Evaluating the reliability of tools for mRNA annotation and IRES studies

This paper identifies critical artifacts in existing tools for studying cellular IRESes, such as false positives from back-splicing plasmids and misidentification of nuclear RNAs by smFISH and qRT-PCR, while proposing and validating more reliable orthogonal methods for accurate transcript annotation and IRES activity assessment.

The Gist

Imagine you are a detective trying to find a secret, hidden door in a massive, complex building (the cell). This door, called an IRES, allows a special delivery truck (the ribosome) to enter a room and start building something (making a protein) without needing the usual front door key (the 5' cap).

For decades, scientists have been hunting for these "cellular IRES" doors in human genes, hoping to find new ways to control how our bodies work. However, the tools they've been using to find these doors have been like faulty metal detectors: they beep loudly at things that aren't treasure, leading to many false alarms.

This paper is a "Reality Check" written by a team of scientists (May, Akirtava, and McManus) who are saying, "Stop! The map you are using is wrong, and the metal detector is broken. Here is how to fix it."

Here is the breakdown of their investigation using simple analogies:

1. The Broken Metal Detector: The "Back-Splicing" Plasmid

The Problem:
Recently, another group of scientists (Koch et al.) proposed a new, fancy metal detector called a "back-splicing circRNA plasmid." They claimed this tool was perfect because it supposedly filtered out noise. They used it to find several new "IRES doors" in genes like Hoxa9.

The Reality Check:
The authors of this paper tested that tool and found it was leaking.

• The Analogy: Imagine you are trying to hear a whisper (the IRES signal) in a noisy room. The new tool was supposed to silence the room. But, the authors discovered that the tool itself was actually creating a loud shout (a linear mRNA artifact) that sounded exactly like the whisper.
• What happened: The "plasmid" (the test tube setup) accidentally created a straight-line message (linear RNA) instead of a circular one. Because this straight message had a normal "front door" (a promoter), the cell's machinery opened it easily and made proteins. The scientists thought, "Wow, the IRES is working!" but really, they were just seeing the result of a broken tool creating a fake signal.
• The Verdict: The "IRES" signals they found were actually just promoters (normal start signals) masquerading as IRESes.

2. The Misleading Map: Bad Annotations

The Problem:
To find an IRES, you need to know exactly where the gene starts. If your map says the gene starts 1,000 miles earlier than it actually does, you might think a random patch of land is the entrance.

The Reality Check:
The authors looked at the Hoxa9 gene, which was the star of the previous study.

• The Analogy: The previous study thought the gene started way back in the "nuclear forest" (a long 5' UTR). They thought there was a secret door there.
• The Truth: The authors showed that the "forest" was actually a different building entirely (a non-coding RNA called pri-mir196b). The "secret door" they found was actually just the front door to that other building.
• The Evidence: When they looked at the data with a high-powered microscope (PacBio sequencing) and compared it to a different type of map (nAnT-iCAGE), they saw that the "long gene" didn't exist as a messenger RNA. It was just a non-coding RNA that stayed locked inside the nucleus (the office), never going out to the factory floor (cytoplasm) to make proteins.

3. The Ghost in the Machine: smFISH and qRT-PCR

The Problem:
The previous study used two other tools to prove their findings:

1. smFISH: Taking a photo of the RNA inside the cell.
2. qRT-PCR: Counting the RNA molecules in the "protein factories" (polysomes).

The Reality Check:

• The Photo (smFISH): The previous study took a picture and saw the "IRES" signal. The authors re-analyzed the photo and realized the signal was glowing only in the nucleus (the office), not in the cytoplasm (the factory). If it's not in the factory, it can't be making proteins! It was just a non-coding RNA hanging out in the office.
• The Count (qRT-PCR): They found RNA in the "protein factories." But the authors showed that you can accidentally drag non-working RNA into the factory just by how you mix the chemicals. It's like finding a toy in a car engine; it doesn't mean the toy is driving the car. The "IRES" RNA was just a passenger, not the driver.

