Mutation allele frequency threshold does not affect prognostic analysis using next-generation sequencing in oral squamous cell carcinoma

Forty-six patients with locally advanced OSCC from 2008 to 2010 were included in this study, which was approved by the Human Research Ethics Committee of the Ninth People’s Hospital Shanghai Jiao Tong University School of Medicine [approval number: 2008(12)]. Cancerous tissue samples were derived from biopsy specimens, while matched non-cancerous tissue samples were derived from neck dissection specimens; all specimens were formalin-fixed and paraffin-embedded (FFPE). All patients received radical surgery and post-operative radiotherapy. Four of the patients received docetaxel, cisplatin, and 5-fluorouracil (TPF) induction chemotherapy prior to definitive treatment with surgery and radiotherapy.

All tissue specimens were reviewed by two pathologists, and tumor cell areas on hematoxylin-eosin (HE) stained slides were determined for manual microdissection and subsequent DNA sequencing. Only tumor cells (appearing as HE-stained) were microdissected as cancerous samples, with an average tumor purity of 85% (ranging from 80 to 95%). Five 10-μm FFPE sections from each block were deparaffinized, and DNA was extracted using the QIAamp DNA FFPE Tissue Kit (Qiagen, Hilden, Germany). Quality and quantity of the purified DNA were measured using the Qubit and Nano-Drop platforms (Thermo Fisher Scientific, Waltham, MA, USA).

Ten nanograms of DNA was used for multiplex PCR amplification. Libraries were constructed using the Ion AmpliSeq Library Kit v2.0 (Thermo Fisher Scientific) according to the manufacturer’s instructions. The quality of the obtained libraries was evaluated using Agilent 2100 Bioanalyzer on-chip electrophoresis (Agilent Technologies, Palo Alto, CA, USA). Emulsion PCR was performed with the OneTouch DL or OneTouch 2 system (Thermo Fisher Scientific). Sequencing was run on the Ion Torrent Personal Genome Machine (Thermo Fisher Scientific), loaded with a 316 or 318v2 chip as per the manufacturer’s protocol. Data analysis, including alignment to the hg19 human reference genome as well as variant calling and filtering, was completed using Torrent Suite Software v3.6 (Thermo Fisher Scientific). Filtered variants were annotated using Ion Reporter software v4.4 (Thermo Fisher Scientific).

Selected mutations were confirmed using Sanger sequencing. Sequence variants were compared with the head and neck squamous cell carcinoma dataset from the Cancer Genome Atlas (TCGA), dbSNP1000 Genomes, ClinVar database, COSMIC, 5000 Exomes, OMIM, and Pfam databases. SIFT, Polyphen, Phylop, and Grantham scores were used to estimate evolutionary conservation and the effects of amino acid substitutions on the structure and function of the protein.

In the context of this analysis, a somatic mutation was considered to be “validated” if: (1) the altered reads were ≥ 10, (2) non-synonymous mutations had an allele frequency ≥ 3%, and (3) non-synonymous mutations were not found in any of the SNP databases. All mutations were visually confirmed using the Integrative Genomics Viewer (IGV v2.3).

Overall survival (OS) was calculated from the date of the pathological diagnosis to the date of death; disease-free survival (DFS), locoregional recurrence-free survival (LRFS), and distant metastasis-free survival (DMFS) were calculated from the date of the pathological diagnosis to recurrence, locoregional recurrence, or distant metastasis or death from any cause, respectively. For descriptive analyses, categorical data were expressed as number and percentage. Survival analysis was conducted using the Kaplan-Meier method with a Log-rank test. Hazard ratios (HRs) were calculated using the Cox proportional hazards model. All hypothesis-generating tests were two-sided at a significance level of 0.05. Data were analyzed with SPSS v. 18.0 for Windows (SPSS Inc., Chicago, IL, USA).

Among the targeted CAGE (for TP53, NOTCH1, CASP8, CDKN2A, CDH1, ANXA1, EGFR, IGFBP3, TGFB1, CTNNB1, PTEN, and TP63), a mean of 3099-fold sequence coverage was achieved, and 99% of the targeted CAGE was represented by at least 10 reads. Compared to matched non-cancerous tissues and reference sequences, a total of 94 non-synonymous (missense, nonsense, splicing site, insertion, and deletion) mutations were detected in the cancerous tissue samples from 40 OSCC patients with an allele frequency threshold of ≥10%. None of the non-synonymous mutations were found in the SNP databases. The average number of non-synonymous mutations per sample was 2.04. Missense mutations made up the majority (70.2%) of the identified variants, followed by nonsense mutations (11.7%), insertions/deletions (10.6%), and splice site mutations (7.5%) (Additional file 1: Table S1).

