The two accompanying Comments1,2 highlight the results of our recent study3 and their implications for understanding the basic biology of DNA repair and for potential future therapies. Given the importance of the conclusions of our study3, further verification was requested to rule out alternative interpretations. One argument made by Egli et al.2 was that the repair of the mutant paternal allele using maternal-homologous sequences is highly unlikely based on the assumption that in early zygotes, parental genomes are physically separated in paternal and maternal pronuclei. This temporary isolation would therefore preclude homologous chromosome interactions required for homology-directed repair (HDR). However, it should be noted that CRISPR–Cas9 ribonucleoprotein (RNP) specific to the mutant paternal allele was delivered into pronuclear-stage zygotes or even earlier during fertilization in our original study3, while the subsequent readouts of targeting and repair outcomes were measured three days later in multicellular embryos3. In late mammalian zygotes, paternal and maternal pronuclei migrate towards each other with subsequent nuclear envelope breakdown and formation of a diploid mitotic spindle4,5. Thus, from this point onwards, parental homologues are presented with ample opportunities to physically interact and recombine. Although we do not exclude the possibility of initial targeting and induction of double-strand breaks (DSBs) in metaphase II (MII) oocytes or zygotes at the time of CRISPR–Cas9 injections, HDR or non-homologous end-joining (NHEJ) could have occurred later during subsequent three mitotic cell cycles. Indeed, we showed that each mosaic 4–8-cell embryo contained blastomeres with two or more different repair outcomes suggesting that CRISPR–Cas9 remains active well beyond the pronuclear stage.
On the basis of the assumption that HDR in pronuclear stage zygotes is impossible, Egli et al.2 and Adikusuma et al.1 suggested that our results could also represent ‘loss’ in the detection of the mutant paternal allele altogether owing to large deletions. Such genetic lesions could prevent PCR primer binding and amplification, thus subsequently escaping detection. Adikusuma et al.1 indicated the possibility of CRISPR–Cas9 inducing large deletions (greater than 100 base pairs (bp)) at the targeted region, particularly, when disrupting both parental alleles simultaneously. Previous mouse studies showed that the most frequent deletions induced by a single single-guide RNA (sgRNA) do not exceed 10 bp and occurrence of larger deletions are rare to account for the high rates of HDR (above 50%) observed in our experiments6. We designed and pre-tested different sgRNAs in patient induced pluripotent stem cells (iPSCs) and selected one with high specificity for the mutant sequence with no evidence of large deletions detected, making it unlikely, in our view, that our selected sgRNA would induce a large deletion at the frequency we observed for HDR. A recent report also demonstrated that pre-testing of several candidate sgRNAs in human embryonic stem (ES) cells could be effective in predicting editing efficacy of disrupting both copies of POU5F1 in human embryos7. In this study, the most frequently observed on-target editing in CRISPR–Cas9-microinjected human embryos were small (2–3 bp) indels. Only one embryo contained a few blastomeres with uncommonly large 330-bp deletions7.
Although species differences may have impacted editing outcomes, Adikusuma et al.1 did not report pre-testing candidate sgRNAs. Moreover, we used CRISPR–Cas9 RNP, whereas Adikusuma et al.1 used Cas9 mRNA, which may have accounted for large deletions.
