Nephroblastoma-specific dysregulated gene SNHG15 with prognostic significance: scRNA-Seq with bulk RNA-Seq data and experimental validation

The scRNA-seq data of GSE200256 was downloaded from the Gene Expression Omnibus (GEO) database(https://​www.​ncbi.​nlm.​nih.​gov/​gds) (Access date: September 11, 2022). Under the management of NCBI's Office of Cancer Genomics and Cancer Therapy Evaluation Program, the TARGET database is a special database for pediatric cancer registration. This database is publicly available, and the data records include clinical data and multiomics data of pediatric cancer patients. The RNA sequencing (RNA-seq) transcriptome data and clinical data of WT patients were retrieved from the TARGET database, containing a total of 129 WT and 6 healthy tissue samples. The uniformly normalized pan-cancer dataset was downloaded from the UCSC database (https://​xenabrowser.​net/​) (Access date: September 15, 2022). The GSE66405 dataset in the GEO database can be employed to verify whether there are differences in the expression of SNHG15 [13, 14].

The original expression data were pre-processed by the Limma function package in R software [15]. The human genome (hg38) and associated annotation file (version 31) were obtained from the Gencode database (https://​www.​gencodegenes.​org) [16]. This annotation file was used to identify lncRNAs. The molecules with gene types being "lincrna", "antisense", "processed transcript", "sense_intronic", "TEC", "bidirectional promoter lncRNA", "sense_overlapping", "macrolncRNA" or "non coding" were defined as lncRNAs. When multiple probes corresponded to the same lncRNA, the average of their expression values. In this study, |Log2 (FC)|> 1 and corrected P < 0.05 were selected as the criteria for identifying DEGs [15].

After abnormal outliers in TARGET data sets were eliminated, and the network construction and module clustering were performed. The WGCNA function package in R software was adopted to construct a scalefree gene coexpression network [11]. Pearson correlation matrix analysis was performed on all gene pairs to construct the weighted adjacency matrix. The adjacency matrix between genes were transformed into the topological matrix (TOM), which was a biologically relevant measure of gene similarity based on the co-expression correlation between two genes. Then, the correlation between a specific gene and all directly or indirectly associated genes was identified. The similarity between genes was calculated by the topological overlap method. These genes were classified by the dynamic tree cutting method, and modules were named by color. Subsequently, Pearson correlation analysis was performed to measure the correlation (P-value) of each module with histological stages, overall survival (OS), and relapse-free survival (RFS). The lncRNA (key lncRNA) in the gene module with the highest correlation coefficient was taken as a candidate prognostic molecular marker for the subsequent analysis.

The candidate lncRNAs were subjected to survival analysis based on the clinical data sets in the TARGET database. WT samples were separated into low- and high-SNHG15 subtypes, based on median SNHG15 expression values in the dataset. The prognosis of patients with WT in the sub-group was scrutinized by Kaplan–Meier analysis, and the value of SNHG15 expression for predicting the prognosis of patients with WT in the cohort was estimated using survival rate, ROC curves, and AUC values. Subsequently, Cox regression analyses were conducted to ascertain whether SNHG15 expression was an independent prognostic biomarker in patients with WT [17]. Based on this, an independent dataset, GSE66405, downloaded from the GEO database, validated the differences in SNHG15 expression in normal kidney tissues versus nephroblastoma. In addition, based on the starBase database (https://​starbase.​sysu.​edu.​cn/​index.​php) and LncACTdb3.0 data-base (http://​bio-bigdata.​hrbmu.​edu.​cn/​LncACTdb/​index.​html), these candidate lncRNAs were analyzed in an attempt to further explore the potential mechanisms by which these lncRNAs affect clinical features [18, 19]. These candidate lncRNAs were subjected to GO functional annotation and KEGG signaling pathway enrichment analyses, and correlation analyses with key pathway molecules.

The ESTIMATE algorithm was utilized to assess the immune cell infiltration and the stromal cell distribution, which contributes to clarifying the effect of the tumor microenvironment (TME) on tumor cells [20]. The CIBERSORT and ssGSEA [21] algorithms were utilized to assess the immune cell infiltration and response between the normal group and the WT group. The enrichment levels of 16 immune cells and 13 immune functions were further quantified with the ssGSEA algorithm to evaluate the immunological characteristics of both groups. Moreover, the potential immune checkpoints were also predicted in this study [22].

