The study, conducted by Hidenori Tani of Yokohama University of Pharmacy, takes an unusually careful comparative approach. Rather than comparing mRNA and lncRNA stability data from different experiments, different laboratories or different measurement platforms, Tani assembled half-life measurements derived from the same metabolic labeling and sequencing assays, allowing messenger RNAs and long non-coding RNAs to be evaluated on a genuinely level playing field. This design matters, because the field has been plagued by comparisons in which differences in assay chemistry, cellular context or transcript abundance could masquerade as biological differences in decay regulation. By holding the measurement method constant, the analysis isolates the one variable of interest: whether the sequence of a transcript encodes its own stability.

The findings for messenger RNAs confirm the power of codon-mediated regulation. When Tani re-estimated codon stabilization coefficients, quantitative measures of how strongly each codon is associated with transcript longevity, inside every cross-validation fold to eliminate information leakage between training and test data, the resulting model predicted mRNA half-life with a cross-validated coefficient of determination of 0.174 and a Spearman rank correlation of 0.44. Critically, the codon stabilization coefficient on its own, yielding an R-squared of 0.084, outperformed a six-member set of translation-independent features, which managed only 0.008. That ordering, with codon optimality beating a battery of sequence characteristics such as GC content, transcript length and known destabilizing elements, held consistently across three additional datasets, two distinct measurement technologies and two different species.

The picture for long non-coding RNAs could hardly be more different. In a carefully matched set of 364 lncRNAs measured in the same HeLa cell assay as the messenger RNAs, sequence-based models of half-life performed no better than chance, yielding an R-squared of minus 0.050, a value meaning the predictions were actually worse than simply guessing the average half-life for every transcript. The result was not a quirk of a small or unrepresentative sample. When Tani turned to the largest published lncRNA stability dataset, encompassing 33,285 transcripts, k-mer composition features achieved an R-squared of essentially zero, at 0.0007. Even the cross-validated rank correlation reported in that dataset’s original study, 0.091, is vanishingly small in absolute terms.

Here the study makes a subtle but important statistical argument. A Spearman correlation of 0.091 in a dataset of more than thirty thousand transcripts does clear the permutation null, meaning it is statistically detectable rather than pure noise. But detectability is not the same as explanatory power, and the contrast between the two RNA classes is one of magnitude, not of presence or absence. Messenger RNA half-life is meaningfully sequence-encoded; long non-coding RNA half-life, by every measure Tani applied, is not. The distinction has practical consequences for anyone attempting to model RNA decay: a signal that is real but negligible cannot support the kind of predictive machinery that works for coding transcripts.

The analysis also confronts one of the most fashionable tools in modern genomics: pretrained genomic language models. These deep neural networks, trained on enormous corpora of genomic sequence, have been touted as universal feature extractors capable of discovering biological signals without explicit programming. Tani tested models including HyenaDNA, which can process nearly complete transcripts at single-nucleotide resolution, and DNABERT-2. Even with one model covering 99.6 percent of transcripts in their entirety, the language models left messenger RNA prediction at an R-squared of just 0.042, and lncRNA prediction at values between 0.0002 and 0.0036. The implication is sobering: whatever the language models learned about genomic sequence, they did not uncover a hidden stability code in non-coding transcripts that simpler approaches had missed.

To rule out confounding, Tani stratified transcripts by predicted coding potential, testing whether a subset of lncRNAs with translated open reading frames might behave like messenger RNAs after all. They did not. Nor did differences in transcript length, GC content or sample size explain the gap: when lncRNAs and mRNAs were matched on all three characteristics and analyzed with an identical feature set, the messenger RNAs yielded an R-squared of 0.033 while the lncRNAs yielded 0.0002. Perhaps most convincingly, a signal-injection experiment established that the analytical pipeline could reliably detect effects explaining as little as 0.5 percent of the variance in half-life, a sensitivity threshold comfortably above every single lncRNA result reported in the study. If a sequence-based stability signal existed in these transcripts at even a modest level, the method would have found it.

