© The Author(s) 2019. Published by Oxford University Press on behalf of the European Society of Human Reproduction and Embryology. All rights reserved.For permissions, please e-mail: journals.permissions@oup.com

Human Reproduction, Vol.34, No.11, pp. 2104–2111, 2019Advance Access Publication on November 14, 2019 doi:10.1093/humrep/dez162

INVITED COMMENTARY

The technical risks of human geneeditingBenjamin Davies*Welcome Centre for Human Genetics, University of Oxford, Oxford OX3 7BN, UK.

*Correspondence. Email: ben.davies@well.ox.ac.uk

Submitted on May 29, 2019; resubmitted on July 2, 2019; editorial decision on July 15, 2019

ABSTRACT: A recent report from Dr He Jiankui concerning the birth of twin girls harbouring mutations engineered by CRISPR/Cas nucleaseshas been met with international condemnation. Beside the serious ethical concerns, there are known technical risks associated with CRISPR/Casgene editing which further raise questions about how these events could have been allowed to occur. Numerous studies have reportedunexpected genomic mutation and mosaicism following the use of CRISPR/Cas nucleases, and it is currently unclear how prevalent thesedisadvantageous events are and how robust and sensitive the strategies to detect these unwanted events may be. Although Dr Jiankui’s studyappears to have involved certain checks to ascertain these risks, the decision to implant the manipulated embryos, given these unknowns, mustnonetheless be considered reckless. Here I review the technical concerns surrounding genome editing and consider the available data from DrJiankui in this context. Although the data remains unpublished, preventing a thorough assessment of what was performed, it seems clear thatthe rationale behind the undertaking was seriously flawed; the procedures involved substantial technical risks which, when added to the seriousethical concerns, fully justify the widespread criticism that the events have received.

Key words: gene editing / CRISPR / nuclease / mutagenesis / Cas9

IntroductionThe development of site-specific nucleases over the last decade nowmakes it possible to introduce precise changes into the DNA sequenceof our cells (Carroll, 2017). In particular, RNA-guided CRISPR/Casnucleases are very easy to design against specific genomic targetsequences and high efficiencies of mutagenesis can be achieved (Sanderand Joung, 2014). These qualities are making the therapeutic applica-tion of CRISPR/Cas nucleases to tackle genetic disease feasible, andthere has been a diverse range of success stories published in pre-clinical models (Porteus, 2019). For example, CRISPR/Cas9 nucleaseshave been designed to ablate the mutation responsible for musculardystrophy, restoring normal gene expression (Long, et al., 2014). Inboth small (Nelson, et al., 2016) and large (Amoasii, et al., 2018) animalmodels, viral delivery of these nucleases into the diseased muscle wasshown to restore muscle condition and strength. Such studies typify theexciting new field of therapeutic gene editing and highlight its potentialin tackling genetic disease.

The target cell for therapeutic gene editing needs to be carefullyconsidered to maximise the therapeutic potential and its longevity.Clearly, editing a stem cell population would have clear advantages fora prolonged therapeutic effect. Indeed, CRISPR/Cas nucleases havebeen introduced into haematopoietic stem cells to correct the under-lying mutations responsible for sickle-cell anaemia (Vakulskas, et al.,

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

2018) and there is thus considerable interest in this approach for long-lasting treatments for genetic diseases of the blood.

The application of gene editing tools in somatic stem cell therapieshas raised the possibility that they could be applied in the ultimatestem cell, the one-cell embryo, allowing the genetic correction to bepermanent and thus present in all cells of the resulting individual.Furthermore, a successful genetic manipulation of the one-cell embryowould lead to the presence of the genetic change in the germ cells ofthe manipulated individual, and thus the inheritance of these genomeedits would be achieved. Early work introducing CRISPR/Cas nucle-ases into the one-cell mouse embryo by microinjection demonstratedthat genetic modification of the whole organism could be achievedefficiently and confirmed the inheritance of the genetic change insubsequent generations (Wang et al., 2013).

CRISPR/Cas nucleases in human embryosReports soon emerged that CRISPR/Cas could indeed be used tomanipulate the human one-cell embryo. The first published stud-ies used discarded tripronuclear zygotes and achieved mutagenesisefficiencies of up to 50%, with specific gene editing (i.e. the incor-poration of information from a co-injected repair template) occur-ring at around 15% (Kang et al., 2016; Liang et al., 2015). This wasquickly followed by a further report of efficient mutagenesis in healthy

The technical risks of human gene editing 2105

Figure 1 Disadvantageous outcomes of CRISPR/Cas mutagenesis within the 1-cell embryo. (A) Off-target mutagenesis. In addition tothe correct mutagenesis event, similar sequences elsewhere in the genome are also mutated leading to unpredictable effects. (B) Mosaicism. Prolongedactivity of the nuclease within the developing embryo can lead to different mutations in different parts of the resulting individual. (C) Large deletions. Theleft panel shows the intended mutagenesis event where a target gene is inactivated. The right panel shows a potential consequence of a large deletionevent, where a neighbouring gene is also inactivated. (D) On-site damage. The top panel shows the intended mutagenesis event, with a CRISPR/Casnuclease specifically recognising only the mutant allele, leading to its inactivation. The bottom panel shows the biallelic mutation that could occur if theCRISPR/Cas nuclease is not able to discriminate the mutant sequence from the normal sequence.

two-pronuclear embryos (Tang, et al., 2017). Two more substantialarticles followed, both published in Nature (Ma et al., 2017; Fogartyet al., 2017). The first demonstrated efficient correction of a dominantpathogenic mutation at the MYBPC3 gene, encoding a cardiac myosin-binding protein, in heterozygous human embryos, and proposed amechanism of inter-homologue repair using the wild-type allele asa repair template (Ma et al., 2017). The second study focussed onthe use of the technology to explore the role of genes involved inpreimplantation human development and achieved targeted mutationof the gene encoding the transcription factor OCT4 (POU5F1) in 71% ofmanipulated embryos (Fogarty et al., 2017). More recent studies havesuccessfully achieved gene editing using exogenous repair templates torepair pathogenic mutations (Tang et al., 2018) or to introduce specificreporter sequences (Yao et al., 2018).

