Article Search-and-replace genome editing without double-strand breaks or donor DNA

https://doi.org/10.1038/s41586-019-1711-4 Andrew V. Anzalone1,2,3, Peyton B. Randolph1,2,3, Jessie R. Davis1,2,3, Alexander A. Sousa1,2,3, Luke W. Koblan1,2,3, Jonathan M. Levy1,2,3, Peter J. Chen1,2,3, Christopher Wilson1,2,3, Received: 26 August 2019 Gregory A. Newby1,2,3, Aditya Raguram1,2,3 & David R. Liu1,2,3* Accepted: 10 October 2019

Published online: 21 October 2019 Most genetic variants that contribute to disease1 are challenging to correct efciently and without excess byproducts2–5. Here we describe prime editing, a versatile and precise genome editing method that directly writes new genetic information into a specifed DNA site using a catalytically impaired Cas9 endonuclease fused to an engineered reverse transcriptase, programmed with a prime editing guide RNA (pegRNA) that both specifes the target site and encodes the desired edit. We performed more than 175 edits in human cells, including targeted insertions, deletions, and all 12 types of point mutation, without requiring double-strand breaks or donor DNA templates. We used prime editing in human cells to correct, efciently and with few byproducts, the primary genetic causes of sickle cell disease (requiring a transversion in HBB) and Tay–Sachs disease (requiring a deletion in HEXA); to install a protective transversion in PRNP; and to insert various tags and epitopes precisely into target loci. Four human cell lines and primary post-mitotic mouse cortical neurons support prime editing with varying efciencies. Prime editing shows higher or similar efciency and fewer byproducts than homology-directed repair, has complementary strengths and weaknesses compared to base editing, and induces much lower of- target editing than Cas9 nuclease at known Cas9 of-target sites. Prime editing substantially expands the scope and capabilities of genome editing, and in principle could correct up to 89% of known genetic variants associated with human diseases.

The ability to make virtually any targeted change in the genome of any organisms, including mammals16–19, but cannot currently perform the living cell or organism is a longstanding aspiration of the life sciences. eight transversion mutations (C→A, C→G, G→C, G→T, A→C, A→T, T→A, Despite rapid advances in genome editing technologies, the majority and T→G), such as the T•A-to-A•T mutation needed to directly correct of the more than 75,000 known disease-associated genetic variants in the most common cause of sickle cell disease (HBB(E6V)). In addition, humans1 remain difficult to correct or install in most therapeutically rel- no DSB-free method has been reported to perform targeted deletions, evant cell types (Fig. 1a). Programmable nucleases such as CRISPR–Cas9 such as the removal of the four-base duplication that causes Tay-Sachs make double-strand DNA breaks (DSBs) that can disrupt genes by induc- disease (HEXA1278+TATC), or targeted insertions, such as the three-base ing mixtures of insertions and deletions (indels) at target sites2–4. DSBs, insertion required to directly correct the most common cause of cystic however, are associated with undesired outcomes, including complex fibrosis (CFTR(ΔF508)). Targeted transversions, insertions, and dele- mixtures of products, translocations5, and activation of p536,7. Moreover, tions are therefore difficult to install or correct efficiently and without the vast majority of pathogenic alleles arise from specific insertions, dele- excess byproducts in most cell types, even though they collectively tions, or base substitutions that require more precise editing technolo- account for most known pathogenic alleles (Fig. 1a). gies to correct (Fig. 1a, Supplementary Discussion). Homology-directed Here we describe the development of prime editing, a ‘search-and- repair (HDR) stimulated by DSBs8 has been widely used to install precise replace’ genome editing technology that mediates targeted insertions, DNA changes. HDR, however, relies on exogenous donor DNA repair tem- deletions, all 12 possible base-to-base conversions, and combinations plates, typically generates an excess of indels from end-joining repair of thereof in human cells without requiring DSBs or donor DNA templates. DSBs, and is inefficient in most therapeutically relevant cell types (T cells Prime editors (PEs), initially exemplified by PE1, use a reverse transcriptase and some types of stem cell being important exceptions)9,10. Whereas (RT) fused to an RNA-programmable nickase and a prime editing guide enhancing the efficiency and precision of DSB-mediated editing remains RNA (pegRNA) to copy genetic information directly from an extension the focus of promising efforts11–15, these challenges motivate the explora- on the pegRNA into the target genomic locus. PE2 uses an engineered RT tion of alternative precision genome editing strategies. to increase editing efficiencies, while PE3 nicks the non-edited strand to Base editing can efficiently install the four transition mutations (C→T, induce its replacement and further increase editing efficiency, typically G→A, A→G, and T→C) without requiring DSBs in many cell types and to 20–50% with 1–10% indel formation in human HEK293T cells. Prime

1Merkin Institute of Transformative Technologies in Healthcare, Broad Institute of Harvard and MIT, Cambridge, MA, USA. 2Department of Chemistry and Chemical Biology, Harvard University, Cambridge, MA, USA. 3Howard Hughes Medical Institute, Harvard University, Cambridge, MA, USA. *e-mail: [email protected]

Nature | Vol 576 | 5 December 2019 | 149 Article

Known human pathogenic genetic variants Prime editor (PE) and pegRNA a b Target DNA 2.0% 2.5% Reverse pegRNA nick site 3.6% 0.9% transcriptase Protospacer PAM domain 6.8% Transition point mutation 5′••• ••• 3′ Transversion point mutation pegRNA 3′ 3 Deletion ′••• •••5′ Duplication Edit position: +1 +4 +12 30% 8.2% Copy number loss Prime editing range Copy number gain for SpCas9 PEs Insertion Prime editing scope: 5′ All 4 transition point mutations 26% Insertion and deletion 20% Other All 8 transversion point mutations Insertions (1 bp to ≥ 44 bp) Cas9 nickase Deletions (1 bp to ≥ 80 bp) domain Combinations of the above Total variants = 75,122

Primer- RT template Direct polymerization of edited binding site including edit DNA from the pegRNA into c the target site 3 3′ 3’ ′ 3’ PE + 5′ L pegRNA 5′••• + DNA Nicking 3′••• Hybridization of Reverse of PAM primer-binding transcription strand 5′ site in pegRNA to PAM strand

3′ Flap 5′ Flap (contains edit) (lacks edit) 3’ Flap 5′ 5 Flap DNA ′ Edited DNA equilibration cleavage repair 5′••• ••• 3′ 5′••• ••• 3′ 5′•••••• 3′ 5′••• ••• 3′ 3′•••••• 5′ 3′••• ••• 5′ 3′••• ••• 5′ 3′•••••• 5′ Ligation

f 3′-extended pegRNA 5′-extended pegRNA d 5′-extended pegRNAs e 3′-extended pegRNAs stop codon correction stop codon correction T•A-to-A•T transversion T•A-to-A•T transversion Non-nicked 56 nt DNA substrate (51 nt) templated RT 49 nt RT products products product 42 nt (51 nt) Pre-nicked Pre-nicked substrate (34 nt) substrate (34 nt) 3′-extended pegRNA 3′-extended pegRNA Pre-nicked dsDNA frameshift correction frameshift correction Pre-nicked dsDNA + + + + + + + + – + + – 1-nt insertion 1-nt deletion dCas9 + + + + ––– Non-nicked dsDNA – + – –+ RT + + + + + + + dCas9 – – + – – sgRNA + ––– ––– Cas9 H840A nickase – – – + + RT – – + + + 5′-extended pegRNA – +++ +++ 3′-extended pegRNA ––+ + + Length of pegRNA – 8 15 22 8 15 22 RT template (nt)

Fig. 1 | Overview of prime editing and feasibility studies in vitro and in yeast labelled PAM strands, dCas9, and a commercial M-MLV RT variant (RT, cells. a, The 75,122 known pathogenic human genetic variants in ClinVar Superscript III). dCas9 was complexed with pegRNAs, then added to DNA (accessed July, 2019), classified by type. b, A prime editing complex consists of substrates along with the indicated components. After 1 h, reactions were a PE protein containing an RNA-guided DNA-nicking domain, such as Cas9 analysed by denaturing PAGE to visualize Cy5 fluorescence. e, Primer extension nickase, fused to an RT domain and complexed with a pegRNA. The PE–pegRNA assays performed as in d using 3′-extended pegRNAs pre-complexed with complex enables a variety of precise DNA edits at a wide range of positions. dCas9 or Cas9(H840A) nickase, and pre-nicked or non-nicked dsDNA spCas9, Streptococcus pyogenes Cas9. c, The PE–pegRNA complex binds the substrates. f, Yeast colonies transformed with GFP–mCherry fusion reporter target DNA and nicks the PAM-containing strand. The resulting 3′ end plasmids edited in vitro with pegRNAs, Cas9 nickase, and RT. Plasmids hybridizes to the PBS, then primes reverse transcription of new DNA containing containing nonsense or frameshift mutations between GFP and mCherry were the desired edit using the RT template of the pegRNA. Equilibration between edited with pegRNAs that restored mCherry translation via transversion, 1-bp the edited 3′ flap and the unedited 5′ flap, cellular 5′ flap cleavage and ligation, insertion, or 1-bp deletion. GFP and mCherry double-positive cells (yellow) and DNA repair results in stably edited DNA. d, In vitro primer extension assays reflect successful editing. Images in d–f are representative of n = 2 independent with 5′-extended pegRNAs, pre-nicked dsDNA substrates containing 5′-Cy5- replicates. For gel source data, see Supplementary Fig. 1. editing offers much lower off-target activity than Cas9 at known Cas9 off- guide RNAs that both specify the DNA target and contain new genetic target loci, far fewer byproducts and higher or similar efficiency compared information that replaces target DNA nucleotides. To transfer informa- to Cas9-initiated HDR, and complementary strengths and weaknesses tion from these engineered guide RNAs to target DNA, we proposed that compared to base editors. By enabling precise targeted insertions, dele- genomic DNA, nicked at the target site to expose a 3′-hydroxyl group, tions, and all 12 possible classes of point mutations without requiring DSBs could be used to prime the reverse transcription of an edit-encoding or donor DNA templates, prime editing has the potential to advance the extension on the engineered guide RNA (the pegRNA) directly into the study and correction of the vast majority of pathogenic alleles. target site (Fig. 1b, c, Supplementary Discussion). These initial steps result in a branched intermediate with two redun- dant single-stranded DNA flaps: a 5′ flap that contains the unedited DNA Prime editing strategy sequence and a 3′ flap that contains the edited sequence copied from Cas9 targets DNA using a guide RNA containing a spacer sequence that the pegRNA (Fig. 1c). Although hybridization of the perfectly comple- hybridizes to the target DNA site2–4,20,21. We envisioned the generation of mentary 5′ flap to the unedited strand is likely to be thermodynamically

150 | Nature | Vol 576 | 5 December 2019 a HEK3, 10-nt RT template EMX1, 13-nt RT template FANCF, 17-nt RT template RNF2, 11-nt RT template HEK4, 13-nt RT template HEK3, 13-nt PBS, +1 T to A +5 G to T +5 G to T +1 C to A +2 G to T 10-nt RT template 10 PE1 15 PE1 20 PE1 5 PE1 10 PE1 30 PE1 PE2 8 PE2 PE2 4 PE2 8 PE2 25 PE2 15 10 20 6 3 6 10 15 4 2 4 5 10 5 2 1 2 5 0 0 0 0 0 0 Total sequencing reads 8910 11 12 13 14 15 16 17 91011121314151617 8910 11 12 13 14 15 16 17 91011121314151617 789101112131415 with the speci ed edit (%) +1 T +1 A +1 CTT PBS length (nt) PBS length (nt) PBS length (nt) PBS length (nt) PBS length (nt) del ins ins b PE2, 13-nt PBS PE2, 13-nt PBS PE2, 13-nt PBS PE2, 15-nt PBS PE2, 11-nt PBS 25 20 HEK3, +1 T to A EMX1, +1 G to C 30 FANCF, +5 G to T 15 RNF2, +1 C to A 20 HEK4, +2 G to T 20 15 15 15 20 10 10 10 10 10 5 5 5 5

0 0 0 0 0

Total sequencing reads 10 11 12 13 14 15 16 17 18 19 20 10 11 12 13 14 15 16 17 18 19 20 10 11 12 13 14 15 16 17 18 19 20 10 11 12 13 14 15 16 17 18 19 20 7891011121314151617181920 with the speci ed edit (%) RT template length (nt) RT template length (nt) RT template length (nt) RT template length (nt) RT template length (nt)

Fig. 2 | Prime editing of genomic DNA in human cells by PE1 and PE2. a, Use of b, PE2 editing efficiencies with varying RT template lengths at five genomic an engineered M-MLV reverse transcriptase (D200N, L603W, T306K, W313F, sites in HEK293T cells. Editing efficiencies reflect sequencing reads that T330P) in PE2 substantially improves prime editing efficiencies at five genomic contain the intended edit and do not contain indels among all treated cells, sites in HEK293T cells, and small insertion and small deletion edits at HEK3. with no sorting. Mean ± s.d. of n = 3 independent biological replicates.

favoured, 5′ flaps are the preferred substrate for structure-specific a conventional sgRNA (Fig. 1d). These results demonstrate that nicked endonucleases such as FEN122, which excises 5′ flaps generated during DNA exposed by dCas9 is competent to prime reverse transcription lagging-strand DNA synthesis and long-patch base excision repair. The from a pegRNA. redundant unedited DNA may also be removed by 5′ exonucleases Next, we tested non-nicked dsDNA substrates with a Cas9(H840A) such as EXO123. nickase that nicks the PAM-containing strand2. In these reactions, We reasoned that preferential 5′ flap excision and 3′ flap ligation 5′-extended pegRNAs generated reverse transcription products inef- could drive the incorporation of the edited DNA strand, creating ficiently (Extended Data Fig. 1f), but 3′-extended pegRNAs enabled heteroduplex DNA containing one edited strand and one unedited efficient Cas9 nicking and reverse transcription (Fig. 1e). The use of strand (Fig. 1c). DNA repair to resolve the heteroduplex by copying the 3′-extended pegRNAs generated only a single apparent product, despite information in the edited strand to the complementary strand would the theoretical possibility that reverse transcription could terminate permanently install the edit (Fig. 1c). On the basis of a similar strategy anywhere within the pegRNA. DNA sequencing of reactions with Cas9 we developed to favourably resolve heteroduplex DNA during base nickase, RT, and 3′-extended pegRNAs revealed that the complete RT editing16–18, we hypothesized that nicking the non-edited DNA strand template sequence was reverse transcribed into the DNA substrate might bias DNA repair to preferentially replace the non-edited strand. (Extended Data Fig. 1g). These experiments establish that 3′-extended pegRNAs can direct Cas9 nickase and template reverse transcription in vitro. Validation in vitro and in yeast To evaluate the eukaryotic cell DNA repair outcomes of 3′ flaps First, we tested whether the 3′ end of the protospacer-adjacent motif produced by pegRNA-programmed reverse transcription in vitro, we (PAM)-containing DNA strand cleaved by the RuvC nuclease domain performed in vitro prime editing on reporter plasmids, then trans- of Cas9 was sufficiently accessible to prime reverse transcription. We formed the reaction products into yeast cells (Extended Data Fig. 2). designed pegRNAs by adding to single guide RNAs (sgRNAs) a primer We constructed reporter plasmids encoding EGFP and mCherry sepa- binding site (PBS) that allows the 3′ end of the nicked DNA strand to rated by a linker containing an in-frame stop codon, +1 frameshift, or hybridize to the pegRNA, and an RT template containing the desired −1 frameshift. When plasmids were edited in vitro with Cas9 nickase, edit (Fig. 1c). We constructed candidate pegRNAs by extending sgR- RT, and 3′-extended pegRNAs encoding a transversion that corrects NAs on either end with a PBS sequence (5–6 nucleotides (nt)) and an the premature stop codon, 37% of yeast transformants expressed both RT template (7–22 nt), and confirmed that 5′-extended pegRNAs sup- GFP and mCherry (Fig. 1f, Extended Data Fig. 2). Reactions edited with port Cas9 binding to target DNA in vitro and that both 5′-extended and 5′-extended pegRNAs yielded fewer GFP and mCherry double-positive 3′-extended pegRNAs support Cas9-mediated DNA nicking in vitro colonies (9%). Productive editing was also observed using 3′-extended and DNA cleavage in mammalian cells (Extended Data Fig. 1a–c). Next, pegRNAs that insert a single nucleotide (15%) or delete a single nucle- we tested the compatibility of these candidate pegRNAs with reverse otide (29%) to correct frameshift mutations (Fig. 1f, Extended Data transcription using pre-nicked 5′-Cy5-labelled double-stranded DNA Fig. 2). These results demonstrate that DNA repair in eukaryotic cells (dsDNA) substrates, catalytically dead Cas9 (dCas9), and a commercial can resolve 3′ DNA flaps from prime editing to incorporate precise Moloney murine leukaemia virus (M-MLV) RT variant (Extended Data transversions, insertions, and deletions. Fig. 1d). When all components were present, the labelled DNA strand was efficiently converted into longer DNA products with gel mobilities consistent with reverse transcription along the RT template (Fig. 1d, Prime editor 1 Extended Data Fig. 1d, e). Omission of dCas9 led to nick translation Encouraged by these observations, we sought to develop a prime edit- products that resulted from RT-mediated DNA polymerization on the ing system with a minimum number of components that could edit DNA template, with no pegRNA information transfer. No DNA polym- genomic DNA in mammalian cells. We transfected HEK293T cells with erization products were observed when the pegRNA was replaced by one plasmid encoding a fusion of the wild-type M-MLV RT through a

Nature | Vol 576 | 5 December 2019 | 151 Article

a Prime editor c and pegRNA RNF2 60 encoding Y 3′ Flap excision Ligation Unchanged sequence Correct edit (without indels) X′ X X′X Y X′X X′X (remains a substrate for Indels 3′ subsequent prime editing) Starting DNA 40 sequence Disfavoured repair Flap of the edited strand equilibration PE3 and PE3b X = Original DNA sequence use an sgRNA Preferential repair 20 to nick the X′ = Complement of X 5′ of the nicked non-edited strand non-edited strand Desired Y = Edited DNA sequence X Y 5′ Flap excision ′ X X′ Y X′ Y Y′ Y prime editing Y′ = Complement of Y Ligation outcome 0 Total sequencing reads with the specified edit or indels (%) PE2 PE3 PE2 PE3 PE3b pegRNA nick site PE3b +5 nick b +5 nick +41 nick PAM +41 nick 5′••• ••• 3′ Edited strand RNF2 +1 C•G to A•T (without indels) EMX1 +5 G•C to T•A (without indels) +4 A to C +4 A to G 3′••• ••• 5′ Complementary strand 60 FANCF +5 G•C to T•A (without indels) FANCF –10 –1 +1 +10 HEK3 +2 G•C to C•G (without indels) 60 50 Complementary-strand nick location HEK4 +2 G•C to T•A (without indels) Correct edit Indels (without indels) 40 40 Indels

