters were ligated, and the sequence library amplified with eight PCR cycles according to the Illumina protocol. Next, the yield was determined with a Bioanalyzer DNA 7500 chip (Agilent Technologies). Libraries were equimolar pooled with 20-21 barcoded samples per lane and sequenced on a HiSeq 2500 (Illumina) in a single-read 50-cycle run mode (SR50). Next, raw reads were trimmed, and mapped to the human reference genome build GRCh37/hg19 with BWA (31). Analysis of the data was performed with the QDNAseq R-script (version 1.14.0) previous described by Scheinin and colleagues (32). The reads were binned in 30 kb bins to generate the copy number (CN) profiles. The most stringent filter was applied that removes problematic genomic regions for sequencing. The data were further filtered against a blacklist of regions known to be copy-number variants (CNV; ref. 33). Log₂ ratios were median-normalized and segmented with the Circular Binary Segmentation algorithm (34) as incorporated in the R-script. Calling of the segments was done using CGHcall (version 2.40.0; ref. 35). Copy-number aberrations (CNA) were scored as focal when the size was ≤3 Mb (36).

The gene panel for target enrichment sequencing was based on the genomic analysis of the TCGA network and included; AJUBA, CASP8, CDKN2A, FAT1, FBXW7, HRAS, MLL2 (KMT2D), NOTCH1, NSD1, PIK3CA, PTEN, and TP53 (37). Target enrichment for sequencing was performed using the HaloPlex Target Enrichment system (protocol version F1, Agilent Technologies). Briefly, 225 ng of gDNA was digested in eight different digestion reactions, each containing two restriction enzymes. DNA was then hybridized to the HaloPlex probes provided with barcodes, and the products were then captured with magnetic beads. Next, the products were ligated and amplified with 21 PCR cycles according to the protocols of Agilent Technologies. The PCR amplicons were purified with AMPure XP beads (Beckman Coulter, cat. no. A63881). Before sequencing, the quality of the samples was analyzed with a Bioanalyzer High Sensitivity DNA chip (Agilent Technologies). Samples were sequenced on the Illumina HiSeq 2500 (125 bp paired end) or the Illumina Miseq (150 bp paired end). Sequencing reads were aligned to the human genome (Human hg19) with SureCall (Agilent Technologies). Synonymous mutations were removed as well as variants that were identified as SNPs in the 1,000 Genome project Phase 3 (EUR_AF>0.01; ref. 38). Sanger sequencing was used to re-sequence the identified variants and to exclude any SNPs or germline mutations by DNA analysis of either blood or normal fibroblasts. In short, PCR products were treated with Shrink Alkaline Phosphastase and Exonucleoase I (Affymetrix, cat. no. 70996) for 30 minutes at 37°C followed by 15 minutes at 80°C. The cycle sequencing reaction was carried out with the BigDyeTerminator 3.1 kit (Applied Biosystems, cat. no. 4337454). In total, 2 pmol of the exon-specific sequencing primers were added and sequencing was performed on the ABI3500 (Applied Biosystems).

UM-SCC-22A, VU-SCC-OE, VU-SCC-147 (HPV+ve), UM-SCC-47 (HPV+ve), VU-SCC-120, FaDu, VU-SCC-1131, UM-SCC-038, VU-SCC-096, UM-SCC-11B, VU-SCC-040, VU-preSCC-M3, D4, D19, D35 and HN433-M1 were subcutaneously injected in both flanks of 8-week-old female athymic Nude-Foxn1^nu mice (Envigo). Per site 2 × 10⁶ cells were injected in three mice per cell line. Tumor growth was monitored by visual inspection twice a week for a period of 3 months. All animal experiments were performed according to Dutch and EU legislations, and the protocol (13-04) was approved by the Institutional Review Board on animal experimentation.

In total 108 biopsies of normal appearing mucosal tissue surrounding HNSCCs of 36 patients were collected and cultured. Because of infection or lack of cell proliferation, 10 cultures (9%) were lost and thus 98 biopsies of 32 patients could be cultured successfully (Supplementary Table S2). To reduce the workload we preselected cultures for further genetic characterization. We focused on cultures with a higher number of PDs, particularly because these cultures were generated for in vitro target identification and drug testing in precancer cells, demanding a suitable proliferation capacity. A culture with ≥20 PDs was considered to have an extended life span, and these might be enriched for cancer-associated genetic alterations. Only 20 (20%) of 99 cultures reached a PD of 20 or more (Fig. 1), which includes VU-preSCC-M3, the previously established precancer cell culture (20). Five other precancer cell lines derived from visible lesions were added to the collection as well (Supplementary Table S1; refs. 21, 22).

