Abstract

   Nasopharyngeal carcinoma (NPC) is a malignant head and neck cancer type
   with high morbidity in Southeast Asia, however the pathogenic mechanism
   of this disease is poorly understood. Using integrative
   pharmacogenomics, we find that NPC subtypes maintain distinct molecular
   features, drug responsiveness, and graded radiation sensitivity. The
   epithelial carcinoma (EC) subtype is characterized by activations of
   microtubule polymerization and defective mitotic spindle checkpoint
   related genes, whereas sarcomatoid carcinoma (SC) and mixed
   sarcomatoid-epithelial carcinoma (MSEC) subtypes exhibit enriched
   epithelial-mesenchymal transition (EMT) and invasion promoting genes,
   which are well correlated with their morphological features.
   Furthermore, patient-derived organoid (PDO)-based drug test identifies
   potential subtype-specific treatment regimens, in that SC and MSEC
   subtypes are sensitive to microtubule inhibitors, whereas EC subtype is
   more responsive to EGFR inhibitors, which is synergistically enhanced
   by combining with radiotherapy. Through combinational chemoradiotherapy
   (CRT) screening, effective CRT regimens are also suggested for patients
   showing less sensitivity to radiation. Altogether, our study provides
   an example of applying integrative pharmacogenomics to establish a
   personalized precision oncology for NPC subtype-guided therapies.

   Subject terms: Cancer, Drug screening, Preclinical research,
   Translational research, Oncology
     __________________________________________________________________

   Nasopharyngeal carcinoma (NPC) is a malignant cancer type with high
   morbidity in Asia, and its current molecular classification is
   insufficient to predict therapy outcomes. Here the authors explore NPC
   subtype-specific response to therapy with a pharmacogenomics strategy
   integrating genomics and drug response of patient-derived organoids.

Introduction

   Nasopharyngeal carcinoma (NPC) is a type of malignant tumor that is
   commonly found in specific geographical locations including Southeast
   Asia, North Africa, southern provinces of China, Hong Kong, and
   Taiwan^[68]1–[69]3. In 2015, 60,600 new NPC cases were identified, and
   34,100 patients died in China^[70]4. Many risk factors are involved in
   NPC, including environmental factors, Epstein–Barr virus (EBV)
   infection, smoking, diet, and personal lifestyle of peoples,
   etc.^[71]5, yet how these factors contribute to NPC formation is poorly
   understood. Genomic variations and familial risks are other important
   causes for NPC, with more young people being affected by NPC than other
   cancer types^[72]6. Unlike other head and neck cancers, the
   asymptomatic nature of NPC is a major challenge, which hinders the
   early diagnosis of this disease. Therefore, many patients are diagnosed
   at advanced stages, which reduces the patient’s overall survival
   (OS)^[73]6,[74]7.

   Earlier studies proposed an array of genetic factors and genetic
   aberrations leading to the development of NPC, including NF-κB pathway
   activating mutations, chromatin modification-related mutations,
   ERBB-PI3K signaling activating mutations, etc.^[75]5,[76]8–[77]11.
   However, many questions, such as those regarding the major causative
   factors, the key driving pathways, druggable genetic targets in NPC,
   especially in different histological subtypes, are still unanswered
   clearly.

   According to the current World Health Organization (WHO) classification
   system, NPC is classified into three major histological subtypes:
   keratinizing squamous cell carcinoma (KSCC), non-keratinizing
   carcinoma, and basaloid squamous cell carcinoma. Non-keratinizing
   tumors are further subcategorized as non-keratinizing undifferentiated
   carcinoma (NKUC) and non-keratinizing differentiated carcinoma
   (NKDC)^[78]12. Although the WHO subtype system is the most common form
   of NPC clinical classification, an increasing number of clinicians
   realize that the current WHO classification is insufficient for
   predicting chemotherapy and radiotherapy (RT) outcomes^[79]13,[80]14.
   The prognosis does not differ significantly between NKUC and NKDC,
   which are the two major subtypes of NPC and account for ~95% of all
   cases in China^[81]13. A new prognostic histopathologic classification
   system of NPC has emerged that classifies NPC into four subtypes based
   on morphologic characteristics: epithelial carcinoma (EC), sarcomatoid
   carcinoma (SC), mixed sarcomatoid-epithelial carcinoma (MSEC), and
   squamous cell carcinoma (SCC), and these NPC subtypes could be linked
   to prognosis^[82]13. However, further information about the genomic
   features and therapeutic response among different subtypes is worthy of
   investigation.

