SERIES | Perspective

SERIES | Perspective
https://doi.org/10.1038/s41556-019-0424-0

Human germline genome editing

Rebecca A. Lea    and Kathy K. Niakan   *

With the advent of efficient, easy-to-use genome editing by CRISPR–Cas9, editing human embryos is now possible, providing
tremendous opportunities to study gene function and cell fate in early human development. The technique can also be used to
modify the human germline. Unresolved questions about pre-implantation human development could be addressed by basic
research using CRISPR–Cas9. In this Perspective, we discuss advances in human genome editing and consider ethical questions
and potential clinical implications of this technology.

T
he ability to simply and efficiently target any region of the Utility of genome editing to understand early human
human genome using the clustered regularly interspaced development
short palindromic repeats (CRISPR)–CRISPR-associated Although the above work aimed to assess the preclinical capabil-
(Cas) gene editing technology1–3 has transformed biology. Targeted ity of CRISPR–Cas9 to correct pathological sequences in human
gene disruption or modification has been achieved in a variety of embryos11,12,16, gene editing has also been applied to investigate the
cells, and diverse uses of catalytically inactive, ‘dead’ Cas9 have basic regulation of early human embryogenesis18. Our lab recently
been developed, including localising fluorescent tags and epigenetic used CRISPR–Cas9-induced insertion or deletion (indel) mutations
regulation of target genes4. The ease of generating target-specific to disrupt the pluripotency factor OCT4 (gene name POU5F1) in
guide RNAs (gRNA) compared to engineering other programmed human zygotes donated as surplus to infertility treatment18. The
nucleases (zinc finger nucleases5, transcription activator-like effec- function of OCT4 has been thoroughly investigated through tradi-
tor nucleases6 and meganucleases7) made the notion of modifying tional genetic approaches in mice19,20 and proven essential for main-
the human germline genome more practicable. Thus, only two years tenance of the pluripotent inner-cell mass by inhibiting acquisition
after CRISPR–Cas9 was described as a molecular biological tool8,9, of trophectoderm fate. With CRISPR–Cas9 technology, we showed
it was applied to human embryos with the intent of assessing the that the function of OCT4 in early embryogenesis is not conserved
clinical feasibility of gene correction10–12. among mice and humans18. Murine Pou5f1-/- embryos develop to
These initial studies focused on understanding how CRISPR– the blastocyst stage but consist of only extraembryonic CDX2-
Cas9 performs in human embryos, assessing mutation efficiency, expressing trophectoderm cells. However, the POU5F1-targeted
off-target editing, rates of mosaicism and compatibility with con- human embryos showed a substantial defect in blastocyst forma-
tinued preimplantation development10–12 (Table 1). Many studies tion, related at the single-cell level to a defect in the formation of all
used non-viable tripronuclear embryos to abate ethical objections. three cell lineages18. Intriguingly, a recent paper applying CRISPR–
However, the use of such material, which would be clinically dis- Cas9 to ablate POU5F1 in bovine embryos21 revealed a strikingly
carded and wherein DNA repair mechanisms may be dysfunctional, similar phenotype to that reported in human. These findings high-
makes it impossible to interpret experimental outcomes, as such light that although rodent studies will continue to be transformative
embryos develop abnormally13–15. Most groups focused on achieving to our understanding of mammalian development, certain species-
homology-directed DNA repair (HDR) to introduce designed edits specific aspects can only be ascertained by performing functional
into the human germline genome, as proof of principle for clini- studies directly in human embryos and non-rodent species. These
cal application10–12,16. However, as we will discuss, many technical results emphasise the power of genome editing to enable the study
limitations exist that make HDR-based editing of human embryos of gene function in previously inaccessible developmental contexts.
difficult, as evidenced by the generally low efficiency of mutation
‘repair’ to date10–12. Alternative methods to understand gene function in
The specificity and efficiency of CRISPR–Cas9-based genome human development
editing is ever-improving17. However, several fundamental aspects Though alternative techniques such as RNA or protein knockdown,
of human development, including timing of early cell cycles and pharmacological inhibition or provision of exogenous dominant-
mechanisms of DNA damage repair, have yet to be elucidated and negative or overexpression constructs can allow modulation of gene
will determine how the human embryo responds to CRISPR–Cas9- expression without modifying the germline genome, there are sev-
based genome editing. Basic research into these mechanisms will eral factors that make CRISPR–Cas9 revolutionary. Foremost is the
be fundamental to improving our proficiency in human-embryo ease with which genes of interest can be targeted for highly efficient
genome editing, which will lead to a better understanding of our and specific disruption, as the reliability of gene knockout reduces
own early biology and inform the debate about potential safe and the number of human embryos required to come to meaningful con-
effective clinical uses of this technology. clusions about gene function. Furthermore, small-molecule inhibi-
Below, we summarise the progress already made in applying tors often have low specificity, perhaps affecting a whole family of
CRISPR–Cas9 to human embryos and consider the current limi- signalling receptors22. The branched nature of signalling pathways
tations to more ambitious applications of this technology, as well complicates analysis, as downstream readouts may reflect effects on
as ethical implications. We discuss potential clinical applications parallel pathways. Additionally, RNA knockdown studies often do
of human germline genome editing, proposing a workflow for safe not recapitulate the phenotype of that seen with a full genetic knock-
and efficacious medical research. Further, we look to the promising out23–25, owing to a combination of effects including off-target gene
future of this technology in elucidating fundamental aspects of early modulation26,27, activation of a p53 response28,29, incomplete inhi-
human biology. bition of gene expression and compensation by redundant genes30.

Human Embryo and Stem Cell Laboratory, The Francis Crick Institute, London, UK. *e-mail: kathy.niakan@crick.ac.uk

Nature Cell Biology | www.nature.com/naturecellbiology

 Table 1 | A summary of human embryo CRISPR–Cas9-mediated genome editing experiments
Basic Preclinical
biology

Technique NHEJ NHEJ with HDR Base editing

two guides
Mode of Indels Large Precise genetic changes Single base-pair changes
editing deletions
Perspective | SERIES

Starting 2PN zygotes 3PN zygotes 2PN zygotes MII-phase 3PN zygotes Nuclear 2PN
material (from ICSI with het sperm donor)b oocytes transfer zygotes
during ICSI embryos (from ICSI
with het
sperm)b
Gene target POU5F1 CCR5 HBB HBB G6PD HBB G6PD MYBPC3 MYBPC3 RNF2 HBB FANCF HBB FBN1
On-target 45% of 15.4% 6.7% 9.7% HDR 10% HDR 20% HDR 25% HDR 100% HDR 44.4% 5.9% HDR 87.5% C-T 36.8% with 100% with 40.9% G-A 100% G-A
efficiency cleavage- with Δ32 with Δ32 with ssDNA proposed with ssODN conversion desired targeted conversion conversion
(≥1 allele)a stage mutation mutation of total HDR with around nonsense mutation (58.8% at correct
embryos of total of total amplified maternal target site mutation include additional position
with analysed analysed alleled mutations)
complete
editingc
% non- 0% 50% 0% 0% 0% 100% 0% 50% 0% 0% Not 0% 17.6% 0% 90.9%
mosaicism reported
Off-target None None None 100% Not Not Not None in None None in two Not Not One off-target in Not None
effects analysed reported reported reported one embryo in three embryos reported reported 33.3% analysed reported
by whole analysed embryos analysed by WGS
exome-seq analysed
Reference Fogarty et al., Kang et al., 2016 Liang et al., Tang et al., 2017 Ma et al., 2017 Li et al., 2017 Zhou et al., 2017 Liang et al., Zeng et al.,
2017 2015 2017 2018
a
If multiple experiments were performed, the maximum efficiency is reported. Efficiency is calculated as the number of embryos containing at least one allele with the desired edit out of the total number of embryos used in that experiment. bWhere a heterozygous sperm
donor is used to create embryos carrying a mutation to be ‘repaired’, efficiency is calculated as the number of embryos containing at least one allele with the desired edit out of the total number of confirmed mutant embryos. cOf 11 embryos arrested or collected during
the cleavage-stage of development (pre-compaction), 5 had no detectable wild-type POU5F1 alleles across all cells. dIn this experiment, it has not been proven unequivocally that HDR with the maternal allele as a repair template has occurred (see main text). NHEJ, non-
homologous end joining; HDR, homology-directed repair; 2PN, two pronuclear; 3PN, tripronuclear; ICSI, intracytoplasmic sperm injection.

