A total of 48 samples were obtained from HNSCC patients and included in the analysis ( Table 1 ) and clinical and pathological data tumors were rigorously classified as hypopharyngeal or laryngeal. Blood samples were collected from 13/48 patients prior to chemo-/radiotherapy or surgical intervention and those blood (normal)-tumor pairs were subject to exome sequencing. All of the diagnostic protocols were reassessed in order to standardize them with the 8^th Edition of the American Joint Committee on Cancer Staging Manual (16). In dubious cases, histologic slides were re-examined. As pathologic TNM classification is applicable to postoperative material, we were not able to perform staging in 6 cases in which only small biopsies were available. Histological types were consistent with the 4th edition of the WHO classification of Head and Neck Tumors (17), however in 5 cases a paucity of material did not allow to distinguish between keratinizing and nonkeratinizing subtypes of conventional squamous cell carcinomas (SCC). In few cases material was not originally properly designated and the samples were no longer available (those cases are marked as not determined in the Table 1 ).

DNA was isolated using either Gentra Puregene kit (Qiagen) from fresh or snap-frozen tumor tissues and cell line pellets or with DNA Blood Mini Kit (Qiagen) from patients’ blood samples. DNA was dissolved in Buffer AE (Qiagen) and stored at +4°C. A single tumor sample was isolated from formalin-fixed paraffin-embedded tissue using E.Z.N.A. FFPE DNA Kit (Omega Bio-tek).

HPV infection status was assessed using PCR with MY11/09 primers as described previously, with modifications (18). A 20µl PCR mixture contained the following components: 1U of HotStartTaq Plus polymerase (Qiagen), 2µl 10x CoralLoad Buffer, 2×0.4µl MY09/MY11 primers (10µM), 0.4µl dNTPs (10mM), 0.8µl DNA (200ng) and 15.88 µl water. Cycling conditions were set as follows: 35 cycles (60 s at 94°C, 60 s at 55°C, 60 s at 72°C) with initial denaturation for 5 min at 95°C and final extension for 7 min at 72°C.

DNA sequences of TERT promoter (pTERT) were obtained in two reactions. The PCR reaction was performed with two primers: forward 5’- CACCCGTCCTGCCCCTTCACCTT and reverse 5’- GGCTTCCCACGTGCGCAGCAGGA (10 µM) (19) and the KAPA HiFi HotStart Ready Mix (KAPA Biosystems). The reaction mixture (20 µl) contained the following components: water - 7.8 µl, HotStart Mix - 10 µl, primers (forward and reverse) - 2×0.6 µl (10µM) and DNA - 1 µl (approx. 100-300ng). PCR was performed for 32 cycles (20 s at 95°C, 30 s at 72°C) with initial denaturation for 3 min at 95°C and final extension for 1 min at 72°C. The reaction resulted in formation of a 147 bp product. Subsequently, amplicon was purified using VAHTS DNA Clean Beads (Vazyme) (with a 1:1.8 DNA to beads ratio) and labeled using BigDye Terminator v3.1 Cycle Sequencing kit (Thermo Fischer Scientific, USA) and primers used in the first reaction. The reaction mixture (10 µl) contained the following components: BigDye - 0.4 µl, Sequencing Buffer - 3.6 µl, water - 2 µl, primer (forward or reverse, 1 µM) - 2 µl and 2 µl of the purified amplicon. PCR was performed in a thermocycler for 55 cycles (10 s at 95°C, 15 s at 50°C, 90 s at 60°C) with an initial denaturation for 5 min at 95°C and a final extension for 5 min at 60°C. All PCR amplicons were analyzed on 1-2% agarose gels stained with ethidium bromide (Sigma-Aldrich, USA).

Finally, 10 µl of labeled product was purified using beads by incubating the product for 5 min with 10 µl beads and with 42 µl of 90% ethanol at room temperature, two washes with 90% ethanol for 30 seconds on a magnet and elution with 10 µl of water. The product was then processed for sequencing on a Genetic Analyzer 3500 (Applied Biosystems, Thermo Fischer, USA). Chromatograms were analyzed using FinchTV software.

