Genome Editing Technologies: Defining a Path to Clinic

The MIT Faculty has made this article openly available. Please share how this access benefits you. Your story matters.

As Published http://dx.doi.org/10.1038/mt.2015.54

Publisher Elsevier BV

Version Final published version

Citable link http://hdl.handle.net/1721.1/112722

Detailed Terms http://creativecommons.org/licenses/by-nc-nd/4.0/ meeting report © The American Society of Gene & Cell Therapy doi:10.1038/mt.2015.54

palindromic repeats (CRISPR) are being investigated as promising tools for human gene therapy. ZFNs are the first class Genome Editing Technologies: of nucleases to have reached the clinic in phase I trials for HIV.2 One key issue that Deﬁning a Path to Clinic will have to be addressed as these technologies move into clinical trials is whether Genomic Editing: Establishing Preclinical technology-specific preclinical evaluations are available that can establish safety. Al- Toxicology Standards though these are targeted editing tools, their precision, and specifically the degree Bethesda, Maryland to which there are off-target actions and 10 June 2014 the clinical implications of such activity are important questions for the field. Jacqueline Corrigan-Curay1, Marina O’Reilly1, Donald B Kohn2, Paula M Cannon3, Gang Bao4, Frederic D Bushman5, Dana Carroll6, Overview of genome editing Toni Cathomen7, J Keith Joung8,9, David Roth10, Michel Sadelain11, technologies Andrew M Scharenberg12, Christof von Kalle13, Feng Zhang14, The fundamental process common to all of these technologies is the use of nucleases to Robert Jambou1, Eugene Rosenthal1, Morad Hassani1, make site-specific double-stranded breaks 1 15 Aparna Singh and Matthew H Porteus (DSBs) in the genome. Several approaches to genome editing have been developed; ecently developed genomic editing ate off-target effects and establish safety. In this summary describes them briefly. Rtechnologies have the potential to be order to begin a dialogue on these issues, powerful tools for gene therapy because the National Institutes of Health (NIH) ZFNs. ZFNs are the most clinically ad- of their ability to inactivate genes, correct Office of Science Policy, in collaboration vanced nuclease platform. Each zinc-fin- mutated sequences, or insert intact genes. with several NIH-funded investigators ger consists of ~30 amino acids that fold While the genomic editing field is advanc- and the NIH Recombinant DNA Advisory into a conserved ββα configuration.3 Each ing at an exceptionally rapid pace, there Committee, organized a workshop on 10 “finger” recognizes about three or four remain key issues regarding development June 2014, in Bethesda, Maryland, to pro- base pairs of DNA using at least six amino of appropriate preclinical assays to evalu- vide a forum to educate the scientific and acids through contacts between specific oversight communities and the public on residues in the second α-helix, also known 1National Institutes of Health, Bethesda, Maryland, USA; 2University of California, Los different genome editing technologies, as the “recognition helix” terminus. Three Angeles, Los Angeles, California, USA; 3Keck clinical experiences to date, and the pre- to six individual fingers can be linked to School of Medicine, University of Southern Cali- clinical assays being developed to examine enable construction of arrays that recog- fornia, Los Angeles, California, USA; 4Georgia the precision of these tools and their suit- nize longer sequences of 9–18 base pairs Institute of Technology, Atlanta, Georgia, USA; ability for clinical application. (bp). Of note, 18 bp of DNA sequence can 5University of Pennsylvania, School of Medicine, Philadelphia, Pennsylvania, USA; 6University of Targeted genome modification by de- confer specificity within 68 billion base Utah, School of Medicine, Salt Lake City, Utah, signer nucleases is an emerging technol- pairs of DNA. Further specificity can be USA; 7University Medical Center, Freiberg, Frei- ogy that can be used to investigate gene engineered by changing critical residues 8 berg, Germany; Massachusetts General Hos- function and could also be used to treat within the recognition helices. In addi- pital, Charlestown, Massachusetts; 9Harvard Medical School, Boston, Massachusetts, USA; genetic or acquired diseases. A wide range tion, individual fingers in an array can 10University of Pennsylvania, School of Medi- of genome alterations has been achieved potentially interact and make base-specific cine, Philadelphia, Pennsylvania, USA; 11Memo- by these nucleases, including localized contacts into the sequences recognized by rial Sloan-Kettering Cancer Center, New York, mutagenesis, local and dispersed sequence adjacent fingers, and optimization of bind- New York, USA; 12Seattle Children’s Research replacements, large and small insertions ing can therefore be potentially altered by Institute and University of Washington, School of Medicine, Seattle, Washington, USA; 13Na- and deletions, and even chromosomal interactions between individual fingers. tional Center for Tumor Diseases and German translocations. The nuclease approach to The nuclease domain of the zinc-finger is Cancer Research Center, Heidelberg, Germany; targeted genome editing has been applied derived from the C-terminus of the FokI 14 Massachusetts Institute of Technology and successfully to more than 50 different or- restriction endonuclease. FokI only cuts Harvard University, Cambridge, Massachusetts, USA; 15Stanford University, California, Stanford, ganisms, including crop plants, livestock, DNA when it dimerizes, so two sets of California, USA and humans.1 zinc-fingers are required (seeFigure 1a). Correspondence: Jacqueline Corrigan-Curay, Recently developed genome editing The natural enzyme generates a 5 over- Ofﬁce of Science Policy, National Institutes technologies such as zinc finger nucleases hang. However, one can engineer these en- of Health, 6705 Rockledge Drive, Bethesda, (ZFNs), transcription activator-like effec- zymes such that within the dimer, one side Maryland 20892, USA. E-mail: [email protected] tor nucleases (TALENs), meganucleases, of the nuclease domain is catalytically ac- and clustered regularly interspaced short tive while its twin is catalytically inactive,

796 www.moleculartherapy.org vol. 23 no. 5 may 2015 © The American Society of Gene & Cell Therapy meeting report

ing up to six mismatches and/or bulges. arrays that recognize 13–20 bp can be Similar to ZFNs, it is possible to make a constructed. The nuclease domain used nick rather than a break with the CRISPR/ in TALENs is also from FokI; however, in Cas9 system. However, unlike the ZFNs contrast to ZFNs and for reasons that are and TALENs, the double-strand cut made not yet understood, TALENs cannot be by CRISPR/Cas9 leaves a blunt end. manipulated to create nicks rather than DSBs. Homing endonucleases (meganucleases). These are natural proteins called CRISPR/CRISPR-associated (Cas) pro- endodeoxyribonucleases that recognize tein 9. CRISPR/Cas9 is distinct from long (>12 bp) DNA sequences with high the previous engineered endonucleases specificity. Described as genetic parasites, in that it uses an RNA-guided system to they target the recognition site in an al- perform genome editing. This platform, lele, make a break in that allele, and target derived from a bacterial innate immune the transfer of their own protein-coding system, was described relatively recently, sequence into that allele by homologous but progress on its development has been recombination. They generate a DSB with rapid. It has captured considerable atten- a four-base 3 overhang, which is proposed tion due to the relative ease of engineering to be a natural substrate for the homolo- its RNA-based targeting component. In gous recombination machinery. Initial Figure 1 Nuclease site recognition features bacteria, type II CRISPR systems process genome editing manipulations were done (a) Zinc-ﬁnger nuclease dimer: recognition sites foreign sequences from invading phages using the I-CreI and the I-SceI endonucle- 9–18 bp × 2. (b) TAL effector nuclease (TALEN): recognition sites 12–20 bp × 2, spacer 12–20 bp. or plasmids into small segments that are ases. Engineering these nucleases is dif- (c) RNA-guided endonuclease CRISPR/Cas9. then introduced into the CRISPR array, ficult primarily because both recognition Courtesy of Matthew Porteus which contains the regularly interspaced and enzymatic activity are intertwined palindromic repeats. These snippets of for- within the protein, often making it difficult which will generate a single-stranded nick eign DNA become templates for CRISPR to alter one without having an effect on the rather than a DSB. RNA (crRNA), which now contains a vari- other. However, these nucleases are small able sequence from the invading DNA. TALENS. TALENs are similar in architec- This crRNA then hybridizes with atrans - ture to ZFNs except that they use a differ- activating RNA (tracrRNA), and the RNAs ent DNA-binding domain. They consist of form complexes with the Cas9 protein. The arrays of single protein modules that each next time this foreign sequence is detected, recognize a single DNA base pair and that it is cleaved and degraded (see Figure 1c).7 are derived from transcription activator- Recognition of the target DNA se- like effectors (TALEs), factors encoded by quence is mediated between the genomic plant pathogenic bacteria4 (see Figure 1b). DNA target and by a 20-nucleotide se- Each of these modules is about 34 amino quence in the crRNA. Another feature Figure 2 Mechanisms of DNA repair after targeted cleavage.HDR, homology-depen- acids long, and they are nearly identical of this system is that the Cas9 protein is dent repair; NHEJ, nonhomologous end except for the identities of amino acids at directed to cleave the complementary joining. Courtesy of Dana Carroll. positions 12 and 13, which together are target-DNA sequence if it is adjacent to a and can be easily contained in commonly known as the “repeat variable di-residue”. short sequence known as the protospacer used vectors. In sum, they offer potentially To engineer DNA-binding domains with adjacent motif (PAM). The PAM sequence high specificity, but barriers to their engi- novel DNA-binding specificities, indi- commonly used from Streptococcus pyo- neering are high.9 vidual TALE repeats are assembled into genes has the sequence [N]GG, although an array that is designed to recognize the [N]AG can also be used. Many bacteria Impact of double-stranded breaks target DNA sequence. Although the sin- have the Cas9 system, but not all of them The mechanism unifying each of these gle-nucleotide specificity of TALE repeats possess the same PAM sequences, provid- technologies is the ability to make targeted potentially offers greater design flexibility ing some additional variability. In addi- DSBs in genomic DNA. The outcome of than do zinc-fingers, their highly repeti- tion, mutations in the PAM sequence will targeted cleavage depends on cellular re- tive nature presents technical challenges in prevent the Cas9 protein from causing a pair pathways of DSBs, which are poten- assembling DNA-encoding arrays of these break at that site. In 2012 it was shown that tially lethal to the cell unless they are re- domains. Different strategies have been the crRNA and the tracrRNA can be com- paired quickly. The principal mechanisms developed to facilitate rapid assembly of bined into a single RNA molecule known of repair are nonhomologous end joining DNA-encoding TALE repeats, includ- as a guide RNA (or gRNA) that can still (NHEJ) and homology-dependent repair ing the “Golden Gate” assembly method5 engage the Cas9 protein.8 As discussed (HDR) (see Figure 2). NHEJ often results and a system called FLASH.6 TALE repeat later, gRNAs recognize 20-bp target sites in sequence changes at the cut site, most but can also recognize off-target sites bear- commonly variable-length insertions or

