Lung and liver editing by lipid nanoparticle delivery of a stable CRISPR-Cas9 ribonucleoprotein

Abstract

Lipid nanoparticle (LNP) delivery of clustered regularly interspaced short palindromic repeat (CRISPR) ribonucleoproteins (RNPs) could enable high-efficiency, low-toxicity and scalable in vivo genome editing if efficacious RNP-LNP complexes can be reliably produced. Here we engineer a thermostable Cas9 from Geobacillus stearothermophilus (GeoCas9) to generate iGeoCas9 variants capable of >100× more genome editing of cells and organs compared with the native GeoCas9 enzyme. Furthermore, iGeoCas9 RNP-LNP complexes edit a variety of cell types and induce homology-directed repair in cells receiving codelivered single-stranded DNA templates. Using tissue-selective LNP formulations, we observe genome-editing levels of 16‒37% in the liver and lungs of reporter mice that receive single intravenous injections of iGeoCas9 RNP-LNPs. In addition, iGeoCas9 RNPs complexed to biodegradable LNPs edit the disease-causing SFTPC gene in lung tissue with 19% average efficiency, representing a major improvement over genome-editing levels observed previously using viral or nonviral delivery strategies. These results show that thermostable Cas9 RNP-LNP complexes can expand the therapeutic potential of genome editing.

Fig. 1. Directed evolution of GeoCas9 improves its editing efficiency by orders of magnitude and broadens its PAM compatibility.

a, Schematic diagram of the direct evolution system used to evolve GeoCas9 based on bacterial selection. AMP^R, ampicillin resistant. b, Evolutionary lineage of GeoCas9 mutants. c, Compared to the wild-type GeoCas9, evolved GeoCas9 (mutant R1W1) adequately preserved its thermostability with a melting temperature much higher than canonical SpyCas9. Melting temperatures of the three Cas9 proteins were measured by a thermal shift assay. d, Schematic diagram of GeoCas9-mediated genome editing of NPCs isolated from Ai9 mice. The spacer and PAM sequences of the GeoCas9 gRNAs were designed to turn tdTomato if successful editing occurs. Guides g7 and g8 target the LoxP sites for the stop cassette deletion. DSB, double-strand break. e, GeoCas9 mutants edit NPCs with significantly higher efficiency than wild-type GeoCas9 after electroporation-mediated delivery. Genome-editing efficiencies quantified based on tdTom(+) signals with the whole lineage of GeoCas9 mutants paired with different sgRNAs. f, PAM specificity is broadened through the further engineering of the GeoCas9 PI domain (n = 4 for each group); data are presented as mean values with individual data points.

Fig. 2. iGeoCas9 edits cells more efficiently than SpyCas9 or iCas12a after LNP-mediated delivery.

a, iGeoCas9, SpyCas9 and iCas12a edit cells with similar efficiency after nucleofection. However, iGeoCas9 edits cells more efficiently after LNP-mediated delivery than either SpyCas9 or iCas12a (n = 4 for each group); data are presented as mean values with individual data points. b, Chemical structures of the lipids used in this study; two formulations were identified that delivered iGeoCas9 RNP efficiently, termed standard and cationic (details can be found in the table). DLS of standard and cationic LNPs demonstrates that they have sizes of 178 nm and 181 nm. iGeoCas9 used in this figure is 2NLS-iGeoCas9(C2)-2NLS.

Fig. 3. iGeoCas9 RNP–LNP complexes can edit a wide range of genomic targets and multiple different cell lines.

