WO2025235828A2 - Engineered prime editors for treating genetic deafness - Google Patents

Abstract

Provided herein are engineered split prime editors (PEs) and optimized pegRNAs (with optimized reverse transcriptase template (RTT) length, primer binding site (PBS) length, engineered pegRNA (epegRNA) composition, silent mutation type and location, RNA linkers, and RT initiation sequences) and their use in correcting a prevalent deafness-causing mutation, GJB2 c35delG.

Description

ENGINEERED PRIME EDITORS FOR TREATING GENETIC DEAFNESS

CLAIM OF PRIORITY

This application claims the benefit of U.S. Provisional Patent Application Serial No. 63/644,379, filed on May 8, 2024. The entire contents of the foregoing are hereby incorporated by reference.

FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT

This invention was made with Government support under Grant No.

HL 142494 awarded by the National Institutes of Health. The Government has certain rights in the invention.

TECHNICAL FIELD

Provided herein are engineered split prime editors (PEs) and optimized prime editor guide RNAs (pegRNAs) (with optimized reverse transcriptase template (RTT) length, primer binding site (PBS) length, engineered pegRNA (epegRNA) composition, silent mutation type and location, RNA linkers, and/or RT initiation sequences) and their use in correcting a prevalent deafness-causing mutation, GJB2 c35delG.

BACKGROUND

Genome editing technologies enable custom changes in living cells and offer hope for the correction of human diseases. Despite the promise of CRISPR enzymes and base editors for precise base pair-level correction^1,2, BEs can lead to unwanted bystander editing of nearby bases, are dependent on appropriately positioned protospacer-adjacent motifs, and may carry risks of persistent expression of deaminase domains³. To overcome these caveats, prime editing approaches were developed for RNA-templated precision genome edits^4,5. Prime editor (PE) constructs are comprised of a nickase SpCas9 protein (nCas9; typically with an H840A mutation) fused to a reverse transcriptase (RT) domain (typically an engineered version of M-MLV)^4,5. The PE construct can write new genetic information into a specified DNA locus by copying sequencing information encoded in the prime editor guide RNA (pegRNA). A secondary guide RNA (gRNA) that directs the PE to nick the non-edited strand (referred to as a nicking gRNA or ngRNA) can improve editing efficiencies by assisting with edit resolution through subversion of DNA repair to preferentially utilize the edited strand for installation (where PE3 ngRNAs are distal from the original pegRNA target site, and PE3b ngRNA target sites overlap and encode the installed edit). PEs are capable of nearly any small sequence change that can be encoded within the reverse transcriptase template (RTT) sequence located on the 3’ end of the pegRNA, including all types of substitutions, and short deletions or insertions.

SUMMARY

Prime editors enable the precise installation of genetic edits encoded on a RNA template molecule. Here we extensively optimized various components of a prime editing approach, including the engineering of split prime editors more capable of efficient intein-mediated construct reconstitution, screening dozens of combinations of pegRNAs (featuring various RTT and PBS lengths), exploring the use of epegRNAs, screening many silent mutations to potentiate prime editing by subverting DNA mismatch repair, modifications to the gRNA scaffold (including termination reducing substitutions, extension of the crRNA:tracrRNA duplex, and linkers between the gRNA scaffold and the pegRNA, etc.), and other approaches. Using our highly optimized PE and pegRNA constructs, we achieved >70% editing in a human cell model of a prevalent pathogenic deafness causing mutation, GJB2 c.35delG (>15,000 births per year). This approach was translated into a new mouse model of this deafness disorder via AAV vector delivery, achieving >70% editing in Gjb2 expressing cells in mouse cochlea or liver after 2 weeks of treatment. Our extensive optimization of prime editing for correction of GJB2 c.35delG should motivate the continued development of novel genetic therapies for this disease of unmet need, while also providing a blueprint to achieve high levels of in vivo correction for a broader scope of genetic mutations.

Thus, provided herein are engineered split prime editors (PEs) comprising a nickase SpCas9 protein (nSpCas9) fused to a reverse transcriptase (RT) domain, wherein the PE is split in the nSpCas9 domain at residues E1068; T1069; G1055; E945; 11057; A1053; T1051; E945; or N946, preferably having a sequence that is at least 95% identical to a sequence shown in Table 4. Also provided are engineered split PEs for use with inteins that are split at SpCas9 amino acid residues E1068; T1069; G1055; E945; 11057; A1053; T1051; E945; or N946, preferably having a sequence that is at least 95% identical to a sequence shown in Table 4.

Additionally provided are optimized prime editor guide RNAs (pegRNAs), preferably engineered prime editor guide RNAs (epegRNAs) having an architecture comprising: a reverse transcriptase template (RTT) length of 10-18 nucleotides, preferably 14 nucleotides, comprising an intended mutation (such mutations can be any change, including insertion, deletion, and/or alteration of one or more nucleotides); a primer binding site (PBS) length of 9-13 nucleotides, preferably 11 or 12 nucleotides; preferably the pegRNA compsises an evopreQ (which can include tevopreQl) or mpknot modification at the 3’ end, e.g., is an epegRNA; an optional sequence comprising one or more silent mutations, optionally within 0-6 codons of the intended mutation ; an optional RNA linker comprising 1-20 nucleotides, preferably 3-5 nts, between between the 3’ of the sgRNA and the RTT; an optional RT initiation sequence, e.g., comprising IS 1, IS3, or IS 11 between PBS and evopreQ modification, preferably IS3; and an sgRNA spacer sequence and scaffold, preferably wherein the epegRNA has a sequence that is at least 95% identical to a sequence shown in Table A or 2.

In some embodiments, the optimized pegRNA comprises a sgRNA spacer sequence of GCACGCUGC AGACGAUCCUGG (SEQ ID NO:378), GGCACGCUGCAGACGAUCCUGG (SEQ ID NO: 379), GCUGCAGACGAUCCUGGGGG (SEQ ID NO: 380), or GCAGACGAUCCUGGGGGUG (SEQ ID NO:381), and/or optionally comprising a scaffold sequence of GUUUUAGAGCUAGAAAUAGCAAGUUAAAAUAAGGCUAGUCCGUUAUCA ACUUGAAAAAGUGGCACCGAGUCGGUGC (SEQ ID NO:42), GUUUCAGAGCUAGAAAUAGCAAGUUGAAAUAAGGCUAGUCCGUUAUCA ACUUGAAAAAGUGGCACCGAGUCGGUGC (SEQ ID NO:43), or GUUUCAGAGCUAUGCUGGAAACAGCAUAGCAAGUUUGAAUAAGGCUAG UCCGUUAUCAACUUGAAAAAGUGGCACCGAGUCGGUGC (SEQ ID NO:44).

Also provided herein are compositions comprising ribonucleoprotein (RNP) complexes comprising a split PE as described herein and an optimized pegRNA as described herien, and preferably a nicking guide RNA (ngRNA) that directs the PE to introduce a nick on the non-edited strand, optionally wherein the RNP complexes are delivered using nanoparticles, preferably lipid nanoparticles (LNPs).

Additionally provided herein are compositions comprising nucleic acids encoding a split PE described herein and an optimized pegRNA as described herein, and optionally a ngRNA that directs the PE to introduce a nick on the non-edited strand, e.g., to introduce a nick within 0-6, or 3-5, codons of the intended mutation on the non-edited strand.

In some embodiments, the compositions comprise one or more adenoassociated viruses (AAVs).

In some embodiments, the nucleic acids comprise a first AAV with sequences encoding a first half of the split PE driven by a promoter and a second AAV with with sequences encoding a second half of the split PE driven by a promoter, preferably the same promoter as the promoter driving expression of the first half, wherein the first half is either the N terminal or C terminal half of the PE, and the second half is the other half, and optionally wherein the first and/or second AAV further comprise sequences encoding the pegRNA and/or ngRNA, preferably driven by a U6 promoter or tRNA. As used herein, “half’ doesn’t refer to 50% of the construct; the “first half’ is the N terminal portion, and the “second half’ refers to the C terminal portion of the construct. The split need not be, and generally is not, in the middle of the construct.

In some embodiments, each of the pegRNA and the ngRNA are driven by a U6 promoter or tRNA, optionally selected from human U6 promoter (hU6), bovine U6 promoter (bU6), murine U6 promoter (mU6), and hCtRNA, optionally wherein (i) the ngRNA is driven by hCtRNA promoter and pegRNA is driven by bU6 promoter, or (ii) the ngRNA is driven by hU6 promoter and a pegRNA driven by hU6 promoter.

In some embodiments, the ngRNA is listed in Table B.

Additionally, provided herein are compositions comprising an optimized pegRNA as described herein, preferably an epegRNA, and a nicking guide RNA (ngRNA) that directs the PE to introduce a nick on the non-edited strand, or nucleic acids encoding the optimized pegRNA and ngRNA. In some embodiments, the optimized pegRNA is listed in Table A, and the ngRNA is listed in Table B. In some embodiments, the optimized pegRNA is LM3178 or LM3180, and the ngRNA is LM1728, LM3188, or LM3194. Also provided herein are methods of treating a subject who has deafnesscausing mutation in the GJB2 gene, preferably a c35delG mutation, the method comprising delivering to the inner ear of the subject a composition as described herein. Additionally provided are methods of correcting a deafness-causing mutation in the GJB2 gene in a cell, preferably a c35delG mutation, the method comprising contacting the cell with a composition as described herein.

Further, provided herein are methods of treating a subject who has deafnesscausing mutation in the GJB2 gene, preferably a c35delG mutation, or for correcting a deafness-causing mutation in the GJB2 gene in a cell, preferably a c35delG mutation, comprising delivering to the inner ear of the subject a prime editor and an optimized pegRNA as described herein, preferably an epegRNA, and a nicking guide RNA (ngRNA) that directs the PE to introduce a nick on the non-edited strand, or nucleic acids encoding the optimized pegRNA and ngRNA. In some embodiments, the optimized pegRNA is listed in Table A, and the ngRNA is listed in Table B. In some embodiments, the optimized pegRNA is LM3178 or LM3180, and the ngRNA is LM1728, LM3188, or LM3194. In some embodiments, the prime editor is PE2, PE4, PEmax, PE6, PE6a, PE6b, PE6c, PE6d, PE6e, PE6a/e, PE7, PE3max, or PE5max, or a variant thereof, optionally with truncation of the RNaseH domain of the RT. In some embodiments, the prime editor is split at SpCas9 amino acid residues E1068; T1069; G1055; E945; 11057; A1053; T1051; E945; N946; K1024; Q844; or K1153.

Also provided herein are methods of treating a subject who has deafnesscausing mutation in the GJB2 gene, or correcting a deafness-causing mutation in the GJB2 gene in a cell, preferably a c35delG mutation, the method comprising delivering to the inner ear of the subject an insertion frame editor (iFE) and a guide RNA listed in Table 2, preferably LM248; LM250; LM254; LM258; LM260; LM262; LM264; LM266; LM268; LM270; LM274; LM276; LM278; LM280; LM282; LM284; LM286; LM288; LM290; LM292; LM294; or LM296.

Additionally, provided herein are optimized pegRNAs having an architecture described herein, e.g., in Table A, and optionally having a sequence that is at least 95% identical to a sequence shown in Table 1 or 2.

Further, provided herein are compositions comprising ribonucleoprotein (RNP) complexes comprising the split PE and the optimized pegRNA as described herein, optionally wherein the RNP complexes are delivered using nanoparticles, e.g., lipid nanoparticles (LNPs).

Also provided herein are compositions comprising nucleic acids encoding a split PE and optimized pegRNA as described herein.

In some embodiments, the compositions are or comprise an AAV. In some embodiments, the AAV comprises a first AAV with sequences encoding one half of the split PE driven by a promoter, and optionally a sequence encoding the pegRNA driven by U6, and a second AAV with a second half of the split PE driven by a promoter, preferably the same promoter as the promoter driving expression of the first half, wherein the first half is either the N terminal or C terminal half of the PE, and the second half is the other half.

Additionally provided herein are methods of treating a subject who has deafness-causing mutation in the GJB2 gene, preferably a c35delG mutation, the method comprising delivering to the inner ear of the subject a composition as described herein.

Further, provided herein are methods of correcting a deafness-causing mutation in the GJB2 gene in a cell, preferably a c35delG mutation, the method comprising contacting the cell with a composition as described herein.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Methods and materials are described herein for use in the present invention; other, suitable methods and materials known in the art can also be used. The materials, methods, and examples are illustrative only and not intended to be limiting. All publications, patent applications, patents, sequences, database entries, and other references mentioned herein are incorporated by reference in their entirety. In case of conflict, the present specification, including definitions, will control.

Other features and advantages of the invention will be apparent from the following detailed description and figures, and from the claims.

DESCRIPTION OF DRAWINGS

Figure 1. Creation of a GJB2 c.35delG model cell line to investigate prime editing strategies for correction, a, Schematic of the sequence of the c.35delG mutation in the GJB2 gene, which creates a Int frameshift leading to a premature stop codon (ATCCTGGGGGGTGTGAAC, SEQ ID NO : 382 ;

ATCCTGGGGGTGTGAAC, SEQ ID NO : 383 ; ) . b, Schematic of the role of the connexin 26 protein produced by the GJB2 gene, which is a gap junction protein that mediates intercellular communication. Mutation of GJB2 can lead to loss of function of cell-to-cell communication. Dark gray connexons represent the wildtype Cx26; light gray connexons represent mutated Cx26. c, Schematic of the design of pegRNA (with reverse transcriptase template (RTT) and primer binding site (PBS) combinations shown) and nicking gRNA target sites for prime editing approaches to create and model the 35delG mutation in HEK 293T cells, highlighting some of the top combinations that led to the highest levels of editing and were thus used to create the GJB2 c.35delG cell line; ATGGATTGGGGCACGCTGCAGACGATCCTGGGGGTGTGAACAAACACT ( SEQ ID NO : 384 ) . d, Precise prime editing efficiencies from experiments in HEK 293T cells, where various pegRNAs and ngRNAs (PE3 or PE3b) were assessed to create the GJB2 c.35delG mutation. The PE3b-l, PE3b-2 and PE3-3 ngRNAs are 3 different target sites for nicking guides. e,f, Representative Sanger sequencing trace (panel e, ACGCTGCAGACGATCCTGGGGGTGTGAACAAACACTCCACC ( SEQ ID

NO : 385 ) .) and targeted sequencing result (analyzed via CRISPResso2; panel f) to confirm the Int deletion in the early coding sequence of the GJB2 gene, sequenced from the homozygous GJB2 c.35delG HEK 293T cell line; CAGACGATCCTGGGGGGTGTGAACAAACACT ( SEQ ID NO : 386 ) ;

CAGACGATCCTGGGGGTGTGAACAAACACT ( SEQ ID NO : 387 ) . g, Schematic of the initial design of pegRNA and ngRNA target sites to correct the single nucleotide 35delG mutation, illustrating the most effective pegRNA design with RTT and PBS combinations shown; the PE3b and PE3 ngRNA spacers are shown, h, Exemplary targeted sequencing result (analyzed using CRISPRresso2) for precise correction of the Int deletion caused by 35delG. i, Precise prime editing efficiencies in homozygous GJB2 c.35delG HEK 293T cells when using pegRNAs targeting target site 1 and various pegRNAs with different RTT and PBS combinations. The ngRNA that was utilized is PE3b-3 ngRNA. j, Analysis of sequencing reads harboring unwanted insertion or deletion mutations (indels) from the experiments performed in panel i. k, Precise prime editing efficiencies in homozygous GJB2 c.35delG HEK 293T cells when using pegRNAs targeting target site 1 and various pegRNAs with different RTT and PBS combinations. The ngRNAthat was utilized is PE3-1 ngRNA. 1, Analysis of sequencing reads harboring unwanted indels from the experiments performed in panel k. Data from panels d,i-k are plotted as mean ± s.d. with individual datapoints shown for n = 3 independent biological replicates.