Nature Cell Biology | www.nature.com/naturecellbiology

NaTURe Cell BIology
NaTURe Cell BIology SERIES | Perspective
An alternative is to use the Trim-Away technique to target pro- overrepresented48,49. Intriguingly, elements of the circadian clock are
teins for ubiquitin-mediated degradation31. Although innovative, it also upregulated in 8-cell embryos, suggesting that cleavage divi-
is unclear whether this method could lead to more than transient sions may proceed periodically, without accounting for the DNA
knockdown of a protein of interest, and nuclear-localised factors status of cells and without requirement for external growth factors48.
may be difficult to target efficiently because of their compartmen- Lack of checkpoint activity is also suggested by the observation that
talisation. In general, any exogenously introduced construct is lim- even amongst embryos with multinucleated blastomeres, caspase
ited by perdurance. For these reasons, genetic knockouts are the activity is rarely detected50. The absence of cell cycle checkpoints
gold-standard for understanding gene function. Such experiments is reminiscent of early pre-EGA embryos of organisms such as the
are now possible in the context of human embryogenesis owing to frog and fly51, and may be advantageous if lack of checkpoint activ-
the simplicity and efficiency of CRISPR–Cas9 technology. ity avoids p53-dependent arrest and cell death seen in somatic cells.
Conversely, it is easy to imagine how such a situation would lead to
Challenges of genome editing in human embryos increased levels of DNA damage and aneuploidy as development
Rates of development of human embryos following assisted repro- progresses, perhaps leading to eventual arrest around the blastocyst
ductive technologies, such as in vitro fertilisation (IVF), are poor. stage, when apoptosis reportedly becomes more active52.
Approximately 42% of fertilised zygotes reach the blastocyst stage32,
and only 51.4% of those lead to successful implantation when trans- DNA repair following CRISPR–Cas9
ferred to a patient33. The initial wave of developmental arrest occurs The above considerations with regard to DNA repair raise the
during the period of cleavage division at the 4-to-8-cell stage, coin- question of how and when Cas9-induced DSBs are repaired in the
ciding with the timing of embryonic genome activation (EGA)34,35. embryo. Although the method of DSB repair depends on cell cycle
Therefore, problems with the initiation of EGA, or within the gene stage53, attempts have been made to bypass cell cycle dependency by
complement of the embryo due to aneuploidy, are most likely to introducing exogenous repair factors or inhibitors to favour HDR.
manifest in arrest at this stage. Moreover, as the optimal in  vitro This has been a promising approach in many cell-types, through
culture requirements for human embryos are still unknown, this inhibition of 53BP1 (ref. 54) or overexpression of RAD51 in human
can further exacerbate issues of development, and promote embryo pluripotent stem cells55 and mouse zygotes56, and may also benefit
arrest, despite normal EGA36,37. human embryo genome editing.
The second point of developmental drop-off can be attributed to To date, the efficacy of introducing designed mutations into
problems around the blastocyst stage, as a properly formed blasto- human embryos by HDR has been very low, and the formation of
cyst must expand and ‘hatch’ to interact with the maternal endome- indel mutations predominates despite provision of a repair tem-
trium and implant for further development38. Natural pregnancies plate10–12,16. Although evidence points towards the expression of
are also thought to often end at this stage39, though the reasons why crucial HDR components such as RAD50 in human embryos47,
are unclear and likely diverse. Of those embryos that do implant whether expression of such components is sufficient for functional
successfully, only roughly 47.7% make it past the third month repair is unknown. Additionally, genome editing of human embryos
of gestation33. is typically performed by microinjection of CRISPR–Cas9 compo-
nents into fertilised zygotes, suggesting that DSBs may occur (and
Aneuploidy presumably be repaired) long before EGA and thus under the con-
Aneuploidy, wherein a cell contains an incorrect chromosomal trol of maternally deposited factors. There is simply not enough
complement, is quite prevalent in human embryos40, initiating data at present to understand the capability of early blastomeres to
either during meiosis or during cleavage in the embryo itself41,42. It repair DNA.
is estimated that 5–20% of human oocytes have undergone incor- An alternative explanation for poor HDR success is lack of con-
rect meiosis43, but mitosis during embryogenesis may be even more trol over when DSBs occur. Microinjection can be performed into
error-prone44,45. The high rates of aneuploidy, persisting through- the cytoplasm or into the two visible pronuclei (Fig. 1a). Pronuclear
out preimplantation development and even beyond46, suggest a appearance and fading typically herald the beginning and end of the
decreased checkpoint activity during embryonic cleavages com- zygotic S phase, respectively, with subsequent progression to G2 and
pared to somatic cell divisions. This point, discussed further below, mitosis57,58. However, depending on the approach (i.e., Cas9 mRNA
is important to consider in the context of CRISPR–Cas9-mediated or protein), there can be a variable delay between injection, tran-
mutagenesis, as human embryos may not respond to DNA damage scription and translation (approximately a 6–12-h delay for Cas9
in a similar way to well-studied cellular models. On the other hand, mRNA versus protein), DSB formation and Cas9 protein degrada-
it seems possible that some level of karyotypic abnormality may be tion (approximately>72 h post-Cas9 mRNA injection compared to
compatible with successful development or that there are corrective 24 h for protein)59. This process may be further hindered by con-
mechanisms in place. densation of sister chromatids in preparation for mitosis, potentially
making the Cas9 target sequence inaccessible, as chromatin state
Expression of cell cycle and DNA repair factors has been demonstrated to impact Cas9 activity60,61. Thus, it is dif-
Indeed, DNA repair may be active in early human embryos47, but ficult to determine when in the cell cycle the DSBs will be formed,
control of cell cycle progression appears to differ greatly from that and therefore whether the HDR machinery would be available for
in somatic cells48,49. Comparative gene expression analysis between repair. One could envision using the geminin-fused Cas9 con-
high- and poor-quality embryos revealed substantial overexpression struct62 in human embryos to promote HDR over non-homologous
of genes involved in DNA repair and cell cycle control in embryos end joining by limiting DSB formation to the S, G2 and M phases.
carrying complex aneuploidies, including RAD50, a component The timing of microinjection has proven useful for favouring
specific to DNA double-strand-break (DSB) repair47. Conversely, HDR in mouse embryos63. Coordinating the injection of CRISPR–
no enrichment of genes involved in cell cycle checkpoints or apop- Cas9 components with EGA, which seems to promote a long HDR-
tosis was observed in these inferior embryos, implying that DNA permissive G2 phase, vastly improved the success of HDR-based
repair is attempted without cell cycle arrest47. Consistent with this gene editing in mouse embryos63. However, it is yet to be deter-
finding, canonical checkpoint genes RB1 and WEE1 are expressed mined whether attempting CRISPR–Cas9 microinjection around
at low levels in human 8-cell embryos, whereas genes involved in the time of human EGA will have the same dramatic effect on the
promoting cell cycle progression (e.g., CCNE1 (encoding cyclin E), cell’s choice of repair mechanism, nor is it known when there is a
CCNB1 (encoding cyclin B), CDC25B, and MYC) are comparatively prolonged G2 phase in human embryos (Fig. 1a).

Nature Cell Biology | www.nature.com/naturecellbiology

Perspective | SERIES NaTURe Cell BIology

a
Extended G2 to increase HDR?

