Clonal Mutations Activate the NF-κB Pathway to Promote Recurrence of Nasopharyngeal Carcinoma

High-throughput next-generation sequencing has provided enormous insight into the genomic landscape of several tumor types, illuminating molecularly defined tumor subtypes, identifying new druggable targets, and providing insights into the heterogeneity of many tumors (1). Previous studies all focused on the genomic landscape of primarily diagnosed nasopharyngeal carcinoma (pNPC; refs. 2–6), identifying significant mutations in negative regulators of NF-κB pathways, ERBB/PI3K signaling, and chromatin modification processes. In addition, frequent chromosomal deletions of Chr 3p, 14q, and 16q were observed. Compared with pNPC, recurrent NPC (rNPC) tumors conceivably undergo extensive genomic evolution during progression and resistance and therefore likely harbor unique genomic drivers compared with pNPC tumors. However, the understanding of the genomic information for recurrent NPC is very poor, with only 11 local recurrent samples profiled hitherto (3). Therefore, a strong need exists to comprehensively characterize the genomic foundations of rNPC.

Accumulating evidence suggests that tumors often evolve through a process of branched evolution, involving genetically distinct subclones (7). Therefore, drug development and precision medicine strategies will likely need not only an understanding of cancer genes and mutational processes but also an appreciation of the clonal status of driver events as well as the order of mutational processes (8). Herein, we report the genomic landscape of rNPC based on large-scale rNPC cohorts using high-coverage whole-exome sequencing (WES) and whole-genome sequencing (WGS). We additionally profiled pNPC exomes and integrated publicly available databases to comprehensively conduct comparisons of the significantly mutated genes (SMG), clonal status of driver events, mutational patterns, and the timing of mutational processes between rNPC and pNPC to illuminate the critical drivers of NPC progression and recurrences for both prognosis and treatment advancement. Consequently, based on these differential mutation events, we attempted to identify novel biomarkers and validate them in pNPC samples to determine patient outcomes and therapeutic options.

All samples for sequencing were histologically confirmed as NPC (WHO I, II, or III) and obtained at Sun Yat-sen University Cancer Center (SYSUCC; Guangzhou, China) between June 1, 2012, and May 1, 2016. The institutional review board of SYSUCC approved the study protocol, and all patients provided written informed consent. The quality of tumor samples was examined by tissue sectioning and hematoxylin and eosin staining to estimate the tumor content. Only the highest quality samples with no less than 30% tumor content were chosen for subsequent study. A total of 55 rNPC and 44 pNPC fresh tissue samples were collected and then analyzed by sequencing. In addition, 7 paired pNPC and rNPC samples were collected and then analyzed by deep sequencing (Supplementary Table S1). Patients with pNPC had not previously received chemotherapy or radiotherapy when diagnosed. Patients with rNPC were diagnosed with time-to-tumor recurrence of at least 6 months from the end of the last primary course of radiotherapy. All patients with rNPC received radiotherapy with a radical dose at the time of initial diagnosis. Among these 60 patients with rNPC, 40 patients received intensity-modulated radiotherapy and the remaining 20 patients received two-dimensional radiotherapy (2D-RT) or 3-dimensional conformal radiotherapy (3D-CRT) at the time of initial diagnosis. Furthermore, 46 of 60 patients were treated with cisplatin-based chemotherapy combined with radiotherapy at the time of initial diagnosis.

In addition, 148 rNPC and 237 pNPC formalin-fixed, paraffin-embedded (FFPE) samples were histologically confirmed as NPC (WHO I, II, or III) and obtained at SYSUCC between January 1, 2002, and December 1, 2010, with approval of the ethics committee of SYSUCC. Almost all patients (more than 97.9%) were treated with 2D-RT at SYSUCC. The follow-up time for all 237 patients with pNPC still alive was more than 5 years (Supplementary Table S2).

pNPC data from publicly available sequencing data were downloaded from the National Center for Biotechnology Information Sequence Read Archive, www.ncbi.nlm.nih.gov/sra (accession nos. SRA288429 and SRA291304) and European Nucleotide Archive (accession no. PRJEB12830). These files consisted of 50 and 69 pNPC samples recently reported by Zheng and colleagues (4) and Li and colleagues (3), respectively.

Genomic DNA from frozen tissues and FFPE samples was extracted using a DNeasy blood and tissue kit (Qiagen) and QIAamp DNA FFPE Tissue Kit (Qiagen), respectively, following the manufacturer's protocol. Degradation and contamination were monitored on a 1% agarose gel, and the concentration was measured by using a Qubit DNA Assay Kit in a Qubit 2.0 Fluorometer (Life Technologies).

For WGS, a total of 0.8-μg genomic DNA per sample for 12 patients with rNPC and 17 patients with pNPC with high molecular weight (>20 Kb single band) was used for DNA library preparation. A sequencing library was generated using a TruSeq Nano DNA HT Sample Prep Kit (Illumina) following the manufacturer's recommendations, and index codes were added to each sample.

For WES, qualified genomic DNA from tumors and matched peripheral blood from 48 patients with rNPC and 34 patients with pNPC were fragmented by Covaris technology with resultant library fragments of 180 to 280 bp, and then adapters were ligated to both ends of the fragments. Extracted DNA was then amplified by ligation-mediated PCR, purified, and hybridized to the Agilent SureSelect Human Exome V6 for enrichment, and nonhybridized fragments were then washed out. Both uncaptured and captured ligation-mediated PCR products were subjected to real-time PCR to estimate the magnitude of enrichment. Each captured library was then loaded onto the Illumina HiSeq X platform, and we performed high-throughput sequencing for each captured library independently to ensure that each sample met the desired average fold coverage.