4. The New, Reliable Tools

So, if the old tools are broken, what should we use? The authors suggest a "Gold Standard" toolkit:

1. The "Tornado" Plasmid: A new, cleaner version of the test tube setup that doesn't accidentally create the fake "shouts" (linear artifacts).
2. Direct RNA Transfection: Instead of putting DNA in a cell and hoping it works, just inject the finished RNA message directly. This bypasses all the messy "construction" steps where errors happen.
3. Better Maps: Use high-quality sequencing (PacBio) combined with a specific "cap-detecting" method (nAnT-iCAGE) to make sure you know exactly where the gene starts.

The Big Picture

This paper is a crucial correction in the scientific community. For years, researchers have been excitedly announcing they found "cellular IRESes," but many of these discoveries were likely false alarms caused by:

• Tools that created their own fake signals.
• Maps that pointed to the wrong building.
• Measurements that counted the wrong type of RNA.

The Takeaway:
Science is self-correcting. This paper doesn't say "IRESes don't exist." It says, "We need to be much more careful with our tools and our maps before we claim we've found them." By cleaning up the noise and using better methods, we can finally find the real secret doors in our cells.

Technical Summary

1. The Problem

The identification of cellular Internal Ribosome Entry Sites (IRESes)—sequences in mammalian mRNA 5' UTRs that allow cap-independent translation initiation—has been plagued by high rates of false positives.

• Historical Context: Researchers have long sought cellular IRESes, but standard plasmid-based reporter assays (bicistronic and back-splicing circular RNA/circRNA systems) are vulnerable to artifacts.
• The Specific Issue: Candidate IRES sequences often contain cryptic promoters or splice sites. In plasmid systems, these can drive the transcription of linear, monocistronic mRNAs that are translated via standard cap-dependent mechanisms, mimicking IRES activity.
• The Target of Critique: The authors focus on a recent study by Koch et al. (2025), which proposed a "versatile toolbox" for IRES discovery. This toolbox relied on:
  1. Back-splicing circRNA plasmids (ZKSCAN1-based) to separate promoter activity from IRES activity.
  2. smFISH and qRT-PCR for 5' UTR annotation and validation.
  3. Polysome profiling to detect translation.
• The Hypothesis: The authors argue that Koch et al.'s methods suffer from the same fundamental artifacts they sought to avoid, leading to incorrect conclusions about cellular IRES activity (specifically regarding Hoxa9, Chrdl1, and Dlx1) and erroneous mRNA annotations.

2. Methodology

The authors employed a multi-pronged approach combining experimental re-evaluation, orthogonal validation, and bioinformatic re-analysis:

• Detection of Trans-splicing Artifacts:

  • They reconstructed the ZKSCAN1 back-splicing plasmid system used by Koch et al. containing Hoxa9 and Chrdl1 candidate IRES sequences.
  • They performed RT-PCR with RNase R treatment (which degrades linear RNA but not circRNA) to distinguish between circular and linear transcripts.
  • They used specific primers to detect trans-spliced linear artifacts (chimeric RNAs formed by the promoter driving transcription of the reporter gene via trans-splicing).

• Orthogonal IRES Validation:

  • Tornado Plasmid System: They tested the candidate sequences in the Tornado circRNA plasmid, a system designed to minimize linear artifacts by using ribozyme cleavage and ligation rather than back-splicing.
  • Direct mRNA Transfection: They synthesized linear mRNAs with an A-cap/stemloop structure (to block cap-dependent scanning) and transfected them directly into cells. This is considered the "gold standard" for IRES validation as it bypasses plasmid promoter artifacts.
  • Controls: They included validated viral IRESes (HCV, EMCV, CrPV) as positive controls and scrambled sequences as negative controls.