Molecular Therapy vol. 23 no. 5 may 2015 797 meeting report © The American Society of Gene & Cell Therapy

deletions, referred to as “indels,” which can insertion applications. more sensitive to mismatches and there- be detected through sequencing or mis- fore reduces off-target site cleavage.18 This match cleavage assays that use enzymes Improving speciﬁcity through truncation approach does not work for all such as CelI or T7 endonuclease I and design gRNAs but appears to work well for the more recently a new method called TIDE Because genome editing is directed at spe- vast majority of gRNAs tested to date. An- (tracking of indels by decomposition).10 cific sites, this technology offers greater other strategy is to combine the specificity There are subtle differences among precision compared to other approaches from different platforms. For example, it is the NHEJ products generated by each to long-term gene modification, e.g., deliv- possible to combine the dimerization-de- of the above nucleases. Small deletions ery of a gene by an integrating viral vector. pendent FokI nuclease used in ZFNs and and insertions are common in NHEJ Most integrating viral vectors have a large- TALENs with the CRISPR/Cas9 system. mutations, but the average deletion size ly random integration pattern. However, By fusing the FokI nuclease to a catalyti- is somewhat larger with TALENs and despite the elegant precision that these ed- cally inactive Cas9, one can use the gRNAs ZFNs compared to CRISPRs, and in- iting tools offer, off-target effects are likely, to direct binding but require dimeriza- sertions are somewhat more frequent depending upon the construct and the tion for cleavage.19,20 In one experiment in with ZFNs.1 Rarely are insertions large length of the target site. Ideally, the inser- which five off-target sites were previously enough that the source of the inserted tions, deletions, inversions, and transloca- observed with the use of a specific gRNA, sequences can be identified.11 Such find- tions that may result from NHEJ or HDR deep sequencing was used to confirm that ings highlight that the mechanisms of at these off-target sites must be minimized the frequency of indel mutations induced DSB repair are still not completely un- before moving to the clinic. by the dimeric CRISPR (or RNA-guided derstood. Improving on the natural design. FokI nuclease) was not greater than back- HDR can incorporate user-provided much has been learned regarding how to ground at all five off-target sites, indicat- sequence changes from a donor DNA. use these tools to target specific sequences, ing the ability of this platform to eliminate NHEJ dominates in almost all systems, and the designs of these naturally occur- activity at the off-target sites of a single but there may be ways to shift the bal- ring tools are being modified to improve gRNA. This hybrid platform provides an ance toward HDR through downregu- the specificity. At the workshop, Keith increase in the length of the binding site by lation of enzymes that are involved in Joung reviewed his research with different utilizing both gRNAs in a dimeric configu- the NHEJ pathway or modulating the platforms—ZFNs, TALENs, and CRISPR/ ration and reduction in potential off-target stage of the cell cycle. In addition, alter- Cas9 nucleases—in which he generally has sites.19 Interestingly, in that experiment ing engineered nucleases to produce a seen stronger binding to on-target sites the authors also failed to see detectable single-stranded nick rather than a DSB compared with off-target sites, but there evidence of off-target activity for these di- will favor HDR, but at an overall lower is still considerable activity at off-target meric RNA-guided FokI nucleases, even at absolute frequency of repair.12,13 sequences. Focusing on the CRISPR/Cas9 the most closely mismatched sites in the If the goal is to incorporate new ge- platform, he examined the off-target ef- genome. netic material rather than just to disrupt fects of first-generation CRISPR/Cas9 Another approach being explored is a particular sequence, the nature of the agents that were directed by six different to mutate the Cas9 protein so that it will donor DNA will also influence the suc- gRNAs and demonstrated that there were only create a nick rather than a DSB. By cess of the approach. With long dou- a number of off-target sites (harboring as then pairing two such mutated Cas9 pro- ble-stranded donor DNAs, successful many as five mismatches relative to the on- teins with gRNAs that are offset, one can incorporation requires several hundred target site) and that the rate of mutagenesis create DSBs while increasing the specific- base pairs of homology on both sides at these sites could be as high as that seen ity of cleavage and significantly reducing of the nuclease-induced break.11,14 Short at the on-target site.17 This analysis was off-target breaks.21 Essentially this ap- single-stranded donor DNAs can also done across different cell types, and the proach attempts to increase the specificity be used, but such templates would be mutations sometimes fell within the cod- in the same way that dimerizing nucleases limited to applications requiring small ing sequences of the genome. does but with the caveat that monomeric changes close to the DSB, such as correc- This finding has led to the development nickases still have the potential to induce tion of point mutations. In some systems of second-generation CRISPR/Cas9 agents mutations.19 Further research is needed to (e.g., Drosophila), sequences throughout with modifications to increase specificity. determine whether these strategies can be the length of a long donor can be cap- Intuitively one might conclude that in- combined to enhance specificity further. tured at the target, but in other cases creasing the length of the binding site for A recent study describes a method for (e.g., cultured mammalian cells) only the nuclease would increase specificity, but gRNA design that significantly enhanced sequences close to the break are rou- Dr Joung found that for the CRISPR/Cas the frequency of genome editing by Cas9 tinely incorporated, although sequences 9 system, specificity can be improved by in Caenorhabditis elegans. The key inno- several kilobases away can occasionally truncating the 5 end of the gRNA by as vation was to design gRNAs with a GG be incorporated.15,16 Further knowledge many as three nucleotides. This truncation motif at the 3 end of their target-specific of the activities that control this feature, generally does not impair the ability of a sequences. All guides designed for all known as “conversion tract length,” is gRNA to direct on-target site cleavage but targets supported robust genome editing, needed as the field moves toward gene appears to make the gRNA/Cas9 complex both imprecise NHEJ events and precise,

