Molecular characterization of EBV-associated primary pulmonary lymphoepithelial carcinoma by multiomics analysis

Abstract

Background

Primary pulmonary lymphoepithelial carcinoma (pLEC) is a subtype of non-small cell lung cancer (NSCLC) characterized by Epstein-Barr virus (EBV) infection. However, the molecular pathogenesis of pLEC remains poorly understood.

Methods

In this study, we explored pLEC using whole-exome sequencing (WES) and RNA-whole-transcriptome sequencing (RNA-seq) technologies. Datasets of normal lung tissue, other types of NSCLC, and EBV-positive nasopharyngeal carcinoma (EBV+-NPC) were obtained from public databases. Furthermore, we described the gene signatures, viral integration, cell quantification, cell death and immune infiltration of pLEC.

Results

Compared with other types of NSCLC and EBV+-NPC, pLEC patients exhibited a lower somatic mutation burden and extensive copy number deletions, including 1p36.23, 3p21.1, 7q11.23, and 11q23.3. Integration of EBV associated dysregulation of gene expression, with CNV-altered regions coinciding with EBV integration sites. Specifically, ZBTB16 and ERRFI1 were downregulated by CNV loss, and the FOXD family genes were overexpressed with CNV gain. Decreased expression of the FOXD family might be associated with a favorable prognosis in pLEC patients, and these patients exhibited enhanced cytotoxicity.

Conclusion

Compared with other types of NSCLC and NPC, pLEC has distinct molecular characteristics. EBV integration, the aberrant expression of genes, as well as the loss of CNVs, may play a crucial role in the pathogenesis of pLEC. However, further research is needed to assess the potential role of the FOXD gene family as a biomarker.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12885-024-13410-3.

Introduction

Lung cancer is the leading cause of cancer-related death worldwide, with an estimated 1.8 million deaths in 2020 [1]. Pulmonary lymphoepithelial carcinoma (pLEC) is a rare lung malignancy associated with Epstein-Barr virus (EBV) [2]. In the 2015 World Health Organization (WHO) classification of lung tumors, pLEC is classified as unclassified non-small cell lung cancer (NSCLC) [3], accounting for less than 1% of NSCLC [4]. As with all neoplastic lesions, the diagnosis of pLEC relies primarily on histopathological examination. However, the histological appearance of pLEC is similar to that of undifferentiated nasopharyngeal carcinoma (NPC) and poorly differentiated lung squamous cell carcinoma (LUSC) histologically [5–8], which poses challenges for accurate diagnosis.

High-frequency mutations common in other types of NSCLC are rarely detected in pLEC, according to prior research [8]. Therefore, personalized targeted therapy for other NSCLC may not benefit pLEC patients [9]. At the genome level, Chen et al.‘s research has identified the presence of EBV in pLEC, as well as in other malignancies such as gastric cancer and nasopharyngeal cancer [10]. This virus was found to exist in cells in a typical episomal form or integrated into the host genome [11]. Nonetheless, the lack of omics data and knowledge on the etiological and molecular events leading to pLEC tumorigenesis poses challenges for personalized targeted therapy for this disease [9, 12, 13].

Currently, molecular subtyping and prognosis-related studies in LUSC and LUAD are relatively extensive, providing patients with treatment guidelines and prognosis assessments for different subtypes. Various prognostic biomarkers, such as EGFR mutations [14], KRAS mutations [15], ALK-rearrangements [16], and KEAP1/NFE2L2 gene mutations [17], have been identified for these cancers and impact patient treatment response and prognosis. In contrast, there is limited research on the molecular subtyping and prognosis of pLEC. A defined molecular classification system and prognostic markers are lacking for pLEC, making personalized treatment strategies and prognosis evaluations challenging.

To better comprehend this unclassified lung cancer, we integrated genomic and transcriptomic data from pLEC for analysis, and further compared it with LUSC, LUAD, and EBV+-NPC data. Here, through an integrated analysis of multiomics data, we identified possible key genes in pLEC and highlighted their possible roles in tumor progression. Multidimensional comparative studies have shown that pLEC has distinct genomic features from LUSC, LUAD, and NPC. Our findings reveal the possible pathogenesis of pLEC, and this discovery may provide insights for personalized treatment options for patients with this rare tumor.

Materials and methods

Study population

In this study, fresh peripheral blood and formalin-fixed paraffin-embedded (FFPE) tissue were retrospectively collected from 15 pLEC patients for WES and RNA-seq integration analysis. All patients were diagnosed at the First Affiliated Hospital of Guangxi Medical University between June 2016 and June 2020. To rule out nasopharyngeal carcinoma lung metastasis in individuals, head CT, nasopharyngoscopy, and magnetic resonance imaging (MRI) were performed. Furthermore, 24 Chinese patients who were diagnosed with pLEC between July 2015 and December 2018 at three Guangdong Province hospitals (the First Affiliated Hospital of Guangzhou Medical University, Nanfang Hospital, and the First People’s Hospital of Foshan) were included [18].

All patients were classified as having undifferentiated carcinomas and were positive for EBV infection [3], consistent with the WHO classification criteria. TNM staging was performed according to the Eighth Edition of the International Association for the Study of Lung Cancer International Staging Project [19]. The expression of PD-L1 in 39 tumor tissues and its clinical significance were detected by IHC. The study was conducted in accordance with the Declaration of Helsinki and was approved by the Bioethics Committee of the First Affiliated Hospital of Guangxi Medical University, and all patients provided written informed consent for the study. (NO. (2021-KY-E-006))

DNA and RNA extraction

Peripheral blood and lung tissue from primary tumors were collected and prepared as FFPE samples. DNA and RNA were isolated from peripheral blood and frozen pLEC samples using the DNeasy Blood & Tissue Kit (Qiagen, Hilden, Germany) and RNeasy FFPE Kit (Qiagen #73504), respectively, following the manufacturer’s instructions.