Nevertheless, to rule out the possibility of large deletions, we decided to carry out a large-scale re-testing of all embryonic blastomere samples from our published study. Originally, we used PCR amplification followed by Sanger sequencing of a 534-bp fragment spanning approximately 250 bp in each direction from the MYBPC3ΔGAGT mutation site. To detect larger deletions, we designed an additional 8 pairs of long-range PCR primers amplifying various lengths of fragments surrounding the MYBPC3ΔGAGT mutation locus ranging from 493 bp to 10,160 bp (PCR1–PCR8 in Fig. 1a; Extended Data Table 1). First, we re-tested 8 blastomeres with wild-type maternal and paternal genotypes (WT/WT) from the 4 mosaic embryos S-phase-injected with CRISPR–Cas9 (Extended Data Table 2), along with 4 WT/WT and 4 wild-type maternal and mutant paternal (WT/Mut) blastomeres from the non-injected control embryos (Extended Data Table 3). PCR products were separated on 1% agarose gels. In all 16 samples, primers PCR1, PCR2, PCR4 and PCR5 amplified a single band of the expected size (Fig. 1b–e). For PCR6 and PCR7, several faint bands of smaller size were also detectable in some corrected and control blastomeres (Fig. 1f, g). However, Sanger sequencing of these faint bands did not produce any readable products suggesting non-specific PCR primer binding. Next, we performed 2 additional long-range PCR amplifications with PCR3 and PCR8 primers on all remaining WT/WT blastomeres (n = 35) from the 13 mosaic embryos along with controls (Extended Data Table 2 and 3). Amplification with PCR3 produced a single band of the expected 1,742 bp size in all experimental and control blastomeres (Fig. 1h). For PCR8, in addition to a major band matching the expected 10,160 bp size, a few faint smaller size bands were also visible in some targeted and control samples, but again, Sanger sequencing indicated non-specific primer binding (Fig. 1i).
We next screened for larger deletions in the M-phase-injected embryos and randomly selected one blastomere from every WT/WT embryo (n = 41) as all individual blastomeres within each embryo in this group carried identical MYBPC3 genotypes. We also tested the only mosaic embryo (M2-WT42) in this group that contained 3 blastomeres with WT/WT genotypes and 4 blastomeres with WT/ssODN (Extended Data Table 4). Again, long-range PCR screening of all samples with primers PCR3 and PCR8 produced a single band of expected 1,742 bp or 10,160 bp size, that is, failing to detect large deletions (Fig. 1j, k).
We also examined whole-exome sequencing (WES) results for large deletion in the 6 human ES cell lines derived from M-phase-injected embryos. Comparisons of the area 5 kb downstream and 5 kb upstream from the mutation site in ES cells and the corresponding egg and sperm donors revealed no differences in sequencing depth, consistent with the absence of any large deletions (Extended Data Fig. 1).
We then designed two additional 16-kb and 20-kb PCR primers in an attempt to screen for even larger deletions; however, these primers failed to produce a detectable response. All DNA samples were extracted from single blastomeres and then pre-amplified by whole-genome amplification. It is likely that whole-genome amplification results in smaller size DNA segments not compatible for amplification with 16-kb and 20-kb PCR.
In summary, all these tests failed to detect the presence of large deletions up to ±5 kb from the mutation site in CRISPR–Cas9-treated human embryos. Although the PCR primers used in this study did not identify much larger deletions, available evidence suggests that most deletions induced by CRISPR–Cas9 should have been detected with our assays. The use of multiple sgRNAs targeting several sites may produce large deletions of up to 24-kb DNA segments; however, the use of single sgRNA has resulted in smaller deletions of less than 600-bp DNA in mouse embryos6.