798 www.moleculartherapy.org vol. 23 no. 5 may 2015 © The American Society of Gene & Cell Therapy meeting report

template HDR events.22 Identifying safe harbors. An alterna- grated transgene on neighboring gene Dr Scharenberg reviewed work he tive strategy to achieve safe, targeted expression. This can be more easily ac- has done with Cellectis Therapeutics and gene delivery and limit off-target activity complished in cell types that can be cloned Seattle Children’s Research Institute com- is to identify sites in the human genome such as pluripotent stem cells or T lym- bining a meganuclease, or homing endo- that are at minimal risk of causing in- phocytes, and may not be feasible in cells nuclease, with a TAL array.23 The goal is to sertional oncogenesis upon integration such as neural cells or hematopoietic stem combine the high binding specificity of the of foreign DNA, while being accessible cells. Moreover, true GSHs would have to TAL with the high cleavage specificity of to a highly specific nuclease with mini- tolerate (i.e., without unintended trans- the meganuclease. By fusing a site-specific mal off-target activity. Such “genomic formation) the integration of a number of meganuclease to a TAL array that binds safe harbors” may be extragenic sites different elements, including promoters, adjacent to the meganuclease target site that are remote from a gene or genomic enhancers, and chromatin determinants. (thus tethering the meganuclease adjacent regulatory sequence, or intragenic sites Finally, validation of the safety of integra- to its desired target site), one can increase (within a gene) whose disruption is tion at these safe harbors should be done in cleavage activity at that target site and deemed to be tolerable. Drawing from animals; however, there are challenges in minimize off-target activity, as tethering human clinical trial data on integration developing appropriate animal models.25 will not occur at off-target sites lacking an sites for retroviral and lentiviral vectors, adjacent sequence capable of being bound several researchers24,25 have proposed Identifying and evaluating the by the TAL array. The homing endonucle- the following criteria that could consti- impact of off-target activity ase has to be engineered, but the specificity tute an extragenic safe harbor for DNA Although innovative designs and explora- can be improved through the engineering integration: A safe harbor should be (i) tion of safe harbors are certainly important of the TAL, which is easier. outside a gene transcription unit; (ii) lo- strategies, any clinical development strat- Scharenberg and colleagues have ex- cated >50 kilobases (kb) from the 5 end egy will have to include identification and plored this construct (called a megaTAL) of any gene; (iii) located >300 kb from evaluation of potential off-target sites. A in T cells, in which the goal is to engineer cancer-related genes; (iv) located >300 complete catalog of off-target sites might the T cell to express a specific chimeric an- kb from any identified microRNA; and be accomplished using whole-genome se- tigen receptor and at the same time disrupt (v) outside ultra-conserved regions and quencing to look for evidence of indels and the native T-cell receptor alpha (TCRα). long noncoding RNAs. In studies of len- translocations. However, this approach This approach would allow for allogeneic tiviral vector integrations in transduced would also be very costly, especially for designer T cells, currently being tested induced pluripotent stem cells, analysis less frequent events, and the sequencing for a number of oncology applications, to of over 5,000 integration sites revealed itself has an error rate.26 For example, if the function without the risk of graft-vs.-host that ~17% of integrations occurred in off-target cleavage occurs at a frequency of disease (GVHD). The activity and speci- safe harbors. The vectors that integrated 0.1% per genome, at least 1,000 genomes ficity of the meganuclease targeting TCRα into these safe harbors were able to ex- may have to be sequenced to capture such was assessed both with and without fusion press therapeutic levels of β-globin from a low-frequency event, adding a significant to a TAL array that could bind a DNA se- their transgene without perturbing en- cost to the analysis. Furthermore, as DSBs quence upstream from the cleavage site for dogenous gene expression.24 and repair can occur in cells as a result of the TCRα-specific meganuclease. Cleav- Several candidate genomic safe-har- culture conditions alone (even in the ab- age of the TCRα gene using the meganu- bor sites (GSHs) have been explored, in- sence of exogenous nucleases), there is the clease alone was ~1.6% but was increased cluding AAVS1, CCR5, and the ROSA26 challenge of distinguishing the actions of by 20-fold with the megaTAL construct. locus. Although there are clinical data nucleases from naturally occurring back- The megaTAL was then tested with co- for CCR5 knockout in T cells and other ground DNA breaks and the spontaneous transfection of each nuclease with Trex2, data showing the safety of integration formation of small indels. In addition, a 3 endonuclease that can trim back the into AAVS1 in human cultured T cells, with ongoing deep-sequencing projects, 3 overhangs that homing endonucleases these sites have not been validated as it is now recognized that any individual’s make, thereby markedly accentuating the universal GSHs.25 In addition, it is not genome can contain up to 750,000 unique generation of indels. Addition of Trex2 known whether the gene-rich loci of indels.27 As a result of these limitations, further increased the rate of disruption, these sites, including some oncogenes, many groups have used a focused ap- yielding rates of TCRα disruption consis- will limit their use when targeting other proach to base prediction of potential tently exceeding 70%.23 cell lines. off-target sites sequence similarity to the This research underscores that the Of course, much remains to be learned on-target site followed by experimental specificity of these tools could be en- about sites that are identified as being confirmation to validate those predictions. hanced further by combining them to take GSHs. Non-protein-coding sequences However, the significant disadvantage advantage of their respective specificity may be more prevalent than currently of such a focused approach is that it will and ease of engineering. Nonetheless, an known, and some data suggest that there potentially miss other potential off-target assessment of safety and specificity using may be low levels of transcription in in- sites that could have clinical significance. preclinical assays will still be necessary, no tragenic sites. Validation of such sites will Bioinformatics tools are being de- matter which nuclease platform is utilized require measuring the effect of the inte- veloped to predict off-target sites. A

Molecular Therapy vol. 23 no. 5 may 2015 799 meeting report © The American Society of Gene & Cell Therapy

number of web-based tools have been target, for others small alterations in a Drawing from work with integrating developed, such as Predicted Report of similar sequence would prevent binding, viral vectors, Christof von Kalle has de- Genome-wide Nuclease Off-Target Sites e.g., a thymine methylation that leads to veloped an assay using integrase-defec- (PROGNOS, http://baolab.bme.gatech. a 1,000-fold diminution in affinity. The tive lentiviral vectors (IDLVs) to identify edu/Research/BioinformaticTools/prog- future for these tools may be to incor- off-target breaks. IDLVs—like any other nos.html). PROGNOS can provide a re- porate biochemical data that will allow extrachromosomal DNA—occasionally port of potential genome-wide nuclease for some ranking of the most likely off- get trapped in a DSB during NHEJ re- target sites for ZFNs and TALENs. Once target sites. With further improvements, pair, thereby stably marking these other- a particular target site is identified, the these tools will be useful to allow rank- wise transient and undetectable events. program can provide a rank list of po- ing of off-target sites; however, the fre- IDLV integration sites in cells treated tential off-target sites. These tools are quency at which indels occur at these with ZFNs targeting the human genes just being developed, and their valida- sites will be determined by the specific CCR5 and IL2RG have been analyzed by tion will require more data on actual cell type being manipulated (because linear amplification–mediated (LAM) off-target sites from specific constructs. different cell types have different muta- PCR. A clustering of IDLV integration However, once validated they have the genic properties) and the duration and sites was detected at the ZFNs on-target potential to offer a roadmap to search level of nuclease expression. site indicative of ZFN activity. However, for off-target sites. a few other genomic positions show such In evaluating the nature of off-target Unbiased analysis of genome clustering of IDLVs indicative of off-tar- sites, the problem is determining what breaks get activity at those loci. Molecular anal- degree of similarity the sequence must In addition to identifying potential bind- yses confirmed that off-target activity possess to lead potentially to binding sites for the nucleases, it is important occurred at genomic positions bearing ing and cleavage. For example, for the to understand whether DSBs are occurring homology to the ZFNs target site. With CRISPR/Cas9 system, the 20-bp bind- and what the implications of those breaks the detection of ZFNs off-target binding sequence can tolerate mismatches are. The field refers to these approaches as ing sites, one could then measure the between the gRNA and its complemen- “unbiased,” as they are attempting to mea- frequency of off-target cleavage at a spe- tary target DNA sequence resulting in sure the off-target DSBs. cific off-target site by deep sequencing binding and nuclease action. The degree Although the first step in the action to identify the exact nucleotide positions to which the CRISPR/Cas9 system will of these nucleases is to bind to the recog- within the ZFNs target sequence that tol- bind to these mismatched sequences nition sequence, the nuclease must then erate nonspecific sequence recognition, may depend on the number, location, cut the DNA, and it is the DSB that is the thereby contributing to off-target activ- and nature of mismatches,21 but larger key off-target activity of interest. Sev- ity. Interestingly, the presence of a highly data sets are needed to discern whether eral biochemical approaches are being homologous sequence did not reliably predictive rules can be derived. Bind- developed not only to look at the poten- predict off-target activity, indicating that ing might occur even if there is an ex- tial binding sites but also to see whether additional unknown cellular factors also tra DNA base pair (sometimes referred there is actual DNA cleavage. Dr Liu and influence target site recognition. Using to as a DNA bulge) or an extra RNA his colleagues examined whether there a similar technique, the specificity of nucleotide (an RNA bulge). Based on the were actual DNA breaks by ZFNs at TALENs targeting the human COL7A1 analysis of how specific CRISPR gRNAs off-target sites using an in vitro method gene was analyzed. Only three off-target could still bind with such mismatches that combined libraries of potential off- positions could be detected by the de- together with DNA or RNA bulges, Gang target binding sites and deep sequencing scribed IDLV capture approach. Thus, Bao and his colleagues developed a new to look for evidence of actual cleavage at experimental determination of the off- tool called CRISPR Off-target Sites with those sites.29 Another method was used target activity for each designer nucle- Mismatches, Insertions and Deletions to determine the nature of off-target se- ase may be required. These experiments (COSMID).28 This tool is now being fur- quences cleaved by two ZFNs (CCR5 224 represent an approach to move toward ther validated using published data on and VF2468) currently in clinical trials. genome-wide determination of designer CRISPR off-target sites to evaluate the They created a series (1012 total) of mu- nuclease–associated off-target activ- accuracy of the predictions. tated half-sites and determined which ity but also demonstrate that ZFNs and While these tools are elegant and pro- of those mutations were recognized and TALENs can modify the host genome vide useful data, they are just a starting cleaved and at what frequency by the two with an extraordinarily high selectivity. point, as experimental data are needed functional ZFNs. Using PCR and deep Although this method is promising, to validate whether the identified targets sequencing, the authors were able to the lower limit of sensitivity remains to are real. As noted by Frederic Bushman, identify specific nucleotide changes that be defined and the use of a viral vec- with any target there will be a signifi- could lead to off-target cleavage by both tor has the potential to introduce some cant number of potential binding sites, ZFNs. They experimentally showed that bias into this analysis. The question then some favored and others less so. While many off-target sequences were present becomes whether it is important from a some nonfavored sites appear to be al- and identified in K562 cells grown in tis- clinical perspective to capture all DSBs most as good a match as the intended sue culture.29 or just those that happen at a particular

