Engineering self-deliverable ribonucleoproteins for genome editing in the brain

CRISPR-Cas systems enable RNA-guided genome editing but broader therapeutic application requires efficient, safe in vivo delivery of the editing machinery. Delivery to the central nervous system is particularly challenging. Viral vectors (for example, AAV) can express Cas proteins in the brain but carry risks of immunogenicity, insertional mutagenesis, and manufacturing constraints. Direct delivery of Cas RNPs could mitigate these issues by providing transient editing activity with reduced off-target and immune responses, yet RNP uptake and intracellular trafficking limit efficiency, especially in neuronal contexts. Prior non-viral strategies used nanoparticles or relied on highly cationic nuclear localization sequences for partial self-delivery and on synthetic endosomolytic peptides provided in trans. The study’s objective is to engineer Cas9 proteins genetically fused to functional cell-penetrating peptides to create self-deliverable RNPs that efficiently enter neural cells and enable robust genome editing in vitro and in vivo, including at disease-relevant loci in the mouse brain.

Non-viral RNP delivery to the brain has been explored using diverse nanoparticles (for example, CRISPR-Gold, amphiphilic peptide nanocomplexes, PEGylated nanocapsules, glucose-conjugated silica nanoparticles), achieving variable editing and disease improvement in mouse models but requiring complex formulation optimization and manufacturing. Previous in vivo neuronal editing using Cas9 RNPs with multiple SV40 NLS motifs demonstrated some self-delivery. Recent studies showed that synthetic amphipathic or endosomolytic peptides added in trans can enhance RNP uptake in primary cells. Collectively, these reports motivate genetically encoding compact CPPs on Cas proteins to simplify manufacturing and potentially improve brain delivery without auxiliary carriers.

• Enzyme comparison: Constructed cell-permeable variants of SpyCas9 and engineered LbCas12a (iCas12a) by appending multiple SV40 NLS copies (e.g., 2x at N-terminus and 2–4x at C-terminus). Assessed in vitro editing in Ai9 tdTomato neural progenitor cells (NPCs) via a stop-cassette excision reporter and in vivo via stereotaxic striatal injections in Ai9 mice at 125–250 pmol RNP.
• CPP screening: Selected 34 CPPs across classes (cationic, anionic, hydrophobic, viral-derived, signaling-derived, antimicrobial, membrane-active). Generated fusions at the Cas9 C-terminus with an N-terminal dual NLS (2x SV40). Expression constructs incorporated CL7 and His6 tags with HRV-3C cleavage sites to enable rapid dual-affinity purification (Ni-NTA and Im7-6B). Purity (>90%) was typically achieved without ion-exchange or SEC. RNPs were assembled with sgRNA and directly applied to Ai9 NPCs; editing quantified by flow cytometry.
• Optimization: Focused on top CPPs (Bac7 and A22p). Systematically varied peptide position (N- vs C-terminus), NLS type (SV40 vs human c-Myc), exposure of C-terminal CPP (removal of residual S3 sequence), copy number (1–≥4), and backbone insertion (Bac7 at Cas9 residue 205 with optimized linkers). Evaluated delivery via direct RNP treatment and maintained catalytic competence via nucleofection controls.
• Mechanistic probes: Tested CPPs supplied in trans (10× peptide) mixed with reference 2x-Cas9-2x RNP. Applied endocytosis inhibitors (monensin 80 nM, cytochalasin D 1.2 µM) prior to RNP administration to assess pathway dependence. Examined effects of growth factors on uptake.
• Lead construct engineering: To mitigate aggregation, mutated two solvent-exposed cysteines in Cas9 to serines (Cas9(dS)). Final lead: 2x-Cas9(dS)-A22p3 with fully exposed C-terminal A22p copies. Endotoxin-free purification incorporated Triton X-114 washes, heparin affinity, and SEC.
• In vitro assays: NPC culture and direct RNP dosing (6.25–100 pmol; typical 12.5–25 pmol for lead); tdTomato activation by flow cytometry; endogenous loci (TH, mGluR5) indels by NGS.
• In vivo assays: Intracranial convection-enhanced delivery to mouse striatum (5 µL; 5–50 µM RNP; typically 25–250 pmol total) with sterile, endotoxin-controlled buffers. Quantified edited striatal volume by whole-brain imaging and ROI analysis; neuron-specific editing by NeuN co-labeling; endogenous gene editing quantified by NGS on dissected tissue and mRNA reduction by qPCR.
• RNP assembly conditions: sgRNA:Cas9 molar ratios of 1.2:1 (reporter) or 1.5:1 (endogenous loci). RNPs prepared at 10–50 µM, sterile filtered, and injected within 6 h of assembly.