Whole-exome sequencing and RNA sequencing

WES was performed as previously reported [20, 21], and DNA purity and concentration were determined using a Nanodrop 2000 spectrophotometer and a Qubit 2.0 fluorometer, respectively (Invitrogen, Carlsbad, CA, USA). Covaris S2 equipment was used to separate a total of 1 g of DNA into 200–250 nucleotide pieces (Woburn, MA, USA). Sequencing libraries were then prepared using the KAPA DNA Library Prep Kit (Kapa BiosSystems, Boston, MA, USA) according to the manufacturer’s protocol and the whole coding and partial noncoding regions of every gene were covered. The captured libraries were subjected to paired-end sequencing with a length of 100 bp using the Geneplus-2000 sequencing platform (Geneplus, Beijing, China). Peripheral blood mononuclear cell DNA served as a control (germline).

The purity of the RNA was determined using a KaiaoK5500^® Spectrophotometer (Kaiao, Beijing, China). Utilizing the RNA Nano 6000 Assay Kit of the Bioanalyzer 2100 system, RNA integrity and concentration were assessed (Agilent Technologies, Palo Alto, CA, USA). Libraries were prepared using the NEBNext^® Ultra™ RNA Library Preparation Kit (NEB, Beverly, MA, USA) following the manufacturer’s protocol. The constructed RNA-seq library was sequenced on the Geneplus-2000 sequencing platform (Beijing, China).

WES analysis

Next-generation sequencing raw data were then filtered to remove linker sequences and low-quality reads (paired-end reads were eliminated if one of the following three criteria was met: (a) half of the bases had a quality ≤ 5; (b) the proportion of N bases exceeded 5%; or (c) the average base quality was less than 0. Mutation calling was performed using the BWA aligner (version 0.7.10) to further map the clean reads to the reference human genome (hg19). The PCR duplicates were identified with Picard (version 1.98, Broad Institute, Cambridge, USA). Relocal alignment and indel mass value calibration were completed by using GATK (version 3.4–46-gbc02625, Broad Institute, Cambridge, MA, USA). Using MuTect, single nucleotide variants (SNVs) were identified (version 1.1.4, Broad Institute, Cambridge, MA, USA). Small insertions and deletions (indels) were identified via GATK. Somatic copy number alterations were consistent with those of ContrA (2.0.8, Jason Li et al., Melbourne, Australia). A mutation was considered a candidate somatic mutation only if: (I) the mutation was detected in no less than 5 high-quality reads, (II) the allele had a mutation frequency > 0.01, (III) No mutation in > 1% of the population in the 1000 Genomes Project (version 3) or database (Single Nucleotide Polymorphism Database, version dbSNP 137) and (IV) there was no mutation in a local database of control samples.

RNA analysis

To obtain high-quality paired-end reads, we removed sequencing reads containing adapter sequences and low-quality reads. Quality-controlled reads were aligned to the human genome (hg19) using HISAT (v2.0.4, Daehwan Kim, et al., Baltimore, USA). Transcript assembly was achieved by StringTie (v1.2.3, Mihaela Pertea et al., Baltimore, USA).

Significant mutations and oncogene/tumor suppressor gene (TSG) analysis

MuSiC2 [22] was used to identify significantly mutated genes in pLEC. Somatic cell characterization was performed according to the methods of the R package “deconstructSigs”. Somatic copy number alterations were analyzed using GISTIC2 software with a residual q-value of 0.1 to determine broad peak limits. In addition, oncogene (tumor-suppressor gene) TSGs and mutational signatures were extracted from the COSMIC database (v90), Comprehensive299, and Bert Vogelstein125 [23–26]. After excluding influencing factors such as sex and age, the nonsynonymous mutation rates and copy number alterations in pLEC patients were also compared to those in other lung cancer subtypes, including 50 LUAD and 50 LUSC patients from the TCGA and 50 EBV+-NPC patients from another group [27, 28].

DEG screening, gene set enrichment, and immune microenvironment analysis

pLEC used the R package Deseq2 to perform gene differential expression analysis with normal lung tissue, with the control group sourced from normal lung tissue rather than adjacent nontumor tissue of pLEC (GSE81089) [29], and records with a P-value < 0.05 were retained for differential screening. Gene set enrichment analysis (GSEA) was used to calculate enrichment scores for certain pathways. In addition, clusterProfiler was used to select KEGG pathways and analyze DEGs for pathway enrichment.

Single-sample gene set enrichment analysis (ssGSEA) associated with immune cells and cell death scores was performed using the GSVA software package [30]. The R package IOBR, which integrates 8 methods including xCell, ESTIMATE, and TIMER, was used to evaluate the immune microenvironment [31]. Additionally, the degree of immune infiltration in LUAD, LUSC, and EBV+-NPC patients was compared to that in pLEC patients [27, 32].

Calculation of the tumor stemness index based on RNA-seq data

To obtain expression data for pluripotent stem cell samples (including embryonic stem cells (ESCs) and induced pluripotent stem cells (iPSCs)) from the Progenitor Cell Biology Consortium (PCBC) [33], the data was downloaded using the R package synapse (syn2701943). The OCLR (one-class logistic regression) method, as described in a previous study [34], was employed to predict and calculate the stemness index using ESC and iPSC expression data. To compute the mRNA expression-based stemness index (mRNAsi), the 229 samples in the PCBC database were divided into two groups: 78 stem cell samples and 151 nonstem cell samples. The expression profile data were normalized to the mean value of each sample. The R package gelnet was utilized to compute the weight vector for mRNAsi using the OCLR method.

