Proof of Concept

Ribosomal stop-codon readthrough (RT), or “readthrough” for short, refers to the phenomenon where a small fraction of ribosomes bypass a normal stop codon during translation and continue elongation due to specific tRNA recognition or the influence of surrounding nucleotide and protein environments. This results in the synthesis of a protein isoform with an extended C-terminal peptide. Functional translational readthrough (FTR) is not a random translational error but a regulated process that occurs under specific sequence contexts and cellular conditions. It represents a conserved mechanism contributing to protein functional diversity and subcellular targeting. Systems biology analyses and experimental validation have shown that the low-abundance C-terminal extensions produced by readthrough can carry functional signal peptides, thereby affecting protein subcellular localization, interactions, and metabolic pathway partitioning

Peroxisomal targeting signals (PTS) are short peptide sequences that mediate the recognition and import of proteins into the peroxisomal matrix. They are mainly classified into two types: PTS1, a tripeptide (or its variants) located at the extreme C-terminus of the protein, with the canonical sequence Ser-Lys-Leu (SKL). PTS2, a conserved nonapeptide located near the N-terminus. PTS1 is recognized in the cytosol by the receptor Pex5, which mediates the formation of a receptor–membrane complex that transports the cargo into the peroxisome matrix. The recycling and quality control of the receptor are coordinated by multiple PEX proteins. The sequence variants of PTS1 and their adjacent C-terminal residues together determine the efficiency and priority of targeting; even minor changes can significantly affect targeting efficiency.

The connection between translational readthrough and peroxisomal targeting has been supported by strong experimental and bioinformatic evidence. Genomic and proteomic analyses have revealed “cryptic” or “downstream-of-stop-codon” PTS1 motifs in several metabolic enzymes of fungi and animals. These PTS1 motifs are incorporated into the mature protein only when readthrough occurs, generating C-terminally extended isoforms capable of peroxisomal import (cryptic PTS). Subsequent studies have confirmed this through fluorescence tagging, site-directed stop codon modification, or readthrough mimicry experiments: readthrough can produce functional PTS1 motifs recognizable by Pex5, thereby directing a portion of the enzyme into the peroxisome. Conversely, deletion of the downstream PTS1 or inhibition of readthrough results in predominantly cytosolic localization. Representative studies include the activation of cryptic PTSs in fungi (Freitag et al., 2012), the generation of peroxisomal isoforms of LDHB (lactate dehydrogenase B) through readthrough in mammalian systems (Schueren et al., 2014; Stiebler et al., 2014), and several reviews and methodological studies on the mechanisms and motifs of readthrough (Schueren & Thoms, 2016). Together, these findings demonstrate that stop-codon readthrough is a broadly utilized biological strategy that produces low-abundance but functionally specific peroxisomal isoforms, enabling subcellular “division of labor” and flexible regulation of metabolic enzymes.

In subcellular localization experiments, mCherry-SKL is widely used as a peroxisomal marker (positive control). Its design is based on fusing the red fluorescent protein mCherry with the canonical PTS1 tripeptide SKL at the C-terminus, allowing the fusion protein to be recognized by Pex5 and imported into the peroxisomal matrix. Advantages of mCherry-SKL include ease of construction and expression, strong and punctate fluorescence, compatibility with green fluorescent fusions for dual-channel colocalization, and reliability as an internal control for verifying peroxisomal integrity and import mechanisms. In colocalization assays, strong overlap between GFP-fusions and mCherry-SKL in merged images indicates successful peroxisomal import of the GFP-fusion protein. In contrast, a diffuse cytoplasmic GFP signal or partial colocalization suggests incomplete targeting or unmet localization conditions. To avoid misinterpretation due to PTS competition or expression-level discrepancies, experiments should include parallel controls of mCherry-SKL and varying levels of GFP-fusions, as well as appropriate negative controls.

Extensive literature thus provides both theoretical and empirical support for the hypothesis that “ribosomal stop-codon readthrough can generate C-terminal extensions carrying cryptic PTS1 motifs, mediating partial peroxisomal import of the protein.” Meanwhile, mCherry-SKL, as a classic PTS1 positive control, reliably indicates peroxisomal localization in fluorescence colocalization experiments to validate readthrough-dependent targeting. In this proof-of-concept study, we use mCherry-SKL as a reference and construct a series of GFP-fusion proteins—retaining, deleting, or mimicking readthrough—to directly test the readthrough-dependent peroxisomal targeting of DAS/DAK in Aureobasidium melanogenum P16. This design represents a rigorous and falsifiable experimental approach grounded in established theory and methodology.

