ARISING FROM H. Ma et al. Nature 548, 413–419 (2017); https://doi.org/10.1038/nature23305
The development and application of methods to prevent the transmission of damaging mutations through the human germ line would have considerable health benefits. In an attempt to correct a paternal pathogenic mutation using CRISPR–Cas9 technology in human embryos, Ma et al.1 assert that the maternal allele is an efficient repair template for gene correction, including when Cas9 is applied in metaphase II (MII) oocytes. As the maternal and paternal genomes undergo distinct developmental programs and are in separate nuclei before the first mitotic division, which would seem to preclude inter-homologue interactions, we believe that it is crucial to provide a comprehensive analysis of the molecular outcomes of double-strand break (DSB) repair in human embryos. In the absence of direct molecular evidence for the inferred events, the consideration of using such methods for correction of the human germ line should proceed with extreme caution. There is a Reply to this Comment by Ma, H. et al. Nature 560, https://doi.org/10.1038/s41586-018-0381-y (2018).
Ma et al.1 use two approaches to attempt gene correction in human embryos. In one approach, which is deemed to be more promising because it is thought to give rise to non-mosaic embryos, MII oocytes were injected with donor sperm from a heterozygous mutation carrier together with Cas9 complexes to direct the cleavage of the mutant paternal allele. About 72% of embryos arising from Cas9 injection were thought to be wild type compared with 50% of control embryos. The authors argue that this excess of apparently wild-type embryos (22%) arose by correction of the paternal allele, by using the maternal allele as a repair template, a process termed inter-homologue homologous recombination (abbreviated here as IH-HR).
In the other approach, Ma et al.1 again used sperm from the mutation carrier to fertilize wild-type oocytes; when the pronuclear-stage zygotes were completing S phase, they were injected with Cas9 complexes, again directed to the mutant paternal allele. In contrast to the previous approach, embryos derived from fertilization with mutant sperm could be conclusively identified because mosaic embryos were obtained. Some cells of these mosaic embryos contained a mutant paternal locus, either unmodified or with small indels, together with the wild-type maternal allele. Other cells in these mosaic embryos contained only a detectable wild-type allele. The authors inferred that these cells arose by IH-HR of the mutant paternal allele using the wild-type maternal allele as a template, leading to gene correction.
Considering the data presented in Ma et al.1, alternatives to IH-HR are possible. Genotyping involved the amplification of an approximately 534-base-pair (bp) fragment in which the MYBPC3ΔGAGT mutation is approximately 200 bp from one of the primer-binding sites. Deletions larger than 200 bp would be sufficient to remove this primer-binding site and lead to amplification of only the maternal allele (Fig. 1a, b), giving the misleading appearance of gene correction of the paternal allele. Although typically not as common as small indels, long deletions and other events have been detected in cultured cells and in both mouse and pig zygotes2,3,4. To detect longer deletions, a matrix of primer pairs needs to be tiled at increasing distances from both sides of the mutation; linkage analysis performed on the long-range PCR products would confirm whether amplification is from both the maternal and the paternal chromosomes. In a study designed to score these events systematically, Cas9-induced double-strand breaks in mouse embryonic stem cells were found to resolve into large deletions (250–9,500 bp) in approximately 20% of edited cells5. This approach remains imperfect to detect all events, however, because very large deletions or other events such as inversions, translocations, chromosome loss and large insertions prevent amplification and thus will escape characterization. Indeed, in 19% of cells edited at an autosomal locus, only one of two alleles could be recovered5. These various outcomes of repair of a DSB could result in genotypes incompatible with normal development, and therefore need to be reliably identified to exclude affected embryos.
Wild-type genotypes in a PCR assay can also arise by the activation of the egg during Cas9 injection, but without successful integration of a sperm genome, resulting in haploid or diploid parthenogenetic cells containing only the maternal genome6 (Fig. 1c). A paternal contribution was verified by cytogenetic analysis in some of the stem-cell lines generated from embryos by Ma et al.1, but the authors did not determine whether wild-type stem-cell lines were from wild-type sperm, or arose by gene correction.
To directly demonstrate gene correction by IH-HR, evidence for a new linkage of maternal and paternal alleles—that is, through the incorporation of the wild-type sequence from one of the maternal homologues into the mutant paternal chromosome at the site of the DSB—is required (Fig. 1a). New DNA linkages can be determined by phased DNA sequencing, or by long-range PCR using allele-specific primers7,8. Such haplotype analysis is particularly crucial in the case of the embryos derived from MII-phase oocyte injections, because which embryos were derived from sperm carrying the mutant allele was not determined.