Immunohistochemistry (IHC)

Immunohistochemical staining was performed on formalin-fixed paraffin-embedded specimens using the PD-L1 IHC 22C3 pharmDx assay (Agilent Technologies/Dako) to evaluate PD-L1 expression in tumor and immune cells (lymphocytes and macrophages). The staining results were interpreted by two independent pathologists. Previous studies estimated the levels of PD-L1 in tumor cells to be < 1%, 1 − 49%, and ≥ 50%. A TPS ≥ 1% was used to characterize PD-L1 positive instances, and PD-L1 was considered to be highly expressed when the TPS was ≥ 50% [35].

Genome alignment, EBV host integration, and variant calling

After removing low-quality and adaptor base sequencing reads, the clean reads were aligned to the 1000 Genomes Phase2 Reference Genome Sequence (hs37d5) and the EBV Aggregate Reference Sequence using Burrows-Wheeler Aligner (BWA)-MEM.

Human somatic SNVs were detected by Mutect2, and VEP annotation was performed. Virus subtype identification and virus insertion site analysis were performed with the virusfinder2 software [36], and SV analysis was performed by SVDetect and CREST software [37, 38]. SV analysis revealed the site of virus insertion in the human genome.

Statistical analysis

Two-tailed Student’s t test and Fisher’s exact tests were used for continuous and discrete variables, respectively. Differences in effective rates were compared using Pearson’s chi-square test. Survival probabilities were analyzed using the Kaplan-Meier method, and comparisons were performed using the log-rank test. Multivariate analysis was performed using a Cox regression model (multivariate analysis). Cosine similarity was used to evaluate the similarity of pLEC to other tumor mutational profiles. A P-value < 0.05 was considered to indicate statistical significance in all tests. All statistical analyses were performed using standard R packages.

Results

Clinical features and survival analysis of patients

The workflow of this study is depicted in Fig. 1 (Figure S1). The study cohort consisted of 39 patients diagnosed with pLEC, with a median age of 49 years (range: 24–76 years) and a male-to-female ratio of 0.95. A smoking history was reported for 15 patients (38%). All patients were positive for EBV infection. The clinical characteristics of the patients are summarized in Table 1. According to the TNM staging system, 4 (10%), 5 (12%), 15 (39%), and 15 (39%) patients were diagnosed with stage I, II, III, and IV disease, respectively. Among the 27 patients who received chemotherapy and/or adjuvant chemotherapy, 5 also received radiotherapy. Among the 25 patients with documented best response, 9 achieved partial response (median OS = 17.00 months), and 16 achieved disease stability (median OS = 23.35 months). Immunohistochemical analysis of 15/39 patients revealed that all patients patients (15/15, 100%) were positive for CK5/6 and P63, while 100% (15/15) were negative for thyroid transcription factor-1 (TTF-1), indicating that pLEC molecularly resembles squamous cell carcinoma rather than adenocarcinoma (Table S1,2) [9, 39].

Fig. 1.

Overview, design and summary statistics. This flowchart describes the inclusion and exclusion of all samples. pLEC Pulmonary lymphoepithelial carcinoma, NPC Nasopharyngeal carcinoma, LUAD Lung adenocarcinoma, LUSC Lung squamous cell carcinoma, WES whole exome sequencing, RNA-seq RNA whole transcriptome sequencing

Table 1.

Clinical characteristics of patients with pLELC

Characteristics		All (n=39)/Best respose(%)
Age	24-76(50.23±10.48)	39(100%)
Gender	Male	19(49%)
	Female	20(51%)
Stage	I	4(10%)
	II	5(12%)
	III	15(39%)
	IV	15(39%)
smoke	Yes	15(38%)
	No	24(62%)
Surgery	Yes	15(38%)
	No	24(62%)
Radiotherapy	Yes	6(15%)
	No	33(85%)
Local or systemic regimen received	neoadjuvant chemotherapy  + surgery + adjuvant chemotherapy + radiotherapy	1(2.6%)/100%
	neoadjuvant chemotherapy + surgery + adjuvant chemotherapy	3(7.7%)/66.6%
	Surgery + adjuvant chemotherapy	4(10%)/100%
	surgery + adjuvant chemotherapy + chemotherapy	1(2.6%)/100%
	surgery + chemotherapy + radiotherapy	1(2.6%)/0%
	surgery + chemotherapy	2(5.1%)/50%
	Chemotherapy alone	18(46%)/72.2%
	Chemotherapy + radiotherapy	3(7.7%)/66.6%
	Surgery alone	3(7.7%)/0%
	Radiotherapy alone	1(2.6%)/100%
	No local or systemic regimen received	2(5.1%)/0%
Best response	Stable disease	16(41%)
	Partial response	9(23%)
	Unknown	14(36%)
PD-L1	<1%	16(41%)
	≥1%	23(59%)

For the 38 patients for whom follow-up information with available, the median follow-up duration was 20.85 months (range: 3.70 to 73.00 months). Clinicopathological factors including age, gender, smoking history, and progression stage were evaluated. Univariate analysis revealed no significant correlation between these clinical factors and overall survival (Fig. 2A). In addition, the effect of PD-L1 expression on patient prognosis was assessed. Notably, patients with higher PD-L1 expression tend to have longer survival (Figure S2A, B). Survival analysis demonstrated that individuals with pLEC had considerably lower mortality rates than those with other non-small cell lung malignancies and NPC (Fig. 2B). Previous research has indicated that pLEC is frequently misdiagnosed due to its radiological and pathological similarities to poorly differentiated LUSC [8, 13]. To differentiate between pLEC and poorly differentiated LUSC (GSE44170), a set of genes differentially expressed in these poorly differentiated LUSC tumors was selected for sample clustering [40]. By utilizing the NMF clustering approach to group the expression profiles and select the optimal number of clusters, the gene set was found to be capable of successfully distinguishing between the two disease samples with an AUC of 0.91 (Fig. 2C, D, Figure S2C, D). These findings indicate that pLEC may represent a distinct subtype of NSCLC and that patients with pLEC have a favorable prognosis. Despite their pathological similarities, the two diseases can be distinguished at the molecular expression level. These results further promote our investigation of pLEC through multiomics analysis.

