Human 8-cell embryos enable efficient induction of disease-preventive mutations without off-target effect by cytosine base editor

Abstract Approximately 140 million people worldwide are homozygous carriers of APOE4 (ε4), a strong genetic risk factor for late onset familial and sporadic Alzheimer’s disease (AD), 91% of whom will develop AD at earlier age than heterozygous carriers and noncarriers. Susceptibility to AD could be reduced by targeted editing of APOE4, but a technical basis for controlling the off-target effects of base editors is necessary to develop low-risk personalized gene therapies. Here, we first screened eight cytosine base editor variants at four injection stages (from 1- to 8-cell stage), and found that FNLS-YE1 variant in 8-cell embryos achieved the comparable base conversion rate (up to 100%) with the lowest bystander effects. In particular, 80% of AD-susceptible ε4 allele copies were converted to the AD-neutral ε3 allele in human ε4-carrying embryos. Stringent control measures combined with targeted deep sequencing, whole genome sequencing, and RNA sequencing showed no DNA or RNA off-target events in FNLS-YE1-treated human embryos or their derived stem cells. Furthermore, base editing with FNLS-YE1 showed no effects on embryo development to the blastocyst stage. Finally, we also demonstrated FNLS-YE1 could introduce known protective variants in human embryos to potentially reduce human susceptivity to systemic lupus erythematosus and familial hypercholesterolemia. Our study therefore suggests that base editing with FNLS-YE1 can efficiently and safely introduce known preventive variants in 8-cell human embryos, a potential approach for reducing human susceptibility to AD or other genetic diseases.


Introduction
Alzheimer's disease (AD) represents one of the most severe and widespread neurodegenerative diseases due to a progressive loss of memory and other cognitive functions (Huang et al., 2012;Selkoe, 1991). Moreover, with no effective treatments for AD currently available, the number of AD patients increases at an astounding doubling rate of 20 years, which if unchecked, will result in 80 million affected individuals by 2040. Mutations in the genes encoding amyloid precursor protein (APP), presenilin-1 (PSEN1), and presenilin-2 (PSEN2) are associated with early-onset familial AD, but only accounts for >1% of all AD cases (Huang and Mucke, 2012;Selkoe, 1991). The majority of AD cases occur at age 65 or older, and are thus commonly referred to as late-onset AD. In humans, the APOE gene has three major isoforms [APOE2 (ε2), APOE3 (ε3), and APOE4 (ε4)] that each confer different effects on lipid and neuronal homeostasis (Tanzi, 2012).
An estimated 140 million people carry homozygous copies of ε4 worldwide, rendering the possibility of using preimplantation genetic diagnosis to identify ε4 couples and eliminate homozygous ε4 embryos during conception unrealistic at best, even among the limited set of carriers undergoing assisted reproductive therapy. Indeed, only a single amino acid difference distinguishes the ε4 (112R/158R; CGC/CGC) allele from ε3 (112C/158R or 112R/158C; TGC/CGC or CGC/TGC), which could be efficiently converted from a C to T without inducing a double-strand break (DSB) in human cells (Komor et al., 2016) and embryos using a cytosine base editor (CBE) such as a Cas9 nickase with a fused deaminase. Since homozygous APOE4 carriers represent a much larger population (i.e., millions of adults) than that of other mutation carriers with severe, premature death-associated monogenic defects, targeted editing of APOE4 in embryos could provide broad therapeutic benefits for a massive swath of high-risk candidates of late-onset sporadic AD.
Conventional CRISPR/Cas9 triggers an error-prone, nonhomologous end joining DNA repair pathway after DSB induction, resulting in complex repairing alleles such as insertion, deletion, substitutions, and other compound mutations potentially with loss-of-function, gain-of-function, and other unknown consequences. Furthermore, a recent study found that CRISPR-Cas9 administration in early human embryos leads to frequent loss of the targeted chromosome and uncontrollable chromosome rearrangement (Zuccaro et al., 2020), raising huge safety concerns about this technology. BE using nickase Cas9 and engineered deaminase could introduce targeted base conversions without DSB, providing a much safer genome editing tool without causing detrimental off-target effects than wildtype Cas9 (Doman et al., 2020;Gaudelli et al., 2017;Komor et al., 2016;Zuo et al., 2020). Previous studies (Dang et al., 2022;Liang et al., 2017;Zeng et al., 2018;Zhang et al., 2019) have demonstrated induction and correction of genetic mutations with base editors (BE) in human embryos. However, whether base editing could install preventive mutations in human embryos with high efficiency and specificity remains to be investigated.
To test whether susceptibility to this widespread and untreatable disease could be reduced in high risk ε4-carrying embryos, we investigated the possibility of converting ε4 to ε3 alleles by targeted base editing while performing stringent control measures to confirm that only the correct edits were introduced. To this end, we first screened eight variants of CBEs by injection at four different embryonic stages (from the 1-to 8-cell stages) and found that the FNLS-YE1 editor in 8-cell embryos resulted in the comparable base conversion rate (i.e., up to 100%) with the lowest bystander [nontargeted nucleotide substitution (NTS) or indel mutation] effects. Further assays demonstrated that FNLS-YE1 could efficiently convert AD-susceptible ε4 allele into the AD-neutral ε3 allele in human embryos carriers for ε4. In addition, we also found that FNLS-YE1 could efficiently introduce protective alleles in TYK2 (Diogo et al., 2015;Harper et al., 2015)/WDFY4 (Yang et al., 2010;Zhang et al., 2014) and PCSK9/ANGPTL4 (Harper et al., 2015;Stitziel et al., 2016), offering a potential approach to reduce susceptivity for systemic lupus erythematosus (SLE) and familial hypercholesterolemia (FH), respectively. Overall, these findings in 8-cell embryos provide a technical basis that can guide the development of low-risk personalized gene therapies in utero, and somatic editing approaches for reduced risk of AD or other genetic diseases.
Next, we checked for bystander effects resulting from either FNLS-BE3 or FNLS-YE1 base editing in the OCT4 and PCSK9 genes of different early-cleavage embryos. We found that both editors induced fewer NTS or indel mutations in 8-cell embryos than in 1/2/4-cell embryos (Figs. 1G, 1H and S1E-H). These findings suggesting the advantages of 8-cell embryos were further confirmed using a panel of 10 other endogenous genes (Fig. 1I). Further comparison of the two editors showed that FNLS-YE1 had a similar editing window and editing efficiency as FNLS-BE3, but with significantly fewer bystander effects (Figs. 1J-L, S1I and S1J). Finally, we verified that the editing effects we observed in 8-cell embryos were not due to the total amount of injected editing agents. For this purpose, we used a 4-fold increase in the concentration of editing agents in 2-cell embryos and found no significant changes in efficiency for FNLS-YE1 (Fig. S1B).
Taken together, our results identified FNLS-YE1 as the most efficient and accurate editor among all eight editors tested here, and that the 8-cell stage was the most suitable for base editing in early human embryos.