30 20 20

10 0 Total sequencing reads with % of total sequencing reads the specified edit or indels (%) 0 with the specified edit or indels PE2 PE3 Complementary- PE2 PE3 PE3b PE3b –95 –52 –26 –57 –78 –50 –27 –17 –38 +52 +74 +41 +67 +14 +27 +53 +80 +21 +48 +26 +37 +63 +90 +7 nick strand nick location +7 nick –116 –108 None None None None None +48 nick +48 nick RNF2 EMX1 FANCF HEK3 HEK4 +5 G to T +7 A to C

Fig. 3 | PE3 and PE3b systems nick the non-edited strand to increase prime effect of complementary strand nicking on prime editing efficiency and indel editing efficiency. a, Overview of prime editing by PE3. After initial synthesis formation. ‘None’ refers to PE2 controls, which do not nick the complementary of the edited strand, 5′ flap excision leaves behind a DNA heteroduplex strand. c, Comparison of editing efficiencies with PE2, PE3, and PE3b (edit- containing one edited strand and one non-edited strand. Mismatch repair specific complementary strand nick). Editing efficiencies reflect sequencing resolves the heteroduplex to give either edited or non-edited products. reads that contain the intended edit and do not contain indels among all Nicking the non-edited strand favours repair of that strand, resulting in treated cells, with no sorting. Mean ± s.d. of n = 3 independent biological preferential generation of duplex DNA containing the desired edit. b, The replicates. flexible linker to either terminus of the Cas9(H840A) nickase, and a and T330P into M-MLV RT, hereafter referred to as M3, led to a 6.8- second plasmid encoding a pegRNA (Extended Data Fig. 3a). Initial fold average increase in transversion and insertion editing efficiency attempts led to no detectable editing. across five genomic loci in HEK293T cells compared to PE1 (Extended Extension of the PBS in the pegRNA to 8–15 bases, however, led to Data Fig. 4). detectable installation of a transversion at the HEK293 site 3 (hereafter We tested additional RT mutations that have been shown to enhance referred to as HEK3) target site, with higher efficiencies when the RT was binding to the template–PBS complex, enzyme processivity, and ther- fused to the C terminus of Cas9 nickase than when it was fused to the N mostability26. Among the 14 additional mutants analysed, addition of terminus (Extended Data Fig. 3b). These results suggest that wild-type T306K and W313F to M3 improved editing efficiency an additional M-MLV RT fused to Cas9 requires longer PBS sequences for genome 1.3-fold to 3.0-fold for six transversion or insertion edits across five editing in human cells compared to what is required in vitro using the genomic sites (Extended Data Fig. 4). This pentamutant RT incor- commercial variant of M-MLV RT supplied in trans. We designated porated into PE1 (Cas9(H840A)–M-MLV RT(D200N/L603W/T330P/ this M-MLV RT fused to the C terminus of Cas9(H840A) nickase as PE1. T306K/W313F)) is hereafter referred to as prime editor 2 (PE2). We tested the ability of PE1 to introduce transversion point mutations PE2 installs single-nucleotide transversion, insertion, and dele- at four additional genomic sites specified by the pegRNA (Fig. 2a). tion mutations with substantially higher efficiency than PE1, and is Editing efficiency at these sites was dependent on PBS length, with compatible with shorter PBS sequences, consistent with enhanced maximal editing efficiencies reaching 0.7–5.5% (Fig. 2a). Indels from PE1 engagement of transient genomic DNA–PBS complexes (Fig. 2a). On were minimal, averaging 0.2 ± 0.1% (mean ± s.d.) for the five sites under average, PE2 led to a 1.6- to 5.1-fold improvement in the efficiency of conditions that maximized each site’s editing efficiency (Extended prime editing point mutations over PE1. PE2 also performed targeted Data Fig. 3a–f). PE1 also mediated targeted insertions and deletions insertions and deletions more efficiently than PE1 (Fig. 2a, Extended with 4–17% efficiency at the HEK3 locus (Fig. 2a). These findings show Data Fig. 4d). that PE1 can directly install targeted transversions, insertions, and deletions without requiring DSBs or DNA templates. Optimization of pegRNAs We systematically probed the relationship between pegRNA structure Prime editor 2 and PE2 editing efficiency. Priming regions with lower G/C content We hypothesized that engineering the RT in PE1 might improve the generally required longer PBS sequences, consistent with the energetic efficiency of DNA synthesis during prime editing. M-MLV RT muta- requirements of hybridization of the nicked DNA strand to the pegRNA tions that increase thermostability24,25, processivity24, and DNA–RNA PBS (Fig. 2a). No PBS length or G/C content level was strictly predictive substrate affinity26, and that inactivate RNaseH activity27, have been of editing efficiency, suggesting that other factors such as DNA primer reported. We constructed 19 variants of PE1 containing a variety of RT or RT template secondary structure also influence editing activity. mutations to evaluate their editing efficiency in human cells. We recommend starting with a PBS length of about 13 nt, and testing First, we investigated M-MLV RT variants that support reverse tran- different PBS lengths during optimization, especially if the priming scription at elevated temperatures24. Introduction of D200N, L603W region deviates from about 40–60% G/C content.

152 | Nature | Vol 576 | 5 December 2019 abPE3, 10-nt RT template, PE3, 34-nt RT template, c PE3, 14-nt RT template, nicking at +90, HEK3 nicking at +90, HEK3 nicking at +41, RNF2 60 60 60 Correct edit (without indels) Correct edit (without indels) Correct edit (without indels) Indels Indels Indels 50 50 50 40 40 40 30 30 30 20 20 20 10 10 10 0 0 0 Total sequencing reads wit h Total sequencing reads wit h Total sequencing reads wit h targeted point mutation or indels (%) targeted point mutation or indels (%) targeted point mutation or indels (%) +3 A to T +8 A to T +4 A to T +4 T to A +1 T to A +1 T to A +2 T to A +4 T to C +7 C to T +1 T to C +7 T to C +1 C to T +4 T to G +5 G to T +6 G to T +1 T to G +2 G to T +2 T to G +5 G to T +3 A to C +8 A to C +4 A to C +7 C to A +1 C to A +3 A to G +8 A to G +4 A to G +5 G to A +6 G to A +2 G to A +6 G to A +5 G to C +6 G to C +7 C to G +2 G to C +1 C to G +3 G to C +14 A to T +24 T to A +12 G to C +17 G to C +20 G to C +23 C to G +26 C to G +30 C to G + 33 C to G dePE3, 15-nt RT template, PE3, 22-nt RT template, g PE3, larger deletions, nicking at +38, RUNX1 nicking at +57, VEGFA nicking at +90, HEK3 50 Correct edit (without indels) 70 Correct edit (without indels) 100 Correct edit (without indels) Indels Indels Indels 60 40 80 50 30 40 60

20 30 40 20 10 20 10 0 0

0 Total sequencing reads wit h Total sequencing reads wit h Total sequencing reads wit h targeted deletion or indels (%) targeted point mutation or indels (%) targeted point mutation or indels (%) del +1–5 +3 A to T +3 A to T +4 T to A +1 T to A +1 C to T +4 T to C +7 C to T +1 T to C +4 T to G +5 G to T +5 G to T +1 T to G +3 A to C +3 A to C +1 C to A +7 C to A +3 A to G +3 A to G +2 G to A +2 G to A +1 C to G +6 G to C +6 G to C +9 C to G del +1–10 del +1–15 del +1–25 del +1–30 del +1–80 f PE3, small insertions and small deletions h PE3, combination edits 70 Correct edit (without indels) 100 Correct edit (without indels) Indels Indels 60 80 50 40 60 30 40 20 20 10

Total sequencing reads wit h 0 Total sequencing reads wit h 0 the speci ed edit or indels (%) the speci ed edit or indels (%) +1 T ins +6 T ins +1 A ins +1 T del +1 C ins +4 A del +3 C ins +4 C ins +3 A del +4 C ins +3 A del +2 G del +6 G del +5 G del +6 G to T +5 G to T +6 G to T +5 G to T +6 G to A +5 G to C +5 G to C +1 CTT ins +1 CTT ins +1 CTT ins +1 TCA ins +1 ATG ins +1 GTA ins +4 GAT ins +2 ACA ins +1 TGC ins +4–5 TG del +3–4 GA del +1 T ins and +1 T ins and +1 T del and +1 A ins and +1–2 CT del +2 C del and and +5 G del +1–3 TGA del +2–4 GAT del and +6 G to T +1 A to C and +1 C to A and +3–5 GAG del +5–7 GGA del +2–4 GAG del +3–5 AGG del +2 AA ins and and +2 G to C +2 G to C and +4–6 GGG del

HEK3 RNF2 FANCF EMX1 RUNX1 VEGFA DNMT1 HEK3 RNF2 FANCF

Fig. 4 | Targeted insertions, deletions, and all 12 types of point mutation (c), RUNX1 (d), and VEGFA (e). f, Targeted 1- and 3-bp insertions, and 1- and 3-bp with PE3 at seven endogenous genomic loci in HEK293T cells. a, All 12 types deletions with PE3 at seven endogenous genomic loci. g, Targeted precise of single-nucleotide edit from position +1 to +8 of the HEK3 site using a 10-nt RT deletions of 5–80 bp at HEK3. h, Combination edits at three endogenous template, counting the first nucleotide following the pegRNA-induced nick as genomic loci. Editing efficiencies reflect sequencing reads that contain the position +1. b, Long-range PE3 edits at HEK3 using a 34-nt RT template. c–e, PE3- intended edit and do not contain indels among all treated cells, with no sorting. mediated transition and transversion edits at the specified positions for RNF2 Mean ± s.d. of n = 3 independent biological replicates.

Next, we systematically evaluated pegRNAs with RT templates nickase already present in PE2 and a simple sgRNA (Fig. 3a). As the 10–20 nt long at five genomic target sites using PE2 (Fig. 2b), and with RT edited DNA strand is also nicked to initiate prime editing, we tested a templates up to 31 nt at three genomic sites (Extended Data Fig. 5a–c). variety of nick locations on the non-edited strand to minimize DSBs As with PBS length, RT template length could also be varied to maxi- that lead to indels. mize prime editing efficiency, although many RT template lengths of We first tested this strategy, designated PE3, at five genomic sites in ten or more nucleotides performed comparably. As some target sites HEK293T cells using sgRNAs that induce nicks 14–116 nt away from the preferred longer RT templates (more than 15 nt; FANCF, EMX1), whereas site of the pegRNA-induced nick. In four of the five sites tested, nicking other loci preferred shorter RT templates (HEK3 and HEK293 site 4, the non-edited strand increased editing efficiency by 1.5- to 4.2-fold hereafter referred to as HEK4) (Fig. 2b), we recommend starting with compared to PE2, to as high as 55% (Fig. 3b). Although the optimal about 10–16 nt and testing shorter and longer RT templates during nicking position varied depending on the genomic site (Supplementary pegRNA optimization. Discussion), nicks positioned 3′ of the edit about 40–90 bp from the Notably, the use of RT templates that place a C adjacent to the 3′ hair- pegRNA-induced nick generally increased editing efficiency (averaging pin of the sgRNA scaffold generally resulted in lower editing efficiency 41%) without excess indel formation (6.8% average indels for the sgRNA (Extended Data Fig. 5a–c). We speculate that a C as the first nucleotide with the highest editing efficiency) (Fig. 3b). We recommend starting of the 3′ extension can disrupt guide RNA structure by pairing with G81, with non-edited strand nicks about 50 bp from the pegRNA-mediated which normally forms a pi stack with Y1356 in Cas9 and a non-canonical nick, and testing alternative nick locations if indel frequencies exceed base pair with A68 of the sgRNA28. Because many RT template lengths acceptable levels. support prime editing, we recommend designing pegRNAs so that the Nicking the non-edited strand only after resolution of the edited first base of the 3′ extension is not C. strand flap should minimize the presence of concurrent nicks, thereby minimizing formation of DSBs and indels. To achieve this goal, we designed sgRNAs with spacers that matched the edited strand, but Prime editor 3 systems not the original allele. Using this strategy, denoted PE3b, mismatches The resolution of heteroduplex DNA from PE2 containing one edited between the spacer and the unedited allele should disfavour sgRNA and one non-edited strand determines long-term editing outcomes. nicking until after editing of the PAM strand has taken place. PE3b To optimize base editing we previously used Cas9 nickase to nick the resulted in a 13-fold decrease in the average number of indels (0.74%) non-edited strand, directing DNA repair to that strand using the edited compared to PE3, without any evident decrease in editing efficiency strand as a template16–18. To apply this strategy to enhance prime edit- (Fig. 3c). When the edit lies within a second protospacer, we recom- ing, we tested nicking the non-edited strand using the Cas9(H840A) mend the PE3b approach.

Nature | Vol 576 | 5 December 2019 | 153 Article Together, these findings establish that PE3 systems improve editing the presence of multiple cytidine or adenine bases within the base efficiencies about threefold compared with PE2, albeit with a higher editing activity window16–18,29, or by the absence of a PAM positioned range of indels than PE2. When it is possible to nick the non-edited about 15 ± 2 nt from the target nucleotide16,30. We anticipated that prime strand with an sgRNA that requires editing before nicking, the PE3b sys- editing could complement base editing when bystander edits are unac- tem offers PE3-like editing levels while greatly reducing indel formation. ceptable or when the target site lacks a suitably positioned PAM. To demonstrate the targeting scope and versatility of prime editing We compared PEs and CBEs at three genomic loci that contain multi- with PE3, we performed all 24 possible single-nucleotide substitutions ple target cytosines in the canonical base editing window (protospacer across the +1 to +8 positions (counting the first base 3′ of the pegRNA- positions 4–8, counting the PAM as positions 21–23) using current-gen- induced nick as position +1) of the HEK3 target site using PE3 and pegR- eration CBEs31 without or with nickase activity (BE2max and BE4max, NAs with 10-nt RT templates (Fig. 4a). These 24 edits collectively cover respectively), or using analogous PE2 and PE3 prime editing systems. all 12 possible transition and transversion mutations, and proceeded Among the nine total cytosines within the base editing windows of the with average editing efficiencies (containing no indels) of 33 ± 7.9%, three sites, BE4max yielded 2.2-fold higher average total C•G-to-T•A with 7.5 ± 1.8% average indels. conversion than PE3 for bases in the centre of the base editing win- Notably, long-distance RT templates can also give rise to efficient dow (protospacer positions 5–7, Extended Data Fig. 6a). However, PE3 prime editing. Using PE3 with a 34-nt RT template, we installed point outperformed BE4max by 2.7-fold at cytosines positioned outside the mutations at positions +12, +14, +17, +20, +23, +24, +26, +30, and +33 in centre of the base editing window. Overall, indel frequencies for PE2 the HEK3 locus with 36 ± 8.7% average efficiency and 8.6 ± 2.0% indels were very low (averaging 0.86 ± 0.47%), and for PE3 were similar to or (Fig. 4b). Other RT templates of 30 or more nucleotides at three other modestly higher than that of BE4max (PE3: 2.5–21%; BE4max: 2.5–14%) genomic sites also supported prime editing (Extended Data Fig. 5a–c). (Extended Data Fig. 6b). As an NGG PAM on either DNA strand occurs on average every 8 bp, far For the installation of precise edits (with no bystander editing), the less than edit-to-PAM distances that support efficient prime editing, efficiency of prime editing greatly exceeded that of base editing at prime editing is not substantially constrained by the availability of a the above sites, which, like most genomic DNA sites, contain multi- nearby PAM sequence, in contrast to other precision editing meth- ple cytosines within the base editing window. BE4max generated few ods11,15,16. Given the presumed relationship between RNA secondary products containing only the single target base-pair conversion with structure and prime editing efficiency, when designing pegRNAs for no bystander edits. By contrast, prime editing at this site could be used long-range edits we recommend testing RT templates of various lengths to selectively install a C•G-to-T•A edit at any position or combination and, if necessary, sequence compositions (for example, using synony- of positions (Extended Data Fig. 6c). mous codons). We also compared nicking and non-nicking adenine base editors To further test the scope and limitations of PE3 for introducing point (ABEs) with PE3 and PE2, with similar results (Extended Data Fig. 6d–f, mutations, we tested 72 additional edits covering all possible types of Supplementary Discussion). Collectively, these results indicate that point mutation across six additional genomic target sites (Fig. 4c–e, base editing and prime editing offer complementary strengths and Extended Data Fig. 5d–f). Editing efficiency averaged 25 ± 14%, while weaknesses for making targeted transition mutations. When a single indel formation averaged 8.3 ± 7.5%. Because the pegRNA RT template target nucleotide is present within the base editing window, or when includes the PAM sequence, prime editing can induce changes in the bystander edits are acceptable, current base editors are typically more PAM sequence. In these cases, we observed higher editing efficiency efficient and generate fewer indels than prime editors. When multiple (averaging 39 ± 9.7%) and lower indel generation (averaging 5.0 ± 2.9%; cytosines or adenines are present and bystander edits are undesirable, Fig. 4, mutations at +5 or +6), potentially due to the inability of Cas9 or when PAMs that position target nucleotides for base editing are not nickase to re-bind and nick the edited strand before the repair of the available, prime editors offer substantial advantages. complementary strand. We recommend editing the PAM, in addition to other desired changes, whenever possible. Next, we performed 28 targeted small insertions and small deletions Off-target prime editing at seven genomic sites using PE3 (Fig. 4f). Targeted 1-bp and 3-bp inser- Prime editing requires target DNA–pegRNA spacer complementarity tions proceeded with an average efficiency of 32 ± 9.8% and 39 ± 16%, for the Cas9 domain to bind, target DNA–pegRNA PBS complemen- respectively. Targeted 1-bp and 3-bp deletions were also efficient, tarity to initiate pegRNA-templated reverse transcription, and target averaging 29 ± 14% and 32 ± 11% editing, respectively. Indel generation DNA–RT product complementarity for flap resolution. To test whether (beyond the target insertion or deletion) averaged 6.8 ± 5.4%. Because these three distinct DNA hybridization steps reduce off-target prime insertions and deletions between positions +1 and +6 alter the location editing compared to editing methods that require only target–guide or structure of the PAM, we speculate that insertions or deletions at RNA complementarity, we treated HEK293T cells with PE3 or PE2 and these positions are more efficient because they prevent re-engagement 16 pegRNAs that target four genomic loci, each of which has at least of the edited strand. four well-characterized Cas9 off-target sites32,33. We also treated cells We also tested PE3 for its ability to mediate larger precise deletions with Cas9 nuclease and the same 16 pegRNAs, or with Cas9 and four of 5–80 bp at the HEK3 site (Fig. 4g). We observed very high editing sgRNAs targeting the same four protospacers (Supplementary Table 1). efficiencies (52–78%) for precise 5-, 10-, 15-, 25-, and 80-bp deletions, Consistent with previous studies32, Cas9 and sgRNAs targeting HEK3, with indels averaging 11 ± 4.8%. Finally, we tested the ability of PE3 to HEK4, EMX1, and FANCF modified the top four known Cas9 off-target mediate 12 combinations of insertions, deletions, and/or point muta- loci for each sgRNA with average frequencies of 16 ± 16%, 60 ± 26%, tions across three genomic sites. These combination edits were also 48 ± 28%, and 4.3 ± 5.6%, respectively (Extended Data Fig. 6g). Cas9 very efficient, averaging 55% editing with 6.4% indels (Fig. 4h). Together, with pegRNAs modified on-target sites with similar efficiency as Cas9 the 156 distinct edits in Fig. 4 and Extended Data Fig. 5d–f establish the with sgRNAs, whereas Cas9 with pegRNAs modified off-target sites at versatility, precision, and targeting flexibility of PE3 systems. 4.4-fold lower average efficiency than Cas9 with sgRNAs. Strikingly, PE3 or PE2 with the same 16 pegRNAs containing these four target spacers resulted in detectable off-target editing at only 3 Prime editing compared with base editing out of 16 off-target sites, with only 1 of 16 showing an off-target editing Cytidine base editors (CBEs) and adenine base editors (ABEs) can install efficiency of 1% or more (Extended Data Fig. 6h). Average off-target transition mutations efficiently and with few indels16–18. The applica- prime editing for pegRNAs targeting HEK3, HEK4, EMX1, and FANCF tion of base editing can be limited by unwanted bystander edits from at the top four known Cas9 off-target sites for each protospacer was