The original expression profile dataset (GSE200256) used for analysis was screened using the GEO public database. We first filtered the scRNA-seq data using the R package Seurat for data processing [23], setting each gene to be expressed in a minimum of 3 cells, and each cell to express at least 250 genes. The percentage of mitochondria and rRNA was calculated by the PercentageFeatureSet function and ensured that each cell expressed more than 500 genes and < 6000 genes, with < 10% mitochondrial content and at least 100 unique molecular identifiers (UMIs) per cell. Then, the data were normalized by log-normalization, and the FindVariableFeatures function was used to find highly variable genes. All the genes were scaled using the ScaleData function, and principal component analysis (PCA) downscaling was performed. Finally, the cells were clustered using the FindNeighbors and FindClusters functions to obtain cell subgroups. SingleR (v1.8.1), CellMarker database [24] and PanglaoDB database [25] were used for cell type annotation. In addition, functional enrichment of "HALLMARK" was performed on cancer cells with high/low SNHG15 expression with the use of "irGSEA" and "GSVA" in R [26]. CellChat was used to explore the potential interactions between core cells. To explain the molecular mechanism of WT progression, pseudo-temporal analysis was performed using "monocle2", and CellChat was used to explore potential interactions between core cells [27].

We downloaded the uniformly normalized pan-cancer dataset: TCGA TARGET GTEx (PANCAN, N = 19,131, G = 60,499) from the UCSC (https://​xenabrowser.​net/​) database, from which we further extracted the expression data of SNHG15 in each sample. Further, we extracted the expression data of SNHG15 in each sample, and further we screened the samples from: Solid Tissue Normal, Primary Solid Tumor, Primary Tumor, Normal Tissue, Primary Blood Derived Cancer - Bone Mar-row, and Primary Blood Derived Cancer - Peripheral, we further performed log2(x + 0.001) transformation for each expression value, and finally we also excluded those with less than 3 samples in a single cancer species, and finally obtained the expression data of 34 cancer species. We further extracted the expression data of SNHG15 and 44 class III RNA modification’s (m1A, m5C, m6A) genes in each sample, and next we calculated the Pearson correlation between SNHG15 and five classes of immune pathways of marker genes [28].

The survival analysis was conducted with the assistance of the survival, survminer, and PROC packages in R software, and the visualization was realized by the ggplot2 and forestplot packages in R software. The log-rank test was carried out, and the Kaplan–Meier survival analysis was performed to compare the survival difference between both groups. The time-dependent ROC analysis was performed to compare the accuracy of lncRNA prediction results. In terms of the Kaplan–Meier curve, the P-value and hazard ratio (HR) with a 95% confidence interval (CI) were obtained by the log-rank test and univariate Cox regression. The significant difference between both groups was determined by the Shapiro–Wilk normality test, independent samples T test, and Wilcoxon test. P < 0.05 indicated that the difference was statistically significant. All the above analysis methods were performed through R software (v4.2.1).

The workflow pertaining to the present study was shown in Fig. 1. The data sets related to nephroblastoma were downloaded from the TARGET database, and a total of 1839 DEGs were identified, including 1087 up-regulated genes and 752 down-regulated genes. Subsequently, the volcanic map and heat map of these DEGs were plotted (Fig. 2A, B). The expression data profiles of DEGs and their clinical data would be used for the subsequent WGCNA.

The gene co-expression network was constructed with the assistance of the WGCNA package in R software. It was found that outliers were not required to be eliminated, and hence all 1904 DEGs were included in WGCNA. Given the construction of the scale-free network and the moderate retention of average connectivity, β = 4 was selected to construct the co-expression network (the correlation coefficient = 0.98 was selected as the standard), as shown in Fig. 2C and D. All DEGs were divided into 9 modules based on WGCNA (Fig. 2E, F). In order to infer the clinical relevance of these genes, these gene modules were combined with the clinical data of patients with nephroblastoma. Tumor staging and survival data of patients were important evaluation indexes for selecting functional modules. The results demonstrated that the eigenvalue of the yellow module highly correlated with the histological type and pathological stage of patients with nephroblastoma, as shown in Fig. 2G. In order to explore the molecular markers related to prognosis, these lncRNAs in this module were selected for the subsequent survival analysis.