Figure 2. Systematic optimization of prime editing efficiencies to correct GJB2 c.35delG. a, Schematic of four spacers nearby the c.35delG mutation and the location of the intended Int G insertion, b, Precise prime editing efficiencies when using pegRNAs targeting target sites 2, 3, and 4 (from panel a) and with different RTT and PBS combinations, c, Analysis of sequencing reads harboring unwanted indels from the experiments performed in panel b. d, Precise prime editing efficiencies when using VRQR and SpG PAM PE variants and pegRNAs with different RTT and PBS combinations, e, Heatmap of systematic screen of pegRNAs targeting target site 1, bearing different RTT and PBS lengths (nt). The ngRNAthat was utilized is PE3b- 3. f, Analysis of sequencing reads harboring unwanted indels from the experiments performed in panel e. The ngRNA that was utilized in panels e and f is PE3b-3. g, Comparison of prime editing from unmodified pegRNA and evopreQ epegRNA with high, medium, and low efficiency from heatmap screen result in panel e. h, Comparison of prime editing from unmodified pegRNA and evopreQ or mpknot epegRNAs with low and medium efficiency. The ngRNAthat was utilized in panels g and h is PE3b-3. i, Schematic of the incorporation of additional ‘silent’ mutations in the RTT of the pegRNA, which can promote prime editing efficiency by evading DNA mismatch repair. Silent mutations are marked in blue (CC), intended G insertion is marked in yellow, j, Prime editing efficiency from adding CC silent mutation to one of the top candidate pegRNAs with 14nt RTT and lint PBS. The ngRNAthat was utilized is PE3b-4. k, Prime editing efficiency from adding CC silent mutation to pegRNAs with 17nt RTT and 13/11 nt PBS. The ngRNAthat was utilized was PE3b-3 or 4 based on whether there was silent mutation in the RTT of the pegRNA. 1, Silent mutation screen on one of the top performing pegRNAs with 14nt RTT and 12nt PBS. ngRNAs are different based on the silent mutation type. The silent mutation templates are named based on the 3 substitutions located at the wobble positions of LIO, Gil, and G12 (e.g., pegRNAs encoding CTxGGxGGx substitutions are named ‘xxx’). m, Silent mutation screen on pegRNAs with 17nt RTT and 13nt PBS (with silent mutation RTTs named as for panel 1). The spacer sequence of the ngRNAs differs based on the silent mutation. All c.35delG mutation correction transfections were performed in homozygous GJB2 c.35delG HEK 293T cells. Data in this figure are plotted as mean ± s.d. with individual datapoints shown for n = 3 independent biological replicates. Transfections performed in HEK 293T cells Editing efficiencies were assessed by targeted amplicon sequencing and analyzed using CRISPResso2.

Figure 3. Optimization of a split PE construct for dual-AAV delivery, a, Schematic of SpCas9 protein domain structure (with amino acids (AAs) numbered on top), highlighting some potential exemplary split sites located on the surface or loop regions of SpCas9 (numbered on the bottom), b, Heatmap of editing efficiency of SpCas9 nuclease at 3 endogenous targets sites in HEK 293T cells when using variant enzymes encoding intein scars at the indicated split sites, c, Schematic of split prime editor architecture mediated by intein reconstitution. For the C-terminal split PE vector, we tested a PEmax system with the full-length reverse transcriptase (RT), a PEmax system with the RNaseH domain of the RT deleted (delRNaseH RT), and PEmax-delRNaseH RT system with a CFN scar sequence between the C-intein and C- terminus of SpCas9-delRNaseH. The CFN scar appended to the C-terminal extein is the preferred amino acid sequence for enhancing intein reconstitution, d, Comparison of prime editing efficiency from full length PEmax and delRNaseH PEmax with SpCas9 split sites of T1051 and G1055 using a pegRNAto correct GJB2 c.35delG in our model HEK 293T cells with 14nt RTT and lint PBS. e, Comparison of prime editing efficiency on candidate split sites without CFN scar sequence using delRNaseH PEmax. f, Comparison of prime editing efficiency on the best candidate sites and previously reported split sites. delRNaseH PEmax with CFN scar was used in this comparison, g, Prime editing efficiency from titration of prime editors split at K1024 and E1068. Split PEs were tested at four different DNA dosages, lx means full dosage, 1 :3 means 1/3 of the full dosage, 1 :9 means 1/9 of the full dosage, 1 :27 means 1/27 of the full dosage. All c.35delG mutation correction transfections were performed in homozygous GJB2 c.35delG HEK 293T cells. Data are plotted as mean ± s.d. with individual datapoints shown for n = 3 independent biological replicates. Editing efficiencies were assessed by targeted amplicon sequencing and analyzed using CRISPResso2.

Figure 4. Genome-wide off-targets profiling via GUIDE-seq2. Putative off- target sites for top performing pegRNAs and nicking gRNAs were nominated via GUIDE-seq2 in HEK 293T cells, a, On and off-target GUIDE-seq2 read counts identified for SpCas9 nuclease paired with a conventional gRNA (top) and with three top candidate pegRNAs that recognize the same protospacer, b, On and off-target sites identified for SpCas9 nuclease paired with the PE3b nicking guide (used for prime editing experiments with pegRNAs without silent mutation), c, On and off-target sites identified for Cas9 paired with three top candidate epegRNAs that recognize the same protospacer with primary guide, d, On- and off-target sites identified for SpCas9 nuclease paired with the PE3b ngRNA with silent mutations (used for prime editing experiments with pegRNAs with the +5 and +8 G-to-C silent mutations). Mismatched positions in the target sites of off-targets are labeled with mismatched nucleotide, and GUIDE-seq read counts shown to the right of the on- and off-target sequences represent a measure of cleavage efficiency at a given site.

Figure 5. In vivo correction of GJB2 c.35delG in a humanized Gjb2 mouse model, a-c, Precise 35delG correction efficiency in DNA extracted from bulk left cochlea (panel a), liver (panel b), and brain (panel c) of humanized 35delG mice treated with AAV.PHPB-PEmax. Tissue samples and genomic DNA were harvested 14 days after injection. The humanized c.35delG floxed mice were either crossed or not with a SoxlO-Cre mouse to indue tissue-specific deletion of the WT Gjb2 allele, d-f, Precise 35delG correction efficiency at the mRNA level, assessed via cDNA sequencing. RNA was extracted from bulk left cochlea (panel d), liver (panel e), and brain (panel f) of humanized 35delG mice treated with AAV.PHPB-PEmax. g, Left, Precise 35delG correction efficiency in DNA extracted from bulk left cochlea, liver, and brain of humanized 35delG mice treated with AAV.PHPB-PEmax 30 days after injection. Right, analysis of indels from right panel, h, Left, Precise 35delG correction efficiency in RNA extracted from bulk left cochlea, liver, and brain of humanized 35delG mice treated with AAV.PHPB-PEmax 30 days after injection. Right, Analysis of indels from left panel. Left cochlea, liver, brain were harvested either 14 days or 30 days after injection and analyzed by target amplicon sequencing. Each bar represents individual mice.

Figure 6. Further engineering of pegRNAs for correction of GJB2 c.35delG. a, Comparison of GJB2 c.35delG correction efficiencies when using different prime editor constructs including PE6a, PE6b, PE6c, PE6d, PEmax, and PEmax-delRNaseH. The pegRNA is designed to target target site 1, with 14nt RTT and 11 or 12nt PBS. The construct silent #1 encodes two silent mismatches in the RTT (+5 G-to-C and +8 T-to-C), silent #2 encodes one silent mismatch in the RTT (+5 G- to-C). b, 35delG correction efficiency from different promoter combinations for pegRNA and ngRNA. hU6: human U6 promoter, bU6: bovine U6 promoter, CtRNA: human Cysteine tRNA. pegRNAs used are targeting target site 1, with 14nt RTT and 11 or 12nt PBS. The construct silent #1 encodes two silent mismatches in the RTT (+5 G-to-C and +8 T-to-C), silent #2 encodes one silent mismatch in the RTT (+5 G-to-C). c, Comparison from unmodified pegRNA, epegRNA, and epegRNA with silent mutations including silent #1 that encodes two silent mismatches in the RTT (+5 G- to-C and +8 T-to-C), silent #2 encodes one silent mismatch in the RTT (+5 G-to-C ) encoded by the top candidate pegRNAs. d, Comparison of prime editing efficiencies when using unmodified and modified SpCas9 gRNA scaffold sequences in the pegRNA and nicking guide. Titration from full dosage (lx), to 1/3 of full dosage (1 :3), 1/9 of full dosage (1 :9), and 1/27 of full dosage (1 :27). WT stands for unmodified conventional gRNA scaffold, GTTTC stands for mutating the fifth T of the scaffold to C (with a compensatory A>G change in the tracrRNA), extension stands for extended stem-loop in the scaffold (along the crRNA/tracrRNA duplex as described by). Data are plotted as mean ± s.d. with individual datapoints shown for n = 3 or 4 independent biological replicates.

Figure 7. Creation of a GJB2 c.35delG model cell line, a, Schematic of the design of pegRNA targeting target site 4 and nicking gRNA target sites for prime editing approaches to create and model the 35delG mutation in HEK 293T cells, b, Analysis of indels from Figure Id. The PE3b-l, PE3b-2, and PE3-3 ngRNAs are 3 different target sites for nicking guides, c, Precise prime editing efficiencies to create the 35delG mutation using PE2-SpG, pegRNAs with different RTT and PBS lengths targeting target site 4, and various ngRNAs. d, Precise prime editing efficiencies to create the 35delG mutation using PE2-VQRQ, pegRNAs with different RTT and PBS lengths targeting target site 4, and various ngRNAs. e, Analysis of indels from panel c. f, Analysis of indels from panel f. g, Workflow used to generate 35delG HEK293T cell line. Data are plotted as mean ± s.d. with individual datapoints shown for n = 3 independent biological replicates.

Figure 8. Testing NGT PAM variant prime editors to correct 35delG mutation, a, Schematic of the spacers for NGA and NGT PAM variant PE3 prime editors using the PE3 system, b, Precise prime editing efficiencies when using VRQR and SpG PAM variant PE and pegRNAs with different RTT and PBS combinations. Editing efficiencies were assessed by targeted amplicon sequencing and analyzed using CRISPResso2. Data are plotted as mean ± s.d. with individual datapoints shown for n = 3 independent biological replicates.

Figure 9. pegRNA screen on target sitel. Bar charts for all the pegRNAs screened for target site 1 for GJB2 c.35delG correction in our HEK 293T cell line, showing intended edit (panel a) and indels (panel b) for each pegRNA from heatmap screen in Figure 2e and 2f. Data are plotted as mean ± s.d. with individual datapoints shown for n = 3 or 4 independent biological replicates.

Figure 10. Testing split prime editors at more genomic sites, a, Precise prime editing efficiency from the split PEmax system at non-GJB2 sites to create exemplary R555Q and R124L mutations in TGFBI, and a third mutation in the HEK3 site of HEK 293T cells. The split points in SpCas9 for intein insertion are indicated on the x-axis; whether a CFN ‘scar’ peptide was added to the N-terminal end of the C- terminal extein is indicated, b, c.35delG correction efficiency by prime editing using previously reported split PEmax constructs (v3em and vlem from Davis et al., Nature Biotechnology, 42:253-264 (2024)); and comparison of unmodified pegRNA, epegRNA, and epegRNA with silent #1 mutations that encode two silent mismatches in the RTT (+5 G-to-C and +8 T-to-C). c, Analysis of indels from panel b. The editing efficiency was analyzed by CRISPResso2 in HDR mode. Data are plotted as mean ± s.d. with individual datapoints shown for n = 3 independent biological replicates.

Figure 11. Assessment of various polIII promoters for gRNA and pegRNA expression, a, Schematic of the various promoters tested, b, Editing efficiencies in experiments using HEK 293 T cells, when each promoter construct was used to express gRNAs targeted to various genomic sites with SpCas9 nuclease. Each promoter was tested in a full concentration and in a titration at 1/10 of the full concentration, c, Editing efficiencies for GJB2 c.35delG correction when expressing pegRNAs (y-axis) and ngRNAs (x-axis) from different promoters tested in combination with PEmax. Design 1 encodes two silent mismatches in the RTT (+5 G- to-C and +8 T-to-C), Design 2 encodes one silent mismatch in the RTT (+5 G-to-C). d, Alternate plot of the precise editing displayed in panel c. Data from panels b and c are plotted as the mean of n = 3 independent biological replicates. Data from panel d is plotted as mean ± s.d. with individual datapoints shown for n = 3 independent biological replicates.

Figure 12. GJB2 c.35delG correction under various experimental conditions, a, GJB2 c.35delG correction when using PEmax at various split points for intein insertion in SpCas9. Experiments were performed with full concentration plasmid dose and several titration doses of 1 :2 (1/2 the full concentration), 1 :4 (1/4 the full concentration), and 1 :10 (1/10 the full concentration). Each point tested was delivered with either a ngRNA driven by hCtRNA and a pegRNA driven by bU6 or a ngRNA driven by a hU6 and a pegRNA driven by a hU6. All split PEmax systems have the RNaseH domain removed from the RT and have an added CFN scar peptide following the Npu C-terminal intein domain (and used to the N-terminal end of the C- terminal SpCas9 extein fragment), b, Comparison of prime editing efficiency from PEmax split at SpCas9 residues KI 024 and El 068 in different DNA dosages. Promoter combinations for pegRNA and ngRNA are hU6/hU6 and hCtRNA/hCtRNA array. Data are plotted as mean ± s.d. with individual datapoints shown for n = 3 independent biological replicates.

Figure 13. Nuclease editing strategies to correct 35delG mutation in

GJB2. a, Representative allele frequency table of gRNAs 2 and 17 using SpG and SpG-LZ3 Cas9 nuclease, respectively, which create a major Ibp insertion profile, b, Amino acid sequence around G35 and a schematic of the editing outcome from guides 2 and 17, restoring the reading frame of the GJB2 protein, c-d, Pie chart of editing profile showing the percentages of unedited, precise edited, and indel reads of guide 17 (c) and guide 2 (d). e. Representative allele frequency table of guide 19 and guide 20 using SpG and SpG-LZ3 Cas9 nuclease, introducing 2 bp deletion, f. The amino acid sequence around G35 and the schematic of the editing outcome from guide 19 and 20, introducing del34G mutation while rescuing the reading frame of GJB2 protein, g-h, Pie chart of editing profile showing the percentages of unedited, in frame, precisely edited, and indel reads of guide 20 (g) and guide 19 (h). i, Representative allele frequency table of guide 10 using SpG and SpG-LZ3 Cas9 nuclease, creating a major Ibp insertion profile, j, Amino acid sequence around G35 and a schematic of the editing outcome from guide 10, introducing G35V and V36L mutations while rescuing the reading frame of the GJB2 protein, k, Pie chart of editing profile showing the percentages of unedited, precisely edited, and indel reads of guide 10. 1, Pie charts of editing profiles from guides 8, 9, 11, 13, 18, 21. Among the editing types, most of the reads are random indels. Sequences for spacers are listed in Table 2.

Figure 14. Frame editing strategies to correct 35delG mutation in GJB2.

SpCas9, LZ3-SpCas9 (addgene ID: 140561), SpG-LZ3, SpRY-LZ3 Cas9 nuclease were compared with SpCas9-iFE (addgene ID: 190138), LZ3-iFE (Addgene ID: 190143) frame editor using different guide RNAs, a, guide 1; b, guide 4; c, guide 5. Pie charts show the percentage of precisely edited, indels, and unedited alleles from above described editors. Sequences for spacers are listed in the table 2.

Figure 15. More guides using frame editing strategies to correct 35delG mutation in GJB2. SpCas9, LZ3-SpCas9 (addgene ID: 140561), SpG-LZ3, SpRY- LZ3 Cas9 nuclease were compared with SpCas9-iFE (addgene ID: 190138), LZ3-iFE (Addgene ID: 190143) frame editor using different guide RNAs, a, guide 6; b, guide 7. Pie charts show the percentage of precisely edited, indels, and unedited alleles from above described editors. Sequences for spacers are listed in the table 2.

Figure 16. Testing AAV construct split PEs in 35delG HEK293T cell line. a, Comparison of split PEmax AAV vectors to correct 35delG mutation. hU6: human U6, mU6: murine U6, bU6: bovine U6, CtRNA: human cysteine tRNA. WT: unmodified Wild Type guide scaffold, M: modified guide RNA scaffold with both the fifth T mutated to C and extended stem-loop duplex. The pegRNA/ngRNA are in opposite direction of the C-terminus of PEmax. b, Testing of prime editing efficiency from AAV vectors with pegRNA/ngRNA in the tandem or opposite direction of the C- terminus of PEmax. For the hU6/mU6 combination, the pegRNA has 14nt RTT/1 Int PBS with two silent mismatches in the RTT (+5 G-to-C and +8 T-to-C). For all other promoter combinations, the pegRNA has 14nt RTT/12nt with one silent mismatch in the RTT (+5 G-to-C). Data are plotted as mean ± s.d. with individual datapoints shown for n = 3 independent biological replicates.