Or Or

Fertilisation Compaction Cavitation

EGA
MII oocyte Zygote 2-cell 4-cell 8-cell Morula Blastocyst
at ICSI

ICSI + CRISPR CRISPR microinjection CRISPR electroporation

b
NANOG GFP
Reporter alleles

Live imaging Dissociation

CDX2 CFP

SOX17 RFP
3-colour Fluorescence signal
blastocyst distinguishes cell identity

OCT4 AID Auxin

E2
Auxin-inducible degron

PolyUb Rbx1
Rbx1 Treatment
AID Cul1
Cul1 AID
OCT4
OCT4 Skp
Skp TIR1
TIR1

OCT4 protein
TIR1 degradation
Safe-harbour
locus

NANOG Cre NANOG Cre

Cre
Cre-lox recombination

Cre
OCT4 OCT4
loxP loxP Onset of Recombination OCT4-/-
Floxed OCT4 NANOG expression

NANOG+ NANOG+
OCT4+ OCT4–

Fig. 1 | Techniques for introducing and utilising genome editing of human embryos. a, CRISPR–Cas9 components, as either a ribonucleoprotein complex
or DNA or RNA templates, can be introduced into human embryos at various stages using differing methods: injection coincident with fertilisation by
intracytoplasmic sperm injection (ICSI) in MII (metaphase II stage) oocytes; microinjection at the zygote stage, into the cytoplasm or the two pronuclei;
or electroporation, which may open the door to genome editing at later stages. Whereas introducing CRISPR–Cas9 during a long G2 phase in mouse
embryogenesis apparently vastly improves rates of homology-directed DNA repair (HDR)63, it is unclear whether and when there is a corresponding event
in human preimplantation development. EGA, embryonic genome activation. b, Examples of how HDR-based genome editing could be applied in future
basic research to generate insight into human preimplantation embryogenesis. Top, targeting of fluorescent reporters to lineage-specific genes (e.g.,
NANOG for the epiblast, SOX17 for the primitive endoderm and CDX2 for the trophectoderm) to investigate cell fate in human blastocysts. This could be
useful for maintaining location and cell identity information following single-cell dissociation. Middle, applying the auxin-inducible degron (AID) system
to genetically tag a gene of interest with an AID sequence and introduce a constitutively expressed TIR1 construct into a safe-harbour site for auxin-
inducible degradation of a target protein114. TIR1 will form a complex with endogenous ubiquitin ligase components (E2, Rbx1, Cul1, Skp) and facilitate
specific ubiquitin-mediated degradation of a protein of interest tagged with AID. Shown is an example of AID tagging of the locus encoding OCT4.
OCT4 is normally expressed in all cells of the blastocyst. Upon auxin treatment of the embryo, OCT4 protein would be reversibly destroyed, allowing
for assessment of its function with temporal specificity. PolyUb, polyubiquitylation. Bottom, leveraging the Cre–lox recombination system to study gene
function in a temporally controlled manner by inducibly deleting a gene of interest flanked by loxP sites (a ‘floxed’ gene)115,116. In this example, the OCT4
locus is ‘floxed’. Introduction of a Cre recombinase transgene under the control of the NANOG promoter leads to specific deletion of OCT4 within the
pluripotent NANOG-positive epiblast cells. OCT4 expression is therefore lost only from the epiblast, allowing assessment of its pluripotency-specific role.

Detailed information about the timing of early cell cycle pro- development, live embryo imaging and cell cycle reporters such as
gression in human embryos is not available. To understand whether the FUCCI system64,65 or chromosome painting (achieved recently
a similar prolonged G2 phase exists in human preimplantation using dCas9 (ref. 66)) would be required. Further consideration is

Nature Cell Biology | www.nature.com/naturecellbiology

NaTURe Cell BIology SERIES | Perspective
needed about the current limitations of such experiments includ- The use of base editors in human embryos
ing both the difficulty of imaging fairly large human preimplanta- In addition to conventional knockouts, CRISPR–Cas9-mediated
tion embryos (the human blastocyst is approximately 0.1–0.2 mm base editing can be used to more precisely alter the genome while
in diameter, ~4× larger than the mouse blastocyst) and maintain- bypassing the need for DSBs that may allow unintended modifica-
ing their normal in  vitro development in a microscope chamber, tions to genomic DNA75,76. Given how little is understood about
as well as the potential toxicity of introducing cell cycle reporters. the repair of genetic lesions in human embryos, base editing is a
Such experiments have, however, been achieved in the mouse by promising alternative, and has been applied recently in this con-
tracking fluorescently labelled chromosomes (H2B-mCherry) and text77–80 (Table 1). As with traditional CRISPR–Cas9, these studies
kinetochores (EGFP-Cenpc)67 or by genetically introducing the had a preclinical focus on disease-associated mutations that could
FUCCI system68. be repaired to wild-type sequences with base editing, providing a
An alternative approach for introducing Cas9-gRNA ribonu- proof of principle of the technique’s utility in the human germline.
cleoprotein complexes and a repair template into oocytes at the These studies detected little or no off-target activity, but variable
second meiotic metaphase (MII) coincident with fertilisation by efficiency of on-target base conversion, with some conversion of
intracytoplasmic sperm injection (ICSI) has recently been sug- nearby nucleotides, some unexpected conversion patterns (e.g.,
gested to increase HDR in human embryos16. The rationale for this C-to-A or C-to-G instead of C-to-T deamination) and a low pro-
experiment was that introducing CRISPR–Cas9 at this early stage portion of indel mutations77–80. Moreover, recent studies have sug-
would lead to DSB induction and repair prior to the first cell divi- gested unexpectedly high off-target editing in rice and in mouse
sion, and indeed, the investigators report a dramatically reduced embryos81,82, necessitating further evaluation in a human context. If
incidence of mosaicism, though this differs from a previous report base editing in human embryos can be shown to be reliable, it could
in mice69. However, the authors also report a considerable increase be an immensely powerful tool for introducing null mutations into
in HDR, with the proportion of homozygous wild-type embryos genes to study their function during early development, as demon-
increased from 66.7% by zygote microinjection to 72.4% by MII strated in the mouse83. Alternatively, it may be possible in the future
injection. Still, 20–30% of targeted embryos carried indels at the to use primed editing, which uses a primed editing guide RNA and
on-target site, regardless of whether a repair template was pro- a modified Cas9-nickase fused to an engineered reverse transcrip-
vided16. This finding may severely limit applicability in a future tase84. Though this method has been used successfully to edit muta-
clinical context, because mosaic embryos may harbour undesired tions in human cells, it has yet to be tested on mammalian embryos,
indel mutations. and mouse embryo studies would be an informative next step.

Unexpected editing outcomes following CRISPR–Cas9 in Available tools for evaluating success
human embryos When genetically engineering model systems, undesirable alleles
Interestingly, in the study discussed above16, the increase in muta- can be selected against to yield an organism that has only the desired
tion-free embryos was not attributed to HDR using the exog- genotype. Working with any human material, however, makes this
enous repair template, but rather was suggested to be the result impossible, and whereas undesired effects may be accounted for in
of interhomologue repair using the wild-type maternal allele16. basic research, they must be strictly identified and avoided in clini-
This suggestion has been met with scepticism by some research- cal applications. Additionally, working with the human germline
ers, with suggestions that alternative explanations may account for presents its own unique set of obstacles. In particular, it is important
the observed result, including allele dropout (ADO), partheno- to consider the small amount of DNA that can be used for ‘diag-
genetic oocyte activation or the introduction of undetected large nosis’ of genome editing events in embryos. In the laboratory, it is
deletions or rearrangements70,71. One reason for these objections possible to use single-cell analysis of whole, dissociated embryos to
is the physical separation of male and female genetic material by assess the range of genotypes introduced following CRISPR–Cas9-
nuclear membranes during the very early stages of post-fertilisa- mediated mutagenesis16,18. However, if preparing genome-edited
tion development70, whereby the parental pronuclei remain dis- embryos for subsequent implantation and establishment of preg-
tinct throughout the first interphase and DNA replication. Recent nancy, preimplantation genetic diagnosis (PGD) would be required,
experiments in mouse zygotes have suggested that the separation typically involving a relatively non-invasive biopsy of a small num-
may persist even longer, with the formation of independent mitotic ber of trophectoderm cells from blastocysts85,86.
spindles72. Ma and colleagues have provided data in support of For more than two decades, the potential for mosaicism to
their interpretation of interhomologue DSB repair by performing impede accurate PGD has been a concern43. A recent study in sheep
long-range PCR and SNP analysis73, but outstanding questions still compared genotyping results from trophectoderm biopsies and
remain, and further studies will be required to determine whether bulk sequencing of the remaining cells from the same embryos fol-
the parental homologues possess the ability to interact prior to lowing CRISPR–Cas9 targeting of the PDX1 gene in MII oocytes
pronuclear fading. and found a distinct lack of correlation87. Less than 50% of troph-
The potential confounding factor of large CRISPR–Cas9-induced ectoderm biopsies were found to be concordant with sequencing
deletions74 also calls into question the interpretation of studies using of the remaining embryo, with a trend for underestimating the
end-joining mechanisms for gene knockout in human embryos and amount of unedited PDX1 (ref. 87). This could also be attributed to
other contexts. Accumulating evidence points towards a surpris- one of the other major issues encountered when sequencing low-
ing incidence of large DNA deletions or rearrangements resulting input DNA, which is ADO88,89. It is standard practice to perform
from CRISPR–Cas9 mutagenesis74. By analysing the consequences whole-genome amplification on low-input DNA samples, but such
of CRISPR–Cas9 targeting to various loci in mouse embryonic stem amplification often favours one allele over the other. Amplification
cells (ESCs) in depth, this study determined that the range of muta- bias can be random, or may clearly relate to allele length as for
tions is more complex than simple insertion or deletion, including PDX1 (ref. 87), in which the shorter mutant allele is preferentially
compound mutations and translocations of nearby sequences74. amplified. The apparent prevalence of large deletions74 makes the
Further, more than 20% of targeted alleles contained large deletions latter especially worrying.
that stretched up to 6 kb away from the CRISPR cut site. The authors It is currently a concern that if mosaicism or ADO leads to
also corroborated these results in additional human and mouse cell misrepresentation of embryo genotypes in a clinical setting, the
lines, suggesting a universality to the occurrence of dramatic DNA outcomes could be highly variable depending on the level of mosa-
lesions following genome editing74. icism and the nature of unidentified editing outcomes. It should be