Fig. 2.

Prognosis of pLEC by Kaplan-Meier survival analysis with log-rank test. A Cox-model hazard ratios (x-axis) for Clinical features (y-axis)). Horizontal bars indicate 95% confidence intervals. B The probability of OS in patients with pLEC compared with patients with LUSC, LUAD, and NPC (P = 0.001). C Cophenetic correlation from NMF analysis of pLEC and poorly differentiated LUSC. D The NMF consensus matrix shows that the cluster analysis identified 2 subtypes. P < 0.05 indicates a significant difference. NPC nasopharyngeal carcinoma, LUAD lung adenocarcinoma, LUSC lung squamous cell carcinoma

Key coding mutations of pLEC and copy number characteristics

To gain further insight into the molecular characteristics of pLEC and its underlying tumorigenic mechanism, we conducted on 15 tumor specimens with an average coverage of 100× (Table S3,4). Our analysis identified a total of 1,034 somatic mutations, including 708 nonsilenced, 259 silent, and 67 short indels (Fig. 3A), indicating a low mutation rate (median: 0.78 mutations per Mb; Figure S3). Notably, the principal somatic mutation types comprise C > T transitions and T > C transitions (Figure S4A). Subsequently, two stable and independent mutation signatures were identified. The primary signature was attributed to the deamination process of 5-methylcytosine (signature 1), followed by improper DNA mismatch repair (signature 6) (Table S5, Figure S4D).

Fig. 3.

Mutation spectrum and CNV landscape of pLEC and comparison with other cancer types. A Mutational landscape of 15 pLEC samples. The dark blue, red, orange, khaki, dark green, light blue, light green, and purple represent missense, nonsynonymous, frameshift deletions, frameshift insertions, splice site mutations, and in-frame insertions, respectively, in-frame deletions, and multiple mutations. The bar graph on the right shows the mutation frequency in the population, and the upper bar graph shows the frequency of non-synonymous mutations in the sample. B High-frequency CNVs of fragments in pLEC. CNV gain (red) is shown on the left and CNV loss (blue) is shown on the right. The figure shows significant amplification or deletion of chromosomes from 1 (top) to 22 (bottom). The green line represents the cutoff value for significance (q = 0.1). C Distribution of non-silent mutation rates in LUSC (red), LUAD (yellow), NPC (cyan) and pLEC (purple). The black lines in the scatterplot represent the median and upper and lower quartiles of sample mutation frequency in each cancer. All cancers were statistically tested by the Wilcoxon rank sum test method. A P value less than 0.05 was defined as a significant difference. D Mutation profiles of the six mutation types for each cancer type. E Mutation spectrum of six mutation types for each cancer type. F Comparison of genomic mutational profiles between pLEC, LUAD, LUSC, and NPC. The bars at the bottom of the graph represent different tumor types. Color blocks represent different types of base changes, and the mutation frequency of the population is shown on the right. G Comparison of CNA in pLEC and other cancer types. H Pathway Alteration Frequencies. The fraction of altered samples for each pathway and tumor subtype. Pathways are ordered by decreasing median frequency of change. Increasing the color intensity reflects a higher percentage. Average mutation counts for each cancer subtype are also provided. Average mutation counts for each cancer subtype are also provided

A total of 31 genes with mutation frequencies exceeding 10% were identified in the pLEC, which exhibited a cooccurring pattern (Fig. 3A and Figure S4B). Consistent with previous research, high-frequency mutations commonly observed in other NSCLC, such as KRAS and EGFR, were not detected. To contribute to a better understanding of the distinctions between pLEC and other diseases, we comprehensively compared its genetic profile with LUAD, LUSC, and EBV + NPC. The nonsynonymous mutation rates of pLEC and NPC were significantly lower than that of LUSC and LUAD (p = 7.7 × 10⁻⁷ and p = 8.9 × 10⁻⁴). pLEC and NPC exhibited a distinctive pattern of C > T mutations, contrasting with the frequent C > A changes observed in other lung cancer subtypes. Meanwhile, cluster analysis based on mutation spectra (Figure S4D) showed that pLEC was comparable to NPC but distinct from other lung tumors (Fig. 3C–F).

To further explore the genomic landscape of pLEC, we conducted somatic copy number alteration analysis using GISTIC2 [41] software with a cutoff value of 0.1. CNV loss was widely observed in pLEC (Figure S4C), with considerable CNV amplification observed at 9q12 (FOXD family) and CNV losses at 1p36.23 (CHD5), 3p21.1 (CCR3), 7q11.23 (FGL2), and 11q23.3 (ZBTB16, Fig. 3B). These findings imply that the development of pLEC is likely driven by unique genomic mechanisms and may facilitate the development of targeted therapies for this rare and aggressive lung cancer subtype.