Figure 17. PE editing efficiency is enhanced by the addition of an Initiation Sequence (IS) to the 3' terminus of pegRNAs. a, Design of IS added to the 3’ end of pegRNAs. The IS sequences are listed in table 2. b-d, Efficiency of PE3- mediated installation of substitution mutations using pegRNAs with ISs in HEK293T cells on the R124L mutation (b), R555Q mutation (c), and EMX mutation (d). e-h, Comparison of 35delG correction efficiency from various pegRNAs with or without IS with different RTT and PBS combinations. WT means pegRNA without modification, IS 1-12 means adding IS 1-12 to the 3’ end of pegRNA, epegRNA means adding evopreQ hairpin to 3’ end of pegRNA, epeg-ISl/3 means adding IS1 or IS3 between PBS and evopreQ hairpin, ncl/2 means adding two different random sequences between PBS and evopreQ hairpin. Data are plotted as mean ± s.d. with individual datapoints shown for n = 3 independent biological replicates.

Figure 18. Other efforts to engineer the pegRNA. a, Schematic adding an RNA linker between the sgRNA scaffold and the RTT region to generate linker pegRNAs that are more flexible, b, Linker size and sequence used in the initial test, c, Prime editing assessing RNA linkers with various lengths on the R124H and R555Q mutations in the TGFBI gene. None means no linker added to pegRNA. d, Prime editing efficiency on R124H mutation from pegRNAs with linker lengths from 2nt to lOnt. e-g, Comparison of 35delG correction efficiency from top candidate pegRNAs with 6 nt RNA linker, h, Schematic of adding both RNA linker and the IS to pegRNAs. i, Schematic of different configurations of modified pegRNAs. The silent mutation in the RTT pf pegRNA is a +5 G-to-C substitution; the sequence for IS3 is AGAAAAAGGGGGGAA; linker size is 6 nt linker sequence is TCTCTC. j, Percent modification using the combined RNA linker and IS strategy on 35delG correction with pegRNA with 13 nt RTT and 10 nt PBS. k-1, Precise editing showing the impact of the order of IS and epegRNA hairpin within the PEmax architecture on the 35delG correction. Data are plotted as mean ± s.d. with individual datapoints shown for n = 3 independent biological replicates.

DETAILED DESCRIPTION

Despite their potential to create nearly any programmable genetic change, prime editors require optimization for use to correct genetic mutations in vivo. Recent studies demonstrated the potential of PEs for in vivo genome editing^6,7, potentiated by improvements to pegRNA stability⁸, modified PE constructs⁹, and the installation of silent mutations into a genomic locus to evade DNA repair^9,10. A complication for delivery of the PE construct in vivo is that the PE coding sequence is too large for the packaging limits of a single AAV. To circumvent this problem it is possible to engineer a novel dual vector AAV system, as has been done for CRISPR-Cas nucleases and base editors^{11 l 4}. A dual vector system would split the coding sequence for the PE and include intein peptide sequences inserted at the end of the N-terminal half and the beginning of the C-terminal half. The coding sequences for each of the two halves can be packaged in separate AAVs. When cells are co-infected, the two proteins auto-catalytically fuse via the intein sequence to reconstitute the full PE by joining of the N- and C-terminal extein sequences. The AAV vectors will also transcribe a pegRNA and ngRNA under a U6 promoter.

Here we described optimization and testing of new PE constructs to examine their ability to correct genetic mutations in cells and in mice. We extensively optimized aspects of the pegRNA (including RTT lengths, PBS lengths, silent mismatched bases, etc.), ngRNA, and the PE construct itself by testing various split points in SpCas9 for intein insertion towards encoding the entire PE/pegRNA/ngRNA constructs in a dual -AAV approach. We thoroughly optimized various parameters of the editor components to insert a missing G base towards correction of a prevalent mutation causative of deafness (the c.35delG mutation in the GJB2 gene).

DFNB1 Hereditary Deafness. The monogenic disorder DFNB1 is by far the most common hereditary deafness, accounting for a quarter to half of all recessive nonsyndromic deafness. In the United States alone, about 3500 children are born each year with mutations in both alleles of the causative gene, Gap Junction beta 2 (GJB2), which encodes the gap junction protein GJB2 also known as connexin26^{15 l 7}. Three mutations including c. 35delG, p.M34T, and p.V37 account for three quarters of all variants in GJ 2G The single-base deletion c.35delG creates a frameshift and stop codon; about 700 infants in the United States, and more than 10,000 worldwide, are bom per year with 35delG-associated deafness¹⁸. Many of the affected children are born with profound hearing loss, but two-thirds have some residual hearing at birth and the majority of those lose hearing over the next few years. This suggests that a window exists for therapeutic intervention¹⁹. For DFNB1 — and the 35delG allele in particular — there is thus an opportunity to prevent hearing loss with gene therapy to correct the underlying mutation.

In the cochlea, GJB2 is expressed in both an epithelial system of supporting cells and a cytoplasmic system of fibrocytes of the lateral wall^{20 22}. In the cochlea, the epithelial system is largely post-mitotic; however in fibroblasts of the cytoplasmic system slow but nonzero cell division has been observed with BrdU labeling²³.

Gene Therapy in Mouse Models for DFNB1. Gene addition to neonatal Gjb2 conditional knockout mice with an AAV vector has thus far failed to rescue function^{24 26}, perhaps because promiscuous expression in all cochlear cell types is toxic. In addition, the slow turnover of fibrocytes means that a therapeutic AAV vector driving GJB2 expression will eventually be diluted, requiring repeated treatment. A more effective treatment would therefore be to correct the underlying genetic cause of DFNB1; the corrected gene would be replicated with cell division and would restore wild-type GJB2 expression in the right cells, at the right time, and at the right levels.

Here we undertook several approaches to optimize prime editing efficiencies in human cells towards use in mice. Development of engineered split PEs led to improved precise editing efficiencies. Similarly, extensive optimization of the pegRNA parameters for a therapeutic target (RTT length, PBS length, epegRNA composition, silent mutation type and location, RNA linkers, RT initiation sequences, etc.) resulted in high levels of +1 nt insertions with few unwanted indel or bystander edits. The innovations developed herein should be extensible to other prime editing approaches.

Despite decades of intensive study, hereditary deafness remains a debilitating class of diseases with few treatment options. DFNB1, caused by mutations in the gap junction gene GJB2, is the most common human hereditary deafness. It accounts for 25-50% of all cases of nonsyndromic hereditary hearing loss, or about 50,000 cases of congenital hearing loss in the world each year. The GJB2 coding sequence is short — easily fitting into a single AAV viral vector — so gene addition should in principle be an attractive therapeutic approach. Yet Gjb2 gene addition in mouse models of DFNB 1 has thus far failed to fully rescue function. In addition, sustained rescue by gene addition relies on persistence of the AAV episomes in the nucleus to continually express the corrective transgene. However, the cells that express GJB2 in the cochlea — epithelial cells near the organ of Corti and fibrocytes in the lateral wall — divide at a slow rate, causing dilution of AAV episomes and thus gene addition may eventually fail. To overcome these challenges, we extensively optimized an in vitro and in vivo prime editing approach to correct one of the most common mutations in GJB2. Gene editing has advantages over gene addition: the edited chromosome is duplicated with cell division to maintain correction even if target cells divide, the corrected gene is expressed under the normal promoter in the right cells and at the right level, and others. The present demonstration of efficacious editing in an animal model should form the basis for an eventual genetic approach to treat DFNB1 and is also a proof-of-concept of how highly optimized prime editors are translatable as potential treatments for other hereditary disorders.

Optimized Split Prime Editors (PEs)

Provided herein are engineered split PEs that are useful for delivery in a dual AAV approach, and that preferably have improved precise editing efficiencies. These PEs can be split utilizing an intein to post-translationally join the two fragments of the protein, e.g., at SpCas9 amino acid residues E1068; T1069; G1055; E945; 11057; A1053; T1051; E945; or N946. Previously described split PEs were split at K1024; Q844; or KI 153.⁶’^7,31’³³’³⁴ Exemplary sequences for the N and C terminal halves of the split PEs can be found in Table 4. The CFN scar can be present or absent.

The compositions and methods described herein can also use PE2, PE4, PEmax, PE6, PE6a, PE6b, PE6c, PE6d, PE6e, PE6a/e, PE7, PE3max, or PE5max, or a variant thereof, optionally with truncation of the RNaseH domain of the RT, e.g., PEmaxARNaseH (Davis et al., Nat Biotechnol. 2024;42:253-64), as well as split versions of those PEs.

Alternatively, an insertion frame editor (iFE), in which DNA polymerase beta (POLB) is fused to a Cas9 nuclease, can be used in place of the prime editor.

Optimized pegRNAs

Also provided herein are optimized pegRNA (with optimized RTT length, PBS length, engineered pegRNA (epegRNA) composition (with evopreQl, ‘trimmed evopreQE (tevopreQl), or mpknot hairpins added to the 3’ end of a pegRNA), silent mutation type and location, RNA linkers, RT initiation sequences, etc.; exemplary sequences are shown in Table 1, and exemplary spacer sequences for pegRNA targeting GJB2 are shown in Table 2).

Exemplary pegRNAs can include the pegRNAs in Table 2, including those with the preferred combinations of parameters as listed in Table A. Other preferred pegRNA include those listed in Table A, but with mpknot epegRNA in place of the tevopreQl, or with no epegRNA. Preferably the pegRNAs target GJB2, and optionally comprise the spacer sequence GCACGCUGCAGACGAUCCUGG (SEQ ID NO: 378), GGCACGCUGCAGACGAUCCUGG (SEQ ID NO: 379), GCUGCAGACGAUCCUGGGGG (SEQ ID NO:380), or GCAGACGAUCCUGGGGGUG (SEQ ID NO:381).

The pegRNAs can introduce silent mutations, e.g., within 0-6 codons of the intended mutation, e.g., re-coding the GJB2 amino acids Leu 10, Glyl 1, and Glyl2, to use for targeting ngRNAs.

TABLE A - Exemplary Optimized pegRNA

Nicking guide RNAs (ngRNAs)

The methods and compositions can also include nicking guide RNAs (ngRNAs), e.g., as shown in Table B, which are specific to the actual edit/silent mutation being installed, and that direct the PE to introduce a nick (preferably within 0-6 codons of the intended mutation) on the non-edited strand. For example, the ngRNA LM226 can be used for most edits/pegRNAs, but when silent mutations are installed, then a different ngRNA is used. When using a pegRNA with the silent mutations ‘GCT mutation’ on the RTT of the pegRNA, then an ngRNA specific to that set of silent mutations (like LM3186 or LM3192) is used.

TABLE B - Most optimal nicking gRNA (ngRNA / PE3(b) gRNA) sequences Methods of Gene Editing and Treatment

The methods described herein can be used, e.g., for prime editing genes in cells, as well as for the treatment of disorders associated with mutations in the GJB2 gene, e.g., a c35delG mutation.

In some embodiments, the disorder is DFNB1 hereditary deafness, as described above. Generally, the methods include administering a therapeutically effective amount of a prime editing system as described herein, to a subject who is in need of, or who has been determined to be in need of, such treatment. The prime editing systems described herein include at least two components adapted from naturally occurring CRISPR systems: a pegRNA and a prime editor, e.g., an optimized split prime editor as described herein. These two components form a complex that is capable of associating with a specific nucleic acid sequence in a cell and editing the DNA in or around that nucleic acid sequence, for example for correction of the GJB2 c.35delG mutation. The prime editor can include mutations that alter PAM specificity or on-target activity, a number of which are known in the art. The methods can include delivering nucleic acids encoding the prime editor and pegRNA, e.g., in naked mRNA or dual AAVs as described herein, or can include delivering ribonucleoprotein (RNP) complexes comprising the PE and pegRNA.

As used in this context, to “treat” means to ameliorate at least one symptom of the disorder associated with a mutation in the GJB2 gene. Often, these mutations result in hearing loss; thus, a treatment comprising administration of a therapeutic prime editing system as described herein can result in a reduction in hearing impairment; a reduction in the rate of progression of hearing loss; and/or a return or approach to normal hearing. Hearing can be tested using known methods, e.g., audiology testing.

The methods can be used to treat any subject (e.g., a mammalian subject, preferably a human subject) who has a mutation in the GJB2 gene, e.g., in one or both alleles of GJB2. As used herein, an “allele” is one of a pair or series of genetic variants of a polymorphism (also referred to as a mutation) at a specific genomic location. As used herein, “genotype” refers to the diploid combination of alleles for a given genetic polymorphism. A homozygous subject carries two copies of the same allele and a heterozygous subject carries two different alleles. Methods for identifying subjects with such mutations are known in the art; see, e.g., Yan et al., J Hum Genet. 2009 Dec; 54(12): 732-738; Leroy et al., Exp Eye Res. 2001 May;72(5):503-9; or Consugar et al., Genet Med. 2015 Apr;17(4):253-261. For example, gel electrophoresis, capillary electrophoresis, size exclusion chromatography, sequencing, and/or arrays can be used to detect the presence or absence of the allele or genotype. Amplification of nucleic acids, where desirable, can be accomplished using methods known in the art, e.g., PCR. In one example, a sample (e.g., a sample comprising genomic DNA), is obtained from a subject. The DNA in the sample is then examined to identify or detect the presence of an allele or genotype as described herein. The allele or genotype can be identified or determined by any method described herein, e.g., by Sanger sequencing or Next Generation Sequencing (NGS). Other methods can include hybridization of the gene in the genomic DNA, RNA, or cDNAto a nucleic acid probe, e.g., a DNA probe (which includes cDNA and oligonucleotide probes) or an RNA probe. The nucleic acid probe can be designed to specifically or preferentially hybridize with a particular mutation (also referred to as a polymorphic variant).

Other methods of nucleic acid analysis can include direct manual sequencing (Church and Gilbert, Proc. Natl. Acad. Sci. USA 81 : 1991-1995 (1988); Sanger et al., Proc. Natl. Acad. Sci. USA 74:5463-5467 (1977); Beavis et al., U.S. Pat. No. 5,288,644); automated fluorescent sequencing; single- stranded conformation polymorphism assays (SSCP) (Schafer et al., Nat. Biotechnol. 15:33-39 (1995)); clamped denaturing gel electrophoresis (CDGE); two-dimensional gel electrophoresis (2DGE or TDGE); conformational sensitive gel electrophoresis (CSGE); denaturing gradient gel electrophoresis (DGGE) (Sheffield et al., Proc. Natl. Acad. Sci. USA 86:232-236 (1989)); denaturing high performance liquid chromatography (DHPLC, Underhill et al., Genome Res. 7:996-1005 (1997)); infrared matrix-assisted laser desorption/ionization (IR-MALDI) mass spectrometry (WO 99/57318); mobility shift analysis (Orita et al., Proc. Natl. Acad. Sci. USA 86:2766-2770 (1989)); restriction enzyme analysis (Flavell et al., Cell 15:25 (1978); Geever et al., Proc. Natl. Acad. Sci. USA 78:5081 (1981)); quantitative real-time PCR (Raca et al., Genet Test 8(4):387-94 (2004)); heteroduplex analysis; chemical mismatch cleavage (CMC) (Cotton et al., Proc. Natl. Acad. Sci. USA 85:4397-4401 (1985)); RNase protection assays (Myers et al., Science 230: 1242 (1985)); use of polypeptides that recognize nucleotide mismatches, e.g., E. coli mutS protein; allele-specific PCR, and combinations of such methods. See, e.g., Gerber et al., U.S. Patent Publication No. 2004/0014095 which is incorporated herein by reference in its entirety.

AAV Delivery Systems

The methods can include delivery of a prime editing system, including a prime editor (e.g., a split PE as described herein) and pegRNA and optional ngRNA, to a subject in need thereof. The delivery methods can include, e.g., viral delivery, e.g., preferably using an adeno-associated virus (AAV) vector that comprises sequences encoding the PE and guide RNA(s). Adeno-associated virus is a naturally occurring defective virus that requires another virus, such as an adenovirus or a herpes virus, as a helper virus for efficient replication and a productive life cycle. (For a review see Muzyczka et al., Curr. Topics in Micro and Immunol.158:97-129 (1992)). AAV vectors efficiently transduce various cell types and can produce long-term expression of transgenes in vivo. AAV vectors have been extensively used for gene augmentation or replacement and have shown therapeutic efficacy in a range of animal models as well as in the clinic; see, e.g., Mingozzi and High, Nature Reviews Genetics 12, 341 - 355 (2011); Deyle and Russell, Curr Opin Mol Then 2009 Aug; 11(4): 442-447; Asokan et al., Mol Then 2012 April; 20(4): 699-708. AAV vectors containing as little as 300 base pairs of AAV can be packaged and can produce recombinant protein expression. For example, AAV2, AAV5, AAV2/5, AAV2/8 and AAV2/7 vectors have been used to introduce DNAinto photoreceptor cells (see, e.g., Pang et al., Vision Research 2008, 48(3):377-385; Khani et al., Invest Ophthalmol Vis Sci. 2007 Sep;48(9):3954-61; Allocca et al., J. Virol. 2007 81(20): 11372-11380). The PHP.eB vector can also be used. In some embodiments, the AAV vector can include (or include a sequence encoding) an AAV capsid polypeptide described in PCT/US2014/060163; for example, a virus particle comprising an AAV capsid polypeptide having an amino acid sequence selected from the group consisting of SEQ ID NOs: 1, 3, 5, 7, 9, 11, 13, 15, and 17 of PCT/US2014/060163, and a PE sequence and pegRNA sequence as described herein. In some embodiments, the AAV capsid polypeptide is an Anc80 polypeptide, e.g., Anc80L27; Anc80L59; Anc80L60; Anc80L62; Anc80L65; Anc80L33; Anc80L36; or Anc80L44. In some embodiments, the AAV incorporates inverted terminal repeats (ITRs) derived from the AAV2 serotype. Exemplary left and right ITRs are presented in Table 6 of WO 2018/026976. It should be noted, however, that numerous modified versions of the AAV2 ITRs are used in the field, and the ITR sequences shown below are exemplary and are not intended to be limiting. Modifications of these sequences are known in the art, or will be evident to skilled artisans, and are thus included in the scope of this disclosure.