The original fluorescence image files obtained from the HiSeq platform were transformed to short reads (raw data) by base calling and recorded in FASTQ format, which contained sequence information and corresponding sequencing quality information. After excluding reads containing adapter contamination and low-quality/unrecognizable nucleotides, clean data were used for downstream bioinformatical analyses. At the same time, the number of total reads, sequencing error rate, percentage of reads with average quality >20 and with average quality >30, and GC content distribution were calculated.

Valid sequencing data were mapped to the human reference genome (UCSC hg19) by Burrows-Wheeler Aligner software to obtain the original mapping results stored in BAM format (9). Then, SAMtools (10) and Picard (http://broadinstitute.github.io/picard/) were used to sort BAM files and perform duplicate marking to generate a final BAM file for computing the sequence coverage and depth.

To call somatic single-nucleotide variations and small insertions and deletions (InDels) from paired tumor-normal samples, MuTect and Strelka were used, respectively (11, 12). Subsequently, the variant call format file was annotated by ANNOVAR (13).

Somatic copy-number variants (CNV) were identified by Control-FREEC (14), whereas the GISTIC (15) algorithm was used to infer recurrently amplified or deleted genomic regions. G-scores were calculated for genomic and gene-coding regions on the basis of the frequency and amplitude of amplifications or deletions affecting each gene. A significant CNV region was defined as having an amplification or a deletion with a G-score of > 0.1, corresponding to a P value threshold of 0.05 from the permutation-derived null distribution.

For somatic structural variant detection based on the soft-clipped reads, CREST (16) was implemented to directly map structural variations at the nucleotide level of resolution. Only break point pairs with at least three supporting clipped reads spanning the break point were selected for further analysis.

SMGs were identified using MuSiC (17, 18). MuSiC estimates the background mutation rate for each gene–patient–category combination based on the observed silent mutations in the gene and noncoding mutations in the surrounding regions. Significant levels (P values) were determined by three tests, and FDRs (q value) were then calculated. SMGs were mutated in at least two tumor samples with an FDR of less than 0.2.

Pathway enrichment analysis was carried out by using the PathScan algorithm to identify known cellular pathways with significant accretions of somatic mutations in rNPC tumors (19). Regardless of the frequency of mutation in specific genes, the entire nonsynonymous mutation was investigated to figure out the distribution of genes within the Kyoto Encyclopedia of Genes and Genomes (KEGG) database.

All somatic point mutations in exonic region were extracted from the WGS and WES data and classified into 96 mutated triple nucleotides (mutated base plus its sequence context). Then, the underlying mutational signatures were deciphered by the nonnegative matrix factorization (NMF) supplied by the Wellcome Trust Sanger Institute Mutational Signature Framework (20). The Gene Pattern platform with NMF module was exploited to obtain several signatures, which were then clustered with previously defined signatures in the Catalogue of Somatic Mutations in Cancer (COSMIC) database.

The cancer cell fraction (CCF) of the variant was determined through an integrative analysis of tumor purity, variant allele frequency, and variant copy number as we and others have described previously (8, 21, 22). Segmented copy-number data along with mutational profiles were fed into the ABSOLUTE program to assess the CCF of each variant, taking into account tumor purity, read depth coverage, and variant allele frequencies (23). In line with the recommended best practice, all ABSOLUTE solutions were reviewed by three researchers, with solutions based on majority vote. A variant was classified as clonal if the upper 95% confidence interval of its CCF was greater than or equal to 1, as previously described (8, 21, 22).

Categorized variables were compared using the χ² test, the correction χ² test, or Fisher exact test, and continuous variables were compared using the Mann–Whitney U test. The events for overall survival (OS) and locoregional relapse-free survival (LRRFS) were death from any cause and locoregional recurrence, respectively. Survival results were calculated using the Kaplan–Meier method, and differences were compared by log-rank test. We then performed a multivariate Cox regression analysis to identify independent prognostic factors for survival outcomes (24), with the assumption of proportional hazards confirmation based on Schoenfeld residuals (25).

All of the analyses were performed using Stata version 10.0, and a two-tailed P value of less than 0.05 was considered statistically significant.

Paraffin sections were stained for NF-κB P65 (Cell Signaling Technology, 8242S), NOTCH1 (Bioss, bs1335R), RPL22 (Biorbyt, orb128654), BAP1 (Abnova, PAB1686), FAM135B (Bioss, bs-9063R), and NCKAP5L (Invitrogen, PA5-59404) by IHC.

The nuclear expression of NF-κB was assessed by measuring the proportion of positive tumor cells with NF-κB translocated to the nucleus. Positive tumor cells were defined as cells with brown or tan particles of NF-κB proteins appearing in the nucleus. Five fields of view in a high-magnification lens (×400) were randomly selected for observation and counting. We counted 200 tumor cells in each field of view and calculated the relative proportion of positive tumor cells. The average value in these five fields of view was defined as the proportion of positive tumor cells with NF-κB translocated to the nucleus in this case.