Our original results suggest that DSB repair on the paternal allele governed by maternal homologue-based HDR extends to the adjacent ΔGAGT deletion site resulting in conversion of the paternal sequence (see extended data figure 2a of the original study3). Therefore, we asked whether DNA proofreading and mismatch repair mechanisms involved in HDR could also contribute to the conversion of neighbouring neutral paternal SNPs resulting in loss-of-heterozygosity (LOH) within the MYBPC3 locus. We postulated that paternal single nucleotide polymorphisms (SNPs) adjacent to the targeted DSB locus would be converted to become maternal-like, while more distant polymorphic sites would be preserved. We searched WES and whole-genome sequencing (WGS) datasets and identified three informative parental SNPs within the MYBPC3 gene distinguishing egg donor 1 from the sperm donor. SNPs #1 (rs2071304) and #2 (rs2856650) were located downstream of the ΔGAGT deletion site (−7,959 bp and −781 bp), whereas SNP #3 (rs2856653) was +3,335 bp upstream from this locus (Fig. 2a). We next genotyped individual blastomeres of the two CRISPR–Cas9-injected mosaic embryos (Mos2 and Mos3 in Table 1) from this parental combination. The ES cell line (ES-C1) derived from the control non-injected, blastocyst from the same parental combination was also genotyped. ES-C1 with the WT/WT genotype at the mutation locus and two blastomeres, Mos2.3 and Mos3.2, from the mosaic embryos with the WT/NHEJ genotype were heterozygous at all three polymorphic sites representing the expected maternal and paternal SNPs (G/C, T/C and G/A for SNPs #1, #2 and #3, respectively) (Table 1 and Fig. 2b). By contrast, the two blastomeres Mos2.1 and Mos3.1 with the WT/HDR genotypes from the same mosaic embryos were homozygous for all three SNP sites carrying exclusively maternal nucleotides. Notably, another blastomere, Mos2.2, also with a WT/HDR genotype, was homozygous at the SNP #2 locus carrying maternal nucleotides, but heterozygous at both SNP #1 (G/C) and SNP #3 (G/A), indicating preservation of these paternal SNP sites (Table 1 and Fig. 2b). These results support the notion that HDR-based conversion can expand beyond the targeted mutant loci, resulting in loss of neutral paternal SNPs across the MYBPC3 region. However, the acquisition of maternal SNPs by something other than long conversion tract cannot be completely ruled out.
In contrast to the mosaic counterparts, the MYBPC3 genotype of the original sperm in uniform WT/WT embryos produced from CRISPR–Cas9-treated zygotes or oocytes cannot be determined. Nevertheless, we suggested that some embryos with MYBPC3WT/WT genotypes could have originated from mutant MYBPC3ΔGAGT sperm, with subsequent HDR correction of the deletion. This original assumption was based on a significant increase in the percentage of WT/WT embryos in the CRISPR–Cas9-treated group compared to non-treated controls3. We reasoned that loss of neutral paternal SNPs in some of these WT/WT embryos could be used as evidence of repair of the mutant MYBPC3ΔGAGT. Among 42 WT/WT, M-phase-injected embryos, six (M2-WT28 to M2-WT33 in Table 1) were derived from the egg donor 1 and the sperm donor, and thus should be heterozygous at the SNP #1, #2 and #3 sites. We randomly genotyped two sister blastomeres from each of these six embryos, and documented LOH in at least one of these polymorphic sites in four embryos. As expected, paternal SNPs were lost at these loci, resulting in homozygous maternal nucleotides (Table 1 and Fig. 3). Notably, genotypes of two sister blastomeres from the same embryo were distinct from each other, suggesting independent HDR events probably occurred at the two-cell stage or later. For example, one blastomere (M2-WT29.3) in embryo M2-WT29 was homozygous at all three SNP loci carrying exclusively maternal nucleotides while the other sister blastomere (M2-WT29.2) was heterozygous at all three SNP sites (Table 1 and Fig. 3). A similar pattern was also observed in embryos M2-WT 30 and 32. By contrast, one blastomere (M2-WT31.1) of embryo M2-WT31 was homozygous containing maternal alleles at the SNP #2 and #3 (T/T and G/G, respectively), whereas the more distant SNP#1was heterozygous (G/C). Its sister blastomere M2-WT31.2 was heterozygous at these three SNP positions.
As indicated above, LOH associated with erasure of paternal SNPs in these four uniform WT/WT embryos provides support for repair of the mutant sperm MYBPC3ΔGAGT deletion following CRISPR–Cas9 treatment. Additional genotyping of distant SNPs throughout chromosome 11 from all parental contributions would be necessary to establish the median conversion tract length. All examined blastomeres in the remaining embryos, M2-WT28 and M2-WT33, were heterozygous at all three SNP sites, suggesting that these embryos were fertilized by wild-type sperm.