The whole process of library construction and enrichment was carried out using SeqCap EZ chemistry (Roche NimbleGen, USA), according to the SeqCap EZ Library SR User’s Guide v.4.2, with minor modifications, briefly described below. DNA from patient samples and FaDu cell lines was converted into DNA fragment libraries using KAPA Lib Prep kit (Kapa Biosystems, USA). 100–1000ng of DNA was sheared on Covaris M220 for 225s (175s for partially fragmented FFPE DNA) and used as an input for library construction, followed by end-repair, adenylation, and adapter ligation steps. The resulting libraries were then subject to dual-sided SPRI size-selection method and PCR amplification for 3-7 cycles, depending on DNA input. Subsequently, libraries were mixed into 8- to 24-plex pools, hybridized to SeqCap EZ capture probes, and reamplified for 9–10 PCR cycles.

For custom sequence capture SeqCap EZ custom probe designs targeting 7 Mb and 10Mb were used and for exome sequencing SeqCapEZ Exome v2 or MedExome (Roche NimbleGen) probe sets were employed. Additionally, TruSeq library preparation and exome enrichment kits (Illumina, USA) were used according to user guides for sequencing of three normal-tumor sample pairs. Regions not overlapping between custom/exome panels were excluded from analyses when needed. A list of probe designs used for individual samples is provided in Supplementary Data 1.2 .

DNA and DNA library concentrations were measured on NanoDrop (Thermo Fisher Scientific) and Qubit fluorometer using dsDNA High Sensitivity kit (Thermo Fisher Scientific). Quality of DNA and DNA library fragment size ranges were assessed on 0.7–2% agarose gels or 2100 Bioanalyser instrument (Agilent Technologies, USA) when needed.

All libraries were sequenced on Illumina HiSeq 1500 instrument using 2 x 100 bp reads. Tumor samples analyzed with custom panels were sequenced to reach mean coverages in a range of 49.2–234x (median 154.93x) with % bp @ 20x in a range of 77.7–97.2% (median 95.6%) while tumor exome samples were sequenced to reach mean coverages in a range of 31.5– 145.2x (median 112.3x) and % bp @ 20x in a range of 71.4–99.2% (median 92.2%). Blood samples were sequenced to reach mean coverages in a range of 52.6–166.9x (median 93.4x) and % bp @ 20x in a range of 80.1–98.5% (median 89.8%).

Raw sequencing data was processed according to Broad Institute recommendations. Variant discovery included the following steps: quality control of raw fastq, adapter trimming and low-quality reads removal using Trimmomatic (20), read mapping to hg19 using BWA-MEM (21), duplication removal, local realignment and quality recalibration using GATK and Picard and variant calling using UnifiedGenotyper, and HaplotypeCaller. Exome data was additionally analyzed using Mutect-2 to identify somatic single nucleotide variants (SNV) and insertions-deletions (indels) through direct comparison of germline and tumor samples (22).

Variants were filtered using public (NHLBI ESP (23), gnomAD (24) and internal databases in order to remove common genetic variation. CADD (25), PolyPhen2 (26), CHASM (27), SIFT (28), FATHMM (29) and Mutation Taster (30) predictions were used to identify possible protein-damaging missense alterations. All variants were also manually curated to avoid including sequencing artifacts and ClinVar (31), Varsome (32) and COSMIC (33) databases were also used to aid final variant classification. Complete lists of classified variants are available in Supplementary Data 3 .

Copy-number calling was performed by sequencing coverage analysis using CNVkit v0.9.5 (34). CNVkit was run with default settings except for 400bp bin size limit. Groups of normal samples were used to create reference coverage models across predefined targets for each custom/exome capture. Two samples were recentered due to clear deviation of basal copy number. Additionally, a GISTIC 2.0.23 (35) analysis was performed on segmented CNV data acquired for genomic regions targeted by all custom captures ( Supplementary Data 1.3 ). GISTIC was run with default parameters except for 5000bp pseudo-markers spacing setting. TCGA CNV data was reanalyzed with identical GISTIC setting. RAW CNV segmentation data for patients and cell lines is available in Supplementary Data 3.2, 3.5 .

Heterozygosity alterations were analyzed across covered regions by plotting a chart of the observed absolute variant allele frequency (VAF) deviations from 0.5 (heterozygous state) for all variants called by HaplotypeCaller, with allele frequencies in range (0.0001, 0.95) in gnomAD and coverage larger than 30x. Normal-tumor pairs were additionally analyzed in a similar manner that also included calculation of VAF shifts between normal and tumor samples. This data was also segmented using CNVkit built-in cbs method after exclusion of homozygous variants to aid in identification of regions with allelic imbalance. Ambiguous results with a VAF deviation lower than 0.15 across segment were considered not altered.