Nature Cell Biology | www.nature.com/naturecellbiology

Perspective | SERIES NaTURe Cell BIology

considered that apparently correctly gene-edited embryos chosen Box 1 | Prerequisites for clinical human embryo genome editing
for transfer could yield unhealthy pregnancies if the percentage of
edited cells overall (or specifically in the disease-affected tissue) is
The ethics of human germline genome editing are widely dis-
too low to ameliorate symptoms or if there are large deletions or
cussed, both by scientists and the general public and media. In
rearrangements that could impact on nearby genes. An additional
light of recent events regarding the use of CRISPR–Cas9 on hu-
concern is that outcomes may be gene specific, both in terms of
man embryos bound for implantation and pregnancy, and the
the edited target and potential collateral damage on neighbouring
guidelines set out in documents published by scientific and ethi-
regions of the genome. The question of how to determine whether
cal bodies, the following points present what, in our view, are
a gene-edited embryo would be healthy is still an open one, and it
the most important prerequisites for any potential future use of
may be that our current technologies are insufficient to answer it
clinical human embryo genome editing:
at present.
1. The treatment must address an unmet medical need and be
Predicting on-target mutations in human embryos judged the most reliable method of safely sparing an indi-
One possibility to simplify the analysis of on-target editing is to vidual from genetic disease105,106;
use predictive algorithms to aid selection of highly specific gRNA 2. The alleles introduced into patients should exist naturally
sequences and preempt the likely mutation spectrum arising from in humans of a similar genetic background, given that it is
CRISPR–Cas9. Three recent papers demonstrate highly accurate very difficult to predict the potential for gene–gene interac-
predictive tools90–92. The success of these algorithms largely relies on tions to impinge upon overall phenotype, and the balance
training data from CRISPR–Cas9 experiments, and all report high of potential benefit to potential harm must be as certain as
cell-type specificity. Given that studies in human embryos are lim- possible105,106. This is relevant in modifying CCR5 for HIV
ited, it would be impossible to generate the required amount of data resistance, because inactivating mutations are sometimes
for accurate prediction. However, the mutation spectrum observed present in northern European populations but very rare in
in human ESCs may be closely reflective of the in  vivo embryo18. others, perhaps pointing towards negative selection131. This
Given this, it is interesting to note that Allen et  al. included plu- argument also concerns the persistence of sickle cell anae-
ripotent stem cell-types in their analyses and found higher rates of mia in African populations, wherein heterozygotes are pro-
large modifications and a prevalence of microhomology-mediated tected from malaria132. When carrying out genome editing
deletions91. Intriguingly, microhomology-mediated small deletions to prevent genetic disease might increase susceptibility to
are favoured at tandem repeats, leading to suggestions of possible other conditions, serious consideration should be given to
therapeutic routes for repeat diseases like Fragile X syndrome and the benefit-to-risk ratio on a case-by-case basis;
Huntington’s disease91,92. 3. There must be adequate proof of safety through in-depth
preclinical research, subject to peer and regulatory review,
Evaluating off-target effects investigating the precise genome edits to be made, and
The issues of mosaicism and ADO also diminish our ability to any off-target effects and how they might affect overall
evaluate off-target effects of CRISPR–Cas9 in human embryos. health105,106;
Although numerous techniques exist to scan for off-target editing 4. There should also be sufficiently robust methods in place to
genome-wide93–98, they usually involve detection of Cas9 cleavage assess on- and off-target effects in modified embryos105,106.
in vitro and give limited information about editing that has actu- Currently available methods are severely limited by small
ally occurred in cells. One exception is Digenome-seq93, in which sample size and mosacism (Table 1), therefore, it is currently
in vitro cleavage of genome-edited cellular DNA pinpoints potential not possible to unambiguously determine the genetic conse-
Cas9 off-target sites by the ‘straight’ alignment of high-throughput quences of genome editing in human embryos;
sequencing read ends flanking cut-sites, and the remaining ‘stag- 5. The informed consent process should always be discussed
gered’ reads spanning the sites can be interrogated for indel muta- with participants by an unbiased third-party agent, who is
tions. Circle-seq has the advantage of experimentally determining trained in the correct procedures and has no conflict of in-
sites of off-target mutations in primary human cells that have been terest, rather than by members of the scientific team them-
edited95 but requires amounts of genomic DNA (~25 μg) that would selves; and
preclude its use on preimplantation human embryos. Although 6. The work must be performed openly and conform to broad
recent methods have been developed to identify cut sites directly societal consensus about acceptable uses of this technol-
in vivo, BLISS97 uses cells and tissues fixed on slides and is therefore ogy. Scientific progress in areas as controversial as human
not applicable to whole embryos, and DISCOVER-seq98 relies on genome editing cannot proceed successfully without the
ChIP methodology, which is also greatly limited by chromatin yield trust and support of the public who it is designed to help,
from embryos. as proven by the general response following the surprising
Low-throughput methods of in silico prediction followed announcement in November 2018133.
by targeted sequencing can be used on PCR-amplified sites of
expected off-target cleavage, but it is possible that the reliability
of this method may be influenced by the use of low-input DNA, Possible clinical applications and alternatives to
particularly by mosaic editing. Additionally, off-target editing is genome editing
suggested to be disproportionately hindered by chromatin com- Even before any experiments applying CRISPR–Cas9 gene editing to
pared to on-target effects, making in vitro assessment less reliable99. human embryos had been confirmed, the scientific community was
Interestingly, however, a number of studies suggest that off-target well aware of the potential implications of using this technology in a
activity of CRISPR–Cas9 may be less extensive than initially feared clinical setting. Some suggested imposing a series of restrictions102,
in both cultured cells100,101 and human embryos16,18. Editing at the whereas others called for a complete moratorium of gene editing
one-cell stage may be advantageous in reducing the likelihood of in human embryos, especially because of fears that it may impact
off-target mutations compared to editing many more cells at later on the approval for somatic cell therapies103. Nevertheless, others
stages of development or in adults. For example, with an infrequent quickly came out in support of such research, stressing the potential
off-target mutation, the likelihood of incurring a mutation is higher benefit that could come from it and suggesting that the scientific
as more cells are targeted. community would indeed be remiss in denying future generations

Nature Cell Biology | www.nature.com/naturecellbiology

NaTURe Cell BIology SERIES | Perspective
Ineffective
Working
g1 g2 g3 Cas9–sgRNA
– + – + – + complex gRNA in vitro screen

Uncleaved amplicon

In silico
guide design Cleavage products

In vitro
cleavage assay On-target efficiency Off-target
• Miseq
• Sanger
Genome wide Targeted
• Digenome-seq, • In silico prediction
Circle-seq, etc. • Miseq
• Sanger
Non-human
primate

Mouse

Compatability with development

within 14-day limit
Screen in animal models
• Conservation of target?