FNLS-YE1 induces no detectable off-target effects or developmental defects
Recent studies have reported that CBEs can induce both Cas9dependent (resulted from sequence similarity) (Fu et al., 2013;Hsu et al., 2013) and Cas9-independent (resulted from deaminase) off-target effects in both genomic DNA (Zuo et al., 2019) and transcriptome-wide mRNA (Grunewald et al., 2019b;Zhou et al., 2019). However, some engineered BEs, including FNLS-YE1, were shown to induce no detectable off-target effects in mouse embryos and human cell lines (Doman et al., 2020;Zuo et al., 2020). We therefore sought to further verify whether FNLS-YE1 induced off-target effects in human embryos. We first examined whether FNLS-YE1 induced any Cas9-dependent off-target alterations. To minimize the differences in genetic background between the gene-edited and control blastomeres, we co-injected FNLS-YE1, GFP mRNA, and PCSK9 sgRNA for g.55046590G>A (G8) or g.55046591G>A (G7) into 4 of the 8 blastomeres of an eight-cell embryo derived from a 2PN zygote ( Fig. 2A). The injected blastomeres were identified by GFP expression (GFP + ) in early morula embryos. We found no evidence indicating the introduction of off-target mutation in targeted deep sequencing analysis of the 13 top off-target sites predicted by Cas-OFFinder in either the injected or un-injected blastomeres (Fig. 2B). Furthermore, we also conducted primer extension mediated (PEM)-seq experiments to identify potential off-target loci of the PCSK9 sgRNA in HEK293T and human embryonic stem cell lines (hESCs). This analysis captured seven off-target mutations in the Cas9/sgRNA-treated cells (Fig. S2). Comparison between injected with un-injected blastomeres showed that none of these seven off-target loci had a significant editing signal in targeted deep sequencing data (Fig. 2B). These results thus indicated that FNLS-YE1 induced no Cas9-dependent off-target alterations apart from its targeted edits in PCSK9 at g.55046590G>A (G8) or g.55046591G>A (G7) in 8-cell embryos.
To next test for Cas9-independent off-target effects on RNA, we performed single-cell RNA sequencing at two days after injecting blastomeres with either the "GFP-only," "FNLS-YE1, PCSK9 sgRNA, and GFP," or "FNLS-BE3, PCSK9 sgRNA, and GFP" RNAs, together with corresponding un-injected control blastomeres ( Fig. 2A). We found no differences in the background levels of RNAs harboring single nucleotide varients (SNVs), nor in the pattern of canonical C-to-U conversions in the blastomeres injected with FNLS-YE1 compared with the GFP only-injected control blastomeres ( Fig. 2C and 2D). By contrast, FNLS-BE3 induced severe off-target effects in the transcriptome and canonical C-to-U pattern in human blastomeres, which was consistent with previous studies (Figs. 2C, 2D, S3 and S4) (Grunewald et al., 2019a(Grunewald et al., , 2019bZhou et al., 2019).
Third, we performed whole genome sequencing (WGS) on hESCs derived from 8-cell human embryos injected or not with FNLS-YE1, together with sgRNAs targeting PCSK9. We also sequenced and analyzed the genomes of peripheral blood cells obtained from sperm and egg donors as a control for genetic background to distinguish FNLS-YE1-induced off-target variants from germline variants (Fig. 2E). Targeted deep sequencing was used to confirm the high targeting efficiency of FNLS-YE1 in four hESCs derived from edited human embryos (Fig. 2F). WGS showed that the numbers of SNVs and indels were similar between FNLS-YE1-edited and nonedited hESCs (Figs. 2G, 2H, S5A and S5B). Furthermore, these SNVs for each edited hESCs exhibited no canonical patterns of C-to-T conversion characteristic of off-target activity by CBEs (Figs. 2I, 2J and S5C) (Zuo et al., 2019(Zuo et al., , 2020. We also performed karyotyping analysis of different hESCs and somatic cells collected from the gamete donors, and found normal karyotypes in all cells, indicating no obvious chromosomal abnormalities after base editing by FNLS-YE1 (Table  S1). In addition, targeted deep sequencing analysis of the top 13 off-target sites predicted by Cas-OFFinder showed no off-target mutations in the FNLS-YE1-edited and nonedited hESCs, nor in the seven potential off-target sites identified by PEM-seq ( Fig.  S5D). Taken together, these PEM-seq, RNA sequencing, and WGS results demonstrated the high targeting specificity of FNLS-YE1 in human embryos without detectable off-target effects.
Finally, to investigate the potential effects of FNLS-YE1 editing on embryonic development, we examined the morphology and development rates of embryos from the 8-cell to blastocyst stages injected or not with FNLS-YE1. We observed no morphological abnormalities in the 8-cell injected embryos compared with that PAM sequences are shown in black and blue, respectively. Target nucleotides within the CBE editing window are shown in red. (D) Single-cell sequencing analysis of base editing efficiency in embryos injected with FNLS-BE3 and FNLS-YE1 targeting the OCT4 locus at the 1-cell, 2-cell, 4-cell, and 8-cell stages. The indels and NTS frequency on the 12 target sites edited with FNLS-BE3 and FNLS-YE1. n = 12 independent experiments for each group in (J-L). Data are presented as the mean ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, unpaired Student's t-test. ns, not significant. Each dot in Fig. 1C-H and 1K-L represents an embryo and the endogenous locus tested, respectively.  (Fig. 2K). Developmental rates also appeared unaffected in the 8-cell injected embryos, implying that base-editing with FNLS-YE1 was safe for use in embryonic editing ( Fig. 2L; Table S2).
Analysis of developmental rates identified no developmental defects in the edited embryos from the 8-cell to blastocyst stages (Fig. 4J). We also used targeted deep sequencing to check the off-target sites for APOE (R158C) and found no evidence of off-target editing (Figs. 5A and S7). Furthermore, we performed single blastomere RNA sequencing of APOE-edited embryos to screen for potential off-target effects in the transcriptome. In contrast to the conventional FNLS-BE3 editor, APOE-targeting by FNLS-YE1 resulted in similar numbers of SNVs and C-to-U patterns as the GFP group at the RNA level (Figs. 5B-E and S8), indicating the absence of detectable off-target effects, in agreement with the results of PCSK9-targeting experiments.
Taken together, these results suggest that it is feasible to reduce the risk of AD associated with the APOE4 allele without inducing off-target effects using the high-fidelity FNLS-YE1 editor via sequential conversion of APOE4 into the neutral APOE3r or APOE3 into the protective APOE2 alleles.