To mitigate potential biases arising from low mutation rates, we conducted CNV comparisons (Fig. 3G). Our analysis revealed that pLEC displayed a distinct CNV distribution compared to other lung tumors. Unlike LUAD and LUSC, which exhibited amplification of 1q and 5p, and deletion of 8p, pLEC exhibited deletions on chromosomes 3p, 13q, 14q, and 16q, as well as amplification of chromosome 12, consistent with previous reports [12]. Moreover, our findings revealed similar frequencies of pLEC and NPC in the cancer pathway (Fig. 3H), yet with disparate rates of genetic mutations within the pathways, highlighting the distinct features of pLEC (Figure S5) [42]. These findings imply that CNV alterations may be strongly implicated in the pathogenesis of pLEC.

Integrative transcriptome analysis of pLEC

To gain further insight into the changes in pLEC at the expression level, we performed RNA-seq analysis on 14 pLEC samples and 14 normal lung tissue samples. We discerned that pLEC exhibited an augmented activation of CD4 T cells, activated CD8 T cells, neutrophils, and regulatory T cells (Figure S6A). We utilized DESeq2 [43, 44] to compare tumor and normal lung tissue (GSE81089) [29], and identified a set of differentially expressed genes (DEGs), including 5518 upregulated genes and 3847 downregulated genes (p < 0.05, Benjamini–Hochberg (BH) method correction; log FC > 1 or < −1) (Figure S6B). The functional pathway enrichment analysis of DEGs demonstrates significant differences in pathways, including DNA transcription factor activation, olfactory transduction, the Epstein-Barr virus infection pathway, and the NF-kappa B signaling pathway, between pLEC and normal tissues (Figure S6C-D, Table S6). Furthermore, immune microenvironment analysis revealed that pLEC highly infiltrated T helper 17 cells (Figure S6E).

Given the similar mutation characteristics of pLEC and NPC, we further investigated of their expression patterns (Fig. 4A). Based on the functional pathway enrichment analysis of DEGs, pLEC was significantly positively linked with olfactory receptor activity (p = 5.0 × 10⁻³⁷) and epidermal cell differentiation (p = 1.4 × 10⁻¹¹), which led us to hypothesize that these two types of cancers share some commonalities in disease progression (Figure S7A-D). Our findings demonstrate that NPC exhibits a significantly greater level of tumor stemness than pLEC (Fig. 4B, p = 9.7 × 10⁻⁴), implying that NPC possesses a greater capacity for tumor development, metastasis, medication resistance, and renewal iterative ability than pLEC. To assess the regulated cell death, we utilized the “ssgsea” method to calculate the enrichment scores. The results revealed that pLEC was enriched in Intrinsic apoptosis, Necroptosis, and Lysosome dependent cell death, in contrast to NPC (Fig. 4C). In the investigation of immune infiltration in comparison to other diseases, pLEC exhibits a higher susceptibility to innate immunity, potentially associated with the immune evasion mechanism of EBV (Fig. 4D, Table S7). Specifically, the EBV immune escape factor BDLF3 reduces the expression of HLA molecules on the cell surface, thereby diminishing the capacity of CD8 + T cells and CD4 + T cells capacity to detect infected cells, and hindering the host’s immunological response.

Fig. 4.

Transcriptome comparison of pLEC with other cancer types. A Differential expression analysis of RNA-seq data, where red and blue correspond to up- and down-regulation of pLEC samples relative to NPC samples, respectively. B Evaluation of tumor stemness score. The Wilcoxon rank-sum test was used for statistical testing of the two groups of data. C Sample Regulatory Cell Death Type Score. D Comparative analysis of immune microenvironment between pLEC and NPC, LUSC and LUAD. The upper bars indicate different assessment methods, blue indicates significant downregulation of pLEC and red indicates significant upregulation. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001

EBV integration promotes tumor development in pLEC

For a comprehensive analysis of pLEC, the genomic data were aligned to the EBV reference genome. We identified 117 EBV integration breakpoints scattered across 23 human chromosomes, excluding the Y chromosome, from nine samples (Fig. 5A, B). EBV integration counts varied considerably among individuals, with six of the nine samples showing breakpoints within the range of 10–32, and the same integration sites were detected in multiple samples. Annotated by integration sites, EBV has a propensity to integrate close to fragile sites that are susceptible to DNA damage, including 27 common regions and 4 rare sites. This increases the possibility of DNA insertion into the host genome through microhomology-mediated DNA repair (Table S8).

Fig. 5.

EBV integrated analysis. A Landscape of EBV integration breakpoints across the genome. Chromosomes are numbered and represented by a series of colors. B Distribution of EBV integration breakpoints across chromosomes. The x-axis is the number of chromosomes, and the y-axis is the number of integration breakpoints. C mRNA cluster analysis of EBV integration sites. The calculation formula of the number (N) of each block is: N = log2 (mRNA expression of a gene in tumor tissue/mRNA expression of a gene in control tissue). Blue represents down-regulated mRNA and red represents up-regulated mRNA

In pursuit of advancing our understanding of pLEC, we undertook an analysis of the expression of 63 genes that were integrated by the virus. The results demonstrated that EBV-integrated genes were more commonly silenced or absent in pLEC tissue relative to normal lung tissue (Fig. 5C). Functional enrichment analysis showed that viral integration genes were significantly correlated with Hippo signaling regulation pathways (p = 9 × 10⁻⁴, Table S9). Intriguingly, a significant number of viral integration sites were observed on chromosomes 1, 3, 7, and 11 (Fig. 5B). Thus, we speculate that EBV viral integration may be a critical driver of the high frequency of copy number deletions observed in pLEC patients and may also underlie the molecular mechanisms of disease progression.

