Common Pitfalls in CircRNA Detection and Quantification

Technical Summary: Common Pitfalls in CircRNA Detection and Quantification

Problem Statement
Circular RNAs (circRNAs) are increasingly recognized for their roles in biological processes and diseases, with their inherent stability making them attractive candidates for biomarkers and therapeutic targets. However, accurate detection and quantification of circRNAs remain technically challenging. While the field generally acknowledges that circRNAs lack a poly(A) tail and are therefore best detected in ribosomal RNA (rRNA)-depleted total RNA sequencing (RNA-seq) data, the practice of analyzing circRNAs in the more ubiquitous poly(A)-enriched RNA-seq datasets persists. This study addresses the critical gap in understanding how library preparation methods—specifically poly(A) enrichment versus rRNA depletion—impact the reliability of circRNA identification.

Methodology
To systematically evaluate the impact of library preparation on circRNA detection, the authors conducted a comparative analysis using matched datasets. They generated and analyzed two distinct types of RNA-seq data from the same biological samples:

1. Poly(A)-enriched RNA-seq: Standard libraries targeting mRNA.
2. Total RNA-seq with rRNA depletion: Libraries designed to retain non-polyadenylated transcripts, including circRNAs.

The study employed multiple circRNA detection algorithms to call back-splice junctions (BSJs) in both datasets. The analysis focused on the overlap of detected circRNAs between tools and datasets, the fraction of ribosomal reads as a proxy for sample processing quality, and the biological plausibility of the detected events.

Key Contributions and Results
The study yields several critical findings regarding the technical validity of circRNA detection:

• Poly(A)-Enriched Data Yields False Positives: The authors demonstrate that the relatively few circRNAs detected in poly(A)-enriched data are likely false positives. There is essentially no agreement between different detection tools when applied to poly(A)-enriched datasets, suggesting that the identified back-splice junctions are predominantly tool-specific artifacts rather than genuine circRNAs.
• Tool Agreement in Total RNA-seq: In contrast, when analyzing total RNA-seq data with effective rRNA depletion, there is moderate agreement between different detection tools. This concordance supports the validity of circRNA calls in this library type.
• Impact of rRNA Contamination: The quality of sample processing, quantified by the fraction of ribosomal reads, significantly influences detection sensitivity. High levels of rRNA contamination (indicating poor depletion) introduce bias and reduce the ability to accurately profile circRNAs.
• Validation Across Contexts: These findings were validated across diverse datasets and tissue types, indicating that the observed pitfalls are generalizable rather than specific to a single experimental condition.

Significance and Recommendations
The paper establishes a set of best practices to ensure robust circRNA analysis, directly challenging the validity of using poly(A)-enriched data for comprehensive circRNA profiling. The authors conclude that:

1. Library Selection: Total RNA sequencing with effective rRNA depletion is the preferred and necessary approach for accurate circRNA profiling. Poly(A)-enriched data should be considered unsuitable for this purpose.
2. Algorithmic Rigor: To enhance confidence in candidate selection, researchers should employ multiple circRNA detection tools and prioritize back-splice junctions that are identified by several algorithms.
3. Quality Control: Attention must be paid to the fraction of ribosomal reads during sample processing, as this technical factor substantially impairs detection accuracy.

By highlighting that ribosomal RNA contamination and inappropriate library selection are major, previously under-attended confounding factors, this work provides a framework to prevent the propagation of false positives in the circRNA field.