The modules with a strong correlation with the prognosis of patients with nephroblastoma were selected. Then, the module significance (MS) of each module was calculated after each module was correlated with clinical features. The higher the MS value, the more important the module. Based on MS comparison, the modules with a strong correlation with a certain clinical feature were regarded as hub modules. Subsequently, the gene significance (GS) and module membership (MM) were calculated. In the hub modules, the genes with │MM│ > 0.8 and │GS│ > 0.2 were regarded as candidate hub genes. As a result, 11 candidate hub genes were obtained, including CTD-2006C1.2, RP4-785G19.5, SNHG1, LRP4-AS1, ZFAS1, RP11-290L1.5, FOXC2-AS1, SNHG5, PXN-AS1, CCNT2-AS1, and SNHG15. The clinical data of patients with nephroblastoma were downloaded from the TARGET database. The Kaplan–Meier survival analysis was performed on the lncRNAs of 11 candidate hub genes with the assistance of the survival package in R software. These patients with nephroblastoma were divided into two groups according to the best cut-off value of each lncRNA expression value. The Kaplan–Meier survival analysis results revealed that the overall survival rate of patients in the SNHG15 high expression group was lower (P = 0.011), as shown in Fig. 3B. The time-dependent ROC curve indicated that SNHG15 was accurate in predicting the 1-year, 2-year, and 3-year survival rates, and the area under the curve (AUC) was 0.618, 0.551, and 0.542, respectively (Fig. 3C). The univariate Cox regression analysis results suggested that SNHG15 (HR = 1.196, P = 0.008) can be regarded as an independent prognostic factor (Fig. 3A).

The expression of SNHG15 in nephroblastoma was verified with another independent data set (GSE66405) from the GEO database (https://​www.​ncbi.​nlm.​nih.​gov/​). The statistical analysis results suggested that the average expression level in the normal group and the WT group was 9.012 ± 0.197 and 9.844 ± 0.555, respectively. The independent samples T test results suggested that the average expression level in the WT group was higher than that in the normal group (Fig. 3D). The difference between both groups was 0.831 (0.253–1.41). The expression level of SNHG15 in nephroblastoma was higher than that in normal kidney tissues, and there was a significant difference (t = 2.934, P = 0.006). The SNHG15-related mRNAs were predicted based on the StarBase database and LncACTdb3.0 database. Besides, the functional enrichment analysis and signal pathway enrichment analysis were carried out on the downstream target genes to infer the function of SNHG15, and the results were visualized (Fig. 3G). According to GO analysis, the down-stream target genes of SNHG15 were mainly involved in methyltransferase activity, DNA modifying enzyme, post-transcriptional regulation of gene expression, regulation of cellular amide metabolism, mRNA synthesis, and the processing, splicing and binding of RNA (Fig. 3E). According to KEGG analysis, the downstream target genes were mainly involved in splice formation, PI3K/AKT signaling pathway [29], IL-17 signaling pathway, AMPK signaling pathway, and cell cycle and apoptosis regulation, as shown in Fig. 3F. The PI3K/AKT signaling pathway is an intracellular signal transduction pathway, and it can respond to extracellular signals and promote metabolism, proliferation, cell survival, growth, and angiogenesis.

To reveal differences of immune infiltration in patients with high and low levels of SNHG15, we evaluated the proportions of stromal and immune cells of patients in the TARGET cohort, and found that stromal and tumor purity were higher in patients with high SNHG15 expression (Fig. 4A, B). Then, the relationship between SNHG15 and the immune component of patients in the WT group was assessed by the ESTIMATE and ssGSEA algorithms. The results indicated that there was a significant difference in the immune cells and immune function between both groups. B cells, M0 and M1 macrophages were found to be fewer in number while M2 macrophages, eosinophils, and neutrophils were more abundant in the patients with high SNHG15 expression. Moreover, the immune checkpoints were identified and there were significant differences in CD80, CD48, IDO2, CD276, CD28, and CD200R1 between both groups (Fig. 4C–E) [30].

