A miRNA expression signature that separates between normal and malignant prostate tissues

Abstract

Background: MicroRNAs (miRNAs) constitute a class of small non-coding RNAs that post-transcriptionally regulate genes involved in several key biological processes and thus are involved in various diseases, including cancer. In this study we aimed to identify a miRNA expression signature that could be used to separate between normal and malignant prostate tissues.

Results: Nine miRNAs were found to be differentially expressed (p <0.00001). With the exception of two samples, this expression signature could be used to separate between the normal and malignant tissues. A cross-validation procedure confirmed the generality of this expression signature. We also identified 16 miRNAs that possibly could be used as a complement to current methods for grading of prostate tumor tissues.

Conclusions: We found an expression signature based on nine differentially expressed miRNAs that with high accuracy (85%) could classify the normal and malignant prostate tissues in patients from the Swedish Watchful Waiting cohort. The results show that there are significant differences in miRNA expression between normal and malignant prostate tissue, indicating that these small RNA molecules might be important in the biogenesis of prostate cancer and potentially useful for clinical diagnosis of the disease.

Figure 1

miRNA expression signature consisting of 30 differentially expressed miRNAs. Differentially expressed miRNAs (p <0.0001) were clustered and the results show that the expression profiles of these 30 miRNAs could be used to separate the normal (N) from the malignant (M) tissue samples with the exception for three malignant samples (6M, 8M and 20M).

Figure 2

miRNA expression signature consisting of nine differentially expressed miRNAs. Differentially expressed miRNAs (p <0.00001) were clustered and the results show that the expression profiles of these nine miRNAs could be used to separate between the normal (N) and malignant (M) tissue samples with the exception of one normal sample (20N) and one malignant sample (6M).

Figure 3

Principal component analysis of the nine miRNA expression signature. Principal component analysis of the nine miRNAs differentially expressed at the 0.00001 significance level. Only two samples (20N and 6M) were misplaced. The percentages on the axes describe the amount of variance that is picked up by the principal components in that direction. Green: malignant tissue sample, red: normal tissue sample.

Figure 4

Principal component analysis to find subgroups within the malignant samples. A set of 16 miRNAs could be used to separate the malignant samples according to Gleason scores (GS) with the exception of three samples: one GS 6 sample placed in the GS 7 group, one GS 7 sample placed in the GS 6 group, and one GS 7 sample placed in the GS 9 group. In the figure, one GS 7 sample is hidden behind another GS 7 sample. The strings between samples correspond to their nearest neighbors. Blue: GS 6, green: GS 7, yellow: GS 9, white: GS 10.

Figure 5

Number of misclassified samples in cross-validation procedure. Number of misclassified samples in each of the 15 repetitions performed in the generalization test. The error rate of the clusterings range from 0 to 40%, with an average error rate of 15.3%, meaning that on average, 1.5 of the 10 test samples was placed in a cluster where the majority of the samples belonged to the opposite class (malignant or normal).

Figure 6

Cross-validation procedure. Overview over the cross-validation procedure performed to test the generality of the clustering results.