800 www.moleculartherapy.org vol. 23 no. 5 may 2015 © The American Society of Gene & Cell Therapy meeting report

frequency. It was noted that as cells are different sites, and (iii) a CRISPR/Cas9 von Kalle with the IDLVs. Southern blot cultured over several days, there is al- to a single target site. Different doses of analysis indicated that 2% of the factor IX ways the potential for DSBs, thus some the nucleases were used. In this experi- alleles (2% of the haploid genomes) were background level is already tolerated. ment, the higher dose of ZFNs resulted modified by homologous recombination. Recently, Dr Joung’s lab described in a significant gamma-H2AX signal However, qPCR detected that 40% of the an unbiased, sensitive, and genome- with a dose response, which was also ob- haploid genomes had an integrated AAV wide approach for identifying DSBs in- served with the use of TALENs, but with vector, much more than those that had un- duced by CRISPR/Cas9 nucleases. This the CRISPR/Cas9 there was no increase dergone homologous recombination. This method, known as genome-wide unbi- in gamma-H2AX. The CRISPR was ac- suggested that there may be up to 20 times ased identification of DSBs enabled by tive at the target site, raising the question as many off-targets as compared to on-tar- sequencing (GUIDE-seq), relies on cap- as to whether there is something differ- get sites. By using deep sequencing the au- ture of short double-stranded oligode- ent about the CRISPR/Cas9 break that is thors were able to examine the NHEJ sites oxynucleotides (dsODNs) into CRISPR/ not well understood and does not result and determined that approximately 3% of Cas9-induced DSBs in cultured human in the formation of gamma-H2AX. the sites were on-target and the remaining cells.19 Fragments of genomic DNA har- It is important to note that many of were off-target.31 Thus, of the 40% of hap- boring the dsODN can be selectively these studies have been done in cell lines loid genomes with AAV vector integration, amplified, sequenced, and mapped back derived from tumors. Therefore, valida- only 1.2% had NHEJ-integrated vectors at to the genome to precisely identify DSBs tion will still have to be performed in cell the knockin site. In addition, they exam- to the nucleotide level. GUIDE-seq pro- lines that are the clinical target. Another ined the number of AAV integrations seen files of 10 different gRNAs show that complexity is that genetic polymor- in cells that were transduced with an AAV the number of off-target DSBs can vary phisms may make it more difficult to vector expressing luciferase compared to widely from more than 150 to none de- predict the potential off-target effects for the cells transduced with the ZFN-AAV tectable. This method provides the first any individual in human applications. vector and detected fewer AAV sites in genome-wide method for defining DSBs The ideal would be to have an assay that the ZFN-AAV transduced cells, suggesting induced by CRISPR/Cas9 nucleases and could assess the potential for off-target that the ZFN-AAV may have some cellular should provide an important tool for sites not only in a particular cell type toxicity compared to the AAV-luciferase. preclinical evaluation of the specificities but perhaps even in an individual’s cells Another set of studies examined chro- of these reagents. This method seems to (e.g., the GUIDE-seq approach could be mosomal translocations generated with be more sensitive than the prior viral performed in the patient’s cells). While exposure to different nucleases, not just based methods of capturing off-target this may be the ideal, it is not necessar- ZFNs. In the event of simultaneous on- DSBs, because significantly more oligo- ily a prerequisite for proceeding to clini- target and off-target breaks, these may nucleotide can be introduced into the cal studies if the cumulative evidence in re-ligate, creating the potential for translo- cell. Since this method is sensitive and a relevant model provides a favorable cations, deletions, or inversions. Dr Bush- does not require any specialized exper- risk–benefit ratio. man described a series of experiments tise in viral production—although it in which cells were exposed to ZFNs, does demand significant bioinformatic Chromosomal rearrangements: TALENs, or CRISPR that targeted differ- expertise to sort out true off-target will they occur and should we be ent genes—those encoding CCR5, VEGF, DSBs from noise—it is likely to become concerned? and β-globin. Each of these nucleases was an important approach to assessing the What are the potential effects of these off- shown to have the expected on-target ef- specificity of any given nuclease or its target breaks? Dr Bushman and David fect. However, they also saw high levels variants. Roth presented their research on chro- of translocations in all of the nuclease- Another assay that has been used to mosomal rearrangements, which may exposed cells and fewer in controls trans- detect DSBs utilizes the gamma-H2AX lead to more severe toxicities than those duced with a GFP-expressing vector. It histone protein. A DSB occurring in this caused by indels. Dr Bushman described remains to be determined to what degree region leads to phosphorylation of H2AX work by Kathy High, president and chief these translocations would have clinically and the formation of a gamma-H2AX scientific officer of Spark Therapeutics, us- significant effects. protein. An antibody to gamma-H2AX ing adeno-associated virus (AAV) vectors To further examine the significance is available and therefore can be used to to deliver ZFNs to correct factor IX in a of translocations, Dr Roth has focused on quantify DSBs that result from off-target hemophilia mouse model. In this experi- well-known naturally occurring nucleases, nuclease activity.30 It is important to note ment, the ZFN nuclease was used to cleave RAG1 and RAG2, that mediate recombi- that there is a background level of gam- the factor IX target site, and then a second nation of the gene segments in T cells to ma-H2AX due to spontaneous DSBs. In AAV vector delivered the wild-type ex- create a diverse repertoire of TCRs, known K562 cells, Matthew Porteus’s lab evalu- ons to recombine into the targeted break as V(D)J recombination. The RAG nucle- ated several different constructs using made by the ZFNs.31 The AAV provided a ase has considerable specificity, cleaving this method, bringing about the binding marker from which to sequence out into only certain sites, known as recombination of (i) a pair of ZFNs to two different tar- the flanking DNA to determine where the signal sites (RSSs).32 Rejoining of the DSB get sites, (ii) a pair of TALENs to three breaks occurred, much like the work of Dr occurs by NHEJ. The system is not perfect,