The staining results were evaluated by two scores for NOTCH1, RPL22, BAP1, FAM135B, and NCKAP5L: the intensity of expression from negative to strongly positive as 0 = negative, 1 = weak positive expression, 2 = moderate positive expression, 3 = strong positive expression, and the proportion of the expression area ranging from 0% to 100% (0 = 0%–5%, 1 = 6%–25%, 2 = 26%–50%, 3 = 51%–75%, and 4 = 76%–100%). The final total score was equal to the multiplied results of the intensity score and its corresponding area score for each patient.

All of the staining results were evaluated independently by two pathologists (C.-Y. Wu and C. Zhang) who were blinded to the clinicopathologic data for the patients. If any evaluated results from the same cases differed greatly between these two pathologists, the case was reevaluated by another supervising pathologist (J.-P. Yun).

The three human NPC cell lines (SUNE1, HONE1, and CNE2) used in this study were provided by Professor Jin-Xin Bei (SYSUCC) and routinely maintained in RPMI1640 medium (Life Technologies) supplemented with 10% FBS (Life Technologies) at 37°C under 5% CO₂. The human NPC cell line (SUNE2) and 293T cell line used in this study were provided by Professor Mu-Sheng Zeng and routinely maintained in DMEM (Life Technologies) supplemented with 10% FBS (Life Technologies) at 37°C under 5% CO₂. All NPC cell lines used in this study were authenticated using short tandem repeat profiling. All cell lines were tested Mycoplasma-free as determined by PCR-based method (16S rDNA-F: 5′-ACTCC TACGGGAGGCAGCAGTA-3′, 16S rDNA-R: 5′-TGCACCATCTGTCACTCTG TTAACCTC-3′). Mycoplasma testing was carried out every 2 or 3 weeks, and the cells were not cultured for more than 2 months.

Briefly, the cells were lysed in RIPA buffer (Sigma-Aldrich) containing protease and phosphatase inhibitor, and the protein concentration was measured with a BCA Protein Assay Kit (Pierce). Equal amounts of protein lysates were electrophoretically separated by 10% SDS-PAGE and transferred to polyvinylidene difluoride membranes. The membranes were blocked with 5% nonfat dried milk for 1 hour at room temperature and then incubated with primary antibodies at the recommended concentration in Tris-Buffered Saline Tween-20 for 1 hour at room temperature. After incubation with horseradish peroxidase–conjugated secondary antibody for 1 hour at room temperature, the protein bands were detected using an ECL Detection System (Pierce). The following antibodies were used for Western blotting: rabbit mAbs against p65 (ab79398; Abcam), H3 (ab18521; Abcam), and HA-Tag (3724; Cell Signaling Technology).

The cells were trypsinized and seeded on glass cover slips in 24-well plates. After 24 hours, the plates with cells were irradiated at a dose of 2 Gy. The immunofluorescence analysis was performed as described previously (26). For γH2AX, the cells were analyzed at 0, 2, and 6 hours after irradiation. The cells were fixed with cold methanol for 10 minutes at room temperature, followed by blocking in 5% BSA for 30 minutes. For γH2AX, the cells were sequentially incubated with a rabbit mAb against phospho-H2A.X (1:1,000; Cell Signaling Technology; #9718) and antirabbit Alexa 488–conjugated secondary antibody (Life Technologies). The nuclei were then counterstained with DAPI solution (Life Technologies), and the coverslips were mounted with ProLong Gold Antifade Mounting Solution (Life Technologies). Images were taken using an Olympus FV100 confocal imaging system. Cells with more than 20 γH2AX foci were defined as γH2AX foci-positive cells, and five random fields were examined to estimate the number of foci-positive cells per field for each coverslip.

For cell viability analysis after treatment with cisplatin, cells were plated in 96-well microplates at a density of 5 × 10³ cells per well and cultured overnight. Culturing was followed by the addition of increasing concentrations of drugs and incubation for 24 hours and then cell viability was determined by an MTT assay.

For the apoptosis assay, the cells were harvested and washed with cold PBS and then the cells were stained first with Annexin V (Nanjing Kaiji Bio-Tek Corporation) for 20 minutes at 4°C in the dark and second with propidium iodide solution (50 μg/mL). Cells were then analyzed using a Cytomics FC 500 instrument (Beckman Coulter). The results were analyzed and displayed with CXP software and FlowJo software.

Male BALB/c nude mice (3–4 weeks of age, 18–20 g) were purchased from GemPharmatech Co., Ltd. All experimental procedures were approved by the Institutional Animal Care and Use Committee of Sun Yat-Sen University (Guangzhou, China). Two in vivo experiments were designed. In experiment 1, CNE2 cells (1 × 10⁷) were injected subcutaneously into the haunch of each mouse. All mice were randomly assigned to the radiotherapy plus JSH-23 group and radiotherapy plus DMSO group (n = 9 per group) on day 10 when the tumors were measurable. Treatment began on day 10. JSH-23 was intraperitoneally injected at a dose of 3 mg/kg once a day. DMSO was intraperitoneally injected at 100 μL once a day as the control. The xenografted tumor was irradiated at a dose of 2 Gy once a day for a total of 7 times after the mice were anesthetized with 1.0% pentobarbital intraperitoneally injected at a dose of 150 μL. In experiment 2, CNE2 cells (1 × 10⁷) were injected subcutaneously into the flank of each mouse. All mice were randomly assigned to the cisplatin plus JSH-23 group and cisplatin plus DMSO group (n = 7 per group) on day 10 when the tumors were measurable. Treatment also began on day 10. JSH-23 was intraperitoneally injected at dose of 3 mg/kg once a day. DMSO was also intraperitoneally injected at 100 μL every day as the control. Cisplatin was administered at a dose of 3 mg/kg once a week for 2 weeks. Tumor volume (V) was measured every 2 days and calculated with the formula V = (Length × Width²)/2.