a, LNP-mediated delivery of iGeoCas9 RNPs edits NPCs with efficiencies comparable to nucleofection. b, Chemical modification of sgRNAs improves the editing efficiency after LNP-mediated delivery (ms, 2′-methoxy and phosphorothioate linkage). c, Comparison of the genome-editing levels in HEK293T cells based on nucleofection and LNP-assisted delivery of iGeoCas9 RNP. Left, schematic diagram of iGeoCas9-mediated genome editing of HEK293T EGFP cells, resulting in the knockout of EGFP fluorescence. Right, genome-editing efficiencies quantified based on EGFP(−) signals using the engineered GeoCas9 paired with different sgRNAs (n = 4 for each group); data are presented as mean values with individual data points. NT-ctrl, non-targeting control. d, iGeoCas9 RNP–LNP complexes exhibit ultrastability, allowing for long-term storage in a neutral buffer at 4 °C. Left, schematic illustration of LNP stability test. Right, genome-editing efficiencies quantified based on tdTom(+) or EGFP(−) signals using iGeoCas9 RNP–LNP complexes stored at 4 °C for certain amounts of time (n = 4 for each group); data are presented as mean values with the s.d. iGeoCas9 used in this figure is 2NLS-iGeoCas9(C2)-2NLS except for editing experiments using 2NLS-iGeoCas9(G)-2NLS with g18 and g19 in a.

Fig. 4. Codelivery of iGeoCas9 RNPs and ssDNA templates with LNPs efficiently generates HDR in cells.

a, Characterization of LNPs encapsulating iGeoCas9 RNPs and ssDNA templates. b, Codelivery of iGeoCas9 RNPs and ssDNA HDR templates with LNPs edits the chromophore of EGFP to BFP in HEK293T cells. Top, target and donor designs for iGeoCas9-mediated chromophore editing. Bottom, genome-editing efficiencies quantified based on EGFP and BFP signals using iGeoCas9 paired with different sgRNAs ± ssDNA templates. iGeoCas9 RNP with ssDNA generates between 20% and 40% HDR in HEK293T cells. GFP, green fluorescent protein. c, Genome-editing efficiencies (indels and HDR) by iGeoCas9 paired with different sgRNAs ± ssDNA templates, as quantified by NGS (n = 4 for each group); data are presented as mean values with individual data points. iGeoCas9 used in this figure is NLS-iGeoCas9(C2)-2NLS. std, standard; cat, cationic.

Fig. 5. Rescreening of ionizable lipids dramatically boosts the delivery efficiency of iGeoCas9 RNP–LNPs assisted by enhDNA.

a, Schematic diagram of procedures for LNP assembly and general lipid compositions of the two sets of LNP formulations (FX and FC) for ionizable lipid rescreening. b, Screening results indicate that ionizable lipids can dramatically affect the RNP delivery efficiency. Editing assays with tdTom NPCs (with tdTom-g3(23ms)) and HEK293 EGFP cells (with EGFP-g2) were used for the rescreening of FX and FC formulations, respectively. LP01 (IL11) and BP lipid 312 (IL12) were identified as the optimal ionizable lipids for the FX formulation and lipid III-45 (IL8) was identified as the optimal ionizable lipid for the FC formulation. The cationic* formulation used ADP-2k as the PEGylated lipid and d-Lin as the ionizable lipid based on the general FC formulation; standard and cationic* LNP formulations were assembled at pH 7.0. c, Characterization of microfluidic-formulated LNPs based on FX12 (FX with IL12) and FC8 (FC with IL8) formula. Top, chemical structures of IL12 and IL8. Bottom left, cryo-TEM imaging of FX12 and FC8 nanoparticles. Bottom right, DLS shows particle size distribution consistent with cryo-TEM imaging. The two formulations had good to high encapsulation efficiency for RNP cargoes and showed minimal cytotoxicity to cultured cells (NPCs and HEK293 cells; n = 4 for each group); data are presented as mean values ± s.d. d, FX12 and FC8 formulations show substantially improved efficiency for RNP delivery (with tdTom-g3(23ms) and EGFP-g6(23ms) as the sgRNAs) compared to the standard and cationic* formulations with different RNP dosages, even at subnanomolar RNP concentrations. Genome-editing efficiencies quantified on the basis of tdTom(+) or EGFP(−) signals using iGeoCas9 RNP–LNP complexes (n = 4 for each group); data are presented as mean values with individual data points. e, iGeoCas9 RNP–LNP delivery outcompetes mRNA+sgRNA–LNP delivery, especially with low cargo dosages. mRNA delivery is sensitive to sgRNA stability and requires hypermodification of sgRNA to enable successful editing at low mRNA and sgRNA dosage, while sgRNA modification does not affect the editing efficiency based on RNP–LNP delivery. iGeoCas9 used in this figure is 2NLS-iGeoCas9(C1)-2NLS. OMe, 2′-Omethyl; PS, phosphorothioate; Conc., concentration.