We extended SNP analysis to four uniform WT/WT embryos from the S-phase-injected group from the same parental combination. Three embryos (WT4, WT5 and WT6) were heterozygous for all three SNPs, whereas both blastomeres examined from WT3 embryo were heterozygous at SNP #1 and #3 but homozygous at SNP #2 (Table 1 and Fig. 3). Thus, this embryo was probably generated from the mutant sperm but subsequently corrected by HDR using the wild-type maternal allele.
To provide further genetic evidence for HDR, we also screened egg donor 2 and identified two informative SNPs within the MYBPC3 gene that would differentiate from the paternal contribution. Egg donor 2 was homozygous (G/G) at the SNP #4 site (positioned −6,189 bp downstream of the ΔGAGT mutation, rs2697920), whereas the sperm donor was heterozygous (A/G) at this locus (Extended Data Fig. 2a). At SNP #5 (+9,514 bp, rs11570115), both parents were heterozygous A/G. We initially genotyped blastomeres with WT/NHEJ or WT/Mut genotypes from seven mosaic embryos (Mos1, Mos7, Mos8, Mos10, Mos11, Mos12, and Mos13) derived from this parental combination, and found that six were heterozygous A/G at the SNP #4 locus (Table 2, footnote symbol b), indicating that mutant sperm contributed the ‘A’ allele at this locus in these embryos. We next sequenced all sister blastomeres with WT/HDR genotypes from these six embryos and found that five (Mos1, Mos7, Mos8, Mos10 and Mos13) contained one or more blastomeres that lost the paternal allele and became homozygous G/G at SNP #4, supporting the gene conversion from maternal allele (Table 2 and Extended Data Fig. 2b). The remaining WT/HDR blastomeres in these embryos retained the paternal allele and were heterozygous A/G at the SNP#4, probably indicating a shorter conversion tract. Both WT/HDR blastomeres from Mos11 embryo were heterozygous A/G at SNP #4. WT/Mut blastomere from Mos12 embryo was homozygous G/G at SNP #4, precluding the need for further genotyping determinations (Table 2).
We next sequenced one randomly selected blastomere from each of the seven uniform WT/WT, M-phase-injected embryos generated from the egg donor 2 and the sperm donor. We found all seven blastomeres were homozygous G/G at SNP #4 (Table 2). In comparison, six out of seven S-phase-injected mosaic embryos generated from the same parental combination were heterozygous A/G (Table 2). Therefore, we think it is possible that some of these G/G homozygous embryos in the M-phase-injected group also lost paternal SNPs owing to gene conversion. Genotyping for the SNP #5 locus showed that two mosaic S-phase-injected embryos (Mos1 and Mos8) were heterozygous A/G and informative for conversion analyses (Table 2). All five sister blastomeres with WT/HDR genotypes in the Mos1 embryo were homozygous G/G at SNP #5, indicating loss of paternal SNPs. Of the two WT/HDR blastomeres in Mos8 embryo, one was homozygous G/G and one was heterozygous A/G at SNP #5. Among M-phase-injected embryos, one out of seven was heterozygous A/G and the remaining six were homozygous G/G at SNP #5 (Table 2).
Together, these results suggest that gene conversion in human embryos induced by HDR may happen and extend considerable distances in both directions from the original target site, resulting in LOH associated with erasure of neutral paternal SNPs. The length of the conversion tract varied among individual blastomeres even from the same embryo. The existence of polymorphic sites and retention of paternal SNPs on some corrected blastomeres also provides a strong suggestion that the mutant paternal MYPBC3 locus was repaired in embryos injected at the S phase and M phase.