The WES and WGS data that support the findings of this study have been deposited in the Genome Sequence Archive (27) in BIG Data Center (28), Beijing Institute of Genomics (BIG), Chinese Academy of Sciences, under accession numbers HRA000052 and HRA000053, which can be accessed at http://bigd.big.ac.cn/gsa-human. The correspondence between the sample ID in the manuscript and the sample alias in the database (HRA000053) is shown in the Supplementary Table S3. The authenticity of this article has been validated by uploading the key raw data onto the Research Data Deposit public platform (http://researchdata.org/), with the approval RDD number RDDB2019000658.

We performed WGS with 12 and 17 germline and tumor pairs in rNPC and pNPC, respectively, and conducted WES in 43 and 27 germline and tumor pairs in rNPC and pNPC, respectively. For tumor samples in rNPC, the sequencing yielded mean genomic and exonic coverage of 63 × (range 59–65 ×) and 491 × (range 380–636 ×), with approximately 99% of targeted bases covered by greater than 10 reads; in pNPC, the sequencing yielded mean genomic and exonic coverage of 108 × (range 48–175 ×) and 227 × (range 173–306 ×), with approximately 99% of targeted bases covered by greater than 10 reads (Supplementary Table S4). We identified 3,238 variants in protein-coding and splicing regions in rNPC. In addition, we identified 2,324 variants in protein-coding and splicing regions in pNPC (Supplementary Tables S5 and S6). Validation of candidate mutations with Sanger sequencing and TA vector clones showed that true positive rates of 93.9% and 100.0% were achieved in rNPC and pNPC, respectively (Supplementary Table S7). A total of six variants were overlapped between these unpaired pNPC and rNPC samples, including MUC17 (c.C2513T:p.T838I; c.G1279A:p.A427T), AHNAK2 (c.A11605G:p.M3869V), CYLD (c.C1103A:p.S368X), FOXD4 (c.C122A:p.A41E), and TP53 (c.G457A:p.E153K).

In addition, we performed WES with seven germline and tumor pairs in paired pNPC and rNPC samples, respectively. The sequencing yielded mean exonic coverages of 408 × (range 350–445 ×) in pNPC samples and 470 × (range 448–514 ×) in rNPC samples, with approximately 99% of targeted bases covered by greater than 10 reads (Supplementary Tables S8 and S9). A total of 127 variants were overlapped between these seven paired rNPC and pNPC samples (Supplementary Table S10).

rNPC tumors exhibited a tumor burden comparable to that of pNPC (P = 0.079) and pNPC reported by Zheng and colleagues (P = 0.870; ref. 4). The tumor burden was significantly higher in rNPC than in pNPC reported by Li and colleagues (P = 1.2 × 10⁻⁶; ref. 3), which might be explained by the strict criteria for somatic variant calling because of the potential artifacts in FFPE samples (Fig. 1A). Furthermore, the increased mutational burdens in rNPC lesions were not correlated with the recurrence rate or remission time (P = 0.624; P = 0.193, respectively). In addition, we observed a significant increase in the number of clonal mutations in the rNPC samples compared with pNPC reported by Li and colleagues (P = 0.019; ref. 3) and pNPC reported by Zheng and colleagues (P = 0.034; ref. 4), albeit the increase in clonal mutations in rNPC samples compared with pNPC samples was observed without statistical significance (P = 0.34), confirming the association between prior treatment and increased clonality (Fig. 1B).

We further explored whether mutational processes vary during tumor evolution between rNPC and pNPC. In these rNPC and pNPC samples, we identified seven robust mutational signatures using combined NMF clustering and correlation with the curated mutational signatures defined by the COSMIC database (cosine similarity >0.5). Specifically, there were high proportions (>40.0%) of samples harboring defective DNA mismatch repair signatures (signature 6) and smoking signatures (signature 4) in rNPC and pNPC, especially in rNPC; however, the defective DNA mismatch repair signatures and smoking signatures were significantly more prevalent in subclonal mutations than in clonal mutations (P < 0.05), suggesting that these mutational processes increase in prevalence later in tumor evolution. In contrast, apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like (APOBEC) signatures (signature 2 and signature 13) harbored by a small subset of samples were detected in both clonal and subclonal mutations with similar fractions (P > 0.05) in rNPC and pNPC, suggesting that APOBEC activity is more than just a transient event and does not simply represent a historical relic within the tumor genome that was active at only one point in time during the disease course (Fig. 1C; ref. 29).

rNPC showed a decrease in CNV size compared with pNPC (P < 0.05; Supplementary Table S11). Large-scale chromosome deletions at 9p21.3 (CDKN2A), 4p16.3, and 11q14.3 (BIRC3, ATM), and amplifications at 11q13.3 (CCND1), 3q26.1 (PIK3CA), 20q13.33, 19p13.3, 8q21.2, 16p11.2, 5p13.2, and 15q26.3 were common in both pNPC and rNPC samples (evaluated by GISTIC), suggesting that these regions or candidate genes were early CNV events. We also identified several significant deletion regions at 3P22.1 (BAP1), 1p36.12 (ARID1A), and 16q12.2 (CYLD) exclusively in rNPC samples, implying that the genes located in these regions are either late events or specifically required for rNPC (Supplementary Fig. S1; ref. 30).