To determine whether Aureobasidium melanogenum P16 proteins DAS and DAK possess potential readthrough-dependent peroxisomal targeting sequences (PTS1), we employed the SL-AttnESM structure-aware localization prediction model and its intelligent agent LocAgent for computational analysis.

SL-AttnESM is built upon the 150-million-parameter ESM-2 language model, generating residue-level embeddings reweighted by ESMFold-derived structural priors—secondary structure, relative solvent accessibility (RSA), and confidence score (pLDDT)—through a structure-biased attention mechanism. This architecture enables multi-label prediction across ten eukaryotic subcellular compartments. The model is specifically optimized for short, surface-exposed, and conformation-dependent targeting motifs and achieves a +0.34 improvement in Matthews correlation coefficient (MCC) for the peroxisome class compared with DeepLoc 2.0, effectively capturing cryptic C-terminal PTS1 patterns.

LocAgent serves as the interactive interface for SL-AttnESM, invoking ESMFold to generate 3D backbone structures, extracting per-residue attention distributions, and cross-validating signal types via SignalP-6, NetGPI-3, and TMHMM-2. This integrated workflow yields localization probabilities, salient sequence regions, and experimental cloning recommendations without requiring multiple sequence alignments (MSAs).

The amino acid sequences of DAS and DAK—including approximately 100 bp downstream of the first stop codon—were submitted to LocAgent, and the following steps were executed:

① Structure Prediction: ESMFold v1 was used to predict the backbone conformation and extract per-residue secondary structure, RSA, and pLDDT.
② Localization Prediction: SL-AttnESM computed the probabilities (p_compartment) for ten subcellular compartments and generated residue-wise attention maps.
③ Signal Cross-Validation: SignalP-6 identified N-terminal signal peptides; TMHMM-2 detected transmembrane segments; and NetGPI-3 predicted potential GPI anchors, allowing exclusion of alternative localization mechanisms.
④ Cryptic PTS Identification: The SL-AttnESM outputs were examined for C-terminal regions showing both high pLDDT and locally enriched attention weights. Residues matching the canonical or variant tripeptide motif (S/A/C)-(K/R/H)-(L/M) were interpreted as putative PTS1 candidates.

For DAS, SL-AttnESM classified the protein predominantly as cytosolic (p_cyto = 0.82), with only weak attention peaks within the last 15 residues of the C-terminus. No canonical (S/A/C)-(K/R/H)-(L/M) tripeptide was detected. ESMFold predicted this segment to form an α-helical structure with a low confidence score (pLDDT < 0.75), suggesting it is likely buried within a hydrophobic core and thus inaccessible for receptor recognition. Accordingly, the model inferred that DAS lacks a functional PTS1-like motif, yielding a peroxisomal import probability of only p_perox = 0.07, and was therefore predicted to be a non-targeted cytosolic enzyme.

For DAK, the native construct was predicted as mainly cytosolic (p_cyto = 0.79) with a low peroxisomal probability (p_perox = 0.12). The attention map revealed a pronounced intensity peak 5–7 residues downstream of the stop codon, corresponding to the C-terminal sequence “–SMCQL–.” Although this motif deviates from the canonical PTS1 consensus, LocAgent identified that a single amino acid substitution (Q→L) would convert it into “–SMCLL–,” predicted to raise the peroxisomal probability to p_perox = 0.68 (Δ + 0.56). Structural superposition of the wild-type and mutant models produced a Cα-RMSD of only 0.4 Å, indicating that the substitution does not perturb global folding. Notably, SL-AttnESM attention weights increased 4.2-fold in residues 448–452, precisely mapping to the emerging “–LXL” motif that forms an exposed PTS1-like signal.

Collectively, these computational results suggest that:
① DAS lacks an exposed C-terminal PTS1 and is unlikely to undergo peroxisomal import via translational readthrough.
② DAK contains a cryptic PTS1-like extension downstream of the first stop codon, whose translation depends on readthrough efficiency.
③ Simulated readthrough or an engineered Q→L substitution creating the “–SMCLL–” motif substantially enhances peroxisomal import probability, supporting a readthrough-dependent PTS1 mechanism.

