Identification of immunotherapy-related lncRNA signature for predicting prognosis, immunotherapy responses and drug candidates in bladder cancer

We downloaded clinical data, miRNA expression data, lncRNA expression data, and somatic mutation data from the Cancer Genome Atlas (TCGA) database (https://​portal.​gdc.​cancer.​gov/​) [10]. All raw RNA-seq and miRNA-seq data are normalized to fragments of one millionth of a thousand bases (FPKM). We downloaded the expression data and clinical features from the GEO database as external data sets for verification. We also downloaded the IMvigor210 dataset, a set of expression data and clinical information from patients with urothelial cancer treated with atezolizumab (PD-L1 blocker) [11].

We set the IMvigor210 dataset CR and PR as the response group to PD-L1 blockers, and SD and PD as non-response groups to PD-L1 blockers. The differential expression of lncRNA and mRNA in immunotherapy was identified by R packet "DESeq2" (bioconductor.org/packages/devel/bioc/html/DESeq2.html) by comparing the response group and non-response group with threshold value (FDR < 0.05, | log Fc |> 0.75). The R package "ggplot2" (https://​www.​rdocumentation.​org/​packages/​ggplot2/​) is then used to visualize the volcano map and the Wayne diagram.

The best prognostic risk model was established by univariate Cox and multivariate Cox regression analysis. Risk score per patient for bladder cancer: risk score = Coef(TFAP2A-AS1) × Expr(TFAP2A-AS1) + Coef(SBF2-AS1) × Expr(SBF2-AS1) + Coef(RRN3P2) × Expr(RRN3P2). Expr represents the expression level of a particular gene, and Coef is obtained by univariate cox regression and then multivariate cox regression analysis, which represents the coefficient of gene Cox analysis in the model.

We used the LncATLAS database (https://​lncatlas.​crg.​eu/​) to identify subcellular localization of lncRNA [12]. The ceRNA network of lncRNA and the LncACTdb database (http://​bio-bigdata.​hrbmu.​edu.​cn/​LncACTdb/​) and ENCORI database (https://​starbase.​sysu.​edu.​cn/​) were predicted. where the threshold for the ENCORI database is set to high stringency [13, 14].

Tumor purity, matrix score, immune score and ESTIMATE score were calculated by R package "ESTIMATE" [15]. Single sample gene set enrichment analysis (ssGSEA) algorithm studies the level of immune infiltration between high-risk and low-risk groups based on different immune cell types and immune functions. We also evaluated the association between HNRNPA2B1 and immunoinfiltrating cells, including B cells, CD8 + T cells, CD4 + T cells, macrophages, neutrophils, and dendritic cells, in the Tumor Immune Evaluation Resource (TIMER, http://​cistrome.​shinyapps.​io/​timer/​).

The Tumor Immune Dysfunction and Exclusion (TIDE) algorithm can be used to infer the patient's effect on immunotherapy [16]. TIDE scores were inversely correlated with immunotherapy efficacy. Download the IPS scores for bladder cancer anti-PD-1 and anti-CTLA4 from the TCIA database (https://​tcia.​at/​home) to evaluate the association between risk scores based on immunotherapy-related lncRNA construction and the efficacy of PD-1 and CTLA4 blockers [17].

We use MOE software to simulate molecular docking of target proteins and small molecule inhibitors. The three-dimensional structure of the target protein is downloaded from the PDB database, and the small molecule inhibitors are FDA-approved drugs, downloaded from the zinc15 database and converted into three-dimensional structures in the MOE software. We optimize proteins, such as removing water molecules and ligands and replenishing hydrogen atoms and protons, and minimizing energy for small molecule inhibitors. Finally, the binding mode of HNRNPA2B1 with small molecule drugs was studied by docking simulation.

We used bladder cancer cell lines T24 and UC3, purchased from the Chinese Academy of Sciences Cell Bank. T24 and UC3 cells were cultured in RPMI-10 medium (Procell) supplemented with 1640% fetal bovine serum. siRNA purchased from JTSBIO Co. (China) was transfected with lipo3000. The bladder cancer cells were spread on the six-well plate 24 h before transfection, and transfection was carried out when the cell density grew to about 70%. First, 7.5 μ L siRNA and 125 μ L serum-free medium were mixed and incubated at room temperature for 5 min, and 8 μ L Lipo3000 and 125 μ L serum-free medium were incubated at room temperature for 5 min. The above two solutions were then mixed and incubated for 15 min. Finally, the mixture was added to the six-well plate, and the transfection efficiency was analyzed after 48 h. The siRNA sequence is as follows: GATCCAGATGGAGGAAACATCTCTGAT.