Induction of various protective mutations for other diseases
We have also examined whether FNLS-YE1-mediated base conversion in human embryos shown above could be generalized to induce preventive mutations for other diseases. SLE is a typical autoimmune disease characterized by autoantibody production and multi-organ damage (Yang et al., 2010). Genetic factors are known to play an important role in this disease (Fig. 6A), with the monozygotic twin concordance rate between 20% and 59%, and the risk for siblings of affected individuals 30 times higher than that for the general population (Arnett et al., 1976;Ramos-Niembro et al., 1978;Sestak et al., 1999). As a proof-of-concept experiment, we repurposed FNLS-YE1 for inducing protective alleles in TYK2 and WDFY4 genes that are associated with reduced risk for SLE (Fig. 6B) (Diogo et al., 2015;Yang et al., 2010;Zhang et al., 2014), via A928V (C > T) (Diogo et al., 2015) and R1816Q (G > A) (Yang et al., 2010) conversion, respectively. Two days after injection of FNLS-YE1 and the corresponding sgRNA in 8-cell human embryos, we analyzed the editing efficiency in single blastomeres from both 3PN and 2PN embryos and found high editing efficiency (up to 90%) at the allelic level was achieved for both TYK2 and  Roles of APOE in Aβ-dependent and Aβ-independent pathways in AD pathology; the detrimental effects of APOE4 reflect its unique structural and biophysical properties. APOE is synthesized by astrocytes, activated microglia, and neurons. In response to stressors and injurious agents, neurons turn on or increase their expression of APOE to repair or remodel the damaged neurons. However, neuronal APOE undergoes proteolytic processing and APOE4 is more susceptible than APOE3 to the cleavage. The APOE fragments cause cytoskeletal disruption, such as tau phosphorylation, neurofibrillary tangle (NFT) formation, and mitochondrial dysfunction, finally leading to neurodegeneration. Figure adapted from (Mahley et al., 2006). (B) APOE structure and genotype frequency in a human population. Data from (Liu et al., 2013a;Raber et al., 2004).    . We also performed targeted deep sequencing to verify the off-target sites for WDFY4 and found no evidence for off-sites targeting (Fig. S10). Furthermore, NTS and indel analysis showed that <5% unwanted mutations were induced by FNLS-YE1 on target loci for TYK2 and WDFY4 genes (Fig. S9C).
FH is a hereditary disease with a high incidence of about 1 in 250 (Akioyamen et al., 2017;Nordestgaard et al., 2013). Previous studies have identified several mutations in LDLR, APOB, and PCSK9 genes as causative factors for FH (Sturm et al., 2018). Natural loss-of-function mutations in PCSK9 (Cohen et al., 2006) and ANGPTL4 (Stitziel et al., 2016) resulted in notable protective effects in reduced LDL-cholesterol level and lower risk of premature coronary artery diseases (CAD) over large-population epidemiological studies (Fig. 6D and 6E). In contrast to the induction of a nonnatural loss-of-function variant W156X of PCSK9 shown in both 3PN and 2PN embryos (Figs. 6F and S9B), we attempted to install natural loss-of-function alleles in PCSK9 (Harper et al., 2015) (Fig. S9E) and found five of them having targetable sequence with both NGG PAM and target base within 5-to-8 editing window for FNLS-YE1 (Fig. S9F). However, none of five alleles have >10% C-to-T conversion rate after injection of FNLS-YE1 and the corresponding sgRNA in 8-cell embryos (Fig. S9F). In addition, since W428X has no NGG PAM sequence, we thus used PAMless-FNLS-YE1 for editing (Fig. S9E). Our results showed 20%-50% conversion rate for loss-of-function mutation W428X (Fig. S9F). Moreover, we performed similar experiments on ANGPTL4 gene previously identified for its protective effect on CAD by lossof-function mutations (Figs. 6F and S9B). In both 3PN and 2PN embryos, 8-cell injection with FNLS-YE1 and the corresponding sgRNA enabled >65% targeted C-to-T substitution, resulting in W350X conversion for ANGPTL4 (Figs. 6F and S9B). Furthermore, NTS and indel analysis showed that <10% unwanted mutations were induced by FNLS-YE1 or PAMless-FNLS-YE1 on target loci for PCSK9 and ANGPTL4 genes ( Fig. S9D and S9G).
Together, these results indicate that FNLS-YE1 could efficiently introduce preventive alleles in TYK2/WDFY4 and PCSK9/ANGPTL4, offering a potential approach to reduce susceptivity for SLE and FH, respectively.