• In vitro enzyme and NLS configuration: Increasing NLS copy number improved self-delivery. iCas12a (2x N-terminal and 4x C-terminal NLS) achieved high in vitro delivery and editing in NPCs, with ~66% tdTomato activation at 6.25 pmol RNP (~50 nM).
• In vivo enzyme comparison: Both SpyCas9 and iCas12a yielded tdTomato+ cells after striatal injection (125–250 pmol), but SpyCas9 produced 3–6× larger edited striatal volumes. Conversely, iCas12a edits were more localized with a higher fraction of NeuN+ edited neurons within the ROI.
• CPP screening: Nine CPPs improved delivery versus 2x-Cas9-2x; effective peptides were cationic. Top performers included Bac7, HBP, CA-Tat, and A22p, each yielding >3× enhancement in NPC editing over the 2x-Cas9-2x reference. Several cationic CPPs (e.g., Penetratin, EB1, LALF) impaired Cas9 function.
• Optimization: A22p required a fully exposed C-terminal fusion for optimal activity; N-terminal placement reduced efficacy. Tripling A22p copies (A22p3) at the Cas9 C-terminus increased NPC editing from ~40% (two copies) to ~70%, though ≥4 copies caused aggregation; Bac7 in three copies was toxic during expression. The Cas9(dS) variant with A22p3 and fully exposed C-terminus achieved ~72% tdTomato activation by direct delivery to NPCs.
• Mechanism: Adding Bac7 or A22p peptides in trans improved editing 1.5–2.5× over reference RNPs. Endocytosis inhibitors (monensin, cytochalasin D) reduced editing 4–5×, implicating endocytic uptake and trafficking in delivery.
• Lead in vitro performance: The engineered 2x-Cas9(dS)-A22p3 construct achieved robust NPC editing at low doses (e.g., ~47% editing at 12.5 pmol, ~100 nM), approximately 2× higher than 4x-Cas9-2x.
• Lead in vivo performance: A22p3-RNPs produced robust striatal editing across 250–25 pmol dosing, outperforming 4x-Cas9-2x at lower doses in both edited volume and fraction of edited neurons.
• Endogenous targets: In NPCs, direct delivery targeting TH and mGluR5 achieved up to ~72% indels (NGS). In mouse striatum (250 pmol RNP), editing averaged ~1.5–5% indels at DNA level with ~15–20% mRNA reduction by qPCR for both TH and mGluR5.

The work demonstrates that compact CPPs genetically fused to Cas9 can endow RNPs with efficient self-delivery to neural cells, reducing reliance on viral vectors or complex nanoparticles. SpyCas9 was superior to iCas12a in achieving broader tissue spread in the brain, though iCas12a edits were more focal. Systematic CPP engineering identified C-terminally fused A22p in three copies as a potent enhancer of neuronal delivery and editing, likely acting via receptor-mediated or pH-sensitive endocytic pathways and improved endosomal escape. The peptide’s derivation from semaphorin-3a and reported affinity to neuropilin receptors suggest a plausible neuron-surface recognition mechanism. The engineered A22p3-Cas9(dS) RNPs enabled robust tdTomato activation and edited endogenous CNS-relevant genes in vivo with measurable mRNA knockdown, supporting their translational potential. Compared to NLS-only constructs, CPP fusions yielded higher efficacy, particularly at lower doses, and maintained the advantages of transient, minimally immunogenic RNP delivery. These findings support CPP-guided, self-deliverable RNPs as a generalizable platform for brain genome editing and motivate exploration of cell-type-specific ligand peptides to target additional tissues.

This study establishes a genetically encoded CPP strategy for self-deliverable Cas9 RNPs and identifies a lead SpyCas9 construct bearing three C-terminal copies of A22p that enables efficient neuronal uptake and genome editing in vitro and in the mouse brain. The platform achieves high editing in NPCs at low doses and edits disease-relevant genes in vivo with associated transcript reduction, outperforming NLS-only constructs, especially at lower doses. The approach offers a simplified, potentially less immunogenic alternative to viral and nanoparticle delivery. Future work should optimize peptide-receptor targeting for specific neuronal subtypes, enhance endosomal escape and brain-wide distribution, improve in vivo editing levels at endogenous loci, and evaluate safety, durability, and efficacy in larger animal models and disease contexts.

• In vivo endogenous editing levels in striatum were modest (~1.5–5% DNA indels) despite robust reporter activation; further optimization is needed for therapeutic thresholds.
• Delivery required intracranial injection; systemic or minimally invasive delivery was not assessed.
• Increasing CPP copy number can compromise protein stability (aggregation with ≥4 A22p copies) or expression (Bac7 toxicity in E. coli), constraining design space.
• Mechanistic understanding is preliminary; endocytosis implicated, but receptor identity and trafficking steps remain to be fully defined.
• Diffusion and cell-type specificity differences (e.g., SpyCas9 vs iCas12a spread) suggest variable biodistribution that may limit uniform editing.
• Experiments were not randomized and investigators were not blinded during allocation per the reporting, which may introduce bias; sample sizes were not predetermined statistically.
• Studies were conducted in mice; generalizability to larger brains and humans remains to be demonstrated; long-term safety and off-target analyses are not detailed here.