m6A-immune-related lncRNA prognostic signature for predicting immune landscape and prognosis of bladder cancer

The mutation data, public transcriptome data, and the corresponding bladder cancer clinical information were obtained from TCGA (https://​portal.​gdc.​cancer.​gov/​) database, and CIBERSORT immune fractions of these data were obtained from https://​gdc.​cancer.​gov/​about-data/​publications/​panimmune. The RNA sequence transcriptome data and GSE154261 clinical data were obtained from GEO. m6A regulators, including readers (LRPPRC, RBMX, HNRNPA2B1, HNRNPC, FMR1, YTHDF3, YTHDF1, YTHDF2, IGF2BP2, IGF2BP3, IGF2BP1, YTHDC1, and YTHDC2), writers (ZC3H13, METTL3, RBM15, RBM15B, VIRMA, WTAP, METTL16, and METTL14), erasers (FTO and ALKBH5), and were obtained from published articles.

Totally 618 m6A-immune-related lncRNAs were identified by Pearson’s correlation analysis between lncRNAs and m6A regulators. Correlations between lncRNAs and CIBERSORT immune fractions were analyzed and 490 immune-related lncRNAs were identified. Through Univariate Cox proportional hazard regression analysis, 49 lncRNAs were extracted, and the intersections of these lncRNAs were used for further analysis.

LASSO Cox regression was used for analysis of 47 shared prognostic m6A-immune-related lncRNAs, and 11 of them were ultimately used for construction of a risk model. The risk score was determined on the basis of the multiplication of lncRNA expression and each coefficient. The genomic location and correlation of these lncRNAs and m6A regulators were visualized by “circlize” package in R software. Patients were classified into low-risk group and high-risk group. The differences between high- and low-risk groups were analyzed in multiple dimensions, including clinical features, expression levels of these lncRNAs and m6A regulators, immune infiltration estimations (TIMER [19], CIBERSORT [20] and TCIA [21]), cell stemness index [22], tumor mutation burden and commonly mutated genes. Visualization was conducted by “pheatmap”, “ggplot2” and “maftools” package.

The prognostic capability of the risk model was evaluated using the Kaplan–Meiler survival curve. The specificity and sensitivity of the risk model were evaluated using the area under curve (AUC). The validation was performed on GSE154261 dataset.

GSEA was performed in R software by using “clusterprofiler” package, and revealed the associated signaling pathways. Visualization of pathways was done by using “pathview” and “ggnet” packages.

T24 and RT-112, two bladder cancer cell lines, were purchased from ATCC and cultured in RPMI-1640 medium supplemented with 10% FBS.

Cells were seeded into six-well plate and scratched with a pipette. The photos were taken at different time points after scratching. In invasion assay, Cells (1 × 10⁵) were seeded into a Boyden chamber (8 mm pore size) pre-coated with matrigel in serum-free medium, and RPMI-1640 containing 10% FBS was added to the bottom chamber. After 48 h, the filter lower surface cells were fixed, stained, and counted under a microscope.

The IC50 of Talazoparib was calculated based on dose–response growth curve. 200 untreated cells were seeded into each well of six-well plate, and cultured with or without Talazoparib for 2 weeks, and the colonies were then analyzed.

All data processing and statistical analysis was performed in R software (version 4.1.0). P value < 0.05 was considered statistically significant.

Figure 1A shows the workflow process. Totally 814 lncRNAs were included in this study. 618 lncRNAs were identified to be significantly correlated with at least one of the m6A regulators as analyzed by Pearson correlation analysis (Fig. 1B). Similarly, 490 lncRNAs were identified as immune-related lncRNAs as analyzed by Pearson correlation analysis between lncRNAs and CIBERSORT immune fractions (Fig. 1C). The overall survival- and progression-free survival-related prognostic was filtered using lncRNAs Univariate Cox regression analysis, and 49 lncRNAs showed prognostic values (Fig. 1D). Figure 1E shows the correlations between these 47 shared lncRNAs and CIBERSORT immune fractions as well as m6A regulators.

LASSO-Cox regression analysis was carried out on the shared lncRNAs that were significantly correlated to both m6A-regulators and CIBERSORT immune fractions to construct a risk score for prediction of overall survival. Figure 2A shows the risk model constructed using 11 selected lncRNAs, Fig. 2B shows the coefficient of each lncRNA, and Fig. 2C shows the forest plot of progression-free survival, overall survival, and disease-specific survival. Additional file 1: Fig. S1 shows the Kaplan–Meier plots of each lncRNA. Additional file 1: Fig. S2 shows the genomic location and correlation between these lncRNAs and m6A regulators. The risk scores were calculated, and based on the median value, they were classified to low- or high-risk subgroup. Figure 2D shows the correlations between risk score and clinicopathological characteristics. Higher risk score indicates later stage of bladder cancer patients, and most high-risk group tumors were basal squamous subtype, while most low-risk group tumors were luminal papillary subtype. Figure 2E shows the high-risk group with poorer prognosis.