In parallel, we attempted to culture tumor cells from patients of the same cohort. We collected 65 tumor biopsies from 36 patients of which we established five in vitro tumor cell cultures (Supplementary Table S2; patients 10, 11, 12, 30, and 33).

First, copy number (CN) profiles were established from two cultures of primary keratinocytes and three cultures of primary fibroblasts of noncancer controls, to check for induction of CNAs by culturing. All primary keratinocyte and fibroblast cultures showed normal CN profiles without any aberrations, as expected. There was one remarkable exception, the DNA of both the keratinocytes and fibroblasts of control UPPP50 revealed a consistent focal loss at chromosome 17p (Supplementary Fig. S1). Loss of this particular region is known to be related to hereditary neuropathy with liability to pressure palsies (HNPP; ref. 39). After consultation with the treating physician, it became clear that HNPP indeed had been diagnosed, and we decided not to consider this focal loss as a somatic or culture induced change.

Next, we analyzed the 20 cultures with an extended life span and nine randomly selected cultures with a normal life span (<20 PDs; Fig. 1), thus 29 in total. Of the 20 cultures with an extended life span, nine cases contained CNAs and somatic mutations (45%; Table 1A and Fig. 2). Of the nine cultures with a normal life span (<20 PDs), remarkably, four samples (44%) displayed several (focal) aberrations at different chromosomes and/or somatic mutations (Table 1A). We analyzed if there was an association between the number of PDs (<20 or ≥20) and the presence of genetic aberrations (yes vs. no), but could not find any correlation (Fisher's exact test; P = 0.6).

The obtained five (erythro)leukoplakia cultures were analyzed likewise for genetic aberrations. These all showed aberrant CN profiles with many HNSCC-associated changes (Table 1B). Analysis of the five established new tumor cell cultures also showed a wide variety of CNAs, see Fig. 2C for a representative example. These CNAs are common for HNSCC, including losses at 3p (3p12.2, 3p14.2, 3p25.3) and 8p (8p23.2), and amplifications of chromosomes 3q (3q26.33) and 8q (8q24.21, 8q11.21; ref. 37). In addition, all five tumor cultures contained at least one mutation in TP53 and 3 out of 5 also displayed a loss of the CDKN2A region on chromosome 9p (Supplementary Table S3).

The number of CNAs varied widely in the collection of precancer cell cultures. The number of segments of all analyzed cultures were determined and varied from 22 till 86, with 22 as normal as the sex chromosomes were excluded (Fig. 3). The cultures originating from the resection margins demonstrated a range of segments from 22 till 33. The cultures originating from dysplastic visible lesions (D4, D19, D20, D34, and D35) contained more aberrations, with segment numbers ranging from 39 to 86 (Fig. 3). The tumor cell cultures from the same patient cohort had a median number of 126 segments, which is comparable with the TCGA tumor profiling data with 136 segments (37). To make a comprehensive comparison, we also depicted the number of segments in several established tumor cell lines in Fig. 3. The specific CN and mutational profiles of these cell lines will be published elsewhere in detail (article in preparation). The HNSCC cell lines that we included showed a median number of 93 segments (Fig. 3).

Most of the identified CNAs in the precancer cell cultures are known to be associated with head and neck cancer development (ref. 1, 37, 40, 41; Supplementary Table S4A). For instance, loss of the locus at chromosome 9p encompassing CDKN2A, was observed in 8 of the 13 (62%) margin cultures containing genetic alterations, in 3 out of 5 (erythro)leukoplakia cultures and in 3 out of the 5 tumor cultures (60%; Table 1; Supplementary Table S3). Of note, the promoter region of CDKN2A in D34 is hypermethylated according to a previous report (22), an epigenetic event that may inactivate this tumor-suppressor gene.

