Comprehensive Genomic Profiling of Patient-Matched Head and Neck Cancer Cells: a Preclinical Pipeline for Metastatic and Recurrent Disease

Published OnlineFirst August 14, 2018; DOI: 10.1158/1541-7786.MCR-18-0056

Genomics Molecular Cancer Research Comprehensive Genomic Proﬁling of Patient- matched Head and Neck Cancer Cells: A Preclinical Pipeline for Metastatic and Recurrent Disease Lluís Nisa1,2,3, David Barras4, Michaela Medova1,2, Daniel M. Aebersold1,2, Matus Medo1,2, Michaela Poliakova1,2, Jonas Koch1,2, Beat Bojaxhiu1, Olgun Elicin¸ 1, Matthias S. Dettmer5, Paolo Angelino4, Roland Giger3, Urs Borner3, Marco D. Caversaccio3, Thomas E. Carey6,7,LizaHo8, Thomas A. McKee8, Mauro Delorenzi4,9,10, and Yitzhak Zimmer1,2

Abstract

Metastases and tumor recurrence have a major prognostic confirmedcommonSPRR2Adownregulationinprimary impact in head and neck squamous cell carcinoma tumors (61.9% of cases) and lymph node metastases (HNSCC); however, cellular models that comprehensively (31.3%), but not in normal tissue. High expression of characterize metastatic and recurrent HNSCC are lacking. To SPRR2A in lymph node metastases was, along with non- this end, we obtained genomic, transcriptomic, and copy oropharyngeal location of the primary tumor, an indepen- number profiles of the UM-SCC cell line panel, encompass- dent prognostic factor for regional disease recurrence after ing patient-matched metastatic and recurrent cells. UM-SCC surgery and radiotherapy (HR 2.81; 95% CI, 1.16–6.79; P ¼ cells recapitulate the most prevalent genomic alterations 0.02). These results suggest that SPRR2A plays a dual role in described in HNSCC, featuring common TP53, PI3K, invasion and therapeutic resistance in HNSCC, respectively NOTCH, and Hippo pathway mutations. This analysis through its downregulation and overexpression. identified a novel F977Y kinase domain PIK3CA mutation exclusively present in a recurrent cell line (UM-SCC14B), Implications: The current study reveals translationally rele- potentially conferring resistance to PI3K inhibitors. Small vant mechanisms underlying metastasis and recurrence in proline-rich protein 2A (SPRR2A), a protein involved in HNSCC and represents an adjuvant tool for preclinical epithelial homeostasis and invasion, was one of the most research in this disease setting. Underlining its discovery consistently downregulated transcripts in metastatic and potential this approach identified a PIK3CA-resistant muta- recurrent UM-SCC cells. Assessment of SPRR2A protein tion as well as SPRR2A as possible theragnostic markers. expression in a clinical cohort of patients with HNSCC Mol Cancer Res; 1–15. 2018 AACR.

1Department of Radiation Oncology, Inselspital, Bern University Hospital, Introduction 2 University of Bern, Bern, Switzerland. Department for BioMedical Research, Development of regional or distant metastases and recurrence Inselspital, Bern University Hospital, University of Bern, Bern, Switzerland. fi 3Department of Otorhinolaryngology – Head and Neck Surgery, Inselspital, after de nitive therapy are arguably the most outstanding issues in Bern University Hospital, University of Bern, Bern, Switzerland. 4SIB Swiss the management of patients with head and neck squamous cell Institute of Bioinformatics, Lausanne, Switzerland. 5Institute of Pathology, carcinoma (HNSCC). Presence of lymph node metastases has a University of Bern, Bern, Switzerland. 6Department of Otolaryngology - Head major prognostic impact, as it reduces survival of HNSCC patients and Neck Surgery, University of Michigan School of Medicine, Ann Arbor, roughly by half (1). Moreover, lymph node metastases do not 7 Michigan. Comprehensive Cancer Center, University of Michigan School of necessarily respond to given therapeutic approaches in the same Medicine, Ann Arbor, Michigan. 8Division of Clinical Pathology, Geneva University Hospitals, Geneva, Switzerland. 9Ludwig Center for Cancer Research, way as their matched primaries do, an observation likely under- University of Lausanne, Lausanne, Switzerland. 10Department of Oncology, lying different intrinsic biological mechanisms (2). Such a het- University of Lausanne, Lausanne, Switzerland. erogeneous behavior poses an obvious challenge to achieve Note: Supplementary data for this article are available at Molecular Cancer simultaneous local (primary tumor) and regional (lymph node Research Online (http://mcr.aacrjournals.org/). metastases) disease control. Indeed, while most early-stage (with L. Nisa and D. Barras are co-first authors of this article. the exception of oral squamous cell carcinoma) and some advanced-stage primaries can be usually managed by radiother- M. Delorenzi and Y. Zimmer are co-senior authors of this article. apy (with or without chemotherapy/cetuximab), presence of Corresponding Author: Yitzhak Zimmer, Department for BioMedical Research, lymph node metastases often requires combined approaches, € Maurice E. Muller-Haus, E-807, Murtenstrasse 35, Bern 3008, Switzerland. encompassing neck dissection and irradiation (3). Distant metas- Phone: 413-1632-2642; Fax: 413-1632-3297; E-mail: [email protected]. tases develop in approximately 10%–30% of HNSCC patients doi: 10.1158/1541-7786.MCR-18-0056 after definitive therapy and carries a meager prognosis (approx- 2018 American Association for Cancer Research. imately 5% overall 5-year survival; ref. 4).

www.aacrjournals.org OF1

Nisa et al.