With an allele frequency threshold of ≥5%, 132 non-synonymous mutations were detected in the cancerous tissue samples from 41 patients. The average number of non-synonymous mutations per sample was 2.87. Missense mutations made up the majority (77.3%) of the identified variants, followed by nonsense mutations (9.9%), insertions/deletions (7.6%), and splice site mutations (5.3%) (Additional file 2: Table S2). With an allele frequency threshold of ≥3%, 239 non-synonymous mutations were detected in the cancerous tissue samples from 42 patients. The average number of non-synonymous mutations per sample was 5.20. Similar to the 10 and 5% thresholds, missense mutations made up the majority (86.2%) of the identified variants followed by nonsense mutations (6.7%), insertions/deletions (4.2%), and splice site mutations (2.9%) (Additional file 3: Table S3).

To verify the reliability of the deep sequencing in our panel, the remaining genomic DNA was used for validation. TP53 mutation was selected as a representative example because it was the most frequently mutated gene with 46 genetic variates in our panel. The TP53 Sanger sequencing was performed to confirm the TP53 variants. Because of the low DNA content, 9 non-synonymous mutations cannot be validated by Sanger sequencing; among the other 37 non-synonymous mutations, 70.3% (26/37) of them were successfully validated by Sanger sequencing in the same DNA samples (Additional file 4: Table S4).

Non-synonymous mutations were identified in all 12 genes when the allele frequency threshold was not defined. However, when the allele frequency threshold was defined as ≥10%, the most frequently mutated genes were TP53(78.3%), NOTCH1(30.4%), CASP8(13.0%), CDKN2A(10.9%), and CDH1(6.5%). Genetic mutations in the most frequently mutated genes were identified in 40 (87.0%) of the patients. The mutation frequency was very low in ANXA1 (2.2%), EGFR (2.2%), IGFBP3 (2.2%), and TGFB1 (2.2%). No non-synonymous mutations were observed at an allele frequency of ≥10% in CTNNB1, PTEN, and TP63. The mutation frequency of each gene when the allele frequency threshold was defined as ≥5% and ≥ 3% is shown in Table 1.

We then compared our data to the mutational patterns reported in The Cancer Genome Atlas (TCGA) head and neck squamous cell carcinoma (HNSCC) database (containing whole-exome sequencing data from 279 samples) (Table 1). With an allele frequency threshold of ≥10%, the frequency of mutations detected in our study was similar to that reported in the TCGA database, with the exception of NOTCH1 and CDH1. The mutation frequencies of NOTCH1 and CDH1 were much higher in our cohort than those reported in the TCGA database (30.4% vs. 18.3%, P = 0.056 and 6.5% vs. 1.4%, P = 0.062 for NOTCH1 and CDH1, respectively). Further analysis on the molecular characteristics of the detected mutations was based on an allele frequency threshold of ≥10%.

TP53 was the most frequently mutated gene, with a total of 46 genetic variants (33 missense mutations, 4 nonsense mutations, 5 insertions/deletions, and 4 splice-site mutations) in 36 patients (78.3%). Thirty-three out of 46 (71.7%) p53 mutations were found to be located in the DNA-binding domain, while 4 (8.7%) mutations were in the tetramerization motif, and 2 mutations (4.4%) were in the transactivation motif (Table 2). When compared to the TCGA database, p.Val216Met, p.Pro151Thr, p.Arg175His, p.Arg337Cys, p.Arg282Trp, p.Ala159Val, p.Arg273His, p.Arg248Gln, p.Arg282Trp, p.His193Leu, p.His178fs, p.Gly245Ser, p.Pro152Leu, p.Tyr220Cys, p.Gln331Ter, p.Pro151His, p.Arg342Ter, p.Glu286Lys, and p.Arg213Ter appeared in the TCGA HNSCC database; p.Cys135Phe, p.Phe113Cys, p.Cys176Phe, p.Trp53Ter, p.His179Leu, p.Val272Leu, p.Cys135Tyr, p.Arg213Gln, p.Pro191del, p.Val274Phe, p.Thr253Ile, p.Asp184His, p.Cys135Phe, p.Val218Glu, p.Ile255Phe, p.Asp148Asn, p.Asp57Asn, p.Pro128Ser, p.Leu93fs, and p.Pro85Ser appeared in the TCGA database of breast, bladder, renal, lung, stomach cancers and melanomas. The other mutations listed in the Additional file 1: Table S1, Additional file 2: Table S2, Additional file 3: Table S3 did not appear in the TCGA database.