The virus can also include one or more sequences that promote expression of a transgene, e.g., one or more promoter sequences; enhancer sequences, e.g. 5’ untranslated region (UTR) or a 3’ UTR; a polyadenylation site; and/or insulator sequences. In some embodiments, the promoter is an inner-ear specific promoter, e.g., a GJB2 promoter. In some embodiments, the promoter is a pan-cell type promoter, e.g., chicken beta-actin (CBA), CAG, CASI, cytomegalovirus (CMV), beta glucuronidase, (GUSB), ubiquitin C (UBC), or Rous sarcoma virus (RSV) promoter. See, eg., WO2021067448 and US20230340038). The woodchuck hepatitis virus posttranscriptional response element (WPRE) can also be used. Nucleotide sequences for each of these promoters are known in the art. Modifications of these sequences may be possible or desirable in certain applications, and such modifications are within the scope of this disclosure.

Expression of the pegRNA in the AAV vector is driven by a promoter known in the art. In some embodiments, a polymerase III promoter, such as a human U6 promoter. An exemplary U6 promoter sequence is presented below:

AAGGTCGGGCAGGAAGAGGGCCTATTTCCCATGATTCCTTCATATTT GCATATACGATACAAGGCTGTTAGAGAGATAATTAGAATTAATTTGACTGTA AACACAAAGATATTAGTACAAAATACGTGACGTAGAAAGTAATAATTTCTTG GGTAGTTTGCAGTTTTAAAATTATGTTTTAAAATGGACTATCATATGCTTACC GTAACTTGAAAGTATTTCGATTTCTTGGCTTTATATATCTTGTGGAAAGGAC GAAACACC. (SEQ ID NO:1)

In some embodiments, the nucleic acid or AAV vector shares at least 80%, 85%, 90%, 95%, 96%, 97%, 98%, 99% or greater sequence identity with one of the nucleic acids or AAV vectors recited above.

The AAV genomes described above can be packaged into AAV capsids, which capsids can be included in compositions (such as pharmaceutical compositions) and/or administered to subjects. An exemplary pharmaceutical composition comprising an AAV capsid according to this disclosure can include a pharmaceutically acceptable carrier such as balanced saline solution (BSS) and one or more surfactants (e.g., Tween 20) and/or a thermosensitive or reverse-thermosensitive polymer (e.g., pluronic). Other pharmaceutical formulation elements known in the art may also be suitable for use in the compositions described here.

Thus present disclosure also provides dual AAV vectors encoding split prime editing systems as described herein, and on the use of such vectors to treat GJB2 mutation-associated disease. Exemplary AAV vector genomes are described in WO2019/183641, which illustrates certain fixed and variable elements of these vectors: inverted terminal repeats (ITRs), a pegRNA sequence and promoter sequences to drive its expression, a coding sequence for each half of the split prime editor and another promoter to drive its expression. Each of these elements is discussed in detail herein. An exemplary pair of constructs for use in the methods described herein could include a first AAV with sequences encoding one half of the split PE driven by a promoter, and a sequence encoding the pegRNA driven by U6, and a second AAV with the second half of the split PE driven by a promoter, preferably the same promoter as the promoter driving expression of the first half). Alternatively, an additional AAV can be used to deliver the pegRNA.

Compositions comprising AAV vectors according to this disclosure can be administered to subjects by any suitable means, including without limitation injection, for example, injection through the round window. The concentration of AAV vector within the composition is selected to ensure, among other things, that a sufficient AAV dose is administered to the inner ear of the subject, taking account of dead volume within the injection apparatus and the relatively limited volume that can be safely administered. Suitable doses may include, for example, IxlO¹¹ viral genomes (vg)/mL, 2xlOⁿ viral genomes (vg)/mL, 3xlOⁿ viral genomes (vg)/mL, 4xlOⁿ viral genomes (vg)/mL, 5xlOⁿ viral genomes (vg)/mL, 6xlOⁿ viral genomes (vg)/mL, 7xlOⁿ viral genomes (vg)/mL, 8xlOⁿ viral genomes (vg)/mL, 9xlOⁿ viral genomes (vg)/mL, IxlO¹² vg/mL, 2xl0¹² viral genomes (vg)/mL, 3xl0¹² viral genomes (vg)/mL, 4xl0¹² viral genomes (vg)/mL, 5xl0¹² viral genomes (vg)/mL, 6xl0¹² viral genomes (vg)/mL, 7xl0¹² viral genomes (vg)/mL, 8xl0¹² viral genomes (vg)/mL, 9xl0¹² viral genomes (vg)/mL, IxlO¹³ vg/mL, 2xl0¹³ viral genomes (vg)/mL, 3xl0¹³ viral genomes (vg)/mL, 4xl0¹³ viral genomes (vg)/mL, 5xl0¹³ viral genomes (vg)/mL, 6xl0¹³ viral genomes (vg)/mL, 7xl0¹³ viral genomes (vg)/mL, 8xl0¹³ viral genomes (vg)/mL, or 9xl0¹³ viral genomes (vg)/mL. Any suitable volume of the composition may be delivered to the cochlear space.

For delivery to the inner ear, injection to the cochlear duct, which is filled with high potassium endolymph fluid, could provide direct access to hair cells. However, alterations to this delicate fluid environment may disrupt the endocochlear potential, heightening the risk for injection- related toxicity. The perilymph-filled spaces surrounding the cochlear duct, scala tympani and scala vestibuli, can be accessed from the middle ear, either through the oval or round window membrane (RWM). The RWM, which is the only non-bony opening into the inner ear, is relatively easily accessible in many animal models and administration of viral vector using this route is well tolerated. Administration through the oval window or across the tympanic membrane can also be used. See, e.g., W02017100791 and US7206639.

For pre-clinical development purposes, systems, compositions, nucleotides and vectors according to this disclosure can be evaluated ex vivo using a cochlear explant system, or in vivo using an animal model such as a mouse, rabbit, pig, nonhuman primate, etc. Cochlear explants are optionally maintained on a support matrix, and AAV vectors can be delivered by injection. Tissue for cochlear explanation can be obtained from human or animal subjects, for example mouse.

Explants are particularly useful for studying the expression of pegRNAs and/or PEs following viral transduction, and for studying genome editing over comparatively short intervals. These models also permit higher throughput than may be possible in animal models, and can be predictive of expression and genome editing in animal models and subjects. Small (mouse, rat) and large animal models (such as rabbit, pig, nonhuman primate) can be used for pharmacological and/or toxicological studies and for testing the systems, nucleotides, vectors and compositions of this disclosure under conditions and at volumes that approximate those that will be used in clinic. Because model systems are selected to recapitulate relevant aspects of human anatomy and/or physiology, the data obtained in these systems will generally (though not necessarily) be predictive of the behavior of AAV vectors and compositions according to this disclosure in human and animal subjects.

Nanoparticles

In some embodiments, the prime editing polynucleotides as disclosed herein for delivery to a target tissue in vivo are encapsulated or associated with in a nanoparticle. Methods for nanoparticle packaging are well known in the art, and are described, for example, in Bose S, et al (Role of Nucleolin in Human Parainfluenza Virus Type 3 Infection of Human Lung Epithelial Cells. J. Virol. 78:8146. 2004); Dong Y et al. Poly(d,l-lactide-co-glycolide)/montmorillonite nanoparticles for oral delivery of anticancer drugs. Biomaterials 26:6068. 2005); Lobenberg R. et al (Improved body distribution of 14C-labelled AZT bound to nanoparticles in rats determined by radioluminography. J Drug Target 5: 171.1998); Sakuma S R et al (Mucoadhesion of polystyrene nanoparticles having surface hydrophilic polymeric chains in the gastrointestinal tract. Int J Pharm 177: 161. 1999); Virovic L et al. Novel delivery methods for treatment of viral hepatitis: an update. Expert Opin Drug Deliv 2:707.2005); and Zimmermann E et al, Electrolyte- and pH-stabilities of aqueous solid lipid nanoparticle (SLN) dispersions in artificial gastrointestinal media. Eur J Pharm Biopharm 52:203. 2001). In some embodiments, one or more polynucleotides is delivered to a target tissue in vivo in a vesicle, e.g., a liposome (see Langer, Science 249: 1527-1533 (1990); Treat et al., in Liposomes in the Therapy of Infectious Disease and Cancer, Lopez-Berestein and Fidler (eds.), Liss, New York, pp. 353-365 (1989); Lopez-Berestein, ibid., pp. 317-327; see generally ibid). In some embodiments, lipid- based nanoparticles (LNP) are used; see, e.g., Robinson et al., Mol Ther. 2018 Aug l;26(8):2034-2046; US9956271B2.

The present methods and compositions can include microvesicles or a preparation thereof that contains one or more therapeutic molecules, e.g., PE RNPs, or polynucleotides or RNA encoding a PE system, as described herein. “Microvesicles”, as the term is used herein, refers to membrane-derived microvesicles, which includes a range of extracellular vesicles, including exosomes, microparticles and shed microvesicles secreted by many cell types under both normal physiological and pathological conditions. See, e.g., EP2010663B1. The methods and compositions described herein can be applied to microvesicles of all sizes. In some embodiments, the microvesicles are 30 to 200 nm, 30 to 800 nm, or up to 2 um. The methods and compositions described herein can also be more broadly applied to all extracellular vesicles, a term which encompasses exosomes, shed microvesicles, oncosomes, ectosomes, and retroviral -like particles. Such a microvesicle or preparation is produced by the herein described methods. As the term is used herein, a microvesicle preparation refers to a population of microvesicles obtained/prepared from the same cellular source. Such a preparation is generated, for example, in vitro, by culturing cells expressing the nucleic acid molecule of the instant invention and isolating microvesicles produced by the cells. Methods of isolating such microvesicles are known in the art (Thery et al., Isolation and characterization of exosomes from cell culture supernatants and biological fluids, in Current Protocols Cell Biology, Chapter 3, 322, (John Wiley, 2006); Palmisano et al., (Mol Cell Proteomics. 2012 August; ll(8):230-43) and Waldenstrom et al., ((2012) PLoS ONE 7(4): e34653)).

EXEMPLARY SEQUENCES AND CONSTRUCTS

In some embodiments, the sequence of a protein or nucleic acid used in a composition or method described herein is at least 80%, 85%, 90%, 95%, 97%, 98%, or 99% identical to a reference sequence set forth herein (e.g., in Tables 1-4). To determine the percent identity of two amino acid sequences, or of two nucleic acid sequences, the sequences are aligned for optimal comparison purposes (e.g., gaps can be introduced in one or both of a first and a second amino acid or nucleic acid sequence for optimal alignment and non-homologous sequences can be disregarded for comparison purposes). In a preferred embodiment, the length of a reference sequence aligned for comparison purposes is at least 80% of the length of the reference sequence, and in some embodiments is at least 90% or 100%. The amino acid residues or nucleotides at corresponding amino acid positions or nucleotide positions are then compared. When a position in the first sequence is occupied by the same amino acid residue or nucleotide as the corresponding position in the second sequence, then the molecules are identical at that position. The percent identity between the two sequences is a function of the number of identical positions shared by the sequences, taking into account the number of gaps, and the length of each gap, which need to be introduced for optimal alignment of the two sequences.

The comparison of sequences and determination of percent identity between two sequences can be accomplished using a mathematical algorithm. For example, the percent identity between two amino acid sequences can be determined using the Needleman and Wunsch ((1970) J. Mol. Biol. 48:444-453) algorithm which has been incorporated into the GAP program in the GCG software package (available on the world wide web at gcg.com), using the default parameters, e.g., a Blossum 62 scoring matrix with a gap penalty of 12, a gap extend penalty of 4, and a frameshift gap penalty of 5.

EXAMPLES

The invention is further described in the following examples, which do not limit the scope of the invention described in the claims.

METHODS

The following materials and methods were used in the Examples below.

Plasmids and oligonucleotides

Plasmid constructs were generated via conventional molecular cloning (restriction digests and ligation), isothermal assembly, or Golden Gate assembly.

All nicking gRNA (ngRNA) plasmids were generated by ligation into BsmBI- digested pUC19-U6-[BsmBI_spacer_entry]-SpCas9_sgRNAscaffold (BPK1520; Addgene plasmid 65777). Expression plasmids for human U6 promoter-driven pegRNAs were generated by phosphorylating, annealing, and ligating 3 sets of duplexed oligos corresponding to (1) the spacer sequence, (2) the SpCas9 gRNA scaffold, and (3) the pegRNA extension (RTT/PBS) into BsmBI-digested pegRNA entry vectors, including pUC19-U6-[BsmBI_entry] (MNW320; Addgene plasmid 208977) for conventional pegRNAs, pUC19-U6-[BsmBI_entry]-tevopreQi (LM1138) for tevopreQi epegRNAs, pUC19-U6-[BsmBI_entry]-mpknot (LM1140) for mpknot epegRNAs, pUC19-modbU6-[BsmBI_entry]- tevopreQi (BKS1042) for bU6 tevopreQi epegRNAs, or pUC19-hCtRNA-[BsmBI_entry]- tevopreQi (BKS1044) for hCtRNA tevopreQi epegRNAs. PolIII promoter sequences, and SpCas9 sgRNA scaffold and epegRNA sequences are available in Table 1. gRNA, ngRNA, pegRNA, and epegRNA sequences are available in Table 2. Descriptions of plasmids used are available in Table 3. Amino acid sequences of constructs are available in Table 4.

Various Npu intein-split PE constructs were cloned into N- and C-terminal AAV vectors (Addgene plasmids 137177 and 137178, respectively).

Oligonucleotide sequences used in this study for amplicon sequencing were purchased from Integrated DNA Technologies (IDT) (Table 1).

Cell culture and transfections

Human HEK 293 T cells (American Type Culture Collection; ATCC) were cultured in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10% heat-inactivated FBS (HI-FBS) and 1% penicillin-streptomycin. Samples of supernatant media from cell culture experiments were analyzed monthly for the presence of mycoplasma using a PCR-based detection method.

All HEK 293 T cell experiments were performed with at least 2 or 3 independent biological or technical replicates, resulting from transfections using separately seeded cells or from the same cell passage, respectively. Approximately 20-24 hours before transfection, HEK 293T cells were seeded into wells of 96-well plates at a density of -20,000 cells per well. Transfections typically contained 70 ng of prime editor (PE) expression plasmid, 38 ng pegRNA expression plasmid, and 13 ng nicking guide express plasmid (unless otherwise noted) mixed with 0.792 pL of TransIT-X2 (Minis) in a total volume of 20 pL Opti-MEM (Thermo Fisher Scientific), incubated for 15 minutes at room temperature, and distributed across the seeded HEK 293T cells. Genomic DNA was extracted from cells -72 hours after transfection, by discarding the media, resuspending the cells in 100 pL of quick lysis buffer (20 mM Hepes pH 7.5, 100 mM KC1, 5 mM MgCl₂, 5% glycerol, 25 mM DTT, 0.1% Triton X-100, and 60 ng/pL Proteinase K (NEB)), and heating the lysate for 6 minutes at 66 °C, then heating at 98 °C for 2 minutes. Amplicon sequencing and data analysis

Editing efficiencies in bulk transfected cells were assessed ~72 hours after transfection by extracting genomic DNA and performing targeted sequencing using a 2-step PCR-based Illumina library construction method. Briefly, genomic loci were amplified using approximately 50-100 ng of gDNA, Q5 High-fidelity DNA Polymerase (NEB), and PCR-1 primers with cycling conditions of 1 cycle at 98 °C for 2 min; 35 cycles of 98 °C for 10 sec, 58 °C for 10 sec, 72 °C for 20 sec; and 1 cycle of 72 °C for 1 min. PCR products were purified using paramagnetic beads at a ratio of lx. Approximately 20 ng of purified PCR-1 products were used as template for a second round of PCR (PCR-2) to add barcodes and Illumina adapter sequences using Q5 and primers (Table 1) and cycling conditions of 1 cycle at 98 °C for 2 min; 10 cycles at 98 °C for 10 sec, 65 °C for 30 sec, 72 °C 30 sec; and 1 cycle at 72 °C for 5 min. PCR products were purified prior to quantification via capillary electrophoresis (Qiagen QIAxcel), normalization, and pooling. Final libraries were quantified by either qPCR using the KAPA Library Quantification Kit (Complete kit; Universal) (Roche) or via Qubit (Thermo Fisher Scientific) prior to sequencing on a MiSeq sequencer using a 300- cycle v2 kit (Illumina). On-target genome editing activities were determined from amplicon sequencing data using CRISPResso2⁵⁴. Using CRISPResso2, amplicon sequences were aligned to a reference sequence in HDR mode using the intended editing outcome as the expected allele (-e) and the parameters “-q 30” and “-discard indel reads”. For each amplicon, the quantification window (-qwc) was defined as the entire sequence between gRNA- and ngRNA- directed cut sites plus an additional 10 bp on either side of each nicking site. The same quantification window was used to analyze data for each amplicon, whether or not a ngRNA was transfected. Editing efficiencies were quantified by determining: (# of reads aligned to HDR / number of total reads). Indel efficiencies were quantified as (number of discarded indel-containing reads / number of total reads).