154 | Nature | Vol 576 | 5 December 2019 ab c d g Installation and correction of the Installation and correction of the Installation of the Insertion of a His6 tag, FLAG tag, pathogenic E6V mutation pathogenic1278insTATC mutation protective G127V mutation PE3, DNMT1, +5 G to T, and extended LoxP sequence in HBB in HEK293T cells in HEXA in HEK293T cells in PRNP in HEK293T cells mouse primary cortical neurons in HEK3 in HEK293T cells Correct edit Correct edit (without indels) Correct edit (without indels) Correct edit 80 Correct edit Indels Indels (without indels) 40 (without indels) (without indels) Indels Indels Indels 70 50 30 60 70 20 60 40 60 10 50 10 30 50 40 8 40 40 30 20 6 30 20 4 10 20 20 10 2

10 Total sequencing reads with 0 0 Total sequencing reads with the speci ed edit or indels (% ) 0 Total sequencing reads with the speci ed edit or indels (% ) Total sequencing reads with WT HBB E6V E6V HBB the speci ed edit or indels (% ) WT HEXA HEXA HEXA

the speci ed edit or indels (% ) 0

Total sequencing reads with 0 PE3 PE3 to E6V HBB to WT to +TATC +TATC +TATC to the speci ed edit or indels (% ) PRNP G127V Cas9 Cas9 His6 FLAG LoxP (install) to WT with silent (install) to WT WT with silent (install) sorted sorted unsorted (correct) (correct) (correct) (correct) unsorted Untreated e PE3 in HEK293T, K562, U2OS, and HeLa cells f Cas9-initiated HDR in HEK293T, K562, U2OS, and HeLa cells 100 100 Correct edit (without indels) Indels 80 80 Correct edit (without indels) Indels 60 60

40 40

20 20 Total sequencing reads with Total sequencing reads with

0 the speci ed edit or indels (% ) 0 the speci ed edit or indels (% ) RNF2 RNF2 RNF2 RNF2 RNF2 RNF2 RNF2 RNF2 HEK3 HEK3 HEK3 HEK3 HEK3 HEK3 HEK3 HEK3 HEK3 HEK3 HEK3 HEK3 HEK3 HEK3 HEK3 HEK3 PRN P PRN P (install) (install) (correct ) (correct ) HBB E6V HBB E6V HBB E6V HBB E6V +1 T to G +1 T to G +1 T to G +1 T to G +1 T to G +1 T to G +1 T to G +1 T to G +6 G to T +6 G to T +1 C to G +1 C to G +1 C to G +1 C to G +1 C to G +1 C to G +1 C to G +1 C to G +1 CTT ins +1 CTT ins +1 CTT ins +1 CTT ins +1 CTT ins +1 CTT ins +1 CTT ins +1 CTT ins HEK293T cells K562 cells U2OS cells HeLa cells HEK293T cells K562 cellsU2OS cells HeLa cells

Fig. 5 | Prime editing of pathogenic mutations, prime editing in primary Installation of a G•C-to-T•A transversion in DNMT1 of mouse primary cortical mouse cortical neurons, and comparison of prime editing and HDR in four neurons using a split-intein PE3 lentivirus system (see Methods). Sorted values human cell lines. a, Installation (via T•A-to-A•T transversion) and correction reflect editing or indels from GFP-positive nuclei, while unsorted values are (via A•T-to-T•A transversion) of the pathogenic E6V-coding mutation in HBB in from all nuclei. e, f, PE3 editing and indels (e) or Cas9-initiated HDR editing and HEK293T cells. Correction either to wild-type HBB, or to HBB containing a PAM- indels (f) at endogenous genomic loci in HEK293T, K562, U2OS, and HeLa cells. disrupting silent mutation, is shown. b, Installation (via 4-bp insertion) and g, Targeted insertion of a His6 tag (18 bp), Flag epitope tag (24 bp), or extended correction (via 4-bp deletion) of the pathogenic HEXA1278+TATC allele in HEK293T loxP site (44 bp) in HEK293T cells by PE3. Editing efficiencies reflect cells. Correction either to wild-type HEXA, or to HEXA containing a PAM- sequencing reads that contain the intended edit and do not contain indels disrupting silent mutation, is shown. c, Installation of the protective G127V- among all treated cells, with no sorting, except where specified in e. Mean ± s.d. coding variant in PRNP in HEK293T cells via G•C-to-T•A transversion. d, of n = 3 independent biological replicates.

<0.1%, <2.2 ± 5.2%, <0.1%, and <0.13 ± 0.11%, respectively (Extended Data PE2 minimally perturbed the transcriptome relative to Cas9 nickase or Fig. 6h). Notably, at the HEK4 off-target 3 site that was edited by Cas9 a control lacking active RT (Supplementary Discussion). with pegRNA1 at 97% efficiency, PE2 with pegRNA1 resulted in only 0.2% off-target editing despite sharing the same pegRNA, demonstrating how the two additional hybridization events required for prime edit- Prime editing pathogenic mutations ing can greatly reduce off-target modification. Together, these results We tested the ability of PE3 to directly install or correct in human cells suggest that prime editing induces much lower off-target editing than transversion, insertion, and deletion mutations that cause genetic Cas9 at known Cas9 off-target sites. diseases. Sickle cell disease is caused by a A•T-to-T•A transversion muta- Reverse transcription of 3′-extended pegRNAs in principle can tion in HBB, resulting in an E6V mutation in β-globin (Supplementary proceed into the guide RNA scaffold, resulting in scaffold sequence Discussion). We used PE3 to install this HBB mutation into HEK293T insertion that contributes to indels at the target locus. We analysed cells with 44% efficiency and 4.8% indels (Fig. 5a) and isolated from 66 PE3 editing experiments at four loci in HEK293T cells and observed a single prime editing experiment six HEK293T cell lines that were 1.7 ± 1.5% average total insertion of any number of pegRNA scaffold homozygous (triploid) for the mutated HBB allele (Supplementary nucleotides (Extended Data Fig. 7). We speculate that inaccessibility Note 1). To correct the mutant HBB allele to wild-type HBB, we treated of the guide RNA scaffold to reverse transcription due to Cas9 domain HEK293T cells homozygous for mutant HBB with PE3 and a pegRNA binding, and cellular excision of the mismatched 3′ end of 3′ flaps that programmed to directly revert the HBB mutation to wild-type HBB. extend into the pegRNA scaffold, minimize products that incorporate All 14 tested pegRNAs mediated efficient correction of mutant HBB to pegRNA scaffold nucleotides. wild-type HBB (26–52% efficiency), and indel levels averaged 2.8 ± 0.70% The presence of endogenous human RTs from retroelements34 and (Extended Data Fig. 9a). Introduction of a PAM-modifying silent muta- telomerase suggests that RT activity is not inherently toxic to human tion improved editing efficiency and product purity to 58% correction cells. Indeed, we observed no differences in the viability of HEK293T with 1.4% indels (Fig. 5a). cells expressing dCas9, Cas9(H840A) nickase, PE2, or PE2 with R110S The most common mutation that causes Tay-Sachs disease is a 4-bp and K103L mutations (PE2-dRT) that inactivate the RT and abolish insertion in HEXA (HEXA1278+TATC). We used PE3 to install this 4-bp inser- prime editing35 (Extended Data Fig. 8a, b). To evaluate changes in the tion into HEXA with 31% efficiency and 0.8% indels (Fig. 5b), and iso- cellular transcriptome that result from prime editing, we performed lated two HEK293T cell lines that were homozygous for HEXA1278+TATC RNA sequencing (RNA-seq) on HEK293T cells expressing PE2, PE2- (Supplementary Note 1). We used these cells to test 43 pegRNAs and dRT, or Cas9(H840A) nickase together with a PRNP-targeting or HEXA- three nicking sgRNAs with PE3 or PE3b systems for correction of the targeting pegRNA (Extended Data Fig. 8c–k), and observed that active pathogenic insertion in HEXA (Extended Data Fig. 9b). Nineteen of the

Nature | Vol 576 | 5 December 2019 | 155 Article 43 pegRNAs tested resulted in editing with an efficiency of 20% or more. example, we used PE3 in HEK293T cells to precisely insert into HEK3 a

Correction to wild-type HEXA with the best pegRNA proceeded with 33% His6 tag (18 bp, 65% efficiency), a Flag epitope tag (24 bp, 18% efficiency), efficiency and 0.32% indels using PE3b (Fig. 5b, Extended Data Fig. 9b). and an extended Cre recombinase loxP site (44 bp, 23% efficiency) with Finally, we used PE3 to install a protective G•C-to-T•A transversion 3.0–5.9% indels (Fig. 5g). We anticipate that the ability to efficiently and into PRNP (resulting in PRNP(G127V)) into HEK293T cells, introducing precisely insert new DNA sequences into target sites in living cells will a mutant allele that confers resistance to prion disease in humans36 and enable many biotechnological and therapeutic applications. mice37 (Supplementary Discussion). We evaluated four pegRNAs and Collectively, the prime editing experiments described here performed three nicking sgRNAs. The most effective pegRNA with PE3 resulted 19 insertions up to 44 bp, 23 deletions up to 80 bp, 119 point mutations in 53% installation of G127V, with 1.7% indels (Fig. 5c). Together, these including 83 transversions, and 18 combination edits at 12 endogenous results establish the ability of prime editing in human cells to install loci in the human and mouse genomes at locations ranging from 3 bp or correct transversion, insertion, or deletion mutations that cause upstream to 29 bp downstream of a PAM without making explicit DSBs. or confer resistance to disease efficiently, and with few byproducts. These results establish prime editing as a remarkably versatile genome editing method. Because 85–99% of insertions, deletions, indels, and duplications in ClinVar are 30 bp in length or smaller (Extended Data Other cell lines and primary neurons Fig. 11), in principle prime editing could correct up to about 89% of the Next, we tested prime editing at endogenous sites in three additional 75,122 pathogenic human genetic variants in ClinVar (Fig. 1a). human cell lines (Extended Data Fig. 10a, Supplementary Discussion). Prime editing offers many possible choices of pegRNA-induced

In K562 cells, PE3 achieved three transversion edits and a His6 tag inser- nick locations, sgRNA-induced second nick locations, PBS lengths, RT tion with 15–30% editing efficiency and 0.85–2.2% indels (Extended template lengths, and which strand to edit first. This flexibility, which Data Fig. 10a). In U2OS cells, we installed transversion mutations, as contrasts with more limited options typically available for other preci- 11,15,16 well as a 3-bp insertion and His6 tag insertion, with 7.9–22% editing sion editing methods , allows editing efficiency, product purity, efficiency and 0.13–2.2% indels (Extended Data Fig. 10a). Finally, in HeLa DNA specificity, and other parameters to be optimized to suit a given cells we performed a 3-bp insertion with 12% average efficiency and application (Extended Data Fig. 9). 1.3% indels (Extended Data Fig. 10a). Collectively, these data indicate Much additional research is needed to further understand and that cell lines other than HEK293T support prime editing, although improve prime editing in a broad range of cell types and organisms, editing efficiencies vary by cell type and are generally less efficient to assess off-target prime editing in a genome-wide manner, and to than in HEK293T cells. Editing:indel ratios remained favourable in all further characterize the extent to which prime editors might affect human cell lines tested. cells. Interfacing prime editing with additional in vitro and in vivo To determine whether prime editing is possible in post-mitotic, ter- delivery strategies is essential for exploring the potential of prime minally differentiated primary cells, we transduced primary cortical editing to enable applications, including the study and treatment of neurons from E18.5 mice with a PE3 lentiviral delivery system in which genetic diseases. By enabling precise targeted transitions, transver- PE2 protein components were expressed from the neuron-specific sions, insertions, and deletions in the genomes of mammalian cells synapsin promoter38 along with a GFP marker (see Methods). Nuclei without requiring DSBs, donor DNA templates, or HDR, however, prime were isolated two weeks after transduction and sequenced directly, editing provides a new search-and-replace capability that substantially or sorted for GFP expression before sequencing. We observed 7.1% expands the scope of genome editing. average prime editing of DNMT1 with 0.58% average indels in sorted cortical neuron nuclei (Fig. 5d). Cas9 nuclease in the same lentivirus system resulted in 31% average indels among sorted nuclei (Fig. 5d). Online content These data indicate that post-mitotic, terminally differentiated primary Any methods, additional references, Nature Research reporting sum- cells can support prime editing. maries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author con- tributions and competing interests; and statements of data and code Prime editing compared with HDR availability are available at https://doi.org/10.1038/s41586-019-1711-4. Finally, we compared the performance of PE3 with that of optimized Cas9-initiated HDR11,14 in mitotic cell lines that support HDR14. We 1. Landrum, M. J. et al. ClinVar: public archive of interpretations of clinically relevant treated HEK293T, HeLa, K562 and U2OS cells with Cas9 nuclease, an variants. Nucleic Acids Res. 44, D862–D868 (2016). 2. Jinek, M. et al. A programmable dual-RNA-guided DNA endonuclease in adaptive sgRNA, and a single-stranded DNA (ssDNA) donor template designed to bacterial immunity. Science 337, 816–821 (2012). install a variety of transversion and insertion edits (Fig. 5e, f, Extended 3. Cong, L. et al. Multiplex genome engineering using CRISPR/Cas systems. Science 339, 819–823 (2013). Data Fig. 10). Cas9-initiated HDR in all cases successfully installed the 4. Mali, P. et al. RNA-guided human genome engineering via Cas9. Science 339, 823–826 desired edit, but with far higher levels of indel byproducts than with PE3, (2013). as expected given that Cas9 induces DSBs. In HEK293T cells, the ratio 5. Kosicki, M., Tomberg, K. & Bradley, A. Repair of double-strand breaks induced by CRISPR– Cas9 leads to large deletions and complex rearrangements. Nat. Biotechnol. 36, 765–771 of editing to indels for installation or correction of the allele encoding (2018). HBB(E6V) or installation of the allele encoding PRNP(G127V) was on 6. Haapaniemi, E., Botla, S., Persson, J., Schmierer, B. & Taipale, J. CRISPR–Cas9 genome average 270-fold higher for PE3 than for Cas9-initiated HDR. editing induces a p53-mediated DNA damage response. Nat. Med. 24, 927–930 (2018). Comparisons between PE3 and HDR in human cell lines other than 7. Ihry, R. J. et al. p53 inhibits CRISPR–Cas9 engineering in human pluripotent stem cells. HEK293T showed similar results, although with lower PE3 editing effi- Nat. Med. 24, 939–946 (2018). ciencies (Fig. 5e, f, Supplementary Discussion). Collectively, these data 8. Rouet, P., Smih, F. & Jasin, M. Expression of a site-specific endonuclease stimulates homologous recombination in mammalian cells. Proc. Natl Acad. Sci. USA 91, 6064– indicate that HDR typically results in similar or lower editing efficiencies 6068 (1994). than PE3 with far more indels in four tested human cell lines (Extended 9. Chapman, J. R., Taylor, M. R. G. & Boulton, S. J. Playing the end game: DNA double-strand Data Fig. 10). break repair pathway choice. Mol. Cell 47, 497–510 (2012). 10. Cox, D. B. T., Platt, R. J. & Zhang, F. Therapeutic genome editing: prospects and challenges. Nat. Med. 21, 121–131 (2015). 11. Paquet, D. et al. Efficient introduction of specific homozygous and heterozygous Discussion and future directions mutations using CRISPR/Cas9. Nature 533, 125–129 (2016). 12. Chu, V. T. et al. Increasing the efficiency of homology-directed repair for CRISPR– The ability to insert arbitrary DNA sequences with single-nucleotide Cas9-induced precise gene editing in mammalian cells. Nat. Biotechnol. 33, 543–548 precision is an especially promising capability of prime editing. For (2015).