With respect to therapeutic resistance, recurrences after defin- the most common aberrations found in HNSCC are recapitulated itive therapy in HNSCC occur in 30%–50% of the cases, depend- in the cell lines. Moreover, the current datasets underline the ing on primary tumor anatomic location, stage at presentation, discovery potential of the UM-SCC panel and represent a unique and human papillomavirus (HPV) status among other factors (5). toolset for the preclinical study of invasion and treatment resis- Management options for treatment failures after multimodal tance mechanisms in HNSCC. therapy, including extensive salvage surgery, reirradiation, or fi palliation, are limited and often of limited ef cacy. Moreover, Materials and Methods such approaches carry almost systematically high rates of severe toxicity and/or functional impairment of breathing, swallowing, Ethical statement airway protection, and speech production (6, 7). Patients' tissue and clinical data were collected after approval by The inescapable corollary of these clinical considerations is the the Bernese Cantonal Ethical Committee (Protocol 050/14). The pressing need to develop and implement more effective thera- UM-SCC cell lines were obtained from surgically excised tissues peutic strategies for metastatic and recurrent HNSCC. from patients treated at the University of Michigan (Ann Arbor, The ability to tailor individual management approaches and to MI) who gave written informed consent for the use of their tissues implement them requires thorough characterization of the molec- for research studies, including the development of permanent cell ular alterations in HNSCC and extensive preclinical validation. lines. In this sense, concerted next-generation sequencing (NGS) approaches in HNSCC have been published over the last few Cell lines and reagents years, providing an unprecedented amount of information on the UM-SCC cell lines were provided by Professor T. Carey most frequent alterations and mechanisms underlying tumori- (University of Michigan, Ann Arbor, MI). Culture, mainte- genesis (8–12). With respect to preclinical validation, cancer cell nance, and cell line identity confirmation were performed as lines and their related in vivo models represent major research described previously (18). For group analysis, cell lines were tools in preclinical oncology when it comes to addressing issues divided in invasive/metastatic (UM-SCC-10B, -14C, -17B and related to tumor biology and target discovery. Comprehensive -22B) and recurrent (UM-SCC-11B, -14B, -74B and -81B), and genomic characterization is often unavailable for commonly used compared with their matched primaries ("A" series). cell lines, a limitation that hinders the generalization of results PI3K was inhibited with pictilisib (GDC-0941, AbMole Bio- and accounts at least to a certain extent for the discordant Science). This drug was dissolved in DMSO and stored at 20 C. responses to novel therapies seen in preclinical versus clinical settings (13). Karyotypes Recognizing the crucial nature of this issue, Barretina and For chromosome analysis, cells were grown as monolayers on colleagues (14) established genomic profiles of a large number coverslips (35 mm dish, 22 22 mm Glass, MatTek corporation) of human cancer cell lines. In their study, the authors explored and harvested in situ. Cultures in exponential growth were correlations between genomic background and responses arrested at metaphase by adding 10 mg/mL colcemid (KaryoMax to particular therapeutic agents, emphasizing the relevance of Colcemid, Gibco) for 20 minutes at 37 C. Cells were then genetically driven management approaches. Several other stud- harvested with hypotonic saline solution (60 mmol/L KCl), fixed ies provide similar datasets for cell lines from different tumor with fresh fixative (3:1 methanol: acetic acid, Scharlau), dried in a entities, and databases such as the catalogue of somatic muta- drying chamber (25 C and 40% humidity, CDS-5, Thermotron), tions in cancer (COSMIC) systematically gather results from aged for 1 hour at 93 C on a hot plate (Pr€azitherm, Huber&Co these studies to facilitate access to the research community AG), and stained using Giemsa (azur eosin methylene blue (www.cancer.sanger.ac.uk/cosmic). With respect to HNSCC, solution, Scharlau) and trypsin (Difco Trypsin 250, Dr. Grogg Martin and colleagues (15) obtained full exome and transcrip- Chemie). Capturing and karyotyping was performed with the tome profiles of 21 HNSCC cell lines, showing that most typical commercial software Genikon (Nikon). genomic alterations found in HNSCC patients were recapitulated in these cell lines. While this study and others cover the Transcriptome and exome profiling spectrum of alterations in primary tumors, patient-derived Nucleic acids were extracted following lysis of cell pellets in models to address research questions pertaining to mechanisms 600 mL RTL lysis buffer (AllPrep DNA/RNA/miRNA Universal Kit, of invasion and therapeutic resistance have so far not been Qiagen). The lysates were homogenized in CK14 Precellys comprehensively characterized from a genomic perspective. homogenization tubes (Labgene Scientific) using the Minilys Paralleling the sequencing approaches performed in primary homogenizer (Bertin Technologies). DNA and RNA were purified HNSCCs, two recent studies investigated the genomic landscape from the homogenized lysates using a robotic workstation of metastatic and recurrent HNSCC (16, 17). Hedberg and col- (Qiacube, Qiagen) following manufacturer's instructions and leagues (16) performed whole-exome sequencing from 13 their quality assessed using the Fragment Analyzer (Advanced HNSCC lymph node metastases and 10 recurrent tumors and Analytical Technologies Inc.). RNA sequencing libraries were their matched primaries, while Morris and colleagues (17) carried prepared using the Illumina TruSeq Stranded Total RNA reagents out a targeted mutational approach in 410 cancer-related genes (catalog number RS-122-2201; Illumina) according to the pro- and copy number profiles of several head and neck tumors, tocol supplied by the manufacturer and using 750 ng of total RNA. including 53 HNSCCs. Exome sequencing libraries were produced as follows: 3 mgof Here we provide comprehensive genomic profiles of a HNSCC genomic DNA was fragmented into 150–200 base pair fragments cell line panel derived from primaries and their patient-matched using the Covaris Sample Preparation System (S-series). Samples metastases/recurrences (18). Our results show that the UM-SCC were prepared using SureSelect library prep kit (Agilent). panel is representative of HNSCC from a molecular perspective, as The resulting adaptor-tagged libraries were then hybridized with

OF2 Mol Cancer Res; 2018 Molecular Cancer Research

Proﬁling of Recurrent and Metastatic HNSCC Cell Lines

exon biotin-coated baits from the SureSelect V5 kit (Agilent) for (FDR) values computed by edgeR were used to select the differ- 24 hours at 65C. Hybrids were captured with streptavidin-coated entially expressed genes. Genes with an FDR < 0.25 were kept. beads, PCR amplified and indexed. Genes/cell lines biclustering and heatmaps were constructed Cluster generation was performed with the RNA and DNA using correlation distance as metrics and the average linkage libraries using the Illumina HiSeq PE Cluster Kit v4 cBot reagents agglomeration algorithm and heatmaps (R package nclust version (catalog number PE-401-400) and sequenced on the Illumina 1.9.0, http://bcf.isb-sib.ch/Resources.html). HiSeq 2500 using HiSeq SBS Kit V4 reagents (catalog number FC- Pathway analyses were performed with the single sample GSEA 401-4002). Sequencing data were processed using the Illumina (ssGSEA) and the preranked GSEA methods available at the Pipeline Software version 1.84. Exome sequencing initial number GenePattern web interface of the Broad Institute (http://www. of reads averaged 164 41 (SD) million per sample. The rate of genepattern.broadinstitute.org/). Gene directional likelihood mapping to the hGRC37 genome was of 99.9% 0.1% (SD). ratios resulting from differential gene expression analyses were RNA sequencing initial number of reads averaged 69 13 (SD) used as weights in the preranked GSEA method and pathways million per sample. Reads were first trimmed to remove polyA with an FDR < 0.25 were kept. The single samples GSEA signature and Illumina TruSeq adapter sequences using cutadapt (19), then score of these significant pathways are presented in heatmaps aligned to the human reference hGRC37 genome using the STAR created with the pheatmap package. We used the Hallmark col- aligner (20). Reads that uniquely mapped to the reference genome lection of the MSigDB portal (http://www.broadinstitute.org/ averaged 92.6% 1.6% (SD). The number of counts were gsea/msigdb). The network of 24 proteins SPRR2A interacts with summarized at the gene level using featureCounts (21). Gene was obtained from the STRING database (medium confidence, all expression values were computed from the RPKM (reads per interaction sources, interactions are weighted with the combined kilobase per million) values produced by the rpkm function of score reported by STRING). the edgeR package (version 3.16.5) by adding a pseudocount of 1 and log2-transforming the results. TCGA data analysis Level 3 TCGA RNA sequencing gene expression data was Somatic mutation and copy number variation calling acquired using the RTCGA package. We used the RSEM normal- Exome sequences were processed according to the Genome ized expression values and log2-transformed them adding a Analysis Toolkit (GATK) best practices pipeline, which involves pseudocount of 1. The gene classifier that allows categorizing trimming, alignment using BWA-MEM, marking of duplicates, TCGA patients into HNSCC molecular subtypes was kindly pro- local realignment around indels, and base recalibration (Supple- vided by V. Walter, University of North Carolina at Chapel Hill, mentary Methods; ref. 22). Single-nucleotide polymorphisms Chapel Hill, NC (30). HNSCC subtype scores are defined as the (SNP) calling was performed against a panel of eight normal correlation of the sample gene expression with the subtype tissue samples taken from the salivary gland of independent centroids. TCGA patients and UM-SCC lines were then classified patients obtained within the frame of a different study, using the as subtypes with the highest correlation score. To analyze corre- algorithms of MuTect (23) and VarScan2 (24) with default set- lation of SPRR2A RNA levels and survival, we included patients tings. The variants identified with both MuTect and Varscan2 were for which both transcriptomic data (FPKM-UQ) of primary annotated using ANNOVAR (25). To specifically find mutations tumors and survival/follow-up information was available to involved in recurrence and metastatic progression, the variant calculate overall survival and disease-free survival. A total of calling was also performed comparing each recurrent or meta- 498 patients out of the 806 available (as of February 27, 2018) static cell line directly against its primary tumor counterpart, by fulfilled these criteria. using the same algorithms. We determined known cancer-relevant genes contained in Real-time PCR and genomic locus sequencing chromosomal regions that were amplified or lost, using the Total RNA was extracted from 80% confluent cell cultures using curated cancer-driver gene database derived from Rubio-Perez TRIzol lysis according to manufacturer's instructions (Roche). and colleagues (https://www.intogen.org/downloads; ref. 26). Reverse transcription of mRNA was performed using the Omnis- The GISTIC2 software was used to find copy number alterations cript RT Kit (Qiagen). Quantitative PCR of SPRR2A was per- recurring in multiple samples. formed using a 7900HT fast real-time PCR system with a TaqMan Somatic copy number variation between primary cell lines assay (Applied Biosystems). The mean Ct was determined from and metastatic/recurrent counterparts was estimated using the triplicate experiments and mRNA levels normalized to those VEGAWES R package (27) and represented in Circos plots (28). obtained for 28S ribosomal protein. Changes in expression were Use of VEGAWES for CNV (copy number variation) calling was determined by calculations of DDCt. previously validated by establishing karyotypes of two randomly A 1 kbp region around the SNP of interest was amplified by selected cell lines (UM-SCC-17A and UM-SCC-17B). Karyotypes PCR, purified on agarose gel, and sequenced by the Sanger and VEGAWES calling was performed and showed a strikingly capillary method. The resulting electropherograms were manually high concordance (Supplementary Fig. S1). analyzed to confirm the presence of the SNP of interest.