Implantation
model On-target Off-target Phenotype?
efficiency • Screen for KO
• Analyse adult tissues
• Long term and intergenerational
follow-up
Chromosome copy number

4.00
Phenotyping if gene Genotyping
3.60
45XY(-chr16) + Directly measure mosaicism
expressed
3.20 + Simultaneous geno/phenotyping
2.80 • IF with G&T-seq
2.40 • RNA-seq
2.00 • Stem cell derivation + DNA is not limiting
1.60 − Not all cells of the embryo
1.20 are represented
0.80
0.40

+ More DNA than single cells

15
17
19
11
13
1

− Confounded by mosaicism
y
21
9
3

Position along each chromosome − Low-frequency events masked

• Low-pass WGS

Fig. 2 | A proposed workflow for preclinical evaluation of human embryo genome-editing experiments. gRNAs (e.g., g1, g2, g3) are designed using online
tools and then tested by in vitro cleavage or Sanger sequencing, revealing the (in)ability of any gRNA sequence to mediate on-target cleavage by Cas9.
Guides that clearly cut the on-target site are carried forward to cellular testing (e.g., human embryonic stem cells (hESCs)). Non-transformed, karyotypically
normal cells should be used to allow interpretation of editing outcomes. On-target and off-target editing can be assessed by targeted sequencing (Sanger or
MiSeq), whereas genome-wide methods (Digenome-seq, Circle-seq or whole-genome sequencing (WGS)) may be useful for detecting unexpected off-target
effects. gRNAs with high on-target efficiency and no evidence of off-target activity can be tested in preclinical models. Depending upon local availability
and target sequence conservation, mouse or non-human primate models may be useful. On- and off-target editing must be screened by DNA sequencing.
Although these events may differ between species, it is essential to evaluate the consequence of sequence changes, especially those that may not have
been previously detectable in the human population, in as many cellular contexts as possible. The phenotype of introduced edits must be assessed in animal
and human cell line models when possible. Alternatively, early expression would allow phenotyping in the embryo model, but adult and intergenerational
follow-up are also required to ensure no long-term adverse effects. If any guides pass these screening stages, preclinical studies would be performed in
human embryos. Compatibility of the desired edit with continued development should be assessed, perhaps up to the 14-d limit, using recently described
methods128,129. On-target efficiency and off-target editing is commonly assessed at the 8-cell or blastocyst stage. Genotyping can be performed at the whole-
embryo or single-cell level, or using a biopsy of a few cells. Pros and cons of these approaches are listed. Special consideration must be taken to account for
potential large deletions. Samples should be assessed for karyotypic abnormalities that may potentially be induced following CRISPR–Cas9-mediated DSBs. If
targeting a gene that is expressed early, preliminary phenotyping in the human embryo (e.g., by immunofluorescence (IF), RNA-seq or G&T-seq130) should be
performed. Alternatively, hESCs could be derived from edited embryos and differentiated to a relevant cell type to investigate phenotype.

the chance to drastically reduce the burden of genetic disease104. Engineering and Medicine105 and the UK Nuffield Council on
Regardless of the stance taken, there is almost universal agreement Bioethics106 (Box 1).
that any potential application of clinical germline genome editing
must come only after in-depth public and policy discussions, and Production of disease-free embryos
should fit a number of strict ethical and safety criteria, as laid out One common argument against the use of genome editing is that
in documents published by the US National Academies of Sciences, IVF with PGD is an already available route to ensuring that couples

Nature Cell Biology | www.nature.com/naturecellbiology

Perspective | SERIES NaTURe Cell BIology

carrying disease genes can bear disease-free children. However, For inducible control at the gene level, the Cre-lox system112,113
in our view, there are notable benefits to trying to correct disease could be applied, using CRISPR–Cas9 to ‘flox’ genes for subsequent
alleles, rather than selecting for disease-free embryos. recombination-based deletion. However, the efficiencies required
For instance, when both parents are carriers of recessive or for this system to work reliably in human embryos may preclude
dominant genetic disease, the number of embryos eligible for trans- its use.
fer to establish pregnancy is greatly reduced if avoiding affected
and carrier embryos. Given the low in vitro development rates of Future potential clinical use
human embryos, and the rates of successful pregnancies following The announcement of the birth of ‘CRISPR babies’ on the eve of
IVF (~12% of fertilised eggs), it is likely that the number of cycles the second international human genome-editing summit in Hong
needed to yield a healthy, disease-free child for such parents is mul- Kong has led to renewed calls for a moratorium on human germ-
tiplied107. This can present future parents with the possibility of line genome editing114. However, it has also apparently inspired oth-
repeated ovarian stimulation and IVF cycles, which comes with the ers to follow in the same footsteps115. We would thus like to stress
risk of complications108–110. the immense importance of proper regulation and oversight of all
This situation is exacerbated in the case of autosomal dominant future endeavors in this field, and provide a suggested workflow
disorders, which may not present until later in life and therefore for rigorous preclinical evaluation (Fig. 2). Since November 2018,
do not always preclude an affected person from reaching repro- great strides have been made to ensure that any potential future
ductive age. In addition, advances in modern medicine mean that use of reproductive genome editing is thoroughly vetted prior to
more patients of once severely life-limiting diseases are surviving approval and is keenly scrutinised while underway. China will soon
to adulthood, increasing the likelihood of passing on disease-asso- introduce regulations to guarantee that accountability for gene edit-
ciated dominant alleles. In cases in which one parent carries two ing in humans lies with those performing the technique, outlaw-
dominant disease-causing alleles, it would be impossible at present ing any experiments that undermine an individual’s well-being or
for them to produce a healthy, genetically related child. This may dignity116. China also recently drafted specific regulations regarding
lead some to pursue IVF with donor gametes, an option which may the approval process for editing human embryos117. Interestingly,
not always be appealing to prospective parents. Japan had previously released draft guidelines with a more permis-
One must also consider the potential morally objectionable sive stance on human embryo genome editing, which did not out-
aspects of current and prospective reproductive technologies. At law germline editing for reproduction118. However, in the wake of
present, selection against disease-carrying embryos involves either the Hong Kong announcement, government officials are now dis-
destruction before implantation or initiation of pregnancy fol- cussing the introduction of strict, unambiguous legislation to pre-
lowed by elective termination after in  utero diagnosis. With safe vent the implantation of genetically modified human embryos119.
and effective human germline genome editing, it should be consid- Further, the World Health Organisation has introduced an expert
ered that many fewer (or feasibly no) embryos would be destroyed advisory committee to oversee global standards in human genome
based on their genotype. Ultimately, providing more options for editing, who quickly suggested the establishment of a global registry
patients empowers them to make the choice that is best for their for such research, to widespread approval120.
family and circumstances. Given the existing limitations in our understanding of the biology
of the human embryo and how the germline genome might inter-
Conclusions and future outlook act with CRISPR–Cas9-mediated DSBs, we and many others feel
Future non-clinical basic biology. We envisage that future devel- that the time to pursue such research clinically is not at hand121,122.
opments in gene editing technology and our understanding of its Much more work is required to solidify our knowledge of the basic
mechanism inside human embryos will open up many more avenues biology of human development before we consider introducing
for understanding the biology of early human development (Fig. 1b). genome modification into the repertoire of treatments offered to
In particular, improving our ability to perform HDR would pro- IVF patients. One must ensure that the outcome will be the birth
vide the opportunity to introduce a variety of genetic changes, such of healthy, disease-free children, without any potential long-term
as genetic tags as reporters of gene expression63. In many species, complications. The regulatory handling of mitochondrial replace-
reporter alleles linked to fluorescent markers have revealed the ment therapy, also a form of heritable genome editing that affects
precise temporal and spatial expression profiles of genes of inter- only the mitochondrial DNA123,124, should be taken as an example.
est, particularly when combined with live-imaging technologies. Many years of preclinical data went into proving the safety and effi-
Recently, three-colour mouse embryos were generated by HDR fol- cacy of this technique prior to its consideration for use in human
lowing CRISPR–Cas9, with genes distinguishing all three cell-types reproduction, and it can only be used when there is a known risk
of the blastocyst63. In our view, having such fluorescent markers to of inheriting serious mitochondrial disease125–127. Applying similar
distinguish cells of different types within the human embryo could rigor to the handling of germline genome editing will be crucial in
vastly improve the efficiency of microdissection procedures needed determining whether this is a viable clinical option in the future.
for successful derivation of human ESCs and single-cell analyses of
lineage-specific gene expression patterns. One concern is that, at Received: 3 April 2019; Accepted: 25 October 2019;
present, such procedures are performed on morphology alone and Published: xx xx xxxx
cell types are assigned retrospectively and often based on transcrip-
tional markers with a  priori assumptions, which is an unreliable References
readout of cell identity. 1. Cong, L. et al. Multiplex genome engineering using CRISPR/Cas systems.
Additionally, the auxin-inducible degron (AID) system has been Science 339, 819–823 (2013).
2. Jinek, M. et al. RNA-programmed genome editing in human cells. eLife 2,
employed in mammalian embryos through CRISPR–Cas9 (ref. 63). e00471 (2013).
By simultaneously introducing an AID tag to a gene-of-interest 3. Mali, P. et al. RNA-guided human genome engineering via Cas9. Science
and a TIR1 construct into a “safe-harbour” locus, the degron sys- 339, 823–826 (2013).
tem can be used to inducibly, yet reversibly, destroy any desired 4. Adli, M. The CRISPR tool kit for genome editing and beyond.
protein upon auxin treatment111. This method allows for temporal Nat. Commun. 9, 1911 (2018).
5. Carroll, D. Genome engineering with zinc-finger nucleases. Genetics 188,
control of protein expression during embryogenesis, but, similar to 773–782 (2011).
knockdown studies, may not lead to 100% protein removal in the 6. Joung, J. K. & Sander, J. D. TALENs: a widely applicable technology for
desired timeframe. targeted genome editing. Nat. Rev. Mol. Cell Biol. 14, 49–55 (2013).

