Three normal NK normal control and 15 NKTCL case (Supplementary Table 1) RNA-Seq raw data files were downloaded from the National Center for Biotechnology Information (NCBI) Sequence Read Archive (SRA) with the ID SRP049695. TrimGalore (http://www.bioinformatics.babraham.ac.uk/projects/trim_galore/) method was used to filter low-quality raw reads as follows: (1) removal of primers sequencing; (2) removal of sequences of low quality in terminal; (3) removal of sequences <35 bp. Clean reads were aligned with HISAT2 to the USCS hg19 human genome assembly. mRNA and lncRNA quantitative expressions [fragments per kilobase million (FPKM)] were performed by Cufflinks (http://cole-trapnell-lab.github.io/cufflinks).

The Cuffdiff was used to screen the differentially expressed genes (DEGs) between normal NK cells and NKTCL cases. The DEGs threshold was set as follows: (1) p < 0.05; (2) log₂ (fold change) > 1 or log₂ (fold change) <-1; (3) q < 0.05. ggplot2 package in R was used to show the heat map and volcano maps.

Gene Ontology (GO) analysis was performed to show the unique biological significance based on differentially expressed genes. The Kyoto Encyclopedia of Genes and Genomes (KEGG) database was carried out to find out the important pathway. The ClusterProfiler packages (18) in R was applied to analysis and demonstrated GO annotations and KEGG pathway.

Expression dataset from SRP049695 was converted to the tab-delimited GCT format as follows: The first column of the GCT file contains gene symbols. The second column was filled with “NA.” Subsequent columns were filled with each sample's expression value. The following operations were carried out according to the protocol (http://www.gsea-msigdb.org/gsea/) (19, 20).

Differentially expressed mRNAs (fold change >2 or fold change < 0.5, P < 0.05) were taken into the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING). The confidence score was set at 0.9. The Molecular Complex Detection (MCODE) was used to analyze the core modules of the protein–protein interaction (PPI) network.

WGCNA package of R software was applied to uncover the correlation among genes. Firstly, expression data of DEGs was input into R software to inspect good genes and samples, SRR1648324 and SRR1648323 were excluded from the analysis due to the poor quality. The power of β was set at 14 to ensure a scale-free network. The minimum number of module genes was set at 30. The hierarchical clustering dendrogram summarized the Gene modules with different colors. Heat map and topological overlap matrix (TOM) plot were used to visualize the module structure. The threshold of output to Cytoscape was set at 0.6.

Gene network files exported from WGCNA analysis were input into Cytoscape software. The MCODE plugin of Cytoscape was used to calculate K-core value of each gene. GSE69406 dataset was used to validate the expression of the hub genes. GSE90597 dataset was used to plot the Kaplan–Meier survival curve in ggplot2 of R software.

In order to elucidate the molecular pathogenesis of NKTCL, heat maps of all mRNAs were shown in Figure 1A. To assess differential gene expressions between NKTCL and the normal NK cells, all genes were plotted in volcano plots. In total, 6,680 mRNAs displayed the differential expressions in NKTCL, including 5,664 upregulated and 1,016 downregulated mRNAs (Supplementary Figure 1A). Of note, 6,005 lncRNAs showed differential expressions, 4,968 lncRNAs were upregulated, and 1,037 lncRNAs were downregulated (Supplementary Figure 1B). Supplementary Figures 1C,D showed the dysregulated mRNAs or lncRNAs that were log₂Foldchange | >10 in NKTCL.