Low expression of FOXD family genes is associated with good prognosis in pLEC

To elucidate the link between CNVs and gene expression, a combined analysis of genomic and transcriptomic data was performed. The results revealed that the FOXD family genes, including FOXD4L2, FOXD4L4, FOXD4L5, FOXD4L3, and FOXD4L6, exhibited elevated gene expression levels in regions of copy number amplification, while genes such as ZBTB16, ERRF11, GPR62 in regions of copy number loss showed reduced gene expression (Fig. 6A).

Fig. 6.

Integrated analysis of genome and transcriptome and prognostic analysis. A Integrated analysis of transcriptome and genomic CNVs. Each column represents a sample, and each row represents genes amplified by high frequency CNVs. At the top of the chart are annotations of sample clinical information, representing OS time as well as clinical stage. The graph shows the mRNA expression of this gene relative to the control log2-fold change in mRNA expression in cancer samples. FOXD4L2, FOXD4L4, and FOXD4L5 family genes were amplified and upregulated in 100% (14/14) of pLEC patients; mRNA log2-fold changes of FOXD4L2, FOXD4L4, and FOXD4L5 ranged from 0.21 to 1.8, 0.47 to 2, and 0.33 to 1.91, respectively ZBTB16, ERRFl1, GPR62 gene copy number loss and down-regulation. B Kaplan-Meier survival analysis of genes with copy number loss or gain. Statistical significance was estimated by two-sided log-rank test

To further confirm the validity of the above results, we conducted an analysis using the Kaplan-Meier method to explore the potential association between copy number alteration-related genes and patient prognosis. We found that expression of genes belonging to the FOXD family might be associated with prognosis in 13 patients with available survival information (Fig. 6B). Specifically, patients with low expression of FOXD4L4, FOXD4L5, or FOXD4L6 in pLEC were all alive during the follow-up period of nearly 70 months. However, no meaningful results were observed for ZBTB16 or ERRFl1. This could be attributed to the fact that ZBTB16 and ERRFl1 do not produce direct effects but rather mediate the disease process through other factors.

High expression of the FOXD family shapes an immune-suppressive tumor microenvironment

To gain initial insights into the relationship between the FOXD gene family and infiltrating immune cells, a correlation analysis was conducted between the FOXD gene family and immune cell markers. These biological markers are widely utilized for immune cell characterization (Table S10). We found a strong correlation between the expression of the FOXD gene family and CD8 + T-cell, suggesting a potential involvement of FOXD gene expression in immune regulation and immune responses. (Fig. 7A, B).

Fig. 7.

Mechanisms underlying an FOXD4L4-associated immune-suppressive tumor microenvironment. A,B FOXD4L4 expression was significantly negatively correlated with infiltrating levels of CD8 + T cells. C The FOXD4L4^high group had significantly greater enrichment scores for activated CD8 T cells and cytotoxic cells. Note: *p < 0.05, **p < 0.01, ***p < 0.001. D-F Boxplot showing cell death scores within FOXD4L4^high and FOXD4L4^low groups

To further investigate this hypothesis, we analyzed the levels of immune cell infiltration in FOXD4L4 low samples and FOXD4L4 high samples [31]. By utilizing different methods to quantify the different immune cell populations in the two groups, we found a high level of consistency. Compared to the FOXD4L4 high group, the FOXD4L4 low group scored higher in CD8 + T cells (p = 2 × 10⁻²) and cytotoxicity (p = 2 × 10⁻²) (Fig. 7C). CD8 + T cells and cytotoxic cells are capable of triggering cell apoptosis and killing virus-infected cells. Additionally, we observed that the FOXD4L4 low group had higher scores in intrinsic cell apoptosis and alkaliptosis compared to the FOXD4L4 high group (Fig. 7D-F, p < 0.05). These findings explain the better survival trend observed in the FOXD4L4 low group. We further compared the expression of FOXD family genes in pLEC, other NSCLC types, and NPC, finding that FOXD family genes tend to be highly expressed in pLEC. Given the similarities between pLEC and NPC, we also examined FOXD family gene expression in tumor and normal tissues of NPC. The results indicated that while NPC also showed an upward trend in FOXD family gene expression, only the FOXD4L1 gene reached statistical significance (Table S11). Therefore, FOXD4L4 may be a biomarker that can be used to predict the prognosis of patients with pLEC.

Discussion

pLEC is a rare type of non-small cell lung cancer that is associated with EBV infection and lacks distinctive pathological features. To comprehensively investigate the molecular characteristics of pLEC, we performed an integrated analysis of 15 cases of pLEC using WES and RNA-seq, and assessed the impact of PD-L1 expression on pLEC prognosis. Our research confirms previous findings that pLEC patients with high PD-L1 expression exhibit better prognosis. Furthermore, compared to other subtypes of NSCLC and EBV+-NPC, pLEC displays distinctive FMGs, extensive CNV deletion, and higher immune infiltration in terms of innate immunity. pLEC and NPC share similarities in terms of mutation characteristics and functional pathways, but pLEC exhibits a lower tumor stemness score. Since the tumorigenesis of pLEC and EBV infection is tightly connected, we performed an analysis of EBV viral integration and discovered that virus integration genes are closely related to Hippo signaling regulation pathways. Furthermore, we found that the CNV high-frequency deletion region and the viral integration site were congruent, and the genes in the region tended to be down-regulated. We also identified a potential association between low expression of FOXD family genes and good prognosis in pLEC. In summary, through multidimensional genome comparison, we revealed the unique genetic characteristics and possible pathogenesis of pLEC.