156 | Nature | Vol 576 | 5 December 2019 13. Maruyama, T. et al. Increasing the efficiency of precise genome editing with CRISPR– 27. Kotewicz, M. L., Sampson, C. M., D’Alessio, J. M. & Gerard, G. F. Isolation of cloned Cas9 by inhibition of nonhomologous end joining. Nat. Biotechnol. 33, 538–542 Moloney murine leukemia virus reverse transcriptase lacking ribonuclease H activity. (2015). Nucleic Acids Res. 16, 265–277 (1988). 14. Rees, H. A., Yeh, W.-H. & Liu, D. R. Development of hRad51-Cas9 nickase fusions that 28. Nishimasu, H. et al. Crystal structure of Cas9 in complex with guide RNA and target DNA. mediate HDR without double-stranded breaks. Nat. Commun. 10, 2212 (2019). Cell 156, 935–949 (2014). 15. Shen, M. W. et al. Predictable and precise template-free CRISPR editing of pathogenic 29. Thuronyi, B. W. et al. Continuous evolution of base editors with expanded target variants. Nature 563, 646–651 (2018). compatibility and improved activity. Nat. Biotechnol. 37, 1070–1079 (2019). 16. Rees, H. A. & Liu, D. R. Base editing: precision chemistry on the genome and 30. Kim, Y. B. et al. Increasing the genome-targeting scope and precision of base editing transcriptome of living cells. Nat. Rev. Genet. 19, 770–788 (2018). with engineered Cas9-cytidine deaminase fusions. Nat. Biotechnol. 35, 371–376 (2017). 17. Komor, A. C., Kim, Y. B., Packer, M. S., Zuris, J. A. & Liu, D. R. Programmable editing of a 31. Koblan, L. W. et al. Improving cytidine and adenine base editors by expression target base in genomic DNA without double-stranded DNA cleavage. Nature 533, 420– optimization and ancestral reconstruction. Nat. Biotechnol. 36, 843–846 (2018). 424 (2016). 32. Tsai, S. Q. et al. GUIDE-seq enables genome-wide profiling of off-target cleavage by 18. Gaudelli, N. M. et al. Programmable base editing of A•T to G•C in genomic DNA without CRISPR-Cas nucleases. Nat. Biotechnol. 33, 187–197 (2015). DNA cleavage. Nature 551, 464–471 (2017). 33. Kleinstiver, B. P. et al. High-fidelity CRISPR-Cas9 nucleases with no detectable genome- 19. Gao, X. et al. Treatment of autosomal dominant hearing loss by in vivo delivery of genome wide off-target effects. Nature 529, 490–495 (2016). editing agents. Nature 553, 217–221 (2018). 34. Bannert, N. & Kurth, R. Retroelements and the human genome: new perspectives on an 20. Marraffini, L. A. & Sontheimer, E. J. CRISPR interference limits horizontal gene transfer in old relation. Proc. Natl Acad. Sci. USA 101 (Suppl. 2), 14572–14579 (2004). staphylococci by targeting DNA. Science 322, 1843–1845 (2008). 35. Halvas, E. K., Svarovskaia, E. S. & Pathak, V. K. Role of murine leukemia virus reverse 21. Barrangou, R. et al. CRISPR provides acquired resistance against viruses in prokaryotes. transcriptase deoxyribonucleoside triphosphate-binding site in retroviral replication and Science 315, 1709–1712 (2007). in vivo fidelity. J. Virol. 74, 10349–10358 (2000). 22. Liu, Y., Kao, H.-I. & Bambara, R. A. Flap endonuclease 1: a central component of DNA 36. Mead, S. et al. A novel protective prion protein variant that colocalizes with kuru metabolism. Annu. Rev. Biochem. 73, 589–615 (2004). exposure. N. Engl. J. Med. 361, 2056–2065 (2009). 23. Keijzers, G., Bohr, V. A. & Rasmussen, L. J. Human exonuclease 1 (EXO1) activity 37. Asante, E. A. et al. A naturally occurring variant of the human prion protein completely characterization and its function on flap structures. Biosci. Rep. 35, e00206 (2015). prevents prion disease. Nature 522, 478–481 (2015). 24. Baranauskas, A. et al. Generation and characterization of new highly thermostable and 38. Kügler, S., Kilic, E. & Bähr, M. Human synapsin 1 gene promoter confers highly neuron- processive M-MuLV reverse transcriptase variants. Protein Eng. Des. Sel. 25, 657–668 specific long-term transgene expression from an adenoviral vector in the adult rat brain (2012). depending on the transduced area. Gene Ther. 10, 337–347 (2003). 25. Gerard, G. F. et al. The role of template-primer in protection of reverse transcriptase from thermal inactivation. Nucleic Acids Res. 30, 3118–3129 (2002). Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in 26. Arezi, B. & Hogrefe, H. Novel mutations in Moloney murine leukemia virus reverse published maps and institutional affiliations. transcriptase increase thermostability through tighter binding to template-primer. Nucleic Acids Res. 37, 473–481 (2009). © The Author(s), under exclusive licence to Springer Nature Limited 2019