CCK8 is used to detect cell viability. After the cells were digested and washed and resuscitated, the cells were counted by cell counting board, and the si-SBF2-AS1-transfected T24 and UC3 cells were inoculated in a 96-well plate, then 2000 cells were inoculated into each well, with 6 multiple holes in each group, and 24 h, 48 h and 72 h groups. Cell counting kit-8 (CCK8) was added to each well at 24 h, 48 h and 72 h, respectively, and then the absorbance of each well was measured at 450 nm using multimode enzyme labeling instrument. Each experiment was repeated three times.

The cells were digested, cleaned and resuscitated, and then counted by cell counting board. About 2000 normal control and si-transfected T24 and UC3 cells were inoculated in petri dish and shook until the cells were evenly distributed in the dish. The cells were cultured in cell incubator for 2 weeks, during which the cell clone formation was observed regularly and the medium was changed after 1 week of culture. When the shape and size of the cell clone is appropriate, discard the culture medium, add PBS and wash for three times, then add 4% paraformaldehyde and fix 30 min at room temperature. The paraformaldehyde was discarded and washed with PBS for three times. Under the condition of avoiding light, the colony was stained with 1 ml 0.1% crystal violet for 30 min. Discard crystal violet, wash with PBS for three times, and then observe and take pictures under microscope after air-drying.

Transwell test was used to determine the invasive ability of cells. The transfected T24 and UC3 cells were re-suspended in serum-free RPMI-10 medium and cultured on the surface of the upper chamber coated with matrix glue. After 24 h, the cells adhered to the lower membrane were fixed with 4% paraformaldehyde and stained with crystal violet.

All statistical analyses were performed in R (http://​www.​r-project.​org/​). Two and more groups were compared between groups using Wilcoxon's test and Kruskal–Wallis test, respectively. Correlation was assessed by Spearman correlation analysis. Statistical analysis was performed using two-tailed unpaired t test, one-way analysis of variance, and two-way analysis of variance using GraphPad Prism software. A p-value ≤ 0.05 was considered statistically significant.

We first screened out genes with significant differences between responses to PD-L1 blockers and non-responses through Imvigor210 data sets (Fig. 1A). Then, we intersected the above genes with lncRNA in TCGA bladder cancer to obtain 6 lncRNAs. They are TFAP2A-AS1, SBF2-AS1, HHLA3, MIRLET7BHG, FLG-AS1 and RRN3P2 respectively (Fig. 1B). We analyzed the difference and prognosis of immune-related lncRNA, followed by multivariate Cox regression analysis. Results of univariate and multivariate cox analysis of 6 immunotherapy-related lncRNAs (Table S1). Finally identified three immunotherapy-related lncRNAs to build risk models, including TFAP2A-AS1, SBF2-AS1 and RRN3P2 (Fig. 1C). According to the median risk score, bladder cancer patients in the TCGA cohort were divided into low-risk group and high-risk group. The Kaplan–Meier survival curve showed that the survival rate of the high risk group was significantly lower than that of the low risk group. (Fig. 1D). We used GSE31684 as an external data set to verify the prognostic characteristics of the model, and the survival curve showed that the prognosis was poor in the high risk group (Fig. 1E). Analysis of the correlation between risk score and survival state heat map showed that the higher the risk score, the higher the mortality of patients (Fig. 1F). Through univariate and multivariate regression analysis of forest map results, risk score was an independent risk factor (Figs. 1G, H). We used the time-dependent and clinical characteristics of the receiver operating characteristics (ROC) to evaluate the sensitivity of the model to prognosis, and evaluated the results according to the area under the ROC curve (AUC). The results showed that ROC was more accurate in predicting the prognosis of BC patients (Fig. 1I, J).

The results of univariate and multivariate regression analysis showed that age, clinical stage and risk score were independent risk factors and could independently predict the survival rate of patients. Therefore, we construct a line map based on age, clinical stage and risk score, in which risk score is the main component of the total score (Fig. 2A). We further constructed a calibration curve for the diagram, and the results showed that the predicted survival rates of patients in the 1st, 3rd and 5th years were consistent with the actual survival rates (Fig. 2B-D).