The reliability and sensitivity of the risk model was assessed using ROC curves. As shown in Additional file 1: Fig. S3, the risk model only dependent on risk score alone may not be sufficient for predicting patients’ prognosis. Therefore, risk score was combined with other key clinical characteristics, including age of diagnosis, pathologic tumor stage, and three comprehensive nomograms to make prediction in OS, PFS and DSS. Figure 3A–C show the Kaplan–Meier plots comparing high-risk group and low-risk group of progression-free survival, overall survival, and disease-specific survival. By using time dependent ROC analysis in “timeROC” package, the AUC of these three nomograms at 1, 3 and 5 years are shown in Fig. 3A–C. All AUCs at 1 year are larger than 0.74, indicating this model shows important prognostic value. To validate the predictive accuracy of this model, it was applied to an independent dataset, which provides overall survival data, from GEO (GSE154261). ROC analysis manifested that the AUC of this model was close to that of the TCGA dataset, suggesting this model is robust in other datasets (Fig. 3D).

The relationships between the risk score and tumor mutation rate, mutation burden, tumor immune microenvironment score and tumor stemness indices were determined. Some commonly mutated genes in bladder cancer showed significant difference between low-risk group and high-risk group. The mutation rate of KDM6A in the low-risk subgroup was 12% lower than that in the high-risk subgroup (Fig. 4A, Additional file 1: Fig. S4). The patients in low-risk group had a higher mutation burden (Fig. 4B). Immunophenoscore was used to predict the potential response of bladder cancer patients to immune checkpoint inhibitors (ICI). The statistical analysis results showed that the immunophenoscore was larger in the low-risk group than in the high-risk group, which means patients in low-risk group might have stronger immunogenicity and might be more sensitive to immunotherapy (Fig. 4C, Additional file 1: Fig. S5). All of the infiltrating immune cells including CD4 + and CD8 + T cells, dendritic cells, neutrophils, B cells, and macrophages in the high-risk group were more abundant than those in the low-risk group as calculated using the Tumor Immune Estimation Resource (TIMER) algorithm (Fig. 4D). Interestingly, low-risk group showed higher stemness indices, suggesting cell stemness and immunogenicity play different roles in bladder cancer (Fig. 4E). Correlation matrix shows the relationship between risk score and these factors (Fig. 4F). The expression heatmap of 21 m6A regulators and 11 m6A-immune-related lncRNAs shows their relationships with risk score (Fig. 4G, H). Sorting patients by risk score, significant transition in tumor microenvironment can be observed in the heatmap of CIBERSORT fractions (Fig. 4I). The infiltration percentage of macrophage in the low-risk group was remarkably lower than that in the high-risk group, while naïve T cells shrank in high-risk group.

GSEA was conducted to identify the pathways associated with this risk model. As shown in Fig. 5A,“GO_STEM_CELL_PROLIFERATION”,“GO_EPITHELIAL_CELL_PROLIFERATION”,“HALLMARK_EPITHELIAL_MESENCHYMAL_TRANSITION”,“GO_REGULATION_OF_STEM_CELL_DIFFERENTIATION”,“GO_NEGATIVE_REGULATION_OF_B_CELL_ACTIVATION”,“GO_NEGATIVE_REGULATION_OF_IMMUNE_RESPONSE”,“GO_NEGATIVE_REGULATION_OF_MACROPHAGE_ACTIVATION”, and “GO_NEGATIVE_REGULATION_OF_T_CELL_MEDIATED_IMMUNITY” were significantly positively correlated to the risk score. This result suggests that the model was highly related to tumor proliferation, migration, invasion and tumor immune microenvironment. The relationships between these pathways were visualized by using “Pathview” and “ggnetwork” package in R (Fig. 5B, C). The relationships between 11 lncRNAs and their most correlated immune genes were also visualized (Fig. 5D).

The risk model was applied to CCLE, and the risk score of bladder cancer cell line was calculated. As shown in Fig. 6A, T24 had the highest risk score while RT-112 had the lowest risk score), consistent with previous studies of bladder cancer, which indicated that T24 had relatively higher malignancy than other cell lines [23‐25]. We further validated these results in vitro by performing colony formation, wound healing, and invasion assays (Fig. 6B–D). Furthermore, by applying drug sensitivity data in GDSC, we performed Pearson correlation analysis between cell line risk scores and the drug IC50 values. A significantly negative correlation was identified between risk score and IC50 of Talazoparib, which means patients with higher risk score are more sensitive to Talazoparib (Fig. 6E). The IC50 (2 µM) of Talazoparib in T24 was calculated (Fig. 6F), and colony formation assay was performed, which showed that T24 cells were sensitive to Talazoparib (Fig. 6G–I). This might provide a theoretical basis for targeted therapy of bladder cancer.