Transcriptome and pathway-level analyses Cell proliferation, clonogenic assays, and cell migration Paired differential gene expression analyses were performed Cell viability was determined using a resazurin sodium salt with the generalized linear modeling functions glmFit and reduction assay (Sigma). Brieﬂy, at the indicated experimental glmLRT of the edgeR package (version 3.16.5) using integer read endpoints cells cultured in 96-well plates were supplemented counts as input (29). Only genes whose expression was consis- with medium containing 44 mmol/L resazurin. Resazurin reduc- tently up- or downregulated in all 4 cell lines of either group tion was colorimetrically estimated 1 and 6 hours later (570/ (metastases and/or recurrences) were kept. The false discovery rate 600 nm) using a Tecan Reader (Tecan Group Ltd.). The

www.aacrjournals.org Mol Cancer Res; 2018 OF3

Nisa et al.

measurements obtained in every treatment condition were nor- or Fisher exact test. Univariate survival analysis was plotted malized to vehicle-treated controls. Results are representative of according to the Kaplan–Meier method and compared with the three independent experiments. log-rank test. Multivariate survival analysis was performed using For clonogenic assays, cells were plated and left to attach Cox proportional hazards regression analysis, using established overnight. Irradiation was delivered using a 137Cs research irra- risk factors for decreased disease control or survival. All P values diator (MDS Nordion) with a dose rate of 0.75 Gy/minute. The were two-sided and statistical significance was set at 0.05. surviving fraction was normalized to the plating efficiency (PE ¼ colonies formed/cells plated). Radiosensitization was evaluated using the radiation enhancement ratio (RER). A RER significantly Results superior to 1 according to one-sample t test was deemed to Genomic profiles of UM-SCC cell lines recapitulate the indicate radiosensitization (31). molecular heterogeneity and most common alterations in Cell migration was assessed with the Oris Cell Migration HNSCC Assembly Kit (AMS Biotechnology) after 3 days of culture. We employed whole-exome sequencing, copy number varia- Briefly, cells were plated in 96-well plates in presence of a tion (CNV), and RNA sequencing to characterize the UM-SCC stopper which creates a 2-mm diameter cell-free circular area in panel. This panel, which includes 15 cell lines derived from 7 the middle of each well. After cells are attached, cell stoppers are different patients (3 females and 4 males), encompasses all removed and cells may freely migrate. Pictures were captured at anatomical HNSCC subsites (Fig. 1A and B; ref. 18). baseline and after 72 hours using a Leica DC 300F microscope. We applied the molecular HNSCC subtype classifier initially The invaded area in each treatment condition was determined reported by Chung and colleagues (32) and subsequently with the ImageJ software (imagej.nih.gov/ij/). refined and used by a number of studies, including the TCGA (9, 30). After validation of the subtype classifier in the TCGA HNSCC patient cohort cohort, UM-SCC cell lines were classified as atypical (n ¼ 9), Validations in patient-derived tissues were based on a retro- mesenchymal (n ¼ 5), and classical (n ¼ 1; Fig. 1C; Supple- spective cohort including HNSCC formalin-fixed paraffin-embed- mentary Fig. S2A and S2B). Importantly, no cell line displayed ded (FFPE) specimens of patient-matched primary tumors, lymph high centroid correlation for any given molecular subtype, node metastases, and normal reference tissue. All tumors were reflecting a high degree of molecular heterogeneity, which fully stage IV-A and IV-B according to the UICC 7th edition (2009). All parallels observations in HNSCC (30). Equally relevant in patients were treated with curative intent in the Departments of terms of expression profiles, we found that most cell lines Radiation Oncology and Otorhinolaryngology-Head and Neck cluster with their patient-matched counterparts, with the excep- Surgery, Inselspital Bern University Hospital (Bern, Switzerland) tion of UM-SCC-11A/B (Supplementary Fig. S2C). UM-SCC- and received radiotherapy, either as primary therapeutic strategy 11B clustered better with UM-SCC-74A/B than with its own or following surgery for primary tumors, with or without con- matched counterpart, an observation most likely because these comitant systemic therapy (either cisplatin/5-fluorouracil or three cell lines have a clearly higher mesenchymal centroid cetuximab). All patients received neck dissections (Supplemen- correlation and phenotype than UM-SCC-11A (Fig. 1B and C). tary Table S1). In addition, gene expression correlation was not necessarily determined by primary anatomic location or specimen's origin Construction of tissue microarrays and IHC (primary, metastatic, etc.), a finding congruent with previous Tissue microarrays (TMAs) were constructed using a new- studies (Supplementary Fig. S2D; refs. 9, 16, 17, 30). generation tissue arrayer. Briefly, FFPE blocks were retrieved from Also consistent with previous NGS studies, the mutational the archives of the Institute of Pathology, University of Bern (Bern, profiles of the UM-SCC panel featured common alterations of Switzerland). Punches with a diameter of 0.6 mm were transferred TP53, members of the Notch, the PI3K and the Hippo pathway, to receptor TMA blocks. IHC staining was performed with an and of the histone methyltransferases KMT2D and NSD1 (Fig. automated system BOND RX (Leica Biosystems). Sections of 4 mm 1D; Table 1; Fig. 2A; Supplementary Document S1; refs. 8, 9, 12, were deparaffinized and rehydrated in dewax solution (Leica 15, 33). Conversely, however, mutations in other frequently Biosystems). Endogenous peroxide activity was blocked in altered genes such as CDKN2A, CASP8, and TGFBR2 were not H2O2 for 4 minutes. Samples were incubated with specific pri- found in this panel (Fig. 2A). Alterations in TRAF3, commonly mary antibodies for 30 minutes at room temperature: SPRR2A associated with HPV-positive HNSCC, were not encountered in (1:500; Novus Biologicals). Slides were scanned using the Pan- this panel of exclusively HPV-negative cell lines (9). noramic Midi digital slide scanner (3DHISTECH Ltd.). Staining To explore recurrently mutated genes in an unbiased manner, intensity was scored using the following scores: 0, no detectable we focused on genes whose mutation rate across cell lines was staining; 1, weak staining; 2, moderate to strong in up to 50% of high and whose length was short to middle, as long genes are tumor cells; and 3, strong staining in more than 50% of tumor statistically more likely to accumulate passenger mutations cells. For subsequent ad hoc group comparisons, scores 0 and 1 (Fig. 1E). Among these were NOD2, a gene involved in immune were considered as "low staining intensity" and 2 and 3 as "high responses whose mutations have been reported in several types of staining intensity." cancers (34). Another interesting finding was presence of ADAMTSL4, member of a gene family involved in cellular adhe- Statistical analysis sion and angiogenesis (35). In addition, our unbiased approach Descriptive and comparative analyses were performed using revealed a number of genes known to be consistently false- GraphPad v5.03 (GraphPad Software, Inc.). Student t test was positive findings in mutational studies (36). performed for intergroup comparison, association between stain- Finally, we performed gene expression pathway analysis using ings and clinicopathologic features were evaluated using x2 test single sample gene set enrichment analysis (ssGSEA), with the

OF4 Mol Cancer Res; 2018 Molecular Cancer Research

Proﬁling of Recurrent and Metastatic HNSCC Cell Lines

Figure 1. General characteristics of the 15 UM-SCC cell lines. A, Clinical, histologic, and therapeutic characteristics of UM-SCC cell lines. PT, primary tumors; LN, lymph node metastasis; PD, poorly differentiated; P-MD, poorly to moderately differentiated; MD, moderately differentiated; M-WD, moderately to well differentiated; N/A, not available. B, Representative pictures of the 15 UM-SCC cell lines (20 magnification). C, HNSCC molecular subtype score for each of the UM-SCC cell lines. The final HNSCC subtype call is defined as the one with the highest subtype centroid correlation score. D, Nonsynonymous mutations of the UM-SCC samples detected in driver genes taken from the TCGA, known HNSCC driver pathways and in genes found recurrently mutated (see Supplementary Fig. S1). A gene was considered mutated (black square) if at least one variant was found. The number of variants per gene and the percentage of samples with mutated genes are reported as side bars. A full list of variants appears in supplementary Fig. 2A. E, Scatterplot showing mutations summarized at the gene level according to their frequency in UM-SCC cell lines and to their gene length. Genes highlighted in green have a good mutation frequency to gene length ratio. Genes in red are known drivers. F, Heatmap of single sample geneset enrichment analysis (ssGSEA) scores. The upper heatmap includes the gene sets of the MSigDB Hallmark collection that displayed low correlation of the gene profile across cell lines (see Materials and Methods section for details) and thus represent pathways with different activation/enrichment patterns across cell lines. The lower heatmap displays ssGSEA results for pathways that are considered as drivers in HNSCC. www.aacrjournals.org Mol Cancer Res; 2018 OF5

Nisa et al.

