Differentially expressed microRNAs in lung adenocarcinoma invert effects of copy number aberrations of prognostic genes

Abstract

In many cancers, significantly down- or upregulated genes are found within chromosomal regions with DNA copy number alteration opposite to the expression changes. Generally, this paradox has been overlooked as noise, but can potentially be a consequence of interference of epigenetic regulatory mechanisms, including microRNA-mediated control of mRNA levels. To explore potential associations between microRNAs and paradoxes in non-small-cell lung cancer (NSCLC) we curated and analyzed lung adenocarcinoma (LUAD) data, comprising gene expressions, copy number aberrations (CNAs) and microRNA expressions. We integrated data from 1,062 tumor samples and 241 normal lung samples, including newly-generated array comparative genomic hybridization (aCGH) data from 63 LUAD samples. We identified 85 "paradoxical" genes whose differential expression consistently contrasted with aberrations of their copy numbers. Paradoxical status of 70 out of 85 genes was validated on sample-wise basis using The Cancer Genome Atlas (TCGA) LUAD data. Of these, 41 genes are prognostic and form a clinically relevant signature, which we validated on three independent datasets. By meta-analysis of results from 9 LUAD microRNA expression studies we identified 24 consistently-deregulated microRNAs. Using TCGA-LUAD data we showed that deregulation of 19 of these microRNAs explains differential expression of the paradoxical genes. Our results show that deregulation of paradoxical genes is crucial in LUAD and their expression pattern is maintained epigenetically, defying gene copy number status.

Keywords: copy number aberrations; gene regulatory network; lung adenocarcinoma; microRNA; prognostic signature.

Figure 1. Flowchart depicting sequence of analyses/computational steps as performed and datasets as used

For more details see Materials and Methods section.

Figure 2. Association between chromosomal copy number aberrations and differential expression of genes

(A) Frequencies of gains (pointing outbound) and losses (pointing inbound) of the given chromosomal region as obtained from integrative analysis of three aCGH datasets. The aberration frequencies are depicted in range from 0-50% and regions with significant frequency of aberrations are highlighted by color (orange – losses, green – gains). Precise chromosomal locations of these paradoxical genes are depicted in the circular plot. (B) Tumor-vs-normal expression fold change of the paradoxical genes, obtained across 10 publicly available datasets. In Figures A and B, symbols of the downregulated genes are labeled red, while the symbols of the upregulated genes are labeled blue. (C) Venn diagram showing overlaps between up-/downregulated genes and genes residing within the regions of chromosomal copy number gain, or loss.

Figure 3. Frequency of deregulation and CNAs of paradoxical genes

Barplot at the left shows frequencies of paradoxical and non-paradoxical co-occurrence of deregulated expression and CNAs. Barplots depicting frequencies of up- and downregulation (middle), gain and losses (right) of 70 validated paradoxical genes, as occur across TCGA LUAD samples. Colors of the gene labels indicate their deregulation/CNA status as obtained from the integrative analysis.

Figure 4. Ranking of the differentially expressed miRNAs as reported across 9 LUAD miRNA studies

The lower the rank the greater the reported significance (and/or expression fold change) of the corresponding miRNA. Height of the bars denotes total number of reported miRNAs in each study.

Figure 5. The network of interactions between deregulated miRNAs and their paradoxical gene targets as obtained from mirDIP

Rectangles and circles represent miRNAs and genes, respectively. Red color denotes downregulated transcripts, while blue denotes upregulated ones. Size of nodes corresponds to number of interactions (degree). Solid red lines indicate miRNA:gene interactions between inversely deregulated transcripts, indicating potential causal associations.

Figure 6. Correlation between deregulated miRNAs and paradoxical genes

(A) Partial correlations between deregulated miRNAs and paradoxical genes as measured across TCGA LUAD data (red denotes negative correlation, blue denotes positive correlation, darker shade indicates significant correlations, p < 0.05). Plus signs denote partial correlation with causal explanation of gene deregulation, minus signs denote correlations that are significant but non-explanatory. (B) Barplot showing multiple correlations between paradoxical genes and en block deregulated miRNAs as calculated across TCGA LUAD data. Curve on the right depicts distribution of the same measure across all the genes in the TCGA LUAD data. Dashed line denotes 95th percentile of the distribution; there are 23 (32.9%) paradoxical genes whose multiple correlation coefficient falls among the top 5% of the highest values.

Figure 7. Clinical significance of the paradoxical genes

(A) Venn diagram showing overlapping subsets of paradoxical genes with up-/ downregulated expression and subsets with positive (HR < 1, FDR < 0.05) and negative (HR > 1, FDR < 0.05) association with LUAD prognosis, as obtained from KMplot. (B–D) Kaplan-Meier plots showing survival curves in the three independent validation cohorts, as stratified based on the Cox proportional hazards calculated from paradoxical genes expression. Numbers in the bottom left, indicate resulting hazard ratio (HR), associated statistical significance of patient stratification (p), concordance index (c-index), area under ROC curve (AUC) calculated at five years.