An alternative strategy for achieving site-specific change within thegenome has been reported: base editing, which relies upon the fusionof enzymatic domains to the CRISPR/Cas machinery, capable ofchemically converting one nucleotide base to another (Rees and Liu2018). Base editing has also been successfully applied within the humanone-cell embryo (Li et al., 2017; Liang et al., 2018), and pathogenicmutations have been successfully corrected using this technology (Lianget al., 2017; Zeng et al., 2018).

These studies demonstrate that it is now feasible to achieve muta-tions and edits in human embryos at manageable frequencies andsuggest that the tools for therapeutic germline editing are now avail-able. There are numerous ethical concerns surrounding this technologywhich have been widely discussed and reviewed elsewhere (van Dijkeet al., 2018), but a major requirement before germline editing can be

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

considered is to assess the safety of the manipulations. Investigationsof CRISPR/Cas9 mutagenesis both in cell culture experiments and inembryos have highlighted a number of disadvantageous consequences,where research is needed to assess and mitigate the risk and optimisemethods permitting the reliable detection of such events. There thusremains an appreciable amount of technology development and opti-misation to be done before therapeutic editing can be considered.

Off-target mutagenesisSoon after the demonstration that CRISPR/Cas9 could be used fortargeted manipulation of the mammalian genome, reports emergedthat its use carries a risk of unintended mutagenesis at closely matchedgenomic sequences (Fu et al., 2013). This so-called off-target muta-genesis is also more pronounced than initially expected as the com-monly used Cas9 enzyme can tolerate certain mismatches within itstargeting sequence (Fig. 1A). Many of the studies addressing off-targetmutagenesis have been performed in cell culture experiments wherethe CRISPR/Cas enzymes are transfected into millions of cells, thegenomic DNA of which is then deep-sequenced to ascertain levelsof accuracy. These types of experiment may overestimate the risk ofoff-target mutagenesis occurring when the CRISPR/Cas nucleases areapplied in a single cell, i.e. the one-cell embryo. Indeed, one carefullycontrolled study in mouse used whole genome sequencing on a trio(sequencing both parental and offspring DNA) to address off-targetmutagenesis resulting from a one-cell embryo microinjection experi-ment and the authors were unable to detect any events in the foundermice analysed (Iyer et al., 2018). In contrast, a larger study investigated

2106 Davies

founder rodent lines generated with multiple CRISPR/Cas9 enzymesaddressing a number of different target sequences and found thatalmost 30% of the mutant lines harboured putative off-target mutations(Anderson et al., 2018). Interestingly, base editors designed to convertcytidine to thymidine residues were also found to have substantialoff-target effects when applied within the mouse one-cell embryo(Zuo et al., 2019). Some of the human studies have also analysedthe resulting embryos for off-target effects. Candidate off-target sites,localised using bioinformatic approaches, have been analysed by eitherSanger or next generation sequencing. One of the studies confirmedan off-target mutation in two of the resulting embryos (Liang et al.,2015). All of the other studies found no evidence for significant levelsof off-target mutation (Fogarty et al., 2017; Kang et al., 2016; Ma et al.,2017; Tang et al., 2017). In contrast with the results from the cytidineto thymidine base editors in mouse (Zuo et al., 2019), only low orentirely absent levels of off-target mutagenesis were detected whenthese reagents were applied in human embryos (Li et al., 2017; Lianget al., 2017; Zeng et al., 2018).

The risk of off-target mutagenesis is thus clearly dependent uponthe target sequence and can be reduced by designing CRISPR/Casnucleases against truly unique genomic sequences; a number of onlinealgorithms are available to facilitate this improved design (Haeussleret al., 2016; Hodgkins et al., 2015). These bioinformatic assessmentsof off-target risk can be somewhat flawed however, as many of theavailable tools do not take into consideration human genetic sequencevariation. A true off-target profile and thus risk assessment for aselected CRISPR/Cas nuclease can only really be accomplished byestablishing a personalised genome. Indeed, studies have suggested thatnaturally occurring human SNPs can alter the off-target landscape ofsite-specific nucleases substantially (Lessard et al., 2017).

It has also been shown that the concentration and persistence ofthe nuclease can increase the chance of off-target cleavage (Kim et al.,2014; Zuris et al., 2015), and subsequently, a number of approachesaimed at limiting the activity of the nuclease have been shown to reducethe level of off-target mutation (Chen et al., 2016; Shen et al., 2018a).Structural investigations and molecular evolution of the Cas9 nucleasehave enabled the design of variant sequences which show increasedlevels of accuracy (Kleinstiver et al., 2016; Slaymaker et al., 2016) andreduced risk of off-target mutagenesis in rodent models (Andersonet al., 2018). Furthermore, orthologues of Cas nucleases from alter-native bacterial species have been shown to have increased levelsof accuracy (Kim et al., 2016; Teng et al., 2018). A recent studyalso improved the accuracy of CRISPR/Cas effectors by altering thestructure of its cofactor guide-RNA (Kocak et al., 2019), highlighting adifferent approach to addressing this problem.

Taken together, there is a measurable risk of off-target mutagenesiswhen applying CRISPR/Cas nucleases in cells and embryos, but therehas been significant technology refinement and bioinformatics tooldevelopment to reduce these risks substantially. Nonetheless, moreresearch is needed in this area to assess the risk, improve the accuracyof the enzymes and explore methods for detecting these off-targetevents.