Molecular Therapy vol. 23 no. 5 may 2015 801 meeting report © The American Society of Gene & Cell Therapy

and occasionally an authentic cleavage site how these DSBs interact with any fragile will be reduced replication or death of is joined to closely related off-target sites, sites that also exist in the cell type of in- the GFP+ cells, leading to a decline in known as cryptic RSSs. More recently, terest. The creation of translocations be- the relative percentage of GFP+ cells over NEHJ has been detected between two off- tween the intended nuclease target and time. This provides a quantitative assess- target events (i.e., breaks made in two dif- random DSBs on other chromosomes has ment of toxicity. When this assay was ferent cryptic RSSs), resulting in leukemia. also been confirmed using a LAM-PCR used with selected ZFNs and TALENs, Dr Roth described an examination of high-throughput, genome-wide, translo- cell viability was inversely related to the the chromosomal abnormalities in a mouse cation sequencing approach.35 These two dose, suggesting that nuclease concen- that had a mutant p53 gene, making it methods of detecting engineered nucle- tration may affect viability. To examine prone to tumors, but expressing wild-type ase-induced chromosomal translocations the effect of nuclease concentration on RAG. In addition to chromosomal chang- are likely to become an important new ap- cell viability in primary stem cells, ZFNs es that would be expected in the mutant proach to assessing the safety of a genome specific for enhanced GFP (eGFP) have p53 background, some tumors seemed to editing process and raises the concern that been employed in keratinocyte stem arise from RAG-mediated translocations, targeting a site that is associated with can- cells derived from an eGFP-transgenic in that they were located near known cryp- cer translocations might have an increased neonatal mouse. ZFN on-target activity tic RSSs. A number of these breaks near safety risk. should eliminate GFP activity. At differ- these cryptic RSSs resulted in deletions ent ZFN doses, high levels of on-target that activated oncogenes. Two recent ar- Functional toxicity assays: activity were seen, although as reported ticles showed that RAG nucleases could be genotoxicity, and cytotoxicity by others, cytotoxicity did increase with the driver of leukemia through off-target Identifying off-target sites and potential a higher dose of ZFNs.36 Under optimal cuts resulting in recombination and dele- chromosomal rearrangements is critical to conditions, the stem cell potential of the tions.33,34 The relevance of this finding to assess the safety of new constructs. For de- keratinocyte stem cells was not altered work with other nucleases is that even with veloping clinical applications, the question by the ZFNs. a well-conserved, specific nuclease, there was raised as to which studies would be GFP tagging can also be used to de- is the potential for off-target activity to most useful for evaluating clinical toxic- tect clonal dominance, which is often, have significant biological consequences. ity. Studies that may be able to identify all but not always, a precursor of genotoxic- Engineered nucleases with arguably genome alterations, including those that ity. Cells were marked with a “barcode” lower specificity might likewise lead to have not been correlated with any cellu- (a short nucleotide sequence that pro- unpredicted and significant genomic rear- lar toxicity or clinical adverse effects, may vides a unique identifier for each cell, rangements. In order to determine wheth- provide important scientific knowledge either by lentiviral transduction or by er this might occur with TALENs, Dr Roth but may not necessarily be most relevant ZFN-mediated targeted integration). By used TALENs designed to correct the for preclinical development. Although using deep sequencing, the clonal dy- β-globin locus. Genome-wide sequencing these basic science studies may be at the namics of the population can be stud- revealed deletions that were of equivalent extreme edge of sensitivity, functional ied over time. When the barcodes were size to those seen in some of the RAG-in- studies are needed to validate potential introduced semi-randomly by lentiviral duced tumors. A number of translocations toxicity for preclinical development. vector integration, there was evidence were also found in chromosome 11 where Dr Cathomen noted that concep- the hemoglobin gene is located. Although tually for a given concentration of a this was a limited analysis, the discovery nuclease, you will have on-target and that off-target activity by RAG can lead to off-target activity. The ideal concentra- tumors and the discovery of similar types tion will be one in which you have high of translocations with a β-globin-specific on-target activity but low off-target ac- TALEN indicate that deletions resulting tivity (Figure 3, bottom). If a nuclease from off-target genome breaks may be bio- has low specificity, then the two curves logically significant. are closer to one another and the abil- In recent work, Dr Joung and col- ity to reach an effective dose without leagues reported that translocations can toxicity is limited (Figure 3, top). There occur between CRISPR/Cas9-induced are two main toxicity concerns specific Figure 3. Plots showing the relationship on-target and off-target DSBs identified to genome editing technologies: cyto- between concentration of a given nucle- .19 ase and on-target vs. off target activity. by GUIDE-seq Interestingly, these inves- toxicity and genotoxicity. A number of A “good” nuclease should exhibit high specific- tigators also observed that translocations approaches are being developed to mea- ity and affinity for its cognate binding site such + that the mass action equilibrium will not shift in could occur between nuclease-induced sure cytotoxicity. In one assay, GFP cells favor of off-target sites with small increases of DSBs and nuclease-independent DSB are used to track cell viability. Cells are nuclease concentrations. “Bad” nucleases with hotspots also identified by GUIDE-seq. co-transfected with a GFP expression low specificity are prone to bind more off-target sites with small changes in nuclease concentra- This latter observation suggests that it is plasmid and a nuclease expression plas- tion and thus limit the ability to reach an effec- important to consider not only the breaks mid (not all cells are transfected). If the tive non-toxic dose. Courtesy of Toni Cathomen. caused by exogenous nucleases but also nuclease activity results in toxicity, there

of spontaneous clonal changes in the They further examined the effect on the is a precursor or stem cell, animal mod- population but none that could be re- cell cycle. Extensive cleavage at off-target els can support cell differentiation, which producibly attributed to the lentiviral sites would probably result in arrest of the may not be possible in vitro, and thereby insertion. In contrast, the population of cell cycle until the DSBs are repaired. Us- make it possible to more rigorously as- cells marked by ZFN-mediated recom- ing HeLa FUCCI cells that express fluo- sess functionality. Furthermore, the cell bination at a single safe harbor showed rescently tagged cell cycle indicators, they expansion that occurs in vivo can amplify greater clonal skewing, with the clonal again compared the ZFNs to the TALENs. genotoxic events and allow the outgrowth dominance occurring reproducibly from Compared with control, the TALENs did of tumorigenic cells resulting from a very the same clones.37 It is unclear why the not affect the cell cycle while the ZFNs did rare event. However, the use of animal integration of the ZFNs led to additional lead to more cells blocked at the G2 stage; models to assess DNA sequence–specific clonal growth, but the use of K562 cells after 3 days, more apoptosis was observed reagents such as nucleases can be com- and high doses of ZFNs may have affect- in those cells where the cycling was dis- plicated by differences between the hu- ed these results. Furthermore, although rupted. man and animal target sequences, so that this assay suggests that changes in clonal What can be concluded about these the reagents used may not recognize the dynamics induced by engineered nucle- results, especially in light of at least analogous nonhuman sequences. As one ases can be detected, the clinical rel- one ZFN against CCR5 moving into approach to circumvent this problem, the evance has not been validated. the clinic successfully with no evidence gene therapy field has adopted the use of Dr Cathomen’s group examined the of clinical toxicity and several years of immunodeficient mice that can support ratio of on-target to off-target events us- follow-up? Dr Cathomen noted that the engraftment and differentiation of ing ZFNs and TALENs specific for CCR5 these results cannot be interpreted as human hematopoietic systems, and these and AAVS1. When assaying the geno- ZFNs being generally less specific than have been used to assess genotoxicity of toxicity of ZFN and TALEN pairs target- TALENs. Rather, it will be important to integrating vectors in these cell tissues.39 ing CCR5, they initially focused on the carefully evaluate the genotoxic poten- Such an approach may now also be used known primary off-target site, CCR2. tial of every designer nuclease intended to assess endonuclease-based approaches Overexpression of a highly specific to enter clinical trials. Dr Porteus noted in hematopoietic stem cells. TALEN pair did not result in detectable that these assays raise interesting ques- Paula Cannon reviewed some key chromosomal rearrangements; however, tions, but there is no evidence that they elements of these types of experiments. these were detected following use of the have predictive power for clinical out- Because the recipient mouse strains are CCR5-specific ZFN. The ratio of CCR5/ comes. The majority of these assays have immunodeficient, if tumors do arise in CCR2 specificity for the different nucle- been done in cell lines using delivery such mice, investigators must character- ases was determined. The CCR5-specific strategies and/or doses that would not be ize these tumors and evaluate whether TALENs had a specificity ratio of either used in clinical applications. Dr Catho- they are of human origin. If tumors are 130:1 or 7:1, depending upon the target men added that the other assays being of human origin, then it will be neces- sequence, whereas for the CCR5-specific used to assess genotoxicity—karyotype sary to further evaluate their clonality ZFNs, it was 3:1. This difference in spec- analysis and array-comparative genomic with respect to a nuclease modification ificity paralleled the detection of large hybridization—are relatively insensitive. signature at any of the on- or off-target chromosomal deletions or inversions in They may best serve as screening and sites. However, clonality observed in a the area of the CCR5/CCR2 loci, where comparison tools between platforms or nuclease-modified cell does not neces- a highly specific TALEN pair did not in- as refinements to current platforms. If sarily equal causality and may instead be duce chromosomal rearrangements, but they are to be adapted to provide data an innocent label that merely reflects the CCR5-specific ZFNs and the TALEN regarding functional toxicity for clinical tumor’s clonal origin. pair that had the lower ratio of specifici- applications, they will have to be con- While the models developed for eval- ty demonstrated high off-target effects.38 ducted in the cell types of interest, with uating mutagenic potential for integrat- Using the bioinformatics tools described the nuclease delivered as it would be in ing vectors can be adopted for nucleases, above (PROGNOS), they found that the a clinical trial. at least for nuclease-modified hematopoi- predicted ratio of on-target to off-target etic stem cells or T cells,40 an important sites for the ZFN pairs targeting CCR5 The role of animal models caveat is that these models have not been and AAVS1 was about 1:2, and the ratio Although in vitro models are important able to recapitulate the clinical tumors for the CCR5-specific TALENs was 60:1; to guide development of new approaches, seen in the human trials with integrat- and for the TALENs targeting AAVS1 the animal models have always been used to ing viral vectors. Although insertions ratio was 27:1. These constructs were then more definitively explore toxicity. In the near oncogenes can be documented, the tested at high doses in HEK293 cells, high- context of these agents, animal models po- mouse models have not demonstrated er than would be done for physiological tentially allow for assessment of the viabil- the vector-driven leukemia seen in some dosing. About 80% of the cells transfected ity and functionality of the modified cells subjects. The lack of toxicity in animal with TALENs expression vectors survived in an environment in which they will com- models will be an important safety check, compared with about half of cells trans- pete with unmodified cells. In addition, as one would not want to proceed in the fected with the ZFN-encoding plasmids.38 for applications where the engineered cell face of animal toxicity. However, the chal-