Nature | Vol 576 | 5 December 2019 | 157 Article Methods products were diluted tenfold with binding buffer (40% saturated aqueous guanidinium chloride and 60% isopropanol) and purified General methods by QIAquick spin column (Qiagen), then used as templates for primer DNA amplification was conducted by PCR using Phusion U Green Mul- extension by Klenow fragment (New England Biolabs) using primer tiplex PCR Master Mix (ThermoFisher Scientific) or Q5 Hot Start High- AVA134 (A-tailed products) or AVA135 (G-tailed products) (Supple- Fidelity 2× Master Mix (New England BioLabs) unless otherwise noted. mentary Table 2). Extensions were amplified by PCR for 10 cycles using DNA oligonucleotides, including Cy5-labelled DNA oligonucleotides, primers AVA110 and AVA122, then sequenced with AVA037 using the dCas9 protein, and Cas9(H840A) protein were obtained from Integrated Sanger method (Supplementary Table 2). DNA Technologies. Yeast reporter plasmids were derived from previ- ously described plasmids39 and cloned by the Gibson assembly method. Yeast fluorescent reporter assays All mammalian editor plasmids used in this work were assembled using Dual fluorescent reporter plasmids containing an in-frame stop codon, the USER cloning method as previously described40. Plasmids expressing a +1 frameshift, or a −1 frameshift were subjected to 5′-extended pegRNA sgRNAs were constructed by ligation of annealed oligonucleotides into or 3′-extended pegRNA prime editing reactions in vitro as described BsmBI-digested acceptor vector (Addgene plasmid no. 65777). Plasmids above using 100 ng of plasmid substrate. Following incubation at 37 °C expressing pegRNAs were constructed by Gibson assembly or Golden for 1 h, the reactions were diluted with water and plasmid DNA was Gate assembly using a custom acceptor plasmid (see Supplementary precipitated with 0.3 M sodium acetate and 70% ethanol. Resuspended Note 3). Sequences of sgRNA and pegRNA constructs used in this work DNA was transformed into Saccharomyces cerevisiae by electroporation are listed in Supplementary Tables 2 and 3. All vectors for mammalian as previously described42 and plated on synthetic complete medium cell experiments were purified using Plasmid Plus Midiprep kits (Qiagen) without leucine (SC(glucose), L−). GFP and mCherry fluorescence or PureYield plasmid miniprep kits (Promega), which include endotoxin signals were visualized from colonies with the Typhoon FLA 7000 removal steps. All experiments using live animals were approved by biomolecular imager. the Broad Institute Institutional and Animal Care and Use Committees. Wild-type C57BL/6 mice were obtained from Charles River (#027). No General mammalian cell culture conditions statistical methods were used to predetermine sample size. The experi- HEK293T (ATCC CRL-3216), U2OS (ATTC HTB-96), K562 (CCL-243), and ments were not randomized, and investigators were not blinded to HeLa (CCL-2) cells were purchased from ATCC and cultured and pas- allocation during experiments and outcome assessment. saged in Dulbecco’s modified Eagle’s medium (DMEM) plus GlutaMAX (ThermoFisher Scientific), McCoy’s 5A medium (Gibco), RPMI medium In vitro biochemical assays 1640 plus GlutaMAX (Gibco), or Eagle’s minimal essential medium pegRNAs and sgRNAs were transcribed in vitro using the HiScribe T7 (EMEM, ATCC), respectively, each supplemented with 10% (v/v) fetal in vitro transcription kit (New England Biolabs) from PCR-amplified bovine serum (Gibco, qualified) and 1× penicillin streptomycin (Corn- templates containing a T7 promoter sequence. RNA was purified by ing). All cell types were incubated, maintained, and cultured at 37 °C denaturing urea PAGE and quality-confirmed by an analytical gel before with 5% CO2. Cell lines were authenticated by their respective suppliers use. 5′-Cy5-labelled DNA duplex substrates were annealed using two and tested negative for mycoplasma. oligonucleotides (Cy5-AVA024 and AVA025; 1:1.1 ratio) for the non- nicked substrate or three oligonucleotides (Cy5-AVA023, AVA025 and HEK293T tissue culture transfection protocol and genomic DNA AVA026; 1:1.1:1.1) for the pre-nicked substrate by heating to 95 °C for preparation 3 min followed by slowly cooling to room temperature (Supplementary HEK293T cells were seeded on 48-well poly-d-lysine coated plates Table 2). Cas9 cleavage and reverse transcription reactions were carried (Corning). Between 16 and 24 h after seeding, cells were transfected out in 1× cleavage buffer41 supplemented with dNTPs (20 mM HEPES-K, at approximately 60% confluency with 1 µl lipofectamine 2000 (Thermo pH 7.5; 100 mM KCl; 5% glycerol; 0.2 mM EDTA, pH 8.0; 3 mM MgCl2; Fisher Scientific) according to the manufacturer’s protocols and 750 ng 0.5 mM dNTP mix; 5 mM DTT). dCas9 or Cas9(H840A) (5 µM final) and PE plasmid, 250 ng pegRNA plasmid, and 83 ng sgRNA plasmid (for the sgRNA or pegRNA (5 µM final) were pre-incubated at room tempera- PE3 and PE3b). Unless otherwise stated, cells were cultured for 3 days ture in a 5-µl reaction mixture for 10 min before the addition of 0.5 μl of following transfection, after which the medium was removed, the 4 μM duplex DNA substrate (400 nM final), followed by the addition of cells were washed with 1× PBS solution (Thermo Fisher Scientific), 0.2 μl of Superscript III reverse transcriptase (ThermoFisher Scientific), and genomic DNA was extracted by the addition of 150 µl of freshly an undisclosed M-MLV RT variant, when applicable. Reactions were prepared lysis buffer (10 mM Tris-HCl, pH 7.5; 0.05% SDS; 25 µg/ml carried out at 37 °C for 1 h, then diluted to a volume of 10 µl with water, proteinase K (ThermoFisher Scientific)) directly into each well of the treated with 0.2 µl of proteinase K solution (20 mg/ml, ThermoFisher tissue culture plate. The genomic DNA mixture was incubated at 37 °C Scientific), and incubated at room temperature for 30 min. Following for 1–2 h, followed by an 80 °C enzyme inactivation step for 30 min. heat inactivation at 95 °C for 10 min, reaction products were combined Primers used for mammalian cell genomic DNA amplification are listed with 2× formamide gel loading buffer (90% formamide; 10% glycerol; in Supplementary Table 4. For HDR experiments in HEK293T cells, 0.01% bromophenol blue), denatured at 95 °C for 5 min, and separated 231 ng Cas9 nuclease-expression plasmid, 69 ng sgRNA-expression by denaturing urea PAGE gel (15% TBE-urea, 55 °C, 200 V). DNA products plasmid and 50 ng (1.51 pmol) of 100-nt ssDNA donor template (PAGE- were visualized by Cy5 fluorescence signal using a Typhoon FLA 7000 purified; Integrated DNA Technologies) was lipofected using 1.4 µl biomolecular imager. lipofectamine 2000 (ThermoFisher) per well. Genomic DNA from all Electrophoretic mobility shift assays were carried out in 1× binding HDR experiments was purified using the Agencourt DNAdvance Kit buffer (1× cleavage buffer with 10 µg/ml heparin) using pre-incubated (Beckman Coulter), according to the manufacturer’s protocol. dCas9–sgRNA or dCas9–pegRNA complexes (concentration between 5 nM and 1 µM final) and Cy5-labelled duplex DNA (Cy5-AVA024 and High-throughput DNA sequencing of genomic DNA samples AVA025; 20 nM final). After 15 min of incubation at 37 °C, the samples were Genomic sites of interest were amplified from genomic DNA samples analysed by native PAGE gel (10% TBE) and imaged for Cy5 fluorescence. and sequenced on an Illumina MiSeq as previously described with the For DNA sequencing of reverse transcription products, fluorescent following modifications17,18. In brief, amplification primers containing bands were excised and purified from urea PAGE gels, then 3′ tailed Illumina forward and reverse adapters (Supplementary Table 4) were with terminal transferase (TdT; New England Biolabs) in the presence used for a first round of PCR (PCR 1) to amplify the genomic region of of dGTP or dATP according to the manufacturer’s protocol. Tailed DNA interest. PCR 1 reactions (25 µl) were performed with 0.5 µM of each forward and reverse primer, 1 µl genomic DNA extract and 12.5 µl Phu- confluency. For base editing with CBE or ABE constructs, cells were sion U Green Multiplex PCR Master Mix. PCR reactions were carried out transfected with 750 ng base editor plasmid, 250 ng sgRNA expression as follows: 98 °C for 2 min, then 30 cycles of [98 °C for 10 s, 61 °C for plasmid, and 1 µl of lipofectamine 2000 (Thermo Fisher Scientific). 20 s, and 72 °C for 30 s], followed by a final 72 °C extension for 2 min. PE transfections were performed as described above. Genomic DNA Unique Illumina barcoding primer pairs were added to each sample in extraction for PE and BE was performed as described above. a secondary PCR reaction (PCR 2). Specifically, 25 µl of a given PCR 2 reaction contained 0.5 µM of each unique forward and reverse Illumina Determination of PE3 activity at known Cas9 off-target sites barcoding primer pair, 1 µl unpurified PCR 1 reaction mixture, and 12.5 µl To evaluate PE3 off-target editing activity at known Cas9 off-target sites, Phusion U Green Multiplex PCR 2× Master Mix. The barcoding PCR 2 genomic DNA extracted from HEK293T cells 3 days after transfection reactions were carried out as follows: 98 °C for 2 min, then 12 cycles with PE3 was used as template for PCR amplification of 16 previously of [98 °C for 10 s, 61 °C for 20 s, and 72 °C for 30 s], followed by a final reported Cas9 off-target genomic sites32,33 (the top four off-target sites 72 °C extension for 2 min. PCR products were evaluated analytically by each for the HEK3, EMX1, FANCF, and HEK4 spacers; primer sequences electrophoresis in a 1.5% agarose gel. PCR 2 products (pooled by com- are listed in Supplementary Table 4). These genomic DNA samples were mon amplicons) were purified by electrophoresis with a 1.5% agarose identical to those used for quantifying on-target PE3 editing activi- gel using a QIAquick Gel Extraction Kit (Qiagen), eluting with 40 µl ties shown in Fig. 4 or Extended Data Fig. 5d, e; pegRNA and nicking water. DNA concentration was measured by fluorometric quantification sgRNA sequences are listed in Supplementary Table 3. Following PCR (Qubit, ThermoFisher Scientific) or qPCR (KAPA Library Quantification amplification of off-target sites, amplicons were sequenced on the Kit-Illumina, KAPA Biosystems) and sequenced on an Illumina MiSeq Illumina MiSeq platform as described above (see ‘High-throughput instrument according to the manufacturer’s protocols. DNA sequencing of genomic DNA samples’ section). To determine the Sequencing reads were demultiplexed using MiSeq Reporter (Illu- on-target and off-target editing activity of Cas9 nuclease, Cas9(H840A) mina). Alignment of amplicon sequences to a reference sequence was nickase, dCas9, and PE2-dRT, we transfected HEK293T cells with 750 ng performed using CRISPResso243. For all prime editing yield quantifica- editor plasmid (Cas9 nuclease, Cas9(H840A) nickase, dCas9, or PE2- tion, prime editing efficiency was calculated as: percentage of (number dRT), 250 ng pegRNA or sgRNA plasmid, and 1 μl lipofectamine 2000. of reads with the desired edit that do not contain indels)/(number of Genomic DNA was isolated from cells 3 days after transfection as total reads). For quantification of point mutation editing, CRISPResso2 described above. On-target and off-target genomic loci were ampli- was run in standard mode with “discard_indel_reads” on. Prime edit- fied by PCR using the primer sequences in Supplementary Table 4 and ing for installation of point mutations was then explicitly calculated sequenced on an Illumina MiSeq. as: (frequency of specified point mutation in non-discarded reads) × High-throughput sequencing (HTS) data analysis was performed (number of non-discarded reads)/(total reads). For insertion or dele- using CRISPResso243. The editing efficiencies of Cas9 nuclease, Cas9 tion edits, CRISPResso2 was run in HDR mode using the desired allele H840A nickase, and dCas9 were quantified as the percentage of total as the expected allele (e flag), and with “discard_indel_reads” on. Editing sequencing reads containing indels. For quantification of PE3 and yield was calculated as: (number of HDR-aligned reads)/(total reads). PE3-dRT off-targets, aligned sequencing reads were examined for For all experiments, indel yields were calculated as: (number of indel- point mutations, insertions, or deletions that were consistent with containing reads)/(total reads). the anticipated product of pegRNA reverse transcription initiated at the Cas9 nick site. Single nucleotide variations occurring at <0.1% over- Nucleofection of U2OS, K562, and HeLa cells all frequency among total reads within a sample were excluded from Nucleofection was used for transfection in all experiments using K562, analysis. For reads containing single nucleotide variations that both HeLa, and U2OS cells. For PE conditions in these cell types, 800 ng occurred at frequencies ≥0.1% and were partially consistent with the prime editor expression plasmid, 200 ng pegRNA expression plasmid, pegRNA-encoded edit, t-tests (unpaired, one-tailed, α = 0.5) were used and 83 ng nicking sgRNA expression plasmid was nucleofected in a to determine whether the variants occurred at significantly higher lev- final volume of 20 μl in a 16-well nucleocuvette strip (Lonza). For HDR els compared to samples treated with pegRNAs that contained the same conditions in these three cell types, 350 ng Cas9 nuclease expression spacer but encoded different edits. To avoid differences in sequencing plasmid, 150 ng sgRNA expression plasmid and 200 pmol (6.6 μg) 100- errors, comparisons were made between samples that were sequenced nt ssDNA donor template (PAGE-purified; Integrated DNA Technolo- simultaneously within the same MiSeq run. Variants that did not meet gies) was nucleofected in a final volume of 20 µl per sample in a 16-well the criteria of P > 0.05 were excluded. Off-target PE3 editing activity Nucleocuvette strip (Lonza). K562 cells were nucleofected using the was then calculated as the percentage of total sequencing reads that SF Cell Line 4D-Nucleofector X Kit (Lonza) with 5 × 105 cells per sample met the above criteria. (program FF-120), according to the manufacturer’s protocol. U2OS cells were nucleofected using the SE Cell Line 4D-Nucleofector X Kit Generation of a HEK293T cell line containing HBB(E6V) using (Lonza) with 3–4 × 105 cells per sample (program DN-100), according Cas9-initiated HDR to the manufacturer’s protocol. HeLa cells were nucleofected using the HEK293T cells were seeded in a 48-well plate and transfected at SE Cell Line 4D-Nucleofector X Kit (Lonza) with 2 × 105 cells per sample approximately 60% confluency with 1.5 µl lipofectamine 2000, 300 ng (program CN-114), according to the manufacturer’s protocol. Cells Cas9(D10A) nickase plasmid, 100 ng sgRNA plasmid, and 200 ng were harvested 72 h after nucleofection for genomic DNA extraction. 100-mer ssDNA donor template (Supplementary Table 5). Three days after transfection, the medium was exchanged for fresh medium. Four Genomic DNA extraction for HDR experiments days after transfection, cells were dissociated using 30 µl TrypLE solu- Genomic DNA from all HDR comparison experiments in HEK293T, tion and suspended in 1.5 ml medium. Single cells were isolated into HEK293T HBB(E6V), K562, U2OS, and HeLa cells was purified using individual wells of two 96-well plates by fluorescence-activated cell the Agencourt DNAdvance Kit (Beckman Coulter), according to the sorting (FACS) (Beckman-Coulter Astrios). See Supplementary Note manufacturer’s protocol. 1 for representative FACS sorting examples. Cells were expanded for 14 days before genomic DNA sequencing as described above. Of the Comparison between PE2, PE3, BE2, BE4max, ABEdmax, and isolated clonal populations, none was found to be homozygous for the ABEmax HBB allele encoding the E6V mutation, so a second round of editing by HEK293T cells were seeded on 48-well poly-d-lysine coated plates lipofection, sorting, and outgrowth was repeated in a partially edited (Corning). After 16–24 h, cells were transfected at approximately 60% cell line to yield a cell line homozygous for the E6V-encoding allele. Article Generation of a HEK293T cell line containing HBB(E6V) using PE3 was resuspended in Opti-MEM (Thermo Fisher Scientific, Waltham, HEK293T cells (2.5 × 104) were seeded on 48-well poly-d-lysine coated MA, USA) using 1% of the original medium volume. Resuspended pellet plates (Corning). Between 16 and 24 h after seeding, cells were trans- was flash-frozen and stored at −80 °C until use. fected at approximately 70% confluency with 1 µl lipofectamine 2000 (Thermo Fisher Scientific) according to the manufacturer’s protocols Mouse primary cortical neuron dissection and culture and 750 ng PE2-P2A-GFP plasmid, 250 ng pegRNA plasmid, and 83 ng E18.5 dissociated cortical cultures were taken from timed-pregnant sgRNA plasmid. After 3 days, cells were washed with 1× PBS (Gibco) and C57BL/6 mice (Charles River). Embryos were removed from pregnant dissociated using TrypLE Express (Gibco). Cells were then diluted with mice after euthanasia by CO2 followed by decapitation. Cortical caps DMEM plus GlutaMax (Thermo Fisher Scientific) supplemented with were dissected in ice-cold Hibernate-E supplemented with penicil- 10% (v/v) FBS (Gibco) and passed through a 35-µm cell strainer (Corn- lin/streptomycin (Life Technologies). Following a rinse with ice-cold ing) before sorting. Flow cytometry was carried out on a LE-MA900 cell Hibernate-E, tissue was digested at 37 °C for 8 min in papain/DNase sorter (Sony). Cells were treated with 3 nM DAPI (BioLegend) 15 min (Worthington/Sigma). Tissue was triturated in NBActiv4 (BrainBits) sup- before sorting. After gating for doublet exclusion, single DAPI-negative plemented with DNase. Cells were counted and plated in 24-well plates at cells with GFP fluorescence above that of a GFP-negative control cell 100,000 cells per well. Half of the medium was changed twice per week. population were sorted into 96-well flat-bottom cell culture plates (Corning) filled with pre-chilled DMEM with GlutaMax supplemented Prime editing in primary neurons and nucleus isolation with 10% FBS. See Supplementary Note 1 for representative FACS sort- At days in vitro (DIV) 1, 15 µl lentivirus was added at a 10:10:1 ratio of ing examples and allele tables. Cells were cultured for 10 days before N-terminal:C-terminal:nicking sgRNA. At DIV 14, neuronal nuclei were genomic DNA extraction and characterization by HTS, as described isolated using the EZ-PREP buffer (Sigma D8938) following the manu- above. A total of six clonal cell lines were identified that are homozygous facturer’s protocol. All steps were performed on ice or at 4 °C. Medium for the E6V-encoding mutation in HBB. was removed from dissociated cultures, and cultures were washed with ice-cold PBS. PBS was aspirated and replaced with 200 µl EZ-PREP Generation of a HEK293T cell line containing the HEXA1278+TATC solution. Following a 5-min incubation on ice, EZ-PREP was pipetted insertion using PE3 across the surface of the well to dislodge remaining cells. The sam- HEK293T cells containing the HEXA1278+TATC allele were generated fol- ple was centrifuged at 500g for 5 min, and the supernatant removed. lowing the protocol described above for creation of the HBB(E6V) cell Samples were washed with 200 µl EZ-PREP and centrifuged again at line; pegRNA and sgRNA sequences are listed in Supplementary Table 3 500g for 5 min. Samples were resuspended with gentle pipetting in under the Fig. 5 subheading. After transfection and sorting, cells were 200 µl ice-cold Nuclei Suspension Buffer (NSB) consisting of 100 µg/ cultured for 10 days before genomic DNA was extracted and character- ml BSA and 3.33 µM Vybrant DyeCycle Ruby (Thermo Fisher) in 1×PBS, ized by HTS, as described above. We recovered two heterozygous cell then centrifuged at 500g for 5 min. The supernatant was removed and lines that contained 50% HEXA1278+TATC alleles and two homozygous cell nuclei were resuspended in 100 µl NSB and sorted into 100 µl Agencourt lines containing 100% HEXA1278+TATC alleles. DNAdvance lysis buffer using a MoFlo Astrios (Beckman Coulter) at the Broad Institute flow cytometry facility. Genomic DNA was purified Cell viability assays according to the manufacturer’s Agencourt DNAdvance instructions. HEK293T cells were seeded in 48-well plates and transfected at approxi- mately 70% confluency with 750 ng editor plasmid (PE2, PE2(R110S/ RNA-seq and data analysis K103L), Cas9(H840A) nickase, or dCas9), 250 ng HEK3-targeting HEK293T cells were co-transfected with PRNP-targeting or HEXA-target- pegRNA plasmid, and 1 μl lipofectamine 2000, as described above. ing pegRNAs and PE2, PE2-dRT, or Cas9(H840A) nickase. Seventy-two Cell viability was measured every 24 h post-transfection for 3 days using hours after transfection, total RNA was harvested from cells using TRI- the CellTiter-Glo 2.0 assay (Promega) according to the manufacturer’s zol reagent (Thermo Fisher) and purified with RNeasy Mini kit (Qiagen) protocol. Luminescence was measured in 96-well flat-bottomed poly- including on-column DNaseI treatment. Ribosomes were depleted from styrene microplates (Corning) using a M1000 Pro microplate reader total RNA using the rRNA removal protocol of the TruSeq Stranded (Tecan) with a 1-s integration time. Total RNA library prep kit (Illumina) and subsequently washed with RNAClean XP beads (Beckman Coulter). Sequencing libraries were Lentivirus production prepared using ribo-depleted RNA on a SMARTer PrepX Apollo NGS Lentivirus was produced as previously described44. T-75 flasks of rapidly library prep system (Takara) following the manufacturer’s protocol. dividing HEK293T cells (ATCC; Manassas, VA, USA) were transfected The resulting libraries were visualized on a 2200 TapeStation (Agilent with lentivirus production helper plasmids pVSV-G and psPAX2 in com- Technologies), normalized using a Qubit dsDNA HS assay (Thermo bination with modified lentiCRISPRv2 genomes carrying intein-split Fisher), and sequenced on a NextSeq 550 using high output v2 flow PE2 editor using FuGENE HD (Promega, Madison, WI, USA) according cell (Illumina) as 75-bp paired-end reads. Fastq files were generated to the manufacturer's protocol. Four split-intein editor constructs with bcl2fastq2 version 2.20 and trimmed using TrimGalore version were designed: 1) a viral genome encoding a U6-pegRNA expression 0.6.2 (https://github.com/FelixKrueger/TrimGalore) to remove low- cassette and the N-terminal portion (1–573) of Cas9(H840A) nickase quality bases, unpaired sequences, and adaptor sequences. Trimmed fused to the Npu N-intein, a self-cleaving P2A peptide, and GFP-KASH; reads were aligned to a Homo sapiens genome assembly GRCh38 with a 2) a viral genome encoding the Npu C-intein fused to the C-terminal custom Cas9(H840A) gene entry using RSEM version 1.3.146. The limma- remainder of PE2; 3) a viral genome encoding the Npu C-intein fused voom47 package was used to normalize gene expression levels and to the C-terminal remainder of Cas9 for the Cas9 control; and 4) a nick- perform differential expression analysis with batch effect correction. ing sgRNA for DNMT1 (derived from Addgene plasmid no. 52963). The Differentially expressed genes were called with FDR-corrected P < 0.05 split-intein45 mediates trans splicing to join the two halves of PE2 or and fold change > 2 cutoffs, and results were visualized in R. Cas9, while the P2A GFP-KASH enables co-translational production of a nuclear membrane-localized GFP. After 48 h, supernatant was collected, ClinVar analysis centrifuged at 500g for 5 min to remove cellular debris, and filtered The ClinVar variant summary was downloaded from NCBI (accessed July using a 0.45-µm filter. Filtered supernatant was concentrated using the 15, 2019), and the information contained therein was used for all down- PEG-it Virus Precipitation Solution (System Biosciences, Palo Alto, CA, stream analysis. The list of all reported variants was filtered by allele USA) according to the manufacturer’s directions. The resulting pellet ID in order to remove duplicates and by clinical significance in order to restrict the analysis to pathogenic variants. The list of pathogenic 39. Anzalone, A. V., Lin, A. J., Zairis, S., Rabadan, R. & Cornish, V. W. Reprogramming variants was filtered sequentially by variant type in order to calculate eukaryotic translation with ligand-responsive synthetic RNA switches. Nat. Methods 13, 453–458 (2016). the fraction of pathogenic variants that are insertions, deletions, and 40. Badran, A. H. et al. Continuous evolution of Bacillus thuringiensis toxins overcomes insect so on. Single nucleotide variants (SNVs) were separated into two cat- resistance. Nature 533, 58–63 (2016). egories (transitions and transversions) on the basis of the reported 41. Anders, C. & Jinek, M. in Methods in Enzymology (eds. Doudna, J. A. & Sontheimer, E. J.) 546, 1–20 (Academic, 2014). reference and alternate alleles. SNVs that did not report reference or 42. Pirakitikulr, N., Ostrov, N., Peralta-Yahya, P. & Cornish, V. W. PCRless library mutagenesis alternate alleles were excluded from the analysis. via oligonucleotide recombination in yeast. Protein Sci. 19, 2336–2346 (2010). The lengths of reported insertions, deletions, and duplications were 43. Clement, K. et al. CRISPResso2 provides accurate and rapid genome editing sequence analysis. Nat. Biotechnol. 37, 224–226 (2019). calculated using reference/alternate alleles, variant start/stop posi- 44. Levy, J. M. & Nicoll, R. A. Membrane-associated guanylate kinase dynamics reveal regional tions, or appropriate identifying information in the variant name. Vari- and developmental specificity of synapse stability. J. Physiol. (Lond.) 595, 1699–1709 (2017). ants that did not report any of the above information were excluded 45. Zettler, J., Schütz, V. & Mootz, H. D. The naturally split Npu DnaE intein exhibits an extraordinarily high rate in the protein trans-splicing reaction. FEBS Lett. 583, 909–914 from the analysis. The lengths of reported indels (single variants that (2009). include both insertions and deletions relative to the reference genome) 46. Li, B. & Dewey, C. N. RSEM: accurate transcript quantification from RNA-seq data with or were calculated by determining the number of mismatches or gaps without a reference genome. BMC Bioinformatics 12, 323 (2011). 47. Ritchie, M. E. et al. limma powers differential expression analyses for RNA-sequencing in the best pairwise alignment between the reference and alternate and microarray studies. Nucleic Acids Res. 43, e47 (2015). alleles. Frequency distributions of variant lengths were calculated using GraphPad Prism 8. Acknowledgements We thank J. M. Madison for neuron cell culture advice. This work was supported by the Merkin Institute of Transformative Technologies in Healthcare, US NIH grants U01AI142756, RM1HG009490, R01EB022376, and R35GM118062, and the HHMI. A.V.A. Reporting summary acknowledges a Jane Coffin Childs postdoctoral fellowship. P.B.R. and A.R. acknowledge NIH Further information on research design is available in the Nature T32 GM095450. A.A.S. acknowledges NIH T32 GM007726. P.J.C. and A.R. acknowledge NSF Research Reporting Summary linked to this paper. graduate fellowships. C.W. acknowledges a Damon Runyon Cancer Research Foundation fellowship (DRG-2343-18). G.A.N acknowledges a Helen Hay Whitney postdoctoral fellowship.

Author contributions A.V.A. designed the research, performed experiments, analysed data, Data availability and wrote the manuscript. P.B.R., J.R.D., A.A.S., and G.A.N. performed human cell experiments and analysed data. L.W.K. and J.M.L. performed neuron experiments. P.J.C. and C.W. performed High-throughput sequencing data have been deposited to the NCBI and analysed RNA-seq experiments. A.R. analysed ClinVar data. D.R.L designed and supervised Sequence Read Archive database under accession PRJNA565979. the research and wrote the manuscript. Plasmids encoding PE1, PE2 (same as PE3), and pegRNA expression Competing interests Authors through the Broad Institute have filed patent applications on vectors are available from Addgene. Previously described plasmids prime editing. D.R.L. is a consultant and co-founder of Prime Medicine, Beam Therapeutics, expressing sgRNAs are also available from Addgene, such as Addgene Pairwise Plants, and Editas Medicine, companies that use genome editing. plasmid no. 65777. Additional information Supplementary information is available for this paper at https://doi.org/10.1038/s41586-019- 1711-4. Code availability Correspondence and requests for materials should be addressed to D.R.L. Peer review information Nature thanks Guangping Gao, Randall Platt and Fyodor Urnov for The script used to quantify pegRNA scaffold insertion is provided their contribution to the peer review of this work. as Supplementary Note 4. Reprints and permissions information is available at http://www.nature.com/reprints.