Discussion
Given the scale of homozygous ε4 distribution among the global population, we first explored a method for reducing the risk of AD through base editing while maintaining focus on minimizing the off-target effects. For this purpose, we used the ICSI method to prepare 2PN embryos carrying the ε4 allele from a heterozygous ε4 male identified by preimplantation genetic screening for in vitro fertilization. Given efficient and specific base substitution capability, base editing system is considered more suitable for treating >50% genetic diseases caused by different point mutations described in the ClinVar database (Doman et al., 2020;Komor et al., 2016;Zuo et al., 2020). In present study, we tested eight BEs at four different stages of human early embryos and demonstrated that the ε4 allele of APOE gene could be efficiently switched to the neutral ε3r variant by converting Arg158 to Cys158 using the FNLS-YE1 editor in 8-cell embryos. To ensure that the benefits of this approach far outweighed the risks, we dedicated a nontrivial level of effort to minimizing the off-target effects while characterizing the on-target effects of editing. Our results indicated that injection at the 8-cell stage led to higher on-target efficiency with fewer bystander edits or indels than other stages, possibly attributable to the less compact structure of chromatin at this stage than in earlier stages (Lee et al., 2014). Furthermore, by performing stringent control measures, our study also provides the first demonstration that FNLS-YE1 can serve as an efficient and high fidelity BE for C-to-T base conversion in human embryos. Previous human and animal studies indicated that APOE4 clearly accelerates Aβ aggregation, resulting in the occurrence and development of AD (Lin et al., 2018;Shi et al., 2017). Thus, it is reasonable to propose that APOE4-targeted embryonic editing could serve as a first-line approach for homozygous ε4 couples to reduce AD susceptibility in their newborns prior to birth. Furthermore, in consideration of guidelines proposed by the National Academy of Sciences and Medicine for future potential clinical applications of gene editing in humans, our results show that this approach is highly versatile in its modification of different genes, and can be used to effectively ensure the preservation of both widely distributed or rare beneficial alleles, as appropriate for different diseases.
It is now clear that reducing plaques does not halt disease progression even in relatively early AD patients. This may be one reason why most clinically tested vaccines and mAbs targeting plaques have been ineffective thus far (Ostrowitzki et al., 2012;Rinne et al., 2010;Sevigny et al., 2016;van Dyck, 2018). In addition, preclinical experiments suggest that preventive rather than therapeutic vaccination may be more effective for limiting AD development (Petrushina et al., 2017). This finding is unsurprising if we consider that removal of plaque deposits simply leaves a hole in the nervous circuits, which is difficult damage to repair. Hence, instead of plaque removal, then prophylactic prevention of plaque accumulation in the first place may represent a more sound strategy against AD.
Further evidence that vaccine-based immunological stimulation against plaques is ineffective for treating dementia can be found in the Phase II/III clinical trial (i.e., the Generation Study) of the Novartis CAD106 vaccine (NCT02565511), which was terminated early. Designed to test if immunotherapy could prevent AD in cognitively healthy, homozygous carriers of the APOE4 allele, aged 60-75, unexpected changes were observed in cognitive function, brain volume loss, and body weight loss. Thus, bypassing the need for a heightened immune response among elderly patients (who typically have higher rates of compromised immunity), we propose that, presuming the elimination of off-target effects in future applications, in utero personalized gene therapies or early-stage somatic editing approaches targeting AD-susceptible APOE4 should result in AD-neutral APOE3, the phenotypes of which are well established in children and adults.
In addition to the ε2, ε3, and ε4 alleles resulting from two different C-to-T SNP haplotypes, a fourth uncommon haplotype, ε3r, has been identified in four families worldwide (Seripa et al., 2011). Although several published studies have proposed that it shares an identical phenotype with that of ε3 (Seripa et al., 2007(Seripa et al., , 2011, the rarity and physiological role of ε3r remains unknown. Since the ε3 allele carrying the R112C mutation in APOE is more prevalent than ε3r in our population, we focused on identifying a between low LDL-C level or CAD risk and clinically relevant loss-of-function alleles in PCSK9. Figure is a visual adaption of data, from (Cohen et al., 2006). (F) Single blastomere genotyping results of human 2PN embryos injected with FNLS-YE1 for the induction of protective mutations on PCSK9 and ANGPTL4 loci. INT, intact. Data are presented as the mean ± SEM. Each dot in Fig. 6C and 6F represents an embryo. gene editor that was most effective at converting the ε4 allele into ε3. In addition, recent studies have reported a coding mutation (A673T) in the APP gene that protects against AD and cognitive decline in the elderly without AD, possibly by reducing the formation of amyloidogenic peptides (Jonsson et al., 2012). However, we observed no or very low C-to-T conversion in tests of various sgR-NAs in APOE (112R) and APP loci ( Fig. S6C and S6D), respectively. Therefore, to exploit this potentially prevalent or protective allele, it is necessary to optimize the current base editing system or use a prime editing system (Anzalone et al., 2019).
We also demonstrated FNLS-YE1 could introduce known protective variants in human embryos to potentially reduce human susceptivity to FH and SLE. However, FDA-approved drugs for FH and weak association of gene variants for SLE protection limit the advantage of inducing protective mutations by human embryonic editing. Therefore, more studies are necessary for these diseases before using human embryonic editing to introduce protective mutations.
In summary, our results firstly demonstrated that base editing with FNLS-YE1 can efficiently introduce known preventive variants without off-target effect in 8-cell human embryos, a potential approach for reducing human susceptibility to AD or other genetic diseases. Based on the increase of efficiency and specificity, we believe FNLS-YE1-mediated disease-preventive mutations can guide the development of somatic gene therapy for some genetic-based diseases.