lenges in developing appropriate, efficient vector into mature T cells. To determine the target. Immunostaining for DSBs was also animal models to evaluate genotoxicity impact of this nuclease on the T cells, San- done using p53-binding protein1 (53BP1), and in particular oncogenicity for inte- gamo undertook a series of assays in which which is recruited to sites of DSBs early grating vectors have been documented,41 the phenotype and the growth kinetics of in their repair and is required for NHEJ.43 and similar challenges may arise as these the modified T cells were compared to those Importantly, there is a background rate of models are used to evaluate genome edit- of unmodified T cells. Cytokine release was positive sites reflecting the physiological in- ing technologies. also determined to be the same in modified cidence of DSBs. Transduction of the cells versus unmodified T cells, and there was no using the adenoviral vector carrying the Bringing it all together: moving skewing of the diversity of the TCR vari- CCR5-specific ZFNs resulted in an increase into the clinic able domain. Finally, it was determined that in DSBs of 1.4–1.6% compared with 4.1% in At the time of this meeting, only ZFNs have the CCR5-modified T cells are stable in the the presence of a chemotherapy agent eto- advanced to clinical trials. The first success- population in the absence of HIV infection, poside. Because CCR2 is in close proximity ful clinical application for genome editing and HIV infection leads to enrichment of to CCR5, it was not possible to visualize two has been seen with a ZFN targeting CCR5, the CCR5-modified T cells. independent 53BP1 foci by staining.43 a receptor expressed on T cells that allows In addition to these experiments, it was Following these molecular assays, a se- HIV to enter.2 Phil Gregory and Dale Ando, important to analyze the fidelity of the ge- ries of in vitro and in vivo assays were con- Sangamo BioSciences, reviewed their path nome editing. The first step was to focus ducted to establish safety and in particular, to the clinic. Development of this product on molecular assays that could identify pri- the absence of oncogenic potential. In vitro began in 2003 when many of the assays mary off-target action, i.e., DSBs. Although oncogenicity assays were based on the expe- reviewed during the meeting were not yet bioinformatics tools were a starting point, rience in previous gene therapy T-cell prod- available. Dr Gregory noted that the path an unbiased approach was also needed. uct characterizations. In addition, because taken for these products is in some ways the One can deep-sequence sites that are the the cancer chemotherapy cytotoxic agents same for any pharmaceutical development, closest matches to the consensus site for the have the greatest genotoxic and carcinogen- including generating therapeutic reagents ZFNs binding sites, as identified using bio- ic potential, standard in vitro studies for pre- that are maximized for potency/specificity informatics tools. This deep sequencing can clinical evaluations of these types of drugs and working within established regulatory detect events as rare as 1:10,000 alleles but were also used. The ability of a primary T frameworks to characterize the safety of is limited by the fact that the initial screen cell to grow without cytokines and cell sig- the product. For genome editing, an addi- is identifying a relatively small number of naling is a feature of carcinogenic transfor- tional challenge that is unique to this class sites per sequence. One can also use im- mation. Culturing of SB-728-T cells without of “drugs” is the need to define specificity in munostaining for DSBs, which is unbiased cytokine support was performed for weeks addition to classical toxicology assessment. but again limited to about 100 individual and demonstrated that normal cell death Also, identification of off-target sites is com- nuclei per conditional time point. As dis- occurred. The classic biological cell trans- plicated by the lack of a clear footprint in the cussed above, genome-wide assessment of formation assay is anchorage-independent genome, unlike those of integrating vectors, ZFNs-induced DSBs using IDLV capture growth of fibroblasts and is a stringent test which could be easily detected. and nonrestrictive LAM-PCR is another of carcinogenesis. These fibroblasts are also One of the first steps with this platform approach, but it is limited by the sensitivity amenable to gene transfer by adenovirus, so development was to maximize the speci- of IDLV capture at rare DSBs. Nonrestric- delivery of the ZFNs into these cells can be ficity of the ZFNs. As discussed above, the tive LAM-PCR does not utilize restriction achieved. The US Food and Drug Admin- ZFN modules each recognize three base enzymes to cleave the genomic DNA, which istration (FDA) asked Sangamo to use as pairs. These modules can be combined so may reduce sensitivity to detect some sites, high a multiplicity of infection as possible that the interface is highly specific. In addi- but its use of sonication shearing should for these tests. Other tests evaluated but not tion, the linkers between the modules and increase sensitivity. Finally, karyotyping of chosen to evaluate for oncogenicity were the links between the Fok1 nuclease and cells can reveal genomic rearrangement, but the Ames test, Mouse micronucleus test, the modules can be altered to maximize en- the number of cells analyzed per sample is and mouse lymphoma TK gene mutation gagement of the preferred sequence. More- low. In addition, nonclonal rearrangements assay. over, the Fok1 domains can be engineered are often present in untreated cells, making SB-728-T in vivo safety studies relied on to require heterodimer binding. Because of it important to also determine the back- the fact that human T cells could be main- these variables, up to 105 ZFN dimers can be ground incidence. tained in immunodeficient NOG mice. generated for a particular exon region. The In the development of the CCR5 prod- After discussion with the FDA, Sangamo selection of a candidate to move forward uct, Sangamo identified 15 potential off- tested a full human dose of modified T cells into the clinic required the use of bioinfor- target sites and then used deep sequencing (using T cells from three different donors), matics and selection technologies such as (454 sequencing) to look for DSBs at the allowed the modified human T cells to live phage display42 to identify the product that identified sites. This allowed for identifi- and expand for months in the NOG model, is maximized for specificity and activity. cation of one off-target site in 1:20,000 al- and tested this against T cells modified us- The ZFN-CCR5 used in clinical tri- leles, and the identification of one other ing a maximal multiplicity of infection in als developed by Sangamo, known as site, CCR2, that had about a 4% frequency order to recreate a “worst case” scenario. SB-728-T, was delivered by an adenoviral of DSBs compared with ~35% at the CCR5 It took some time to develop a model with