Article

ab c HEK3

Electrophoretic mobility shift assays s 100 HEK293T cells 5’-extended pegRNAs 1.0 dsDNA substrate 80 sgRNA pegRNA 1 60 pegRNA 2 bound pegRNA 3 40

pegRNA 4 with indels 0.5 pegRNA 5 20 Nicked products

% of total sequencing read 0 d d Fraction of DNA dsDNA substrate + + + 0.0 Cas9 H840A nickase – + + sgRN A pegRNA 110 100 1000 5’-extended pegRNA – + – pegRNA 5’-extende [dCas9:gRNA] (nM) 3’-extended pegRNA – – + 3’-extende d Pre-nicked Cy5-labeled DNA substrate Cy5 sgRNA or ReReverseverse 5’ Protospacer PAM 5’-extended pegRNA or 3’-extended trtranscriptasanscripta e 3’-extended pegRNA pegRNA 3’ 5’ 3’ + 3’ dCas9 or Cas9 H840A 3’ 5’ Cy5 Cy5 5’5 5’5 OR OR 5’ 5’ M-MLV RT 3’ 3’ Non-nicked Cy5-labeled DNA substrate dNTPs Cy5 37 ˚C, 1 hr Protospacer PAM Separate primer extension 5’-extended products by denaturing Cas9 pegRNA 5’ 3’ urea PAGE and visualize 3’ 5’ Cy5 fluorescence

e 5’-extended pegRNAs f 5’-extended pegRNAs

Non-nicked DNA substrate RT RT products products

Pre-nicked DNA substrate Cleavage products

Nicked ssDNA marker + – – – – – – – Nicked ssDNA marker – + – – – – – – Pre-nicked DNA – + + + + + + + dsDNA substrate + – + + + + + + dCas9 – – + + + + + + Cas9 H840A nickase – – + + + + + + M-MLV RT – – + + + + + + M-MLV RT – – – + + + + + sgRNA – – + – – – – – sgRNA – – + – – – – – 5’-extended pegRNA – – – + + + + + 5’-extended pegRNA – – – + + + + + RT template length (nt) – – – 7 7 8 15 22 RT template length (nt) – – – 7 7 8 15 22 Linker ––– ABBBB Linker – – – ABBBB g gRNA RT template scaffold homopolymer tail 3’-extended pegRNA

RT product poly-G tailing

Band excision, DNA elution, TdT tailing, primer extension, Sanger sequencing poly-A tailing

Extended Data Fig. 1 | See next page for caption. Extended Data Fig. 1 | In vitro prime editing validation studies with extension reactions using 5′-extended pegRNAs, pre-nicked DNA substrates, fluorescently labelled DNA substrates. a, Electrophoretic mobility shift and dCas9 lead to substantial conversion to RT products. f, Primer extension assays with dCas9, 5′-extended pegRNAs and 5′-Cy5-labelled DNA substrates. reactions using 5′-extended pegRNAs as in b with non-nicked DNA substrate pegRNAs 1–5 contain a 15-nt linker sequence (linker A for pegRNA 1, linker B for and Cas9(H840A) nickase. Product yields are greatly reduced by comparison pegRNAs 2–5) between the spacer and the PBS, a 5-nt PBS sequence, and RT to pre-nicked substrate. g, An in vitro primer extension reaction using a templates of 7 nt (pegRNAs 1 and 2), 8 nt (pegRNA 3), 15 nt (pegRNA 4), and 22 nt 3′-pegRNA generates a single apparent product by denaturing urea PAGE. The (pegRNA 5). pegRNAs are those used in e and f; full sequences are listed RT product band was excised, eluted from the gel, then subjected to in Supplementary Table 2. b, In vitro nicking assays of Cas9(H840A) using homopolymer tailing with terminal transferase (TdT) using either dGTP or 5′-extended and 3′-extended pegRNAs. Data in a, b are representative of n = 2 dATP. Tailed products were extended using poly-T or poly-C primers, and the independent replicates. c, Cas9-mediated indel formation in HEK293T cells at resulting DNA was sequenced. Sanger traces indicate that three nucleotides HEK3 using 5′-extended and 3′-extended pegRNAs. Mean ± s.d. of n = 3 derived from the pegRNA scaffold were reverse-transcribed (added as the final independent biological replicates. d, Overview of prime editing in vitro 3′ nucleotides to the DNA product). Note that pegRNA scaffold insertion is biochemical assays. 5′-Cy5-labelled pre-nicked and non-nicked dsDNA much rarer in mammalian cell prime editing experiments than in vitro substrates were tested. sgRNAs, 5′-extended pegRNAs, or 3′-extended (Extended Data Fig. 6), potentially owing to the inability of the tethered RT to pegRNAs were pre-complexed with dCas9 or Cas9(H840A) nickase, then access the Cas9-bound guide RNA scaffold, and/or cellular excision of combined with dsDNA substrate, Superscript III M-MLV RT, and dNTPs. mismatched 3′ ends of 3′ flaps containing pegRNA scaffold sequences. Data in Reactions were allowed to proceed at 37 °C for 1 h before separation by e–g are representative of n = 2 independent replicates. For gel source data, denaturing urea PAGE and visualization by Cy5 fluorescence. e, Primer see Supplementary Fig. 1. Article a b Stop codon, +1 frameshift, or 1. Cas9 H840A, No GFP only Unedited -1 frameshift pegRNA, dNTPs, editing GFP-STOP-mCherry M-MLV RT negative control

GFP mCherry 2. Transform into yeast cells Editing GFP–mCherry Pre-edited fusion protein GFP–mCherry positive control Start c d 3’-Extended pegRNA stop codon correction (T•A to A•T) 5’-Extended pegRNA stop codon correction (T•A to A•T)

mCherry GFP merge mCherry GFP merge ef 3’-Extended pegRNA +1 frameshift correction (1-bp deletion) 3’-Extended pegRNA -1 frameshift correction (1-bp insertion)

mCherry GFP merge mCherry GFP merge g Sequence of additional GFP and pegRNA-encoded T-to-A edit mCherry double-positive colonies

Unedited plasmid transformant (GFP-only colony)

Edited plasmid transformant (GFP and mCherry double-positive colony)

Extended Data Fig. 2 | Cellular repair in yeast of 3′ DNA flaps from in vitro negative control, top), or containing no stop codon or frameshift between GFP prime editing reactions. a, Dual fluorescent protein reporter plasmids and mCherry (pre-edited positive control, bottom). c–f, Visualization of contain GFP and mCherry open reading frames separated by a target site mCherry and GFP fluorescence from yeast colonies transformed with in vitro encoding an in-frame stop codon, a +1 frameshift, or a −1 frameshift. Prime prime editing reaction products. c, d, Stop codon correction via T•A-to-A•T editing reactions were carried out in vitro with Cas9(H840A) nickase, pegRNA, transversion using a 3′-extended pegRNA (c) or a 5′-extended pegRNA (d). e, +1 dNTPs, and M-MLV RT, then transformed into yeast. Colonies that contain frameshift correction via a 1-bp deletion using a 3′-extended pegRNA. f, −1 unedited plasmids produce GFP but not mCherry. Yeast colonies containing frameshift correction via a 1-bp insertion using a 3′-extended pegRNA. edited plasmids produce both GFP and mCherry as a fusion protein. b, Overlay g, Sanger DNA sequencing traces from plasmids isolated from GFP-only of GFP and mCherry fluorescence for yeast colonies transformed with reporter colonies in b and GFP and mCherry double-positive colonies in c. Data in b–g plasmids containing a stop codon between GFP and mCherry (unedited are representative of n = 2 independent replicates. Extended Data Fig. 3 | Prime editing of genomic DNA in human cells by PE1. sequences ranging from 9 to 17 nt. e, G•C-to-T•A transversion editing efficiency a, pegRNAs contain a spacer sequence, an sgRNA scaffold, and a 3′ extension and indel generation by PE1 at the +5 position of FANCF using pegRNAs containing an RT template (purple), which contains the edited base(s) (red), containing 17-nt RT templates and PBS sequences ranging from 8 to 17 nt. and a primer-binding site (PBS, green). The primer-binding site hybridizes to f, C•G-to-A•T transversion editing efficiency and indel generation by PE1 at the the nicked target DNA strand. The RT template is homologous to the DNA +1 position of RNF2 using pegRNAs containing 11-nt RT templates and PBS sequence downstream of the nick, with the exception of the encoded edited sequences ranging from 9 to 17 nt. g, G•C-to-T•A transversion editing efficiency base(s). b, Installation of a T•A-to-A•T transversion at the HEK3 site in HEK293T and indel generation by PE1 at the +2 position of HEK4 using pegRNAs cells using Cas9(H840A) nickase fused to wild-type M-MLV RT (PE1) and containing 13-nt RT templates and PBS sequences ranging from 7 to 15 nt. pegRNAs with varying PBS lengths. c, T•A-to-A•T transversion editing h, PE1-mediated +1 T deletion, +1 A insertion, and +1 CTT insertion at the HEK3 efficiency and indel generation by PE1 at the +1 position of HEK3 using pegRNAs site using a 13-nt PBS and a 10-nt RT template. Sequences of pegRNAs are as in containing 10-nt RT templates and PBS sequences ranging from 8 to 17 nt. Fig. 2a (Supplementary Table 3). Editing efficiencies reflect sequencing reads d, G•C-to-T•A transversion editing efficiency and indel generation by PE1 at the that contain the intended edit and do not contain indels among all treated cells, +5 position of EMX1 using pegRNAs containing 13-nt RT templates and PBS with no sorting. Mean ± s.d. of n = 3 independent biological replicates. Article

a b 20 PE1, HEK3 PE1 Cas9 H840A–M-MLV RT (wt) PE1-M1 Cas9 H840A–M-MLV RT (D200N) 15 PE1-M2 Cas9 H840A–M-MLV RT (D200N+L603W) Correct edit (w/o indels) PE1-M3 Cas9 H840A–M-MLV RT (D200N+L603W+T330P) Indels PE1-M6 Cas9 H840A–M-MLV RT (D200N+L603W+T330P+D524G+E562Q+D583N) 10 PE1-M15 Cas9 H840A–M-MLV RT (P51L+S67R+E67K+T197A+H204R+E302K+F309N+W313F+T330P+L435G+N454K+D524G+D583N+H594Q+D653N) PE1-M3inv M-MLV RT (D200N+L603W+T330P)–Cas9 H840A 5

PE2 Cas9 H840A–M-MLV RT (D200N+L603W+T330P+T306K+W313F) % of total sequencing reads wit h the specified edit or indels 0 A ins T de l +1 c d e +1 +1 CT T ins 20 HEK3 +2 G to C 6 HEK3 FLAG-tag insertion (24 bp) 4 RNF2 +1 C to A

15 A• T 3 4

10 2

2 5 1 targeted FLAG insertion target C•G converted to target G•C converted to C• G % of total sequencing reads with % of total sequencing reads with % of total sequencing reads with

0 0 0 v v v PE 1 PE 1 PE 1 PE1-M 6 PE1-M 6 PE1-M 6 PE1-M1 PE1-M2 PE1-M3 PE1-M1 PE1-M2 PE1-M3 PE1-M1 PE1-M2 PE1-M3 PE1-M1 5 PE1-M1 5 PE1-M1 5 PE1-M3 + PE1-M3 + PE1-M3 + PE1-M3+ PE1-M3+ PE1-M3+ untreate d untreate d untreate d PE1-M3in PE1-M3in PE1-M3in E607K+L139P E607K+L139P E607K+L139P PE1-M3+E69K PE1-M3+E69K PE1-M3+E69K E302R+W313F E302R+W313F E302R+W313F PE1-M3+L139P PE1-M3+L139P PE1-M3+L139P PE1-M3+T306K PE1-M3+T306K PE1-M3+T306K PE1-M3+L435G PE1-M3+L435G PE1-M3+L435G PE1-M3+E607K PE1-M3+E607K PE1-M3+E607K PE1-M3+N454K PE1-M3+N454K PE1-M3+N454K PE1-M3+E302R PE1-M3+E302R PE1-M3+E302R PE1-M3+W313F PE1-M3+W313F PE1-M3+W313F f PE1-M15+D200N ghPE1-M15+D200N PE1-M15+D200N 15 EMX1 +1 G to C 10 HBB +2 T to A 6 FANCF +1 G to C

8 G A• T

10 4 6 converted to 4

5 T• A 2 arget t arget G•C converted to C• G 2 arget G•C converted to C• t t % of total sequencing reads with % of total sequencing reads with % of total sequencing reads with

0 0 0 v v v PE 1 PE 1 PE 1 PE1-M 6 PE1-M 6 PE1-M 6 PE1-M1 PE1-M2 PE1-M3 PE1-M1 PE1-M2 PE1-M3 PE1-M1 PE1-M2 PE1-M3 PE1-M1 5 PE1-M1 5 PE1-M1 5 PE1-M3 + PE1-M3 + PE1-M3 + PE1-M3 + PE1-M3 + PE1-M3 + untreate d untreate d untreate d PE1-M3in PE1-M3in PE1-M3in E607K+L139P E607K+L139P E607K+L139P PE1-M3+E69K PE1-M3+E69K PE1-M3+E69K E302R+W313F E302R+W313F E302R+W313F PE1-M3+L139P PE1-M3+L139P PE1-M3+L139P PE1-M3+T306K PE1-M3+T306K PE1-M3+T306K PE1-M3+L435G PE1-M3+L435G PE1-M3+L435G PE1-M3+E607K PE1-M3+E607K PE1-M3+E607K PE1-M3+E302R PE1-M3+E302R PE1-M3+E302R PE1-M3+N454K PE1-M3+N454K PE1-M3+N454K PE1-M3+W313F PE1-M3+W313F PE1-M3+W313F PE1-M15+D200N PE1-M15+D200N PE1-M15+D200N ij kl mn HEK3 +2 G to C HEK3 FLAG-tag insertion (24 bp) RNF2 +1 C to A EMX1 +1 G to C HBB +2 T to A FANCF +1 G to C 20 6 8 15 8 5

G 4 n A•T 15 6 A• T 6 4 10 3 10 4 4 converted to 2

2 5 T• A 5 2 2 targeted FLAG insertio arget G•C converted to C• target

target C•G converted to 1 t target G•C converted to C• G target G•C converted to C• G % of total sequencing reads with % of total sequencing reads with % of total sequencing reads with % of total sequencing reads with % of total sequencing reads with % of total sequencing reads with

0 0 0 0 0 PE2 PE2 PE2 PE2 PE2 PE2 PE1-M3 PE1-M3 PE1-M3 PE1-M3 PE1-M3 PE1-M3 PE1-M3 + PE1-M3 + PE1-M3 + PE1-M3 + PE1-M3 + PE1-M3 + Untreated Untreated Untreated Untreated Untreated Untreated E302R+W313F E302R+W313F E302R+W313F E302R+W313F E302R+W313F E302R+W313F PE1-M3+T306K PE1-M3+T306K PE1-M3+T306K PE1-M3+T306K PE1-M3+T306K PE1-M3+T306K opqr s PE2, HEK3 +1 T to A PE2, EMX1 +5 G to T PE2, FANCF +5 G to T PE2, RNF2 +1 C to A PE2, HEK4 +2 G to T

10 15 20 4 10 Indels Correct edit Correct edit Correct edit Correct edit (w/o indels) Correct edit (w/o indels) (w/o indels) (w/o indels) (w/o indels) Indels 8 Indels Indels 8 Indels 15 3 10 6 6 10 2 4 4 5 5 1 the specified edit or indels the specified edit or indels the specified edit or indels the specified edit or indels 2 the specified edit or indels 2 % of total sequencing reads % of total sequencing reads % of total sequencing reads % of total sequencing reads % of total sequencing reads with with with with with

0 0 0 0 819 011121314151617 91011121314151617 819 011121314151617 91011121314151617 789101112131415 PBS length (nucleotides) PBS length (nucleotides) PBS length (nucleotides) PBS length (nucleotides) PBS length (nucleotides)

Extended Data Fig. 4 | Evaluation of M-MLV RT variants for prime editing. a, prime editor constructs containing M-MLV variants for their ability to install Abbreviations for prime editor variants used in this figure. b, Targeted the edits shown in c–h in a second round of independent experiments. o–s, PE2 insertion and deletion edits with PE1 at the HEK3 locus. c–h, Comparison of 18 editing efficiency at five genomic loci with varying PBS lengths. o, +1 T•A-to-A•T prime editor constructs containing M-MLV RT variants for their ability to install at HEK3. p, +5 G•C-to-T•A at EMX1. q, +5 G•C-to-T•A at FANCF. r, +1 C•G-to-A•T at a +2 G•C-to-C•G transversion edit at HEK3 (c), a 24-bp Flag insertion at the +1 RNF2. s, +2 G•C-to-T•A at HEK4. Editing efficiencies reflect sequencing reads position of HEK3 (d), a +1 C•G-to-A•T transversion edit at RNF2 (e), a +1 G•C-to- that contain the intended edit and do not contain indels among all treated cells, C•G transversion edit at EMX1 (f), a +2 T•A-to-A•T transversion edit at HBB (g), with no sorting. Mean ± s.d. of n = 3 independent biological replicates. and a +1 G•C-to-C•G transversion edit at FANCF (h). i–n, Comparison of four a 40 PE2, 13-nt PBS, VEGFA +5 G to T s Correct edit (w/o indels) 30 Indels

20

10

% of total sequencing read 0 wit h the specified edit or indels 8910 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 RT template length (nucleotides) Last templated nucleotide T C C A G A T G G C A C A T T G T C A G A G G G b 15 PE2, 13-nt PBS, DNMT1 +5 G to T s Correct edit (w/o indels) Indels 10

5

% of total sequencing read 0 wit h the specified edit or indels 8910 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 RT template length (nucleotides) Last templated nucleotide T G A C G G G A G G G C A G A A C T A G T C C T c PE2, 15-nt PBS, RUNX1 +5 G to T

s 25 Correct edit (w/o indels) 20 Indels 15

10

5

% of total sequencing read 0 wit h the specified edit or indels 8910 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 RT template length (nucleotides) Last templated T T C A G A C A G C A T A T T T G A G T C A T T nucleotide de f PE3, 17-nt RT template, PE3, 16-nt template, PE3, 11-nt RT template, nicking at +48, FANCF nicking at +53, EMX1 nicking at +49, DNMT1 60 40 50 Correct edit s s s (w/o indels) Correct edit Correct edit (w/o indels) 50 Indels 40 (w/o indels) 30 Indels Indels 40 30 30 20 20 20 10 10 10 % of total sequencing read % of total sequencing read % of total sequencing read wit h the specified edit or indels wit h the specified edit or indels wit h the specified edit or indels 0 0 0 T T T T T T A T A A A T A C T A C A A A C to to C T to A to T A to T to A to T to C A to C T T to G A to C A to G T to G T to C A to G A to C T to G A to G +2 +8 +3 +8 +1 +2 +4 +8 +9 C to +1 +3 +7 C to +8 +8 +2 C to +5 G to +1 +2 C to +7 +5 G to +8 +8 +5 G to +3 C to +4 G to +1 +2 C to +4 +5 G to +4 G to +9 C to G +6 G to +2 C to G +3 C to G +6 G to +10 G to

Extended Data Fig. 5 | Design features of pegRNA PBS and RT template pegRNA) are highlighted in red; RT templates that end in C should be avoided sequences, and additional editing examples with PE3. a, PE2-mediated +5 during pegRNA design to maximize prime editing efficiency. b, +5 G•C-to-T•A G•C-to-T•A transversion editing efficiency (blue line) at VEGFA in HEK293T cells transversion editing and indels for DNMT1 as in a. c, +5 G•C-to-T•A transversion as a function of RT template length. Indels (grey line) are plotted for editing and indels for RUNX1 as in a. d–f, PE3-mediated transition and comparison. The sequence below the graph shows the last nucleotide transversion edits at the specified positions for FANCF (d), EMX1 (e), and DNMT1 templated for synthesis by the pegRNA. G nucleotides (templated by a C in the (f). Mean ± s.d. of n = 3 independent biological replicates. Article

a b

HEK3 total C•G-to-T•A conversion FANCF tota C•G-to-T•A conversion EMX1 total C•G-to-T•A conversion Indels 80 80 80 BE2max BE4max PE2 80 BE2max BE4max BE2max PE2 PE3 T• A T• A T• A PE3 PE2 PE3 PE2 BE2max BE4max 60 60 60 60 BE4max

40 40 PE3 40 40

20 20 20 reads with indels 20 % of total sequencing target C•G converted to target C•G converted to 0 target C•G converted to 0 0 0 % of total sequencing reads with % of total sequencing reads with % of total sequencing reads with C3 C4 C7 C3 C4 C7 C3 C7 C8 C5 C6 C7 C3 C7 C8 C5 C6 C7 CB E CB E CB E c HEK3 FANCF EMX1