Generation of a c.35delG HEK293T cell line

To generate HEK 293T cell lines bearing the GJB2 c.35delG mutation, transfections were performed as described above. HEK 293T cells were seeded and transfected as described above with the following plasmids: (1) PE2, (2) a pegRNA targeting spacer 1 with PBS length 11 nt and RTT length 13 nt, and (3) ngRNA PE3b- 1-20 (Table 2). Transfected cells were grown for approximately 72 hours prior to dilution plating for single cell clones into 96-well plates, which were then grown for ~2 weeks until confluent. Cells were transferred into 48-well plates with some cell mass reserved to extract genomic DNA (gDNA) for genotyping via PCR and Sanger sequencing or targeted amplicon sequencing (analyzed using CRISPResso2) to verify genotypes of cell colonies.

Specificity assessment using GUIDE-seq2

Approximately 20,000 HEK 293T cells were seeded per well in 96-well plates ~ 20 hours prior to transfection, performed using 29 ng of nuclease expression plasmid, 12.5 ng of gRNA expression plasmid, 1 pmol of the GUIDE-seq doublestranded oligodeoxynucleotide tag (dsODN; oSQT685/686)⁸⁸, and 0.3 pL of TransIT- X2 (Minis). Genomic DNA was extracted ~72 hours post transfection using the DNAdvance Kit (Beckman Coulter) according to manufacturer's instructions, and then quantified by Qubit (Thermo Fisher). On-target dsODN integration was assessed by PCR amplification, library preparation, and next-generation sequencing as described above, with data analysis via CRISPResso2 run in non-pooled mode by supplying the target site spacer, the reference amplicon, and both the forward and reverse dsODN-containing amplicons as ‘HDR’ alleles with custom parameters: -w 25 -g GUIDE — plot window size 50. The fraction of alleles bearing an integrated dsODN was calculated as the number of reads mapped to the forward dsODN amplicon plus the number of reads mapped to the reverse dsODN amplicon divided by the sum of the total reads mapped to all three amplicons.

GUIDE-seq2 reactions were performed essentially as described (Lazzarotto & Li et al, in preparation) with minor modifications. Briefly, the Tn5 transposase was prepared by combining 36 pL hyperactive Tn5 (1.85 mg/mL, purified as previously described⁵⁵), 15 pL annealed i5 adapter oligos encoding 8 nucleotide (nt) barcodes and 10-nt unique molecular indexes (UMIs) (as previously described; Walton, 2020, Science; 368(6488):290-296), with 52 pL 2x Tn5 dialysis buffer (100 mM HEPES- KOH pH 7.2, 200 mM NaCl, 0.2 mM EDTA, 2 mM DTT, 0.2% Triton X-100, and 20% glycerol) for 60 minutes at 24 °C. Tagmentation reactions were performed in 40 pL reactions for 7 minutes at 55 °C, containing approximately 250 ng of genomic DNA, 8 pL of the assembled Tn5/i5 -transposome, and 8 pL of freshly prepared 5x TAPS-DMF buffer (50 mM TAPS-NaOH, 25 mM MgCh, and 50% dimethylformamide (DMF)). Tagmentation reactions were halted using 5 pL of a 50% proteinase K (NEB) solution (mixed with H2O) with incubation at 55 °C for 15 minutes, purified using SPRI-guanidine magnetic beads, and analyzed via TapeStation with High Sensitivity D5000 tapes (Agilent). Separate PCR reactions were performed using dsODN sense- and antisense-specific primers (Table 1) using Platinum Taq (Thermo Fisher), with a thermocycler program of 95 °C for 5 minutes, followed by 15 cycles of temperature cycling (95 °C for 30 s, 70 °C (-1 °C per cycle) for 120 s, and 72 °C for 30 s), 20 constant cycles (95 °C for 30 s, 55 °C for 60 s, and 72 °C for 30 s), an a final extension at 72 °C for 5 minutes. PCR products were purified using SPRI beads and analyzed via QIAxcel (Qiagen) prior to sample pooling to form single sense- and antisense- libraries. Libraries were purified using the Pippin Prep (Sage Science) DNA size selection system to achieve a size range of 250-500 base pairs. Sense- and antisense- libraries were quantified using Qubit (Thermo Fisher) and pooled in equal amounts to achieve a final concentration of 2 nM. The library was sequenced using NextSeql000/2000 P3 kit (Illumina) with cycle settings of 146, 8, 18, 146. Demultiplexed sequencing reads were down sampled to ensure equal numbers of reads for samples being compared using the same gRNA. Data analysis was performed using an updated version of the open-source GUTDE-seq2 analysis software³⁸ (github.com/tsailabSJ/guideseq/tree/V2) with max_mismatches parameter set to 6.

AAV production

Adeno-associated viral (AAV) vectors were generated via Packgene.

Gjb235delG mouse model generation

CRISPR-Cas9 technology was used to modify the Gjb2 gene to generate the Gjb2" mouse strain via assistance of the Genome Modification Facility (GMF, Cambridge, MA, USA). SoxlO-Cre mice were purchased from Jackson Laboratory (Bar Harbor, MN, USA). Specific gRNAs were developed to target intron 1 and exon 2 of the Gjb2 gene in the C57BL/6 mouse strain (spacer sequences AUGUACCCGGAACCAGAGAU (SEQ ID NO:2) and AGCCAGGCAAUGCAUUAGAC (SEQ ID NO:3); Twist Biosciences). For Gjb2^fl/ the donor template was a small portion of human intron 1, coding sequence (CDS), comprising the 35delG mutation, and the rest of exon 2 flanked by a pair of loxP sites: GACCTCAGCCAAGAAACTACCGGGAAGCGACACGGGGTCCTGTGGGATTTCACAAAT TCTGCCATACGAGCTGGGCAGCCCTGCCTCTACGGTGAGGTTGGCCCAAGCTTTTTT CTGTTACTGT GAT AC AC T G C AAAG CTTGCTACTACT GAC AC AAC C C AC T GAG T GAC C TGTACAGAAATGCGAACATATGGGAGCAGGCTTAGCCAGAACTCGGTTCTGCCTTTA TAGTAACAAGTACTGTGCTTTGGTGCACCCCGTAACCACATCTTTATTTTGTGCTAA AGAC T AG G T GAAT T AAC T GAG G T T AAAAC AAAAAC AAAAAC TTTTTTTTTTT GAGAA CTGTACACAGAAATGTGTTGGTGATGGTTTGAGAGCCACACACTTAACCCTATGGCA AAGCAAATTGTGTTGTCACCTATCAGCAGCCTAGAGGAGGCTGTGTGTGCTTTGGTG TTTGCCCGG GAAGAC AG T T AAGAAT AT G T AC C C G GAAC C AGAGAT AC GAC C T AC T T T CCGGCCAAACCAATGATATTATGTTTGTCTTCTCCAGTGCCAACCATCCAGAGGACA AGATGGATTGGGGCACGCTGCAGACGATCCTGGGGGTGTGAACAAACACTCCACCAG CATTGGAAAGATCTGGCTCACCGTCCTCTTCATTTTTCGCATTATGATCCTCGTTGT GGCTGCAAAGGAGGTGTGGGGAGATGAGCAGGCCGACTTTGTCTGCAACACCCTGCA GCCAGGCTGCAAGAACGTGTGCTACGATCACTACTTCCCCATCTCCCACATCCGGCT ATGGGCCCTGCAGCTGATCTTCGTGTCCACGCCAGCGCTCCTAGTGGCCATGCACGT GGCCTACCGGAGACATGAGAAGAAGAGGAAGTTCATCAAGGGGGAGATAAAGAGTGA ATTTAAGGACATCGAGGAGATCAAAACCCAGAAGGTCCGCATCGAAGGCTCCCTGTG GTGGACCTACACAAGCAGCATCTTCTTCCGGGTCATCTTCGAAGCCGCCTTCATGTA CGTCTTCTATGTCATGTACGACGGCTTCTCCATGCAGCGGCTGGTGAAGTGCAACGC CTGGCCTTGTCCCAACACTGTGGACTGCTTTGTGTCCCGGCCCACGGAGAAGACTGT CTTCACAGTGTTCATGATTGCAGTGTCTGGAATTTGCATCCTGCTGAATGTCACTGA ATTGTGTTATTTGCTAATTAGATATTGTTCTGGGAAGTCAAAAAAGCCAGTTTACCC ATACGATGTTCCAGATTACGCTTAATGATAATGCATTGCCTGGCTGCTAGAGCAAAG ATGGAGGGAGAGGATGAGGCAACCCATGCTTAGTCGCTTAGTCGGCAGAGCTCAGCC AC C AG C AG T T C C C AAGACAAAC AT T C C C AT C T AAAAT G C C AC C AT T T GAAG T C C C T G GAGGCCTCCTATGAAACTCCAGAAGCCTCCGTGGGCCTCCCTTCCCCCAAAGCTCCC AAACAAAGGCCCAATTCTATGCCTGTATTAATGGGTTCTAAAGTTAGTTCCACTGAG ACCCCGTGCTGGTAGCGACCAAGCTTGGAGGATACATTCACAGTTTAAACAAAGGGA TCTCACATTGTTTCTCTTCCTCTGAGGACAGGAGAGAGGAACCCGGTCCTGAGGAAG GTGCCACTGAGAAAGTCCCCCCCCCCAAGTTACCCCTAGCTAAAGGTAAAGAAATGC CATACTGTTATTTGCTTTGATAGTTTACGTGTGCAACAATGGACAAAAAGC ( SEQ ID NO : 4 ) .

The gRNAs, CRISPR-Cas9 nuclease mRNA, and ssDonor templates were delivered simultaneously into the zygote of a C57BL/6 mouse to generate the founders. Four male founder mice were obtained for Gjb2'^v . By mating Gjb2'^v and SoxlO-Cre mice, heterozygous conditional KO mice were obtained. Crossbreeding between these heterozygous conditional KO mice generated Gjb2'¹ '¹ SoxlO-Cre progenies. To be noted, based on previous reports, early-maturation of Cre would happen during embryonic stage if the SoxlO-Cre was transferred by paternal side that will cause the allele deletion in all of the cell types in the following generations.

Therefore, to avoid this situation, we preferred to set up mating with female SoxlO- Cre carrier.

Genotyping

Genotyping was performed using the following primers: Gjb2_flx IF: 5’- GAAATGTGTTGGTGATGGTTTG-3’ (SEQ ID N0:5) and Gjb2_flx 1R: 5’- ACCGTGAGCCAGATCTTTCC-3’ (SEQ ID NO:6) check for existence of floxed Gjb2 Gjb2_hs IF: 5’-GCCCGGGAAGACAGTTAAG-3’ (SEQ ID N0:7) and Gjb2_hs 1R: 5’-TCTCCCCACACCTCCTTTG-3’ (SEQ ID N0:8) check for the human GJB2 CDS; control F: 5’- GAC AAAATGGTGAAGGTCGG-3 ’ (SEQ ID NO: 9), control R: 5’- CAAAGGCGGAGTTACCAGAG-3’ (SEQ ID NO: 10), SoxlO: 5’-CACCTAGGGTCTGGCATGTG-3’ (SEQ ID NO: 11), and Cre: 5’- AGGCAAATTTTGGTGTACGG-3’ (SEQ ID NO: 12) are for SoxlO-Cre checking. The thermal cycle is set up as below for floxed Gjb2 and humanized c.35delG: 95°C for 5 mins, followed by 35 cycles of 95°C for 30 s, 55°C for 30 secs, and 72°C for 30 secs, then 72°C for 10 mins. For human GJB2 CDS, PstI enzyme digestion is conducted following the PCR program. The thermal cycle for SoxlO-Cre is different: 95°C for 5 mins, followed by 10 cycles of 95°C for 30 s, 65°C for 30 secs (0.5°C per cycle decrease), and 68°C for 30 secs, then 28 cycles of 95°C for 30 s, 60°C for 30 secs, and 72°C for 30 secs, and finally 72°C for 10 mins. The PCR products can be separated on an agarose gel. Floxed Gjb2 allele shows a 335-bp band; wild-type shows a 301-bp band. Hs_35delG shows two bands — 133-bp and 116-bp — after PstI digestion; wild-type shows a 250-bp band because there is no PstI cutting sites in the amplicon of wild-type genome. SoxlO-Cre allele shows a 300-bp band; wild-type shows a 217-bp band.

Neonatal mouse round window membrane (RWM) injections

The RWM injections were performed under a stereomicroscope (Nikon SMZ1500). PO-pl pups were anesthetized using hypothermia by exposure on ice and then kept on an ice pack during the surgery. An incision was made behind and below the pinna on left ear. After removing some muscles, the position of RWM was localized visually by recognizing the facial nerve and the shade of RWM covered by connective tissues. 1.5 pl of the mix of two AAV-packaged PE vector solution (l*10¹⁴) _Was injected with a micropipette needle at a rate of 120 nl/min using a Nanoliter 2000 Injector (World Precision Instruments). After injection, the wound was closed with a 7-0 Vycril surgical suture, then the pups were marked with clipped toes for the following genotyping. Standard postoperative care was applied after the injection.

Mouse cochlear histology and imaging

After harvesting the cochleae from injected mice, they were immediately fixed with 4% formaldehyde in Hank’s balanced salt solution (HBSS) overnight at room temperature, then washed with HBSS and transferred to fresh 10% EDTA for 2 days. After decalcification, the organs of Corti were microdissected, blocked, and permeabilized with 10% donkey/10% goat serum with 0.5% Triton X-100 for 1 h at room temperature. Samples were stained with rabbit polyclonal anti-Cx26 (ThermoFisher)/ rabbit polyclonal anti-HAtag (Cell Signaling Technology), and guinea-pig anti-parvalbumin antibody (Synaptic System). The antibodies were diluted 1 :200 for anti-HAtag, 1 :50 for anti-Cx26, and 1 :200 for anti-parvalbumin in 10% goat serum supplemented with PBS and incubated overnight at 4°C followed by several rinses in HBSS. Next, samples were incubated in blocking solution for 30 mins at room temperature and incubated overnight at room temperature with a goat anti-rabbit IgG secondary antibody conjugated to Alexa Fluor 488, and a goat anti -guinea pig IgG secondary antibody conjugated to Alexa Fluor 647 in a 1 :500 dilution in blocking solution. To label hair bundle actin, we used phalloidin-Alexa Fluor 568.