Chromosome 3 is also often affected in HNSCC, about 58% of tumors previously studied contained a gain at 3q encompassing the PIK3CA gene, and 3p is often lost in tumor tissue (76%) whereas the relevant gene is still unknown (42). In our collection, one or more of the recurrent areas on chromosome 3; 3q26.33, 3q28 and 3p12.2, 3p14.2, 3p25.3 (37) were affected in only 1 out of 29 (3%) margin cultures tested, while changes were found in all five (erythro)leukoplakia cultures and in all five tumor cell cultures. This observation suggests that these chromosome 3 changes arise later during (pre)cancer development.

Remarkably, some other aberrations frequently detected in HNSCC such as EGFR and CCND1 amplifications were almost absent in our precancer collection. These aberrations, present in 15% and 31% of the HNSCC patients, respectively (37), are apparently also not involved in early carcinogenesis, but likely develop later during progression.

In summary, the surgical margin cultures contained multiple HNSCC-associated CNAs (Supplementary Fig. S2), where loss of the CDKN2A region at chromosome 9p was most abundant. The (erythro)leukoplakia cultures showed a more HNSCC-like CN profile, and apparently represent more advanced lesions (Supplementary Fig. S2).

The quantification and ranking of the number of segments as described above raised another question. Generally, precancer cell cultures from the surgical margins display less changed segments and can be separated from the tumor cell lines based on the number of segments, whereas the cultures originating from the dysplastic visible lesions contained higher numbers of segments, and can hardly be distinguished from the tumor cell lines (Fig. 3). We wondered on the basis of these findings whether the latter had been microinvasive or even early stage tumors instead of advanced precancers. To examine their tumor engraftment capacity, cells of D4, D19, D35, VU-preSCC-M3 and HN433-M1 were injected in nude mice. As a control several established tumor cell lines were injected in parallel. Tumor growth was monitored during three months. From the 11 HNSCC cell lines, nine showed tumor growth with a take rate of 70% or more (Table 2). From the five tested precancer cell culture models, none showed take during the three months follow-up, obviously a highly significant difference (Fisher's Exact Test; P < 0.01). Hence, despite the “tumor-like” genetic profiles of some of these precancer cell cultures, they still lack the invasive characteristics and engraftment capacity of tumor cells.

In addition to lcWGS for CNAs, we also applied HaloPlex target-enrichment sequencing of a custom designed cancer gene panel. Here, we included all 20 samples with an extended life span, including VU-preSCC-M3, three samples with a life span below 20 PDs (Fig. 1), and four of the obtained (erythro)leukoplakia cultures. The target regions were sequenced with a mean coverage of 3,704x or 18,501x on the Illumina Miseq or Hiseq, respectively. All germline mutations/SNPs were excluded on the basis of filtering and sequencing of either blood or normal fibroblasts from the same patients. All somatic variants were re-sequenced by Sanger sequencing and 30 of the 31 (97%) somatic mutations were confirmed (Supplementary Table S4B). The identified and verified mutations are listed in Table 1.

In 9 out of the 23 analyzed margin cultures we detected somatic mutations in TP53, NSD1, NOTCH1, and/or FAT1. Most frequently mutated genes were TP53 and NOTCH1. In addition, all four analyzed (erythro)leukoplakia cell cultures contained one or two mutations in TP53, and mutations in CASP8, PIK3CA, FBXW7, FAT1, CDKN2A, and MLL2. In all precancer cell cultures the AJUBA, HRAS, and PTEN genes were wild type.

Of the 27 analyzed cultures, we identified TP53 mutations in 9 out of 27 cultures (33%), including missense, nonsense, and frameshift mutations (Fig. 4, Table 1, Supplementary Table S4B). This is a rather low frequency compared with the 72% mutations identified in tumors (37). Furthermore, in our precancer collection we identified six NOTCH1 mutations (22%); three missense and three frameshift mutations (Fig. 4, Table 1, Supplementary Table S4B), indicating that mutation of NOTCH1 might also be an early event in HNSCC development. Interestingly, the NOTCH1 mutations were all present in cultures that originate from biopsies of surgical margins, and none in those that arose from (erythro)leukoplakia lesions. In HNSCC, NOTCH1 is mutated in 19% of the patients (37), a very similar frequency to the 22% we found, also supporting that NOTCH1 mutations are an early event but apparently not or less in visible lesions.