HEK3 precise C•G-to-T•A edits FANCF precise C•G-to-T•A edits EMX1 precise C•G-to-T•A edits T• A T• A

BE2max 60 BE2max BE4max 60 T 30 BE2max PE2 PE2 PE3 PE2 BE4max 40 40 BE4max 20 PE3 PE3

20 20 10 ase pair(s) converted to A• with ony the specified C•G % of total sequencing reads 0 0 b 0 % of total sequencing reads with % of total sequencing reads with ony the target C•G converted to only the target C•G converted to C3 C4 C7 C3 C7 C8 C5 C6 C7 C5+C6 C5+C7 C6+C7 C5+C6+C7 def HEK3 tota A•T-to-G•C conversion FANCF tota A•T-to-G•C conversion HEK3 precise A•T-to-G•C edits FANCF precise A•T-to-G•C edits Indels

80 ABEdmax ABEmax 50 ABEdmax 80 ABEdmax ABEmax 50 ABEdmax 80 ABEdmax PE2 PE3 PE2 PE2 PE3 PE2 PE2 40 40 60 ABEmax 60 ABEmax 60 ABEmax 30 PE3 30 PE3 PE3 converted to G• C converted to G• C

40 40 40 A•T converted to G• C converted to G• C 20 A• T 20 A• T A• T 20 20 eads with indels 20 10 10 r % of total sequencing target target 0 0 0 0 0 % of total sequencing reads with % of total sequencing reads with % of total sequencing reads with % of total sequencing reads with A5 A8 A5 only the target A5 A8 A5 A5 A8 only the target A5 AB E AB E HEK3 FANCF g Cas9-mediated % indels HEK3 HEK4 EMX1 FANCF pegRNA pegRNA pegRNA pegRNA NA Guide RNA R 1234 1234 1234 1234 sgRNA sgRNA sgRNA sg 100 On-target 91.8 87.5 89.2 89.1 86.8 71.8 68.6 72.8 72.8 70.9 85.6 79.7 70.6 76.6 76.0 78.7 55.9 58.3 51.8 52.0 80 Off-target site 1 17.2 1.9 5.5 5.2 1.8 54.2 39.5 48.4 49.7 49.2 81.1 63.5 48.1 53.0 59.6 12.6 1.9 1.9 1.7 1.7 60 Off-target site 2 38.0 6.5 12.6 11.8 4.7 42.5 19.5 29.4 27.3 30.3 58.3 12.0 6.0 8.2 12.9 1.1 0.2 0.2 0.2 0.1 40 Off-target site 3 8.8 0.6 1.7 1.5 0.5 98.1 96.9 97.3 97.6 97.5 14.8 4.2 3.1 3.6 4.8 2.4 0.2 0.1 0.2 0.2 20 Off-target site 4 0.3 0.1 0.1 0.1 < 0.1 45.3 16.9 28.0 27.5 29.7 39.5 1.3 0.9 0.6 1.3 1.0 0.2 0.2 0.2 0.2 0 h PE-mediated % editing and % indels (in parentheses) HEK3 (PE3) HEK4 (PE2) EMX1 (PE3) FANCF (PE3) pegRNA pegRNA pegRNA pegRNA Guide RNA 1234 1234 1 234 1234 100 44.2 61.2 40.4 48.4 18.2 14.4 9.8 7.9 28.6 14.1 35.7 15.4 56.9 32.4 42.8 47.6 On-target (11.9) (8.8) (16.5) (3.3) (0.9) (1.8) (2.0) (2.2) (3.5) (2.4) (3.3) (2.9) (9.3) (16.7) (13.6) (12.0) 80 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 0.4 < 0.1 0.4 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 0.6 < 0.1 < 0.1 Off-target site 1 (< 0.1) (< 0.1) (< 0.1) (< 0.1) (< 0.1) (< 0.1) (< 0.1) (< 0.1) (0.1) (0.1) (0.1) (0.1) (0.1) (0.1) (0.1) (0.1) 60 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 Off-target site 2 (< 0.1) (< 0.1) (< 0.1) (< 0.1) (0.1) (0.1) (0.1) (0.1) (< 0.1) (0.1) (0.1) (0.1) (< 0.1) (< 0.1) (0.1) (< 0.1) 40 < 0.1 < 0.1 < 0.1 < 0.1 0.2 6.8 19.2 7.9 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 Off-target site 3 (< 0.1) (< 0.1) (< 0.1) (< 0.1) (0.5) (1.9) (0.5) (3.5) (0.3) (0.3) (0.3) (0.3) (< 0.1) (< 0.1) (< 0.1) (< 0.1) 20 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 < 0.1 Off-target site 4 (0.1) (0.1) (0.1) (< 0.1) (< 0.1) (0.1) (0.2) (< 0.1) (0.1) (0.1) (0.2) (0.1) (< 0.1) (< 0.1) (< 0.1) (< 0.1) 0

Extended Data Fig. 6 | Comparison of prime editing and base editing, and HEK293T cells at four endogenous on-target sites and their 16 known top off- off-target editing by Cas9 and prime editors at known Cas9 off-target sites. target sites32,33. For each on-target site, Cas9 was paired with an sgRNA or with a, C•G-to-T•A editing efficiency at the same target nucleotides for PE2, PE3, each of four pegRNAs that recognize the same protospacer. h, Average BE2max, and BE4max at endogenous HEK3, FANCF, and EMX1 sites in HEK293T triplicate on-target and off-target editing efficiencies and indel efficiencies cells. b, Indel frequency from treatments in a. c, Editing efficiency of precise (below in parentheses) in HEK293T cells for PE2 or PE3 paired with each C•G-to-T•A edits (without bystander edits or indels) at HEK3, FANCF, and EMX1. pegRNA in g. Editing efficiencies reflect sequencing reads that contain the d, Total A•T-to-G•C editing efficiency for PE2, PE3, ABEdmax, and ABEmax at intended edit and do not contain indels among all treated cells, with no sorting. HEK3 and FANCF. e, Precise A•T-to-G•C editing efficiency without bystander Off-target editing efficiencies in h reflect off-target locus modification edits or indels at HEK3 and FANCF. f, Indel frequency from treatments in d. consistent with prime editing. Mean ± s.d. of n = 3 independent biological g, Average triplicate Cas9 nuclease editing efficiencies (indel frequencies) in replicates. sequence nucleotides within an insertion adjacent to the RT template (left); the the (left); template RT the to adjacent insertion an within nucleotides sequence scaffold pegRNA more or one containing reads sequencing total of percentage 4. a Note in Supplementary described target loci. into sequence scaffold pegRNA of 7|Incorporation Fig. Data Extended d c b a

% of total sequencing reads with % of total sequencing reads with % of total sequencing reads with % of total sequencing reads with scaffold insertion of any length scaffold insertion of any length scaffold insertion of any length scaffold insertion of any length 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.5 1.0 1.5 0 2 4 6 8 0 2 4 6

HTS data were analysed for pegRNA scaffold sequence insertion as as insertion sequence scaffold pegRNA for analysed were data HTS +1 T to A +1 A to G +1 C to A +1 TGC ins +1 T to C +1 A to T +1 C to G +2 A to C +1 T to G +2 C to A +1 C to T +2 A to T +2 G to A +3 C to G pegRNA +1 T ins +2 G to C +3 A to G +1 GTA ins +3 C to T

+2 G to T pegRNA +4 G to C

+3 C ins EMX1 FA +2 T to A +3 A to C HEK3 pegRNA +4-6 GGG del RNF2 +4 T to A

+2 T to G +3 A to G pegRNA

NCF +5 G del +4 T to G +3 G to C +3 A to T +5 G to A +4 T to A +4 GAT ins +3-5 GAG del +5 G to T +4 T to C +5 G to A +4 A to C +6 T ins +4 T to G +5-7 GGA del +4 A to G +7 C to A +5 G to A +6 G to C +4 A to T +8 T to A +5 G to C +6 G del +4 A del +5 G to T +8 T to C +7 A to C +5 G to T +6 G to A +8 T to G +6 G to A +6 G to C +8 T to C , Analysis for the EMX1 the for , Analysis +9 C to G +10 G to T +7 T to C +6 G to T +9 C to T

% of total sequencing reads with % of total sequencing reads with % of total sequencing reads with % of total sequencing reads with pegRNA / scaffold insertion of the pegRNA / scaffold insertion of the pegRNA / scaffold insertion of the pegRNA / scaffold insertion of the exact length specified exact length specified

0.0 0.2 0.4 0.6 exact length specified exact length specified 0.0 0.5 1.0 1.5 0.0 0.5 1.0 0.0 0.2 0.4 0.6 02 02 02 02 04 04 locus. Shown is the the is Shown locus. 04 04 pegRNA pegRNA pegRNA pegRN A scaf scaff scaf scaf 06 06 06 06 fold insertionlength(nt fold insertionlength(nt) fold insertionlength(nt old insertionlength(nt) +9 CtoG +7 Cto +2 +3 +2 +5 Gto +4 Gto +5 Gto +3 CtoG +2 Cto +1 +1 +7 +10 Gto +6 Gto +5 Gto +1 CtoG +1 Cto +4 +4 +4 +5 Gto +3 GtoC +6 Gto A A A +3 +3 +2 Gto +2 Gto +1 +1 +1 +2 Gto +3 A A A to to to 08 08 A A A 08 A A A T T T 08 to to to T G C A A to to to T C to to to to to to T G C A A C T G C T G A T A G C C A T A T C in in of pegRNA insertion up to and including the length specified on the the on specified length the including and to up insertion pegRNA of percentage total cumulative the (middle); and length specified the of insertion sequence scaffold a pegRNA containing reads sequencing total of percentage independent biological replicates. for FANCF a for ) ) +4-6 GGGdel +5 Gdel +1 +6 +8 +8 +8 +9 Cto +5-7 GGA +6 Gdel +4 GA +3 Cins +4 +8 +4 +3 Cto 0 01 0 0 TGC ins T T T T +5 Gto +5 Gto +4 +4 +4 +6 Gto +6 Gto +6 Gto +5 Gto +3-5 GAGdel +4 +1 GT +1 +2 +7 +2 +1 Cto T T T ins to to to to to to T T T A T T T T A G C T T to to to del ins to to to A G C T ins A A A G C G C A A T T T C C ins del . 100 100 100 c 00 , As in in , As

Cumulative % of total sequencing reads Cumulative % of total sequencing reads Cumulative % of total sequencing reads Cumulative % of total sequencing reads

for HEK3 a for with pegRNA scaffold insertion with pegRNA scaffold insertion with pegRNA scaffold insertion 0.0 0.5 1.0 1.5 with pegRNA scaffold insertion 0.0 0.5 1.0 1.5 0 2 4 6 8 0 2 4 6 8 02 02 02 02 +2 Gto +1 +1 +1 +2 Gto +7 +10 Gto +6 Gto +5 Gto +4 +5 Gto +3 Gto +6 Gto . +2 +5 Gto +4 Gto +5 Gto T T T d A A to to to , As in in , As A to 04 to 04 04 04 A G C to A C C C A pegRNA C pegRNA A T pegRNA C pegRNA C A T T C for RNF2 a for +5 Gto +4 +4 +4 +5 Gto scaff scaf scaf scaff +4 +8 +4 +3 Cto +2 +7 +2 +1 Cto +8 +8 +8 +9 Cto 06 06 T T T 06 06 T T T fold insertionlength(nt fold insertionlength(nt) to to to T T T old insertionlength(nt T T T old insertionlength(nt) to to to to to to to to to A G C A C A G C T A G C A G C T T = 3 n = 3 of . Mean ± s.d. +3 +3 +2 Gto 08 +3 08 08 +3 CtoG +2 Cto +1 +1 08 +9 CtoG +7 Cto +2 +3 +1 CtoG +1 Cto +4 +4 A A A A A to to to A A to to A A to to T G C to to T G T A T G A T G A ) ) 0 0 01 01 +6 Gto +6 Gto +6 Gto +5 Gto +5-7 GGA +6 Gdel +4 GA +3 Cins +4-6 GGGdel +5 Gdel +1 +6 x +3-5 GAGdel +4 +1 GT +1 -axis. -axis. TGC ins T A T T ins A T T C del ins ins 100 100 A 00 00 del ins b , As , As Article

ab HEK3, +5 G to A 6 15 8 10 PE2 Correct edit (w/o indels) ) PE2 R110S K103L (dRT) 6 106 10 Indels Cas9 H840A nickase dCas9 4 106 5 Untransfected

6 ell viability (RLU 2 10 % of total sequencing reads C 0 with the specified edit or indels T PE 2

0 dCas9 nickase

123 PE2-dR

Days post-transfection Cas9 H840A Untransfecte d

cde pegRNA targeting PRNP pegRNA targeting PRNP pegRNA targeting PRNP (adjusted p-value) (adjusted p-value) (adjusted p-value) 10 10 10 –log –log –log

log2 (PE2 / PE2-dRT) log2 (PE2 / Cas9 H840A)log2 (PE2-dRT / Cas9 H840A) f gh pegRNA targeting HEXA pegRNA targeting HEXA pegRNA targeting HEXA (adjusted p-value) (adjusted p-value) (adjusted p-value) 10 10 10 –log –log –log

log2 (PE2 / PE2-dRT) log2 (PE2 / Cas9 H840A)log2 (PE2-dRT / Cas9 H840A) ijk PRNP HEXA PRNP HEXA PRNP HEXA

Upregulated Upregulated 13 11 45 40 18 32 Upregulated 32

Downregulated 97 15 Downregulated 19 25 18 Downregulated 28 16 4

PE2 / PE2-dRT PE2 / Cas9 H840A PE2-dRT / Cas9 H840A

Extended Data Fig. 8 | See next page for caption. Extended Data Fig. 8 | Effects of PE2, PE2-dRT, Cas9(H840A) nickase, and genes and 14,368 genes were detected in PRNP and HEXA samples, respectively. dCas9 on cell viability and on transcriptome-wide RNA abundance. c–h, Volcano plot displaying the −log10 FDR-adjusted P value versus log2-fold HEK293T cells were transiently transfected with plasmids encoding PE2, change in transcript abundance for each RNA, comparing PE2 versus PE2-dRT PE2(R110S/K103L), Cas9(H840A) nickase, or dCas9, together with a HEK3- with PRNP-targeting pegRNA (c), PE2 versus Cas9(H840A) with PRNP-targeting targeting pegRNA plasmid. Cell viability was measured for the bulk cellular pegRNA (d), PE2-dRT versus Cas9(H840A) with PRNP-targeting pegRNA (e), population every 24 h after transfection for 3 days using the CellTiter-Glo 2.0 PE2 versus PE2-dRT with HEXA-targeting pegRNA(f), PE2 versus Cas9(H840A) assay (Promega). a, Viability, as measured by luminescence, at 1, 2, or 3 days with HEXA-targeting pegRNA (g), PE2-dRT versus Cas9(H840A) with HEXA- after transfection. Mean ± s.e.m. of n = 3 independent biological replicates, targeting pegRNA (h). Red dots indicate genes that show twofold or more each performed in technical triplicate. b, Percentage editing and indels for PE2, changes in relative abundance that are statistically significant (FDR-adjusted PE2(R110S/K103L), Cas9(H840A) nickase, or dCas9, together with a HEK3- P < 0.05). i–k, Venn diagrams of upregulated and downregulated transcripts targeting pegRNA plasmid that encodes a +5 G-to-A edit. Editing efficiencies (twofold change or more) comparing PRNP and HEXA samples for PE2 versus were measured on day 3 after transfection from cells treated alongside those PE2-dRT (i), PE2 versus Cas9(H840A) (j), and PE2-dRT versus Cas9(H840A) (k). used for assaying viability in a. Mean ± s.d. of n = 3 independent biological Values for each RNA-seq condition reflect the mean of n = 5 biological replicates. c–k, Analysis of cellular RNA, depleted for ribosomal RNA, isolated replicates. Differential expression was assessed using a two-sided t-test with from HEK293T cells expressing PE2, PE2-dRT, or Cas9(H840A) nickase and a empirical Bayesian variance estimation. PRNP-targeting or HEXA-targeting pegRNA. RNAs corresponding to 14,410 Article a Correction of HBB E6V in HEK293T cells with PE3

h Correct edit 60 (w/o indels) Indels

40

20

0 non-pathogenic allele or indels % of total sequencing reads wit HBB 3. 5 HBB 3. 7 HBB 5. 2 HBB 5. 3 HBB 5. 4 HBB 5. 5 HBB 5. 6 HBB 5. 7 HBB 5. 8 HBB 5. 9 HBB 5. 11 HBB 5.10 HBB 5.12 HBB 5.13 Untreate d pegRNA b 60 Correction of HEXA 1278+TATC in HEK293T cells with PE3 h Correct edit (w/o indels) Indels 40

20 non-pathogenic allele or indels % of total sequencing reads wit 0 b EXA 11 EXAs 11 HEXA 5 HEXA 6 HEXA 7 HEXA 8 HEXA 9 H EXA 11 HEXAs 1 HEXAs 2 HEXAs 3 HEXAs 4 HEXAs 5 HEXAs 6 HEXAs 7 HEXAs 8 HEXAs 9 H HEXA 10 HEXA 12 HEXA 13 HEXA 14 HEXA 15 HEXA 16 HEXA 17 HEXA 18 HEXA 19 HEXA 20 HEXA 21 H HEXAs 10 HEXAs 12 HEXAs 13 HEXAs 14 HEXAs 15 HEXAs 16 HEXAs 17 HEXAs 18 HEXAs 19 HEXAs 20 HEXAs 21 HEXAs 22 HEXAs 23 HEXAs 24 HEXAs 25 HEXAs 26 untreated 1 untreated 2 HEXAs 24 b HEXAs 26 b pegRNA