External mutational data for HPV-negative (HPVneg) head and neck cancers was downloaded via cBioPortal (36, 37) for TCGA PanCancer Atlas (38), Agrawal et al. (13) and Stransky et al. (14) projects or from the journal’s site for Vossen et al. (15) project (referred also as “NKI dataset”). Datasets from those projects combined with a current dataset referred to as “Medical University of Warsaw (MUW) dataset” are further referred to as “combined dataset”.

Text data was parsed using python 2.7 and pandas 0.22. Plots and statistical analyses for NGS data were generated using R library maftools v.2.3.40 (39) and matplotlib 2.2.4 (40).

All in vitro experiments were performed on the human cell line derived from squamous cell carcinoma of the hypopharynx, namely FaDu (HTB-43; ATCC). Cells were typically cultured in 75 cm² adherent cell flasks in DMEM D6429 medium (Sigma) supplemented with 10% HyClone calf serum (FBS, SH30072.03, GE Healthcare) and 1x antibiotic/antimycotic-solution (30-004-Cl, Corning) or penicillin-streptomycin solution (P4333, Sigma) during lentiviral infections. Cells were detached into suspension with 1x Trypsin (15090-046, Gibco), passaged in 2-3 day intervals and cryopreserved in liquid nitrogen in DMEM supplemented with 10% DMSO and 50% FBS. For routine passaging, cells were counted in trypan blue in Bürker chamber. For in vitro experiments, Count and Viability Kit for Muse Cell Analyser instrument (Luminex, USA) was used for more accurate counting. All cell cultures were monitored for Mycoplasma contamination by PCR weekly.

CRISPR-Cas9 system was used to generate KMT2C-mutant (KMT2C ^mut) FaDu cells. pLenti CRISPR v2 plasmid [Addgene #52961 (41)] was used along with lentiviral transfection to acquire stable expression of two different sgRNAs (sg2 and sg4), targeting exons 3 and 12 of KMT2C, respectively, as well as a non-targeting sgRNA control (sgNTC) ( Table 2 ). KMT2C-targeting sgRNAs were designed using E-CRISPR (42) and CHOPCHOP tools (43); sg2 sequence was also previously included in the GeCKO v2 library (41). All used sgRNA sequences were predicted to affect the curated KMT2C transcript variant NM_170606.3 as well as most of putative transcripts except for XM_011516454.2, XM_017012489.1 and XM_017012490.2 lacking several exons at the N-terminus. HIV-SFFV-mRFP (received courtesy of dr Els Verhoeyen, Centre International de Recherche en Infectiologie, Lyon, France) plasmid was used as a transfection control. Lentiviral particles were assembled in HEK293 cells grown in DMEM supplemented with 10% FBS and PenStrep, using psPAX2 (Addgene #12260) and pMD2.G (Addgene #12259) plasmids. Virus-containing medium was acquired twice after subsequent overnight incubations and each time it was filtered with 0.45µm syringe filter, concentrated by overnight centrifugation, mixed in 1:1 proportion with fresh DMEM medium and added to FaDu cells. Then, 24h after second infection, puromycin selection was started at 2 µg/ml concentration previously measured to decrease viability of unmodified cells below 5%, with gradual decrease to 0.5 µg/ml over 1 week, resulting in a complete detachment of control HIV-SFFV-mRFP-expressing cells and acquisition of 100% or near 100% pLentiCRISPR transfected cell population as confirmed during further clonal selection.

Induction of mutations by sgRNAs was confirmed by gel electrophoresis of PCR products amplified with primers flanking the sgRNA binding/Cas9 cut sites and subsequently using Sanger sequencing and NGS ( Supplementary Data 2.4 and data not shown).

FaDu cells expressing sg2 and sg4 were diluted to achieve an average concentration of 0.8 cells/per well when seeded on 96-well plates. Clones were detached from wells where single cells formed colonies and propagated through 24- and 6-well plates and 20 cm² and 75 cm² flasks when clone cell lines were frozen and DNA/RNA isolation was performed.