MuSiC was used to identify SMGs that were mutated in at least two tumor samples with an FDR of less than 0.2 in rNPC and pNPC cases. As a result, 9 genes were found to be significantly mutated in rNPC. Confirming previous findings, the majority of SMGs were well-established NPC genes, with highly similar mutational frequencies across different cohorts. Notably, of these 9 genes, 4 genes were significantly mutated in both rNPC and pNPC, including TP53, NF-κB pathway genes, TRAF3, CYLD, and NFKBIA. In addition, NOTCH1, NAPA, RPL22, FAM135B, and NCKAP5L were significantly mutated in rNPC alone (Fig. 2A). The mutation sites of these somatic mutations in rNPC and pNPC are shown in Supplementary Fig. S2.

Notably, mutations in the NF-κB pathway genes TRAF3, CYLD, and NFKBIA were all clonal in rNPC, whereas 55.6% to 57.9% of them were clonal in pNPC (P = 0.021) and pNPC reported by Li and colleagues (P = 0.012; Fig. 2B; ref. 3). We have constructed wild-type and mutant TRAF3 (c.627_630del:p.H209fs, c.635_636del:p.T212fs, c.A1144G:p.M382V, c.G1190A:p.G397E, and c.1402_1404del:p.468_468del), mutant NFKBIA (c.428_429insTTTCG:p.R143fs), and mutant CYLD (c.475_484del:p.S159fs, c.C562T:p.Q188X, c.G1048T:p.E350X, and c.C1103A:p.S368X), and transiently transfected them into SUNE2 and 293T cells. The wild-type TRAF3, NFKBIA, and CYLD led to lower nuclear NF-κB expressions and constitutive inactivation of NF-κB signals in SUNE2 and 293T cell lines, whereas most mutant TRAF3, NFKBIA, and CYLD resulted in significantly higher nuclear NF-κB expression and constitutive activation of NF-κB signals in SUNE2 and 293T cell lines (Supplementary Fig. S3). In addition, mutations in putative tumor suppressor genes (NOTCH1 and RPL22) were all clonal in rNPC, which may reflect drivers during rNPC tumorigenesis. Only insertions and deletions (InDels) were identified somatic mutations in ribosomal protein RPL22 in rNPC (c.56_63del:p.L 19fs; c.44dupA:P.K15fs). Previous studies revealed that monoallelic inactivation of RPL22 enhances development of thymic lymphoma by activation of NF-κB (31). Interestingly, mutations in FAM135B and NCKAP5L were partially clonal in rNPC. FAM135B has been demonstrated as one tumor-associated gene in esophageal squamous carcinoma (32). However, NCKAP5L has not been previously linked to cancer (Fig. 2C). Furthermore, we conducted IHC staining to examine the expression of these specific SMGs in 148 rNPC samples and 122 pNPC samples and found that all of these specific SMGs were expressed in both rNPC and pNPC samples. Moreover, there were similar expression levels in these specific SMGs between rNPC and pNPC samples, except for NOTCH1 and NCKAP5L (Supplementary Fig. S4).

The top 50 genes from MuSiC ranking in rNPC were used for pathway and KEGG analysis, which identified several pathways with enriched somatic mutations, including the NF-κB pathway, PI3K-AKT, JAK-STAT, and cell cycle (Fig. 3A).

We identified 15 altered genes in 14 cases with rNPC (14/55, 25.5%) affecting PI3K-AKT, whereas altered genes in PI3K-AKT were detected in 6 of 44 pNPC cases (13.6%), 8 of 69 pNPC cases (11.6%) reported by Li and colleagues (3), and 9 of 50 pNPC cases (18.0%) reported by Zheng and colleagues (4). The fraction of clonal mutations associated with PI3K–AKT pathway in rNPC was similar to that in pNPC (57.1% vs. 50.0%, P = 0.730), that in pNPC reported by Li and colleagues (57.1% vs. 50.0%, P = 0.730; ref. 3), and that in pNPC reported by Zheng and colleagues (57.1% vs. 36.3%, P = 0.264; Fig. 3B; ref. 4). Overall, 42.9% to 63.7% of mutations in the PI3K–AKT pathway were subclonal in rNPC and pNPC, suggesting that these mutations could be linked to tumor development and progression. In agreement with the results in pNPC reported by Lin and colleagues (2), this category of alterations showed strong associations with poor OS in univariate but not multivariate analysis in rNPC (Fig. 3D; Supplementary Table S12), also indicating that dysregulation of PI3K-AKT might contribute to the aggressiveness of rNPC. The presence of mutations that activate the PI3K–AKT pathway and contribute to carcinogenesis has engendered much interest in inhibitors of this axis (33). Notably, with the exception of ERBB2 and AKT3, for every gene that has been linked with a targeted therapy approach, a clonal mutation was identified in at least one tumor in rNPC (Fig. 3C).

