Template switching causes artificial junction formation and false identification of circular RNAs

Chong Tang, Tian Yu, Yeming Xie, Zhuqing Wang, Hayden McSwiggin, Ying Zhang, Huili Zheng, Wei Yan

Preprint posted on February 06, 2018

Magic circles: the illusion of circular RNAs created by template-switching of MMLV-derived reverse transcriptases during RNA-Seq library preparation

Selected by Fabio Liberante


The preprint and its context

The study of circular RNAs has grown massively in the last decade or so, not least because of their increasing association with a huge array of both developmental and disease-related processes in cells. As demand to profile and understand these mysterious RNAs grows, so does the demand for pipelines to analyse the huge amount of RNA-Seq data that the scientific community continues to generate. However, such a rapid expansion of tools and techniques can often be at risk of overlooking caveats in the subtleties of the molecular processes used at the lab bench.

This preprint has highlighted such a caveat in using standard MMLV-derived reverse transcriptases (RT) to create cDNA libraries for RNA-Seq. The template-switching activity of these enzymes means that they add a small number of extra nucleotides at the 3’ end of transcripts. These can then anneal to a different part of the same RNA template and continue extending the transcript. The created transcript now contains junctions that were absent from the original RNA pool and which could be identified as circRNAs by bioinformatics analysis.


MMLV enzyme creating artificial junctions and "circRNAs"

Figure 1A from the preprint: Schematic of the mechanism through which the terminal transferase activity of MMLV-derived reverse transcriptase can add several non-templated nucleotides to the 3’ end of cDNA. Template-switching can then generate artificial junctions derived from linear RNA templates.


The authors examined whether detection of circular RNAs in samples might be an artefact of the reverse transcriptase used during cDNA synthesis. This hypothesis was tested through a number of different experiments, but the simplest one generated synthetic circular RNAs from linear ones using T4 RNA ligase and used these as positive and negative controls, respectively, in standard cDNA protocols. They indeed found that using an RT enzyme typically provided with RNA-Seq kits created these artificial junctions from linear RNAs, which could be detected by divergent primers, as illustrated in the figure above.

The authors also demonstrated that RNase R treatment before cDNA synthesis is not sufficient to remove all linear RNAs and can actually degrade circRNAs, likely through hydrolysis. Worryingly, this is something many RNA-Seq protocols have begun to include to enrich for circRNAs.

If these findings hold up under scrutiny by other groups, this could be a not insignificant setback for circRNA research. All is not lost, however. While they showed that their methodology identified considerably fewer “circRNAs” than a previous comparable study, the ones that were identified showed significant overlap, suggesting that the true positive results in existing studies are still there to be found, despite the artefacts that the datasets contain.


What makes this study interesting

This preprint initially caught my attention because “false identification of circular RNAs” is quite an alarming phrase to read in a title. Unfortunately, studies that highlight flaws in widely used protocols can still be difficult to sell convincingly to editors. This makes it an excellent example of the benefits of preprint publishing; it made the findings public quickly so that others don’t fall victim to the same problem whilst the paper undergoes the lengthy review and editorial process.

Part of its usefulness is in its simplicity – it’s only 4 figures. The authors, demonstrate the problem with a simple artificial template, then commonly-employed RNA spike-ins and finally a set of real-world samples. Finally, like any form of criticism, it is best when constructive. The authors offer a very simple solution to the problem: swap RT enzyme. They also offer a new method of validating identified circRNAs using High-resolution melting (HRM) analysis, something that could be implemented in most labs.


Questions and future directions

It would be interesting to know what alerted the authors to the issue of false junctions. Was it an unexpected result or was it based on their insight into the biochemistry of virus-derived reverse transcriptases?

The authors only describe MonsterScript as a suitable enzyme that can be easily substituted to solve the issue they identified. However, some cursory research seems to suggest that this product is no longer available. It would have been helpful if the authors provided more detail or speculation on which domains of this enzyme were likely modified by the supplier to eliminate the template-switching activity. This would allow other alternatives to be easily identified. Even more useful, they could have tested a number of other RT enzymes and perhaps even the more modern non-MMLV derived types, such as TGIRT enzymes or the recently published MarathonRT.

Further, the authors only used a single suite of tools, namely bwa and CIRI2, to analyse the generated RNA-Seq data. It would have been interesting to compare how other tools handled or filtered the low junction ratios that the authors say are usually ascribed to false junctions.


Further reading

An ultraprocessive, accurate reverse transcriptase encoded by a metazoan group II intron
RNA. 2018 Feb;24(2):183-195. doi: 10.1261/rna.063479.117
Zhao C, Liu F, Pyle AM

Emerging roles and context of circular RNAs
Wiley Interdiscip Rev RNA. 2017 Mar;8(2). doi: 10.1002/wrna.1386
Panda AC, Grammatikakis I, Munk R, Gorospe M, Abdelmohsen K

Tags: circrna, circrnas, circular rna, hrm, mmlv, monsterscript, rna-seq, rt, superscript

Posted on: 23rd July 2018 , updated on: 24th July 2018

Read preprint (No Ratings Yet)

  • Have your say

    Your email address will not be published. Required fields are marked *

    This site uses Akismet to reduce spam. Learn how your comment data is processed.

    Sign up to customise the site to your preferences and to receive alerts

    Register here