Nature Cell Biology | www.nature.com/naturecellbiology

NaTURe Cell BIology SERIES | Perspective
7. Silva, G. et al. Meganucleases and other tools for targeted genome 39. Hertig, A. T., Rock, J., Adams, E. C. & Menkin, M. C. Thirty-four fertilized
engineering: perspectives and challenges for gene therapy. Curr. Gene Ther. human ova, good, bad and indifferent, recovered from 210 women of
11, 11–27 (2011). known fertility; a study of biologic wastage in early human pregnancy.
8. Gasiunas, G., Barrangou, R., Horvath, P. & Siksnys, V. Cas9-crRNA Pediatrics 23, 202–211 (1959).
ribonucleoprotein complex mediates specific DNA cleavage for adaptive 40. Lee, A. & Kiessling, A. A. Early human embryos are naturally aneuploid-
immunity in bacteria. Proc. Natl Acad. Sci. USA 109, E2579–E2586 (2012). can that be corrected? J. Assist. Reprod. Genet. 34, 15–21 (2017).
9. Jinek, M. et al. A programmable dual-RNA-guided DNA endonuclease in 41. Scott, R. T. Jr. et al. Blastocyst biopsy with comprehensive chromosome
adaptive bacterial immunity. Science 337, 816–821 (2012). screening and fresh embryo transfer significantly increases in vitro
10. Liang, P. et al. CRISPR/Cas9-mediated gene editing in human tripronuclear fertilization implantation and delivery rates: a randomized controlled trial.
zygotes. Protein Cell 6, 363–372 (2015). Fertil. Steril. 100, 697–703 (2013).
11. Kang, X. et al. Introducing precise genetic modifications into human 3PN 42. Mantikou, E., Wong, K. M., Repping, S. & Mastenbroek, S. Molecular origin
embryos by CRISPR/Cas-mediated genome editing. J. Assist. Reprod. Genet. of mitotic aneuploidies in preimplantation embryos. Biochim. Biophys. Acta
33, 581–588 (2016). 1822, 1921–1930 (2012).
12. Tang, L. et al. CRISPR/Cas9-mediated gene editing in human zygotes using 43. Delhanty, J. D. & Handyside, A. H. The origin of genetic defects in the
Cas9 protein. Mol. Genet. Genomics 292, 525–533 (2017). human and their detection in the preimplantation embryo. Hum. Reprod.
13. Joergensen, M. W. et al. Altered cleavage patterns in human tripronuclear Update 1, 201–215 (1995).
embryos and their association to fertilization method: a time-lapse study. 44. Maurer, M. et al. Chromosomal aneuploidies and early embryonic
J. Assist. Reprod. Genet. 31, 435–442 (2014). developmental arrest. Int. J. Fertil. Steril. 9, 346–353 (2015).
14. Lammers, J., Splingart, C., Barrière, P. & Fréour, T. Morphokinetic 45. Vanneste, E. et al. Chromosome instability is common in human
parameters of ICSI tripronucleated embryos observed using time lapse. cleavage-stage embryos. Nat. Med. 15, 577–583 (2009).
Reprod. Biomed. Online 28, 658–660 (2014). 46. Hassold, T. et al. A cytogenetic study of 1000 spontaneous abortions. Ann.
15. Yao, G. et al. Developmental potential of clinically discarded human Hum. Genet. 44, 151–178 (1980).
embryos and associated chromosomal analysis. Sci. Rep. 6, 23995 (2016). 47. Bazrgar, M. et al. DNA repair signalling pathway genes are overexpressed in
16. Ma, H. et al. Correction of a pathogenic gene mutation in human embryos. poor-quality pre-implantation human embryos with complex aneuploidy.
Nature 548, 413–419 (2017). Eur. J. Obstet. Gynecol. Reprod. Biol. 175, 152–156 (2014).
17. Safari, F., Farajnia, S., Ghasemi, Y. & Zarghami, N. New developments 48. Kiessling, A. A. et al. Evidence that human blastomere cleavage is under
in CRISPR technology: improvements in specificity and efficiency. unique cell cycle control. J. Assist. Reprod. Genet. 26, 187–195 (2009).
Curr. Pharm. Biotechnol. 18, 1038–1054 (2017). 49. Kiessling, A. A. et al. Genome-wide microarray evidence that 8-cell human
18. Fogarty, N. M. E. et al. Genome editing reveals a role for OCT4 in human blastomeres over-express cell cycle drivers and under-express checkpoints.
embryogenesis. Nature 550, 67–73 (2017). J. Assist. Reprod. Genet. 27, 265–276 (2010).
19. Frum, T. et al. Oct4 cell-autonomously promotes primitive endoderm 50. Martinez, F. et al. Caspase activity in preimplantation human embryos is
development in the mouse blastocyst. Dev. Cell 25, 610–622 (2013). not associated with apoptosis. Hum. Reprod. 17, 1584–1590 (2002).
20. Nichols, J. et al. Formation of pluripotent stem cells in the mammalian 51. Kermi, C., Aze, A. & Maiorano, D. Preserving genome integrity during the
embryo depends on the POU transcription factor Oct4. Cell 95, early embryonic DNA replication cycles. Genes (Basel) 10, (E398 (2019).
379–391 (1998). 52. Hardy, K. Apoptosis in the human embryo. Rev. Reprod. 4, 125–134 (1999).
21. Daigneault, B. W., Rajput, S., Smith, G. W. & Ross, P. J. Embryonic POU5F1 53. Cubbon, A., Ivancic-Bace, I. & Bolt, E. L. CRISPR-Cas immunity, DNA
is required for expanded bovine blastocyst formation. Sci. Rep. 8, 7753 (2018). repair and genome stability. Biosci. Rep. 38, BSR20180457 (2018).
22. Davis, M. I. et al. Comprehensive analysis of kinase inhibitor selectivity. 54. Canny, M. D. et al. Inhibition of 53BP1 favors homology-dependent DNA
Nat. Biotechnol. 29, 1046–1051 (2011). repair and increases CRISPR-Cas9 genome-editing efficiency. Nat.
23. De Souza, A. T. et al. Transcriptional and phenotypic comparisons Biotechnol. 36, 95–102 (2018).
of Ppara knockout and siRNA knockdown mice. Nucleic Acids Res. 34, 55. Takayama, K. et al. Highly efficient biallelic genome editing of human ES/
4486–4494 (2006). iPS cells using a CRISPR/Cas9 or TALEN system. Nucleic Acids Res. 45,
24. Blij, S., Frum, T., Akyol, A., Fearon, E. & Ralston, A. Maternal Cdx2 is 5198–5207 (2017).
dispensable for mouse development. Development 139, 3969–3972 (2012). 56. Wilde, J.J. et al. Efficient zygotic genome editing via RAD51-enhanced
25. Kok, F. O. et al. Reverse genetic screening reveals poor correlation between interhomolog repair. Preprint at bioRxiv https://doi.org/10.1101/263699
morpholino-induced and mutant phenotypes in zebrafish. Dev. Cell 32, (2018).
97–108 (2015). 57. Balakier, H., MacLusky, N. J. & Casper, R. F. Characterization of the first
26. Jackson, A. L. et al. Expression profiling reveals off-target gene regulation cell cycle in human zygotes: implications for cryopreservation. Fertil. Steril.
by RNAi. Nat. Biotechnol. 21, 635–637 (2003). 59, 359–365 (1993).
27. Jackson, A. L. & Linsley, P. S. Noise amidst the silence: off-target effects of 58. Capmany, G., Taylor, A., Braude, P. R. & Bolton, V. N. The timing of
siRNAs? Trends Genet. 20, 521–524 (2004). pronuclear formation, DNA synthesis and cleavage in the human 1-cell
28. Scacheri, P. C. et al. Short interfering RNAs can induce unexpected and embryo. Mol. Hum. Reprod. 2, 299–306 (1996).
divergent changes in the levels of untargeted proteins in mammalian cells. 59. Kim, S., Kim, D., Cho, S. W., Kim, J. & Kim, J. S. Highly efficient
Proc. Natl Acad. Sci. USA 101, 1892–1897 (2004). RNA-guided genome editing in human cells via delivery of purified Cas9
29. Robu, M. E. et al. p53 activation by knockdown technologies. PLoS Genet. ribonucleoproteins. Genome Res. 24, 1012–1019 (2014).
3, e78 (2007). 60. Horlbeck, M. A. et al. Nucleosomes impede Cas9 access to DNA in vivo
30. Rossi, A. et al. Genetic compensation induced by deleterious mutations but and in vitro. eLife 5, e12677 (2016).
not gene knockdowns. Nature 524, 230–233 (2015). 61. Isaac, R. S. et al. Nucleosome breathing and remodeling constrain
31. Clift, D., So, C., McEwan, W. A., James, L. C. & Schuh, M. Acute and rapid CRISPR-Cas9 function. eLife 5, e13450 (2016).
degradation of endogenous proteins by Trim-Away. Nat. Protoc. 13, 62. Gutschner, T., Haemmerle, M., Genovese, G., Draetta, G. F. & Chin, L.
2149–2175 (2018). Post-translational regulation of Cas9 during G1 enhances homology-
32. Hardy, K., Handyside, A. H. & Winston, R. M. The human blastocyst: cell directed repair. Cell Rep. 14, 1555–1566 (2016).
number, death and allocation during late preimplantation development 63. Gu, B., Posfai, E. & Rossant, J. Efficient generation of targeted large
in vitro. Development 107, 597–604 (1989). insertions by microinjection into two-cell-stage mouse embryos.
33. Boomsma, C. M. et al. Endometrial secretion analysis identifies a cytokine Nat. Biotechnol. 36, 632–637 (2018).
profile predictive of pregnancy in IVF. Hum. Reprod. 24, 1427–1435 (2009). 64. Sakaue-Sawano, A. et al. Visualizing spatiotemporal dynamics of
34. Braude, P., Bolton, V. & Moore, S. Human gene expression first occurs multicellular cell-cycle progression. Cell 132, 487–498 (2008).
between the four- and eight-cell stages of preimplantation development. 65. Sakaue-Sawano, A., Kobayashi, T., Ohtawa, K. & Miyawaki, A. Drug-
Nature 332, 459–461 (1988). induced cell cycle modulation leading to cell-cycle arrest, nuclear
35. Vassena, R. et al. Waves of early transcriptional activation and pluripotency mis-segregation, or endoreplication. BMC Cell Biol. 12, 2 (2011).
program initiation during human preimplantation development. 66. Zhou, Y. et al. Painting a specific chromosome with CRISPR/Cas9 for
Development 138, 3699–3709 (2011). live-cell imaging. Cell Res. 27, 298–301 (2017).
36. Artley, J. K., Braude, P. R. & Johnson, M. H. Gene activity and cleavage 67. Reichmann, J., Eguren, M., Lin, Y., Schneider, I. & Ellenberg, J. Live
arrest in human pre-embryos. Hum. Reprod. 7, 1014–1021 (1992). imaging of cell division in preimplantation mouse embryos using inverted
37. Jarvis, G. E. Early embryo mortality in natural human reproduction: What light-sheet microscopy. Methods Cell Biol. 145, 279–292 (2018).
the data say. F1000Res. 5, 2765 (2016). 68. Abe, T. et al. Visualization of cell cycle in mouse embryos with Fucci2
38. Koot, Y. E., Teklenburg, G., Salker, M. S., Brosens, J. J. & Macklon, N. S. reporter directed by Rosa26 promoter. Development 140, 237–246 (2013).
Molecular aspects of implantation failure. Biochim. Biophys. Acta 1822, 69. Suzuki, T., Asami, M. & Perry, A. C. Asymmetric parental genome
1943–1950 (2012). engineering by Cas9 during mouse meiotic exit. Sci. Rep. 4, 7621 (2014).