In our original study3, we demonstrated that early exposure to CRISPR–Cas9 RNP during fertilization (M phase) could considerably reduce or completely eliminate mosaicism in cleaving embryos. These results are irrespective of whether repair occurred via HDR or NHEJ because mosaic embryos may include blastomeres with different NHEJ-derived indel genotypes. Egli et al.2 dispute our interpretation and speculate that the decrease in mosaicism in our M-phase-injected group could be due to fertilization failure resulting in parthenogenetic development of oocytes. We showed that only 1 out of 58 (1.7%) cleaving embryos produced by M-phase-injection was mosaic, and 16 out of 58 (27.6%) were uniformly heterozygous, carrying NHEJ-derived indels in the mutant paternal allele (MYBPC3WT/∆GAGT-indel; figure 3b of the original study3). These heterozygous embryos could not originate from parthenogenesis because they all carry the paternal MYBPC3 deletion. By contrast, when CRISPR–Cas9 was injected one day after fertilization into late S-phase zygotes, 13 out of 54 (24%) embryos were mosaics (figure 2a of the original study3). Egli et al.2 argue that the increased yield of WT/WT embryos (22%) in the M-phase-injected group could also be due to parthenogenetic development. Of 75 M-phase-injected oocytes, 2 were lysed during intracytoplasmic sperm injection and 10 failed to fertilize. The remaining 63 (84%) exhibited normal fertilization morphology with two pronuclei and two polar bodies, inconsistent with parthenogenic activation. Similar results were obtained from non-injected controls and S-phase-injected embryos (extended data table 2 of the original study3). Moreover, SNP analyses provided in Tables 1 and 2 for WT/WT embryos in the M-phase-injected group clearly demonstrate retention of paternal SNPs. To further exclude the possibility of parthenogenetic development, we confirmed the paternal contribution in WT/WT ES cell lines derived from M-phase-injected embryos by short tandem repeat (STR) assay. As expected, all six ES cell lines derived by M-phase injection and one non-injected control (extended data figure 3a of the original study3) contained both maternal and paternal STR alleles (Extended Data Table 5). Thus, in all samples we analysed, paternal contribution was detected and parthenogenesis could be ruled out.
Mounting evidence suggests that the two parental homologues provide more than a genetic diversity contributed by parents8. Recent developments in custom-designed nucleases allowing selective targeting of one of the two parental alleles have provided evidence for inter-chromosomal pairing, interaction and contribution to DNA repair across plant and animal species. Among the possible interactions are DNA DSB repair governed by mitotic recombination or homologue-template-based repair contributing to LOH9. A more recent study using mutant tomato plants with different fruit colours concluded that in heterozygous plants, CRISPR–Cas9-induced DSBs in the targeted allele were repaired using the intact allele as a template at a frequency up to 14% and that HDR between homologues occurred in the absence of the meiotic machinery10. Specific targeting of the mutant paternal allele in heterozygous mice also demonstrated that DSB repair, via HDR using the wild-type maternal allele, resulted in the birth of viable WT/WT offspring11. DSB induction in both parental alleles simultaneously could also induce template-mediated repair using endogenous genomic sequences from close homologous gene families. In human zygotes, CRISPR–Cas9 based bi-allelic targeting of the β-globin gene (HBB) resulted in HDR using the endogenous δ-globin gene (HBD)12.
On the basis of our original results3 and those present here, we suggest that human embryos have the capacity for non-meiotic homologous chromosome-based DNA repair. This endogenous repair competence must be further explored, reproduced with different founder mutations, and perhaps evaluated for future germline gene therapeutic applications. Many questions remain concerning the precise mechanisms involved during homologous chromosome-based HDR and cell cycle timing. However, given that such DNA repair seems widely conserved across different species, in depth mechanistic studies can potentially be addressed in model organisms. Indeed, a recent article in bioRxiv confirmed our findings in the mouse using more rigorous analyses13. Wilde et al.13 validated high frequency of inter-homologue HDR mechanism in mouse heterozygous zygotes and demonstrated a significant increase of this process by complementing with the HDR-associated strand exchange factor RAD51. In the interim, we hope that the questions raised by Egli et al.2 and Adikusuma et al.1 and our new results presented here will contribute to a better understanding of the complex nature of DNA repair and serve as a useful platform for further discussions and studies.