Although IH-HR in fertilized oocytes and zygotes cannot be excluded, there are several obstacles to this mechanism. IH-HR after the induction of a DSB in mitotic mammalian cells has been described11, and was also seen in a recent study using CRISPR–Cas9 in embryonic stem cells5, though it was less frequent than inter-sister HR or NHEJ. In mammals, IH-HR is essential for the reductional division to form gametes, and is promoted by the large number of DSBs that are programmed to form on each chromosome9. It is important to note, however, that meiotic IH-HR occurs during fetal development in females10 and so it is temporally removed from the events described in Ma et al1.
The physical separation of maternal and paternal genomes in fertilized eggs would be expected to be a substantial impediment to IH-HR during the first cell cycle. After fertilization, distinct maternal and paternal nuclei form (pronuclei), such that the two genomes are separate in a cell that is more than 100 µm in diameter (Fig. 1d–g). This separation may prevent the incorporation of paternal chromosomes into the oocyte MII spindle (Fig. 1e). During the first interphase, maternal and paternal pronuclei migrate from the site of their formation towards the centre of the zygote, but their integrity persists throughout interphase (Fig. 1f, g), during which individual nuclei can be manipulated12. In both human and mouse zygotes, maternal and paternal genomes undergo DNA replication in separate nuclei, and enter the first mitosis as separate entities, at which time they can still be individually manipulated (Fig. 1f, g). Microtubule action assembles maternal and paternal genomes on a common metaphase plate at the first mitosis13,14, although they remain in distinct groups15. Therefore, direct contact between maternal and paternal genomes required for inter-homologue repair does not seemingly occur until the first mitosis or later when embryos enter the two-cell stage and the two genomes are packaged within the same nucleus. Relative to the application of Cas9, this is 24–30 h after the MII injections, and 6–12 h after the zygotic injections.
It is important to note the different outcomes Ma et al.1 obtained with regards to mosaicism depending on whether CRISPR–Cas9 was injected into zygotes (frequent mosaicism) or together with sperm into MII-phase eggs (lack of apparent mosaicism). Mosaicism from zygotic injection is consistent with DSB repair occurring after DNA replication, whereas the lack of mosaicism after MII-phase injection indicates that repair occurs before DNA replication and thus before the first mitosis. Consistent with this, the injection of CRISPR–Cas9 together with sperm in mouse MII oocytes results in non-mosaic modification of the paternal genome within only 3 h owing to NHEJ during sperm chromatin decondensation16. Notably, the maternal genome seems to be refractory to editing during this time of meiotic exit. Different outcomes depending on the time of CRISPR–Cas9 injection indicate that gene editing in the paternal genome during decondensation or after the formation of a nucleus may involve different repair mechanisms.
In summary, the direct verification of gene correction and exclusion of other possible outcomes is an imperative for any embryo that would be considered for future implantation. Gene editing has the potential to reduce disease-causing alleles, but inadvertent changes to the human germ line, including rearrangements, long deletions, and loss of heterozygosity, for example, from IH-HR, could have serious consequences that affect development, predisposition to cancer and fertility. Our discussion of Ma et al.1 demonstrates the need for a more comprehensive characterization of the repair mechanisms in the early embryo, and identifies a key challenge for the therapeutic use of gene editing in the human germ line: the development of reliable assays to distinguish between different repair outcomes when DNA is limiting.
Mouse oocytes were obtained from 5–7-week-old B6D2F1/J females (Jackson Laboratories stock 10006) by superovulation. Oocytes were removed from oviducts 14 h after injection of human chorionic gonadotropin, and fertilized with mouse sperm injection by ICSI on an inverted Olympus IX73 equipped with a Narishige micromanipulator. Images were taken using an Olympus IX73 equipped with an Olympus DP80 camera (Fig. 1f) or a Zeiss 710 confocal microscope at indicated time points (Fig. 1e). Immunostaining was performed using a monoclononal beta tubulin antibody (clone AA2, Millipore 05-661, dilution 1:1,000, lot 2370698) in PBS with 10% FBS for 3 h at room temperature. Secondary antibody Invitrogen 488 donkey anti mouse (A21202) at 1:500 dilution in PBS for 45 min at room temperature. Hoechst 33342 was used for DNA staining at 5 μg ml−1 (Life Technologies H3570). All animal research was reviewed and approved by the Columbia University IACUC, and performed in accordance with animal use guidelines and applicable ethical regulations.
All available data are included in this manuscript and available from the corresponding authors upon reasonable request.