The biological interpretation is as interesting as the statistical one. The mechanistic chain linking codon optimality to decay runs through translation itself: slow-moving ribosomes on unoptimized codons recruit decay factors, and the DEAD-box helicase Dhh1p in yeast monitors codon optimality directly to couple translation to destruction. Long non-coding RNAs, which are largely untranslated, simply cannot participate in this feedback loop. Their lifespans are instead governed by other forces, including RNA-binding proteins, nuclear retention mechanisms, structural elements and, in many organisms, the nonsense-mediated decay pathway, which can eliminate lncRNAs carrying premature stop codons. A growing literature documents hundreds of short-lived non-coding transcripts in mammalian cells and shows that nonsense-mediated decay restricts lncRNA expression even in RNAi-capable budding yeasts, underscoring that non-coding transcript turnover has its own logic, written in a different biochemical language.

The study’s broader lesson extends beyond RNA biology into the methodology of machine-learning-driven science. By re-estimating features within each cross-validation fold, Tani guarded against data leakage, the insidious practice pattern that has inflated reported performance across computational biology, and the work aligns with recent guidance on reproducibility in machine-learning applications. The bottom-line recommendation is blunt: mRNA-derived decay models should not be transferred to long non-coding RNAs without re-validation. For researchers designing RNA therapeutics, engineering synthetic transcripts or interpreting lncRNA dysregulation in disease, the message is clear. The stability of the non-coding transcriptome cannot be read from codon usage tables, and understanding how these molecules are scheduled for destruction will require looking elsewhere, at the proteins and structures that shepherd them through the cell.

Subject of Research: Codon optimality as a predictor of mRNA half-life and its failure to predict long non-coding RNA stability

Article Title: Codon optimality predicts mRNA half-life but does not transfer to lncRNAs

Article References: Tani, H. (2026). Codon optimality predicts mRNA half-life but does not transfer to lncRNAs. Molecular Genetics and Genomics, 301(1), Article 196. https://doi.org/10.1007/s00438-026-02520-1

Image Credits: AI Generated

DOI: 10.1007/s00438-026-02520-1

Keywords: codon optimality, mRNA stability, long non-coding RNA, RNA decay, RNA half-life, machine learning, cross-validation, genomic language models, nonsense-mediated decay, transcriptome, RNA-binding proteins, gene regulation

VPS41 encodes a component of the HOPS complex, a six-part molecular machine that acts as a tether, physically bringing lysosomes together with late endosomes and autophagosomes so these membrane-bound compartments can fuse. This fusion step is central to autophagy, the cellular recycling program that clears damaged proteins and organelles, and to endolysosomal trafficking, the logistics network that shuttles cargo through the cell’s degradation compartments. When this pathway falters, cellular waste accumulates, and the consequences are felt most severely in neurons, which are long-lived cells with limited capacity for self-renewal. Disruption of lysosomal function has been implicated in lysosomal storage disorders, neurodevelopmental syndromes, and neurodegenerative conditions, making genes like VPS41 prime suspects in unexplained neurological disease. Since VPS41-related disorders were first reported in 2020, the handful of documented patients have shown cerebellar ataxia, cognitive impairment, and dystonia, inherited in an autosomal recessive pattern, meaning both copies of the gene must be impaired for disease to emerge.

The patient at the center of the new report presented a clinical picture that was both characteristic and perplexing. He had cerebellar ataxia, hypotonia, and intellectual disability dating from early childhood, together with a full-scale IQ of 49 on formal cognitive assessment. Brain magnetic resonance imaging revealed anterior-predominant atrophy of the cerebellum, the movement-coordination structure at the back of the brain. Yet he also displayed features never before recorded in this syndrome: progressive swan-neck deformities of the second to fourth fingers on both hands and pes cavus, a high-arched foot deformity. Because pes cavus often signals peripheral nerve disease, clinicians repeatedly investigated that possibility. Motor nerve conduction studies performed at ages ten, fifteen, and twenty years consistently showed preserved conduction velocities in both arms and legs, and spinal cord imaging of the cervical and upper thoracic regions showed no abnormalities suggestive of posterior column degeneration. A 103-gene sequencing panel covering Charcot-Marie-Tooth disease and related inherited neuropathies, along with chromosomal microarray analysis, all came back unrevealing.