   RT is established as the definitive treatment for nonmetastatic NPC at
   an early stage, which leads to favorable clinical and survival outcomes
   with a 5-year OS rate of 87.3–93%, for stage I NPC
   patients^[83]15–[84]18. However, due to the intrinsic invasiveness and
   asymptomatic nature of the disease, the majority of NPC patients
   (60–70%) are diagnosed at an advanced stage, with local spread or
   regional lymph node metastasis^[85]5. For NPC patients with
   recurrent/metastatic tumor, the outcome is very poor, with a median OS
   of ~20 months^[86]19, and chemoradiotherapy (CRT) is the
   standard-of-care treatment at this stage as recommended by the National
   Comprehensive Cancer Network guidelines (v2.2018). If the cancer cells
   have spread to distant organs, chemotherapy is the only
   option^[87]19,[88]20. The commonly used chemotherapeutics for treating
   NPC includes cisplatin, fluorouracil, docetaxel, paclitaxel,
   gemcitabine, capecitabine, irinotecan, doxorubicin, vinorelbine,
   carboplatin and oxaliplatin. Several studies have indicated that CRT
   can significantly improve therapeutic outcome compared with RT
   alone^[89]21,[90]22. To investigate new strategies identifying the
   personalized optimal CRT and chemotherapy regimens is promising to
   improve the prognosis of NPC patients. Furthermore, since only a few
   chemical drugs have been approved for NPC, to explore more therapeutic
   drugs that can be used for the treatment of NPC is necessary.

   Recent studies have employed the patient-derived organoid (PDO) culture
   system for drug screening against various cancers, including
   breast^[91]23, colorectal^[92]24, gastric^[93]25,
   gastrointestinal^[94]26, prostate^[95]27, esophageal^[96]28,
   liver^[97]29, pancreatic^[98]30, and bladder cancers^[99]31 etc.
   Vlachogiannis et al., had demonstrated the good potential of this
   system to accurately predict the clinical responses of cancer drugs
   using organoids derived from metastatic, heavily pretreated colorectal
   and gastroesophageal cancer patients. Their data indicated that the
   PDOs showed as high as 88% positive predictive value and 100% negative
   predictive value in cancer patients^[100]26. Similar high accuracy in
   predicting clinical outcomes were also revealed by other
   studies^[101]32–[102]34.

   In this study, to explore the application of the PDO system in
   screening chemotherapy drugs that could be used alone or in CRT for
   NPC, we have conducted a pharmacogenomics-based precision medicine
   approach by integrating genomics with a drug sensitivity test on PDOs
   to comprehensively investigate effective regimens for individual
   patients. Our integrative approach uncovers potential subtype-related
   and patient-specific strategies for the use of drugs and CRT
   combinations against this deadly disease.

Results

Clinical information and histological subtyping of NPCs

   A total of 106 NPC tumors were collected from the hospital with routine
   records (Supplementary Table [103]1; Supplementary Data [104]1,
   [105]2), and we further studied their histological features. All NPC
   patients were of Chinese origin with a median diagnosed age of 48
   years. EBV status was examined on 40 tumor samples, the results
   indicated that 36 samples showed strongly positive, and 4 samples were
   low positive (Supplementary Data [106]1). With characterization on WHO
   subtypes, 73.58% tumors were NKUC, 23.58% were NKDC, and 2.83% were
   KSCC (Supplementary Table [107]1; Supplementary Data [108]1). Based on
   the standard of the new NPC classification system^[109]13, all tumors
   were classified into four subtypes: (1) epithelial carcinoma (EC)
   (57/106, 53.77%), which is mainly composed of morphologically round
   epithelial cells; (2) sarcomatoid carcinoma (SC) (20/106, 18.87%),
   which contains large proportions of spindle sarcomatoid cells; (3)
   mixed sarcomatoid-epithelial carcinoma (MSEC) (26/106, 24.53%), which
   shares both features of EC and SC; and (4) squamous cell carcinoma
   (SCC) (3/106, 2.83%), which is characterized by keratinizing phenotype
   that is rarely found in other subtypes (Fig. [110]1a). More percent of
   distant metastasis was diagnosed in SC subtype patients (35%) than in
   EC subtype patients (12.28%) (Supplementary Data [111]1).