MosaicismThe microinjection of CRISPR/Cas nucleases into mouse zygotessoon revealed that the CRISPR/Cas nucleases frequently retain activ-

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

ity after the first embryo cleavage event. Consequently, there is anappreciable risk that individual cells of the two-cell or even four-cell embryo harbour different combinations of wild-type and mutantalleles. The resulting organism is thus frequently a genetic mosaic, withdifferent combinations of mutations in different parts of the animal(Fig. 1B). This phenomenon is easily demonstrated by microinjectionexperiments in mouse embryos which use CRISPR/Cas nucleasesdesigned against coat colour genes. Addressing the tyrosinase gene,the loss of function of which leads to the albino phenotype, clearsomatic mosaicism was evident as the majority of founder offspringshow a speckled, patchy coat colour, rather than complete albinism(Yen et al., 2014). Clear evidence of mosaicism was also found in thehuman studies (Fogarty et al., 2017; Ma et al., 2017; Tang et al., 2017;Yao et al., 2018).

Mosaicism may be tackled by altering the timing of the nucleaseactivity within the one-cell embryo, effectively restricting its activityto the one-cell stage. Reducing the half-life of the nuclease throughthe use of a destabilised version of Cas9 was shown to reducemosaicism whilst editing non-human primate embryos (Tu et al., 2017).Another improvement was achieved by delivering the nuclease to invitro fertilised embryos by electroporation, permitting the delivery ofthe CRISPR/Cas machinery at a very early developmental stage, evenbefore pronuclei have formed (Hashimoto et al., 2016). One studyin human embryos was able to effectively eliminate mosaicism byintroducing the CRISPR/Cas reagents at the same time as performingthe fertilisation by intracytoplasmic sperm injection (Ma et al., 2017).

Large deletions and rearrangementsThe mutagenesis occurring following the application of CRISPR/Casnucleases relies upon the innate DNA repair machinery of the tar-get cells. Most frequently, the induced double-strand break (DSB) isrepaired by non-homologous end joining, which can lead to the intro-duction of small deletions and insertions. The mutations are frequentlysmall in size, the most common being a single-nucleotide insertion ordeletion (Chakrabarti et al., 2018; Taheri-Ghahfarokhi et al., 2018).Nonetheless, larger deletions do occur and there is evidence to suggestthat, on occasion, the repair event can result in large kilobase-scaledeletions. In one in vivo study, introducing CRISPR/Cas nucleases asa virus to correct a mutation in the Otc gene, an appreciable rate(6.5%) of disruptive large deletions was found (Yang et al., 2016). Morerecently, various studies have observed large deletions occurring atsignificant levels following the use of CRISPR/Cas9 in vitro (Kosickiet al., 2018) and when applied within the one-cell embryo (Parikhet al., 2015; Shin et al., 2017). These unexpectedly large deletions havethe capacity to delete whole genes or cause misregulation of nearbyexpressed sequences (Fig. 1C).

In addition to large deletions, a mouse study has revealed thatcomplex rearrangements can occur following the application of nucle-ases in the one-cell embryo (Boroviak et al., 2017). These eventsseem particularly prevalent when using multiple CRISPR/Cas nucleaseswhich cleave in cis. The repair of the resulting two or more DSBs canresult in the deletion of the intervening sequence (which is often theaim of the experiment), but also the inversion of the sequences canoccur, as well as many surprising duplications and insertion events.

Interestingly, the prevalence of these deletions and rearrangementsmay have been underestimated since these complex events are often

The technical risks of human gene editing 2107

invisible to the molecular assays used to genotype the resulting muta-tions. Simple PCR-based genotyping strategies can be compromisedby the deletion encompassing the primer binding sites. Short-read nextgeneration sequencing technologies are not well equipped to detectand assess genomic inversions and duplications. Genomic technolo-gies based on long-reads or more traditional assessment of targetlocus integrity by Southern blotting or fluorescent in situ hybridisa-tion analysis may help detection, but these methods are difficult toapply and may not very applicable for genotyping embryo biopsymaterial.

The difficulty in detecting large deletions has fuelled controversysurrounding the inter-homologue repair mechanism proposed whenCRISPR/Cas9 nucleases are used to selectively ablate a pathogenicmutation present heterozygously (Ma et al., 2017). It has been sug-gested that the inability to detect a rearranged or damaged mutantallele could lead to the misinterpretation that the allele has beenrepaired from the intact wild-type allele (Adikusuma et al., 2018;Egli et al., 2018), although follow-up analysis of the original studyprovided evidence arguing against this explanation (Ma et al., 2018).At the moment, we know too little about the dominant DNA repairmachinery active within the early preimplantation embryos and thesediscussions highlight the requirement for further research to fullyestablish what repair events are likely, how they can be harnessedfor therapeutic effect and how disadvantageous large deletions andrearrangements of the target locus can be detected.

On-site damage and biallelic modificationThe high efficiency of CRISPR/Cas nucleases frequently leads to themutagenesis of both autosomal copies of a target gene. Where a com-plete loss of function is therapeutic, this is, of course, advantageous;however, there are frequent situations where only one copy of a geneneeds to be addressed, in particular when trying to correct or ablatedominant heterozygous mutations. Although CRISPR/Cas nucleasescan be designed against the mutated copy of a gene, the tolerance ofCas9 for small mismatches may make it challenging to design nucleasesthat can discriminate between a mutant copy and a normal copy ofa gene (Fig. 1D). Interestingly, in the human study which successfullyapplied CRISPR/Cas nucleases to correct a dominant heterozygousmutation in the MYBPC3 gene (Ma et al., 2017), the mutation chosenfor this proof-of-concept study was a 4-bp deletion. This relativelylarge mutation allowed the nuclease to be designed specifically againstthis mutant allele, thus eliminating the risk of mutating the wild-typeallele. The majority of disease-associated mutations, however, aresingle-nucleotide changes where discrimination may be challenging, andmutagenesis of the normal copy of the gene or even reprocessingand subsequent mutagenesis of the correctly repaired mutation wouldbe expected to occur at appreciable frequencies. Development ofenzymes with a higher level of discrimination may help the selectivecorrection of the mutant alleles. The developed Cas enzymes andorthologues with reported higher accuracy may be very useful in thiscontext (Kim et al., 2016; Kleinstiver et al., 2016; Slaymaker et al., 2016;Teng et al., 2018).