Exome sequencing eventually flagged four variants in VPS41, each confirmed by Sanger sequencing. One was a splice-site variant, c.385-2A>G, at a canonical position where the splicing machinery recognizes an exon boundary. The other three were missense variants that each change a single amino acid: p.Val137Met, p.Thr294Met, and p.Arg416His. Segregation analysis in the family added a complication. The healthy older brother carried none of the variants, and the father carried two of them, Thr294Met and Arg416His, on what was presumed to be one chromosome. But the inheritance pattern of the splice-site variant and the Val137Met variant could not be established, because the patient’s mother had died and her DNA was unavailable. Under the American College of Medical Genetics and Genomics classification framework, the splice-site variant was judged likely pathogenic based on its predicted severe effect on splicing and its rarity in population databases, while the three missense variants remained variants of uncertain significance. Without knowing which variants sat on which parental chromosomes, the team could not confirm that the patient had one damaging mutation on each of his two VPS41 copies, the configuration required for recessive disease.

That is where long-read genome sequencing made the decisive difference. Unlike standard short-read sequencing, which chops DNA into fragments of a few hundred bases and struggles to bridge complex haplotypes, long-read technology can span entire haplotype blocks in single DNA molecules, directly revealing which variants travel together on the same chromosome. The analysis showed that Val137Met sat in cis with the splice-site variant, meaning both occupied the same maternal allele, while Thr294Met and Arg416His lay in trans, on the opposite paternal copy. Because the splice-site variant disrupts splicing and the missense variants on the other chromosome were candidates for functional damage, the phasing narrowed the field to two alleles requiring laboratory validation. In addition, computational splicing prediction with SpliceAI assigned the Arg416His variant a high probability of disrupting the adjacent exon boundary, with delta scores of 0.96 for donor loss and 0.93 for acceptor loss, elevating it to the primary suspect on the paternal allele. With maternal DNA absent, this resolution would have been impossible using conventional approaches.

The researchers then examined RNA extracted from patient-derived lymphoblastoid cell lines, immortalized white blood cells that provide a renewable window into the patient’s gene expression. Transcriptome analysis revealed an abnormal splice junction consistent with skipping of exon 7, a defect the team attributed to the c.385-2A>G variant on the maternal copy. Quantitative RT-PCR showed that total VPS41 messenger RNA was significantly reduced in the patient’s cells compared with controls, and targeted RT-PCR with sequencing confirmed the exon 7 skip, which deletes 66 bases while preserving the reading frame. Because the deletion keeps triplets intact, the transcript evades nonsense-mediated decay, the cellular quality-control system that normally destroys messages carrying premature stop codons, allowing a shortened protein to be made. The paternal allele told a different story. A second primer set detected a low-abundance transcript lacking exon 15, and when the researchers treated the cells with cycloheximide, a drug that indirectly blocks nonsense-mediated decay, this aberrant band grew clearly visible. Exon 15 skipping removes 62 bases and shifts the reading frame, creating a premature stop codon, which explains why the defective message is normally degraded almost completely.