human T-cell xeno-GVHD, defining that 2 activity. to allogeneic transplants—the standard of months was a maximal time for prolifera- One question is whether it is necessary care—then the risk of an off-target event tion of cells before animals died of GVHD, to detect every potential indel. Is there a that might have an unforeseen clinical out- and defining a dose and donors that gave re- level below which one could be confident come should be weighed against the known liable GVHD in the NOG mice. Three stud- that the frequency would not lead to a clini- 20–30% risk of GVHD and a 5–10% risk of ies were performed (one for each donor), cally significant lesion? Does a focus on the death. with a duration of 2 months, and then ani- most frequent off-target sites, perhaps with Another safety feature may be the mals were euthanized; all organs were eval- a particular focus on translocations, suffi- choice of initial cells to target with these uated by histology for neoplasms, immu- ciently help ensure safety? It is unlikely that tools. Sangamo Biosciences conducted their nostaining to detect human cells, and PCR the nature of the repair product can reliably initial ZFNs clinical trial in a terminally for detection of ZFNs CCR5 modification predict function, as not all deletions are differentiated T cell. New trials are explor- at on-target and off-target sites. Histology benign; moreover, although translocations ing the same construct in stem cells. If one was consistent with GVHD in liver, lung, are frequent in tumors, there may be many looks at the experience in gene therapy, the gut, and spleen, showing intense inflamma- translocations that do not lead to transfor- same type of retroviral vector that caused tory infiltration. Human DNA was present mation events. Therefore, it may not be nec- leukemia when used in hematopoietic stem in all animals, and there was equivalent en- essary to identify all off-target activity but cells in trials for X-linked severe combined graftment of modified and control cells. No rather to develop assays that reliably iden- immunodeficiency, chronic granulomatous CCR5-related CD4 T-cell neoplasms were tify transformation of cells. In addition, as disease, and Wiskott-Aldrich syndrome, found. the specificity of these nucleases improves, has not led to leukemia in other protocols These studies led to the first genome ed- the frequency of off-target DSBs may de- that have used terminally differentiated T iting phase I trial conducted by Pablo Tebas crease to a level that may reasonably allow cells, even though the ability to transform and Carl June at the University of Penn- the analyses to focus on the most prevalent T cells in a preclinical setting has been sylvania.2 In that study, administration of of the off-target sites. demonstrated.44 Monitoring the behavior CCR5-modified T cells was safe, marked in- Assay development may be an essential of these nucleases in differentiated cells, creases in total CD4 T cells were observed, milestone in the development pathway be- and looking for normal activity and dif- and the modified T cells trafficked normally fore a commitment is made to a particular ferentiation, provides some data that these to the gut. One subject achieved controlled engineered nuclease. One could begin with nucleases are not disrupting cellular activity HIV viral load below levels of detection af- using biochemistry/bioinformatics/molec- through action at other loci. However, the ter an antiviral treatment interruption of 12 ular assays to evaluate candidate nucleases counterargument to this is that safety in the weeks. The ZFNs-modified T cells persisted and their potential for off-target and cellu- differentiated cell does not guarantee safety at a level >50 modified CD4 T cells/μl for lar toxicity. These results may then be used in the stem cell. 252 weeks. to optimize the products. Once the optimal Several new investigators in this field specificity has been engineered and any off- come from the gene therapy field, and in Putting it all together for future target sites identified, there will be a need particular the area that uses integrating clinical applications for more classical measurements of toxicity vectors to accomplish long-term gene cor- The meeting reviewed a number of assays and genotoxicity, both in vitro and in vivo. rection. The genome editing field draws on that are being developed but are yet to be Importantly, these assays may have to be the experience of those studying integrat- validated. The question remains how to best adapted to focus not only on the nuclease ing vectors. However, there are limits in the integrate these assays into the preclinical but also on the proposed method of vector ability to extrapolate from the experience development strategy. A goal for the field transduction and the cell context. with integrating vectors and their assays. may be to build assays prospectively for a On a more practical level, this is a rapid- For example, in the field of gene therapy, particular outcome, defining the sensitivity ly developing field, with new ways to assess one group looked at integration sites for and cutoff values in advance. It was noted the activity of these constructs emerging at the same vector in T cells vs. stem cells and that although the meeting proposed to fo- an equally fast pace. Ultimately decisions found different integration patterns. How- cus on “establishing preclinical toxicology will need to be made regarding which assays ever, when the ZFN for CCR5 was exam- standards,” the main focus was on detecting are scientifically valid and which are needed ined in T cells and hematopoietic stem cells, and evaluating off-target effects. It was ac- to satisfy the legal-regulatory framework. there were no differences in the off-target knowledged that one way to evaluate these There are a number of assays that may sites for the CCR5 in T or CD34+ cells, un- technologies is using an unbiased approach provide some data on safety, but each has derscoring that while vector integration is such as whole-genome sequencing to un- limitations in terms of sensitivity. When as- largely random, the action of these products derstand where these nucleases act in the sessing the impact of rare events, the ques- is more directed, including any off-target genome, but the cost and potentially low tion is often how many negative readouts events. sensitivity makes this approach impracti- are needed to provide confidence that one It is as yet unclear if the preclinical path- cal for preclinical development. Instead, the can proceed into the clinic. This risk–ben- way for these technologies will differ de- goal has been to find predictive tools that efit calculus must also take into account pending upon whether the goal is gene dis- allow for a more focused evaluation of the the proposed disease target. For example, ruption, correction, or insertion. Certainly most likely off-target sites with significant if these approaches will offer alternatives there would be additional challenges when