Table 1. Comparison of mutation rates in top mutated genes in UM-SCC cells DDR2 mutations in cell lines rendered them exquisitely sensitive UM-SCC Public databases/Publications (%) to dasatinib (16), underlying possible implementation of geno- panel TCGA Stransky Agrawal Pickering mically guided personalized approaches in metastatic/recurrent Gene % (9) (12) (8) (10) HNSCC. KMT2D 66.7 16 11 0 10 TP53 53.3 72 62 69 60 In spite of the small number of samples analyzed in this study, fi DCHS2 46.7 6 1.4 0 0 we identi ed a novel kinase domain PIK3CA mutation in FAT1 46.7 23 12 0 30 UM-SCC-14B (recurrence), absent in UM-SCC-14A (primary) FAT4 46.7 8 11 6 2.5 and UM-SCC-14C (metastasis; Fig. 2A). The allelic frequency of FAT2 46.7 7 11 3 2.5 this mutated variant was 25% (Fig. 2B). Presence of the PIK3CA NOTCH1 40 18 12 13 10 T2930A mutation (resulting in a F977Y protein change) was PIK3C2A 26.7 1.2 1.4 0 0 fi NOTCH3 26.7 4 4 0 0 con rmed by direct sequencing of the region of interest (Fig. NSD1 26.7 12 9 0 8 2C). Next, to assess the impact of this mutation on cell prolifer- PTEN 26.7 2.7 5 0 0 ation, UM-SCC-14A, UM-SCC-14B, and UM-SCC-14C cells were FAT3 26.7 9 12 0 5 exposed for 72 hours to the pan-class I PI3K inhibitor pictilisib. NF1 26.7 2.7 4 0 0 Interestingly, pictilisib was less effective at impairing proliferation PIK3CG 204530in UM-SCC-14B than in the other two matched cell lines (Fig. 2D). PIK3CA 20 18 8 9 15 PIK3C2G 20 2.2 1.4 0 0 Given that the PI3K pathway is one of the most altered in AJUBA 206000HNSCC and PIK3CA among the top mutated oncogenes, RICTOR 13.3 1.8 0 0 0 PI3K signaling is considered a potential therapeutic target in WWC1 13.3 2 0 0 5 HNSCC (9, 33). Our observation suggests existence of drug- PIK3CD 13.3 1.6 1.4 0 0 resistant PIK3CA mutations and strongly stresses the importance DCHS1 13.3 4 8 0 0 of mutationally guided pretherapeutic patient stratification. We HRAS 13.3 6 5 9 13 KRAS 13.3 0.2 1.4 0 0 consulted the TCGA datasets through the cBioPortal for Cancer MAML3 6.7 0.6 0 0 2.5 Genomics (www.cbioportal.org) to assess presence of this muta- MAML1 6.7 1.2 0 0 0 tion. Although this specific mutation was not found, as UM-SCC- LATS1 6.7 1.4 0 0 0 14B was derived from a recurrent tumor following radiotherapy, it MST1 6.70.2000is possible that the F977Y mutation be treatment-induced. TEAD4 6.7 0.8 1.4 0 0 Differentially distributed mutations in more than one cell line NOTE: Data were retrieved from the cBioPortal for Cancer Genomics. were only found in metastatic lines (Supplementary Table S2). More specifically, DCHS2 mutations were found in three metastatic cell lines. DCHS2 encodes for the dachsous homolog 2 hallmark gene signature collections available from the MSigDB protein, also known as cadherin-27 or protocadherin-23, and is molecular signature database. We selected pathways whose cor- related to Hippo pathway signaling (37). In addition, three relation matrices displayed at least one low correlation coefficient different mutations in the dystrophin gene (DMD) were found between two cell lines (r 0.5), to identify variable activation in UM-SCC-14C and UM-SCC-22B. Nevertheless, as both DCHS2 patterns across cell lines (Fig. 1F). Frequently altered HNSCC and DMD are very large genes, the likelihood that identified pathways were equally included. Mirroring previously published mutations are mere passenger events is high, especially in the data, gene expression pathway analysis displayed very heteroge- case of DMD, the largest human gene. Most relevant perhaps was neous degrees of reliance on specific pathways (Fig. 1F). Consid- the A1G variant of TLR8 (Toll-like receptor 8) in 2 metastatic cell ering such patterns of expression is most relevant in preclinical lines (Supplementary Table S2). This variant has been reported to research when considering genomic correlates of responses to activate NFkB signaling, as well as modify cytokine secretion given therapies and allows selection models to address specific and metabolic patterns in a number of infectious diseases, result- research questions in a genomically guided manner (9, 14). ing in more efficient protection against disease. Along the same Altogether, these observations show that the UM-SCC panel is lines, TLR8 is thought to potentially confer enhanced tumor genomically representative of HNSCC. invasiveness by stimulating an NFkB-mediated proinflammatory response (38). Our results warrant further characterization of the Landscapes of UM-SCC cell lines display differentially role of TLR8 in metastatic HNSCC. distributed mutations with potential therapeutic implications Two recent sequencing studies demonstrated that metastatic Comparative gene-copy number variation profiles in UM-SCC and recurrent HNSCC specimens are mutationally similar to their cell lines patient-matched primaries (16, 17). Mirroring such findings, As next step, we performed somatic CNV calling to find whether variant allele frequency (VAF) in the UM-SCC panel displayed specific chromosomal aberrations were recurrently present in high concordance between cell lines derived from matched pairs metastatic or recurrent cell lines (Fig. 3A and B; Supplementary of primaries and recurrences/metastases (Supplementary Fig. S3A Document 2). Using GISTIC2, we found three chromosomal and S3B). regions significantly amplified in metastatic cell lines (4q35.1, Even though mutations exclusively present in metastatic/recur- 9q34.3, and 14q24.1), and only one significantly deleted in rent tumors seem to be overall rare, such differential distribution recurrent cell lines (21q22.3; Supplementary Fig. S4). The ampli- is potentially relevant for implementation of therapeutic fied regions in metastatic cell lines contained genes encoding for approaches. For instance, Hedberg and colleagues (16) found established drivers such as IRF2, FAT1, NOTCH1, MAX, and DDR2 mutations in some metastatic/recurrent specimens, absent MLH3. With the exception of FAT1, all genes were overexpressed in their patient-matched primaries. Ectopic expression of such at least in 3 of 4 metastatic cell lines (Fig. 3A). The third amplified

OF6 Mol Cancer Res; 2018 Molecular Cancer Research

Proﬁling of Recurrent and Metastatic HNSCC Cell Lines

Figure 2. Variant detection and characterization of a novel PIK3CA kinase domain mutation in UM-SCC cell lines. A, Nonsynonymous mutations of the UM-SCC samples detected in driver genes taken from the TCGA, known HNSCC driver pathways and in genes found recurrently mutated (see B). Black squares represent mutation found in both primary tumor and matched recurrent/metastasis counterpart while blue and red squares represent those found only in primary, respectively recurrence/metastasis. B, Allelic frequency of the PIK3CA T2930A mutation according to exome- and RNA-sequencing data. C, Presence of the T2930A (F977Y) PIK3CA mutation in the UM-SCC-14 cell lines was conﬁrmed by Sanger sequencing. D, UM-SCC-14A/B/C cells were treated with increasing concentrations of the PI3K inhibitor pictilisib for 72 hours. After this period, cell viability was assessed (RFU, relative ﬂuorescence units; P values: , P < 0.01; , P < 0.001). www.aacrjournals.org Mol Cancer Res; 2018 OF7

Nisa et al.

