Abstract
RNA editing is a mutational mechanism that specifically alters the nucleotide content in transcribed RNA. However, editing rates vary widely, and could result from equivalent editing amongst individual cells, or represent an average of variable editing within a population. Here we present a hierarchical Bayesian model that quantifies the variance of editing rates at specific sites using RNA-seq data from both single cells, and a cognate bulk sample to distinguish between these two possibilities. The model predicts high variance for specific edited sites in murine macrophages and dendritic cells, findings that we validated experimentally by using targeted amplification of specific editable transcripts from single cells. The model also predicts changes in variance in editing rates for specific sites in dendritic cells during the course of LPS stimulation. Our data demonstrate substantial variance in editing signatures amongst single cells, supporting the notion that RNA editing generates diversity within cellular populations.
Introduction: The Diversity Problem in RNA Editing
The "central dogma" of molecular biology traditionally describes a linear and faithful transmission of genetic information from DNA to RNA to protein. However, post-transcriptional modifications like RNA editing challenge this simplicity. Specifically, C-to-U conversion mediated by the enzyme APOBEC1 can alter nucleotide sequences after transcription has occurred.
A significant question arises when analyzing "bulk" RNA-seq data, which averages signals from millions of cells. If a bulk sample shows a 20% editing rate at a specific site, does this mean every cell is edited 20% of the time (distributed editing), or does it mean 20% of the cells are fully edited while the rest remain untouched (concentrated editing)? If editing is concentrated, it creates "functional heterogeneity," where subsets of cells within a seemingly homogeneous population possess different protein-coding instructions despite sharing identical DNA.
Methodology: A Hierarchical Bayesian Approach
To resolve this ambiguity, the researchers developed a hierarchical Bayesian model designed to quantify the variance of editing rates across individual cells. The model operates on two levels:
- Single-Cell Layer: Uses a Binomial distribution to model the probability of observing edited reads within an individual cell.
- Population Layer: Uses a Beta distribution to couple the editing rates of different cells, allowing for the estimation of population-wide variance.
To prevent overfitting and handle cases with zero variance (the null hypothesis), the team utilized Penalized Complexity (PC) priors. This ensures the model favors simplicity unless the data provides strong evidence for high variability. Because standard single-cell RNA-seq (scRNA-seq) often suffers from low coverage, the model's predictions were validated using targeted RT-PCR with barcoded primers and Sanger sequencing, providing high-sensitivity "zoomed-in" views of specific transcripts.
Key Results and Biological Insights
The study utilized newly generated scRNA-seq and bulk data from murine bone marrow-derived macrophages (BMDMs) and publicly available datasets from dendritic cells stimulated with lipopolysaccharide (LPS).
Editing as a Generator of Population Diversity
The model confirmed that for many sites, bulk editing rates are indeed an average of highly variable single-cell rates. For example, in macrophages, the gene Cybb exhibited high variance: while the bulk average was 23%, one specific cell showed a 61.5% editing rate while eleven others showed 0%. In contrast, the gene Anxa5 showed much more uniform (distributed) editing across the population.
Dynamic Shifts During Immune Response
In dendritic cells, the variance in editing rates was found to be dynamic. Variance was highest in the "resting" state (0h) and decreased significantly after 2 hours of LPS stimulation. This suggests that unstimulated cell populations are more "plastic" or diverse in their editing signatures, potentially pre-determining which cells act as "early responders" to pathogens.
Hyperediting and Coverage Requirements
Simulations indicated that the model requires at least 20 reads per site per cell to accurately distinguish between high and low variance. Furthermore, the study highlighted "hyperediting," where transcripts are edited at multiple sites simultaneously—a phenomenon often overlooked by standard bioinformatics pipelines but captured by this single-cell approach.
Discussion and Future Directions
While powerful, the approach is currently limited by the inherent sparsity of scRNA-seq data; most transcripts lack the depth required for this analysis. Technical biases, such as GC-content bias in library preparation, also remain a factor.
The authors suggest that future research utilizing single-molecule long-read sequencing (e.g., PacBio or Oxford Nanopore) would be ideal. This would allow for the capture of the full "editing signature" of a single contiguous RNA molecule, providing an even clearer picture of how RNA editing generates functional diversity within cellular populations.
Summary updated