Tumor mutation load (TMB) plays an important role in clinical practice, and we further study the difference of TMB between high-risk group and low-risk group. The results showed that the TMB of high risk group was significantly higher than that of low risk group (Fig. 3A). Survival analysis showed that patients with high TMB had more significant survival advantages than patients with low TMB (Fig. 3B). When the risk score was integrated with TMB, the bladder cancer patients in the TCGA cohort were divided into four groups. The high TMB and low risk groups had the most significant survival advantage, while the low TMB and high risk groups showed poor prognosis (Fig. 3C). In addition, we also showed the mutation frequency of the first 20 genes in different risk groups of bladder cancer by waterfall map (Fig. 3D, E).

Tumor microenvironment (TME) is not only a complex assembly of tumor cells, immune infiltrating cells, stromal cells and extracellular components, but also a key factor in tumor-immune interaction, affecting tumor response to immunotherapy [18]. We analyzed the difference of TME composition between the high risk group and the low risk group. The results showed that the tumor purity of the high risk group was significantly higher than that of the low risk group, while the ESTIMATEScore, immune score and matrix score of the low risk group were significantly lower than those of the high risk group (Fig. 4A-D). We further analyzed the differences of immune infiltrating cells and immune function between the two groups. Most of the immune infiltrating cells and immune function were significantly decreased in the low risk group, such as DCs, B cells, CD8 + T cells, T helper cells and immune checkpoints. (Fig. 4E). After that, we analyzed the immune checkpoint, and compared with the patients in the high-risk group, the low-risk group showed a higher level of immune checkpoint expression, suggesting the difference in the therapeutic effect of immune checkpoint inhibitor (ICI) between the two groups (Fig. 4F).

We then further studied the correlation between the risk score and the therapeutic effect of ICI. First, by calculating the TIDE score of each patient, the results showed that the patients in the low-risk group had a lower TIDE score, suggesting that the patients in the low-risk group had a better effect on immunotherapy (Fig. 4G). After verifying the correlation between immunotherapy and risk score on Imvigor210 dataset, we found that the risk score of patients who responded to PD-L1 blockers was significantly lower than that of non-response groups, that is, patients with low risk had better efficacy on PD-L1 blockers (Fig. 4H). We also evaluated the response of bladder cancer patients to PD-1 and CTLA4 blockers by IPS score. The results showed that IPS was higher in patients with low risk, suggesting that our low risk group was more effective in treating single or combined blockers of PD-1 and CTLA4 (Fig. 4I).

To explore the potential use of drugs in bladder cancer chemotherapy, we evaluated the semi-maximum inhibitory concentration (IC50) of chemotherapeutic drugs between the high-risk group and the low-risk group. The results showed that the IC50 values of Erlotinib and Lapatinib in the high risk group were lower, suggesting that the patients in the high risk group may benefit more from Erlotinib and Lapatinib. The low risk group has lower IC50 values for Bortezomin, Dasatinib, Elesclomol, Nilotinib, Rapamycin, Roscovitine, Sunitinib and Vinblastine, which means that the low risk group may benefit more from the above chemotherapeutic drugs (Fig. 5A). Using cellMiner database to analyze the correlation between genes and drug sensitivity, we found that RRN3P2 was positively correlated with most drug sensitivity, that is, the stronger the expression of RRN3P2, the more sensitive patients were to these drugs (Fig. 5B) [19].

According to the theory of ceRNA network mechanism, lncRNA can regulate mRNA expression by interacting with miRNA. Through lncATLAS localization analysis of immunotherapy-related lncRNA cells, it was found that SBF2-AS1 was mainly located in the cytoplasm and could be used as ceRNA to regulate mRNA expression through spongification (Fig. 6A). We further predicted the mRNA and miRNA regulated by SBF2-AS1 through LncACTdb and ENCORI databases, and intersected the predicted mRNA and Imvigor210 datasets of immunotherapy differential genes to get two mRNAs, namely ATP2C1 and HNRNPA2B1 (Fig. 6B). After that, we further increased the predicted threshold, found that the correlation of SBF2-AS1/has-miR-582-5p/HNRNPA2B1 signaling pathway was the highest, and predicted the base pairing of has-miR-582-5p associated with SBF2-AS1 and HNRNPA2B1 (Fig. 6C).

We further studied the prognostic significance of the expression of SBF2-AS1 and HNRNPA2B1, and determined the characteristics of OS by univariate and multivariate Cox regression analysis.

In the TCGA cohort of bladder cancer patients, the expression of SBF2-AS1 and HNRNPA2B1 is significantly correlated with the prognosis, and may be an independent prognostic factor in patients with Bladder cancer (Fig. 6D-G). In addition, the prognostic significance of hsa-miR-582-5p in bladder cancer was analyzed by COX regression and Kaplan–Meier curve, and it was found that it had no significant significance in the prognosis of bladder cancer (Table S2 and Figure S1A). The expression of SBF2-AS1 and hnRNPA2B1 in bladder cancer cells was significantly higher than that in normal tissues, but there was no significant difference in hsa-miR-582-5p between normal bladder tissues and bladder cancer cells (Figure S1B-C).