Generally, all cultures containing mutations also displayed CNAs, with one exception HN439-M1 (Table 1A). However, the VAF score for the NOTCH1 mutation in this culture was only 10%, suggesting that only a fraction of the cells were of precancer origin that might have obscured CNA detection. Very remarkable, in three precancer cell cultures we only detected somatic CNAs and no somatic mutations in the sequenced driver genes (Table 1A, Fig. 2B; Supplementary Figs. S3B and S4A), suggesting that carcinogenesis may initiate with CNAs, although other driver genes might be in play.

The presence of dysplasia in the resection margins was revised by an experienced HNSCC pathologist, according to the WHO criteria. In addition, the follow-up data of all patients was documented, as well as smoking habits (Supplementary Table S2). Next, we searched for associations between dysplasia (scored as yes vs. no, or no/mild vs. moderate/severe) and the number of PDs (<20 or ≥20) of all patients (n = 33) (Supplementary Table S2). We could not find any association for both ways of dysplasia scoring (Fisher's Exact Tests; P = 1.0 and P = 0.7, respectively). Also smoking habits (yes vs. no) or the development of disease relapses were not associated with the number of PDs (Fisher's Exact Tests; P = 1.0 and P = 0.7, respectively). To analyze whether there are associations between the presence of genetic alterations in the cultures (yes vs. no) and the presence of dysplasia in the surgical margins of the resection specimen they were established from, we included the 18 patients of which we genetically characterized one or more cultures. There was no association when dysplasia was stratified as yes vs. no (Fisher's Exact Test; P = 0.6), but when dysplasia was stratified as either no/mild vs. moderate/severe, we did find an association with the presence of genetic changes in the cultures (Fisher's Exact Test; P = 0.04). Moderate/severe dysplastic margins were more likely to result in a culture with genetic changes. Of note, the margin biopsies that were used to establish the cultures and the margins sampled for routine histological analyses could not be geographically oriented, which may have impacted the analysis. Finally, no associations were found between the presence or absence of genetic alterations and the occurrence of a disease relapse or smoking habits (both stratified as yes vs. no; Fisher's Exact Tests; both P = 1.0).

From two patients, we established multiple precancer cell cultures from the surgical margins. We collected four biopsies surrounding the tumor of patient HN439, a 73-year old male with a T1N0 laryngeal carcinoma. Interestingly, the CN profiles as well as the mutational profiles demonstrated the presence of three genetically independent precancerous fields (Fig. 5, Table 1A). HN439-M2 and M4 both have an identical homozygous deletion at the location of CDKN2A, whereas M3 showed only a hemizygous loss. In addition, M2 and M4 both contained identical losses at chromosome 11p (Fig. 5A and C), whereas M3 displayed losses at chromosome 6q and 12p (Fig. 5B). Culture M1 contained no aberrations in the CN profile, but did contain a somatic missense mutation in NOTCH1 in a small fraction of the cells. Mutations in M2 and M4, also identical, were found in FAT1 and NOTCH1, while M3 contained mutations in the genes NSD1 and TP53 (Table 1A).

From a second patient HN472, a 64-year old male with a T1N0 mobile tongue carcinoma, we were able to culture four biopsies and established a successful tumor culture as well. CN profiles were generated for the tumor, for one culture with an extended life span (M4) and for 1 of the 3 cultures with less than 20 PDs (M1; Supplementary Fig. S3). CNAs were detected in all three samples; however, none were related to each other. In addition, the TP53 mutations found in the tumor cells and in mucosal culture M1 were also not identical and M4 was TP53 wild type (Table 1A; Supplementary Table S3). Based on these data, it seems that none of the two fields are related to the tumor, and these fields are most likely independent from each other as well.

In this study, we describe the generation of a collection of precancer cell cultures suitable for analysis of inhibitors and functional genetic screens with siRNAs and CRISPR/Cas9, and the genomic characterization thereof. Because a longer life span would be required for subsequent experiments, we set a cutoff value at 20 PDs as a way to select precancer cell cultures for genetic characterization. We expected that cultures with cancer-associated genetic alterations would be able to circumvent cell-cycle control and proliferate at least longer than normal primary cells. Our data indicate that life span in vitro is not a key characteristic of precancer cells. A random selection of cultures with less than 20 PDs showed a comparable fraction with genetic changes. Therefore, only the presence of somatic genetic changes should be used as definition.

