Abstract

   The aim of this study was to use whole‐exome sequencing to derive a
   molecular classifier for nasopharyngeal carcinoma (NPC) and evaluate
   its clinical performance. We performed whole‐exome sequencing on 82
   primary NPC tumors from Sun Yat‐sen University Cancer Center (Guangzhou
   cohort) to obtain somatic single‐nucleotide variants, indels, and copy
   number variants. A novel molecular classifier was then developed and
   validated in another NPC cohort (Hong Kong cohort, n = 99). Survival
   analysis was estimated by the Kaplan‐Meier method and compared using
   the log‐rank test. Cox proportional hazards model was adopted for
   univariate and multivariate analyses. We identified three prominent NPC
   genetic subtypes: RAS/PI3K/AKT (based on RAS, AKT1, and PIK3CA
   mutations), cell‐cycle (based on CDKN2A/CDKN2B deletions, and CDKN1B
   and CCND1 amplifications), and unclassified (based on dominant
   mutations in epigenetic regulators, such as KMT2C/2D, or the Notch
   signaling pathway, such as NOTCH1/2). These subtypes differed in
   survival analysis, with good, intermediate, and poor progression‐free
   survival in the unclassified, cell‐cycle, and RAS/PI3K/AKT subgroups,
   respectively, among the Guangzhou, Hong Kong, and combined cohorts
   (n = 82, P = 0.0342; n = 99, P = 0.0372; and n = 181, P = 0.0023;
   log‐rank test). We have uncovered genetic subtypes of NPC with distinct
   mutations and/or copy number changes, reflecting discrete paths of NPC
   tumorigenesis and providing a roadmap for developing new prognostic
   biomarkers and targeted therapies.

   Keywords: copy number variants, indels, molecular classifier,
   nasopharyngeal carcinoma, single‐nucleotide variants

1. INTRODUCTION

   Nasopharyngeal carcinoma (NPC) has a unique geographical distribution
   that is distinct from other head and neck cancers. Currently, the
   tumor‐lymph node‐metastasis (TNM) staging system is the key clinical
   tool for prognostication, risk stratification, and making treatment
   decisions. Although patients of the same TNM stage receive similar
   treatments, their clinical outcomes vary greatly. Similarly, a previous
   study found comparable relative survival rates between undifferentiated
   and differentiated subtypes according to the World Health Organization
   histological classification.[40]1 Thus, current staging and
   histological classification systems are insufficient for predicting
   patient survival. One current hypothesis is that differences in
   prognosis and treatment efficacy might be attributed to biological
   heterogeneity.

   Proper classification is essential for physicians to properly assign
   treatments and evaluate clinical outcomes. In the past 2 decades,
   researchers have found gene expression profiling is altered between
   cancer patients and healthy controls.[41]2, [42]3, [43]4 Despite its
   early promise as a diagnostic and prognostic tool, gene expression
   profiling remains cost‐prohibitive and challenging to implement in a
   clinical setting. In NPC patients, several studies have reported
   molecular classifications only based on the expression of
   protein‐coding genes and microRNAs (7‐9), which have not been widely
   applied in clinical settings, even though they are relevant to
   prognosis. Recently, molecular profiling has been achieved by
   high‐throughput analyses, making molecular classifications based on
   genetic lesions more comprehensive and prevalent in clinical cancer
   management. In colorectal cancer, BRAF V600E and activating KRAS
   mutations are associated with metastasis, leading to poor
   survival.[44]5 The Cancer Genome Atlas has proposed a molecular
   classification mainly according to genomic mutations, amplifications,
   and fusion genes that divide gastric cancer into four subtypes,
   providing a roadmap for patient stratification and trials of targeted
   therapies.[45]6 A large‐scale international database demonstrated that
   four consensus molecular subtypes with distinguishable features
   including mutations and copy number changes facilitated clinical
   treatment in colorectal cancer.[46]7 Notably, the advent of large‐scale
   DNA molecular profiling has been helpful to identify novel molecular
   targets that can be applied to the treatment of particular cancer
   patients.[47]8, [48]9 Thus, developing a molecular classifier from DNA
   high‐throughput sequencing data that can identify dysregulated pathways
   and candidate drivers in NPC is an urgent priority.

   To gain further insight into the genetic heterogeneity of primary NPC
   and to establish a DNA‐based molecular classifier capable of performing
   multiple‐gene classification for prognostic and therapeutic
   stratification, we performed whole‐exome sequencing (WES) on 82
   formalin‐fixed paraffin‐embedded NPC tumors as a discovery cohort
   (Guangzhou NPC Cohort [GZNPC]). The WES data of 99 external NPC cases
   (Hong Kong NPC Cohort [HKNPC]) were used as an independent validation
   cohort.[49]10 This study revealed that a molecular classifier derived
   from somatic single nucleotide variants (SNVs), indels, and copy number
   variants (CNVs) could better illustrate NPC tumorigenesis, and thus,
   could be used as a tool to explore different therapeutic strategies.