In this study, we performed immunohistochemical examinations on tumor tissues, revealing that all samples were positive for CK5/6 and P63, while negative for TTF-1. These results indicated that pLEC is more inclined to LUSC instead of LUAD in terms of omics. Previous studies have indicated that pLEC morphologically resembled poorly differentiated LUSC and undifferentiated NPC, complicating clinical differential diagnosis [5–7]. Concurrently, the exploration of tumor occurrence and molecular-level changes in pLEC has been limited due to a predominant reliance on genomic data in prior studies [9, 12, 18]. We further expanded the control cohort of pLEC, which demonstrated superior patient outcomes compared to LUSC, LUAD, and EBV+-NPC. Furthermore, pLEC lacks clear prognostic markers to stratify patients, thereby impacting the development of personalized treatment strategies.

To comprehend the molecular characteristics of pLEC, we initially compared the genomic mutation profiles of pLEC with those of LUSC, LUAD, and EBV+-NPC. Notably, unlike other lung cancer subtypes and NPC, pLEC had a low frequency of alterations in frequently mutated driver genes such as EGFR (0/15), KRAS (0/15), and BRAF (1/15), as previously reported [28, 45–49]. Only one patient harbored a MET frameshift loss, which was distinct from the typical MET exon 14 skipping mutation [50, 51]. Thus, the absence of strong mutational drivers in pLEC may reduce the aggressiveness of tumor cells and could be a contributing factor to improved survival rates among pLEC patients. Chen et al. made the observation that pLEC was associated with signatures 1, 3 + 15, which were determined to be substantially equivalent to the signals 1 and 6 that were discovered in our study [11]. These signatures are related to biological functions associated with 5-methylcytosine and DNA mismatch repair. The difference in numbers is due to different versions of the SIG database. We also found that pLEC manifested a distinct distribution of CNV in comparison to other lung cancers.

In the context of transcriptome analysis, pLEC exhibited higher scores in neutrophils, activated CD4 and CD8, and regulatory T cells (Treg), indicating a potential involvement of these cell types in pLEC pathogenesis. Functional enrichment of DEGs revealed similar functional pathways to NPC, encompassing pathways such as protein digestion and absorption, olfactory transduction, among others [52–54]. Interestingly, pLEC showed a negative correlation with the Epstein-Barr virus infection pathway and NF-kappa B signaling pathway, which have been previously associated with decreased gene expression following EBV infection [55]. Additionally, we found that pLEC and LUSC can be distinguished by molecular expression levels, enabling more precise disease typing and underscoring fundamental differences between these malignancies. Given the genetic similarities between pLEC and NPC, we further assessed tumor stemness. In terms of quantitative analysis of tumor sample stemness, the findings revealed that, pLEC demonstrated significantly lower stemness compared to NPC, potentially explaining its relatively less aggressive traits. Furthermore, pLEC displayed higher infiltration of innate immune cells compared to other cancers, highlighting its potential immune-related features. These results also further promote our study of pLEC through multi-omics analysis, which may offer valuable insights for disease diagnosis and treatment.

In integrated genome and transcriptome analysis, we observed significant downregulation of tumor-associated genes ZBTB16 and ERRFI1, which was accompanied by concomitant CNV loss. Conversely, the FOXD4 family genes showed significant upregulation with CNV gain. ZBTB16, also referred to as the promyelocytic leukemia zinc finger gene (PLZF), is a transcriptional repressor that plays a pivotal role in tumor progression [56]. Downregulation of ZBTB16 has been reported in various diseases, including melanoma, liver cancer, and prostate cancer [57–63]. Previous research has demonstrated a substantial correlation between high ZBTB16 expression and prolonged OS in cancer patients [64, 65]. ERRFI1 is known as a tumor suppressor due to its direct inhibition of epidermal growth factor receptors [66–68]. Downregulation of ERRFI1 leads to higher mobility and enhanced cell growth under conditions of low EGFR expression [69, 70]. The FOXD family belongs to the evolutionarily conserved transcription factor FOX superfamily, which participates in the cancer process as an oncogene or tumor suppressor gene [71]. The FOXD family includes nine members including FOXD1, FOXD4, FOXDL4, and FOXDL5, which plays crucial regulatory roles in normal cellular development and disease progression [72–76]. Altogether, we hypothesized that the deregulation of ZBTB16 and ERRFI1 expression, as well as the overexpression of the FOXD family, contribute to the development of pLEC.

EBV is related to a variety of human malignancies and can facilitate tumorigenesis and progression [10]. The EBV genome is typically integrated into the host genome through integrated microhomologies or exists as episomes [77]. In the present study, all examined tumor tissues of pLEC exhibited positivity for EBV. A total of 117 EBV integration breakpoints were identified by aligning WES data to the EBV reference genome. Integrative analysis of the transcriptome data revealed that nearly all integration site genes, including HLA-DRB1, were underexpressed, further supporting the notion that EBV infection may lead to gene expression suppression. Furthermore, we discovered a substantial correlation between the Hippo signaling pathway and viral integration genes. Dysregulation and inactivation of the highly conserved Hippo pathway frequently contribute to tumorigenesis and progression in various cancer types by regulating cell proliferation and apoptosis [78]. We also uncovered that the region of copy number loss (1p36.23, 3p21.1, 7q11.23, and 11q23.3) coincided with the integration region of EBV, with three of these regions occurring in the fragile region of the chromosome. Previous studies have revealed that the EBV immune escape factor BDLF3 can reduce the expression of HLA class I and II molecules on the cell surface through ubiquitination, thereby hindering the recognition of CD8 + T cells and CD4 + T cells to virus-infected cells, and evading the host’s immune response [79]. This finding also demonstrates why pLEC is more susceptible to innate immune infiltration than other NSCLC and NPC.