To verify the hypothesis derived from bioinformatic predictions and wet-lab findings — that ribosomal stop-codon readthrough can expose a downstream cryptic PTS1, thereby mediating the partial peroxisomal localization of target proteins — this study designed and constructed three expression variants for each of the two target enzymes, DAS and DAK: Full-length readthrough candidate fragment (Das, Dak): containing the complete coding sequence including the downstream region predicted to harbor the cryptic PTS1. Cytosolic truncated version (Das cyt, Dak cyt): containing only the sequence upstream of the first stop codon. These serve as negative controls lacking the downstream cryptic PTS1. Readthrough-mimicking mutant (Das pex, Dak pex): constructed by replacing the first stop codon with a sense codon, thereby simulating constant translational readthrough and ensuring that the downstream cryptic PTS1 is always translated. These function as positive controls. All constructs were fused with a GFP reporter either at the N- or C-terminus to enable fluorescence-based subcellular localization analysis. In addition, each was co-expressed with the classical peroxisomal marker mCherry-SKL, allowing for dual-channel colocalization imaging to assess whether the GFP-fusion proteins were imported into peroxisomes.

Fragments were cloned into GFP fusion expression vectors using a restriction enzyme digestion–ligation strategy. For the readthrough-mimicking mutants, site-directed mutagenesis was performed to replace the first stop codon (typically TAG/UGA/UAA) with a sense codon (TAG → TCG), thereby simulating translation elongation beyond the normal termination site. All constructs were confirmed by DNA sequencing.

The mCherry-SKL expression plasmid was electroporated into Aureobasidium melanogenum P16 to obtain a stable background strain with red peroxisomal labeling. Subsequently, different GFP-fusion constructs were introduced into this background strain via electroporation to achieve co-expression of both fluorescent markers.

Live-cell fluorescence imaging was performed using a fluorescence microscope.

Both GFP-Das and GFP-Das cyt displayed diffuse cytosolic fluorescence with almost no colocalization with the punctate peroxisomal signal of mCherry-SKL. Similarly, the readthrough-mimicking construct GFP-Das pex did not show any discernible punctate fluorescence or clear colocalization with peroxisomes.

In summary, under the present design and expression conditions in the P16 background, no experimental evidence was obtained to support the hypothesis that DAS generates a functional PTS through stop-codon readthrough and is imported into peroxisomes. The localization of DAS was predominantly cytosolic under these experimental conditions.

GFP-Dak exhibited a mixed intracellular distribution: a portion of the signal appeared diffusely in the cytosol, while distinct punctate green fluorescence showed strong colocalization with mCherry-SKL, indicating that part of the protein population was targeted to peroxisomes. In contrast, the cytosolic truncated variant GFP-Dak cyt, which lacks the downstream cryptic PTS, displayed only diffuse cytosolic fluorescence with no detectable peroxisomal colocalization. The readthrough-mimicking mutant GFP-Dak pex showed strong and well-defined punctate green fluorescence, highly overlapping with the red puncta of mCherry-SKL (with a clear merge in the overlay channel), demonstrating that this mutant was efficiently imported into peroxisomes.

These results indicate that, in Aureobasidium melanogenum P16, the peroxisomal localization of DAK depends on a cryptic PTS located downstream of the first stop codon, and the exposure of this PTS is regulated by stop-codon readthrough. Mimicking readthrough to constitutively translate this downstream PTS enables efficient targeting of the protein to peroxisomes.

The experiment clearly demonstrates that the peroxisomal localization of DAK is “readthrough-dependent.” The consistent evidence chain—partial peroxisomal localization of GFP-Dak, complete cytosolic distribution of GFP-Dak cyt, and strong peroxisomal localization of GFP-Dak pex—supports this conclusion. The disappearance of peroxisomal targeting upon removal of the downstream sequence and its restoration/enhancement upon mimicking readthrough indicates that translation of the downstream cryptic PTS, triggered by stop-codon readthrough, is a prerequisite for DAK import into peroxisomes.

In this study, DAS did not exhibit the expected readthrough-dependent peroxisomal localization, consistent with the results predicted by the computational model. Possible explanations include low translational readthrough efficiency of DAS in the P16 strain, non-functional or non-recognized downstream cryptic sequences, or interference from sequence context and post-translational modifications affecting signal recognition.

1. [1] Freitag J., Ast J., Bölker M. Cryptic peroxisomal targeting via alternative splicing and stop codon read-through in fungi. Nature 2012;485:522–525.
2. [2] Schueren F., Lingner T., George R., et al. Peroxisomal lactate dehydrogenase is generated by translational readthrough in mammals. eLife 2014;3:e03640.
3. [3] Stiebler AC., Freitag J., Schleiff E., et al. Ribosomal readthrough at a short UGA stop codon context generates peroxisomal isoforms in animals and fungi. PLoS Genet. 2014;10:e1004685.
4. [4] Schueren F., Thoms S. Functional Translational Readthrough: A Systems Biology Perspective. PLoS Genet. 2016;12:e1006196.
5. [5] Mohan KVK, et al. Identification of a type 1 peroxisomal targeting signal. J Bacteriol / Mol Biol (classic PTS1 descriptions).