ABR and DPOAE testing

Auditory measurements were conducted with the protocol described in previous publication. ABRs and DPOAEs were recorded using an EPL acoustic system (Massachusetts Eye and Ear, Boston, MA, USA) in an acoustically and electrically insulated chamber. Adult mice (from p30 to p90) were anesthetized with an intraperitoneal injection of ketamine (100 mg/kg)- xylazine (20 mg/kg) cocktail and placed on a temperature-controlled heating pad set to 37°C during the experiment. Acoustic stimuli were delivered via a custom acoustic assembly consisting of two electrostatic drivers as sound sources and a miniature microphone at the end of a probe tube to measure sound pressure in situ. ABRs were recorded using three subdermal needle electrodes: reference electrode in the scalp between the ears, recording electrode just behind the pinna, and ground electrode in the back near the tail. For ABRs, 5-ms tone-pip stimuli with a 0.5 ms rise-fall time at frequencies from 5.6-32 kHz were delivered in alternating polarity at 30 s'¹. The response was amplified (x 10,000), band-pass filtered (0.3-3 kHz), and averaged (x512) with a PCbased data acquisition system using the Cochlear Function Test Suite software package (Massachusetts Eye and Ear, Boston, MA, USA). Sound levels were incremented in 5-dB steps, from ~20 dB below threshold up to 120 dB sound pressure level (SPL). ABR Peak Analysis software (vl .1.1.9, Massachusetts Eye and Ear, Boston, MA, USA) was used to determine the ABR thresholds. ABR thresholds were confirmed by visual examination, as the lowest stimulus level in which a repeatable waveform could be observed. DPOAEs were recorded for primary tones (frequency ratio f2/fl = 1.2, level ratio LI = L2 + 10), where f2 varied from 5.6 to 32 kHz in half-octave steps. Primary tones were swept in 5 dB steps from 10 to 70 dB SPL (for f2). DPOAE threshold was determined from the average spectra as the fl level required to produce a DPOAE of 5-dB SPL.

Analysis of in vivo editing

Genomic DNA of left cochlea, liver, and brain were extracted from mouse tissue samples using Agencourt DNAdvance Kit (BeckmanCoulter, A48705). Editing efficiencies were assessed by targeted amplicon sequencing as described above and using primers listed in Table 1. RNA of left cochlear, liver, and brain were extracted from mouse tissue samples using the RNeasy Plus Mini Kit (Qiagen, 74136, Germany). 150 ng Purified RNA was reverse transcribed using the ProtoScript® First Strand cDNA Synthesis Kit (NEB, E6300S, US). cDNA then used as template in the targeted amplicon sequencing as described above and using primers listed in Table 1.

Statistics

Data are presented as mean±SD. Statistical analyses were conducted using unpaired Student’s t-test with Bonferroni correction for continuous variables. A p value < 0.05 indicates significance. All analyses were performed using SPSS/Windows software 15.0 (SPSS Inc., Chicago, IL, USA).

Example 1. Creation of a GJB2 c.35delG model cell line to investigate prime editing strategies for correction.

To explore prime editor-mediated correction of the GJB2 c.35delG mutation, we first created a human HEK 293T cell model bearing the c.35delG mutation. We tested a prime editor version 3 strategy (PE3) to generate the c.35delG cell line using pegRNAs targeting the endogenous GJB2 locus (at target site 1) with various lengths of reverse transcription templates (RTTs) and primer binding sites (PBSs), paired with different nicking guide RNAs (ngRNAs) targeting the non-target strand to enhance prime editing efficiency (Fig. 1c). When using ngRNAs targeting different distances from the nicking site on the non-target strand (PE3b-l, PE3b-2, PE3-3) and truncated the spacer length of nicking guide PE3b-l (20, 18, 17, or 16 nt spacers; Fig. Id and Fig. 7a). We also tested the efficiency of PE3 constructs using SpCas9 PAM variant enzymes (SpG-PE2^27,28 and VRQR-PE2^{29 30}) when targeted with pegRNAs to target site 4 (encoding an NGA protospacer-adjacent motif; PAM) with various lengths of RTT and PBS, paired with various ngRNAs targeting different distances from the first nick site (PE3b-l, PE3b-2, PE3b-4) and truncating the spacer length of ngRNAs PE3b-l (20, 18, or 16 nt) and PE3b-2 (21 or 17 nt) (Fig. 7b-7f). We performed transfections with each of these constructs and examined precise deletion efficiency of c.35G.

The pegRNA with a RTT of 13 nt and PBS of 13 nt paired with ngRNA PE3b- 1-20 led to the highest precise editing efficiency (-35%; Fig. Id), which we then used to make the c.35delG cell line (using the workflow is described in the Methods and Fig. 7g). We isolated single cell clones and confirmed the sequence of a homozygous GJB2 c.35delG cell line by Sanger sequencing (Fig. le) and Next Generation Sequencing (NGS) (Fig. If). The homozygous GJB2 c.35delG cell line was utilized for subsequent prime editing studies.

Example 2. Systematic optimization of prime editing efficiency for correction of GJB2 c.35delG

We performed an initial test to investigate c.35delG correction using the SpCas9 PE2 construct when targeting target site 1 with various pegRNAs, and when testing two PE3 ngRNAs (one PE3b and another PE3 ngRNA targeting 30 bp downstream, PE3b-3 and PE3-1, respectively; Fig. 1). We tested pegRNAs encoding different combinations of RTT and PBS lengths, including RTTs of 11, 13, or 17 nts, and PBSs of 11, 13, or 16 nts. After -3 days of cell culture post-transfection, we extracted genomic DNA (gDNA) and sequenced the locus via NGS. We observed that precise 35delG correction varied greatly depending on the pegRNA utilized, even when the difference in length of the RTT or PBS was only 2 nt (Figs, lh-11). The use of a PE3b ngRNA led to higher precise 35delG correction efficiency with lower undesirable insertion/deletion (indel) rates compared to the PE3 ngRNA located distal to the edit (Figs, li and Ij for PE3b versus Figs. Ik and 11 for PE3 ngRNA).

To explore additional parameters to further improve 35delG correction efficiency, we tested other possible target sites (site 2, site 3, site 4) near the intended insertion (Fig. 2a). At these target sites, we also varied the length of the RTT and PBS to test 9 different pegRNAs at each site. However, the 35delG correction efficiency at target sites 2-4 was lower than that of target site 1 (Fig. 2b-c compared to Fig. li). We also tested other target sites using prime editors encoding PAM variant enzymes including VRQR-PE2 and SpG-PE2 targeting an NGA PAM (Fig. 2d and Fig. 8a) and SpG-PE2 and VRAVQL-PE2 (described in US 11286468) targeting an NGT PAM (Fig. 2d and Fig. 8b) and detected a maximum of around 35% correction for both the NGA PAM variant PEs and a maximum of around 5% correction for both the NGT PAM variant PEs.

We therefore selected target site 1 paired with PE3b ngRNAs for further optimization. We performed a systematic screen of the RTT length (investigating lengths of 9-20 nt) and PBS length (9, 10 ,11, 12, or 17 nt) to determine the most optimal combinations for 35delG correction. Results from the screen revealed that the pegRNA with 14 nt RTT length and 12 or 11 nt PBS length, henceforth written using nomenclature as RTT#/PBS#, led to the highest correction of 35delG (Fig. 2e and Fig. 9a). Analysis of insertion or deletion mutations (indels) for each condition in the pegRNA screen revealed generally low unwanted editing with indels under 3% (Fig. 2f and Fig. 9b).

The extended 3’ end of pegRNAs is subject to degradation by exonucleases present in the cells, which decreases the amount of functional pegRNA available to initiate the reverse transcription process⁸. By adding tevopreQi or mpknot hairpins to the 3’ end of a pegRNA, engineered pegRNAs (epegRNAs) were reported to improve prime editing efficiencies⁸. To explore whether epegRNAs could improve our observed levels of 35delG correction, we first tested tevopreQi epegRNAs using combinations of RTT and PBS lengths with low 35delG correction activity (RTT8/PBS10, RTT10/PBS12), medium activity (RTT11/PBS11, RTT13/PBS10, RTT14/PBS11), and high activity (RTT14/PBS11, RTT14/PBS12, RTT15/PBS12, RTT16/PBS12) (Fig. 2g). Overall, the tevopreQi epegRNA improved 35delG correction by -20-40% across all pegRNAs that we examined. We also compared the mpknot and tevopreQi epegRNAs and found that for these few pegRNAs with low and medium 35delG correction efficiency, the mpknot epegRNA was generally more capable than the tevopreQi epegRNA at improving editing (Fig. 2h).

Since adding translationally silent mutations near the intended edit has been shown to improve prime editing efficiencies through mismatch repair evasion, we next sought to test the impact of silent substitutions to improve 35delG correction. We focused on the wobble positions of codons for amino acids G11 and G12 (Fig. 2i) and investigated the impact of silent mutations when using one of the top pegRNA candidates with RTT14/PBS11 in experiments with PEmax. Compared to our previous top performing pegRNA without a 3’ hairpin or mismatches, addition of two silent G-to-C mismatches in the wobble positions of G11 and G12 (leading to +5 and +8 G-to-C installed edits relative to the PE nicking site) increased 35delG correction from 35% to 50% (Fig. 2j). The increase observed with the mismatch-encoding pegRNAs was comparable in magnitude to what we had observed when using epegRNA (Figs. 2g-2h). When combining the doubly mismatched pegRNA encoding +5 and +8 G-to-C edits with epegRNAs further improved 35delG correction to over 60% (Fig. 2j). On other medium efficiency pegRNAs with RTT17/PBS11 or RTT17/PBS13, compared to the unmodified pegRNAs, the addition of the +5 and +8 G-to-C mismatches increased the efficiencies to around 50% and was further improved when also using tevopreQi or mpknot epegRNAs (Fig. 2k).

Given the large potentiation of 35delG correction by the incorporation of the +5 and +8 G-to-C silent substitutions in the RTT of our pegRNA, we then performed a more thorough screen of silent mutations within the RTT of one of the top pegRNA candidates with RTT14/PBS12 (Fig. 21). We introduced various silent mutations to the RTT within the wobble positions of GJB2 codons LIO, G11, and G12 (at +2, +5 and +8 from the PE nick site) and explored the impact of these mutations in the RTT of the tevopreQi epegRNAs on prime editing efficiency. We screened 17 pegRNAs encoding different silent mutations along with their corresponding PE3b ngRNA. The silent mutation templates are named based on the 3 substitutions located at the wobble positions of LIO, Gi l, and G12 (e.g. pegRNAs encoding CTxGGxGGx subsitutions are named ‘xxx’; for example, an RTT template with three silent mutations CTTGGCGGC (SEQ ID NO: 13), compared to the original sequence CTGGGGGGT (SEQ ID NO: 14), is known as ‘TCC’). Among all the silent mutations, the RTT encoding the GCT sequence (+5 G-to-C only) led to the highest correction at levels that surpassed the original GCC substitutions (5 and +8 G-to-C) that we had initially tested (Fig. 21). As the tevopreQi epegRNA with RTT17/PBS13 and the GCC silent mutation also showed over 60% correction, we also did a silent mutation screen for this unmodified pegRNA without the tevopreQi extension (Fig. 2m). Example 3. Optimization of prime editors for dual AAV mediated in vivo delivery

To explore if our prime editing strategy can rescue the hearing loss in a 35delG mouse model, we developed and optimized a dual AAV approach for in vivo delivery of the prime editor, pegRNA, and ngRNA. The coding sequence of the SpCas9-based prime editor is over 7 kb, which is beyond the 4.7-5 kb packaging capacity of an AAV vector genome. We therefore explored various split points in SpCas9 necessary to separate the PE construct into two ‘halves.’ Similar to as previously described for base editors¹³, we utilized the Npu intein to mediate the reconstitution of the N-terminal and C-terminal proteins. To facilitate packaging into AAV given the asymmetry of the nCas9-RT PE fusion protein, putative split sites at SpCas9 amino acids 930-1200 could be within the packaging capacity of an AAV genome (Fig. 3a). Based on surface accessible regions or loops of SpCas9 that might best facilitate intein rejoining and/or also accommodate post-splicing amino acid ‘scars’, we first selected potential split sites from this 920-1200 AA region and tested them in the context of SpCas9 nuclease on three different endogenous loci (target sites in AAVS1, PCSK9, ^w DNMTl) (Fig. 3b). Compared to the reported split site of E573 which is commonly used in nuclease and base editing experiments^11,13’¹⁴, our novel candidate sites of E1068, T1069, and G1055 exhibited comparable on-target indel activities (Fig. 3b).

We then constructed split PE constructs encoded on AAV vectors utilizing various split sites of SpCas9 and tested their potential for 35delG correction in the c.35delG HEK 293T cell line (Figs. 3d-3g), and also against other non-GJB2 genomic sites (Fig. 10a). We assembled three versions of C-terminal vectors, with either full- length M-MLV or a truncated C-terminal version lacking the RNaseH domain (delRNaseH), and with or without a CFN scar between the C-intein and N-terminal end of the C-terminal SpCas9-RT extein (Fig. 3c). Compared to the conditions of split PEs with the full-length M-MLV RT, those with the delRNaseH-RT exhibited comparable or improved 35delG correction efficiency (Fig. 3d). Given that the RNaseH domain was not necessary for prime editing³¹, we proceeded with the delRNaseH constructs for further optimization of PE split points. The Npu intein reaction is most optimal when the N-terminal end of the C-terminal extein encodes a CFN peptide sequence³². We therefore tested if adding the amino acids CFN in between the C-intein and N-terminal end of the C-terminal PE fragment could permit better reconstitution of the N-terminal and C-terminal split PE. Addition of the CFN scar peptide generally led to somewhat higher prime editing efficiencies for most split sites (e.g., A1053 and E1068) (Fig. 3e). We also compared candidate splits sites of G1055 and E1068 with previously reported split sites Q844, K1024, and KI 153^6,7’³¹’³³’³⁴, with fully constituted PEmax as a control (Fig. 3f). The El 068 split construct outperformed K1024 and had comparable activity to Q844 and KI 153, although the latter two are not located in an ideal region to generate dual AAV vectors. When titrating split PEs with E1068 and K1024 to further compare their activities, we observed that E1068 was more effective than K1024 at lower DNA dosages that might be more reflective of low expression levels observed in vivo following AAV delivery (Fig. 5g). Two previously reported architectures (v3em and vlem⁶) of split prime editors were tested using a tevopreQi epegRNA with the dual silent +5/+8 G-to-C mutations in the RTT, a tevopreQi epegRNA without silent mutations, and a canonical pegRNA without 3’ protection or silent substitutions. The v3em architecture has both the ngRNA and the pegRNA on the C-terminal vector, while the vlem architecture has the ngRNA on the N-terminal vector and the pegRNA on the C-terminal vector⁶. In the original report from Davis et al., both the vlem and v3em constructs are split at SpCas9 KI 024 on the N-terminal vector and have a CFN scar on the C-terminal vector before resuming normal SpCas9 sequence at E1028⁶. Our experiments revealed that the v3em architecture performed more efficiently overall for 35delG correction, with the largest improvement compared to vlem observed when using a tevopreQi epegRNA without the +5/+8 G-to-C silent mutations (Fig. 10b). Indels resulting from these constructs were minimal, reaching a maximum mean of 1.72% (Fig. 10c). Thus, we selected the v3em architecture to proceed with a tevopreQi epegRNA encoding RTT14/PBS12 and silent mutations.

To determine if improved pegRNA or ngRNA expression might improve in vivo prime editing efficiencies, we tested additional RNA polymerase III (polIII) promoters other than the canonically used human U6 (hU6). We first cloned entry plasmids containing the various promoters to express identical gRNAs (bearing the same spacers and scaffolds; Fig. Ila). The promoters tested in addition to hU6 were bovine U6 (bU6), murine U6 (mU6), a human cysteine transfer RNA (hCtRNA), human 7SK (h7SK), and human Hl (hH I )^{35 37}. These promoters were first tested using SpCas9 nuclease with gRNAs targeting various genomic sites, both in full concentration and in a 1 : 10 titration (Fig. 11b). The polIII promoters that led to comparable on-target editing relative to hU6 expressed gRNAs were bU6 and hCtRNA, which both exhibited much higher editing efficiencies compared to mU6, h7SK, or hHl (Fig. 11b). These polIII promoters were then tested with the v3em PE architecture and an optimal epegRNA and ngRNA for 35delG correction (where the epegRNA was tevopreQi RTT14/PBS12 and either encoded two silent mismatches in the RTT (+5 G-to-C and +8 T-to-C; Design 1), or one silent mismatch in the RTT (+5 G-to-C; Design 2)). The pegRNAs and ngRNAs were cloned for expression from each of the hU6, bU6, and CtRNA promoters. Experiments using these constructs revealed that the top performing promoter combinations typically utilized bU6 promoter to express the pegRNA, and that the pegRNA design with only one silent mutation in the RTT outperformed the design with two silent mutations (Figs, llc-d).