We analyzed the difference of TME composition between HNRNPA2B1 high expression group and low expression group. The results showed that the tumor purity of high expression group was significantly higher than that of low expression group, while the ESTIMATEScore, immune score and matrix score of low expression group were significantly higher than those of high expression group (Fig. 7A-D). We then evaluated the relationship between HNRNPA2B1 expression and immune infiltration of bladder cancer, and found that HNRNPA2B1 expression was positively correlated with the infiltration of CD8 + T cells, neutrophils and dendritic cells (Fig. 7E). The difference analysis of immune cell infiltration between HNRNPA2B1 high expression group and low expression group showed that CD8 + T cells in HNRNPA2B1 high expression group were significantly higher than those in low expression group (Fig. 7F). We also analyzed the difference of common immune checkpoints between HNRNPA2B1 high expression group and low expression group. Compared with HNRNPA2B1 low expression group, CD274 (PD-L1) increased significantly in HNRNPA2B1 high expression group (Fig. 7G). Finally, we observed in the Imvigor210 immunotherapy data set that the expression of HNRNPA2B1 in the response group was significantly higher than that in the non-response group to PD-L1 blockers (Fig. 7H). There was significant difference in SBF2-AS1 between anti-PD-L1-responsive group and non-reactive group (p < 0.05), and the expression level of SBF2-AS1 in anti-PD-L1-responsive group was significantly higher than that in non-reactive group (Figure S1D). By analyzing the relationship between SBF2-AS1, has-miR-582-5p and TME, we found that there was no difference in immune cell infiltration among SBF2-AS1 differential expression groups, but the level of immune infiltration in low has-miR-582-5p expression group was significantly higher than that in high has-miR-582-5p group (Figure S1E-I).

According to previous studies, targeted HNRNPA2B1 can reduce drug resistance of cancer cells to endocrine therapy [20]. Therefore, we docked HNRNPA2B1 with small molecules to find small molecules that can target HNRNPA2B1.

We use MOE software to simulate the binding posture of HNRNPA2B1 to small molecular drugs, in which the interactions between target proteins and small molecules are summarized in Table S3, and show the first eight small molecular drugs with the strongest binding to HNRNPA2B1, including Itc, Naloxegol, Dihydroergotamine, Trypan Blue, Eovist, Ceftolozane, Sqv and Ceftriaxone (Figs. 8A-H). For example, Trypan Blue (ZINC000169289767) forms hydrogen bond and ion bond interaction with HNRNPA2B1 amino acid residues Lys-173, Lys104, Lys186, Gly-106, Arg-185, Arg-190 and Met-193, in which Gly106 is used as hydrogen bond receptor and Lys-173, Lys104, Lys186, Arg-185, Arg-190 and Met-193 as hydrogen bond donors. Ceftriaxone (ZINC000028467879) forms hydrogen bond and ionic bond interaction with HNRNPA2B1 amino acid residues Lys-186, Lys-173, Asp-167 and Lys104, in which Asp-167 is used as hydrogen bond acceptor and Lys-186, Lys-173 and Lys-104 as hydrogen bond donors.

We verified the protein expression level of HNRNPA2B1 in normal tissue and tumor tissue by HPA database, the darker the color, the higher the expression in tumor tissue, suggesting that HNRNPA2B1 is highly expressed not only in normal tissue but also in tumor tissue, and the expression level is higher in tumor tissue (Fig. 9A). To investigate the role of SBF2-AS1 in bladder cancer cells, we designed siRNA to silence SBF2-AS1 expression in T24 and UC3 cells. Subsequently, CCK8, Transwell and colony formation assays were performed on T24 and UC3 cells transfected with si-SBF2-AS1. CCK8 results showed that the cell proliferation ability of NC group was significantly stronger than that of si-SBF2-AS1 group at 24, 48 and 72 h (Fig. 9B). Transwell results showed that knocking down SBF2-AS1 significantly limited the migration and invasion ability of T24 and UC3 cells (Fig. 9C). In addition, the colony formation assay of tumor cells showed that the proliferation ability of T24 and UC3 cells transfected with si-SBF2-AS1 was significantly lower than that of NC group (Fig. 9D). These results suggest that knocking down the expression of SBF2-AS1 can inhibit the proliferation, migration and invasion of bladder cancer cells.