Fig. 6. iGeoCas9 RNP–LNPs efficiently edit the liver and lungs of mice.

a, Schematic diagram of the experimental design used to evaluate iGeoCas9 RNP–LNP-mediated editing in Ai9 mice. b, Schematic presentation of LNP preparation procedures. c, The modified FX12 LNP formulation (FX12m, with lipid compositions indicated in the table) primarily edits the liver tissue with 37% efficiency. In vivo genome-editing levels in different tissues and different cell types in the liver were quantified by tdTom(+) signals using flow cytometry. d, The modified FC8 LNP formulation (FC8m, with lipid compositions indicated in the table) primarily edits the lung tissue with 16% efficiency. In vivo genome-editing levels in different tissues and different cell types in the lungs were quantified by tdTom(+) signals using flow cytometry. For c and d, n = 5 for each group; data are presented as mean values with individual data points and the s.d.; IVT sgRNA, tdTom-g3(23), was used. e,f, Nuclear staining with DAPI (blue) and imaging of tdTomato (red) in the edited and nonedited liver (e) or lung (f) tissues. Editing signals were observed with the tissues from experimental mice (n = 5). RFP, red fluorescent protein. g, sgRNA target designs for PCSK9 and SFTPC gene editing with iGeoCas9 in the liver and lungs, respectively. h,i, In vivo PCSK9 and SFTPC gene-editing levels (indels) in the liver and lung tissues using FX12m and FC8m LNP formulations, respectively, as quantified by NGS (n = 5 for each group); data are presented as mean values with individual data points and the s.d.; PBS-only injections are included as negative controls and the indels in the liver (FX12m) or in the lungs (FC8m) are shown as the blank editing levels. iGeoCas9 used in this figure is 2NLS-iGeoCas9(C1)-2NLS. Neg ctrl, negative control.

Extended Data Fig. 1. Directed evolution of GeoCas9.

a. Modelled GeoCas9 structure with mutations highlighted. b. Schematic illustration on two sequential rounds of selection to identify improved GeoCas9 mutants. c. Mutants and beneficial mutations identified in each round of selection. d. Target cleavage activities of WT-GeoCas9 and R1W1 mutant in the bacterial assay using different spacer and PAM sequences, as reflected by the bacterial survival rates.

Extended Data Fig. 2. Off-target effect analysis for iGeoCas9-mediated genome editing.

a. Schematic illustration on the analysis of on-target and off-target editing activities by iGeoCas9. b. On-target and off-target sequences listed in the tables, editing levels shown in the bar graphs. iGeoCas9 shows overall minimal off-target editing. Editing efficiencies quantified by NGS. n = 2 for each group, data are presented as mean values with individual data points. Target 1 = AAVS1 site1; target 2 = AAVS1 site2; target 3 = EMX1 site3. iGeoCas9 used in this figure is NLS-iGeoCas9(C2)-2NLS.

Extended Data Fig. 3. Optimization of LNP formulation for iGeoCas9 RNP delivery.

a. Comparison of three different genome editors for Ai9 NPC editing based on RNP delivery by LNPs. b. Optimization of the percentage of pegylated lipid ADP-2k in LNP formulations. c. Comparison of different pegylated lipids in LNP formulations for iGeoCas9 RNP delivery efficiency and cytotoxicity with NPCs. n = 4 for each group, data are presented as mean values with individual data points. iGeoCas9 used in this figure is 2NLS-iGeoCas9(C2)-2NLS.