• Transcript Annotation Re-analysis:

  • PacBio Iso-Seq vs. nAnT-iCAGE: They re-analyzed PacBio Iso-Seq data from mouse embryos (E11.5) and compared it against nAnT-iCAGE data (an orthogonal, cap-dependent method) to verify Transcription Start Sites (TSSs).
  • Isoquant Re-running: They re-ran the transcript assembly software Isoquant on the raw PacBio data without providing the user-defined, potentially erroneous annotations used by Koch et al., allowing for de novo assembly.
  • smFISH Re-evaluation: They re-analyzed the smFISH images from Koch et al. to determine the subcellular localization (nuclear vs. cytoplasmic) of the signals.
  • Polysome Contamination Test: They mixed non-translating "scramble" RNA into polysome extracts to test if non-translating RNAs can artifactually migrate to heavy polysome fractions.

3. Key Contributions and Results

A. Back-splicing Plasmids Generate False Positives

• Finding: The ZKSCAN1 back-splicing plasmid produces abundant linear, trans-spliced mRNA artifacts.
• Evidence: RNase R treatment eliminated the GFP signal, proving the reporter mRNA was linear, not circular. PCR confirmed these were chimeric transcripts formed by trans-splicing driven by the candidate IRES promoters.
• Implication: The GFP signal observed by Koch et al. was likely due to cap-dependent translation of these linear artifacts, not IRES-mediated translation of circRNAs.

B. Orthogonal Assays Show No Cellular IRES Activity

• Finding: When tested in the Tornado plasmid (low artifact background) and via direct mRNA transfection, the putative IRESes from Hoxa9, Chrdl1, and Dlx1 showed negligible activity.
• Quantification: The HCV viral IRES produced >1,400-fold more luciferase than the candidate cellular sequences. The candidates performed no better than random scrambled sequences.
• Conclusion: The "IRES" activity reported by Koch et al. was an artifact of the plasmid system, not a biological feature.

C. Flaws in 5' UTR Annotation and smFISH Interpretation

• PacBio Reliability: Contrary to Koch et al.'s claim that PacBio Iso-Seq suffers from 5' truncation, the authors found a high correlation (Pearson's R 0.8–0.9) between PacBio 5' ends and nAnT-iCAGE data. The "truncation" was actually due to reliance on incorrect GENCODE annotations.
• Hoxa9 Isoform Reality: Re-analysis of PacBio data (using de novo assembly) revealed that the long "IRES" isoform (3,226 nt) does not exist as a mature mRNA. Instead, the region is transcribed as pri-miR196b (a long non-coding RNA) and fusion transcripts.
• smFISH Specificity: The "IRES" probe used by Koch et al. targets the Hoxa9 promoter region, which is part of the nuclear-retained pri-miR196b. Re-analysis of the images showed 96% of the signal was nuclear, not cytoplasmic. Since IRESes function in the cytoplasm, this proves the signal was not from a translating mRNA.

D. Polysome Profiling Artifacts

• Finding: Non-translating RNAs (like the scramble control) can migrate into heavy polysome fractions during sucrose gradient centrifugation.
• Implication: Detecting RNA in polysome fractions via qRT-PCR does not prove active translation; it may reflect non-specific association or contamination.

4. Significance and Recommendations

• Debunking the "Toolbox": The paper demonstrates that the "toolbox" proposed by Koch et al. is fundamentally flawed. The back-splicing plasmid is susceptible to the same promoter-driven artifacts as bicistronic plasmids, and smFISH/qRT-PCR lacks the specificity to distinguish between coding mRNAs and nuclear non-coding RNAs.
• Establishing Best Practices: The authors propose a robust, artifact-free workflow for IRES studies:
  1. Annotation: Use PacBio Iso-Seq validated by nAnT-iCAGE (or similar cap-dependent methods) to define accurate 5' UTRs. Avoid relying solely on user-provided annotations in assembly software like Isoquant.
  2. Validation: Use direct mRNA transfection (with A-cap/stemloop controls) or the Tornado circRNA plasmid (which minimizes linear artifacts) to test for IRES activity.
  3. Controls: Always include validated viral IRESes (positive) and scrambled sequences (negative).
• Broader Impact: This work cautions against the over-interpretation of "cellular IRESes" and highlights the critical need for rigorous orthogonal validation to avoid decades of potential misannotation in the field of translational control. It suggests that many previously reported cellular IRESes may be artifacts of promoter activity and annotation errors.