The scRNA-seq data (GSE200256) used for analysis was filtered using the GEO database. We first filtered the scRNA-seq data using the R package "Seurat" for data processing, which filtered out unqualified cells for subsequent analysis (Fig. 5A, B). Then, the FindVariableFeatures function was used to find highly variable genes, and we found that 1500 genes were highly variable (Fig. 5C). The data were normalized by log-normalization. All genes were scaled using the ScaleData function and subjected to principal component analysis (PCA) downscaling, single-cell samples were scattered and distributed with logical results (Fig. 5D). Meanwhile, in PCA, we also selected 20 principal components (PCs) [31] with P.value less than 0.05 for subsequent analysis (Fig. 5E, F). Then, the core cells were classified into 19 independent cell clusters using the Uniform Manifold Approximation and Projection (UMAP) and t-distributed Stochastic Neighbor Embedding (t-SNE) algorithm (Fig. 5G, H) [32].

The different cell clusters were annotated by finding marker genes through the "singleR" package (v1.8.1), CellMarker database, and PanglaoDB database (Fig. 6A), resulting in 10 cell clusters, namely B cells, cancer cells, M2 macrophages, regulatory T cells, plasma cells, monocytes, CD8+ T cells, monocytes, fibroblasts, and epithelial cells (Fig. 6B, C) [33]. Specific gene marker expression in each cell type was shown in the form of a Heatmap (Fig. 6D).

In addition, functional enrichment of "HALLMARK" was performed on cancer cells with high/low SNHG15 expression with the use of "irGSEA" and "GSVA" in R. GSVA analysis implied that “wnt/β-catenin signaling”, “PI3K‐AKT‐MTOR signaling”, and “TGF-β signaling” were enriched in the cells with high SNHG15 expression (Fig. 6E). These results suggested that the high expression of SNHG15 in cancer cells might play a role in promoting M2 macrophage infiltration through these pathways, thus affecting prognosis.

Epithelial-mesenchymal transition (EMT) occurs normally throughout development, and dysregulation of EMT can lead to tumorigenesis [34]. EMT was shown to exhibit heterogeneity among different cell types (Fig. 6F). In this study, we found that SNHG15 exhibited heterogeneity among different cell types (Fig. 6G), with the largest proportion of tumor cells and M2 macrophages. The difference in the expression levels of tumor cells between the SNHG15 high expression group and the SNHG15 low expression group was statistically significant, and differentially expressed genes were visualized in the two groups (Fig. 6H).

Pseudo-temporal analysis was performed separately for all clusters annotated in order to explore their differentiation directions with the Monocle 2 algorithm. The results showed that WT cells gradually followed 3 directions of differentiation (Fig. 7A, B). Epithelial cells divided earlier than other cells and differentiated into two branches, one of which was dominated by EMT (Fig. 7E), and the other branch gradually differentiated in multiple directions, dominated by tumor cells highly ex-pressing SNHG15 and M2 macrophages (Fig. 7D). Furthermore, we inferred intercellular communication networks to predict intercellular communication according to specific pathways and ligand receptors. The Heatmap of the number of ligand-receptor pairs showed that cellular communication occurred more frequently in M2 macrophages, B cells, tumor cells, and epithelial cells (Fig. 7C, G).

Specifically, the frequency and intensity of interactions between cancer cells and M2 macrophages, and between cancer cells and epithelial cells were high (Fig. 7G). In addition, plasma cells, CD8+ T cells had relatively few interactions with other cells. The study of ligand-receptor pairs showed that the key receptor-ligand pairs between macrophages and tumor cells were mainly NRXN3-NLGN1/NRXN1-NLGN1/NRG3-ERBB4/NRG1-ERBB4 (Fig. 7F).