Figure 3. Somatic copy number variation in metastatic and recurrent cell lines. Circos representations showing somatic chromosomal copy number gains (in green) or losses (in red) in metastatic (A) and recurrent (B) HNSSC cell lines. Chromosomal regions found to be signiﬁcantly recurrently gained or lost by GISTIC2 are highlighted. The potential driver genes found in these regions are highlighted and their expression in the respective cell lines is shown in inset panels.

OF8 Mol Cancer Res; 2018 Molecular Cancer Research

Proﬁling of Recurrent and Metastatic HNSCC Cell Lines

region (14q24.1) in metastatic cell lines contains, among others, ation and homeostasis (PRRX1, ALPK2, FLG, and IVL; Fig. 5A). A MYC associated factor X gene (MAX)andMLH3,twogenes full list of differentially regulated genes (FDR < 1) is provided in whose mutations have been involved in pheochromocytoma Supplementary Document S3. and colorectal cancer development, respectively (39, 40). The Interestingly, small proline-rich protein 2A (SPRR2A) was the only significantly deleted region in recurrent cell lines only gene differentially expressed in both metastatic (FDR ¼ (21q22.3) contains RUNX1, a tumor suppressor that poten- 0.089) and recurrent (FDR ¼ 0.004) cell lines (Fig. 5A). Down- tially interacts with p53 (41). regulation was confirmed in 3 of 4 cell lines tested by means of Taken together, our findings suggest that, while differences in qPCR (Fig. 5B). SPRR2A is one of the 14 SPRR-family of proteins CNV between primary and matched metastatic/recurrent tumors and acts as a keratinocyte cross-linking protein, ensuring tissue seem to be overall discrete, genes involved in progression and integrity when facing injury and oxidative stress (47). therapeutic resistance may be often altered and most likely play To gain further mechanistic insights into the potential roles of relevant roles in these processes. SPRR2A in metastasis and recurrence, we used Cytoscape to visualize the network of established SPRR2A interactors according Differentially regulated pathway analysis in metastatic and to the STRING database and their fold-change results from the recurrent cell lines highlights the relevance of epithelial-to- transcriptomic analysis. Genes of the SPRR family were coordi- mesenchymal transition as a mechanism of therapeutic nately downregulated both in metastatic and recurrent cell lines resistance when compared with their matched primaries (Fig. 5C and D). To characterize differentially regulated pathways in metastatic Other genes involved in epithelial homeostasis, such as POU2F2 and recurrent cell lines, we performed pathway enrichment anal- and CNFN, were similarly downregulated. ysis using the gene likelihood ratios obtained by differential gene The oncogenic role of SPRR2A is complex. Specht and collea- analysis as weight factors (Fig. 4A). Among the frequently gues (48) demonstrated that stable SPRR2A transfection of cho- enriched pathways in recurrent tumors, we identified several langiocarcinoma cell lines injected intrasplenically led to inability proinflammatory signatures (IFNa/g response, IL6/JAK/STAT3 to form liver metastases, however increasing local tumor inva- and IL2/STAT5 signaling, and TNFa signaling via NFkB). Altera- siveness. Conversely, cell lines with downregulated SPRR2A read- tions of proinflammatory signaling are frequent in HNSCC ily developed liver metastases. From a mechanistic perspective, tumors and cell lines, and have been linked to potential thera- SPRR2A binds SH3 domain–containing kinases, including Src, peutic resistance (9, 15, 42). Another process involved in resis- Yes1, and Abl (47). In our transcriptomic dataset, SPRR2A down- tance to chemotherapeutic agents, xenobiotic metabolism, was regulation did not seem to have an impact at the transcriptional equally upregulated in 3 of 4 recurrent cell lines. Metabolic level on Src-related kinases (Fig. 5C and D). However, given that reprogramming is indeed thought to confer a selection advantage interactions between SPRR2A and Src-related kinases occur at the to tumor cells upon genotoxic therapy (43). protein-protein level, it is anticipated that SPRR2A downregula- Most prominently, epithelial-to-mesenchymal transition tion may have an impact on prosurvival signaling mediated by (EMT) was upregulated in all recurrent cell lines, while no these kinases. consistently enriched pathways could be identified in metastatic While a limited number of studies have addressed the role of cell lines (Fig. 4A). In line with the observed EMT upregulation, we SPRR2A in hepatobiliary malignancies, the relevance of SPRR2A found that HNSCC subtype score progression in recurrent but not in HNSCC has not been explored. Thus, we proceeded to assess metastatic cell lines is characterized by an increase in the mesen- expression of SPRR2A by qPCR in 51 patient-matched sets of chymal score (Fig. 4B). To further assess the differences in EMT's HNSCC primary tumors and lymph node metastases, demon- enrichment in metastases and recurrent cell lines, we performed strating significant downregulation of SPRR2A mRNA levels in differential expression analysis and compared the results for genes metastatic specimens (Fig. 6A). Next, to explore the potential in the EMT group versus all genes in metastatic and recurrent cell clinical relevance of SPRR2A, protein expression levels were lines (Fig. 4C). In contrast to recurrent cell lines, EMT genes with P assessed by IHC in 426 primary tumors from 181 patients, 530 values < 10 3 were absent in metastatic cell lines. Focusing on 14 lymph node metastases from 236 patients, and 295 normal EMT genes with an FDR < 0.25 in the differential expression tissues (mainly salivary gland tissue) from 237 patients. SPRR2A analysis, we found these genes to be well connected and consis- was highly expressed in 49.53% of primaries and 26.98% of tently upregulated in recurrent cell lines (Fig. 4D). lymph node metastases, while its expression was not detected in The consistent upregulation of EMT in recurrent but not met- normal tissues (Fig. 6B). astatic cell lines is congruent with recent findings in pancreatic and Matched sets of primary tumor, lymph node metastasis, and mammary carcinomas, supporting the notion that EMT may have normal tissue, as well as full clinical and survival data were a more important role in therapeutic resistance than in invasive- available for 147 patients. Patients were classified according to ness (44–46). SPRR2A expression (low: scores 0 and 1; vs. high: scores 2 and 3), in primary tumors and lymph node metastases (Supplementary Comparative expression profiles of UM-SCC cell lines unveil a Table S3). In this set of matched samples, SPRR2A was expressed potential role of SPRR2A as effector of therapeutic resistance according to 4 different patterns (Fig. 6C). In 44.9% of the cases, and potential prognostic marker in regionally invasive HNSCC SPRR2A was highly expressed in primary tumors only, in 14.29% The differential expression analysis results were further used to in lymph node metastases only, and in 17.01% in both primary identify individual genes involved in metastasis and/or recurrence and matched lymph node metastases. SPRR2A was not detected in in the UM-SCC panel. The ten most differentially expressed genes 23.81% of the cases (Fig. 6C). SPRR2A expression was not in metastatic and/or recurrent UM-SCC cell lines include ECM associated with tumoral stage, site, or development of distant components (SRGN), genes involved in extracellular matrix metastases. SPRR2A low expression in primary tumors was sig- (ECM) remodeling (MMP2 and KLK5), or epidermal differenti- nificantly associated with local recurrences (P < 0.01) in the

www.aacrjournals.org Mol Cancer Res; 2018 OF9

Nisa et al.

Figure 4. Pathway analysis comparisons in UM-SCC cell lines. A, Preranked GSEA using the likelihood ratio statistics of the differential expression testing as weight vector. The MSigDB hallmark geneset collection was used and ssGSEA score were used to draw the heatmap. Only pathways with a FDR below 0.25 in the preranked GSEA analysis were kept for plotting. B, Progression of the HNSCC subtype score from primary tumor cell lines to their recurrent/metastatic counterparts. The progression score is computed as the difference between the recurrence/metastasis score and the primary tumor score for each cell line. C, Fraction of EMT versus all genes in metastatic and recurrent cell lines. D, Interaction networks of differentially expressed EMT genes with an FDR < 0.25 and their estimated logarithmic fold change values.

