TL;DR
Long non-coding RNAs (lncRNAs) are increasingly recognized for their regulatory roles, but their precise impact on gene expression remains unclear. This study quantifies how much gene expression variation in breast cancer can be attributed to lncRNAs, revealing that they explain a significant portion, especially when acting in cis (nearby the target gene). The work also constructs a high-confidence regulatory network, moving beyond simple co-expression to incorporate mechanistic evidence.
What Problem Does This Paper Address?
While the existence of long non-coding RNAs (lncRNAs) has been known for some time, their functional roles are still being uncovered. A major question in the field is: to what extent do these lncRNAs actually drive the variation we see in gene expression across different individuals, particularly in the context of diseases like cancer? Previous approaches often relied on simple co-expression analysis, which can be misleading because it doesn't distinguish between direct regulation and shared regulation by other factors (like transcription factors). This paper aims to provide a more robust, quantitative estimate of lncRNA regulatory influence and to build a more reliable network of these interactions by accounting for confounding factors and incorporating mechanistic evidence.
Methods
The authors employed a multivariable linear regression approach to model gene expression. A key innovation was the strategy to first account for the influence of transcription factors (TFs) by regressing them out. This allowed them to isolate the variation in gene expression that could be attributed to lncRNAs. They used an Elastic Net regularization technique to select a parsimonious set of candidate lncRNA regulators for each gene, preventing overfitting. To further refine their findings, they also incorporated orthogonal criteria such as genomic proximity (cis-acting effects), ceRNA (competing endogenous RNA) mechanisms, and statistical interaction terms involving TFs and RNA-binding proteins (RBPs). Finally, they integrated these different lines of evidence to construct a high-confidence, multi-evidence lncRNA-gene regulatory network.
Datasets
The primary dataset used in this study was transcriptomic data from the The Cancer Genome Atlas (TCGA) breast cancer cohort (TCGA-BRCA project). This dataset included RNA-Seq expression measurements for 1,217 primary tumor samples. They also utilized data from the HumanTFDB for transcription factor identities, RBPDB V1.3.1 for RNA-binding protein information, Ensembl BioMart for gene and lncRNA locations, MCF10A breast cancer cell line Hi-C data for topologically associating domain (TAD) boundaries, DIANA-TarBase v8 and IncBase v.2 for miRNA-gene and miRNA-lncRNA interactions (for ceRNA prediction), and GRID-seq data for RNA-chromatin interactions. For cross-tissue validation, they used RNA-seq data from GTEx.
Key Insights and Ideas
- LncRNAs are significant regulators, even after accounting for TFs: The study demonstrates that lncRNAs can explain a substantial portion of gene expression variation, even after controlling for the influence of transcription factors, suggesting a direct regulatory role beyond shared regulation.
- Cis-acting lncRNAs are particularly influential: LncRNAs located physically close to their target genes (overlapping, same TAD, proximal) show a stronger regulatory impact compared to those acting in trans.
- Beyond co-expression: The construction of a "multi-evidence" network, integrating expression modeling with genomic location and protein interaction data, provides a more robust and mechanistically informed view of lncRNA regulation than co-expression alone.
- Protein interactions reveal mechanistic hypotheses: The identification of statistical interaction terms between lncRNAs and TFs/RBPs offers concrete hypotheses about how lncRNAs might exert their influence, for example, by acting as scaffolds.
Key Results
- For 16% of analyzed genes, lncRNAs could explain more than 20% of expression variation, even after accounting for TF co-regulation.
- Approximately 25-50% of putative lncRNA regulators were found to be in cis (overlapping or proximal) to their target genes.
- The study identified 643 high-confidence lncRNA-gene regulatory relationships, with a significant fraction showing cis-acting characteristics.
- The constructed multi-evidence regulatory network included 1,157 IncRNA-mRNA relationships, with an average of 2.3 targets per IncRNA and 5.0 regulators per mRNA.
- ceRNA mechanisms showed limited predictive value in this dataset, suggesting potential limitations in current ceRNA prediction methods or the actual prevalence of this mechanism in driving expression variation.
Limitations and Future Work
- Context Specificity: The authors acknowledge that their findings are specific to the breast cancer context and may not generalize to other biological contexts. Further work is needed to assess generalizability.
- ceRNA Prediction Limitations: The limited predictive power of ceRNA mechanisms might stem from the reliance on computational predictions for miRNA-gene and miRNA-lncRNA interactions. Future work could benefit from experimentally validated ceRNA pairs.
- TCGA Data Processing: The use of HTSeq-counts, which may have limitations in mapping reads to overlapping genes, was noted. While they performed re-calculation with updated methods, this is a potential area for refinement.
- Future Directions: The authors propose integrating more diverse mechanistic information into network inference, similar to how TF regulatory networks are built. They also suggest prioritizing lncRNAs with extensive global regulatory roles and enriching networks with causal relationships to generate testable hypotheses.
Summary updated