Another approach to help improve the predictability of gene editingoutcomes is emerging from the analysis of large numbers of mutage-nesis events. It has become clear that, for a specific target sequence,certain mutational outcomes can be quite common. Part of the expla-

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

nation for this lies in regions of microhomology that lie upstream anddownstream of the target site (Bae et al., 2014), but recent studies haveidentified other key attributes of the underlying primary sequence thatcan be used to predict the predominant pattern of repair (Allen et al.,2018; Chakrabarti et al., 2018; Shen et al., 2018b; Taheri-Ghahfarokhiet al., 2018). These studies demonstrate the power of using machinelearning tools and large data set analysis to help make CRISPR/Casmutagenesis more predictive. With the help of these new tools, itcould be possible to achieve the desired repair of a mutation simply bydesign considerations alone. Indeed, one of the recent papers in thisarea has already confirmed the feasibility of this approach by using thepredictive outcome of DSB processing to repair a pathogenic mutation(Shen et al., 2018b).

The reported birth of gene-edited twinsDespite technical improvements addressing the shortcomings, thereremains uncertainty about the prevalence, extent and detection ofgenomic damage and mosaicism. Considerable research and technicaldevelopment are needed to quantify and address the issues beforetherapeutic gene editing can be considered. Given these unresolvedsafety concerns, it was alarming to hear the reports emerging latelast year from the Southern University of Science and Technology,Shenzhen, China, which suggested that human embryos had beenmanipulated by CRISPR/Cas9, reimplanted into the mother and car-ried to term. The principal scientist involved in this study, Dr He Jiankui,reported his results at the Second International Summit on HumanGenome Editing in Hong Kong (National Academies of Sciences,2019) and described the birth of twin girls, Lula and Nana, who bothcarried CRISPR/Cas9-engineered mutations. Dr Jiankui attemptedto introduce loss-of-function mutations into the gene encoding theCCR5 chemokine receptor, a co-receptor for certain subtypes ofHIV virus.

A naturally occurring CCR5 variant involving a deletion of 32 bpintroduces a premature STOP codon into the gene, resulting in theexpression of a truncated protein which is not able to act as anHIV co-receptor. Subsequently, individuals homozygous for this socalled �32 mutation are resistant to infection by certain HIV subtypes.There has been wide interest in this mutation, since an HIV-infectedpatient has been effectively cured of viral infection by an allogeneicstem-cell transplantation with haematopoietic stem cells from a donorhomozygous for this �32 CCR5 variant (Hutter et al., 2009). The aimof Dr Jiankui’s study was to engineer loss-of-function mutations withinthe CCR5 gene in human embryos generated by IVF from parents wherethe father was infected with HIV. In doing so, the goal was to protectthe resulting embryos from HIV infection.

Flawed scientific rationale and experimentaldesignThe scientific rationale behind the study is questionable for a numberof reasons. Firstly, there are established protocols involving semenwashing which can be used to reduce the risk of infection when usingHIV-infected semen in assisted reproductive therapy (Zafer et al.,2016). There appears no need to invoke genome editing for thispurpose. Secondly, CCR5 is a co-receptor for one subtype of HIV;a different chemokine receptor, CXCR4, can also act as a co-receptor

2108 Davies

for different classes of HIV. In patients with CCR5 mutations, CXCR4-tropic HIV subtypes can, albeit inefficiently, enter the cells through thisalternative receptor which would continue to be expressed (Agrawalet al., 2004). Potentially the edited offspring would thus still be sus-ceptible to HIV infection, despite engineered mutations in their CCR5receptor gene.

With respect to the experimental design, the genome engineeringstrategy adopted did not involve the incorporation of the naturallyoccurring �32 mutation, despite the fact that a previous study inhuman embryos showed successfully that the �32 mutation could beincorporated at the CCR5 gene with a repair template (Kang et al.,2016). Instead, a random mutagenesis approach was adopted, albeit atthe same position within the gene, which would be expected to lead tothe incorporation of de novo mutations within the gene. The biologicalconsequences of these novel mutations are impossible to predict andcould lead to a global CCR5 knockout by affecting mRNA or proteinstability.

Importantly, it has been suggested that stable expression of the CCR5�32 variant may be important for HIV resistance (Agrawal et al.,2007); thus, mutations that cause a global knockout might not leadto the immunity which was the primary goal of the study. Moreworrying, results from Ccr5 knockout mouse models are revealingthat there may be other consequences of CCR5 loss of function,besides HIV entry. Ccr5 loss of function led to an increased severityfollowing infection of influenza virus (Falcon et al., 2015) and WestNile virus (Durrant et al., 2015) and have implicated CCR5 functionin neuronal plasticity (Zhou et al., 2016) and recovery after brain injury(Joy et al., 2019). Another recent study explored human UK Biobankdata to assess the impact of the CCR5 �32 variant on longevityand reported an estimated 21% increase in mortality for individualshomozygous for this mutation (Wei & Nielsen, 2019). Mutations atCCR5, especially those with uncharacterised consequences on proteinstability and expression, might thus be expected to have unpredictableand disadvantageous consequences.