Clone pools and selected clones were compared to NTC-treated and non-treated cells using NGS to exclude the presence of significant off-target mutations and copy-number variations induced by CRISPR-Cas9.

All subsequent in vitro experiments were performed at least two times.

Clonogenic assays for FaDu cell lines were carried out in standard 6-well plates with 625 cells per well. Cells were seeded in triplicates and cultured for approximately 14 days or until colonies started to merge, rinsed with PBS, and fixed and stained with methanol and 20% crystal violet solution.

After staining, plates were scanned with BioRad GS-800 scanner. Colonies were counted with ImageJ software with Fiji package (44) using analyze particles function (size: 0.0001–infinity, circularity 0.01–1.00), and area covered by colonies was measured for those automatically counted.

For DNA synthesis measurements Click-iT EdU Imaging Kit (Thermo Fisher Scientific) was used and EdU incorporation was measured. Cells were seeded on 6-well plates in 0.15×10⁶ cells/well density in triplicates, incubated with 10 µM EdU for 1h the next day, detached with tripsin, fixed and stained with Alexa Fluor 488. Percentage of EdU-positive cells was measured on FACSCanto II cytometer (Becton Dickinson, USA). Exemplary gating is provided in Supplementary Data 1.4 .

CellBlue and CellTiterGlo assays (Promega, USA) were used to measure cell sensitivity to cisplatin. Cells were seeded in 96-well plates in 5 replicates per each group in a density of 3×10³ cells per well and incubated for 24h prior to cisplatin addition. Cisplatin was added to achieve 2 µg/ml and 10 µg/ml final concentrations and the cells were treated for 48h. Viability measurements were taken as recommended by the manufacturer example 10 min after addition of CellTiter-Glo and 4h after addition of CellTiter-Blue. Fluorescence and luminescence were measured on Victor X4 instrument (Perkin Elmer), the latter in white opaque plates. Cell-free medium was used to measure background signal.

All in vitro data was analyzed using GraphPad Prism 6. For clonogenic and DNA synthesis assays, results for KMT2C-mutant cells were compared against NTC cells using Dunnett’s test. In cisplatin-sensitivity assays, average background-subtracted readouts from cisplatin-treated cells were recalculated as fractions of their corresponding controls and compared to NTC cells using Dunnett’s test.

To define the mutational profile in our cohort of patients (MUW dataset) comprised of laryngeal and hypopharyngeal cancers, we analyzed 37 genes previously described as significantly mutated in head and neck and esophageal carcinoma (14, 45–51) and covered by the probe designs used to enrich genomic libraries ( Supplementary Data 1.1–2 ). Variants were manually curated to exclude benign and likely benign. Copy-number and loss-of-hetereozygosity (LOH) data was reanalyzed with the small-scale mutation data to investigate the interplay between these types of genetic aberrations.

In this approach, 32 genes were altered by small-scale mutations in our dataset ( Supplementary Data 2.1, 3.1 ) among which 13 genes were mutated in at least 3 patients ( Figure 1 ). The top six mutated genes were TP53, FAT1, NOTCH1, KMT2C, CDKN2A and KMT2D. TP53 was typically biallelically altered in most cases, either by multiple mutations or combinations of mutations and shallow deletions or copy-neutral losses-of-heterozygosity (CN-LOH), most frequently involving deletion of the short arm of chromosome 17. In some cases, putative amplification of mutated allele was found. FAT1 was altered mostly by frameshift and nonsense mutations, accompanied by shallow deletions. NOTCH1, KMT2C, CDKN2A and KMT2D were similarly altered by truncating mutations and/or shallow deletions or CN-LOH. In particular, deep deletions of CDKN2A were detected in 20.8% of patients (10/48). Finally, 7 other genes were recurrently mutated in more than 2 patients ( Figure 1 and Supplementary Data 2.1 ).

Since TERT promoter was not covered by any of our sequencing panels, we used Sanger sequencing for this analysis. Mutations of pTERT were detected in 6.4% (3/47) patients, for which DNA was available. None of pTERT-mutated patients had any significant CNV of TERT gene, however, three other patients had TERT amplification (4–5 copies) ( Figure 1 ).