Extended Data Fig. 4. Mechanistic rationalization for promoted RNP delivery with acid-degradable lipids.

a. pH-sensitive acetyl linker used in synthetic lipid design. b. Endocytosis pathway in LNP-based delivery promoted by the pH-sensitive acetyl linker in the lipids.

Extended Data Fig. 5. Editing of pathogenic mutations in the CFTR gene through HDR.

a. Target and donor designs for iGeoCas9-mediated editing of pathogenic mutations. b. Genome editing efficiencies quantified by NGS. n = 4 for each group, data are presented as mean values with individual data points. iGeoCas9 used in this figure is NLS-iGeoCas9(C2)-2NLS.

Extended Data Fig. 6. Rescreening of ionizable lipids to improve LNP delivery efficiency of iGeoCas9 RNP.

a. Lipid compositions for LNP formulations shown in the tables. b. Structures of ionizable lipids (IL1 to IL13). c. Screening of ionizable lipids for the FX formulation to deliver iGeoCas9 RNP to Ai9 tdTom NPC and HEK293T EGFP cells for genome editing. Genome-editing efficiencies quantified based on tdTom(+) or EGFP(‒) signals using iGeoCas9 RNP:LNP complexes in two doses. n = 4 for each group, data are presented as mean values with individual data points. d. Screening of ionizable lipids for the FC formulation to deliver iGeoCas9 RNP to HEK293T EGFP cells for EGFP knock-down. Genome-editing efficiencies quantified based on EGFP(‒) signal using iGeoCas9 RNP:LNP complexes. n = 4 for each group, data are presented as mean values with individual data points. iGeoCas9 used in this figure is 2NLS-iGeoCas9(C1)-2NLS.

Extended Data Fig. 7. Comparison of the editing efficiency of iGeoCas9 and SpyCas9 in Ai9 tdTom NPCs based on FX12-LNP delivery of corresponding RNP.

LNP characterization shows similar encapsulation properties for iGeoCas9 and SpyCas9 RNP cargoes, but SpyCas9 has much lower efficiency compared to iGeoCas9, especially at low RNP dosages. n = 4 for each group, data are presented as mean values with individual data points or ± standard deviation (encapsulation rates). Imaging of NPC cultures suggests that SpyCas9 RNP:LNP complexes tend to form aggregates in the culture media, probably due to the instability of SpyCas9 RNP, while no LNP aggregates were visibly observed for iGeoCas9 RNP. iGeoCas9 used in this figure is 2NLS-iGeoCas9(C1)-2NLS.

Extended Data Fig. 8. Comparison of the editing efficiency of iGeoCas9 genome editors delivered as mRNA+sgRNA versus RNP using FX12-LNP in Ai9 tdTom NPCs.

mRNA delivery is sensitive to sgRNA stability and requires hyper-modification of sgRNA to enable successful editing at low mRNA+sgRNA dosage, while sgRNA with modification or not does not affect the editing efficiency based on RNP:LNP delivery. Overall, RNP delivery outperforms mRNA+sgRNA delivery, especially with low cargo dosages. n = 4 for each group, data are presented as mean values with individual data points. iGeoCas9 used in this figure is 2NLS-iGeoCas9(C1)-2NLS.

Extended Data Fig. 9. PCR validation of liver and lung editing with Ai9 tdTomato mice.

a. Schematic illustration of iGeoCas9-mediated transgene editing in Ai9 mouse models to turn on tdTomato expression. b. PCR validation of genome edits in the liver and lungs of Ai9 tdTomato mice following IV injections of FX12m and FC8m LNPs.

Extended Data Fig. 10. LNP delivery of prime editor RNP.

Preliminary results of LNP delivery of prime editor (PE2, based on SpyCas9) showed 1% efficiency in achieving the desired GFP-to-BFP conversion in HEK293T cells. Optimization of the RNP:LNP complex by using a more stable prime-editor RNP, along with an improved LNP formulation, is expected to further enhance the editing efficiency.