Although we cannot formally exclude the possibility that the identified genetic alterations are (in part) induced by the culturing itself, there are several strong arguments that the identified cancer-associated genetic alterations are authentic and somatic. First, about half of the cultures that underwent 20 PDs or more contained no genetic changes at all. For instance, there were cultures with 26, 27 and 28 PDs that displayed no genetic aberrations and one of the cultures without genetic alterations even exceeded 40 PDs. Hence, even in 40 doublings not a single change was induced. In contrast, a culture that underwent only 5 PDs already showed several CNAs as well as a mutation in TP53. Secondly, the keratinocyte and fibroblast cultures of several controls (n = 5) did not contain any somatic cancer-associated genetic alterations. Thirdly, in our collection we established two separate cultures from one patient that were completely identical, both for the CN profiles as well as for the presence of two somatic mutations. Hence, even with these changes, genetic profiles remained identical during culture propagation. Fourthly, for HN433-M1 we ran the lcWGS twice on two samples, one at the moment of 6 PDs and one at 30 PDs. Both CN profiles where identical, and apparently stable (Fig. 2B; Supplementary Fig. S4B). Altogether, these results support that our genetic data are genuine, and reflect the changes that were present in the tissue.

The precancer cell cultures described here, displayed a wide range of CNAs resulting in a variable number of segments, suggesting differences in stage of progression. On the basis of the engraftment experiments, they are all clearly precancerous, even when they contain multiple CNAs and are hardly distinguishable from cancer cells. The (erythro)leukoplakia cultures that where included in this study, were all immortal and derived from moderate to severe dysplastic lesions or were even already carcinoma in situ (21, 22, 43). This might explain why these cultures appear to have many more genetic aberrations, but nonetheless they do not engraft in nude mice.

Finally, the following observations were most interesting in our view. The pattern of genetic changes varied widely, most likely reflecting the complex and gradual development of squamous cell carcinomas in patients. Obviously, we cannot conclude whether all of the precancerous changes reflected in these cultures might have resulted in the development of cancer. Possibly some might have disappeared in vivo. Genetic profiling of progressing and non-progressing lesions in the future will reveal this. In addition, somatic changes in TP53 and CDKN2A were expected, but the frequent mutations of NOTCH1 as an early change was a novel finding. Also intriguing is the finding of three precancer cell cultures that only displayed (cancer-associated) CNAs, but no somatic driver mutations as far as tested. Exome sequencing of these cases might reveal whether other driver mutations, less frequently found in HNSCC, are present, or that carcinogenesis may initiate with CNAs instead of mutations. In most cases mutations and CNAs were identified in combination. Finally, the lack of 3p loss in the very large majority of cultures is quite surprising because this is claimed to be an early event (7). We cannot exclude the presence of allelic losses with intact copy numbers, although in the TCGA data the losses at 3p are usually numerical, at least in tumors (37).

The established cultures seem to form a representative collection of the precursor stages of HNSCC based on genetic analysis. Thereby they form a suitable in vitro model to develop targeted treatments to prevent HNSCC (relapse) formation. Previously, we already presented an approach to identify new targeted treatments by using siRNA-screening methods on these precancer cell cultures (44), and came up with PLK1 as druggable target.

No potential conflicts of interest were disclosed.

Conception and design: D.V. de Boer, R.H. Brakenhoff

Development of methodology: D.V. de Boer, K.D. Hunter, B. Ylstra, R.H. Brakenhoff

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D.V. de Boer, A. Brink, M. Buijze, K.D. Hunter, E. Bloemena

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D.V. de Boer, A. Brink, M. Buijze, C.R. Leemans, R.H. Brakenhoff

Writing, review, and/or revision of the manuscript: D.V. de Boer, A. Brink, K.D. Hunter, B. Ylstra, E. Bloemena, C.R. Leemans, R.H. Brakenhoff

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D.V. de Boer, A. Brink, M. Stigter-van Walsum

Study supervision: R.H. Brakenhoff

The authors would like to thank Daoud Sie for the use of his blacklist set to filter our copy number data and Boudewijn Braakhuis for his contribution to the initial study design. This work and D.V. de Boer were supported by a grant from the VUmc CCA Foundation (Grant nr.: CCA 2012-2-06).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.