We identified 15 altered genes in 22 cases with rNPC (22/55, 40.0%) affecting the NF-κB pathway, whereas altered genes in the NF-κB pathway were detected in 16 of 44 pNPC cases (36.4%), 28 of 69 pNPC cases (40.1%) reported by Li and colleagues (3), and 14 of 50 pNPC cases (28.0%) reported by Zheng and colleagues (4), indicating a broad and important influence of NF-κB signaling in the occurrence and progression of NPC (Fig. 3E and F). Importantly, a significant increase in the number of clonal mutations in genes associated with NF-κB in rNPC was observed compared with those in pNPC (88.5% vs. 57.8%, P = 0.018), pNPC reported by Li and colleagues (88.5% vs. 54.5%, P = 0.005; ref. 3), and pNPC reported by Zheng and colleagues (88.5% vs. 50.0%, P = 0.003; Fig. 3B; ref. 4). Notably, 88.5% of mutations in genes in the NF-κB pathway were clonal in rNPC.

From the seven paired pNPC and rNPC samples, we identified 6 altered genes in 5 cases with rNPC (5/7, 71.4%) affecting the NF-κB pathway and 5 altered genes in 4 cases with pNPC (4/7, 57.1%). In addition, 14 altered genes were identified in 5 cases with rNPC (5/7, 71.4%) affecting PI3K–AKT pathway and 14 altered genes were identified in 7 cases with pNPC (7/7, 100.0%). For these 7 paired pNPC and rNPC samples, variants were categorized into four groups during recurrence based on the clonality in both rNPC and the pNPC tumors: (i) variants that were selected (“selected” variants, defined as clonal in recurrent tumors but subclonal or not found in primary tumors); (ii) variants that were novel (“novel” variants, defined as subclonal in recurrent tumors but not found in primary tumors); (iii) variants that were founding (“founding” variants, defined as clonal in both recurrent and primary tumors); and (iv) variants that were unselected (“unselected” variants, defined as not found in recurrent tumors but clonal or subclonal in primary tumors). According to this classification, selected and founding mutants might contain drivers of local relapse in NPC. Interestingly, we found that the proportion of “selected” and “founding” variants in the NF-κB pathway (44.4%, 4/9) in patients with NPC was higher than that in the JAK–STAT pathway (0%, 0/12, P = 0.023), PI3K-AKT (15.4%, 4/26, P = 0.162), or cell cycle (16.7%, 2/12, P = 0.331; Fig. 4A). In the NF-κB pathway, nine mutants were observed among the pNPC and rNPC samples in these 7 patients. Among these nine mutants, the CCF values of five mutants were increased in the rNPC samples compared with the pNPC samples, whereas the CCF value of one mutant in rNPC was equivalent to that of one mutant in pNPC. Interestingly, TRAF3 (c.627_630del:p.H209fs) and NFKBIA (c.700_701insGT:p.S234fs) were both observed in paired pNPC and rNPC samples in 2 patients (PP06 and PP04). TRAF3 (c.627_630del:p.H209fs) was clonal in both pNPC and rNPC samples in PP06, suggesting that the patient harboring this mutant might be prone to experience local recurrence. NFKBIA (c.700_701insGT:p.S234fs) was clonal in the rNPC sample, but subclonal in pNPC sample in PP04, suggesting that the treatment selects and enriches this mutation as clonal, potentially contributing to local recurrence. BIRC3 (c.G1219A:p.G407R) and PLCG2 (c.G1701C:p.E567D) were clonal in the rNPC samples, but not observed in pNPC samples in PP01 and PP02, suggesting that these two mutations were de novo and became clonal attributed to treatment including radiotherapy and chemotherapy, potentially contributing to local recurrence. These two mutants might also be present but under detection limit in pNPC samples. In contrast, PRKCQ (c.G43A:p.G15R), PLCG1 (c.2581-2→AGATGTATTTGTCCCAT), and PARP1 (c.C14T:p.S5L) were subclonal in the pNPC samples but not detected in the rNPC samples in PP01, PP03, and PP06 patients. However, the CCF values of these three mutants were also much low in the pNPC samples (0.02–0.06; Fig. 4B).

Given the strong tendency for mutations in driver genes to be clonal compared with mutations in noncancer genes (8), we hypothesized that mutations within the NF-κB pathway arise earlier during the development of rNPC compared with those in pNPC and may contain drivers of local relapse in NPC. According to previous studies (3, 4, 34, 35), somatic mutations were the most critical genomic force driving NF-κB activation in NPC and other cancers, including HPV(+) head and neck squamous cell carcinoma, lymphoma etc. Indeed, we have demonstrated that the amount of NF-κB translocated to the cell nucleus in samples with clonal NF-κB mutants was significantly higher than those in such samples with subclonal NF-κB mutants (Fig. 4D). Therefore, we conducted IHC staining on 237 pNPC and 148 rNPC tissues to explore the difference in the translocation of NF-κB to the cell nucleus between pNPC and rNPC (Fig. 4C). Importantly, the nuclear level of NF-κB protein was significantly greater in rNPC than in pNPC, with a median number of 10.0% (IQR: 0%–40%) and 5% (IQR: 0%–10%; P = 0.0016; Fig. 4E). Furthermore, among the 237 patients with pNPC, the nuclear expression of NF-κB was significantly greater in 47 pNPC with locoregional relapse than in 190 pNPC without such relapse, with a median number of 10.0% (IQR: 5%–20%) and 5% (IQR: 0%–10%; P < 0.001; Fig. 4F). These data are consistent with our hypothesis that NF-κB pathway activation driven by clonal mutations promotes local relapse in NPC and suggest that the nuclear accumulation of NF-κB serves as a potential biomarker to predict tumor recurrence in NPC.