A GISTIC analysis of CNV data from MUW dataset yielded a similar pattern of amplification and deletion peaks as in previously published TCGA HPVneg cohort ( Supplementary Data 2.2 ). In the significantly amplified regions, multiple genes were found to have 4 or more copies recurrently, including CCND1 (29%, 14/48 patients), BIRC2/3 (8.3%, 4/48), FGFR1 (10.4%, 5/48), TP63 (14.6%, 7/48), EGFR (6.3%, 3/48), ERBB2 (4.2%, 2/48), MDM2 (4.2%, 2/48), TERT (6.3%, 3/48), PIK3CA (5/48), MYC (8.3%, 4/48), MET (4.2%, 2/48). ERBB2, BIRC2/3, EGFR and CCND1 were highly amplified in some patients (in range of 39–52, 6–16, 5–15 and 4–11 copies, respectively). In the significantly deleted regions, CDKN2A/B (20.8% of patients, 10/48), FAT1 (2.1%, 1/48) and PTEN (2.1%, 1/48) were affected by deep deletion.

A total of 18.8% of patients (9/48) harbored PIK3CA mutations and/or 4-5 gene copies and 6.3% (3/48) harbored PTEN mutation or deep deletion. In addition to the aforementioned copy number alterations, other well-known RAS pathway-activating events were detected in 8.3% patients (4/48): hotspot mutations in HRAS (1 patient), PTPN11 (1) and FGFR3 (1) and NF1 homozygous deletion (1). Finally, one patient carried KRAS amplification (5 copies).

Genes implicated in DNA repair were affected in 18.8% of patients (9/48), either by mutation and deletion or CN-LOH (BAP1: 1 patient; BRCA2: 2; CHEK2: 1; PMS2: 1) or by mutation only (ARID1A: 4 patients, BRCA1: 1; RECQL: 1). Additionally, 25% of patients (12/48) were found to carry shallow deletions of ATM also identified as a significant GISTIC deletion peak yet without any associated ATM mutations.

Also, 4.2% (2/48) of patients carried potentially inactivating CREBBP mutations, similar to those previously described in leukemia (52). Truncating mutations in ASXL1 in 8.3% (4/48) and TET2 in 6.3% (3/48) patients were also detected.

We then analyzed NGS data for potentially druggable or prognostic genetic alterations. We selected only previously described mutations or those with strong prediction of pathogenicity, deep deletions and amplifications to at least 5 copies, while eliminating shallow deletions without a second hit, low-level amplifications and other ambiguous alterations. With this strict approach, we found that over three-quarters of patients harbored at least single alterations of genes involved in RTK/RAS/PI3K, cell-cycle or DNA damage repair pathways ( Figure 2 and Supplementary Data S3.3 ).

While we could not reliably assess tumor mutational burden due to limited availability of paired germline samples, we analyzed copy numbers of CD274/PDCD1LG2 and found that 29.2% (14/48) patients had shallow deletions and 12.5% (6/48) had copy gains (3–4 copies) of these genes, potentially affecting PD-L1 and PD-L2 expression (53) and sensitivity to immune checkpoint inhibitors.

Finally, to better determine the mutational profile of hypopharyngeal cancers we used data from the MUW dataset as well as previously published data for hypopharyngeal cancer samples with confirmed negative HPV infection status (combined dataset, Supplementary Data 3.6 ). In this analysis, we included 18 significantly mutated genes common to all datasets (listed in Supplementary Data 1.1 ). To minimize the differences in variant filtering between datasets we added 6 variants detected in the MUW cohort that we excluded from previous analyses due to low likelihood of pathogenicity. In total, 595 non-hypopharyngeal and 67 hypopharyngeal cancers were compared using maftools. We did not find any significant differences ( Figures 3A, B and Supplementary Data 2.3 ) in mutation frequency except for CASP8 mutations, which were very rare (p = 0.0058, OR 0.12, CI 0.0028–0.69) and HRAS mutation, which were absent (p = 0.025, OR = 0, CI 0–0.87) in hypopharyngeal cancers.