Ethical statement
The regulatory framework surrounding the use of human gametes and embryos in this research was based on the Management of Human Assisted Reproductive Technology (2001) Editing on 3 September 2020), the Helsinki Declaration, and other laws and regulations. This study was peer-reviewed with deliberating consideration given to the use of gene editing technology in human embryos for basic research by Institutional Review Board of the Shanghai Jiao Tong University Renji Hospital Assisted Reproductive Ethics Committee (RHAREC). The committee comprised of seven members from internal and external sources: a committee director/clinician, a clinical geneticist, an ethics expert, a legal expert, a director nurse, and two director doctors. Upon completion of the review, the committee approved this novel research (research license number 2017112406) using gene editing technology under the condition of strictly complying with current guidelines and policies in China. The approved study was guided by the genetic expert and monitored regularly by the RHAREC since the license was granted.
Before oocyte retrieval, all donors understood and signed the informed consent for donation of gametes for scientific research. All details of the proposed research project were clearly presented to donors, including the use of gene editing tools on gametes, embryos, and their derivatives (human embryonic stem cell lines, hESCs) to evaluate the safety and efficacy. In accordance with internationally accepted standards that developmental progression should not exceed 14 days, all human embryos used in this project were cultured in vitro for no >6 days, and then the subsequent genotyping analysis and the derivation of hESC were performed. Additionally, donors were informed that the donated gametes would not be used for other purposes, including but not limited to, conception for other people by assisted reproduction methods without informed consent of the study participants.