Regulations for studies involving human research participants
Oregon Health & Science University (OHSU) has established a strong regulatory framework that allowed for approval and continued monitoring of this study protocol. Approval was granted by all relevant regulatory bodies; OHSU Institutional IRB (IRB), OHSU Scientific Review Committee (SRC), OHSU Innovative Research Advisory Panel (IRAP), and Data Safety Monitoring Committee (DSMC). These committees are informed by national and international guidelines published by the Hinxton group, the International Society for Stem Cell Research (ISSCR) and the National Academy of Science and Medicine committee reports.
Healthy gamete donors were recruited locally via print and web-based advertising and underwent ovarian stimulation at OHSU. A single sperm donor with a heritable MYBPC3 mutation was identified by physicians from the OHSU Knight Cardiovascular Institute and referred to the research team. Written informed consent was obtained from all participants before study-related procedures. The current study complied with all relevant ethical regulations.
Long-range PCR and Sanger sequencing
Long-range PCR (PCR1, PCR2 and PCR4–PCR7) was performed using PrimeSTAR GXL DNA Polymerase, whereas the long-range PCR3 and PCR8 were performed with TaKaRa LA Taq DNA Polymerase (Clontech) according to manufacturer’s procedure. In brief, PCR conditions were 10 s at 98 °C, 15 s at 60 °C, and 1 min kb−1 at 68 °C (30–35 cycles). PCR products were resolved with 1% agarose gel electrophoresis and were visualized with EtBr staining.
For Sanger sequencing, targeted region PCR for each single nucleotide polymorphisms (SNPs) was carried out using the PCR Platinum SuperMix High Fidelity Kit (Life Technologies). The PCR products were Sanger sequenced and analysed by Sequencher v5.0 (GeneCodes).
Parentage analysis by STR assay
DNA was extracted from blood of egg and sperm donors and individual ES cell lines using commercial kits (Gentra). STR microsatellite parentage analysis was conducted by the Genetics Laboratory at University of California, Davis as described previously14.
SNP searching and calling using WES and WGS
WES sequencing data were first processed by filtering adaptor sequences and removing low quality reads or reads with a high percentage of N bases using SOAPnuke (1.5.2) software (http://soap.genomics.org.cn/) developed by BGI. Clean reads were generated for each library. Clean data were paired-end aligned using the Burrows-Wheeler Aligner15 (BWA) program version 0.7.12 to the human genome assembly hg19. Duplicate reads in alignment BAM files were identified using MarkDuplicates in Picard v1.54 (https://broadinstitute.github.io/picard/). The alignment results were processed by RealignerTargetCreator, IndelRealigner and BaseRecalibrator modules in GATK16 (3.3.0) and variants detection was performed by HaplotypeCaller tool in GATK according to GATK Best Practices recommendations17,18. SNV and InDel information were extracted and filtered by VQSR in GATK and annotated by AnnoDB v3 (http://www.igm.columbia.edu/resources/bioinformatics).
The datasets including Sanger sequencing, STR, WES and WGS generated and analysed during this study are not publicly available to protect the identity and privacy of study participants. However, the data will be available from the corresponding author upon individual requests and after OHSU IRB/DSMC approvals.
The authors acknowledge the OHSU Institutional Review Board (IRB), Innovative Research Advisory Panel (IRAP), Scientific Review Committee (SRC) and Data Safety Monitoring Committee (DSMC) for oversight and guidance on this study. We thank the Sequencing Core of the Vollum Institute at OHSU for assistance with Sanger sequencing. Studies conducted at OHSU were supported by the OHSU institutional funds. Work in the laboratory of J.S.K. was supported by the Institute for Basic Science (IBS-R021-D1). Work in the laboratory of J.C.I.B. was supported by the G. Harold and Leila Y. Mathers Charitable Foundation, the Moxie Foundation and the Leona M. and Harry B. Helmsley Charitable Trust. Work at BGI was supported by the Shenzhen Municipal Government of China (DRC-SZ  884).