NOTCH1 was also frequently mutated, with a total of 25 genetic variants (22 missense mutations, 2 insertions and deletions, and 1 splice site mutation) in 14 patients (30.4%). Nineteen out of 25 (76.0%) mutations were found to be located in the EGF-like repeat domains, and 5 (20.0%) mutations were in the intracellular domain, including 1 mutation in the ankyrin repeat domain (Table 2). When compared to the TCGA database, p.Ala465Thr and p.Asp338Asn appeared in the TCGA HNSCC database; p.Asn718Ser, p.Glu488Lys, p.Pro2332Leu, p.Arg2272Cys, p.Arg365Cys, p.Val1229Ile, p.Gly1195Arg, p.Pro1097Ser, p.Ala1338Thr, p.Ser2336Asn, p.Ser836Asn, p.Pro668Ser, p.Phe436Leu, p.Thr767Ile, p.Met2362Ile, p.Ser1541Asn, p.Pro148Leu, p.Arg1758Cys, p.Arg1211Gln, and p.Thr588Ile appeared in the TCGA breast, esophageal, colon, bladder, prostate, lung cancers and melanomas.

Eight CASP8 mutations were identified in six patients (13.0%). Among them, 6 mutations were inactivating mutations (4 nonsense mutations and 2 deletions); the other 2 mutations were missense mutations. Seven (87.5%) mutations were located in the caspase homology domain, and the other was in the death effector domain (Table 2). When compared to the TCGA database, p.Arg494Ter and p.Arg472Ter appeared in the TCGA HNSCC database; p.Lys532fs and p.Gln417Ter appeared in the TCGA colorectal and bladder cancer database, respectively.

Five CDKN2A mutations were identified in five patients (10.4%). Most of them were inactivating mutations, including 2 nonsense mutations, 1 insertion, and 1 splice site mutation; the other was missense mutation. Four CDKN2A mutations were located in ankyrin repeats, and the other was a splice site mutation. When compared to the TCGA database, p.Glu120Ter appeared in the TCGA HNSCC database.

Three CDH1 mutations were identified in three patients (6.5%). Two mutations were missense mutations and the other was a nonsense mutation. Two CDH1 mutations were located in the cadherin repeat domain, and the other was in the cadherin cytoplasmic region. p.Trp532Ter and p.Arg90Trp appeared in the TCGA cervical and mixed cancer database, respectively.

Three EGFR mutations were identified in one patient (2.2%), including 2 missense mutations and 1 splice site mutation. The 2 missense mutations were located in the furin-like repeats of EGFR. Two missense IGFBP3 mutations were identified in one patient (2.2%), and these mutations were located in the thyroglobulin type-1 repeat domain and the IGFBP homology domain, respectively. One missense ANXA1 mutation was identified in one patient (2.2%), but it was not located in the conservative region of annexin a1. One missense TGFB1 mutation was identified in one patient (2.2%), which was in the latency-associated peptide of TGFβ-1. Some of the mutations also appeared in the TCGA database.

The p53 DNA-binding domain was the major conserved domain that contained a mutation in 28 patients (60.9%), and the NOTCH1 EGF-like repeats domain was the second major conserved domain, with a mutation in 13 patients (28.3%), followed by the caspase homology domain of caspase 8 that was mutated in five patients (10.9%), the tetramerization motif of p53 with a mutation in four patients (8.7%), and the ankyrin repeat domains of p16, which contained mutations in four patients (8.7%) (Table 2).

There was one patient (2.2%) who had 8 mutations in conserved domains of various genes, and one patient (2.2%) had 4 mutations in these conserved domains, followed by three patients (6.5%) with 3 mutations, 13 patients (28.3%) with 2 mutations, and 19 patients (41.3%) with just 1 mutation in a conserved domain.

Correlation analysis between non-synonymous mutant status (including all targeted genes of CAGE) and baseline characteristics was performed; no significant correlations were found (Table 3). Survival analysis found no significant differences in outcomes with regard to OS, DFS, LRFS, and DMFS between patients with non-synonymous mutations and wild type carriers (Fig. 1).

When the allele frequency thresholds of 5 and 3% were used, the difference in outcomes between patients with non-synonymous mutations and wild type carriers was non-significant (Additional file 5: Figure S1 and Additional file 6: Figure S2).

Five genes (TP53, NOTCH1, CASP8, CDKN2A, and CDH1) with an allele frequency of non-synonymous mutations above 10% were selected for further relationship analysis, and no significant correlations were found between patients with non-synonymous mutations in each gene and baseline characteristics, with the exception of NOTCH1 (Additional file 7: Table S5, Additional file 8: Table S6, Additional file 9: Table S7, Additional file 10: Table S8, Additional file 11: Table S9). The NOTCH1 non-synonymous mutation rate was higher in T1/T2 patients than that in T3/T4 patients. To investigate the impact of allele frequency threshold on survival analysis, a univariate Cox model was used with allele frequency thresholds of 10, 5, and 3%; the difference in OS (Fig. 2) and DFS (Fig. 3) between patients with non-synonymous mutations and wild type carriers was not significant.