Derivation and culture of embryonic stem cells from human embryos
A total of four CBE-edited stem cell lines and one wild-type stem cell line from oocyte donors (age 25-35 at donation) and sperm from the same donor, were obtained and used for experiments to determine on-target efficiency and off-target effects after FNLS-YE1 mediated C-to-T conversion of the PCSK9 allele. hESCs were derived from 8-cell injected and un-injected embryos after trophectoderm biopsy and placing the inner cell mass into a 4-well plate as previously described (Yamada et al., 2014). Briefly, mural trophectoderm was ablated using a laser-assisted method with pulses of 400 ms, at 100% intensity (Hamilton Thorne; Chen et al., 2009). This method spares polar trophectoderm, which usually results in trophectoderm growth that is then ablated with additional pulses. Derivation was performed on irradiated mouse embryonic fibroblasts in knockout-Dulbecco's modified Eagle's medium (KO-DMEM, Gibco) with 25% knockout-serum replacement (KO-SR) and 10 mmol/L Rock inhibitor Y-27632, 10 ng/mL basic fibroblast growth factor, and 2% ES grade fetal bovine serum (FBS). Subsequently, hESCs were cultured in mTeSR1 (Stem Cell Technologies) with feeder-and serum-free conditions on growth factor-reduced Matrigel-coated dishes (Corning). When the confluence reached 70%, cells were passaged at a ratio of 1:5, or cryopreserved in a freezing solution containing 50% medium, 40% FBS (Gibco), and 10% dimethyl sulfoxide (DMSO, Sigma Aldrich). Cells were dissociated into a single-cell suspension using TrypLE (Gibco) for 3 min at 37°C, and plated into media containing Rock inhibitor Y-27632; media was replaced within 24 h. Ultimately, cells were harvested from the 6-cm dish, and genomic DNA was extracted with the DNeasy ® Blood & Tissue Kit (Qiagen) for WGS analysis. Furthermore, all stem cell lines underwent routine mycoplasma screening and karyotyping.

Gametes retrieval and in vitro fertilization (IVF)
Semen samples were collected by masturbation after 3−5 days of abstinence. Semen was kept at 37°C for 30 min for liquefaction, followed by an established density-gradient separation method. Briefly, after the second centrifugation, the pellet was resuspended with 0.5 mL of G-IVF (Vitrolife) and incubated for standard swim-up for 30 min. The supernatant was used for insemination. For the acquisition of oocytes, cumulus-corona oocyte complexes were isolated from the follicle fluid accurately and rapidly, and then cultured in G-IVF for 3 h. Next, each oocyte was inseminated in 4-well plates with approximately 100,000 motile spermatozoa. Approximately 18-20 h after fertilization, clinically abandoned three pronuclei (3PN) embryos were collected for experiments.

Sperm cryopreservation and thawing
Semen samples were collected by masturbation from donors into sterile containers after 3-5 days of sexual abstinence and left to liquefy at 37°C. Semen samples were placed in 15 mL centrifuge tubes, diluted 1:1 with cryoprotectant in a slow dropwise manner, and gently mixed to form a homogeneous solution. The homogeneous solutions of semen and cryoprotectant were equally aliquoted into 0.25 mL straws. Then, the straws were placed 4 cm (−170ºC to −180ºC) above the surface of liquid nitrogen for 15 min before being transferred into liquid nitrogen for preservation.
Straws containing homogeneous solution of semen and cryoprotectant were warmed in a water bath at 37°C for 1 min. The solution was then washed with 1 mL prewarmed G-MOPS medium at least twice. The mixture was centrifuged at 360 ×g for 5 min to discard the supernatant, and then the pellet was resuspended in 100 μL G-MOPS medium, and stored at 37°C until it was used for intracytoplasmic sperm injection (ICSI).

Generation of 2PN embryos by ICSI
Immature metaphase I (MI) oocytes were collected from patients for IVF or ICSI treatment. All oocytes were retrieved with informed consent from the donor or patient. MI oocytes were cultured in a maturation medium (TCM199 + 10% FBS + 10 μg/mL sodium pyruvate + 10 μg/mL FSH + 5 μg/mL LH + 1 μg/mL E2 + 1 nmol/L melatonin) until the first polar body appeared. Mature metaphase II (MII) oocytes were placed into a 50 µL micromanipulation droplet of HTF (modified human tubal fluid) with HEPES 10% buffer solution. The droplet was covered with tissue culture oil, and the dish was then mounted on the stage of an inverted microscope (Olympus IX73) equipped with a stage warmer and Narishige micromanipulators. Oocytes were fertilized by ICSI using frozen and thawed sperm. Fertilization was determined approximately 18 h after ICSI by noting the presence of two pronuclei and extrusion of the second polar body.

In vitro transcription of CBE-variant mRNA and sgRNA
The pX330 and pCMV-CBE-variant plasmids were constructed in our lab. The pCMV-CBE-variant plasmids were linearized, purified, and the template was in vitro transcribed using the mMES-SAGE mMACHINE T7 Ultra kit (Life Technologies). T7 promoter and sgRNA target sequence were added to sgRNA template by PCR amplification of pX330, using primers listed in Table S3. The T7-sgRNA PCR product was purified, and the template was in vitro transcribed using MEGA shortscript T7 Kit (Life Technologies). The sgRNA and CBE-variant mRNAs were purified using the MEGAclear kit (Life Technologies) and eluted in RNase-free water. In vitro transcribed RNAs were aliquoted and stored at −80°C until use. Prior to microinjection, the CBE-variant mRNA and sgRNA mixture was prepared by centrifuging for 10 min at 14,000 rpm at 4°C and then transferring the supernatant to a fresh 0.2 mL PCR tube for injection.