Additionally, EBV integration has been reported to occur near common fragile sites (CFS), which are susceptible to DNA damage, thus increasing the likelihood of EBV DNA insertion into the host genome [80]. Our study identified 27 integration sites near CFS and 4 sites near rare fragile sites, with the genes at these integration sites displaying low expression. Interestingly, almost all regions of pLEC copy number loss overlapped with vulnerable regions of the genome. Therefore, we inferred that the development of pLEC may have been caused by copy number loss resulting from EBV integration. Our findings highlight the importance of understanding the molecular mechanisms underlying EBV-associated cancers, particularly in the context of viral integration and genomic instability.

To further confirm the role of these genes in pLEC, we performed survival analysis on the expression of the genes. The findings suggested that the FOXD family would be helpful in predicting the prognosis of pLEC patients, and that low levels of FOXD family expression may be associated with a superior prognosis. Differential expression analysis demonstrated that the low-FOXD4L4 group exhibited significant enrichment in pathways related to Epstein-Barr virus infection (NES = 1.579, p = 6.3 × 10⁻⁵), natural killer cell-mediated cytotoxicity (NES = 2.2, p = 1.1 × 10⁻⁶), and TNFA signaling via NFKB (NES = 2.36, p = 4.1 × 10⁻¹⁶). Compared to other NSCLC and NPC, pLEC exhibited higher expression. These findings highlight the potential of FOXD4L4 as a biomarker for predicting prognosis in pLEC patients.

Indeed, this study has several limitations. First, we must acknowledge that the small sample size in our study may reduce the representativeness of some findings, especially for such a rare EBV-associated pulmonary malignancy. To address this limitation, we expanded the comparative scope by analyzing the relationship between pLEC and other NSCLC and NPC from genomic and transcriptomic perspectives. This revealed significant CNV losses and unique immune microenvironment characteristics in pLEC, which further enhanced our understanding of its disease features. Second, although we found that patients with low expression of FOXD family genes tend to have a better prognosis, we lack a deeper exploration of their underlying mechanisms. Furthermore, we have not validated whether FOXD family gene expression grouping is helpful for individualized diagnosis and treatment. Therefore, future studies with larger sample sizes are needed to confirm the potential of FOXD family gene expression as a reliable prognostic biomarker for pLEC. Additionally, future research should explore the mechanisms by which FOXD family genes promote pLEC progression and assess their applicability in personalized treatment strategies. Finally, translating these molecular findings into clinical practice will require careful validation and evaluation in larger cohorts, considering factors such as inter-patient variability, treatment response, and long-term outcomes. We hope that future large-scale studies will validate these results and further investigate the mechanisms of pLEC as well as the role of FOXD family genes in disease diagnosis and therapy.

Conclusions

Our study comprehensively explored genomic and transcriptomic data of pLEC and performed a whole-exome analysis of EBV integration in pLEC. By assessing molecular expression levels, we successfully differentiated pLEC from poorly differentiated LUSC. Extensive CNV deletions were present in pLEC patients and coincided with EBV integration sites. ZBTB16, ERRFI1, and FOXD family were identified as putative oncogenes in pLEC, accompanied by copy number alterations and expression dysregulation. Collectively, the loss of CNV, decreased expression of ZBTB16 and ERRFI1, and increased expression of the FOXD family may be triggers for tumorigenesis. Meanwhile, the FOXD family may serve as a valuable prognostic biomarker for pLEC.

Abbreviations

pLEC

Pulmonary lymphoepithelial carcinoma

NSCLC

non-small cell lung cancer

EBV

Epstein-Barr virus

WES

whole-exome sequencing

RNA-seq

RNA-whole-transcriptome sequencing

NPC

nasopharyngeal carcinoma

FFPE

formalin-fixed paraffin-embedded

MRI

magnetic resonance imaging

SNV

single nucleotide variant

TSG

tumor-suppressor gene

GSEA

Gene Set Enrichment Analysis

ssgsea

Single-sample gene set enrichment analysis

ESCs

mbryonic stem cells

iPSCs

induced pluripotent stem cells

mRNAsi

mRNA expression-based stemness index

IHC

Immunohistochemical

Authors’ contributions

Jing Bai: Funding acquisition, Project administration. Meiling Yang and Jing Bai: Conceptualization, Supervision. Meiling Yang, Guixian Zheng, Haijuan Tang, Fukun Chen, Yu Huang, Zili Lv, Benhua Li, Yaoyao Liu, Xuan Gao, Maolin Yang, Qing Bu, Lixia Zhu, Pengli Yu, Zengyu Huo, Xinyan Wei, Xiaoli Chen, Yanbing Huang, Zhiyi He, and Xuefeng Xia: Data curation, Software, Investigation. Meiling Yang, Guixian Zheng, Haijuan Tang, Fukun Chen, Yaoyao Liu, Xuan Gao, Qing Bu, Pengli Yu, Lixia Zhu, Zengyu Huo, Xinyan Wei, Xiaoli Chen and Zhiyi He: Formal Analysis, Methodology. Meiling Yang, Guixian Zheng, Haijuan Tang, Fukun Chen, Yaoyao Liu, Gao Xuan and Xuefeng Xia: Validation. Meiling Yang, Fukun Chen and Xuan Gao: Visualization, Writing – original draft All authors: Writing – review & editing.

Funding

This study is supported by the National Natural Science Foundation of China (82260009, 81560008, 81960010) and the Natural Science Foundation of Guangxi (2018GXNSFAA050056, 2020GXNSFAA297143).

Data availability

The sequencing data generated in this study have been deposited in the Genome Sequence Archive (GSA) at the National Genomics Data Center, part of the China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences, under accession number HRA008178 (https://ngdc.cncb.ac.cn/gsa-human/).The results analyzed in this study are deposited in 10.6084/m9.figshare.23259305.

Declarations

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.