Next, we explored the nuclear accumulation of NF-κB and its association with clinical parameters using IHC in the 237 patients with pNPC. No significant correlation was observed between the nuclear level of NF-κB and gender, age, T stage, N stage, or treatment regimens. Notably, the nuclear level of NF-κB was significantly correlated with shorter LRRFS (5-year LRRFS, 83.3% vs. 64.0%, P = 0.002; Fig. 5A and B). Moreover, multivariate analyses revealed that low nuclear expression of NF-κB and concurrent chemotherapy were independent favorable prognostic indicators for LRRFS (Supplementary Table S13). These results suggest that nuclear abundance of NF-κB is clinically relevant to the locoregional relapse of pNPC and that the nuclear amount of NF-κB may be used as an independent prognostic biomarker of LRRFS in patients with pNPC.

Moreover, we found that adding concurrent chemotherapy to patients with high nuclear expression of NF-κB resulted in improved control of locoregional relapse (Fig. 5C). In contrast, patients with lower expression did not benefit from combined concurrent chemotherapy (Fig. 5D). These findings suggest that nuclear accumulation of NF-κB might be effective in stratifying patients who are at a high risk of locoregional relapse and likely to benefit from concurrent chemotherapy.

We next tested whether inhibition of the activity of NF-κB regulates radiosensitivity and chemosensitivity in NPC cells in vitro. Two NF-κB inhibitors [JSH-23 and ammonium pyrrolidine dithiocarbamate (PDTC)] that could effectively inhibit the role of NF-κB and block NF-κB translocation to the nucleus were examined (36, 37).

DNA double-strand breaks (DSB) are major lesions induced by irradiation (38), and the DSB repair capacity is closely related to radiosensitivity (39). We evaluated the DNA damage response induced by 2 Gy of irradiation in NPC cells. Specifically, we performed immunofluorescence assays at 0, 2, and 6 hours after irradiation to examine the phosphorylation of H2A.X at Ser139 [γH2AX, a biomarker of DSBs (40)]. Importantly, inhibition of NF-κB with JSH-23 and PDTC accelerated the absorption of γH2AX foci in SUNE1 and HONE1 cells induced by irradiation (Fig. 6A and B). These data suggested that inhibition of NF-κB enhances radiosensitivity by regulating DNA DSB repair.

An MTT assay was performed after SUNE1 and HONE1 cells treated with or without JSH-23 and PDTC were treated with increasing doses of cisplatin, a commonly used chemotherapeutic drug for patients with NPC. As shown in Fig. 6C, SUNE1 and HONE1 cells with JSH-23 and PDTC displayed decreased viability in the presence of varying concentrations of cisplatin. In addition, staining with Annexin V and PI showed that inhibition of NF-κB with JSH-23 and PDTC promoted apoptosis in cisplatin-treated NPC cells (Fig. 6D). Collectively, the results suggest that inhibition of NF-κB activity enhances the sensitivity of NPC cells to cisplatin.

In Fig. 6E–G, we found that inhibition of NF-κB activity with JSH-23 enhanced the anti-NPC effect of radiotherapy, whereas radiotherapy alone did not effectively control tumor growth. Similarly, cisplatin chemotherapy alone did not show efficacy in restraining NPC growth, whereas cisplatin plus JSH-23 markedly inhibited tumor proliferation (Fig. 6H–J). These data suggested that inhibition of NF-κB activity with JSH-23 enhances sensitivity of tumor to either chemotherapy or radiotherapy in NPC.

To the best of our knowledge, this is the first study to comprehensively describe the genomic characteristics of rNPC based on a large-scale rNPC cohort using high-coverage WES and WGS and to compare them with the genomic characteristics of pNPC using our own and previously reported sequencing data. The major findings in this study can be summarized as follows:

rNPC and pNPC tissues had similar mutational burdens; however, the number of clonal mutations was increased in rNPC samples. The number of clonal mutations in NF-κB pathway–associated genes was significantly higher in rNPC than in pNPC. The NF-κB mutational clonality was selected and/or enriched during NPC recurrence. High nuclear NF-κB level was an independent negative prognostic marker for LRRFS in pNPC. Finally, we functionally validated that inhibition of NF-κB enhanced both radiosensitivity and chemosensitivity both in vivo and in vitro.

Precision medicine will ultimately require not only a catalog of cancer genes and mutational processes but also an understanding of their spatial and temporal dynamics during a tumor's evolution (8). Notwithstanding cooperative subclonal interactions (41, 42), targeting clonally dominant truncal somatic events may represent one therapeutic approach to optimize tumor control (43). Therefore, defining whether driver mutations are found in all or only a subset of cancer cells is likely to become increasingly relevant in cancer treatment development. Previous genomic analyses of NPC mostly focused on the mutated genes and did not distinguish the clonal status of mutated genes, especially SMGs. Although the majority of SMGs were the same in pNPC and rNPC, the clonality of these genes was vastly different between pNPC and rNPC. Notably, mutations in the NF-κB pathway genes TRAF3, CYLD, and NFKBIA were all clonal in rNPC; however, 55.6% to 57.9% of them were clonal in pNPC (P < 0.05).