Extended Data Fig. 9 | PE3-mediated correction of E6V-encoding HBB Those pegRNAs labelled HEXAs correct the pathogenic allele by a shifted 4-bp mutation and HEXA1278+TATC by various pegRNAs. a, Screen of 14 pegRNAs for deletion that disrupts the PAM and leaves a silent mutation. Those pegRNAs correction of the HBB E6V-encoding allele in HEK293T cells with PE3. All labelled HEXA correct the pathogenic allele back to wild-type. Entries ending in pegRNAs evaluated convert the mutant HBB allele back to wild-type HBB b use an edit-specific nicking sgRNA in combination with the pegRNA (the PE3b without the introduction of any silent PAM mutation. b, Screen of 41 pegRNAs system). Mean ± s.d. of n = 3 independent biological replicates. for correction of the HEXA1278+TATC allele in HEK293T cells with PE3 or PE3b. a b HEK293T cells PE3, K562 cells PE3, U2OS cells PE3, HeLa cells 100 Correct edit (w/o indels) s Indels Correct edit Correct edit Correct edit 80 (w/o indels) (w/o indels) (w/o indels) s 40 Indels s 25 15 Indels Indels 60 20 30 10 40 15 20 20 10 5 % of total sequencing reads 10 with the specified edit or indel 5 0 R R R R R R R R R R PE 3 PE 3 PE 3 PE 3 PE 3 PE 3 PE 3 PE 3 PE 3 PE 3 % of total sequencing reads % of total sequencing reads % of total sequencing reads 0 0 with the specified edit or indel 0 with the specified edit or indel with the specified edit or indels s s T T T C C Cas9 HD Cas9 HD Cas9 HD Cas9 HD Cas9 HD Cas9 HD Cas9 HD Cas9 HD Cas9 HD Cas9 HD NC F NC F HEK3 HEK3 HEK3 HEK3 HEK3 HEK3 EMX1 non-targeting non-targeting non-targeting non-targeting non-targeting FA FA +5 G to +5 G to +5 G to +1 G to +2 G to +1 CT T in s +1 CT T in s HEK3 HEK3 HEK3 HEK3 HEK3 RNF2 RNF2 RNF2 HBB PRNP +1 ins 6xHi +1 ins 6xHi +1 T to G +3 A to C +3 A to T +3 A to T, +1 CTT ins +1 C to A +1 C to G +1 GTAins +4 A to T +6 G to T +5-6 GG to TT c d K562 cells

80 100 U2OS cells

s Correct edit (w/o indels) Correct edit (w/o indels) s Indels Indels 60 80

60 40 40 20 20 % of total sequencing read with the specified edit or indels % of total sequencing read

0 with the specified edit or indels 0 R R R R R R R R R R R R R R PE 3 PE 3 PE 3 PE 3 PE 3 PE 3 PE 3 PE 3 PE 3 PE 3 PE 3 PE 3 PE 3 PE 3 Cas9 HD Cas9 HD Cas9 HD Cas9 HD Cas9 HD Cas9 HD Cas9 HD Cas9 HD Cas9 HD Cas9 HD Cas9 HD Cas9 HD Cas9 HD Cas9 HD non-targeting non-targeting non-targeting non-targeting non-targeting non-targeting non-targeting non-targeting non-targeting non-targeting non-targeting non-targeting

HEK3 HEK3 HEK3 RNF2 RNF2 RNF2 HBB HEK3 HEK3 HEK3 RNF2 RNF2 RNF2 HBB +1 T to G +3 A to C +1 CTT ins +1 C to A +1 C to G +1 GTA ins +4 A to T +1 T to G +3 A to C +1 CTT ins +1 C to A +1 C to G +1 GTA ins +4 A to T ef HeLa cells 100 HDR ssDNA template carryover controls 100 Correct edit (w/o indels) s Correct edit (w/o indels) s 80 80 Indels Indels

60 60

40 40

20

% of total sequencing read 20 % of total sequencing read with the specified edit or indels

0 with the specified edit or indels R R R R R R R PE 3 PE 3 PE 3 PE 3 PE 3 PE 3 PE 3 0 R Cas9 HD Cas9 HD Cas9 HD Cas9 HD Cas9 HD Cas9 HD Cas9 HD non-targeting non-targeting non-targeting non-targeting non-targeting non-targeting dCas9 HD dCas9 HDR dCas9 HDR dCas9 HDR dCas9 HDR dCas9 HDR

HEK3 non-targeting non-targeting non-targeting non-targeting non-targeting HEK3 non-targeting HEK3 RNF2 RNF2 RNF2 HBB wt Cas9 HDR wt Cas9 HD R wt Cas9 HD R wt Cas9 HD R wt Cas9 HD R wt Cas9 HDR +1 T to G +3 A to C +1 CTT ins +1 C to A +1 C to G +1 GTA ins +4 A to T K562, HEK3 U2OS, HEK3 HeLa, HEK3 HEK293T, HEK3 HEK293T, HBB K562, RNF2 +3 A to T +3 A to T +3 A to T +3 A to T +4 T to A +1 C to G g (E6V correction) TGGGGCCCAGACTGAGCACGTGATGGCAGAGGAAAGGAAG Reference K562 HEK3 negative control (non-targeting pegRNA) Substitutions TGGGGCCCAGACTGAGCACGTGATGGCAGAGGAAAGGAAG 93.54% (47190 reads) Insertions TGGGGCCCAGACTGAGCACGTGATGGCAGAGGAAAGGAAG 0.47% (238 reads) Deletions TGGGGCCCAGACTGAGCACGTGATGGCAG C GGAAAGGAAG 0.21% (108 reads) Predicted K562 HEK3 PE3 +1 CTT insertion cleavage position TGGGGCCCAGACTGAGCACGTGATGGCAGAGGAAAGGAAG 69.66% (39061 reads) TGGGGCCCAGACTGAGCACGCTTTGATGGCAGAGGAAAGG 22.80% (12784 reads) TGGGGCCCAGACTGAGCACGTGATGGCAGAGGAAAGGAAG 0.32% (179 reads) TGGGGCCCAGACTGAGCACGCTTTGATGGCAGAGGAAAGG 0.21% (120 reads) TGGGGCCCAGACTGAGCACGCTTTTGATGGCAGAGGAAAG 0.21% (116 reads) K562 HEK3 Cas9-initiated HDR +1 CTT insertion TGGGGCCCAGACTGAGCAC-TGATGGCAGAGGAAAGGAAG 17.15% (7256 reads) TGGGGCCCAGACTGAGCACGTGATGGCAGAGGAAAGGAAG 13.86% (5864 reads) TGGGGCCCAGACTGAGCACGCTTTGATGGCAGAGGAAAGG 10.55% (4461 reads) TGGGGCCCAGACTGAGCA --TGATGGCAGAGGAAAGGAAG 9.72% (4111 reads) TGGGGCCCAGACTGAGCACG --ATGGCAGAGGAAAGGAAG 5.14% (2174 reads) TGGGGCCCAGACTGAGCACGGTGATGGCAGAGGAAAGGAA 4.77% (2016 reads) TGGGGCCCAGACTGAGCACG -GATGGCAGAGGAAAGGAAG 3.82% (1616 reads) TGGGGCCCAGACTGAGCACGTTGATGGCAGAGGAAAGGAA 3.17% (1342 reads) TGGGGCCCAGACTGA------TGGCAGAGGAAAGGAAG 1.92% (814 reads) TGGGGCCCAGACTGAGCACGATGATGGCAGAGGAAAGGAA 1.63% (689 reads) TGGGGCCCAGACTGAGCACG ------CAGAGGAAAGGAAG 1.26% (535 reads) TGGGGC ------TGATGGCAGAGGAAAGGAAG 0.78% (329 reads) TGGGGCCCAGACTGAGCA ------GAGGAAAGGAAG 0.61% (257 reads) TGGGGCCCAGACTGAGCACGCTGATGGCAGAGGAAAGGAA 0.59% (250 reads) TGGGGCC ------TGATGGCAGAGGAAAGGAAG 0.56% (237 reads) TGGGG------TGATGGCAGAGGAAAGGAAG 0.52% (219 reads) TGGGGCCCAGACTGAGCACG ------AGGAAAGGAAG 0.47% (197 reads) TGGGGCCCAGACTGAGCACG ------AGAGGAAAGGAAG 0.43% (181 reads) TGGGGCCCAGACTGAGCACG ----GGCAGAGGAAAGGAAG 0.41% (173 reads) TGGGGCCCAGACTGAG ------GAAAGGAAG 0.40% (169 reads) TGGGGC ------AGAGGAAAGGAAG 0.38% (159 reads) TGGGGCCCAGACTGAGCACG ------GAAAGGAAG 0.37% (156 reads) TGGGGCCCAGA------TGGCAGAGGAAAGGAAG 0.35% (148 reads) TGGGGCCCAGACTGAGCACG ------AAAGGAAG 0.35% (146 reads) TG ------ATGGCAGAGGAAAGGAAG 0.31% (132 reads) TGGGGCCCAGACTGAGC TT -TGATGGCAGAGGAAAGGAAG 0.30% (125 reads) TGGGGCCCAGACTGAGCACG ------AAG 0.30% (125 reads) TGGGGCCCAGACTGAGCACG ------GAGGAAAGGAAG 0.29% (122 reads) TGGGGCCCAGACTGAGCACG ------GAAG 0.29% (121 reads) TGGGGCCCAGA------GGAAAGGAAG 0.27% (114 reads) TGGGGCCCAG ------TGATGGCAGAGGAAAGGAAG 0.26% (112 reads) TGG------TGATGGCAGAGGAAAGGAAG 0.26% (112 reads) TGGGGCCCAGACTGAGCACGTGTGATGGCAGAGGAAAGGA 0.26% (110 reads) ------ATGGCAGAGGAAAGGAAG 0.24% (100 reads) TGGGGCCCAGACTG ------TGATGGCAGAGGAAAGGAAG 0.23% (96 reads) TGGGGCCCAGACTGAG ----TGATGGCAGAGGAAAGGAAG 0.23% (96 reads) TGGGGCCC ------TGATGGCAGAGGAAAGGAAG 0.22% (93 reads) TGGGGCCCAGACTGAGCACG -----GCAGAGGAAAGGAAG 0.21% (87 reads)

Extended Data Fig. 10 | See next page for caption. Article

Extended Data Fig. 10 | PE3 activity in human cell lines and comparison of does not contribute to the HDR measurements in a–d. g, Example HEK3 site PE3 and Cas9-initiated HDR. a, Prime editing in K562 (leukaemic bone allele tables from genomic DNA samples isolated from K562 cells after editing marrow), U2OS (osteosarcoma), and HeLa (cervical cancer) cells. with PE3 or with Cas9-initiated HDR. Alleles were sequenced on an Illumina b–e, Efficiency of generating the correct edit (without indels) and indel MiSeq and analysed using CRISPResso243. The reference HEK3 sequence from frequency for PE3 and Cas9-initiated HDR in HEK293T cells (b), K562 cells this region is at the top. Allele tables are shown for a non-targeting pegRNA (c), U2OS cells (d), and HeLa cells (e). Each bracketed editing comparison negative control, a +1 CTT insertion at HEK3 using PE3, and a +1 CTT insertion at installs identical edits with PE3 and Cas9-initiated HDR. Non-targeting controls HEK3 using Cas9-initiated HDR. Allele frequencies and corresponding Illumina are PE3 and a pegRNA that targets a non-target locus. (f) Control experiments sequencing read counts are shown for each allele. All alleles observed with with non-targeting pegRNA + PE3, and with dCas9 + sgRNA, compared with frequency ≥0.20% are shown. Mean ± s.d. of n = 3 independent biological wild-type Cas9 HDR experiments confirming that ssDNA donor HDR template, replicates. a common contaminant that artificially elevates apparent HDR efficiencies, ab Insertions (1,844 total) Duplications (6,154 total) Cumulative frequenc 100 100 100 100 C umulative frequenc 80 80 80 80 r ar

Va 60 60 60 60

40 40 40 40 in Clin in ClinV y (% ) 20 20 20 20 y (% ) % of pathogenic insertions

0 0 % of pathogenic duplications 0 0 0510 15 20 25 0510 15 20 25 Insertion length (bp) Duplication length (bp)

cd Deletions (19,706 total) Indels (1,530 total)

100 100 100 100 Cumulative frequenc Cumulative frequency

80 80 80 80 r ar

Va 60 60 60 60

40 40 40 40 in Clin in ClinV y (% )

20 20 (% ) 20 20 % of pathogenic indels % of pathogenic deletions 0 0 0 0 0510 15 20 25 0510 15 20 25 Deletion length (bp) Indel length (bp)

Extended Data Fig. 11 | Distribution by length of pathogenic insertions, reported indels (single variants that include both insertions and deletions duplications, deletions, and indels in the ClinVar database. The ClinVar relative to the reference genome) were calculated by determining the number variant summary was downloaded from NCBI on 15 July 2019. The lengths of of mismatches or gaps in the best pairwise alignment between the reference reported insertions, deletions, and duplications were calculated using and alternate alleles. a, Length distribution of insertions. b, Length reference and alternate alleles, variant start and stop positions, or appropriate distribution of duplications. c, Length distribution of deletions. d, Length identifying information in the variant name. Variants that did not report any of distribution of indels. the above information were excluded from the analysis. The lengths of nature research | reporting summary

Corresponding author(s): David R. Liu

Last updated by author(s): Oct 8, 2019 Reporting Summary Nature Research wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Research policies, see Authors & Referees and the Editorial Policy Checklist.

Statistics For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section. n/a Confirmed The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one- or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section. A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals)

For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted Give P values as exact values whenever suitable.

For Bayesian analysis, information on the choice of priors and Markov chain Monte Carlo settings For hierarchical and complex designs, identification of the appropriate level for tests and full reporting of outcomes Estimates of effect sizes (e.g. Cohen's d, Pearson's r), indicating how they were calculated

Our web collection on statistics for biologists contains articles on many of the points above. Software and code Policy information about availability of computer code Data collection Illumina Miseq Control software (3.1) was used on the Illumina Miseq sequencers to collect the high-throughput sequencing data

Data analysis Crispresso2 was used to analyze HTS data for quantifying editing activity at genomic sites. Cell Sorter Software Version 3.0.5 was used for flow cytometry analysis. RNA-seq demultiplexing was performed with bcl2fastq2 version 2.20, and sequences were trimmed with TrimGalore v. 0.6.2. Alignment of RNA-seq reads to the human genome was performed with RSEM version 1.3.1. RNA-seq data output was genearted with limma-voom and visualized in R. Frequency, mean, and standard deviations were calculated using GraphPad Prism 8. Custom python scripts provided in Supplementary Note 4 were used to analyze and quantify guide RNA scaffold insertion. For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors/reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Research guidelines for submitting code & software for further information. Data Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: - Accession codes, unique identifiers, or web links for publicly available datasets October 2018 - A list of figures that have associated raw data - A description of any restrictions on data availability

High-throughput sequencing data have been deposited in the NCBI Sequence Read Archive database under accession code PRJNA565979.

1 nature research | reporting summary Field-specific reporting Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection. Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf

Life sciences study design All studies must disclose on these points even when the disclosure is negative. Sample size Sample sizes were determined based on literature precedence for genome editing experiments.

Data exclusions No data was excluded.

Replication All experiments were repeated at least once. All attempts at replication were successful.

Randomization Yeast and mammalian cells used in this study were grown under identical conditions; no randomization was used.

Blinding Yeast and mammalian cells used in this study were grown under identical conditions; blinding was not used.

Reporting for specific materials, systems and methods We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material, system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response. Materials & experimental systems Methods n/a Involved in the study n/a Involved in the study Antibodies ChIP-seq Eukaryotic cell lines Flow cytometry Palaeontology MRI-based neuroimaging Animals and other organisms Human research participants Clinical data

Eukaryotic cell lines Policy information about cell lines Cell line source(s) HEK293T (ATCC), U2OS (ATCC), K562 (ATCC), HeLa (ATCC).

Authentication Cells were authenticated by the supplier using STR analysis.

Mycoplasma contamination All cell lines tested negative for mycoplasma.

Commonly misidentified lines None used. (See ICLAC register)

Animals and other organisms Policy information about studies involving animals; ARRIVE guidelines recommended for reporting animal research Laboratory animals To generate dissociated neuronal cultures, timed-pregnant C57BL/6 mice were provided by Charles River. Pregnant mice were euthanized at E18.5, and tissue for dissociated cultures was harvested from all embryos. October 2018

Wild animals The study did not involve wild animals.

Field-collected samples The study did not involve samples collected from the field.

Ethics oversight The Broad IACUC provided ethical guidance. Note that full information on the approval of the study protocol must also be provided in the manuscript.

2 Flow Cytometry nature research | reporting summary Plots Confirm that: The axis labels state the marker and fluorochrome used (e.g. CD4-FITC). The axis scales are clearly visible. Include numbers along axes only for bottom left plot of group (a 'group' is an analysis of identical markers). All plots are contour plots with outliers or pseudocolor plots. A numerical value for number of cells or percentage (with statistics) is provided.

Methodology Sample preparation 2.5 x 104HEK293T cells grown in the absence of antibiotic were seeded on 48-well poly-D-lysine coated plates (Corning). 16-24 h post-seeding, cells were transfected at approximately 70% confluency with 1 μL of Lipofectamine 2000 (Thermo Fisher Scientific) according to the manufacturer’s protocols and 750 ng of PE2-P2A-GFP plasmid, 250 ng of pegRNA plasmid, and 83 ng of sgRNA plasmid. After 3 days post transfection, cells were washed with phosphate-buffered saline (Gibco) and dissociated using TrypLE Express (Gibco). Cells were then diluted with DMEM plus GlutaMax (Thermo Fisher Scientific) supplemented with 10% (v/v) FBS (Gibco) and passed through a 35-μm cell strainer (Corning) prior to sorting. Cells were treated with 3 nM DAPI (BioLegend) 15 minutes prior to sorting.

Instrument Sony LE-MA900 Cell Sorter

Software Cell Sorter Software Version 3.0.5 (Sony)

Cell population abundance Of the surviving single sorted HEK293T cells edited to have HEXA 1278+TATC, 3.02% were homozygous. Of the surviving single sorted HEK293T cells edited to have HBB E6V, 25% were homozygous. Cells were genotyped using next-generation sequencing (Illumina).

Gating strategy HEK293T cells were initially gated on population using FSC-A/BSC-A (Gate A) and then sorted for singlets using FSC-A/FSC-H (Gate B). Live cells were sorted for by gating for DAPI-negative cells (Gate C). Finally the upper 50% of GFP expressing cells were sorted for using eGFP as the fluorochrome (Gate D). Tick this box to confirm that a figure exemplifying the gating strategy is provided in the Supplementary Information. October 2018

3