REFERENCES 24. Papapetrou, EP, Lee, G, Malani, N, Setty, M, Riviere, moving beyond gene disruption—which 1. Carroll, D (2014). Genome engineering with targe- I, Tirunagari, LM et al. (2011). Genomic safe harbors has been the focus of clinical applications table nucleases. Annu Rev Biochem 83: 409–439. permit high beta-globin transgene expression in 2. Tebas, P, Stein, D, Tang, WW, Frank, I, Wang, SQ, thalassemia-induced pluripotent stem cells. Nat to date—to gene replacement. Gene cor- Lee, G et al. (2014). Gene editing of CCR5 in autolo- Biotechnol 29: 73–78. rection faces the challenge that DSBs could gous CD4 T cells of persons infected with HIV. N Engl 25. Sadelain, M, Papapetrou, EP and Bushman, FD J Med 370: 901–910. (2012). Safe harbours for the integration of new DNA also be repaired by NHEJ; therefore, if this 3. Gaj, T, Gersbach, CA and Barbas, CF 3rd (2013). in the human genome. Nat Rev Cancer 12: 51–58. ZFN, TALEN, and CRISPR/Cas-based methods for 26. Tsai, SQ, Iafrate, AJ and Joung, JK (2014). Genome approach is to be safe, site selection must be genome engineering. Trends Biotechnol 31: 397–405. editing: a tool for research and therapy: towards a such that disruption would not lead to ad- 4. Bogdanove, AJ and Voytas, DF (2011). TAL effectors: functional understanding of variants for molecular customizable proteins for DNA targeting. Science diagnostics using genome editing. Nat Med 20: verse effects. 333: 1843–1846. 1103–1104. It remains to be determined whether 5. Cermak, T, Starker, CG and Voytas, DF (2015). Ef- 27. Snyder, M, Du, J and Gerstein, M (2010). Personal ficient design and assembly of custom TALENs using genome sequencing: current approaches and chal- genome editing to achieve gene addition the Golden Gate platform. Methods Mol Biol 1239: lenges. Genes Dev 24: 423–431. 133–159. 28. Lin, Y, Cradick, TJ, Brown, MT, Deshmukh, H, Ranjan, will be safer compared to gene therapy 6. Reyon, D, Maeder, ML, Khayter, C, Tsai, SQ, Foley, P, Sarode, N et al. (2014). CRISPR/Cas9 systems have approaches using randomly integrating JE, Sander, JD et al. (2013). Engineering customized off-target activity with insertions or deletions be- TALE nucleases (TALENs) and TALE transcription fac- tween target DNA and guide RNA sequences. Nucleic vectors. Although there is considerable tors by fast ligation-based automatable solid-phase Acids Res 42: 7473–7485. experience with gene delivery via integrat- high-throughput (FLASH) assembly. Curr Protoc Mol 29. Pattanayak, V, Ramirez, CL, Joung, JK and Liu, DR Biol Chapter 12: Unit 12.16. (2011). Revealing off-target cleavage specificities of ing vectors, efficient gene delivery into a 7. Sander, JD and Joung, JK (2014). CRISPR-Cas systems zinc-finger nucleases by in vitro selection. Nat Meth- predetermined site using nuclease technol- for editing, regulating and targeting genomes. Nat ods 8: 765–770. Biotechnol 32: 347–355. 30. Pruett-Miller, SM, Connelly, JP, Maeder, ML, Joung, JK ogy and homologous recombination is still 8. Jinek, M, Chylinski, K, Fonfara, I, Hauer, M, Doudna, and Porteus, MH (2008). Comparison of zinc finger JA and Charpentier, E (2012). A programmable dual- nucleases for use in gene targeting in mammalian being developed. The therapy that will pre- RNA-guided DNA endonuclease in adaptive bacterial cells. Mol Ther 16: 707–717. vail may not be the most elegant but must immunity. Science 337: 816–821. 31. Li, H, Haurigot, V, Doyon, Y, Li, T, Wong, SY, Bhag- 9. Silva, G, Poirot, L, Galetto, R, Smith, J, Montoya, G, wat, AS et al. (2011). In vivo genome editing restores be proven to be safe, effective, and easy to Duchateau, P et al. (2011). Meganucleases and other haemostasis in a mouse model of haemophilia. implement at multiple manufacturing sites, tools for targeted genome engineering: perspectives Nature 475: 217–221. and challenges for gene therapy. Curr Gene Ther 11: 32. Sadofsky, MJ (2001). The RAG proteins in V(D)J re- and provide an advantage over treatments 11–27. combination: more than just a nuclease. Nucleic Acids 29 or approaches that currently exist. Targeting 10. Brinkman, EK, Chen, T, Amendola, M and van Res : 1399–1409. Steensel, B (2014). Easy quantitative assessment of 33. Papaemmanuil, E, Rapado, I, Li, Y, Potter, NE, Wedge, expression from a specific site, even if well genome editing by sequence trace decomposition. DC, Tubio, J et al. (2014). RAG-mediated recom- Nucleic Acids Res 42: e168. bination is the predominant driver of oncogenic characterized, may not provide more effica- 11. Hendel, A, Kildebeck, EJ, Fine, EJ, Clark, JT, Punjya, N, rearrangement in ETV6-RUNX1 acute lymphoblastic cious levels of expression than the expres- Sebastiano, V et al. (2014). Quantifying genome- leukemia. Nat Genet 46: 116–125. editing outcomes at endogenous loci with SMRT 34. Mendes, RD, Sarmento, LM, Cante-Barrett, K, Zuur- sion from multiple random sites achieved sequencing. Cell Rep 7: 293–305. bier, L, Buijs-Gladdines, JG, Povoa, V et al. (2014). with integrating vectors. 12. Gilbert, LA, Larson, MH, Morsut, L, Liu, Z, Brar, GA, PTEN microdeletions in T-cell acute lymphoblastic Torres, SE et al. (2013). CRISPR-mediated modular leukemia are caused by illegitimate RAG-mediated There remains a gap in our current un- RNA-guided regulation of transcription in eukaryotes. recombination events. Blood 124: 567–578. Cell 154: 442–451. 35. Frock, RL, Hu, J, Meyers, RM, Ho, YJ, Kii, E and Alt, derstanding of certain aspects of these tech- 13. Wang, J, Friedman, G, Doyon, Y, Wang, NS, Li, CJ, FW (2015). Genome-wide detection of DNA double- nologies that may be important in evaluat- Miller, JC et al. (2012). Targeted gene addition to stranded breaks induced by engineered nucleases. a predetermined site in the human genome using Nat Biotechnol 33: 179–186. ing clinical applications. For example, in a ZFN-based nicking enzyme. Genome Res 22: 36. Hoher, T, Wallace, L, Khan, K, Cathomen, T and developing new TALENs and CRISPRs, 1316–1326. Reichelt, J (2012). Highly efficient zinc-finger 14. Beumer, KJ and Carroll, D (2014). Targeted genome nuclease-mediated disruption of an eGFP transgene when a reagent does not work at the ex- engineering techniques in Drosophila. Methods 68: in keratinocyte stem cells without impairment of pected target, it has been relatively easy to 29–37. stem cell properties. Stem Cell Rev 8: 426–434. 15. Elliott, B, Richardson, C, Winderbaum, J, Nickoloff, 37. Porter, SN, Baker, LC, Mittelman, D and Porteus, MH adjust the sequence and manufacture a new JA and Jasin, M (1998). Gene conversion tracts from (2014). Lentiviral and targeted cellular barcoding double-strand break repair in mammalian cells. Mol reveals ongoing clonal dynamics of cell lines in vitro one rather than exploring why the agent did Cell Biol 18: 93–101. and in vivo. Genome Biol 15: R75. not work. 16. Stark, JM and Jasin, M (2003). Extensive loss of het- 38. Mussolino, C, Alzubi, J, Fine, EJ, Morbitzer, R, erozygosity is suppressed during homologous repair Cradick, TJ, Lahaye, T et al. (2014). TALENs facilitate An ultimate goal would be to establish of chromosomal breaks. Mol Cell Biol 23: 733–743. targeted genome editing in human cells with high regulatory pathways requiring well-under- 17. Fu, Y, Foden, JA, Khayter, C, Maeder, ML, Reyon, D, specificity and low cytotoxicity. Nucleic Acids Research Joung, JK et al. (2013). High-frequency off-target 42: 6762–6773. stood, standard assays, so that one could mutagenesis induced by CRISPR-Cas nucleases in 39. Zhou, S, Ma, Z, Lu, T, Janke, L, Gray, JT and Sor- human cells. Nat Biotechnol 31: 822–826. rentino, BP (2013). Mouse transplant models for sequence the genome of a patient or infant, 18. Fu, Y, Reyon, D and Joung, JK (2014). Targeted evaluating the oncogenic risk of a self-inactivating identify a target, make sequence-specific genome editing in human cells using CRISPR/Cas XSCID lentiviral vector. PloS One 8: e62333. nucleases and truncated guide RNAs. Methods Enzy- 40. Maier, DA, Brennan, AL, Jiang, S, Binder-Scholl, GK, reagents, and (because of the similarity to mol 546: 21–45. Lee, G, Plesa, G et al. (2013). Efficient clinical scale the approaches that have been previously 19. Tsai, SQ, Zheng, Z, Nguyen, NT, Liebers, M, Topkar, gene modification via zinc finger nuclease-targeted VV, Thapar, V et al. (2014). GUIDE-seq enables disruption of the HIV co-receptor CCR5. Hum Gene successful in the clinic) develop a person- genome-wide profiling of off-target cleavage by Ther 24: 245–258. alized reagent that is ready for clinical use CRISPR-Cas nucleases. Nat Biotechnol 33: 187–197. 41. Corrigan-Curay, J, Cohen-Haguenauer, O, O’Reilly, 20. Guilinger, JP, Thompson, DB and Liu, DR (2014). M, Ross, SR, Fan, H, Rosenberg, N et al. (2012). in 2–3 months. The development of such a Fusion of catalytically inactive Cas9 to FokI nuclease Challenges in vector and trial design using retroviral improves the specificity of genome modification. Nat vectors for long-term gene correction in hematopoi- well-defined validated process may be the Biotechnol 32: 577–582. etic stem cell gene therapy. Mol Ther 20: 1084–1094. ultimate path for precision medicine. 21. Ran, FA, Hsu, PD, Lin, CY, Gootenberg, JS, Koner- 42. Greisman, HA and Pabo, CO (1997). A general mann, S, Trevino, AE et al. (2013). Double nicking strategy for selecting high-affinity zinc finger proteins by RNA-guided CRISPR Cas9 for enhanced genome for diverse DNA target sites. Science 275: 657–661. ACKNOWLEDGMENTS editing specificity. Cell 154: 1380–1389. 43. Perez, EE, Wang, J, Miller, JC, Jouvenot, Y, Kim, KA, 22. Farboud, B and Meyer, BJ (2015). Dramatic enhance- Liu, O et al. (2008). Establishment of HIV-1 resistance The authors thank all the participants for their + ment of genome editing by CRISPR/Cas9 through in CD4 T cells by genome editing using zinc-finger contributions and Dale Ando for assistance in improved guide RNA design. Genetics, e-pub ahead nucleases. Nat Biotechnol 26: 808–816. summarizing his presentation. The views ex- of print 18 February 2015. 44. Newrzela, S, Cornils, K, Heinrich, T, Schlager, J, Yi, JH, pressed in this report are those of the individual 23. Boissel, S, Jarjour, J, Astrakhan, A, Adey, A, Gouble, A, Lysenko, O et al. (2011). Retroviral insertional muta- authors and do not represent the views of the Duchateau, P et al. (2014). megaTALs: a rare-cleaving genesis can contribute to immortalization of mature T lymphocytes. Mol Med 17: 1223–1232. NIH. nuclease architecture for therapeutic genome engineering. Nucleic Acids Res 42: 2591–2601.

806 www.moleculartherapy.org vol. 23 no. 5 may 2015