The protein-level consequences were equally informative. Western blot analysis showed that VPS41 protein was present in the patient’s cells but reduced to roughly 30 percent of control levels. Simple loss of one allele through nonsense-mediated decay would be expected to halve expression, so the deeper reduction suggests the exon-7-skipped protein produced by the maternal allele is itself partially unstable or targeted for accelerated degradation. To probe that idea, the team built structural models of the HOPS complex with AlphaFold and compared the wild-type machine against a version carrying the 22-amino-acid in-frame deletion, p.Ile129_Lys150del, that the exon 7 skip produces. Although the deleted region falls outside the annotated WD40 repeat domain spanning roughly amino acids 302 to 747, it lies near TPR-like and CHCR motifs that help shape the scaffold-like beta-propeller architecture of VPS41. The modeling showed a reshaped interaction landscape: some intersubunit interfaces, such as the A-E pairing, gained contact area and predicted stability, while the B-E and B-F interfaces weakened substantially, with destabilizing free-energy changes of 6.6 and 4.3 kilocalories per mole. The deletion, in other words, does not simply amputate part of the protein; it redistributes stress across the entire six-subunit complex and may undermine its overall integrity.

Transmission electron microscopy of the patient’s lymphoblastoid cells then provided the cellular corroboration. The images revealed characteristic endolysosomal abnormalities, including multiple multivesicular bodies and multilamellar bodies, precisely the kind of membrane-compartment pileup expected when the HOPS tethering machinery cannot complete fusion events efficiently. Taken together, the RNA findings, the protein reduction, the structural perturbation, and the ultrastructural phenotype supplied functional evidence for both alleles. Applying the American College of Medical Genetics and Genomics and Association for Molecular Pathology guidelines alongside Clinical Genome Resource sequence variant interpretation recommendations, the team reclassified the maternal splice-site variant as likely pathogenic based on PVS1-moderate, PM2, and PM3 evidence, and the paternal Arg416His variant as likely pathogenic based on PVS1 and PM2 evidence, formally establishing a compound heterozygous diagnosis of autosomal recessive spinocerebellar ataxia 29.

The case carries lessons that reach well beyond a single family. It demonstrates that long-read sequencing can resolve haplotypes in rare disease diagnostics even when parental samples are unavailable, a situation that arises frequently given the age at which many such patients are evaluated. It also confirms that RNA-based functional assays, performed directly on patient-derived cells, can convert variants of uncertain significance into actionable diagnoses by revealing the exact molecular consequence of each change. The newly reported swan-neck deformities and pes cavus expand the phenotypic spectrum of VPS41-related disease, though the authors caution that, in a single case, it remains uncertain whether these features are specific to the syndrome or coincidental, and additional cases will be needed to settle that question. Limitations acknowledged by the team include the single-patient design, the current cost and limited clinical availability of long-read sequencing, and the absence of updated spinal cord imaging in adulthood. Even so, the study marks a clear demonstration of how third-generation sequencing and transcript-level analysis together can close diagnostic gaps that once seemed permanent, offering a template for the thousands of rare disease patients still waiting for an answer.

Subject of Research: Biallelic VPS41 variants causing autosomal recessive spinocerebellar ataxia 29 resolved by long-read sequencing and RNA analysis

Article Title: Biallelic VPS41 Variants in Autosomal Recessive Spinocerebellar Ataxia 29 Resolved by Long‐Read Sequencing and RNA Analysis

Article References: Nakamura, N., Nishio, Y., Nyuzuki, H., Fukushima, A., Miura, M., Kobayashi, Y., Ishioka, R., Tsukada, K., Oka, Y., Tsujikawa, K., Morinaga, H., Inaba, M., Tohyama, J., Nakazawa, Y., Ikeuchi, T., Ono, T., Saitoh, S., & Ogi, T. (2026). Biallelic VPS41 Variants in Autosomal Recessive Spinocerebellar Ataxia 29 Resolved by Long‐Read Sequencing and RNA Analysis. Molecular Genetics & Genomic Medicine, 14(9), Article e70285. https://doi.org/10.1002/mgg3.70285

Image Credits: AI Generated

DOI: 10.1002/mgg3.70285

Keywords: VPS41, HOPS complex, spinocerebellar ataxia, long-read sequencing, haplotype phasing, lysosomal trafficking, autophagy, splicing defect, nonsense-mediated decay, rare disease genetics, compound heterozygosity, exon skipping