Genotyping data from Dr Jiankui’s studyThe results of Dr Jiankui’s study remain unpublished, and thus, theprimary data has not been peer-reviewed, making it difficult to assessthe study. However, from his presentation at the Second InternationalSummit on Human Genome Editing, we can learn some of the geno-typing approaches he adopted in an attempt to mitigate the knownproblems of CRISPR/Cas nucleases outlined above.

Following CRISPR/Cas9 injection, the embryos were cultured to theblastocyst stage in vitro and the CCR5 genotype was then assessed bypreimplantation genetic diagnosis (PGD) using whole genome sequenc-ing on trophectoderm biopsies. Two embryos were selected for trans-plantation, the first of which (Lulu) harboured +1-bp and −4-bpframeshift alleles, predicted to encode truncation mutations that aresimilar but different from the natural occurring �32 mutation. Asdiscussed above, the biological activity of these mutations is completelyunclear.

The second embryo (Nana) harboured a 15-bp deletion on oneallele, whilst the other allele remained unedited. This 15-bp deletionresults in an in-frame deletion that would effectively remove five aminoacids from the mature CCR5 peptide chain. Dr Jiankui hypothesisedthat this allele might destabilise the CCR5 structure near the HIV-

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

binding site, but no experimental validation of this assumption waspresented. The in-frame deletion would almost certainly result in amature protein of completely unknown function; one could envisagea dominant effect which might compromise the function of organsystems in which CCR5 is known to play a role.

The fact that an embryo with an in-frame deletion and with aremaining intact copy of the CCR5 gene was selected for implantation ishighly questionable. The unedited CCR5 allele would still be expressed,and thus, a normal version of this chemokine receptor would remainon the surface of T-cells, thus rendering the cells infectable by HIV, atodds with the primary goal of the study.

Dr Jiankui highlighted in his presentation that the parents madethe decision concerning whether this embryo should be implanted,although it is entirely unclear how the parents were advised or coun-selled. Indeed, in the question-and-answer session that followed hispresentation, it became clear that the scientists directly involved inthe study may have performed the genetic counselling of the parents,rather than correctly qualified and trained genetic counsellors.

Known problems of CRISPR/Casmutagenesis were addressedConcerning off-target site mutagenesis, theoretical off-target sites werebioinformatically assessed using the parental genomes to allow theconsequence of SNP variation to be considered. As mentioned above,naturally occurring genome variation has the potential to alter the off-target landscape substantially (Lessard, et al., 2017) and it is interestingthat this study did indeed ascertain a personalised genome for this pur-pose. These theoretical off-targets were combined with experimentallyreported off-target sites from published in vitro experiments whichused the same target site to establish a panel of risk sites within thegenome. PGD using whole genome sequencing on biopsied cells fromthe embryos revealed the presence of a single intergenic off-targetsite within Lulu’s genome. A decision was thus made to knowinglyimplant an embryo harbouring a CRISPR/Cas-induced mutation atan off-target site; although localised to an intergenic region of thegenome, the functional consequences of this off-target mutation areunclear. Intriguingly, this off-target was no longer detected in fetalDNA analysis obtained from maternal blood during gestation and inthe cord and placental samples obtained at birth, indicating that theinitial PGD result might have been an artefact of whole genome ampli-fication or reflect a low-level mosaicism in the trophectoderm cellsbiopsied.

Concerning mosaicism, the CRISPR/Cas9 reagents were appliedduring the ICSI fertilisation procedure using the same approach asadopted previously (Ma et al., 2017). Similarly, despite what appears tobe mosaic sequence traces in the PGD Sanger sequencing, the resultsof the whole genome sequences revealed equal proportions of twoalleles in each of the implanted embryos, suggesting that the embryoswere not genetic mosaics.

Concerning large deletion analysis, the presence of large deletionswas investigated by searching for chimeric sequencing reads arisingfrom two regions of the genome in cis. Interestingly, in one editedembryo, not selected for implantation, a 6-kb deletion at the CCR5target site was indeed found, confirming the prevalence of this kindof repair outcome. The embryos chosen for implantation revealed noevidence of large deletions.

The technical risks of human gene editing 2109

Despite these relatively thorough sequencing experiments which,to a degree, seek to mitigate the known problems of CRISPR/Casmutagenesis outlined above, it is unclear how thorough and completethe analysis was. Without an in-depth assessment of the primary data,it is impossible to know whether the investigations were sufficient tocompletely rule out non-specific mutagenesis events, large deletions orgenomic rearrangements.

ConclusionsGiven the above technical and scientific issues, combined with the graveethical concerns (Krimsky, 2019), it is not surprising that Dr Jiankui’sstudy has been widely condemned as being a reckless and prematureuse of the technology. As a direct result, experts in the field have calledfor a moratorium on germline editing (Lander et al., 2019), which hasbeen widely supported by the scientific community. It is important torecognise however that this moratorium concerns the implantation ofedited embryos. Indeed, many scientists, clinicians, patient groups andethicists support that research is needed to understand and addressthe risks involved. There is thus an understanding that this researchmay necessitate the use of human embryos, and the argument has beenmade that intentionally refraining from engaging in life-saving research isnot morally defensible (Savulescu et al., 2015). However, at this point intime, at the very beginning of this emerging field with many unknowns,the implantation of edited embryos cannot be justified.

Research is needed to fully understand the repair outcomes occur-ring following the action of CRISPR/Cas9 nucleases within the earlyembryo, especially those involving repair templates. Further improve-ments in the accuracy of nucleases by either mutagenesis or molec-ular evolution, or by mining the bacterial and archeal kingdoms foralternative more accurate enzymes, would be advantageous. Improvingmethodologies for the detection of off-target mutation and largedeletion and rearrangement events is also an area where continuedresearch would be beneficial. Furthermore, a more thorough exam-ination of the effects of base editing technology would also be ofconsiderable interest.