Furthermore, we demonstrated that rNPC was significantly clonally enriched in mutations in NF-κB. In addition, previous studies demonstrated that somatic mutations were the most critical genomic force driving NF-κB activation in NPC, HPV(+) head and neck squamous cell carcinoma, lymphoma, etc (3, 4, 34, 35). Therefore, we reasoned that NF-κB activation induced by clonal mutations in this pathway potentially plays a key role in the recurrence of NPC after radical treatment. The activation of the NF-κB pathway was due to the release and translocation of NF-κB to the nucleus to drive gene transcription of candidates (44). Therefore, we conducted IHC staining on pNPC tissues and found that the translocation of NF-κB to the cell nucleus in pNPC with locoregional relapse was significantly greater than in pNPC without locoregional relapse, further supporting this hypothesis.

Only a small number of studies have been aimed at finding molecular biomarkers to predict LRRFS in patients with NPC because a relatively high local control rate was observed in pNPC. However, approximately 10% to 20% of newly diagnosed patients with NPC experienced recurrence after radical radiotherapy (45–47). Once patients experienced recurrence, the survival outcomes and quality of life were poor; thus, the effective control of locoregional relapse deserves increased attention, especially in these patients with high risk of locoregional relapse. Interestingly, we demonstrated a significant correlation between the nuclear expression of NF-κB and locoregional relapse in patients with pNPC. High nuclear expression of NF-κB was also demonstrated as an independent negative prognostic factor of LRRFS in pNPC. In addition, we also demonstrated that concurrent chemotherapy resulted in improved control of locoregional relapse in patients with high nuclear expression of NF-κB. Therefore, the nuclear expression of NF-κB might correctly predict the LRRFS of patients with pNPC and aid clinicians in choosing appropriate patients for concurrent chemotherapy. Indeed, the better survival in patients with concurrent chemotherapy in high nuclear expression of NF-κB could be confounded by many other factors. Therefore, prospective, randomized trials and further studies are strongly needed to validate the result.

NF-κB is increasingly recognized as a crucial player in many steps of cancer initiation and progression (48). Indeed, in radioadaptive resistance, the prosurvival network has been reported to be initiated by NF-κB. Activation of NF-κB has also been reported to reduce the therapeutic effect of radiotherapy in cancer cells (49). Moreover, blocking NF-κB activation increases the apoptotic response, decreases growth and clonogenic survival, and enhances radiosensitivity in several human cancer cells (50–52). Notably, we also demonstrated that inhibition of NF-κB could enhance radiosensitivity and chemosensitivity both in vitro and in vivo, which might explain the fact that excessive activation of NF-κB was attributed to the recurrence of NPC due to radioresistance and chemoresistance in tumor cells. In addition, this finding might also explain the fact that concurrent chemotherapy could be needed for patients with high nuclear expression of NF-κB because of the effect of cytotoxic drugs, such as cisplatin, on increasing the radiosensitivity of tumor cells.

In summary, our results demonstrate that NF-κB pathway activation by clonal mutations plays an important role in promoting the recurrence of NPC. Moreover, nuclear accumulation of NF-κB is a prominent biomarker for the prediction of LRRFS.

No potential conflicts of interest were disclosed.

Conception and design: R. You, Q. Li, J.-Q. Wang, A.-H. Zhuang, Y.-X. Zeng, M.-Y. Chen

Development of methodology: R. You, Q. Li, X.-L. Zhang, G.-P. He, C.-Y. Wu, Z.-X. Zuo, M.-Y. Chen

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R. You, Y.-P. Liu, Q. Li, T. Yu, G.-P. He, Q. Yang, Y.-N. Zhang, Y.-L. Xie, R. Jiang, C. Cui, Y.-J. Hua, R. Sun, Y. Sun, M.-S. Zeng, M.-Y. Chen

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R. You, Y.-P. Liu, D.-C. Lin, Q. Li, T. Yu, X. Zou, M. Lin, X.-L. Zhang, C.-Y. Wu, C. Cui, Y. Wang, Y.-J. Hua, Z.-X. Liu, X.-F. Zhu, C.-N. Qian, L. Feng, M.-Y. Chen

Writing, review, and/or revision of the manuscript: R. You, Y.-P. Liu, Q. Li, T. Yu, Y.-N. Zhang, Y.-L. Xie, C.-Y. Wu, C. Zhang, C. Cui, A.-H. Zhuang, R. Sun, J.-P. Yun, Z.-X. Zuo, H.-Q. Mai, M.-S. Zeng, M.-Y. Chen

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y.-P. Liu, T. Yu, X. Zou, J.-Q. Wang, J.-P. Yun, T.-B. Kang, H.-Q. Mai, Y. Sun, M.-Y. Chen

Study supervision: G.-F. Guo, J.-P. Yun, H.-Q. Mai, M.-S. Zeng, M.-Y. Chen

Funding was provided by the National Natural Science Foundation of China (nos. 81572912, 81772895, 81572848, and 81874134), the Major Project of Sun Yat-Sen University for the New Cross Subject, the Special Support Program for High-level Talents in Sun Yat-Sen University Cancer Center (to M.-Y. Chen), Guangdong Province Science and Technology Development Special Funds (Frontier and Key Technology Innovation Direction—Major Science and Technology Project; no. 703040078088), and Guangzhou Science and Technology Planning Project—Production and Research Collaborative Innovation Major Project (no. 201505012235268).

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