Fig. 1. Histological and molecular landscape of NPC subtypes.

   [112]Fig. 1
   [113]Open in a new tab

   a Representative histological view of four NPC subtypes. The EC subtype
   is characterized by round epithelial tumor cells, while the SC subtype
   is characterized by spindle sarcomatoid tumor cells. The MSEC subtype
   encompasses both round epithelial and spindle sarcomatoid tumor cells.
   SCC is a minor subtype of NPC characterized by a keratinizing
   phenotype. Scale bar, 100 μm. The H&E staining images are
   representatives of 106 tumors consist of 57 EC subtype, 20 SC subtype,
   26 MSEC subtype, and 3 SCC subtype tumors. b Immunohistochemical
   staining of NPC markers. The EC subtype is positive for the epithelial
   cell marker AE1/3 and negative for the sarcomatoid cell marker
   vimentin, whereas the SC subtype is positive for the sarcomatoid cell
   marker vimentin and negative to epithelial cell marker AE1/3. The MSEC
   subtype exhibited mixed pattern of both epithelial and sarcomatoid cell
   markers. Scale bar, 100 μm. The immunohistochemical images are
   representatives of 31 tumors consist of 12 EC subtype, 9 MSEC subtype,
   7 SC subtype, and 3 SCC subtype tumors. c Summary table of NPC marker
   expression levels among subtypes. The top panel shows levels of AE1/3
   positive cells in EC (n = 12), MSEC (n = 9), SC (n = 7), and SCC
   (n = 3) tumors. The bottom panel presents levels of vimentin-positive
   cells in EC (n = 12), MSEC (n = 9), SC (n = 7), and SCC (n = 3) tumors.
   The corresponding immunohistochemical images are demonstrated in
   Fig. 1b and Supplementary Fig. [114]1a. d Top recurrent protein coding
   SNVs identified from this cohort and represented in pathway wise,
   including cell cycle, NF-κB signaling, receptor tyrosine kinase (RTK),
   chromatin remodeling, apoptosis, microtubule polymerization, mitosis
   regulation, DNA repair and EMT/invasion. SNV mutational rate for each
   tumor is shown at the top of the panel. Information of subtype, sample
   type, WHO subtype, gender, age of diagnosis, tumor stage, node
   classification, metastasis and smoking status are shown at the bottom.
   Only paired tumor samples (n = 88) were assigned for SNV
   identification. e Detailed somatic variation types and amounts
   discovered from this NPC cohort. A total of 2662 somatic mutations
   included 2306 missense mutations, 191 nonsense mutations, 47 splice
   site, 51 frame shift deletion, 31 in frame deletion, 23 frame shift
   insertion, 8 in frame insertion, and 5 others (n = 88 tumors). f
   Overall mutational signature of this NPC cohort revealed C > A and
   C > T base substitutions are predominant mutational signature in NPC
   (n = 88 tumors).

   Next, we conducted the immunofluorescence assay using several
   antibodies against tumor subtype markers. The data indicated that the
   EC subtype extensively expressed pan-epithelial markers, including
   AE1/3, CK5/6, and P63, and the SC subtype was largely positive for
   vimentin, while MSEC exhibited a mixed pattern of both epithelial and
   sarcomatoid cell markers (Fig. [115]1b, c; Supplementary Fig. [116]1a,
   b). These data uncovered distinct molecular features among EC, SC, and
   MSEC subtypes. Notably, highly proliferating Ki67-positive tumor cells
   and CD3e-positive lymphocytes were found across all subtypes, with no
   clear difference among subtypes (Supplementary Fig. [117]1c).

