Individualized detection of TMPRSS2-ERG fusion status in prostate cancer: a rank-based qualitative transcriptome signature

Abstract

Background: TMPRSS2-ERG (T2E) fusion is highly related to aggressive clinical features in prostate cancer (PC), which guides individual therapy. However, current fusion prediction tools lacked enough accuracy and biomarkers were unable to be applied to individuals across different platforms due to their quantitative nature. This study aims to identify a transcriptome signature to detect the T2E fusion status of PC at the individual level.

Methods: Based on 272 high-throughput mRNA expression profiles from the Sboner dataset, we developed a rank-based algorithm to identify a qualitative signature to detect T2E fusion in PC. The signature was validated in 1223 samples from three external datasets (Setlur, Clarissa, and TCGA).

Results: A signature, composed of five mRNAs coupled to ERG (five ERG-mRNA pairs, 5-ERG-mRPs), was developed to distinguish T2E fusion status in PC. 5-ERG-mRPs reached 84.56% accuracy in Sboner dataset, which was verified in Setlur dataset (n = 455, accuracy = 82.20%) and Clarissa dataset (n = 118, accuracy = 81.36%). Besides, for 495 samples from TCGA, two subtypes classified by 5-ERG-mRPs showed a higher level of significance in various T2E fusion features than subtypes obtained through current fusion prediction tools, such as STAR-Fusion.

Conclusions: Overall, 5-ERG-mRPs can robustly detect T2E fusion in PC at the individual level, which can be used on any gene measurement platform without specific normalization procedures. Hence, 5-ERG-mRPs may serve as an auxiliary tool for PC patient management.

Keywords: 5-ERG-mRPs; Cross-platform; Prostate cancer; Qualitative signature; TMPRSS2-ERG fusion.

Fig. 1

Workflow of this study. A Prioritizing reference genes based on five metrics of variability for genes by stingscore method. B Identifying signature candidate gene pairs combined with ERG based on reverse and discrimination degrees of gene pairs. C A qualitative signature consisting of 5 gene pairs was developed by performing a forward selection approach, denoted as 5-ERG-mRPs

Fig. 2

Performance of 5-ERG-mRPs in the training and validation datasets. A Confusion matrix and ROC curve for the Sboner dataset by 5-ERG-mRPs. B Confusion matrix and ROC curve for the Setlur dataset by 5-ERG-mRPs. C Confusion matrix and ROC curve for the Clarissa dataset by 5-ERG-mRPs

Fig. 3

The comparison of performance between 5-ERG-mRPs and fusion prediction tools. Raincloud plots illustrate the distribution of molecular features between TFP and TFN samples grouped by the combination of 5-ERG-mRPs and fusion prediction tools or 5-ERG-mRPs alone, including A, B the activity score of estrogen response pathway, C, D the expression of ER-alpha protein, E, F the expression of ANO7 and ERG. G Oncoplot showing the difference of molecular characteristics above between TFP and TFN samples classified by fusion prediction tools. The left shows the label information of prediction tools and molecular features, and grouping annotations are provided at the bottom

Fig. 4

The difference in clinicopathological characteristics between groups based on 5-ERG-mRPs. A Kaplan–Meier survival analysis of groups classified by 5-ERG-mRPs in the Sboner dataset. B The difference of Gleason score between the TFP and TFN samples classified by 5-ERG-mRPs in the Sboner dataset. C The proportions of TFP and TFN tumors in lethal or indolent disease. D The Kaplan–Meier curves of OS for samples in the TCGA dataset. E Multivariate Cox analyses for 5-ERG-mRPs, Gleason score, age, PSA level, and stage were performed in TCGA. Solid circles represent the HR of death, and the open-ended horizontal lines represent the 95% confidence interval (CI). HR and 95% CIs were generated using multivariate Cox regression models

Fig. 5

Immune infiltration analysis of TFP tumors. A Identifying the relative infiltration of immune cell populations for 435 PC of reference samples in TCGA using ssGSEA method. The relative infiltration of each cell type is normalized into a z-score. B Relationship between infiltration of cell types executing anti-tumor immunity and cell types executing pro-tumor, immune suppressive functions. R coefficient and p value were calculated by Pearson’s correlation method. C Abundance of M1 and M2 immune cells between the TFP and TFN tumors from the TCGA dataset. D Immune cytolytic (CYT) score across 435 PC samples derived from TCGA sequencing data stratified by TFP (red) and TFN (blue) status. Immune score and stromal score computed by E estimated method and F xCell method between the TFP and TFN samples. p values in the heatmap and box plot were determined by one-sided Wilcoxon test. *p < 0.05, **p < 0.01, ***p < 0.001

Fig. 6

Single-cell analysis between TFP and TFN tumors. A–E UMAP visualization of 8469 cells among five PC samples. F Heatmap of differentially expressed genes (rows) between cells classified into inferred seven cell subsets. Bars on the top of the heatmap indicated the cell type corresponding to those of A with selected genes indicated. G Distribution of immune-related cells subpopulations in TFP vs TFN tumors. H–K UMAP of tumor cells identified by the Copycat method. L Histogram showed the prediction results of PC samples by 5-ERG-mRPs. The row represented the times that a sample was diagnosed as TFP or TFN sampler by a random method. *p < 0.05, **p < 0.01, ***p < 0.001. M Violin plots showing expression of ERG, TNPO1, EXTL2, DPP4, and ANG in among four samples

Fig. 7

Identification of universal dysregulated genes and potential therapeutic drugs. A, B Volcano diagram of differentially expressed genes in TCGA and Clarissa datasets. Limma with FDR less than 0.05 was considered to be significant. C Venn map of DEGs between TCGA and Clarissa cohorts. D Correlation heatmap between 39 universal genes and ssGSEA scores of fifty oncogenic pathways. Red (blue) dots represented a negative (positive) relationship. E The dysregulated frequency of 39 genes in TCGA and Clarissa datasets. F, G The dysregulation frequency of 39 universal DEGs in TCGA and Clarissa datasets. H Drug connectivity analysis using alteration-specific transcriptional signatures (CLUE, L1000). Twenty compounds that most strongly reverse or enhance the signature are highlighted