OF10 Mol Cancer Res; 2018 Molecular Cancer Research

Proﬁling of Recurrent and Metastatic HNSCC Cell Lines

Figure 5. Differentially expressed genes and relevance of SPRR2A expression in metastatic and recurrent HNSCC cell lines. A, Expression of the ten most differentially expressed genes (all with FDR < 0.25) separately in recurrent and metastatic cell lines versus their respective primaries. Only genes that were consistently up- or downregulated in all four cell lines are shown. B, SPRR2A mRNA levels in metastatic cell lines. C and D, SPRR2A interaction networks based on transcriptomic data in metastatic and recurrent cell lines. Link width is proportional to the combined evidence score reported by STRING, and node color indicates the estimated logarithmic fold change in the primary-metastasis and primary-recurrent comparisons, respectively.

www.aacrjournals.org Mol Cancer Res; 2018 OF11

Nisa et al.

Figure 6. Patterns of SPRR2A expression and impact on survival in regionally metastatic HNSCC tissue samples. A, Comparison of SPRR2A mRNA levels in 51 patient- matched primary tumors versus lymph node metastases. B, SPRR2A IHC staining intensity in HNSCC samples. C, SPRR2A staining patterns in patient-matched primaries, lymph node (LN) metastases, and normal tissue. D, Kaplan–Meier plots illustrating regional disease-free survival as a function of SPRR2A expression. E, Migration assay with UM-SCC-10A/B and UM-SCC-22A/B cell lines. Cells were allowed to migrate for 72 hours, after which the invaded area was measured. F, Colony-forming assays. RER, radiation enhancement ratio.

OF12 Mol Cancer Res; 2018 Molecular Cancer Research

Proﬁling of Recurrent and Metastatic HNSCC Cell Lines

univariate time-independent analysis. SPRR2A low expression in background. For instance, it was recently demonstrated that lymph node metastases was significantly more prevalent in exclusive presence of DDR2 mutations in lymph node metastases patients older than 60 years old (P ¼ 0.02), while SPRR2A high resulted in increased sensitization of HNSCC preclinical models expression in lymph node metastases was more prevalent in to dasatinib (16). In this study and hinting at a potential resistance patients that experienced regional recurrences (P ¼ 0.04) and mechanism, we identified a novel kinase domain PIK3CA muta- those who got concomitant systemic therapy (P ¼ 0.04) (Sup- tion (F977Y) that impairs responses to PI3K inhibitors in cell lines plementary Table S3). (Fig. 2D). This mutation was found in a cell line derived from a In the univariate survival analysis, only high SPRR2A expres- recurrent HNSCC specimen and is possibly treatment induced. sion in lymph node metastases was significantly associated with Thus, given the potential relevance of the PI3K pathway as a reduced regional recurrence-free survival (RRFS; HR ¼ 2.80, 95% therapeutic target along with the high rate of alterations along this CI: 1.16–6.79, P ¼ 0.02; Fig. 6D; Supplementary Table S4). In the pathway, elucidating the implications of such alterations in terms multivariate analysis, nonoropharyngeal location of the primary of treatment responses is of utmost importance prior to imple- tumor and high SPRR2A expression in lymph node metastases mentation of PI3K pathway inhibitors in recurrent HNSCC. were both independent predictors of poor regional recurrence- Mutations of components of the Hippo pathway were found in free survival (Supplementary Table S4). Our multivariate models all cell lines (Figs. 1D and 2A). Aberrant Hippo pathway activation could not identify significant predictors of overall survival or local has been reported in several types of human malignancies, recurrence-free survival. Surgical resection of the primary tumor including lung, colorectal, ovarian, and liver cancer (49). Most and oropharyngeal primary site predicted improved distant prevalently mutated were DCHS2 and FAT isoforms 1 to 4. FAT1 is metastasis-free survival (Supplementary Table S4). one of the most commonly mutated genes in HNSCC and is We next performed similar analyses using the TCGA datasets thought to act upstream of the Hippo pathway mediating cross- available for 498 patients and found no correlation between talk with the Wnt pathway, thus contributing to acquisition of SPRR2A RNA levels and survival (Supplementary Fig. S5). This invasive features (9, 49). Nevertheless, the oncogenic significance observation is not surprising, as data is derived from primary of FAT1, DCHS1, and DCHS2 mutations is not fully clear, as these tumors and protein levels are not available in the TCGA cohorts. genes are large and high prevalence of mutations in these genes Because the clinical data seems to suggest that high expression may be due, in part, to their size. Moreover, while FAT and DCHS of SPRR2A may contribute to radioresistance and low expression are well established Hippo pathway modulators in Drosophila may facilitate metastasis, we performed two supplementary melanogaster, their precise role in mammalians are not fully clear in vitro approaches. For this purpose, UM-SCC-10A/B and UM- (49). Besides mutations, we found hemicentin 1 (HMCN1), an SCC-22A/B cells were plated and allowed to migrate for 72 hours. increasingly recognized key upstream regulator of Hippo pathway The UM-SCC-10 pair migrated faster than UM-SCC-22 cells. core kinases MST1/MST2 and LATS1/LATS2, to be consistently Nevertheless, for both pairs of cell lines, the SPRR2A low expres- upregulated in metastatic cell lines (Fig. 5A). Given the marked sing 10B and 22B migrated significantly faster than 10A and 22A, differences in Hippo pathway members' mutations in primary respectively (Fig. 6E). To assess radiosensitivity, cells were irradi- versus metastatic and recurrent HNSCCs, the impact of Hippo ated with increasing doses (0, 2, and 4 Gy), and left to proliferate signaling in such settings warrant further investigation. for 7 days. The SPRR2A-high line UM-SCC-10A was significantly CNV profiles were highly preserved in metastatic and recurrent more resistant to irradiation than its matched SPRR2A-low cell versus primary tumors, with only 3 significant peaks in metastatic line UM-SCC-10B. We did not find significant differences between and 1 in recurrent cell lines (Fig. 3A and B). Such discrete UM-SCC-22A and 22B (Fig. 6F). differences are consistent with previously findings (16). Never- theless, altered regions contained relevant drivers or tumor sup- pressors and underline potentially relevant mechanisms in devel- Discussion opment of metastases or recurrences. In this study, we provide the most comprehensive molecular Expression data revealed that transcriptomic differences characterization of a panel of cell lines derived from metastatic between primary tumors and their matched metastases were more and recurrent HNSCC specimens to date with the aim to provide a marked than in the case of recurrences. Among the frequently suitable tool for preclinical research in the metastatic and recur- upregulated pathways in recurrent cell lines were pathways rent setting of HNSCC. Molecular portraits of HNSCC, including involved in the inflammatory response and, most significantly, comparisons between HPV-positive and HPV-negative tumors, as EMT (Fig. 4A). While EMT is considered a hallmark process of well as primaries versus lymph node metastases and recurrent metastasizing tumors, emergent evidence links EMT to treatment tumors, have emerged in recent years (8, 9, 11, 12, 16, 17). resistance (45). Indeed, recent studies in mammary and pancre- Implementation of novel findings in clinical practice, however, atic tumor models convincingly show that impairment of EMT by requires thorough preclinical validation in well-characterized blockade of either ZEB1 and ZEB2 or Snail and Twist does not lead models (14). to decreased metastatic ability (44, 46). Importantly however, Our current results show that the UM-SCC panel is represen- inability to undergo EMT resulted in increased sensitivity to tative of the most common and most characteristic molecular cyclophosphamide and gemcitabine in the same models (44, 46). alterations found in HNSCC. Indeed, mutational profiling of UM- When considering individual cell lines, EMT signatures fea- SCC cells revealed typical alterations affecting TP53, PI3K, and the tured most prominently in UM-SCC-74A, UM-SCC-74B, and NOTCH pathways. A major aspect of interest was the observation UM-SCC-11B, all derived from patients previously managed of mutations exclusively present in metastases or recurrences, and with chemoradiotherapy and with phenotypical characteristics the subsequent possibility to study differential sensitivities to of EMT (Fig. 1A and B; ref. 18). diverse management approaches in primaries versus their The heterogeneity in terms of pathway analysis seen in the UM- matched metastases and/or recurrences based on their mutational SCC panel reflects the heterogeneity found in actual tumors and

www.aacrjournals.org Mol Cancer Res; 2018 OF13

Nisa et al.