Mutational landscapes of NPC

   To further investigate subtype-specific genomic features and
   therapeutic prognosis, we performed whole-exome sequencing (WES) to
   study the genomic landscape of NPC subtypes. A total of 2662 somatic
   mutations including 2306 missense mutations, 191 nonsense mutations, 82
   deletions, 31 insertions, and 52 other types of mutations were detected
   from 88 paired tumors by overlapping results from three different
   callers (MuTect2, Strelka2 and LANCET), and 2148 genes were affected
   (Fig. [118]1d, e; Supplementary Fig. [119]2a; Supplementary
   Data [120]3). Validation of candidate mutations with Sanger sequencing
   showed that a true positive rate of 100% was achieved (Supplementary
   Fig. [121]2b; Supplementary Data [122]4). The somatic mutation rate in
   NPC is relatively low compared to other types of cancers, with the
   somatic mutation rate of less than one per megabase (Supplementary
   Fig. [123]2c), which is consistent with previous
   studies^[124]35–[125]37. Averagely, we identified 30.3 somatic SNVs per
   sample. The recurred mutations presented in Fig. [126]1d revealed the
   oncogenic drivers of NPC, such as TP53 (mutational frequency at 8.0%),
   CYLD (10.2%), KMT2C (5.7%), NOTCH2 (6.8%), NFKBIA (4.5%), FBXW7 (4.5%),
   ARID1A (2.3%), PTEN (2.3%), and BAP1(2.3%). The mutation frequencies
   identified here were comparable with those reported by several previous
   studies (Supplementary Table [127]2)^[128]5,[129]8–[130]11, although
   they were, in general, lower than that of some other cancer
   types^[131]35. The recurrently mutated genes aggregated into signaling
   pathways involving in cell cycle, NF-κB signaling, receptor tyrosine
   kinase (RTK), chromatin remodeling, apoptosis, microtubule
   polymerization, mitosis regulation, DNA repair and EMT/invasion
   (Fig. [132]1d). Among these top mutated pathways, NF-κB signaling (CYLD
   (mutational frequency at 15.2% in EC), NFKBIA (8.7% in EC), TRAF3 (6.5%
   in EC), and RIPK2 (4.3% in EC)), mitosis regulation (WEE1 (4.3% in EC),
   NOTCH2 (8.7% in EC), and FBXW7 (8.7% in EC)), DNA repair (ATM (6.5% in
   EC) and NBN/NBS1 (4.3% in EC)), and microtubule polymerization (FMN2
   (4.3% in EC)) exhibited relatively high mutational frequencies in EC
   subtype, which indicated that they might be the contributive factors to
   subtype-specific oncogenic mechanisms.

   Mutational signatures analysis revealed that C > T base substitutions
   was the predominant signatures in NPC without obvious subtype
   difference (Fig. [133]1f; Supplementary Fig. [134]2d). The second
   frequent signature in our NPC cohort was the C > A transition, which
   was associated with smoking exposure (Supplementary
   Fig. [135]2e)^[136]38. Consistent conclusion was also demonstrated by
   previous report^[137]39. Among top frequent COSMIC signatures in NPC,
   signatures 2 and 13 were related to APOBEC family, signatures 6, 15,
   20, and 16 were related to DNA mismatch repair, signature 5 was of
   unknown aetiology, signatures 4 and 29 were due to tobacco, and
   signature 1 was associated with methylcytosine (Supplementary
   Fig. [138]2f). There was no significant difference on these NPC-related
   COSMIC mutational signatures among subtypes, but slightly higher APOBEC
   family and methylcytosine related signatures were observed in EC
   subtype than MSEC and SC subtypes, whereas tobacco associated
   signatures were higher in SC subtype than other subtypes (Supplementary
   Fig. [139]2g–k).

Driver pathways and networks revealed by copy number variation (CNV) analysis

   Further investigation on somatic CNVs revealed that frequent
   chromosomal deletions of Chr. 3p, 9p, 14q, 16q, and amplifications of
   3q, 8q, 12p, 12q, and 18q were the major features of NPC across all
   subtypes (Fig. [140]2a; Supplementary Fig. [141]3a), which suggested
   that such recurrent changes were critical genetic events leading to NPC
   tumorigenesis. The highest frequency of chromosome 3p deletion, locus
   of MST1R and BAP1, indicated that the inactivation of tumor suppressor
   genes in this chromosome might be an early event contributing to
   transformation from nasopharyngeal epithelium to NPC. There were no
   significant differences on overall chromosomal gain and loss
   frequencies among EC, MSEC and SC subtypes, except EC vs. SC on
   chromosomal gain frequencies (Fig. [142]2b). Age also acted as an
   important contributive factor during NPC progression. Accumulation of
   chromosomal abnormality was observed with increasing age, and several
   unique chromosomal variations presented in patients over 50 years of
   age, such as chromosomal gain of Chr. 3q and losses of Chr. 16q and 19p
   (Supplementary Fig. [143]3b).