To get an insight into the function of DEGs of NKTCL, the upregulated and downregulated DEGs were analyzed by enrichGO or enrichKEGG function of clusterProfiler packages in R software, respectively. GO analysis results were enriched in the biological process (BP), the upregulated DEGs significantly enriched in the extracellular structure organization, circulatory system process, blood circulation, extracellular matrix (ECM) organization (Figure 1B), and the downregulated DEGs significantly enriched in T-cell activation, lymphocyte differentiation, regulation of hemopoiesis, positive regulation of cell adhesion, myeloid cell differentiation, and leukocyte cell–cell adhesion (Figure 1C). KEGG analysis showed that the upregulated DEGs significantly were enriched in ribosome, ECM–receptor interaction, and cell adhesion molecules (CAMs) pathways (Figure 1D), and the downregulated DEGs were enriched in T-cell receptor signaling pathway, chemokine signaling pathway, and FoxO signaling pathway (Figure 1E). To illustrate the interactions of the upregulated and downregulated pathways in the progression of NKTCL, the top 10 of upregulated and downregulated enriched pathways and their genes were used to construct a network. The T-cell receptor signaling pathway and CAMs pathway were considered to be at the central network, ECM–receptor interaction pathway was found to be the linker between the upregulated pathways and the downregulated pathways (Supplementary Figure 2A). In order to further understand the changes of the pathogenesis in NKTCL, all DEGs, including the upregulated and downregulated genes, were put into KEGG pathway analysis, the molecular network constructed by top 20 of enriched pathways demonstrated that the phosphoinositide 3-kinase (PI3K)/Akt pathway is the most important signal in NKTCL (Supplementary Figure 2B).

Considering that the number of genes input in GO and KEGG analysis is our artificially defined standard (p < 0.05 and |log₂ (FC)|>1), the results may be different under different standards (such as p < 0.01 and |log₂(FC)|>1). Here we used all expression data sets of SRP049695 for Gene Set Enrichment Analysis (GSEA) analysis, and the results show that complement and coagulation cascades, ECM receptor interaction, and focal adhesion pathways were enriched in the tumor samples (Figure 2A). T-cell receptor pathway, NK cell-mediated cytotoxicity, and adipocytokine signaling pathway were enriched in the control samples (Figure 2B).

To further investigate the DEGs and the potential protein levels, the STRING database was applied for revealing the core PPI network. As shown in Figure 3A, 982 nodes and 1,076 edges constructed the core PPI network. The network data were reanalyzed by MCODE to get the k-core which reflected the importance of each gene. The 40 highest k-core genes make up two important networks, one is the G protein-coupled receptor signaling pathway, the other is ECM receptor signaling pathway (Figure 3B), which may suggest the underlying mechanism in the NKTCL pathogenesis.

To clarify the key modules and genes in NKTCL, WGCNA was used to uncover the highly correlated genes and the co-expression networks of NKTCL. SRR1648323 and SRR1638324 samples were excluded from analysis after quality assessment (Figure 4A). The power of β was automatically set at 14 to ensure a scale-free network (Figure 4B). Gene modules were calculated, and the gray module represents genes that cannot be clustered into any other modules (Figure 4C). Finally, 18 gene modules were identified by the hierarchical clustering dendrogram. The interactions between gene modules were subsequently analyzed, and the TOM plot of a gene network was generated with the corresponding hierarchical clustering dendrogram and the modules (Figure 4D).

As WGCNA generated a huge gene network, we narrowed the network by set edges weight >0.6 to localize the hub genes. As a result, 325 nodes and 6,900 edges were screened out for further analysis. Figure 5A showed two crucial subnetworks. The most important genes were constructed by the edges degree and k-core. The subnetwork contains mRNAs, such as LMO3, PEG3, GRB14, ASB9, and lncRNAs, such as lnc-FRG2-13, lnc-F11-2, lnc-GALNT9-4, LINC01206, probably constituting the core of the networks. To visualize these hub genes, a total of 57 genes (k-core >60) and the top 10 core genes were listed in the volcano plots (Figure 5B). We observed the highest fold changes in LMO3 among these genes, and PEG3 is one of the most significantly changed genes. Further, the co-expression genes of LMO3 and PEG3 were plotted in Figures 5C,D.

In order to validate the possible key genes which screened from WGCNA, we explored the top 10 genes, namely, LMO3, PEG3, ASB9, GRB14, REEP1, SLC30A3, LY6K, TMEM59L, and LDHAL6B, in GEO database with the ID GSE69406, which contains three control cell line samples and five NKTCL samples. As a result, LMO3, PEG3, GRB14, and LDHAL6B were significantly upregulated in NKTCL, whereas ASB9 and others displayed no significant changes between NKTCL and the controls (Figure 6A). To reveal the relationship between the expression of these genes and the survival rate, another GEO dataset with the ID GSE90597 was used to plot the Kaplan–Meier survival curve. Figures 6B–J showed the OS between the high expression group and the median/low expression group. As a result, the overexpression of LMO3 (p = 0.044) and GRB14 (p = 0.027) are significantly related to the shorter period of OS.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.