PGD provides one alternative strategy for combatting genetic dis-ease. However, there are concerns that, with reproductive age increas-ing in the Western world and given that PGD is known to impactreproductive success (Steffann et al., 2018), the number of viableembryos from which healthy individuals can be selected may frequentlybe too low for PGD selection to provide an effective solution. Thereare, of course, also situations where PGD cannot provide a solution,for example, where one patient is homozygous for a mutation. It isthus not too far-fetched to imagine a growing necessity to considerhuman germline editing in the future. It is thus clear that, in parallel withresearch addressing the safety concerns, the debate exploring ethicalaspects of human germline editing must continue.

FundingWellcome Trust Core Award Grant (203141/Z/16/Z).

Conflict of interestThe author has nothing to declare.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

ReferencesAdikusuma F, Piltz S, Corbett MA, Turvey M, McColl SR, Helbig KJ,

Beard MR, Hughes J, Pomerantz RT, Thomas PQ. Large deletionsinduced by Cas9 cleavage. Nature 2018;560:E8–E9.

Agrawal L, Jin Q, Altenburg J, Meyer L, Tubiana R, Theodorou I,Alkhatib G. CCR5Delta32 protein expression and stability are criticalfor resistance to human immunodeficiency virus type 1 in vivo. J Virol2007;81:8041–8049.

Agrawal L, Lu X, Qingwen J, VanHorn-Ali Z, Nicolescu IV, McDermottDH, Murphy PM, Alkhatib G. Role for CCR5Delta32 protein inresistance to R5, R5X4, and X4 human immunodeficiency virus type1 in primary CD4+ cells. J Virol 2004;78:2277–2287.

Allen F, Crepaldi L, Alsinet C, Strong AJ, Kleshchevnikov V, De AngeliP, Palenikova P, Khodak A, Kiselev V, Kosicki M et al. Predictingthe mutations generated by repair of Cas9-induced double-strandbreaks. Nat Biotechnol 2018.

Amoasii L, Hildyard JCW, Li H, Sanchez-Ortiz E, Mireault A, CaballeroD, Harron R, Stathopoulou TR, Massey C, Shelton JM et al.Gene editing restores dystrophin expression in a canine model ofDuchenne muscular dystrophy. Science 2018;362:86–91.

Anderson KR, Haeussler M, Watanabe C, Janakiraman V, Lund J,Modrusan Z, Stinson J, Bei Q, Buechler A, Yu C et al. CRISPR off-target analysis in genetically engineered rats and mice. Nat Methods2018;15:512–514.

Bae S, Kweon J, Kim HS, Kim JS. Microhomology-based choice of Cas9nuclease target sites. Nat Methods 2014;11:705–706.

Boroviak K, Fu B, Yang F, Doe B, Bradley A. Revealing hidden com-plexities of genomic rearrangements generated with Cas9. Sci Rep2017;7:12867.

Carroll D. Genome Editing: Past, Present, and Future. Yale J Biol Med2017;90:653–659.

Chakrabarti AM, Henser-Brownhill T, Monserrat J, Poetsch AR,Luscombe NM, Scaffidi P. Target-specific precision of CRISPR-mediated genome editing. Mol Cell 2018.

Chen Y, Liu X, Zhang Y, Wang H, Ying H, Liu M, Li D, Lui KO, DingQ. A self-restricted CRISPR system to reduce off-target effects. MolTher 2016;24:1508–1510.

Durrant DM, Daniels BP, Pasieka T, Dorsey D, Klein RS. CCR5 limitscortical viral loads during West Nile virus infection of the centralnervous system. J Neuroinflammation 2015;12:233.

Egli D, Zuccaro MV, Kosicki M, Church GM, Bradley A, Jasin M. Inter-homologue repair in fertilized human eggs? Nature 2018;560:E5–E7.

Falcon A, Cuevas MT, Rodriguez-Frandsen A, Reyes N, Pozo F,Moreno S, Ledesma J, Martinez-Alarcon J, Nieto A, Casas I. CCR5deficiency predisposes to fatal outcome in influenza virus infection.J Gen Virol 2015;96:2074–2078.

Fogarty NME, McCarthy A, Snijders KE, Powell BE, Kubikova N,Blakeley P, Lea R, Elder K, Wamaitha SE, Kim D et al. Genomeediting reveals a role for OCT4 in human embryogenesis. Nature2017;550:67–73.

Fu Y, Foden JA, Khayter C, Maeder ML, Reyon D, Joung JK. High-frequency off-target mutagenesis induced by CRISPR-Cas nucleasesin human cells. Nat Biotechnol 2013;31.

Haeussler M, Schonig K, Eckert H, Eschstruth A, Mianne J, RenaudJB, Schneider-Maunoury S, Shkumatava A, Teboul L, Kent J et al.Evaluation of off-target and on-target scoring algorithms and

2110 Davies

integration into the guide RNA selection tool CRISPOR. Genome Biol2016;17:148.

Hashimoto M, Yamashita Y, Takemoto T. Electroporation of Cas9protein/sgRNA into early pronuclear zygotes generates non-mosaicmutants in the mouse. Dev Biol 2016;418:1–9.

Hodgkins A, Farne A, Perera S, Grego T, Parry-Smith DJ, SkarnesWC, Iyer V. WGE: a CRISPR database for genome engineering.Bioinformatics 2015;31:3078–3080.

Hutter G, Nowak D, Mossner M, Ganepola S, Mussig A, Allers K,Schneider T, Hofmann J, Kucherer C, Blau O et al. Long-term controlof HIV by CCR5 Delta32/Delta32 stem-cell transplantation. N EnglJ Med 2009;360:692–698.

Iyer V, Boroviak K, Thomas M, Doe B, Riva L, Ryder E, Adams DJ.No unexpected CRISPR-Cas9 off-target activity revealed by triosequencing of gene-edited mice. PLoS Genet 2018;14:e1007503.