Next, we tested combinations of SpCas9 splite points along with alternate promoter architectures. One of the top candidate polIII promoter combinations (hCtRNA driving the ngRNA and bU6 driving the pegRNA) was tested against hU6 driving both gRNAs on various split points of PEmax-delRNaseH (each with a CFN scar on the N-terminal end of the C-terminal SpCas9 fragment) compared to fully constituted PEmax with and without the M-MLV RNaseH domain as controls. All conditions were transfected to cells in full concentration and in 1 :2, 1 :4, and 1 : 10 titrations. Both combinations of promoters worked comparably on all split sites, but hCtRNA driving the ngRNA with bU6 driving the pegRNA performed better than hU6 driving both gRNAs in fully constituted PEmax with the RNaseH domain removed (Fig. 12a). Next, two top performing split sites, 1024 and 1068, were compared with either hU6 or CtRNA expression of the pegRNA/ngRNA. While both promoter combinations and split sites worked comparably in the full dosage transfection, when the constructs were titrated down 1 :3, 1 :9, and 1 :27, the N-terminal 1068 site with the hCtRNA promoter outperformed the N-terminal 1024 site with both promoter conditions as well as the N-terminal 1068 site with the hU6 promoter (Fig. 12b)

Example 4. Genome-wide specificity of GJB2 correction approach.

We investigated the genome-wide potential for off-target editing using some of our most efficacious pegRNAs and ngRNAs. To nominate potential off-target sites, we performed a modified version of GUTDE-seq^38,39, called GUIDE-seq2 (Lazzarotto et al., in preparation) in our homozygous c.35delG HEK 293T cell line. The GUTDE- seq2 assay utilizes SpCas9 nuclease as a proxy to nominate off-target sites for SpCas9-based PE binding events. We performed GUIDE-seq2 using SpCas9 nuclease paired with a conventional gRNA scaffold (no RTT/PBS extension as used for pegRNAs) and three top candidate pegRNAs that recognize the GJB2 35delG target site 1 sequence (Fig. 4a). For each of these four experiments, we largely only recovered GUIDE-seq2 read counts at the intended on-target site, suggesting that the GJB2 35delG target site 1 spacer is highly specific with only low probability off- target edits. We also performed GUIDE-seq2 using the PE3b gRNA that recognizes the corrected allele to nick the non-edited strand. We identified six low-level off- target sites that were targeted >10-fold less efficiently than the on-target site (Fig. 4b), none of which were located in coding exons of other genes.

We then performed GUIDE-seq2 to identify off-targets for SpCas9 paired with three top evopreQi epegRNAs with the dual +5 and +8 G-to-C silent mutations (Fig. 4c). Similar to our results with the conventional pegRNAs, the more efficient epegRNAs did not lead to detectable off-target editing. Finally, we performed GUTDE-seq2 using the PE3b ngRNA targeting the corrected allele harboring both +5 and +8 G-to-C silent mutations, which does not exist in the native human genome or in our GJB2 c35delG cell line, leading to an inability to recover on-target site read counts (but also no off-target site read counts; Fig. 4d). Together, these results demonstrate that our prime editing approach is likely to be highly specific with a low probability for off-target editing.

Example 5. Development of a humanized Gjb2 mouse model to explore in vivo correction

Next, we sought to translate our GJB2 35delG precise correction approach into an in vivo mouse model. Because the mouse endogenous genomic sequence flanking the 35delG mutation in Gjb2 is different from that of human, we engineered a mouse in which 150 nucleotides of the human sequence, with the 35delG mutation, was knocked into the mouse locus. In this region, most differences are silent, so the mouse mutant amino acid sequence will be identical to human except for an S>T substitution. To create the mouse model, the Genome Modification Facility at Harvard performed pronuclear injection of Cas9, flanking guide RNAs, and donor DNA vector containing the mutant human sequence with flanking homology arms. Mice were sequenced to confirm the correct insertion. Because mice lacking functional GJB2 are embryonic lethal, these Gjb2 35delG mice were maintained as compound heterozygotes with our existing line of floxed Gjb2 for later crossing with SoxlO-Cre mice, where Cre recombination deletes Gjb2 only in the inner ear and is viable.

Next, we performed in vivo experiments using our previously optimized prime editing approach, including v3em PEmax split at El 068, tevopreQi epegRNAs with RTT14/PBS11 and 2x silent mutations (+5 and +8 G-to-C) using a hU6 promoter for epegRNA expression and the mU6 promoter for ngRNA expression. We packaged our initial in-vitro optimized PE and pegRNAs in AAV9-PHP.B vectors and injected them through the round-window membrane (0.6 uL each, IxlO¹⁴ GC/mL) at Pl into cochleae of Qji>2^35delG/fl mice that had either been crossed or not crossed with SoxlO- Cre mice. At 14 days post injection, we sacrificed the mice, harvested tissues (cochlea, liver, and brain), extracted genomic DNA, and performed NGS to assess editing. In cochlea, we observed approximately 1-5% precise editing, which may be a consequence of the AAV9-PHP.B capsid being unable to transduce most cells in the cochlea (Fig. 5a). In liver samples from treated mice, we observed much higher levels of editing in the range of 20-40% (Fig. 5b). Finally, in brain samples our editing approach reached -2-10% precise editing (Fig. 5c). To assess prime editing in cells that express Gjb2, we extracted RNA from the same samples, generated cDNA, and performed NGS on the Gjb2 transcript. We observed -60-70% precise correction of the 35delG human mutation at the RNA level in mouse cochlea and liver (Figs. 5d and 5e, respectively), and -20-40% editing in brain (Fig. 5f), suggesting that our prime editing approach was capable of editing Gjb2 expressing cells that were transduced by the AAV9-PHP.B vectors.

We also assessed editing in the cochlea, brain, and liver of treated mice that were scarified at 30 days post injection. We observed similar levels of precise editing in the day 30 samples from genomic DNA or mRNA (Figs. 5g and 5h, respectively) as compared to the day 14 samples with only low levels of unwanted indels. Together, these results suggest that our extensive optimization of the PE construct, pegRNA, and ngRNA resulted in highly efficacious constructs capable of robust in vivo editing in various mouse tissues. Example 6. Exploration of nuclease-mediated approaches to correct GJB2 35delG

In addition to the prime editing strategy to correct 35delG, we also explored nuclease-based strategies. Previous studies have shown that in certain sequence contexts, SpCas9, when programmed with appropriate gRNAs, is capable of enriched edit outcomes including +1 insertions, -1 deletions, or other indels^{40 43}. To explore whether SpCas9 programmed with GJB2 35delG targeted gRNAs could restore the GJB2 reading frame (e.g. through +1 insertions, or -2 or -5 nt deletions, etc.), we cloned many possible gRNAs that target sites bearing different NGN PAMs near the 35delG site. We utilized the SpCas9 PAM variant enzymes SpG and SpG-LZ3^27,44, where the LZ3 enzyme had previously been shown to lead to enriched +1 edit outcomes at certain sites. Among all the gRNAs tested, gRNAs 2 and guide 17 showed a major (22.73%) 1 bp insertion editing profile, which will restore the reading frame of the GJB2 protein (Figs. 13a-c). Furthermore, gRNAs 19 and 20 exhibited two major editing types that introduced 1 bp or 2 bp deletions, which introduced a 34delG mutation while rescuing the reading frame of the GJB2 protein (Fig. 13e-h). We also observed that gRNA 10 could restore the reading frame while introducing two additional mutations (G12V, V13L) (Figs. 13i-k). However, with all of these approaches, there was still a high percentage of random indels introduced by these nucleases and gRNAs. For other gRNAs tested (e.g. gRNAs 8, 9, 11, 13, and 18) there were no major enriched edit outcomes, leading to heterogenous indels (Fig. 131).

Recently, an insertion frame editor (iFE) was reported to introduce a major insertion profile⁴⁵. In the iFE system, DNA polymerase beta (POLB) is fused to an Cas9 nuclease and when programmed with a guide RNA was reported to predominantly introduce insertions at a target site⁴⁵. Therefore, we assessed the potential of iFEs to correct the 35delG mutation. We compared SpCas9 nuclease, SpCas9-iFE, and LZ3-iFE variants of wild-type SpCas9, SpG, and SpRY variants using multiple gRNAs targeting sides encoding different PAMs. Our results with the iFE or LZ3 constructs did not demonstrate enrichment of insertions that would restore the reading frame of GJB2 35delG, with most editing types resulting in heterogenous indels (Figs. 14 and 15). Example 7. Further optimization of our PE-mediated GJB2 35delG correction approach

Returning to the PE-based method to correct GJB2 35delG, we explored various other approaches to improve activity of correction. We performed a systematic comparison of two top pegRNA configurations (RTT14/PBS11 or RTT14/PBS12), using our El 068 split PE construct, and optimized promoter combinations from our previous experiments. We also explored whether the direction of the pegRNA and ngRNA expression cassette encoded in the C-terminal v3em construct would impact editing (where the conventional vector has the pegRNA/ngRNA cassette facing towards (opposite) the SpCas9-RT expression construct, and the tandem orientation has the pegRNA/ngRNA cassette and SpCas9- RT expression construct being translated in the same direction). The results from these experiments demonstrated highly efficient and precise correction of 35delG with various constructs, with the most efficacious promoter combinations of hU6/bU6 for pegRNA/ngRNA, and efficient editing with either silent mutation construct of GCC and GCT in the RTT (Fig. 13a and b).

We also investigated if the recently described PE6 editors, which were reported to have higher processivity than conventional PEs⁴⁶, could further improve 35delG correction efficiency. We compared PE6a, PE6b, PE6c, PE6d, and PEmax, together with PEmax-delRNaseH programmed with two top pegRNA candidates (RTT14-PBS11-GCC silent mutation and RTT14-PBS12-GCT silent mutation). The correction efficiencies from this comparison revealed that none of the PE6a-6d editors could achieve higher 35delG correction compared to our originally utilized PEmax or PEmax-delRNaseH (Figs. 6a-b). Our results are consistent with what has been reported for these PE6a-6d variant enzymes, where they are typically capable of improving prime editing efficiencies for longer or more structured edits⁴⁶; however, our 35delG correction requires only a short 1 bp insertion.

To explore the ceiling of editing efficiency for correcting 35delG, we performed another systematic optimization on the top five pegRNA candidates using an unmodified pegRNA (peg), an evopreQi epegRNA (evopreQ), an mpknot epegRNA (mpknot), and either epegRNA with the double silent mutation RTT (+5 and +8 G-to-C substitutions; “silent 1”) or the single silent mutation RTT (+5 G-to-C; “silent 2”) (Fig. 6c). These constructs were compared in our homozygous 35delG HEK 293 T cell line using various pegRNA permutations of RTT and PBS lengths (17/13, 16/12, 15/12, 14/12, or 14/11) and across a titration from canonical dose and 1 :3, 1 :9, and 1 :27 dilutions. In general, both the tevopreQi and mpknot epegRNAs led to higher precise editing than the unmodified pegRNA. Addition of either silent mutation type to the epegRNAs further improved 35delG correction efficiency, while the impact of different silent mutations and different epegRNAs varied depending on the specific lengths of RTT/PBS utilized in the pegRNA extension (Fig. 6c). In general, the single +5 silent substitution was superior to the +5/+8 dual silent mutation approach for the RTT17/PBS13 and RTT14/PBS11 pegRNAs, while the converse was evident for the other three pegRNAs. For the top pegRNAs, 35delG correction was most efficient when using the tevopreQi epegRNA compared to the mpknot epegRNA (Fig. 6c).

We also investigated if previous optimized gRNA scaffold sequences^47,48 could further increase 35delG correction efficiency. There are 4 thymine bases encoded in the 5’ end of the SpCas9 crRNA sequence, which has previously been shown to act as a transcription terminator sequence⁴⁹ and thus result in lower levels of gRNA expression^47,48. To investigate the impact of the 4xT sequence of the crRNA on prime editing efficiency, we tested if mutating a thymine (T) base to a cytosine (C) could reduce potential transcription termination and improve prime editing efficiency (Fig. 6d) We also determined whether a 5nt extension of the crRNA/tracrRNA pairing region could improve editing, similar to as previously described⁴⁷. Using the modified pegRNA and ngRNA scaffolds, we then tested 35delG correction using the RTT14/PBS12 combination with the +5 G-to-C silent mutation. We performed a titration of four different plasmid DNA dosages of prime editing components. The results demonstrate that the 35delG correction efficiency from the modified gRNA scaffolds (GTTTC or GTTTC & extension) is comparable to that of the unmodified gRNA scaffold (Fig. 6d).

We also attempted other methods to engineer our pegRNAs to improve the efficiency of prime editing. For a productive prime editing event, the PBS on the 3’ extension of the pegRNA must first anneal with the nicked DNA strand of the Cas9 target site, allowing the 3 ’-OH on the nicked target DNA flap of the RNA:DNA duplex to initiate the reverse transcription process. The efficiency with which the 3’- OH group of the nicked ssDNA recruits the reverse transcriptase domain may be a limiting factor for overall PE efficiency. To determine whether modifications of the 3’ extension of a pegRNA might enhance reverse transcription, we explored the addition of specific sequences to the 3’ end of the pegRNA (Fig. 17a). The reverse transcription of retroviruses usually starts from a specific DNA Initiation Sequence (IS) region on the virus genome which is recognized by the reverse transcriptase as a replication origin^50,51. We transposed the DNA sequences into RNA to make use of ISs at various lengths (IS 1 -IS 12; Table 1) to engineer pegRNAs with 3’ extensions that may preferentially recruit the RT domain of a PE. These modified pegRNAs were tested across three endogenous genomic loci (two target sites to create mutations R124L and R555Q in TGFBI, and a target site in EMX1). Among all 12 ISs, IS3 was capable of consistent improvement compared to an unmodified pegRNA (Figs. 17b- d). Of note is that the improvement seen by adding IS3 to the pegRNA is comparable to using the tevopreQi epegRNA (Figs. 17b-d). We then combined the IS3 sequences with the tevopreQi epegRNA but did not observe synergistic improvement in editing. When replacing the IS3 with two different random sequences (ncl and nc2) of the same length, prime editing efficiencies decreased to levels similar to those observed with the canonical pegRNA (with no 3’ extension; Fig. 17b), suggesting that the specific sequence of IS3 is responsible for the increase in prime editing. We then added IS3 to the top pegRNA candidates to correct c.35delG mutation, and all four pegRNAs experienced an approximate 25% increase in 35delG correction (Figs. 17e- h)

The accessibility and flexibility of the pegRNA PBS to pair with the nicked DNA strand of the target site is another potential rate limiting step for prime editing. In the structure of the pegRNA, the PBS and RTT sequences are directly linked to the 3’ end of the gRNA scaffold, which is located underneath the RuvC domain of SpCas9 and not in close proximity to the nicked NTS^52,53. We therefore wondered whether the addition of an RNA linker between the 3’ of the gRNA scaffold and the RTT of the pegRNA might permit improved annealing between the pegRNA PBS and nicked target NTS via enhanced flexibility (Fig. 18a). By assessing various lengths of RNA linkers (Fig. 18b), we observed that a 5 nt linker substantially improved the installation of an A-to-T mutation in the TGFBI gene in HEK 293 T cells (to create an R124H substitution), and also improved the installation of an R555Q mutation in TGFBI (Fig. 18c). When installing either mutation, the use of longer RNA linkers was deleterious to prime editing efficiencies. We also assessed the impact of a wider range of linker lengths and compositions (Fig. 18d).

Next, we explored whether a 6 nt TCTCTC linker could improve editing when combined with a top pegRNA for correcting the 35delG mutation. Although the linker improved editing with a pegRNA encoding RTT14/PBS11, the linker had an inconsistent impact on other pegRNAs with different RTT/PBS lengths (Figs. 18e- 12g). We next sought to investigate the effect of adding the IS and linker to pegRNAs in various combinations (Figs. 18h and 18i). For the pegRNAs with RTT13/PBS10 (Fig. 18j), RTT14/PBS12 (Fig. 18k), and RTT14/PBS11 (Fig. 181), we observed some synergistic improvements when combining IS and epegRNA, although the addition of linkers did not impact editing. Together, these results highlight various methods to improve prime editing efficiencies, towards the precise correction of GJB2 35delG.

Table 1 - nucleotide sequences of primers, polIII promoters, sgRNA scaffolds, epegRNAs, and initiation sequences.

Table 2 - sequences of gRNA spacers and epegRNA extensions sgRNAs, ngRNAs, pegRNAs, epegRNAs, and sequences used for this study [sgRNA] indicates a sgRNA sequence, e.g., SEQ ID NO:42-44 above (“[sgRNA]” or

[sgRNA GTTTC], or "[sgRNA GTTTC&ext]", as shown in Table 1, or other sgRNA.

Table 3 - list of plasmids

Plasmids and sequences of constructs used in this study.

Table 4 - Amino acid sequences of major constructs used. To avoid redundancy, we used abbreviations for domains that are common amongst the constructs (meaning,

stating the AA sequence of M-MLV-RT once, and then having it listed as “-[M-MLV- RT]” in other cases.

REFERENCES

1. Rees, H. A. & Liu, D. R. Base editing: precision chemistry on the genome and transcriptome of living cells. Nature Reviews Genetics 19, 770-788 (2018).