Nature Cell Biology | www.nature.com/naturecellbiology

Perspective | SERIES NaTURe Cell BIology
70. Egli, D. et al. Inter-homologue repair in fertilized human eggs? Nature 560, 103. Lanphier, E., Urnov, F., Haecker, S. E., Werner, M. & Smolenski, J. Don’t
E5–E7 (2018). edit the human germ line. Nature 519, 410–411 (2015).
71. Adikusuma, F. et al. Large deletions induced by Cas9 cleavage. Nature 560, 104. Savulescu, J., Pugh, J., Douglas, T. & Gyngell, C. The moral imperative
E8–E9 (2018). to continue gene editing research on human embryos. Protein Cell 6,
72. Reichmann, J. et al. Dual-spindle formation in zygotes keeps 476–479 (2015).
parental genomes apart in early mammalian embryos. Science 361, 105. National Academies of Sciences Engineering and Medicine. Human Genome
189–193 (2018). Editing: Science, Ethics, and Governance. (The National Academies Press,
73. Ma, H. et al. Ma et al. reply. Nature 560, E10–E23 (2018). Washington, DC, 2017).
74. Kosicki, M., Tomberg, K. & Bradley, A. Repair of double-strand breaks 106. Nuffield Council on Bioethics. Genome Editing and Human Reproduction:
induced by CRISPR-Cas9 leads to large deletions and complex social and ethical issues. (Nuffield Council on Bioethics, London, 2018).
rearrangements. Nat. Biotechnol. 36, 765–771 (2018). 107. De Rycke, M. et al. ESHRE PGD Consortium data collection XIV-XV:
75. Komor, A. C., Kim, Y. B., Packer, M. S., Zuris, J. A. & Liu, D. R. cycles from January 2011 to December 2012 with pregnancy follow-up to
Programmable editing of a target base in genomic DNA without October 2013. Hum. Reprod. 32, 1974–1994 (2017).
double-stranded DNA cleavage. Nature 533, 420–424 (2016). 108. Jacobs, H. S. & Agrawal, R. Complications of ovarian stimulation. Baillieres
76. Rees, H. A. & Liu, D. R. Base editing: precision chemistry on the genome Clin. Obstet. Gynaecol. 12, 565–579 (1998).
and transcriptome of living cells. Nat. Rev. Genet. 19, 770–788 (2018). 109. Baron, K. T., Babagbemi, K. T., Arleo, E. K., Asrani, A. V. & Troiano, R. N.
77. Li, G. et al. Highly efficient and precise base editing in discarded human Emergent complications of assisted reproduction: expecting the unexpected.
tripronuclear embryos. Protein Cell 8, 776–779 (2017). Radiographics 33, 229–244 (2013).
78. Liang, P. et al. Correction of β-thalassemia mutant by base editor in human 110. Steffann, J., Jouannet, P., Bonnefont, J. P., Chneiweiss, H. & Frydman, N.
embryos. Protein Cell 8, 811–822 (2017). Could failure in preimplantation genetic diagnosis justify editing the human
79. Zhou, C. et al. Highly efficient base editing in human tripronuclear zygotes. embryo genome? Cell Stem Cell 22, 481–482 (2018).
Protein Cell 8, 772–775 (2017). 111. Natsume, T., Kiyomitsu, T., Saga, Y. & Kanemaki, M. T. Rapid protein
80. Zeng, Y. et al. Correction of the Marfan syndrome pathogenic FBN1 depletion in human cells by auxin-inducible degron tagging with short
mutation by base editing in human cells and heterozygous embryos. homology donors. Cell Rep. 15, 210–218 (2016).
Mol. Ther. 26, 2631–2637 (2018). 112. Sternberg, N. & Hamilton, D. Bacteriophage P1 site-specific
81. Jin, S. et al. Cytosine, but not adenine, baseeditors induce genome-wide recombination. I. Recombination between loxP sites. J. Mol. Biol. 150,
off-target mutations in rice. Science 364, 292–295 (2019). 467–486 (1981).
82. Zuo, E. et al. Cytosine baseeditor generates substantial off-target single- 113. Sauer, B. & Henderson, N. Site-specific DNA recombination in mammalian
nucleotide variants in mouse embryos. Science 364, (289–292 (2019). cells by the Cre recombinase of bacteriophage P1. Proc. Natl Acad. Sci. USA
83. Yang, G. et al. Base-editing-mediated R17H substitution in histone H3 85, 5166–5170 (1988).
reveals methylation-dependent regulation of Yap signaling and early mouse 114. Lander, E. S. et al. Adopt a moratorium on heritable genome editing.
embryo development. Cell Rep. 26, 302–312.e4 (2019). Nature 567, 165–168 (2019).
84. Anzalone, A. V. et al. Search-and-replace genome editing without 115. Cyranoski, D. Russian biologist plans more CRISPR-edited babies. Nature
double-strand breaks or donor DNA. Nature https://doi.org/10.1038/ 570, 145–146 (2019).
s41586-019-1711-4 (2019). 116. Cyranoski, D. China set to introduce gene-editing regulation following
85. McArthur, S. J. et al. Blastocyst trophectoderm biopsy and preimplantation CRISPR-baby furore. Nature https://doi.org/10.1038/d41586-019-01580-1
genetic diagnosis for familial monogenic disorders and chromosomal (2019).
translocations. Prenat. Diagn. 28, 434–442 (2008). 117. Normile, D. China tightens rules on gene editing. Science 363,
86. Adler, A. et al. Blastocyst culture selects for euploid embryos: comparison 1023 (2019).
of blastomere and trophectoderm biopsies. Reprod. Biomed. Online 28, 118. Cyranoski, D. Japan set to allow gene editing in human embryos. Nature
485–491 (2014). https://doi.org/10.1038/d41586-018-06847-7 (2018).
87. Vilarino, M. et al. Mosaicism diminishes the value of pre-implantation 119. Chiba, N. Japanese government considering legally restricting genome
embryo biopsies for detecting CRISPR/Cas9 induced mutations in sheep. editing technology. https://mainichi.jp/english/articles/20190412/
Transgenic Res. 27, 525–537 (2018). p2a/00m/0na/004000c (2019).
