The motivation for the work is that programmable RNA systems commonly rely on nonhuman protein scaffolds. Such components may produce immune responses and unintended intracellular effects that are difficult to predict, creating obstacles to translation. Cas proteins and other scaffolds have nevertheless enabled a wide range of programmable RNA functions. ALTER instead begins with AGO2, a human protein already associated with RNA interference, and engineers it into a modular recognition platform. The researchers describe the approach as a way to redirect an intact human protein toward a function it does not normally perform while adding only a human RNA effector domain.

AGO2 normally uses a guide RNA to recognize a complementary RNA sequence and can help mediate target cleavage as part of RNA interference. ALTER separates sequence recognition from that destructive activity. The researchers fuse AGO2 to a deaminase that performs C-to-U editing and modify the AGO2–guide interaction to reduce nuclease activity. A further guide-design feature creates a loop in the target RNA. Instead of pairing with one uninterrupted complementary segment, the guide reshapes the local transcript so that the nucleotide selected for editing is more accessible to the deaminase. The authors selected a loop-inducing guide that did not produce substantial downregulation of the target.

This arrangement allows the short guide RNA to perform several tasks at once: directing AGO2 to the transcript, limiting the normal slicing behavior and influencing the structure around the editing site. The work also shows that AGO2 engineering cannot simply be copied from RNA-interference studies. Mutations or guide features that favor one activity may not favor another, and the authors found that guide designs suitable for editing need not resemble those optimized for interference. They suggest that comparing the two contexts could help reveal previously unrecognized features of AGO2 activity and regulation.

Across the messenger RNA targets tested in HEK293T cells, ALTER achieved an average 1.9-fold increase in C-to-U editing relative to CURE and RESCUE. This result describes the average performance of the tested targets rather than a uniform advantage for every sequence. The researchers also observed that ALTER and the Cas13 systems sometimes exhibited different, and in some settings complementary, cellular activities. They characterize the determinants of editing performance as target dependent, multifactorial and difficult to predict. Guide design was likewise context specific: the experiments provide a set of practical starting points, but the authors say systematic testing across many guides and targets will be needed to develop more general design principles.

To assess unintended activity, the team generated transcriptome-wide RNA-sequencing data and compared editor-treated cells with transfection controls, using three biological replicates per condition. Reads were aligned to a human reference genome, and candidate C-to-U variants were filtered according to read depth, genotype quality and agreement with control reads. This analysis found similar or lower levels of off-target editing for ALTER than for the Cas13-based comparison systems. The investigators also analyzed differentially expressed genes, defining them with an adjusted P value below 0.01 and an absolute log2 fold change greater than one. They cautioned that some functions available to Cas-based effectors might remain difficult to achieve with human proteins alone and that ALTER could produce more differentially expressed genes in some settings.

The researchers separately evaluated the potential immune consequences of nonnative sequences in the engineered protein. Using the MARIA computational tool, they enumerated overlapping 15-amino-acid peptides containing engineered substitutions, inserted sequences or residues at domain junctions, then estimated their likelihood of being presented by MHC class II. Relative to the Cas-based editor, this model predicted a substantially lower rate of MHC II-presented exogenous peptides from modified AGO2. The authors also adapted a validated artificial-intelligence-directed deimmunization workflow to search for single- and double-amino-acid changes that could remove predicted peptide presentation. The workflow produced modified designs that advanced to experimental benchmarking, demonstrating that ALTER could be incorporated into an established computational process for reducing predicted immune recognition.

Those immune findings remain predictions rather than evidence of safety in patients. ALTER can be supplied as a ribonucleoprotein complex or as a gene, and the investigators also examined virus-like particles and adeno-associated virus delivery in cultured HEK293T cells carrying an EGFP Y66H reporter. These experiments showed that the engineered editor could be incorporated into experimental delivery platforms. They did not establish therapeutic performance in animals or humans. The authors noted that virus-like particles and AAV vectors retain immunogenic risks of their own; advances in deimmunizing delivery systems could shift the primary immune concern from the vector to the protein cargo. The current work also did not identify a specific effect of target-directed messenger RNA degradation on ALTER under the tested conditions.

Additional design characteristics discussed by the authors include minimal protein fusions, short RNA guides, editing without requiring nuclear entry and the absence of A-to-I off-target activity associated with engineered ADAR-based technologies. They also reported no detectable activity of the A3A variant used in ALTER on DNA. These properties make ALTER a distinct design within programmable RNA manipulation, but the study does not establish that they will translate into better clinical outcomes. Functional experiments were performed in a human embryonic kidney cell line, much of the editing assessment used reporter proteins or selected transcripts, and the immune analysis relied on computational peptide presentation. The system therefore remains an engineered-cell proof of concept rather than a validated medicine.

Even with those limitations, the study expands the available set of human-protein tools for programmable RNA manipulation. By turning AGO2 into a modular targeting domain, the researchers show that sequence recognition, control of target structure and enzymatic editing can be combined within a system built around a familiar human protein. ALTER produced higher average editing than the two tested Cas13 comparators, while transcriptome-wide sequencing detected similar or lower off-target editing and the immune model predicted fewer exogenous peptides from modified AGO2. The authors conclude that further AGO2-based effectors could benefit from the extensive body of knowledge about human RNA interference. The next challenges are empirical rather than conceptual: testing many targets, establishing reliable guide-design rules, measuring guide-dependent effects across contexts and determining how the platform behaves with delivery systems and in biological systems beyond engineered cell lines.

Subject of Research: Human AGO2-based programmable C-to-U RNA base editing in HEK293T cells

Article Title: Engineered human Argonaute 2 for targeted C-to-U RNA base editing

Article References: Lusi, R. F., Moncayo von Hase, S., Model, C., & Banik, S. M. (2026). Engineered human Argonaute 2 for targeted C-to-U RNA base editing. Nature Biotechnology. https://doi.org/10.1038/s41587-026-03359-y

Image Credits: AI Generated

DOI: 10.1038/s41587-026-03359-y

Keywords: RNA base editing, Argonaute 2, C-to-U conversion, gene editing, Cas13 comparators, HEK293T cells, off-target editing, protein engineering, viral delivery, immunogenicity, Engineered, human