needs to be considered when exploring novel therapeutic therapy and performing prospective validation of SPRR2A as a approaches (9, 14, 30, 32). Relevant to this point, the character- biomarker for risk stratification. ization provided in UM-SCC panel allows investigation of Limitations of this study include the absence of HPV-positive mechanisms and target validation taking into consideration the cell lines and limited amount of cell lines included. However, it molecular background of such cell lines. An important implica- should be noted that availability of metastatic/recurrent and tion of such approach is the possibility to unveil heterogeneity in primary tumor patient-matched HNSCC cell lines, especially treatment responses and potentially identifying predictive HPV-positive ones, remains limited. biomarkers. fi Finally, the nding that most clearly illustrates the discovery Disclosure of Potential Conflicts of Interest and validation potential of the UM-SCC panel concerns the fi O. Elicin is a consultant/advisory board member for AstraZeneca, Merck, and identi cation of SPRR2A as potential effector of therapeutic Merck Serono. T.E. Carey has ownership interest (including stock, patents, etc.) resistance and biomarker in HNSCC. In cholangiocarcinoma and in licensed cell lines to Millipore EMD. No potential conflicts of interest were liver cancer models SPRR2A is activated by STAT-3 both in biliary disclosed by the other authors. epithelial cells and cholangiocarcinoma, coorchestrating EMT through interaction with ZEB-1 (47, 50). In addition, SPRR2A Authors' Contributions reduces p53 acetylation by impairing p300-p53 interactions, Conception and design: L. Nisa, D. Barras, M. Medova, D.M. Aebersold, leading to inhibition of p53 transcriptional targets (50). More- R. Giger, M. Delorenzi, Y. Zimmer over, SPRR2A acts as an SH3 domain ligand, hence promoting Development of methodology: L. Nisa, D. Barras, M. S. Dettmer, R. Giger, L. Ho protection against oxidative stress and DNA damage (47). In our Acquisition of data (provided animals, acquired and managed patients, FFPE validation cohort, most primary tumors (61.91%) highly provided facilities, etc.): L. Nisa, M. Paliakova, J. Koch, B. Bojaxhiu, O. Elicin,¸ expressed SPRR2A, while only a minority of lymph node metas- M. S. Dettmer, R. Giger, M.D. Caversaccio, T.E. Carey, T.A. McKee fl Analysis and interpretation of data (e.g., statistical analysis, biostatistics, tases did (31.30%), perhaps re ecting necessary SPRR2A down- computational analysis): L. Nisa, D. Barras, M. Medova, M. Medo, regulation in certain clones within the primary tumor to give rise M. Paliakova, J. Koch, P. Angelino, R. Giger, M. Delorenzi , Y. Zimmer to metastases (48). Importantly, however, our results emphasize Writing, review, and/or revision of the manuscript: L. Nisa, D. Barras, an additional aspect on the role of SPRR2A in therapeutic resis- M. Medova, D.M. Aebersold, O. Elicin,¸ R. Giger, U. Borner, M.D. Caversaccio, tance. Indeed, as high SPRR2A expression in lymph node metas- T.E. Carey, T.A. McKee, M. Delorenzi, Y. Zimmer tases is an independent risk factor for development of regional Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Nisa, D. Barras, M. Medova, J. Koch, recurrence after therapy, it could be hypothesized that after B. Bojaxhiu, O. Elicin,¸ M. S. Dettmer, R. Giger, U. Borner, M.D. Caversaccio, metastatic colonization SPRR2A expression may be restored in M. Delorenzi neoplastic cells. In this context, SPRR2A would contribute to the Study supervision: L. Nisa, D. Barras, M. Medova, R. Giger, M. Delorenzi, stress response upon radiotherapy/chemoradiation by virtue of its Y. Zimmer antioxidative stress ability and through protein–protein interactions with multiple prosurvival signaling molecules through its Acknowledgments SH3 domain ligand activity (including Src, Yes1, and others) (47). This work was supported by Swiss National Science Foundation In vitro migration and radiosensitization assays seem to suggest (grant no. 31003A_156816, to Y. Zimmer) and by a Bernische Krebsliga grant that indeed high SPRR2A expression may contribute to radio- (to M. Medova). The authors gratefully acknowledge the help of Sabina Gallati, resistance, while low SPRR2A expression may stimulate invasive PhD and Nijas Aliu, MSc, from the Division of Human Genetics, Department of features. Pediatrics (Inselspital, Bern, Switzerland) for establishing karyotypes. Factors that determine SPRR2A reestablished expression in The costs of publication of this article were defrayed in part by the payment of metastases are yet to be fully investigated and may provide novel page charges. This article must therefore be hereby marked advertisement in mechanistic insights with potential therapeutic relevance, as accordance with 18 U.S.C. Section 1734 solely to indicate this fact. SPRR2A cannot be directly targeted. In addition, it would be of high interest to determine SPRR2A Received January 17, 2018; revised April 28, 2018; accepted August 7, 2018; expression levels in recurrent lymph node metastases after radio- published first August 14, 2018.

References 1. Mamelle G, Pampurik J, Luboinski B, Lancar R, Lusinchi A, Bosq J. Lymph 6. Jayaram SC, Muzaffar SJ, Ahmed I, Dhanda J, Paleri V, Mehanna H. Efficacy, node prognostic factors in head and neck squamous cell carcinomas. Am J outcomes, and complication rates of different surgical and nonsurgical Surg 1994;168:494–8. treatment modalities for recurrent/residual oropharyngeal carcinoma: a 2. Leemans CR, Braakhuis BJ, Brakenhoff RH. The molecular biology of head systematic review and meta-analysis. Head Neck 2016;38:1855–61. and neck cancer. Nat Rev Cancer 2011;11:9–22. 7. Tan HK, Giger R, Auperin A, Bourhis J, Janot F, Temam S. Salvage surgery 3. Elicin O, Nisa L, Dal Pra A, Bojaxhiu B, Caversaccio M, Schmucking M, et al. after concomitant chemoradiation in head and neck squamous cell carci- Up-front neck dissection followed by definitive (chemo)-radiotherapy in nomas - stratification for postsalvage survival. Head Neck 2010;32:139–47. head and neck squamous cell carcinoma: rationale, complications, toxicity 8. Agrawal N, Frederick MJ, Pickering CR, Bettegowda C, Chang K, Li RJ, et al. rates, and oncological outcomes - A systematic review. Radiother Oncol Exome sequencing of head and neck squamous cell carcinoma reveals 2016;119:185–93. inactivating mutations in NOTCH1. Science 2011;333:1154–7. 4. Young ER, Diakos E, Khalid-Raja M, Mehanna H. Resection of subsequent 9. Cancer Genome Atlas Network. Comprehensive genomic characterization pulmonary metastases from treated head and neck squamous cell carcinoma: of head and neck squamous cell carcinomas. Nature 2015;517:576–82. systematic review and meta-analysis. Clin Otolaryngol 2015;40:208–18. 10. Pickering CR, Zhang J, Yoo SY, Bengtsson L, Moorthy S, Neskey DM, et al. 5. Crozier E, Sumer BD. Head and neck cancer. Med Clin North Am 2010;94: Integrative genomic characterization of oral squamous cell carcinoma 1031–46. identifies frequent somatic drivers. Cancer Discov 2013;3:770–81.