In our dataset, we identified mutations in KMT2C gene, encoding Histone-Lysine N-Methyltransferase 2C, in 14 (29%) or 9 (18.8%) patients, after elimination of benign variants. Seven (14.6%) patients carried KMT2C truncating mutations and we also observed frequent shallow deletions or combinations of mutation and CNV/CN-LOH ( Figure 1 ). In the combined HPVneg dataset (excluding NKI dataset, Vossen et al., 2018) KMT2C mutations were present in 8.7% (48/551) of patients ( Supplementary Data 2.3, 3.7 ) and 3.4% (19/551) of them harbored truncating mutations that were scattered across the gene ( Figure 3C ). Moreover, we found that hypopharyngeal cancers harbored more KMT2C mutations than cancers from other sites ( Figures 3A, B and Supplementary Data 2.3 ), but this difference is likely not biologically relevant as discussed below.

KMT2C was already characterized as a tumor suppressor gene in acute myeloid leukemia (54). Based on above-mentioned data and our findings in laryngeal and hypopharyngeal cancer, we decided to assess the biological effects of KMT2C mutation in commercially available HPV-negative hypopharyngeal cell line FaDu (ATCC HTB-43). First, we characterized FaDu cells using targeted NGS and found multiple pathogenic genetic aberrations, including mutations of TP53 (missense and splice-site), CDKN2A (splice-site, homozygous), FAT1 (frameshift, homozygous), SYNE1 (missense, frameshift), deep deletions of AJUBA and SMAD4, high-level amplification of CCND1, as well as mutations in ERBB3, VHL and other genes ( Supplementary Data 3.4 ). Hence, we found that FaDu harbors multiple genetic lesions frequently occurring in head and neck and esophageal cancer.

To study the role of truncating mutations in KMT2C in HNSCC, we used CRISPR-Cas9 to induce mutations in the N-terminal quarter of the coding sequence ( Figure 3C ). We generated KMT2C ^mut FaDu cells stably expressing two different KMT2C-targeting sgRNAs (sg2 and sg4) as well as a non-targeting control sgRNA (sgNTC). Using PCR, Sanger sequencing (data not shown) and NGS ( Supplementary Data 2.4 ) we confirmed that both sgRNAs effectively induced indels while sgNTC did not, additionally revealing that most clones derived from pools harbored more than 2 copies of KMT2C, in concordance with the expected hyperdiploidy of FaDu cells and ploidy analysis in selected cell populations (data not shown). Based on initial growth observations we selected clones sg2-7 and sg4-14 along with clone pools sg2 and sg4 for further experiments.

Targeted NGS analysis of CRISPR-Cas9 modified cells confirmed the specific induction of various KMT2C mutations at exons 3 and 12 ( Supplementary Data 2.4 ). Furthermore, through Mutect-2 comparisons and manual analysis of NGS data, we did not find any additional, significant genetic point mutations or CNVs unique to any of the modified cell pools, except for clone sg4-14 which could be distinguished by lack of several variants ( Supplementary Data 3.4 ) as well as by more pronounced deletion of 3p11.1/EPHA3.

We checked whether induction of KMT2C mutations affects phenotypic features of FaDu cells and, therefore, firstly assessed the clonogenic potential of KMT2C ^mut FaDu cells by the clonogenic assay on 6-well plates. We noticed a marked increase in clonogenicity of the modified cells as compared to sgNTC-expressing cells, as manifested not only by significant differences in number of colonies but also in the growth surface area. These effects were confirmed in sg2 and sg4 clone pools (1.19x and 1,47x more colonies and 2.72x and 2.61x larger growth area, respectively) and clones sg2-7 and sg4-14 (1.97x and 2.41x more colonies and 3.34x and 3.44x larger growth area, respectively) ( Figures 4A, B ).

We could not reliably analyze the cell cycle using propidium iodide due to variable ploidy among clone pools and clones (data not shown). Instead, we analyzed DNA synthesis rate using EdU incorporation assay. Significantly increased uptake was observed for KMT2C ^mut cells (8.5%, 5.5% and 18.5% increase for sg2, sg4 and sg2-7, respectively), except for sg4-14 clone (5.1% decrease), confirming their proliferative advantage over sgNTC cells ( Figure 4C ).

Finally, to test if KMT2C mutation could affect response to standard chemotherapy, we measured viability of FaDu cells exposed for 2 days to two different cisplatin concentrations using CellTiter-Blue and CellTiter-Glo (Promega) assays. While some statistically significant differences in cisplatin sensitivity were observed for clones sg2-7 and sg4-14, they could not be clearly associated with KMT2C mutation, given their magnitude and distribution ( Figure 4D ).