Joy MT, Ben Assayag E, Shabashov-Stone D, Liraz-Zaltsman S, MazzitelliJ, Arenas M, Abduljawad N, Kliper E, Korczyn AD, Thareja NS et al.CCR5 is a therapeutic target for recovery after stroke and traumaticbrain injury. Cell 2019;176:1143–1157.e1113.

Kang X, He W, Huang Y, Yu Q, Chen Y, Gao X, Sun X, Fan Y.Introducing precise genetic modifications into human 3PN embryosby CRISPR/Cas-mediated genome editing. J Assist Reprod Genet2016;33:581–588.

Kim D, Kim J, Hur JK, Been KW, Yoon SH, Kim JS. Genome-wideanalysis reveals specificities of Cpf1 endonucleases in human cells.Nat Biotechnol 2016;34:863–868.

Kim S, Kim D, Cho SW, Kim J, Kim JS. Highly efficient RNA-guidedgenome editing in human cells via delivery of purified Cas9 ribonu-cleoproteins. Genome Res 2014;24:1012–1019.

Kleinstiver BP, Pattanayak V, Prew MS, Tsai SQ, Nguyen NT,Zheng Z, Joung JK. High-fidelity CRISPR–Cas9 nucleases withno detectable genome-wide off-target effects. Nature 2016;529:490–495.

Kocak DD, Josephs EA, Bhandarkar V, Adkar SS, Kwon JB, GersbachCA. Increasing the specificity of CRISPR systems with engineeredRNA secondary structures. Nat Biotechnol 2019.

Kosicki M, Tomberg K, Bradley A. Repair of double-strand breaksinduced by CRISPR-Cas9 leads to large deletions and complexrearrangements. Nat Biotechnol 2018;36:765–771.

Krimsky S. Ten ways in which He Jiankui violated ethics. Nat Biotechnol2019;37:19–20.

Lander ES, Baylis F, Zhang F, Charpentier E, Berg P, Bourgain C,Friedrich B, Joung JK, Li J, Liu D et al. Adopt a moratorium onheritable genome editing. Nature 2019;567:165–168.

Lessard S, Francioli L, Alfoldi J, Tardif JC, Ellinor PT, MacArthurDG, Lettre G, Orkin SH, Canver MC. Human genetic variationalters CRISPR-Cas9 on- and off-targeting specificity at therapeuticallyimplicated loci. Proc Natl Acad Sci U S A 2017;114:E11257–e11266.

Li G, Liu Y, Zeng Y, Li J, Wang L, Yang G, Chen D, Shang X, Chen J,Huang X et al. Highly efficient and precise base editing in discardedhuman tripronuclear embryos. Protein Cell 2017;8:776–779.

Liang P, Ding C, Sun H, Xie X, Xu Y, Zhang X, Sun Y, Xiong Y, Ma W,Liu Y et al. Correction of beta-thalassemia mutant by base editor inhuman embryos. Protein Cell 2017;8:811–822.

Liang P, Sun H, Zhang X, Xie X, Zhang J, Bai Y, Ouyang X, Zhi S, XiongY, Ma W et al. Effective and precise adenine base editing in mousezygotes. Protein Cell 2018;9:808–813.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Liang P, Xu Y, Zhang X, Ding C, Huang R, Zhang Z, Lv J, Xie X,Chen Y, Li Y et al. CRISPR/Cas9-mediated gene editing in humantripronuclear zygotes. Protein Cell 2015;6:363–372.

Long C, McAnally JR, Shelton JM, Mireault AA, Bassel-Duby R,Olson EN. Prevention of muscular dystrophy in mice by CRISPR/Cas9-mediated editing of germline DNA. Science 2014;345:1184–1188.

Ma H, Marti-Gutierrez N, Park S-W, Wu J, Hayama T, Darby H,Van Dyken C, Li Y, Koski A, Liang D et al. Ma et al. reply. Nature2018;560:E10–E23.

Ma H, Marti-Gutierrez N, Park SW, Wu J, Lee Y, Suzuki K, Koski A,Ji D, Hayama T, Ahmed R et al. Correction of a pathogenic genemutation in human embryos. Nature 2017;548:413–419.

National Academies of Sciences, Engineering, and Medicine, Policy andGlobal Affairs. The National Academies Collection: Reports fundedby National Institutes of Health. In: Second International Summit onHuman Genome Editing: Continuing the Global Discussion: Proceedingsof a Workshop-in Brief . US: National Academies Press, 2019

Nelson CE, Hakim CH, Ousterout DG, Thakore PI, Moreb EA,Castellanos Rivera RM, Madhavan S, Pan X, Ran FA, Yan WX et al.In vivo genome editing improves muscle function in a mouse modelof Duchenne muscular dystrophy. Science 2016;351:403–407.

Parikh BA, Beckman DL, Patel SJ, White JM, Yokoyama WM.Detailed phenotypic and molecular analyses of genetically modi-fied mice generated by CRISPR-Cas9-mediated editing. PLoS One2015;10:e0116484.

Porteus MH. A new class of medicines through DNA editing. N Engl JMed 2019;380:947–959.

Rees HA, Liu DR. Base editing: precision chemistry on the genome andtranscriptome of living cells. Nat Rev Genet 2018;19:770–788.

Sander JD, Joung JK. CRISPR-Cas systems for editing, regulating andtargeting genomes. Nat Biotechnol 2014;32:347–355.

Savulescu J, Pugh J, Douglas T, Gyngell C. The moral imperativeto continue gene editing research on human embryos. Protein Cell2015;6:476–479.

Shen C-C, Hsu M-N, Chang C-W, Lin M-W, Hwu J-R, Tu Y, HuY-C. Synthetic switch to minimize CRISPR off-target effects byself-restricting Cas9 transcription and translation. Nucleic Acids Res2018a.