OF14 Mol Cancer Res; 2018 Molecular Cancer Research

Proﬁling of Recurrent and Metastatic HNSCC Cell Lines

11. Seiwert TY, Zuo Z, Keck MK, Khattri A, Pedamallu CS, Stricker T, et al. chromosomal gain and loss of canonical cancer genes. PLoS One 2013;8: Integrative and comparative genomic analysis of HPV-positive and HPV- e56823. negative head and neck squamous cell carcinomas. Clin Cancer Res 31. Fertil B, Dertinger H, Courdi A, Malaise EP. Mean inactivation dose: a useful 2015;21:632–41. concept for intercomparison of human cell survival curves. 1984. Radiat 12. Stransky N, Egloff AM, Tward AD, Kostic AD, Cibulskis K, Sivachenko A, Res 2012;178:AV237–43. et al. The mutational landscape of head and neck squamous cell carcinoma. 32. Chung CH, Parker JS, Karaca G, Wu J, Funkhouser WK, Moore D, et al. Science 2011;333:1157–60. Molecular classification of head and neck squamous cell carcinomas using 13. Haibe-Kains B, El-Hachem N, Birkbak NJ, Jin AC, Beck AH, Aerts HJ, et al. patterns of gene expression. Cancer Cell 2004;5:489–500. Inconsistency in large pharmacogenomic studies. Nature 2013;504: 33. Lui VW, Hedberg ML, Li H, Vangara BS, Pendleton K, Zeng Y, et al. Frequent 389–93. mutation of the PI3K pathway in head and neck cancer defines predictive 14. Barretina J, Caponigro G, Stransky N, Venkatesan K, Margolin AA, Kim S, biomarkers. Cancer Discov 2013;3:761–9. et al. The Cancer Cell Line Encyclopedia enables predictive modelling of 34. Nej K, Bartsch DK, Sina-Frey M, Rieder H, Hahn SA, Lubinski J. The NOD2 anticancer drug sensitivity. Nature 2012;483:603–7. 3020insC mutation and the risk of familial pancreatic cancer? Hered 15. Martin D, Abba MC, Molinolo AA, Vitale-Cross L, Wang Z, Zaida M, et al. Cancer Clin Pract 2004;2:149–50. The head and neck cancer cell oncogenome: a platform for the develop- 35. Wagstaff L, Kelwick R, Decock J, Edwards DR. The roles of ADAMTS ment of precision molecular therapies. Oncotarget 2014;5:8906–23. metalloproteinases in tumorigenesis and metastasis. Front Biosci 2011; 16. Hedberg ML, Goh G, Chiosea SI, Bauman JE, Freilino ML, Zeng Y, et al. 16:1861–72. Genetic landscape of metastatic and recurrent head and neck squamous cell 36. Lawrence MS, Stojanov P, Polak P, Kryukov GV, Cibulskis K, Sivachenko A, carcinoma. J Clin Invest 2016;126:169–80. et al. Mutational heterogeneity in cancer and the search for new cancer- 17. Morris LG, Chandramohan R, West L, Zehir A, Chakravarty D, Pfister DG, associated genes. Nature 2013;499:214–8. et al. The molecular landscape of recurrent and metastatic head and neck 37. An CH, Je EM, Yoo NJ, Lee SH. Frameshift mutations of cadherin genes cancers: insights from a precision oncology sequencing platform. JAMA DCHS2, CDH10 and CDH24 genes in gastric and colorectal cancers with Oncol 2017;3:244–55 high microsatellite instability. Pathol Oncol Res 2015;21:181–5. 18. Brenner JC, Graham MP, Kumar B, Saunders LM, Kupfer R, Lyons RH, et al. 38. Schon MP, Schon M. TLR7 and TLR8 as targets in cancer therapy. Oncogene Genotyping of 73 UM-SCC head and neck squamous cell carcinoma cell 2008;27:190–9. lines. Head Neck 2010;32:417–26. 39. Burnichon N, Cascon A, Schiavi F, Morales NP, Comino-Mendez I, Abermil 19. Martin M. Cutadapt removes adapter sequences from high-throughput N, et al. MAX mutations cause hereditary and sporadic pheochromocytoma sequencing reads. EMBnet Journal 2011;17:10–2. and paraganglioma. Clin Cancer Res 2012;18:2828–37. 20. Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, et al. STAR: 40. Korhonen MK, Vuorenmaa E, Nystrom M. The first functional study of ultrafast universal RNA-seq aligner. Bioinformatics 2013;29:15–21. MLH3 mutations found in cancer patients. Genes Chromosomes Cancer 21. Liao Y, Smyth GK, Shi W. featureCounts: an efficient general purpose 2008;47:803–9. program for assigning sequence reads to genomic features. Bioinformatics 41. Ito Y, Bae SC, Chuang LS. The RUNX family: developmental regulators in 2014;30:923–30. cancer. Nat Rev Cancer 2015;15:81–95. 22. DePristo MA, Banks E, Poplin R, Garimella KV, Maguire JR, Hartl C, et al. A 42. Freudlsperger C, Bian Y, Contag Wise S, Burnett J, Coupar J, Yang X, et al. framework for variation discovery and genotyping using next-generation TGF-beta and NF-kappaB signal pathway cross-talk is mediated through DNA sequencing data. Nat Genet 2011;43:491–8. TAK1 and SMAD7 in a subset of head and neck cancers. Oncogene 2013; 23. Cibulskis K, Lawrence MS, Carter SL, Sivachenko A, Jaffe D, Sougnez C, et al. 32:1549–59. Sensitive detection of somatic point mutations in impure and heteroge- 43. Tamasi V, Monostory K, Prough RA, Falus A. Role of xenobiotic metab- neous cancer samples. Nat Biotechnol 2013;31:213–9. olism in cancer: involvement of transcriptional and miRNA regulation of 24. Koboldt DC, Zhang Q, Larson DE, Shen D, McLellan MD, Lin L, et al. P450s. Cell Mol Life Sci 2011;68:1131–46. VarScan 2: somatic mutation and copy number alteration discovery in 44. Fischer KR, Durrans A, Lee S, Sheng J, Li F, Wong ST, et al. Epithelial-to- cancer by exome sequencing. Genome Res 2012;22:568–76. mesenchymal transition is not required for lung metastasis but contributes 25. Wang K, Li M, Hakonarson H. ANNOVAR: functional annotation of genetic to chemoresistance. Nature 2015;527:472–6. variants from high-throughput sequencing data. Nucleic Acids Res 2010; 45. Seton-Rogers S. Epithelial-mesenchymal transition: Untangling EMT's 38:e164. functions. Nat Rev Cancer 2016;16:1. 26. Rubio-Perez C, Tamborero D, Schroeder MP, Antolin AA, Deu-Pons J, 46. Zheng X, Carstens JL, Kim J, Scheible M, Kaye J, Sugimoto H, et al. Perez-Llamas C, et al. In silico prescription of anticancer drugs to cohorts of Epithelial-to-mesenchymal transition is dispensable for metastasis but 28 tumor types reveals targeting opportunities. Cancer Cell 2015;27: induces chemoresistance in pancreatic cancer. Nature 2015;527:525–30. 382–96. 47. Demetris AJ, Specht S, Nozaki I, Lunz JG 3rd, Stolz DB, Murase N, et al. 27. Anjum S, Morganella S, D'Angelo F, Iavarone A, Ceccarelli M. VEGAWES: Small proline-rich proteins (SPRR) function as SH3 domain ligands, variational segmentation on whole exome sequencing for copy number increase resistance to injury and are associated with epithelial-mesenchy- detection. BMC Bioinformatics 2015;16:315. mal transition (EMT) in cholangiocytes. J Hepatol 2008;48:276–88. 28. Krzywinski M, Schein J, Birol I, Connors J, Gascoyne R, Horsman D, et al. 48. Specht S, Isse K, Nozaki I, Lunz JG 3rd, Demetris AJ. SPRR2A expression in Circos: an information aesthetic for comparative genomics. Genome Res cholangiocarcinoma increases local tumor invasiveness but prevents 2009;19:1639–45. metastasis. Clin Exp Metastasis 2013;30:877–90. 29. Robinson MD, McCarthy DJ, Smyth GK. edgeR: a Bioconductor package 49. Harvey KF, Zhang X, Thomas DM. The Hippo pathway and human cancer. for differential expression analysis of digital gene expression data. Nat Rev Cancer 2013;13:246–57. Bioinformatics 2010;26:139–40. 50. Mizuguchi Y, Specht S, Lunz JG 3rd, Isse K, Corbitt N, Takizawa T, et al. 30. Walter V, Yin X, Wilkerson MD, Cabanski CR, Zhao N, Du Y, et al. SPRR2A enhances p53 deacetylation through HDAC1 and down regulates Molecular subtypes in head and neck cancer exhibit distinct patterns of p21 promoter activity. BMC Mol Biol 2012;13:20.

www.aacrjournals.org Mol Cancer Res; 2018 OF15

Comprehensive Genomic Profiling of Patient-matched Head and Neck Cancer Cells: A Preclinical Pipeline for Metastatic and Recurrent Disease

Lluís Nisa, David Barras, Michaela Medová, et al.

Mol Cancer Res Published OnlineFirst August 14, 2018.

Updated version Access the most recent version of this article at: doi:10.1158/1541-7786.MCR-18-0056

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://mcr.aacrjournals.org/content/early/2018/09/26/1541-7786.MCR-18-0056. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.