We calculated the expression differences between normal and tumor samples in each tumor using R software and performed differential significance analysis using unpaired Wilcoxon Rank Sum and Signed Rank Tests (Fig. 8A). We observed significant upregulation in 20 tumors such as GBM (Tumor:7.39 ± 0.53, Normal:5.73 ± 1.46, p = 8.0e−63), GBMLGG (Tumor: 7.12 ± 0.52, Normal: 5.73 ± 1.46, p = 3.8e−156), LGG (Tumor:7.04 ± 0.49, Normal:5.73 ± 1.46, p = 5.1e−123), BRCA (Tumor. 7.36 ± 0.81, Normal: 7.14 ± 0.72, p = 5.9e−9), LUAD (Tumor: 7.40 ± 0.78, Normal: 7.09 ± 0.93, p = 9.9e−12), ESCA (Tumor: 7.11 ± 0.80, Normal: 6.24 ± 1.32, p = 2.0e−38), STES (Tumor: 6.97 ± 0.86, Normal: 6.09 ± 1.44, p = 1.9e−78), COAD (Tumor: 7.50 ± 0.55, Normal:6.34 ± 1.80, p = 2.8e−75), COADREAD (Tumor: 7.53 ± 0.56. Normal: 6.37 ± 1.78, p = 1.6e−87), STAD (Tumor: 6.90 ± 0.87, Normal: 5.59 ± 1.68, p = 9.8e−48), LIHC (Tumor:5.80 ± 0.90, Normal: 4.99 ± 0.56, p = 2.1e−27), WT (Tumor: 7.16 ± 0.86, Normal: 6.43 ± 1.55, p = 5.9e−9), SKCM (Tumor:7.69 ± 1.16, Normal: 6.49 ± 0.38, p = 1.5e−31), THCA (Tumor:7.49 ± 0.78, Normal:6.99 ± 1.00. p = 6.9e−33), READ (Tumor: 7.64 ± 0.58, Normal: 7.10 ± 0.47, p = 2.0e−3), PAAD (Tumor:7.19 ± 0.68, Normal: 4.76 ± 1.71, p = 1.9e−53), TGCT (Tumor: 6.35 ± 1.10. Normal: 5.75 ± 0.47, p = 3.0e−13), ALL (Tumor: 5.34 ± 1.10, Normal: 3.66 ± 1.34, p = 1.3e−29), LAML (Tumor:6.79 ± 0.62, Normal: 3.66 ± 1.34, p = 5.3e−74), CHOL (Tumor:6.10 ± 0.94, Normal: 5.22 ± 0.25,p = 2.8e−3), we observed significant downregulation in seven tumors such as UCEC (Tumor: 6.45 ± 0.99, Normal: 7.04 ± 0.46, p = 5.7e−3), CESC(Tumor:6.56 ± 0.89. Normal: 7.33 ± 0.56, p = 1.2e−3), KIRP (Tumor: 6.03 ± 0.85, Normal: 6.43 ± 1.55, p = 1.8e−12), KIPAN (Tumor: 6.40 ± 0.94, Normal: 6.43 ± 1.55, p = 0.01), OV (Tumor: 6.44 ± 1.18, Normal: 7.15 ± 0.42, p = 1.2e−13), UCS (Tumor: 6.70 ± 0.93, Normal: 7.18 ± 0.39, p = 4.8e−3), KICH (Tumor: 5.68 ± 0.92, Normal: 6.43 ± 1.55, p = 1.8e−12). SNHG15 is closely linked to marker genes for three classes of RNA modifications (m1A (10), m5C (13), and m6A (21)) genes, and the exact mechanism needs to be further explored, which confirms that SNHG15 plays a major role in tumors (Fig. 8D).

The expression of SNHG15 was measured by qRT-PCR in WT tissues and paired normal tissues (Fig. 8B). The siRNA interference of SNHG15 significantly inhibited the expression of SNHG15 compared with the control group (P < 0.05), as shown in Fig. 8C.

In addition, a series of experiments were performed to investigate the effects of SNHG15 on the proliferation, invasion and migration of nephroblastoma cells. We investigated the effect of SNHG15 on the proliferation and viability of nephroblastoma cells using CCK-8 assay (Fig. 8F). Wound healing assay showed that siRNA interference with SNHG15 expression significantly inhibited G401 cell proliferation migration and promoted apoptosis compared with control (Fig. 8G). These results were further supported by Transwell assay, siSNHG15 reduced the invasion and migration of nephroblastoma cells, as shown in Fig. 8E. In addition, we also detected the knock-down of SNHG15 by Western blotting for the expression of EMT-related core proteins. Compared with the siNC group, knockdown of SNHG15 decreased the expression of EMT-associated core protein (Fig. 8H). According to the results of cell cycle analysis, knocking down the expression of SNHG15 resulted in its blockage in the S phase, which in turn inhibited the proliferation of tumor cells (Supplementary figure 1).