2. Porto, E. M., Komor, A. C., Slaymaker, I. M. & Yeo, G. W. Base editing: advances and therapeutic opportunities. Nature Reviews Drug Discovery 19, 839-859 (2020).

3. Huang, T. P., Newby, G. A. & Liu, D. R. Precision genome editing using cytosine and adenine base editors in mammalian cells. Nat Protoc 16, 1089- 1128 (2021).

4. Anzalone, A. V. et al. Search-and-replace genome editing without double-strand breaks or donor DNA. Nature 576, 149-157 (2019). 5. Anzalone, A. V., Koblan, L. W. & Liu, D. R. Genome editing with CRISPR-Cas nucleases, base editors, transposases and prime editors. Nature Biotechnology 38, 824-844 (2020).

6. Davis, J. R. et al. Efficient prime editing in mouse brain, liver and heart with dual AAVs. Nat Biotechnol 1-12 (2023) doi:10.1038/s41587-023-01758-z.

7. Bock, D. et al. In vivo prime editing of a metabolic liver disease in mice. Science Translational Medicine 14, eabl9238 (2022).

8. Nelson, J. W. et al. Engineered pegRNAs improve prime editing efficiency. Nat Biotechnol 1-9 (2021) doi: 10.1038/s41587-021-01039-7.

9. Chen, P. J. et al. Enhanced prime editing systems by manipulating cellular determinants of editing outcomes. Cell 184, 5635-5652. e29 (2021).

10. Ferreira da Silva, J. et al. Prime editing efficiency and fidelity are enhanced in the absence of mismatch repair. Nat Commun 13, 760 (2022).

11. Zetsche, B., Volz, S. E. & Zhang, F. A split-Cas9 architecture for inducible genome editing and transcription modulation. Nat Biotechnol 33, 139-142 (2015).

12. Wright, A. V. et al. Rational design of a split-Cas9 enzyme complex. Proceedings of the National Academy of Sciences 112, 2984-2989 (2015).

13. Levy, J. M. et al. Cytosine and adenine base editing of the brain, liver, retina, heart and skeletal muscle of mice via adeno-associated viruses. Nature Biomedical Engineering 4, 97-110 (2020).

14. Truong, D.-J. J. et al. Development of an intein-mediated split-Cas9 system for gene therapy. Nucleic Acids Research 43, 6450-6458 (2015).

15. Kelsell, D. P. et al. Connexin 26 mutations in hereditary non- syndromic sensorineural deafness. Nature 387, 80-83 (1997).

16. Zelante, L. et al. Connexin26 Mutations Associated with the Most Common Form of Non- Syndromic Neurosensory Autosomal Recessive Deafness (DFNB1) in Mediterraneans. Human Molecular Genetics 6, 1605-1609 (1997).

17. Azaiez, H. et al. Genomic Landscape and Mutational Signatures of Deafness-Associated Genes. The American Journal of Human Genetics 103, 484-497 (2018).

18. Azaiez, H. et al. GJB2: The spectrum of deafness-causing allele variants and their phenotype. Human Mutation 24, 305-311 (2004). 19. Kenna, M. A. et al. Audiologic Phenotype and Progression in GJB2 (Connexin 26) Hearing Loss. Archives of Otolaryngology-Head & Neck Surgery 136, 81-87 (2010).

20. Kikuchi, T., Kimura, R. S., Paul, D. L. & Adams, J. C. Gap junctions in the rat cochlea: immunohistochemical and ultrastructural analysis. Anat Embryol 191, 101-118 (1995).

21. Forge, A. et al. Gap junctions and connexin expression in the inner ear. Novartis Found Symp 219, 134-150; discussion 151-156 (1999).

22. Forge, A. et al. Gap junctions in the inner ear: Comparison of distribution patterns in different vertebrates and assessement of connexin composition in mammals. Journal of Comparative Neurology 467, 207-231 (2003).

23. Lang, H., Schulte, B. A. & Schmiedt, R. A. Effects of Chronic Furosemide Treatment and Age on Cell Division in the Adult Gerbil Inner Ear. JARO 4, 164-175 (2003).

24. Crispino, G. et al. BAAV Mediated GJB2 Gene Transfer Restores Gap Junction Coupling in Cochlear Organotypic Cultures from Deaf Cx26SoxlOCre Mice. PLOS ONE 6, e23279 (2011).

25. Yu, Q. et al. Virally expressed connexin26 restores gap junction function in the cochlea of conditional Gjb2 knockout mice. Gene Ther 21, 71-80 (2014).

26. lizuka, T. et al. Perinatal Gjb2 gene transfer rescues hearing in a mouse model of hereditary deafness. Human Molecular Genetics 24, 3651-3661 (2015).

27. Walton, R. T., Christie, K. A., Whittaker, M. N. & Kleinstiver, B. P. Unconstrained genome targeting with near-PAMless engineered CRISPR-Cas9 variants. Science 368, 290-296 (2020).

28. Kweon, J. et al. Engineered Prime Editors with PAM flexibility. Molecular Therapy (2021) doi: 10.1016/j.ymthe.2021.02.022.

29. Kleinstiver, B. P. et al. Engineered CRISPR-Cas9 nucleases with altered PAM specificities. Nature 523, 481-485 (2015).

30. Kleinstiver, B. P. et al. High-fidelity CRISPR-Cas9 nucleases with no detectable genome-wide off-target effects. Nature 529, 490-495 (2016). 31. Zheng, C. et al. A flexible split prime editor using truncated reverse transcriptase improves dual AAV delivery in mouse liver. Molecular Therapy 0, (2022).

32. Cheriyan, M., Pedamallu, C. S., Tori, K. & Perler, F. Faster Protein Splicing with the Nostoc punctiforme DnaE Intein Using Non-native Extein Residues. Journal of Biological Chemistry 288, 6202-6211 (2013).

33. Zhi, S. et al. Dual -AAV delivering split prime editor system for in vivo genome editing. Molecular Therapy 30, 283-294 (2022).

34. She, K. et al. Dual-AAV split prime editor corrects the mutation and phenotype in mice with inherited retinal degeneration. Sig Transduct Target Ther 8, 1-12 (2023).

35. Adamson, B. et al. A Multiplexed Single-Cell CRISPR Screening Platform Enables Systematic Dissection of the Unfolded Protein Response. Cell 167, 1867-1882. e21 (2016).

36. Yuan, Q. & Gao, X. Multiplex base- and prime-editing with drive-and- process CRISPR arrays. Nat Commun 13, 2771 (2022).

37. Kabadi, A. M., Ousterout, D. G., Hilton, I. B. & Gersbach, C. A. Multiplex CRISPR/Cas9-based genome engineering from a single lentiviral vector. Nucleic Acids Research 42, el47 (2014).

38. Tsai, S. Q. et al. GUIDE-seq enables genome-wide profiling of off- target cleavage by CRISPR-Cas nucleases. Nature Biotechnology 33, 187-197 (2015).

39. Malinin, N. L. et al. Defining genome-wide CRISPR-Cas genomeediting nuclease activity with GUIDE-seq. Nat Protoc 1-24 (2021) doi : 10.1038/s41596-021 -00626-x.

40. van Overbeek, M. et al. DNA Repair Profiling Reveals Nonrandom Outcomes at Cas9-Mediated Breaks. Molecular Cell 63, 633-646 (2016).

41. Shen, M. W. et al. Predictable and precise template-free CRISPR editing of pathogenic variants. Nature 563, 646-651 (2018).

42. Leenay, R. T. et al. Large dataset enables prediction of repair after CRISPR-Cas9 editing in primary T cells. Nature Biotechnology 37, 1034-1037 (2019). 43. Kim, H. K. et al. SpCas9 activity prediction by DeepSpCas9, a deep learning-based model with high generalization performance. Science Advances 5, eaax9249 (2019).

44. Schmid-Burgk, J. L. et al. Highly Parallel Profiling of Cas9 Variant Specificity. Molecular Cell 78, 794-800. e8 (2020).

45. Nakade, S. et al. Frame Editors for Precise, Template-Free Frameshifting. 2022.12.05.518807 Preprint at https://doi.org/10.1101/2022.12.05.518807 (2022).

46. Doman, J. L. et al. Phage-assisted evolution and protein engineering yield compact, efficient prime editors. Cell 186, 3983-4002. e26 (2023).

47. Chen, B. et al. Dynamic Imaging of Genomic Loci in Living Human Cells by an Optimized CRISPR/Cas System. Cell 155, 1479-1491 (2013).

48. Dang, Y. et al. Optimizing sgRNA structure to improve CRISPR-Cas9 knockout efficiency. Genome Biology 16, 280 (2015).

49. Gao, Z., Herrera-Carrillo, E. & Berkhout, B. Delineation of the Exact Transcription Termination Signal for Type 3 Polymerase III. Molecular Therapy - Nucleic Acids 10, 36-44 (2018).

50. Finston, W. I. & Champoux, J. J. RNA-primed initiation of Moloney murine leukemia virus plus strands by reverse transcriptase in vitro. Journal of Virology 51, 26-33 (1984).

51. Sutcliffe, J. G., Shinnick, T. M., Verma, I. M. & Lerner, R. A. Nucleotide sequence of Moloney leukemia virus: 3’ end reveals details of replications, analogy to bacterial transposons, and an unexpected gene. Proceedings of the National Academy of Sciences 77, 3302-3306 (1980).

52. Nishimasu, H. et al. Crystal Structure of Cas9 in Complex with Guide RNA and Target DNA. Cell 156, 935-949 (2014).

53. Huai, C. et al. Structural insights into DNA cleavage activation of CRISPR-Cas9 system. Nat Commun 8, (2017).

54. Clement, K. et al. CRISPResso2 provides accurate and rapid genome editing sequence analysis. Nature Biotechnology 37, 224-226 (2019).

55. Xu, T., Xiao, M. & Yu, L. Method for efficient soluble expression and purification of recombinant hyperactive Tn5 transposase. Protein Expression and Purification 183, 105866 (2021). OTHER EMBODIMENTS

It is to be understood that while the invention has been described in conjunction with the detailed description thereof, the foregoing description is intended to illustrate and not limit the scope of the invention, which is defined by the scope of the appended claims. Other aspects, advantages, and modifications are within the scope of the following claims.

Claims

WHAT IS CLAIMED IS:

1. An engineered split prime editor (PE) comprising a nickase SpCas9 protein (nSpCas9) fused to a reverse transcriptase (RT) domain, wherein the PE is split in the nSpCas9 domain at residues E1068; T1069; G1055; E945; 11057; A1053; T1051; E945; or N946, preferably having a sequence that is at least 95% identical to a sequence shown in Table 4.

2. An optimized engineered prime editor guide RNA (epegRNA) having an architecture comprising: a reverse transcriptase template (RTT) length of 10-18 nucleotides, preferably 14 nucleotides, comprising an intended mutation; a primer binding site (PBS) length of 9-13 nucleotides, preferably 11 or 12 nucleotides; an evopreQ or mpknot modification at the 3’ end of the pegRNA; an optional sequence comprising one or more silent mutations, optionally within 0-6 codons of the intended mutation ; an optional RNA linker comprising 1-20 nucleotides between between the 3’ of the sgRNA and the RTT; an optional RT initiation sequence of IS1, IS3, or IS 11 between PBS and evopreQ modification, preferably IS3; and an sgRNA spacer sequence and scaffold, preferably wherein the epegRNA has a sequence that is at least 95% identical to a sequence shown in Table A or 2.

3. The optimized epegRNA of claim 2, comprising a sgRNA spacer sequence of GCACGCUGCAGACGAUCCUGG (SEQ ID NO: 378), GGCACGCUGCAGACGAUCCUGG (SEQ ID NO: 379), GCUGCAGACGAUCCUGGGGG (SEQ ID NO: 380), or GCAGACGAUCCUGGGGGUG (SEQ ID NO:381), and/or optionally comprising a scaffold sequence of GUUUUAGAGCUAGAAAUAGCAAGUUAAAAUAAGGCUAGUCCGUUAUCAACUUGAA AAAGUGGCACCGAGUCGGUGC ( SEQ ID NO : 2 ) , GUUUCAGAGCUAGAAAUAGCAAGUUGAAAUAAGGCUAGUCCGUUAUCAACUUGAA AAAGUGGCACCGAGUCGGUGC ( SEQ ID NO : 3 ) , or GUUUCAGAGCUAUGCUGGAAACAGCAUAGCAAGUUUGAAUAAGGCUAGUCCGUUA

UCAACUUGAAAAAGUGGCACCGAGUCGGUGC ( SEQ ID NO : 44 ) .

4. A composition comprising ribonucleoprotein (RNP) complexes comprising the split PE of claim 1 and the optimized pegRNA of claim 2, and preferably a nicking guide RNA (ngRNA) that directs the PE to introduce a nick on the nonedited strand, optionally wherein the RNP complexes are delivered using nanoparticles, preferably lipid nanoparticles (LNPs).

5. A composition comprising nucleic acids encoding the split PE of claim 1 and the optimized pegRNA of claim 2, and optionally a ngRNA that directs the PE to introduce a nick on the non-edited strand.

6. The composition of claim 5, which comprises adenoassociated viruses (AAVs).

7. The composition of claim 6, wherein the nucleic acids comprise a first AAV with sequences encoding a first half of the split PE driven by a promoter and a second AAV with with sequences encoding a second half of the split PE driven by a promoter, preferably the same promoter as the promoter driving expression of the first half, wherein the first half is either the N terminal or C terminal half of the PE, and the second half is the other half, and optionally wherein the first and/or second AAV further comprise sequences encoding the pegRNA and/or ngRNA, preferably driven by a U6 promoter or tRNA.

8. The composition of claim 7, wherein each of the pegRNA and the ngRNA are driven by a U6 promoter or tRNA, optionally selected from human U6 promoter (hU6), bovine U6 promoter (bU6), murine U6 promoter (mU6), and hCtRNA, optionally wherein (i) the ngRNA is driven by hCtRNA promoter and pegRNA is driven by bU6 promoter, or (ii) the ngRNA is driven by hU6 promoter and a pegRNA driven by hU6 promoter.

9. The composition of any of claims 4-8, wherein the ngRNA is listed in Table B.

10. A composition comprising the optimized pegRNA of claim 2 or 3, and a nicking guide RNA (ngRNA) that directs the PE to introduce a nick on the non-edited strand, or nucleic acids encoding the optimized pegRNA and ngRNA.

11. The composition of claim 10, wherein the optimized pegRNA is listed in Table A, and the ngRNAis listed in Table B.

12. The composition of claim 11, wherein the optimized pegRNA is LM3178 or LM3180, and the ngRNA is LM1728, LM3188, or LM3194.

13. A method of treating a subject who has deafness-causing mutation in the GJB2 gene, preferably a c35delG mutation, the method comprising delivering to the inner ear of the subject the composition of any of claims 4-9.

14. A method of correcting a deafness-causing mutation in the GJB2 gene in a cell, preferably a c35delG mutation, the method comprising contacting the cell with the composition of any of claims 4-9.

15. A method of treating a subject who has deafness-causing mutation in the GJB2 gene, preferably a c35delG mutation, the method comprising delivering to the inner ear of the subject a prime editor and the composition of any of claims 10-13.

16. A method of correcting a deafness-causing mutation in the GJB2 gene in a cell, preferably a c35delG mutation, the method comprising contacting the cell with a prime editor and the composition of any of claims 10-13.

17. The method of claims 16 or 17, wherein the prime editor is PE2, PE4, PEmax, PE6, PE6a, PE6b, PE6c, PE6d, PE6e, PE6a/e, PE7, PE3max, or PE5max, or a variant thereof, optionally with truncation of the RNaseH domain of the RT.

18. The method of claim 18, wherein the prime editor is split at SpCas9 amino acid residues E1068; T1069; G1055; E945; 11057; A1053; T1051; E945; N946; K1024; Q844; or K1153.

19. A method of treating a subject who has deafness-causing mutation in the GJB2 gene, preferably a c35delG mutation, the method comprising delivering to the inner ear of the subject an insertion frame editor (iFE) and a guide RNA listed in Table 2, preferably LM248; LM250; LM254; LM258; LM260; LM262; LM264; LM266; LM268; LM270; LM274; LM276; LM278; LM280; LM282; LM284; LM286; LM288; LM290; LM292; LM294; or LM296.

0. A method of correcting a deafness-causing mutation in the GJB2 gene in a cell, preferably a c35delG mutation, the method comprising contacting the cell with an IFE and a guide RNA listed in Table 2, preferable LM248; LM250; LM254; LM258; LM260; LM262; LM264; LM266; LM268; LM270; LM274; LM276;

LM278; LM280; LM282; LM284; LM286; LM288; LM290; LM292; LM294; or LM296.