Validation of a microRNA profile in urine liquid biopsy with diagnostic and stratification value for bladder cancer classification, available through the open app BladdermiRaCan

Abstract

We aimed to identify a profile of urine microRNAs (miRNAs) with diagnostic and stratification potential in the whole range of bladder cancer (BC) categories, to avoid current invasive, harmful and expensive procedures. We collected a first morning urine sample from the screening (35 BC patients and 15 age- and gender-matched controls) and validation cohorts (172 BC and 94 controls). In the screening stage we analyzed the expression level of 179 miRNAs by real-time reverse transcription quantitative PCR in urine supernatants. miRNA levels in each sample were normalized by the levels of the previously identified and stably expressed miR-29c-3p. We performed an ordinal regression for each miRNA with False Discovery Rate (FDR) adjustment to identify dysregulated miRNAs, and an ordinal elastic net logistic regression model to identify a miRNA profile for BC diagnosis and stratification with the software R (v3.5.1). Next, we validated the most dysregulated miRNAs, and empirically identified the real miRNA targets in BC cells by miR-eCLIP immunoprecipitation and sequencing. We identified 70 dysregulated miRNAs in BC patients (p < 0.05 FDR-adjusted). With the expression level of 7 miRNAs in urine (miR-221-3p, miR-93-5p, miR-362-3p, miR-191-5p, miR-200c-3p, miR-192-5p, miR-21-5p) we could stratify BC patients and control subjects. To enable the global use of our model, we developed the free BladdermiRaCan online tool. Furthermore, we identified miR-21-5p, miR-425-5p and miR-99a-5p as follow-up markers for BC relapse, and miR-21-5p and miR-221-3p as markers for metastasis. These miRNAs were also dysregulated in BC tissue sections from a subgroup of patients from which urine samples were studied. In conclusion, we have validated and patented a 7-miRNAs urine profile able to diagnose and stratify BC patients; BladdermiRaCan will enable the global use of our model. The experimentally verified target proteins identified for these miRNAs may unravel novel therapeutic targets.

Keywords: Biomarker; Bladder cancer; BladdermiRaCan; Diagnosis; Liquid biopsy; Prognosis; Stratification; Urine; miR-eCLIP Immunoprecipitation; microRNA.

Fig. 1

Expression level of the dysregulated miRNAs in the screening and validation cohorts. Expression of the 7 miRNAs comprised in the elastic net logistic regression model for BC diagnosis and staging. A Heatmap, green represents relative miRNA underexpression and red represents relative overexpression. B Relative miRNA expression in the different BC categories and controls. Due to the normalization strategy employed, a lower relative expression value indicates a higher expression of the miRNA of interest (see Statistical analysis, Supplementary Material). Global FDR-adjusted p-values of the ordinal regression model are depicted for each miRNA, and pairwise differences among groups are detailed in Table S3. C The formula to calculate the probability of belonging to a given category (T ≥ 2GX, TaT1G3, TaG1 or control). Expression of the 6 main dysregulated miRNAs and the 7 predictor miRNAs in the validation cohort. D Heatmap, green represents relative miRNA underexpression and red represents relative overexpression. E Relative expression in the different BC categories and controls (follow the interpretation detailed for panel B). Global FDR-adjusted p-values of the ordinal regression model are depicted for each miRNA, and pairwise differences among groups are detailed in Table S4. F Bangdiwala agreement plot that represents the agreement between predicted and observed values obtained in the validation cohort with our ordinal regression model. Perfect matches between predicted and observed values are resented as black squares, adjacent categories are represented as grey squares and lack of match are represented as white squares. Rho value of the validated model = 0.61, Bangdiwala weighted value = 0.62, Bangdiwala unweighted value = 0.42, reflecting a high coincidence between predicted and observed values for each individual studied. When selecting the most probable category with the model and considering only controls vs. any BC group, sensitivity and specificity were 0.75 and 0.70, respectively