2. MATERIALS AND METHODS

2.1. Clinical specimens

   Between July 2007 and December 2012, we obtained 82 primary NPC tumor
   tissues and corresponding blood samples from the Department of
   Pathology and Biobank at Sun Yat‐sen University Cancer Center (GZNPC
   cohort, Guangzhou, China). All specimens were independently reviewed by
   two pathologists to determine World Health Organization histological
   classification and tumor cellularity. NPC specimens with >50% tumor
   cellularity were used for sectioning, nucleic acid extraction, and
   library preparation. Additionally, 99 NPC patients from a Hong Kong
   study were used as the validation cohort (HKNPC).[50]10 Detailed
   clinical data of all patients are summarized in Table [51]1. This study
   was approved by the institutional review board of Sun Yat‐sen
   University Cancer Center (RDDA2019001009).

Table 1.

   Clinical characteristics of the surveyed NPC patients including 82
   patients from Guangzhou and 99[52]^a patients from Hong Kong,
   respectively
     Discovery cohort (n = 82, %) External validation cohort (n = 99, %)
   Total (n = 181, %) P value
   Age at diagnosis (years)
   Median 47 49 48 0.0721[53]^b
   Range 19‐71 23‐80 19‐80
   OS, months
   Median 48 56 50 0.1029[54]^b
   Range 7‐94 2‐122 2‐122
   PFS, months
   Median 31 22 29 0.3034[55]^b
   Range 1‐63 2‐96 1‐96
   Sex
   Male 62 (75.6) 71 (73.2) 133 (74.3) 0.7346[56]^c
   Female 20 (24.4) 26 (26.8) 46 (25.7)
   Unknown 0 2 2
   Smoking status
   Nonsmoker 45 (61.6) 47 (51.7) 92 (57.9) 0.0809[57]^c
   Smoker 28 (38.4) 44 (48.3) 72 (42.1)
   Unknown 9 8 17
   Clinical stage
   Early stage (I + II) 7 (8.5) 25 (26.1) 32 (18.0) 0.0030[58]^c
   Advanced stage (III + IV) 75 (91.5) 71 (73.9) 146 (82.0)
   Unknown 0 3 3
   WHO classification
   NKUC 73 (89.0) 91 (93.8) 164 (91.6) 0.0205[59]^c
   NKDC 7 (8.5) 0 7 (3.3)
   KSCC 2 (2.5) 6 (6.2) 8 (5.1)
   Unknown 0 2 2
   PFS rate (5 year, 95% CI) (%) 45.4 (26.4‐62.7) 41.5 (30.1‐52.5) 46.4
   (37.8‐54.5)
   OS rate (5 year, 95% CI) (%) 78.2 (63.5‐87.5) 65.9 (53.7‐75.5) 71.8
   (62.9‐78.9)
   [60]Open in a new tab

   NPC, nasopharyngeal carcinoma; OS, overall survival; PFS,
   progression‐free survival; CI, confidence interval; NKUC,
   nonkeratinizing undifferentiated carcinoma; NKDC, nonkeratinizing
   differentiated carcinoma; KSCC, keratinizing squamous cell carcinoma.
   ^a

   All NPC cases are from Asia.
   ^b

   Wilcoxon rank sum test.
   ^c

   Pearson's x ^2‐test.

2.2. The whole‐exome sequencing

   Details of DNA isolation methods are provided in the online Data
   [61]S1. DNA library preparation for NPC tumors and matched controls was
   performed according to the Agilent SureSelect^XT protocol (Santa Clara,
   CA) with minor modifications. Briefly, 2 μg of tumor DNA and 200 ng of
   control DNA were fragmented by ultrasonication (M220; Covaris, Woburn,
   MA), and fragments were captured using SureSelect^XT Human All Exon
   V5+UTRs 75M (Cat no. 5190‐6214; Agilent). Quantities and sizes of the
   libraries were determined using a Qubit fluorescence detector (Life
   Technologies, Carlsbad, CA) and an Agilent Bioanalyzer 2100,
   respectively. Finally, WES libraries were hybridized to an Illumina
   HiSeq PE Cluster Kit and SBS kit v4 for enrichment and were sequenced
   by 150 paired‐end read lengths on an Illumina Hiseq 1500 sequencing
   platform including a dual eight‐base index barcode according to the
   manufacturer's protocols (Illumina, San Diego, CA). The WES raw
   sequence data reported have been deposited in the Genome Sequence
   Archive, Beijing Institute of Genomics (BIG), Chinese Academy of
   Sciences,[62]11, [63]12 under accession number CRA001397 that are
   publicly accessible at [64]http://bigd.big.ac.cn/gsa/s/29XtNNXW.

2.3. Variant calling

   Analyses of WES data were performed to identify somatic alterations in
   each tumor, including somatic SNVs, indels, and CNVs. DNA from
   peripheral blood were used as references to filter germline variants.