Single blastomere genotyping analysis
About 48 h after injection, zonae pellucidae from 8 to 16 cell stage embryos were removed by brief exposure to acidic Tyrode solution (Sigma Aldrich). Zona-free embryos were briefly (30 s) exposed to the 0.05% Trypsin-EDTA solution (Gibco) before manual disaggregation into single blastomeres with a small bore pipette. Subsequently, individual blastomeres were placed into PCR tubes with 2.5 μL embryo lysis buffer (0.1% Tween-20, 0.1% Triton X-100, 20mg/mL proteinase K, and ddH 2 O, 1:1:3:5) and incubated at 56°C for 30 min, followed by heat inactivation at 95°C for 10 min. PCR amplification was performed using nested primer sets and Phanta Max Super-Fidelity DNA Polymerase (Vazyme Biotech Co., Ltd). The first round PCR program was set as follows: 95°C for 30 s, 60°C for 30 s, and 72°C for 30 s, with a final extension at 72°C for 5 min. The second round PCR was performed using 0.5 μL PCR product as the template and nested inner PCR primers, and carried out with the same program as the first round. The PCR product was analyzed by Sanger sequencing or targeted deep sequencing to detect efficiency of base editing. It was regarded as homozygous (0% or 100% editing efficiency) for wildtype or edited genotypes when sequencing results for each blastomere detected only one type of allele. Moreover, it was regarded as heterozygous (33.3% or 66.7% editing efficiency in 3PN embryos and 50% editing efficiency in 2PN embryos) for edited genotypes when sequencing results for each blastomere detected two types of allele. PCR primers for genotype analysis are indicated in Table S3.

Off-target analysis with targeted deep sequencing
About 50-60 h after ICSI, half of the 8-cell embryos were injected with a mixture of GFP mRNA, FNLS-YE1 mRNA and sgRNA, and the other half were not injected as a control. Individual GFP + and GFPblastomeres were transferred into 0.2 mL PCR tubes containing 4 µL PBS and placed into a freezer at −80°C until further use. Whole genome amplification of the blastomeres was then performed using the REPLI-g Single Cell Kit (Qiagen). Briefly, samples frozen at −80°C were thawed and transferred into PCR tubes containing reconstituted buffer D2 (7 μL), and then incubated at 65°C for 10 min before the addition of stop solution (3.5 μL) and master mix (40 μL), followed by incubation at 30°C for 8 h. The DNA preparation was diluted with ddH 2 O (1:30), and 1 μL of the diluted DNA was used for PCR analysis and targeted deep sequencing.
Target sites were amplified by nested PCR from genomic DNA using Phanta Max Super-Fidelity DNA Polymerase (Vazyme Biotech Co., Ltd). The paired-end sequencing of PCR amplicons was performed with GENEWIZ Co., Ltd using the NovaSeq 6000 platform. The sequencing data were subsequently demultiplexed using fastq-multx (v1.4.1) with the PCR primers. Next, sequence alignment was performed between the demultiplexed sequencing data with each of the on-and off-target sites using CRISPResso2 (v2.0.32), and mapping statistics was generated using in-house scripts with Perl (v5.26.2) and R (v4.1.0).
Potential off-target sites of sgRNAs were predicted as previously reported (Bae et al., 2014). We extracted the top off-target sites with no >3 mismatches for PCSK9 and APOE sgRNA. We then mapped the nonsynonymous SNVs and frameshift indels to the predicted off-targets by chromosome positions. PCR primers for off-target analysis are indicated in Table S3.

Single-cell RNA sequencing analysis
Individual human blastomeres were manually picked, lysed, and subjected to one-step cDNA synthesis and amplification following the Smart-seq ® HT instructions (Takara). The sequencing library was constructed (Vazyme Biotech Co., Ltd), quality checked and sequenced with paired-end 150-bp reads on an Illumina NovaSeq 6000 platform. After quality checks of all libraries, 77 single-cell libraries passed our criteria and were subjected to deep sequencing. Raw reads of single-cell RNA-seq data were first trimmed and aligned to the GRCh38 human transcriptome (STAR v2.7.4a, two-pass mode). After de-duplication, RNA SNVs from individual cells were identified using the GATK pipeline (v4.2) for discovering variants in RNA-seq data. Briefly, the alignments spanning introns were first reformatted by SplitNCigarReads. Base quality was then re-calibrated using BaseRecalibrator and SNVs were called using HaplotypeCaller. Those SNVs detected in single cells with read depth ≥ 20.0, Fisher strand values ≤ 30.0, RMS mapping quality ≥ 20, and quality by depth values ≥ 2.0 were retained for downstream analysis. Further filtering was done by removing the SNV sites that overlapped with the dbSNP (v150) database downloaded from NCBI.
As the mRNAs were converted into cDNA before sequencing, both the nucleotide and its complementary base could be sequenced. For example, if there is a C in the mRNA, the cDNA has both C and G at the specific site. When the reference genome was C, the sequence would be read as C, and if the reference was G at the site, G would be read. Therefore, we counted the sum of C to T + G to A mutations for the editing of FNLS-YE1 and FNLS-BE3. Any high confidence variants found in un-injected blastomeres in each 2PN embryo were considered to be SNPs and were filtered out from the only GFP, FNLS-YE1/GFP-injected, and FNLS-BE3/GFP-injected groups for off-target analysis. The editing rate was calculated as the number of mutated reads divided by the sequencing depth for each site. Gene expression was quantified as log 2 (reads per kilobase of transcript per million mapped reads (RPKM) + 1) using HTSeq (v0.13.5).