Shen MW, Arbab M, Hsu JY, Worstell D, Culbertson SJ, KrabbeO, Cassa CA, Liu DR, Gifford DK, Sherwood RI. Predictable andprecise template-free CRISPR editing of pathogenic variants. Nature2018b;563:646–651.

Shin HY, Wang C, Lee HK, Yoo KH, Zeng X, Kuhns T, Yang CM, MohrT, Liu C, Hennighausen L. CRISPR/Cas9 targeting events causecomplex deletions and insertions at 17 sites in the mouse genome.Nat Commun 2017;8:15464.

Slaymaker IM, Gao L, Zetsche B, Scott DA, Yan WX, Zhang F. Ratio-nally engineered Cas9 nucleases with improved specificity. Science2016;351:84–88.

Steffann J, Jouannet P, Bonnefont JP, Chneiweiss H, Frydman N. Couldfailure in preimplantation genetic diagnosis justify editing the humanembryo genome? Cell Stem Cell 2018;22:481–482.

Taheri-Ghahfarokhi A, Taylor BJM, Nitsch R, Lundin A, Cavallo AL,Madeyski-Bengtson K, Karlsson F, Clausen M, Hicks R, Mayr LM et al.Decoding non-random mutational signatures at Cas9 targeted sites.Nucleic Acids Res 2018;46:8417–8434.

The technical risks of human gene editing 2111

Tang L, Zeng Y, Du H, Gong M, Peng J, Zhang B, Lei M, Zhao F, WangW, Li X et al. CRISPR/Cas9-mediated gene editing in human zygotesusing Cas9 protein. Mol Gen Genomics 2017;292:525–533.

Tang L, Zeng Y, Zhou X, Du H, Li C, Liu J, Zhang P. Highly effi-cient ssODN-mediated homology-directed repair of DSBs gener-ated by CRISPR/Cas9 in human 3PN zygotes. Mol Reprod Dev2018;85:461–463.

Teng F, Cui T, Feng G, Guo L, Xu K, Gao Q, Li T, Li J, Zhou Q, Li W.Repurposing CRISPR-Cas12b for mammalian genome engineering.Cell Discov 2018;4:63.

Tu Z, Yang W, Yan S, Yin A, Gao J, Liu X, Zheng Y, Zheng J, Li Z,Yang S et al. Promoting Cas9 degradation reduces mosaic mutationsin non-human primate embryos. Sci Rep 2017;7:42081.

Vakulskas CA, Dever DP, Rettig GR, Turk R, Jacobi AM, CollingwoodMA, Bode NM, McNeill MS, Yan S, Camarena J et al. A high-fidelityCas9 mutant delivered as a ribonucleoprotein complex enablesefficient gene editing in human hematopoietic stem and progenitorcells. Nat Med 2018;24:1216–1224.

van Dijke I, Bosch L, Bredenoord AL, Cornel M, Repping S, Hendriks S.The ethics of clinical applications of germline genome modification:a systematic review of reasons. Hum Reprod 2018;33:1777–1796.

Wang H, Yang H, Shivalila CS, Dawlaty MM, Cheng AW, ZhangF, Jaenisch R. One-step generation of mice carrying mutations inmultiple genes by CRISPR/Cas-mediated genome engineering. Cell2013;153:910–918.

Wei X, Nielsen R. CCR5-�32 is deleterious in the homozygous statein humans. Nat Med 2019;25:909–910.

Yang Y, Wang L, Bell P, McMenamin D, He Z, White J, Yu H, Xu C,Morizono H, Musunuru K et al. A dual AAV system enables the

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Cas9-mediated correction of a metabolic liver disease in newbornmice. Nat Biotechnol 2016;34:334–338.

Yao X, Zhang M, Wang X, Ying W, Hu X, Dai P, Meng F, Shi L,Sun Y, Yao N et al. Tild-CRISPR allows for efficient and precise geneknockin in mouse and human cells. Dev Cell 2018;45:526–536.e525.

Yen S-T, Zhang M, Deng JM, Usman SJ, Smith CN, Parker-ThornburgJ, Swinton PG, Martin JF, Behringer RR. Somatic mosaicism andallele complexity induced by CRISPR/Cas9 RNA injections in mousezygotes. Dev Biol 2014;393:3–9.

Zafer M, Horvath H, Mmeje O, van der Poel S, Semprini AE,Rutherford G, Brown J. Effectiveness of semen washing to preventhuman immunodeficiency virus (HIV) transmission and assist preg-nancy in HIV-discordant couples: a systematic review and meta-analysis. Fertil Steril 2016;105:645–655.e642.

Zeng Y, Li J, Li G, Huang S, Yu W, Zhang Y, Chen D, Chen J, Liu J, HuangX. Correction of the Marfan syndrome pathogenic FBN1 mutationby base editing in human cells and heterozygous embryos. Mol Ther2018;26:2631–2637.

Zhou M, Greenhill S, Huang S, Silva TK, Sano Y, Wu S, Cai Y, NagaokaY, Sehgal M, Cai DJ et al. CCR5 is a suppressor for cortical plasticityand hippocampal learning and memory. Elife 2016;5.

Zuo E, Sun Y, Wei W, Yuan T, Ying W, Sun H, Yuan L,Steinmetz LM, Li Y, Yang H. Cytosine base editor generates substan-tial off-target single-nucleotide variants in mouse embryos. Science2019;364:289–292.

Zuris JA, Thompson DB, Shu Y, Guilinger JP, Bessen JL, Hu JH, MaederML, Joung JK, Chen Z-Y, Liu DR. Cationic lipid-mediated delivery ofproteins enables efficient protein-based genome editing in vitro andin vivo. Nat Biotechnol 2015;33:73.