Fig. 2. NPC cancer driver pathways and networks revealed by somatic CNVs.

   [144]Fig. 2
   [145]Open in a new tab

   a Global chromosomal gains (red) and deletions (blue) revealing
   frequent chromosomal deletions in Chr. 3p, 9p, 14q, 16q, and
   amplification of Chr. 3q, 8q, 12p, 12q, 18q as major feature of NPC
   (n = 99 tumors). b Average chromosomal gain (red) and loss (blue)
   frequencies among different NPC subtypes (n = 99 tumors). Violin plots
   show the kernel probability density of the data at different values.
   Inside boxplots represent median levels, first and third percentiles.
   Whiskers indicate 1.5× inter-quartile range (IQR) extending from the
   hinges. Points above the upper and lower whiskers represent the
   outliers (>1.5× IQR). Statistical significance was calculated using
   two-sided t-test. No significant subtype-specific difference on either
   gain or loss was observed (p > 0.05), except EC vs. SC on chromosomal
   gain frequencies (p = 0.039). 55 EC, 24 MESC and 17 SC tumors were
   analyzed. c Top frequent CNVs identified from this cohort revealing
   common and subtype-specific driver mutational pathways of NPC. CNVs
   involving in G1-S transition, NF-κB signaling, receptor tyrosine kinase
   (RTK), chromatin remodeling, microtubule polymerization, mitosis
   regulation, DNA repair, and EMT/invasion were suggested as NPC
   oncogenic drivers. Information of subtype, sample type, WHO subtype,
   gender, age of diagnosis, tumor stage, node classification, metastasis,
   and smoking status are shown at the bottom. 99 tumor samples were used
   for CNV identification. d Subtype-specific chromosomal gains (red) and
   deletions (blue). Chromosomal amplifications of Chr. 1q, 5p, 7p, 15q,
   17q, 20q, 21q, 22q and deletions in Chr. 5q, 13q exhibited significant
   differences among NPC subtypes. Important subtype-specific CNVs are
   indicated at top and bottom region. e Subtype-specific aberrant cancer
   driver pathways and networks in EC and SC subtypes. Important signaling
   pathways, including microtubule polymerization, mitosis regulation,
   NF-κB signaling, and EMT/invasion, were suggested to contribute to NPC
   subtype specificity.

   Somatic CNVs in gene level were called by Sequenza and CNVkit, and only
   consensus results shown in both callers were retained for further
   analysis (Fig. [146]2c; Supplementary Fig. [147]4a, b). The detected
   driver mutational events aggregated into several molecular mechanisms,
   including defective G1-S checkpoint surveillance (CDKN2A/B, TP53, and
   CCND1), activated NF-κB signaling (CYLD, TRAF3, NFKBIA, NLRC5, LTBR,
   TNFRSF1A, RELA, NIBP, RELA, RIPK2, IKBKB, and BIRC2/3), aberrant RTK
   (PIK3CA, PTEN, ERBB3, KRAS, MET, BRAF, and MST1R), and chromatin
   remodeling (KMT2C/D, BAP1, ARID1A, and TET1) (Fig. [148]2c). The
   mutation frequencies of these NPC drivers were in consistent with
   previous studies (Supplementary Tables [149]3, [150]4)^[151]5. They
   were the most frequently mutated pathways across all subtypes uncovered
   by CNV analysis.

   One of the interesting findings on these top frequent mutations was
   that macrophage stimulating 1 receptor (MST1R) was detected with
   recurrent deletions in 55.6% of tumors (Fig. [152]2c). MST1R expression
   in wild-type and mutant samples was further confirmed by IHC staining
   (Supplementary Fig. [153]5a, b). MST1R was also known as c-Met-related
   tyrosine kinase, and normally harbored activation/gain mutations and/or
   overexpression in other cancer types^[154]40–[155]43. In addition to
   playing an oncogenic role as tyrosine kinase to enhance activation of
   Ras/MAPK and other signaling cascades, MST1R also plays a vital
   function in host defense against viral infection, including
   Epstein–Barr virus (EBV) and human immunodeficiency virus
   (HIV)^[156]6,[157]44,[158]45. Considering the high frequent loss of
   MST1R had no obvious subtype and age preferences (Supplementary