WGS analysis
WGS was carried out using Illumina NovaSeq 6000 at mean coverage of 100×-150×. Trimmomatic (v0.39) was used to trim low quality reads and adapter sequences from the FASTQ files. Qualified reads were mapped to the human mitochondrial reference genome (hg19) by BWA (v0.7.12) with mem -M, and Picardtools (v2.3.0) was used to reorder, sort, add read groups, and mark duplicates of the aligned BAM files. Then, Mutect2 (v4.1.5), Lofreq (v2.1.2), and Strelka (v2.7.1) were run separately on the aligned BAM file of hESCs for de novo SNVs detection, with blood samples from both parents as control. Whole genome de novo indels were detected using Mutect2 (v4.1.5), Scalpel (v0.5.3), and Strelka (v2.7.1) in the same way (Cibulskis et al., 2013;Fang et al., 2016;Saunders et al., 2012;Wilm et al., 2012). For example, if the WT allele is G at a certain coordinate, the hESC carries A, and blood samples from both parents carry G, then the mutant A will be called a de novo mutation. In contrast, if blood samples from both parents carry A, then the mutant could not be identified.
Briefly, Strelka was first used to identify the genome-wide SNVs and Indels. The regions 200 bp upstream and downstream of the variants identified by Strelka were treated as candidate regions. Second, Lofreq and Mutect2 were separately used to calculate SNVs of the candidate regions identified by Strelka. Scalpel and Mutect2 were then separately used to identify the indels of the candidate regions. The adjacent 400 bp regions of these variants were also used as candidate regions for Indel detection by Scalpel.
Only variants identified by all three algorithms and with >10% allele frequencies were used for the following analysis. To strictly control the quality of the variants, we removed variants that overlapped with UCSC repeat regions or that were reported in the dbSNP151 database.
PEM-seq was performed as described previously (Yin et al., 2019). In brief, 20 μg genomic DNA was extracted and sonicated to about 300-700 bp. 5ʹ biotinylated primer was placed within 200 bp from the cleavage site. For primer extension, the process was performed as follows: 95°C 3 min; 95°C 2 min, T a (annealing temperature) 3 min, 5 cycles; T a 3 min. Then, Bst polymerase 3.0 (NEB) was added to perform primer extension: 65°C 10 min, 80°C 5 min. Excessive biotinylated primers were depleted by 1.2× AxyPrep Mag PCR Clean-Up beads (Axygen). Purified products were heated to 95°C for 5 min and then quickly chilled on ice for 5 min for DNA denaturation. Biotinylated PCR products were enriched by Dynabeads ™ MyOne ™ Streptavidin C1 (Thermo Fisher Scientific). Then, PCR products on Streptavidin C1 beads were washed twice with 400 μL 1× B&W buffer [1 mol/NaCl, 5 mmol/L Tris-HCl (pH 7.4), and 1 mmol/L EDTA (pH 8.0)] followed by washing with 400 μL dH 2 O. The DNA-beads complex was then resuspended with 42.4 μL ddH 2 O. In the bridge adapter ligation step, the ligation reaction was performed in 15% PEG8000 (Sigma Aldrich) with T4 DNA ligase (Thermo Fisher Scientific) at room temperature overnight. Ligation products were washed twice with 400 μL 1× B&W buffer, then the 400 μL ddH 2 O was added, and the products were resuspended with 80 μL dH 2 O. The beads-DNA complex underwent on beads nested PCR (Taq, Transgen Biotech) with I5 and I7 sequence primers for 16 cycles. Then PCR products were recovered by size-selection beads (Axygen) followed by PCR (Fastpfu, Transgen Biotech) tagged with Illumina P5 and P7 sequences. All PEM-seq libraries were subjected to WGS performed at Macrogen using an Illumina HiSeq X Ten at a sequencing depth of 300 million reads. PEM-seq analysis and off-target hotspot identification were calculated using a previously described scoring system (Yin et al., 2019).

Statistical analysis
All statistical values are presented as means ± SEM. Differences between datasets were considered to be significant at P value <0.05. All statistical tests were conducted with the unpaired student's t-test (two-tailed), unless otherwise stated.