88. Thornhill, A. R. & Snow, K. Molecular diagnostics in preimplantation 120. Reardon, S. World Health Organization panel weighs in on CRISPR-babies
genetic diagnosis. J. Mol. Diagn. 4, 11–29 (2002). debate. Nature 567, 444–445 (2019).
89. Blais, J. et al. Risk of misdiagnosis due to allele dropout and false-positive 121. Dzau, V. J., McNutt, M. & Ramakrishnan, V. Academies’ action plan for
PCR artifacts in molecular diagnostics: analysis of 30,769 genotypes. germline editing. Nature 567, 175 (2019).
J. Mol. Diagn. 17, 505–514 (2015). 122. Ledford, H. CRISPR babies: when will the world be ready? Nature 570,
90. Chakrabarti, A. M. et al. Target-specific precision of CRISPR-mediated 293–296 (2019).
genome editing. Mol. Cell 73, 699–713.e6 (2019). 123. Cree, L. & Loi, P. Mitochondrial replacement: from basic research to
91. Allen, F. et al. Predicting the mutations generated by repair of Cas9-induced assisted reproductive technology portfolio tool-technicalities and possible
double-strand breaks. Nat. Biotechnol. 37, 64–72 (2018). risks. Mol. Hum. Reprod. 21, 3–10 (2015).
92. Shen, M. W. et al. Predictable and precise template-free CRISPR editing of 124. Claiborne, A. B., English, R. A., Kahn, J. P. & ETHICS, O. F. NEW
pathogenic variants. Nature 563, 646–651 (2018). TECHNOLOGIES. Finding an ethical path forward for mitochondrial
93. Kim, D. et al. Digenome-seq: genome-wide profiling of CRISPR-Cas9 replacement. Science 351, 668–670 (2016).
off-target effects in human cells. Nat. Methods 12, 237–243 (2015). 125. Human Fertilisation and Embryology Authority. Scientific Review of the
94. Tsai, S. Q. et al. GUIDE-seq enables genome-wide profiling of off-target Safety and Efficacy of Methods to Avoid Mitochondrial Disease Through
cleavage by CRISPR-Cas nucleases. Nat. Biotechnol. 33, 187–197 (2015). Assisted Conception: 2016 update. (Human Fertilisation and Embryology
95. Tsai, S. Q. et al. CIRCLE-seq: a highly sensitive in vitro screen for Authority, London, 2016).
genome-wide CRISPR-Cas9 nuclease off-targets. Nat. Methods 14, 126. Hyslop, L. A. et al. Towards clinical application of pronuclear transfer to
607–614 (2017). prevent mitochondrial DNA disease. Nature 534, 383–386 (2016).
96. Cameron, P. et al. Mapping the genomic landscape of CRISPR-Cas9 127. Committee on the Ethical and Social Policy Considerations of Novel
cleavage. Nat. Methods 14, 600–606 (2017). Techniques for Prevention of Maternal Transmission of Mitochondrial
97. Yan, W. X. et al. BLISS is a versatile and quantitative method for DNA Diseases; Board on Health Sciences Policy; Institute of Medicine;
genome-wide profiling of DNA double-strand breaks. Nat. Commun. 8, National Academies of Sciences, Engineering, and Medicine (eds Claiborne,
15058 (2017). A., English, R. & Kahn, J.) Mitochondrial replacement techniques: Ethical,
98. Wienert, B. et al. Unbiased detection of CRISPR off-targets in vivo using social, and policy considerations. (National Academies Press, Washington
DISCOVER-Seq. Science 364, 286–289 (2019). DC, 2016).
99. Kim, D. & Kim, J. S. DIG-seq: a genome-wide CRISPR off-target profiling 128. Deglincerti, A. et al. Self-organization of the in vitro attached human
method using chromatin DNA. Genome Res. 28, 1894–1900 (2018). embryo. Nature 533, 251–254 (2016).
100. Smith, C. et al. Whole-genome sequencing analysis reveals high specificity 129. Shahbazi, M. N. et al. Self-organization of the human embryo in the
of CRISPR/Cas9 and TALEN-based genome editing in human iPSCs. absence of maternal tissues. Nat. Cell Biol. 18, 700–708 (2016).
Cell Stem Cell 15, 12–13 (2014). 130. Macaulay, I. C. et al. G&T-seq: parallel sequencing of single-cell genomes
101. Veres, A. et al. Low incidence of off-target mutations in individual and transcriptomes. Nat. Methods 12, 519–522 (2015).
CRISPR-Cas9 and TALEN targeted human stem cell clones detected by 131. de Silva, E. & Stumpf, M. P. HIV and the CCR5-Delta32 resistance allele.
whole-genome sequencing. Cell Stem Cell 15, 27–30 (2014). FEMS Microbiol. Lett. 241, 1–12 (2004).
102. Baltimore, D. et al. Biotechnology. A prudent path forward for genomic 132. Luzzatto, L. Sickle cell anaemia and malaria. Mediterr. J. Hematol. Infect.
engineering and germline gene modification. Science 348, 36–38 (2015). Dis. 4, e2012065 (2012).

Nature Cell Biology | www.nature.com/naturecellbiology

NaTURe Cell BIology SERIES | Perspective
133. Dzau, V. J., McNutt, M. & Bai, C. Wake-up call from Hong Kong. Science Competing interests
362, 1215 (2018). The authors declare no competing interests.

Acknowledgements Additional information

We thank members of the Niakan, J. Turner and R. Lovell-Badge laboratories for Correspondence should be addressed to K.K.N.
helpful discussions and comments on the Perspective. We are grateful to J. Brock in the Reprints and permissions information is available at www.nature.com/reprints.
Scientific Illustrations team for generating figures. Work in the Niakan laboratory is
supported by the Francis Crick Institute, which receives its core funding from Cancer Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
Research UK (FC001120), the UK Medical Research Council (FC001120), and the published maps and institutional affiliations.
Wellcome Trust (FC001120). © Springer Nature Limited 2019

Nature Cell Biology | www.nature.com/naturecellbiology